From 029dc67524ad1d2fc9737de4824810e61ce201b0 Mon Sep 17 00:00:00 2001 From: leblane Date: Fri, 28 Apr 2023 15:47:06 +0300 Subject: [PATCH] Initial code commit --- LICENCE.txt | 109 + README.md | 14 + Source/HalfFloat.cpp | 1039 +++ Source/HalfFloat.h | 8 + Source/KHR/khr_df.h | 619 ++ Source/Main.cpp | 619 ++ Source/createdfd.cpp | 659 ++ Source/dfd.h | 173 + Source/ispc_texcomp/ispc_texcomp.cpp | 557 ++ Source/ispc_texcomp/ispc_texcomp.def | 30 + Source/ispc_texcomp/ispc_texcomp.h | 128 + Source/ispc_texcomp/ispc_texcomp.vcxproj | 177 + .../ispc_texcomp/ispc_texcomp.vcxproj.filters | 62 + Source/ispc_texcomp/ispc_texcomp_astc.cpp | 564 ++ Source/ispc_texcomp/kernel.ispc | 3798 ++++++++ Source/ispc_texcomp/kernel_astc.ispc | 2272 +++++ Source/meson.build | 26 + Source/stb_image.cpp | 2 + Source/stb_image.h | 7987 +++++++++++++++++ Source/stb_image_resize.cpp | 2 + Source/stb_image_resize.h | 2634 ++++++ Source/vk2dfd.cpp | 33 + Source/vk2dfd.inl | 294 + build/.keep | 0 meson.build | 3 + 25 files changed, 21809 insertions(+) create mode 100644 LICENCE.txt create mode 100644 README.md create mode 100644 Source/HalfFloat.cpp create mode 100644 Source/HalfFloat.h create mode 100644 Source/KHR/khr_df.h create mode 100644 Source/Main.cpp create mode 100644 Source/createdfd.cpp create mode 100644 Source/dfd.h create mode 100644 Source/ispc_texcomp/ispc_texcomp.cpp create mode 100644 Source/ispc_texcomp/ispc_texcomp.def create mode 100644 Source/ispc_texcomp/ispc_texcomp.h create mode 100644 Source/ispc_texcomp/ispc_texcomp.vcxproj create mode 100644 Source/ispc_texcomp/ispc_texcomp.vcxproj.filters create mode 100644 Source/ispc_texcomp/ispc_texcomp_astc.cpp create mode 100644 Source/ispc_texcomp/kernel.ispc create mode 100644 Source/ispc_texcomp/kernel_astc.ispc create mode 100644 Source/meson.build create mode 100644 Source/stb_image.cpp create mode 100644 Source/stb_image.h create mode 100644 Source/stb_image_resize.cpp create mode 100644 Source/stb_image_resize.h create mode 100644 Source/vk2dfd.cpp create mode 100644 Source/vk2dfd.inl create mode 100644 build/.keep create mode 100644 meson.build diff --git a/LICENCE.txt b/LICENCE.txt new file mode 100644 index 0000000..7ebf941 --- /dev/null +++ b/LICENCE.txt @@ -0,0 +1,109 @@ +TextureTaffy +============ + +MIT License + +Copyright (c) 2023 leblane + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. + + +stb_image and stb_image_resize +------------------------------ +[stb_image and stb_image_resize] are available under 2 licenses -- choose whichever you prefer. + +ALTERNATIVE A - MIT License +Copyright (c) 2017 Sean Barrett +Permission is hereby granted, free of charge, to any person obtaining a copy of +this software and associated documentation files (the "Software"), to deal in +the Software without restriction, including without limitation the rights to +use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies +of the Software, and to permit persons to whom the Software is furnished to do +so, subject to the following conditions: +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +------------------------------------------------------------------------------ +ALTERNATIVE B - Public Domain (www.unlicense.org) +This is free and unencumbered software released into the public domain. +Anyone is free to copy, modify, publish, use, compile, sell, or distribute this +software, either in source code form or as a compiled binary, for any purpose, +commercial or non-commercial, and by any means. +In jurisdictions that recognize copyright laws, the author or authors of this +software dedicate any and all copyright interest in the software to the public +domain. We make this dedication for the benefit of the public at large and to +the detriment of our heirs and successors. We intend this dedication to be an +overt act of relinquishment in perpetuity of all present and future rights to +this software under copyright law. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN +ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. +------------------------------------------------------------------------------ + + +ISPC Texture Compressor +----------------------- + +Copyright 2017 Intel Corporation + +Permission is hereby granted, free of charge, to any person obtaining a copy of +this software and associated documentation files (the "Software"), to deal in +the Software without restriction, including without limitation the rights to +use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies +of the Software, and to permit persons to whom the Software is furnished to do +so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. + + +Khronos DFD Components +---------------------- + +** Copyright 2015-2020 The Khronos Group Inc. +** SPDX-License-Identifier: Apache-2.0 + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. \ No newline at end of file diff --git a/README.md b/README.md new file mode 100644 index 0000000..e74b8a1 --- /dev/null +++ b/README.md @@ -0,0 +1,14 @@ +# TextureTaffy + +A utility to create compressed textures, in BC1 (DXT1), BC3 (DXT5), BC4, BC5, BC6(U)H and BC7 compression formats, with the [KTX File Format Version 2.0](https://registry.khronos.org/KTX/specs/2.0/ktxspec.v2.html) (KTX2). + +## Requirements + +* [The Meson build system](https://mesonbuild.com/) +* [IntelĀ® Implicit SPMD Program Compiler](https://ispc.github.io/) + +## Building + + + +## Notes and limitations \ No newline at end of file diff --git a/Source/HalfFloat.cpp b/Source/HalfFloat.cpp new file mode 100644 index 0000000..658e984 --- /dev/null +++ b/Source/HalfFloat.cpp @@ -0,0 +1,1039 @@ +#include "HalfFloat.h" + +namespace HalfFloat { + static const uint16_t basetable[512] = { + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0000, + 0x0001, + 0x0002, + 0x0004, + 0x0008, + 0x0010, + 0x0020, + 0x0040, + 0x0080, + 0x0100, + 0x0200, + 0x0400, + 0x0800, + 0x0c00, + 0x1000, + 0x1400, + 0x1800, + 0x1c00, + 0x2000, + 0x2400, + 0x2800, + 0x2c00, + 0x3000, + 0x3400, + 0x3800, + 0x3c00, + 0x4000, + 0x4400, + 0x4800, + 0x4c00, + 0x5000, + 0x5400, + 0x5800, + 0x5c00, + 0x6000, + 0x6400, + 0x6800, + 0x6c00, + 0x7000, + 0x7400, + 0x7800, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x7c00, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8000, + 0x8001, + 0x8002, + 0x8004, + 0x8008, + 0x8010, + 0x8020, + 0x8040, + 0x8080, + 0x8100, + 0x8200, + 0x8400, + 0x8800, + 0x8c00, + 0x9000, + 0x9400, + 0x9800, + 0x9c00, + 0xa000, + 0xa400, + 0xa800, + 0xac00, + 0xb000, + 0xb400, + 0xb800, + 0xbc00, + 0xc000, + 0xc400, + 0xc800, + 0xcc00, + 0xd000, + 0xd400, + 0xd800, + 0xdc00, + 0xe000, + 0xe400, + 0xe800, + 0xec00, + 0xf000, + 0xf400, + 0xf800, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00, + 0xfc00 + }; + + static const uint8_t shifttable[512] = { + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x17, + 0x16, + 0x15, + 0x14, + 0x13, + 0x12, + 0x11, + 0x10, + 0x0f, + 0x0e, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x0d, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x17, + 0x16, + 0x15, + 0x14, + 0x13, + 0x12, + 0x11, + 0x10, + 0x0f, + 0x0e, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x0d, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x18, + 0x0d + }; + + uint16_t FromFloat(float x) + { + uint32_t f = *reinterpret_cast(&x); + return basetable[(f >> 23) & 0x1ff] + ((f & 0x007fffff) >> shifttable[(f >> 23) & 0x1ff]); + } +} diff --git a/Source/HalfFloat.h b/Source/HalfFloat.h new file mode 100644 index 0000000..85a7303 --- /dev/null +++ b/Source/HalfFloat.h @@ -0,0 +1,8 @@ +#pragma once + +#include + +namespace HalfFloat +{ + uint16_t FromFloat(float x); +}; \ No newline at end of file diff --git a/Source/KHR/khr_df.h b/Source/KHR/khr_df.h new file mode 100644 index 0000000..bbd0d14 --- /dev/null +++ b/Source/KHR/khr_df.h @@ -0,0 +1,619 @@ +/* The Khronos Data Format Specification (version 1.3) */ +/* +** Copyright 2015-2020 The Khronos Group Inc. +** SPDX-License-Identifier: Apache-2.0 +*/ + +/* This header defines a structure that can describe the layout of image + formats in memory. This means that the data format is transparent to + the application, and the expectation is that this should be used when + the layout is defined external to the API. Many Khronos APIs deliberately + keep the internal layout of images opaque, to allow proprietary layouts + and optimisations. This structure is not appropriate for describing + opaque layouts. */ + +/* We stick to standard C89 constructs for simplicity and portability. */ + +#ifndef _KHR_DATA_FORMAT_H_ +#define _KHR_DATA_FORMAT_H_ + +/* Accessors */ +typedef enum _khr_word_e { + KHR_DF_WORD_VENDORID = 0U, + KHR_DF_WORD_DESCRIPTORTYPE = 0U, + KHR_DF_WORD_VERSIONNUMBER = 1U, + KHR_DF_WORD_DESCRIPTORBLOCKSIZE = 1U, + KHR_DF_WORD_MODEL = 2U, + KHR_DF_WORD_PRIMARIES = 2U, + KHR_DF_WORD_TRANSFER = 2U, + KHR_DF_WORD_FLAGS = 2U, + KHR_DF_WORD_TEXELBLOCKDIMENSION0 = 3U, + KHR_DF_WORD_TEXELBLOCKDIMENSION1 = 3U, + KHR_DF_WORD_TEXELBLOCKDIMENSION2 = 3U, + KHR_DF_WORD_TEXELBLOCKDIMENSION3 = 3U, + KHR_DF_WORD_BYTESPLANE0 = 4U, + KHR_DF_WORD_BYTESPLANE1 = 4U, + KHR_DF_WORD_BYTESPLANE2 = 4U, + KHR_DF_WORD_BYTESPLANE3 = 4U, + KHR_DF_WORD_BYTESPLANE4 = 5U, + KHR_DF_WORD_BYTESPLANE5 = 5U, + KHR_DF_WORD_BYTESPLANE6 = 5U, + KHR_DF_WORD_BYTESPLANE7 = 5U, + KHR_DF_WORD_SAMPLESTART = 6U, + KHR_DF_WORD_SAMPLEWORDS = 4U +} khr_df_word_e; + +typedef enum _khr_df_shift_e { + KHR_DF_SHIFT_VENDORID = 0U, + KHR_DF_SHIFT_DESCRIPTORTYPE = 17U, + KHR_DF_SHIFT_VERSIONNUMBER = 0U, + KHR_DF_SHIFT_DESCRIPTORBLOCKSIZE = 16U, + KHR_DF_SHIFT_MODEL = 0U, + KHR_DF_SHIFT_PRIMARIES = 8U, + KHR_DF_SHIFT_TRANSFER = 16U, + KHR_DF_SHIFT_FLAGS = 24U, + KHR_DF_SHIFT_TEXELBLOCKDIMENSION0 = 0U, + KHR_DF_SHIFT_TEXELBLOCKDIMENSION1 = 8U, + KHR_DF_SHIFT_TEXELBLOCKDIMENSION2 = 16U, + KHR_DF_SHIFT_TEXELBLOCKDIMENSION3 = 24U, + KHR_DF_SHIFT_BYTESPLANE0 = 0U, + KHR_DF_SHIFT_BYTESPLANE1 = 8U, + KHR_DF_SHIFT_BYTESPLANE2 = 16U, + KHR_DF_SHIFT_BYTESPLANE3 = 24U, + KHR_DF_SHIFT_BYTESPLANE4 = 0U, + KHR_DF_SHIFT_BYTESPLANE5 = 8U, + KHR_DF_SHIFT_BYTESPLANE6 = 16U, + KHR_DF_SHIFT_BYTESPLANE7 = 24U +} khr_df_shift_e; + +typedef enum _khr_df_mask_e { + KHR_DF_MASK_VENDORID = 0x1FFFFU, + KHR_DF_MASK_DESCRIPTORTYPE = 0x7FFFU, + KHR_DF_MASK_VERSIONNUMBER = 0xFFFFU, + KHR_DF_MASK_DESCRIPTORBLOCKSIZE = 0xFFFFU, + KHR_DF_MASK_MODEL = 0xFFU, + KHR_DF_MASK_PRIMARIES = 0xFFU, + KHR_DF_MASK_TRANSFER = 0xFFU, + KHR_DF_MASK_FLAGS = 0xFFU, + KHR_DF_MASK_TEXELBLOCKDIMENSION0 = 0xFFU, + KHR_DF_MASK_TEXELBLOCKDIMENSION1 = 0xFFU, + KHR_DF_MASK_TEXELBLOCKDIMENSION2 = 0xFFU, + KHR_DF_MASK_TEXELBLOCKDIMENSION3 = 0xFFU, + KHR_DF_MASK_BYTESPLANE0 = 0xFFU, + KHR_DF_MASK_BYTESPLANE1 = 0xFFU, + KHR_DF_MASK_BYTESPLANE2 = 0xFFU, + KHR_DF_MASK_BYTESPLANE3 = 0xFFU, + KHR_DF_MASK_BYTESPLANE4 = 0xFFU, + KHR_DF_MASK_BYTESPLANE5 = 0xFFU, + KHR_DF_MASK_BYTESPLANE6 = 0xFFU, + KHR_DF_MASK_BYTESPLANE7 = 0xFFU +} khr_df_mask_e; + +/* Helper macro: + Extract field X from basic descriptor block BDB */ +#define KHR_DFDVAL(BDB, X) \ + (((BDB)[KHR_DF_WORD_ ## X] >> (KHR_DF_SHIFT_ ## X)) \ + & (KHR_DF_MASK_ ## X)) + +/* Helper macro: + Set field X of basic descriptor block BDB */ +#define KHR_DFDSETVAL(BDB, X, val) \ + ((BDB)[KHR_DF_WORD_ ## X] = \ + ((BDB)[KHR_DF_WORD_ ## X] & \ + ~((KHR_DF_MASK_ ## X) << (KHR_DF_SHIFT_ ## X))) | \ + (((val) & (KHR_DF_MASK_ ## X)) << (KHR_DF_SHIFT_ ## X))) + +/* Offsets relative to the start of a sample */ +typedef enum _khr_df_sampleword_e { + KHR_DF_SAMPLEWORD_BITOFFSET = 0U, + KHR_DF_SAMPLEWORD_BITLENGTH = 0U, + KHR_DF_SAMPLEWORD_CHANNELID = 0U, + KHR_DF_SAMPLEWORD_QUALIFIERS = 0U, + KHR_DF_SAMPLEWORD_SAMPLEPOSITION0 = 1U, + KHR_DF_SAMPLEWORD_SAMPLEPOSITION1 = 1U, + KHR_DF_SAMPLEWORD_SAMPLEPOSITION2 = 1U, + KHR_DF_SAMPLEWORD_SAMPLEPOSITION3 = 1U, + KHR_DF_SAMPLEWORD_SAMPLEPOSITION_ALL = 1U, + KHR_DF_SAMPLEWORD_SAMPLELOWER = 2U, + KHR_DF_SAMPLEWORD_SAMPLEUPPER = 3U +} khr_df_sampleword_e; + +typedef enum _khr_df_sampleshift_e { + KHR_DF_SAMPLESHIFT_BITOFFSET = 0U, + KHR_DF_SAMPLESHIFT_BITLENGTH = 16U, + KHR_DF_SAMPLESHIFT_CHANNELID = 24U, + /* N.B. Qualifiers are defined as an offset into a byte */ + KHR_DF_SAMPLESHIFT_QUALIFIERS = 24U, + KHR_DF_SAMPLESHIFT_SAMPLEPOSITION0 = 0U, + KHR_DF_SAMPLESHIFT_SAMPLEPOSITION1 = 8U, + KHR_DF_SAMPLESHIFT_SAMPLEPOSITION2 = 16U, + KHR_DF_SAMPLESHIFT_SAMPLEPOSITION3 = 24U, + KHR_DF_SAMPLESHIFT_SAMPLEPOSITION_ALL = 0U, + KHR_DF_SAMPLESHIFT_SAMPLELOWER = 0U, + KHR_DF_SAMPLESHIFT_SAMPLEUPPER = 0U +} khr_df_sampleshift_e; + +typedef enum _khr_df_samplemask_e { + KHR_DF_SAMPLEMASK_BITOFFSET = 0xFFFFU, + KHR_DF_SAMPLEMASK_BITLENGTH = 0xFF, + KHR_DF_SAMPLEMASK_CHANNELID = 0xF, + /* N.B. Qualifiers are defined as an offset into a byte */ + KHR_DF_SAMPLEMASK_QUALIFIERS = 0xF0, + KHR_DF_SAMPLEMASK_SAMPLEPOSITION0 = 0xFF, + KHR_DF_SAMPLEMASK_SAMPLEPOSITION1 = 0xFF, + KHR_DF_SAMPLEMASK_SAMPLEPOSITION2 = 0xFF, + KHR_DF_SAMPLEMASK_SAMPLEPOSITION3 = 0xFF, + /* ISO C restricts enum values to range of int hence the + cast. We do it verbosely instead of using -1 to ensure + it is a 32-bit value even if int is 64 bits. */ + KHR_DF_SAMPLEMASK_SAMPLEPOSITION_ALL = (int) 0xFFFFFFFFU, + KHR_DF_SAMPLEMASK_SAMPLELOWER = (int) 0xFFFFFFFFU, + KHR_DF_SAMPLEMASK_SAMPLEUPPER = (int) 0xFFFFFFFFU +} khr_df_samplemask_e; + +/* Helper macro: + Extract field X of sample S from basic descriptor block BDB */ +#define KHR_DFDSVAL(BDB, S, X) \ + (((BDB)[KHR_DF_WORD_SAMPLESTART + \ + ((S) * KHR_DF_WORD_SAMPLEWORDS) + \ + KHR_DF_SAMPLEWORD_ ## X] >> (KHR_DF_SAMPLESHIFT_ ## X)) \ + & (KHR_DF_SAMPLEMASK_ ## X)) + +/* Helper macro: + Set field X of sample S of basic descriptor block BDB */ +#define KHR_DFDSETSVAL(BDB, S, X, val) \ + ((BDB)[KHR_DF_WORD_SAMPLESTART + \ + ((S) * KHR_DF_WORD_SAMPLEWORDS) + \ + KHR_DF_SAMPLEWORD_ ## X] = \ + ((BDB)[KHR_DF_WORD_SAMPLESTART + \ + ((S) * KHR_DF_WORD_SAMPLEWORDS) + \ + KHR_DF_SAMPLEWORD_ ## X] & \ + ~((uint32_t)(KHR_DF_SAMPLEMASK_ ## X) << (KHR_DF_SAMPLESHIFT_ ## X))) | \ + (((val) & (uint32_t)(KHR_DF_SAMPLEMASK_ ## X)) << (KHR_DF_SAMPLESHIFT_ ## X))) + +/* Helper macro: + Number of samples in basic descriptor block BDB */ +#define KHR_DFDSAMPLECOUNT(BDB) \ + (((KHR_DFDVAL(BDB, DESCRIPTORBLOCKSIZE) >> 2) - \ + KHR_DF_WORD_SAMPLESTART) \ + / KHR_DF_WORD_SAMPLEWORDS) + +/* Helper macro: + Size in words of basic descriptor block for S samples */ +#define KHR_DFDSIZEWORDS(S) \ + (KHR_DF_WORD_SAMPLESTART + \ + (S) * KHR_DF_WORD_SAMPLEWORDS) + +/* Vendor ids */ +typedef enum _khr_df_vendorid_e { + /* Standard Khronos descriptor */ + KHR_DF_VENDORID_KHRONOS = 0U, + KHR_DF_VENDORID_MAX = 0x1FFFFU +} khr_df_vendorid_e; + +/* Descriptor types */ +typedef enum _khr_df_khr_descriptortype_e { + /* Default Khronos basic descriptor block */ + KHR_DF_KHR_DESCRIPTORTYPE_BASICFORMAT = 0U, + /* Extension descriptor block for additional planes */ + KHR_DF_KHR_DESCRIPTORTYPE_ADDITIONAL_PLANES = 0x6001U, + /* Extension descriptor block for additional dimensions */ + KHR_DF_KHR_DESCRIPTORTYPE_ADDITIONAL_DIMENSIONS = 0x6002U, + /* Bit indicates modifying requires understanding this extension */ + KHR_DF_KHR_DESCRIPTORTYPE_NEEDED_FOR_WRITE_BIT = 0x2000U, + /* Bit indicates processing requires understanding this extension */ + KHR_DF_KHR_DESCRIPTORTYPE_NEEDED_FOR_DECODE_BIT = 0x4000U, + KHR_DF_KHR_DESCRIPTORTYPE_MAX = 0x7FFFU +} khr_df_khr_descriptortype_e; + +/* Descriptor block version */ +typedef enum _khr_df_versionnumber_e { + /* Standard Khronos descriptor */ + KHR_DF_VERSIONNUMBER_1_0 = 0U, /* Version 1.0 of the specification */ + KHR_DF_VERSIONNUMBER_1_1 = 0U, /* Version 1.1 did not bump the version number */ + KHR_DF_VERSIONNUMBER_1_2 = 1U, /* Version 1.2 increased the version number */ + KHR_DF_VERSIONNUMBER_1_3 = 2U, /* Version 1.3 increased the version number */ + KHR_DF_VERSIONNUMBER_LATEST = KHR_DF_VERSIONNUMBER_1_3, + KHR_DF_VERSIONNUMBER_MAX = 0xFFFFU +} khr_df_versionnumber_e; + +/* Model in which the color coordinate space is defined. + There is no requirement that a color format use all the + channel types that are defined in the color model. */ +typedef enum _khr_df_model_e { + /* No interpretation of color channels defined */ + KHR_DF_MODEL_UNSPECIFIED = 0U, + /* Color primaries (red, green, blue) + alpha, depth and stencil */ + KHR_DF_MODEL_RGBSDA = 1U, + /* Color differences (Y', Cb, Cr) + alpha, depth and stencil */ + KHR_DF_MODEL_YUVSDA = 2U, + /* Color differences (Y', I, Q) + alpha, depth and stencil */ + KHR_DF_MODEL_YIQSDA = 3U, + /* Perceptual color (CIE L*a*b*) + alpha, depth and stencil */ + KHR_DF_MODEL_LABSDA = 4U, + /* Subtractive colors (cyan, magenta, yellow, black) + alpha */ + KHR_DF_MODEL_CMYKA = 5U, + /* Non-color coordinate data (X, Y, Z, W) */ + KHR_DF_MODEL_XYZW = 6U, + /* Hue, saturation, value, hue angle on color circle, plus alpha */ + KHR_DF_MODEL_HSVA_ANG = 7U, + /* Hue, saturation, lightness, hue angle on color circle, plus alpha */ + KHR_DF_MODEL_HSLA_ANG = 8U, + /* Hue, saturation, value, hue on color hexagon, plus alpha */ + KHR_DF_MODEL_HSVA_HEX = 9U, + /* Hue, saturation, lightness, hue on color hexagon, plus alpha */ + KHR_DF_MODEL_HSLA_HEX = 10U, + /* Lightweight approximate color difference (luma, orange, green) */ + KHR_DF_MODEL_YCGCOA = 11U, + /* ITU BT.2020 constant luminance YcCbcCrc */ + KHR_DF_MODEL_YCCBCCRC = 12U, + /* ITU BT.2100 constant intensity ICtCp */ + KHR_DF_MODEL_ICTCP = 13U, + /* CIE 1931 XYZ color coordinates (X, Y, Z) */ + KHR_DF_MODEL_CIEXYZ = 14U, + /* CIE 1931 xyY color coordinates (X, Y, Y) */ + KHR_DF_MODEL_CIEXYY = 15U, + + /* Compressed formats start at 128. */ + /* These compressed formats should generally have a single sample, + sited at the 0,0 position of the texel block. Where multiple + channels are used to distinguish formats, these should be cosited. */ + /* Direct3D (and S3) compressed formats */ + /* Note that premultiplied status is recorded separately */ + /* DXT1 "channels" are RGB (0), Alpha (1) */ + /* DXT1/BC1 with one channel is opaque */ + /* DXT1/BC1 with a cosited alpha sample is transparent */ + KHR_DF_MODEL_DXT1A = 128U, + KHR_DF_MODEL_BC1A = 128U, + /* DXT2/DXT3/BC2, with explicit 4-bit alpha */ + KHR_DF_MODEL_DXT2 = 129U, + KHR_DF_MODEL_DXT3 = 129U, + KHR_DF_MODEL_BC2 = 129U, + /* DXT4/DXT5/BC3, with interpolated alpha */ + KHR_DF_MODEL_DXT4 = 130U, + KHR_DF_MODEL_DXT5 = 130U, + KHR_DF_MODEL_BC3 = 130U, + /* BC4 - single channel interpolated 8-bit data */ + /* (The UNORM/SNORM variation is recorded in the channel data) */ + KHR_DF_MODEL_BC4 = 131U, + /* BC5 - two channel interpolated 8-bit data */ + /* (The UNORM/SNORM variation is recorded in the channel data) */ + KHR_DF_MODEL_BC5 = 132U, + /* BC6H - DX11 format for 16-bit float channels */ + KHR_DF_MODEL_BC6H = 133U, + /* BC7 - DX11 format */ + KHR_DF_MODEL_BC7 = 134U, + /* Gap left for future desktop expansion */ + + /* Mobile compressed formats follow */ + /* A format of ETC1 indicates that the format shall be decodable + by an ETC1-compliant decoder and not rely on ETC2 features */ + KHR_DF_MODEL_ETC1 = 160U, + /* A format of ETC2 is permitted to use ETC2 encodings on top of + the baseline ETC1 specification */ + /* The ETC2 format has channels "red", "green", "RGB" and "alpha", + which should be cosited samples */ + /* Punch-through alpha can be distinguished from full alpha by + the plane size in bytes required for the texel block */ + KHR_DF_MODEL_ETC2 = 161U, + /* Adaptive Scalable Texture Compression */ + /* ASTC HDR vs LDR is determined by the float flag in the channel */ + /* ASTC block size can be distinguished by texel block size */ + KHR_DF_MODEL_ASTC = 162U, + /* ETC1S is a simplified subset of ETC1 */ + KHR_DF_MODEL_ETC1S = 163U, + /* PowerVR Texture Compression */ + KHR_DF_MODEL_PVRTC = 164U, + KHR_DF_MODEL_PVRTC2 = 165U, + KHR_DF_MODEL_UASTC = 166U, + /* Proprietary formats (ATITC, etc.) should follow */ + KHR_DF_MODEL_MAX = 0xFFU +} khr_df_model_e; + +/* Definition of channel names for each color model */ +typedef enum _khr_df_model_channels_e { + /* Unspecified format with nominal channel numbering */ + KHR_DF_CHANNEL_UNSPECIFIED_0 = 0U, + KHR_DF_CHANNEL_UNSPECIFIED_1 = 1U, + KHR_DF_CHANNEL_UNSPECIFIED_2 = 2U, + KHR_DF_CHANNEL_UNSPECIFIED_3 = 3U, + KHR_DF_CHANNEL_UNSPECIFIED_4 = 4U, + KHR_DF_CHANNEL_UNSPECIFIED_5 = 5U, + KHR_DF_CHANNEL_UNSPECIFIED_6 = 6U, + KHR_DF_CHANNEL_UNSPECIFIED_7 = 7U, + KHR_DF_CHANNEL_UNSPECIFIED_8 = 8U, + KHR_DF_CHANNEL_UNSPECIFIED_9 = 9U, + KHR_DF_CHANNEL_UNSPECIFIED_10 = 10U, + KHR_DF_CHANNEL_UNSPECIFIED_11 = 11U, + KHR_DF_CHANNEL_UNSPECIFIED_12 = 12U, + KHR_DF_CHANNEL_UNSPECIFIED_13 = 13U, + KHR_DF_CHANNEL_UNSPECIFIED_14 = 14U, + KHR_DF_CHANNEL_UNSPECIFIED_15 = 15U, + /* MODEL_RGBSDA - red, green, blue, stencil, depth, alpha */ + KHR_DF_CHANNEL_RGBSDA_RED = 0U, + KHR_DF_CHANNEL_RGBSDA_R = 0U, + KHR_DF_CHANNEL_RGBSDA_GREEN = 1U, + KHR_DF_CHANNEL_RGBSDA_G = 1U, + KHR_DF_CHANNEL_RGBSDA_BLUE = 2U, + KHR_DF_CHANNEL_RGBSDA_B = 2U, + KHR_DF_CHANNEL_RGBSDA_STENCIL = 13U, + KHR_DF_CHANNEL_RGBSDA_S = 13U, + KHR_DF_CHANNEL_RGBSDA_DEPTH = 14U, + KHR_DF_CHANNEL_RGBSDA_D = 14U, + KHR_DF_CHANNEL_RGBSDA_ALPHA = 15U, + KHR_DF_CHANNEL_RGBSDA_A = 15U, + /* MODEL_YUVSDA - luma, Cb, Cr, stencil, depth, alpha */ + KHR_DF_CHANNEL_YUVSDA_Y = 0U, + KHR_DF_CHANNEL_YUVSDA_CB = 1U, + KHR_DF_CHANNEL_YUVSDA_U = 1U, + KHR_DF_CHANNEL_YUVSDA_CR = 2U, + KHR_DF_CHANNEL_YUVSDA_V = 2U, + KHR_DF_CHANNEL_YUVSDA_STENCIL = 13U, + KHR_DF_CHANNEL_YUVSDA_S = 13U, + KHR_DF_CHANNEL_YUVSDA_DEPTH = 14U, + KHR_DF_CHANNEL_YUVSDA_D = 14U, + KHR_DF_CHANNEL_YUVSDA_ALPHA = 15U, + KHR_DF_CHANNEL_YUVSDA_A = 15U, + /* MODEL_YIQSDA - luma, in-phase, quadrature, stencil, depth, alpha */ + KHR_DF_CHANNEL_YIQSDA_Y = 0U, + KHR_DF_CHANNEL_YIQSDA_I = 1U, + KHR_DF_CHANNEL_YIQSDA_Q = 2U, + KHR_DF_CHANNEL_YIQSDA_STENCIL = 13U, + KHR_DF_CHANNEL_YIQSDA_S = 13U, + KHR_DF_CHANNEL_YIQSDA_DEPTH = 14U, + KHR_DF_CHANNEL_YIQSDA_D = 14U, + KHR_DF_CHANNEL_YIQSDA_ALPHA = 15U, + KHR_DF_CHANNEL_YIQSDA_A = 15U, + /* MODEL_LABSDA - CIELAB/L*a*b* luma, red-green, blue-yellow, stencil, depth, alpha */ + KHR_DF_CHANNEL_LABSDA_L = 0U, + KHR_DF_CHANNEL_LABSDA_A = 1U, + KHR_DF_CHANNEL_LABSDA_B = 2U, + KHR_DF_CHANNEL_LABSDA_STENCIL = 13U, + KHR_DF_CHANNEL_LABSDA_S = 13U, + KHR_DF_CHANNEL_LABSDA_DEPTH = 14U, + KHR_DF_CHANNEL_LABSDA_D = 14U, + KHR_DF_CHANNEL_LABSDA_ALPHA = 15U, + /* NOTE: KHR_DF_CHANNEL_LABSDA_A is not a synonym for alpha! */ + /* MODEL_CMYKA - cyan, magenta, yellow, key/blacK, alpha */ + KHR_DF_CHANNEL_CMYKSDA_CYAN = 0U, + KHR_DF_CHANNEL_CMYKSDA_C = 0U, + KHR_DF_CHANNEL_CMYKSDA_MAGENTA = 1U, + KHR_DF_CHANNEL_CMYKSDA_M = 1U, + KHR_DF_CHANNEL_CMYKSDA_YELLOW = 2U, + KHR_DF_CHANNEL_CMYKSDA_Y = 2U, + KHR_DF_CHANNEL_CMYKSDA_KEY = 3U, + KHR_DF_CHANNEL_CMYKSDA_BLACK = 3U, + KHR_DF_CHANNEL_CMYKSDA_K = 3U, + KHR_DF_CHANNEL_CMYKSDA_ALPHA = 15U, + KHR_DF_CHANNEL_CMYKSDA_A = 15U, + /* MODEL_XYZW - coordinates x, y, z, w */ + KHR_DF_CHANNEL_XYZW_X = 0U, + KHR_DF_CHANNEL_XYZW_Y = 1U, + KHR_DF_CHANNEL_XYZW_Z = 2U, + KHR_DF_CHANNEL_XYZW_W = 3U, + /* MODEL_HSVA_ANG - value (luma), saturation, hue, alpha, angular projection, conical space */ + KHR_DF_CHANNEL_HSVA_ANG_VALUE = 0U, + KHR_DF_CHANNEL_HSVA_ANG_V = 0U, + KHR_DF_CHANNEL_HSVA_ANG_SATURATION = 1U, + KHR_DF_CHANNEL_HSVA_ANG_S = 1U, + KHR_DF_CHANNEL_HSVA_ANG_HUE = 2U, + KHR_DF_CHANNEL_HSVA_ANG_H = 2U, + KHR_DF_CHANNEL_HSVA_ANG_ALPHA = 15U, + KHR_DF_CHANNEL_HSVA_ANG_A = 15U, + /* MODEL_HSLA_ANG - lightness (luma), saturation, hue, alpha, angular projection, double conical space */ + KHR_DF_CHANNEL_HSLA_ANG_LIGHTNESS = 0U, + KHR_DF_CHANNEL_HSLA_ANG_L = 0U, + KHR_DF_CHANNEL_HSLA_ANG_SATURATION = 1U, + KHR_DF_CHANNEL_HSLA_ANG_S = 1U, + KHR_DF_CHANNEL_HSLA_ANG_HUE = 2U, + KHR_DF_CHANNEL_HSLA_ANG_H = 2U, + KHR_DF_CHANNEL_HSLA_ANG_ALPHA = 15U, + KHR_DF_CHANNEL_HSLA_ANG_A = 15U, + /* MODEL_HSVA_HEX - value (luma), saturation, hue, alpha, hexagonal projection, conical space */ + KHR_DF_CHANNEL_HSVA_HEX_VALUE = 0U, + KHR_DF_CHANNEL_HSVA_HEX_V = 0U, + KHR_DF_CHANNEL_HSVA_HEX_SATURATION = 1U, + KHR_DF_CHANNEL_HSVA_HEX_S = 1U, + KHR_DF_CHANNEL_HSVA_HEX_HUE = 2U, + KHR_DF_CHANNEL_HSVA_HEX_H = 2U, + KHR_DF_CHANNEL_HSVA_HEX_ALPHA = 15U, + KHR_DF_CHANNEL_HSVA_HEX_A = 15U, + /* MODEL_HSLA_HEX - lightness (luma), saturation, hue, alpha, hexagonal projection, double conical space */ + KHR_DF_CHANNEL_HSLA_HEX_LIGHTNESS = 0U, + KHR_DF_CHANNEL_HSLA_HEX_L = 0U, + KHR_DF_CHANNEL_HSLA_HEX_SATURATION = 1U, + KHR_DF_CHANNEL_HSLA_HEX_S = 1U, + KHR_DF_CHANNEL_HSLA_HEX_HUE = 2U, + KHR_DF_CHANNEL_HSLA_HEX_H = 2U, + KHR_DF_CHANNEL_HSLA_HEX_ALPHA = 15U, + KHR_DF_CHANNEL_HSLA_HEX_A = 15U, + /* MODEL_YCGCOA - luma, green delta, orange delta, alpha */ + KHR_DF_CHANNEL_YCGCOA_Y = 0U, + KHR_DF_CHANNEL_YCGCOA_CG = 1U, + KHR_DF_CHANNEL_YCGCOA_CO = 2U, + KHR_DF_CHANNEL_YCGCOA_ALPHA = 15U, + KHR_DF_CHANNEL_YCGCOA_A = 15U, + /* MODEL_CIEXYZ - CIE 1931 X, Y, Z */ + KHR_DF_CHANNEL_CIEXYZ_X = 0U, + KHR_DF_CHANNEL_CIEXYZ_Y = 1U, + KHR_DF_CHANNEL_CIEXYZ_Z = 2U, + /* MODEL_CIEXYY - CIE 1931 x, y, Y */ + KHR_DF_CHANNEL_CIEXYY_X = 0U, + KHR_DF_CHANNEL_CIEXYY_YCHROMA = 1U, + KHR_DF_CHANNEL_CIEXYY_YLUMA = 2U, + + /* Compressed formats */ + /* MODEL_DXT1A/MODEL_BC1A */ + KHR_DF_CHANNEL_DXT1A_COLOR = 0U, + KHR_DF_CHANNEL_BC1A_COLOR = 0U, + KHR_DF_CHANNEL_DXT1A_ALPHAPRESENT = 1U, + KHR_DF_CHANNEL_DXT1A_ALPHA = 1U, + KHR_DF_CHANNEL_BC1A_ALPHAPRESENT = 1U, + KHR_DF_CHANNEL_BC1A_ALPHA = 1U, + /* MODEL_DXT2/3/MODEL_BC2 */ + KHR_DF_CHANNEL_DXT2_COLOR = 0U, + KHR_DF_CHANNEL_DXT3_COLOR = 0U, + KHR_DF_CHANNEL_BC2_COLOR = 0U, + KHR_DF_CHANNEL_DXT2_ALPHA = 15U, + KHR_DF_CHANNEL_DXT3_ALPHA = 15U, + KHR_DF_CHANNEL_BC2_ALPHA = 15U, + /* MODEL_DXT4/5/MODEL_BC3 */ + KHR_DF_CHANNEL_DXT4_COLOR = 0U, + KHR_DF_CHANNEL_DXT5_COLOR = 0U, + KHR_DF_CHANNEL_BC3_COLOR = 0U, + KHR_DF_CHANNEL_DXT4_ALPHA = 15U, + KHR_DF_CHANNEL_DXT5_ALPHA = 15U, + KHR_DF_CHANNEL_BC3_ALPHA = 15U, + /* MODEL_BC4 */ + KHR_DF_CHANNEL_BC4_DATA = 0U, + /* MODEL_BC5 */ + KHR_DF_CHANNEL_BC5_RED = 0U, + KHR_DF_CHANNEL_BC5_R = 0U, + KHR_DF_CHANNEL_BC5_GREEN = 1U, + KHR_DF_CHANNEL_BC5_G = 1U, + /* MODEL_BC6H */ + KHR_DF_CHANNEL_BC6H_COLOR = 0U, + KHR_DF_CHANNEL_BC6H_DATA = 0U, + /* MODEL_BC7 */ + KHR_DF_CHANNEL_BC7_DATA = 0U, + KHR_DF_CHANNEL_BC7_COLOR = 0U, + /* MODEL_ETC1 */ + KHR_DF_CHANNEL_ETC1_DATA = 0U, + KHR_DF_CHANNEL_ETC1_COLOR = 0U, + /* MODEL_ETC2 */ + KHR_DF_CHANNEL_ETC2_RED = 0U, + KHR_DF_CHANNEL_ETC2_R = 0U, + KHR_DF_CHANNEL_ETC2_GREEN = 1U, + KHR_DF_CHANNEL_ETC2_G = 1U, + KHR_DF_CHANNEL_ETC2_COLOR = 2U, + KHR_DF_CHANNEL_ETC2_ALPHA = 15U, + KHR_DF_CHANNEL_ETC2_A = 15U, + /* MODEL_ASTC */ + KHR_DF_CHANNEL_ASTC_DATA = 0U, + /* MODEL_ETC1S */ + KHR_DF_CHANNEL_ETC1S_RGB = 0U, + KHR_DF_CHANNEL_ETC1S_RRR = 3U, + KHR_DF_CHANNEL_ETC1S_GGG = 4U, + KHR_DF_CHANNEL_ETC1S_AAA = 15U, + /* MODEL_PVRTC */ + KHR_DF_CHANNEL_PVRTC_DATA = 0U, + KHR_DF_CHANNEL_PVRTC_COLOR = 0U, + /* MODEL_PVRTC2 */ + KHR_DF_CHANNEL_PVRTC2_DATA = 0U, + KHR_DF_CHANNEL_PVRTC2_COLOR = 0U, + /* MODEL UASTC */ + KHR_DF_CHANNEL_UASTC_DATA = 0U, + KHR_DF_CHANNEL_UASTC_RGB = 0U, + KHR_DF_CHANNEL_UASTC_RGBA = 3U, + KHR_DF_CHANNEL_UASTC_RRR = 4U, + KHR_DF_CHANNEL_UASTC_RRRG = 5U, + KHR_DF_CHANNEL_UASTC_RG = 6U, + + /* Common channel names shared by multiple formats */ + KHR_DF_CHANNEL_COMMON_LUMA = 0U, + KHR_DF_CHANNEL_COMMON_L = 0U, + KHR_DF_CHANNEL_COMMON_STENCIL = 13U, + KHR_DF_CHANNEL_COMMON_S = 13U, + KHR_DF_CHANNEL_COMMON_DEPTH = 14U, + KHR_DF_CHANNEL_COMMON_D = 14U, + KHR_DF_CHANNEL_COMMON_ALPHA = 15U, + KHR_DF_CHANNEL_COMMON_A = 15U +} khr_df_model_channels_e; + +/* Definition of the primary colors in color coordinates. + This is implicitly responsible for defining the conversion + between RGB an YUV color spaces. + LAB and related absolute color models should use + KHR_DF_PRIMARIES_CIEXYZ. */ +typedef enum _khr_df_primaries_e { + /* No color primaries defined */ + KHR_DF_PRIMARIES_UNSPECIFIED = 0U, + /* Color primaries of ITU-R BT.709 and sRGB */ + KHR_DF_PRIMARIES_BT709 = 1U, + /* Synonym for KHR_DF_PRIMARIES_BT709 */ + KHR_DF_PRIMARIES_SRGB = 1U, + /* Color primaries of ITU-R BT.601 (625-line EBU variant) */ + KHR_DF_PRIMARIES_BT601_EBU = 2U, + /* Color primaries of ITU-R BT.601 (525-line SMPTE C variant) */ + KHR_DF_PRIMARIES_BT601_SMPTE = 3U, + /* Color primaries of ITU-R BT.2020 */ + KHR_DF_PRIMARIES_BT2020 = 4U, + /* CIE theoretical color coordinate space */ + KHR_DF_PRIMARIES_CIEXYZ = 5U, + /* Academy Color Encoding System primaries */ + KHR_DF_PRIMARIES_ACES = 6U, + /* Color primaries of ACEScc */ + KHR_DF_PRIMARIES_ACESCC = 7U, + /* Legacy NTSC 1953 primaries */ + KHR_DF_PRIMARIES_NTSC1953 = 8U, + /* Legacy PAL 525-line primaries */ + KHR_DF_PRIMARIES_PAL525 = 9U, + /* Color primaries of Display P3 */ + KHR_DF_PRIMARIES_DISPLAYP3 = 10U, + /* Color primaries of Adobe RGB (1998) */ + KHR_DF_PRIMARIES_ADOBERGB = 11U, + KHR_DF_PRIMARIES_MAX = 0xFFU +} khr_df_primaries_e; + +/* Definition of the optical to digital transfer function + ("gamma correction"). Most transfer functions are not a pure + power function and also include a linear element. + LAB and related absolute color representations should use + KHR_DF_TRANSFER_UNSPECIFIED. */ +typedef enum _khr_df_transfer_e { + /* No transfer function defined */ + KHR_DF_TRANSFER_UNSPECIFIED = 0U, + /* Linear transfer function (value proportional to intensity) */ + KHR_DF_TRANSFER_LINEAR = 1U, + /* Perceptually-linear transfer function of sRGH (~2.4) */ + KHR_DF_TRANSFER_SRGB = 2U, + /* Perceptually-linear transfer function of ITU BT.601, BT.709 and BT.2020 (~1/.45) */ + KHR_DF_TRANSFER_ITU = 3U, + /* SMTPE170M (digital NTSC) defines an alias for the ITU transfer function (~1/.45) */ + KHR_DF_TRANSFER_SMTPE170M = 3U, + /* Perceptually-linear gamma function of original NTSC (simple 2.2 gamma) */ + KHR_DF_TRANSFER_NTSC = 4U, + /* Sony S-log used by Sony video cameras */ + KHR_DF_TRANSFER_SLOG = 5U, + /* Sony S-log 2 used by Sony video cameras */ + KHR_DF_TRANSFER_SLOG2 = 6U, + /* ITU BT.1886 EOTF */ + KHR_DF_TRANSFER_BT1886 = 7U, + /* ITU BT.2100 HLG OETF */ + KHR_DF_TRANSFER_HLG_OETF = 8U, + /* ITU BT.2100 HLG EOTF */ + KHR_DF_TRANSFER_HLG_EOTF = 9U, + /* ITU BT.2100 PQ EOTF */ + KHR_DF_TRANSFER_PQ_EOTF = 10U, + /* ITU BT.2100 PQ OETF */ + KHR_DF_TRANSFER_PQ_OETF = 11U, + /* DCI P3 transfer function */ + KHR_DF_TRANSFER_DCIP3 = 12U, + /* Legacy PAL OETF */ + KHR_DF_TRANSFER_PAL_OETF = 13U, + /* Legacy PAL 625-line EOTF */ + KHR_DF_TRANSFER_PAL625_EOTF = 14U, + /* Legacy ST240 transfer function */ + KHR_DF_TRANSFER_ST240 = 15U, + /* ACEScc transfer function */ + KHR_DF_TRANSFER_ACESCC = 16U, + /* ACEScct transfer function */ + KHR_DF_TRANSFER_ACESCCT = 17U, + /* Adobe RGB (1998) transfer function */ + KHR_DF_TRANSFER_ADOBERGB = 18U, + KHR_DF_TRANSFER_MAX = 0xFFU +} khr_df_transfer_e; + +typedef enum _khr_df_flags_e { + KHR_DF_FLAG_ALPHA_STRAIGHT = 0U, + KHR_DF_FLAG_ALPHA_PREMULTIPLIED = 1U +} khr_df_flags_e; + +typedef enum _khr_df_sample_datatype_qualifiers_e { + KHR_DF_SAMPLE_DATATYPE_LINEAR = 1U << 4U, + KHR_DF_SAMPLE_DATATYPE_EXPONENT = 1U << 5U, + KHR_DF_SAMPLE_DATATYPE_SIGNED = 1U << 6U, + KHR_DF_SAMPLE_DATATYPE_FLOAT = 1U << 7U +} khr_df_sample_datatype_qualifiers_e; + +#endif diff --git a/Source/Main.cpp b/Source/Main.cpp new file mode 100644 index 0000000..6170b1e --- /dev/null +++ b/Source/Main.cpp @@ -0,0 +1,619 @@ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "stb_image_resize.h" +#include "stb_image.h" +#include "dfd.h" +#include "ispc_texcomp/ispc_texcomp.h" +#include "HalfFloat.h" + +const std::vector formatOrder = { + "BC1", + "BC1_SRGB", + "BC4", + "BC5", + "BC3", + "BC3_SRGB", + "BC6H", + "BC7", + "BC7_SRGB" +}; + +const std::map> formats = { + {"BC1", {"(DXT1) 5:6:5 Color, 1 bit alpha. 8 bytes per block.", 8, vk::Format::eBc1RgbUnormBlock}}, + {"BC1_SRGB", {"(DXT1) 5:6:5 Color, 1 bit alpha. 8 bytes per block.", 8, vk::Format::eBc1RgbSrgbBlock}}, + {"BC4", {"Greyscale, 8 bytes per block.", 8, vk::Format::eBc4UnormBlock}}, + {"BC5", {"2x BC4 images. 16 bytes per block.", 16, vk::Format::eBc5UnormBlock}}, + {"BC3", {"(DXT5) BC1 Color, BC4 Alpha, 16 bytes per block.", 16, vk::Format::eBc3UnormBlock}}, + {"BC3_SRGB", {"(DXT5) BC1 Color, BC4 Alpha, 16 bytes per block.", 16, vk::Format::eBc3SrgbBlock}}, + {"BC6H", {"16 bit RGB, no alpha. Signed. 16 bytes per block.", 16, vk::Format::eBc6HUfloatBlock}}, + {"BC7", {"8 bit RGBA - Good general purpose. 16 bytes per block.", 16, vk::Format::eBc7UnormBlock}}, + {"BC7_SRGB", {"8 bit RGBA - Good general purpose. 16 bytes per block.", 16, vk::Format::eBc7SrgbBlock}} +}; + +const std::string usage = "Usage: TextureConverter [cube|array] [input2, input3...] [fast|slow|veryslow]"; + +int main(int argc, char ** argv) +{ + ISPCInit(); + + if (argc < 4) { + std::cout << usage << std::endl; + std::cout << "Formats:" << std::endl; + for (auto & formatName : formatOrder) { + auto format = formats.at(formatName); + std::cout << " " << formatName << " - " << std::get<0>(format) << std::endl; + } + return 1; + } + + int numInputs = argc - 3; + unsigned int inputsStart = 1; + std::vector inputs; + std::string output; + std::string option(argv[1]); + if (option == "cube" || option == "array") { + numInputs -= 1; + inputsStart += 1; + } else { + option = "none"; + } + + std::string speed(argv[argc - 1]); + std::string formatString; + bool fastMode = true; + bool verySlow = false; + if (speed == "fast" || speed == "slow" || speed == "veryslow") { + formatString = std::string(argv[argc - 2]); + numInputs -= 1; + + if (speed == "slow") { + fastMode = false; + } else if (speed == "veryslow") { + fastMode = false; + verySlow = true; + } + } else { + formatString = std::string(argv[argc - 1]); + } + + if (numInputs < 1) { + std::cout << usage << std::endl; + return 1; + } + + if (option == "cube" && numInputs != 6) { + std::cout << "Cube maps must have 6 inputs." << std::endl; + return 1; + } + + if (option == "array" && numInputs < 2) { + std::cout << "Array maps must have at least 2 inputs." << std::endl; + return 1; + } + + if (formats.find(formatString) == formats.end()) { + std::cout << "Invalid format: " << formatString << std::endl; + std::cout << usage << std::endl; + std::cout << "Formats:" << std::endl; + for (auto & formatName : formatOrder) { + auto format = formats.at(formatName); + std::cout << " " << formatName << " - " << std::get<0>(format) << std::endl; + } + return 1; + } + + /* Check if it ends in SRGB */ + bool srgb = false; + if (formatString.length() > 5 && formatString.substr(formatString.length() - 5, 5) == "_SRGB") { + srgb = true; + } + + bool hdr = false; + if (formatString.substr(0, 3) == "BC6") { + hdr = true; + } + + for (int i = inputsStart; i < (int)(inputsStart + numInputs); i++) { + inputs.push_back(std::string(argv[i])); + } + + output = std::string(argv[inputsStart + numInputs]); + + /* Print inputs and output */ + std::cout << "Inputs: " << std::endl; + for (auto & input : inputs) { + std::cout << " " << input << std::endl; + } + std::cout << "Output: " << output << std::endl; + std::cout << "Format: " << formatString << std::endl; + std::cout << "Speed: " << (fastMode ? "Fast" : verySlow ? "Very slow" : "Slow") << std::endl; + + int isa; + isa = ISPCIsa(); + + std::string isaName; + switch(isa) { + case 0: + isaName = "SSE2"; + break; + case 1: + isaName = "SSE4"; + break; + case 2: + isaName = "AVX2"; + break; + default: + isaName = "Unknown"; + }; + + std::cout << "ISPC ISA: " << isaName << std::endl; + + unsigned char * ldrBufferA, * ldrBufferB, * ldrBufferMain, * ldrBufferOther; + float * hdrBufferA, * hdrBufferB, * hdrBufferMain, * hdrBufferOther; + int width, height, channels; + + int copyChannels = 4; + int forcedChannels = 4; + if (formatString == "BC4") { + copyChannels = 1; + } else if (formatString == "BC5") { + copyChannels = 2; + } + + std::vector>> ldrLevels; + std::vector>> hdrLevels; + + std::vector>>> ldrLevelBlocks; + std::vector>>> hdrLevelBlocks; + + if (hdr) { + hdrLevels.resize(numInputs); + hdrLevelBlocks.resize(numInputs); + } else { + ldrLevels.resize(numInputs); + ldrLevelBlocks.resize(numInputs); + } + + uint32_t levelCount; + + for (int input = 0; input < numInputs; input++) { + int level = 0; + + std::cout << "Loading/scaling " << input << ": " << inputs[input] << std::endl; + + if (hdr) { + hdrBufferA = stbi_loadf(inputs[input].c_str(), &width, &height, &channels, forcedChannels); + if (hdrBufferA == nullptr) { + std::cout << "Failed to load image: " << inputs[input] << std::endl; + return 1; + } + + hdrBufferB = new float[width * height * forcedChannels]; + hdrBufferMain = hdrBufferA; + hdrBufferOther = hdrBufferB; + } else { + ldrBufferA = stbi_load(inputs[input].c_str(), &width, &height, &channels, forcedChannels); + if (ldrBufferA == nullptr) { + std::cout << "Failed to load image: " << inputs[input] << std::endl; + return 1; + } + + ldrBufferB = new unsigned char[width * height * forcedChannels]; + ldrBufferMain = ldrBufferA; + ldrBufferOther = ldrBufferB; + } + + int oldWidth = width; + int oldHeight = height; + + levelCount = 1; + { + int levelWidth = width; + int levelHeight = height; + while (levelWidth > 1 || levelHeight > 1) { + levelWidth = std::max(1, (int)floorf((float)levelWidth / 2)); + levelHeight = std::max(1, (int)floorf((float)levelHeight / 2)); + levelCount++; + } + } + + while(1) { + if (hdr) { + hdrLevels[input].push_back(std::vector(hdrBufferMain, hdrBufferMain + oldWidth * oldHeight * forcedChannels)); + } else { + ldrLevels[input].push_back(std::vector(ldrBufferMain, ldrBufferMain + oldWidth * oldHeight * forcedChannels)); + } + + if (oldWidth == 1 && oldHeight == 1) { + break; + } + + int newWidth = std::max(1, (int)floorf((float)oldWidth / 2)); + int newHeight = std::max(1, (int)floorf((float)oldHeight / 2)); + + stbir_colorspace colorspace = srgb ? STBIR_COLORSPACE_SRGB : STBIR_COLORSPACE_LINEAR; + int alphaChannel = channels == 4 ? 3 : STBIR_ALPHA_CHANNEL_NONE; + + if (hdr) { + int rv = stbir_resize_float_generic(hdrBufferMain, oldWidth, oldHeight, 0, hdrBufferOther, newWidth, newHeight, 0, forcedChannels, alphaChannel, 0, STBIR_EDGE_CLAMP, STBIR_FILTER_MITCHELL, colorspace, nullptr); + if (rv != 1) { + std::cerr << "Error resizing" << std::endl; + } + std::swap(hdrBufferMain, hdrBufferOther); + } else { + int rv = stbir_resize_uint8_generic(ldrBufferMain, oldWidth, oldHeight, 0, ldrBufferOther, newWidth, newHeight, 0, forcedChannels, alphaChannel, 0, STBIR_EDGE_CLAMP, STBIR_FILTER_MITCHELL, colorspace, nullptr); + if (rv != 1) { + std::cerr << "Error resizing" << std::endl; + } + std::swap(ldrBufferMain, ldrBufferOther); + } + + oldWidth = newWidth; + oldHeight = newHeight; + level++; + } + + oldWidth = width; + oldHeight = height; + level = 0; + if (hdr) { + free(hdrBufferA); + delete[] hdrBufferB; + } else { + free(ldrBufferA); + delete[] ldrBufferB; + } + } + + bc6h_enc_settings bc6henc; + bc7_enc_settings bc7enc; + + if (formatString == "BC6H") { + if (fastMode) { + GetProfile_bc6h_basic(&bc6henc); + } else { + if (verySlow) { + GetProfile_bc6h_veryslow(&bc6henc); + } else { + GetProfile_bc6h_slow(&bc6henc); + } + } + } else { + if (channels == 3) { + if (fastMode) { + GetProfile_basic(&bc7enc); + } else { + GetProfile_slow(&bc7enc); + } + } else { + if (fastMode) { + GetProfile_alpha_basic(&bc7enc); + } else { + GetProfile_alpha_slow(&bc7enc); + } + } + } + + std::vector>> levelBlocksCompressed(numInputs); + std::tuple format = formats.find(formatString)->second; + size_t blockSize = std::get<1>(format); + + for (int input = 0; input < numInputs; input++) { + if (numInputs > 1) { + if (option == "cube") { + std::cout << "Face " << input << std::endl; + } else { + std::cout << "Layer " << input << std::endl; + } + } + + if (hdr) { + hdrLevelBlocks[input].resize(levelCount); + } else { + ldrLevelBlocks[input].resize(levelCount); + } + int level = 0; + + int oldWidth = width; + int oldHeight = height; + + while(1) { + unsigned int blocksWidth = (oldWidth + 3) / 4; + unsigned int blocksHeight = (oldHeight + 3) / 4; + + if (hdr) { + hdrLevelBlocks[input][level].resize(blocksWidth * blocksHeight); + + for (unsigned int y = 0; y < blocksHeight; y++) { + for (unsigned int x = 0; x < blocksWidth; x++) { + std::vector block(16 * copyChannels); + + for (unsigned int pixelY = y * 4; pixelY < y * 4 + 4; pixelY++) { + for (unsigned int pixelX = x * 4; pixelX < x * 4 + 4; pixelX++) { + unsigned int clampedY = std::min(pixelY, (unsigned int)oldHeight - 1); + unsigned int clampedX = std::min(pixelX, (unsigned int)oldWidth - 1); + + for (int channel = 0; channel < copyChannels; channel++) { + float value = hdrLevels[input][level][(clampedY * oldWidth + clampedX) * forcedChannels + channel]; + if (value < 0.0f) { + value = 0.0f; + } + + if (value > 65504.0f) { + value = 65504.0f; + } + + block[((pixelY % 4) * 4 + (pixelX % 4)) * copyChannels + channel] = HalfFloat::FromFloat(value); + } + } + } + + hdrLevelBlocks[input][level][blocksWidth * y + x] = block; + } + } + } else { + ldrLevelBlocks[input][level].resize(blocksWidth * blocksHeight); + + for (unsigned int y = 0; y < blocksHeight; y++) { + for (unsigned int x = 0; x < blocksWidth; x++) { + std::vector block(16 * copyChannels); + + for (unsigned int pixelY = y * 4; pixelY < y * 4 + 4; pixelY++) { + for (unsigned int pixelX = x * 4; pixelX < x * 4 + 4; pixelX++) { + unsigned int clampedY = std::min(pixelY, (unsigned int)oldHeight - 1); + unsigned int clampedX = std::min(pixelX, (unsigned int)oldWidth - 1); + + for (int channel = 0; channel < copyChannels; channel++) { + block[((pixelY % 4) * 4 + (pixelX % 4)) * copyChannels + channel] = ldrLevels[input][level][(clampedY * oldWidth + clampedX) * forcedChannels + channel]; + } + } + } + + ldrLevelBlocks[input][level][blocksWidth * y + x] = block; + } + } + } + + if (oldWidth == 1 && oldHeight == 1) { + break; + } + + oldWidth = std::max(1, (int)floorf((float)oldWidth / 2)); + oldHeight = std::max(1, (int)floorf((float)oldHeight / 2)); + level++; + } + + + /* Compress */ + levelBlocksCompressed[input].resize(levelCount); + for (unsigned int l = 0; l < levelCount; l++) { + if (hdr) { + levelBlocksCompressed[input][l].resize(hdrLevelBlocks[input][l].size() * blockSize); + } else { + levelBlocksCompressed[input][l].resize(ldrLevelBlocks[input][l].size() * blockSize); + } + } + + std::mutex mutex; + std::vector completedBlocks(levelCount, 0); + + unsigned int numThreads = std::thread::hardware_concurrency(); + std::vector threads; + unsigned int maxLevel = 0; + for (unsigned t = 0; t < numThreads; t++) { + threads.push_back(std::thread([&, t](){ + for (unsigned int l = 0; l < levelCount; l++) { + unsigned int blocksPerThread; + if (hdr) { + blocksPerThread = hdrLevelBlocks[input][l].size() / numThreads; + } else { + blocksPerThread = ldrLevelBlocks[input][l].size() / numThreads; + } + unsigned int startBlock = t * blocksPerThread; + unsigned int endBlock = startBlock + blocksPerThread; + + if (hdr) { + if (t == numThreads - 1) { + endBlock = hdrLevelBlocks[input][l].size(); + } + } else { + if (t == numThreads - 1) { + endBlock = ldrLevelBlocks[input][l].size(); + } + } + + for (unsigned int b = startBlock; b < endBlock; b++) { + if (formatString == "BC6H") { + rgba_surface surface; + surface.ptr = (uint8_t *)hdrLevelBlocks[input][l][b].data(); + surface.width = 4; + surface.height = 4; + surface.stride = copyChannels * 4 * 2; + + CompressBlocksBC6H(&surface, &levelBlocksCompressed[input][l][b * blockSize], &bc6henc); + } else { + rgba_surface surface; + surface.ptr = ldrLevelBlocks[input][l][b].data(); + surface.width = 4; + surface.height = 4; + surface.stride = copyChannels * 4; + + if (formatString == "BC1" || formatString == "BC1_SRGB") { + CompressBlocksBC1(&surface, &levelBlocksCompressed[input][l][b * blockSize]); + } else if (formatString == "BC3" || formatString == "BC3_SRGB") { + CompressBlocksBC3(&surface, &levelBlocksCompressed[input][l][b * blockSize]); + } else if (formatString == "BC4") { + CompressBlocksBC4(&surface, &levelBlocksCompressed[input][l][b * blockSize]); + } else if (formatString == "BC5") { + CompressBlocksBC5(&surface, &levelBlocksCompressed[input][l][b * blockSize]); + } else if (formatString == "BC7" || formatString == "BC7_SRGB") { + CompressBlocksBC7(&surface, &levelBlocksCompressed[input][l][b * blockSize], &bc7enc); + } + } + + std::lock_guard lock(mutex); + completedBlocks[l]++; + if (completedBlocks[l] % 100 == 0) { + maxLevel = std::max(l, maxLevel); + if (l == maxLevel) { + float progress; + if (hdr) { + progress = (float)completedBlocks[l] / hdrLevelBlocks[input][l].size(); + } else { + progress = (float)completedBlocks[l] / ldrLevelBlocks[input][l].size(); + } + int barWidth = 70; + + std::cout << std::setw(2) << l << " ["; + int pos = barWidth * progress; + for (int i = 0; i < barWidth; ++i) { + if (i < pos) std::cout << "="; + else if (i == pos) std::cout << ">"; + else std::cout << " "; + } + std::cout << "] " << std::setw(2) << int(progress * 100.0) << " %\r"; + std::cout.flush(); + } + } + } + } + })); + } + + for (unsigned t = 0; t < numThreads; t++) { + threads[t].join(); + } + + if (hdr) { + hdrLevelBlocks[input].clear(); + } else { + ldrLevelBlocks[input].clear(); + } + + int barWidth = 70; + std::cout << std::setw(2) << (levelCount - 1) << " ["; + int pos = barWidth; + for (int i = 0; i < barWidth; ++i) { + if (i < pos) std::cout << "="; + else if (i == pos) std::cout << ">"; + else std::cout << " "; + } + std::cout << "] " << std::setw(2) << 100 << " %\r"; + std::cout.flush(); + + std::cout << std::endl; + } + + /* Write KTX2 */ + std::ofstream fh (output, std::ios::out | std::ios::binary); + if (!fh.is_open()) { + std::cout << "Failed to open output file: " << output << std::endl; + return 1; + } + + const uint8_t identifier[] = {0xAB, 0x4B, 0x54, 0x58, 0x20, 0x32, 0x30, 0xBB, 0x0D, 0x0A, 0x1A, 0x0A}; + fh.write((char *)identifier, sizeof(identifier)); + vk::Format vkformat = std::get<2>(formats.at(formatString)); + fh.write((char *)&vkformat, sizeof(vkformat)); + uint32_t typeSize = 1; // Fix for uncompressed vkformats, size of an individual component + fh.write((char *)&typeSize, sizeof(typeSize)); + uint32_t pixelWidth = width; + uint32_t pixelHeight = height; + uint32_t pixelDepth = 0; + + uint32_t layerCount; + uint32_t faceCount; + if (numInputs > 1) { + if (option == "cube") { + layerCount = 0; + faceCount = 6; + } else { + layerCount = numInputs; + faceCount = 1; + } + } else { + layerCount = 0; + faceCount = 1; + } + uint32_t supercompressionScheme = 0; + fh.write((char *)&pixelWidth, sizeof(pixelWidth)); + fh.write((char *)&pixelHeight, sizeof(pixelHeight)); + fh.write((char *)&pixelDepth, sizeof(pixelDepth)); + fh.write((char *)&layerCount, sizeof(layerCount)); + fh.write((char *)&faceCount, sizeof(faceCount)); + fh.write((char *)&levelCount, sizeof(levelCount)); + fh.write((char *)&supercompressionScheme, sizeof(supercompressionScheme)); + + uint32_t * dfd = vk2dfd(*(VkFormat *)&vkformat); + + uint32_t dfdByteOffset = 0; + uint32_t dfdByteLength = dfd[0]; + uint32_t kvdByteOffset = 0; + uint32_t kvdByteLength = 0; + uint64_t sgdByteOffset = 0; + uint64_t sgdByteLength = 0; + + auto dfdByteOffsetPosition = fh.tellp(); + + fh.write((char *)&dfdByteOffset, sizeof(dfdByteOffset)); + fh.write((char *)&dfdByteLength, sizeof(dfdByteLength)); + fh.write((char *)&kvdByteOffset, sizeof(kvdByteOffset)); + fh.write((char *)&kvdByteLength, sizeof(kvdByteLength)); + fh.write((char *)&sgdByteOffset, sizeof(sgdByteOffset)); + fh.write((char *)&sgdByteLength, sizeof(sgdByteLength)); + + auto levelOffsetBytePosition = fh.tellp(); + for (unsigned int i = 0; i < levelCount; i++) { + uint64_t byteOffset = 0; + uint64_t byteLength = 0; + uint64_t uncompressedByteLength = 0; + fh.write((char *)&byteOffset, sizeof(byteOffset)); + fh.write((char *)&byteLength, sizeof(byteLength)); + fh.write((char *)&uncompressedByteLength, sizeof(uncompressedByteLength)); + } + + dfdByteOffset = fh.tellp(); + fh.write((char *)dfd, dfdByteLength); + free(dfd); + + kvdByteOffset = fh.tellp(); + fh.seekp(dfdByteOffsetPosition); + fh.write((char *)&dfdByteOffset, sizeof(dfdByteOffset)); + fh.seekp(kvdByteOffset); + + size_t alignment = std::lcm((size_t)4, (size_t)std::get<1>(formats.at(formatString))); + + for (int level = levelCount - 1; level >= 0; level--) { + // Alignment + while (fh.tellp() % alignment != 0) { + uint8_t padding = 0; + fh.write((char *)&padding, sizeof(padding)); + } + + auto levelBytePosition = fh.tellp(); + fh.seekp(levelOffsetBytePosition + (std::ofstream::pos_type)(level * 24)); + uint64_t byteOffset = levelBytePosition; + fh.write((char *)&byteOffset, sizeof(byteOffset)); + uint64_t byteLength = levelBlocksCompressed[0][level].size() * numInputs; + fh.write((char *)&byteLength, sizeof(byteLength)); + fh.write((char *)&byteLength, sizeof(byteLength)); + fh.seekp(levelBytePosition); + + for (int input = 0; input < numInputs; input++) { + fh.write((char *)levelBlocksCompressed[input][level].data(), levelBlocksCompressed[input][level].size()); + } + } + + return 0; +} \ No newline at end of file diff --git a/Source/createdfd.cpp b/Source/createdfd.cpp new file mode 100644 index 0000000..ea00d8d --- /dev/null +++ b/Source/createdfd.cpp @@ -0,0 +1,659 @@ +/* -*- tab-width: 4; -*- */ +/* vi: set sw=2 ts=4 expandtab: */ + +/* Copyright 2019-2020 The Khronos Group Inc. + * SPDX-License-Identifier: Apache-2.0 + */ + +/** + * @file + * @~English + * @brief Utilities for creating data format descriptors. + */ + +/* + * Author: Andrew Garrard + */ + +#include +#include + +#include "dfd.h" + +typedef enum { i_COLOR, i_NON_COLOR } channels_infotype; + +static uint32_t *writeHeader(int numSamples, int bytes, int suffix, + channels_infotype infotype) +{ + uint32_t *DFD = (uint32_t *) malloc(sizeof(uint32_t) * + (1 + KHR_DF_WORD_SAMPLESTART + + numSamples * KHR_DF_WORD_SAMPLEWORDS)); + uint32_t* BDFD = DFD+1; + DFD[0] = sizeof(uint32_t) * + (1 + KHR_DF_WORD_SAMPLESTART + + numSamples * KHR_DF_WORD_SAMPLEWORDS); + BDFD[KHR_DF_WORD_VENDORID] = + (KHR_DF_VENDORID_KHRONOS << KHR_DF_SHIFT_VENDORID) | + (KHR_DF_KHR_DESCRIPTORTYPE_BASICFORMAT << KHR_DF_SHIFT_DESCRIPTORTYPE); + BDFD[KHR_DF_WORD_VERSIONNUMBER] = + (KHR_DF_VERSIONNUMBER_LATEST << KHR_DF_SHIFT_VERSIONNUMBER) | + (((uint32_t)sizeof(uint32_t) * + (KHR_DF_WORD_SAMPLESTART + + numSamples * KHR_DF_WORD_SAMPLEWORDS) + << KHR_DF_SHIFT_DESCRIPTORBLOCKSIZE)); + BDFD[KHR_DF_WORD_MODEL] = + ((KHR_DF_MODEL_RGBSDA << KHR_DF_SHIFT_MODEL) | /* Only supported model */ + (KHR_DF_FLAG_ALPHA_STRAIGHT << KHR_DF_SHIFT_FLAGS)); + if (infotype == i_COLOR) { + BDFD[KHR_DF_WORD_PRIMARIES] |= KHR_DF_PRIMARIES_BT709 << KHR_DF_SHIFT_PRIMARIES; /* Assumed */ + } else { + BDFD[KHR_DF_WORD_PRIMARIES] |= KHR_DF_PRIMARIES_UNSPECIFIED << KHR_DF_SHIFT_PRIMARIES; + } + if (suffix == s_SRGB) { + BDFD[KHR_DF_WORD_TRANSFER] |= KHR_DF_TRANSFER_SRGB << KHR_DF_SHIFT_TRANSFER; + } else { + BDFD[KHR_DF_WORD_TRANSFER] |= KHR_DF_TRANSFER_LINEAR << KHR_DF_SHIFT_TRANSFER; + } + BDFD[KHR_DF_WORD_TEXELBLOCKDIMENSION0] = 0; /* Only 1x1x1x1 texel blocks supported */ + BDFD[KHR_DF_WORD_BYTESPLANE0] = bytes; /* bytesPlane0 = bytes, bytesPlane3..1 = 0 */ + BDFD[KHR_DF_WORD_BYTESPLANE4] = 0; /* bytesPlane7..5 = 0 */ + return DFD; +} + +static uint32_t setChannelFlags(uint32_t channel, enum VkSuffix suffix) +{ + switch (suffix) { + case s_UNORM: break; + case s_SNORM: + channel |= + KHR_DF_SAMPLE_DATATYPE_SIGNED; + break; + case s_USCALED: break; + case s_SSCALED: + channel |= + KHR_DF_SAMPLE_DATATYPE_SIGNED; + break; + case s_UINT: break; + case s_SINT: + channel |= + KHR_DF_SAMPLE_DATATYPE_SIGNED; + break; + case s_SFLOAT: + channel |= + KHR_DF_SAMPLE_DATATYPE_FLOAT | + KHR_DF_SAMPLE_DATATYPE_SIGNED; + break; + case s_UFLOAT: + channel |= + KHR_DF_SAMPLE_DATATYPE_FLOAT; + break; + case s_SRGB: + if (channel == KHR_DF_CHANNEL_RGBSDA_ALPHA) { + channel |= KHR_DF_SAMPLE_DATATYPE_LINEAR; + } + break; + } + return channel; +} + +static void writeSample(uint32_t *DFD, int sampleNo, int channel, + int bits, int offset, + int topSample, int bottomSample, enum VkSuffix suffix) +{ + // Use this to avoid type-punning complaints from the gcc optimizer + // with -Wall. + union { + uint32_t i; + float f; + } lower, upper; + uint32_t *sample = DFD + 1 + KHR_DF_WORD_SAMPLESTART + sampleNo * KHR_DF_WORD_SAMPLEWORDS; + if (channel == 3) channel = KHR_DF_CHANNEL_RGBSDA_ALPHA; + + if (channel == 3) channel = KHR_DF_CHANNEL_RGBSDA_ALPHA; + channel = setChannelFlags(channel, suffix); + + sample[KHR_DF_SAMPLEWORD_BITOFFSET] = + (offset << KHR_DF_SAMPLESHIFT_BITOFFSET) | + ((bits - 1) << KHR_DF_SAMPLESHIFT_BITLENGTH) | + (channel << KHR_DF_SAMPLESHIFT_CHANNELID); + + sample[KHR_DF_SAMPLEWORD_SAMPLEPOSITION_ALL] = 0; + + switch (suffix) { + case s_UNORM: + case s_SRGB: + default: + if (bits > 32) { + upper.i = 0xFFFFFFFFU; + } else { + upper.i = (uint32_t)((1U << bits) - 1U); + } + lower.i = 0U; + break; + case s_SNORM: + if (bits > 32) { + upper.i = 0x7FFFFFFF; + } else { + upper.i = topSample ? (1U << (bits - 1)) - 1 : (1U << bits) - 1; + } + lower.i = ~upper.i; + if (bottomSample) lower.i += 1; + break; + case s_USCALED: + case s_UINT: + upper.i = bottomSample ? 1U : 0U; + lower.i = 0U; + break; + case s_SSCALED: + case s_SINT: + upper.i = bottomSample ? 1U : 0U; + lower.i = ~0U; + break; + case s_SFLOAT: + upper.f = 1.0f; + lower.f = -1.0f; + break; + case s_UFLOAT: + upper.f = 1.0f; + lower.f = 0.0f; + break; + } + sample[KHR_DF_SAMPLEWORD_SAMPLELOWER] = lower.i; + sample[KHR_DF_SAMPLEWORD_SAMPLEUPPER] = upper.i; +} + +/** + * @~English + * @brief Create a Data Format Descriptor for an unpacked format. + * + * @param bigEndian Set to 1 for big-endian byte ordering and + 0 for little-endian byte ordering. + * @param numChannels The number of color channels. + * @param bytes The number of bytes per channel. + * @param redBlueSwap Normally channels appear in consecutive R, G, B, A order + * in memory; redBlueSwap inverts red and blue, allowing + * B, G, R, A. + * @param suffix Indicates the format suffix for the type. + * + * @return A data format descriptor in malloc'd data. The caller is responsible + * for freeing the descriptor. + **/ +uint32_t *createDFDUnpacked(int bigEndian, int numChannels, int bytes, + int redBlueSwap, enum VkSuffix suffix) +{ + uint32_t *DFD; + if (bigEndian) { + int channelCounter, channelByte; + /* Number of samples = number of channels * bytes per channel */ + DFD = writeHeader(numChannels * bytes, numChannels * bytes, suffix, i_COLOR); + /* First loop over the channels */ + for (channelCounter = 0; channelCounter < numChannels; ++channelCounter) { + int channel = channelCounter; + if (redBlueSwap && (channel == 0 || channel == 2)) { + channel ^= 2; + } + /* Loop over the bytes that constitute a channel */ + for (channelByte = 0; channelByte < bytes; ++channelByte) { + writeSample(DFD, channelCounter * bytes + channelByte, channel, + 8, 8 * (channelCounter * bytes + bytes - channelByte - 1), + channelByte == bytes-1, channelByte == 0, suffix); + } + } + + } else { /* Little-endian */ + + int sampleCounter; + /* One sample per channel */ + DFD = writeHeader(numChannels, numChannels * bytes, suffix, i_COLOR); + for (sampleCounter = 0; sampleCounter < numChannels; ++sampleCounter) { + int channel = sampleCounter; + if (redBlueSwap && (channel == 0 || channel == 2)) { + channel ^= 2; + } + writeSample(DFD, sampleCounter, channel, + 8 * bytes, 8 * sampleCounter * bytes, + 1, 1, suffix); + } + } + return DFD; +} + +/** + * @~English + * @brief Create a Data Format Descriptor for a packed format. + * + * @param bigEndian Big-endian flag: Set to 1 for big-endian byte ordering and + * 0 for little-endian byte ordering. + * @param numChannels The number of color channels. + * @param bits[] An array of length numChannels. + * Each entry is the number of bits composing the channel, in + * order starting at bit 0 of the packed type. + * @param channels[] An array of length numChannels. + * Each entry enumerates the channel type: 0 = red, 1 = green, + * 2 = blue, 15 = alpha, in order starting at bit 0 of the + * packed type. These values match channel IDs for RGBSDA in + * the Khronos Data Format header. To simplify iteration + * through channels, channel id 3 is a synonym for alpha. + * @param suffix Indicates the format suffix for the type. + * + * @return A data format descriptor in malloc'd data. The caller is responsible + * for freeing the descriptor. + **/ +uint32_t *createDFDPacked(int bigEndian, int numChannels, + int bits[], int channels[], + enum VkSuffix suffix) +{ + uint32_t *DFD = 0; + if (numChannels == 6) { + /* Special case E5B9G9R9 */ + DFD = writeHeader(numChannels, 4, s_UFLOAT, i_COLOR); + writeSample(DFD, 0, 0, + 9, 0, + 1, 1, s_UNORM); + KHR_DFDSETSVAL((DFD+1), 0, SAMPLEUPPER, 8448); + writeSample(DFD, 1, 0 | KHR_DF_SAMPLE_DATATYPE_EXPONENT, + 5, 27, + 1, 1, s_UNORM); + KHR_DFDSETSVAL((DFD+1), 1, SAMPLELOWER, 15); + KHR_DFDSETSVAL((DFD+1), 1, SAMPLEUPPER, 31); + writeSample(DFD, 2, 1, + 9, 9, + 1, 1, s_UNORM); + KHR_DFDSETSVAL((DFD+1), 2, SAMPLEUPPER, 8448); + writeSample(DFD, 3, 1 | KHR_DF_SAMPLE_DATATYPE_EXPONENT, + 5, 27, + 1, 1, s_UNORM); + KHR_DFDSETSVAL((DFD+1), 3, SAMPLELOWER, 15); + KHR_DFDSETSVAL((DFD+1), 3, SAMPLEUPPER, 31); + writeSample(DFD, 4, 2, + 9, 18, + 1, 1, s_UNORM); + KHR_DFDSETSVAL((DFD+1), 4, SAMPLEUPPER, 8448); + writeSample(DFD, 5, 2 | KHR_DF_SAMPLE_DATATYPE_EXPONENT, + 5, 27, + 1, 1, s_UNORM); + KHR_DFDSETSVAL((DFD+1), 5, SAMPLELOWER, 15); + KHR_DFDSETSVAL((DFD+1), 5, SAMPLEUPPER, 31); + } else if (bigEndian) { + /* No packed format is larger than 32 bits. */ + /* No packed channel crosses more than two bytes. */ + int totalBits = 0; + int bitChannel[32]; + int beChannelStart[4]; + int channelCounter; + int bitOffset = 0; + int BEMask; + int numSamples = numChannels; + int sampleCounter; + for (channelCounter = 0; channelCounter < numChannels; ++channelCounter) { + beChannelStart[channelCounter] = totalBits; + totalBits += bits[channelCounter]; + } + BEMask = (totalBits - 1) & 0x18; + for (channelCounter = 0; channelCounter < numChannels; ++channelCounter) { + bitChannel[bitOffset ^ BEMask] = channelCounter; + if (((bitOffset + bits[channelCounter] - 1) & ~7) != (bitOffset & ~7)) { + /* Continuation sample */ + bitChannel[((bitOffset + bits[channelCounter] - 1) & ~7) ^ BEMask] = channelCounter; + numSamples++; + } + bitOffset += bits[channelCounter]; + } + DFD = writeHeader(numSamples, totalBits >> 3, suffix, i_COLOR); + + sampleCounter = 0; + for (bitOffset = 0; bitOffset < totalBits;) { + if (bitChannel[bitOffset] == -1) { + /* Done this bit, so this is the lower half of something. */ + /* We must therefore jump to the end of the byte and continue. */ + bitOffset = (bitOffset + 8) & ~7; + } else { + /* Start of a channel? */ + int thisChannel = bitChannel[bitOffset]; + if ((beChannelStart[thisChannel] ^ BEMask) == bitOffset) { + /* Must be just one sample if we hit it first. */ + writeSample(DFD, sampleCounter++, channels[thisChannel], + bits[thisChannel], bitOffset, + 1, 1, suffix); + bitOffset += bits[thisChannel]; + } else { + /* Two samples. Move to the end of the first one we hit when we're done. */ + int firstSampleBits = 8 - (beChannelStart[thisChannel] & 0x7); /* Rest of the byte */ + int secondSampleBits = bits[thisChannel] - firstSampleBits; /* Rest of the bits */ + writeSample(DFD, sampleCounter++, channels[thisChannel], + firstSampleBits, beChannelStart[thisChannel] ^ BEMask, + 0, 1, suffix); + /* Mark that we've already handled this sample */ + bitChannel[beChannelStart[thisChannel] ^ BEMask] = -1; + writeSample(DFD, sampleCounter++, channels[thisChannel], + secondSampleBits, bitOffset, + 1, 0, suffix); + bitOffset += secondSampleBits; + } + } + } + + } else { /* Little-endian */ + + int sampleCounter; + int totalBits = 0; + int bitOffset = 0; + for (sampleCounter = 0; sampleCounter < numChannels; ++sampleCounter) { + totalBits += bits[sampleCounter]; + } + + /* One sample per channel */ + DFD = writeHeader(numChannels, totalBits >> 3, suffix, i_COLOR); + for (sampleCounter = 0; sampleCounter < numChannels; ++sampleCounter) { + writeSample(DFD, sampleCounter, channels[sampleCounter], + bits[sampleCounter], bitOffset, + 1, 1, suffix); + bitOffset += bits[sampleCounter]; + } + } + return DFD; +} + +static khr_df_model_e compModelMapping[] = { + KHR_DF_MODEL_BC1A, /*!< BC1, aka DXT1, no alpha. */ + KHR_DF_MODEL_BC1A, /*!< BC1, aka DXT1, punch-through alpha. */ + KHR_DF_MODEL_BC2, /*!< BC2, aka DXT2 and DXT3. */ + KHR_DF_MODEL_BC3, /*!< BC3, aka DXT4 and DXT5. */ + KHR_DF_MODEL_BC4, /*!< BC4. */ + KHR_DF_MODEL_BC5, /*!< BC5. */ + KHR_DF_MODEL_BC6H, /*!< BC6h HDR format. */ + KHR_DF_MODEL_BC7, /*!< BC7. */ + KHR_DF_MODEL_ETC2, /*!< ETC2 no alpha. */ + KHR_DF_MODEL_ETC2, /*!< ETC2 punch-through alpha. */ + KHR_DF_MODEL_ETC2, /*!< ETC2 independent alpha. */ + KHR_DF_MODEL_ETC2, /*!< R11 ETC2 single-channel. */ + KHR_DF_MODEL_ETC2, /*!< R11G11 ETC2 dual-channel. */ + KHR_DF_MODEL_ASTC, /*!< ASTC. */ + KHR_DF_MODEL_ETC1S, /*!< ETC1S. */ + KHR_DF_MODEL_PVRTC, /*!< PVRTC(1). */ + KHR_DF_MODEL_PVRTC2 /*!< PVRTC2. */ +}; + +static uint32_t compSampleCount[] = { + 1U, /*!< BC1, aka DXT1, no alpha. */ + 1U, /*!< BC1, aka DXT1, punch-through alpha. */ + 2U, /*!< BC2, aka DXT2 and DXT3. */ + 2U, /*!< BC3, aka DXT4 and DXT5. */ + 1U, /*!< BC4. */ + 2U, /*!< BC5. */ + 1U, /*!< BC6h HDR format. */ + 1U, /*!< BC7. */ + 1U, /*!< ETC2 no alpha. */ + 2U, /*!< ETC2 punch-through alpha. */ + 2U, /*!< ETC2 independent alpha. */ + 1U, /*!< R11 ETC2 single-channel. */ + 2U, /*!< R11G11 ETC2 dual-channel. */ + 1U, /*!< ASTC. */ + 1U, /*!< ETC1S. */ + 1U, /*!< PVRTC. */ + 1U /*!< PVRTC2. */ +}; + +static khr_df_model_channels_e compFirstChannel[] = { + KHR_DF_CHANNEL_BC1A_COLOR, /*!< BC1, aka DXT1, no alpha. */ + KHR_DF_CHANNEL_BC1A_ALPHAPRESENT, /*!< BC1, aka DXT1, punch-through alpha. */ + KHR_DF_CHANNEL_BC2_ALPHA, /*!< BC2, aka DXT2 and DXT3. */ + KHR_DF_CHANNEL_BC3_ALPHA, /*!< BC3, aka DXT4 and DXT5. */ + KHR_DF_CHANNEL_BC4_DATA, /*!< BC4. */ + KHR_DF_CHANNEL_BC5_RED, /*!< BC5. */ + KHR_DF_CHANNEL_BC6H_COLOR, /*!< BC6h HDR format. */ + KHR_DF_CHANNEL_BC7_COLOR, /*!< BC7. */ + KHR_DF_CHANNEL_ETC2_COLOR, /*!< ETC2 no alpha. */ + KHR_DF_CHANNEL_ETC2_COLOR, /*!< ETC2 punch-through alpha. */ + KHR_DF_CHANNEL_ETC2_ALPHA, /*!< ETC2 independent alpha. */ + KHR_DF_CHANNEL_ETC2_RED, /*!< R11 ETC2 single-channel. */ + KHR_DF_CHANNEL_ETC2_RED, /*!< R11G11 ETC2 dual-channel. */ + KHR_DF_CHANNEL_ASTC_DATA, /*!< ASTC. */ + KHR_DF_CHANNEL_ETC1S_RGB, /*!< ETC1S. */ + KHR_DF_CHANNEL_PVRTC_COLOR, /*!< PVRTC. */ + KHR_DF_CHANNEL_PVRTC2_COLOR /*!< PVRTC2. */ +}; + +static khr_df_model_channels_e compSecondChannel[] = { + KHR_DF_CHANNEL_BC1A_COLOR, /*!< BC1, aka DXT1, no alpha. */ + KHR_DF_CHANNEL_BC1A_ALPHAPRESENT, /*!< BC1, aka DXT1, punch-through alpha. */ + KHR_DF_CHANNEL_BC2_COLOR, /*!< BC2, aka DXT2 and DXT3. */ + KHR_DF_CHANNEL_BC3_COLOR, /*!< BC3, aka DXT4 and DXT5. */ + KHR_DF_CHANNEL_BC4_DATA, /*!< BC4. */ + KHR_DF_CHANNEL_BC5_GREEN, /*!< BC5. */ + KHR_DF_CHANNEL_BC6H_COLOR, /*!< BC6h HDR format. */ + KHR_DF_CHANNEL_BC7_COLOR, /*!< BC7. */ + KHR_DF_CHANNEL_ETC2_COLOR, /*!< ETC2 no alpha. */ + KHR_DF_CHANNEL_ETC2_ALPHA, /*!< ETC2 punch-through alpha. */ + KHR_DF_CHANNEL_ETC2_COLOR, /*!< ETC2 independent alpha. */ + KHR_DF_CHANNEL_ETC2_RED, /*!< R11 ETC2 single-channel. */ + KHR_DF_CHANNEL_ETC2_GREEN, /*!< R11G11 ETC2 dual-channel. */ + KHR_DF_CHANNEL_ASTC_DATA, /*!< ASTC. */ + KHR_DF_CHANNEL_ETC1S_RGB, /*!< ETC1S. */ + KHR_DF_CHANNEL_PVRTC_COLOR, /*!< PVRTC. */ + KHR_DF_CHANNEL_PVRTC2_COLOR /*!< PVRTC2. */ +}; + +static uint32_t compSecondChannelOffset[] = { + 0U, /*!< BC1, aka DXT1, no alpha. */ + 0U, /*!< BC1, aka DXT1, punch-through alpha. */ + 64U, /*!< BC2, aka DXT2 and DXT3. */ + 64U, /*!< BC3, aka DXT4 and DXT5. */ + 0U, /*!< BC4. */ + 64U, /*!< BC5. */ + 0U, /*!< BC6h HDR format. */ + 0U, /*!< BC7. */ + 0U, /*!< ETC2 no alpha. */ + 0U, /*!< ETC2 punch-through alpha. */ + 64U, /*!< ETC2 independent alpha. */ + 0U, /*!< R11 ETC2 single-channel. */ + 64U, /*!< R11G11 ETC2 dual-channel. */ + 0U, /*!< ASTC. */ + 0U, /*!< ETC1S. */ + 0U, /*!< PVRTC. */ + 0U /*!< PVRTC2. */ +}; + +static uint32_t compChannelBits[] = { + 64U, /*!< BC1, aka DXT1, no alpha. */ + 64U, /*!< BC1, aka DXT1, punch-through alpha. */ + 64U, /*!< BC2, aka DXT2 and DXT3. */ + 64U, /*!< BC3, aka DXT4 and DXT5. */ + 64U, /*!< BC4. */ + 64U, /*!< BC5. */ + 128U, /*!< BC6h HDR format. */ + 128U, /*!< BC7. */ + 64U, /*!< ETC2 no alpha. */ + 64U, /*!< ETC2 punch-through alpha. */ + 64U, /*!< ETC2 independent alpha. */ + 64U, /*!< R11 ETC2 single-channel. */ + 64U, /*!< R11G11 ETC2 dual-channel. */ + 128U, /*!< ASTC. */ + 64U, /*!< ETC1S. */ + 64U, /*!< PVRTC. */ + 64U /*!< PVRTC2. */ +}; + +static uint32_t compBytes[] = { + 8U, /*!< BC1, aka DXT1, no alpha. */ + 8U, /*!< BC1, aka DXT1, punch-through alpha. */ + 16U, /*!< BC2, aka DXT2 and DXT3. */ + 16U, /*!< BC3, aka DXT4 and DXT5. */ + 8U, /*!< BC4. */ + 16U, /*!< BC5. */ + 16U, /*!< BC6h HDR format. */ + 16U, /*!< BC7. */ + 8U, /*!< ETC2 no alpha. */ + 8U, /*!< ETC2 punch-through alpha. */ + 16U, /*!< ETC2 independent alpha. */ + 8U, /*!< R11 ETC2 single-channel. */ + 16U, /*!< R11G11 ETC2 dual-channel. */ + 16U, /*!< ASTC. */ + 8U, /*!< ETC1S. */ + 8U, /*!< PVRTC. */ + 8U /*!< PVRTC2. */ +}; + +/** + * @~English + * @brief Create a Data Format Descriptor for a compressed format. + * + * @param compScheme Vulkan-style compression scheme enumeration. + * @param bwidth Block width in texel coordinates. + * @param bheight Block height in texel coordinates. + * @param bdepth Block depth in texel coordinates. + * @author Mark Callow, Edgewise Consulting. + * @param suffix Indicates the format suffix for the type. + * + * @return A data format descriptor in malloc'd data. The caller is responsible + * for freeing the descriptor. + **/ +uint32_t *createDFDCompressed(enum VkCompScheme compScheme, int bwidth, int bheight, int bdepth, + enum VkSuffix suffix) +{ + uint32_t *DFD = 0; + uint32_t numSamples = compSampleCount[compScheme]; + uint32_t* BDFD; + uint32_t *sample; + uint32_t channel; + // Use union to avoid type-punning complaints from gcc optimizer + // with -Wall. + union { + uint32_t i; + float f; + } lower, upper; + + DFD = (uint32_t *) malloc(sizeof(uint32_t) * + (1 + KHR_DF_WORD_SAMPLESTART + + numSamples * KHR_DF_WORD_SAMPLEWORDS)); + BDFD = DFD+1; + DFD[0] = sizeof(uint32_t) * + (1 + KHR_DF_WORD_SAMPLESTART + + numSamples * KHR_DF_WORD_SAMPLEWORDS); + BDFD[KHR_DF_WORD_VENDORID] = + (KHR_DF_VENDORID_KHRONOS << KHR_DF_SHIFT_VENDORID) | + (KHR_DF_KHR_DESCRIPTORTYPE_BASICFORMAT << KHR_DF_SHIFT_DESCRIPTORTYPE); + BDFD[KHR_DF_WORD_VERSIONNUMBER] = + (KHR_DF_VERSIONNUMBER_LATEST << KHR_DF_SHIFT_VERSIONNUMBER) | + (((uint32_t)sizeof(uint32_t) * + (KHR_DF_WORD_SAMPLESTART + + numSamples * KHR_DF_WORD_SAMPLEWORDS) + << KHR_DF_SHIFT_DESCRIPTORBLOCKSIZE)); + BDFD[KHR_DF_WORD_MODEL] = + ((compModelMapping[compScheme] << KHR_DF_SHIFT_MODEL) | + (KHR_DF_PRIMARIES_BT709 << KHR_DF_SHIFT_PRIMARIES) | /* Assumed */ + (KHR_DF_FLAG_ALPHA_STRAIGHT << KHR_DF_SHIFT_FLAGS)); + + if (suffix == s_SRGB) { + BDFD[KHR_DF_WORD_TRANSFER] |= KHR_DF_TRANSFER_SRGB << KHR_DF_SHIFT_TRANSFER; + } else { + BDFD[KHR_DF_WORD_TRANSFER] |= KHR_DF_TRANSFER_LINEAR << KHR_DF_SHIFT_TRANSFER; + } + BDFD[KHR_DF_WORD_TEXELBLOCKDIMENSION0] = + (bwidth - 1) | ((bheight - 1) << KHR_DF_SHIFT_TEXELBLOCKDIMENSION1) | ((bdepth - 1) << KHR_DF_SHIFT_TEXELBLOCKDIMENSION2); + /* bytesPlane0 = bytes, bytesPlane3..1 = 0 */ + BDFD[KHR_DF_WORD_BYTESPLANE0] = compBytes[compScheme]; + BDFD[KHR_DF_WORD_BYTESPLANE4] = 0; /* bytesPlane7..5 = 0 */ + + sample = BDFD + KHR_DF_WORD_SAMPLESTART; + channel = compFirstChannel[compScheme]; + channel = setChannelFlags(channel, suffix); + + sample[KHR_DF_SAMPLEWORD_BITOFFSET] = + (0 << KHR_DF_SAMPLESHIFT_BITOFFSET) | + ((compChannelBits[compScheme] - 1) << KHR_DF_SAMPLESHIFT_BITLENGTH) | + (channel << KHR_DF_SAMPLESHIFT_CHANNELID); + + sample[KHR_DF_SAMPLEWORD_SAMPLEPOSITION_ALL] = 0; + switch (suffix) { + case s_UNORM: + case s_SRGB: + default: + upper.i = 0xFFFFFFFFU; + lower.i = 0U; + break; + case s_SNORM: + upper.i = 0x7FFFFFFF; + lower.i = ~upper.i; + break; + case s_USCALED: + case s_UINT: + upper.i = 1U; + lower.i = 0U; + break; + case s_SSCALED: + case s_SINT: + upper.i = 1U; + lower.i = ~0U; + break; + case s_SFLOAT: + upper.f = 1.0f; + lower.f = -1.0f; + break; + case s_UFLOAT: + upper.f = 1.0f; + lower.f = 0.0f; + break; + } + sample[KHR_DF_SAMPLEWORD_SAMPLELOWER] = lower.i; + sample[KHR_DF_SAMPLEWORD_SAMPLEUPPER] = upper.i; + + if (compSampleCount[compScheme] > 1) { + sample += KHR_DF_WORD_SAMPLEWORDS; + channel = compSecondChannel[compScheme]; + channel = setChannelFlags(channel, suffix); + + sample[KHR_DF_SAMPLEWORD_BITOFFSET] = + (compSecondChannelOffset[compScheme] << KHR_DF_SAMPLESHIFT_BITOFFSET) | + ((compChannelBits[compScheme] - 1) << KHR_DF_SAMPLESHIFT_BITLENGTH) | + (channel << KHR_DF_SAMPLESHIFT_CHANNELID); + + sample[KHR_DF_SAMPLEWORD_SAMPLEPOSITION_ALL] = 0; + + sample[KHR_DF_SAMPLEWORD_SAMPLELOWER] = lower.i; + sample[KHR_DF_SAMPLEWORD_SAMPLEUPPER] = upper.i; + } + return DFD; +} + +/** + * @~English + * @brief Create a Data Format Descriptor for a depth-stencil format. + * + * @param depthBits The numeber of bits in the depth channel. + * @param stencilBits The numeber of bits in the stencil channel. + * @param sizeBytes The total byte size of the texel. + * + * @return A data format descriptor in malloc'd data. The caller is responsible + * for freeing the descriptor. + **/ +uint32_t *createDFDDepthStencil(int depthBits, + int stencilBits, + int sizeBytes) +{ + /* N.B. Little-endian is assumed. */ + uint32_t *DFD = 0; + DFD = writeHeader((depthBits > 0) + (stencilBits > 0), + sizeBytes, s_UNORM, i_NON_COLOR); + if (depthBits == 32) { + writeSample(DFD, 0, KHR_DF_CHANNEL_RGBSDA_DEPTH, + 32, 0, + 1, 1, s_SFLOAT); + } else if (depthBits > 0) { + writeSample(DFD, 0, KHR_DF_CHANNEL_RGBSDA_DEPTH, + depthBits, 0, + 1, 1, s_UNORM); + } + if (stencilBits > 0) { + if (depthBits > 0) { + writeSample(DFD, 1, KHR_DF_CHANNEL_RGBSDA_STENCIL, + stencilBits, depthBits, + 1, 1, s_UINT); + } else { + writeSample(DFD, 0, KHR_DF_CHANNEL_RGBSDA_STENCIL, + stencilBits, 0, + 1, 1, s_UINT); + } + } + return DFD; +} diff --git a/Source/dfd.h b/Source/dfd.h new file mode 100644 index 0000000..633b475 --- /dev/null +++ b/Source/dfd.h @@ -0,0 +1,173 @@ +/* -*- tab-width: 4; -*- */ +/* vi: set sw=2 ts=4 expandtab: */ + +/* Copyright 2019-2020 The Khronos Group Inc. + * SPDX-License-Identifier: Apache-2.0 + */ + +/** + * @file + * @~English + * @brief Header file defining the data format descriptor utilities API. + */ + +/* + * Author: Andrew Garrard + */ + +#ifndef _DFD_H_ +#define _DFD_H_ + +#include +#include + +#ifdef __cplusplus +extern "C" { +#endif + +/** Qualifier suffix to the format, in Vulkan terms. */ +enum VkSuffix { + s_UNORM, /*!< Unsigned normalized format. */ + s_SNORM, /*!< Signed normalized format. */ + s_USCALED, /*!< Unsigned scaled format. */ + s_SSCALED, /*!< Signed scaled format. */ + s_UINT, /*!< Unsigned integer format. */ + s_SINT, /*!< Signed integer format. */ + s_SFLOAT, /*!< Signed float format. */ + s_UFLOAT, /*!< Unsigned float format. */ + s_SRGB /*!< sRGB normalized format. */ +}; + +/** Compression scheme, in Vulkan terms. */ +enum VkCompScheme { + c_BC1_RGB, /*!< BC1, aka DXT1, no alpha. */ + c_BC1_RGBA, /*!< BC1, aka DXT1, punch-through alpha. */ + c_BC2, /*!< BC2, aka DXT2 and DXT3. */ + c_BC3, /*!< BC3, aka DXT4 and DXT5. */ + c_BC4, /*!< BC4. */ + c_BC5, /*!< BC5. */ + c_BC6H, /*!< BC6h HDR format. */ + c_BC7, /*!< BC7. */ + c_ETC2_R8G8B8, /*!< ETC2 no alpha. */ + c_ETC2_R8G8B8A1, /*!< ETC2 punch-through alpha. */ + c_ETC2_R8G8B8A8, /*!< ETC2 independent alpha. */ + c_EAC_R11, /*!< R11 ETC2 single-channel. */ + c_EAC_R11G11, /*!< R11G11 ETC2 dual-channel. */ + c_ASTC, /*!< ASTC. */ + c_ETC1S, /*!< ETC1S. */ + c_PVRTC, /*!< PVRTC(1). */ + c_PVRTC2 /*!< PVRTC2. */ +}; + +#if !defined(uint32_t) +typedef unsigned int uint32_t; +#endif + +#if !defined(LIBKTX) +#include +#else +#include "../vkformat_enum.h" +#endif + +uint32_t* vk2dfd(enum VkFormat format); + +/* Create a Data Format Descriptor for an unpacked format. */ +uint32_t *createDFDUnpacked(int bigEndian, int numChannels, int bytes, + int redBlueSwap, enum VkSuffix suffix); + +/* Create a Data Format Descriptor for a packed format. */ +uint32_t *createDFDPacked(int bigEndian, int numChannels, + int bits[], int channels[], + enum VkSuffix suffix); + +/* Create a Data Format Descriptor for a compressed format. */ +uint32_t *createDFDCompressed(enum VkCompScheme compScheme, + int bwidth, int bheight, int bdepth, + enum VkSuffix suffix); + +/* Create a Data Format Descriptor for a depth/stencil format. */ +uint32_t *createDFDDepthStencil(int depthBits, + int stencilBits, + int sizeBytes); + +/** @brief Result of interpreting the data format descriptor. */ +enum InterpretDFDResult { + i_LITTLE_ENDIAN_FORMAT_BIT = 0, /*!< Confirmed little-endian (default for 8bpc). */ + i_BIG_ENDIAN_FORMAT_BIT = 1, /*!< Confirmed big-endian. */ + i_PACKED_FORMAT_BIT = 2, /*!< Packed format. */ + i_SRGB_FORMAT_BIT = 4, /*!< sRGB transfer function. */ + i_NORMALIZED_FORMAT_BIT = 8, /*!< Normalized (UNORM or SNORM). */ + i_SIGNED_FORMAT_BIT = 16, /*!< Format is signed. */ + i_FLOAT_FORMAT_BIT = 32, /*!< Format is floating point. */ + i_UNSUPPORTED_ERROR_BIT = 64, /*!< Format not successfully interpreted. */ + /** "NONTRIVIAL_ENDIANNESS" means not big-endian, not little-endian + * (a channel has bits that are not consecutive in either order). **/ + i_UNSUPPORTED_NONTRIVIAL_ENDIANNESS = i_UNSUPPORTED_ERROR_BIT, + /** "MULTIPLE_SAMPLE_LOCATIONS" is an error because only single-sample + * texel blocks (with coordinates 0,0,0,0 for all samples) are supported. **/ + i_UNSUPPORTED_MULTIPLE_SAMPLE_LOCATIONS = i_UNSUPPORTED_ERROR_BIT + 1, + /** "MULTIPLE_PLANES" is an error because only contiguous data is supported. */ + i_UNSUPPORTED_MULTIPLE_PLANES = i_UNSUPPORTED_ERROR_BIT + 2, + /** Only channels R, G, B and A are supported. */ + i_UNSUPPORTED_CHANNEL_TYPES = i_UNSUPPORTED_ERROR_BIT + 3, + /** Only channels with the same flags are supported + * (e.g. we don't support float red with integer green). */ + i_UNSUPPORTED_MIXED_CHANNELS = i_UNSUPPORTED_ERROR_BIT + 4 +}; + +/** @brief Interpretation of a channel from the data format descriptor. */ +typedef struct _InterpretedDFDChannel { + uint32_t offset; /*!< Offset in bits for packed, bytes for unpacked. */ + uint32_t size; /*!< Size in bits for packed, bytes for unpacked. */ +} InterpretedDFDChannel; + +/* Interpret a Data Format Descriptor. */ +enum InterpretDFDResult interpretDFD(const uint32_t *DFD, + InterpretedDFDChannel *R, + InterpretedDFDChannel *G, + InterpretedDFDChannel *B, + InterpretedDFDChannel *A, + uint32_t *wordBytes); + +/* Print a human-readable interpretation of a data format descriptor. */ +void printDFD(uint32_t *DFD); + +/* Get the number of components & component size from a DFD for an + * unpacked format. + */ +void +getDFDComponentInfoUnpacked(const uint32_t* DFD, uint32_t* numComponents, + uint32_t* componentByteLength); + +/* Return the number of components described by a DFD. */ +uint32_t getDFDNumComponents(const uint32_t* DFD); + +/* Recreate and return the value of bytesPlane0 as it should be for the data + * post-inflation from variable-rate compression. + */ +void +recreateBytesPlane0FromSampleInfo(const uint32_t* DFD, uint32_t* bytesPlane0); + +/** @brief Colourspace primaries information. + * + * Structure to store the 1931 CIE x,y chromaticities of the red, green, and blue + * display primaries and the reference white point of a colourspace. + */ +typedef struct _Primaries { + float Rx; /*!< Red x. */ + float Ry; /*!< Red y. */ + float Gx; /*!< Green x. */ + float Gy; /*!< Green y. */ + float Bx; /*!< Blue x. */ + float By; /*!< Blue y. */ + float Wx; /*!< White x. */ + float Wy; /*!< White y. */ +} Primaries; + +khr_df_primaries_e findMapping(Primaries *p, float latitude); + +#ifdef __cplusplus +} +#endif + +#endif /* _DFD_H_ */ diff --git a/Source/ispc_texcomp/ispc_texcomp.cpp b/Source/ispc_texcomp/ispc_texcomp.cpp new file mode 100644 index 0000000..3824020 --- /dev/null +++ b/Source/ispc_texcomp/ispc_texcomp.cpp @@ -0,0 +1,557 @@ +//////////////////////////////////////////////////////////////////////////////// +// Copyright (c) 2016-2019, Intel Corporation +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. +//////////////////////////////////////////////////////////////////////////////// + +#include "ispc_texcomp.h" +#include "kernel_ispc.h" +#include // memcpy + +namespace ispc { +extern "C" { + extern int32_t ISPCIsa_ispc_sse4(); + extern "C" void CompressBlocksBC1_ispc_sse4(const rgba_surface* src, uint8_t* dst); + extern "C" void CompressBlocksBC3_ispc_sse4(const rgba_surface* src, uint8_t* dst); + extern "C" void CompressBlocksBC4_ispc_sse4(const rgba_surface* src, uint8_t* dst); + extern "C" void CompressBlocksBC5_ispc_sse4(const rgba_surface* src, uint8_t* dst); + extern "C" void CompressBlocksBC6H_ispc_sse4(const rgba_surface* src, uint8_t* dst, bc6h_enc_settings* settings); + extern "C" void CompressBlocksBC7_ispc_sse4(const rgba_surface* src, uint8_t* dst, bc7_enc_settings* settings); + extern "C" void CompressBlocksETC1_ispc_sse4(const rgba_surface* src, uint8_t* dst, etc_enc_settings* settings); + extern "C" void CompressBlocksASTC_ispc_sse4(const rgba_surface* src, uint8_t* dst, astc_enc_settings* settings); +} +} + +static bool isAmd = false; + +void ISPCInit() +{ +#if defined(_MSC_VER) +{ + int CPUInfo[4]; + __cpuid(CPUInfo, 0x80000001); + isAmd = (CPUInfo[2] & (1 << 6)) != 0; +} +#else +{ + unsigned int eax = 0x80000001; + unsigned int ebx = 0; + unsigned int ecx = 0; + unsigned int edx = 0; + asm volatile("cpuid" + : "=a"(eax), "=b"(ebx), "=c"(ecx), "=d"(edx) + : "a"(eax)); + isAmd = (ecx & (1 << 6)) != 0; +} +#endif +} + +void GetProfile_ultrafast(bc7_enc_settings* settings) +{ + settings->channels = 3; + + // mode02 + settings->mode_selection[0] = false; + settings->skip_mode2 = true; + + settings->refineIterations[0] = 2; + settings->refineIterations[2] = 2; + + // mode13 + settings->mode_selection[1] = false; + settings->fastSkipTreshold_mode1 = 3; + settings->fastSkipTreshold_mode3 = 1; + settings->fastSkipTreshold_mode7 = 0; + + settings->refineIterations[1] = 2; + settings->refineIterations[3] = 1; + + // mode45 + settings->mode_selection[2] = false; + + settings->mode45_channel0 = 0; + settings->refineIterations_channel = 0; + settings->refineIterations[4] = 2; + settings->refineIterations[5] = 2; + + // mode6 + settings->mode_selection[3] = true; + + settings->refineIterations[6] = 1; +} + +void GetProfile_veryfast(bc7_enc_settings* settings) +{ + settings->channels = 3; + + // mode02 + settings->mode_selection[0] = false; + settings->skip_mode2 = true; + + settings->refineIterations[0] = 2; + settings->refineIterations[2] = 2; + + // mode13 + settings->mode_selection[1] = true; + settings->fastSkipTreshold_mode1 = 3; + settings->fastSkipTreshold_mode3 = 1; + settings->fastSkipTreshold_mode7 = 0; + + settings->refineIterations[1] = 2; + settings->refineIterations[3] = 1; + + // mode45 + settings->mode_selection[2] = false; + + settings->mode45_channel0 = 0; + settings->refineIterations_channel = 0; + settings->refineIterations[4] = 2; + settings->refineIterations[5] = 2; + + // mode6 + settings->mode_selection[3] = true; + + settings->refineIterations[6] = 1; +} + +void GetProfile_fast(bc7_enc_settings* settings) +{ + settings->channels = 3; + + // mode02 + settings->mode_selection[0] = false; + settings->skip_mode2 = true; + + settings->refineIterations[0] = 2; + settings->refineIterations[2] = 2; + + // mode13 + settings->mode_selection[1] = true; + settings->fastSkipTreshold_mode1 = 12; + settings->fastSkipTreshold_mode3 = 4; + settings->fastSkipTreshold_mode7 = 0; + + settings->refineIterations[1] = 2; + settings->refineIterations[3] = 1; + + // mode45 + settings->mode_selection[2] = false; + + settings->mode45_channel0 = 0; + settings->refineIterations_channel = 0; + settings->refineIterations[4] = 2; + settings->refineIterations[5] = 2; + + // mode6 + settings->mode_selection[3] = true; + + settings->refineIterations[6] = 2; +} + +void GetProfile_basic(bc7_enc_settings* settings) +{ + settings->channels = 3; + + // mode02 + settings->mode_selection[0] = true; + settings->skip_mode2 = true; + + settings->refineIterations[0] = 2; + settings->refineIterations[2] = 2; + + // mode13 + settings->mode_selection[1] = true; + settings->fastSkipTreshold_mode1 = 8+4; + settings->fastSkipTreshold_mode3 = 8; + settings->fastSkipTreshold_mode7 = 0; + + settings->refineIterations[1] = 2; + settings->refineIterations[3] = 2; + + // mode45 + settings->mode_selection[2] = true; + + settings->mode45_channel0 = 0; + settings->refineIterations_channel = 2; + settings->refineIterations[4] = 2; + settings->refineIterations[5] = 2; + + // mode6 + settings->mode_selection[3] = true; + + settings->refineIterations[6] = 2; +} + +void GetProfile_slow(bc7_enc_settings* settings) +{ + settings->channels = 3; + + int moreRefine = 2; + // mode02 + settings->mode_selection[0] = true; + settings->skip_mode2 = false; + + settings->refineIterations[0] = 2+moreRefine; + settings->refineIterations[2] = 2+moreRefine; + + // mode13 + settings->mode_selection[1] = true; + settings->fastSkipTreshold_mode1 = 64; + settings->fastSkipTreshold_mode3 = 64; + settings->fastSkipTreshold_mode7 = 0; + + settings->refineIterations[1] = 2+moreRefine; + settings->refineIterations[3] = 2+moreRefine; + + // mode45 + settings->mode_selection[2] = true; + + settings->mode45_channel0 = 0; + settings->refineIterations_channel = 2+moreRefine; + settings->refineIterations[4] = 2+moreRefine; + settings->refineIterations[5] = 2+moreRefine; + + // mode6 + settings->mode_selection[3] = true; + + settings->refineIterations[6] = 2+moreRefine; +} + +void GetProfile_alpha_ultrafast(bc7_enc_settings* settings) +{ + settings->channels = 4; + + // mode02 + settings->mode_selection[0] = false; + settings->skip_mode2 = true; + + settings->refineIterations[0] = 2; + settings->refineIterations[2] = 2; + + // mode137 + settings->mode_selection[1] = false; + settings->fastSkipTreshold_mode1 = 0; + settings->fastSkipTreshold_mode3 = 0; + settings->fastSkipTreshold_mode7 = 4; + + settings->refineIterations[1] = 1; + settings->refineIterations[3] = 1; + settings->refineIterations[7] = 2; + + // mode45 + settings->mode_selection[2] = true; + + settings->mode45_channel0 = 3; + settings->refineIterations_channel = 1; + settings->refineIterations[4] = 1; + settings->refineIterations[5] = 1; + + // mode6 + settings->mode_selection[3] = true; + + settings->refineIterations[6] = 2; +} + +void GetProfile_alpha_veryfast(bc7_enc_settings* settings) +{ + settings->channels = 4; + + // mode02 + settings->mode_selection[0] = false; + settings->skip_mode2 = true; + + settings->refineIterations[0] = 2; + settings->refineIterations[2] = 2; + + // mode137 + settings->mode_selection[1] = true; + settings->fastSkipTreshold_mode1 = 0; + settings->fastSkipTreshold_mode3 = 0; + settings->fastSkipTreshold_mode7 = 4; + + settings->refineIterations[1] = 1; + settings->refineIterations[3] = 1; + settings->refineIterations[7] = 2; + + // mode45 + settings->mode_selection[2] = true; + + settings->mode45_channel0 = 3; + settings->refineIterations_channel = 2; + settings->refineIterations[4] = 2; + settings->refineIterations[5] = 2; + + // mode6 + settings->mode_selection[3] = true; + + settings->refineIterations[6] = 2; +} + +void GetProfile_alpha_fast(bc7_enc_settings* settings) +{ + settings->channels = 4; + + // mode02 + settings->mode_selection[0] = false; + settings->skip_mode2 = true; + + settings->refineIterations[0] = 2; + settings->refineIterations[2] = 2; + + // mode137 + settings->mode_selection[1] = true; + settings->fastSkipTreshold_mode1 = 4; + settings->fastSkipTreshold_mode3 = 4; + settings->fastSkipTreshold_mode7 = 8; + + settings->refineIterations[1] = 1; + settings->refineIterations[3] = 1; + settings->refineIterations[7] = 2; + + // mode45 + settings->mode_selection[2] = true; + + settings->mode45_channel0 = 3; + settings->refineIterations_channel = 2; + settings->refineIterations[4] = 2; + settings->refineIterations[5] = 2; + + // mode6 + settings->mode_selection[3] = true; + + settings->refineIterations[6] = 2; +} + +void GetProfile_alpha_basic(bc7_enc_settings* settings) +{ + settings->channels = 4; + + // mode02 + settings->mode_selection[0] = true; + settings->skip_mode2 = true; + + settings->refineIterations[0] = 2; + settings->refineIterations[2] = 2; + + // mode137 + settings->mode_selection[1] = true; + settings->fastSkipTreshold_mode1 = 8+4; + settings->fastSkipTreshold_mode3 = 8; + settings->fastSkipTreshold_mode7 = 8; + + settings->refineIterations[1] = 2; + settings->refineIterations[3] = 2; + settings->refineIterations[7] = 2; + + // mode45 + settings->mode_selection[2] = true; + + settings->mode45_channel0 = 0; + settings->refineIterations_channel = 2; + settings->refineIterations[4] = 2; + settings->refineIterations[5] = 2; + + // mode6 + settings->mode_selection[3] = true; + + settings->refineIterations[6] = 2; +} + +void GetProfile_alpha_slow(bc7_enc_settings* settings) +{ + settings->channels = 4; + + int moreRefine = 2; + // mode02 + settings->mode_selection[0] = true; + settings->skip_mode2 = false; + + settings->refineIterations[0] = 2+moreRefine; + settings->refineIterations[2] = 2+moreRefine; + + // mode137 + settings->mode_selection[1] = true; + settings->fastSkipTreshold_mode1 = 64; + settings->fastSkipTreshold_mode3 = 64; + settings->fastSkipTreshold_mode7 = 64; + + settings->refineIterations[1] = 2+moreRefine; + settings->refineIterations[3] = 2+moreRefine; + settings->refineIterations[7] = 2+moreRefine; + + // mode45 + settings->mode_selection[2] = true; + + settings->mode45_channel0 = 0; + settings->refineIterations_channel = 2+moreRefine; + settings->refineIterations[4] = 2+moreRefine; + settings->refineIterations[5] = 2+moreRefine; + + // mode6 + settings->mode_selection[3] = true; + + settings->refineIterations[6] = 2+moreRefine; +} + +void GetProfile_bc6h_veryfast(bc6h_enc_settings* settings) +{ + settings->slow_mode = false; + settings->fast_mode = true; + settings->fastSkipTreshold = 0; + settings->refineIterations_1p = 0; + settings->refineIterations_2p = 0; +} + +void GetProfile_bc6h_fast(bc6h_enc_settings* settings) +{ + settings->slow_mode = false; + settings->fast_mode = true; + settings->fastSkipTreshold = 2; + settings->refineIterations_1p = 0; + settings->refineIterations_2p = 1; +} + +void GetProfile_bc6h_basic(bc6h_enc_settings* settings) +{ + settings->slow_mode = false; + settings->fast_mode = false; + settings->fastSkipTreshold = 4; + settings->refineIterations_1p = 2; + settings->refineIterations_2p = 2; +} + +void GetProfile_bc6h_slow(bc6h_enc_settings* settings) +{ + settings->slow_mode = true; + settings->fast_mode = false; + settings->fastSkipTreshold = 10; + settings->refineIterations_1p = 2; + settings->refineIterations_2p = 2; +} + +void GetProfile_bc6h_veryslow(bc6h_enc_settings* settings) +{ + settings->slow_mode = true; + settings->fast_mode = false; + settings->fastSkipTreshold = 32; + settings->refineIterations_1p = 2; + settings->refineIterations_2p = 2; +} + +void GetProfile_etc_slow(etc_enc_settings* settings) +{ + settings->fastSkipTreshold = 6; +} + +void ReplicateBorders(rgba_surface* dst_slice, const rgba_surface* src_tex, int start_x, int start_y, int bpp) +{ + int bytes_per_pixel = bpp >> 3; + + bool aliasing = false; + if (&src_tex->ptr[src_tex->stride * start_y + bytes_per_pixel * start_x] == dst_slice->ptr) aliasing = true; + + for (int y = 0; y < dst_slice->height; y++) + for (int x = 0; x < dst_slice->width; x++) + { + int xx = start_x + x; + int yy = start_y + y; + + if (aliasing && xx < src_tex->width && yy < src_tex->height) continue; + + if (xx >= src_tex->width) xx = src_tex->width - 1; + if (yy >= src_tex->height) yy = src_tex->height - 1; + + void* dst = &dst_slice->ptr[dst_slice->stride * y + bytes_per_pixel * x]; + void* src = &src_tex->ptr[src_tex->stride * yy + bytes_per_pixel * xx]; + + memcpy(dst, src, bytes_per_pixel); + } +} + +void CompressBlocksBC1(const rgba_surface* src, uint8_t* dst) +{ + if (isAmd) { + ispc::CompressBlocksBC1_ispc_sse4((ispc::rgba_surface*)src, dst); + } else { + ispc::CompressBlocksBC1_ispc((ispc::rgba_surface*)src, dst); + } +} + +void CompressBlocksBC3(const rgba_surface* src, uint8_t* dst) +{ + if (isAmd) { + ispc::CompressBlocksBC3_ispc_sse4((ispc::rgba_surface*)src, dst); + } else { + ispc::CompressBlocksBC3_ispc((ispc::rgba_surface*)src, dst); + } +} + +void CompressBlocksBC4(const rgba_surface* src, uint8_t* dst) +{ + if (isAmd) { + ispc::CompressBlocksBC4_ispc_sse4((ispc::rgba_surface*)src, dst); + } else { + ispc::CompressBlocksBC4_ispc((ispc::rgba_surface*)src, dst); + } +} + +void CompressBlocksBC5(const rgba_surface* src, uint8_t* dst) +{ + if (isAmd) { + ispc::CompressBlocksBC5_ispc_sse4((ispc::rgba_surface*)src, dst); + } else { + ispc::CompressBlocksBC5_ispc((ispc::rgba_surface*)src, dst); + } +} + +void CompressBlocksBC7(const rgba_surface* src, uint8_t* dst, bc7_enc_settings* settings) +{ + if (isAmd) { + ispc::CompressBlocksBC7_ispc_sse4((ispc::rgba_surface*)src, dst, (ispc::bc7_enc_settings*)settings); + } else { + ispc::CompressBlocksBC7_ispc((ispc::rgba_surface*)src, dst, (ispc::bc7_enc_settings*)settings); + } +} + +void CompressBlocksBC6H(const rgba_surface* src, uint8_t* dst, bc6h_enc_settings* settings) +{ + if (isAmd) { + ispc::CompressBlocksBC6H_ispc_sse4((ispc::rgba_surface*)src, dst, (ispc::bc6h_enc_settings*)settings); + } else { + ispc::CompressBlocksBC6H_ispc((ispc::rgba_surface*)src, dst, (ispc::bc6h_enc_settings*)settings); + } +} + +void CompressBlocksETC1(const rgba_surface* src, uint8_t* dst, etc_enc_settings* settings) +{ + if (isAmd) { + ispc::CompressBlocksETC1_ispc_sse4((ispc::rgba_surface*)src, dst, (ispc::etc_enc_settings*)settings); + } else { + ispc::CompressBlocksETC1_ispc((ispc::rgba_surface*)src, dst, (ispc::etc_enc_settings*)settings); + } +} + +int ISPCIsa() +{ + if (isAmd) { + return ispc::ISPCIsa_ispc_sse4(); + } else { + return ispc::ISPCIsa_ispc(); + } +} \ No newline at end of file diff --git a/Source/ispc_texcomp/ispc_texcomp.def b/Source/ispc_texcomp/ispc_texcomp.def new file mode 100644 index 0000000..8f4c26c --- /dev/null +++ b/Source/ispc_texcomp/ispc_texcomp.def @@ -0,0 +1,30 @@ +LIBRARY ispc_texcomp +EXPORTS + CompressBlocksBC1 + CompressBlocksBC3 + CompressBlocksBC4 + CompressBlocksBC5 + CompressBlocksBC6H + CompressBlocksBC7 + CompressBlocksETC1 + CompressBlocksASTC + GetProfile_ultrafast + GetProfile_veryfast + GetProfile_fast + GetProfile_basic + GetProfile_slow + GetProfile_alpha_ultrafast + GetProfile_alpha_veryfast + GetProfile_alpha_fast + GetProfile_alpha_basic + GetProfile_alpha_slow + GetProfile_bc6h_veryfast + GetProfile_bc6h_fast + GetProfile_bc6h_basic + GetProfile_bc6h_slow + GetProfile_bc6h_veryslow + GetProfile_etc_slow + GetProfile_astc_fast + GetProfile_astc_alpha_fast + GetProfile_astc_alpha_slow + ReplicateBorders diff --git a/Source/ispc_texcomp/ispc_texcomp.h b/Source/ispc_texcomp/ispc_texcomp.h new file mode 100644 index 0000000..6caf828 --- /dev/null +++ b/Source/ispc_texcomp/ispc_texcomp.h @@ -0,0 +1,128 @@ +//////////////////////////////////////////////////////////////////////////////// +// Copyright (c) 2016-2019, Intel Corporation +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. +//////////////////////////////////////////////////////////////////////////////// + +#include + +struct rgba_surface +{ + uint8_t* ptr; + int32_t width; + int32_t height; + int32_t stride; // in bytes +}; + +struct bc7_enc_settings +{ + bool mode_selection[4]; + int refineIterations[8]; + + bool skip_mode2; + int fastSkipTreshold_mode1; + int fastSkipTreshold_mode3; + int fastSkipTreshold_mode7; + + int mode45_channel0; + int refineIterations_channel; + + int channels; +}; + +struct bc6h_enc_settings +{ + bool slow_mode; + bool fast_mode; + int refineIterations_1p; + int refineIterations_2p; + int fastSkipTreshold; +}; + +struct etc_enc_settings +{ + int fastSkipTreshold; +}; + +struct astc_enc_settings +{ + int block_width; + int block_height; + int channels; + + int fastSkipTreshold; + int refineIterations; +}; + +// profiles for RGB data (alpha channel will be ignored) +extern "C" void GetProfile_ultrafast(bc7_enc_settings* settings); +extern "C" void GetProfile_veryfast(bc7_enc_settings* settings); +extern "C" void GetProfile_fast(bc7_enc_settings* settings); +extern "C" void GetProfile_basic(bc7_enc_settings* settings); +extern "C" void GetProfile_slow(bc7_enc_settings* settings); + +// profiles for RGBA inputs +extern "C" void GetProfile_alpha_ultrafast(bc7_enc_settings* settings); +extern "C" void GetProfile_alpha_veryfast(bc7_enc_settings* settings); +extern "C" void GetProfile_alpha_fast(bc7_enc_settings* settings); +extern "C" void GetProfile_alpha_basic(bc7_enc_settings* settings); +extern "C" void GetProfile_alpha_slow(bc7_enc_settings* settings); + +// profiles for BC6H (RGB HDR) +extern "C" void GetProfile_bc6h_veryfast(bc6h_enc_settings* settings); +extern "C" void GetProfile_bc6h_fast(bc6h_enc_settings* settings); +extern "C" void GetProfile_bc6h_basic(bc6h_enc_settings* settings); +extern "C" void GetProfile_bc6h_slow(bc6h_enc_settings* settings); +extern "C" void GetProfile_bc6h_veryslow(bc6h_enc_settings* settings); + +// profiles for ETC +extern "C" void GetProfile_etc_slow(etc_enc_settings* settings); + +// profiles for ASTC +extern "C" void GetProfile_astc_fast(astc_enc_settings* settings, int block_width, int block_height); +extern "C" void GetProfile_astc_alpha_fast(astc_enc_settings* settings, int block_width, int block_height); +extern "C" void GetProfile_astc_alpha_slow(astc_enc_settings* settings, int block_width, int block_height); + +// helper function to replicate border pixels for the desired block sizes (bpp = 32 or 64) +extern "C" void ReplicateBorders(rgba_surface* dst_slice, const rgba_surface* src_tex, int x, int y, int bpp); + +/* +Notes: + - input width and height need to be a multiple of block size + - LDR input is 32 bit/pixel (sRGB), HDR is 64 bit/pixel (half float) + - for BC4 input is 8bit/pixel (R8), for BC5 input is 16bit/pixel (RG8) + - dst buffer must be allocated with enough space for the compressed texture: + - 8 bytes/block for BC1/BC4/ETC1, + - 16 bytes/block for BC3/BC5/BC6H/BC7/ASTC + - the blocks are stored in raster scan order (natural CPU texture layout) + - use the GetProfile_* functions to select various speed/quality tradeoffs + - the RGB profiles are slightly faster as they ignore the alpha channel +*/ + +extern "C" void CompressBlocksBC1(const rgba_surface* src, uint8_t* dst); +extern "C" void CompressBlocksBC3(const rgba_surface* src, uint8_t* dst); +extern "C" void CompressBlocksBC4(const rgba_surface* src, uint8_t* dst); +extern "C" void CompressBlocksBC5(const rgba_surface* src, uint8_t* dst); +extern "C" void CompressBlocksBC6H(const rgba_surface* src, uint8_t* dst, bc6h_enc_settings* settings); +extern "C" void CompressBlocksBC7(const rgba_surface* src, uint8_t* dst, bc7_enc_settings* settings); +extern "C" void CompressBlocksETC1(const rgba_surface* src, uint8_t* dst, etc_enc_settings* settings); +extern "C" void CompressBlocksASTC(const rgba_surface* src, uint8_t* dst, astc_enc_settings* settings); + +extern "C" void ISPCInit(); +extern "C" int ISPCIsa(); \ No newline at end of file diff --git a/Source/ispc_texcomp/ispc_texcomp.vcxproj b/Source/ispc_texcomp/ispc_texcomp.vcxproj new file mode 100644 index 0000000..0713ed1 --- /dev/null +++ b/Source/ispc_texcomp/ispc_texcomp.vcxproj @@ -0,0 +1,177 @@ +ļ»æ + + + + Debug + Win32 + + + Debug + x64 + + + Release + Win32 + + + Release + x64 + + + + {9B44F7B9-A9AF-45A4-8695-96792A18B052} + ispc_texcomp + 10.0 + + + + DynamicLibrary + true + MultiByte + v142 + + + DynamicLibrary + true + MultiByte + v142 + + + DynamicLibrary + false + true + MultiByte + v142 + + + DynamicLibrary + false + true + MultiByte + v142 + + + + + + + + + + + + + + + + + + + $(SolutionDir);$(ExecutablePath) + + + $(SolutionDir);$(ExecutablePath) + + + $(SolutionDir);$(ExecutablePath) + + + $(SolutionDir);$(ExecutablePath) + + + + Level3 + Disabled + + + true + Windows + ispc_texcomp.def + + + + + Level3 + Disabled + + + true + ispc_texcomp.def + + + + + Level3 + MaxSpeed + true + true + + + true + true + true + ispc_texcomp.def + false + + + + + Level3 + MaxSpeed + true + true + + + true + true + true + ispc_texcomp.def + + + + + + + + + + + + + + + + + + + + + + Document + "$(ProjectDir)..\ISPC\win\ispc.exe" -O2 "%(Filename).ispc" -o "$(TargetDir)%(Filename).obj" -h "$(ProjectDir)%(Filename)_ispc.h" --arch=x86 --target=sse2,sse4,avx,avx2 --opt=fast-math + "$(ProjectDir)..\ISPC\win\ispc.exe" -O2 "%(Filename).ispc" -o "$(TargetDir)%(Filename).obj" -h "$(ProjectDir)%(Filename)_ispc.h" --target=sse2,sse4,avx,avx2 --opt=fast-math + $(TargetDir)%(Filename).obj;$(TargetDir)%(Filename)_sse2.obj;$(TargetDir)%(Filename)_sse4.obj;$(TargetDir)%(Filename)_avx.obj;$(TargetDir)%(Filename)_avx2.obj; + $(TargetDir)%(Filename).obj;$(TargetDir)%(Filename)_sse2.obj;$(TargetDir)%(Filename)_sse4.obj;$(TargetDir)%(Filename)_avx.obj;$(TargetDir)%(Filename)_avx2.obj; + "$(ProjectDir)..\ISPC\win\ispc.exe" -O2 "%(Filename).ispc" -o "$(TargetDir)%(Filename).obj" -h "$(ProjectDir)%(Filename)_ispc.h" --arch=x86 --target=sse2,sse4,avx,avx2 --opt=fast-math + $(TargetDir)%(Filename).obj;$(TargetDir)%(Filename)_sse2.obj;$(TargetDir)%(Filename)_sse4.obj;$(TargetDir)%(Filename)_avx.obj;$(TargetDir)%(Filename)_avx2.obj; + "$(ProjectDir)..\ISPC\win\ispc.exe" -O2 "%(Filename).ispc" -o "$(TargetDir)%(Filename).obj" -h "$(ProjectDir)%(Filename)_ispc.h" --target=sse2,sse4,avx,avx2 --opt=fast-math + $(TargetDir)%(Filename).obj;$(TargetDir)%(Filename)_sse2.obj;$(TargetDir)%(Filename)_sse4.obj;$(TargetDir)%(Filename)_avx.obj;$(TargetDir)%(Filename)_avx2.obj; + + + + + Document + "$(ProjectDir)..\ISPC\win\ispc.exe" -O2 "%(Filename).ispc" -o "$(TargetDir)%(Filename).obj" -h "$(ProjectDir)%(Filename)_ispc.h" --target=sse2,sse4,avx,avx2 --opt=fast-math + $(TargetDir)%(Filename).obj;$(TargetDir)%(Filename)_sse2.obj;$(TargetDir)%(Filename)_sse4.obj;$(TargetDir)%(Filename)_avx.obj;$(TargetDir)%(Filename)_avx2.obj; + "$(ProjectDir)..\ISPC\win\ispc.exe" -O2 "%(Filename).ispc" -o "$(TargetDir)%(Filename).obj" -h "$(ProjectDir)%(Filename)_ispc.h" --target=avx --opt=fast-math + $(TargetDir)%(Filename).obj; + "$(ProjectDir)..\ISPC\win\ispc.exe" -O2 "%(Filename).ispc" -o "$(TargetDir)%(Filename).obj" -h "$(ProjectDir)%(Filename)_ispc.h" --arch=x86 --target=sse2,sse4,avx,avx2 --opt=fast-math + $(TargetDir)%(Filename).obj;$(TargetDir)%(Filename)_sse2.obj;$(TargetDir)%(Filename)_sse4.obj;$(TargetDir)%(Filename)_avx.obj;$(TargetDir)%(Filename)_avx2.obj; + "$(ProjectDir)..\ISPC\win\ispc.exe" -O2 "%(Filename).ispc" -o "$(TargetDir)%(Filename).obj" -h "$(ProjectDir)%(Filename)_ispc.h" --arch=x86 --target=avx --opt=fast-math + $(TargetDir)%(Filename).obj; + + + + + + \ No newline at end of file diff --git a/Source/ispc_texcomp/ispc_texcomp.vcxproj.filters b/Source/ispc_texcomp/ispc_texcomp.vcxproj.filters new file mode 100644 index 0000000..7a98ac4 --- /dev/null +++ b/Source/ispc_texcomp/ispc_texcomp.vcxproj.filters @@ -0,0 +1,62 @@ +ļ»æ + + + + {4FC737F1-C7A5-4376-A066-2A32D752A2FF} + + + {93995380-89BD-4b04-88EB-625FBE52EBFB} + + + + + Source Files + + + Source Files + + + + + Source Files + + + Source Files + + + + + Source Files + + + Generated Header Files + + + Generated Header Files + + + Generated Header Files + + + Generated Header Files + + + Generated Header Files + + + Generated Header Files + + + Generated Header Files + + + Generated Header Files + + + Generated Header Files + + + Generated Header Files + + + diff --git a/Source/ispc_texcomp/ispc_texcomp_astc.cpp b/Source/ispc_texcomp/ispc_texcomp_astc.cpp new file mode 100644 index 0000000..045651a --- /dev/null +++ b/Source/ispc_texcomp/ispc_texcomp_astc.cpp @@ -0,0 +1,564 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// Copyright (c) 2016, Intel Corporation +// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated +// documentation files (the "Software"), to deal in the Software without restriction, including without limitation +// the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to +// permit persons to whom the Software is furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in all copies or substantial portions of +// the Software. +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO +// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, +// TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +// SOFTWARE. +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// + +#include "ispc_texcomp.h" +#include "kernel_astc_ispc.h" +#include +#include +#include +#include +#include + +void GetProfile_astc_fast(astc_enc_settings* settings, int block_width, int block_height) +{ + settings->block_width = block_width; + settings->block_height = block_height; + settings->channels = 3; + + settings->fastSkipTreshold = 5; + settings->refineIterations = 2; +} + +void GetProfile_astc_alpha_fast(astc_enc_settings* settings, int block_width, int block_height) +{ + settings->block_width = block_width; + settings->block_height = block_height; + settings->channels = 4; + + settings->fastSkipTreshold = 5; + settings->refineIterations = 2; +} + +void GetProfile_astc_alpha_slow(astc_enc_settings* settings, int block_width, int block_height) +{ + settings->block_width = block_width; + settings->block_height = block_height; + settings->channels = 4; + + settings->fastSkipTreshold = 64; + settings->refineIterations = 2; +} + +struct astc_block +{ + int width; + int height; + uint8_t dual_plane; + int weight_range; + uint8_t weights[64]; + int color_component_selector; + + int partitions; + int partition_id; + int color_endpoint_pairs; + int channels; + int color_endpoint_modes[4]; + int endpoint_range; + uint8_t endpoints[18]; +}; + +bool can_store(int value, int bits) +{ + if (value < 0) return false; + if (value >= 1 << bits) return false; + return true; +} + +int pack_block_mode(astc_block* block) +{ + int block_mode = 0; + + int D = block->dual_plane; + int H = block->weight_range >= 6; + int DH = D * 2 + H; + int R = block->weight_range + 2 - ((H > 0) ? 6 : 0); + R = R / 2 + R % 2 * 4; + + if (can_store(block->width - 4, 2) && can_store(block->height - 2, 2)) + { + int B = block->width - 4; + int A = block->height - 2; + + block_mode = (DH << 9) | (B << 7) | (A << 5) | ((R & 4) << 2) | (R & 3); + } + + if (can_store(block->width - 8, 2) && can_store(block->height - 2, 2)) + { + int B = block->width - 8; + int A = block->height - 2; + + block_mode = (DH << 9) | (B << 7) | (A << 5) | ((R & 4) << 2) | 4 | (R & 3); + } + + if (can_store(block->width - 2, 2) && can_store(block->height - 8, 2)) + { + int A = block->width - 2; + int B = block->height - 8; + + block_mode = (DH << 9) | (B << 7) | (A << 5) | ((R & 4) << 2) | 8 | (R & 3); + } + + if (can_store(block->width - 2, 2) && can_store(block->height - 6, 1)) + { + int A = block->width - 2; + int B = block->height - 6; + + block_mode = (DH << 9) | (B << 7) | (A << 5) | ((R & 4) << 2) | 12 | (R & 3); + } + + if (can_store(block->width - 2, 1) && can_store(block->height - 2, 2)) + { + int B = block->width; + int A = block->height - 2; + + block_mode = (DH << 9) | (B << 7) | (A << 5) | ((R & 4) << 2) | 12 | (R & 3); + } + + if (DH == 0 && can_store(block->width - 6, 2) && can_store(block->height - 6, 2)) + { + int A = block->width - 6; + int B = block->height - 6; + + block_mode = (B << 9) | 256 | (A << 5) | (R << 2); + } + + return block_mode; +} + +int range_table[][3] = +{ + //2^ 3^ 5^ + { 1, 0, 0 }, // 0..1 + { 0, 1, 0 }, // 0..2 + { 2, 0, 0 }, // 0..3 + + { 0, 0, 1 }, // 0..4 + { 1, 1, 0 }, // 0..5 + { 3, 0, 0 }, // 0..7 + + { 1, 0, 1 }, // 0..9 + { 2, 1, 0 }, // 0..11 + { 4, 0, 0 }, // 0..15 + + { 2, 0, 1 }, // 0..19 + { 3, 1, 0 }, // 0..23 + { 5, 0, 0 }, // 0..31 + + { 3, 0, 1 }, // 0..39 + { 4, 1, 0 }, // 0..47 + { 6, 0, 0 }, // 0..63 + + { 4, 0, 1 }, // 0..79 + { 5, 1, 0 }, // 0..95 + { 7, 0, 0 }, // 0..127 + + { 5, 0, 1 }, // 0..159 + { 6, 1, 0 }, // 0..191 + { 8, 0, 0 }, // 0..255 +}; + +int get_levels(int range) +{ + return (1 + 2 * range_table[range][1] + 4 * range_table[range][2]) << range_table[range][0]; +} + +int sequence_bits(int count, int range) +{ + int bits = count * range_table[range][0]; + bits += (count * range_table[range][1] * 8 + 4) / 5; + bits += (count * range_table[range][2] * 7 + 2) / 3; + return bits; +} + +void set_bits(uint32_t data[4], int* pos, int bits, uint32_t value) +{ + assert(bits <= 25); + uint32_t word = *(uint32_t*)(((uint8_t*)data) + *pos / 8); + + uint32_t mask = (1 << bits) - 1; + word |= value << (*pos % 8); + + *(uint32_t*)(((uint8_t*)data) + *pos / 8) = word; + *pos += bits; +} + +uint32_t get_field(uint32_t input, int a, int b) +{ + assert(a >= b); + return (input >> b) & ((1 << (a - b + 1)) - 1); +} + +uint32_t get_bit(uint32_t input, int a) +{ + return get_field(input, a, a); +} + +void pack_five_trits(uint32_t data[4], int sequence[5], int* pos, int n) +{ + int t[5]; + int m[5]; + + for (int i = 0; i < 5; i++) + { + t[i] = sequence[i] >> n; + m[i] = sequence[i] - (t[i] << n); + } + + int C; + + if (t[1] == 2 && t[2] == 2) + { + C = 3 * 4 + t[0]; + } + else if (t[2] == 2) + { + C = t[1] * 16 + t[0] * 4 + 3; + } + else + { + C = t[2] * 16 + t[1] * 4 + t[0]; + } + + int T; + + if (t[3] == 2 && t[4] == 2) + { + T = get_field(C, 4, 2) * 32 + 7 * 4 + get_field(C, 1, 0); + } + else + { + T = get_field(C, 4, 0); + if (t[4] == 2) + { + T += t[3] * 128 + 3 * 32; + } + else + { + T += t[4] * 128 + t[3] * 32; + } + } + + uint32_t pack1 = 0; + pack1 |= m[0]; + pack1 |= get_field(T, 1, 0) << n; + pack1 |= m[1] << (2 + n); + + uint32_t pack2 = 0; + pack2 |= get_field(T, 3, 2); + pack2 |= m[2] << 2; + pack2 |= get_field(T, 4, 4) << (2 + n); + pack2 |= m[3] << (3 + n); + pack2 |= get_field(T, 6, 5) << (3 + n * 2); + pack2 |= m[4] << (5 + n * 2); + pack2 |= get_field(T, 7, 7) << (5 + n * 3); + + set_bits(data, pos, 2 + n * 2, pack1); + set_bits(data, pos, 6 + n * 3, pack2); +} + +void pack_three_quint(uint32_t data[4], int sequence[3], int* pos, int n) +{ + int q[3]; + int m[3]; + + for (int i = 0; i < 3; i++) + { + q[i] = sequence[i] >> n; + m[i] = sequence[i] - (q[i] << n); + } + + int Q; + + if (q[0] == 4 && q[1] == 4) + { + Q = get_field(q[2], 1, 0) * 8 + 3 * 2 + get_bit(q[2], 2); + } + else + { + int C; + if (q[1] == 4) + { + C = (q[0] << 3) + 5; + } + else + { + C = (q[1] << 3) + q[0]; + } + + if (q[2] == 4) + { + Q = get_field(~C, 2, 1) * 32 + get_field(C, 4, 3) * 8 + 3 * 2 + get_bit(C, 0); + } + else + { + Q = q[2] * 32 + get_field(C, 4, 0); + } + } + + uint32_t pack = 0; + pack |= m[0]; + pack |= get_field(Q, 2, 0) << n; + pack |= m[1] << (3 + n); + pack |= get_field(Q, 4, 3) << (3 + n * 2); + pack |= m[2] << (5 + n * 2); + pack |= get_field(Q, 6, 5) << (5 + n * 3); + + set_bits(data, pos, 7 + n * 3, pack); +} + +void pack_integer_sequence(uint32_t output_data[4], uint8_t sequence[], int pos, int count, int range) +{ + int n = range_table[range][0]; + int bits = sequence_bits(count, range); + int pos0 = pos; + + uint32_t data[5] = { 0 }; + if (range_table[range][1] == 1) + { + for (int j = 0; j < (count + 4) / 5; j++) + { + int temp[5] = { 0 }; + for (int i = 0; i < std::min(count - j * 5, 5); i++) temp[i] = sequence[j * 5 + i]; + pack_five_trits(data, temp, &pos, n); + } + } + else if (range_table[range][2] == 1) + { + for (int j = 0; j < (count + 2) / 3; j++) + { + int temp[3] = { 0 }; + for (int i = 0; i < std::min(count - j * 3, 3); i++) temp[i] = sequence[j * 3 + i]; + pack_three_quint(data, temp, &pos, n); + } + } + else + { + for (int i = 0; i < count; i++) + { + set_bits(data, &pos, n, sequence[i]); + } + } + + if (pos0 + bits < 96) data[3] = 0; + if (pos0 + bits < 64) data[2] = 0; + if (pos0 + bits < 32) data[1] = 0; + data[(pos0 + bits) / 32] &= (1 << ((pos0 + bits) % 32)) - 1; + + for (int k = 0; k < 4; k++) output_data[k] |= data[k]; +} + +uint32_t reverse_bits_32(uint32_t input) +{ + uint32_t t = input; + t = (t << 16) | (t >> 16); + t = ((t & 0x00FF00FF) << 8) | ((t & 0xFF00FF00) >> 8); + t = ((t & 0x0F0F0F0F) << 4) | ((t & 0xF0F0F0F0) >> 4); + t = ((t & 0x33333333) << 2) | ((t & 0xCCCCCCCC) >> 2); + t = ((t & 0x55555555) << 1) | ((t & 0xAAAAAAAA) >> 1); + + return t; +} + +void pack_block(uint32_t data[4], astc_block* block) +{ + memset(data, 0, 16); + + int pos = 0; + set_bits(data, &pos, 11, pack_block_mode(block)); + + int num_weights = block->width * block->height * (block->dual_plane ? 2 : 1); + int weight_bits = sequence_bits(num_weights, block->weight_range); + int extra_bits = 0; + + assert(num_weights <= 64); + assert(24 <= weight_bits && weight_bits <= 96); + + set_bits(data, &pos, 2, block->partitions - 1); + if (block->partitions > 1) + { + set_bits(data, &pos, 10, block->partition_id); + + int min_cem = 16; + int max_cem = 0; + for (int j = 0; j < block->partitions; j++) + { + min_cem = std::min(min_cem, block->color_endpoint_modes[j]); + max_cem = std::max(max_cem, block->color_endpoint_modes[j]); + } + assert(max_cem / 4 <= min_cem / 4 + 1); + + int CEM = block->color_endpoint_modes[0] << 2; + if (max_cem != min_cem) + { + CEM = std::min(3, min_cem / 4 + 1); + for (int j = 0; j < block->partitions; j++) + { + int c = block->color_endpoint_modes[j] / 4 - ((CEM & 3) - 1); + int m = block->color_endpoint_modes[j] % 4; + assert(c == 0 || c == 1); + CEM |= c << (2 + j); + CEM |= m << (2 + block->partitions + 2 * j); + } + extra_bits = 3 * block->partitions - 4; + int pos2 = 128 - weight_bits - extra_bits; + set_bits(data, &pos2, extra_bits, CEM >> 6); + } + + set_bits(data, &pos, 6, CEM & 63); + } + else + { + set_bits(data, &pos, 4, block->color_endpoint_modes[0]); + } + + if (block->dual_plane) + { + assert(block->partitions < 4); + extra_bits += 2; + int pos2 = 128 - weight_bits - extra_bits; + set_bits(data, &pos2, 2, block->color_component_selector); + } + + int config_bits = pos + extra_bits; + int remaining_bits = 128 - config_bits - weight_bits; + + int num_cem_pairs = 0; + for (int j = 0; j < block->partitions; j++) num_cem_pairs += 1 + block->color_endpoint_modes[j] / 4; + + assert(num_cem_pairs <= 9); + + int endpoint_range = -1; + for (int range = 20; range>0; range--) + { + int bits = sequence_bits(2 * num_cem_pairs, range); + if (bits <= remaining_bits) + { + endpoint_range = range; + break; + } + } + + assert(endpoint_range >= 4); + assert(block->endpoint_range == endpoint_range); + + pack_integer_sequence(data, block->endpoints, pos, 2 * num_cem_pairs, endpoint_range); + + uint32_t rdata[4] = { 0, 0, 0, 0 }; + pack_integer_sequence(rdata, block->weights, 0, num_weights, block->weight_range); + + for (int i = 0; i < 4; i++) data[i] |= reverse_bits_32(rdata[3 - i]); +} + +void atsc_rank(const rgba_surface* src, int xx, int yy, uint32_t* mode_buffer, astc_enc_settings* settings) +{ + ispc::astc_rank_ispc((ispc::rgba_surface*)src, xx, yy, mode_buffer, (ispc::astc_enc_settings*)settings); +} + +extern "C" void pack_block_c(uint32_t data[4], ispc::astc_block* block) +{ + assert(sizeof(ispc::astc_block) == sizeof(astc_block)); + pack_block(data, (astc_block*)block); +} + +void setup_list_context(ispc::astc_enc_context* ctx, uint32_t packed_mode) +{ + ctx->width = 2 + get_field(packed_mode, 15, 13); // 2..8 <= 2^3 + ctx->height = 2 + get_field(packed_mode, 18, 16); // 2..8 <= 2^3 + ctx->dual_plane = get_field(packed_mode, 19, 19); // 0 or 1 + ctx->partitions = 1; + + int color_endpoint_modes0 = get_field(packed_mode, 7, 6) * 2 + 6; // 6, 8, 10 or 12 + ctx->color_endpoint_pairs = 1 + (color_endpoint_modes0 / 4); + + ctx->channels = (color_endpoint_modes0 > 8) ? 4 : 3; +} + +void astc_encode(const rgba_surface* src, float* block_scores, uint8_t* dst, uint64_t* list, astc_enc_settings* settings) +{ + ispc::astc_enc_context list_context; + setup_list_context(&list_context, uint32_t(list[1] & 0xFFFFFFFF)); + + assert(sizeof(ispc::rgba_surface) == sizeof(rgba_surface)); + assert(sizeof(ispc::astc_enc_settings) == sizeof(astc_enc_settings)); + ispc::astc_encode_ispc((ispc::rgba_surface*)src, block_scores, dst, list, &list_context, (ispc::astc_enc_settings*)settings); +} + +void CompressBlocksASTC(const rgba_surface* src, uint8_t* dst, astc_enc_settings* settings) +{ + assert(src->height % settings->block_height == 0); + assert(src->width % settings->block_width == 0); + + assert(settings->block_height <= 8); + assert(settings->block_width <= 8); + + int tex_width = src->width / settings->block_width; + int programCount = ispc::get_programCount(); + + std::vector block_scores(tex_width * src->height / settings->block_height); + + for (int yy = 0; yy < src->height / settings->block_height; yy++) + for (int xx = 0; xx < tex_width; xx++) + { + block_scores[yy * tex_width + xx] = std::numeric_limits::infinity(); + } + + int mode_list_size = 3334; + int list_size = programCount; + std::vector mode_lists(list_size * mode_list_size); + std::vector mode_buffer(programCount * settings->fastSkipTreshold); + + for (int yy = 0; yy < src->height / settings->block_height; yy++) + for (int _x = 0; _x < (tex_width + programCount - 1) / programCount; _x++) + { + int xx = _x * programCount; + atsc_rank(src, xx, yy, mode_buffer.data(), settings); + + for (int i = 0; i < settings->fastSkipTreshold; i++) + for (int k = 0; k < programCount; k++) + { + if (xx + k >= tex_width) continue; + + uint32_t offset = (yy << 16) + (xx + k); + uint32_t mode = mode_buffer[programCount * i + k]; + int mode_bin = mode >> 20; + uint64_t* mode_list = &mode_lists[list_size * mode_bin]; + + if (*mode_list < programCount - 1) + { + int index = int(mode_list[0] + 1); + mode_list[0] = index; + + mode_list[index] = (uint64_t(offset) << 32) + mode; + } + else + { + mode_list[0] = (uint64_t(offset) << 32) + mode; + + astc_encode(src, block_scores.data(), dst, mode_list, settings); + memset(mode_list, 0, list_size * sizeof(uint64_t)); + } + } + } + + for (int mode_bin = 0; mode_bin < mode_list_size; mode_bin++) + { + uint64_t* mode_list = &mode_lists[list_size * mode_bin]; + if (mode_list[0] == 0) continue; + mode_list[0] = 0; + + astc_encode(src, block_scores.data(), dst, mode_list, settings); + memset(mode_list, 0, list_size * sizeof(uint64_t)); + } +} diff --git a/Source/ispc_texcomp/kernel.ispc b/Source/ispc_texcomp/kernel.ispc new file mode 100644 index 0000000..9a284b5 --- /dev/null +++ b/Source/ispc_texcomp/kernel.ispc @@ -0,0 +1,3798 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// Copyright (c) 2016, Intel Corporation +// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated +// documentation files (the "Software"), to deal in the Software without restriction, including without limitation +// the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to +// permit persons to whom the Software is furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in all copies or substantial portions of +// the Software. +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO +// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, +// TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +// SOFTWARE. +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// + +#ifndef ISPC_UINT_IS_DEFINED +//these are defined in ISPC version 1.13.0 and later +typedef unsigned int8 uint8; +typedef unsigned int32 uint32; +typedef unsigned int64 uint64; +#endif + +/////////////////////////// +// generic helpers + +inline float RCP(float x) +{ + return 1.0f/x; // uses rcp when compiled with --opt=fast-math + //return rcp(x); + //return rcp_fast(x); +} + +inline float RSQRT(float x) +{ + return 1.0f/sqrt(x); // uses rsqrt when compiled with --opt=fast-math + //return rsqrt(x); + //return rsqrt_fast(x); +} + +inline void swap_ints(int u[], int v[], uniform int n) +{ + for (uniform int i=0; i>bits; // (perf warning expected) +} + +/////////////////////////////////////////////////////////// +// BC1/BC7 shared + +struct rgba_surface +{ + uint8* ptr; + int width, height, stride; +}; + +inline void load_block_interleaved(float block[48], uniform rgba_surface* uniform src, int xx, uniform int yy) +{ + for (uniform int y = 0; y<4; y++) + for (uniform int x = 0; x<4; x++) + { + uniform unsigned int32* uniform src_ptr = (unsigned int32*)&src->ptr[(yy * 4 + y)*src->stride]; + unsigned int32 rgba = gather_uint(src_ptr, xx * 4 + x); + + block[16 * 0 + y * 4 + x] = (int)((rgba >> 0) & 255); + block[16 * 1 + y * 4 + x] = (int)((rgba >> 8) & 255); + block[16 * 2 + y * 4 + x] = (int)((rgba >> 16) & 255); + } +} + +inline void load_block_interleaved_rgba(float block[64], uniform rgba_surface* uniform src, int xx, uniform int yy) +{ + for (uniform int y=0; y<4; y++) + for (uniform int x=0; x<4; x++) + { + uniform unsigned int32* uniform src_ptr = (unsigned int32*)&src->ptr[(yy*4+y)*src->stride]; + unsigned int32 rgba = gather_uint(src_ptr, xx*4+x); + + block[16*0+y*4+x] = (int)((rgba>> 0)&255); + block[16*1+y*4+x] = (int)((rgba>> 8)&255); + block[16*2+y*4+x] = (int)((rgba>>16)&255); + block[16*3+y*4+x] = (int)((rgba>>24)&255); + } +} + +inline void load_block_interleaved_16bit(float block[48], uniform rgba_surface* uniform src, int xx, uniform int yy) +{ + for (uniform int y = 0; y<4; y++) + for (uniform int x = 0; x<4; x++) + { + uniform unsigned int32* uniform src_ptr_r = (unsigned int32*)&src->ptr[(yy * 4 + y)*src->stride + 0]; + uniform unsigned int32* uniform src_ptr_g = (unsigned int32*)&src->ptr[(yy * 4 + y)*src->stride + 2]; + uniform unsigned int32* uniform src_ptr_b = (unsigned int32*)&src->ptr[(yy * 4 + y)*src->stride + 4]; + unsigned int32 xr = gather_uint(src_ptr_r, (xx * 4 + x) * 2); + unsigned int32 xg = gather_uint(src_ptr_g, (xx * 4 + x) * 2); + unsigned int32 xb = gather_uint(src_ptr_b, (xx * 4 + x) * 2); + + block[16 * 0 + y * 4 + x] = (int)(xr & 0xFFFF); + block[16 * 1 + y * 4 + x] = (int)(xg & 0xFFFF); + block[16 * 2 + y * 4 + x] = (int)(xb & 0xFFFF); + block[16 * 3 + y * 4 + x] = 0; + } +} + +inline void load_block_r_8bit(float block[16], uniform rgba_surface* uniform src, int xx, uniform int yy) +{ + for (uniform int y=0; y<4; y++) + { + uniform unsigned int32* uniform src_ptr = (unsigned int32*)&src->ptr[(yy*4+y)*src->stride]; + unsigned int32 rrrr = gather_uint(src_ptr, xx); + + block[y*4+0] = (int)((rrrr>> 0)&255); + block[y*4+1] = (int)((rrrr>> 8)&255); + block[y*4+2] = (int)((rrrr>>16)&255); + block[y*4+3] = (int)((rrrr>>24)&255); + } +} + +inline void load_block_interleaved_rg_8bit(float block[32], uniform rgba_surface* uniform src, int xx, uniform int yy) +{ + for (uniform int y=0; y<4; y++) + { + uniform unsigned int32* uniform src_ptr = (unsigned int32*)&src->ptr[(yy*4+y)*src->stride]; + unsigned int32 rgrg0 = gather_uint(src_ptr, xx * 2 + 0); + unsigned int32 rgrg1 = gather_uint(src_ptr, xx * 2 + 1); + + // r + block[16*0+y*4+0] = (int)((rgrg0>> 0)&255); + block[16*0+y*4+1] = (int)((rgrg0>>16)&255); + block[16*0+y*4+2] = (int)((rgrg1>> 0)&255); + block[16*0+y*4+3] = (int)((rgrg1>>16)&255); + // g + block[16*1+y*4+0] = (int)((rgrg0>> 8)&255); + block[16*1+y*4+1] = (int)((rgrg0>>24)&255); + block[16*1+y*4+2] = (int)((rgrg1>> 8)&255); + block[16*1+y*4+3] = (int)((rgrg1>>24)&255); + } +} + +inline void store_data(uniform uint8 dst[], int width, int xx, uniform int yy, uint32 data[], int data_size) +{ + for (uniform int k=0; k> 8)) >> 8; +} + +inline unsigned int16 stb__As16Bit(int r, int g, int b) +{ + return (stb__Mul8Bit(r,31) << 11) + (stb__Mul8Bit(g,63) << 5) + stb__Mul8Bit(b,31); +} + +inline unsigned int16 enc_rgb565(float c[3]) +{ + return stb__As16Bit((int)c[0], (int)c[1], (int)c[2]); +} + +inline void dec_rgb565(float c[3], int p) +{ + int c2 = (p>>0)&31; + int c1 = (p>>5)&63; + int c0 = (p>>11)&31; + + c[0] = (c0<<3)+(c0>>2); + c[1] = (c1<<2)+(c1>>4); + c[2] = (c2<<3)+(c2>>2); +} + +inline void pick_endpoints_dc(int c0[3], int c1[3], int block[48], int iaxis[3]) +{ + for (uniform int p=0; p<3; p++) + for (uniform int y=0; y<4; y++) + for (uniform int x=0; x<4; x++) + { + c0[p] += block[p*16+y*4+x]; + } + + for (uniform int p=0; p<3; p++) + c0[p] >>= 4; +} + +inline void pick_endpoints(float c0[3], float c1[3], float block[48], float axis[3], float dc[3]) +{ + float min_dot = 256*256; + float max_dot = 0; + + for (uniform int y=0; y<4; y++) + for (uniform int x=0; x<4; x++) + { + float dot = 0; + for (uniform int p=0; p<3; p++) + dot += (block[p*16+y*4+x]-dc[p])*axis[p]; + + min_dot = min(min_dot, dot); + max_dot = max(max_dot, dot); + } + + if (max_dot-min_dot < 1.0f) + { + min_dot -= 0.5f; + max_dot += 0.5f; + } + + float norm_sq = 0; + for (uniform int p=0; p<3; p++) + norm_sq += axis[p]*axis[p]; + + float rnorm_sq = RCP(norm_sq); + for (uniform int p=0; p<3; p++) + { + c0[p] = clamp(dc[p]+min_dot*rnorm_sq*axis[p], 0, 255); + c1[p] = clamp(dc[p]+max_dot*rnorm_sq*axis[p], 0, 255); + } +} + +inline uint32 fast_quant(float block[48], int p0, int p1) +{ + float c0[3]; + float c1[3]; + dec_rgb565(c0, p0); + dec_rgb565(c1, p1); + + float dir[3]; + for (uniform int p=0; p<3; p++) dir[p] = c1[p]-c0[p]; + + float sq_norm = 0; + for (uniform int p=0; p<3; p++) sq_norm += sq(dir[p]); + + float rsq_norm = RCP(sq_norm); + + for (uniform int p=0; p<3; p++) dir[p] *= rsq_norm*3; + + float bias = 0.5; + for (uniform int p=0; p<3; p++) bias -= c0[p]*dir[p]; + + uint32 bits = 0; + uint32 scaler = 1; + for (uniform int k=0; k<16; k++) + { + float dot = 0; + for (uniform int p=0; p<3; p++) + dot += block[k+p*16]*dir[p]; + + int q = clamp((int)(dot+bias), 0, 3); + + //bits += q<<(k*2); + bits += q*scaler; + scaler *= 4; + } + + return bits; +} + +inline void compute_covar_dc(float covar[6], float dc[3], float block[48]) +{ + for (uniform int i=0; i<6; i++) covar[i] = 0; + for (uniform int p=0; p<3; p++) dc[p] = 0; + + for (uniform int k=0; k<16; k++) + { + for (uniform int p=0; p<3; p++) + dc[p] += block[k+p*16]; + } + + for (uniform int p=0; p<3; p++) dc[p] /= 16; + + for (uniform int k=0; k<16; k++) + { + float rgb[3]; + for (uniform int p=0; p<3; p++) + rgb[p] = block[k+p*16]-dc[p]; + + covar[0] += rgb[0]*rgb[0]; + covar[1] += rgb[0]*rgb[1]; + covar[2] += rgb[0]*rgb[2]; + + covar[3] += rgb[1]*rgb[1]; + covar[4] += rgb[1]*rgb[2]; + + covar[5] += rgb[2]*rgb[2]; + } +} + +// ugly, but makes BC1 compression 20% faster overall +inline void compute_covar_dc_ugly(float covar[6], float dc[3], float block[48]) +{ + for (uniform int p=0; p<3; p++) + { + float acc = 0; + for (uniform int k=0; k<16; k++) + acc += block[k+p*16]; + dc[p] = acc/16; + } + + float covar0 = 0.0f; + float covar1 = 0.0f; + float covar2 = 0.0f; + float covar3 = 0.0f; + float covar4 = 0.0f; + float covar5 = 0.0f; + + for (uniform int k=0; k<16; k++) + { + float rgb0, rgb1, rgb2; + rgb0 = block[k+0*16]-dc[0]; + rgb1 = block[k+1*16]-dc[1]; + rgb2 = block[k+2*16]-dc[2]; + + covar0 += rgb0*rgb0; + covar1 += rgb0*rgb1; + covar2 += rgb0*rgb2; + + covar3 += rgb1*rgb1; + covar4 += rgb1*rgb2; + + covar5 += rgb2*rgb2; + } + + covar[0] = covar0; + covar[1] = covar1; + covar[2] = covar2; + covar[3] = covar3; + covar[4] = covar4; + covar[5] = covar5; +} + +inline void bc1_refine(int pe[2], float block[48], unsigned int32 bits, float dc[3]) +{ + float c0[3]; + float c1[3]; + + if ((bits ^ (bits*4)) < 4) + { + // single color + for (uniform int p=0; p<3; p++) + { + c0[p] = dc[p]; + c1[p] = dc[p]; + } + } + else + { + float Atb1[3] = {0,0,0}; + float sum_q = 0; + float sum_qq = 0; + unsigned int32 shifted_bits = bits; + + for (uniform int k=0; k<16; k++) + { + float q = (int)(shifted_bits&3); + shifted_bits >>= 2; + + float x = 3-q; + float y = q; + + sum_q += q; + sum_qq += q*q; + + for (uniform int p=0; p<3; p++) Atb1[p] += x*block[k+p*16]; + } + + float sum[3]; + float Atb2[3]; + + for (uniform int p=0; p<3; p++) + { + sum[p] = dc[p]*16; + Atb2[p] = 3*sum[p]-Atb1[p]; + } + + float Cxx = 16*sq(3)-2*3*sum_q+sum_qq; + float Cyy = sum_qq; + float Cxy = 3*sum_q-sum_qq; + float scale = 3.0f * RCP(Cxx*Cyy - Cxy*Cxy); + + for (uniform int p=0; p<3; p++) + { + c0[p] = (Atb1[p]*Cyy - Atb2[p]*Cxy)*scale; + c1[p] = (Atb2[p]*Cxx - Atb1[p]*Cxy)*scale; + + c0[p] = clamp(c0[p], 0, 255); + c1[p] = clamp(c1[p], 0, 255); + } + } + + pe[0] = enc_rgb565(c0); + pe[1] = enc_rgb565(c1); +} + +inline uint32 fix_qbits(uint32 qbits) +{ + uniform const uint32 mask_01b = 0x55555555; + uniform const uint32 mask_10b = 0xAAAAAAAA; + + uint32 qbits0 = qbits&mask_01b; + uint32 qbits1 = qbits&mask_10b; + qbits = (qbits1>>1) + (qbits1 ^ (qbits0<<1)); + + return qbits; +} + +inline void CompressBlockBC1_core(float block[48], uint32 data[2]) +{ + uniform const int powerIterations = 4; + uniform const int refineIterations = 1; + + float covar[6]; + float dc[3]; + compute_covar_dc_ugly(covar, dc, block); + + float eps = 0.001; + covar[0] += eps; + covar[3] += eps; + covar[5] += eps; + + float axis[3]; + compute_axis3(axis, covar, powerIterations); + + float c0[3]; + float c1[3]; + pick_endpoints(c0, c1, block, axis, dc); + + int p[2]; + p[0] = enc_rgb565(c0); + p[1] = enc_rgb565(c1); + if (p[0] 0) q++; + if (q==8) q = 1; + + qblock[k/8] |= q << ((k%8)*3); + } + + // (could be improved by refinement) + + data[0] = clamp((int)ep[0], 0, 255)*256+clamp((int)ep[1], 0, 255); + data[0] |= qblock[0]<<16; + data[1] = qblock[0]>>16; + data[1] |= qblock[1]<<8; +} + +inline void CompressBlockBC1(uniform rgba_surface src[], int xx, uniform int yy, uniform uint8 dst[]) +{ + float block[48]; + uint32 data[2]; + + load_block_interleaved(block, src, xx, yy); + + CompressBlockBC1_core(block, data); + + store_data(dst, src->width, xx, yy, data, 2); +} + +inline void CompressBlockBC3(uniform rgba_surface src[], int xx, uniform int yy, uniform uint8 dst[]) +{ + float block[64]; + uint32 data[4]; + + load_block_interleaved_rgba(block, src, xx, yy); + + CompressBlockBC3_alpha(&block[48], &data[0]); + CompressBlockBC1_core(block, &data[2]); + + store_data(dst, src->width, xx, yy, data, 4); +} + +inline void CompressBlockBC4(uniform rgba_surface src[], int xx, uniform int yy, uniform uint8 dst[]) +{ + float block[16]; + uint32 data[2]; + + load_block_r_8bit(block, src, xx, yy); + + CompressBlockBC3_alpha(block, data); + + store_data(dst, src->width, xx, yy, data, 2); +} + +inline void CompressBlockBC5(uniform rgba_surface src[], int xx, uniform int yy, uniform uint8 dst[]) +{ + float block[32]; + uint32 data[4]; + + load_block_interleaved_rg_8bit(block, src, xx, yy); + + CompressBlockBC3_alpha(block, data); + CompressBlockBC3_alpha(&block[16], &data[2]); + + store_data(dst, src->width, xx, yy, data, 4); +} + +export void CompressBlocksBC1_ispc(uniform rgba_surface src[], uniform uint8 dst[]) +{ + for (uniform int yy = 0; yyheight/4; yy++) + foreach (xx = 0 ... src->width/4) + { + CompressBlockBC1(src, xx, yy, dst); + } +} + +export void CompressBlocksBC3_ispc(uniform rgba_surface src[], uniform uint8 dst[]) +{ + for (uniform int yy = 0; yyheight/4; yy++) + foreach (xx = 0 ... src->width/4) + { + CompressBlockBC3(src, xx, yy, dst); + } +} + +export void CompressBlocksBC4_ispc(uniform rgba_surface src[], uniform uint8 dst[]) +{ + for (uniform int yy = 0; yyheight/4; yy++) + foreach (xx = 0 ... src->width/4) + { + CompressBlockBC4(src, xx, yy, dst); + } +} + +export void CompressBlocksBC5_ispc(uniform rgba_surface src[], uniform uint8 dst[]) +{ + for (uniform int yy = 0; yyheight/4; yy++) + foreach (xx = 0 ... src->width/4) + { + CompressBlockBC5(src, xx, yy, dst); + } +} + +/////////////////////////////////////////////////////////// +// BC7 encoding + +struct bc7_enc_settings +{ + bool mode_selection[4]; + int refineIterations[8]; + + bool skip_mode2; + int fastSkipTreshold_mode1; + int fastSkipTreshold_mode3; + int fastSkipTreshold_mode7; + + int mode45_channel0; + int refineIterations_channel; + + int channels; +}; + +struct bc7_enc_state +{ + float block[64]; + + float opaque_err; // error for coding alpha=255 + float best_err; + uint32 best_data[5]; // 4, +1 margin for skips + + // settings + uniform bool mode_selection[4]; + uniform int refineIterations[8]; + + uniform bool skip_mode2; + uniform int fastSkipTreshold_mode1; + uniform int fastSkipTreshold_mode3; + uniform int fastSkipTreshold_mode7; + + uniform int mode45_channel0; + uniform int refineIterations_channel; + + uniform int channels; +}; + +struct mode45_parameters +{ + int qep[8]; + uint32 qblock[2]; + int aqep[2]; + uint32 aqblock[2]; + int rotation; + int swap; +}; + +void bc7_code_mode01237(uint32 data[5], int qep[6], uint32 qblock[2], int part_id, uniform int mode); +void bc7_code_mode45(uint32 data[5], mode45_parameters params[], uniform int mode); +void bc7_code_mode6(uint32 data[5], int qep[8], uint32 qblock[2]); + +/////////////////////////// +// BC7 format data + +inline uniform const int* uniform get_unquant_table(uniform int bits) +{ + assert(bits>=2 && bits<=4); // invalid bit size + + static uniform const int unquant_table_2bits[] = { 0, 21, 43, 64 }; + static uniform const int unquant_table_3bits[] = { 0, 9, 18, 27, 37, 46, 55, 64 }; + static uniform const int unquant_table_4bits[] = { 0, 4, 9, 13, 17, 21, 26, 30, 34, 38, 43, 47, 51, 55, 60, 64 }; + + uniform const int* uniform unquant_tables[] = {unquant_table_2bits, unquant_table_3bits, unquant_table_4bits}; + + return unquant_tables[bits-2]; +} + +inline uint32 get_pattern(int part_id) +{ + static uniform const uint32 pattern_table[] = { + 0x50505050u, 0x40404040u, 0x54545454u, 0x54505040u, 0x50404000u, 0x55545450u, 0x55545040u, 0x54504000u, + 0x50400000u, 0x55555450u, 0x55544000u, 0x54400000u, 0x55555440u, 0x55550000u, 0x55555500u, 0x55000000u, + 0x55150100u, 0x00004054u, 0x15010000u, 0x00405054u, 0x00004050u, 0x15050100u, 0x05010000u, 0x40505054u, + 0x00404050u, 0x05010100u, 0x14141414u, 0x05141450u, 0x01155440u, 0x00555500u, 0x15014054u, 0x05414150u, + 0x44444444u, 0x55005500u, 0x11441144u, 0x05055050u, 0x05500550u, 0x11114444u, 0x41144114u, 0x44111144u, + 0x15055054u, 0x01055040u, 0x05041050u, 0x05455150u, 0x14414114u, 0x50050550u, 0x41411414u, 0x00141400u, + 0x00041504u, 0x00105410u, 0x10541000u, 0x04150400u, 0x50410514u, 0x41051450u, 0x05415014u, 0x14054150u, + 0x41050514u, 0x41505014u, 0x40011554u, 0x54150140u, 0x50505500u, 0x00555050u, 0x15151010u, 0x54540404u, + 0xAA685050u, 0x6A5A5040u, 0x5A5A4200u, 0x5450A0A8u, 0xA5A50000u, 0xA0A05050u, 0x5555A0A0u, 0x5A5A5050u, + 0xAA550000u, 0xAA555500u, 0xAAAA5500u, 0x90909090u, 0x94949494u, 0xA4A4A4A4u, 0xA9A59450u, 0x2A0A4250u, + 0xA5945040u, 0x0A425054u, 0xA5A5A500u, 0x55A0A0A0u, 0xA8A85454u, 0x6A6A4040u, 0xA4A45000u, 0x1A1A0500u, + 0x0050A4A4u, 0xAAA59090u, 0x14696914u, 0x69691400u, 0xA08585A0u, 0xAA821414u, 0x50A4A450u, 0x6A5A0200u, + 0xA9A58000u, 0x5090A0A8u, 0xA8A09050u, 0x24242424u, 0x00AA5500u, 0x24924924u, 0x24499224u, 0x50A50A50u, + 0x500AA550u, 0xAAAA4444u, 0x66660000u, 0xA5A0A5A0u, 0x50A050A0u, 0x69286928u, 0x44AAAA44u, 0x66666600u, + 0xAA444444u, 0x54A854A8u, 0x95809580u, 0x96969600u, 0xA85454A8u, 0x80959580u, 0xAA141414u, 0x96960000u, + 0xAAAA1414u, 0xA05050A0u, 0xA0A5A5A0u, 0x96000000u, 0x40804080u, 0xA9A8A9A8u, 0xAAAAAA44u, 0x2A4A5254u + }; + + return gather_uint(pattern_table, part_id); +} + +inline int get_pattern_mask(int part_id, int j) +{ + static uniform const uint32 pattern_mask_table[] = { + 0xCCCC3333u, 0x88887777u, 0xEEEE1111u, 0xECC81337u, 0xC880377Fu, 0xFEEC0113u, 0xFEC80137u, 0xEC80137Fu, + 0xC80037FFu, 0xFFEC0013u, 0xFE80017Fu, 0xE80017FFu, 0xFFE80017u, 0xFF0000FFu, 0xFFF0000Fu, 0xF0000FFFu, + 0xF71008EFu, 0x008EFF71u, 0x71008EFFu, 0x08CEF731u, 0x008CFF73u, 0x73108CEFu, 0x3100CEFFu, 0x8CCE7331u, + 0x088CF773u, 0x3110CEEFu, 0x66669999u, 0x366CC993u, 0x17E8E817u, 0x0FF0F00Fu, 0x718E8E71u, 0x399CC663u, + 0xAAAA5555u, 0xF0F00F0Fu, 0x5A5AA5A5u, 0x33CCCC33u, 0x3C3CC3C3u, 0x55AAAA55u, 0x96966969u, 0xA55A5AA5u, + 0x73CE8C31u, 0x13C8EC37u, 0x324CCDB3u, 0x3BDCC423u, 0x69969669u, 0xC33C3CC3u, 0x99666699u, 0x0660F99Fu, + 0x0272FD8Du, 0x04E4FB1Bu, 0x4E40B1BFu, 0x2720D8DFu, 0xC93636C9u, 0x936C6C93u, 0x39C6C639u, 0x639C9C63u, + 0x93366CC9u, 0x9CC66339u, 0x817E7E81u, 0xE71818E7u, 0xCCF0330Fu, 0x0FCCF033u, 0x774488BBu, 0xEE2211DDu, + 0x08CC0133u, 0x8CC80037u, 0xCC80006Fu, 0xEC001331u, 0x330000FFu, 0x00CC3333u, 0xFF000033u, 0xCCCC0033u, + 0x0F0000FFu, 0x0FF0000Fu, 0x00F0000Fu, 0x44443333u, 0x66661111u, 0x22221111u, 0x136C0013u, 0x008C8C63u, + 0x36C80137u, 0x08CEC631u, 0x3330000Fu, 0xF0000333u, 0x00EE1111u, 0x88880077u, 0x22C0113Fu, 0x443088CFu, + 0x0C22F311u, 0x03440033u, 0x69969009u, 0x9960009Fu, 0x03303443u, 0x00660699u, 0xC22C3113u, 0x8C0000EFu, + 0x1300007Fu, 0xC4003331u, 0x004C1333u, 0x22229999u, 0x00F0F00Fu, 0x24929249u, 0x29429429u, 0xC30C30C3u, + 0xC03C3C03u, 0x00AA0055u, 0xAA0000FFu, 0x30300303u, 0xC0C03333u, 0x90900909u, 0xA00A5005u, 0xAAA0000Fu, + 0x0AAA0555u, 0xE0E01111u, 0x70700707u, 0x6660000Fu, 0x0EE01111u, 0x07707007u, 0x06660999u, 0x660000FFu, + 0x00660099u, 0x0CC03333u, 0x03303003u, 0x60000FFFu, 0x80807777u, 0x10100101u, 0x000A0005u, 0x08CE8421u + }; + + uint32 mask_packed = gather_uint(pattern_mask_table, part_id); + int mask0 = mask_packed&0xFFFF; + int mask1 = mask_packed>>16; + + int mask = (j==2) ? (~mask0)&(~mask1) : ( (j==0) ? mask0 : mask1 ); + return mask; +} + +inline void get_skips(int skips[3], int part_id) +{ + static uniform const int skip_table[] = { + 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, + 0xf0u, 0x20u, 0x80u, 0x20u, 0x20u, 0x80u, 0x80u, 0xf0u, 0x20u, 0x80u, 0x20u, 0x20u, 0x80u, 0x80u, 0x20u, 0x20u, + 0xf0u, 0xf0u, 0x60u, 0x80u, 0x20u, 0x80u, 0xf0u, 0xf0u, 0x20u, 0x80u, 0x20u, 0x20u, 0x20u, 0xf0u, 0xf0u, 0x60u, + 0x60u, 0x20u, 0x60u, 0x80u, 0xf0u, 0xf0u, 0x20u, 0x20u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0xf0u, 0x20u, 0x20u, 0xf0u, + 0x3fu, 0x38u, 0xf8u, 0xf3u, 0x8fu, 0x3fu, 0xf3u, 0xf8u, 0x8fu, 0x8fu, 0x6fu, 0x6fu, 0x6fu, 0x5fu, 0x3fu, 0x38u, + 0x3fu, 0x38u, 0x8fu, 0xf3u, 0x3fu, 0x38u, 0x6fu, 0xa8u, 0x53u, 0x8fu, 0x86u, 0x6au, 0x8fu, 0x5fu, 0xfau, 0xf8u, + 0x8fu, 0xf3u, 0x3fu, 0x5au, 0x6au, 0xa8u, 0x89u, 0xfau, 0xf6u, 0x3fu, 0xf8u, 0x5fu, 0xf3u, 0xf6u, 0xf6u, 0xf8u, + 0x3fu, 0xf3u, 0x5fu, 0x5fu, 0x5fu, 0x8fu, 0x5fu, 0xafu, 0x5fu, 0xafu, 0x8fu, 0xdfu, 0xf3u, 0xcfu, 0x3fu, 0x38u + }; + + int skip_packed = gather_int(skip_table, part_id); + skips[0] = 0; + skips[1] = skip_packed>>4; + skips[2] = skip_packed&15; +} + +/////////////////////////// +// PCA helpers + +inline void compute_stats_masked(float stats[15], float block[64], int mask, uniform int channels) +{ + for (uniform int i=0; i<15; i++) stats[i] = 0; + + int mask_shifted = mask<<1; + for (uniform int k=0; k<16; k++) + { + mask_shifted >>= 1; + //if ((mask_shifted&1) == 0) continue; + int flag = (mask_shifted&1); + + float rgba[4]; + for (uniform int p=0; p>= 1; + if ((mask_shifted&1) == 0) continue; + + float dot = 0; + for (uniform int p=0; p= 4); + int vv = v<<(8-bits); + return vv + shift_right(vv, bits); +} + +void ep_quant0367(int qep[], float ep[], uniform int mode, uniform int channels) +{ + uniform int bits = 7; + if (mode == 0) bits = 4; + if (mode == 7) bits = 5; + + uniform int levels = 1 << bits; + uniform int levels2 = levels*2-1; + + for (uniform int i=0; i<2; i++) + { + int qep_b[8]; + + for (uniform int b=0; b<2; b++) + for (uniform int p=0; p<4; p++) + { + int v = (int)((ep[i*4+p]/255.0f*levels2-b)/2+0.5)*2+b; + qep_b[b*4+p] = clamp(v, b, levels2-1+b); + } + + float ep_b[8]; + for (uniform int j=0; j<8; j++) + ep_b[j] = qep_b[j]; + + if (mode==0) + for (uniform int j=0; j<8; j++) + ep_b[j] = unpack_to_byte(qep_b[j], 5); + + float err0 = 0.0f; + float err1 = 0.0f; + for (uniform int p=0; p>= 2; + + float proj = 0; + float div = 0; + for (uniform int p=0; p=0 && best_q<=levels-1); + + qblock[k/8] += ((uint32)best_q) << 4*(k%8); + total_err += best_err; + } + + return total_err; +} + +/////////////////////////// +// LS endpoint refinement + +void opt_endpoints(float ep[], float block[64], uniform int bits, uint32 qblock[2], int mask, uniform int channels) +{ + uniform int levels = 1 << bits; + + float Atb1[4] = {0,0,0,0}; + float sum_q = 0; + float sum_qq = 0; + float sum[5] = {0,0,0,0,0}; + + int mask_shifted = mask<<1; + for (uniform int k1=0; k1<2; k1++) + { + uint32 qbits_shifted = qblock[k1]; + for (uniform int k2=0; k2<8; k2++) + { + uniform int k = k1*8+k2; + float q = (int)(qbits_shifted&15); + qbits_shifted >>= 4; + + mask_shifted >>= 1; + if ((mask_shifted&1) == 0) continue; + + int x = (levels-1)-q; + int y = q; + + sum_q += q; + sum_qq += q*q; + + sum[4] += 1; + for (uniform int p=0; pblock, part_id, mode); + + if (errrefineIterations[mode]; + for (uniform int _=0; _block, bits, best_qblock, mask, channels); + } + + int qep[24]; + uint32 qblock[2]; + + ep_quant_dequant(qep, ep, mode, channels); + + uint32 pattern = get_pattern(best_part_id); + float err = block_quant(qblock, state->block, bits, ep, pattern, channels); + + if (erropaque_err; // take into account alpha channel + + if (best_errbest_err) + { + state->best_err = best_err; + bc7_code_mode01237(state->best_data, best_qep, best_qblock, best_part_id, mode); + } +} + +void partial_sort_list(int list[], uniform int length, uniform int partial_count) +{ + for (uniform int k=0; k list[i]) + { + best_value = list[i]; + best_idx = i; + } + } + + // swap + scatter_int(list, best_idx, list[k]); + list[k] = best_value; + } +} + +void bc7_enc_mode02(bc7_enc_state state[]) +{ + int part_list[64]; + for (uniform int part=0; part<64; part++) + part_list[part] = part; + + bc7_enc_mode01237(state, 0, part_list, 16); + if (!state->skip_mode2) bc7_enc_mode01237(state, 2, part_list, 64); // usually not worth the time +} + +void bc7_enc_mode13(bc7_enc_state state[]) +{ + if (state->fastSkipTreshold_mode1 == 0 && state->fastSkipTreshold_mode3 == 0) return; + + float full_stats[15]; + compute_stats_masked(full_stats, state->block, -1, 3); + + int part_list[64]; + for (uniform int part=0; part<64; part++) + { + int mask = get_pattern_mask(part+0, 0); + float bound12 = block_pca_bound_split(state->block, mask, full_stats, 3); + int bound = (int)(bound12); + part_list[part] = part+bound*64; + } + + partial_sort_list(part_list, 64, max(state->fastSkipTreshold_mode1, state->fastSkipTreshold_mode3)); + bc7_enc_mode01237(state, 1, part_list, state->fastSkipTreshold_mode1); + bc7_enc_mode01237(state, 3, part_list, state->fastSkipTreshold_mode3); +} + +void bc7_enc_mode7(bc7_enc_state state[]) +{ + if (state->fastSkipTreshold_mode7 == 0) return; + + float full_stats[15]; + compute_stats_masked(full_stats, state->block, -1, state->channels); + + int part_list[64]; + for (uniform int part=0; part<64; part++) + { + int mask = get_pattern_mask(part+0, 0); + float bound12 = block_pca_bound_split(state->block, mask, full_stats, state->channels); + int bound = (int)(bound12); + part_list[part] = part+bound*64; + } + + partial_sort_list(part_list, 64, state->fastSkipTreshold_mode7); + bc7_enc_mode01237(state, 7, part_list, state->fastSkipTreshold_mode7); +} + +void channel_quant_dequant(int qep[2], float ep[2], uniform int epbits) +{ + int elevels = (1<>= 4; + + int x = (levels-1)-q; + int y = q; + + sum_q += q; + sum_qq += q*q; + + sum += block[k]; + Atb1 += x*block[k]; + } + } + + float Atb2 = (levels-1)*sum-Atb1; + + float Cxx = 16*sq(levels-1)-2*(levels-1)*sum_q+sum_qq; + float Cyy = sum_qq; + float Cxy = (levels-1)*sum_q-sum_qq; + float scale = (levels-1) / (Cxx*Cyy - Cxy*Cxy); + + ep[0] = (Atb1*Cyy - Atb2*Cxy)*scale; + ep[1] = (Atb2*Cxx - Atb1*Cxy)*scale; + + ep[0] = clamp(ep[0], 0, 255); + ep[1] = clamp(ep[1], 0, 255); + + if (abs(Cxx*Cyy - Cxy*Cxy) < 0.001) + { + ep[0] = sum/16; + ep[1] = ep[0]; + } +} + +float channel_opt_quant(uint32 qblock[2], float block[16], uniform int bits, float ep[]) +{ + uniform const int* uniform unquant_table = get_unquant_table(bits); + int levels = (1<refineIterations_channel; + for (uniform int i=0; iblock[k+p*16]; + + if (rotation < 3) + { + // apply channel rotation + if (state->channels == 4) block[k+rotation*16] = state->block[k+3*16]; + if (state->channels == 3) block[k+rotation*16] = 255; + } + } + + float ep[8]; + block_segment(ep, block, -1, 3); + + int qep[8]; + ep_quant_dequant(qep, ep, mode, 3); + + uint32 qblock[2]; + float err = block_quant(qblock, block, bits, ep, 0, 3); + + // refine + uniform int refineIterations = state->refineIterations[mode]; + for (uniform int i=0; iblock[rotation*16], abits, aepbits); + + if (err<*best_err) + { + + swap_ints(best_candidate->qep, qep, 8); + swap_uints(best_candidate->qblock, qblock, 2); + swap_ints(best_candidate->aqep, aqep, 2); + swap_uints(best_candidate->aqblock, aqblock, 2); + best_candidate->rotation = rotation; + best_candidate->swap = swap; + *best_err = err; + } +} + +void bc7_enc_mode45(bc7_enc_state state[]) +{ + mode45_parameters best_candidate; + float best_err = state->best_err; + + memset(&best_candidate, 0, sizeof(mode45_parameters)); + + uniform int channel0 = state->mode45_channel0; + for (uniform int p=channel0; pchannels; p++) + { + bc7_enc_mode45_candidate(state, &best_candidate, &best_err, 4, p, 0); + bc7_enc_mode45_candidate(state, &best_candidate, &best_err, 4, p, 1); + } + + // mode 4 + if (best_errbest_err) + { + state->best_err = best_err; + bc7_code_mode45(state->best_data, &best_candidate, 4); + } + + for (uniform int p=channel0; pchannels; p++) + { + bc7_enc_mode45_candidate(state, &best_candidate, &best_err, 5, p, 0); + } + + // mode 5 + if (best_errbest_err) + { + state->best_err = best_err; + bc7_code_mode45(state->best_data, &best_candidate, 5); + } +} + +void bc7_enc_mode6(bc7_enc_state state[]) +{ + uniform int mode = 6; + uniform int bits = 4; + float ep[8]; + block_segment(ep, state->block, -1, state->channels); + + if (state->channels == 3) + { + ep[3] = ep[7] = 255; + } + + int qep[8]; + ep_quant_dequant(qep, ep, mode, state->channels); + + uint32 qblock[2]; + float err = block_quant(qblock, state->block, bits, ep, 0, state->channels); + + // refine + uniform int refineIterations = state->refineIterations[mode]; + for (uniform int i=0; iblock, bits, qblock, -1, state->channels); + ep_quant_dequant(qep, ep, mode, state->channels); + err = block_quant(qblock, state->block, bits, ep, 0, state->channels); + } + + if (errbest_err) + { + state->best_err = err; + bc7_code_mode6(state->best_data, qep, qblock); + } +} + +////////////////////////// +// BC7 bitstream coding + +void bc7_code_apply_swap_mode456(int qep[], uniform int channels, uint32 qblock[2], uniform int bits) +{ + uniform int levels = 1 << bits; + if ((qblock[0]&15)>=levels/2) + { + swap_ints(&qep[0], &qep[channels], channels); + + for (uniform int k=0; k<2; k++) + qblock[k] = (uint32)(0x11111111*(levels-1)) - qblock[k]; + } + + assert((qblock[0]&15) < levels/2); +} + +int bc7_code_apply_swap_mode01237(int qep[], uint32 qblock[2], uniform int mode, int part_id) +{ + uniform int bits = 2; if (mode == 0 || mode == 1) bits = 3; + uniform int pairs = 2; if (mode == 0 || mode == 2) pairs = 3; + + int flips = 0; + uniform int levels = 1 << bits; + int skips[3]; + get_skips(skips, part_id); + + for (uniform int j=0; j>((k0%8)*4))&15; + int q = ((gather_uint(qblock, k0>>3)<<(28-(k0&7)*4))>>28); + + if (q>=levels/2) + { + swap_ints(&qep[8*j], &qep[8*j+4], 4); + uint32 pmask = get_pattern_mask(part_id, j); + flips |= pmask; + } + } + + return flips; +} + +void put_bits(uint32 data[5], uniform int* uniform pos, uniform int bits, int v) +{ + assert(v32) + { + data[*pos/32+1] |= shift_right(v, 32-*pos%32); + } + *pos += bits; +} + +inline void data_shl_1bit_from(uint32 data[5], int from) +{ + if (from < 96) + { + assert(from > 64+10); + + uint32 shifted = (data[2]>>1) | (data[3]<<31); + uint32 mask = (pow2(from-64)-1)>>1; + data[2] = (mask&data[2]) | (~mask&shifted); + data[3] = (data[3]>>1) | (data[4]<<31); + data[4] = data[4]>>1; + } + else if (from < 128) + { + uint32 shifted = (data[3]>>1) | (data[4]<<31); + uint32 mask = (pow2(from-96)-1)>>1; + data[3] = (mask&data[3]) | (~mask&shifted); + data[4] = data[4]>>1; + } +} + +void bc7_code_qblock(uint32 data[5], uniform int* uniform pPos, uint32 qblock[2], uniform int bits, int flips) +{ + uniform int levels = 1 << bits; + int flips_shifted = flips; + for (uniform int k1=0; k1<2; k1++) + { + uint32 qbits_shifted = qblock[k1]; + for (uniform int k2=0; k2<8; k2++) + { + int q = qbits_shifted&15; + if ((flips_shifted&1)>0) q = (levels-1)-q; + + if (k1==0 && k2==0) put_bits(data, pPos, bits-1, q); + else put_bits(data, pPos, bits , q); + qbits_shifted >>= 4; + flips_shifted >>= 1; + } + } +} + +void bc7_code_adjust_skip_mode01237(uint32 data[5], uniform int mode, int part_id) +{ + uniform int bits = 2; if (mode == 0 || mode == 1) bits = 3; + uniform int pairs = 2; if (mode == 0 || mode == 2) pairs = 3; + + int skips[3]; + get_skips(skips, part_id); + + if (pairs>2 && skips[1] < skips[2]) + { + int t = skips[1]; skips[1] = skips[2]; skips[2] = t; + } + + for (uniform int j=1; j>1); + } + else if (mode == 1) + { + put_bits(data, &pos, 6, qep[j*4+0+p]>>1); + } + else if (mode == 2) + { + put_bits(data, &pos, 5, qep[j*4+0+p]); + } + else if (mode == 3) + { + put_bits(data, &pos, 7, qep[j*4+0+p]>>1); + } + else if (mode == 7) + { + put_bits(data, &pos, 5, qep[j*4+0+p]>>1); + } + else + { + assert(false); + } + } + + // p bits + if (mode == 1) + for (uniform int j=0; j<2; j++) + { + put_bits(data, &pos, 1, qep[j*8]&1); + } + + if (mode == 0 || mode == 3 || mode == 7) + for (uniform int j=0; jqep, qep, 8); + swap_uints(params->qblock, qblock, 2); + swap_ints(params->aqep, aqep, 2); + swap_uints(params->aqblock, aqblock, 2); + int rotation = params->rotation; + int swap = params->swap; + + uniform int bits = 2; + uniform int abits = 2; if (mode==4) abits = 3; + uniform int epbits = 7; if (mode==4) epbits = 5; + uniform int aepbits = 8; if (mode==4) aepbits = 6; + + if (!swap) + { + bc7_code_apply_swap_mode456(qep, 4, qblock, bits); + bc7_code_apply_swap_mode456(aqep, 1, aqblock, abits); + } + else + { + swap_uints(qblock, aqblock, 2); + bc7_code_apply_swap_mode456(aqep, 1, qblock, bits); + bc7_code_apply_swap_mode456(qep, 4, aqblock, abits); + } + + for (uniform int k=0; k<5; k++) data[k] = 0; + uniform int pos = 0; + + // mode 4-5 + put_bits(data, &pos, mode+1, 1<>1); + put_bits(data, &pos, 7, qep[4+p]>>1); + } + + // p bits + put_bits(data, &pos, 1, qep[0]&1); + put_bits(data, &pos, 1, qep[4]&1); + + // quantized values + bc7_code_qblock(data, &pos, qblock, 4, 0); +} + + +////////////////////////// +// BC7 core + +inline void CompressBlockBC7_core(bc7_enc_state state[]) +{ + if (state->mode_selection[0]) bc7_enc_mode02(state); + if (state->mode_selection[1]) bc7_enc_mode13(state); + if (state->mode_selection[1]) bc7_enc_mode7(state); + if (state->mode_selection[2]) bc7_enc_mode45(state); + if (state->mode_selection[3]) bc7_enc_mode6(state); +} + +void bc7_enc_copy_settings(bc7_enc_state state[], uniform bc7_enc_settings settings[]) +{ + state->channels = settings->channels; + + // mode02 + state->mode_selection[0] = settings->mode_selection[0]; + state->skip_mode2 = settings->skip_mode2; + + state->refineIterations[0] = settings->refineIterations[0]; + state->refineIterations[2] = settings->refineIterations[2]; + + // mode137 + state->mode_selection[1] = settings->mode_selection[1]; + state->fastSkipTreshold_mode1 = settings->fastSkipTreshold_mode1; + state->fastSkipTreshold_mode3 = settings->fastSkipTreshold_mode3; + state->fastSkipTreshold_mode7 = settings->fastSkipTreshold_mode7; + + state->refineIterations[1] = settings->refineIterations[1]; + state->refineIterations[3] = settings->refineIterations[3]; + state->refineIterations[7] = settings->refineIterations[7]; + + // mode45 + state->mode_selection[2] = settings->mode_selection[2]; + + state->mode45_channel0 = settings->mode45_channel0; + state->refineIterations_channel = settings->refineIterations_channel; + state->refineIterations[4] = settings->refineIterations[4]; + state->refineIterations[5] = settings->refineIterations[5]; + + // mode6 + state->mode_selection[3] = settings->mode_selection[3]; + + state->refineIterations[6] = settings->refineIterations[6]; +} + +inline void CompressBlockBC7(uniform rgba_surface src[], int xx, uniform int yy, uniform uint8 dst[], + uniform bc7_enc_settings settings[]) +{ + bc7_enc_state _state; + varying bc7_enc_state* uniform state = &_state; + + bc7_enc_copy_settings(state, settings); + load_block_interleaved_rgba(state->block, src, xx, yy); + state->best_err = 1e99; + state->opaque_err = compute_opaque_err(state->block, state->channels); + + CompressBlockBC7_core(state); + + store_data(dst, src->width, xx, yy, state->best_data, 4); +} + +export void CompressBlocksBC7_ispc(uniform rgba_surface src[], uniform uint8 dst[], uniform bc7_enc_settings settings[]) +{ + for (uniform int yy = 0; yyheight/4; yy++) + foreach (xx = 0 ... src->width/4) + { + CompressBlockBC7(src, xx, yy, dst, settings); + } +} + +/////////////////////////////////////////////////////////// +// BC6H encoding + +struct bc6h_enc_settings +{ + bool slow_mode; + bool fast_mode; + int refineIterations_1p; + int refineIterations_2p; + int fastSkipTreshold; +}; + +struct bc6h_enc_state +{ + float block[64]; + + float best_err; + uint32 best_data[5]; // 4, +1 margin for skips + + float rgb_bounds[6]; + float max_span; + int max_span_idx; + + int mode; + int epb; + int qbounds[8]; + + // settings + uniform bool slow_mode; + uniform bool fast_mode; + uniform int refineIterations_1p; + uniform int refineIterations_2p; + uniform int fastSkipTreshold; +}; + +void bc6h_code_2p(uint32 data[5], int pqep[], uint32 qblock[2], int part_id, int mode); +void bc6h_code_1p(uint32 data[5], int qep[8], uint32 qblock[2], int mode); + +/////////////////////////// +// BC6H format data + +inline uniform int get_mode_prefix(uniform int mode) +{ + static uniform const int mode_prefix_table[] = + { + 0, 1, 2, 6, 10, 14, 18, 22, 26, 30, 3, 7, 11, 15 + }; + + return mode_prefix_table[mode]; +} + +inline uniform float get_span(uniform int mode) +{ + static uniform const float span_table[] = + { + 0.9 * 0xFFFF / 64, // (0) 4 / 10 + 0.9 * 0xFFFF / 4, // (1) 5 / 7 + 0.8 * 0xFFFF / 256, // (2) 3 / 11 + -1, -1, + 0.9 * 0xFFFF / 32, // (5) 4 / 9 + 0.9 * 0xFFFF / 16, // (6) 4 / 8 + -1, -1, + 0xFFFF, // (9) absolute + 0xFFFF, // (10) absolute + 0.95 * 0xFFFF / 8, // (11) 8 / 11 + 0.95 * 0xFFFF / 32, // (12) 7 / 12 + 6, // (13) 3 / 16 + }; + + uniform int span = span_table[mode]; + assert(span > 0); + return span; +} + +inline uniform int get_mode_bits(uniform int mode) +{ + static uniform const int mode_bits_table[] = + { + 10, 7, 11, -1, -1, + 9, 8, -1, -1, 6, + 10, 11, 12, 16, + }; + + uniform int mode_bits = mode_bits_table[mode]; + assert(mode_bits > 0); + return mode_bits; +} + +/////////////////////////// +// endpoint quantization + +inline int unpack_to_uf16(uint32 v, int bits) +{ + if (bits >= 15) return v; + if (v == 0) return 0; + if (v == (1<epb; + ep_quant_bc6h(qep, ep, bits, pairs); + + for (uniform int i = 0; i < 2 * pairs; i++) + for (uniform int p = 0; p < 3; p++) + { + qep[i * 4 + p] = clamp(qep[i * 4 + p], state->qbounds[p], state->qbounds[4 + p]); + } + + ep_dequant_bc6h(ep, qep, bits, pairs); + +} + +////////////////////////// +// parameter estimation + +float bc6h_enc_2p_part_fast(bc6h_enc_state state[], int qep[16], uint32 qblock[2], int part_id) +{ + uint32 pattern = get_pattern(part_id); + uniform int bits = 3; + uniform int pairs = 2; + uniform int channels = 3; + + float ep[16]; + for (uniform int j = 0; jblock, mask, channels); + } + + ep_quant_dequant_bc6h(state, qep, ep, 2); + + float total_err = block_quant(qblock, state->block, bits, ep, pattern, channels); + return total_err; + +} + +void bc6h_enc_2p_list(bc6h_enc_state state[], int part_list[], uniform int part_count) +{ + if (part_count == 0) return; + uniform int bits = 3; + uniform int pairs = 2; + uniform int channels = 3; + + int best_qep[24]; + uint32 best_qblock[2]; + int best_part_id = -1; + float best_err = 1e99; + + for (uniform int part = 0; partrefineIterations_2p; + for (uniform int _ = 0; _block, bits, best_qblock, mask, channels); + } + + int qep[24]; + uint32 qblock[2]; + ep_quant_dequant_bc6h(state, qep, ep, 2); + + uint32 pattern = get_pattern(best_part_id); + float err = block_quant(qblock, state->block, bits, ep, pattern, channels); + + if (errbest_err) + { + state->best_err = best_err; + bc6h_code_2p(state->best_data, best_qep, best_qblock, best_part_id, state->mode); + } +} + +void bc6h_enc_2p(bc6h_enc_state state[]) +{ + float full_stats[15]; + compute_stats_masked(full_stats, state->block, -1, 3); + + int part_list[32]; + for (uniform int part = 0; part < 32; part++) + { + int mask = get_pattern_mask(part, 0); + float bound12 = block_pca_bound_split(state->block, mask, full_stats, 3); + int bound = (int)(bound12); + part_list[part] = part + bound * 64; + } + + partial_sort_list(part_list, 32, state->fastSkipTreshold); + bc6h_enc_2p_list(state, part_list, state->fastSkipTreshold); +} + +void bc6h_enc_1p(bc6h_enc_state state[]) +{ + float ep[8]; + block_segment_core(ep, state->block, -1, 3); + + int qep[8]; + ep_quant_dequant_bc6h(state, qep, ep, 1); + + uint32 qblock[2]; + float err = block_quant(qblock, state->block, 4, ep, 0, 3); + + // refine + uniform int refineIterations = state->refineIterations_1p; + for (uniform int i = 0; iblock, 4, qblock, -1, 3); + ep_quant_dequant_bc6h(state, qep, ep, 1); + err = block_quant(qblock, state->block, 4, ep, 0, 3); + } + + if (err < state->best_err) + { + state->best_err = err; + bc6h_code_1p(state->best_data, qep, qblock, state->mode); + } +} + +inline void compute_qbounds(bc6h_enc_state state[], float rgb_span[3]) +{ + float bounds[8]; + for (uniform int p = 0; p < 3; p++) + { + float middle = (state->rgb_bounds[p] + state->rgb_bounds[3 + p]) / 2; + + bounds[ p] = middle - rgb_span[p] / 2; + bounds[4+p] = middle + rgb_span[p] / 2; + } + + ep_quant_bc6h(state->qbounds, bounds, state->epb, 1); +} + +void compute_qbounds(bc6h_enc_state state[], float span) +{ + float rgb_span[3] = { span, span, span }; + compute_qbounds(state, rgb_span); +} + +void compute_qbounds2(bc6h_enc_state state[], float span, int max_span_idx) +{ + float rgb_span[3] = { span, span, span }; + for (uniform int p = 0; p < 3; p++) + { + rgb_span[p] *= (p == max_span_idx) ? 2 : 1; + } + compute_qbounds(state, rgb_span); +} + +void bc6h_test_mode(bc6h_enc_state state[], uniform int mode, uniform bool enc, uniform float margin) +{ + uniform int mode_bits = get_mode_bits(mode); + uniform float span = get_span(mode); + float max_span = state->max_span; + int max_span_idx = state->max_span_idx; + + if (max_span * margin > span) return; + + if (mode >= 10) + { + state->epb = mode_bits; + state->mode = mode; + + compute_qbounds(state, span); + if (enc) bc6h_enc_1p(state); + } + else if (mode <= 1 || mode == 5 || mode == 9) + { + state->epb = mode_bits; + state->mode = mode; + + compute_qbounds(state, span); + if (enc) bc6h_enc_2p(state); + } + else + { + state->epb = mode_bits; + state->mode = mode + max_span_idx; + + compute_qbounds2(state, span, max_span_idx); + if (enc) bc6h_enc_2p(state); + } +} + +////////////////////////// +// BC6H bitstream coding + +int bit_at(int v, uniform int pos) +{ + return (v >> pos) & 1; +} + +uint32 reverse_bits(uint32 v, uniform int bits) +{ + if (bits == 2) + { + return (v >> 1) + (v & 1) * 2; + } + if (bits == 6) + { + v = (v & 0x5555) * 2 + ((v >> 1) & 0x5555); + return (v >> 4) + ((v >> 2) & 3) * 4 + (v & 3) * 16; + } + else + { + assert(false); + } +} + +void bc6h_pack(uint32 packed[], int qep[], int mode) +{ + if (mode == 0) + { + int pred_qep[16]; + for (uniform int p = 0; p < 3; p++) + { + pred_qep[ p] = qep[p]; + pred_qep[ 4 + p] = (qep[ 4 + p] - qep[p]) & 31; + pred_qep[ 8 + p] = (qep[ 8 + p] - qep[p]) & 31; + pred_qep[12 + p] = (qep[12 + p] - qep[p]) & 31; + } + + for (uniform int i = 1; i < 4; i++) + for (uniform int p = 0; p < 3; p++) + { + assert( qep[i * 4 + p] - qep[p] <= 15); + assert(-16 <= qep[i * 4 + p] - qep[p]); + } + + /* + g2[4], b2[4], b3[4], + r0[9:0], + g0[9:0], + b0[9:0], + r1[4:0], g3[4], g2[3:0], + g1[4:0], b3[0], g3[3:0], + b1[4:0], b3[1], b2[3:0], + r2[4:0], b3[2], + r3[4:0], b3[3] + */ + + uint32 pqep[10]; + + pqep[4] = pred_qep[4] + (pred_qep[ 8 + 1] & 15) * 64; + pqep[5] = pred_qep[5] + (pred_qep[12 + 1] & 15) * 64; + pqep[6] = pred_qep[6] + (pred_qep[ 8 + 2] & 15) * 64; + + pqep[4] += bit_at(pred_qep[12 + 1], 4) << 5; + pqep[5] += bit_at(pred_qep[12 + 2], 0) << 5; + pqep[6] += bit_at(pred_qep[12 + 2], 1) << 5; + + pqep[8] = pred_qep[ 8] + bit_at(pred_qep[12 + 2], 2) * 32; + pqep[9] = pred_qep[12] + bit_at(pred_qep[12 + 2], 3) * 32; + + packed[0] = get_mode_prefix(0); + packed[0] += bit_at(pred_qep[ 8 + 1], 4) << 2; + packed[0] += bit_at(pred_qep[ 8 + 2], 4) << 3; + packed[0] += bit_at(pred_qep[12 + 2], 4) << 4; + + packed[1] = (pred_qep[2] << 20) + (pred_qep[1] << 10) + pred_qep[0]; + packed[2] = (pqep[6] << 20) + (pqep[5] << 10) + pqep[4]; + packed[3] = (pqep[9] << 6) + pqep[8]; + } + else if (mode == 1) + { + int pred_qep[16]; + for (uniform int p = 0; p < 3; p++) + { + pred_qep[ p] = qep[p]; + pred_qep[ 4 + p] = (qep[ 4 + p] - qep[p]) & 63; + pred_qep[ 8 + p] = (qep[ 8 + p] - qep[p]) & 63; + pred_qep[12 + p] = (qep[12 + p] - qep[p]) & 63; + } + + for (uniform int i = 1; i < 4; i++) + for (uniform int p = 0; p < 3; p++) + { + assert( qep[i * 4 + p] - qep[p] <= 31); + assert(-32 <= qep[i * 4 + p] - qep[p]); + } + + /* + g2[5], g3[4], g3[5], + r0[6:0], b3[0], b3[1], b2[4], + g0[6:0], b2[5], b3[2], g2[4], + b0[6:0], b3[3], b3[5], b3[4], + r1[5:0], g2[3:0], + g1[5:0], g3[3:0], + b1[5:0], b2[3:0], + r2[5:0], + r3[5:0] + */ + + uint32 pqep[8]; + + pqep[0] = pred_qep[0]; + pqep[0] += bit_at(pred_qep[12 + 2], 0) << 7; + pqep[0] += bit_at(pred_qep[12 + 2], 1) << 8; + pqep[0] += bit_at(pred_qep[ 8 + 2], 4) << 9; + + pqep[1] = pred_qep[1]; + pqep[1] += bit_at(pred_qep[ 8 + 2], 5) << 7; + pqep[1] += bit_at(pred_qep[12 + 2], 2) << 8; + pqep[1] += bit_at(pred_qep[ 8 + 1], 4) << 9; + + pqep[2] = pred_qep[2]; + pqep[2] += bit_at(pred_qep[12 + 2], 3) << 7; + pqep[2] += bit_at(pred_qep[12 + 2], 5) << 8; + pqep[2] += bit_at(pred_qep[12 + 2], 4) << 9; + + pqep[4] = pred_qep[4] + (pred_qep[ 8 + 1] & 15) * 64; + pqep[5] = pred_qep[5] + (pred_qep[12 + 1] & 15) * 64; + pqep[6] = pred_qep[6] + (pred_qep[ 8 + 2] & 15) * 64; + + packed[0] = get_mode_prefix(1); + packed[0] += bit_at(pred_qep[ 8 + 1], 5) << 2; + packed[0] += bit_at(pred_qep[12 + 1], 4) << 3; + packed[0] += bit_at(pred_qep[12 + 1], 5) << 4; + + packed[1] = (pqep[2] << 20) + (pqep[1] << 10) + pqep[0]; + packed[2] = (pqep[6] << 20) + (pqep[5] << 10) + pqep[4]; + packed[3] = (pred_qep[12] << 6) + pred_qep[8]; + } + else if (mode == 2 || mode == 3 || mode == 4) + { + /* + r0[9:0], g0[9:0], b0[9:0], + r1[3:0], xx[y], xx[y], g2[3:0], + g1[3:0], xx[y], xx[y], g3[3:0], + b1[3:0], xx[y], xx[y], b2[3:0], + r2[3:0], xx[y], xx[y], + r3[3:0], xx[y], xx[y] + */ + + int dqep[16]; + for (uniform int p = 0; p < 3; p++) + { + int mask = 15; + if (p == mode - 2) mask = 31; + dqep[p] = qep[p]; + dqep[ 4 + p] = (qep[ 4 + p] - qep[p]) & mask; + dqep[ 8 + p] = (qep[ 8 + p] - qep[p]) & mask; + dqep[12 + p] = (qep[12 + p] - qep[p]) & mask; + } + + for (uniform int i = 1; i < 4; i++) + for (uniform int p = 0; p < 3; p++) + { + int bits = 4; + if (p == mode - 2) bits = 5; + //assert( qep[i * 4 + p] - qep[p] <= (1<> 10) * 512; + pqep[5] = dqep[5] + (dqep[1] >> 10) * 512; + pqep[6] = dqep[6] + (dqep[2] >> 10) * 512; + + packed[0] = get_mode_prefix(11); + packed[1] = (pqep[2] << 20) + (pqep[1] << 10) + pqep[0]; + packed[2] = (pqep[6] << 20) + (pqep[5] << 10) + pqep[4]; + } + else if (mode == 12) + { + int dqep[8]; + for (uniform int p = 0; p < 3; p++) + { + dqep[p] = qep[p]; + dqep[4 + p] = (qep[4 + p] - qep[p]) & 255; + } + + for (uniform int i = 1; i < 2; i++) + for (uniform int p = 0; p < 3; p++) + { + assert( qep[i * 4 + p] - qep[p] <= 127); + assert(-128 <= qep[i * 4 + p] - qep[p]); + } + + /* + r0[9:0], g0[9:0], b0[9:0], + r1[7:0], r0[10:11], + g1[7:0], g0[10:11], + b1[7:0], b0[10:11] + */ + + uint32 pqep[8]; + + pqep[0] = dqep[0] & 1023; + pqep[1] = dqep[1] & 1023; + pqep[2] = dqep[2] & 1023; + + pqep[4] = dqep[4] + reverse_bits(dqep[0] >> 10, 2) * 256; + pqep[5] = dqep[5] + reverse_bits(dqep[1] >> 10, 2) * 256; + pqep[6] = dqep[6] + reverse_bits(dqep[2] >> 10, 2) * 256; + + packed[0] = get_mode_prefix(12); + packed[1] = (pqep[2] << 20) + (pqep[1] << 10) + pqep[0]; + packed[2] = (pqep[6] << 20) + (pqep[5] << 10) + pqep[4]; + } + else if (mode == 13) + { + int dqep[8]; + for (uniform int p = 0; p < 3; p++) + { + dqep[p] = qep[p]; + dqep[4 + p] = (qep[4 + p] - qep[p]) & 15; + } + + for (uniform int i = 1; i < 2; i++) + for (uniform int p = 0; p < 3; p++) + { + assert( qep[i * 4 + p] - qep[p] <= 7); + assert(-8 <= qep[i * 4 + p] - qep[p]); + } + + /* + r0[9:0], g0[9:0], b0[9:0], + r1[3:0], r0[10:15], + g1[3:0], g0[10:15], + b1[3:0], b0[10:15] + */ + + uint32 pqep[8]; + + pqep[0] = dqep[0] & 1023; + pqep[1] = dqep[1] & 1023; + pqep[2] = dqep[2] & 1023; + + pqep[4] = dqep[4] + reverse_bits(dqep[0] >> 10, 6) * 16; + pqep[5] = dqep[5] + reverse_bits(dqep[1] >> 10, 6) * 16; + pqep[6] = dqep[6] + reverse_bits(dqep[2] >> 10, 6) * 16; + + packed[0] = get_mode_prefix(13); + packed[1] = (pqep[2] << 20) + (pqep[1] << 10) + pqep[0]; + packed[2] = (pqep[6] << 20) + (pqep[5] << 10) + pqep[4]; + } + else + { + assert(false); + } +} + +void bc6h_code_2p(uint32 data[5], int qep[], uint32 qblock[2], int part_id, int mode) +{ + uniform int bits = 3; + uniform int pairs = 2; + uniform int channels = 3; + + int flips = bc7_code_apply_swap_mode01237(qep, qblock, 1, part_id); + + for (uniform int k=0; k<5; k++) data[k] = 0; + uniform int pos = 0; + + uint32 packed[4]; + bc6h_pack(packed, qep, mode); + + // mode + put_bits(data, &pos, 5, packed[0]); + + // endpoints + put_bits(data, &pos, 30, packed[1]); + put_bits(data, &pos, 30, packed[2]); + put_bits(data, &pos, 12, packed[3]); + + // partition + put_bits(data, &pos, 5, part_id); + + // quantized values + bc7_code_qblock(data, &pos, qblock, bits, flips); + bc7_code_adjust_skip_mode01237(data, 1, part_id); +} + +void bc6h_code_1p(uint32 data[5], int qep[8], uint32 qblock[2], int mode) +{ + bc7_code_apply_swap_mode456(qep, 4, qblock, 4); + + for (uniform int k = 0; k<5; k++) data[k] = 0; + uniform int pos = 0; + + uint32 packed[4]; + bc6h_pack(packed, qep, mode); + + // mode + put_bits(data, &pos, 5, packed[0]); + + // endpoints + put_bits(data, &pos, 30, packed[1]); + put_bits(data, &pos, 30, packed[2]); + + // quantized values + bc7_code_qblock(data, &pos, qblock, 4, 0); +} + +////////////////////////// +// BC6H core + +void bc6h_setup(bc6h_enc_state state[]) +{ + for (uniform int p = 0; p < 3; p++) + { + state->rgb_bounds[p ] = 0xFFFF; + state->rgb_bounds[3+p] = 0; + } + + // uf16 conversion, min/max + for (uniform int p = 0; p < 3; p++) + for (uniform int k = 0; k < 16; k++) + { + state->block[p * 16 + k] = (state->block[p * 16 + k] / 31) * 64; + + state->rgb_bounds[p ] = min(state->rgb_bounds[p ], state->block[p * 16 + k]); + state->rgb_bounds[3+p] = max(state->rgb_bounds[3+p], state->block[p * 16 + k]); + } + + state->max_span = 0; + state->max_span_idx = 0; + + float rgb_span[0] = { 0, 0, 0 }; + for (uniform int p = 0; p < 3; p++) + { + rgb_span[p] = state->rgb_bounds[3+p] - state->rgb_bounds[p]; + if (rgb_span[p] > state->max_span) + { + state->max_span_idx = p; + state->max_span = rgb_span[p]; + } + } +} + +inline void CompressBlockBC6H_core(bc6h_enc_state state[]) +{ + bc6h_setup(state); + + if (state->slow_mode) + { + bc6h_test_mode(state, 0, true, 0); + bc6h_test_mode(state, 1, true, 0); + bc6h_test_mode(state, 2, true, 0); + bc6h_test_mode(state, 5, true, 0); + bc6h_test_mode(state, 6, true, 0); + bc6h_test_mode(state, 9, true, 0); + bc6h_test_mode(state, 10, true, 0); + bc6h_test_mode(state, 11, true, 0); + bc6h_test_mode(state, 12, true, 0); + bc6h_test_mode(state, 13, true, 0); + } + else + { + if (state->fastSkipTreshold > 0) + { + bc6h_test_mode(state, 9, false, 0); + if (state->fast_mode) bc6h_test_mode(state, 1, false, 1); + bc6h_test_mode(state, 6, false, 1 / 1.2); + bc6h_test_mode(state, 5, false, 1 / 1.2); + bc6h_test_mode(state, 0, false, 1 / 1.2); + bc6h_test_mode(state, 2, false, 1); + + bc6h_enc_2p(state); + if (!state->fast_mode) bc6h_test_mode(state, 1, true, 0); + } + + bc6h_test_mode(state, 10, false, 0); + bc6h_test_mode(state, 11, false, 1); + bc6h_test_mode(state, 12, false, 1); + bc6h_test_mode(state, 13, false, 1); + bc6h_enc_1p(state); + } +} + +void bc6h_enc_copy_settings(bc6h_enc_state state[], uniform bc6h_enc_settings settings[]) +{ + state->slow_mode = settings->slow_mode; + state->fast_mode = settings->fast_mode; + state->fastSkipTreshold = settings->fastSkipTreshold; + state->refineIterations_1p = settings->refineIterations_1p; + state->refineIterations_2p = settings->refineIterations_2p; +} + +inline void CompressBlockBC6H(uniform rgba_surface src[], int xx, uniform int yy, uniform uint8 dst[], uniform bc6h_enc_settings settings[]) +{ + bc6h_enc_state _state; + varying bc6h_enc_state* uniform state = &_state; + + bc6h_enc_copy_settings(state, settings); + load_block_interleaved_16bit(state->block, src, xx, yy); + state->best_err = 1e99; + + CompressBlockBC6H_core(state); + + store_data(dst, src->width, xx, yy, state->best_data, 4); +} + +export void CompressBlocksBC6H_ispc(uniform rgba_surface src[], uniform uint8 dst[], uniform bc6h_enc_settings settings[]) +{ + for (uniform int yy = 0; yyheight / 4; yy++) + foreach(xx = 0 ... src->width / 4) + { + CompressBlockBC6H(src, xx, yy, dst, settings); + } +} + +/////////////////////////////////////////////////////////// +// ETC encoding + +struct etc_enc_settings +{ + int fastSkipTreshold; +}; + +struct etc_enc_state +{ + float block[64]; + int prev_qcenter[3]; + + float best_err; + uint32 best_data[2]; + + uniform bool diff; + + // settings + uniform int fastSkipTreshold; +}; + +inline uniform int get_etc1_dY(uniform int table, uniform int q) +{ + static uniform const int etc_codeword_table[8][4] = + { + { -8, -2, 2, 8 }, + { -17, -5, 5, 17 }, + { -29, -9, 9, 29 }, + { -42, -13, 13, 42 }, + { -60, -18, 18, 60 }, + { -80, -24, 24, 80 }, + { -106, -33, 33, 106 }, + { -183, -47, 47, 183 }, + }; + + return etc_codeword_table[table][q]; +} + +uniform int remap_q[] = { 2, 3, 1, 0 }; + +int get_remap2_q(int x) +{ + x -= 2; + if (x < 0) x = 1 - x; + return x; +} + +int extend_4to8bits(int value) +{ + return (value << 4) | value; +} + +int extend_5to8bits(int value) +{ + return (value << 3) | (value >> 2); +} + +int quantize_4bits(float value) +{ + return clamp((value / 255.0f) * 15 + 0.5, 0, 15); +} + +int quantize_5bits(float value) +{ + return clamp((value / 255.0f) * 31 + 0.5, 0, 31); +} + +void center_quant_dequant(int qcenter[3], float center[3], uniform bool diff, int prev_qcenter[3]) +{ + if (diff) + { + for (uniform int p = 0; p < 3; p++) + { + qcenter[p] = quantize_5bits(center[p]); + + if (prev_qcenter[0] >= 0) + { + if (qcenter[p] - prev_qcenter[p] > 3) qcenter[p] = prev_qcenter[p] + 3; + if (qcenter[p] - prev_qcenter[p] < -4) qcenter[p] = prev_qcenter[p] - 4; + } + + center[p] = extend_5to8bits(qcenter[p]); + } + } + else + { + for (uniform int p = 0; p < 3; p++) + { + qcenter[p] = quantize_4bits(center[p]); + center[p] = extend_4to8bits(qcenter[p]); + } + } +} + +float quantize_pixels_etc1_half(uint32 qblock[1], float block[48], float center[3], uniform int table) +{ + float total_err = 0; + uint32 bits = 0; + + for (uniform int y = 0; y < 2; y++) + for (uniform int x = 0; x < 4; x++) + { + float best_err = sq(255) * 3; + int best_q = -1; + + for (uniform int q = 0; q < 4; q++) + { + int dY = get_etc1_dY(table, remap_q[q]); + + float err = 0; + for (int p = 0; p < 3; p++) + err += sq(block[16 * p + y*4+x] - clamp(center[p] + dY, 0, 255)); + + if (err < best_err) + { + best_err = err; + best_q = q; + } + } + + assert(best_q >= 0); + + bits |= (best_q & 1) << (x * 4 + y); + bits |= (best_q >> 1) << (x * 4 + y + 16); + total_err += best_err; + } + + qblock[0] = bits; + return total_err; +} + +float compress_etc1_half_1(uint32 out_qbits[1], int out_table[1], int out_qcenter[3], + float half_pixels[], uniform bool diff, int prev_qcenter[3]) +{ + float dc[3]; + + for (uniform int p = 0; p<3; p++) dc[p] = 0; + + for (uniform int k = 0; k<8; k++) + { + for (uniform int p = 0; p<3; p++) + dc[p] += half_pixels[k + p * 16]; + } + + float best_error = sq(255) * 3 * 8.0f; + int best_table = -1; + int best_qcenter[3]; + uint32 best_qbits; + + for (uniform int table_level = 0; table_level < 8; table_level++) + { + float center[3]; + int qcenter[3]; + uint32 qbits; + + for (uniform int p = 0; p < 3; p++) center[p] = dc[p] / 8 - get_etc1_dY(table_level, 2); + center_quant_dequant(qcenter, center, diff, prev_qcenter); + + float err = quantize_pixels_etc1_half(&qbits, half_pixels, center, table_level); + + if (err < best_error) + { + best_error = err; + best_table = table_level; + best_qbits = qbits; + for (uniform int p = 0; p < 3; p++) best_qcenter[p] = qcenter[p]; + } + } + + out_table[0] = best_table; + out_qbits[0] = best_qbits; + for (uniform int p = 0; p < 3; p++) out_qcenter[p] = best_qcenter[p]; + return best_error; +} + +float optimize_center(float colors[4][10], uniform int p, uniform int table_level) +{ + float best_center = 0; + for (uniform int q = 0; q < 4; q++) + { + best_center += (colors[q][7 + p] - get_etc1_dY(table_level, q)) * colors[q][3]; + } + best_center /= 8; + + float best_err = 0; + for (uniform int q = 0; q < 4; q++) + { + float dY = get_etc1_dY(table_level, q); + best_err += sq(clamp(best_center + dY, 0, 255) - colors[q][7 + p]) * colors[q][3]; + } + + for (uniform int branch = 0; branch < 4; branch++) + { + float new_center = 0; + float sum = 0; + for (uniform int q = 0; q < 4; q++) + { + if (branch <= 1 && q <= branch) continue; + if (branch >= 2 && q >= branch) continue; + new_center += (colors[q][7 + p] - get_etc1_dY(table_level, q)) * colors[q][3]; + sum += colors[q][3]; + } + + new_center /= sum; + + float err = 0; + for (uniform int q = 0; q < 4; q++) + { + float dY = get_etc1_dY(table_level, q); + err += sq(clamp(new_center + dY, 0, 255) - colors[q][7 + p]) * colors[q][3]; + } + + if (err < best_err) + { + best_err = err; + best_center = new_center; + } + } + + return best_center; +} + +float compress_etc1_half_7(uint32 out_qbits[1], int out_table[1], int out_qcenter[3], + float half_pixels[], etc_enc_state state[]) +{ + int err_list[165]; + int y_sorted_inv[8]; + float y_sorted[8]; + + { + int y_sorted_idx[8]; + for (uniform int k = 0; k < 8; k++) + { + float value = 0; + for (uniform int p = 0; p < 3; p++) + value += half_pixels[k + p * 16]; + + y_sorted_idx[k] = (((int)value) << 4) + k; + } + + partial_sort_list(y_sorted_idx, 8, 8); + + for (uniform int k = 0; k < 8; k++) + y_sorted_inv[k] = ((y_sorted_idx[k] & 0xF) << 4) + k; + + for (uniform int k = 0; k < 8; k++) + y_sorted[k] = (y_sorted_idx[k] >> 4) / 3.0f; + + partial_sort_list(y_sorted_inv, 8, 8); + } + + uniform int idx = -1; + for (uniform int level1 = 0; level1 <= 8; level1++) + for (uniform int level2 = level1; level2 <= 8; level2++) + for (uniform int level3 = level2; level3 <= 8; level3++) + { + idx++; + assert(idx < 165); + + float sum[4]; + float sum_sq[4]; + float count[4]; + float inv_count[4]; + + for (uniform int q = 0; q < 4; q++) + { + sum[q] = 0; + sum_sq[q] = 0; + count[q] = 0; + inv_count[q] = 0; + } + + for (uniform int k = 0; k < 8; k++) + { + uniform int q = 0; + if (k >= level1) q = 1; + if (k >= level2) q = 2; + if (k >= level3) q = 3; + + sum[q] += y_sorted[k]; + sum_sq[q] += sq(y_sorted[k]); + count[q] += 1; + } + + for (uniform int q = 0; q < 4; q++) + { + if (count[q] > 0) inv_count[q] = 1 / count[q]; + } + + float base_err = 0; + for (uniform int q = 0; q < 4; q++) base_err += sum_sq[q] - sq(sum[q]) * inv_count[q]; + + float t_err = sq(256) * 8; + for (uniform int table_level = 0; table_level < 8; table_level++) + { + float center = 0; + for (uniform int q = 0; q < 4; q++) center += sum[q] - get_etc1_dY(table_level, q) * count[q]; + center /= 8; + + float err = base_err; + for (uniform int q = 0; q < 4; q++) + { + err += sq(center + get_etc1_dY(table_level, q) - sum[q] * inv_count[q])*count[q]; + } + + t_err = min(t_err, err); + } + + int packed = (level1 * 16 + level2) * 16 + level3; + + err_list[idx] = (((int)t_err) << 12) + packed; + } + + partial_sort_list(err_list, 165, state->fastSkipTreshold); + + float best_error = sq(255) * 3 * 8.0f; + int best_table = -1; + int best_qcenter[3]; + uint32 best_qbits; + + for (uniform int i = 0; i < state->fastSkipTreshold; i++) + { + int packed = err_list[i] & 0xFFF; + int level1 = (packed >> 8) & 0xF; + int level2 = (packed >> 4) & 0xF; + int level3 = (packed >> 0) & 0xF; + + float colors[4][10]; + + for (uniform int p = 0; p < 7; p++) + for (uniform int q = 0; q < 4; q++) colors[q][p] = 0; + + uint32 qbits = 0; + for (uniform int kk = 0; kk < 8; kk++) + { + int k = y_sorted_inv[kk] & 0xF; + + int qq = 0; + if (k >= level1) qq = 1; + if (k >= level2) qq = 2; + if (k >= level3) qq = 3; + + uniform int xx = kk & 3; + uniform int yy = kk >> 2; + + int qqq = get_remap2_q(qq); + qbits |= (qqq & 1) << (yy + xx * 4); + qbits |= (qqq >> 1) << (16 + yy + xx * 4); + + float qvec[4]; + for (uniform int q = 0; q < 4; q++) + { + qvec[q] = q == qq ? 1.0 : 0.0; + colors[q][3] += qvec[q]; + } + + for (uniform int p = 0; p < 3; p++) + { + float value = half_pixels[16 * p + kk]; + for (uniform int q = 0; q < 4; q++) + { + colors[q][p] += value * qvec[q]; + colors[q][4 + p] += sq(value) * qvec[q]; + } + } + } + + float base_err = 0; + for (uniform int q = 0; q < 4; q++) + { + if (colors[q][3] > 0) + for (uniform int p = 0; p < 3; p++) + { + colors[q][7 + p] = colors[q][p] / colors[q][3]; + base_err += colors[q][4 + p] - sq(colors[q][7 + p])*colors[q][3]; + } + } + + for (uniform int table_level = 0; table_level < 8; table_level++) + { + float center[3]; + int qcenter[3]; + + for (uniform int p = 0; p < 3; p++) + { + center[p] = optimize_center(colors, p, table_level); + } + + center_quant_dequant(qcenter, center, state->diff, state->prev_qcenter); + + float err = base_err; + for (uniform int q = 0; q < 4; q++) + { + int dY = get_etc1_dY(table_level, q); + for (uniform int p = 0; p < 3; p++) + err += sq(clamp(center[p] + dY, 0, 255) - colors[q][7 + p])*colors[q][3]; + } + + if (err < best_error) + { + best_error = err; + best_table = table_level; + best_qbits = qbits; + for (uniform int p = 0; p < 3; p++) best_qcenter[p] = qcenter[p]; + } + } + } + + out_table[0] = best_table; + out_qbits[0] = best_qbits; + for (uniform int p = 0; p < 3; p++) out_qcenter[p] = best_qcenter[p]; + return best_error; +} + +float compress_etc1_half(uint32 qbits[1], int table[1], int qcenter[3], float half_pixels[], etc_enc_state state[]) +{ + float err = compress_etc1_half_7(qbits, table, qcenter, half_pixels, state); + + for (uniform int p = 0; p < 3; p++) + state->prev_qcenter[p] = qcenter[p]; + + return err; +} + +////////////////////////// +// ETC1 core + +inline uint32 bswap32(uint32 v) +{ + uint32 r = 0; + r += ((v >> 24) & 255) << 0; + r += ((v >> 16) & 255) << 8; + r += ((v >> 8) & 255) << 16; + r += ((v >> 0) & 255) << 24; + return r; +} + +void etc_pack(uint32 data[], uint32 qbits[2], int tables[2], int qcenters[2][3], uniform int diff, uniform int flip) +{ + for (uniform int k = 0; k < 2; k++) data[k] = 0; + uniform int pos = 0; + + if (diff == 0) + { + put_bits(data, &pos, 4, qcenters[1][0]); + put_bits(data, &pos, 4, qcenters[0][0]); + + put_bits(data, &pos, 4, qcenters[1][1]); + put_bits(data, &pos, 4, qcenters[0][1]); + + put_bits(data, &pos, 4, qcenters[1][2]); + put_bits(data, &pos, 4, qcenters[0][2]); + } + else + { + put_bits(data, &pos, 3, (qcenters[1][0] - qcenters[0][0]) & 7); + put_bits(data, &pos, 5, qcenters[0][0]); + + put_bits(data, &pos, 3, (qcenters[1][1] - qcenters[0][1]) & 7); + put_bits(data, &pos, 5, qcenters[0][1]); + + put_bits(data, &pos, 3, (qcenters[1][2] - qcenters[0][2]) & 7); + put_bits(data, &pos, 5, qcenters[0][2]); + } + + put_bits(data, &pos, 1, flip); + put_bits(data, &pos, 1, diff); + put_bits(data, &pos, 3, tables[1]); + put_bits(data, &pos, 3, tables[0]); + + uint32 all_qbits_flipped = (qbits[1] << 2) | qbits[0]; + uint32 all_qbits = 0; + + if (flip != 0) all_qbits = all_qbits_flipped; + + if (flip == 0) + for (uniform int k = 0; k < 2; k++) + for (uniform int y = 0; y < 4; y++) + for (uniform int x = 0; x < 4; x++) + { + int bit = (all_qbits_flipped >> (k * 16 + x * 4 + y)) & 1; + all_qbits += bit << (k * 16 + y * 4 + x); + } + + data[1] = bswap32(all_qbits); +} + +inline void CompressBlockETC1_core(etc_enc_state state[]) +{ + float flipped_block[48]; + + for (uniform int y = 0; y < 4; y++) + for (uniform int x = 0; x < 4; x++) + for (uniform int p = 0; p < 3; p++) + { + flipped_block[16 * p + x * 4 + y] = state->block[16 * p + y * 4 + x]; + } + + for (uniform int flip = 0; flip < 2; flip++) + for (uniform int diff = 1; diff >= 0; diff--) + { + state->diff = diff == 1; + state->prev_qcenter[0] = -1; + + varying float * uniform pixels = state->block; + if (flip == 0) pixels = flipped_block; + + uint32 qbits[2]; + int tables[2]; + int qcenters[2][3]; + + float err = 0; + err += compress_etc1_half(&qbits[0], &tables[0], qcenters[0], &pixels[0], state); + err += compress_etc1_half(&qbits[1], &tables[1], qcenters[1], &pixels[8], state); + + if (err < state->best_err) + { + state->best_err = err; + etc_pack(state->best_data, qbits, tables, qcenters, diff, flip); + } + } +} + +void etc_enc_copy_settings(etc_enc_state state[], uniform etc_enc_settings settings[]) +{ + state->fastSkipTreshold = settings->fastSkipTreshold; +} + +inline void CompressBlockETC1(uniform rgba_surface src[], int xx, uniform int yy, uniform uint8 dst[], uniform etc_enc_settings settings[]) +{ + etc_enc_state _state; + varying etc_enc_state* uniform state = &_state; + + etc_enc_copy_settings(state, settings); + load_block_interleaved(state->block, src, xx, yy); + state->best_err = 1e99; + + CompressBlockETC1_core(state); + + store_data(dst, src->width, xx, yy, state->best_data, 2); +} + +export void CompressBlocksETC1_ispc(uniform rgba_surface src[], uniform uint8 dst[], uniform etc_enc_settings settings[]) +{ + for (uniform int yy = 0; yyheight / 4; yy++) + foreach(xx = 0 ... src->width / 4) + { + CompressBlockETC1(src, xx, yy, dst, settings); + } +} + +export uniform int ISPCIsa_ispc() +{ +#if defined(ISPC_TARGET_SSE2) + return 0; +#elif defined(ISPC_TARGET_SSE4) + return 1; +#elif defined(ISPC_TARGET_AVX2) + return 2; +#else + return -1; +#endif +} \ No newline at end of file diff --git a/Source/ispc_texcomp/kernel_astc.ispc b/Source/ispc_texcomp/kernel_astc.ispc new file mode 100644 index 0000000..98f3e5f --- /dev/null +++ b/Source/ispc_texcomp/kernel_astc.ispc @@ -0,0 +1,2272 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// Copyright (c) 2016, Intel Corporation +// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated +// documentation files (the "Software"), to deal in the Software without restriction, including without limitation +// the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to +// permit persons to whom the Software is furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in all copies or substantial portions of +// the Software. +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO +// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, +// TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +// SOFTWARE. +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// + +typedef int8 int8_t; +typedef int32 int32_t; +typedef int64 int64_t; + +typedef unsigned int8 uint8_t; +typedef unsigned int32 uint32_t; +typedef unsigned int64 uint64_t; + +/////////////////////////// +// generic helpers + +inline float RCP(float x) +{ + return 1.0f/x; // uses rcp when compiled with --opt=fast-math + //return rcp(x); + //return rcp_fast(x); +} + +inline float RSQRT(float x) +{ + return 1.0f/sqrt(x); // uses rsqrt when compiled with --opt=fast-math + //return rsqrt(x); + //return rsqrt_fast(x); +} + +void swap(float& a, float& b) +{ + int t = a; + a = b; b = t; +} + +void swap(int& a, int& b) +{ + int t = a; + a = b; b = t; +} + +void swap(uint32_t& a, uint32_t& b) +{ + uint32_t t = a; + a = b; b = t; +} + +void swap(uint8_t& a, uint8_t& b) +{ + uint8_t t = a; + a = b; b = t; +} + +inline float sq(float v) +{ + return v*v; +} + +inline float clamp(float v, int a, int b) +{ + return clamp(v, (float)a, (float)b); +} + +inline float dot3(float a[3], float b[3]) +{ + return a[0] * b[0] + a[1] * b[1] + a[2] * b[2]; +} + +inline float dot4(float a[4], float b[4]) +{ + return a[0] * b[0] + a[1] * b[1] + a[2] * b[2] + a[3] * b[3]; +} + +// the following helpers isolate performance warnings + +inline uint32_t gather_uint(const uniform uint32_t* const uniform ptr, int idx) +{ + return ptr[idx]; // (perf warning expected) +} + +inline float gather_float(uniform float* uniform ptr, int idx) +{ + return ptr[idx]; // (perf warning expected) +} + +inline float gather_float(varying float* uniform ptr, int idx) +{ + return ptr[idx]; // (perf warning expected) +} + +inline void scatter_uint(uniform uint32_t* ptr, int idx, uint32_t value) +{ + ptr[idx] = value; // (perf warning expected) +} + +inline void scatter_float(uniform float* uniform ptr, int idx, float value) +{ + ptr[idx] = value; // (perf warning expected) +} + +inline void scatter_float(varying float* uniform ptr, int idx, float value) +{ + ptr[idx] = value; // (perf warning expected) +} + +/////////////////////////////////////////////////////////// +// ASTC shared functions + +struct rgba_surface +{ + uint8_t* ptr; + int width, height, stride; +}; + +inline void set_pixel(float pixels[], uniform int p, uniform int x, uniform int y, float value); + +inline void load_block_interleaved(float pixels[], uniform rgba_surface src[], int xx, int yy, uniform int width, uniform int height) +{ + uniform int pitch = width * height; + for (uniform int y = 0; y < height; y++) + for (uniform int x = 0; x < width; x++) + { + uint32_t rgba = gather_uint((uint32_t*)src->ptr, ((yy * height + y)*src->stride + (xx * width + x) * 4)/4); + + set_pixel(pixels, 0, x, y, (int)((rgba >> 0) & 255)); + set_pixel(pixels, 1, x, y, (int)((rgba >> 8) & 255)); + set_pixel(pixels, 2, x, y, (int)((rgba >> 16) & 255)); + set_pixel(pixels, 3, x, y, (int)((rgba >> 24) & 255)); + } +} + +struct astc_enc_settings +{ + int block_width; + int block_height; + int channels; + + int fastSkipTreshold; + int refineIterations; +}; + +export uniform int get_programCount() +{ + return programCount; +} + +inline float get_pixel(float pixels[], uniform int p, uniform int x, uniform int y) +{ + uniform static const int ystride = 8; + uniform static const int pstride = 64; + + return pixels[pstride * p + ystride * y + x]; +} + +inline void set_pixel(float pixels[], uniform int p, uniform int x, uniform int y, float value) +{ + uniform static const int ystride = 8; + uniform static const int pstride = 64; + + pixels[pstride * p + ystride * y + x] = value; +} + +struct pixel_set +{ + varying float* uniform pixels; + + uniform int width; + uniform int height; +}; + +inline void clear_alpha(float pixels[], uniform int width, uniform int height) +{ + for (uniform int y = 0; y < height; y++) + for (uniform int x = 0; x < width; x++) + { + set_pixel(pixels, 3, x, y, 255); + } +} + +void rotate_plane(pixel_set block[], int p) +{ + uniform int pitch = block->height * block->width; + + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + float r = get_pixel(block->pixels, 0, x, y); + float g = get_pixel(block->pixels, 1, x, y); + float b = get_pixel(block->pixels, 2, x, y); + float a = get_pixel(block->pixels, 3, x, y); + + if (p == 0) swap(a, r); + if (p == 1) swap(a, g); + if (p == 2) swap(a, b); + + set_pixel(block->pixels, 0, x, y, r); + set_pixel(block->pixels, 1, x, y, g); + set_pixel(block->pixels, 2, x, y, b); + set_pixel(block->pixels, 3, x, y, a); + } +} + +inline void compute_moments(float stats[15], pixel_set block[], uniform int channels) +{ + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + float rgba[4]; + for (uniform int p = 0; p < channels; p++) rgba[p] = get_pixel(block->pixels, p, x, y); + + stats[10] += rgba[0]; + stats[11] += rgba[1]; + stats[12] += rgba[2]; + + stats[0] += rgba[0] * rgba[0]; + stats[1] += rgba[0] * rgba[1]; + stats[2] += rgba[0] * rgba[2]; + + stats[4] += rgba[1] * rgba[1]; + stats[5] += rgba[1] * rgba[2]; + + stats[7] += rgba[2] * rgba[2]; + + if (channels == 4) + { + stats[13] += rgba[3]; + + stats[3] += rgba[0] * rgba[3]; + stats[6] += rgba[1] * rgba[3]; + stats[8] += rgba[2] * rgba[3]; + stats[9] += rgba[3] * rgba[3]; + } + } + + stats[14] += block->height * block->width; +} + +inline void covar_from_stats(float covar[10], float stats[15], uniform int channels) +{ + covar[0] = stats[0] - stats[10 + 0] * stats[10 + 0] / stats[14]; + covar[1] = stats[1] - stats[10 + 0] * stats[10 + 1] / stats[14]; + covar[2] = stats[2] - stats[10 + 0] * stats[10 + 2] / stats[14]; + + covar[4] = stats[4] - stats[10 + 1] * stats[10 + 1] / stats[14]; + covar[5] = stats[5] - stats[10 + 1] * stats[10 + 2] / stats[14]; + + covar[7] = stats[7] - stats[10 + 2] * stats[10 + 2] / stats[14]; + + if (channels == 4) + { + covar[3] = stats[3] - stats[10 + 0] * stats[10 + 3] / stats[14]; + covar[6] = stats[6] - stats[10 + 1] * stats[10 + 3] / stats[14]; + covar[8] = stats[8] - stats[10 + 2] * stats[10 + 3] / stats[14]; + covar[9] = stats[9] - stats[10 + 3] * stats[10 + 3] / stats[14]; + } +} + +inline void compute_covar_dc(float covar[], float dc[], pixel_set block[], bool zero_based, uniform int channels) +{ + float stats[15] = { 0 }; + compute_moments(stats, block, channels); + + if (zero_based) + for (uniform int p = 0; p < 4; p++) stats[10 + p] = 0; + + covar_from_stats(covar, stats, channels); + for (uniform int p = 0; p < channels; p++) dc[p] = stats[10 + p] / stats[14]; +} + +inline void ssymv3(float a[4], float covar[10], float b[4]) +{ + a[0] = covar[0] * b[0] + covar[1] * b[1] + covar[2] * b[2]; + a[1] = covar[1] * b[0] + covar[4] * b[1] + covar[5] * b[2]; + a[2] = covar[2] * b[0] + covar[5] * b[1] + covar[7] * b[2]; +} + +inline void ssymv4(float a[4], float covar[10], float b[4]) +{ + a[0] = covar[0] * b[0] + covar[1] * b[1] + covar[2] * b[2] + covar[3] * b[3]; + a[1] = covar[1] * b[0] + covar[4] * b[1] + covar[5] * b[2] + covar[6] * b[3]; + a[2] = covar[2] * b[0] + covar[5] * b[1] + covar[7] * b[2] + covar[8] * b[3]; + a[3] = covar[3] * b[0] + covar[6] * b[1] + covar[8] * b[2] + covar[9] * b[3]; +} + +inline void compute_axis(float axis[4], float covar[10], uniform const int powerIterations, uniform int channels) +{ + float vec[4] = { 1, 1, 1, 1 }; + + for (uniform int i = 0; i < powerIterations; i++) + { + if (channels == 3) ssymv3(axis, covar, vec); + if (channels == 4) ssymv4(axis, covar, vec); + for (uniform int p = 0; p < channels; p++) vec[p] = axis[p]; + + if (i % 2 == 1) // renormalize every other iteration + { + float norm_sq = 0; + for (uniform int p = 0; p < channels; p++) + norm_sq += axis[p] * axis[p]; + + float rnorm = RSQRT(norm_sq); + for (uniform int p = 0; p < channels; p++) vec[p] *= rnorm; + } + } + + for (uniform int p = 0; p < channels; p++) axis[p] = vec[p]; +} + +void compute_pca_endpoints(float ep[8], pixel_set block[], bool zero_based, uniform int channels) +{ + float dc[4]; + float cov[10]; + compute_covar_dc(cov, dc, block, zero_based, channels); + + uniform int powerIterations = 10; + + float eps = sq(0.001) * 1000; + cov[0] += eps; + cov[4] += eps; + cov[7] += eps; + cov[9] += eps; + + float dir[4]; + compute_axis(dir, cov, powerIterations, channels); + + float ext[2] = { 1000, -1000 }; + + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + float proj = 0; + for (uniform int p = 0; p < channels; p++) proj += (get_pixel(block->pixels, p, x, y) - dc[p]) * dir[p]; + + ext[0] = min(ext[0], proj); + ext[1] = max(ext[1], proj); + } + + if (ext[1] - 1.0f < ext[0]) + { + ext[1] += 0.5f; + ext[0] -= 0.5f; + } + + for (uniform int i = 0; i < 2; i++) + for (uniform int p = 0; p < channels; p++) + { + ep[p * 2 + i] = dc[p] + dir[p] * ext[i]; + } +} + +uniform static const int range_table[][3] = +{ + //2^ 3^ 5^ + { 1, 0, 0 }, // 0..1 + { 0, 1, 0 }, // 0..2 + { 2, 0, 0 }, // 0..3 + + { 0, 0, 1 }, // 0..4 + { 1, 1, 0 }, // 0..5 + { 3, 0, 0 }, // 0..7 + + { 1, 0, 1 }, // 0..9 + { 2, 1, 0 }, // 0..11 + { 4, 0, 0 }, // 0..15 + + { 2, 0, 1 }, // 0..19 + { 3, 1, 0 }, // 0..23 + { 5, 0, 0 }, // 0..31 + + { 3, 0, 1 }, // 0..39 + { 4, 1, 0 }, // 0..47 + { 6, 0, 0 }, // 0..63 + + { 4, 0, 1 }, // 0..79 + { 5, 1, 0 }, // 0..95 + { 7, 0, 0 }, // 0..127 + + { 5, 0, 1 }, // 0..159 + { 6, 1, 0 }, // 0..191 + { 8, 0, 0 }, // 0..255 +}; + +uniform int get_levels(uniform int range) +{ + return (1 + 2 * range_table[range][1] + 4 * range_table[range][2]) << range_table[range][0]; +} + +struct range_values +{ + int levels_m; + int levels_m_rcp; + int levels_e; + int levels; +}; + +void fill_range_values(range_values values[], int _range[]) +{ + int range = *_range; + int range_div3 = (range * 21846) >> 16; + int range_mod3 = range - range_div3 * 3; + + int levels_m = max(2, 5 - range_mod3 * 2); + int levels_e = max(0, range_mod3 + range_div3 - 1); + if (range == 0) levels_m = 2; + + int levels_m_rcp = 0x10000 / 2 + 1; + if (levels_m == 3) levels_m_rcp = 0x10000 / 3 + 1; + if (levels_m == 5) levels_m_rcp = 0x10000 / 5 + 1; + + values->levels_e = levels_e; + values->levels_m = levels_m; + values->levels_m_rcp = levels_m_rcp; + + values->levels = levels_m << levels_e; +} + +range_values get_range_values(int range) +{ + range_values values; + fill_range_values(&values, &range); + return values; +} + +int get_levels(int range) +{ + int range_div3 = (range * 21846) >> 16; + int range_mod3 = range - range_div3 * 3; + + int levels_m = max(2, 5 - range_mod3 * 2); + int levels_e = range_mod3 + range_div3 - 1; + + return (levels_m << (levels_e + 1)) >> 1; +} + +uniform float get_sq_rcp_levels(uniform int range) +{ + uniform static const float table[] = + { + 1.000000, 0.250000, 0.111111, + 0.062500, 0.040000, 0.020408, + 0.012346, 0.008264, 0.004444, + 0.002770, 0.001890, 0.001041, + 0.000657, 0.000453, 0.000252, + 0.000160, 0.000111, 0.000062, + 0.000040, 0.000027, 0.000015, + }; + + return table[range]; +} + +/////////////////////////////////////////////////////////// +// ASTC candidate ranking + +struct astc_rank_state +{ + float pixels[256]; + + float pca_error[2][5]; + float alpha_error[2][5]; + float sq_norm[2][5]; + float scale_error[7][7]; // 2x2 to 8x8 + + float best_scores[64]; + uint32_t best_modes[64]; + + // settings + uniform int block_width; + uniform int block_height; + uniform int pitch; + + uniform int fastSkipTreshold; +}; + +struct astc_mode +{ + int width; + int height; + bool dual_plane; + int weight_range; + int color_component_selector; + int partitions; + int partition_id; + int color_endpoint_pairs; + int color_endpoint_modes[2]; + int endpoint_range; +}; + +void dct_4(float values[], uniform int stride) +{ + uniform static const float scale[] = { 0.5, 0.707106769 }; + uniform static const float c[5] = { 1, 0.923879533, 0.707106769, 0.382683432, 0 }; + + float data[4]; + for (uniform int i = 0; i < 2; i++) + { + float a = values[stride * i]; + float b = values[stride * (3 - i)]; + data[0 + i] = a + b; + data[2 + i] = a - b; + } + + for (uniform int i = 0; i < 4; i++) + { + float acc = 0; + varying float* uniform input = &data[(i % 2) * 2]; + for (uniform int j = 0; j < 2; j++) + { + uniform int e = (2 * j + 1)*i; + e = e % (4 * 4); + uniform float w = 1; + if (e>8) { e = 16 - e; } + if (e>4) { w = -1; e = 8 - e; } + w *= c[e]; + acc += w * input[j]; + } + + values[stride * i] = acc * scale[i > 0]; + } +} + +void dct_6(float values[], uniform int stride) +{ + uniform static const float scale[] = { 0.408248290, 0.577350269 }; + uniform static const float c[7] = + { 1, 0.965925813, 0.866025388, 0.707106769, 0.500000000, 0.258819044, 0 }; + + float data[6]; + for (uniform int i = 0; i < 3; i++) + { + float a = values[stride * i]; + float b = values[stride * (5 - i)]; + data[0 + i] = a + b; + data[3 + i] = a - b; + } + + for (uniform int i = 0; i < 6; i++) + { + float acc = 0; + varying float* uniform input = &data[(i % 2) * 3]; + for (uniform int j = 0; j < 3; j++) + { + uniform int e = (2 * j + 1)*i; + e = e % (4 * 6); + uniform float w = 1; + if (e>12) { e = 24 - e; } + if (e>6) { w = -1; e = 12 - e; } + w *= c[e]; + acc += w * input[j]; + } + + values[stride * i] = acc * scale[i > 0]; + } +} + +void dct_n(float values[], uniform int stride, uniform int n) +{ + uniform static const float pi = 3.14159265358979323846; + + assert(n <= 16); + uniform float c[16 + 1]; + for (uniform int i = 0; i <= n; i++) + c[i] = cos((i / (4.0 * n) * 2 * pi)); + + uniform float scale[] = { 1 / sqrt(1.0*n), 1 / sqrt(n / 2.0), }; + + float data[16]; + for (uniform int i = 0; i < n; i++) + data[i] = values[stride * i]; + + for (uniform int i = 0; i < n; i++) + { + float acc = 0; + for (uniform int j = 0; j < n; j++) + { + uniform int e = (2 * j + 1)*i; + e = e % (4 * n); + float w = 1; + if (e > 2 * n) { e = 4 * n - e; } + if (e > n) { w = -1; e = 2 * n - e; } + assert(e <= n); + w *= c[e]; + acc += w * data[j]; + } + + values[stride * i] = acc * scale[i > 0]; + } +} + +void dct(float values[], uniform int stride, uniform int n) +{ + if (false) {} + else if (n == 8) dct_n(values, stride, 8); + else if (n == 6) dct_6(values, stride); + else if (n == 5) dct_n(values, stride, 5); + else if (n == 4) dct_4(values, stride); + else + { + assert(false); + } +} + +void compute_dct_inplace(pixel_set block[], uniform int channels) +{ + uniform static const int stride = 8; + uniform static const int pitch = 64; + + for (uniform int p = 0; p < channels; p++) + { + for (uniform int y = 0; y < block->height; y++) + dct(&block->pixels[pitch * p + y * stride], 1, block->width); + + for (uniform int x = 0; x < block->width; x++) + dct(&block->pixels[pitch * p + x], stride, block->height); + } +} + +void compute_metrics(astc_rank_state state[]) +{ + float temp_pixels[256]; + pixel_set _pset; varying pixel_set* uniform pset = &_pset; + pset->pixels = temp_pixels; + pset->width = state->block_width; + pset->height = state->block_height; + + for (uniform int p = 0; p < 4; p++) + for (uniform int y = 0; y < state->block_height; y++) + for (uniform int x = 0; x < state->block_width; x++) + { + float value = get_pixel(state->pixels, p, x, y); + set_pixel(pset->pixels, p, x, y, value); + } + + for (uniform int i = 0; i < 2; i++) + { + bool zero_based = (i == 1); + float endpoints[8]; + compute_pca_endpoints(endpoints, pset, zero_based, 4); + + float base[4], dir[4]; + for (int p = 0; p < 4; p++) dir[p] = endpoints[p * 2 + 1] - endpoints[p * 2]; + for (int p = 0; p < 4; p++) base[p] = endpoints[p * 2]; + float sq_norm = dot4(dir, dir) + 0.00001; + + float pca_error = 0; + float alpha_error = 0; + float pca_alpha_error = 0; + for (uniform int y = 0; y < state->block_height; y++) + for (uniform int x = 0; x < state->block_width; x++) + { + float pixel[4]; + for (uniform int p = 0; p < 4; p++) pixel[p] = get_pixel(pset->pixels, p, x, y) - base[p]; + float proj = dot4(pixel, dir) / sq_norm; + for (uniform int p = 0; p < 3; p++) pca_error += sq(get_pixel(pset->pixels, p, x, y) - (proj * dir[p] + base[p])); + pca_alpha_error += sq(get_pixel(pset->pixels, 3, x, y) - (proj * dir[3] + base[3])); + alpha_error += sq(get_pixel(pset->pixels, 3, x, y) - 255); + } + + state->pca_error[i][0] = pca_error + pca_alpha_error; + state->alpha_error[i][0] = alpha_error - pca_alpha_error; + state->sq_norm[i][0] = sq_norm; + } + + for (uniform int i = 0; i < 2; i++) + for (uniform int c = 1; c < 5; c++) + { + rotate_plane(pset, c - 1); + + bool zero_based = (i == 1); + float endpoints[8]; + compute_pca_endpoints(endpoints, pset, zero_based, 3); + + float base[3], dir[3]; + for (int p = 0; p < 3; p++) dir[p] = endpoints[p * 2 + 1] - endpoints[p * 2]; + for (int p = 0; p < 3; p++) base[p] = endpoints[p * 2]; + float sq_norm = dot3(dir, dir) + 0.00001; + + float pca_error = 0; + float alpha_error = 0; + float pca_alpha_error = 0; + float ext[2] = { 1000, -1000 }; + for (uniform int y = 0; y < state->block_height; y++) + for (uniform int x = 0; x < state->block_width; x++) + { + float pixel[3]; + for (uniform int p = 0; p < 3; p++) pixel[p] = get_pixel(pset->pixels, p, x, y) - base[p]; + float proj = dot3(pixel, dir) / sq_norm; + for (uniform int p = 0; p < 3; p++) + { + if (p == c - 1) + { + pca_alpha_error += sq(get_pixel(pset->pixels, p, x, y) - (proj * dir[p] + base[p])); + alpha_error += sq(get_pixel(pset->pixels, p, x, y) - 255); + } + else + { + pca_error += sq(get_pixel(pset->pixels, p, x, y) - (proj * dir[p] + base[p])); + } + } + + float value = get_pixel(pset->pixels, 3, x, y); + ext[0] = min(ext[0], value); + ext[1] = max(ext[1], value); + } + + state->pca_error[i][c] = pca_error + pca_alpha_error; + state->alpha_error[i][c] = alpha_error - pca_alpha_error; + state->sq_norm[i][c] = sq_norm + sq(ext[1] - ext[0]); + + // rotate back + rotate_plane(pset, c - 1); + } + + compute_dct_inplace(pset, 4); + + for (uniform int h = 2; h <= state->block_height; h++) + for (uniform int w = 2; w <= state->block_width; w++) + { + uniform int stride = 8; + uniform int pitch = 64; + + float sq_sum = 0; + + for (uniform int y = 0; y < state->block_height; y++) + for (uniform int x = 0; x < state->block_width; x++) + { + if (y < h && x < w) continue; + + for (uniform int p = 0; p < 4; p++) + sq_sum += sq(pset->pixels[pitch * p + stride * y + x]); + } + + state->scale_error[h - 2][w - 2] = sq_sum; + } +} + +float estimate_error(astc_rank_state state[], uniform astc_mode mode[]) +{ + uniform int c = 0; + if (mode->dual_plane) c = 1 + mode->color_component_selector; + + float scale_error = state->scale_error[mode->height - 2][mode->width - 2]; + + uniform bool zero_based = (mode->color_endpoint_modes[0] % 4) == 2; + float pca_error = state->pca_error[zero_based][c]; + float sq_norm = state->sq_norm[zero_based][c]; + + if (mode->color_endpoint_modes[0] <= 8) pca_error += state->alpha_error[zero_based][c]; + + uniform float sq_rcp_w_levels = get_sq_rcp_levels(mode->weight_range); + uniform float sq_rcp_ep_levels = get_sq_rcp_levels(mode->endpoint_range); + float quant_error = 0; + + quant_error += 2 * sq_norm * sq_rcp_w_levels; + quant_error += 9000 * (state->block_height * state->block_width) * sq_rcp_ep_levels; + + float error = 0; + error += scale_error; + error += pca_error; + error += quant_error; + + return error; +} + +void insert_element(astc_rank_state state[], float error, uint32_t packed_mode, float threshold_error[]) +{ + float max_error = 0; + + for (uniform int k = 0; k < state->fastSkipTreshold; k++) + { + if (state->best_scores[k] > error) + { + swap(state->best_scores[k], error); + swap(state->best_modes[k], packed_mode); + } + + max_error = max(max_error, state->best_scores[k]); + } + + *threshold_error = max_error; +} + +uniform static const int packed_modes_count = 3334; +uniform static const uint32_t packed_modes[3334] = +{ + 0x0006D400, 0x0016D340, 0x0026D380, 0x0036CDC0, 0x00469400, 0x00569401, 0x00668702, 0x00769440, + 0x00868D41, 0x00969480, 0x00A68D81, 0x00B693C0, 0x00C688C1, 0x00D4D400, 0x00E4D401, 0x00F4C702, + 0x0104D440, 0x0114CD41, 0x0124D480, 0x0134CD81, 0x0144D3C0, 0x0154C8C1, 0x01667400, 0x01767401, + 0x01867302, 0x01966803, 0x01A67440, 0x01B67341, 0x01C66B42, 0x01D66443, 0x01E67480, 0x01F67381, + 0x02066B82, 0x02166483, 0x022674C0, 0x02366DC1, 0x024667C2, 0x0253D400, 0x0263D401, 0x0273D302, + 0x0283C803, 0x0293D440, 0x02A3D341, 0x02B3CB42, 0x02C3C443, 0x02D3D480, 0x02E3D381, 0x02F3CB82, + 0x0303C483, 0x0313D4C0, 0x0323CDC1, 0x0333C7C2, 0x03465400, 0x03565401, 0x03665402, 0x03765403, + 0x03865004, 0x03964705, 0x03A65440, 0x03B65441, 0x03C65342, 0x03D64E43, 0x03E64944, 0x03F65480, + 0x04065481, 0x04165382, 0x04264E83, 0x04364984, 0x044654C0, 0x045652C1, 0x04664DC2, 0x047649C3, + 0x048646C4, 0x049E5400, 0x04AE5240, 0x04BE5280, 0x04CE4DC0, 0x04DE5410, 0x04EE5250, 0x04FE5290, + 0x050E4DD0, 0x051E5420, 0x052E5260, 0x053E52A0, 0x054E4DE0, 0x055E52B0, 0x056E4DF0, 0x0572D400, + 0x0582D401, 0x0592D402, 0x05A2D403, 0x05B2D004, 0x05C2C705, 0x05D2D440, 0x05E2D441, 0x05F2D342, + 0x0602CE43, 0x0612C944, 0x0622D480, 0x0632D481, 0x0642D382, 0x0652CE83, 0x0662C984, 0x0672D4C0, + 0x0682D2C1, 0x0692CDC2, 0x06A2C9C3, 0x06B2C6C4, 0x06CAD400, 0x06DAD240, 0x06EAD280, 0x06FACDC0, + 0x070AD410, 0x071AD250, 0x072AD290, 0x073ACDD0, 0x074AD420, 0x075AD260, 0x076AD2A0, 0x077ACDE0, + 0x078AD2B0, 0x079ACDF0, 0x07A63400, 0x07B63401, 0x07C63402, 0x07D63403, 0x07E63404, 0x07F63405, + 0x08063306, 0x08162E07, 0x08262708, 0x08363440, 0x08463441, 0x08563442, 0x08663443, 0x08763444, + 0x08862F45, 0x08962B46, 0x08A62847, 0x08B63480, 0x08C63481, 0x08D63482, 0x08E63483, 0x08F63484, + 0x09062F85, 0x09162B86, 0x09262887, 0x093634C0, 0x094634C1, 0x095633C2, 0x096630C3, 0x09762EC4, + 0x09862AC5, 0x099627C6, 0x09A625C7, 0x09BE3400, 0x09CE3401, 0x09DE2502, 0x09EE3440, 0x09FE2C41, + 0x0A0E3480, 0x0A1E2C81, 0x0A2E33C0, 0x0A3E28C1, 0x0A4E3410, 0x0A5E3411, 0x0A6E2512, 0x0A7E3450, + 0x0A8E2C51, 0x0A9E3490, 0x0AAE2C91, 0x0ABE33D0, 0x0ACE28D1, 0x0ADE3420, 0x0AEE3421, 0x0AFE2522, + 0x0B0E3460, 0x0B1E2C61, 0x0B2E34A0, 0x0B3E2CA1, 0x0B4E33E0, 0x0B5E28E1, 0x0B6E34B0, 0x0B7E2CB1, + 0x0B8E33F0, 0x0B9E28F1, 0x0BA1D400, 0x0BB1D401, 0x0BC1D402, 0x0BD1D403, 0x0BE1D404, 0x0BF1D405, + 0x0C01D306, 0x0C11CE07, 0x0C21C708, 0x0C31D440, 0x0C41D441, 0x0C51D442, 0x0C61D443, 0x0C71D444, + 0x0C81CF45, 0x0C91CB46, 0x0CA1C847, 0x0CB1D480, 0x0CC1D481, 0x0CD1D482, 0x0CE1D483, 0x0CF1D484, + 0x0D01CF85, 0x0D11CB86, 0x0D21C887, 0x0D31D4C0, 0x0D41D4C1, 0x0D51D3C2, 0x0D61D0C3, 0x0D71CEC4, + 0x0D81CAC5, 0x0D91C7C6, 0x0DA1C5C7, 0x0DB9D400, 0x0DC9D401, 0x0DD9C502, 0x0DE9D440, 0x0DF9CC41, + 0x0E09D480, 0x0E19CC81, 0x0E29D3C0, 0x0E39C8C1, 0x0E49D410, 0x0E59D411, 0x0E69C512, 0x0E79D450, + 0x0E89CC51, 0x0E99D490, 0x0EA9CC91, 0x0EB9D3D0, 0x0EC9C8D1, 0x0ED9D420, 0x0EE9D421, 0x0EF9C522, + 0x0F09D460, 0x0F19CC61, 0x0F29D4A0, 0x0F39CCA1, 0x0F49D3E0, 0x0F59C8E1, 0x0F69D4B0, 0x0F79CCB1, + 0x0F89D3F0, 0x0F99C8F1, 0x0FA61401, 0x0FB61402, 0x0FC61403, 0x0FD61404, 0x0FE61405, 0x0FF61406, + 0x10061407, 0x10161408, 0x10261409, 0x1036140A, 0x1046130B, 0x10561441, 0x10661442, 0x10761443, + 0x10861444, 0x10961445, 0x10A61446, 0x10B61447, 0x10C61348, 0x10D61049, 0x10E60E4A, 0x10F60B4B, + 0x11061481, 0x11161482, 0x11261483, 0x11361484, 0x11461485, 0x11561486, 0x11661487, 0x11761388, + 0x11861089, 0x11960E8A, 0x11A60B8B, 0x11B614C1, 0x11C614C2, 0x11D614C3, 0x11E614C4, 0x11F613C5, + 0x120611C6, 0x121610C7, 0x12260DC8, 0x12360BC9, 0x12460ACA, 0x125607CB, 0x126E1400, 0x127E1401, + 0x128E1402, 0x129E1403, 0x12AE0E04, 0x12BE0505, 0x12CE1440, 0x12DE1441, 0x12EE1242, 0x12FE0D43, + 0x130E0844, 0x131E1480, 0x132E1481, 0x133E1282, 0x134E0D83, 0x135E0884, 0x136E14C0, 0x137E11C1, + 0x138E0DC2, 0x139E08C3, 0x13AE05C4, 0x13BE1410, 0x13CE1411, 0x13DE1412, 0x13EE1413, 0x13FE0E14, + 0x140E0515, 0x141E1450, 0x142E1451, 0x143E1252, 0x144E0D53, 0x145E0854, 0x146E1490, 0x147E1491, + 0x148E1292, 0x149E0D93, 0x14AE0894, 0x14BE14D0, 0x14CE11D1, 0x14DE0DD2, 0x14EE08D3, 0x14FE05D4, + 0x150E1420, 0x151E1421, 0x152E1422, 0x153E1423, 0x154E0E24, 0x155E0525, 0x156E1460, 0x157E1461, + 0x158E1262, 0x159E0D63, 0x15AE0864, 0x15BE14A0, 0x15CE14A1, 0x15DE12A2, 0x15EE0DA3, 0x15FE08A4, + 0x160E14E0, 0x161E11E1, 0x162E0DE2, 0x163E08E3, 0x164E05E4, 0x165E14B0, 0x166E14B1, 0x167E12B2, + 0x168E0DB3, 0x169E08B4, 0x16AE14F0, 0x16BE11F1, 0x16CE0DF2, 0x16DE08F3, 0x16EE05F4, 0x16F0D401, + 0x1700D402, 0x1710D403, 0x1720D404, 0x1730D405, 0x1740D406, 0x1750D407, 0x1760D408, 0x1770D409, + 0x1780D40A, 0x1790D30B, 0x17A0D441, 0x17B0D442, 0x17C0D443, 0x17D0D444, 0x17E0D445, 0x17F0D446, + 0x1800D447, 0x1810D348, 0x1820D049, 0x1830CE4A, 0x1840CB4B, 0x1850D481, 0x1860D482, 0x1870D483, + 0x1880D484, 0x1890D485, 0x18A0D486, 0x18B0D487, 0x18C0D388, 0x18D0D089, 0x18E0CE8A, 0x18F0CB8B, + 0x1900D4C1, 0x1910D4C2, 0x1920D4C3, 0x1930D4C4, 0x1940D3C5, 0x1950D1C6, 0x1960D0C7, 0x1970CDC8, + 0x1980CBC9, 0x1990CACA, 0x19A0C7CB, 0x19B8D400, 0x19C8D401, 0x19D8D402, 0x19E8D403, 0x19F8CE04, + 0x1A08C505, 0x1A18D440, 0x1A28D441, 0x1A38D242, 0x1A48CD43, 0x1A58C844, 0x1A68D480, 0x1A78D481, + 0x1A88D282, 0x1A98CD83, 0x1AA8C884, 0x1AB8D4C0, 0x1AC8D1C1, 0x1AD8CDC2, 0x1AE8C8C3, 0x1AF8C5C4, + 0x1B08D410, 0x1B18D411, 0x1B28D412, 0x1B38D413, 0x1B48CE14, 0x1B58C515, 0x1B68D450, 0x1B78D451, + 0x1B88D252, 0x1B98CD53, 0x1BA8C854, 0x1BB8D490, 0x1BC8D491, 0x1BD8D292, 0x1BE8CD93, 0x1BF8C894, + 0x1C08D4D0, 0x1C18D1D1, 0x1C28CDD2, 0x1C38C8D3, 0x1C48C5D4, 0x1C58D420, 0x1C68D421, 0x1C78D422, + 0x1C88D423, 0x1C98CE24, 0x1CA8C525, 0x1CB8D460, 0x1CC8D461, 0x1CD8D262, 0x1CE8CD63, 0x1CF8C864, + 0x1D08D4A0, 0x1D18D4A1, 0x1D28D2A2, 0x1D38CDA3, 0x1D48C8A4, 0x1D58D4E0, 0x1D68D1E1, 0x1D78CDE2, + 0x1D88C8E3, 0x1D98C5E4, 0x1DA8D4B0, 0x1DB8D4B1, 0x1DC8D2B2, 0x1DD8CDB3, 0x1DE8C8B4, 0x1DF8D4F0, + 0x1E08D1F1, 0x1E18CDF2, 0x1E28C8F3, 0x1E38C5F4, 0x1E449400, 0x1E549401, 0x1E649402, 0x1E749003, + 0x1E848804, 0x1E949440, 0x1EA49441, 0x1EB48F42, 0x1EC48943, 0x1ED48444, 0x1EE49480, 0x1EF49481, + 0x1F048F82, 0x1F148983, 0x1F248484, 0x1F3494C0, 0x1F4490C1, 0x1F548AC2, 0x1F6486C3, 0x1F747400, + 0x1F847401, 0x1F947402, 0x1FA47403, 0x1FB47404, 0x1FC46B05, 0x1FD47440, 0x1FE47441, 0x1FF47442, + 0x20047043, 0x20146C44, 0x20246645, 0x20347480, 0x20447481, 0x20547482, 0x20647083, 0x20746C84, + 0x20846685, 0x209474C0, 0x20A473C1, 0x20B46FC2, 0x20C46BC3, 0x20D468C4, 0x20E464C5, 0x20FC7400, + 0x210C6501, 0x211C7440, 0x212C7480, 0x213C6EC0, 0x214C7410, 0x215C6511, 0x216C7450, 0x217C7490, + 0x218C6ED0, 0x219C7420, 0x21AC6521, 0x21BC7460, 0x21CC74A0, 0x21DC6EE0, 0x21EC74B0, 0x21FC6EF0, + 0x22039400, 0x22139401, 0x22239402, 0x22339403, 0x22439404, 0x22538B05, 0x22639440, 0x22739441, + 0x22839442, 0x22939043, 0x22A38C44, 0x22B38645, 0x22C39480, 0x22D39481, 0x22E39482, 0x22F39083, + 0x23038C84, 0x23138685, 0x232394C0, 0x233393C1, 0x23438FC2, 0x23538BC3, 0x236388C4, 0x237384C5, + 0x238B9400, 0x239B8501, 0x23AB9440, 0x23BB9480, 0x23CB8EC0, 0x23DB9410, 0x23EB8511, 0x23FB9450, + 0x240B9490, 0x241B8ED0, 0x242B9420, 0x243B8521, 0x244B9460, 0x245B94A0, 0x246B8EE0, 0x247B94B0, + 0x248B8EF0, 0x24937400, 0x24A37401, 0x24B37402, 0x24C37403, 0x24D37404, 0x24E37405, 0x24F37006, + 0x25036B07, 0x25137440, 0x25237441, 0x25337442, 0x25437443, 0x25537344, 0x25636E45, 0x25736946, + 0x25836647, 0x25937480, 0x25A37481, 0x25B37482, 0x25C37483, 0x25D37384, 0x25E36E85, 0x25F36986, + 0x26036687, 0x261374C0, 0x262374C1, 0x263373C2, 0x26436FC3, 0x26536DC4, 0x266369C5, 0x267366C6, + 0x268364C7, 0x269B7400, 0x26AB7101, 0x26BB7440, 0x26CB6A41, 0x26DB7480, 0x26EB6A81, 0x26FB72C0, + 0x270B67C1, 0x271B7410, 0x272B7111, 0x273B7450, 0x274B6A51, 0x275B7490, 0x276B6A91, 0x277B72D0, + 0x278B67D1, 0x279B7420, 0x27AB7121, 0x27BB7460, 0x27CB6A61, 0x27DB74A0, 0x27EB6AA1, 0x27FB72E0, + 0x280B67E1, 0x281B74B0, 0x282B6AB1, 0x283B72F0, 0x284B67F1, 0x28545400, 0x28645401, 0x28745402, + 0x28845403, 0x28945404, 0x28A45405, 0x28B45306, 0x28C44E07, 0x28D44708, 0x28E45440, 0x28F45441, + 0x29045442, 0x29145443, 0x29245444, 0x29344F45, 0x29444B46, 0x29544847, 0x29645480, 0x29745481, + 0x29845482, 0x29945483, 0x29A45484, 0x29B44F85, 0x29C44B86, 0x29D44887, 0x29E454C0, 0x29F454C1, + 0x2A0453C2, 0x2A1450C3, 0x2A244EC4, 0x2A344AC5, 0x2A4447C6, 0x2A5445C7, 0x2A6C5400, 0x2A7C5401, + 0x2A8C4502, 0x2A9C5440, 0x2AAC4C41, 0x2ABC5480, 0x2ACC4C81, 0x2ADC53C0, 0x2AEC48C1, 0x2AFC5410, + 0x2B0C5411, 0x2B1C4512, 0x2B2C5450, 0x2B3C4C51, 0x2B4C5490, 0x2B5C4C91, 0x2B6C53D0, 0x2B7C48D1, + 0x2B8C5420, 0x2B9C5421, 0x2BAC4522, 0x2BBC5460, 0x2BCC4C61, 0x2BDC54A0, 0x2BEC4CA1, 0x2BFC53E0, + 0x2C0C48E1, 0x2C1C54B0, 0x2C2C4CB1, 0x2C3C53F0, 0x2C4C48F1, 0x2C529400, 0x2C629401, 0x2C729402, + 0x2C829403, 0x2C929404, 0x2CA29405, 0x2CB29306, 0x2CC28E07, 0x2CD28708, 0x2CE29440, 0x2CF29441, + 0x2D029442, 0x2D129443, 0x2D229444, 0x2D328F45, 0x2D428B46, 0x2D528847, 0x2D629480, 0x2D729481, + 0x2D829482, 0x2D929483, 0x2DA29484, 0x2DB28F85, 0x2DC28B86, 0x2DD28887, 0x2DE294C0, 0x2DF294C1, + 0x2E0293C2, 0x2E1290C3, 0x2E228EC4, 0x2E328AC5, 0x2E4287C6, 0x2E5285C7, 0x2E6A9400, 0x2E7A9401, + 0x2E8A8502, 0x2E9A9440, 0x2EAA8C41, 0x2EBA9480, 0x2ECA8C81, 0x2EDA93C0, 0x2EEA88C1, 0x2EFA9410, + 0x2F0A9411, 0x2F1A8512, 0x2F2A9450, 0x2F3A8C51, 0x2F4A9490, 0x2F5A8C91, 0x2F6A93D0, 0x2F7A88D1, + 0x2F8A9420, 0x2F9A9421, 0x2FAA8522, 0x2FBA9460, 0x2FCA8C61, 0x2FDA94A0, 0x2FEA8CA1, 0x2FFA93E0, + 0x300A88E1, 0x301A94B0, 0x302A8CB1, 0x303A93F0, 0x304A88F1, 0x30535401, 0x30635402, 0x30735403, + 0x30835404, 0x30935405, 0x30A35406, 0x30B35407, 0x30C35308, 0x30D34E09, 0x30E34A0A, 0x30F35441, + 0x31035442, 0x31135443, 0x31235444, 0x31335445, 0x31435246, 0x31534F47, 0x31634B48, 0x31734849, + 0x3183454A, 0x31935481, 0x31A35482, 0x31B35483, 0x31C35484, 0x31D35485, 0x31E35286, 0x31F34F87, + 0x32034B88, 0x32134889, 0x3223458A, 0x323354C1, 0x324354C2, 0x325354C3, 0x326352C4, 0x32734FC5, + 0x32834CC6, 0x32934AC7, 0x32A347C8, 0x32B345C9, 0x32CB5400, 0x32DB5401, 0x32EB5102, 0x32FB4703, + 0x330B5440, 0x331B5241, 0x332B4A42, 0x333B5480, 0x334B5281, 0x335B4A82, 0x336B54C0, 0x337B4DC1, + 0x338B47C2, 0x339B5410, 0x33AB5411, 0x33BB5112, 0x33CB4713, 0x33DB5450, 0x33EB5251, 0x33FB4A52, + 0x340B5490, 0x341B5291, 0x342B4A92, 0x343B54D0, 0x344B4DD1, 0x345B47D2, 0x346B5420, 0x347B5421, + 0x348B5122, 0x349B4723, 0x34AB5460, 0x34BB5261, 0x34CB4A62, 0x34DB54A0, 0x34EB52A1, 0x34FB4AA2, + 0x350B54E0, 0x351B4DE1, 0x352B47E2, 0x353B54B0, 0x354B52B1, 0x355B4AB2, 0x356B54F0, 0x357B4DF1, + 0x358B47F2, 0x35927401, 0x35A27402, 0x35B27403, 0x35C27404, 0x35D27405, 0x35E27406, 0x35F27407, + 0x36027308, 0x36126E09, 0x36226A0A, 0x36327441, 0x36427442, 0x36527443, 0x36627444, 0x36727445, + 0x36827246, 0x36926F47, 0x36A26B48, 0x36B26849, 0x36C2654A, 0x36D27481, 0x36E27482, 0x36F27483, + 0x37027484, 0x37127485, 0x37227286, 0x37326F87, 0x37426B88, 0x37526889, 0x3762658A, 0x377274C1, + 0x378274C2, 0x379274C3, 0x37A272C4, 0x37B26FC5, 0x37C26CC6, 0x37D26AC7, 0x37E267C8, 0x37F265C9, + 0x380A7400, 0x381A7401, 0x382A7102, 0x383A6703, 0x384A7440, 0x385A7241, 0x386A6A42, 0x387A7480, + 0x388A7281, 0x389A6A82, 0x38AA74C0, 0x38BA6DC1, 0x38CA67C2, 0x38DA7410, 0x38EA7411, 0x38FA7112, + 0x390A6713, 0x391A7450, 0x392A7251, 0x393A6A52, 0x394A7490, 0x395A7291, 0x396A6A92, 0x397A74D0, + 0x398A6DD1, 0x399A67D2, 0x39AA7420, 0x39BA7421, 0x39CA7122, 0x39DA6723, 0x39EA7460, 0x39FA7261, + 0x3A0A6A62, 0x3A1A74A0, 0x3A2A72A1, 0x3A3A6AA2, 0x3A4A74E0, 0x3A5A6DE1, 0x3A6A67E2, 0x3A7A74B0, + 0x3A8A72B1, 0x3A9A6AB2, 0x3AAA74F0, 0x3ABA6DF1, 0x3ACA67F2, 0x3AD25401, 0x3AE25402, 0x3AF25403, + 0x3B025404, 0x3B125405, 0x3B225406, 0x3B325407, 0x3B425408, 0x3B525409, 0x3B62540A, 0x3B72530B, + 0x3B825441, 0x3B925442, 0x3BA25443, 0x3BB25444, 0x3BC25445, 0x3BD25446, 0x3BE25447, 0x3BF25348, + 0x3C025049, 0x3C124E4A, 0x3C224B4B, 0x3C325481, 0x3C425482, 0x3C525483, 0x3C625484, 0x3C725485, + 0x3C825486, 0x3C925487, 0x3CA25388, 0x3CB25089, 0x3CC24E8A, 0x3CD24B8B, 0x3CE254C1, 0x3CF254C2, + 0x3D0254C3, 0x3D1254C4, 0x3D2253C5, 0x3D3251C6, 0x3D4250C7, 0x3D524DC8, 0x3D624BC9, 0x3D724ACA, + 0x3D8247CB, 0x3D9A5400, 0x3DAA5401, 0x3DBA5402, 0x3DCA5403, 0x3DDA4E04, 0x3DEA4505, 0x3DFA5440, + 0x3E0A5441, 0x3E1A5242, 0x3E2A4D43, 0x3E3A4844, 0x3E4A5480, 0x3E5A5481, 0x3E6A5282, 0x3E7A4D83, + 0x3E8A4884, 0x3E9A54C0, 0x3EAA51C1, 0x3EBA4DC2, 0x3ECA48C3, 0x3EDA45C4, 0x3EEA5410, 0x3EFA5411, + 0x3F0A5412, 0x3F1A5413, 0x3F2A4E14, 0x3F3A4515, 0x3F4A5450, 0x3F5A5451, 0x3F6A5252, 0x3F7A4D53, + 0x3F8A4854, 0x3F9A5490, 0x3FAA5491, 0x3FBA5292, 0x3FCA4D93, 0x3FDA4894, 0x3FEA54D0, 0x3FFA51D1, + 0x400A4DD2, 0x401A48D3, 0x402A45D4, 0x403A5420, 0x404A5421, 0x405A5422, 0x406A5423, 0x407A4E24, + 0x408A4525, 0x409A5460, 0x40AA5461, 0x40BA5262, 0x40CA4D63, 0x40DA4864, 0x40EA54A0, 0x40FA54A1, + 0x410A52A2, 0x411A4DA3, 0x412A48A4, 0x413A54E0, 0x414A51E1, 0x415A4DE2, 0x416A48E3, 0x417A45E4, + 0x418A54B0, 0x419A54B1, 0x41AA52B2, 0x41BA4DB3, 0x41CA48B4, 0x41DA54F0, 0x41EA51F1, 0x41FA4DF2, + 0x420A48F3, 0x421A45F4, 0x42243401, 0x42343402, 0x42443403, 0x42543404, 0x42643405, 0x42743406, + 0x42843407, 0x42943408, 0x42A43409, 0x42B4310A, 0x42C42B0B, 0x42D43441, 0x42E43442, 0x42F43443, + 0x43043444, 0x43143445, 0x43243446, 0x43343347, 0x43442F48, 0x43542C49, 0x43642A4A, 0x4374264B, + 0x43843481, 0x43943482, 0x43A43483, 0x43B43484, 0x43C43485, 0x43D43486, 0x43E43387, 0x43F42F88, + 0x44042C89, 0x44142A8A, 0x4424268B, 0x443434C1, 0x444434C2, 0x445434C3, 0x446434C4, 0x447431C5, + 0x44842FC6, 0x44942DC7, 0x44A42AC8, 0x44B428C9, 0x44C426CA, 0x44D424CB, 0x44EC3400, 0x44FC3401, + 0x450C3402, 0x451C2E03, 0x452C2704, 0x453C3440, 0x454C3441, 0x455C2E42, 0x456C2843, 0x457C3480, + 0x458C3481, 0x459C2E82, 0x45AC2883, 0x45BC34C0, 0x45CC2FC1, 0x45DC2AC2, 0x45EC25C3, 0x45FC3410, + 0x460C3411, 0x461C3412, 0x462C2E13, 0x463C2714, 0x464C3450, 0x465C3451, 0x466C2E52, 0x467C2853, + 0x468C3490, 0x469C3491, 0x46AC2E92, 0x46BC2893, 0x46CC34D0, 0x46DC2FD1, 0x46EC2AD2, 0x46FC25D3, + 0x470C3420, 0x471C3421, 0x472C3422, 0x473C2E23, 0x474C2724, 0x475C3460, 0x476C3461, 0x477C2E62, + 0x478C2863, 0x479C34A0, 0x47AC34A1, 0x47BC2EA2, 0x47CC28A3, 0x47DC34E0, 0x47EC2FE1, 0x47FC2AE2, + 0x480C25E3, 0x481C34B0, 0x482C34B1, 0x483C2EB2, 0x484C28B3, 0x485C34F0, 0x486C2FF1, 0x487C2AF2, + 0x488C25F3, 0x48919401, 0x48A19402, 0x48B19403, 0x48C19404, 0x48D19405, 0x48E19406, 0x48F19407, + 0x49019408, 0x49119409, 0x4921910A, 0x49318B0B, 0x49419441, 0x49519442, 0x49619443, 0x49719444, + 0x49819445, 0x49919446, 0x49A19347, 0x49B18F48, 0x49C18C49, 0x49D18A4A, 0x49E1864B, 0x49F19481, + 0x4A019482, 0x4A119483, 0x4A219484, 0x4A319485, 0x4A419486, 0x4A519387, 0x4A618F88, 0x4A718C89, + 0x4A818A8A, 0x4A91868B, 0x4AA194C1, 0x4AB194C2, 0x4AC194C3, 0x4AD194C4, 0x4AE191C5, 0x4AF18FC6, + 0x4B018DC7, 0x4B118AC8, 0x4B2188C9, 0x4B3186CA, 0x4B4184CB, 0x4B599400, 0x4B699401, 0x4B799402, + 0x4B898E03, 0x4B998704, 0x4BA99440, 0x4BB99441, 0x4BC98E42, 0x4BD98843, 0x4BE99480, 0x4BF99481, + 0x4C098E82, 0x4C198883, 0x4C2994C0, 0x4C398FC1, 0x4C498AC2, 0x4C5985C3, 0x4C699410, 0x4C799411, + 0x4C899412, 0x4C998E13, 0x4CA98714, 0x4CB99450, 0x4CC99451, 0x4CD98E52, 0x4CE98853, 0x4CF99490, + 0x4D099491, 0x4D198E92, 0x4D298893, 0x4D3994D0, 0x4D498FD1, 0x4D598AD2, 0x4D6985D3, 0x4D799420, + 0x4D899421, 0x4D999422, 0x4DA98E23, 0x4DB98724, 0x4DC99460, 0x4DD99461, 0x4DE98E62, 0x4DF98863, + 0x4E0994A0, 0x4E1994A1, 0x4E298EA2, 0x4E3988A3, 0x4E4994E0, 0x4E598FE1, 0x4E698AE2, 0x4E7985E3, + 0x4E8994B0, 0x4E9994B1, 0x4EA98EB2, 0x4EB988B3, 0x4EC994F0, 0x4ED98FF1, 0x4EE98AF2, 0x4EF985F3, + 0x4F033401, 0x4F133402, 0x4F233403, 0x4F333404, 0x4F433405, 0x4F533406, 0x4F633407, 0x4F733408, + 0x4F833409, 0x4F93340A, 0x4FA3340B, 0x4FB33441, 0x4FC33442, 0x4FD33443, 0x4FE33444, 0x4FF33445, + 0x50033446, 0x50133447, 0x50233448, 0x50333349, 0x5043314A, 0x50532E4B, 0x50633481, 0x50733482, + 0x50833483, 0x50933484, 0x50A33485, 0x50B33486, 0x50C33487, 0x50D33488, 0x50E33389, 0x50F3318A, + 0x51032E8B, 0x511334C1, 0x512334C2, 0x513334C3, 0x514334C4, 0x515334C5, 0x516333C6, 0x517331C7, + 0x51832FC8, 0x51932DC9, 0x51A32BCA, 0x51B329CB, 0x51CB3400, 0x51DB3401, 0x51EB3402, 0x51FB3403, + 0x520B3304, 0x521B2A05, 0x522B3440, 0x523B3441, 0x524B3442, 0x525B2F43, 0x526B2B44, 0x527B2545, + 0x528B3480, 0x529B3481, 0x52AB3482, 0x52BB2F83, 0x52CB2B84, 0x52DB2585, 0x52EB34C0, 0x52FB33C1, + 0x530B2EC2, 0x531B2AC3, 0x532B27C4, 0x533B3410, 0x534B3411, 0x535B3412, 0x536B3413, 0x537B3314, + 0x538B2A15, 0x539B3450, 0x53AB3451, 0x53BB3452, 0x53CB2F53, 0x53DB2B54, 0x53EB2555, 0x53FB3490, + 0x540B3491, 0x541B3492, 0x542B2F93, 0x543B2B94, 0x544B2595, 0x545B34D0, 0x546B33D1, 0x547B2ED2, + 0x548B2AD3, 0x549B27D4, 0x54AB3420, 0x54BB3421, 0x54CB3422, 0x54DB3423, 0x54EB3324, 0x54FB2A25, + 0x550B3460, 0x551B3461, 0x552B3462, 0x553B2F63, 0x554B2B64, 0x555B2565, 0x556B34A0, 0x557B34A1, + 0x558B34A2, 0x559B2FA3, 0x55AB2BA4, 0x55BB25A5, 0x55CB34E0, 0x55DB33E1, 0x55EB2EE2, 0x55FB2AE3, + 0x560B27E4, 0x561B34B0, 0x562B34B1, 0x563B34B2, 0x564B2FB3, 0x565B2BB4, 0x566B25B5, 0x567B34F0, + 0x568B33F1, 0x569B2EF2, 0x56AB2AF3, 0x56BB27F4, 0x56C17401, 0x56D17402, 0x56E17403, 0x56F17404, + 0x57017405, 0x57117406, 0x57217407, 0x57317408, 0x57417409, 0x5751740A, 0x5761740B, 0x57717441, + 0x57817442, 0x57917443, 0x57A17444, 0x57B17445, 0x57C17446, 0x57D17447, 0x57E17448, 0x57F17349, + 0x5801714A, 0x58116E4B, 0x58217481, 0x58317482, 0x58417483, 0x58517484, 0x58617485, 0x58717486, + 0x58817487, 0x58917488, 0x58A17389, 0x58B1718A, 0x58C16E8B, 0x58D174C1, 0x58E174C2, 0x58F174C3, + 0x590174C4, 0x591174C5, 0x592173C6, 0x593171C7, 0x59416FC8, 0x59516DC9, 0x59616BCA, 0x597169CB, + 0x59897400, 0x59997401, 0x59A97402, 0x59B97403, 0x59C97304, 0x59D96A05, 0x59E97440, 0x59F97441, + 0x5A097442, 0x5A196F43, 0x5A296B44, 0x5A396545, 0x5A497480, 0x5A597481, 0x5A697482, 0x5A796F83, + 0x5A896B84, 0x5A996585, 0x5AA974C0, 0x5AB973C1, 0x5AC96EC2, 0x5AD96AC3, 0x5AE967C4, 0x5AF97410, + 0x5B097411, 0x5B197412, 0x5B297413, 0x5B397314, 0x5B496A15, 0x5B597450, 0x5B697451, 0x5B797452, + 0x5B896F53, 0x5B996B54, 0x5BA96555, 0x5BB97490, 0x5BC97491, 0x5BD97492, 0x5BE96F93, 0x5BF96B94, + 0x5C096595, 0x5C1974D0, 0x5C2973D1, 0x5C396ED2, 0x5C496AD3, 0x5C5967D4, 0x5C697420, 0x5C797421, + 0x5C897422, 0x5C997423, 0x5CA97324, 0x5CB96A25, 0x5CC97460, 0x5CD97461, 0x5CE97462, 0x5CF96F63, + 0x5D096B64, 0x5D196565, 0x5D2974A0, 0x5D3974A1, 0x5D4974A2, 0x5D596FA3, 0x5D696BA4, 0x5D7965A5, + 0x5D8974E0, 0x5D9973E1, 0x5DA96EE2, 0x5DB96AE3, 0x5DC967E4, 0x5DD974B0, 0x5DE974B1, 0x5DF974B2, + 0x5E096FB3, 0x5E196BB4, 0x5E2965B5, 0x5E3974F0, 0x5E4973F1, 0x5E596EF2, 0x5E696AF3, 0x5E7967F4, + 0x5E823402, 0x5E923403, 0x5EA23404, 0x5EB23405, 0x5EC23406, 0x5ED23407, 0x5EE23408, 0x5EF23409, + 0x5F02340A, 0x5F12340B, 0x5F223442, 0x5F323443, 0x5F423444, 0x5F523445, 0x5F623446, 0x5F723447, + 0x5F823448, 0x5F923449, 0x5FA2344A, 0x5FB2344B, 0x5FC23482, 0x5FD23483, 0x5FE23484, 0x5FF23485, + 0x60023486, 0x60123487, 0x60223488, 0x60323489, 0x6042348A, 0x6052348B, 0x606234C2, 0x607234C3, + 0x608234C4, 0x609234C5, 0x60A234C6, 0x60B234C7, 0x60C233C8, 0x60D232C9, 0x60E230CA, 0x60F22FCB, + 0x610A3400, 0x611A3401, 0x612A3402, 0x613A3403, 0x614A3404, 0x615A3405, 0x616A3106, 0x617A2C07, + 0x618A2508, 0x619A3440, 0x61AA3441, 0x61BA3442, 0x61CA3443, 0x61DA3344, 0x61EA2E45, 0x61FA2A46, + 0x620A2747, 0x621A3480, 0x622A3481, 0x623A3482, 0x624A3483, 0x625A3384, 0x626A2E85, 0x627A2A86, + 0x628A2787, 0x629A34C0, 0x62AA34C1, 0x62BA33C2, 0x62CA30C3, 0x62DA2DC4, 0x62EA2AC5, 0x62FA27C6, + 0x630A24C7, 0x631A3410, 0x632A3411, 0x633A3412, 0x634A3413, 0x635A3414, 0x636A3415, 0x637A3116, + 0x638A2C17, 0x639A2518, 0x63AA3450, 0x63BA3451, 0x63CA3452, 0x63DA3453, 0x63EA3354, 0x63FA2E55, + 0x640A2A56, 0x641A2757, 0x642A3490, 0x643A3491, 0x644A3492, 0x645A3493, 0x646A3394, 0x647A2E95, + 0x648A2A96, 0x649A2797, 0x64AA34D0, 0x64BA34D1, 0x64CA33D2, 0x64DA30D3, 0x64EA2DD4, 0x64FA2AD5, + 0x650A27D6, 0x651A24D7, 0x652A3420, 0x653A3421, 0x654A3422, 0x655A3423, 0x656A3424, 0x657A3425, + 0x658A3126, 0x659A2C27, 0x65AA2528, 0x65BA3460, 0x65CA3461, 0x65DA3462, 0x65EA3463, 0x65FA3364, + 0x660A2E65, 0x661A2A66, 0x662A2767, 0x663A34A0, 0x664A34A1, 0x665A34A2, 0x666A34A3, 0x667A33A4, + 0x668A2EA5, 0x669A2AA6, 0x66AA27A7, 0x66BA34E0, 0x66CA34E1, 0x66DA33E2, 0x66EA30E3, 0x66FA2DE4, + 0x670A2AE5, 0x671A27E6, 0x672A24E7, 0x673A34B0, 0x674A34B1, 0x675A34B2, 0x676A34B3, 0x677A33B4, + 0x678A2EB5, 0x679A2AB6, 0x67AA27B7, 0x67BA34F0, 0x67CA34F1, 0x67DA33F2, 0x67EA30F3, 0x67FA2DF4, + 0x680A2AF5, 0x681A27F6, 0x682A24F7, 0x68315402, 0x68415403, 0x68515404, 0x68615405, 0x68715406, + 0x68815407, 0x68915408, 0x68A15409, 0x68B1540A, 0x68C1540B, 0x68D15442, 0x68E15443, 0x68F15444, + 0x69015445, 0x69115446, 0x69215447, 0x69315448, 0x69415449, 0x6951544A, 0x6961544B, 0x69715482, + 0x69815483, 0x69915484, 0x69A15485, 0x69B15486, 0x69C15487, 0x69D15488, 0x69E15489, 0x69F1548A, + 0x6A01548B, 0x6A1154C2, 0x6A2154C3, 0x6A3154C4, 0x6A4154C5, 0x6A5154C6, 0x6A6154C7, 0x6A7153C8, + 0x6A8152C9, 0x6A9150CA, 0x6AA14FCB, 0x6AB95400, 0x6AC95401, 0x6AD95402, 0x6AE95403, 0x6AF95404, + 0x6B095405, 0x6B195106, 0x6B294C07, 0x6B394508, 0x6B495440, 0x6B595441, 0x6B695442, 0x6B795443, + 0x6B895344, 0x6B994E45, 0x6BA94A46, 0x6BB94747, 0x6BC95480, 0x6BD95481, 0x6BE95482, 0x6BF95483, + 0x6C095384, 0x6C194E85, 0x6C294A86, 0x6C394787, 0x6C4954C0, 0x6C5954C1, 0x6C6953C2, 0x6C7950C3, + 0x6C894DC4, 0x6C994AC5, 0x6CA947C6, 0x6CB944C7, 0x6CC95410, 0x6CD95411, 0x6CE95412, 0x6CF95413, + 0x6D095414, 0x6D195415, 0x6D295116, 0x6D394C17, 0x6D494518, 0x6D595450, 0x6D695451, 0x6D795452, + 0x6D895453, 0x6D995354, 0x6DA94E55, 0x6DB94A56, 0x6DC94757, 0x6DD95490, 0x6DE95491, 0x6DF95492, + 0x6E095493, 0x6E195394, 0x6E294E95, 0x6E394A96, 0x6E494797, 0x6E5954D0, 0x6E6954D1, 0x6E7953D2, + 0x6E8950D3, 0x6E994DD4, 0x6EA94AD5, 0x6EB947D6, 0x6EC944D7, 0x6ED95420, 0x6EE95421, 0x6EF95422, + 0x6F095423, 0x6F195424, 0x6F295425, 0x6F395126, 0x6F494C27, 0x6F594528, 0x6F695460, 0x6F795461, + 0x6F895462, 0x6F995463, 0x6FA95364, 0x6FB94E65, 0x6FC94A66, 0x6FD94767, 0x6FE954A0, 0x6FF954A1, + 0x700954A2, 0x701954A3, 0x702953A4, 0x70394EA5, 0x70494AA6, 0x705947A7, 0x706954E0, 0x707954E1, + 0x708953E2, 0x709950E3, 0x70A94DE4, 0x70B94AE5, 0x70C947E6, 0x70D944E7, 0x70E954B0, 0x70F954B1, + 0x710954B2, 0x711954B3, 0x712953B4, 0x71394EB5, 0x71494AB6, 0x715947B7, 0x716954F0, 0x717954F1, + 0x718953F2, 0x719950F3, 0x71A94DF4, 0x71B94AF5, 0x71C947F6, 0x71D944F7, 0x71E13404, 0x71F13405, + 0x72013406, 0x72113407, 0x72213408, 0x72313409, 0x7241340A, 0x7251340B, 0x72613444, 0x72713445, + 0x72813446, 0x72913447, 0x72A13448, 0x72B13449, 0x72C1344A, 0x72D1344B, 0x72E13484, 0x72F13485, + 0x73013486, 0x73113487, 0x73213488, 0x73313489, 0x7341348A, 0x7351348B, 0x736134C4, 0x737134C5, + 0x738134C6, 0x739134C7, 0x73A134C8, 0x73B134C9, 0x73C134CA, 0x73D134CB, 0x73E93401, 0x73F93402, + 0x74093403, 0x74193404, 0x74293405, 0x74393406, 0x74493407, 0x74593408, 0x74693309, 0x74792F0A, + 0x74892A0B, 0x74993441, 0x74A93442, 0x74B93443, 0x74C93444, 0x74D93445, 0x74E93446, 0x74F93247, + 0x75092E48, 0x75192B49, 0x7529294A, 0x7539254B, 0x75493481, 0x75593482, 0x75693483, 0x75793484, + 0x75893485, 0x75993486, 0x75A93287, 0x75B92E88, 0x75C92B89, 0x75D9298A, 0x75E9258B, 0x75F934C1, + 0x760934C2, 0x761934C3, 0x762933C4, 0x763930C5, 0x76492EC6, 0x76592CC7, 0x76692AC8, 0x767927C9, + 0x768925CA, 0x76993411, 0x76A93412, 0x76B93413, 0x76C93414, 0x76D93415, 0x76E93416, 0x76F93417, + 0x77093418, 0x77193319, 0x77292F1A, 0x77392A1B, 0x77493451, 0x77593452, 0x77693453, 0x77793454, + 0x77893455, 0x77993456, 0x77A93257, 0x77B92E58, 0x77C92B59, 0x77D9295A, 0x77E9255B, 0x77F93491, + 0x78093492, 0x78193493, 0x78293494, 0x78393495, 0x78493496, 0x78593297, 0x78692E98, 0x78792B99, + 0x7889299A, 0x7899259B, 0x78A934D1, 0x78B934D2, 0x78C934D3, 0x78D933D4, 0x78E930D5, 0x78F92ED6, + 0x79092CD7, 0x79192AD8, 0x792927D9, 0x793925DA, 0x79493421, 0x79593422, 0x79693423, 0x79793424, + 0x79893425, 0x79993426, 0x79A93427, 0x79B93428, 0x79C93329, 0x79D92F2A, 0x79E92A2B, 0x79F93461, + 0x7A093462, 0x7A193463, 0x7A293464, 0x7A393465, 0x7A493466, 0x7A593267, 0x7A692E68, 0x7A792B69, + 0x7A89296A, 0x7A99256B, 0x7AA934A1, 0x7AB934A2, 0x7AC934A3, 0x7AD934A4, 0x7AE934A5, 0x7AF934A6, + 0x7B0932A7, 0x7B192EA8, 0x7B292BA9, 0x7B3929AA, 0x7B4925AB, 0x7B5934E1, 0x7B6934E2, 0x7B7934E3, + 0x7B8933E4, 0x7B9930E5, 0x7BA92EE6, 0x7BB92CE7, 0x7BC92AE8, 0x7BD927E9, 0x7BE925EA, 0x7BF934B1, + 0x7C0934B2, 0x7C1934B3, 0x7C2934B4, 0x7C3934B5, 0x7C4934B6, 0x7C5932B7, 0x7C692EB8, 0x7C792BB9, + 0x7C8929BA, 0x7C9925BB, 0x7CA934F1, 0x7CB934F2, 0x7CC934F3, 0x7CD933F4, 0x7CE930F5, 0x7CF92EF6, + 0x7D092CF7, 0x7D192AF8, 0x7D2927F9, 0x7D3925FA, 0x7D441402, 0x7D541403, 0x7D641404, 0x7D741405, + 0x7D841406, 0x7D941407, 0x7DA41408, 0x7DB41409, 0x7DC4140A, 0x7DD4140B, 0x7DE41442, 0x7DF41443, + 0x7E041444, 0x7E141445, 0x7E241446, 0x7E341447, 0x7E441448, 0x7E541449, 0x7E64144A, 0x7E74144B, + 0x7E841482, 0x7E941483, 0x7EA41484, 0x7EB41485, 0x7EC41486, 0x7ED41487, 0x7EE41488, 0x7EF41489, + 0x7F04148A, 0x7F14148B, 0x7F2414C2, 0x7F3414C3, 0x7F4414C4, 0x7F5414C5, 0x7F6414C6, 0x7F7414C7, + 0x7F8413C8, 0x7F9412C9, 0x7FA410CA, 0x7FB40FCB, 0x7FCC1400, 0x7FDC1401, 0x7FEC1402, 0x7FFC1403, + 0x800C1404, 0x801C1405, 0x802C1106, 0x803C0C07, 0x804C0508, 0x805C1440, 0x806C1441, 0x807C1442, + 0x808C1443, 0x809C1344, 0x80AC0E45, 0x80BC0A46, 0x80CC0747, 0x80DC1480, 0x80EC1481, 0x80FC1482, + 0x810C1483, 0x811C1384, 0x812C0E85, 0x813C0A86, 0x814C0787, 0x815C14C0, 0x816C14C1, 0x817C13C2, + 0x818C10C3, 0x819C0DC4, 0x81AC0AC5, 0x81BC07C6, 0x81CC04C7, 0x81DC1410, 0x81EC1411, 0x81FC1412, + 0x820C1413, 0x821C1414, 0x822C1415, 0x823C1116, 0x824C0C17, 0x825C0518, 0x826C1450, 0x827C1451, + 0x828C1452, 0x829C1453, 0x82AC1354, 0x82BC0E55, 0x82CC0A56, 0x82DC0757, 0x82EC1490, 0x82FC1491, + 0x830C1492, 0x831C1493, 0x832C1394, 0x833C0E95, 0x834C0A96, 0x835C0797, 0x836C14D0, 0x837C14D1, + 0x838C13D2, 0x839C10D3, 0x83AC0DD4, 0x83BC0AD5, 0x83CC07D6, 0x83DC04D7, 0x83EC1420, 0x83FC1421, + 0x840C1422, 0x841C1423, 0x842C1424, 0x843C1425, 0x844C1126, 0x845C0C27, 0x846C0528, 0x847C1460, + 0x848C1461, 0x849C1462, 0x84AC1463, 0x84BC1364, 0x84CC0E65, 0x84DC0A66, 0x84EC0767, 0x84FC14A0, + 0x850C14A1, 0x851C14A2, 0x852C14A3, 0x853C13A4, 0x854C0EA5, 0x855C0AA6, 0x856C07A7, 0x857C14E0, + 0x858C14E1, 0x859C13E2, 0x85AC10E3, 0x85BC0DE4, 0x85CC0AE5, 0x85DC07E6, 0x85EC04E7, 0x85FC14B0, + 0x860C14B1, 0x861C14B2, 0x862C14B3, 0x863C13B4, 0x864C0EB5, 0x865C0AB6, 0x866C07B7, 0x867C14F0, + 0x868C14F1, 0x869C13F2, 0x86AC10F3, 0x86BC0DF4, 0x86CC0AF5, 0x86DC07F6, 0x86EC04F7, 0x86F09402, + 0x87009403, 0x87109404, 0x87209405, 0x87309406, 0x87409407, 0x87509408, 0x87609409, 0x8770940A, + 0x8780940B, 0x87909442, 0x87A09443, 0x87B09444, 0x87C09445, 0x87D09446, 0x87E09447, 0x87F09448, + 0x88009449, 0x8810944A, 0x8820944B, 0x88309482, 0x88409483, 0x88509484, 0x88609485, 0x88709486, + 0x88809487, 0x88909488, 0x88A09489, 0x88B0948A, 0x88C0948B, 0x88D094C2, 0x88E094C3, 0x88F094C4, + 0x890094C5, 0x891094C6, 0x892094C7, 0x893093C8, 0x894092C9, 0x895090CA, 0x89608FCB, 0x89789400, + 0x89889401, 0x89989402, 0x89A89403, 0x89B89404, 0x89C89405, 0x89D89106, 0x89E88C07, 0x89F88508, + 0x8A089440, 0x8A189441, 0x8A289442, 0x8A389443, 0x8A489344, 0x8A588E45, 0x8A688A46, 0x8A788747, + 0x8A889480, 0x8A989481, 0x8AA89482, 0x8AB89483, 0x8AC89384, 0x8AD88E85, 0x8AE88A86, 0x8AF88787, + 0x8B0894C0, 0x8B1894C1, 0x8B2893C2, 0x8B3890C3, 0x8B488DC4, 0x8B588AC5, 0x8B6887C6, 0x8B7884C7, + 0x8B889410, 0x8B989411, 0x8BA89412, 0x8BB89413, 0x8BC89414, 0x8BD89415, 0x8BE89116, 0x8BF88C17, + 0x8C088518, 0x8C189450, 0x8C289451, 0x8C389452, 0x8C489453, 0x8C589354, 0x8C688E55, 0x8C788A56, + 0x8C888757, 0x8C989490, 0x8CA89491, 0x8CB89492, 0x8CC89493, 0x8CD89394, 0x8CE88E95, 0x8CF88A96, + 0x8D088797, 0x8D1894D0, 0x8D2894D1, 0x8D3893D2, 0x8D4890D3, 0x8D588DD4, 0x8D688AD5, 0x8D7887D6, + 0x8D8884D7, 0x8D989420, 0x8DA89421, 0x8DB89422, 0x8DC89423, 0x8DD89424, 0x8DE89425, 0x8DF89126, + 0x8E088C27, 0x8E188528, 0x8E289460, 0x8E389461, 0x8E489462, 0x8E589463, 0x8E689364, 0x8E788E65, + 0x8E888A66, 0x8E988767, 0x8EA894A0, 0x8EB894A1, 0x8EC894A2, 0x8ED894A3, 0x8EE893A4, 0x8EF88EA5, + 0x8F088AA6, 0x8F1887A7, 0x8F2894E0, 0x8F3894E1, 0x8F4893E2, 0x8F5890E3, 0x8F688DE4, 0x8F788AE5, + 0x8F8887E6, 0x8F9884E7, 0x8FA894B0, 0x8FB894B1, 0x8FC894B2, 0x8FD894B3, 0x8FE893B4, 0x8FF88EB5, + 0x90088AB6, 0x901887B7, 0x902894F0, 0x903894F1, 0x904893F2, 0x905890F3, 0x90688DF4, 0x90788AF5, + 0x908887F6, 0x909884F7, 0x90A31403, 0x90B31404, 0x90C31405, 0x90D31406, 0x90E31407, 0x90F31408, + 0x91031409, 0x9113140A, 0x9123140B, 0x91331443, 0x91431444, 0x91531445, 0x91631446, 0x91731447, + 0x91831448, 0x91931449, 0x91A3144A, 0x91B3144B, 0x91C31483, 0x91D31484, 0x91E31485, 0x91F31486, + 0x92031487, 0x92131488, 0x92231489, 0x9233148A, 0x9243148B, 0x925314C3, 0x926314C4, 0x927314C5, + 0x928314C6, 0x929314C7, 0x92A314C8, 0x92B314C9, 0x92C314CA, 0x92D313CB, 0x92EB1401, 0x92FB1402, + 0x930B1403, 0x931B1404, 0x932B1405, 0x933B1406, 0x934B1407, 0x935B1108, 0x936B0C09, 0x937B080A, + 0x938B1441, 0x939B1442, 0x93AB1443, 0x93BB1444, 0x93CB1445, 0x93DB1146, 0x93EB0E47, 0x93FB0A48, + 0x940B0749, 0x941B044A, 0x942B1481, 0x943B1482, 0x944B1483, 0x945B1484, 0x946B1485, 0x947B1186, + 0x948B0E87, 0x949B0A88, 0x94AB0789, 0x94BB048A, 0x94CB14C1, 0x94DB14C2, 0x94EB13C3, 0x94FB11C4, + 0x950B0EC5, 0x951B0BC6, 0x952B0AC7, 0x953B07C8, 0x954B04C9, 0x955B1411, 0x956B1412, 0x957B1413, + 0x958B1414, 0x959B1415, 0x95AB1416, 0x95BB1417, 0x95CB1118, 0x95DB0C19, 0x95EB081A, 0x95FB1451, + 0x960B1452, 0x961B1453, 0x962B1454, 0x963B1455, 0x964B1156, 0x965B0E57, 0x966B0A58, 0x967B0759, + 0x968B045A, 0x969B1491, 0x96AB1492, 0x96BB1493, 0x96CB1494, 0x96DB1495, 0x96EB1196, 0x96FB0E97, + 0x970B0A98, 0x971B0799, 0x972B049A, 0x973B14D1, 0x974B14D2, 0x975B13D3, 0x976B11D4, 0x977B0ED5, + 0x978B0BD6, 0x979B0AD7, 0x97AB07D8, 0x97BB04D9, 0x97CB1421, 0x97DB1422, 0x97EB1423, 0x97FB1424, + 0x980B1425, 0x981B1426, 0x982B1427, 0x983B1128, 0x984B0C29, 0x985B082A, 0x986B1461, 0x987B1462, + 0x988B1463, 0x989B1464, 0x98AB1465, 0x98BB1166, 0x98CB0E67, 0x98DB0A68, 0x98EB0769, 0x98FB046A, + 0x990B14A1, 0x991B14A2, 0x992B14A3, 0x993B14A4, 0x994B14A5, 0x995B11A6, 0x996B0EA7, 0x997B0AA8, + 0x998B07A9, 0x999B04AA, 0x99AB14E1, 0x99BB14E2, 0x99CB13E3, 0x99DB11E4, 0x99EB0EE5, 0x99FB0BE6, + 0x9A0B0AE7, 0x9A1B07E8, 0x9A2B04E9, 0x9A3B14B1, 0x9A4B14B2, 0x9A5B14B3, 0x9A6B14B4, 0x9A7B14B5, + 0x9A8B11B6, 0x9A9B0EB7, 0x9AAB0AB8, 0x9ABB07B9, 0x9ACB04BA, 0x9ADB14F1, 0x9AEB14F2, 0x9AFB13F3, + 0x9B0B11F4, 0x9B1B0EF5, 0x9B2B0BF6, 0x9B3B0AF7, 0x9B4B07F8, 0x9B5B04F9, 0x9B607403, 0x9B707404, + 0x9B807405, 0x9B907406, 0x9BA07407, 0x9BB07408, 0x9BC07409, 0x9BD0740A, 0x9BE0740B, 0x9BF07443, + 0x9C007444, 0x9C107445, 0x9C207446, 0x9C307447, 0x9C407448, 0x9C507449, 0x9C60744A, 0x9C70744B, + 0x9C807483, 0x9C907484, 0x9CA07485, 0x9CB07486, 0x9CC07487, 0x9CD07488, 0x9CE07489, 0x9CF0748A, + 0x9D00748B, 0x9D1074C3, 0x9D2074C4, 0x9D3074C5, 0x9D4074C6, 0x9D5074C7, 0x9D6074C8, 0x9D7074C9, + 0x9D8074CA, 0x9D9073CB, 0x9DA87401, 0x9DB87402, 0x9DC87403, 0x9DD87404, 0x9DE87405, 0x9DF87406, + 0x9E087407, 0x9E187108, 0x9E286C09, 0x9E38680A, 0x9E487441, 0x9E587442, 0x9E687443, 0x9E787444, + 0x9E887445, 0x9E987146, 0x9EA86E47, 0x9EB86A48, 0x9EC86749, 0x9ED8644A, 0x9EE87481, 0x9EF87482, + 0x9F087483, 0x9F187484, 0x9F287485, 0x9F387186, 0x9F486E87, 0x9F586A88, 0x9F686789, 0x9F78648A, + 0x9F8874C1, 0x9F9874C2, 0x9FA873C3, 0x9FB871C4, 0x9FC86EC5, 0x9FD86BC6, 0x9FE86AC7, 0x9FF867C8, + 0xA00864C9, 0xA0187411, 0xA0287412, 0xA0387413, 0xA0487414, 0xA0587415, 0xA0687416, 0xA0787417, + 0xA0887118, 0xA0986C19, 0xA0A8681A, 0xA0B87451, 0xA0C87452, 0xA0D87453, 0xA0E87454, 0xA0F87455, + 0xA1087156, 0xA1186E57, 0xA1286A58, 0xA1386759, 0xA148645A, 0xA1587491, 0xA1687492, 0xA1787493, + 0xA1887494, 0xA1987495, 0xA1A87196, 0xA1B86E97, 0xA1C86A98, 0xA1D86799, 0xA1E8649A, 0xA1F874D1, + 0xA20874D2, 0xA21873D3, 0xA22871D4, 0xA2386ED5, 0xA2486BD6, 0xA2586AD7, 0xA26867D8, 0xA27864D9, + 0xA2887421, 0xA2987422, 0xA2A87423, 0xA2B87424, 0xA2C87425, 0xA2D87426, 0xA2E87427, 0xA2F87128, + 0xA3086C29, 0xA318682A, 0xA3287461, 0xA3387462, 0xA3487463, 0xA3587464, 0xA3687465, 0xA3787166, + 0xA3886E67, 0xA3986A68, 0xA3A86769, 0xA3B8646A, 0xA3C874A1, 0xA3D874A2, 0xA3E874A3, 0xA3F874A4, + 0xA40874A5, 0xA41871A6, 0xA4286EA7, 0xA4386AA8, 0xA44867A9, 0xA45864AA, 0xA46874E1, 0xA47874E2, + 0xA48873E3, 0xA49871E4, 0xA4A86EE5, 0xA4B86BE6, 0xA4C86AE7, 0xA4D867E8, 0xA4E864E9, 0xA4F874B1, + 0xA50874B2, 0xA51874B3, 0xA52874B4, 0xA53874B5, 0xA54871B6, 0xA5586EB7, 0xA5686AB8, 0xA57867B9, + 0xA58864BA, 0xA59874F1, 0xA5A874F2, 0xA5B873F3, 0xA5C871F4, 0xA5D86EF5, 0xA5E86BF6, 0xA5F86AF7, + 0xA60867F8, 0xA61864F9, 0xA6221405, 0xA6321406, 0xA6421407, 0xA6521408, 0xA6621409, 0xA672140A, + 0xA682140B, 0xA6921445, 0xA6A21446, 0xA6B21447, 0xA6C21448, 0xA6D21449, 0xA6E2144A, 0xA6F2144B, + 0xA7021485, 0xA7121486, 0xA7221487, 0xA7321488, 0xA7421489, 0xA752148A, 0xA762148B, 0xA77214C5, + 0xA78214C6, 0xA79214C7, 0xA7A214C8, 0xA7B214C9, 0xA7C214CA, 0xA7D214CB, 0xA7EA1401, 0xA7FA1402, + 0xA80A1403, 0xA81A1404, 0xA82A1405, 0xA83A1406, 0xA84A1407, 0xA85A1408, 0xA86A1409, 0xA87A140A, + 0xA88A110B, 0xA89A1441, 0xA8AA1442, 0xA8BA1443, 0xA8CA1444, 0xA8DA1445, 0xA8EA1446, 0xA8FA1447, + 0xA90A1248, 0xA91A0F49, 0xA92A0D4A, 0xA93A0A4B, 0xA94A1481, 0xA95A1482, 0xA96A1483, 0xA97A1484, + 0xA98A1485, 0xA99A1486, 0xA9AA1487, 0xA9BA1288, 0xA9CA0F89, 0xA9DA0D8A, 0xA9EA0A8B, 0xA9FA14C1, + 0xAA0A14C2, 0xAA1A14C3, 0xAA2A14C4, 0xAA3A13C5, 0xAA4A10C6, 0xAA5A0FC7, 0xAA6A0DC8, 0xAA7A0AC9, + 0xAA8A09CA, 0xAA9A07CB, 0xAAAA1411, 0xAABA1412, 0xAACA1413, 0xAADA1414, 0xAAEA1415, 0xAAFA1416, + 0xAB0A1417, 0xAB1A1418, 0xAB2A1419, 0xAB3A141A, 0xAB4A111B, 0xAB5A1451, 0xAB6A1452, 0xAB7A1453, + 0xAB8A1454, 0xAB9A1455, 0xABAA1456, 0xABBA1457, 0xABCA1258, 0xABDA0F59, 0xABEA0D5A, 0xABFA0A5B, + 0xAC0A1491, 0xAC1A1492, 0xAC2A1493, 0xAC3A1494, 0xAC4A1495, 0xAC5A1496, 0xAC6A1497, 0xAC7A1298, + 0xAC8A0F99, 0xAC9A0D9A, 0xACAA0A9B, 0xACBA14D1, 0xACCA14D2, 0xACDA14D3, 0xACEA14D4, 0xACFA13D5, + 0xAD0A10D6, 0xAD1A0FD7, 0xAD2A0DD8, 0xAD3A0AD9, 0xAD4A09DA, 0xAD5A07DB, 0xAD6A1421, 0xAD7A1422, + 0xAD8A1423, 0xAD9A1424, 0xADAA1425, 0xADBA1426, 0xADCA1427, 0xADDA1428, 0xADEA1429, 0xADFA142A, + 0xAE0A112B, 0xAE1A1461, 0xAE2A1462, 0xAE3A1463, 0xAE4A1464, 0xAE5A1465, 0xAE6A1466, 0xAE7A1467, + 0xAE8A1268, 0xAE9A0F69, 0xAEAA0D6A, 0xAEBA0A6B, 0xAECA14A1, 0xAEDA14A2, 0xAEEA14A3, 0xAEFA14A4, + 0xAF0A14A5, 0xAF1A14A6, 0xAF2A14A7, 0xAF3A12A8, 0xAF4A0FA9, 0xAF5A0DAA, 0xAF6A0AAB, 0xAF7A14E1, + 0xAF8A14E2, 0xAF9A14E3, 0xAFAA14E4, 0xAFBA13E5, 0xAFCA10E6, 0xAFDA0FE7, 0xAFEA0DE8, 0xAFFA0AE9, + 0xB00A09EA, 0xB01A07EB, 0xB02A14B1, 0xB03A14B2, 0xB04A14B3, 0xB05A14B4, 0xB06A14B5, 0xB07A14B6, + 0xB08A14B7, 0xB09A12B8, 0xB0AA0FB9, 0xB0BA0DBA, 0xB0CA0ABB, 0xB0DA14F1, 0xB0EA14F2, 0xB0FA14F3, + 0xB10A14F4, 0xB11A13F5, 0xB12A10F6, 0xB13A0FF7, 0xB14A0DF8, 0xB15A0AF9, 0xB16A09FA, 0xB17A07FB, + 0xB1805405, 0xB1905406, 0xB1A05407, 0xB1B05408, 0xB1C05409, 0xB1D0540A, 0xB1E0540B, 0xB1F05445, + 0xB2005446, 0xB2105447, 0xB2205448, 0xB2305449, 0xB240544A, 0xB250544B, 0xB2605485, 0xB2705486, + 0xB2805487, 0xB2905488, 0xB2A05489, 0xB2B0548A, 0xB2C0548B, 0xB2D054C5, 0xB2E054C6, 0xB2F054C7, + 0xB30054C8, 0xB31054C9, 0xB32054CA, 0xB33054CB, 0xB3485401, 0xB3585402, 0xB3685403, 0xB3785404, + 0xB3885405, 0xB3985406, 0xB3A85407, 0xB3B85408, 0xB3C85409, 0xB3D8540A, 0xB3E8510B, 0xB3F85441, + 0xB4085442, 0xB4185443, 0xB4285444, 0xB4385445, 0xB4485446, 0xB4585447, 0xB4685248, 0xB4784F49, + 0xB4884D4A, 0xB4984A4B, 0xB4A85481, 0xB4B85482, 0xB4C85483, 0xB4D85484, 0xB4E85485, 0xB4F85486, + 0xB5085487, 0xB5185288, 0xB5284F89, 0xB5384D8A, 0xB5484A8B, 0xB55854C1, 0xB56854C2, 0xB57854C3, + 0xB58854C4, 0xB59853C5, 0xB5A850C6, 0xB5B84FC7, 0xB5C84DC8, 0xB5D84AC9, 0xB5E849CA, 0xB5F847CB, + 0xB6085411, 0xB6185412, 0xB6285413, 0xB6385414, 0xB6485415, 0xB6585416, 0xB6685417, 0xB6785418, + 0xB6885419, 0xB698541A, 0xB6A8511B, 0xB6B85451, 0xB6C85452, 0xB6D85453, 0xB6E85454, 0xB6F85455, + 0xB7085456, 0xB7185457, 0xB7285258, 0xB7384F59, 0xB7484D5A, 0xB7584A5B, 0xB7685491, 0xB7785492, + 0xB7885493, 0xB7985494, 0xB7A85495, 0xB7B85496, 0xB7C85497, 0xB7D85298, 0xB7E84F99, 0xB7F84D9A, + 0xB8084A9B, 0xB81854D1, 0xB82854D2, 0xB83854D3, 0xB84854D4, 0xB85853D5, 0xB86850D6, 0xB8784FD7, + 0xB8884DD8, 0xB8984AD9, 0xB8A849DA, 0xB8B847DB, 0xB8C85421, 0xB8D85422, 0xB8E85423, 0xB8F85424, + 0xB9085425, 0xB9185426, 0xB9285427, 0xB9385428, 0xB9485429, 0xB958542A, 0xB968512B, 0xB9785461, + 0xB9885462, 0xB9985463, 0xB9A85464, 0xB9B85465, 0xB9C85466, 0xB9D85467, 0xB9E85268, 0xB9F84F69, + 0xBA084D6A, 0xBA184A6B, 0xBA2854A1, 0xBA3854A2, 0xBA4854A3, 0xBA5854A4, 0xBA6854A5, 0xBA7854A6, + 0xBA8854A7, 0xBA9852A8, 0xBAA84FA9, 0xBAB84DAA, 0xBAC84AAB, 0xBAD854E1, 0xBAE854E2, 0xBAF854E3, + 0xBB0854E4, 0xBB1853E5, 0xBB2850E6, 0xBB384FE7, 0xBB484DE8, 0xBB584AE9, 0xBB6849EA, 0xBB7847EB, + 0xBB8854B1, 0xBB9854B2, 0xBBA854B3, 0xBBB854B4, 0xBBC854B5, 0xBBD854B6, 0xBBE854B7, 0xBBF852B8, + 0xBC084FB9, 0xBC184DBA, 0xBC284ABB, 0xBC3854F1, 0xBC4854F2, 0xBC5854F3, 0xBC6854F4, 0xBC7853F5, + 0xBC8850F6, 0xBC984FF7, 0xBCA84DF8, 0xBCB84AF9, 0xBCC849FA, 0xBCD847FB, 0xBCE11408, 0xBCF11409, + 0xBD01140A, 0xBD11140B, 0xBD211448, 0xBD311449, 0xBD41144A, 0xBD51144B, 0xBD611488, 0xBD711489, + 0xBD81148A, 0xBD91148B, 0xBDA114C8, 0xBDB114C9, 0xBDC114CA, 0xBDD114CB, 0xBDE91402, 0xBDF91403, + 0xBE091404, 0xBE191405, 0xBE291406, 0xBE391407, 0xBE491408, 0xBE591409, 0xBE69140A, 0xBE79140B, + 0xBE891442, 0xBE991443, 0xBEA91444, 0xBEB91445, 0xBEC91446, 0xBED91447, 0xBEE91448, 0xBEF91449, + 0xBF09144A, 0xBF19144B, 0xBF291482, 0xBF391483, 0xBF491484, 0xBF591485, 0xBF691486, 0xBF791487, + 0xBF891488, 0xBF991489, 0xBFA9148A, 0xBFB9148B, 0xBFC914C2, 0xBFD914C3, 0xBFE914C4, 0xBFF914C5, + 0xC00914C6, 0xC01914C7, 0xC02913C8, 0xC03911C9, 0xC04910CA, 0xC0590ECB, 0xC0691412, 0xC0791413, + 0xC0891414, 0xC0991415, 0xC0A91416, 0xC0B91417, 0xC0C91418, 0xC0D91419, 0xC0E9141A, 0xC0F9141B, + 0xC1091452, 0xC1191453, 0xC1291454, 0xC1391455, 0xC1491456, 0xC1591457, 0xC1691458, 0xC1791459, + 0xC189145A, 0xC199145B, 0xC1A91492, 0xC1B91493, 0xC1C91494, 0xC1D91495, 0xC1E91496, 0xC1F91497, + 0xC2091498, 0xC2191499, 0xC229149A, 0xC239149B, 0xC24914D2, 0xC25914D3, 0xC26914D4, 0xC27914D5, + 0xC28914D6, 0xC29914D7, 0xC2A913D8, 0xC2B911D9, 0xC2C910DA, 0xC2D90EDB, 0xC2E91422, 0xC2F91423, + 0xC3091424, 0xC3191425, 0xC3291426, 0xC3391427, 0xC3491428, 0xC3591429, 0xC369142A, 0xC379142B, + 0xC3891462, 0xC3991463, 0xC3A91464, 0xC3B91465, 0xC3C91466, 0xC3D91467, 0xC3E91468, 0xC3F91469, + 0xC409146A, 0xC419146B, 0xC42914A2, 0xC43914A3, 0xC44914A4, 0xC45914A5, 0xC46914A6, 0xC47914A7, + 0xC48914A8, 0xC49914A9, 0xC4A914AA, 0xC4B914AB, 0xC4C914E2, 0xC4D914E3, 0xC4E914E4, 0xC4F914E5, + 0xC50914E6, 0xC51914E7, 0xC52913E8, 0xC53911E9, 0xC54910EA, 0xC5590EEB, 0xC56914B2, 0xC57914B3, + 0xC58914B4, 0xC59914B5, 0xC5A914B6, 0xC5B914B7, 0xC5C914B8, 0xC5D914B9, 0xC5E914BA, 0xC5F914BB, + 0xC60914F2, 0xC61914F3, 0xC62914F4, 0xC63914F5, 0xC64914F6, 0xC65914F7, 0xC66913F8, 0xC67911F9, + 0xC68910FA, 0xC6990EFB, 0xC6A03408, 0xC6B03409, 0xC6C0340A, 0xC6D0340B, 0xC6E03448, 0xC6F03449, + 0xC700344A, 0xC710344B, 0xC7203488, 0xC7303489, 0xC740348A, 0xC750348B, 0xC76034C8, 0xC77034C9, + 0xC78034CA, 0xC79034CB, 0xC7A83402, 0xC7B83403, 0xC7C83404, 0xC7D83405, 0xC7E83406, 0xC7F83407, + 0xC8083408, 0xC8183409, 0xC828340A, 0xC838340B, 0xC8483442, 0xC8583443, 0xC8683444, 0xC8783445, + 0xC8883446, 0xC8983447, 0xC8A83448, 0xC8B83449, 0xC8C8344A, 0xC8D8344B, 0xC8E83482, 0xC8F83483, + 0xC9083484, 0xC9183485, 0xC9283486, 0xC9383487, 0xC9483488, 0xC9583489, 0xC968348A, 0xC978348B, + 0xC98834C2, 0xC99834C3, 0xC9A834C4, 0xC9B834C5, 0xC9C834C6, 0xC9D834C7, 0xC9E833C8, 0xC9F831C9, + 0xCA0830CA, 0xCA182ECB, 0xCA283412, 0xCA383413, 0xCA483414, 0xCA583415, 0xCA683416, 0xCA783417, + 0xCA883418, 0xCA983419, 0xCAA8341A, 0xCAB8341B, 0xCAC83452, 0xCAD83453, 0xCAE83454, 0xCAF83455, + 0xCB083456, 0xCB183457, 0xCB283458, 0xCB383459, 0xCB48345A, 0xCB58345B, 0xCB683492, 0xCB783493, + 0xCB883494, 0xCB983495, 0xCBA83496, 0xCBB83497, 0xCBC83498, 0xCBD83499, 0xCBE8349A, 0xCBF8349B, + 0xCC0834D2, 0xCC1834D3, 0xCC2834D4, 0xCC3834D5, 0xCC4834D6, 0xCC5834D7, 0xCC6833D8, 0xCC7831D9, + 0xCC8830DA, 0xCC982EDB, 0xCCA83422, 0xCCB83423, 0xCCC83424, 0xCCD83425, 0xCCE83426, 0xCCF83427, + 0xCD083428, 0xCD183429, 0xCD28342A, 0xCD38342B, 0xCD483462, 0xCD583463, 0xCD683464, 0xCD783465, + 0xCD883466, 0xCD983467, 0xCDA83468, 0xCDB83469, 0xCDC8346A, 0xCDD8346B, 0xCDE834A2, 0xCDF834A3, + 0xCE0834A4, 0xCE1834A5, 0xCE2834A6, 0xCE3834A7, 0xCE4834A8, 0xCE5834A9, 0xCE6834AA, 0xCE7834AB, + 0xCE8834E2, 0xCE9834E3, 0xCEA834E4, 0xCEB834E5, 0xCEC834E6, 0xCED834E7, 0xCEE833E8, 0xCEF831E9, + 0xCF0830EA, 0xCF182EEB, 0xCF2834B2, 0xCF3834B3, 0xCF4834B4, 0xCF5834B5, 0xCF6834B6, 0xCF7834B7, + 0xCF8834B8, 0xCF9834B9, 0xCFA834BA, 0xCFB834BB, 0xCFC834F2, 0xCFD834F3, 0xCFE834F4, 0xCFF834F5, + 0xD00834F6, 0xD01834F7, 0xD02833F8, 0xD03831F9, 0xD04830FA, 0xD0582EFB, +}; + +uniform int get_bits(uniform uint32_t value, uniform int from, uniform int to) +{ + return (value >> from) & ((1 << (to + 1 - from)) - 1); +} + +void load_mode_parameters(uniform astc_mode* uniform mode, uniform uint32_t packed_mode) +{ + mode->width = 2 + get_bits(packed_mode, 13, 15); // 2..8 <= 2^3 + mode->height = 2 + get_bits(packed_mode, 16, 18); // 2..8 <= 2^3 + mode->dual_plane = get_bits(packed_mode, 19, 19); // 0 or 1 + mode->partitions = 1; + + mode->weight_range = get_bits(packed_mode, 0, 3); // 0..11 <= 2^4 + mode->color_component_selector = get_bits(packed_mode, 4, 5); // 0..2 <= 2^2 + mode->partition_id = 0; + mode->color_endpoint_modes[0] = get_bits(packed_mode, 6, 7) * 2 + 6; // 6 or 8 + mode->color_endpoint_pairs = 1 + (mode->color_endpoint_modes[0] / 4); + mode->endpoint_range = get_bits(packed_mode, 8, 12); // 0..20 <= 2^5 +} + +export void astc_rank_ispc(uniform rgba_surface src[], uniform int xx, uniform int yy, uniform uint32_t mode_buffer[], uniform astc_enc_settings settings[]) +{ + int tex_width = src->width / settings->block_width; + if (xx + programIndex >= tex_width) return; + + astc_rank_state _state; + varying astc_rank_state* uniform state = &_state; + + state->block_width = settings->block_width; + state->block_height = settings->block_height; + state->fastSkipTreshold = settings->fastSkipTreshold; + + assert(state->fastSkipTreshold <= 64); + + load_block_interleaved(state->pixels, src, xx + programIndex, yy, state->block_width, state->block_height); + if (settings->channels == 3) clear_alpha(state->pixels, state->block_width, state->block_height); + + compute_metrics(state); + + float threshold_error = 0; + int count = -1; + + for (uniform int id = 0; id < packed_modes_count; id++) + { + uniform uint32_t packed_mode = packed_modes[id]; + + uniform astc_mode _mode; + uniform astc_mode* uniform mode = &_mode; + load_mode_parameters(mode, packed_mode); + + if (mode->height > state->block_height) continue; + if (mode->width > state->block_width) continue; + + if (settings->channels == 3 && mode->color_endpoint_modes[0] > 8) continue; + + float error = estimate_error(state, mode); + count += 1; + + if (count < state->fastSkipTreshold) + { + state->best_modes[count] = packed_mode; + state->best_scores[count] = error; + + threshold_error = max(threshold_error, error); + } + else if (error < threshold_error) + { + insert_element(state, error, packed_mode, &threshold_error); + } + } + + assert(count >= 0); + + for (uniform int i = 0; i < state->fastSkipTreshold; i++) + { + mode_buffer[programCount * i + programIndex] = state->best_modes[i]; + } +} + +/////////////////////////////////////////////////////////// +// ASTC candidate encoding + +struct astc_block +{ + uniform int width; + uniform int height; + uniform uint8_t dual_plane; + int weight_range; + uint8_t weights[64]; + int color_component_selector; + + uniform int partitions; + int partition_id; + uniform int color_endpoint_pairs; + uniform int channels; + int color_endpoint_modes[4]; + int endpoint_range; + uint8_t endpoints[18]; +}; + +struct astc_enc_state +{ + float pixels[256]; + float scaled_pixels[256]; + uint32_t data[4]; + + // settings + uniform int block_width; + uniform int block_height; + uniform int pitch; + + uniform int refineIterations; +}; + +struct astc_enc_context +{ + // uniform parameters + int width; + int height; + int channels; + bool dual_plane; + int partitions; + int color_endpoint_pairs; +}; + +uniform static const float filter_data[309] = +{ + 0.688356,-0.188356, 0.414384, 0.085616, 0.085616, 0.414384,-0.188356, 0.688356, + 0.955516,-0.227273, 0.044484, 0.142349, 0.727273,-0.142349,-0.142349, 0.727273, + 0.142349, 0.044484,-0.227273, 0.955516, 0.600000,-0.200000, 0.400000, 0.000000, + 0.200000, 0.200000, 0.000000, 0.400000,-0.200000, 0.600000, 0.828571,-0.142857, + 0.028571, 0.342857, 0.285714,-0.057143,-0.142857, 0.714286,-0.142857,-0.057143, + 0.285714, 0.342857, 0.028571,-0.142857, 0.828571, 0.985714,-0.252381, 0.080952, + -0.014286, 0.057143, 1.009524,-0.323810, 0.057143,-0.085714, 0.485714, 0.485714, + -0.085714, 0.057143,-0.323810, 1.009524, 0.057143,-0.014286, 0.080952,-0.252381, + 0.985714, 0.510753,-0.177419, 0.381720,-0.048387, 0.252688, 0.080645, 0.080645, + 0.252688,-0.048387, 0.381720,-0.177419, 0.510753, 0.754228,-0.194882, 0.052858, + 0.398312, 0.147638,-0.040044,-0.016924, 0.547244,-0.148431,-0.148431, 0.547244, + -0.016924,-0.040044, 0.147638, 0.398312, 0.052858,-0.194882, 0.754228, 0.921235, + -0.216677, 0.063615,-0.013072, 0.210040, 0.577804,-0.169641, 0.034858,-0.164122, + 0.798726,-0.053828, 0.011061, 0.011061,-0.053828, 0.798726,-0.164122, 0.034858, + -0.169641, 0.577804, 0.210040,-0.013072, 0.063615,-0.216677, 0.921235, 0.996932, + -0.209923, 0.069231,-0.020846, 0.003068, 0.016362, 1.119589,-0.369231, 0.111180, + -0.016362,-0.035452, 0.240891, 0.800000,-0.240891, 0.035452, 0.035452,-0.240891, + 0.800000, 0.240891,-0.035452,-0.016362, 0.111180,-0.369231, 1.119589, 0.016362, + 0.003068,-0.020846, 0.069231,-0.209923, 0.996932, 0.415909,-0.165909, 0.343182, + -0.093182, 0.234091, 0.015909, 0.161364, 0.088636, 0.088636, 0.161364, 0.015909, + 0.234091,-0.093182, 0.343182,-0.165909, 0.415909, 0.653807,-0.172170, 0.058458, + 0.395689, 0.040094,-0.013613, 0.189195, 0.209906,-0.071270,-0.068923, 0.422170, + -0.143341,-0.143341, 0.422170,-0.068923,-0.071270, 0.209906, 0.189195,-0.013613, + 0.040094, 0.395689, 0.058458,-0.172170, 0.653807, 0.805363,-0.204713, 0.061406, + -0.016387, 0.363455, 0.220460,-0.066129, 0.017647,-0.078453, 0.645632,-0.193664, + 0.051682,-0.121551, 0.455481, 0.081527,-0.021756,-0.021756, 0.081527, 0.455481, + -0.121551, 0.051682,-0.193664, 0.645632,-0.078453, 0.017647,-0.066129, 0.220460, + 0.363455,-0.016387, 0.061406,-0.204713, 0.805363, 0.881593,-0.204539, 0.075065, + -0.021559, 0.004453, 0.270644, 0.467517,-0.171576, 0.049278,-0.010179,-0.169588, + 0.821023,-0.159819, 0.045902,-0.009481,-0.012311, 0.059603, 0.756331,-0.217226, + 0.044870, 0.044870,-0.217226, 0.756331, 0.059603,-0.012311,-0.009481, 0.045902, + -0.159819, 0.821023,-0.169588,-0.010179, 0.049278,-0.171576, 0.467517, 0.270644, + 0.004453,-0.021559, 0.075065,-0.204539, 0.881593, 0.967275,-0.287351, 0.076902, + -0.018670, 0.005432,-0.000959, 0.104719, 0.919524,-0.246087, 0.059743,-0.017382, + 0.003067,-0.127990, 0.653915, 0.300773,-0.073019, 0.021245,-0.003749, 0.064956, + -0.331864, 1.007366,-0.105833, 0.030792,-0.005434,-0.006723, 0.034349,-0.104266, + 0.996397,-0.289905, 0.051160,-0.005112, 0.026120,-0.079287, 0.323158, 0.571013, + -0.100767, 0.003834,-0.019590, 0.059465,-0.242368, 0.905074, 0.075575,-0.000959, + 0.004898,-0.014866, 0.060592,-0.226268, 0.981106, +}; + +uniform static const int filterbank[5][5] = +{ + { 0, 8, -1, -1, -1 }, + { 20, 30, 45, -1, -1 }, + { 65, 77, 95, 119, -1 }, + { -1, -1, -1, -1, -1 }, + { 149, 165, 189, 221, 261 }, +}; + +void scale_pixels(astc_enc_state state[], uniform astc_enc_context ctx[]) +{ + uniform int channels = ctx->channels; + uniform const float* uniform yfilter = &filter_data[filterbank[state->block_height - 4][ctx->height - 2]]; + uniform const float* uniform xfilter = &filter_data[filterbank[state->block_width - 4][ctx->width - 2]]; + + for (uniform int y = 0; y < ctx->height; y++) + { + float line[8][4]; + + if (state->block_height == ctx->height) + { + for (uniform int x = 0; x < state->block_width; x++) + for (uniform int p = 0; p < channels; p++) + line[x][p] = get_pixel(state->pixels, p, x, y); + } + else + for (uniform int x = 0; x < state->block_width; x++) + { + uniform int n = ctx->height; + + for (uniform int p = 0; p < channels; p++) line[x][p] = 0; + + for (uniform int k = 0; k < state->block_height; k++) + for (uniform int p = 0; p < channels; p++) + line[x][p] += yfilter[k * n + y] * get_pixel(state->pixels, p, x, k); + } + + if (state->block_width == ctx->width) + { + for (uniform int x = 0; x < ctx->width; x++) + for (uniform int p = 0; p < channels; p++) + set_pixel(state->scaled_pixels, p, x, y, clamp(line[x][p], 0, 255)); + } + else + for (uniform int x = 0; x < ctx->width; x++) + { + uniform int n = ctx->width; + + float value[4] = { 0, 0, 0, 0 }; + + for (uniform int k = 0; k < state->block_width; k++) + for (uniform int p = 0; p < channels; p++) + value[p] += xfilter[k * n + x] * line[k][p]; + + for (uniform int p = 0; p < channels; p++) + set_pixel(state->scaled_pixels, p, x, y, clamp(value[p], 0, 255)); + } + } +} + +inline int clamp_unorm8(int value) +{ + if (value < 0) return 0; + if (value > 255) return 255; + return value; +} + +inline void apply_blue_contract(int& r, int& g, int& b) +{ + r = (r + b) >> 1; + g = (g + b) >> 1; +} + +void decode_endpoints(float endpoints[8], uint8_t coded_endpoints[], int mode) +{ + if ((mode % 4) == 2) + { + int v0 = coded_endpoints[0]; + int v1 = coded_endpoints[1]; + int v2 = coded_endpoints[2]; + int v3 = coded_endpoints[3]; + int v4 = coded_endpoints[4]; + int v5 = coded_endpoints[5]; + + endpoints[0] = (v0 * v3) >> 8; + endpoints[1] = (v1 * v3) >> 8; + endpoints[2] = (v2 * v3) >> 8; + endpoints[3] = 0xFF; + + endpoints[4] = v0; + endpoints[5] = v1; + endpoints[6] = v2; + endpoints[7] = 0xFF; + + if (mode > 8) + { + endpoints[3] = clamp_unorm8(v4); + endpoints[7] = clamp_unorm8(v5); + } + } + + if ((mode % 4) == 0) + { + int v0 = coded_endpoints[0]; + int v1 = coded_endpoints[1]; + int v2 = coded_endpoints[2]; + int v3 = coded_endpoints[3]; + int v4 = coded_endpoints[4]; + int v5 = coded_endpoints[5]; + int v6 = coded_endpoints[6]; + int v7 = coded_endpoints[7]; + + bool swap_endpoints = v1 + v3 + v5 < v0 + v2 + v4; + + if (swap_endpoints) + { + swap(v0, v1); + swap(v2, v3); + swap(v4, v5); + swap(v6, v7); + + apply_blue_contract(v0, v2, v4); + apply_blue_contract(v1, v3, v5); + } + + endpoints[0] = clamp_unorm8(v0); + endpoints[1] = clamp_unorm8(v2); + endpoints[2] = clamp_unorm8(v4); + endpoints[3] = 0xFF; + + endpoints[4] = clamp_unorm8(v1); + endpoints[5] = clamp_unorm8(v3); + endpoints[6] = clamp_unorm8(v5); + endpoints[7] = 0xFF; + + if (mode > 8) + { + endpoints[3] = clamp_unorm8(v6); + endpoints[7] = clamp_unorm8(v7); + } + } +} + +void dequant_decode_endpoints(float endpoints[8], uint8_t block_endpoints[], int mode, int range) +{ + int levels = get_levels(range); + int num_cem_pairs = 1 + mode / 4; + + uint8_t dequant_endpoints[8]; + for (uniform int k = 0; k < 2 * num_cem_pairs; k++) + { + dequant_endpoints[k] = (int)(((int)block_endpoints[k]) * 255.0f / (levels - 1) + 0.5); + } + + decode_endpoints(endpoints, dequant_endpoints, mode); +} + +bool compare_endpoints(uint8_t endpoints[8], astc_block block[]) +{ + int sum = 0; + for (uniform int p = 0; p < 3; p++) + { + sum += endpoints[p * 2 + 0]; + sum -= endpoints[p * 2 + 1]; + } + + if (-2 <= sum && sum <= 2) + { + // avoid being too close so we don't need proper rounding + for (uniform int p = 0; p < 3; p++) + { + if (sum<=0) + endpoints[p * 2 + 0] = clamp(endpoints[p * 2 + 0] - 1, 0, get_levels(block->endpoint_range) - 1); + if (sum>0) + endpoints[p * 2 + 1] = clamp(endpoints[p * 2 + 1] - 1, 0, get_levels(block->endpoint_range) - 1); + } + + sum = 0; + for (uniform int p = 0; p < 3; p++) + { + sum += endpoints[p * 2 + 0]; + sum -= endpoints[p * 2 + 1]; + } + } + + return sum > 0; +} + +void reorder_endpoints(uint8_t endpoints[8], astc_block block[], bool blue_contract) +{ + if (compare_endpoints(endpoints, block) == !blue_contract) + for (uniform int p = 0; p < 4; p++) swap(endpoints[p * 2], endpoints[p * 2 + 1]); +} + +inline int quant_endpoint(float value, int levels) +{ + return clamp(value / 255.0f * (levels - 1) + 0.5, 0, levels - 1); +} + +void quantize_endpoints_scale(astc_block block[], float endpoints[4]) +{ + int ep_levels = get_levels(block->endpoint_range); + + float near[3]; + float far[3]; + for (uniform int p = 0; p < 3; p++) + { + near[p] = endpoints[p * 2 + 0]; + far[p] = endpoints[p * 2 + 1]; + } + + for (uniform int p = 0; p < 3; p++) + block->endpoints[p] = quant_endpoint(far[p], ep_levels); + + float sq_norm = dot3(far, far) + 0.00001; + float scale = dot3(far, near) / sq_norm; + + block->endpoints[3] = quant_endpoint(scale * 256, ep_levels); + + if (block->color_endpoint_modes[0] > 8) + { + block->endpoints[4] = quant_endpoint(endpoints[3 * 2 + 0], ep_levels); + block->endpoints[5] = quant_endpoint(endpoints[3 * 2 + 1], ep_levels); + } +} + +void quantize_endpoints_pair(astc_block block[], float endpoints[6]) +{ + int ep_levels = get_levels(block->endpoint_range); + + bool blue_contract = true; + + float blue_compressed[6]; + for (uniform int i = 0; i < 2; i++) + { + blue_compressed[i + 0] = endpoints[i + 0] * 2 - endpoints[i + 4]; + blue_compressed[i + 2] = endpoints[i + 2] * 2 - endpoints[i + 4]; + blue_compressed[i + 4] = endpoints[i + 4]; + + if (blue_compressed[i + 0] < 0) blue_contract = false; + if (blue_compressed[i + 0] > 255) blue_contract = false; + if (blue_compressed[i + 2] < 0) blue_contract = false; + if (blue_compressed[i + 2] > 255) blue_contract = false; + } + + if (blue_contract) + { + for (uniform int p = 0; p < 3; p++) + { + block->endpoints[p * 2 + 0] = quant_endpoint(blue_compressed[p * 2 + 0], ep_levels); + block->endpoints[p * 2 + 1] = quant_endpoint(blue_compressed[p * 2 + 1], ep_levels); + } + } + else + { + for (uniform int p = 0; p < 3; p++) + { + block->endpoints[p * 2 + 0] = quant_endpoint(endpoints[p * 2 + 0], ep_levels); + block->endpoints[p * 2 + 1] = quant_endpoint(endpoints[p * 2 + 1], ep_levels); + } + } + + if (block->color_endpoint_modes[0] > 8) + { + block->endpoints[6] = quant_endpoint(endpoints[3 * 2 + 0], ep_levels); + block->endpoints[7] = quant_endpoint(endpoints[3 * 2 + 1], ep_levels); + } + + reorder_endpoints(block->endpoints, block, blue_contract); +} + +void quantize_endpoints(astc_block block[], float endpoints[]) +{ + bool zero_based = (block->color_endpoint_modes[0] % 4) == 2; + + if (zero_based) + { + quantize_endpoints_scale(block, endpoints); + } + else + { + quantize_endpoints_pair(block, endpoints); + } +} + +void opt_weights(float scaled_pixels[], astc_block block[]) +{ + uniform int channels = 4; + if (block->dual_plane) channels = 3; + + float rec_endpoints[8]; + dequant_decode_endpoints(rec_endpoints, block->endpoints, block->color_endpoint_modes[0], block->endpoint_range); + + int w_levels = get_levels(block->weight_range); + + float dir[4]; dir[3] = 0; + for (uniform int p = 0; p < channels; p++) dir[p] = rec_endpoints[4 + p] - rec_endpoints[0 + p]; + float sq_norm = dot4(dir, dir) + 0.00001; + for (uniform int p = 0; p < channels; p++) dir[p] *= (w_levels - 1) / sq_norm; + + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + float pixel[4]; pixel[3] = 0; + for (uniform int p = 0; p < channels; p++) pixel[p] = get_pixel(scaled_pixels, p, x, y) - rec_endpoints[0 + p]; + + int q = clamp(dot4(pixel, dir) + 0.5, 0, w_levels - 1); + + block->weights[y * block->width + x] = q; + } +} + +bool sgesv2(float A[4], float x[2], float b[2]) +{ + float det = (A[0] * A[3] - A[1] * A[2]); + if(det == 0) + return false; + float inv_det = 1.0f / det; + x[0] = (b[0] * +A[3] + b[1] * -A[2])*inv_det; + x[1] = (b[0] * -A[1] + b[1] * +A[0])*inv_det; + + return true; +} + +void ls_refine_scale(float endpoints[8], float scaled_pixels[], astc_block block[]) +{ + int levels = get_levels(block->weight_range); + float levels_rcp = 1.0f / (levels - 1); + + // In this mode, the endpoints are on a line through 0, and the first endpoint + // is a scaled version of the second endpoint with a scale factor 0 <= s < 1. + // + // Determining optimal s and endpoints is a non-linear problem; approximate it + // by first solving for the scale factor and then separately solving for the + // endpoint value. + // + // The scale factor solve starts from (where e_0, e_1 are RGB endpoints) + // e_0 = s * e_1 + // + // thus for an interpolation weight w + // lerp(e_0, e_1, w) + // = lerp(s * e_1, e_1, w) + // = lerp(s, 1, w) * e_1 + // = (s + (1-s)*w) * e_1 + // + // and if we take look at the 2-norm (Euclidean length) of that vector + // + // len(lerp(e_0, e_1, w)) = (s + (1-s)*w) * len(e_1) + // + // if we consider s and (1-s) as separate unknowns xx_0 and xx_1, we get + // an overdetermined linear system for the 2D vector lengths of the pixels + // (d_i in the following) that we can solve in a Least-Squares sense via + // the Normal Equations and Cramer's rule: + // + // [1 w_1] [d_1] + // [1 w_2] [xx_0] [d_2] + // [1 w_3] [xx_1] = [d_3] + // [ ... ] [...] + // [1 w_N] [d_N] + // + // we then recover s (the ratio between the lengths of e_0 and e_1) as + // xx_0 / (xx_0 + xx_1). + float sum_w = 0; + float sum_ww = 0; + float sum_d = 0; + float sum_wd = 0; + + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + float w = (int)block->weights[y * block->width + x] * levels_rcp; + float d = 0; + + for (uniform int p = 0; p < 3; p++) d += sq(get_pixel(scaled_pixels, p, x, y)); + d = sqrt(d+0.01f); + + sum_w += w; + sum_ww += w*w; + sum_d += d; + sum_wd += w*d; + } + + float sum_1 = 1.0f * block->height * block->width; + + float C[4] = { sum_1, sum_w, sum_w, sum_ww }; + float b[2] = { sum_d, sum_wd }; + float xx[2]; + + // Singular configurations are precisely those where all weights are equal, i.e. constant color. + // Used to set scale=0 in this case but that's really bad when that weight is 0 (we're now making + // the whole block black). scale=1 isn't ideal (and in fact not something we can hit exactly) + // but at least leaves the endpoints in roughly the right place. + float scale = 1; + if(sgesv2(C, xx, b)) + { + scale = xx[0] / (xx[1] + xx[0]); + if (xx[1] + xx[0] < 1) scale = 1; + scale = clamp(scale, 0.0f, 0.9999f); // note: clamp also takes care of possible NaNs + } + + // Now, solve another Least Squares system for the actual endpoint values given the previously + // determined scale. This time we're trying to solve, for every pixel p_i + // + // p_i = lerp(e_0, e_1, w) = (s + (1-s)*w_i) * e_1 + // + // let z_i := s + (1 - s)*w_i, then we get the overdetermined linear system + // + // [z_1] [p_1^T] + // [z_2] [p_2^T] + // [z_3] [e_1]^T = [p_3^T] + // [...] [ ... ] + // [z_N] [p_N^T] + // + // which we solve in a least-squares sense for e_1. + float sum_zz = 0; + float sum_zp[3] = { 0, 0, 0 }; + + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + float w = (int)block->weights[y * block->width + x] * levels_rcp; + float z = scale + (1 - scale)*w; + + sum_zz += z * z; + for (uniform int p = 0; p < 3; p++) sum_zp[p] += z * get_pixel(scaled_pixels, p, x, y); + } + + for (uniform int p = 0; p < 3; p++) endpoints[2 * p + 0] = scale * sum_zp[p] / sum_zz; + for (uniform int p = 0; p < 3; p++) endpoints[2 * p + 1] = sum_zp[p] / sum_zz; + + if (block->channels == 4) + { + float Atb1 = 0; + float sum_q = 0; + float sum_qq = 0; + float sum[2] = { 0, 0 }; + + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + int q = block->weights[y * block->width + x]; + int z = (levels - 1) - q; + + sum_q += q; + sum_qq += q*q; + + sum[1] += 1; + sum[0] += get_pixel(scaled_pixels, 3, x, y); + Atb1 += z * get_pixel(scaled_pixels, 3, x, y); + } + + float Atb2 = (levels - 1)*sum[0] - Atb1; + + float Cxx = sum[1] * sq(levels - 1) - 2 * (levels - 1)*sum_q + sum_qq; + float Cyy = sum_qq; + float Cxy = (levels - 1)*sum_q - sum_qq; + float scale = 1.0f / (Cxx*Cyy - Cxy*Cxy); + + float ep[8]; + ep[0 + 3] = (levels - 1)*(Atb1 * Cyy - Atb2 * Cxy)*scale; + ep[4 + 3] = (levels - 1)*(Atb2 * Cxx - Atb1 * Cxy)*scale; + + if (abs(Cxx*Cyy - Cxy*Cxy) < 0.001) + { + ep[0 + 3] = sum[0] / sum[1]; + ep[4 + 3] = ep[0 + 3]; + } + + endpoints[6 + 0] = ep[0 + 3]; + endpoints[6 + 1] = ep[4 + 3]; + } +} + +void ls_refine_pair(float endpoints[6], float scaled_pixels[], astc_block block[]) +{ + uniform int channels = block->channels; + int levels = get_levels(block->weight_range); + + float Atb1[4] = { 0, 0, 0, 0 }; + float sum_q = 0; + float sum_qq = 0; + float sum[5] = { 0, 0, 0, 0, 0 }; + + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + int q = block->weights[y * block->width + x]; + + int z = (levels - 1) - q; + + sum_q += q; + sum_qq += q*q; + + sum[4] += 1; + for (uniform int p = 0; p < channels; p++) sum[p] += get_pixel(scaled_pixels, p, x, y); + for (uniform int p = 0; p < channels; p++) Atb1[p] += z * get_pixel(scaled_pixels, p, x, y); + } + + float Atb2[4]; + for (uniform int p = 0; p < channels; p++) + { + Atb2[p] = (levels - 1)*sum[p] - Atb1[p]; + } + + float Cxx = sum[4] * sq(levels - 1) - 2 * (levels - 1)*sum_q + sum_qq; + float Cyy = sum_qq; + float Cxy = (levels - 1)*sum_q - sum_qq; + float scale = 1.0f / (Cxx*Cyy - Cxy*Cxy); + + float ep[8]; + for (uniform int p = 0; p < channels; p++) + { + ep[0 + p] = (levels - 1)*(Atb1[p] * Cyy - Atb2[p] * Cxy)*scale; + ep[4 + p] = (levels - 1)*(Atb2[p] * Cxx - Atb1[p] * Cxy)*scale; + } + + if (abs(Cxx*Cyy - Cxy*Cxy) < 0.001) + { + // flatten + for (int p = 0; p < channels; p++) + { + ep[0 + p] = sum[p] / sum[4]; + ep[4 + p] = ep[0 + p]; + } + } + + for (uniform int p = 0; p < channels; p++) + { + endpoints[2 * p + 0] = ep[0 + p]; + endpoints[2 * p + 1] = ep[4 + p]; + } +} + +void ls_refine(float endpoints[], float scaled_pixels[], astc_block block[]) +{ + if (block->color_endpoint_modes[0] % 4 == 2) + { + ls_refine_scale(endpoints, scaled_pixels, block); + } + else + { + ls_refine_pair(endpoints, scaled_pixels, block); + } +} + +float optimize_alt_plane(uint8_t alt_weights[], float scaled_pixels[], astc_block block[]) +{ + int ccs = block->color_component_selector; + + float ext[2] = { 1000, -1000 }; + + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + float value = get_pixel(scaled_pixels, 3, x, y); + ext[0] = min(ext[0], value); + ext[1] = max(ext[1], value); + } + + block->endpoints[3 * 2 + 0] = 0; + block->endpoints[3 * 2 + 1] = 255; + + float _rec_endpoints[8]; + dequant_decode_endpoints(_rec_endpoints, block->endpoints, block->color_endpoint_modes[0], block->endpoint_range); + + float endpoints[8]; + for (int p = 0; p < 3; p++) + { + endpoints[p * 2 + 0] = _rec_endpoints[0 + p]; + endpoints[p * 2 + 1] = _rec_endpoints[4 + p]; + } + + endpoints[3 * 2 + 0] = gather_float(endpoints, ccs * 2 + 0); + endpoints[3 * 2 + 1] = gather_float(endpoints, ccs * 2 + 1); + + scatter_float(endpoints, ccs * 2 + 0, ext[0]); + scatter_float(endpoints, ccs * 2 + 1, ext[1]); + + quantize_endpoints(block, endpoints); + + float rec_endpoints[8]; + dequant_decode_endpoints(rec_endpoints, block->endpoints, block->color_endpoint_modes[0], block->endpoint_range); + + float base = gather_float(rec_endpoints, 0 + ccs); + float dir = gather_float(rec_endpoints, 4 + ccs) - base; + float sq_norm = sq(dir) + 0.00001; + + int w_levels = get_levels(block->weight_range); + dir *= (w_levels - 1) / sq_norm; + + float err = 0; + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + float value = get_pixel(scaled_pixels, 3, x, y) - base; + + int q = clamp(value * dir + 0.5, 0, w_levels - 1); + + alt_weights[y * block->width + x] = q; + } + + if (dir < 0) + for (uniform int y = 0; y < block->height; y++) + for (uniform int x = 0; x < block->width; x++) + { + int q = block->weights[y * block->width + x]; + + block->weights[y * block->width + x] = w_levels - 1 - q; + } + + return err; +} + +void optimize_block(float scaled_pixels[], astc_block block[], astc_enc_state state[]) +{ + pixel_set pset; + pset.pixels = scaled_pixels; + pset.width = block->width; + pset.height = block->height; + + float ep[8]; + bool zero_based = (block->color_endpoint_modes[0] % 4) == 2; + compute_pca_endpoints(ep, &pset, zero_based, 4); + + quantize_endpoints(block, ep); + opt_weights(scaled_pixels, block); + + for (uniform int i = 0; i < state->refineIterations; i++) + { + ls_refine(ep, scaled_pixels, block); + quantize_endpoints(block, ep); + opt_weights(scaled_pixels, block); + } + + if (block->dual_plane) + { + uint8_t alt_weights[64]; + optimize_alt_plane(alt_weights, scaled_pixels, block); + + uint8_t block_weights[64]; + for (uniform int i = 0; i < block->width * block->height; i++) + { + block_weights[i] = block->weights[i]; + } + + for (uniform int i = 0; i < block->width * block->height; i++) + { + block->weights[i * 2 + 0] = block_weights[i]; + block->weights[i * 2 + 1] = alt_weights[i]; + } + } +} + +float measure_error(astc_block block[], astc_enc_state state[]) +{ + uniform int pitch = state->block_height * state->block_width; + assert(pitch <= 64); + + // dequant values + uniform int num_weights = block->width * block->height * (block->dual_plane ? 2 : 1); + + range_values weight_range_values = get_range_values(block->weight_range); + + int block_weights[64]; + for (int i = 0; i < num_weights; i++) + { + block_weights[i] = ((int)block->weights[i] * 64.0f / (weight_range_values.levels - 1) + 0.5); + } + + float rgba_endpoints[8]; + dequant_decode_endpoints(rgba_endpoints, block->endpoints, block->color_endpoint_modes[0], block->endpoint_range); + + uniform int stride = block->width; + uniform int Ds = (1024 + state->block_width / 2) / (state->block_width - 1); + uniform int Dt = (1024 + state->block_height / 2) / (state->block_height - 1); + + uint8_t main_weights[64]; + uint8_t alt_weights[64]; + + for (uniform int i = 0; i < num_weights; i++) main_weights[i] = block_weights[i]; + + if (block->dual_plane) + for (uniform int i = 0; i < num_weights/2; i++) + { + main_weights[i] = block_weights[i * 2 + 0]; + alt_weights[i] = block_weights[i * 2 + 1]; + } + + float sq_error = 0; + + for (uniform int y = 0; y < state->block_height; y++) + for (uniform int x = 0; x < state->block_width; x++) + { + uniform int gs = (x * Ds * (block->width - 1) + 32) >> 6; + uniform int gt = (y * Dt * (block->height - 1) + 32) >> 6; + + uniform int js = gs >> 4; + uniform int jt = gt >> 4; + + uniform int fs = gs & 0x0F; + uniform int ft = gt & 0x0F; + uniform int w11 = ((fs*ft + 8) >> 4); + + int filled_weight = 0; + int alt_filled_weight = 0; + + { + int acc = 0; + acc += main_weights[stride * (jt + 0) + js + 0] * (16 - ft - fs + w11); + acc += main_weights[stride * (jt + 0) + js + 1] * (fs - w11); + acc += main_weights[stride * (jt + 1) + js + 0] * (ft - w11); + acc += main_weights[stride * (jt + 1) + js + 1] * w11; + filled_weight = (acc + 8) >> 4; + } + + if (block->dual_plane) + { + int acc = 0; + acc += alt_weights[stride * (jt + 0) + js + 0] * (16 - ft - fs + w11); + acc += alt_weights[stride * (jt + 0) + js + 1] * (fs - w11); + acc += alt_weights[stride * (jt + 1) + js + 0] * (ft - w11); + acc += alt_weights[stride * (jt + 1) + js + 1] * w11; + alt_filled_weight = (acc + 8) >> 4; + } + + for (uniform int p = 0; p < block->channels; p++) + { + int C0 = rgba_endpoints[0 + p] * 256 + 128; + int C1 = rgba_endpoints[4 + p] * 256 + 128; + int w = filled_weight; + + if (block->dual_plane && block->color_component_selector == p) + { + w = alt_filled_weight; + } + + int C = (C0 * (64 - w) + C1 * w + 32) / 64; + + float diff = (C >> 8) - get_pixel(state->pixels, p, x, y); + sq_error += diff * diff; + } + } + + return sq_error; +} + +int code_value(int value, range_values range) +{ + int coded = value; + + if (range.levels_m != 2 && range.levels > 5) + { + int value2 = value; + if (value >= range.levels / 2) value2 = (range.levels - 1) - value; + int q = (value2 * range.levels_m_rcp) >> 16; + int r = value2 - q * range.levels_m; + coded = q + r * (1 << (range.levels_e - 1)); + coded = coded * 2 + ((value >= range.levels / 2) ? 1 : 0); + } + + return coded; +} + +void code_block(astc_block block[]) +{ + uniform int num_weights = block->width * block->height * (block->dual_plane ? 2 : 1); + + range_values weight_range_values = get_range_values(block->weight_range); + for (uniform int i = 0; i < num_weights; i++) + { + block->weights[i] = code_value(block->weights[i], weight_range_values); + } + + range_values endpoint_range_values = get_range_values(block->endpoint_range); + for (uniform int i = 0; i < 2 * block->color_endpoint_pairs; i++) + { + block->endpoints[i] = code_value(block->endpoints[i], endpoint_range_values); + } +} + +extern "C" void pack_block_c(uniform uint32_t data[4], uniform astc_block block[]); + +void pack_block(astc_block block[], astc_enc_state state[]) +{ + code_block(block); + + foreach_active (instance) + { + uniform astc_block ublock; + + ublock.width = block->width; + ublock.height = block->height; + ublock.dual_plane = block->dual_plane; + ublock.partitions = block->partitions; + ublock.color_endpoint_pairs = block->color_endpoint_pairs; + + ublock.weight_range = extract(block->weight_range, instance); + ublock.color_component_selector = extract(block->color_component_selector, instance); + ublock.partition_id = extract(block->partition_id, instance); + ublock.endpoint_range = extract(block->endpoint_range, instance); + ublock.color_endpoint_modes[0] = extract(block->color_endpoint_modes[0], instance); + + uniform int num_weights = block->width * block->height * (block->dual_plane ? 2 : 1); + for (uniform int i = 0; i < num_weights; i++) + ublock.weights[i] = extract(block->weights[i], instance); + + for (uniform int i = 0; i < 8; i++) + ublock.endpoints[i] = extract(block->endpoints[i], instance); + + uniform uint32_t data[4]; + pack_block_c(data, &ublock); + + for (uniform int i = 0; i < 4; i++) state->data[i] = insert(state->data[i], instance, data[i]); + } +} + +int get_bits(uint32_t value, uniform int from, uniform int to) +{ + return (value >> from) & ((1 << (to + 1 - from)) - 1); +} + +void load_block_parameters(astc_block block[], uint32_t mode, uniform astc_enc_context ctx[]) +{ + // uniform parameters + block->width = ctx->width; + block->height = ctx->height; + block->dual_plane = ctx->dual_plane; + block->partitions = ctx->partitions; + block->color_endpoint_pairs = ctx->color_endpoint_pairs; + block->channels = ctx->channels; + + // varying parameters + block->weight_range = get_bits(mode, 0, 3); // 0..11 <= 2^4 + block->color_component_selector = get_bits(mode, 4, 5); // 0..2 <= 2^2 + block->partition_id = 0; + block->color_endpoint_modes[0] = get_bits(mode, 6, 7) * 2 + 6; // 6, 8, 10 or 12 + block->endpoint_range = get_bits(mode, 8, 12); // 0..20 <= 2^5 +} + +export void astc_encode_ispc(uniform rgba_surface src[], uniform float block_scores[], uniform uint8_t dst[], uniform uint64_t list[], uniform astc_enc_context list_context[], uniform astc_enc_settings settings[]) +{ + uint64_t entry = list[programIndex]; + uint32_t offset = entry >> 32; + uint32_t mode = (entry & 0xFFFFFFFF); + if (mode == 0) return; + int yy = offset >> 16; + int xx = offset & 0xFFFF; + + int tex_width = src->width / settings->block_width; + + astc_enc_state _state; + varying astc_enc_state* uniform state = &_state; + + state->block_width = settings->block_width; + state->block_height = settings->block_height; + state->refineIterations = settings->refineIterations; + + load_block_interleaved(state->pixels, src, xx, yy, state->block_width, state->block_height); + + astc_block _block; + varying astc_block* uniform block = &_block; + + load_block_parameters(block, mode, list_context); + + scale_pixels(state, list_context); + if (block->channels == 3) clear_alpha(state->scaled_pixels, block->width, block->height); + + if (block->dual_plane) + { + pixel_set pset; + pset.pixels = state->scaled_pixels; + pset.width = block->width; + pset.height = block->height; + + rotate_plane(&pset, block->color_component_selector); + } + + optimize_block(state->scaled_pixels, block, state); + float error = measure_error(block, state); + + if (error < gather_float(block_scores, yy * tex_width + xx)) + { + pack_block(block, state); + + scatter_float(block_scores, yy * tex_width + xx, error); + + for (uniform int i = 0; i < 4; i++) + scatter_uint((uint32_t*)dst, (yy * tex_width + xx) * 4 + i, state->data[i]); + } +} diff --git a/Source/meson.build b/Source/meson.build new file mode 100644 index 0000000..cd61c27 --- /dev/null +++ b/Source/meson.build @@ -0,0 +1,26 @@ +sources = files([ + 'createdfd.cpp', + 'HalfFloat.cpp', + 'Main.cpp', + 'stb_image_resize.cpp', + 'stb_image.cpp', + 'vk2dfd.cpp' +]) + +incdirs = include_directories([ +]) + +dependencies = [ +] + +ispc_kernel = custom_target('ipsc_kernel', input: ['ispc_texcomp/kernel.ispc'], output: ['kernel_ispc.o', 'kernel_ispc_avx2.o', 'kernel_ispc_sse4.o', 'kernel_ispc.h'], command: ['ispc', '-O3', '--arch=x86_64', '--target=sse4,avx2', '--opt=fast-math', '--pic', '@INPUT@', '-h', '@OUTDIR@/kernel_ispc.h', '-o', '@OUTPUT0@']) + +ispc_sources = [ + ispc_kernel, + 'ispc_texcomp/ispc_texcomp.cpp', + 'ispc_texcomp/ispc_texcomp.h' +] + +ispc_texcomp = static_library('ispc_texcomp', ispc_sources) + +executable('TextureTaffy', sources, dependencies: dependencies, include_directories: incdirs, install: true, install_dir: '', install_tag: 'exe', link_with: ispc_texcomp) \ No newline at end of file diff --git a/Source/stb_image.cpp b/Source/stb_image.cpp new file mode 100644 index 0000000..badb3ef --- /dev/null +++ b/Source/stb_image.cpp @@ -0,0 +1,2 @@ +#define STB_IMAGE_IMPLEMENTATION +#include "stb_image.h" \ No newline at end of file diff --git a/Source/stb_image.h b/Source/stb_image.h new file mode 100644 index 0000000..5e807a0 --- /dev/null +++ b/Source/stb_image.h @@ -0,0 +1,7987 @@ +/* stb_image - v2.28 - public domain image loader - http://nothings.org/stb + no warranty implied; use at your own risk + + Do this: + #define STB_IMAGE_IMPLEMENTATION + before you include this file in *one* C or C++ file to create the implementation. + + // i.e. it should look like this: + #include ... + #include ... + #include ... + #define STB_IMAGE_IMPLEMENTATION + #include "stb_image.h" + + You can #define STBI_ASSERT(x) before the #include to avoid using assert.h. + And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free + + + QUICK NOTES: + Primarily of interest to game developers and other people who can + avoid problematic images and only need the trivial interface + + JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib) + PNG 1/2/4/8/16-bit-per-channel + + TGA (not sure what subset, if a subset) + BMP non-1bpp, non-RLE + PSD (composited view only, no extra channels, 8/16 bit-per-channel) + + GIF (*comp always reports as 4-channel) + HDR (radiance rgbE format) + PIC (Softimage PIC) + PNM (PPM and PGM binary only) + + Animated GIF still needs a proper API, but here's one way to do it: + http://gist.github.com/urraka/685d9a6340b26b830d49 + + - decode from memory or through FILE (define STBI_NO_STDIO to remove code) + - decode from arbitrary I/O callbacks + - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON) + + Full documentation under "DOCUMENTATION" below. + + +LICENSE + + See end of file for license information. + +RECENT REVISION HISTORY: + + 2.28 (2023-01-29) many error fixes, security errors, just tons of stuff + 2.27 (2021-07-11) document stbi_info better, 16-bit PNM support, bug fixes + 2.26 (2020-07-13) many minor fixes + 2.25 (2020-02-02) fix warnings + 2.24 (2020-02-02) fix warnings; thread-local failure_reason and flip_vertically + 2.23 (2019-08-11) fix clang static analysis warning + 2.22 (2019-03-04) gif fixes, fix warnings + 2.21 (2019-02-25) fix typo in comment + 2.20 (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs + 2.19 (2018-02-11) fix warning + 2.18 (2018-01-30) fix warnings + 2.17 (2018-01-29) bugfix, 1-bit BMP, 16-bitness query, fix warnings + 2.16 (2017-07-23) all functions have 16-bit variants; optimizations; bugfixes + 2.15 (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC + 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs + 2.13 (2016-12-04) experimental 16-bit API, only for PNG so far; fixes + 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes + 2.11 (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64 + RGB-format JPEG; remove white matting in PSD; + allocate large structures on the stack; + correct channel count for PNG & BMP + 2.10 (2016-01-22) avoid warning introduced in 2.09 + 2.09 (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED + + See end of file for full revision history. + + + ============================ Contributors ========================= + + Image formats Extensions, features + Sean Barrett (jpeg, png, bmp) Jetro Lauha (stbi_info) + Nicolas Schulz (hdr, psd) Martin "SpartanJ" Golini (stbi_info) + Jonathan Dummer (tga) James "moose2000" Brown (iPhone PNG) + Jean-Marc Lienher (gif) Ben "Disch" Wenger (io callbacks) + Tom Seddon (pic) Omar Cornut (1/2/4-bit PNG) + Thatcher Ulrich (psd) Nicolas Guillemot (vertical flip) + Ken Miller (pgm, ppm) Richard Mitton (16-bit PSD) + github:urraka (animated gif) Junggon Kim (PNM comments) + Christopher Forseth (animated gif) Daniel Gibson (16-bit TGA) + socks-the-fox (16-bit PNG) + Jeremy Sawicki (handle all ImageNet JPGs) + Optimizations & bugfixes Mikhail Morozov (1-bit BMP) + Fabian "ryg" Giesen Anael Seghezzi (is-16-bit query) + Arseny Kapoulkine Simon Breuss (16-bit PNM) + John-Mark Allen + Carmelo J Fdez-Aguera + + Bug & warning fixes + Marc LeBlanc David Woo Guillaume George Martins Mozeiko + Christpher Lloyd Jerry Jansson Joseph Thomson Blazej Dariusz Roszkowski + Phil Jordan Dave Moore Roy Eltham + Hayaki Saito Nathan Reed Won Chun + Luke Graham Johan Duparc Nick Verigakis the Horde3D community + Thomas Ruf Ronny Chevalier github:rlyeh + Janez Zemva John Bartholomew Michal Cichon github:romigrou + Jonathan Blow Ken Hamada Tero Hanninen github:svdijk + Eugene Golushkov Laurent Gomila Cort Stratton github:snagar + Aruelien Pocheville Sergio Gonzalez Thibault Reuille github:Zelex + Cass Everitt Ryamond Barbiero github:grim210 + Paul Du Bois Engin Manap Aldo Culquicondor github:sammyhw + Philipp Wiesemann Dale Weiler Oriol Ferrer Mesia github:phprus + Josh Tobin Neil Bickford Matthew Gregan github:poppolopoppo + Julian Raschke Gregory Mullen Christian Floisand github:darealshinji + Baldur Karlsson Kevin Schmidt JR Smith github:Michaelangel007 + Brad Weinberger Matvey Cherevko github:mosra + Luca Sas Alexander Veselov Zack Middleton [reserved] + Ryan C. Gordon [reserved] [reserved] + DO NOT ADD YOUR NAME HERE + + Jacko Dirks + + To add your name to the credits, pick a random blank space in the middle and fill it. + 80% of merge conflicts on stb PRs are due to people adding their name at the end + of the credits. +*/ + +#ifndef STBI_INCLUDE_STB_IMAGE_H +#define STBI_INCLUDE_STB_IMAGE_H + +// DOCUMENTATION +// +// Limitations: +// - no 12-bit-per-channel JPEG +// - no JPEGs with arithmetic coding +// - GIF always returns *comp=4 +// +// Basic usage (see HDR discussion below for HDR usage): +// int x,y,n; +// unsigned char *data = stbi_load(filename, &x, &y, &n, 0); +// // ... process data if not NULL ... +// // ... x = width, y = height, n = # 8-bit components per pixel ... +// // ... replace '0' with '1'..'4' to force that many components per pixel +// // ... but 'n' will always be the number that it would have been if you said 0 +// stbi_image_free(data); +// +// Standard parameters: +// int *x -- outputs image width in pixels +// int *y -- outputs image height in pixels +// int *channels_in_file -- outputs # of image components in image file +// int desired_channels -- if non-zero, # of image components requested in result +// +// The return value from an image loader is an 'unsigned char *' which points +// to the pixel data, or NULL on an allocation failure or if the image is +// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels, +// with each pixel consisting of N interleaved 8-bit components; the first +// pixel pointed to is top-left-most in the image. There is no padding between +// image scanlines or between pixels, regardless of format. The number of +// components N is 'desired_channels' if desired_channels is non-zero, or +// *channels_in_file otherwise. If desired_channels is non-zero, +// *channels_in_file has the number of components that _would_ have been +// output otherwise. E.g. if you set desired_channels to 4, you will always +// get RGBA output, but you can check *channels_in_file to see if it's trivially +// opaque because e.g. there were only 3 channels in the source image. +// +// An output image with N components has the following components interleaved +// in this order in each pixel: +// +// N=#comp components +// 1 grey +// 2 grey, alpha +// 3 red, green, blue +// 4 red, green, blue, alpha +// +// If image loading fails for any reason, the return value will be NULL, +// and *x, *y, *channels_in_file will be unchanged. The function +// stbi_failure_reason() can be queried for an extremely brief, end-user +// unfriendly explanation of why the load failed. Define STBI_NO_FAILURE_STRINGS +// to avoid compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly +// more user-friendly ones. +// +// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized. +// +// To query the width, height and component count of an image without having to +// decode the full file, you can use the stbi_info family of functions: +// +// int x,y,n,ok; +// ok = stbi_info(filename, &x, &y, &n); +// // returns ok=1 and sets x, y, n if image is a supported format, +// // 0 otherwise. +// +// Note that stb_image pervasively uses ints in its public API for sizes, +// including sizes of memory buffers. This is now part of the API and thus +// hard to change without causing breakage. As a result, the various image +// loaders all have certain limits on image size; these differ somewhat +// by format but generally boil down to either just under 2GB or just under +// 1GB. When the decoded image would be larger than this, stb_image decoding +// will fail. +// +// Additionally, stb_image will reject image files that have any of their +// dimensions set to a larger value than the configurable STBI_MAX_DIMENSIONS, +// which defaults to 2**24 = 16777216 pixels. Due to the above memory limit, +// the only way to have an image with such dimensions load correctly +// is for it to have a rather extreme aspect ratio. Either way, the +// assumption here is that such larger images are likely to be malformed +// or malicious. If you do need to load an image with individual dimensions +// larger than that, and it still fits in the overall size limit, you can +// #define STBI_MAX_DIMENSIONS on your own to be something larger. +// +// =========================================================================== +// +// UNICODE: +// +// If compiling for Windows and you wish to use Unicode filenames, compile +// with +// #define STBI_WINDOWS_UTF8 +// and pass utf8-encoded filenames. Call stbi_convert_wchar_to_utf8 to convert +// Windows wchar_t filenames to utf8. +// +// =========================================================================== +// +// Philosophy +// +// stb libraries are designed with the following priorities: +// +// 1. easy to use +// 2. easy to maintain +// 3. good performance +// +// Sometimes I let "good performance" creep up in priority over "easy to maintain", +// and for best performance I may provide less-easy-to-use APIs that give higher +// performance, in addition to the easy-to-use ones. Nevertheless, it's important +// to keep in mind that from the standpoint of you, a client of this library, +// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all. +// +// Some secondary priorities arise directly from the first two, some of which +// provide more explicit reasons why performance can't be emphasized. +// +// - Portable ("ease of use") +// - Small source code footprint ("easy to maintain") +// - No dependencies ("ease of use") +// +// =========================================================================== +// +// I/O callbacks +// +// I/O callbacks allow you to read from arbitrary sources, like packaged +// files or some other source. Data read from callbacks are processed +// through a small internal buffer (currently 128 bytes) to try to reduce +// overhead. +// +// The three functions you must define are "read" (reads some bytes of data), +// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end). +// +// =========================================================================== +// +// SIMD support +// +// The JPEG decoder will try to automatically use SIMD kernels on x86 when +// supported by the compiler. For ARM Neon support, you must explicitly +// request it. +// +// (The old do-it-yourself SIMD API is no longer supported in the current +// code.) +// +// On x86, SSE2 will automatically be used when available based on a run-time +// test; if not, the generic C versions are used as a fall-back. On ARM targets, +// the typical path is to have separate builds for NEON and non-NEON devices +// (at least this is true for iOS and Android). Therefore, the NEON support is +// toggled by a build flag: define STBI_NEON to get NEON loops. +// +// If for some reason you do not want to use any of SIMD code, or if +// you have issues compiling it, you can disable it entirely by +// defining STBI_NO_SIMD. +// +// =========================================================================== +// +// HDR image support (disable by defining STBI_NO_HDR) +// +// stb_image supports loading HDR images in general, and currently the Radiance +// .HDR file format specifically. You can still load any file through the existing +// interface; if you attempt to load an HDR file, it will be automatically remapped +// to LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1; +// both of these constants can be reconfigured through this interface: +// +// stbi_hdr_to_ldr_gamma(2.2f); +// stbi_hdr_to_ldr_scale(1.0f); +// +// (note, do not use _inverse_ constants; stbi_image will invert them +// appropriately). +// +// Additionally, there is a new, parallel interface for loading files as +// (linear) floats to preserve the full dynamic range: +// +// float *data = stbi_loadf(filename, &x, &y, &n, 0); +// +// If you load LDR images through this interface, those images will +// be promoted to floating point values, run through the inverse of +// constants corresponding to the above: +// +// stbi_ldr_to_hdr_scale(1.0f); +// stbi_ldr_to_hdr_gamma(2.2f); +// +// Finally, given a filename (or an open file or memory block--see header +// file for details) containing image data, you can query for the "most +// appropriate" interface to use (that is, whether the image is HDR or +// not), using: +// +// stbi_is_hdr(char *filename); +// +// =========================================================================== +// +// iPhone PNG support: +// +// We optionally support converting iPhone-formatted PNGs (which store +// premultiplied BGRA) back to RGB, even though they're internally encoded +// differently. To enable this conversion, call +// stbi_convert_iphone_png_to_rgb(1). +// +// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per +// pixel to remove any premultiplied alpha *only* if the image file explicitly +// says there's premultiplied data (currently only happens in iPhone images, +// and only if iPhone convert-to-rgb processing is on). +// +// =========================================================================== +// +// ADDITIONAL CONFIGURATION +// +// - You can suppress implementation of any of the decoders to reduce +// your code footprint by #defining one or more of the following +// symbols before creating the implementation. +// +// STBI_NO_JPEG +// STBI_NO_PNG +// STBI_NO_BMP +// STBI_NO_PSD +// STBI_NO_TGA +// STBI_NO_GIF +// STBI_NO_HDR +// STBI_NO_PIC +// STBI_NO_PNM (.ppm and .pgm) +// +// - You can request *only* certain decoders and suppress all other ones +// (this will be more forward-compatible, as addition of new decoders +// doesn't require you to disable them explicitly): +// +// STBI_ONLY_JPEG +// STBI_ONLY_PNG +// STBI_ONLY_BMP +// STBI_ONLY_PSD +// STBI_ONLY_TGA +// STBI_ONLY_GIF +// STBI_ONLY_HDR +// STBI_ONLY_PIC +// STBI_ONLY_PNM (.ppm and .pgm) +// +// - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still +// want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB +// +// - If you define STBI_MAX_DIMENSIONS, stb_image will reject images greater +// than that size (in either width or height) without further processing. +// This is to let programs in the wild set an upper bound to prevent +// denial-of-service attacks on untrusted data, as one could generate a +// valid image of gigantic dimensions and force stb_image to allocate a +// huge block of memory and spend disproportionate time decoding it. By +// default this is set to (1 << 24), which is 16777216, but that's still +// very big. + +#ifndef STBI_NO_STDIO +#include +#endif // STBI_NO_STDIO + +#define STBI_VERSION 1 + +enum +{ + STBI_default = 0, // only used for desired_channels + + STBI_grey = 1, + STBI_grey_alpha = 2, + STBI_rgb = 3, + STBI_rgb_alpha = 4 +}; + +#include +typedef unsigned char stbi_uc; +typedef unsigned short stbi_us; + +#ifdef __cplusplus +extern "C" { +#endif + +#ifndef STBIDEF +#ifdef STB_IMAGE_STATIC +#define STBIDEF static +#else +#define STBIDEF extern +#endif +#endif + +////////////////////////////////////////////////////////////////////////////// +// +// PRIMARY API - works on images of any type +// + +// +// load image by filename, open file, or memory buffer +// + +typedef struct +{ + int (*read) (void *user,char *data,int size); // fill 'data' with 'size' bytes. return number of bytes actually read + void (*skip) (void *user,int n); // skip the next 'n' bytes, or 'unget' the last -n bytes if negative + int (*eof) (void *user); // returns nonzero if we are at end of file/data +} stbi_io_callbacks; + +//////////////////////////////////// +// +// 8-bits-per-channel interface +// + +STBIDEF stbi_uc *stbi_load_from_memory (stbi_uc const *buffer, int len , int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk , void *user, int *x, int *y, int *channels_in_file, int desired_channels); + +#ifndef STBI_NO_STDIO +STBIDEF stbi_uc *stbi_load (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_uc *stbi_load_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels); +// for stbi_load_from_file, file pointer is left pointing immediately after image +#endif + +#ifndef STBI_NO_GIF +STBIDEF stbi_uc *stbi_load_gif_from_memory(stbi_uc const *buffer, int len, int **delays, int *x, int *y, int *z, int *comp, int req_comp); +#endif + +#ifdef STBI_WINDOWS_UTF8 +STBIDEF int stbi_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input); +#endif + +//////////////////////////////////// +// +// 16-bits-per-channel interface +// + +STBIDEF stbi_us *stbi_load_16_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_us *stbi_load_16_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels); + +#ifndef STBI_NO_STDIO +STBIDEF stbi_us *stbi_load_16 (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_us *stbi_load_from_file_16(FILE *f, int *x, int *y, int *channels_in_file, int desired_channels); +#endif + +//////////////////////////////////// +// +// float-per-channel interface +// +#ifndef STBI_NO_LINEAR + STBIDEF float *stbi_loadf_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels); + STBIDEF float *stbi_loadf_from_callbacks (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels); + + #ifndef STBI_NO_STDIO + STBIDEF float *stbi_loadf (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels); + STBIDEF float *stbi_loadf_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels); + #endif +#endif + +#ifndef STBI_NO_HDR + STBIDEF void stbi_hdr_to_ldr_gamma(float gamma); + STBIDEF void stbi_hdr_to_ldr_scale(float scale); +#endif // STBI_NO_HDR + +#ifndef STBI_NO_LINEAR + STBIDEF void stbi_ldr_to_hdr_gamma(float gamma); + STBIDEF void stbi_ldr_to_hdr_scale(float scale); +#endif // STBI_NO_LINEAR + +// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR +STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user); +STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len); +#ifndef STBI_NO_STDIO +STBIDEF int stbi_is_hdr (char const *filename); +STBIDEF int stbi_is_hdr_from_file(FILE *f); +#endif // STBI_NO_STDIO + + +// get a VERY brief reason for failure +// on most compilers (and ALL modern mainstream compilers) this is threadsafe +STBIDEF const char *stbi_failure_reason (void); + +// free the loaded image -- this is just free() +STBIDEF void stbi_image_free (void *retval_from_stbi_load); + +// get image dimensions & components without fully decoding +STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp); +STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp); +STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const *buffer, int len); +STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const *clbk, void *user); + +#ifndef STBI_NO_STDIO +STBIDEF int stbi_info (char const *filename, int *x, int *y, int *comp); +STBIDEF int stbi_info_from_file (FILE *f, int *x, int *y, int *comp); +STBIDEF int stbi_is_16_bit (char const *filename); +STBIDEF int stbi_is_16_bit_from_file(FILE *f); +#endif + + + +// for image formats that explicitly notate that they have premultiplied alpha, +// we just return the colors as stored in the file. set this flag to force +// unpremultiplication. results are undefined if the unpremultiply overflow. +STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply); + +// indicate whether we should process iphone images back to canonical format, +// or just pass them through "as-is" +STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert); + +// flip the image vertically, so the first pixel in the output array is the bottom left +STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip); + +// as above, but only applies to images loaded on the thread that calls the function +// this function is only available if your compiler supports thread-local variables; +// calling it will fail to link if your compiler doesn't +STBIDEF void stbi_set_unpremultiply_on_load_thread(int flag_true_if_should_unpremultiply); +STBIDEF void stbi_convert_iphone_png_to_rgb_thread(int flag_true_if_should_convert); +STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip); + +// ZLIB client - used by PNG, available for other purposes + +STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen); +STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header); +STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen); +STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen); + +STBIDEF char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen); +STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen); + + +#ifdef __cplusplus +} +#endif + +// +// +//// end header file ///////////////////////////////////////////////////// +#endif // STBI_INCLUDE_STB_IMAGE_H + +#ifdef STB_IMAGE_IMPLEMENTATION + +#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \ + || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \ + || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \ + || defined(STBI_ONLY_ZLIB) + #ifndef STBI_ONLY_JPEG + #define STBI_NO_JPEG + #endif + #ifndef STBI_ONLY_PNG + #define STBI_NO_PNG + #endif + #ifndef STBI_ONLY_BMP + #define STBI_NO_BMP + #endif + #ifndef STBI_ONLY_PSD + #define STBI_NO_PSD + #endif + #ifndef STBI_ONLY_TGA + #define STBI_NO_TGA + #endif + #ifndef STBI_ONLY_GIF + #define STBI_NO_GIF + #endif + #ifndef STBI_ONLY_HDR + #define STBI_NO_HDR + #endif + #ifndef STBI_ONLY_PIC + #define STBI_NO_PIC + #endif + #ifndef STBI_ONLY_PNM + #define STBI_NO_PNM + #endif +#endif + +#if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB) +#define STBI_NO_ZLIB +#endif + + +#include +#include // ptrdiff_t on osx +#include +#include +#include + +#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) +#include // ldexp, pow +#endif + +#ifndef STBI_NO_STDIO +#include +#endif + +#ifndef STBI_ASSERT +#include +#define STBI_ASSERT(x) assert(x) +#endif + +#ifdef __cplusplus +#define STBI_EXTERN extern "C" +#else +#define STBI_EXTERN extern +#endif + + +#ifndef _MSC_VER + #ifdef __cplusplus + #define stbi_inline inline + #else + #define stbi_inline + #endif +#else + #define stbi_inline __forceinline +#endif + +#ifndef STBI_NO_THREAD_LOCALS + #if defined(__cplusplus) && __cplusplus >= 201103L + #define STBI_THREAD_LOCAL thread_local + #elif defined(__GNUC__) && __GNUC__ < 5 + #define STBI_THREAD_LOCAL __thread + #elif defined(_MSC_VER) + #define STBI_THREAD_LOCAL __declspec(thread) + #elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 201112L && !defined(__STDC_NO_THREADS__) + #define STBI_THREAD_LOCAL _Thread_local + #endif + + #ifndef STBI_THREAD_LOCAL + #if defined(__GNUC__) + #define STBI_THREAD_LOCAL __thread + #endif + #endif +#endif + +#if defined(_MSC_VER) || defined(__SYMBIAN32__) +typedef unsigned short stbi__uint16; +typedef signed short stbi__int16; +typedef unsigned int stbi__uint32; +typedef signed int stbi__int32; +#else +#include +typedef uint16_t stbi__uint16; +typedef int16_t stbi__int16; +typedef uint32_t stbi__uint32; +typedef int32_t stbi__int32; +#endif + +// should produce compiler error if size is wrong +typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1]; + +#ifdef _MSC_VER +#define STBI_NOTUSED(v) (void)(v) +#else +#define STBI_NOTUSED(v) (void)sizeof(v) +#endif + +#ifdef _MSC_VER +#define STBI_HAS_LROTL +#endif + +#ifdef STBI_HAS_LROTL + #define stbi_lrot(x,y) _lrotl(x,y) +#else + #define stbi_lrot(x,y) (((x) << (y)) | ((x) >> (-(y) & 31))) +#endif + +#if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED)) +// ok +#elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED) +// ok +#else +#error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)." +#endif + +#ifndef STBI_MALLOC +#define STBI_MALLOC(sz) malloc(sz) +#define STBI_REALLOC(p,newsz) realloc(p,newsz) +#define STBI_FREE(p) free(p) +#endif + +#ifndef STBI_REALLOC_SIZED +#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz) +#endif + +// x86/x64 detection +#if defined(__x86_64__) || defined(_M_X64) +#define STBI__X64_TARGET +#elif defined(__i386) || defined(_M_IX86) +#define STBI__X86_TARGET +#endif + +#if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD) +// gcc doesn't support sse2 intrinsics unless you compile with -msse2, +// which in turn means it gets to use SSE2 everywhere. This is unfortunate, +// but previous attempts to provide the SSE2 functions with runtime +// detection caused numerous issues. The way architecture extensions are +// exposed in GCC/Clang is, sadly, not really suited for one-file libs. +// New behavior: if compiled with -msse2, we use SSE2 without any +// detection; if not, we don't use it at all. +#define STBI_NO_SIMD +#endif + +#if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD) +// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET +// +// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the +// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant. +// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not +// simultaneously enabling "-mstackrealign". +// +// See https://github.com/nothings/stb/issues/81 for more information. +// +// So default to no SSE2 on 32-bit MinGW. If you've read this far and added +// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2. +#define STBI_NO_SIMD +#endif + +#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET)) +#define STBI_SSE2 +#include + +#ifdef _MSC_VER + +#if _MSC_VER >= 1400 // not VC6 +#include // __cpuid +static int stbi__cpuid3(void) +{ + int info[4]; + __cpuid(info,1); + return info[3]; +} +#else +static int stbi__cpuid3(void) +{ + int res; + __asm { + mov eax,1 + cpuid + mov res,edx + } + return res; +} +#endif + +#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name + +#if !defined(STBI_NO_JPEG) && defined(STBI_SSE2) +static int stbi__sse2_available(void) +{ + int info3 = stbi__cpuid3(); + return ((info3 >> 26) & 1) != 0; +} +#endif + +#else // assume GCC-style if not VC++ +#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16))) + +#if !defined(STBI_NO_JPEG) && defined(STBI_SSE2) +static int stbi__sse2_available(void) +{ + // If we're even attempting to compile this on GCC/Clang, that means + // -msse2 is on, which means the compiler is allowed to use SSE2 + // instructions at will, and so are we. + return 1; +} +#endif + +#endif +#endif + +// ARM NEON +#if defined(STBI_NO_SIMD) && defined(STBI_NEON) +#undef STBI_NEON +#endif + +#ifdef STBI_NEON +#include +#ifdef _MSC_VER +#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name +#else +#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16))) +#endif +#endif + +#ifndef STBI_SIMD_ALIGN +#define STBI_SIMD_ALIGN(type, name) type name +#endif + +#ifndef STBI_MAX_DIMENSIONS +#define STBI_MAX_DIMENSIONS (1 << 24) +#endif + +/////////////////////////////////////////////// +// +// stbi__context struct and start_xxx functions + +// stbi__context structure is our basic context used by all images, so it +// contains all the IO context, plus some basic image information +typedef struct +{ + stbi__uint32 img_x, img_y; + int img_n, img_out_n; + + stbi_io_callbacks io; + void *io_user_data; + + int read_from_callbacks; + int buflen; + stbi_uc buffer_start[128]; + int callback_already_read; + + stbi_uc *img_buffer, *img_buffer_end; + stbi_uc *img_buffer_original, *img_buffer_original_end; +} stbi__context; + + +static void stbi__refill_buffer(stbi__context *s); + +// initialize a memory-decode context +static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len) +{ + s->io.read = NULL; + s->read_from_callbacks = 0; + s->callback_already_read = 0; + s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer; + s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len; +} + +// initialize a callback-based context +static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user) +{ + s->io = *c; + s->io_user_data = user; + s->buflen = sizeof(s->buffer_start); + s->read_from_callbacks = 1; + s->callback_already_read = 0; + s->img_buffer = s->img_buffer_original = s->buffer_start; + stbi__refill_buffer(s); + s->img_buffer_original_end = s->img_buffer_end; +} + +#ifndef STBI_NO_STDIO + +static int stbi__stdio_read(void *user, char *data, int size) +{ + return (int) fread(data,1,size,(FILE*) user); +} + +static void stbi__stdio_skip(void *user, int n) +{ + int ch; + fseek((FILE*) user, n, SEEK_CUR); + ch = fgetc((FILE*) user); /* have to read a byte to reset feof()'s flag */ + if (ch != EOF) { + ungetc(ch, (FILE *) user); /* push byte back onto stream if valid. */ + } +} + +static int stbi__stdio_eof(void *user) +{ + return feof((FILE*) user) || ferror((FILE *) user); +} + +static stbi_io_callbacks stbi__stdio_callbacks = +{ + stbi__stdio_read, + stbi__stdio_skip, + stbi__stdio_eof, +}; + +static void stbi__start_file(stbi__context *s, FILE *f) +{ + stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f); +} + +//static void stop_file(stbi__context *s) { } + +#endif // !STBI_NO_STDIO + +static void stbi__rewind(stbi__context *s) +{ + // conceptually rewind SHOULD rewind to the beginning of the stream, + // but we just rewind to the beginning of the initial buffer, because + // we only use it after doing 'test', which only ever looks at at most 92 bytes + s->img_buffer = s->img_buffer_original; + s->img_buffer_end = s->img_buffer_original_end; +} + +enum +{ + STBI_ORDER_RGB, + STBI_ORDER_BGR +}; + +typedef struct +{ + int bits_per_channel; + int num_channels; + int channel_order; +} stbi__result_info; + +#ifndef STBI_NO_JPEG +static int stbi__jpeg_test(stbi__context *s); +static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PNG +static int stbi__png_test(stbi__context *s); +static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp); +static int stbi__png_is16(stbi__context *s); +#endif + +#ifndef STBI_NO_BMP +static int stbi__bmp_test(stbi__context *s); +static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_TGA +static int stbi__tga_test(stbi__context *s); +static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PSD +static int stbi__psd_test(stbi__context *s); +static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc); +static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp); +static int stbi__psd_is16(stbi__context *s); +#endif + +#ifndef STBI_NO_HDR +static int stbi__hdr_test(stbi__context *s); +static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PIC +static int stbi__pic_test(stbi__context *s); +static void *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_GIF +static int stbi__gif_test(stbi__context *s); +static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static void *stbi__load_gif_main(stbi__context *s, int **delays, int *x, int *y, int *z, int *comp, int req_comp); +static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PNM +static int stbi__pnm_test(stbi__context *s); +static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp); +static int stbi__pnm_is16(stbi__context *s); +#endif + +static +#ifdef STBI_THREAD_LOCAL +STBI_THREAD_LOCAL +#endif +const char *stbi__g_failure_reason; + +STBIDEF const char *stbi_failure_reason(void) +{ + return stbi__g_failure_reason; +} + +#ifndef STBI_NO_FAILURE_STRINGS +static int stbi__err(const char *str) +{ + stbi__g_failure_reason = str; + return 0; +} +#endif + +static void *stbi__malloc(size_t size) +{ + return STBI_MALLOC(size); +} + +// stb_image uses ints pervasively, including for offset calculations. +// therefore the largest decoded image size we can support with the +// current code, even on 64-bit targets, is INT_MAX. this is not a +// significant limitation for the intended use case. +// +// we do, however, need to make sure our size calculations don't +// overflow. hence a few helper functions for size calculations that +// multiply integers together, making sure that they're non-negative +// and no overflow occurs. + +// return 1 if the sum is valid, 0 on overflow. +// negative terms are considered invalid. +static int stbi__addsizes_valid(int a, int b) +{ + if (b < 0) return 0; + // now 0 <= b <= INT_MAX, hence also + // 0 <= INT_MAX - b <= INTMAX. + // And "a + b <= INT_MAX" (which might overflow) is the + // same as a <= INT_MAX - b (no overflow) + return a <= INT_MAX - b; +} + +// returns 1 if the product is valid, 0 on overflow. +// negative factors are considered invalid. +static int stbi__mul2sizes_valid(int a, int b) +{ + if (a < 0 || b < 0) return 0; + if (b == 0) return 1; // mul-by-0 is always safe + // portable way to check for no overflows in a*b + return a <= INT_MAX/b; +} + +#if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR) +// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow +static int stbi__mad2sizes_valid(int a, int b, int add) +{ + return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a*b, add); +} +#endif + +// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow +static int stbi__mad3sizes_valid(int a, int b, int c, int add) +{ + return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) && + stbi__addsizes_valid(a*b*c, add); +} + +// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow +#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) || !defined(STBI_NO_PNM) +static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add) +{ + return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) && + stbi__mul2sizes_valid(a*b*c, d) && stbi__addsizes_valid(a*b*c*d, add); +} +#endif + +#if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR) +// mallocs with size overflow checking +static void *stbi__malloc_mad2(int a, int b, int add) +{ + if (!stbi__mad2sizes_valid(a, b, add)) return NULL; + return stbi__malloc(a*b + add); +} +#endif + +static void *stbi__malloc_mad3(int a, int b, int c, int add) +{ + if (!stbi__mad3sizes_valid(a, b, c, add)) return NULL; + return stbi__malloc(a*b*c + add); +} + +#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) || !defined(STBI_NO_PNM) +static void *stbi__malloc_mad4(int a, int b, int c, int d, int add) +{ + if (!stbi__mad4sizes_valid(a, b, c, d, add)) return NULL; + return stbi__malloc(a*b*c*d + add); +} +#endif + +// returns 1 if the sum of two signed ints is valid (between -2^31 and 2^31-1 inclusive), 0 on overflow. +static int stbi__addints_valid(int a, int b) +{ + if ((a >= 0) != (b >= 0)) return 1; // a and b have different signs, so no overflow + if (a < 0 && b < 0) return a >= INT_MIN - b; // same as a + b >= INT_MIN; INT_MIN - b cannot overflow since b < 0. + return a <= INT_MAX - b; +} + +// returns 1 if the product of two signed shorts is valid, 0 on overflow. +static int stbi__mul2shorts_valid(short a, short b) +{ + if (b == 0 || b == -1) return 1; // multiplication by 0 is always 0; check for -1 so SHRT_MIN/b doesn't overflow + if ((a >= 0) == (b >= 0)) return a <= SHRT_MAX/b; // product is positive, so similar to mul2sizes_valid + if (b < 0) return a <= SHRT_MIN / b; // same as a * b >= SHRT_MIN + return a >= SHRT_MIN / b; +} + +// stbi__err - error +// stbi__errpf - error returning pointer to float +// stbi__errpuc - error returning pointer to unsigned char + +#ifdef STBI_NO_FAILURE_STRINGS + #define stbi__err(x,y) 0 +#elif defined(STBI_FAILURE_USERMSG) + #define stbi__err(x,y) stbi__err(y) +#else + #define stbi__err(x,y) stbi__err(x) +#endif + +#define stbi__errpf(x,y) ((float *)(size_t) (stbi__err(x,y)?NULL:NULL)) +#define stbi__errpuc(x,y) ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL)) + +STBIDEF void stbi_image_free(void *retval_from_stbi_load) +{ + STBI_FREE(retval_from_stbi_load); +} + +#ifndef STBI_NO_LINEAR +static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp); +#endif + +#ifndef STBI_NO_HDR +static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp); +#endif + +static int stbi__vertically_flip_on_load_global = 0; + +STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip) +{ + stbi__vertically_flip_on_load_global = flag_true_if_should_flip; +} + +#ifndef STBI_THREAD_LOCAL +#define stbi__vertically_flip_on_load stbi__vertically_flip_on_load_global +#else +static STBI_THREAD_LOCAL int stbi__vertically_flip_on_load_local, stbi__vertically_flip_on_load_set; + +STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip) +{ + stbi__vertically_flip_on_load_local = flag_true_if_should_flip; + stbi__vertically_flip_on_load_set = 1; +} + +#define stbi__vertically_flip_on_load (stbi__vertically_flip_on_load_set \ + ? stbi__vertically_flip_on_load_local \ + : stbi__vertically_flip_on_load_global) +#endif // STBI_THREAD_LOCAL + +static void *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc) +{ + memset(ri, 0, sizeof(*ri)); // make sure it's initialized if we add new fields + ri->bits_per_channel = 8; // default is 8 so most paths don't have to be changed + ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order + ri->num_channels = 0; + + // test the formats with a very explicit header first (at least a FOURCC + // or distinctive magic number first) + #ifndef STBI_NO_PNG + if (stbi__png_test(s)) return stbi__png_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_BMP + if (stbi__bmp_test(s)) return stbi__bmp_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_GIF + if (stbi__gif_test(s)) return stbi__gif_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_PSD + if (stbi__psd_test(s)) return stbi__psd_load(s,x,y,comp,req_comp, ri, bpc); + #else + STBI_NOTUSED(bpc); + #endif + #ifndef STBI_NO_PIC + if (stbi__pic_test(s)) return stbi__pic_load(s,x,y,comp,req_comp, ri); + #endif + + // then the formats that can end up attempting to load with just 1 or 2 + // bytes matching expectations; these are prone to false positives, so + // try them later + #ifndef STBI_NO_JPEG + if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_PNM + if (stbi__pnm_test(s)) return stbi__pnm_load(s,x,y,comp,req_comp, ri); + #endif + + #ifndef STBI_NO_HDR + if (stbi__hdr_test(s)) { + float *hdr = stbi__hdr_load(s, x,y,comp,req_comp, ri); + return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp); + } + #endif + + #ifndef STBI_NO_TGA + // test tga last because it's a crappy test! + if (stbi__tga_test(s)) + return stbi__tga_load(s,x,y,comp,req_comp, ri); + #endif + + return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt"); +} + +static stbi_uc *stbi__convert_16_to_8(stbi__uint16 *orig, int w, int h, int channels) +{ + int i; + int img_len = w * h * channels; + stbi_uc *reduced; + + reduced = (stbi_uc *) stbi__malloc(img_len); + if (reduced == NULL) return stbi__errpuc("outofmem", "Out of memory"); + + for (i = 0; i < img_len; ++i) + reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling + + STBI_FREE(orig); + return reduced; +} + +static stbi__uint16 *stbi__convert_8_to_16(stbi_uc *orig, int w, int h, int channels) +{ + int i; + int img_len = w * h * channels; + stbi__uint16 *enlarged; + + enlarged = (stbi__uint16 *) stbi__malloc(img_len*2); + if (enlarged == NULL) return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory"); + + for (i = 0; i < img_len; ++i) + enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff + + STBI_FREE(orig); + return enlarged; +} + +static void stbi__vertical_flip(void *image, int w, int h, int bytes_per_pixel) +{ + int row; + size_t bytes_per_row = (size_t)w * bytes_per_pixel; + stbi_uc temp[2048]; + stbi_uc *bytes = (stbi_uc *)image; + + for (row = 0; row < (h>>1); row++) { + stbi_uc *row0 = bytes + row*bytes_per_row; + stbi_uc *row1 = bytes + (h - row - 1)*bytes_per_row; + // swap row0 with row1 + size_t bytes_left = bytes_per_row; + while (bytes_left) { + size_t bytes_copy = (bytes_left < sizeof(temp)) ? bytes_left : sizeof(temp); + memcpy(temp, row0, bytes_copy); + memcpy(row0, row1, bytes_copy); + memcpy(row1, temp, bytes_copy); + row0 += bytes_copy; + row1 += bytes_copy; + bytes_left -= bytes_copy; + } + } +} + +#ifndef STBI_NO_GIF +static void stbi__vertical_flip_slices(void *image, int w, int h, int z, int bytes_per_pixel) +{ + int slice; + int slice_size = w * h * bytes_per_pixel; + + stbi_uc *bytes = (stbi_uc *)image; + for (slice = 0; slice < z; ++slice) { + stbi__vertical_flip(bytes, w, h, bytes_per_pixel); + bytes += slice_size; + } +} +#endif + +static unsigned char *stbi__load_and_postprocess_8bit(stbi__context *s, int *x, int *y, int *comp, int req_comp) +{ + stbi__result_info ri; + void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8); + + if (result == NULL) + return NULL; + + // it is the responsibility of the loaders to make sure we get either 8 or 16 bit. + STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16); + + if (ri.bits_per_channel != 8) { + result = stbi__convert_16_to_8((stbi__uint16 *) result, *x, *y, req_comp == 0 ? *comp : req_comp); + ri.bits_per_channel = 8; + } + + // @TODO: move stbi__convert_format to here + + if (stbi__vertically_flip_on_load) { + int channels = req_comp ? req_comp : *comp; + stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi_uc)); + } + + return (unsigned char *) result; +} + +static stbi__uint16 *stbi__load_and_postprocess_16bit(stbi__context *s, int *x, int *y, int *comp, int req_comp) +{ + stbi__result_info ri; + void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16); + + if (result == NULL) + return NULL; + + // it is the responsibility of the loaders to make sure we get either 8 or 16 bit. + STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16); + + if (ri.bits_per_channel != 16) { + result = stbi__convert_8_to_16((stbi_uc *) result, *x, *y, req_comp == 0 ? *comp : req_comp); + ri.bits_per_channel = 16; + } + + // @TODO: move stbi__convert_format16 to here + // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision + + if (stbi__vertically_flip_on_load) { + int channels = req_comp ? req_comp : *comp; + stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi__uint16)); + } + + return (stbi__uint16 *) result; +} + +#if !defined(STBI_NO_HDR) && !defined(STBI_NO_LINEAR) +static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp) +{ + if (stbi__vertically_flip_on_load && result != NULL) { + int channels = req_comp ? req_comp : *comp; + stbi__vertical_flip(result, *x, *y, channels * sizeof(float)); + } +} +#endif + +#ifndef STBI_NO_STDIO + +#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8) +STBI_EXTERN __declspec(dllimport) int __stdcall MultiByteToWideChar(unsigned int cp, unsigned long flags, const char *str, int cbmb, wchar_t *widestr, int cchwide); +STBI_EXTERN __declspec(dllimport) int __stdcall WideCharToMultiByte(unsigned int cp, unsigned long flags, const wchar_t *widestr, int cchwide, char *str, int cbmb, const char *defchar, int *used_default); +#endif + +#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8) +STBIDEF int stbi_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input) +{ + return WideCharToMultiByte(65001 /* UTF8 */, 0, input, -1, buffer, (int) bufferlen, NULL, NULL); +} +#endif + +static FILE *stbi__fopen(char const *filename, char const *mode) +{ + FILE *f; +#if defined(_WIN32) && defined(STBI_WINDOWS_UTF8) + wchar_t wMode[64]; + wchar_t wFilename[1024]; + if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, filename, -1, wFilename, sizeof(wFilename)/sizeof(*wFilename))) + return 0; + + if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, mode, -1, wMode, sizeof(wMode)/sizeof(*wMode))) + return 0; + +#if defined(_MSC_VER) && _MSC_VER >= 1400 + if (0 != _wfopen_s(&f, wFilename, wMode)) + f = 0; +#else + f = _wfopen(wFilename, wMode); +#endif + +#elif defined(_MSC_VER) && _MSC_VER >= 1400 + if (0 != fopen_s(&f, filename, mode)) + f=0; +#else + f = fopen(filename, mode); +#endif + return f; +} + + +STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp) +{ + FILE *f = stbi__fopen(filename, "rb"); + unsigned char *result; + if (!f) return stbi__errpuc("can't fopen", "Unable to open file"); + result = stbi_load_from_file(f,x,y,comp,req_comp); + fclose(f); + return result; +} + +STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp) +{ + unsigned char *result; + stbi__context s; + stbi__start_file(&s,f); + result = stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp); + if (result) { + // need to 'unget' all the characters in the IO buffer + fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR); + } + return result; +} + +STBIDEF stbi__uint16 *stbi_load_from_file_16(FILE *f, int *x, int *y, int *comp, int req_comp) +{ + stbi__uint16 *result; + stbi__context s; + stbi__start_file(&s,f); + result = stbi__load_and_postprocess_16bit(&s,x,y,comp,req_comp); + if (result) { + // need to 'unget' all the characters in the IO buffer + fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR); + } + return result; +} + +STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *comp, int req_comp) +{ + FILE *f = stbi__fopen(filename, "rb"); + stbi__uint16 *result; + if (!f) return (stbi_us *) stbi__errpuc("can't fopen", "Unable to open file"); + result = stbi_load_from_file_16(f,x,y,comp,req_comp); + fclose(f); + return result; +} + + +#endif //!STBI_NO_STDIO + +STBIDEF stbi_us *stbi_load_16_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__load_and_postprocess_16bit(&s,x,y,channels_in_file,desired_channels); +} + +STBIDEF stbi_us *stbi_load_16_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *)clbk, user); + return stbi__load_and_postprocess_16bit(&s,x,y,channels_in_file,desired_channels); +} + +STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp); +} + +STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user); + return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp); +} + +#ifndef STBI_NO_GIF +STBIDEF stbi_uc *stbi_load_gif_from_memory(stbi_uc const *buffer, int len, int **delays, int *x, int *y, int *z, int *comp, int req_comp) +{ + unsigned char *result; + stbi__context s; + stbi__start_mem(&s,buffer,len); + + result = (unsigned char*) stbi__load_gif_main(&s, delays, x, y, z, comp, req_comp); + if (stbi__vertically_flip_on_load) { + stbi__vertical_flip_slices( result, *x, *y, *z, *comp ); + } + + return result; +} +#endif + +#ifndef STBI_NO_LINEAR +static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp) +{ + unsigned char *data; + #ifndef STBI_NO_HDR + if (stbi__hdr_test(s)) { + stbi__result_info ri; + float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp, &ri); + if (hdr_data) + stbi__float_postprocess(hdr_data,x,y,comp,req_comp); + return hdr_data; + } + #endif + data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp); + if (data) + return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp); + return stbi__errpf("unknown image type", "Image not of any known type, or corrupt"); +} + +STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__loadf_main(&s,x,y,comp,req_comp); +} + +STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user); + return stbi__loadf_main(&s,x,y,comp,req_comp); +} + +#ifndef STBI_NO_STDIO +STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp) +{ + float *result; + FILE *f = stbi__fopen(filename, "rb"); + if (!f) return stbi__errpf("can't fopen", "Unable to open file"); + result = stbi_loadf_from_file(f,x,y,comp,req_comp); + fclose(f); + return result; +} + +STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_file(&s,f); + return stbi__loadf_main(&s,x,y,comp,req_comp); +} +#endif // !STBI_NO_STDIO + +#endif // !STBI_NO_LINEAR + +// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is +// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always +// reports false! + +STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len) +{ + #ifndef STBI_NO_HDR + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__hdr_test(&s); + #else + STBI_NOTUSED(buffer); + STBI_NOTUSED(len); + return 0; + #endif +} + +#ifndef STBI_NO_STDIO +STBIDEF int stbi_is_hdr (char const *filename) +{ + FILE *f = stbi__fopen(filename, "rb"); + int result=0; + if (f) { + result = stbi_is_hdr_from_file(f); + fclose(f); + } + return result; +} + +STBIDEF int stbi_is_hdr_from_file(FILE *f) +{ + #ifndef STBI_NO_HDR + long pos = ftell(f); + int res; + stbi__context s; + stbi__start_file(&s,f); + res = stbi__hdr_test(&s); + fseek(f, pos, SEEK_SET); + return res; + #else + STBI_NOTUSED(f); + return 0; + #endif +} +#endif // !STBI_NO_STDIO + +STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user) +{ + #ifndef STBI_NO_HDR + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user); + return stbi__hdr_test(&s); + #else + STBI_NOTUSED(clbk); + STBI_NOTUSED(user); + return 0; + #endif +} + +#ifndef STBI_NO_LINEAR +static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f; + +STBIDEF void stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; } +STBIDEF void stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; } +#endif + +static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f; + +STBIDEF void stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; } +STBIDEF void stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; } + + +////////////////////////////////////////////////////////////////////////////// +// +// Common code used by all image loaders +// + +enum +{ + STBI__SCAN_load=0, + STBI__SCAN_type, + STBI__SCAN_header +}; + +static void stbi__refill_buffer(stbi__context *s) +{ + int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen); + s->callback_already_read += (int) (s->img_buffer - s->img_buffer_original); + if (n == 0) { + // at end of file, treat same as if from memory, but need to handle case + // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file + s->read_from_callbacks = 0; + s->img_buffer = s->buffer_start; + s->img_buffer_end = s->buffer_start+1; + *s->img_buffer = 0; + } else { + s->img_buffer = s->buffer_start; + s->img_buffer_end = s->buffer_start + n; + } +} + +stbi_inline static stbi_uc stbi__get8(stbi__context *s) +{ + if (s->img_buffer < s->img_buffer_end) + return *s->img_buffer++; + if (s->read_from_callbacks) { + stbi__refill_buffer(s); + return *s->img_buffer++; + } + return 0; +} + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_HDR) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM) +// nothing +#else +stbi_inline static int stbi__at_eof(stbi__context *s) +{ + if (s->io.read) { + if (!(s->io.eof)(s->io_user_data)) return 0; + // if feof() is true, check if buffer = end + // special case: we've only got the special 0 character at the end + if (s->read_from_callbacks == 0) return 1; + } + + return s->img_buffer >= s->img_buffer_end; +} +#endif + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) +// nothing +#else +static void stbi__skip(stbi__context *s, int n) +{ + if (n == 0) return; // already there! + if (n < 0) { + s->img_buffer = s->img_buffer_end; + return; + } + if (s->io.read) { + int blen = (int) (s->img_buffer_end - s->img_buffer); + if (blen < n) { + s->img_buffer = s->img_buffer_end; + (s->io.skip)(s->io_user_data, n - blen); + return; + } + } + s->img_buffer += n; +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_TGA) && defined(STBI_NO_HDR) && defined(STBI_NO_PNM) +// nothing +#else +static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n) +{ + if (s->io.read) { + int blen = (int) (s->img_buffer_end - s->img_buffer); + if (blen < n) { + int res, count; + + memcpy(buffer, s->img_buffer, blen); + + count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen); + res = (count == (n-blen)); + s->img_buffer = s->img_buffer_end; + return res; + } + } + + if (s->img_buffer+n <= s->img_buffer_end) { + memcpy(buffer, s->img_buffer, n); + s->img_buffer += n; + return 1; + } else + return 0; +} +#endif + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC) +// nothing +#else +static int stbi__get16be(stbi__context *s) +{ + int z = stbi__get8(s); + return (z << 8) + stbi__get8(s); +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC) +// nothing +#else +static stbi__uint32 stbi__get32be(stbi__context *s) +{ + stbi__uint32 z = stbi__get16be(s); + return (z << 16) + stbi__get16be(s); +} +#endif + +#if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) +// nothing +#else +static int stbi__get16le(stbi__context *s) +{ + int z = stbi__get8(s); + return z + (stbi__get8(s) << 8); +} +#endif + +#ifndef STBI_NO_BMP +static stbi__uint32 stbi__get32le(stbi__context *s) +{ + stbi__uint32 z = stbi__get16le(s); + z += (stbi__uint32)stbi__get16le(s) << 16; + return z; +} +#endif + +#define STBI__BYTECAST(x) ((stbi_uc) ((x) & 255)) // truncate int to byte without warnings + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM) +// nothing +#else +////////////////////////////////////////////////////////////////////////////// +// +// generic converter from built-in img_n to req_comp +// individual types do this automatically as much as possible (e.g. jpeg +// does all cases internally since it needs to colorspace convert anyway, +// and it never has alpha, so very few cases ). png can automatically +// interleave an alpha=255 channel, but falls back to this for other cases +// +// assume data buffer is malloced, so malloc a new one and free that one +// only failure mode is malloc failing + +static stbi_uc stbi__compute_y(int r, int g, int b) +{ + return (stbi_uc) (((r*77) + (g*150) + (29*b)) >> 8); +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM) +// nothing +#else +static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y) +{ + int i,j; + unsigned char *good; + + if (req_comp == img_n) return data; + STBI_ASSERT(req_comp >= 1 && req_comp <= 4); + + good = (unsigned char *) stbi__malloc_mad3(req_comp, x, y, 0); + if (good == NULL) { + STBI_FREE(data); + return stbi__errpuc("outofmem", "Out of memory"); + } + + for (j=0; j < (int) y; ++j) { + unsigned char *src = data + j * x * img_n ; + unsigned char *dest = good + j * x * req_comp; + + #define STBI__COMBO(a,b) ((a)*8+(b)) + #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b) + // convert source image with img_n components to one with req_comp components; + // avoid switch per pixel, so use switch per scanline and massive macros + switch (STBI__COMBO(img_n, req_comp)) { + STBI__CASE(1,2) { dest[0]=src[0]; dest[1]=255; } break; + STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=255; } break; + STBI__CASE(2,1) { dest[0]=src[0]; } break; + STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=src[1]; } break; + STBI__CASE(3,4) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2];dest[3]=255; } break; + STBI__CASE(3,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break; + STBI__CASE(3,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); dest[1] = 255; } break; + STBI__CASE(4,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break; + STBI__CASE(4,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); dest[1] = src[3]; } break; + STBI__CASE(4,3) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2]; } break; + default: STBI_ASSERT(0); STBI_FREE(data); STBI_FREE(good); return stbi__errpuc("unsupported", "Unsupported format conversion"); + } + #undef STBI__CASE + } + + STBI_FREE(data); + return good; +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) +// nothing +#else +static stbi__uint16 stbi__compute_y_16(int r, int g, int b) +{ + return (stbi__uint16) (((r*77) + (g*150) + (29*b)) >> 8); +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) +// nothing +#else +static stbi__uint16 *stbi__convert_format16(stbi__uint16 *data, int img_n, int req_comp, unsigned int x, unsigned int y) +{ + int i,j; + stbi__uint16 *good; + + if (req_comp == img_n) return data; + STBI_ASSERT(req_comp >= 1 && req_comp <= 4); + + good = (stbi__uint16 *) stbi__malloc(req_comp * x * y * 2); + if (good == NULL) { + STBI_FREE(data); + return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory"); + } + + for (j=0; j < (int) y; ++j) { + stbi__uint16 *src = data + j * x * img_n ; + stbi__uint16 *dest = good + j * x * req_comp; + + #define STBI__COMBO(a,b) ((a)*8+(b)) + #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b) + // convert source image with img_n components to one with req_comp components; + // avoid switch per pixel, so use switch per scanline and massive macros + switch (STBI__COMBO(img_n, req_comp)) { + STBI__CASE(1,2) { dest[0]=src[0]; dest[1]=0xffff; } break; + STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=0xffff; } break; + STBI__CASE(2,1) { dest[0]=src[0]; } break; + STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=src[1]; } break; + STBI__CASE(3,4) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2];dest[3]=0xffff; } break; + STBI__CASE(3,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break; + STBI__CASE(3,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); dest[1] = 0xffff; } break; + STBI__CASE(4,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break; + STBI__CASE(4,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); dest[1] = src[3]; } break; + STBI__CASE(4,3) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2]; } break; + default: STBI_ASSERT(0); STBI_FREE(data); STBI_FREE(good); return (stbi__uint16*) stbi__errpuc("unsupported", "Unsupported format conversion"); + } + #undef STBI__CASE + } + + STBI_FREE(data); + return good; +} +#endif + +#ifndef STBI_NO_LINEAR +static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp) +{ + int i,k,n; + float *output; + if (!data) return NULL; + output = (float *) stbi__malloc_mad4(x, y, comp, sizeof(float), 0); + if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); } + // compute number of non-alpha components + if (comp & 1) n = comp; else n = comp-1; + for (i=0; i < x*y; ++i) { + for (k=0; k < n; ++k) { + output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale); + } + } + if (n < comp) { + for (i=0; i < x*y; ++i) { + output[i*comp + n] = data[i*comp + n]/255.0f; + } + } + STBI_FREE(data); + return output; +} +#endif + +#ifndef STBI_NO_HDR +#define stbi__float2int(x) ((int) (x)) +static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp) +{ + int i,k,n; + stbi_uc *output; + if (!data) return NULL; + output = (stbi_uc *) stbi__malloc_mad3(x, y, comp, 0); + if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); } + // compute number of non-alpha components + if (comp & 1) n = comp; else n = comp-1; + for (i=0; i < x*y; ++i) { + for (k=0; k < n; ++k) { + float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f; + if (z < 0) z = 0; + if (z > 255) z = 255; + output[i*comp + k] = (stbi_uc) stbi__float2int(z); + } + if (k < comp) { + float z = data[i*comp+k] * 255 + 0.5f; + if (z < 0) z = 0; + if (z > 255) z = 255; + output[i*comp + k] = (stbi_uc) stbi__float2int(z); + } + } + STBI_FREE(data); + return output; +} +#endif + +////////////////////////////////////////////////////////////////////////////// +// +// "baseline" JPEG/JFIF decoder +// +// simple implementation +// - doesn't support delayed output of y-dimension +// - simple interface (only one output format: 8-bit interleaved RGB) +// - doesn't try to recover corrupt jpegs +// - doesn't allow partial loading, loading multiple at once +// - still fast on x86 (copying globals into locals doesn't help x86) +// - allocates lots of intermediate memory (full size of all components) +// - non-interleaved case requires this anyway +// - allows good upsampling (see next) +// high-quality +// - upsampled channels are bilinearly interpolated, even across blocks +// - quality integer IDCT derived from IJG's 'slow' +// performance +// - fast huffman; reasonable integer IDCT +// - some SIMD kernels for common paths on targets with SSE2/NEON +// - uses a lot of intermediate memory, could cache poorly + +#ifndef STBI_NO_JPEG + +// huffman decoding acceleration +#define FAST_BITS 9 // larger handles more cases; smaller stomps less cache + +typedef struct +{ + stbi_uc fast[1 << FAST_BITS]; + // weirdly, repacking this into AoS is a 10% speed loss, instead of a win + stbi__uint16 code[256]; + stbi_uc values[256]; + stbi_uc size[257]; + unsigned int maxcode[18]; + int delta[17]; // old 'firstsymbol' - old 'firstcode' +} stbi__huffman; + +typedef struct +{ + stbi__context *s; + stbi__huffman huff_dc[4]; + stbi__huffman huff_ac[4]; + stbi__uint16 dequant[4][64]; + stbi__int16 fast_ac[4][1 << FAST_BITS]; + +// sizes for components, interleaved MCUs + int img_h_max, img_v_max; + int img_mcu_x, img_mcu_y; + int img_mcu_w, img_mcu_h; + +// definition of jpeg image component + struct + { + int id; + int h,v; + int tq; + int hd,ha; + int dc_pred; + + int x,y,w2,h2; + stbi_uc *data; + void *raw_data, *raw_coeff; + stbi_uc *linebuf; + short *coeff; // progressive only + int coeff_w, coeff_h; // number of 8x8 coefficient blocks + } img_comp[4]; + + stbi__uint32 code_buffer; // jpeg entropy-coded buffer + int code_bits; // number of valid bits + unsigned char marker; // marker seen while filling entropy buffer + int nomore; // flag if we saw a marker so must stop + + int progressive; + int spec_start; + int spec_end; + int succ_high; + int succ_low; + int eob_run; + int jfif; + int app14_color_transform; // Adobe APP14 tag + int rgb; + + int scan_n, order[4]; + int restart_interval, todo; + +// kernels + void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]); + void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step); + stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs); +} stbi__jpeg; + +static int stbi__build_huffman(stbi__huffman *h, int *count) +{ + int i,j,k=0; + unsigned int code; + // build size list for each symbol (from JPEG spec) + for (i=0; i < 16; ++i) { + for (j=0; j < count[i]; ++j) { + h->size[k++] = (stbi_uc) (i+1); + if(k >= 257) return stbi__err("bad size list","Corrupt JPEG"); + } + } + h->size[k] = 0; + + // compute actual symbols (from jpeg spec) + code = 0; + k = 0; + for(j=1; j <= 16; ++j) { + // compute delta to add to code to compute symbol id + h->delta[j] = k - code; + if (h->size[k] == j) { + while (h->size[k] == j) + h->code[k++] = (stbi__uint16) (code++); + if (code-1 >= (1u << j)) return stbi__err("bad code lengths","Corrupt JPEG"); + } + // compute largest code + 1 for this size, preshifted as needed later + h->maxcode[j] = code << (16-j); + code <<= 1; + } + h->maxcode[j] = 0xffffffff; + + // build non-spec acceleration table; 255 is flag for not-accelerated + memset(h->fast, 255, 1 << FAST_BITS); + for (i=0; i < k; ++i) { + int s = h->size[i]; + if (s <= FAST_BITS) { + int c = h->code[i] << (FAST_BITS-s); + int m = 1 << (FAST_BITS-s); + for (j=0; j < m; ++j) { + h->fast[c+j] = (stbi_uc) i; + } + } + } + return 1; +} + +// build a table that decodes both magnitude and value of small ACs in +// one go. +static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h) +{ + int i; + for (i=0; i < (1 << FAST_BITS); ++i) { + stbi_uc fast = h->fast[i]; + fast_ac[i] = 0; + if (fast < 255) { + int rs = h->values[fast]; + int run = (rs >> 4) & 15; + int magbits = rs & 15; + int len = h->size[fast]; + + if (magbits && len + magbits <= FAST_BITS) { + // magnitude code followed by receive_extend code + int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits); + int m = 1 << (magbits - 1); + if (k < m) k += (~0U << magbits) + 1; + // if the result is small enough, we can fit it in fast_ac table + if (k >= -128 && k <= 127) + fast_ac[i] = (stbi__int16) ((k * 256) + (run * 16) + (len + magbits)); + } + } + } +} + +static void stbi__grow_buffer_unsafe(stbi__jpeg *j) +{ + do { + unsigned int b = j->nomore ? 0 : stbi__get8(j->s); + if (b == 0xff) { + int c = stbi__get8(j->s); + while (c == 0xff) c = stbi__get8(j->s); // consume fill bytes + if (c != 0) { + j->marker = (unsigned char) c; + j->nomore = 1; + return; + } + } + j->code_buffer |= b << (24 - j->code_bits); + j->code_bits += 8; + } while (j->code_bits <= 24); +} + +// (1 << n) - 1 +static const stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535}; + +// decode a jpeg huffman value from the bitstream +stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h) +{ + unsigned int temp; + int c,k; + + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + + // look at the top FAST_BITS and determine what symbol ID it is, + // if the code is <= FAST_BITS + c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1); + k = h->fast[c]; + if (k < 255) { + int s = h->size[k]; + if (s > j->code_bits) + return -1; + j->code_buffer <<= s; + j->code_bits -= s; + return h->values[k]; + } + + // naive test is to shift the code_buffer down so k bits are + // valid, then test against maxcode. To speed this up, we've + // preshifted maxcode left so that it has (16-k) 0s at the + // end; in other words, regardless of the number of bits, it + // wants to be compared against something shifted to have 16; + // that way we don't need to shift inside the loop. + temp = j->code_buffer >> 16; + for (k=FAST_BITS+1 ; ; ++k) + if (temp < h->maxcode[k]) + break; + if (k == 17) { + // error! code not found + j->code_bits -= 16; + return -1; + } + + if (k > j->code_bits) + return -1; + + // convert the huffman code to the symbol id + c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k]; + if(c < 0 || c >= 256) // symbol id out of bounds! + return -1; + STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]); + + // convert the id to a symbol + j->code_bits -= k; + j->code_buffer <<= k; + return h->values[c]; +} + +// bias[n] = (-1<code_bits < n) stbi__grow_buffer_unsafe(j); + if (j->code_bits < n) return 0; // ran out of bits from stream, return 0s intead of continuing + + sgn = j->code_buffer >> 31; // sign bit always in MSB; 0 if MSB clear (positive), 1 if MSB set (negative) + k = stbi_lrot(j->code_buffer, n); + j->code_buffer = k & ~stbi__bmask[n]; + k &= stbi__bmask[n]; + j->code_bits -= n; + return k + (stbi__jbias[n] & (sgn - 1)); +} + +// get some unsigned bits +stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n) +{ + unsigned int k; + if (j->code_bits < n) stbi__grow_buffer_unsafe(j); + if (j->code_bits < n) return 0; // ran out of bits from stream, return 0s intead of continuing + k = stbi_lrot(j->code_buffer, n); + j->code_buffer = k & ~stbi__bmask[n]; + k &= stbi__bmask[n]; + j->code_bits -= n; + return k; +} + +stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j) +{ + unsigned int k; + if (j->code_bits < 1) stbi__grow_buffer_unsafe(j); + if (j->code_bits < 1) return 0; // ran out of bits from stream, return 0s intead of continuing + k = j->code_buffer; + j->code_buffer <<= 1; + --j->code_bits; + return k & 0x80000000; +} + +// given a value that's at position X in the zigzag stream, +// where does it appear in the 8x8 matrix coded as row-major? +static const stbi_uc stbi__jpeg_dezigzag[64+15] = +{ + 0, 1, 8, 16, 9, 2, 3, 10, + 17, 24, 32, 25, 18, 11, 4, 5, + 12, 19, 26, 33, 40, 48, 41, 34, + 27, 20, 13, 6, 7, 14, 21, 28, + 35, 42, 49, 56, 57, 50, 43, 36, + 29, 22, 15, 23, 30, 37, 44, 51, + 58, 59, 52, 45, 38, 31, 39, 46, + 53, 60, 61, 54, 47, 55, 62, 63, + // let corrupt input sample past end + 63, 63, 63, 63, 63, 63, 63, 63, + 63, 63, 63, 63, 63, 63, 63 +}; + +// decode one 64-entry block-- +static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi__uint16 *dequant) +{ + int diff,dc,k; + int t; + + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + t = stbi__jpeg_huff_decode(j, hdc); + if (t < 0 || t > 15) return stbi__err("bad huffman code","Corrupt JPEG"); + + // 0 all the ac values now so we can do it 32-bits at a time + memset(data,0,64*sizeof(data[0])); + + diff = t ? stbi__extend_receive(j, t) : 0; + if (!stbi__addints_valid(j->img_comp[b].dc_pred, diff)) return stbi__err("bad delta","Corrupt JPEG"); + dc = j->img_comp[b].dc_pred + diff; + j->img_comp[b].dc_pred = dc; + if (!stbi__mul2shorts_valid(dc, dequant[0])) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + data[0] = (short) (dc * dequant[0]); + + // decode AC components, see JPEG spec + k = 1; + do { + unsigned int zig; + int c,r,s; + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1); + r = fac[c]; + if (r) { // fast-AC path + k += (r >> 4) & 15; // run + s = r & 15; // combined length + if (s > j->code_bits) return stbi__err("bad huffman code", "Combined length longer than code bits available"); + j->code_buffer <<= s; + j->code_bits -= s; + // decode into unzigzag'd location + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) ((r >> 8) * dequant[zig]); + } else { + int rs = stbi__jpeg_huff_decode(j, hac); + if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG"); + s = rs & 15; + r = rs >> 4; + if (s == 0) { + if (rs != 0xf0) break; // end block + k += 16; + } else { + k += r; + // decode into unzigzag'd location + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]); + } + } + } while (k < 64); + return 1; +} + +static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b) +{ + int diff,dc; + int t; + if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + + if (j->succ_high == 0) { + // first scan for DC coefficient, must be first + memset(data,0,64*sizeof(data[0])); // 0 all the ac values now + t = stbi__jpeg_huff_decode(j, hdc); + if (t < 0 || t > 15) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + diff = t ? stbi__extend_receive(j, t) : 0; + + if (!stbi__addints_valid(j->img_comp[b].dc_pred, diff)) return stbi__err("bad delta", "Corrupt JPEG"); + dc = j->img_comp[b].dc_pred + diff; + j->img_comp[b].dc_pred = dc; + if (!stbi__mul2shorts_valid(dc, 1 << j->succ_low)) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + data[0] = (short) (dc * (1 << j->succ_low)); + } else { + // refinement scan for DC coefficient + if (stbi__jpeg_get_bit(j)) + data[0] += (short) (1 << j->succ_low); + } + return 1; +} + +// @OPTIMIZE: store non-zigzagged during the decode passes, +// and only de-zigzag when dequantizing +static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac) +{ + int k; + if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + + if (j->succ_high == 0) { + int shift = j->succ_low; + + if (j->eob_run) { + --j->eob_run; + return 1; + } + + k = j->spec_start; + do { + unsigned int zig; + int c,r,s; + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1); + r = fac[c]; + if (r) { // fast-AC path + k += (r >> 4) & 15; // run + s = r & 15; // combined length + if (s > j->code_bits) return stbi__err("bad huffman code", "Combined length longer than code bits available"); + j->code_buffer <<= s; + j->code_bits -= s; + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) ((r >> 8) * (1 << shift)); + } else { + int rs = stbi__jpeg_huff_decode(j, hac); + if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG"); + s = rs & 15; + r = rs >> 4; + if (s == 0) { + if (r < 15) { + j->eob_run = (1 << r); + if (r) + j->eob_run += stbi__jpeg_get_bits(j, r); + --j->eob_run; + break; + } + k += 16; + } else { + k += r; + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) (stbi__extend_receive(j,s) * (1 << shift)); + } + } + } while (k <= j->spec_end); + } else { + // refinement scan for these AC coefficients + + short bit = (short) (1 << j->succ_low); + + if (j->eob_run) { + --j->eob_run; + for (k = j->spec_start; k <= j->spec_end; ++k) { + short *p = &data[stbi__jpeg_dezigzag[k]]; + if (*p != 0) + if (stbi__jpeg_get_bit(j)) + if ((*p & bit)==0) { + if (*p > 0) + *p += bit; + else + *p -= bit; + } + } + } else { + k = j->spec_start; + do { + int r,s; + int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh + if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG"); + s = rs & 15; + r = rs >> 4; + if (s == 0) { + if (r < 15) { + j->eob_run = (1 << r) - 1; + if (r) + j->eob_run += stbi__jpeg_get_bits(j, r); + r = 64; // force end of block + } else { + // r=15 s=0 should write 16 0s, so we just do + // a run of 15 0s and then write s (which is 0), + // so we don't have to do anything special here + } + } else { + if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG"); + // sign bit + if (stbi__jpeg_get_bit(j)) + s = bit; + else + s = -bit; + } + + // advance by r + while (k <= j->spec_end) { + short *p = &data[stbi__jpeg_dezigzag[k++]]; + if (*p != 0) { + if (stbi__jpeg_get_bit(j)) + if ((*p & bit)==0) { + if (*p > 0) + *p += bit; + else + *p -= bit; + } + } else { + if (r == 0) { + *p = (short) s; + break; + } + --r; + } + } + } while (k <= j->spec_end); + } + } + return 1; +} + +// take a -128..127 value and stbi__clamp it and convert to 0..255 +stbi_inline static stbi_uc stbi__clamp(int x) +{ + // trick to use a single test to catch both cases + if ((unsigned int) x > 255) { + if (x < 0) return 0; + if (x > 255) return 255; + } + return (stbi_uc) x; +} + +#define stbi__f2f(x) ((int) (((x) * 4096 + 0.5))) +#define stbi__fsh(x) ((x) * 4096) + +// derived from jidctint -- DCT_ISLOW +#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \ + int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \ + p2 = s2; \ + p3 = s6; \ + p1 = (p2+p3) * stbi__f2f(0.5411961f); \ + t2 = p1 + p3*stbi__f2f(-1.847759065f); \ + t3 = p1 + p2*stbi__f2f( 0.765366865f); \ + p2 = s0; \ + p3 = s4; \ + t0 = stbi__fsh(p2+p3); \ + t1 = stbi__fsh(p2-p3); \ + x0 = t0+t3; \ + x3 = t0-t3; \ + x1 = t1+t2; \ + x2 = t1-t2; \ + t0 = s7; \ + t1 = s5; \ + t2 = s3; \ + t3 = s1; \ + p3 = t0+t2; \ + p4 = t1+t3; \ + p1 = t0+t3; \ + p2 = t1+t2; \ + p5 = (p3+p4)*stbi__f2f( 1.175875602f); \ + t0 = t0*stbi__f2f( 0.298631336f); \ + t1 = t1*stbi__f2f( 2.053119869f); \ + t2 = t2*stbi__f2f( 3.072711026f); \ + t3 = t3*stbi__f2f( 1.501321110f); \ + p1 = p5 + p1*stbi__f2f(-0.899976223f); \ + p2 = p5 + p2*stbi__f2f(-2.562915447f); \ + p3 = p3*stbi__f2f(-1.961570560f); \ + p4 = p4*stbi__f2f(-0.390180644f); \ + t3 += p1+p4; \ + t2 += p2+p3; \ + t1 += p2+p4; \ + t0 += p1+p3; + +static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64]) +{ + int i,val[64],*v=val; + stbi_uc *o; + short *d = data; + + // columns + for (i=0; i < 8; ++i,++d, ++v) { + // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing + if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0 + && d[40]==0 && d[48]==0 && d[56]==0) { + // no shortcut 0 seconds + // (1|2|3|4|5|6|7)==0 0 seconds + // all separate -0.047 seconds + // 1 && 2|3 && 4|5 && 6|7: -0.047 seconds + int dcterm = d[0]*4; + v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm; + } else { + STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56]) + // constants scaled things up by 1<<12; let's bring them back + // down, but keep 2 extra bits of precision + x0 += 512; x1 += 512; x2 += 512; x3 += 512; + v[ 0] = (x0+t3) >> 10; + v[56] = (x0-t3) >> 10; + v[ 8] = (x1+t2) >> 10; + v[48] = (x1-t2) >> 10; + v[16] = (x2+t1) >> 10; + v[40] = (x2-t1) >> 10; + v[24] = (x3+t0) >> 10; + v[32] = (x3-t0) >> 10; + } + } + + for (i=0, v=val, o=out; i < 8; ++i,v+=8,o+=out_stride) { + // no fast case since the first 1D IDCT spread components out + STBI__IDCT_1D(v[0],v[1],v[2],v[3],v[4],v[5],v[6],v[7]) + // constants scaled things up by 1<<12, plus we had 1<<2 from first + // loop, plus horizontal and vertical each scale by sqrt(8) so together + // we've got an extra 1<<3, so 1<<17 total we need to remove. + // so we want to round that, which means adding 0.5 * 1<<17, + // aka 65536. Also, we'll end up with -128 to 127 that we want + // to encode as 0..255 by adding 128, so we'll add that before the shift + x0 += 65536 + (128<<17); + x1 += 65536 + (128<<17); + x2 += 65536 + (128<<17); + x3 += 65536 + (128<<17); + // tried computing the shifts into temps, or'ing the temps to see + // if any were out of range, but that was slower + o[0] = stbi__clamp((x0+t3) >> 17); + o[7] = stbi__clamp((x0-t3) >> 17); + o[1] = stbi__clamp((x1+t2) >> 17); + o[6] = stbi__clamp((x1-t2) >> 17); + o[2] = stbi__clamp((x2+t1) >> 17); + o[5] = stbi__clamp((x2-t1) >> 17); + o[3] = stbi__clamp((x3+t0) >> 17); + o[4] = stbi__clamp((x3-t0) >> 17); + } +} + +#ifdef STBI_SSE2 +// sse2 integer IDCT. not the fastest possible implementation but it +// produces bit-identical results to the generic C version so it's +// fully "transparent". +static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64]) +{ + // This is constructed to match our regular (generic) integer IDCT exactly. + __m128i row0, row1, row2, row3, row4, row5, row6, row7; + __m128i tmp; + + // dot product constant: even elems=x, odd elems=y + #define dct_const(x,y) _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y)) + + // out(0) = c0[even]*x + c0[odd]*y (c0, x, y 16-bit, out 32-bit) + // out(1) = c1[even]*x + c1[odd]*y + #define dct_rot(out0,out1, x,y,c0,c1) \ + __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \ + __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \ + __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \ + __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \ + __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \ + __m128i out1##_h = _mm_madd_epi16(c0##hi, c1) + + // out = in << 12 (in 16-bit, out 32-bit) + #define dct_widen(out, in) \ + __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \ + __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4) + + // wide add + #define dct_wadd(out, a, b) \ + __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \ + __m128i out##_h = _mm_add_epi32(a##_h, b##_h) + + // wide sub + #define dct_wsub(out, a, b) \ + __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \ + __m128i out##_h = _mm_sub_epi32(a##_h, b##_h) + + // butterfly a/b, add bias, then shift by "s" and pack + #define dct_bfly32o(out0, out1, a,b,bias,s) \ + { \ + __m128i abiased_l = _mm_add_epi32(a##_l, bias); \ + __m128i abiased_h = _mm_add_epi32(a##_h, bias); \ + dct_wadd(sum, abiased, b); \ + dct_wsub(dif, abiased, b); \ + out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \ + out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \ + } + + // 8-bit interleave step (for transposes) + #define dct_interleave8(a, b) \ + tmp = a; \ + a = _mm_unpacklo_epi8(a, b); \ + b = _mm_unpackhi_epi8(tmp, b) + + // 16-bit interleave step (for transposes) + #define dct_interleave16(a, b) \ + tmp = a; \ + a = _mm_unpacklo_epi16(a, b); \ + b = _mm_unpackhi_epi16(tmp, b) + + #define dct_pass(bias,shift) \ + { \ + /* even part */ \ + dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \ + __m128i sum04 = _mm_add_epi16(row0, row4); \ + __m128i dif04 = _mm_sub_epi16(row0, row4); \ + dct_widen(t0e, sum04); \ + dct_widen(t1e, dif04); \ + dct_wadd(x0, t0e, t3e); \ + dct_wsub(x3, t0e, t3e); \ + dct_wadd(x1, t1e, t2e); \ + dct_wsub(x2, t1e, t2e); \ + /* odd part */ \ + dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \ + dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \ + __m128i sum17 = _mm_add_epi16(row1, row7); \ + __m128i sum35 = _mm_add_epi16(row3, row5); \ + dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \ + dct_wadd(x4, y0o, y4o); \ + dct_wadd(x5, y1o, y5o); \ + dct_wadd(x6, y2o, y5o); \ + dct_wadd(x7, y3o, y4o); \ + dct_bfly32o(row0,row7, x0,x7,bias,shift); \ + dct_bfly32o(row1,row6, x1,x6,bias,shift); \ + dct_bfly32o(row2,row5, x2,x5,bias,shift); \ + dct_bfly32o(row3,row4, x3,x4,bias,shift); \ + } + + __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f)); + __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f)); + __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f)); + __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f)); + __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f)); + __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f)); + __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f)); + __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f)); + + // rounding biases in column/row passes, see stbi__idct_block for explanation. + __m128i bias_0 = _mm_set1_epi32(512); + __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17)); + + // load + row0 = _mm_load_si128((const __m128i *) (data + 0*8)); + row1 = _mm_load_si128((const __m128i *) (data + 1*8)); + row2 = _mm_load_si128((const __m128i *) (data + 2*8)); + row3 = _mm_load_si128((const __m128i *) (data + 3*8)); + row4 = _mm_load_si128((const __m128i *) (data + 4*8)); + row5 = _mm_load_si128((const __m128i *) (data + 5*8)); + row6 = _mm_load_si128((const __m128i *) (data + 6*8)); + row7 = _mm_load_si128((const __m128i *) (data + 7*8)); + + // column pass + dct_pass(bias_0, 10); + + { + // 16bit 8x8 transpose pass 1 + dct_interleave16(row0, row4); + dct_interleave16(row1, row5); + dct_interleave16(row2, row6); + dct_interleave16(row3, row7); + + // transpose pass 2 + dct_interleave16(row0, row2); + dct_interleave16(row1, row3); + dct_interleave16(row4, row6); + dct_interleave16(row5, row7); + + // transpose pass 3 + dct_interleave16(row0, row1); + dct_interleave16(row2, row3); + dct_interleave16(row4, row5); + dct_interleave16(row6, row7); + } + + // row pass + dct_pass(bias_1, 17); + + { + // pack + __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7 + __m128i p1 = _mm_packus_epi16(row2, row3); + __m128i p2 = _mm_packus_epi16(row4, row5); + __m128i p3 = _mm_packus_epi16(row6, row7); + + // 8bit 8x8 transpose pass 1 + dct_interleave8(p0, p2); // a0e0a1e1... + dct_interleave8(p1, p3); // c0g0c1g1... + + // transpose pass 2 + dct_interleave8(p0, p1); // a0c0e0g0... + dct_interleave8(p2, p3); // b0d0f0h0... + + // transpose pass 3 + dct_interleave8(p0, p2); // a0b0c0d0... + dct_interleave8(p1, p3); // a4b4c4d4... + + // store + _mm_storel_epi64((__m128i *) out, p0); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride; + _mm_storel_epi64((__m128i *) out, p2); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride; + _mm_storel_epi64((__m128i *) out, p1); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride; + _mm_storel_epi64((__m128i *) out, p3); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e)); + } + +#undef dct_const +#undef dct_rot +#undef dct_widen +#undef dct_wadd +#undef dct_wsub +#undef dct_bfly32o +#undef dct_interleave8 +#undef dct_interleave16 +#undef dct_pass +} + +#endif // STBI_SSE2 + +#ifdef STBI_NEON + +// NEON integer IDCT. should produce bit-identical +// results to the generic C version. +static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64]) +{ + int16x8_t row0, row1, row2, row3, row4, row5, row6, row7; + + int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f)); + int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f)); + int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f)); + int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f)); + int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f)); + int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f)); + int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f)); + int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f)); + int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f)); + int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f)); + int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f)); + int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f)); + +#define dct_long_mul(out, inq, coeff) \ + int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \ + int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff) + +#define dct_long_mac(out, acc, inq, coeff) \ + int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \ + int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff) + +#define dct_widen(out, inq) \ + int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \ + int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12) + +// wide add +#define dct_wadd(out, a, b) \ + int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \ + int32x4_t out##_h = vaddq_s32(a##_h, b##_h) + +// wide sub +#define dct_wsub(out, a, b) \ + int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \ + int32x4_t out##_h = vsubq_s32(a##_h, b##_h) + +// butterfly a/b, then shift using "shiftop" by "s" and pack +#define dct_bfly32o(out0,out1, a,b,shiftop,s) \ + { \ + dct_wadd(sum, a, b); \ + dct_wsub(dif, a, b); \ + out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \ + out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \ + } + +#define dct_pass(shiftop, shift) \ + { \ + /* even part */ \ + int16x8_t sum26 = vaddq_s16(row2, row6); \ + dct_long_mul(p1e, sum26, rot0_0); \ + dct_long_mac(t2e, p1e, row6, rot0_1); \ + dct_long_mac(t3e, p1e, row2, rot0_2); \ + int16x8_t sum04 = vaddq_s16(row0, row4); \ + int16x8_t dif04 = vsubq_s16(row0, row4); \ + dct_widen(t0e, sum04); \ + dct_widen(t1e, dif04); \ + dct_wadd(x0, t0e, t3e); \ + dct_wsub(x3, t0e, t3e); \ + dct_wadd(x1, t1e, t2e); \ + dct_wsub(x2, t1e, t2e); \ + /* odd part */ \ + int16x8_t sum15 = vaddq_s16(row1, row5); \ + int16x8_t sum17 = vaddq_s16(row1, row7); \ + int16x8_t sum35 = vaddq_s16(row3, row5); \ + int16x8_t sum37 = vaddq_s16(row3, row7); \ + int16x8_t sumodd = vaddq_s16(sum17, sum35); \ + dct_long_mul(p5o, sumodd, rot1_0); \ + dct_long_mac(p1o, p5o, sum17, rot1_1); \ + dct_long_mac(p2o, p5o, sum35, rot1_2); \ + dct_long_mul(p3o, sum37, rot2_0); \ + dct_long_mul(p4o, sum15, rot2_1); \ + dct_wadd(sump13o, p1o, p3o); \ + dct_wadd(sump24o, p2o, p4o); \ + dct_wadd(sump23o, p2o, p3o); \ + dct_wadd(sump14o, p1o, p4o); \ + dct_long_mac(x4, sump13o, row7, rot3_0); \ + dct_long_mac(x5, sump24o, row5, rot3_1); \ + dct_long_mac(x6, sump23o, row3, rot3_2); \ + dct_long_mac(x7, sump14o, row1, rot3_3); \ + dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \ + dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \ + dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \ + dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \ + } + + // load + row0 = vld1q_s16(data + 0*8); + row1 = vld1q_s16(data + 1*8); + row2 = vld1q_s16(data + 2*8); + row3 = vld1q_s16(data + 3*8); + row4 = vld1q_s16(data + 4*8); + row5 = vld1q_s16(data + 5*8); + row6 = vld1q_s16(data + 6*8); + row7 = vld1q_s16(data + 7*8); + + // add DC bias + row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0)); + + // column pass + dct_pass(vrshrn_n_s32, 10); + + // 16bit 8x8 transpose + { +// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively. +// whether compilers actually get this is another story, sadly. +#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; } +#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); } +#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); } + + // pass 1 + dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6 + dct_trn16(row2, row3); + dct_trn16(row4, row5); + dct_trn16(row6, row7); + + // pass 2 + dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4 + dct_trn32(row1, row3); + dct_trn32(row4, row6); + dct_trn32(row5, row7); + + // pass 3 + dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0 + dct_trn64(row1, row5); + dct_trn64(row2, row6); + dct_trn64(row3, row7); + +#undef dct_trn16 +#undef dct_trn32 +#undef dct_trn64 + } + + // row pass + // vrshrn_n_s32 only supports shifts up to 16, we need + // 17. so do a non-rounding shift of 16 first then follow + // up with a rounding shift by 1. + dct_pass(vshrn_n_s32, 16); + + { + // pack and round + uint8x8_t p0 = vqrshrun_n_s16(row0, 1); + uint8x8_t p1 = vqrshrun_n_s16(row1, 1); + uint8x8_t p2 = vqrshrun_n_s16(row2, 1); + uint8x8_t p3 = vqrshrun_n_s16(row3, 1); + uint8x8_t p4 = vqrshrun_n_s16(row4, 1); + uint8x8_t p5 = vqrshrun_n_s16(row5, 1); + uint8x8_t p6 = vqrshrun_n_s16(row6, 1); + uint8x8_t p7 = vqrshrun_n_s16(row7, 1); + + // again, these can translate into one instruction, but often don't. +#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; } +#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); } +#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); } + + // sadly can't use interleaved stores here since we only write + // 8 bytes to each scan line! + + // 8x8 8-bit transpose pass 1 + dct_trn8_8(p0, p1); + dct_trn8_8(p2, p3); + dct_trn8_8(p4, p5); + dct_trn8_8(p6, p7); + + // pass 2 + dct_trn8_16(p0, p2); + dct_trn8_16(p1, p3); + dct_trn8_16(p4, p6); + dct_trn8_16(p5, p7); + + // pass 3 + dct_trn8_32(p0, p4); + dct_trn8_32(p1, p5); + dct_trn8_32(p2, p6); + dct_trn8_32(p3, p7); + + // store + vst1_u8(out, p0); out += out_stride; + vst1_u8(out, p1); out += out_stride; + vst1_u8(out, p2); out += out_stride; + vst1_u8(out, p3); out += out_stride; + vst1_u8(out, p4); out += out_stride; + vst1_u8(out, p5); out += out_stride; + vst1_u8(out, p6); out += out_stride; + vst1_u8(out, p7); + +#undef dct_trn8_8 +#undef dct_trn8_16 +#undef dct_trn8_32 + } + +#undef dct_long_mul +#undef dct_long_mac +#undef dct_widen +#undef dct_wadd +#undef dct_wsub +#undef dct_bfly32o +#undef dct_pass +} + +#endif // STBI_NEON + +#define STBI__MARKER_none 0xff +// if there's a pending marker from the entropy stream, return that +// otherwise, fetch from the stream and get a marker. if there's no +// marker, return 0xff, which is never a valid marker value +static stbi_uc stbi__get_marker(stbi__jpeg *j) +{ + stbi_uc x; + if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; } + x = stbi__get8(j->s); + if (x != 0xff) return STBI__MARKER_none; + while (x == 0xff) + x = stbi__get8(j->s); // consume repeated 0xff fill bytes + return x; +} + +// in each scan, we'll have scan_n components, and the order +// of the components is specified by order[] +#define STBI__RESTART(x) ((x) >= 0xd0 && (x) <= 0xd7) + +// after a restart interval, stbi__jpeg_reset the entropy decoder and +// the dc prediction +static void stbi__jpeg_reset(stbi__jpeg *j) +{ + j->code_bits = 0; + j->code_buffer = 0; + j->nomore = 0; + j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0; + j->marker = STBI__MARKER_none; + j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff; + j->eob_run = 0; + // no more than 1<<31 MCUs if no restart_interal? that's plenty safe, + // since we don't even allow 1<<30 pixels +} + +static int stbi__parse_entropy_coded_data(stbi__jpeg *z) +{ + stbi__jpeg_reset(z); + if (!z->progressive) { + if (z->scan_n == 1) { + int i,j; + STBI_SIMD_ALIGN(short, data[64]); + int n = z->order[0]; + // non-interleaved data, we just need to process one block at a time, + // in trivial scanline order + // number of blocks to do just depends on how many actual "pixels" this + // component has, independent of interleaved MCU blocking and such + int w = (z->img_comp[n].x+7) >> 3; + int h = (z->img_comp[n].y+7) >> 3; + for (j=0; j < h; ++j) { + for (i=0; i < w; ++i) { + int ha = z->img_comp[n].ha; + if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0; + z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data); + // every data block is an MCU, so countdown the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + // if it's NOT a restart, then just bail, so we get corrupt data + // rather than no data + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } else { // interleaved + int i,j,k,x,y; + STBI_SIMD_ALIGN(short, data[64]); + for (j=0; j < z->img_mcu_y; ++j) { + for (i=0; i < z->img_mcu_x; ++i) { + // scan an interleaved mcu... process scan_n components in order + for (k=0; k < z->scan_n; ++k) { + int n = z->order[k]; + // scan out an mcu's worth of this component; that's just determined + // by the basic H and V specified for the component + for (y=0; y < z->img_comp[n].v; ++y) { + for (x=0; x < z->img_comp[n].h; ++x) { + int x2 = (i*z->img_comp[n].h + x)*8; + int y2 = (j*z->img_comp[n].v + y)*8; + int ha = z->img_comp[n].ha; + if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0; + z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data); + } + } + } + // after all interleaved components, that's an interleaved MCU, + // so now count down the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } + } else { + if (z->scan_n == 1) { + int i,j; + int n = z->order[0]; + // non-interleaved data, we just need to process one block at a time, + // in trivial scanline order + // number of blocks to do just depends on how many actual "pixels" this + // component has, independent of interleaved MCU blocking and such + int w = (z->img_comp[n].x+7) >> 3; + int h = (z->img_comp[n].y+7) >> 3; + for (j=0; j < h; ++j) { + for (i=0; i < w; ++i) { + short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w); + if (z->spec_start == 0) { + if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n)) + return 0; + } else { + int ha = z->img_comp[n].ha; + if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha])) + return 0; + } + // every data block is an MCU, so countdown the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } else { // interleaved + int i,j,k,x,y; + for (j=0; j < z->img_mcu_y; ++j) { + for (i=0; i < z->img_mcu_x; ++i) { + // scan an interleaved mcu... process scan_n components in order + for (k=0; k < z->scan_n; ++k) { + int n = z->order[k]; + // scan out an mcu's worth of this component; that's just determined + // by the basic H and V specified for the component + for (y=0; y < z->img_comp[n].v; ++y) { + for (x=0; x < z->img_comp[n].h; ++x) { + int x2 = (i*z->img_comp[n].h + x); + int y2 = (j*z->img_comp[n].v + y); + short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w); + if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n)) + return 0; + } + } + } + // after all interleaved components, that's an interleaved MCU, + // so now count down the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } + } +} + +static void stbi__jpeg_dequantize(short *data, stbi__uint16 *dequant) +{ + int i; + for (i=0; i < 64; ++i) + data[i] *= dequant[i]; +} + +static void stbi__jpeg_finish(stbi__jpeg *z) +{ + if (z->progressive) { + // dequantize and idct the data + int i,j,n; + for (n=0; n < z->s->img_n; ++n) { + int w = (z->img_comp[n].x+7) >> 3; + int h = (z->img_comp[n].y+7) >> 3; + for (j=0; j < h; ++j) { + for (i=0; i < w; ++i) { + short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w); + stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]); + z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data); + } + } + } + } +} + +static int stbi__process_marker(stbi__jpeg *z, int m) +{ + int L; + switch (m) { + case STBI__MARKER_none: // no marker found + return stbi__err("expected marker","Corrupt JPEG"); + + case 0xDD: // DRI - specify restart interval + if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG"); + z->restart_interval = stbi__get16be(z->s); + return 1; + + case 0xDB: // DQT - define quantization table + L = stbi__get16be(z->s)-2; + while (L > 0) { + int q = stbi__get8(z->s); + int p = q >> 4, sixteen = (p != 0); + int t = q & 15,i; + if (p != 0 && p != 1) return stbi__err("bad DQT type","Corrupt JPEG"); + if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG"); + + for (i=0; i < 64; ++i) + z->dequant[t][stbi__jpeg_dezigzag[i]] = (stbi__uint16)(sixteen ? stbi__get16be(z->s) : stbi__get8(z->s)); + L -= (sixteen ? 129 : 65); + } + return L==0; + + case 0xC4: // DHT - define huffman table + L = stbi__get16be(z->s)-2; + while (L > 0) { + stbi_uc *v; + int sizes[16],i,n=0; + int q = stbi__get8(z->s); + int tc = q >> 4; + int th = q & 15; + if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG"); + for (i=0; i < 16; ++i) { + sizes[i] = stbi__get8(z->s); + n += sizes[i]; + } + if(n > 256) return stbi__err("bad DHT header","Corrupt JPEG"); // Loop over i < n would write past end of values! + L -= 17; + if (tc == 0) { + if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0; + v = z->huff_dc[th].values; + } else { + if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0; + v = z->huff_ac[th].values; + } + for (i=0; i < n; ++i) + v[i] = stbi__get8(z->s); + if (tc != 0) + stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th); + L -= n; + } + return L==0; + } + + // check for comment block or APP blocks + if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) { + L = stbi__get16be(z->s); + if (L < 2) { + if (m == 0xFE) + return stbi__err("bad COM len","Corrupt JPEG"); + else + return stbi__err("bad APP len","Corrupt JPEG"); + } + L -= 2; + + if (m == 0xE0 && L >= 5) { // JFIF APP0 segment + static const unsigned char tag[5] = {'J','F','I','F','\0'}; + int ok = 1; + int i; + for (i=0; i < 5; ++i) + if (stbi__get8(z->s) != tag[i]) + ok = 0; + L -= 5; + if (ok) + z->jfif = 1; + } else if (m == 0xEE && L >= 12) { // Adobe APP14 segment + static const unsigned char tag[6] = {'A','d','o','b','e','\0'}; + int ok = 1; + int i; + for (i=0; i < 6; ++i) + if (stbi__get8(z->s) != tag[i]) + ok = 0; + L -= 6; + if (ok) { + stbi__get8(z->s); // version + stbi__get16be(z->s); // flags0 + stbi__get16be(z->s); // flags1 + z->app14_color_transform = stbi__get8(z->s); // color transform + L -= 6; + } + } + + stbi__skip(z->s, L); + return 1; + } + + return stbi__err("unknown marker","Corrupt JPEG"); +} + +// after we see SOS +static int stbi__process_scan_header(stbi__jpeg *z) +{ + int i; + int Ls = stbi__get16be(z->s); + z->scan_n = stbi__get8(z->s); + if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG"); + if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG"); + for (i=0; i < z->scan_n; ++i) { + int id = stbi__get8(z->s), which; + int q = stbi__get8(z->s); + for (which = 0; which < z->s->img_n; ++which) + if (z->img_comp[which].id == id) + break; + if (which == z->s->img_n) return 0; // no match + z->img_comp[which].hd = q >> 4; if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG"); + z->img_comp[which].ha = q & 15; if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG"); + z->order[i] = which; + } + + { + int aa; + z->spec_start = stbi__get8(z->s); + z->spec_end = stbi__get8(z->s); // should be 63, but might be 0 + aa = stbi__get8(z->s); + z->succ_high = (aa >> 4); + z->succ_low = (aa & 15); + if (z->progressive) { + if (z->spec_start > 63 || z->spec_end > 63 || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13) + return stbi__err("bad SOS", "Corrupt JPEG"); + } else { + if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG"); + if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG"); + z->spec_end = 63; + } + } + + return 1; +} + +static int stbi__free_jpeg_components(stbi__jpeg *z, int ncomp, int why) +{ + int i; + for (i=0; i < ncomp; ++i) { + if (z->img_comp[i].raw_data) { + STBI_FREE(z->img_comp[i].raw_data); + z->img_comp[i].raw_data = NULL; + z->img_comp[i].data = NULL; + } + if (z->img_comp[i].raw_coeff) { + STBI_FREE(z->img_comp[i].raw_coeff); + z->img_comp[i].raw_coeff = 0; + z->img_comp[i].coeff = 0; + } + if (z->img_comp[i].linebuf) { + STBI_FREE(z->img_comp[i].linebuf); + z->img_comp[i].linebuf = NULL; + } + } + return why; +} + +static int stbi__process_frame_header(stbi__jpeg *z, int scan) +{ + stbi__context *s = z->s; + int Lf,p,i,q, h_max=1,v_max=1,c; + Lf = stbi__get16be(s); if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG + p = stbi__get8(s); if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline + s->img_y = stbi__get16be(s); if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG + s->img_x = stbi__get16be(s); if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + c = stbi__get8(s); + if (c != 3 && c != 1 && c != 4) return stbi__err("bad component count","Corrupt JPEG"); + s->img_n = c; + for (i=0; i < c; ++i) { + z->img_comp[i].data = NULL; + z->img_comp[i].linebuf = NULL; + } + + if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG"); + + z->rgb = 0; + for (i=0; i < s->img_n; ++i) { + static const unsigned char rgb[3] = { 'R', 'G', 'B' }; + z->img_comp[i].id = stbi__get8(s); + if (s->img_n == 3 && z->img_comp[i].id == rgb[i]) + ++z->rgb; + q = stbi__get8(s); + z->img_comp[i].h = (q >> 4); if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG"); + z->img_comp[i].v = q & 15; if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG"); + z->img_comp[i].tq = stbi__get8(s); if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG"); + } + + if (scan != STBI__SCAN_load) return 1; + + if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0)) return stbi__err("too large", "Image too large to decode"); + + for (i=0; i < s->img_n; ++i) { + if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h; + if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v; + } + + // check that plane subsampling factors are integer ratios; our resamplers can't deal with fractional ratios + // and I've never seen a non-corrupted JPEG file actually use them + for (i=0; i < s->img_n; ++i) { + if (h_max % z->img_comp[i].h != 0) return stbi__err("bad H","Corrupt JPEG"); + if (v_max % z->img_comp[i].v != 0) return stbi__err("bad V","Corrupt JPEG"); + } + + // compute interleaved mcu info + z->img_h_max = h_max; + z->img_v_max = v_max; + z->img_mcu_w = h_max * 8; + z->img_mcu_h = v_max * 8; + // these sizes can't be more than 17 bits + z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w; + z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h; + + for (i=0; i < s->img_n; ++i) { + // number of effective pixels (e.g. for non-interleaved MCU) + z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max; + z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max; + // to simplify generation, we'll allocate enough memory to decode + // the bogus oversized data from using interleaved MCUs and their + // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't + // discard the extra data until colorspace conversion + // + // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier) + // so these muls can't overflow with 32-bit ints (which we require) + z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8; + z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8; + z->img_comp[i].coeff = 0; + z->img_comp[i].raw_coeff = 0; + z->img_comp[i].linebuf = NULL; + z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15); + if (z->img_comp[i].raw_data == NULL) + return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory")); + // align blocks for idct using mmx/sse + z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15); + if (z->progressive) { + // w2, h2 are multiples of 8 (see above) + z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8; + z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8; + z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15); + if (z->img_comp[i].raw_coeff == NULL) + return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory")); + z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15); + } + } + + return 1; +} + +// use comparisons since in some cases we handle more than one case (e.g. SOF) +#define stbi__DNL(x) ((x) == 0xdc) +#define stbi__SOI(x) ((x) == 0xd8) +#define stbi__EOI(x) ((x) == 0xd9) +#define stbi__SOF(x) ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2) +#define stbi__SOS(x) ((x) == 0xda) + +#define stbi__SOF_progressive(x) ((x) == 0xc2) + +static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan) +{ + int m; + z->jfif = 0; + z->app14_color_transform = -1; // valid values are 0,1,2 + z->marker = STBI__MARKER_none; // initialize cached marker to empty + m = stbi__get_marker(z); + if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG"); + if (scan == STBI__SCAN_type) return 1; + m = stbi__get_marker(z); + while (!stbi__SOF(m)) { + if (!stbi__process_marker(z,m)) return 0; + m = stbi__get_marker(z); + while (m == STBI__MARKER_none) { + // some files have extra padding after their blocks, so ok, we'll scan + if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG"); + m = stbi__get_marker(z); + } + } + z->progressive = stbi__SOF_progressive(m); + if (!stbi__process_frame_header(z, scan)) return 0; + return 1; +} + +static int stbi__skip_jpeg_junk_at_end(stbi__jpeg *j) +{ + // some JPEGs have junk at end, skip over it but if we find what looks + // like a valid marker, resume there + while (!stbi__at_eof(j->s)) { + int x = stbi__get8(j->s); + while (x == 255) { // might be a marker + if (stbi__at_eof(j->s)) return STBI__MARKER_none; + x = stbi__get8(j->s); + if (x != 0x00 && x != 0xff) { + // not a stuffed zero or lead-in to another marker, looks + // like an actual marker, return it + return x; + } + // stuffed zero has x=0 now which ends the loop, meaning we go + // back to regular scan loop. + // repeated 0xff keeps trying to read the next byte of the marker. + } + } + return STBI__MARKER_none; +} + +// decode image to YCbCr format +static int stbi__decode_jpeg_image(stbi__jpeg *j) +{ + int m; + for (m = 0; m < 4; m++) { + j->img_comp[m].raw_data = NULL; + j->img_comp[m].raw_coeff = NULL; + } + j->restart_interval = 0; + if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0; + m = stbi__get_marker(j); + while (!stbi__EOI(m)) { + if (stbi__SOS(m)) { + if (!stbi__process_scan_header(j)) return 0; + if (!stbi__parse_entropy_coded_data(j)) return 0; + if (j->marker == STBI__MARKER_none ) { + j->marker = stbi__skip_jpeg_junk_at_end(j); + // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0 + } + m = stbi__get_marker(j); + if (STBI__RESTART(m)) + m = stbi__get_marker(j); + } else if (stbi__DNL(m)) { + int Ld = stbi__get16be(j->s); + stbi__uint32 NL = stbi__get16be(j->s); + if (Ld != 4) return stbi__err("bad DNL len", "Corrupt JPEG"); + if (NL != j->s->img_y) return stbi__err("bad DNL height", "Corrupt JPEG"); + m = stbi__get_marker(j); + } else { + if (!stbi__process_marker(j, m)) return 1; + m = stbi__get_marker(j); + } + } + if (j->progressive) + stbi__jpeg_finish(j); + return 1; +} + +// static jfif-centered resampling (across block boundaries) + +typedef stbi_uc *(*resample_row_func)(stbi_uc *out, stbi_uc *in0, stbi_uc *in1, + int w, int hs); + +#define stbi__div4(x) ((stbi_uc) ((x) >> 2)) + +static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + STBI_NOTUSED(out); + STBI_NOTUSED(in_far); + STBI_NOTUSED(w); + STBI_NOTUSED(hs); + return in_near; +} + +static stbi_uc* stbi__resample_row_v_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate two samples vertically for every one in input + int i; + STBI_NOTUSED(hs); + for (i=0; i < w; ++i) + out[i] = stbi__div4(3*in_near[i] + in_far[i] + 2); + return out; +} + +static stbi_uc* stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate two samples horizontally for every one in input + int i; + stbi_uc *input = in_near; + + if (w == 1) { + // if only one sample, can't do any interpolation + out[0] = out[1] = input[0]; + return out; + } + + out[0] = input[0]; + out[1] = stbi__div4(input[0]*3 + input[1] + 2); + for (i=1; i < w-1; ++i) { + int n = 3*input[i]+2; + out[i*2+0] = stbi__div4(n+input[i-1]); + out[i*2+1] = stbi__div4(n+input[i+1]); + } + out[i*2+0] = stbi__div4(input[w-2]*3 + input[w-1] + 2); + out[i*2+1] = input[w-1]; + + STBI_NOTUSED(in_far); + STBI_NOTUSED(hs); + + return out; +} + +#define stbi__div16(x) ((stbi_uc) ((x) >> 4)) + +static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate 2x2 samples for every one in input + int i,t0,t1; + if (w == 1) { + out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2); + return out; + } + + t1 = 3*in_near[0] + in_far[0]; + out[0] = stbi__div4(t1+2); + for (i=1; i < w; ++i) { + t0 = t1; + t1 = 3*in_near[i]+in_far[i]; + out[i*2-1] = stbi__div16(3*t0 + t1 + 8); + out[i*2 ] = stbi__div16(3*t1 + t0 + 8); + } + out[w*2-1] = stbi__div4(t1+2); + + STBI_NOTUSED(hs); + + return out; +} + +#if defined(STBI_SSE2) || defined(STBI_NEON) +static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate 2x2 samples for every one in input + int i=0,t0,t1; + + if (w == 1) { + out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2); + return out; + } + + t1 = 3*in_near[0] + in_far[0]; + // process groups of 8 pixels for as long as we can. + // note we can't handle the last pixel in a row in this loop + // because we need to handle the filter boundary conditions. + for (; i < ((w-1) & ~7); i += 8) { +#if defined(STBI_SSE2) + // load and perform the vertical filtering pass + // this uses 3*x + y = 4*x + (y - x) + __m128i zero = _mm_setzero_si128(); + __m128i farb = _mm_loadl_epi64((__m128i *) (in_far + i)); + __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i)); + __m128i farw = _mm_unpacklo_epi8(farb, zero); + __m128i nearw = _mm_unpacklo_epi8(nearb, zero); + __m128i diff = _mm_sub_epi16(farw, nearw); + __m128i nears = _mm_slli_epi16(nearw, 2); + __m128i curr = _mm_add_epi16(nears, diff); // current row + + // horizontal filter works the same based on shifted vers of current + // row. "prev" is current row shifted right by 1 pixel; we need to + // insert the previous pixel value (from t1). + // "next" is current row shifted left by 1 pixel, with first pixel + // of next block of 8 pixels added in. + __m128i prv0 = _mm_slli_si128(curr, 2); + __m128i nxt0 = _mm_srli_si128(curr, 2); + __m128i prev = _mm_insert_epi16(prv0, t1, 0); + __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7); + + // horizontal filter, polyphase implementation since it's convenient: + // even pixels = 3*cur + prev = cur*4 + (prev - cur) + // odd pixels = 3*cur + next = cur*4 + (next - cur) + // note the shared term. + __m128i bias = _mm_set1_epi16(8); + __m128i curs = _mm_slli_epi16(curr, 2); + __m128i prvd = _mm_sub_epi16(prev, curr); + __m128i nxtd = _mm_sub_epi16(next, curr); + __m128i curb = _mm_add_epi16(curs, bias); + __m128i even = _mm_add_epi16(prvd, curb); + __m128i odd = _mm_add_epi16(nxtd, curb); + + // interleave even and odd pixels, then undo scaling. + __m128i int0 = _mm_unpacklo_epi16(even, odd); + __m128i int1 = _mm_unpackhi_epi16(even, odd); + __m128i de0 = _mm_srli_epi16(int0, 4); + __m128i de1 = _mm_srli_epi16(int1, 4); + + // pack and write output + __m128i outv = _mm_packus_epi16(de0, de1); + _mm_storeu_si128((__m128i *) (out + i*2), outv); +#elif defined(STBI_NEON) + // load and perform the vertical filtering pass + // this uses 3*x + y = 4*x + (y - x) + uint8x8_t farb = vld1_u8(in_far + i); + uint8x8_t nearb = vld1_u8(in_near + i); + int16x8_t diff = vreinterpretq_s16_u16(vsubl_u8(farb, nearb)); + int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2)); + int16x8_t curr = vaddq_s16(nears, diff); // current row + + // horizontal filter works the same based on shifted vers of current + // row. "prev" is current row shifted right by 1 pixel; we need to + // insert the previous pixel value (from t1). + // "next" is current row shifted left by 1 pixel, with first pixel + // of next block of 8 pixels added in. + int16x8_t prv0 = vextq_s16(curr, curr, 7); + int16x8_t nxt0 = vextq_s16(curr, curr, 1); + int16x8_t prev = vsetq_lane_s16(t1, prv0, 0); + int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7); + + // horizontal filter, polyphase implementation since it's convenient: + // even pixels = 3*cur + prev = cur*4 + (prev - cur) + // odd pixels = 3*cur + next = cur*4 + (next - cur) + // note the shared term. + int16x8_t curs = vshlq_n_s16(curr, 2); + int16x8_t prvd = vsubq_s16(prev, curr); + int16x8_t nxtd = vsubq_s16(next, curr); + int16x8_t even = vaddq_s16(curs, prvd); + int16x8_t odd = vaddq_s16(curs, nxtd); + + // undo scaling and round, then store with even/odd phases interleaved + uint8x8x2_t o; + o.val[0] = vqrshrun_n_s16(even, 4); + o.val[1] = vqrshrun_n_s16(odd, 4); + vst2_u8(out + i*2, o); +#endif + + // "previous" value for next iter + t1 = 3*in_near[i+7] + in_far[i+7]; + } + + t0 = t1; + t1 = 3*in_near[i] + in_far[i]; + out[i*2] = stbi__div16(3*t1 + t0 + 8); + + for (++i; i < w; ++i) { + t0 = t1; + t1 = 3*in_near[i]+in_far[i]; + out[i*2-1] = stbi__div16(3*t0 + t1 + 8); + out[i*2 ] = stbi__div16(3*t1 + t0 + 8); + } + out[w*2-1] = stbi__div4(t1+2); + + STBI_NOTUSED(hs); + + return out; +} +#endif + +static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // resample with nearest-neighbor + int i,j; + STBI_NOTUSED(in_far); + for (i=0; i < w; ++i) + for (j=0; j < hs; ++j) + out[i*hs+j] = in_near[i]; + return out; +} + +// this is a reduced-precision calculation of YCbCr-to-RGB introduced +// to make sure the code produces the same results in both SIMD and scalar +#define stbi__float2fixed(x) (((int) ((x) * 4096.0f + 0.5f)) << 8) +static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step) +{ + int i; + for (i=0; i < count; ++i) { + int y_fixed = (y[i] << 20) + (1<<19); // rounding + int r,g,b; + int cr = pcr[i] - 128; + int cb = pcb[i] - 128; + r = y_fixed + cr* stbi__float2fixed(1.40200f); + g = y_fixed + (cr*-stbi__float2fixed(0.71414f)) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000); + b = y_fixed + cb* stbi__float2fixed(1.77200f); + r >>= 20; + g >>= 20; + b >>= 20; + if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; } + if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; } + if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; } + out[0] = (stbi_uc)r; + out[1] = (stbi_uc)g; + out[2] = (stbi_uc)b; + out[3] = 255; + out += step; + } +} + +#if defined(STBI_SSE2) || defined(STBI_NEON) +static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step) +{ + int i = 0; + +#ifdef STBI_SSE2 + // step == 3 is pretty ugly on the final interleave, and i'm not convinced + // it's useful in practice (you wouldn't use it for textures, for example). + // so just accelerate step == 4 case. + if (step == 4) { + // this is a fairly straightforward implementation and not super-optimized. + __m128i signflip = _mm_set1_epi8(-0x80); + __m128i cr_const0 = _mm_set1_epi16( (short) ( 1.40200f*4096.0f+0.5f)); + __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f)); + __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f)); + __m128i cb_const1 = _mm_set1_epi16( (short) ( 1.77200f*4096.0f+0.5f)); + __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128); + __m128i xw = _mm_set1_epi16(255); // alpha channel + + for (; i+7 < count; i += 8) { + // load + __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i)); + __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i)); + __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i)); + __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128 + __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128 + + // unpack to short (and left-shift cr, cb by 8) + __m128i yw = _mm_unpacklo_epi8(y_bias, y_bytes); + __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased); + __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased); + + // color transform + __m128i yws = _mm_srli_epi16(yw, 4); + __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw); + __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw); + __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1); + __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1); + __m128i rws = _mm_add_epi16(cr0, yws); + __m128i gwt = _mm_add_epi16(cb0, yws); + __m128i bws = _mm_add_epi16(yws, cb1); + __m128i gws = _mm_add_epi16(gwt, cr1); + + // descale + __m128i rw = _mm_srai_epi16(rws, 4); + __m128i bw = _mm_srai_epi16(bws, 4); + __m128i gw = _mm_srai_epi16(gws, 4); + + // back to byte, set up for transpose + __m128i brb = _mm_packus_epi16(rw, bw); + __m128i gxb = _mm_packus_epi16(gw, xw); + + // transpose to interleave channels + __m128i t0 = _mm_unpacklo_epi8(brb, gxb); + __m128i t1 = _mm_unpackhi_epi8(brb, gxb); + __m128i o0 = _mm_unpacklo_epi16(t0, t1); + __m128i o1 = _mm_unpackhi_epi16(t0, t1); + + // store + _mm_storeu_si128((__m128i *) (out + 0), o0); + _mm_storeu_si128((__m128i *) (out + 16), o1); + out += 32; + } + } +#endif + +#ifdef STBI_NEON + // in this version, step=3 support would be easy to add. but is there demand? + if (step == 4) { + // this is a fairly straightforward implementation and not super-optimized. + uint8x8_t signflip = vdup_n_u8(0x80); + int16x8_t cr_const0 = vdupq_n_s16( (short) ( 1.40200f*4096.0f+0.5f)); + int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f)); + int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f)); + int16x8_t cb_const1 = vdupq_n_s16( (short) ( 1.77200f*4096.0f+0.5f)); + + for (; i+7 < count; i += 8) { + // load + uint8x8_t y_bytes = vld1_u8(y + i); + uint8x8_t cr_bytes = vld1_u8(pcr + i); + uint8x8_t cb_bytes = vld1_u8(pcb + i); + int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip)); + int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip)); + + // expand to s16 + int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4)); + int16x8_t crw = vshll_n_s8(cr_biased, 7); + int16x8_t cbw = vshll_n_s8(cb_biased, 7); + + // color transform + int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0); + int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0); + int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1); + int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1); + int16x8_t rws = vaddq_s16(yws, cr0); + int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1); + int16x8_t bws = vaddq_s16(yws, cb1); + + // undo scaling, round, convert to byte + uint8x8x4_t o; + o.val[0] = vqrshrun_n_s16(rws, 4); + o.val[1] = vqrshrun_n_s16(gws, 4); + o.val[2] = vqrshrun_n_s16(bws, 4); + o.val[3] = vdup_n_u8(255); + + // store, interleaving r/g/b/a + vst4_u8(out, o); + out += 8*4; + } + } +#endif + + for (; i < count; ++i) { + int y_fixed = (y[i] << 20) + (1<<19); // rounding + int r,g,b; + int cr = pcr[i] - 128; + int cb = pcb[i] - 128; + r = y_fixed + cr* stbi__float2fixed(1.40200f); + g = y_fixed + cr*-stbi__float2fixed(0.71414f) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000); + b = y_fixed + cb* stbi__float2fixed(1.77200f); + r >>= 20; + g >>= 20; + b >>= 20; + if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; } + if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; } + if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; } + out[0] = (stbi_uc)r; + out[1] = (stbi_uc)g; + out[2] = (stbi_uc)b; + out[3] = 255; + out += step; + } +} +#endif + +// set up the kernels +static void stbi__setup_jpeg(stbi__jpeg *j) +{ + j->idct_block_kernel = stbi__idct_block; + j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row; + j->resample_row_hv_2_kernel = stbi__resample_row_hv_2; + +#ifdef STBI_SSE2 + if (stbi__sse2_available()) { + j->idct_block_kernel = stbi__idct_simd; + j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd; + j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd; + } +#endif + +#ifdef STBI_NEON + j->idct_block_kernel = stbi__idct_simd; + j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd; + j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd; +#endif +} + +// clean up the temporary component buffers +static void stbi__cleanup_jpeg(stbi__jpeg *j) +{ + stbi__free_jpeg_components(j, j->s->img_n, 0); +} + +typedef struct +{ + resample_row_func resample; + stbi_uc *line0,*line1; + int hs,vs; // expansion factor in each axis + int w_lores; // horizontal pixels pre-expansion + int ystep; // how far through vertical expansion we are + int ypos; // which pre-expansion row we're on +} stbi__resample; + +// fast 0..255 * 0..255 => 0..255 rounded multiplication +static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y) +{ + unsigned int t = x*y + 128; + return (stbi_uc) ((t + (t >>8)) >> 8); +} + +static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp) +{ + int n, decode_n, is_rgb; + z->s->img_n = 0; // make stbi__cleanup_jpeg safe + + // validate req_comp + if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error"); + + // load a jpeg image from whichever source, but leave in YCbCr format + if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; } + + // determine actual number of components to generate + n = req_comp ? req_comp : z->s->img_n >= 3 ? 3 : 1; + + is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif)); + + if (z->s->img_n == 3 && n < 3 && !is_rgb) + decode_n = 1; + else + decode_n = z->s->img_n; + + // nothing to do if no components requested; check this now to avoid + // accessing uninitialized coutput[0] later + if (decode_n <= 0) { stbi__cleanup_jpeg(z); return NULL; } + + // resample and color-convert + { + int k; + unsigned int i,j; + stbi_uc *output; + stbi_uc *coutput[4] = { NULL, NULL, NULL, NULL }; + + stbi__resample res_comp[4]; + + for (k=0; k < decode_n; ++k) { + stbi__resample *r = &res_comp[k]; + + // allocate line buffer big enough for upsampling off the edges + // with upsample factor of 4 + z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3); + if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); } + + r->hs = z->img_h_max / z->img_comp[k].h; + r->vs = z->img_v_max / z->img_comp[k].v; + r->ystep = r->vs >> 1; + r->w_lores = (z->s->img_x + r->hs-1) / r->hs; + r->ypos = 0; + r->line0 = r->line1 = z->img_comp[k].data; + + if (r->hs == 1 && r->vs == 1) r->resample = resample_row_1; + else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2; + else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2; + else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel; + else r->resample = stbi__resample_row_generic; + } + + // can't error after this so, this is safe + output = (stbi_uc *) stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1); + if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); } + + // now go ahead and resample + for (j=0; j < z->s->img_y; ++j) { + stbi_uc *out = output + n * z->s->img_x * j; + for (k=0; k < decode_n; ++k) { + stbi__resample *r = &res_comp[k]; + int y_bot = r->ystep >= (r->vs >> 1); + coutput[k] = r->resample(z->img_comp[k].linebuf, + y_bot ? r->line1 : r->line0, + y_bot ? r->line0 : r->line1, + r->w_lores, r->hs); + if (++r->ystep >= r->vs) { + r->ystep = 0; + r->line0 = r->line1; + if (++r->ypos < z->img_comp[k].y) + r->line1 += z->img_comp[k].w2; + } + } + if (n >= 3) { + stbi_uc *y = coutput[0]; + if (z->s->img_n == 3) { + if (is_rgb) { + for (i=0; i < z->s->img_x; ++i) { + out[0] = y[i]; + out[1] = coutput[1][i]; + out[2] = coutput[2][i]; + out[3] = 255; + out += n; + } + } else { + z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n); + } + } else if (z->s->img_n == 4) { + if (z->app14_color_transform == 0) { // CMYK + for (i=0; i < z->s->img_x; ++i) { + stbi_uc m = coutput[3][i]; + out[0] = stbi__blinn_8x8(coutput[0][i], m); + out[1] = stbi__blinn_8x8(coutput[1][i], m); + out[2] = stbi__blinn_8x8(coutput[2][i], m); + out[3] = 255; + out += n; + } + } else if (z->app14_color_transform == 2) { // YCCK + z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n); + for (i=0; i < z->s->img_x; ++i) { + stbi_uc m = coutput[3][i]; + out[0] = stbi__blinn_8x8(255 - out[0], m); + out[1] = stbi__blinn_8x8(255 - out[1], m); + out[2] = stbi__blinn_8x8(255 - out[2], m); + out += n; + } + } else { // YCbCr + alpha? Ignore the fourth channel for now + z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n); + } + } else + for (i=0; i < z->s->img_x; ++i) { + out[0] = out[1] = out[2] = y[i]; + out[3] = 255; // not used if n==3 + out += n; + } + } else { + if (is_rgb) { + if (n == 1) + for (i=0; i < z->s->img_x; ++i) + *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]); + else { + for (i=0; i < z->s->img_x; ++i, out += 2) { + out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]); + out[1] = 255; + } + } + } else if (z->s->img_n == 4 && z->app14_color_transform == 0) { + for (i=0; i < z->s->img_x; ++i) { + stbi_uc m = coutput[3][i]; + stbi_uc r = stbi__blinn_8x8(coutput[0][i], m); + stbi_uc g = stbi__blinn_8x8(coutput[1][i], m); + stbi_uc b = stbi__blinn_8x8(coutput[2][i], m); + out[0] = stbi__compute_y(r, g, b); + out[1] = 255; + out += n; + } + } else if (z->s->img_n == 4 && z->app14_color_transform == 2) { + for (i=0; i < z->s->img_x; ++i) { + out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]); + out[1] = 255; + out += n; + } + } else { + stbi_uc *y = coutput[0]; + if (n == 1) + for (i=0; i < z->s->img_x; ++i) out[i] = y[i]; + else + for (i=0; i < z->s->img_x; ++i) { *out++ = y[i]; *out++ = 255; } + } + } + } + stbi__cleanup_jpeg(z); + *out_x = z->s->img_x; + *out_y = z->s->img_y; + if (comp) *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output + return output; + } +} + +static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + unsigned char* result; + stbi__jpeg* j = (stbi__jpeg*) stbi__malloc(sizeof(stbi__jpeg)); + if (!j) return stbi__errpuc("outofmem", "Out of memory"); + memset(j, 0, sizeof(stbi__jpeg)); + STBI_NOTUSED(ri); + j->s = s; + stbi__setup_jpeg(j); + result = load_jpeg_image(j, x,y,comp,req_comp); + STBI_FREE(j); + return result; +} + +static int stbi__jpeg_test(stbi__context *s) +{ + int r; + stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg)); + if (!j) return stbi__err("outofmem", "Out of memory"); + memset(j, 0, sizeof(stbi__jpeg)); + j->s = s; + stbi__setup_jpeg(j); + r = stbi__decode_jpeg_header(j, STBI__SCAN_type); + stbi__rewind(s); + STBI_FREE(j); + return r; +} + +static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp) +{ + if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) { + stbi__rewind( j->s ); + return 0; + } + if (x) *x = j->s->img_x; + if (y) *y = j->s->img_y; + if (comp) *comp = j->s->img_n >= 3 ? 3 : 1; + return 1; +} + +static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp) +{ + int result; + stbi__jpeg* j = (stbi__jpeg*) (stbi__malloc(sizeof(stbi__jpeg))); + if (!j) return stbi__err("outofmem", "Out of memory"); + memset(j, 0, sizeof(stbi__jpeg)); + j->s = s; + result = stbi__jpeg_info_raw(j, x, y, comp); + STBI_FREE(j); + return result; +} +#endif + +// public domain zlib decode v0.2 Sean Barrett 2006-11-18 +// simple implementation +// - all input must be provided in an upfront buffer +// - all output is written to a single output buffer (can malloc/realloc) +// performance +// - fast huffman + +#ifndef STBI_NO_ZLIB + +// fast-way is faster to check than jpeg huffman, but slow way is slower +#define STBI__ZFAST_BITS 9 // accelerate all cases in default tables +#define STBI__ZFAST_MASK ((1 << STBI__ZFAST_BITS) - 1) +#define STBI__ZNSYMS 288 // number of symbols in literal/length alphabet + +// zlib-style huffman encoding +// (jpegs packs from left, zlib from right, so can't share code) +typedef struct +{ + stbi__uint16 fast[1 << STBI__ZFAST_BITS]; + stbi__uint16 firstcode[16]; + int maxcode[17]; + stbi__uint16 firstsymbol[16]; + stbi_uc size[STBI__ZNSYMS]; + stbi__uint16 value[STBI__ZNSYMS]; +} stbi__zhuffman; + +stbi_inline static int stbi__bitreverse16(int n) +{ + n = ((n & 0xAAAA) >> 1) | ((n & 0x5555) << 1); + n = ((n & 0xCCCC) >> 2) | ((n & 0x3333) << 2); + n = ((n & 0xF0F0) >> 4) | ((n & 0x0F0F) << 4); + n = ((n & 0xFF00) >> 8) | ((n & 0x00FF) << 8); + return n; +} + +stbi_inline static int stbi__bit_reverse(int v, int bits) +{ + STBI_ASSERT(bits <= 16); + // to bit reverse n bits, reverse 16 and shift + // e.g. 11 bits, bit reverse and shift away 5 + return stbi__bitreverse16(v) >> (16-bits); +} + +static int stbi__zbuild_huffman(stbi__zhuffman *z, const stbi_uc *sizelist, int num) +{ + int i,k=0; + int code, next_code[16], sizes[17]; + + // DEFLATE spec for generating codes + memset(sizes, 0, sizeof(sizes)); + memset(z->fast, 0, sizeof(z->fast)); + for (i=0; i < num; ++i) + ++sizes[sizelist[i]]; + sizes[0] = 0; + for (i=1; i < 16; ++i) + if (sizes[i] > (1 << i)) + return stbi__err("bad sizes", "Corrupt PNG"); + code = 0; + for (i=1; i < 16; ++i) { + next_code[i] = code; + z->firstcode[i] = (stbi__uint16) code; + z->firstsymbol[i] = (stbi__uint16) k; + code = (code + sizes[i]); + if (sizes[i]) + if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG"); + z->maxcode[i] = code << (16-i); // preshift for inner loop + code <<= 1; + k += sizes[i]; + } + z->maxcode[16] = 0x10000; // sentinel + for (i=0; i < num; ++i) { + int s = sizelist[i]; + if (s) { + int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s]; + stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i); + z->size [c] = (stbi_uc ) s; + z->value[c] = (stbi__uint16) i; + if (s <= STBI__ZFAST_BITS) { + int j = stbi__bit_reverse(next_code[s],s); + while (j < (1 << STBI__ZFAST_BITS)) { + z->fast[j] = fastv; + j += (1 << s); + } + } + ++next_code[s]; + } + } + return 1; +} + +// zlib-from-memory implementation for PNG reading +// because PNG allows splitting the zlib stream arbitrarily, +// and it's annoying structurally to have PNG call ZLIB call PNG, +// we require PNG read all the IDATs and combine them into a single +// memory buffer + +typedef struct +{ + stbi_uc *zbuffer, *zbuffer_end; + int num_bits; + stbi__uint32 code_buffer; + + char *zout; + char *zout_start; + char *zout_end; + int z_expandable; + + stbi__zhuffman z_length, z_distance; +} stbi__zbuf; + +stbi_inline static int stbi__zeof(stbi__zbuf *z) +{ + return (z->zbuffer >= z->zbuffer_end); +} + +stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z) +{ + return stbi__zeof(z) ? 0 : *z->zbuffer++; +} + +static void stbi__fill_bits(stbi__zbuf *z) +{ + do { + if (z->code_buffer >= (1U << z->num_bits)) { + z->zbuffer = z->zbuffer_end; /* treat this as EOF so we fail. */ + return; + } + z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits; + z->num_bits += 8; + } while (z->num_bits <= 24); +} + +stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n) +{ + unsigned int k; + if (z->num_bits < n) stbi__fill_bits(z); + k = z->code_buffer & ((1 << n) - 1); + z->code_buffer >>= n; + z->num_bits -= n; + return k; +} + +static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z) +{ + int b,s,k; + // not resolved by fast table, so compute it the slow way + // use jpeg approach, which requires MSbits at top + k = stbi__bit_reverse(a->code_buffer, 16); + for (s=STBI__ZFAST_BITS+1; ; ++s) + if (k < z->maxcode[s]) + break; + if (s >= 16) return -1; // invalid code! + // code size is s, so: + b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s]; + if (b >= STBI__ZNSYMS) return -1; // some data was corrupt somewhere! + if (z->size[b] != s) return -1; // was originally an assert, but report failure instead. + a->code_buffer >>= s; + a->num_bits -= s; + return z->value[b]; +} + +stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z) +{ + int b,s; + if (a->num_bits < 16) { + if (stbi__zeof(a)) { + return -1; /* report error for unexpected end of data. */ + } + stbi__fill_bits(a); + } + b = z->fast[a->code_buffer & STBI__ZFAST_MASK]; + if (b) { + s = b >> 9; + a->code_buffer >>= s; + a->num_bits -= s; + return b & 511; + } + return stbi__zhuffman_decode_slowpath(a, z); +} + +static int stbi__zexpand(stbi__zbuf *z, char *zout, int n) // need to make room for n bytes +{ + char *q; + unsigned int cur, limit, old_limit; + z->zout = zout; + if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG"); + cur = (unsigned int) (z->zout - z->zout_start); + limit = old_limit = (unsigned) (z->zout_end - z->zout_start); + if (UINT_MAX - cur < (unsigned) n) return stbi__err("outofmem", "Out of memory"); + while (cur + n > limit) { + if(limit > UINT_MAX / 2) return stbi__err("outofmem", "Out of memory"); + limit *= 2; + } + q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit); + STBI_NOTUSED(old_limit); + if (q == NULL) return stbi__err("outofmem", "Out of memory"); + z->zout_start = q; + z->zout = q + cur; + z->zout_end = q + limit; + return 1; +} + +static const int stbi__zlength_base[31] = { + 3,4,5,6,7,8,9,10,11,13, + 15,17,19,23,27,31,35,43,51,59, + 67,83,99,115,131,163,195,227,258,0,0 }; + +static const int stbi__zlength_extra[31]= +{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 }; + +static const int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193, +257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0}; + +static const int stbi__zdist_extra[32] = +{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13}; + +static int stbi__parse_huffman_block(stbi__zbuf *a) +{ + char *zout = a->zout; + for(;;) { + int z = stbi__zhuffman_decode(a, &a->z_length); + if (z < 256) { + if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes + if (zout >= a->zout_end) { + if (!stbi__zexpand(a, zout, 1)) return 0; + zout = a->zout; + } + *zout++ = (char) z; + } else { + stbi_uc *p; + int len,dist; + if (z == 256) { + a->zout = zout; + return 1; + } + if (z >= 286) return stbi__err("bad huffman code","Corrupt PNG"); // per DEFLATE, length codes 286 and 287 must not appear in compressed data + z -= 257; + len = stbi__zlength_base[z]; + if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]); + z = stbi__zhuffman_decode(a, &a->z_distance); + if (z < 0 || z >= 30) return stbi__err("bad huffman code","Corrupt PNG"); // per DEFLATE, distance codes 30 and 31 must not appear in compressed data + dist = stbi__zdist_base[z]; + if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]); + if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG"); + if (zout + len > a->zout_end) { + if (!stbi__zexpand(a, zout, len)) return 0; + zout = a->zout; + } + p = (stbi_uc *) (zout - dist); + if (dist == 1) { // run of one byte; common in images. + stbi_uc v = *p; + if (len) { do *zout++ = v; while (--len); } + } else { + if (len) { do *zout++ = *p++; while (--len); } + } + } + } +} + +static int stbi__compute_huffman_codes(stbi__zbuf *a) +{ + static const stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 }; + stbi__zhuffman z_codelength; + stbi_uc lencodes[286+32+137];//padding for maximum single op + stbi_uc codelength_sizes[19]; + int i,n; + + int hlit = stbi__zreceive(a,5) + 257; + int hdist = stbi__zreceive(a,5) + 1; + int hclen = stbi__zreceive(a,4) + 4; + int ntot = hlit + hdist; + + memset(codelength_sizes, 0, sizeof(codelength_sizes)); + for (i=0; i < hclen; ++i) { + int s = stbi__zreceive(a,3); + codelength_sizes[length_dezigzag[i]] = (stbi_uc) s; + } + if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0; + + n = 0; + while (n < ntot) { + int c = stbi__zhuffman_decode(a, &z_codelength); + if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG"); + if (c < 16) + lencodes[n++] = (stbi_uc) c; + else { + stbi_uc fill = 0; + if (c == 16) { + c = stbi__zreceive(a,2)+3; + if (n == 0) return stbi__err("bad codelengths", "Corrupt PNG"); + fill = lencodes[n-1]; + } else if (c == 17) { + c = stbi__zreceive(a,3)+3; + } else if (c == 18) { + c = stbi__zreceive(a,7)+11; + } else { + return stbi__err("bad codelengths", "Corrupt PNG"); + } + if (ntot - n < c) return stbi__err("bad codelengths", "Corrupt PNG"); + memset(lencodes+n, fill, c); + n += c; + } + } + if (n != ntot) return stbi__err("bad codelengths","Corrupt PNG"); + if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0; + if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0; + return 1; +} + +static int stbi__parse_uncompressed_block(stbi__zbuf *a) +{ + stbi_uc header[4]; + int len,nlen,k; + if (a->num_bits & 7) + stbi__zreceive(a, a->num_bits & 7); // discard + // drain the bit-packed data into header + k = 0; + while (a->num_bits > 0) { + header[k++] = (stbi_uc) (a->code_buffer & 255); // suppress MSVC run-time check + a->code_buffer >>= 8; + a->num_bits -= 8; + } + if (a->num_bits < 0) return stbi__err("zlib corrupt","Corrupt PNG"); + // now fill header the normal way + while (k < 4) + header[k++] = stbi__zget8(a); + len = header[1] * 256 + header[0]; + nlen = header[3] * 256 + header[2]; + if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG"); + if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG"); + if (a->zout + len > a->zout_end) + if (!stbi__zexpand(a, a->zout, len)) return 0; + memcpy(a->zout, a->zbuffer, len); + a->zbuffer += len; + a->zout += len; + return 1; +} + +static int stbi__parse_zlib_header(stbi__zbuf *a) +{ + int cmf = stbi__zget8(a); + int cm = cmf & 15; + /* int cinfo = cmf >> 4; */ + int flg = stbi__zget8(a); + if (stbi__zeof(a)) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec + if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec + if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png + if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png + // window = 1 << (8 + cinfo)... but who cares, we fully buffer output + return 1; +} + +static const stbi_uc stbi__zdefault_length[STBI__ZNSYMS] = +{ + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,8,8,8,8,8,8,8,8 +}; +static const stbi_uc stbi__zdefault_distance[32] = +{ + 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5 +}; +/* +Init algorithm: +{ + int i; // use <= to match clearly with spec + for (i=0; i <= 143; ++i) stbi__zdefault_length[i] = 8; + for ( ; i <= 255; ++i) stbi__zdefault_length[i] = 9; + for ( ; i <= 279; ++i) stbi__zdefault_length[i] = 7; + for ( ; i <= 287; ++i) stbi__zdefault_length[i] = 8; + + for (i=0; i <= 31; ++i) stbi__zdefault_distance[i] = 5; +} +*/ + +static int stbi__parse_zlib(stbi__zbuf *a, int parse_header) +{ + int final, type; + if (parse_header) + if (!stbi__parse_zlib_header(a)) return 0; + a->num_bits = 0; + a->code_buffer = 0; + do { + final = stbi__zreceive(a,1); + type = stbi__zreceive(a,2); + if (type == 0) { + if (!stbi__parse_uncompressed_block(a)) return 0; + } else if (type == 3) { + return 0; + } else { + if (type == 1) { + // use fixed code lengths + if (!stbi__zbuild_huffman(&a->z_length , stbi__zdefault_length , STBI__ZNSYMS)) return 0; + if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance, 32)) return 0; + } else { + if (!stbi__compute_huffman_codes(a)) return 0; + } + if (!stbi__parse_huffman_block(a)) return 0; + } + } while (!final); + return 1; +} + +static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header) +{ + a->zout_start = obuf; + a->zout = obuf; + a->zout_end = obuf + olen; + a->z_expandable = exp; + + return stbi__parse_zlib(a, parse_header); +} + +STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen) +{ + stbi__zbuf a; + char *p = (char *) stbi__malloc(initial_size); + if (p == NULL) return NULL; + a.zbuffer = (stbi_uc *) buffer; + a.zbuffer_end = (stbi_uc *) buffer + len; + if (stbi__do_zlib(&a, p, initial_size, 1, 1)) { + if (outlen) *outlen = (int) (a.zout - a.zout_start); + return a.zout_start; + } else { + STBI_FREE(a.zout_start); + return NULL; + } +} + +STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen) +{ + return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen); +} + +STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header) +{ + stbi__zbuf a; + char *p = (char *) stbi__malloc(initial_size); + if (p == NULL) return NULL; + a.zbuffer = (stbi_uc *) buffer; + a.zbuffer_end = (stbi_uc *) buffer + len; + if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) { + if (outlen) *outlen = (int) (a.zout - a.zout_start); + return a.zout_start; + } else { + STBI_FREE(a.zout_start); + return NULL; + } +} + +STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen) +{ + stbi__zbuf a; + a.zbuffer = (stbi_uc *) ibuffer; + a.zbuffer_end = (stbi_uc *) ibuffer + ilen; + if (stbi__do_zlib(&a, obuffer, olen, 0, 1)) + return (int) (a.zout - a.zout_start); + else + return -1; +} + +STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen) +{ + stbi__zbuf a; + char *p = (char *) stbi__malloc(16384); + if (p == NULL) return NULL; + a.zbuffer = (stbi_uc *) buffer; + a.zbuffer_end = (stbi_uc *) buffer+len; + if (stbi__do_zlib(&a, p, 16384, 1, 0)) { + if (outlen) *outlen = (int) (a.zout - a.zout_start); + return a.zout_start; + } else { + STBI_FREE(a.zout_start); + return NULL; + } +} + +STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen) +{ + stbi__zbuf a; + a.zbuffer = (stbi_uc *) ibuffer; + a.zbuffer_end = (stbi_uc *) ibuffer + ilen; + if (stbi__do_zlib(&a, obuffer, olen, 0, 0)) + return (int) (a.zout - a.zout_start); + else + return -1; +} +#endif + +// public domain "baseline" PNG decoder v0.10 Sean Barrett 2006-11-18 +// simple implementation +// - only 8-bit samples +// - no CRC checking +// - allocates lots of intermediate memory +// - avoids problem of streaming data between subsystems +// - avoids explicit window management +// performance +// - uses stb_zlib, a PD zlib implementation with fast huffman decoding + +#ifndef STBI_NO_PNG +typedef struct +{ + stbi__uint32 length; + stbi__uint32 type; +} stbi__pngchunk; + +static stbi__pngchunk stbi__get_chunk_header(stbi__context *s) +{ + stbi__pngchunk c; + c.length = stbi__get32be(s); + c.type = stbi__get32be(s); + return c; +} + +static int stbi__check_png_header(stbi__context *s) +{ + static const stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 }; + int i; + for (i=0; i < 8; ++i) + if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG"); + return 1; +} + +typedef struct +{ + stbi__context *s; + stbi_uc *idata, *expanded, *out; + int depth; +} stbi__png; + + +enum { + STBI__F_none=0, + STBI__F_sub=1, + STBI__F_up=2, + STBI__F_avg=3, + STBI__F_paeth=4, + // synthetic filters used for first scanline to avoid needing a dummy row of 0s + STBI__F_avg_first, + STBI__F_paeth_first +}; + +static stbi_uc first_row_filter[5] = +{ + STBI__F_none, + STBI__F_sub, + STBI__F_none, + STBI__F_avg_first, + STBI__F_paeth_first +}; + +static int stbi__paeth(int a, int b, int c) +{ + int p = a + b - c; + int pa = abs(p-a); + int pb = abs(p-b); + int pc = abs(p-c); + if (pa <= pb && pa <= pc) return a; + if (pb <= pc) return b; + return c; +} + +static const stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 }; + +// create the png data from post-deflated data +static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color) +{ + int bytes = (depth == 16? 2 : 1); + stbi__context *s = a->s; + stbi__uint32 i,j,stride = x*out_n*bytes; + stbi__uint32 img_len, img_width_bytes; + int k; + int img_n = s->img_n; // copy it into a local for later + + int output_bytes = out_n*bytes; + int filter_bytes = img_n*bytes; + int width = x; + + STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1); + a->out = (stbi_uc *) stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into + if (!a->out) return stbi__err("outofmem", "Out of memory"); + + if (!stbi__mad3sizes_valid(img_n, x, depth, 7)) return stbi__err("too large", "Corrupt PNG"); + img_width_bytes = (((img_n * x * depth) + 7) >> 3); + img_len = (img_width_bytes + 1) * y; + + // we used to check for exact match between raw_len and img_len on non-interlaced PNGs, + // but issue #276 reported a PNG in the wild that had extra data at the end (all zeros), + // so just check for raw_len < img_len always. + if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG"); + + for (j=0; j < y; ++j) { + stbi_uc *cur = a->out + stride*j; + stbi_uc *prior; + int filter = *raw++; + + if (filter > 4) + return stbi__err("invalid filter","Corrupt PNG"); + + if (depth < 8) { + if (img_width_bytes > x) return stbi__err("invalid width","Corrupt PNG"); + cur += x*out_n - img_width_bytes; // store output to the rightmost img_len bytes, so we can decode in place + filter_bytes = 1; + width = img_width_bytes; + } + prior = cur - stride; // bugfix: need to compute this after 'cur +=' computation above + + // if first row, use special filter that doesn't sample previous row + if (j == 0) filter = first_row_filter[filter]; + + // handle first byte explicitly + for (k=0; k < filter_bytes; ++k) { + switch (filter) { + case STBI__F_none : cur[k] = raw[k]; break; + case STBI__F_sub : cur[k] = raw[k]; break; + case STBI__F_up : cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break; + case STBI__F_avg : cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); break; + case STBI__F_paeth : cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(0,prior[k],0)); break; + case STBI__F_avg_first : cur[k] = raw[k]; break; + case STBI__F_paeth_first: cur[k] = raw[k]; break; + } + } + + if (depth == 8) { + if (img_n != out_n) + cur[img_n] = 255; // first pixel + raw += img_n; + cur += out_n; + prior += out_n; + } else if (depth == 16) { + if (img_n != out_n) { + cur[filter_bytes] = 255; // first pixel top byte + cur[filter_bytes+1] = 255; // first pixel bottom byte + } + raw += filter_bytes; + cur += output_bytes; + prior += output_bytes; + } else { + raw += 1; + cur += 1; + prior += 1; + } + + // this is a little gross, so that we don't switch per-pixel or per-component + if (depth < 8 || img_n == out_n) { + int nk = (width - 1)*filter_bytes; + #define STBI__CASE(f) \ + case f: \ + for (k=0; k < nk; ++k) + switch (filter) { + // "none" filter turns into a memcpy here; make that explicit. + case STBI__F_none: memcpy(cur, raw, nk); break; + STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); } break; + STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break; + STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); } break; + STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],prior[k],prior[k-filter_bytes])); } break; + STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); } break; + STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],0,0)); } break; + } + #undef STBI__CASE + raw += nk; + } else { + STBI_ASSERT(img_n+1 == out_n); + #define STBI__CASE(f) \ + case f: \ + for (i=x-1; i >= 1; --i, cur[filter_bytes]=255,raw+=filter_bytes,cur+=output_bytes,prior+=output_bytes) \ + for (k=0; k < filter_bytes; ++k) + switch (filter) { + STBI__CASE(STBI__F_none) { cur[k] = raw[k]; } break; + STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k- output_bytes]); } break; + STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break; + STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k- output_bytes])>>1)); } break; + STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],prior[k],prior[k- output_bytes])); } break; + STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k- output_bytes] >> 1)); } break; + STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],0,0)); } break; + } + #undef STBI__CASE + + // the loop above sets the high byte of the pixels' alpha, but for + // 16 bit png files we also need the low byte set. we'll do that here. + if (depth == 16) { + cur = a->out + stride*j; // start at the beginning of the row again + for (i=0; i < x; ++i,cur+=output_bytes) { + cur[filter_bytes+1] = 255; + } + } + } + } + + // we make a separate pass to expand bits to pixels; for performance, + // this could run two scanlines behind the above code, so it won't + // intefere with filtering but will still be in the cache. + if (depth < 8) { + for (j=0; j < y; ++j) { + stbi_uc *cur = a->out + stride*j; + stbi_uc *in = a->out + stride*j + x*out_n - img_width_bytes; + // unpack 1/2/4-bit into a 8-bit buffer. allows us to keep the common 8-bit path optimal at minimal cost for 1/2/4-bit + // png guarante byte alignment, if width is not multiple of 8/4/2 we'll decode dummy trailing data that will be skipped in the later loop + stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range + + // note that the final byte might overshoot and write more data than desired. + // we can allocate enough data that this never writes out of memory, but it + // could also overwrite the next scanline. can it overwrite non-empty data + // on the next scanline? yes, consider 1-pixel-wide scanlines with 1-bit-per-pixel. + // so we need to explicitly clamp the final ones + + if (depth == 4) { + for (k=x*img_n; k >= 2; k-=2, ++in) { + *cur++ = scale * ((*in >> 4) ); + *cur++ = scale * ((*in ) & 0x0f); + } + if (k > 0) *cur++ = scale * ((*in >> 4) ); + } else if (depth == 2) { + for (k=x*img_n; k >= 4; k-=4, ++in) { + *cur++ = scale * ((*in >> 6) ); + *cur++ = scale * ((*in >> 4) & 0x03); + *cur++ = scale * ((*in >> 2) & 0x03); + *cur++ = scale * ((*in ) & 0x03); + } + if (k > 0) *cur++ = scale * ((*in >> 6) ); + if (k > 1) *cur++ = scale * ((*in >> 4) & 0x03); + if (k > 2) *cur++ = scale * ((*in >> 2) & 0x03); + } else if (depth == 1) { + for (k=x*img_n; k >= 8; k-=8, ++in) { + *cur++ = scale * ((*in >> 7) ); + *cur++ = scale * ((*in >> 6) & 0x01); + *cur++ = scale * ((*in >> 5) & 0x01); + *cur++ = scale * ((*in >> 4) & 0x01); + *cur++ = scale * ((*in >> 3) & 0x01); + *cur++ = scale * ((*in >> 2) & 0x01); + *cur++ = scale * ((*in >> 1) & 0x01); + *cur++ = scale * ((*in ) & 0x01); + } + if (k > 0) *cur++ = scale * ((*in >> 7) ); + if (k > 1) *cur++ = scale * ((*in >> 6) & 0x01); + if (k > 2) *cur++ = scale * ((*in >> 5) & 0x01); + if (k > 3) *cur++ = scale * ((*in >> 4) & 0x01); + if (k > 4) *cur++ = scale * ((*in >> 3) & 0x01); + if (k > 5) *cur++ = scale * ((*in >> 2) & 0x01); + if (k > 6) *cur++ = scale * ((*in >> 1) & 0x01); + } + if (img_n != out_n) { + int q; + // insert alpha = 255 + cur = a->out + stride*j; + if (img_n == 1) { + for (q=x-1; q >= 0; --q) { + cur[q*2+1] = 255; + cur[q*2+0] = cur[q]; + } + } else { + STBI_ASSERT(img_n == 3); + for (q=x-1; q >= 0; --q) { + cur[q*4+3] = 255; + cur[q*4+2] = cur[q*3+2]; + cur[q*4+1] = cur[q*3+1]; + cur[q*4+0] = cur[q*3+0]; + } + } + } + } + } else if (depth == 16) { + // force the image data from big-endian to platform-native. + // this is done in a separate pass due to the decoding relying + // on the data being untouched, but could probably be done + // per-line during decode if care is taken. + stbi_uc *cur = a->out; + stbi__uint16 *cur16 = (stbi__uint16*)cur; + + for(i=0; i < x*y*out_n; ++i,cur16++,cur+=2) { + *cur16 = (cur[0] << 8) | cur[1]; + } + } + + return 1; +} + +static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced) +{ + int bytes = (depth == 16 ? 2 : 1); + int out_bytes = out_n * bytes; + stbi_uc *final; + int p; + if (!interlaced) + return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color); + + // de-interlacing + final = (stbi_uc *) stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0); + if (!final) return stbi__err("outofmem", "Out of memory"); + for (p=0; p < 7; ++p) { + int xorig[] = { 0,4,0,2,0,1,0 }; + int yorig[] = { 0,0,4,0,2,0,1 }; + int xspc[] = { 8,8,4,4,2,2,1 }; + int yspc[] = { 8,8,8,4,4,2,2 }; + int i,j,x,y; + // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1 + x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p]; + y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p]; + if (x && y) { + stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y; + if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) { + STBI_FREE(final); + return 0; + } + for (j=0; j < y; ++j) { + for (i=0; i < x; ++i) { + int out_y = j*yspc[p]+yorig[p]; + int out_x = i*xspc[p]+xorig[p]; + memcpy(final + out_y*a->s->img_x*out_bytes + out_x*out_bytes, + a->out + (j*x+i)*out_bytes, out_bytes); + } + } + STBI_FREE(a->out); + image_data += img_len; + image_data_len -= img_len; + } + } + a->out = final; + + return 1; +} + +static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n) +{ + stbi__context *s = z->s; + stbi__uint32 i, pixel_count = s->img_x * s->img_y; + stbi_uc *p = z->out; + + // compute color-based transparency, assuming we've + // already got 255 as the alpha value in the output + STBI_ASSERT(out_n == 2 || out_n == 4); + + if (out_n == 2) { + for (i=0; i < pixel_count; ++i) { + p[1] = (p[0] == tc[0] ? 0 : 255); + p += 2; + } + } else { + for (i=0; i < pixel_count; ++i) { + if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2]) + p[3] = 0; + p += 4; + } + } + return 1; +} + +static int stbi__compute_transparency16(stbi__png *z, stbi__uint16 tc[3], int out_n) +{ + stbi__context *s = z->s; + stbi__uint32 i, pixel_count = s->img_x * s->img_y; + stbi__uint16 *p = (stbi__uint16*) z->out; + + // compute color-based transparency, assuming we've + // already got 65535 as the alpha value in the output + STBI_ASSERT(out_n == 2 || out_n == 4); + + if (out_n == 2) { + for (i = 0; i < pixel_count; ++i) { + p[1] = (p[0] == tc[0] ? 0 : 65535); + p += 2; + } + } else { + for (i = 0; i < pixel_count; ++i) { + if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2]) + p[3] = 0; + p += 4; + } + } + return 1; +} + +static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n) +{ + stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y; + stbi_uc *p, *temp_out, *orig = a->out; + + p = (stbi_uc *) stbi__malloc_mad2(pixel_count, pal_img_n, 0); + if (p == NULL) return stbi__err("outofmem", "Out of memory"); + + // between here and free(out) below, exitting would leak + temp_out = p; + + if (pal_img_n == 3) { + for (i=0; i < pixel_count; ++i) { + int n = orig[i]*4; + p[0] = palette[n ]; + p[1] = palette[n+1]; + p[2] = palette[n+2]; + p += 3; + } + } else { + for (i=0; i < pixel_count; ++i) { + int n = orig[i]*4; + p[0] = palette[n ]; + p[1] = palette[n+1]; + p[2] = palette[n+2]; + p[3] = palette[n+3]; + p += 4; + } + } + STBI_FREE(a->out); + a->out = temp_out; + + STBI_NOTUSED(len); + + return 1; +} + +static int stbi__unpremultiply_on_load_global = 0; +static int stbi__de_iphone_flag_global = 0; + +STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply) +{ + stbi__unpremultiply_on_load_global = flag_true_if_should_unpremultiply; +} + +STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert) +{ + stbi__de_iphone_flag_global = flag_true_if_should_convert; +} + +#ifndef STBI_THREAD_LOCAL +#define stbi__unpremultiply_on_load stbi__unpremultiply_on_load_global +#define stbi__de_iphone_flag stbi__de_iphone_flag_global +#else +static STBI_THREAD_LOCAL int stbi__unpremultiply_on_load_local, stbi__unpremultiply_on_load_set; +static STBI_THREAD_LOCAL int stbi__de_iphone_flag_local, stbi__de_iphone_flag_set; + +STBIDEF void stbi_set_unpremultiply_on_load_thread(int flag_true_if_should_unpremultiply) +{ + stbi__unpremultiply_on_load_local = flag_true_if_should_unpremultiply; + stbi__unpremultiply_on_load_set = 1; +} + +STBIDEF void stbi_convert_iphone_png_to_rgb_thread(int flag_true_if_should_convert) +{ + stbi__de_iphone_flag_local = flag_true_if_should_convert; + stbi__de_iphone_flag_set = 1; +} + +#define stbi__unpremultiply_on_load (stbi__unpremultiply_on_load_set \ + ? stbi__unpremultiply_on_load_local \ + : stbi__unpremultiply_on_load_global) +#define stbi__de_iphone_flag (stbi__de_iphone_flag_set \ + ? stbi__de_iphone_flag_local \ + : stbi__de_iphone_flag_global) +#endif // STBI_THREAD_LOCAL + +static void stbi__de_iphone(stbi__png *z) +{ + stbi__context *s = z->s; + stbi__uint32 i, pixel_count = s->img_x * s->img_y; + stbi_uc *p = z->out; + + if (s->img_out_n == 3) { // convert bgr to rgb + for (i=0; i < pixel_count; ++i) { + stbi_uc t = p[0]; + p[0] = p[2]; + p[2] = t; + p += 3; + } + } else { + STBI_ASSERT(s->img_out_n == 4); + if (stbi__unpremultiply_on_load) { + // convert bgr to rgb and unpremultiply + for (i=0; i < pixel_count; ++i) { + stbi_uc a = p[3]; + stbi_uc t = p[0]; + if (a) { + stbi_uc half = a / 2; + p[0] = (p[2] * 255 + half) / a; + p[1] = (p[1] * 255 + half) / a; + p[2] = ( t * 255 + half) / a; + } else { + p[0] = p[2]; + p[2] = t; + } + p += 4; + } + } else { + // convert bgr to rgb + for (i=0; i < pixel_count; ++i) { + stbi_uc t = p[0]; + p[0] = p[2]; + p[2] = t; + p += 4; + } + } + } +} + +#define STBI__PNG_TYPE(a,b,c,d) (((unsigned) (a) << 24) + ((unsigned) (b) << 16) + ((unsigned) (c) << 8) + (unsigned) (d)) + +static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp) +{ + stbi_uc palette[1024], pal_img_n=0; + stbi_uc has_trans=0, tc[3]={0}; + stbi__uint16 tc16[3]; + stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0; + int first=1,k,interlace=0, color=0, is_iphone=0; + stbi__context *s = z->s; + + z->expanded = NULL; + z->idata = NULL; + z->out = NULL; + + if (!stbi__check_png_header(s)) return 0; + + if (scan == STBI__SCAN_type) return 1; + + for (;;) { + stbi__pngchunk c = stbi__get_chunk_header(s); + switch (c.type) { + case STBI__PNG_TYPE('C','g','B','I'): + is_iphone = 1; + stbi__skip(s, c.length); + break; + case STBI__PNG_TYPE('I','H','D','R'): { + int comp,filter; + if (!first) return stbi__err("multiple IHDR","Corrupt PNG"); + first = 0; + if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG"); + s->img_x = stbi__get32be(s); + s->img_y = stbi__get32be(s); + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + z->depth = stbi__get8(s); if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16) return stbi__err("1/2/4/8/16-bit only","PNG not supported: 1/2/4/8/16-bit only"); + color = stbi__get8(s); if (color > 6) return stbi__err("bad ctype","Corrupt PNG"); + if (color == 3 && z->depth == 16) return stbi__err("bad ctype","Corrupt PNG"); + if (color == 3) pal_img_n = 3; else if (color & 1) return stbi__err("bad ctype","Corrupt PNG"); + comp = stbi__get8(s); if (comp) return stbi__err("bad comp method","Corrupt PNG"); + filter= stbi__get8(s); if (filter) return stbi__err("bad filter method","Corrupt PNG"); + interlace = stbi__get8(s); if (interlace>1) return stbi__err("bad interlace method","Corrupt PNG"); + if (!s->img_x || !s->img_y) return stbi__err("0-pixel image","Corrupt PNG"); + if (!pal_img_n) { + s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0); + if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode"); + } else { + // if paletted, then pal_n is our final components, and + // img_n is # components to decompress/filter. + s->img_n = 1; + if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG"); + } + // even with SCAN_header, have to scan to see if we have a tRNS + break; + } + + case STBI__PNG_TYPE('P','L','T','E'): { + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG"); + pal_len = c.length / 3; + if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG"); + for (i=0; i < pal_len; ++i) { + palette[i*4+0] = stbi__get8(s); + palette[i*4+1] = stbi__get8(s); + palette[i*4+2] = stbi__get8(s); + palette[i*4+3] = 255; + } + break; + } + + case STBI__PNG_TYPE('t','R','N','S'): { + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG"); + if (pal_img_n) { + if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; } + if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG"); + if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG"); + pal_img_n = 4; + for (i=0; i < c.length; ++i) + palette[i*4+3] = stbi__get8(s); + } else { + if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG"); + if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG"); + has_trans = 1; + // non-paletted with tRNS = constant alpha. if header-scanning, we can stop now. + if (scan == STBI__SCAN_header) { ++s->img_n; return 1; } + if (z->depth == 16) { + for (k = 0; k < s->img_n; ++k) tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is + } else { + for (k = 0; k < s->img_n; ++k) tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger + } + } + break; + } + + case STBI__PNG_TYPE('I','D','A','T'): { + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG"); + if (scan == STBI__SCAN_header) { + // header scan definitely stops at first IDAT + if (pal_img_n) + s->img_n = pal_img_n; + return 1; + } + if (c.length > (1u << 30)) return stbi__err("IDAT size limit", "IDAT section larger than 2^30 bytes"); + if ((int)(ioff + c.length) < (int)ioff) return 0; + if (ioff + c.length > idata_limit) { + stbi__uint32 idata_limit_old = idata_limit; + stbi_uc *p; + if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096; + while (ioff + c.length > idata_limit) + idata_limit *= 2; + STBI_NOTUSED(idata_limit_old); + p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory"); + z->idata = p; + } + if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG"); + ioff += c.length; + break; + } + + case STBI__PNG_TYPE('I','E','N','D'): { + stbi__uint32 raw_len, bpl; + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (scan != STBI__SCAN_load) return 1; + if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG"); + // initial guess for decoded data size to avoid unnecessary reallocs + bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component + raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */; + z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone); + if (z->expanded == NULL) return 0; // zlib should set error + STBI_FREE(z->idata); z->idata = NULL; + if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans) + s->img_out_n = s->img_n+1; + else + s->img_out_n = s->img_n; + if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace)) return 0; + if (has_trans) { + if (z->depth == 16) { + if (!stbi__compute_transparency16(z, tc16, s->img_out_n)) return 0; + } else { + if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0; + } + } + if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2) + stbi__de_iphone(z); + if (pal_img_n) { + // pal_img_n == 3 or 4 + s->img_n = pal_img_n; // record the actual colors we had + s->img_out_n = pal_img_n; + if (req_comp >= 3) s->img_out_n = req_comp; + if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n)) + return 0; + } else if (has_trans) { + // non-paletted image with tRNS -> source image has (constant) alpha + ++s->img_n; + } + STBI_FREE(z->expanded); z->expanded = NULL; + // end of PNG chunk, read and skip CRC + stbi__get32be(s); + return 1; + } + + default: + // if critical, fail + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if ((c.type & (1 << 29)) == 0) { + #ifndef STBI_NO_FAILURE_STRINGS + // not threadsafe + static char invalid_chunk[] = "XXXX PNG chunk not known"; + invalid_chunk[0] = STBI__BYTECAST(c.type >> 24); + invalid_chunk[1] = STBI__BYTECAST(c.type >> 16); + invalid_chunk[2] = STBI__BYTECAST(c.type >> 8); + invalid_chunk[3] = STBI__BYTECAST(c.type >> 0); + #endif + return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type"); + } + stbi__skip(s, c.length); + break; + } + // end of PNG chunk, read and skip CRC + stbi__get32be(s); + } +} + +static void *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp, stbi__result_info *ri) +{ + void *result=NULL; + if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error"); + if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) { + if (p->depth <= 8) + ri->bits_per_channel = 8; + else if (p->depth == 16) + ri->bits_per_channel = 16; + else + return stbi__errpuc("bad bits_per_channel", "PNG not supported: unsupported color depth"); + result = p->out; + p->out = NULL; + if (req_comp && req_comp != p->s->img_out_n) { + if (ri->bits_per_channel == 8) + result = stbi__convert_format((unsigned char *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y); + else + result = stbi__convert_format16((stbi__uint16 *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y); + p->s->img_out_n = req_comp; + if (result == NULL) return result; + } + *x = p->s->img_x; + *y = p->s->img_y; + if (n) *n = p->s->img_n; + } + STBI_FREE(p->out); p->out = NULL; + STBI_FREE(p->expanded); p->expanded = NULL; + STBI_FREE(p->idata); p->idata = NULL; + + return result; +} + +static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi__png p; + p.s = s; + return stbi__do_png(&p, x,y,comp,req_comp, ri); +} + +static int stbi__png_test(stbi__context *s) +{ + int r; + r = stbi__check_png_header(s); + stbi__rewind(s); + return r; +} + +static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp) +{ + if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) { + stbi__rewind( p->s ); + return 0; + } + if (x) *x = p->s->img_x; + if (y) *y = p->s->img_y; + if (comp) *comp = p->s->img_n; + return 1; +} + +static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp) +{ + stbi__png p; + p.s = s; + return stbi__png_info_raw(&p, x, y, comp); +} + +static int stbi__png_is16(stbi__context *s) +{ + stbi__png p; + p.s = s; + if (!stbi__png_info_raw(&p, NULL, NULL, NULL)) + return 0; + if (p.depth != 16) { + stbi__rewind(p.s); + return 0; + } + return 1; +} +#endif + +// Microsoft/Windows BMP image + +#ifndef STBI_NO_BMP +static int stbi__bmp_test_raw(stbi__context *s) +{ + int r; + int sz; + if (stbi__get8(s) != 'B') return 0; + if (stbi__get8(s) != 'M') return 0; + stbi__get32le(s); // discard filesize + stbi__get16le(s); // discard reserved + stbi__get16le(s); // discard reserved + stbi__get32le(s); // discard data offset + sz = stbi__get32le(s); + r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124); + return r; +} + +static int stbi__bmp_test(stbi__context *s) +{ + int r = stbi__bmp_test_raw(s); + stbi__rewind(s); + return r; +} + + +// returns 0..31 for the highest set bit +static int stbi__high_bit(unsigned int z) +{ + int n=0; + if (z == 0) return -1; + if (z >= 0x10000) { n += 16; z >>= 16; } + if (z >= 0x00100) { n += 8; z >>= 8; } + if (z >= 0x00010) { n += 4; z >>= 4; } + if (z >= 0x00004) { n += 2; z >>= 2; } + if (z >= 0x00002) { n += 1;/* >>= 1;*/ } + return n; +} + +static int stbi__bitcount(unsigned int a) +{ + a = (a & 0x55555555) + ((a >> 1) & 0x55555555); // max 2 + a = (a & 0x33333333) + ((a >> 2) & 0x33333333); // max 4 + a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits + a = (a + (a >> 8)); // max 16 per 8 bits + a = (a + (a >> 16)); // max 32 per 8 bits + return a & 0xff; +} + +// extract an arbitrarily-aligned N-bit value (N=bits) +// from v, and then make it 8-bits long and fractionally +// extend it to full full range. +static int stbi__shiftsigned(unsigned int v, int shift, int bits) +{ + static unsigned int mul_table[9] = { + 0, + 0xff/*0b11111111*/, 0x55/*0b01010101*/, 0x49/*0b01001001*/, 0x11/*0b00010001*/, + 0x21/*0b00100001*/, 0x41/*0b01000001*/, 0x81/*0b10000001*/, 0x01/*0b00000001*/, + }; + static unsigned int shift_table[9] = { + 0, 0,0,1,0,2,4,6,0, + }; + if (shift < 0) + v <<= -shift; + else + v >>= shift; + STBI_ASSERT(v < 256); + v >>= (8-bits); + STBI_ASSERT(bits >= 0 && bits <= 8); + return (int) ((unsigned) v * mul_table[bits]) >> shift_table[bits]; +} + +typedef struct +{ + int bpp, offset, hsz; + unsigned int mr,mg,mb,ma, all_a; + int extra_read; +} stbi__bmp_data; + +static int stbi__bmp_set_mask_defaults(stbi__bmp_data *info, int compress) +{ + // BI_BITFIELDS specifies masks explicitly, don't override + if (compress == 3) + return 1; + + if (compress == 0) { + if (info->bpp == 16) { + info->mr = 31u << 10; + info->mg = 31u << 5; + info->mb = 31u << 0; + } else if (info->bpp == 32) { + info->mr = 0xffu << 16; + info->mg = 0xffu << 8; + info->mb = 0xffu << 0; + info->ma = 0xffu << 24; + info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0 + } else { + // otherwise, use defaults, which is all-0 + info->mr = info->mg = info->mb = info->ma = 0; + } + return 1; + } + return 0; // error +} + +static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info) +{ + int hsz; + if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP"); + stbi__get32le(s); // discard filesize + stbi__get16le(s); // discard reserved + stbi__get16le(s); // discard reserved + info->offset = stbi__get32le(s); + info->hsz = hsz = stbi__get32le(s); + info->mr = info->mg = info->mb = info->ma = 0; + info->extra_read = 14; + + if (info->offset < 0) return stbi__errpuc("bad BMP", "bad BMP"); + + if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown"); + if (hsz == 12) { + s->img_x = stbi__get16le(s); + s->img_y = stbi__get16le(s); + } else { + s->img_x = stbi__get32le(s); + s->img_y = stbi__get32le(s); + } + if (stbi__get16le(s) != 1) return stbi__errpuc("bad BMP", "bad BMP"); + info->bpp = stbi__get16le(s); + if (hsz != 12) { + int compress = stbi__get32le(s); + if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE"); + if (compress >= 4) return stbi__errpuc("BMP JPEG/PNG", "BMP type not supported: unsupported compression"); // this includes PNG/JPEG modes + if (compress == 3 && info->bpp != 16 && info->bpp != 32) return stbi__errpuc("bad BMP", "bad BMP"); // bitfields requires 16 or 32 bits/pixel + stbi__get32le(s); // discard sizeof + stbi__get32le(s); // discard hres + stbi__get32le(s); // discard vres + stbi__get32le(s); // discard colorsused + stbi__get32le(s); // discard max important + if (hsz == 40 || hsz == 56) { + if (hsz == 56) { + stbi__get32le(s); + stbi__get32le(s); + stbi__get32le(s); + stbi__get32le(s); + } + if (info->bpp == 16 || info->bpp == 32) { + if (compress == 0) { + stbi__bmp_set_mask_defaults(info, compress); + } else if (compress == 3) { + info->mr = stbi__get32le(s); + info->mg = stbi__get32le(s); + info->mb = stbi__get32le(s); + info->extra_read += 12; + // not documented, but generated by photoshop and handled by mspaint + if (info->mr == info->mg && info->mg == info->mb) { + // ?!?!? + return stbi__errpuc("bad BMP", "bad BMP"); + } + } else + return stbi__errpuc("bad BMP", "bad BMP"); + } + } else { + // V4/V5 header + int i; + if (hsz != 108 && hsz != 124) + return stbi__errpuc("bad BMP", "bad BMP"); + info->mr = stbi__get32le(s); + info->mg = stbi__get32le(s); + info->mb = stbi__get32le(s); + info->ma = stbi__get32le(s); + if (compress != 3) // override mr/mg/mb unless in BI_BITFIELDS mode, as per docs + stbi__bmp_set_mask_defaults(info, compress); + stbi__get32le(s); // discard color space + for (i=0; i < 12; ++i) + stbi__get32le(s); // discard color space parameters + if (hsz == 124) { + stbi__get32le(s); // discard rendering intent + stbi__get32le(s); // discard offset of profile data + stbi__get32le(s); // discard size of profile data + stbi__get32le(s); // discard reserved + } + } + } + return (void *) 1; +} + + +static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi_uc *out; + unsigned int mr=0,mg=0,mb=0,ma=0, all_a; + stbi_uc pal[256][4]; + int psize=0,i,j,width; + int flip_vertically, pad, target; + stbi__bmp_data info; + STBI_NOTUSED(ri); + + info.all_a = 255; + if (stbi__bmp_parse_header(s, &info) == NULL) + return NULL; // error code already set + + flip_vertically = ((int) s->img_y) > 0; + s->img_y = abs((int) s->img_y); + + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + mr = info.mr; + mg = info.mg; + mb = info.mb; + ma = info.ma; + all_a = info.all_a; + + if (info.hsz == 12) { + if (info.bpp < 24) + psize = (info.offset - info.extra_read - 24) / 3; + } else { + if (info.bpp < 16) + psize = (info.offset - info.extra_read - info.hsz) >> 2; + } + if (psize == 0) { + // accept some number of extra bytes after the header, but if the offset points either to before + // the header ends or implies a large amount of extra data, reject the file as malformed + int bytes_read_so_far = s->callback_already_read + (int)(s->img_buffer - s->img_buffer_original); + int header_limit = 1024; // max we actually read is below 256 bytes currently. + int extra_data_limit = 256*4; // what ordinarily goes here is a palette; 256 entries*4 bytes is its max size. + if (bytes_read_so_far <= 0 || bytes_read_so_far > header_limit) { + return stbi__errpuc("bad header", "Corrupt BMP"); + } + // we established that bytes_read_so_far is positive and sensible. + // the first half of this test rejects offsets that are either too small positives, or + // negative, and guarantees that info.offset >= bytes_read_so_far > 0. this in turn + // ensures the number computed in the second half of the test can't overflow. + if (info.offset < bytes_read_so_far || info.offset - bytes_read_so_far > extra_data_limit) { + return stbi__errpuc("bad offset", "Corrupt BMP"); + } else { + stbi__skip(s, info.offset - bytes_read_so_far); + } + } + + if (info.bpp == 24 && ma == 0xff000000) + s->img_n = 3; + else + s->img_n = ma ? 4 : 3; + if (req_comp && req_comp >= 3) // we can directly decode 3 or 4 + target = req_comp; + else + target = s->img_n; // if they want monochrome, we'll post-convert + + // sanity-check size + if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0)) + return stbi__errpuc("too large", "Corrupt BMP"); + + out = (stbi_uc *) stbi__malloc_mad3(target, s->img_x, s->img_y, 0); + if (!out) return stbi__errpuc("outofmem", "Out of memory"); + if (info.bpp < 16) { + int z=0; + if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); } + for (i=0; i < psize; ++i) { + pal[i][2] = stbi__get8(s); + pal[i][1] = stbi__get8(s); + pal[i][0] = stbi__get8(s); + if (info.hsz != 12) stbi__get8(s); + pal[i][3] = 255; + } + stbi__skip(s, info.offset - info.extra_read - info.hsz - psize * (info.hsz == 12 ? 3 : 4)); + if (info.bpp == 1) width = (s->img_x + 7) >> 3; + else if (info.bpp == 4) width = (s->img_x + 1) >> 1; + else if (info.bpp == 8) width = s->img_x; + else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); } + pad = (-width)&3; + if (info.bpp == 1) { + for (j=0; j < (int) s->img_y; ++j) { + int bit_offset = 7, v = stbi__get8(s); + for (i=0; i < (int) s->img_x; ++i) { + int color = (v>>bit_offset)&0x1; + out[z++] = pal[color][0]; + out[z++] = pal[color][1]; + out[z++] = pal[color][2]; + if (target == 4) out[z++] = 255; + if (i+1 == (int) s->img_x) break; + if((--bit_offset) < 0) { + bit_offset = 7; + v = stbi__get8(s); + } + } + stbi__skip(s, pad); + } + } else { + for (j=0; j < (int) s->img_y; ++j) { + for (i=0; i < (int) s->img_x; i += 2) { + int v=stbi__get8(s),v2=0; + if (info.bpp == 4) { + v2 = v & 15; + v >>= 4; + } + out[z++] = pal[v][0]; + out[z++] = pal[v][1]; + out[z++] = pal[v][2]; + if (target == 4) out[z++] = 255; + if (i+1 == (int) s->img_x) break; + v = (info.bpp == 8) ? stbi__get8(s) : v2; + out[z++] = pal[v][0]; + out[z++] = pal[v][1]; + out[z++] = pal[v][2]; + if (target == 4) out[z++] = 255; + } + stbi__skip(s, pad); + } + } + } else { + int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0; + int z = 0; + int easy=0; + stbi__skip(s, info.offset - info.extra_read - info.hsz); + if (info.bpp == 24) width = 3 * s->img_x; + else if (info.bpp == 16) width = 2*s->img_x; + else /* bpp = 32 and pad = 0 */ width=0; + pad = (-width) & 3; + if (info.bpp == 24) { + easy = 1; + } else if (info.bpp == 32) { + if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000) + easy = 2; + } + if (!easy) { + if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); } + // right shift amt to put high bit in position #7 + rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr); + gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg); + bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb); + ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma); + if (rcount > 8 || gcount > 8 || bcount > 8 || acount > 8) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); } + } + for (j=0; j < (int) s->img_y; ++j) { + if (easy) { + for (i=0; i < (int) s->img_x; ++i) { + unsigned char a; + out[z+2] = stbi__get8(s); + out[z+1] = stbi__get8(s); + out[z+0] = stbi__get8(s); + z += 3; + a = (easy == 2 ? stbi__get8(s) : 255); + all_a |= a; + if (target == 4) out[z++] = a; + } + } else { + int bpp = info.bpp; + for (i=0; i < (int) s->img_x; ++i) { + stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s)); + unsigned int a; + out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount)); + out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount)); + out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount)); + a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255); + all_a |= a; + if (target == 4) out[z++] = STBI__BYTECAST(a); + } + } + stbi__skip(s, pad); + } + } + + // if alpha channel is all 0s, replace with all 255s + if (target == 4 && all_a == 0) + for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4) + out[i] = 255; + + if (flip_vertically) { + stbi_uc t; + for (j=0; j < (int) s->img_y>>1; ++j) { + stbi_uc *p1 = out + j *s->img_x*target; + stbi_uc *p2 = out + (s->img_y-1-j)*s->img_x*target; + for (i=0; i < (int) s->img_x*target; ++i) { + t = p1[i]; p1[i] = p2[i]; p2[i] = t; + } + } + } + + if (req_comp && req_comp != target) { + out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y); + if (out == NULL) return out; // stbi__convert_format frees input on failure + } + + *x = s->img_x; + *y = s->img_y; + if (comp) *comp = s->img_n; + return out; +} +#endif + +// Targa Truevision - TGA +// by Jonathan Dummer +#ifndef STBI_NO_TGA +// returns STBI_rgb or whatever, 0 on error +static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16) +{ + // only RGB or RGBA (incl. 16bit) or grey allowed + if (is_rgb16) *is_rgb16 = 0; + switch(bits_per_pixel) { + case 8: return STBI_grey; + case 16: if(is_grey) return STBI_grey_alpha; + // fallthrough + case 15: if(is_rgb16) *is_rgb16 = 1; + return STBI_rgb; + case 24: // fallthrough + case 32: return bits_per_pixel/8; + default: return 0; + } +} + +static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp) +{ + int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp; + int sz, tga_colormap_type; + stbi__get8(s); // discard Offset + tga_colormap_type = stbi__get8(s); // colormap type + if( tga_colormap_type > 1 ) { + stbi__rewind(s); + return 0; // only RGB or indexed allowed + } + tga_image_type = stbi__get8(s); // image type + if ( tga_colormap_type == 1 ) { // colormapped (paletted) image + if (tga_image_type != 1 && tga_image_type != 9) { + stbi__rewind(s); + return 0; + } + stbi__skip(s,4); // skip index of first colormap entry and number of entries + sz = stbi__get8(s); // check bits per palette color entry + if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) { + stbi__rewind(s); + return 0; + } + stbi__skip(s,4); // skip image x and y origin + tga_colormap_bpp = sz; + } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE + if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) { + stbi__rewind(s); + return 0; // only RGB or grey allowed, +/- RLE + } + stbi__skip(s,9); // skip colormap specification and image x/y origin + tga_colormap_bpp = 0; + } + tga_w = stbi__get16le(s); + if( tga_w < 1 ) { + stbi__rewind(s); + return 0; // test width + } + tga_h = stbi__get16le(s); + if( tga_h < 1 ) { + stbi__rewind(s); + return 0; // test height + } + tga_bits_per_pixel = stbi__get8(s); // bits per pixel + stbi__get8(s); // ignore alpha bits + if (tga_colormap_bpp != 0) { + if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) { + // when using a colormap, tga_bits_per_pixel is the size of the indexes + // I don't think anything but 8 or 16bit indexes makes sense + stbi__rewind(s); + return 0; + } + tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL); + } else { + tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL); + } + if(!tga_comp) { + stbi__rewind(s); + return 0; + } + if (x) *x = tga_w; + if (y) *y = tga_h; + if (comp) *comp = tga_comp; + return 1; // seems to have passed everything +} + +static int stbi__tga_test(stbi__context *s) +{ + int res = 0; + int sz, tga_color_type; + stbi__get8(s); // discard Offset + tga_color_type = stbi__get8(s); // color type + if ( tga_color_type > 1 ) goto errorEnd; // only RGB or indexed allowed + sz = stbi__get8(s); // image type + if ( tga_color_type == 1 ) { // colormapped (paletted) image + if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9 + stbi__skip(s,4); // skip index of first colormap entry and number of entries + sz = stbi__get8(s); // check bits per palette color entry + if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd; + stbi__skip(s,4); // skip image x and y origin + } else { // "normal" image w/o colormap + if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE + stbi__skip(s,9); // skip colormap specification and image x/y origin + } + if ( stbi__get16le(s) < 1 ) goto errorEnd; // test width + if ( stbi__get16le(s) < 1 ) goto errorEnd; // test height + sz = stbi__get8(s); // bits per pixel + if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index + if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd; + + res = 1; // if we got this far, everything's good and we can return 1 instead of 0 + +errorEnd: + stbi__rewind(s); + return res; +} + +// read 16bit value and convert to 24bit RGB +static void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out) +{ + stbi__uint16 px = (stbi__uint16)stbi__get16le(s); + stbi__uint16 fiveBitMask = 31; + // we have 3 channels with 5bits each + int r = (px >> 10) & fiveBitMask; + int g = (px >> 5) & fiveBitMask; + int b = px & fiveBitMask; + // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later + out[0] = (stbi_uc)((r * 255)/31); + out[1] = (stbi_uc)((g * 255)/31); + out[2] = (stbi_uc)((b * 255)/31); + + // some people claim that the most significant bit might be used for alpha + // (possibly if an alpha-bit is set in the "image descriptor byte") + // but that only made 16bit test images completely translucent.. + // so let's treat all 15 and 16bit TGAs as RGB with no alpha. +} + +static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + // read in the TGA header stuff + int tga_offset = stbi__get8(s); + int tga_indexed = stbi__get8(s); + int tga_image_type = stbi__get8(s); + int tga_is_RLE = 0; + int tga_palette_start = stbi__get16le(s); + int tga_palette_len = stbi__get16le(s); + int tga_palette_bits = stbi__get8(s); + int tga_x_origin = stbi__get16le(s); + int tga_y_origin = stbi__get16le(s); + int tga_width = stbi__get16le(s); + int tga_height = stbi__get16le(s); + int tga_bits_per_pixel = stbi__get8(s); + int tga_comp, tga_rgb16=0; + int tga_inverted = stbi__get8(s); + // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?) + // image data + unsigned char *tga_data; + unsigned char *tga_palette = NULL; + int i, j; + unsigned char raw_data[4] = {0}; + int RLE_count = 0; + int RLE_repeating = 0; + int read_next_pixel = 1; + STBI_NOTUSED(ri); + STBI_NOTUSED(tga_x_origin); // @TODO + STBI_NOTUSED(tga_y_origin); // @TODO + + if (tga_height > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (tga_width > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + // do a tiny bit of precessing + if ( tga_image_type >= 8 ) + { + tga_image_type -= 8; + tga_is_RLE = 1; + } + tga_inverted = 1 - ((tga_inverted >> 5) & 1); + + // If I'm paletted, then I'll use the number of bits from the palette + if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16); + else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16); + + if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency + return stbi__errpuc("bad format", "Can't find out TGA pixelformat"); + + // tga info + *x = tga_width; + *y = tga_height; + if (comp) *comp = tga_comp; + + if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0)) + return stbi__errpuc("too large", "Corrupt TGA"); + + tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0); + if (!tga_data) return stbi__errpuc("outofmem", "Out of memory"); + + // skip to the data's starting position (offset usually = 0) + stbi__skip(s, tga_offset ); + + if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) { + for (i=0; i < tga_height; ++i) { + int row = tga_inverted ? tga_height -i - 1 : i; + stbi_uc *tga_row = tga_data + row*tga_width*tga_comp; + stbi__getn(s, tga_row, tga_width * tga_comp); + } + } else { + // do I need to load a palette? + if ( tga_indexed) + { + if (tga_palette_len == 0) { /* you have to have at least one entry! */ + STBI_FREE(tga_data); + return stbi__errpuc("bad palette", "Corrupt TGA"); + } + + // any data to skip? (offset usually = 0) + stbi__skip(s, tga_palette_start ); + // load the palette + tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0); + if (!tga_palette) { + STBI_FREE(tga_data); + return stbi__errpuc("outofmem", "Out of memory"); + } + if (tga_rgb16) { + stbi_uc *pal_entry = tga_palette; + STBI_ASSERT(tga_comp == STBI_rgb); + for (i=0; i < tga_palette_len; ++i) { + stbi__tga_read_rgb16(s, pal_entry); + pal_entry += tga_comp; + } + } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) { + STBI_FREE(tga_data); + STBI_FREE(tga_palette); + return stbi__errpuc("bad palette", "Corrupt TGA"); + } + } + // load the data + for (i=0; i < tga_width * tga_height; ++i) + { + // if I'm in RLE mode, do I need to get a RLE stbi__pngchunk? + if ( tga_is_RLE ) + { + if ( RLE_count == 0 ) + { + // yep, get the next byte as a RLE command + int RLE_cmd = stbi__get8(s); + RLE_count = 1 + (RLE_cmd & 127); + RLE_repeating = RLE_cmd >> 7; + read_next_pixel = 1; + } else if ( !RLE_repeating ) + { + read_next_pixel = 1; + } + } else + { + read_next_pixel = 1; + } + // OK, if I need to read a pixel, do it now + if ( read_next_pixel ) + { + // load however much data we did have + if ( tga_indexed ) + { + // read in index, then perform the lookup + int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s); + if ( pal_idx >= tga_palette_len ) { + // invalid index + pal_idx = 0; + } + pal_idx *= tga_comp; + for (j = 0; j < tga_comp; ++j) { + raw_data[j] = tga_palette[pal_idx+j]; + } + } else if(tga_rgb16) { + STBI_ASSERT(tga_comp == STBI_rgb); + stbi__tga_read_rgb16(s, raw_data); + } else { + // read in the data raw + for (j = 0; j < tga_comp; ++j) { + raw_data[j] = stbi__get8(s); + } + } + // clear the reading flag for the next pixel + read_next_pixel = 0; + } // end of reading a pixel + + // copy data + for (j = 0; j < tga_comp; ++j) + tga_data[i*tga_comp+j] = raw_data[j]; + + // in case we're in RLE mode, keep counting down + --RLE_count; + } + // do I need to invert the image? + if ( tga_inverted ) + { + for (j = 0; j*2 < tga_height; ++j) + { + int index1 = j * tga_width * tga_comp; + int index2 = (tga_height - 1 - j) * tga_width * tga_comp; + for (i = tga_width * tga_comp; i > 0; --i) + { + unsigned char temp = tga_data[index1]; + tga_data[index1] = tga_data[index2]; + tga_data[index2] = temp; + ++index1; + ++index2; + } + } + } + // clear my palette, if I had one + if ( tga_palette != NULL ) + { + STBI_FREE( tga_palette ); + } + } + + // swap RGB - if the source data was RGB16, it already is in the right order + if (tga_comp >= 3 && !tga_rgb16) + { + unsigned char* tga_pixel = tga_data; + for (i=0; i < tga_width * tga_height; ++i) + { + unsigned char temp = tga_pixel[0]; + tga_pixel[0] = tga_pixel[2]; + tga_pixel[2] = temp; + tga_pixel += tga_comp; + } + } + + // convert to target component count + if (req_comp && req_comp != tga_comp) + tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height); + + // the things I do to get rid of an error message, and yet keep + // Microsoft's C compilers happy... [8^( + tga_palette_start = tga_palette_len = tga_palette_bits = + tga_x_origin = tga_y_origin = 0; + STBI_NOTUSED(tga_palette_start); + // OK, done + return tga_data; +} +#endif + +// ************************************************************************************************* +// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB + +#ifndef STBI_NO_PSD +static int stbi__psd_test(stbi__context *s) +{ + int r = (stbi__get32be(s) == 0x38425053); + stbi__rewind(s); + return r; +} + +static int stbi__psd_decode_rle(stbi__context *s, stbi_uc *p, int pixelCount) +{ + int count, nleft, len; + + count = 0; + while ((nleft = pixelCount - count) > 0) { + len = stbi__get8(s); + if (len == 128) { + // No-op. + } else if (len < 128) { + // Copy next len+1 bytes literally. + len++; + if (len > nleft) return 0; // corrupt data + count += len; + while (len) { + *p = stbi__get8(s); + p += 4; + len--; + } + } else if (len > 128) { + stbi_uc val; + // Next -len+1 bytes in the dest are replicated from next source byte. + // (Interpret len as a negative 8-bit int.) + len = 257 - len; + if (len > nleft) return 0; // corrupt data + val = stbi__get8(s); + count += len; + while (len) { + *p = val; + p += 4; + len--; + } + } + } + + return 1; +} + +static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc) +{ + int pixelCount; + int channelCount, compression; + int channel, i; + int bitdepth; + int w,h; + stbi_uc *out; + STBI_NOTUSED(ri); + + // Check identifier + if (stbi__get32be(s) != 0x38425053) // "8BPS" + return stbi__errpuc("not PSD", "Corrupt PSD image"); + + // Check file type version. + if (stbi__get16be(s) != 1) + return stbi__errpuc("wrong version", "Unsupported version of PSD image"); + + // Skip 6 reserved bytes. + stbi__skip(s, 6 ); + + // Read the number of channels (R, G, B, A, etc). + channelCount = stbi__get16be(s); + if (channelCount < 0 || channelCount > 16) + return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image"); + + // Read the rows and columns of the image. + h = stbi__get32be(s); + w = stbi__get32be(s); + + if (h > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (w > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + // Make sure the depth is 8 bits. + bitdepth = stbi__get16be(s); + if (bitdepth != 8 && bitdepth != 16) + return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit"); + + // Make sure the color mode is RGB. + // Valid options are: + // 0: Bitmap + // 1: Grayscale + // 2: Indexed color + // 3: RGB color + // 4: CMYK color + // 7: Multichannel + // 8: Duotone + // 9: Lab color + if (stbi__get16be(s) != 3) + return stbi__errpuc("wrong color format", "PSD is not in RGB color format"); + + // Skip the Mode Data. (It's the palette for indexed color; other info for other modes.) + stbi__skip(s,stbi__get32be(s) ); + + // Skip the image resources. (resolution, pen tool paths, etc) + stbi__skip(s, stbi__get32be(s) ); + + // Skip the reserved data. + stbi__skip(s, stbi__get32be(s) ); + + // Find out if the data is compressed. + // Known values: + // 0: no compression + // 1: RLE compressed + compression = stbi__get16be(s); + if (compression > 1) + return stbi__errpuc("bad compression", "PSD has an unknown compression format"); + + // Check size + if (!stbi__mad3sizes_valid(4, w, h, 0)) + return stbi__errpuc("too large", "Corrupt PSD"); + + // Create the destination image. + + if (!compression && bitdepth == 16 && bpc == 16) { + out = (stbi_uc *) stbi__malloc_mad3(8, w, h, 0); + ri->bits_per_channel = 16; + } else + out = (stbi_uc *) stbi__malloc(4 * w*h); + + if (!out) return stbi__errpuc("outofmem", "Out of memory"); + pixelCount = w*h; + + // Initialize the data to zero. + //memset( out, 0, pixelCount * 4 ); + + // Finally, the image data. + if (compression) { + // RLE as used by .PSD and .TIFF + // Loop until you get the number of unpacked bytes you are expecting: + // Read the next source byte into n. + // If n is between 0 and 127 inclusive, copy the next n+1 bytes literally. + // Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times. + // Else if n is 128, noop. + // Endloop + + // The RLE-compressed data is preceded by a 2-byte data count for each row in the data, + // which we're going to just skip. + stbi__skip(s, h * channelCount * 2 ); + + // Read the RLE data by channel. + for (channel = 0; channel < 4; channel++) { + stbi_uc *p; + + p = out+channel; + if (channel >= channelCount) { + // Fill this channel with default data. + for (i = 0; i < pixelCount; i++, p += 4) + *p = (channel == 3 ? 255 : 0); + } else { + // Read the RLE data. + if (!stbi__psd_decode_rle(s, p, pixelCount)) { + STBI_FREE(out); + return stbi__errpuc("corrupt", "bad RLE data"); + } + } + } + + } else { + // We're at the raw image data. It's each channel in order (Red, Green, Blue, Alpha, ...) + // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image. + + // Read the data by channel. + for (channel = 0; channel < 4; channel++) { + if (channel >= channelCount) { + // Fill this channel with default data. + if (bitdepth == 16 && bpc == 16) { + stbi__uint16 *q = ((stbi__uint16 *) out) + channel; + stbi__uint16 val = channel == 3 ? 65535 : 0; + for (i = 0; i < pixelCount; i++, q += 4) + *q = val; + } else { + stbi_uc *p = out+channel; + stbi_uc val = channel == 3 ? 255 : 0; + for (i = 0; i < pixelCount; i++, p += 4) + *p = val; + } + } else { + if (ri->bits_per_channel == 16) { // output bpc + stbi__uint16 *q = ((stbi__uint16 *) out) + channel; + for (i = 0; i < pixelCount; i++, q += 4) + *q = (stbi__uint16) stbi__get16be(s); + } else { + stbi_uc *p = out+channel; + if (bitdepth == 16) { // input bpc + for (i = 0; i < pixelCount; i++, p += 4) + *p = (stbi_uc) (stbi__get16be(s) >> 8); + } else { + for (i = 0; i < pixelCount; i++, p += 4) + *p = stbi__get8(s); + } + } + } + } + } + + // remove weird white matte from PSD + if (channelCount >= 4) { + if (ri->bits_per_channel == 16) { + for (i=0; i < w*h; ++i) { + stbi__uint16 *pixel = (stbi__uint16 *) out + 4*i; + if (pixel[3] != 0 && pixel[3] != 65535) { + float a = pixel[3] / 65535.0f; + float ra = 1.0f / a; + float inv_a = 65535.0f * (1 - ra); + pixel[0] = (stbi__uint16) (pixel[0]*ra + inv_a); + pixel[1] = (stbi__uint16) (pixel[1]*ra + inv_a); + pixel[2] = (stbi__uint16) (pixel[2]*ra + inv_a); + } + } + } else { + for (i=0; i < w*h; ++i) { + unsigned char *pixel = out + 4*i; + if (pixel[3] != 0 && pixel[3] != 255) { + float a = pixel[3] / 255.0f; + float ra = 1.0f / a; + float inv_a = 255.0f * (1 - ra); + pixel[0] = (unsigned char) (pixel[0]*ra + inv_a); + pixel[1] = (unsigned char) (pixel[1]*ra + inv_a); + pixel[2] = (unsigned char) (pixel[2]*ra + inv_a); + } + } + } + } + + // convert to desired output format + if (req_comp && req_comp != 4) { + if (ri->bits_per_channel == 16) + out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, 4, req_comp, w, h); + else + out = stbi__convert_format(out, 4, req_comp, w, h); + if (out == NULL) return out; // stbi__convert_format frees input on failure + } + + if (comp) *comp = 4; + *y = h; + *x = w; + + return out; +} +#endif + +// ************************************************************************************************* +// Softimage PIC loader +// by Tom Seddon +// +// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format +// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/ + +#ifndef STBI_NO_PIC +static int stbi__pic_is4(stbi__context *s,const char *str) +{ + int i; + for (i=0; i<4; ++i) + if (stbi__get8(s) != (stbi_uc)str[i]) + return 0; + + return 1; +} + +static int stbi__pic_test_core(stbi__context *s) +{ + int i; + + if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) + return 0; + + for(i=0;i<84;++i) + stbi__get8(s); + + if (!stbi__pic_is4(s,"PICT")) + return 0; + + return 1; +} + +typedef struct +{ + stbi_uc size,type,channel; +} stbi__pic_packet; + +static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest) +{ + int mask=0x80, i; + + for (i=0; i<4; ++i, mask>>=1) { + if (channel & mask) { + if (stbi__at_eof(s)) return stbi__errpuc("bad file","PIC file too short"); + dest[i]=stbi__get8(s); + } + } + + return dest; +} + +static void stbi__copyval(int channel,stbi_uc *dest,const stbi_uc *src) +{ + int mask=0x80,i; + + for (i=0;i<4; ++i, mask>>=1) + if (channel&mask) + dest[i]=src[i]; +} + +static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *comp, stbi_uc *result) +{ + int act_comp=0,num_packets=0,y,chained; + stbi__pic_packet packets[10]; + + // this will (should...) cater for even some bizarre stuff like having data + // for the same channel in multiple packets. + do { + stbi__pic_packet *packet; + + if (num_packets==sizeof(packets)/sizeof(packets[0])) + return stbi__errpuc("bad format","too many packets"); + + packet = &packets[num_packets++]; + + chained = stbi__get8(s); + packet->size = stbi__get8(s); + packet->type = stbi__get8(s); + packet->channel = stbi__get8(s); + + act_comp |= packet->channel; + + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (reading packets)"); + if (packet->size != 8) return stbi__errpuc("bad format","packet isn't 8bpp"); + } while (chained); + + *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel? + + for(y=0; ytype) { + default: + return stbi__errpuc("bad format","packet has bad compression type"); + + case 0: {//uncompressed + int x; + + for(x=0;xchannel,dest)) + return 0; + break; + } + + case 1://Pure RLE + { + int left=width, i; + + while (left>0) { + stbi_uc count,value[4]; + + count=stbi__get8(s); + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pure read count)"); + + if (count > left) + count = (stbi_uc) left; + + if (!stbi__readval(s,packet->channel,value)) return 0; + + for(i=0; ichannel,dest,value); + left -= count; + } + } + break; + + case 2: {//Mixed RLE + int left=width; + while (left>0) { + int count = stbi__get8(s), i; + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (mixed read count)"); + + if (count >= 128) { // Repeated + stbi_uc value[4]; + + if (count==128) + count = stbi__get16be(s); + else + count -= 127; + if (count > left) + return stbi__errpuc("bad file","scanline overrun"); + + if (!stbi__readval(s,packet->channel,value)) + return 0; + + for(i=0;ichannel,dest,value); + } else { // Raw + ++count; + if (count>left) return stbi__errpuc("bad file","scanline overrun"); + + for(i=0;ichannel,dest)) + return 0; + } + left-=count; + } + break; + } + } + } + } + + return result; +} + +static void *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp, stbi__result_info *ri) +{ + stbi_uc *result; + int i, x,y, internal_comp; + STBI_NOTUSED(ri); + + if (!comp) comp = &internal_comp; + + for (i=0; i<92; ++i) + stbi__get8(s); + + x = stbi__get16be(s); + y = stbi__get16be(s); + + if (y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pic header)"); + if (!stbi__mad3sizes_valid(x, y, 4, 0)) return stbi__errpuc("too large", "PIC image too large to decode"); + + stbi__get32be(s); //skip `ratio' + stbi__get16be(s); //skip `fields' + stbi__get16be(s); //skip `pad' + + // intermediate buffer is RGBA + result = (stbi_uc *) stbi__malloc_mad3(x, y, 4, 0); + if (!result) return stbi__errpuc("outofmem", "Out of memory"); + memset(result, 0xff, x*y*4); + + if (!stbi__pic_load_core(s,x,y,comp, result)) { + STBI_FREE(result); + result=0; + } + *px = x; + *py = y; + if (req_comp == 0) req_comp = *comp; + result=stbi__convert_format(result,4,req_comp,x,y); + + return result; +} + +static int stbi__pic_test(stbi__context *s) +{ + int r = stbi__pic_test_core(s); + stbi__rewind(s); + return r; +} +#endif + +// ************************************************************************************************* +// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb + +#ifndef STBI_NO_GIF +typedef struct +{ + stbi__int16 prefix; + stbi_uc first; + stbi_uc suffix; +} stbi__gif_lzw; + +typedef struct +{ + int w,h; + stbi_uc *out; // output buffer (always 4 components) + stbi_uc *background; // The current "background" as far as a gif is concerned + stbi_uc *history; + int flags, bgindex, ratio, transparent, eflags; + stbi_uc pal[256][4]; + stbi_uc lpal[256][4]; + stbi__gif_lzw codes[8192]; + stbi_uc *color_table; + int parse, step; + int lflags; + int start_x, start_y; + int max_x, max_y; + int cur_x, cur_y; + int line_size; + int delay; +} stbi__gif; + +static int stbi__gif_test_raw(stbi__context *s) +{ + int sz; + if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0; + sz = stbi__get8(s); + if (sz != '9' && sz != '7') return 0; + if (stbi__get8(s) != 'a') return 0; + return 1; +} + +static int stbi__gif_test(stbi__context *s) +{ + int r = stbi__gif_test_raw(s); + stbi__rewind(s); + return r; +} + +static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp) +{ + int i; + for (i=0; i < num_entries; ++i) { + pal[i][2] = stbi__get8(s); + pal[i][1] = stbi__get8(s); + pal[i][0] = stbi__get8(s); + pal[i][3] = transp == i ? 0 : 255; + } +} + +static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info) +{ + stbi_uc version; + if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') + return stbi__err("not GIF", "Corrupt GIF"); + + version = stbi__get8(s); + if (version != '7' && version != '9') return stbi__err("not GIF", "Corrupt GIF"); + if (stbi__get8(s) != 'a') return stbi__err("not GIF", "Corrupt GIF"); + + stbi__g_failure_reason = ""; + g->w = stbi__get16le(s); + g->h = stbi__get16le(s); + g->flags = stbi__get8(s); + g->bgindex = stbi__get8(s); + g->ratio = stbi__get8(s); + g->transparent = -1; + + if (g->w > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + if (g->h > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + + if (comp != 0) *comp = 4; // can't actually tell whether it's 3 or 4 until we parse the comments + + if (is_info) return 1; + + if (g->flags & 0x80) + stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1); + + return 1; +} + +static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp) +{ + stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif)); + if (!g) return stbi__err("outofmem", "Out of memory"); + if (!stbi__gif_header(s, g, comp, 1)) { + STBI_FREE(g); + stbi__rewind( s ); + return 0; + } + if (x) *x = g->w; + if (y) *y = g->h; + STBI_FREE(g); + return 1; +} + +static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code) +{ + stbi_uc *p, *c; + int idx; + + // recurse to decode the prefixes, since the linked-list is backwards, + // and working backwards through an interleaved image would be nasty + if (g->codes[code].prefix >= 0) + stbi__out_gif_code(g, g->codes[code].prefix); + + if (g->cur_y >= g->max_y) return; + + idx = g->cur_x + g->cur_y; + p = &g->out[idx]; + g->history[idx / 4] = 1; + + c = &g->color_table[g->codes[code].suffix * 4]; + if (c[3] > 128) { // don't render transparent pixels; + p[0] = c[2]; + p[1] = c[1]; + p[2] = c[0]; + p[3] = c[3]; + } + g->cur_x += 4; + + if (g->cur_x >= g->max_x) { + g->cur_x = g->start_x; + g->cur_y += g->step; + + while (g->cur_y >= g->max_y && g->parse > 0) { + g->step = (1 << g->parse) * g->line_size; + g->cur_y = g->start_y + (g->step >> 1); + --g->parse; + } + } +} + +static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g) +{ + stbi_uc lzw_cs; + stbi__int32 len, init_code; + stbi__uint32 first; + stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear; + stbi__gif_lzw *p; + + lzw_cs = stbi__get8(s); + if (lzw_cs > 12) return NULL; + clear = 1 << lzw_cs; + first = 1; + codesize = lzw_cs + 1; + codemask = (1 << codesize) - 1; + bits = 0; + valid_bits = 0; + for (init_code = 0; init_code < clear; init_code++) { + g->codes[init_code].prefix = -1; + g->codes[init_code].first = (stbi_uc) init_code; + g->codes[init_code].suffix = (stbi_uc) init_code; + } + + // support no starting clear code + avail = clear+2; + oldcode = -1; + + len = 0; + for(;;) { + if (valid_bits < codesize) { + if (len == 0) { + len = stbi__get8(s); // start new block + if (len == 0) + return g->out; + } + --len; + bits |= (stbi__int32) stbi__get8(s) << valid_bits; + valid_bits += 8; + } else { + stbi__int32 code = bits & codemask; + bits >>= codesize; + valid_bits -= codesize; + // @OPTIMIZE: is there some way we can accelerate the non-clear path? + if (code == clear) { // clear code + codesize = lzw_cs + 1; + codemask = (1 << codesize) - 1; + avail = clear + 2; + oldcode = -1; + first = 0; + } else if (code == clear + 1) { // end of stream code + stbi__skip(s, len); + while ((len = stbi__get8(s)) > 0) + stbi__skip(s,len); + return g->out; + } else if (code <= avail) { + if (first) { + return stbi__errpuc("no clear code", "Corrupt GIF"); + } + + if (oldcode >= 0) { + p = &g->codes[avail++]; + if (avail > 8192) { + return stbi__errpuc("too many codes", "Corrupt GIF"); + } + + p->prefix = (stbi__int16) oldcode; + p->first = g->codes[oldcode].first; + p->suffix = (code == avail) ? p->first : g->codes[code].first; + } else if (code == avail) + return stbi__errpuc("illegal code in raster", "Corrupt GIF"); + + stbi__out_gif_code(g, (stbi__uint16) code); + + if ((avail & codemask) == 0 && avail <= 0x0FFF) { + codesize++; + codemask = (1 << codesize) - 1; + } + + oldcode = code; + } else { + return stbi__errpuc("illegal code in raster", "Corrupt GIF"); + } + } + } +} + +// this function is designed to support animated gifs, although stb_image doesn't support it +// two back is the image from two frames ago, used for a very specific disposal format +static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp, stbi_uc *two_back) +{ + int dispose; + int first_frame; + int pi; + int pcount; + STBI_NOTUSED(req_comp); + + // on first frame, any non-written pixels get the background colour (non-transparent) + first_frame = 0; + if (g->out == 0) { + if (!stbi__gif_header(s, g, comp,0)) return 0; // stbi__g_failure_reason set by stbi__gif_header + if (!stbi__mad3sizes_valid(4, g->w, g->h, 0)) + return stbi__errpuc("too large", "GIF image is too large"); + pcount = g->w * g->h; + g->out = (stbi_uc *) stbi__malloc(4 * pcount); + g->background = (stbi_uc *) stbi__malloc(4 * pcount); + g->history = (stbi_uc *) stbi__malloc(pcount); + if (!g->out || !g->background || !g->history) + return stbi__errpuc("outofmem", "Out of memory"); + + // image is treated as "transparent" at the start - ie, nothing overwrites the current background; + // background colour is only used for pixels that are not rendered first frame, after that "background" + // color refers to the color that was there the previous frame. + memset(g->out, 0x00, 4 * pcount); + memset(g->background, 0x00, 4 * pcount); // state of the background (starts transparent) + memset(g->history, 0x00, pcount); // pixels that were affected previous frame + first_frame = 1; + } else { + // second frame - how do we dispose of the previous one? + dispose = (g->eflags & 0x1C) >> 2; + pcount = g->w * g->h; + + if ((dispose == 3) && (two_back == 0)) { + dispose = 2; // if I don't have an image to revert back to, default to the old background + } + + if (dispose == 3) { // use previous graphic + for (pi = 0; pi < pcount; ++pi) { + if (g->history[pi]) { + memcpy( &g->out[pi * 4], &two_back[pi * 4], 4 ); + } + } + } else if (dispose == 2) { + // restore what was changed last frame to background before that frame; + for (pi = 0; pi < pcount; ++pi) { + if (g->history[pi]) { + memcpy( &g->out[pi * 4], &g->background[pi * 4], 4 ); + } + } + } else { + // This is a non-disposal case eithe way, so just + // leave the pixels as is, and they will become the new background + // 1: do not dispose + // 0: not specified. + } + + // background is what out is after the undoing of the previou frame; + memcpy( g->background, g->out, 4 * g->w * g->h ); + } + + // clear my history; + memset( g->history, 0x00, g->w * g->h ); // pixels that were affected previous frame + + for (;;) { + int tag = stbi__get8(s); + switch (tag) { + case 0x2C: /* Image Descriptor */ + { + stbi__int32 x, y, w, h; + stbi_uc *o; + + x = stbi__get16le(s); + y = stbi__get16le(s); + w = stbi__get16le(s); + h = stbi__get16le(s); + if (((x + w) > (g->w)) || ((y + h) > (g->h))) + return stbi__errpuc("bad Image Descriptor", "Corrupt GIF"); + + g->line_size = g->w * 4; + g->start_x = x * 4; + g->start_y = y * g->line_size; + g->max_x = g->start_x + w * 4; + g->max_y = g->start_y + h * g->line_size; + g->cur_x = g->start_x; + g->cur_y = g->start_y; + + // if the width of the specified rectangle is 0, that means + // we may not see *any* pixels or the image is malformed; + // to make sure this is caught, move the current y down to + // max_y (which is what out_gif_code checks). + if (w == 0) + g->cur_y = g->max_y; + + g->lflags = stbi__get8(s); + + if (g->lflags & 0x40) { + g->step = 8 * g->line_size; // first interlaced spacing + g->parse = 3; + } else { + g->step = g->line_size; + g->parse = 0; + } + + if (g->lflags & 0x80) { + stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1); + g->color_table = (stbi_uc *) g->lpal; + } else if (g->flags & 0x80) { + g->color_table = (stbi_uc *) g->pal; + } else + return stbi__errpuc("missing color table", "Corrupt GIF"); + + o = stbi__process_gif_raster(s, g); + if (!o) return NULL; + + // if this was the first frame, + pcount = g->w * g->h; + if (first_frame && (g->bgindex > 0)) { + // if first frame, any pixel not drawn to gets the background color + for (pi = 0; pi < pcount; ++pi) { + if (g->history[pi] == 0) { + g->pal[g->bgindex][3] = 255; // just in case it was made transparent, undo that; It will be reset next frame if need be; + memcpy( &g->out[pi * 4], &g->pal[g->bgindex], 4 ); + } + } + } + + return o; + } + + case 0x21: // Comment Extension. + { + int len; + int ext = stbi__get8(s); + if (ext == 0xF9) { // Graphic Control Extension. + len = stbi__get8(s); + if (len == 4) { + g->eflags = stbi__get8(s); + g->delay = 10 * stbi__get16le(s); // delay - 1/100th of a second, saving as 1/1000ths. + + // unset old transparent + if (g->transparent >= 0) { + g->pal[g->transparent][3] = 255; + } + if (g->eflags & 0x01) { + g->transparent = stbi__get8(s); + if (g->transparent >= 0) { + g->pal[g->transparent][3] = 0; + } + } else { + // don't need transparent + stbi__skip(s, 1); + g->transparent = -1; + } + } else { + stbi__skip(s, len); + break; + } + } + while ((len = stbi__get8(s)) != 0) { + stbi__skip(s, len); + } + break; + } + + case 0x3B: // gif stream termination code + return (stbi_uc *) s; // using '1' causes warning on some compilers + + default: + return stbi__errpuc("unknown code", "Corrupt GIF"); + } + } +} + +static void *stbi__load_gif_main_outofmem(stbi__gif *g, stbi_uc *out, int **delays) +{ + STBI_FREE(g->out); + STBI_FREE(g->history); + STBI_FREE(g->background); + + if (out) STBI_FREE(out); + if (delays && *delays) STBI_FREE(*delays); + return stbi__errpuc("outofmem", "Out of memory"); +} + +static void *stbi__load_gif_main(stbi__context *s, int **delays, int *x, int *y, int *z, int *comp, int req_comp) +{ + if (stbi__gif_test(s)) { + int layers = 0; + stbi_uc *u = 0; + stbi_uc *out = 0; + stbi_uc *two_back = 0; + stbi__gif g; + int stride; + int out_size = 0; + int delays_size = 0; + + STBI_NOTUSED(out_size); + STBI_NOTUSED(delays_size); + + memset(&g, 0, sizeof(g)); + if (delays) { + *delays = 0; + } + + do { + u = stbi__gif_load_next(s, &g, comp, req_comp, two_back); + if (u == (stbi_uc *) s) u = 0; // end of animated gif marker + + if (u) { + *x = g.w; + *y = g.h; + ++layers; + stride = g.w * g.h * 4; + + if (out) { + void *tmp = (stbi_uc*) STBI_REALLOC_SIZED( out, out_size, layers * stride ); + if (!tmp) + return stbi__load_gif_main_outofmem(&g, out, delays); + else { + out = (stbi_uc*) tmp; + out_size = layers * stride; + } + + if (delays) { + int *new_delays = (int*) STBI_REALLOC_SIZED( *delays, delays_size, sizeof(int) * layers ); + if (!new_delays) + return stbi__load_gif_main_outofmem(&g, out, delays); + *delays = new_delays; + delays_size = layers * sizeof(int); + } + } else { + out = (stbi_uc*)stbi__malloc( layers * stride ); + if (!out) + return stbi__load_gif_main_outofmem(&g, out, delays); + out_size = layers * stride; + if (delays) { + *delays = (int*) stbi__malloc( layers * sizeof(int) ); + if (!*delays) + return stbi__load_gif_main_outofmem(&g, out, delays); + delays_size = layers * sizeof(int); + } + } + memcpy( out + ((layers - 1) * stride), u, stride ); + if (layers >= 2) { + two_back = out - 2 * stride; + } + + if (delays) { + (*delays)[layers - 1U] = g.delay; + } + } + } while (u != 0); + + // free temp buffer; + STBI_FREE(g.out); + STBI_FREE(g.history); + STBI_FREE(g.background); + + // do the final conversion after loading everything; + if (req_comp && req_comp != 4) + out = stbi__convert_format(out, 4, req_comp, layers * g.w, g.h); + + *z = layers; + return out; + } else { + return stbi__errpuc("not GIF", "Image was not as a gif type."); + } +} + +static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi_uc *u = 0; + stbi__gif g; + memset(&g, 0, sizeof(g)); + STBI_NOTUSED(ri); + + u = stbi__gif_load_next(s, &g, comp, req_comp, 0); + if (u == (stbi_uc *) s) u = 0; // end of animated gif marker + if (u) { + *x = g.w; + *y = g.h; + + // moved conversion to after successful load so that the same + // can be done for multiple frames. + if (req_comp && req_comp != 4) + u = stbi__convert_format(u, 4, req_comp, g.w, g.h); + } else if (g.out) { + // if there was an error and we allocated an image buffer, free it! + STBI_FREE(g.out); + } + + // free buffers needed for multiple frame loading; + STBI_FREE(g.history); + STBI_FREE(g.background); + + return u; +} + +static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp) +{ + return stbi__gif_info_raw(s,x,y,comp); +} +#endif + +// ************************************************************************************************* +// Radiance RGBE HDR loader +// originally by Nicolas Schulz +#ifndef STBI_NO_HDR +static int stbi__hdr_test_core(stbi__context *s, const char *signature) +{ + int i; + for (i=0; signature[i]; ++i) + if (stbi__get8(s) != signature[i]) + return 0; + stbi__rewind(s); + return 1; +} + +static int stbi__hdr_test(stbi__context* s) +{ + int r = stbi__hdr_test_core(s, "#?RADIANCE\n"); + stbi__rewind(s); + if(!r) { + r = stbi__hdr_test_core(s, "#?RGBE\n"); + stbi__rewind(s); + } + return r; +} + +#define STBI__HDR_BUFLEN 1024 +static char *stbi__hdr_gettoken(stbi__context *z, char *buffer) +{ + int len=0; + char c = '\0'; + + c = (char) stbi__get8(z); + + while (!stbi__at_eof(z) && c != '\n') { + buffer[len++] = c; + if (len == STBI__HDR_BUFLEN-1) { + // flush to end of line + while (!stbi__at_eof(z) && stbi__get8(z) != '\n') + ; + break; + } + c = (char) stbi__get8(z); + } + + buffer[len] = 0; + return buffer; +} + +static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp) +{ + if ( input[3] != 0 ) { + float f1; + // Exponent + f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8)); + if (req_comp <= 2) + output[0] = (input[0] + input[1] + input[2]) * f1 / 3; + else { + output[0] = input[0] * f1; + output[1] = input[1] * f1; + output[2] = input[2] * f1; + } + if (req_comp == 2) output[1] = 1; + if (req_comp == 4) output[3] = 1; + } else { + switch (req_comp) { + case 4: output[3] = 1; /* fallthrough */ + case 3: output[0] = output[1] = output[2] = 0; + break; + case 2: output[1] = 1; /* fallthrough */ + case 1: output[0] = 0; + break; + } + } +} + +static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + char buffer[STBI__HDR_BUFLEN]; + char *token; + int valid = 0; + int width, height; + stbi_uc *scanline; + float *hdr_data; + int len; + unsigned char count, value; + int i, j, k, c1,c2, z; + const char *headerToken; + STBI_NOTUSED(ri); + + // Check identifier + headerToken = stbi__hdr_gettoken(s,buffer); + if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0) + return stbi__errpf("not HDR", "Corrupt HDR image"); + + // Parse header + for(;;) { + token = stbi__hdr_gettoken(s,buffer); + if (token[0] == 0) break; + if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1; + } + + if (!valid) return stbi__errpf("unsupported format", "Unsupported HDR format"); + + // Parse width and height + // can't use sscanf() if we're not using stdio! + token = stbi__hdr_gettoken(s,buffer); + if (strncmp(token, "-Y ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format"); + token += 3; + height = (int) strtol(token, &token, 10); + while (*token == ' ') ++token; + if (strncmp(token, "+X ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format"); + token += 3; + width = (int) strtol(token, NULL, 10); + + if (height > STBI_MAX_DIMENSIONS) return stbi__errpf("too large","Very large image (corrupt?)"); + if (width > STBI_MAX_DIMENSIONS) return stbi__errpf("too large","Very large image (corrupt?)"); + + *x = width; + *y = height; + + if (comp) *comp = 3; + if (req_comp == 0) req_comp = 3; + + if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0)) + return stbi__errpf("too large", "HDR image is too large"); + + // Read data + hdr_data = (float *) stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0); + if (!hdr_data) + return stbi__errpf("outofmem", "Out of memory"); + + // Load image data + // image data is stored as some number of sca + if ( width < 8 || width >= 32768) { + // Read flat data + for (j=0; j < height; ++j) { + for (i=0; i < width; ++i) { + stbi_uc rgbe[4]; + main_decode_loop: + stbi__getn(s, rgbe, 4); + stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp); + } + } + } else { + // Read RLE-encoded data + scanline = NULL; + + for (j = 0; j < height; ++j) { + c1 = stbi__get8(s); + c2 = stbi__get8(s); + len = stbi__get8(s); + if (c1 != 2 || c2 != 2 || (len & 0x80)) { + // not run-length encoded, so we have to actually use THIS data as a decoded + // pixel (note this can't be a valid pixel--one of RGB must be >= 128) + stbi_uc rgbe[4]; + rgbe[0] = (stbi_uc) c1; + rgbe[1] = (stbi_uc) c2; + rgbe[2] = (stbi_uc) len; + rgbe[3] = (stbi_uc) stbi__get8(s); + stbi__hdr_convert(hdr_data, rgbe, req_comp); + i = 1; + j = 0; + STBI_FREE(scanline); + goto main_decode_loop; // yes, this makes no sense + } + len <<= 8; + len |= stbi__get8(s); + if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); } + if (scanline == NULL) { + scanline = (stbi_uc *) stbi__malloc_mad2(width, 4, 0); + if (!scanline) { + STBI_FREE(hdr_data); + return stbi__errpf("outofmem", "Out of memory"); + } + } + + for (k = 0; k < 4; ++k) { + int nleft; + i = 0; + while ((nleft = width - i) > 0) { + count = stbi__get8(s); + if (count > 128) { + // Run + value = stbi__get8(s); + count -= 128; + if ((count == 0) || (count > nleft)) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); } + for (z = 0; z < count; ++z) + scanline[i++ * 4 + k] = value; + } else { + // Dump + if ((count == 0) || (count > nleft)) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); } + for (z = 0; z < count; ++z) + scanline[i++ * 4 + k] = stbi__get8(s); + } + } + } + for (i=0; i < width; ++i) + stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp); + } + if (scanline) + STBI_FREE(scanline); + } + + return hdr_data; +} + +static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp) +{ + char buffer[STBI__HDR_BUFLEN]; + char *token; + int valid = 0; + int dummy; + + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + + if (stbi__hdr_test(s) == 0) { + stbi__rewind( s ); + return 0; + } + + for(;;) { + token = stbi__hdr_gettoken(s,buffer); + if (token[0] == 0) break; + if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1; + } + + if (!valid) { + stbi__rewind( s ); + return 0; + } + token = stbi__hdr_gettoken(s,buffer); + if (strncmp(token, "-Y ", 3)) { + stbi__rewind( s ); + return 0; + } + token += 3; + *y = (int) strtol(token, &token, 10); + while (*token == ' ') ++token; + if (strncmp(token, "+X ", 3)) { + stbi__rewind( s ); + return 0; + } + token += 3; + *x = (int) strtol(token, NULL, 10); + *comp = 3; + return 1; +} +#endif // STBI_NO_HDR + +#ifndef STBI_NO_BMP +static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp) +{ + void *p; + stbi__bmp_data info; + + info.all_a = 255; + p = stbi__bmp_parse_header(s, &info); + if (p == NULL) { + stbi__rewind( s ); + return 0; + } + if (x) *x = s->img_x; + if (y) *y = s->img_y; + if (comp) { + if (info.bpp == 24 && info.ma == 0xff000000) + *comp = 3; + else + *comp = info.ma ? 4 : 3; + } + return 1; +} +#endif + +#ifndef STBI_NO_PSD +static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp) +{ + int channelCount, dummy, depth; + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + if (stbi__get32be(s) != 0x38425053) { + stbi__rewind( s ); + return 0; + } + if (stbi__get16be(s) != 1) { + stbi__rewind( s ); + return 0; + } + stbi__skip(s, 6); + channelCount = stbi__get16be(s); + if (channelCount < 0 || channelCount > 16) { + stbi__rewind( s ); + return 0; + } + *y = stbi__get32be(s); + *x = stbi__get32be(s); + depth = stbi__get16be(s); + if (depth != 8 && depth != 16) { + stbi__rewind( s ); + return 0; + } + if (stbi__get16be(s) != 3) { + stbi__rewind( s ); + return 0; + } + *comp = 4; + return 1; +} + +static int stbi__psd_is16(stbi__context *s) +{ + int channelCount, depth; + if (stbi__get32be(s) != 0x38425053) { + stbi__rewind( s ); + return 0; + } + if (stbi__get16be(s) != 1) { + stbi__rewind( s ); + return 0; + } + stbi__skip(s, 6); + channelCount = stbi__get16be(s); + if (channelCount < 0 || channelCount > 16) { + stbi__rewind( s ); + return 0; + } + STBI_NOTUSED(stbi__get32be(s)); + STBI_NOTUSED(stbi__get32be(s)); + depth = stbi__get16be(s); + if (depth != 16) { + stbi__rewind( s ); + return 0; + } + return 1; +} +#endif + +#ifndef STBI_NO_PIC +static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp) +{ + int act_comp=0,num_packets=0,chained,dummy; + stbi__pic_packet packets[10]; + + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + + if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) { + stbi__rewind(s); + return 0; + } + + stbi__skip(s, 88); + + *x = stbi__get16be(s); + *y = stbi__get16be(s); + if (stbi__at_eof(s)) { + stbi__rewind( s); + return 0; + } + if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) { + stbi__rewind( s ); + return 0; + } + + stbi__skip(s, 8); + + do { + stbi__pic_packet *packet; + + if (num_packets==sizeof(packets)/sizeof(packets[0])) + return 0; + + packet = &packets[num_packets++]; + chained = stbi__get8(s); + packet->size = stbi__get8(s); + packet->type = stbi__get8(s); + packet->channel = stbi__get8(s); + act_comp |= packet->channel; + + if (stbi__at_eof(s)) { + stbi__rewind( s ); + return 0; + } + if (packet->size != 8) { + stbi__rewind( s ); + return 0; + } + } while (chained); + + *comp = (act_comp & 0x10 ? 4 : 3); + + return 1; +} +#endif + +// ************************************************************************************************* +// Portable Gray Map and Portable Pixel Map loader +// by Ken Miller +// +// PGM: http://netpbm.sourceforge.net/doc/pgm.html +// PPM: http://netpbm.sourceforge.net/doc/ppm.html +// +// Known limitations: +// Does not support comments in the header section +// Does not support ASCII image data (formats P2 and P3) + +#ifndef STBI_NO_PNM + +static int stbi__pnm_test(stbi__context *s) +{ + char p, t; + p = (char) stbi__get8(s); + t = (char) stbi__get8(s); + if (p != 'P' || (t != '5' && t != '6')) { + stbi__rewind( s ); + return 0; + } + return 1; +} + +static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi_uc *out; + STBI_NOTUSED(ri); + + ri->bits_per_channel = stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n); + if (ri->bits_per_channel == 0) + return 0; + + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + *x = s->img_x; + *y = s->img_y; + if (comp) *comp = s->img_n; + + if (!stbi__mad4sizes_valid(s->img_n, s->img_x, s->img_y, ri->bits_per_channel / 8, 0)) + return stbi__errpuc("too large", "PNM too large"); + + out = (stbi_uc *) stbi__malloc_mad4(s->img_n, s->img_x, s->img_y, ri->bits_per_channel / 8, 0); + if (!out) return stbi__errpuc("outofmem", "Out of memory"); + if (!stbi__getn(s, out, s->img_n * s->img_x * s->img_y * (ri->bits_per_channel / 8))) { + STBI_FREE(out); + return stbi__errpuc("bad PNM", "PNM file truncated"); + } + + if (req_comp && req_comp != s->img_n) { + if (ri->bits_per_channel == 16) { + out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, s->img_n, req_comp, s->img_x, s->img_y); + } else { + out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y); + } + if (out == NULL) return out; // stbi__convert_format frees input on failure + } + return out; +} + +static int stbi__pnm_isspace(char c) +{ + return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r'; +} + +static void stbi__pnm_skip_whitespace(stbi__context *s, char *c) +{ + for (;;) { + while (!stbi__at_eof(s) && stbi__pnm_isspace(*c)) + *c = (char) stbi__get8(s); + + if (stbi__at_eof(s) || *c != '#') + break; + + while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' ) + *c = (char) stbi__get8(s); + } +} + +static int stbi__pnm_isdigit(char c) +{ + return c >= '0' && c <= '9'; +} + +static int stbi__pnm_getinteger(stbi__context *s, char *c) +{ + int value = 0; + + while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) { + value = value*10 + (*c - '0'); + *c = (char) stbi__get8(s); + if((value > 214748364) || (value == 214748364 && *c > '7')) + return stbi__err("integer parse overflow", "Parsing an integer in the PPM header overflowed a 32-bit int"); + } + + return value; +} + +static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp) +{ + int maxv, dummy; + char c, p, t; + + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + + stbi__rewind(s); + + // Get identifier + p = (char) stbi__get8(s); + t = (char) stbi__get8(s); + if (p != 'P' || (t != '5' && t != '6')) { + stbi__rewind(s); + return 0; + } + + *comp = (t == '6') ? 3 : 1; // '5' is 1-component .pgm; '6' is 3-component .ppm + + c = (char) stbi__get8(s); + stbi__pnm_skip_whitespace(s, &c); + + *x = stbi__pnm_getinteger(s, &c); // read width + if(*x == 0) + return stbi__err("invalid width", "PPM image header had zero or overflowing width"); + stbi__pnm_skip_whitespace(s, &c); + + *y = stbi__pnm_getinteger(s, &c); // read height + if (*y == 0) + return stbi__err("invalid width", "PPM image header had zero or overflowing width"); + stbi__pnm_skip_whitespace(s, &c); + + maxv = stbi__pnm_getinteger(s, &c); // read max value + if (maxv > 65535) + return stbi__err("max value > 65535", "PPM image supports only 8-bit and 16-bit images"); + else if (maxv > 255) + return 16; + else + return 8; +} + +static int stbi__pnm_is16(stbi__context *s) +{ + if (stbi__pnm_info(s, NULL, NULL, NULL) == 16) + return 1; + return 0; +} +#endif + +static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp) +{ + #ifndef STBI_NO_JPEG + if (stbi__jpeg_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PNG + if (stbi__png_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_GIF + if (stbi__gif_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_BMP + if (stbi__bmp_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PSD + if (stbi__psd_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PIC + if (stbi__pic_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PNM + if (stbi__pnm_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_HDR + if (stbi__hdr_info(s, x, y, comp)) return 1; + #endif + + // test tga last because it's a crappy test! + #ifndef STBI_NO_TGA + if (stbi__tga_info(s, x, y, comp)) + return 1; + #endif + return stbi__err("unknown image type", "Image not of any known type, or corrupt"); +} + +static int stbi__is_16_main(stbi__context *s) +{ + #ifndef STBI_NO_PNG + if (stbi__png_is16(s)) return 1; + #endif + + #ifndef STBI_NO_PSD + if (stbi__psd_is16(s)) return 1; + #endif + + #ifndef STBI_NO_PNM + if (stbi__pnm_is16(s)) return 1; + #endif + return 0; +} + +#ifndef STBI_NO_STDIO +STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp) +{ + FILE *f = stbi__fopen(filename, "rb"); + int result; + if (!f) return stbi__err("can't fopen", "Unable to open file"); + result = stbi_info_from_file(f, x, y, comp); + fclose(f); + return result; +} + +STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp) +{ + int r; + stbi__context s; + long pos = ftell(f); + stbi__start_file(&s, f); + r = stbi__info_main(&s,x,y,comp); + fseek(f,pos,SEEK_SET); + return r; +} + +STBIDEF int stbi_is_16_bit(char const *filename) +{ + FILE *f = stbi__fopen(filename, "rb"); + int result; + if (!f) return stbi__err("can't fopen", "Unable to open file"); + result = stbi_is_16_bit_from_file(f); + fclose(f); + return result; +} + +STBIDEF int stbi_is_16_bit_from_file(FILE *f) +{ + int r; + stbi__context s; + long pos = ftell(f); + stbi__start_file(&s, f); + r = stbi__is_16_main(&s); + fseek(f,pos,SEEK_SET); + return r; +} +#endif // !STBI_NO_STDIO + +STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__info_main(&s,x,y,comp); +} + +STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user); + return stbi__info_main(&s,x,y,comp); +} + +STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const *buffer, int len) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__is_16_main(&s); +} + +STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const *c, void *user) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user); + return stbi__is_16_main(&s); +} + +#endif // STB_IMAGE_IMPLEMENTATION + +/* + revision history: + 2.20 (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs + 2.19 (2018-02-11) fix warning + 2.18 (2018-01-30) fix warnings + 2.17 (2018-01-29) change sbti__shiftsigned to avoid clang -O2 bug + 1-bit BMP + *_is_16_bit api + avoid warnings + 2.16 (2017-07-23) all functions have 16-bit variants; + STBI_NO_STDIO works again; + compilation fixes; + fix rounding in unpremultiply; + optimize vertical flip; + disable raw_len validation; + documentation fixes + 2.15 (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode; + warning fixes; disable run-time SSE detection on gcc; + uniform handling of optional "return" values; + thread-safe initialization of zlib tables + 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs + 2.13 (2016-11-29) add 16-bit API, only supported for PNG right now + 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes + 2.11 (2016-04-02) allocate large structures on the stack + remove white matting for transparent PSD + fix reported channel count for PNG & BMP + re-enable SSE2 in non-gcc 64-bit + support RGB-formatted JPEG + read 16-bit PNGs (only as 8-bit) + 2.10 (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED + 2.09 (2016-01-16) allow comments in PNM files + 16-bit-per-pixel TGA (not bit-per-component) + info() for TGA could break due to .hdr handling + info() for BMP to shares code instead of sloppy parse + can use STBI_REALLOC_SIZED if allocator doesn't support realloc + code cleanup + 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA + 2.07 (2015-09-13) fix compiler warnings + partial animated GIF support + limited 16-bpc PSD support + #ifdef unused functions + bug with < 92 byte PIC,PNM,HDR,TGA + 2.06 (2015-04-19) fix bug where PSD returns wrong '*comp' value + 2.05 (2015-04-19) fix bug in progressive JPEG handling, fix warning + 2.04 (2015-04-15) try to re-enable SIMD on MinGW 64-bit + 2.03 (2015-04-12) extra corruption checking (mmozeiko) + stbi_set_flip_vertically_on_load (nguillemot) + fix NEON support; fix mingw support + 2.02 (2015-01-19) fix incorrect assert, fix warning + 2.01 (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2 + 2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG + 2.00 (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg) + progressive JPEG (stb) + PGM/PPM support (Ken Miller) + STBI_MALLOC,STBI_REALLOC,STBI_FREE + GIF bugfix -- seemingly never worked + STBI_NO_*, STBI_ONLY_* + 1.48 (2014-12-14) fix incorrectly-named assert() + 1.47 (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb) + optimize PNG (ryg) + fix bug in interlaced PNG with user-specified channel count (stb) + 1.46 (2014-08-26) + fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG + 1.45 (2014-08-16) + fix MSVC-ARM internal compiler error by wrapping malloc + 1.44 (2014-08-07) + various warning fixes from Ronny Chevalier + 1.43 (2014-07-15) + fix MSVC-only compiler problem in code changed in 1.42 + 1.42 (2014-07-09) + don't define _CRT_SECURE_NO_WARNINGS (affects user code) + fixes to stbi__cleanup_jpeg path + added STBI_ASSERT to avoid requiring assert.h + 1.41 (2014-06-25) + fix search&replace from 1.36 that messed up comments/error messages + 1.40 (2014-06-22) + fix gcc struct-initialization warning + 1.39 (2014-06-15) + fix to TGA optimization when req_comp != number of components in TGA; + fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite) + add support for BMP version 5 (more ignored fields) + 1.38 (2014-06-06) + suppress MSVC warnings on integer casts truncating values + fix accidental rename of 'skip' field of I/O + 1.37 (2014-06-04) + remove duplicate typedef + 1.36 (2014-06-03) + convert to header file single-file library + if de-iphone isn't set, load iphone images color-swapped instead of returning NULL + 1.35 (2014-05-27) + various warnings + fix broken STBI_SIMD path + fix bug where stbi_load_from_file no longer left file pointer in correct place + fix broken non-easy path for 32-bit BMP (possibly never used) + TGA optimization by Arseny Kapoulkine + 1.34 (unknown) + use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case + 1.33 (2011-07-14) + make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements + 1.32 (2011-07-13) + support for "info" function for all supported filetypes (SpartanJ) + 1.31 (2011-06-20) + a few more leak fixes, bug in PNG handling (SpartanJ) + 1.30 (2011-06-11) + added ability to load files via callbacks to accomidate custom input streams (Ben Wenger) + removed deprecated format-specific test/load functions + removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway + error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha) + fix inefficiency in decoding 32-bit BMP (David Woo) + 1.29 (2010-08-16) + various warning fixes from Aurelien Pocheville + 1.28 (2010-08-01) + fix bug in GIF palette transparency (SpartanJ) + 1.27 (2010-08-01) + cast-to-stbi_uc to fix warnings + 1.26 (2010-07-24) + fix bug in file buffering for PNG reported by SpartanJ + 1.25 (2010-07-17) + refix trans_data warning (Won Chun) + 1.24 (2010-07-12) + perf improvements reading from files on platforms with lock-heavy fgetc() + minor perf improvements for jpeg + deprecated type-specific functions so we'll get feedback if they're needed + attempt to fix trans_data warning (Won Chun) + 1.23 fixed bug in iPhone support + 1.22 (2010-07-10) + removed image *writing* support + stbi_info support from Jetro Lauha + GIF support from Jean-Marc Lienher + iPhone PNG-extensions from James Brown + warning-fixes from Nicolas Schulz and Janez Zemva (i.stbi__err. Janez (U+017D)emva) + 1.21 fix use of 'stbi_uc' in header (reported by jon blow) + 1.20 added support for Softimage PIC, by Tom Seddon + 1.19 bug in interlaced PNG corruption check (found by ryg) + 1.18 (2008-08-02) + fix a threading bug (local mutable static) + 1.17 support interlaced PNG + 1.16 major bugfix - stbi__convert_format converted one too many pixels + 1.15 initialize some fields for thread safety + 1.14 fix threadsafe conversion bug + header-file-only version (#define STBI_HEADER_FILE_ONLY before including) + 1.13 threadsafe + 1.12 const qualifiers in the API + 1.11 Support installable IDCT, colorspace conversion routines + 1.10 Fixes for 64-bit (don't use "unsigned long") + optimized upsampling by Fabian "ryg" Giesen + 1.09 Fix format-conversion for PSD code (bad global variables!) + 1.08 Thatcher Ulrich's PSD code integrated by Nicolas Schulz + 1.07 attempt to fix C++ warning/errors again + 1.06 attempt to fix C++ warning/errors again + 1.05 fix TGA loading to return correct *comp and use good luminance calc + 1.04 default float alpha is 1, not 255; use 'void *' for stbi_image_free + 1.03 bugfixes to STBI_NO_STDIO, STBI_NO_HDR + 1.02 support for (subset of) HDR files, float interface for preferred access to them + 1.01 fix bug: possible bug in handling right-side up bmps... not sure + fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all + 1.00 interface to zlib that skips zlib header + 0.99 correct handling of alpha in palette + 0.98 TGA loader by lonesock; dynamically add loaders (untested) + 0.97 jpeg errors on too large a file; also catch another malloc failure + 0.96 fix detection of invalid v value - particleman@mollyrocket forum + 0.95 during header scan, seek to markers in case of padding + 0.94 STBI_NO_STDIO to disable stdio usage; rename all #defines the same + 0.93 handle jpegtran output; verbose errors + 0.92 read 4,8,16,24,32-bit BMP files of several formats + 0.91 output 24-bit Windows 3.0 BMP files + 0.90 fix a few more warnings; bump version number to approach 1.0 + 0.61 bugfixes due to Marc LeBlanc, Christopher Lloyd + 0.60 fix compiling as c++ + 0.59 fix warnings: merge Dave Moore's -Wall fixes + 0.58 fix bug: zlib uncompressed mode len/nlen was wrong endian + 0.57 fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available + 0.56 fix bug: zlib uncompressed mode len vs. nlen + 0.55 fix bug: restart_interval not initialized to 0 + 0.54 allow NULL for 'int *comp' + 0.53 fix bug in png 3->4; speedup png decoding + 0.52 png handles req_comp=3,4 directly; minor cleanup; jpeg comments + 0.51 obey req_comp requests, 1-component jpegs return as 1-component, + on 'test' only check type, not whether we support this variant + 0.50 (2006-11-19) + first released version +*/ + + +/* +------------------------------------------------------------------------------ +This software is available under 2 licenses -- choose whichever you prefer. +------------------------------------------------------------------------------ +ALTERNATIVE A - MIT License +Copyright (c) 2017 Sean Barrett +Permission is hereby granted, free of charge, to any person obtaining a copy of +this software and associated documentation files (the "Software"), to deal in +the Software without restriction, including without limitation the rights to +use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies +of the Software, and to permit persons to whom the Software is furnished to do +so, subject to the following conditions: +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +------------------------------------------------------------------------------ +ALTERNATIVE B - Public Domain (www.unlicense.org) +This is free and unencumbered software released into the public domain. +Anyone is free to copy, modify, publish, use, compile, sell, or distribute this +software, either in source code form or as a compiled binary, for any purpose, +commercial or non-commercial, and by any means. +In jurisdictions that recognize copyright laws, the author or authors of this +software dedicate any and all copyright interest in the software to the public +domain. We make this dedication for the benefit of the public at large and to +the detriment of our heirs and successors. We intend this dedication to be an +overt act of relinquishment in perpetuity of all present and future rights to +this software under copyright law. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN +ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. +------------------------------------------------------------------------------ +*/ diff --git a/Source/stb_image_resize.cpp b/Source/stb_image_resize.cpp new file mode 100644 index 0000000..c5371c2 --- /dev/null +++ b/Source/stb_image_resize.cpp @@ -0,0 +1,2 @@ +#define STB_IMAGE_RESIZE_IMPLEMENTATION +#include "stb_image_resize.h" \ No newline at end of file diff --git a/Source/stb_image_resize.h b/Source/stb_image_resize.h new file mode 100644 index 0000000..ef9e6fe --- /dev/null +++ b/Source/stb_image_resize.h @@ -0,0 +1,2634 @@ +/* stb_image_resize - v0.97 - public domain image resizing + by Jorge L Rodriguez (@VinoBS) - 2014 + http://github.com/nothings/stb + + Written with emphasis on usability, portability, and efficiency. (No + SIMD or threads, so it be easily outperformed by libs that use those.) + Only scaling and translation is supported, no rotations or shears. + Easy API downsamples w/Mitchell filter, upsamples w/cubic interpolation. + + COMPILING & LINKING + In one C/C++ file that #includes this file, do this: + #define STB_IMAGE_RESIZE_IMPLEMENTATION + before the #include. That will create the implementation in that file. + + QUICKSTART + stbir_resize_uint8( input_pixels , in_w , in_h , 0, + output_pixels, out_w, out_h, 0, num_channels) + stbir_resize_float(...) + stbir_resize_uint8_srgb( input_pixels , in_w , in_h , 0, + output_pixels, out_w, out_h, 0, + num_channels , alpha_chan , 0) + stbir_resize_uint8_srgb_edgemode( + input_pixels , in_w , in_h , 0, + output_pixels, out_w, out_h, 0, + num_channels , alpha_chan , 0, STBIR_EDGE_CLAMP) + // WRAP/REFLECT/ZERO + + FULL API + See the "header file" section of the source for API documentation. + + ADDITIONAL DOCUMENTATION + + SRGB & FLOATING POINT REPRESENTATION + The sRGB functions presume IEEE floating point. If you do not have + IEEE floating point, define STBIR_NON_IEEE_FLOAT. This will use + a slower implementation. + + MEMORY ALLOCATION + The resize functions here perform a single memory allocation using + malloc. To control the memory allocation, before the #include that + triggers the implementation, do: + + #define STBIR_MALLOC(size,context) ... + #define STBIR_FREE(ptr,context) ... + + Each resize function makes exactly one call to malloc/free, so to use + temp memory, store the temp memory in the context and return that. + + ASSERT + Define STBIR_ASSERT(boolval) to override assert() and not use assert.h + + OPTIMIZATION + Define STBIR_SATURATE_INT to compute clamp values in-range using + integer operations instead of float operations. This may be faster + on some platforms. + + DEFAULT FILTERS + For functions which don't provide explicit control over what filters + to use, you can change the compile-time defaults with + + #define STBIR_DEFAULT_FILTER_UPSAMPLE STBIR_FILTER_something + #define STBIR_DEFAULT_FILTER_DOWNSAMPLE STBIR_FILTER_something + + See stbir_filter in the header-file section for the list of filters. + + NEW FILTERS + A number of 1D filter kernels are used. For a list of + supported filters see the stbir_filter enum. To add a new filter, + write a filter function and add it to stbir__filter_info_table. + + PROGRESS + For interactive use with slow resize operations, you can install + a progress-report callback: + + #define STBIR_PROGRESS_REPORT(val) some_func(val) + + The parameter val is a float which goes from 0 to 1 as progress is made. + + For example: + + static void my_progress_report(float progress); + #define STBIR_PROGRESS_REPORT(val) my_progress_report(val) + + #define STB_IMAGE_RESIZE_IMPLEMENTATION + #include "stb_image_resize.h" + + static void my_progress_report(float progress) + { + printf("Progress: %f%%\n", progress*100); + } + + MAX CHANNELS + If your image has more than 64 channels, define STBIR_MAX_CHANNELS + to the max you'll have. + + ALPHA CHANNEL + Most of the resizing functions provide the ability to control how + the alpha channel of an image is processed. The important things + to know about this: + + 1. The best mathematically-behaved version of alpha to use is + called "premultiplied alpha", in which the other color channels + have had the alpha value multiplied in. If you use premultiplied + alpha, linear filtering (such as image resampling done by this + library, or performed in texture units on GPUs) does the "right + thing". While premultiplied alpha is standard in the movie CGI + industry, it is still uncommon in the videogame/real-time world. + + If you linearly filter non-premultiplied alpha, strange effects + occur. (For example, the 50/50 average of 99% transparent bright green + and 1% transparent black produces 50% transparent dark green when + non-premultiplied, whereas premultiplied it produces 50% + transparent near-black. The former introduces green energy + that doesn't exist in the source image.) + + 2. Artists should not edit premultiplied-alpha images; artists + want non-premultiplied alpha images. Thus, art tools generally output + non-premultiplied alpha images. + + 3. You will get best results in most cases by converting images + to premultiplied alpha before processing them mathematically. + + 4. If you pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED, the + resizer does not do anything special for the alpha channel; + it is resampled identically to other channels. This produces + the correct results for premultiplied-alpha images, but produces + less-than-ideal results for non-premultiplied-alpha images. + + 5. If you do not pass the flag STBIR_FLAG_ALPHA_PREMULTIPLIED, + then the resizer weights the contribution of input pixels + based on their alpha values, or, equivalently, it multiplies + the alpha value into the color channels, resamples, then divides + by the resultant alpha value. Input pixels which have alpha=0 do + not contribute at all to output pixels unless _all_ of the input + pixels affecting that output pixel have alpha=0, in which case + the result for that pixel is the same as it would be without + STBIR_FLAG_ALPHA_PREMULTIPLIED. However, this is only true for + input images in integer formats. For input images in float format, + input pixels with alpha=0 have no effect, and output pixels + which have alpha=0 will be 0 in all channels. (For float images, + you can manually achieve the same result by adding a tiny epsilon + value to the alpha channel of every image, and then subtracting + or clamping it at the end.) + + 6. You can suppress the behavior described in #5 and make + all-0-alpha pixels have 0 in all channels by #defining + STBIR_NO_ALPHA_EPSILON. + + 7. You can separately control whether the alpha channel is + interpreted as linear or affected by the colorspace. By default + it is linear; you almost never want to apply the colorspace. + (For example, graphics hardware does not apply sRGB conversion + to the alpha channel.) + + CONTRIBUTORS + Jorge L Rodriguez: Implementation + Sean Barrett: API design, optimizations + Aras Pranckevicius: bugfix + Nathan Reed: warning fixes + + REVISIONS + 0.97 (2020-02-02) fixed warning + 0.96 (2019-03-04) fixed warnings + 0.95 (2017-07-23) fixed warnings + 0.94 (2017-03-18) fixed warnings + 0.93 (2017-03-03) fixed bug with certain combinations of heights + 0.92 (2017-01-02) fix integer overflow on large (>2GB) images + 0.91 (2016-04-02) fix warnings; fix handling of subpixel regions + 0.90 (2014-09-17) first released version + + LICENSE + See end of file for license information. + + TODO + Don't decode all of the image data when only processing a partial tile + Don't use full-width decode buffers when only processing a partial tile + When processing wide images, break processing into tiles so data fits in L1 cache + Installable filters? + Resize that respects alpha test coverage + (Reference code: FloatImage::alphaTestCoverage and FloatImage::scaleAlphaToCoverage: + https://code.google.com/p/nvidia-texture-tools/source/browse/trunk/src/nvimage/FloatImage.cpp ) +*/ + +#ifndef STBIR_INCLUDE_STB_IMAGE_RESIZE_H +#define STBIR_INCLUDE_STB_IMAGE_RESIZE_H + +#ifdef _MSC_VER +typedef unsigned char stbir_uint8; +typedef unsigned short stbir_uint16; +typedef unsigned int stbir_uint32; +#else +#include +typedef uint8_t stbir_uint8; +typedef uint16_t stbir_uint16; +typedef uint32_t stbir_uint32; +#endif + +#ifndef STBIRDEF +#ifdef STB_IMAGE_RESIZE_STATIC +#define STBIRDEF static +#else +#ifdef __cplusplus +#define STBIRDEF extern "C" +#else +#define STBIRDEF extern +#endif +#endif +#endif + +////////////////////////////////////////////////////////////////////////////// +// +// Easy-to-use API: +// +// * "input pixels" points to an array of image data with 'num_channels' channels (e.g. RGB=3, RGBA=4) +// * input_w is input image width (x-axis), input_h is input image height (y-axis) +// * stride is the offset between successive rows of image data in memory, in bytes. you can +// specify 0 to mean packed continuously in memory +// * alpha channel is treated identically to other channels. +// * colorspace is linear or sRGB as specified by function name +// * returned result is 1 for success or 0 in case of an error. +// #define STBIR_ASSERT() to trigger an assert on parameter validation errors. +// * Memory required grows approximately linearly with input and output size, but with +// discontinuities at input_w == output_w and input_h == output_h. +// * These functions use a "default" resampling filter defined at compile time. To change the filter, +// you can change the compile-time defaults by #defining STBIR_DEFAULT_FILTER_UPSAMPLE +// and STBIR_DEFAULT_FILTER_DOWNSAMPLE, or you can use the medium-complexity API. + +STBIRDEF int stbir_resize_uint8( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + int num_channels); + +STBIRDEF int stbir_resize_float( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + float *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + int num_channels); + + +// The following functions interpret image data as gamma-corrected sRGB. +// Specify STBIR_ALPHA_CHANNEL_NONE if you have no alpha channel, +// or otherwise provide the index of the alpha channel. Flags value +// of 0 will probably do the right thing if you're not sure what +// the flags mean. + +#define STBIR_ALPHA_CHANNEL_NONE -1 + +// Set this flag if your texture has premultiplied alpha. Otherwise, stbir will +// use alpha-weighted resampling (effectively premultiplying, resampling, +// then unpremultiplying). +#define STBIR_FLAG_ALPHA_PREMULTIPLIED (1 << 0) +// The specified alpha channel should be handled as gamma-corrected value even +// when doing sRGB operations. +#define STBIR_FLAG_ALPHA_USES_COLORSPACE (1 << 1) + +STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + int num_channels, int alpha_channel, int flags); + + +typedef enum +{ + STBIR_EDGE_CLAMP = 1, + STBIR_EDGE_REFLECT = 2, + STBIR_EDGE_WRAP = 3, + STBIR_EDGE_ZERO = 4, +} stbir_edge; + +// This function adds the ability to specify how requests to sample off the edge of the image are handled. +STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_wrap_mode); + +////////////////////////////////////////////////////////////////////////////// +// +// Medium-complexity API +// +// This extends the easy-to-use API as follows: +// +// * Alpha-channel can be processed separately +// * If alpha_channel is not STBIR_ALPHA_CHANNEL_NONE +// * Alpha channel will not be gamma corrected (unless flags&STBIR_FLAG_GAMMA_CORRECT) +// * Filters will be weighted by alpha channel (unless flags&STBIR_FLAG_ALPHA_PREMULTIPLIED) +// * Filter can be selected explicitly +// * uint16 image type +// * sRGB colorspace available for all types +// * context parameter for passing to STBIR_MALLOC + +typedef enum +{ + STBIR_FILTER_DEFAULT = 0, // use same filter type that easy-to-use API chooses + STBIR_FILTER_BOX = 1, // A trapezoid w/1-pixel wide ramps, same result as box for integer scale ratios + STBIR_FILTER_TRIANGLE = 2, // On upsampling, produces same results as bilinear texture filtering + STBIR_FILTER_CUBICBSPLINE = 3, // The cubic b-spline (aka Mitchell-Netrevalli with B=1,C=0), gaussian-esque + STBIR_FILTER_CATMULLROM = 4, // An interpolating cubic spline + STBIR_FILTER_MITCHELL = 5, // Mitchell-Netrevalli filter with B=1/3, C=1/3 +} stbir_filter; + +typedef enum +{ + STBIR_COLORSPACE_LINEAR, + STBIR_COLORSPACE_SRGB, + + STBIR_MAX_COLORSPACES, +} stbir_colorspace; + +// The following functions are all identical except for the type of the image data + +STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space, + void *alloc_context); + +STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space, + void *alloc_context); + +STBIRDEF int stbir_resize_float_generic( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + float *output_pixels , int output_w, int output_h, int output_stride_in_bytes, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space, + void *alloc_context); + + + +////////////////////////////////////////////////////////////////////////////// +// +// Full-complexity API +// +// This extends the medium API as follows: +// +// * uint32 image type +// * not typesafe +// * separate filter types for each axis +// * separate edge modes for each axis +// * can specify scale explicitly for subpixel correctness +// * can specify image source tile using texture coordinates + +typedef enum +{ + STBIR_TYPE_UINT8 , + STBIR_TYPE_UINT16, + STBIR_TYPE_UINT32, + STBIR_TYPE_FLOAT , + + STBIR_MAX_TYPES +} stbir_datatype; + +STBIRDEF int stbir_resize( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + void *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + stbir_datatype datatype, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical, + stbir_filter filter_horizontal, stbir_filter filter_vertical, + stbir_colorspace space, void *alloc_context); + +STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + void *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + stbir_datatype datatype, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical, + stbir_filter filter_horizontal, stbir_filter filter_vertical, + stbir_colorspace space, void *alloc_context, + float x_scale, float y_scale, + float x_offset, float y_offset); + +STBIRDEF int stbir_resize_region( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + void *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + stbir_datatype datatype, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical, + stbir_filter filter_horizontal, stbir_filter filter_vertical, + stbir_colorspace space, void *alloc_context, + float s0, float t0, float s1, float t1); +// (s0, t0) & (s1, t1) are the top-left and bottom right corner (uv addressing style: [0, 1]x[0, 1]) of a region of the input image to use. + +// +// +//// end header file ///////////////////////////////////////////////////// +#endif // STBIR_INCLUDE_STB_IMAGE_RESIZE_H + + + + + +#ifdef STB_IMAGE_RESIZE_IMPLEMENTATION + +#ifndef STBIR_ASSERT +#include +#define STBIR_ASSERT(x) assert(x) +#endif + +// For memset +#include + +#include + +#ifndef STBIR_MALLOC +#include +// use comma operator to evaluate c, to avoid "unused parameter" warnings +#define STBIR_MALLOC(size,c) ((void)(c), malloc(size)) +#define STBIR_FREE(ptr,c) ((void)(c), free(ptr)) +#endif + +#ifndef _MSC_VER +#ifdef __cplusplus +#define stbir__inline inline +#else +#define stbir__inline +#endif +#else +#define stbir__inline __forceinline +#endif + + +// should produce compiler error if size is wrong +typedef unsigned char stbir__validate_uint32[sizeof(stbir_uint32) == 4 ? 1 : -1]; + +#ifdef _MSC_VER +#define STBIR__NOTUSED(v) (void)(v) +#else +#define STBIR__NOTUSED(v) (void)sizeof(v) +#endif + +#define STBIR__ARRAY_SIZE(a) (sizeof((a))/sizeof((a)[0])) + +#ifndef STBIR_DEFAULT_FILTER_UPSAMPLE +#define STBIR_DEFAULT_FILTER_UPSAMPLE STBIR_FILTER_CATMULLROM +#endif + +#ifndef STBIR_DEFAULT_FILTER_DOWNSAMPLE +#define STBIR_DEFAULT_FILTER_DOWNSAMPLE STBIR_FILTER_MITCHELL +#endif + +#ifndef STBIR_PROGRESS_REPORT +#define STBIR_PROGRESS_REPORT(float_0_to_1) +#endif + +#ifndef STBIR_MAX_CHANNELS +#define STBIR_MAX_CHANNELS 64 +#endif + +#if STBIR_MAX_CHANNELS > 65536 +#error "Too many channels; STBIR_MAX_CHANNELS must be no more than 65536." +// because we store the indices in 16-bit variables +#endif + +// This value is added to alpha just before premultiplication to avoid +// zeroing out color values. It is equivalent to 2^-80. If you don't want +// that behavior (it may interfere if you have floating point images with +// very small alpha values) then you can define STBIR_NO_ALPHA_EPSILON to +// disable it. +#ifndef STBIR_ALPHA_EPSILON +#define STBIR_ALPHA_EPSILON ((float)1 / (1 << 20) / (1 << 20) / (1 << 20) / (1 << 20)) +#endif + + + +#ifdef _MSC_VER +#define STBIR__UNUSED_PARAM(v) (void)(v) +#else +#define STBIR__UNUSED_PARAM(v) (void)sizeof(v) +#endif + +// must match stbir_datatype +static unsigned char stbir__type_size[] = { + 1, // STBIR_TYPE_UINT8 + 2, // STBIR_TYPE_UINT16 + 4, // STBIR_TYPE_UINT32 + 4, // STBIR_TYPE_FLOAT +}; + +// Kernel function centered at 0 +typedef float (stbir__kernel_fn)(float x, float scale); +typedef float (stbir__support_fn)(float scale); + +typedef struct +{ + stbir__kernel_fn* kernel; + stbir__support_fn* support; +} stbir__filter_info; + +// When upsampling, the contributors are which source pixels contribute. +// When downsampling, the contributors are which destination pixels are contributed to. +typedef struct +{ + int n0; // First contributing pixel + int n1; // Last contributing pixel +} stbir__contributors; + +typedef struct +{ + const void* input_data; + int input_w; + int input_h; + int input_stride_bytes; + + void* output_data; + int output_w; + int output_h; + int output_stride_bytes; + + float s0, t0, s1, t1; + + float horizontal_shift; // Units: output pixels + float vertical_shift; // Units: output pixels + float horizontal_scale; + float vertical_scale; + + int channels; + int alpha_channel; + stbir_uint32 flags; + stbir_datatype type; + stbir_filter horizontal_filter; + stbir_filter vertical_filter; + stbir_edge edge_horizontal; + stbir_edge edge_vertical; + stbir_colorspace colorspace; + + stbir__contributors* horizontal_contributors; + float* horizontal_coefficients; + + stbir__contributors* vertical_contributors; + float* vertical_coefficients; + + int decode_buffer_pixels; + float* decode_buffer; + + float* horizontal_buffer; + + // cache these because ceil/floor are inexplicably showing up in profile + int horizontal_coefficient_width; + int vertical_coefficient_width; + int horizontal_filter_pixel_width; + int vertical_filter_pixel_width; + int horizontal_filter_pixel_margin; + int vertical_filter_pixel_margin; + int horizontal_num_contributors; + int vertical_num_contributors; + + int ring_buffer_length_bytes; // The length of an individual entry in the ring buffer. The total number of ring buffers is stbir__get_filter_pixel_width(filter) + int ring_buffer_num_entries; // Total number of entries in the ring buffer. + int ring_buffer_first_scanline; + int ring_buffer_last_scanline; + int ring_buffer_begin_index; // first_scanline is at this index in the ring buffer + float* ring_buffer; + + float* encode_buffer; // A temporary buffer to store floats so we don't lose precision while we do multiply-adds. + + int horizontal_contributors_size; + int horizontal_coefficients_size; + int vertical_contributors_size; + int vertical_coefficients_size; + int decode_buffer_size; + int horizontal_buffer_size; + int ring_buffer_size; + int encode_buffer_size; +} stbir__info; + + +static const float stbir__max_uint8_as_float = 255.0f; +static const float stbir__max_uint16_as_float = 65535.0f; +static const double stbir__max_uint32_as_float = 4294967295.0; + + +static stbir__inline int stbir__min(int a, int b) +{ + return a < b ? a : b; +} + +static stbir__inline float stbir__saturate(float x) +{ + if (x < 0) + return 0; + + if (x > 1) + return 1; + + return x; +} + +#ifdef STBIR_SATURATE_INT +static stbir__inline stbir_uint8 stbir__saturate8(int x) +{ + if ((unsigned int) x <= 255) + return x; + + if (x < 0) + return 0; + + return 255; +} + +static stbir__inline stbir_uint16 stbir__saturate16(int x) +{ + if ((unsigned int) x <= 65535) + return x; + + if (x < 0) + return 0; + + return 65535; +} +#endif + +static float stbir__srgb_uchar_to_linear_float[256] = { + 0.000000f, 0.000304f, 0.000607f, 0.000911f, 0.001214f, 0.001518f, 0.001821f, 0.002125f, 0.002428f, 0.002732f, 0.003035f, + 0.003347f, 0.003677f, 0.004025f, 0.004391f, 0.004777f, 0.005182f, 0.005605f, 0.006049f, 0.006512f, 0.006995f, 0.007499f, + 0.008023f, 0.008568f, 0.009134f, 0.009721f, 0.010330f, 0.010960f, 0.011612f, 0.012286f, 0.012983f, 0.013702f, 0.014444f, + 0.015209f, 0.015996f, 0.016807f, 0.017642f, 0.018500f, 0.019382f, 0.020289f, 0.021219f, 0.022174f, 0.023153f, 0.024158f, + 0.025187f, 0.026241f, 0.027321f, 0.028426f, 0.029557f, 0.030713f, 0.031896f, 0.033105f, 0.034340f, 0.035601f, 0.036889f, + 0.038204f, 0.039546f, 0.040915f, 0.042311f, 0.043735f, 0.045186f, 0.046665f, 0.048172f, 0.049707f, 0.051269f, 0.052861f, + 0.054480f, 0.056128f, 0.057805f, 0.059511f, 0.061246f, 0.063010f, 0.064803f, 0.066626f, 0.068478f, 0.070360f, 0.072272f, + 0.074214f, 0.076185f, 0.078187f, 0.080220f, 0.082283f, 0.084376f, 0.086500f, 0.088656f, 0.090842f, 0.093059f, 0.095307f, + 0.097587f, 0.099899f, 0.102242f, 0.104616f, 0.107023f, 0.109462f, 0.111932f, 0.114435f, 0.116971f, 0.119538f, 0.122139f, + 0.124772f, 0.127438f, 0.130136f, 0.132868f, 0.135633f, 0.138432f, 0.141263f, 0.144128f, 0.147027f, 0.149960f, 0.152926f, + 0.155926f, 0.158961f, 0.162029f, 0.165132f, 0.168269f, 0.171441f, 0.174647f, 0.177888f, 0.181164f, 0.184475f, 0.187821f, + 0.191202f, 0.194618f, 0.198069f, 0.201556f, 0.205079f, 0.208637f, 0.212231f, 0.215861f, 0.219526f, 0.223228f, 0.226966f, + 0.230740f, 0.234551f, 0.238398f, 0.242281f, 0.246201f, 0.250158f, 0.254152f, 0.258183f, 0.262251f, 0.266356f, 0.270498f, + 0.274677f, 0.278894f, 0.283149f, 0.287441f, 0.291771f, 0.296138f, 0.300544f, 0.304987f, 0.309469f, 0.313989f, 0.318547f, + 0.323143f, 0.327778f, 0.332452f, 0.337164f, 0.341914f, 0.346704f, 0.351533f, 0.356400f, 0.361307f, 0.366253f, 0.371238f, + 0.376262f, 0.381326f, 0.386430f, 0.391573f, 0.396755f, 0.401978f, 0.407240f, 0.412543f, 0.417885f, 0.423268f, 0.428691f, + 0.434154f, 0.439657f, 0.445201f, 0.450786f, 0.456411f, 0.462077f, 0.467784f, 0.473532f, 0.479320f, 0.485150f, 0.491021f, + 0.496933f, 0.502887f, 0.508881f, 0.514918f, 0.520996f, 0.527115f, 0.533276f, 0.539480f, 0.545725f, 0.552011f, 0.558340f, + 0.564712f, 0.571125f, 0.577581f, 0.584078f, 0.590619f, 0.597202f, 0.603827f, 0.610496f, 0.617207f, 0.623960f, 0.630757f, + 0.637597f, 0.644480f, 0.651406f, 0.658375f, 0.665387f, 0.672443f, 0.679543f, 0.686685f, 0.693872f, 0.701102f, 0.708376f, + 0.715694f, 0.723055f, 0.730461f, 0.737911f, 0.745404f, 0.752942f, 0.760525f, 0.768151f, 0.775822f, 0.783538f, 0.791298f, + 0.799103f, 0.806952f, 0.814847f, 0.822786f, 0.830770f, 0.838799f, 0.846873f, 0.854993f, 0.863157f, 0.871367f, 0.879622f, + 0.887923f, 0.896269f, 0.904661f, 0.913099f, 0.921582f, 0.930111f, 0.938686f, 0.947307f, 0.955974f, 0.964686f, 0.973445f, + 0.982251f, 0.991102f, 1.0f +}; + +static float stbir__srgb_to_linear(float f) +{ + if (f <= 0.04045f) + return f / 12.92f; + else + return (float)pow((f + 0.055f) / 1.055f, 2.4f); +} + +static float stbir__linear_to_srgb(float f) +{ + if (f <= 0.0031308f) + return f * 12.92f; + else + return 1.055f * (float)pow(f, 1 / 2.4f) - 0.055f; +} + +#ifndef STBIR_NON_IEEE_FLOAT +// From https://gist.github.com/rygorous/2203834 + +typedef union +{ + stbir_uint32 u; + float f; +} stbir__FP32; + +static const stbir_uint32 fp32_to_srgb8_tab4[104] = { + 0x0073000d, 0x007a000d, 0x0080000d, 0x0087000d, 0x008d000d, 0x0094000d, 0x009a000d, 0x00a1000d, + 0x00a7001a, 0x00b4001a, 0x00c1001a, 0x00ce001a, 0x00da001a, 0x00e7001a, 0x00f4001a, 0x0101001a, + 0x010e0033, 0x01280033, 0x01410033, 0x015b0033, 0x01750033, 0x018f0033, 0x01a80033, 0x01c20033, + 0x01dc0067, 0x020f0067, 0x02430067, 0x02760067, 0x02aa0067, 0x02dd0067, 0x03110067, 0x03440067, + 0x037800ce, 0x03df00ce, 0x044600ce, 0x04ad00ce, 0x051400ce, 0x057b00c5, 0x05dd00bc, 0x063b00b5, + 0x06970158, 0x07420142, 0x07e30130, 0x087b0120, 0x090b0112, 0x09940106, 0x0a1700fc, 0x0a9500f2, + 0x0b0f01cb, 0x0bf401ae, 0x0ccb0195, 0x0d950180, 0x0e56016e, 0x0f0d015e, 0x0fbc0150, 0x10630143, + 0x11070264, 0x1238023e, 0x1357021d, 0x14660201, 0x156601e9, 0x165a01d3, 0x174401c0, 0x182401af, + 0x18fe0331, 0x1a9602fe, 0x1c1502d2, 0x1d7e02ad, 0x1ed4028d, 0x201a0270, 0x21520256, 0x227d0240, + 0x239f0443, 0x25c003fe, 0x27bf03c4, 0x29a10392, 0x2b6a0367, 0x2d1d0341, 0x2ebe031f, 0x304d0300, + 0x31d105b0, 0x34a80555, 0x37520507, 0x39d504c5, 0x3c37048b, 0x3e7c0458, 0x40a8042a, 0x42bd0401, + 0x44c20798, 0x488e071e, 0x4c1c06b6, 0x4f76065d, 0x52a50610, 0x55ac05cc, 0x5892058f, 0x5b590559, + 0x5e0c0a23, 0x631c0980, 0x67db08f6, 0x6c55087f, 0x70940818, 0x74a007bd, 0x787d076c, 0x7c330723, +}; + +static stbir_uint8 stbir__linear_to_srgb_uchar(float in) +{ + static const stbir__FP32 almostone = { 0x3f7fffff }; // 1-eps + static const stbir__FP32 minval = { (127-13) << 23 }; + stbir_uint32 tab,bias,scale,t; + stbir__FP32 f; + + // Clamp to [2^(-13), 1-eps]; these two values map to 0 and 1, respectively. + // The tests are carefully written so that NaNs map to 0, same as in the reference + // implementation. + if (!(in > minval.f)) // written this way to catch NaNs + in = minval.f; + if (in > almostone.f) + in = almostone.f; + + // Do the table lookup and unpack bias, scale + f.f = in; + tab = fp32_to_srgb8_tab4[(f.u - minval.u) >> 20]; + bias = (tab >> 16) << 9; + scale = tab & 0xffff; + + // Grab next-highest mantissa bits and perform linear interpolation + t = (f.u >> 12) & 0xff; + return (unsigned char) ((bias + scale*t) >> 16); +} + +#else +// sRGB transition values, scaled by 1<<28 +static int stbir__srgb_offset_to_linear_scaled[256] = +{ + 0, 40738, 122216, 203693, 285170, 366648, 448125, 529603, + 611080, 692557, 774035, 855852, 942009, 1033024, 1128971, 1229926, + 1335959, 1447142, 1563542, 1685229, 1812268, 1944725, 2082664, 2226148, + 2375238, 2529996, 2690481, 2856753, 3028870, 3206888, 3390865, 3580856, + 3776916, 3979100, 4187460, 4402049, 4622919, 4850123, 5083710, 5323731, + 5570236, 5823273, 6082892, 6349140, 6622065, 6901714, 7188133, 7481369, + 7781466, 8088471, 8402427, 8723380, 9051372, 9386448, 9728650, 10078021, + 10434603, 10798439, 11169569, 11548036, 11933879, 12327139, 12727857, 13136073, + 13551826, 13975156, 14406100, 14844697, 15290987, 15745007, 16206795, 16676389, + 17153826, 17639142, 18132374, 18633560, 19142734, 19659934, 20185196, 20718552, + 21260042, 21809696, 22367554, 22933648, 23508010, 24090680, 24681686, 25281066, + 25888850, 26505076, 27129772, 27762974, 28404716, 29055026, 29713942, 30381490, + 31057708, 31742624, 32436272, 33138682, 33849884, 34569912, 35298800, 36036568, + 36783260, 37538896, 38303512, 39077136, 39859796, 40651528, 41452360, 42262316, + 43081432, 43909732, 44747252, 45594016, 46450052, 47315392, 48190064, 49074096, + 49967516, 50870356, 51782636, 52704392, 53635648, 54576432, 55526772, 56486700, + 57456236, 58435408, 59424248, 60422780, 61431036, 62449032, 63476804, 64514376, + 65561776, 66619028, 67686160, 68763192, 69850160, 70947088, 72053992, 73170912, + 74297864, 75434880, 76581976, 77739184, 78906536, 80084040, 81271736, 82469648, + 83677792, 84896192, 86124888, 87363888, 88613232, 89872928, 91143016, 92423512, + 93714432, 95015816, 96327688, 97650056, 98982952, 100326408, 101680440, 103045072, + 104420320, 105806224, 107202800, 108610064, 110028048, 111456776, 112896264, 114346544, + 115807632, 117279552, 118762328, 120255976, 121760536, 123276016, 124802440, 126339832, + 127888216, 129447616, 131018048, 132599544, 134192112, 135795792, 137410592, 139036528, + 140673648, 142321952, 143981456, 145652208, 147334208, 149027488, 150732064, 152447968, + 154175200, 155913792, 157663776, 159425168, 161197984, 162982240, 164777968, 166585184, + 168403904, 170234160, 172075968, 173929344, 175794320, 177670896, 179559120, 181458992, + 183370528, 185293776, 187228736, 189175424, 191133888, 193104112, 195086128, 197079968, + 199085648, 201103184, 203132592, 205173888, 207227120, 209292272, 211369392, 213458480, + 215559568, 217672656, 219797792, 221934976, 224084240, 226245600, 228419056, 230604656, + 232802400, 235012320, 237234432, 239468736, 241715280, 243974080, 246245120, 248528464, + 250824112, 253132064, 255452368, 257785040, 260130080, 262487520, 264857376, 267239664, +}; + +static stbir_uint8 stbir__linear_to_srgb_uchar(float f) +{ + int x = (int) (f * (1 << 28)); // has headroom so you don't need to clamp + int v = 0; + int i; + + // Refine the guess with a short binary search. + i = v + 128; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i; + i = v + 64; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i; + i = v + 32; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i; + i = v + 16; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i; + i = v + 8; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i; + i = v + 4; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i; + i = v + 2; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i; + i = v + 1; if (x >= stbir__srgb_offset_to_linear_scaled[i]) v = i; + + return (stbir_uint8) v; +} +#endif + +static float stbir__filter_trapezoid(float x, float scale) +{ + float halfscale = scale / 2; + float t = 0.5f + halfscale; + STBIR_ASSERT(scale <= 1); + + x = (float)fabs(x); + + if (x >= t) + return 0; + else + { + float r = 0.5f - halfscale; + if (x <= r) + return 1; + else + return (t - x) / scale; + } +} + +static float stbir__support_trapezoid(float scale) +{ + STBIR_ASSERT(scale <= 1); + return 0.5f + scale / 2; +} + +static float stbir__filter_triangle(float x, float s) +{ + STBIR__UNUSED_PARAM(s); + + x = (float)fabs(x); + + if (x <= 1.0f) + return 1 - x; + else + return 0; +} + +static float stbir__filter_cubic(float x, float s) +{ + STBIR__UNUSED_PARAM(s); + + x = (float)fabs(x); + + if (x < 1.0f) + return (4 + x*x*(3*x - 6))/6; + else if (x < 2.0f) + return (8 + x*(-12 + x*(6 - x)))/6; + + return (0.0f); +} + +static float stbir__filter_catmullrom(float x, float s) +{ + STBIR__UNUSED_PARAM(s); + + x = (float)fabs(x); + + if (x < 1.0f) + return 1 - x*x*(2.5f - 1.5f*x); + else if (x < 2.0f) + return 2 - x*(4 + x*(0.5f*x - 2.5f)); + + return (0.0f); +} + +static float stbir__filter_mitchell(float x, float s) +{ + STBIR__UNUSED_PARAM(s); + + x = (float)fabs(x); + + if (x < 1.0f) + return (16 + x*x*(21 * x - 36))/18; + else if (x < 2.0f) + return (32 + x*(-60 + x*(36 - 7*x)))/18; + + return (0.0f); +} + +static float stbir__support_zero(float s) +{ + STBIR__UNUSED_PARAM(s); + return 0; +} + +static float stbir__support_one(float s) +{ + STBIR__UNUSED_PARAM(s); + return 1; +} + +static float stbir__support_two(float s) +{ + STBIR__UNUSED_PARAM(s); + return 2; +} + +static stbir__filter_info stbir__filter_info_table[] = { + { NULL, stbir__support_zero }, + { stbir__filter_trapezoid, stbir__support_trapezoid }, + { stbir__filter_triangle, stbir__support_one }, + { stbir__filter_cubic, stbir__support_two }, + { stbir__filter_catmullrom, stbir__support_two }, + { stbir__filter_mitchell, stbir__support_two }, +}; + +stbir__inline static int stbir__use_upsampling(float ratio) +{ + return ratio > 1; +} + +stbir__inline static int stbir__use_width_upsampling(stbir__info* stbir_info) +{ + return stbir__use_upsampling(stbir_info->horizontal_scale); +} + +stbir__inline static int stbir__use_height_upsampling(stbir__info* stbir_info) +{ + return stbir__use_upsampling(stbir_info->vertical_scale); +} + +// This is the maximum number of input samples that can affect an output sample +// with the given filter +static int stbir__get_filter_pixel_width(stbir_filter filter, float scale) +{ + STBIR_ASSERT(filter != 0); + STBIR_ASSERT(filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); + + if (stbir__use_upsampling(scale)) + return (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2); + else + return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2 / scale); +} + +// This is how much to expand buffers to account for filters seeking outside +// the image boundaries. +static int stbir__get_filter_pixel_margin(stbir_filter filter, float scale) +{ + return stbir__get_filter_pixel_width(filter, scale) / 2; +} + +static int stbir__get_coefficient_width(stbir_filter filter, float scale) +{ + if (stbir__use_upsampling(scale)) + return (int)ceil(stbir__filter_info_table[filter].support(1 / scale) * 2); + else + return (int)ceil(stbir__filter_info_table[filter].support(scale) * 2); +} + +static int stbir__get_contributors(float scale, stbir_filter filter, int input_size, int output_size) +{ + if (stbir__use_upsampling(scale)) + return output_size; + else + return (input_size + stbir__get_filter_pixel_margin(filter, scale) * 2); +} + +static int stbir__get_total_horizontal_coefficients(stbir__info* info) +{ + return info->horizontal_num_contributors + * stbir__get_coefficient_width (info->horizontal_filter, info->horizontal_scale); +} + +static int stbir__get_total_vertical_coefficients(stbir__info* info) +{ + return info->vertical_num_contributors + * stbir__get_coefficient_width (info->vertical_filter, info->vertical_scale); +} + +static stbir__contributors* stbir__get_contributor(stbir__contributors* contributors, int n) +{ + return &contributors[n]; +} + +// For perf reasons this code is duplicated in stbir__resample_horizontal_upsample/downsample, +// if you change it here change it there too. +static float* stbir__get_coefficient(float* coefficients, stbir_filter filter, float scale, int n, int c) +{ + int width = stbir__get_coefficient_width(filter, scale); + return &coefficients[width*n + c]; +} + +static int stbir__edge_wrap_slow(stbir_edge edge, int n, int max) +{ + switch (edge) + { + case STBIR_EDGE_ZERO: + return 0; // we'll decode the wrong pixel here, and then overwrite with 0s later + + case STBIR_EDGE_CLAMP: + if (n < 0) + return 0; + + if (n >= max) + return max - 1; + + return n; // NOTREACHED + + case STBIR_EDGE_REFLECT: + { + if (n < 0) + { + if (n < max) + return -n; + else + return max - 1; + } + + if (n >= max) + { + int max2 = max * 2; + if (n >= max2) + return 0; + else + return max2 - n - 1; + } + + return n; // NOTREACHED + } + + case STBIR_EDGE_WRAP: + if (n >= 0) + return (n % max); + else + { + int m = (-n) % max; + + if (m != 0) + m = max - m; + + return (m); + } + // NOTREACHED + + default: + STBIR_ASSERT(!"Unimplemented edge type"); + return 0; + } +} + +stbir__inline static int stbir__edge_wrap(stbir_edge edge, int n, int max) +{ + // avoid per-pixel switch + if (n >= 0 && n < max) + return n; + return stbir__edge_wrap_slow(edge, n, max); +} + +// What input pixels contribute to this output pixel? +static void stbir__calculate_sample_range_upsample(int n, float out_filter_radius, float scale_ratio, float out_shift, int* in_first_pixel, int* in_last_pixel, float* in_center_of_out) +{ + float out_pixel_center = (float)n + 0.5f; + float out_pixel_influence_lowerbound = out_pixel_center - out_filter_radius; + float out_pixel_influence_upperbound = out_pixel_center + out_filter_radius; + + float in_pixel_influence_lowerbound = (out_pixel_influence_lowerbound + out_shift) / scale_ratio; + float in_pixel_influence_upperbound = (out_pixel_influence_upperbound + out_shift) / scale_ratio; + + *in_center_of_out = (out_pixel_center + out_shift) / scale_ratio; + *in_first_pixel = (int)(floor(in_pixel_influence_lowerbound + 0.5)); + *in_last_pixel = (int)(floor(in_pixel_influence_upperbound - 0.5)); +} + +// What output pixels does this input pixel contribute to? +static void stbir__calculate_sample_range_downsample(int n, float in_pixels_radius, float scale_ratio, float out_shift, int* out_first_pixel, int* out_last_pixel, float* out_center_of_in) +{ + float in_pixel_center = (float)n + 0.5f; + float in_pixel_influence_lowerbound = in_pixel_center - in_pixels_radius; + float in_pixel_influence_upperbound = in_pixel_center + in_pixels_radius; + + float out_pixel_influence_lowerbound = in_pixel_influence_lowerbound * scale_ratio - out_shift; + float out_pixel_influence_upperbound = in_pixel_influence_upperbound * scale_ratio - out_shift; + + *out_center_of_in = in_pixel_center * scale_ratio - out_shift; + *out_first_pixel = (int)(floor(out_pixel_influence_lowerbound + 0.5)); + *out_last_pixel = (int)(floor(out_pixel_influence_upperbound - 0.5)); +} + +static void stbir__calculate_coefficients_upsample(stbir_filter filter, float scale, int in_first_pixel, int in_last_pixel, float in_center_of_out, stbir__contributors* contributor, float* coefficient_group) +{ + int i; + float total_filter = 0; + float filter_scale; + + STBIR_ASSERT(in_last_pixel - in_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(1/scale) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical. + + contributor->n0 = in_first_pixel; + contributor->n1 = in_last_pixel; + + STBIR_ASSERT(contributor->n1 >= contributor->n0); + + for (i = 0; i <= in_last_pixel - in_first_pixel; i++) + { + float in_pixel_center = (float)(i + in_first_pixel) + 0.5f; + coefficient_group[i] = stbir__filter_info_table[filter].kernel(in_center_of_out - in_pixel_center, 1 / scale); + + // If the coefficient is zero, skip it. (Don't do the <0 check here, we want the influence of those outside pixels.) + if (i == 0 && !coefficient_group[i]) + { + contributor->n0 = ++in_first_pixel; + i--; + continue; + } + + total_filter += coefficient_group[i]; + } + + // NOTE(fg): Not actually true in general, nor is there any reason to expect it should be. + // It would be true in exact math but is at best approximately true in floating-point math, + // and it would not make sense to try and put actual bounds on this here because it depends + // on the image aspect ratio which can get pretty extreme. + //STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(in_last_pixel + 1) + 0.5f - in_center_of_out, 1/scale) == 0); + + STBIR_ASSERT(total_filter > 0.9); + STBIR_ASSERT(total_filter < 1.1f); // Make sure it's not way off. + + // Make sure the sum of all coefficients is 1. + filter_scale = 1 / total_filter; + + for (i = 0; i <= in_last_pixel - in_first_pixel; i++) + coefficient_group[i] *= filter_scale; + + for (i = in_last_pixel - in_first_pixel; i >= 0; i--) + { + if (coefficient_group[i]) + break; + + // This line has no weight. We can skip it. + contributor->n1 = contributor->n0 + i - 1; + } +} + +static void stbir__calculate_coefficients_downsample(stbir_filter filter, float scale_ratio, int out_first_pixel, int out_last_pixel, float out_center_of_in, stbir__contributors* contributor, float* coefficient_group) +{ + int i; + + STBIR_ASSERT(out_last_pixel - out_first_pixel <= (int)ceil(stbir__filter_info_table[filter].support(scale_ratio) * 2)); // Taken directly from stbir__get_coefficient_width() which we can't call because we don't know if we're horizontal or vertical. + + contributor->n0 = out_first_pixel; + contributor->n1 = out_last_pixel; + + STBIR_ASSERT(contributor->n1 >= contributor->n0); + + for (i = 0; i <= out_last_pixel - out_first_pixel; i++) + { + float out_pixel_center = (float)(i + out_first_pixel) + 0.5f; + float x = out_pixel_center - out_center_of_in; + coefficient_group[i] = stbir__filter_info_table[filter].kernel(x, scale_ratio) * scale_ratio; + } + + // NOTE(fg): Not actually true in general, nor is there any reason to expect it should be. + // It would be true in exact math but is at best approximately true in floating-point math, + // and it would not make sense to try and put actual bounds on this here because it depends + // on the image aspect ratio which can get pretty extreme. + //STBIR_ASSERT(stbir__filter_info_table[filter].kernel((float)(out_last_pixel + 1) + 0.5f - out_center_of_in, scale_ratio) == 0); + + for (i = out_last_pixel - out_first_pixel; i >= 0; i--) + { + if (coefficient_group[i]) + break; + + // This line has no weight. We can skip it. + contributor->n1 = contributor->n0 + i - 1; + } +} + +static void stbir__normalize_downsample_coefficients(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, int input_size, int output_size) +{ + int num_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size); + int num_coefficients = stbir__get_coefficient_width(filter, scale_ratio); + int i, j; + int skip; + + for (i = 0; i < output_size; i++) + { + float scale; + float total = 0; + + for (j = 0; j < num_contributors; j++) + { + if (i >= contributors[j].n0 && i <= contributors[j].n1) + { + float coefficient = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0); + total += coefficient; + } + else if (i < contributors[j].n0) + break; + } + + STBIR_ASSERT(total > 0.9f); + STBIR_ASSERT(total < 1.1f); + + scale = 1 / total; + + for (j = 0; j < num_contributors; j++) + { + if (i >= contributors[j].n0 && i <= contributors[j].n1) + *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i - contributors[j].n0) *= scale; + else if (i < contributors[j].n0) + break; + } + } + + // Optimize: Skip zero coefficients and contributions outside of image bounds. + // Do this after normalizing because normalization depends on the n0/n1 values. + for (j = 0; j < num_contributors; j++) + { + int range, max, width; + + skip = 0; + while (*stbir__get_coefficient(coefficients, filter, scale_ratio, j, skip) == 0) + skip++; + + contributors[j].n0 += skip; + + while (contributors[j].n0 < 0) + { + contributors[j].n0++; + skip++; + } + + range = contributors[j].n1 - contributors[j].n0 + 1; + max = stbir__min(num_coefficients, range); + + width = stbir__get_coefficient_width(filter, scale_ratio); + for (i = 0; i < max; i++) + { + if (i + skip >= width) + break; + + *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i) = *stbir__get_coefficient(coefficients, filter, scale_ratio, j, i + skip); + } + + continue; + } + + // Using min to avoid writing into invalid pixels. + for (i = 0; i < num_contributors; i++) + contributors[i].n1 = stbir__min(contributors[i].n1, output_size - 1); +} + +// Each scan line uses the same kernel values so we should calculate the kernel +// values once and then we can use them for every scan line. +static void stbir__calculate_filters(stbir__contributors* contributors, float* coefficients, stbir_filter filter, float scale_ratio, float shift, int input_size, int output_size) +{ + int n; + int total_contributors = stbir__get_contributors(scale_ratio, filter, input_size, output_size); + + if (stbir__use_upsampling(scale_ratio)) + { + float out_pixels_radius = stbir__filter_info_table[filter].support(1 / scale_ratio) * scale_ratio; + + // Looping through out pixels + for (n = 0; n < total_contributors; n++) + { + float in_center_of_out; // Center of the current out pixel in the in pixel space + int in_first_pixel, in_last_pixel; + + stbir__calculate_sample_range_upsample(n, out_pixels_radius, scale_ratio, shift, &in_first_pixel, &in_last_pixel, &in_center_of_out); + + stbir__calculate_coefficients_upsample(filter, scale_ratio, in_first_pixel, in_last_pixel, in_center_of_out, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0)); + } + } + else + { + float in_pixels_radius = stbir__filter_info_table[filter].support(scale_ratio) / scale_ratio; + + // Looping through in pixels + for (n = 0; n < total_contributors; n++) + { + float out_center_of_in; // Center of the current out pixel in the in pixel space + int out_first_pixel, out_last_pixel; + int n_adjusted = n - stbir__get_filter_pixel_margin(filter, scale_ratio); + + stbir__calculate_sample_range_downsample(n_adjusted, in_pixels_radius, scale_ratio, shift, &out_first_pixel, &out_last_pixel, &out_center_of_in); + + stbir__calculate_coefficients_downsample(filter, scale_ratio, out_first_pixel, out_last_pixel, out_center_of_in, stbir__get_contributor(contributors, n), stbir__get_coefficient(coefficients, filter, scale_ratio, n, 0)); + } + + stbir__normalize_downsample_coefficients(contributors, coefficients, filter, scale_ratio, input_size, output_size); + } +} + +static float* stbir__get_decode_buffer(stbir__info* stbir_info) +{ + // The 0 index of the decode buffer starts after the margin. This makes + // it okay to use negative indexes on the decode buffer. + return &stbir_info->decode_buffer[stbir_info->horizontal_filter_pixel_margin * stbir_info->channels]; +} + +#define STBIR__DECODE(type, colorspace) ((int)(type) * (STBIR_MAX_COLORSPACES) + (int)(colorspace)) + +static void stbir__decode_scanline(stbir__info* stbir_info, int n) +{ + int c; + int channels = stbir_info->channels; + int alpha_channel = stbir_info->alpha_channel; + int type = stbir_info->type; + int colorspace = stbir_info->colorspace; + int input_w = stbir_info->input_w; + size_t input_stride_bytes = stbir_info->input_stride_bytes; + float* decode_buffer = stbir__get_decode_buffer(stbir_info); + stbir_edge edge_horizontal = stbir_info->edge_horizontal; + stbir_edge edge_vertical = stbir_info->edge_vertical; + size_t in_buffer_row_offset = stbir__edge_wrap(edge_vertical, n, stbir_info->input_h) * input_stride_bytes; + const void* input_data = (char *) stbir_info->input_data + in_buffer_row_offset; + int max_x = input_w + stbir_info->horizontal_filter_pixel_margin; + int decode = STBIR__DECODE(type, colorspace); + + int x = -stbir_info->horizontal_filter_pixel_margin; + + // special handling for STBIR_EDGE_ZERO because it needs to return an item that doesn't appear in the input, + // and we want to avoid paying overhead on every pixel if not STBIR_EDGE_ZERO + if (edge_vertical == STBIR_EDGE_ZERO && (n < 0 || n >= stbir_info->input_h)) + { + for (; x < max_x; x++) + for (c = 0; c < channels; c++) + decode_buffer[x*channels + c] = 0; + return; + } + + switch (decode) + { + case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR): + for (; x < max_x; x++) + { + int decode_pixel_index = x * channels; + int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels; + for (c = 0; c < channels; c++) + decode_buffer[decode_pixel_index + c] = ((float)((const unsigned char*)input_data)[input_pixel_index + c]) / stbir__max_uint8_as_float; + } + break; + + case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB): + for (; x < max_x; x++) + { + int decode_pixel_index = x * channels; + int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels; + for (c = 0; c < channels; c++) + decode_buffer[decode_pixel_index + c] = stbir__srgb_uchar_to_linear_float[((const unsigned char*)input_data)[input_pixel_index + c]]; + + if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE)) + decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned char*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint8_as_float; + } + break; + + case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR): + for (; x < max_x; x++) + { + int decode_pixel_index = x * channels; + int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels; + for (c = 0; c < channels; c++) + decode_buffer[decode_pixel_index + c] = ((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float; + } + break; + + case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB): + for (; x < max_x; x++) + { + int decode_pixel_index = x * channels; + int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels; + for (c = 0; c < channels; c++) + decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((float)((const unsigned short*)input_data)[input_pixel_index + c]) / stbir__max_uint16_as_float); + + if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE)) + decode_buffer[decode_pixel_index + alpha_channel] = ((float)((const unsigned short*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint16_as_float; + } + break; + + case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR): + for (; x < max_x; x++) + { + int decode_pixel_index = x * channels; + int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels; + for (c = 0; c < channels; c++) + decode_buffer[decode_pixel_index + c] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float); + } + break; + + case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB): + for (; x < max_x; x++) + { + int decode_pixel_index = x * channels; + int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels; + for (c = 0; c < channels; c++) + decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear((float)(((double)((const unsigned int*)input_data)[input_pixel_index + c]) / stbir__max_uint32_as_float)); + + if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE)) + decode_buffer[decode_pixel_index + alpha_channel] = (float)(((double)((const unsigned int*)input_data)[input_pixel_index + alpha_channel]) / stbir__max_uint32_as_float); + } + break; + + case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR): + for (; x < max_x; x++) + { + int decode_pixel_index = x * channels; + int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels; + for (c = 0; c < channels; c++) + decode_buffer[decode_pixel_index + c] = ((const float*)input_data)[input_pixel_index + c]; + } + break; + + case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB): + for (; x < max_x; x++) + { + int decode_pixel_index = x * channels; + int input_pixel_index = stbir__edge_wrap(edge_horizontal, x, input_w) * channels; + for (c = 0; c < channels; c++) + decode_buffer[decode_pixel_index + c] = stbir__srgb_to_linear(((const float*)input_data)[input_pixel_index + c]); + + if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE)) + decode_buffer[decode_pixel_index + alpha_channel] = ((const float*)input_data)[input_pixel_index + alpha_channel]; + } + + break; + + default: + STBIR_ASSERT(!"Unknown type/colorspace/channels combination."); + break; + } + + if (!(stbir_info->flags & STBIR_FLAG_ALPHA_PREMULTIPLIED)) + { + for (x = -stbir_info->horizontal_filter_pixel_margin; x < max_x; x++) + { + int decode_pixel_index = x * channels; + + // If the alpha value is 0 it will clobber the color values. Make sure it's not. + float alpha = decode_buffer[decode_pixel_index + alpha_channel]; +#ifndef STBIR_NO_ALPHA_EPSILON + if (stbir_info->type != STBIR_TYPE_FLOAT) { + alpha += STBIR_ALPHA_EPSILON; + decode_buffer[decode_pixel_index + alpha_channel] = alpha; + } +#endif + for (c = 0; c < channels; c++) + { + if (c == alpha_channel) + continue; + + decode_buffer[decode_pixel_index + c] *= alpha; + } + } + } + + if (edge_horizontal == STBIR_EDGE_ZERO) + { + for (x = -stbir_info->horizontal_filter_pixel_margin; x < 0; x++) + { + for (c = 0; c < channels; c++) + decode_buffer[x*channels + c] = 0; + } + for (x = input_w; x < max_x; x++) + { + for (c = 0; c < channels; c++) + decode_buffer[x*channels + c] = 0; + } + } +} + +static float* stbir__get_ring_buffer_entry(float* ring_buffer, int index, int ring_buffer_length) +{ + return &ring_buffer[index * ring_buffer_length]; +} + +static float* stbir__add_empty_ring_buffer_entry(stbir__info* stbir_info, int n) +{ + int ring_buffer_index; + float* ring_buffer; + + stbir_info->ring_buffer_last_scanline = n; + + if (stbir_info->ring_buffer_begin_index < 0) + { + ring_buffer_index = stbir_info->ring_buffer_begin_index = 0; + stbir_info->ring_buffer_first_scanline = n; + } + else + { + ring_buffer_index = (stbir_info->ring_buffer_begin_index + (stbir_info->ring_buffer_last_scanline - stbir_info->ring_buffer_first_scanline)) % stbir_info->ring_buffer_num_entries; + STBIR_ASSERT(ring_buffer_index != stbir_info->ring_buffer_begin_index); + } + + ring_buffer = stbir__get_ring_buffer_entry(stbir_info->ring_buffer, ring_buffer_index, stbir_info->ring_buffer_length_bytes / sizeof(float)); + memset(ring_buffer, 0, stbir_info->ring_buffer_length_bytes); + + return ring_buffer; +} + + +static void stbir__resample_horizontal_upsample(stbir__info* stbir_info, float* output_buffer) +{ + int x, k; + int output_w = stbir_info->output_w; + int channels = stbir_info->channels; + float* decode_buffer = stbir__get_decode_buffer(stbir_info); + stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors; + float* horizontal_coefficients = stbir_info->horizontal_coefficients; + int coefficient_width = stbir_info->horizontal_coefficient_width; + + for (x = 0; x < output_w; x++) + { + int n0 = horizontal_contributors[x].n0; + int n1 = horizontal_contributors[x].n1; + + int out_pixel_index = x * channels; + int coefficient_group = coefficient_width * x; + int coefficient_counter = 0; + + STBIR_ASSERT(n1 >= n0); + STBIR_ASSERT(n0 >= -stbir_info->horizontal_filter_pixel_margin); + STBIR_ASSERT(n1 >= -stbir_info->horizontal_filter_pixel_margin); + STBIR_ASSERT(n0 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin); + STBIR_ASSERT(n1 < stbir_info->input_w + stbir_info->horizontal_filter_pixel_margin); + + switch (channels) { + case 1: + for (k = n0; k <= n1; k++) + { + int in_pixel_index = k * 1; + float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++]; + STBIR_ASSERT(coefficient != 0); + output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient; + } + break; + case 2: + for (k = n0; k <= n1; k++) + { + int in_pixel_index = k * 2; + float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++]; + STBIR_ASSERT(coefficient != 0); + output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient; + output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient; + } + break; + case 3: + for (k = n0; k <= n1; k++) + { + int in_pixel_index = k * 3; + float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++]; + STBIR_ASSERT(coefficient != 0); + output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient; + output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient; + output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient; + } + break; + case 4: + for (k = n0; k <= n1; k++) + { + int in_pixel_index = k * 4; + float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++]; + STBIR_ASSERT(coefficient != 0); + output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient; + output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient; + output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient; + output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient; + } + break; + default: + for (k = n0; k <= n1; k++) + { + int in_pixel_index = k * channels; + float coefficient = horizontal_coefficients[coefficient_group + coefficient_counter++]; + int c; + STBIR_ASSERT(coefficient != 0); + for (c = 0; c < channels; c++) + output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient; + } + break; + } + } +} + +static void stbir__resample_horizontal_downsample(stbir__info* stbir_info, float* output_buffer) +{ + int x, k; + int input_w = stbir_info->input_w; + int channels = stbir_info->channels; + float* decode_buffer = stbir__get_decode_buffer(stbir_info); + stbir__contributors* horizontal_contributors = stbir_info->horizontal_contributors; + float* horizontal_coefficients = stbir_info->horizontal_coefficients; + int coefficient_width = stbir_info->horizontal_coefficient_width; + int filter_pixel_margin = stbir_info->horizontal_filter_pixel_margin; + int max_x = input_w + filter_pixel_margin * 2; + + STBIR_ASSERT(!stbir__use_width_upsampling(stbir_info)); + + switch (channels) { + case 1: + for (x = 0; x < max_x; x++) + { + int n0 = horizontal_contributors[x].n0; + int n1 = horizontal_contributors[x].n1; + + int in_x = x - filter_pixel_margin; + int in_pixel_index = in_x * 1; + int max_n = n1; + int coefficient_group = coefficient_width * x; + + for (k = n0; k <= max_n; k++) + { + int out_pixel_index = k * 1; + float coefficient = horizontal_coefficients[coefficient_group + k - n0]; + output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient; + } + } + break; + + case 2: + for (x = 0; x < max_x; x++) + { + int n0 = horizontal_contributors[x].n0; + int n1 = horizontal_contributors[x].n1; + + int in_x = x - filter_pixel_margin; + int in_pixel_index = in_x * 2; + int max_n = n1; + int coefficient_group = coefficient_width * x; + + for (k = n0; k <= max_n; k++) + { + int out_pixel_index = k * 2; + float coefficient = horizontal_coefficients[coefficient_group + k - n0]; + output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient; + output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient; + } + } + break; + + case 3: + for (x = 0; x < max_x; x++) + { + int n0 = horizontal_contributors[x].n0; + int n1 = horizontal_contributors[x].n1; + + int in_x = x - filter_pixel_margin; + int in_pixel_index = in_x * 3; + int max_n = n1; + int coefficient_group = coefficient_width * x; + + for (k = n0; k <= max_n; k++) + { + int out_pixel_index = k * 3; + float coefficient = horizontal_coefficients[coefficient_group + k - n0]; + output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient; + output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient; + output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient; + } + } + break; + + case 4: + for (x = 0; x < max_x; x++) + { + int n0 = horizontal_contributors[x].n0; + int n1 = horizontal_contributors[x].n1; + + int in_x = x - filter_pixel_margin; + int in_pixel_index = in_x * 4; + int max_n = n1; + int coefficient_group = coefficient_width * x; + + for (k = n0; k <= max_n; k++) + { + int out_pixel_index = k * 4; + float coefficient = horizontal_coefficients[coefficient_group + k - n0]; + output_buffer[out_pixel_index + 0] += decode_buffer[in_pixel_index + 0] * coefficient; + output_buffer[out_pixel_index + 1] += decode_buffer[in_pixel_index + 1] * coefficient; + output_buffer[out_pixel_index + 2] += decode_buffer[in_pixel_index + 2] * coefficient; + output_buffer[out_pixel_index + 3] += decode_buffer[in_pixel_index + 3] * coefficient; + } + } + break; + + default: + for (x = 0; x < max_x; x++) + { + int n0 = horizontal_contributors[x].n0; + int n1 = horizontal_contributors[x].n1; + + int in_x = x - filter_pixel_margin; + int in_pixel_index = in_x * channels; + int max_n = n1; + int coefficient_group = coefficient_width * x; + + for (k = n0; k <= max_n; k++) + { + int c; + int out_pixel_index = k * channels; + float coefficient = horizontal_coefficients[coefficient_group + k - n0]; + for (c = 0; c < channels; c++) + output_buffer[out_pixel_index + c] += decode_buffer[in_pixel_index + c] * coefficient; + } + } + break; + } +} + +static void stbir__decode_and_resample_upsample(stbir__info* stbir_info, int n) +{ + // Decode the nth scanline from the source image into the decode buffer. + stbir__decode_scanline(stbir_info, n); + + // Now resample it into the ring buffer. + if (stbir__use_width_upsampling(stbir_info)) + stbir__resample_horizontal_upsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n)); + else + stbir__resample_horizontal_downsample(stbir_info, stbir__add_empty_ring_buffer_entry(stbir_info, n)); + + // Now it's sitting in the ring buffer ready to be used as source for the vertical sampling. +} + +static void stbir__decode_and_resample_downsample(stbir__info* stbir_info, int n) +{ + // Decode the nth scanline from the source image into the decode buffer. + stbir__decode_scanline(stbir_info, n); + + memset(stbir_info->horizontal_buffer, 0, stbir_info->output_w * stbir_info->channels * sizeof(float)); + + // Now resample it into the horizontal buffer. + if (stbir__use_width_upsampling(stbir_info)) + stbir__resample_horizontal_upsample(stbir_info, stbir_info->horizontal_buffer); + else + stbir__resample_horizontal_downsample(stbir_info, stbir_info->horizontal_buffer); + + // Now it's sitting in the horizontal buffer ready to be distributed into the ring buffers. +} + +// Get the specified scan line from the ring buffer. +static float* stbir__get_ring_buffer_scanline(int get_scanline, float* ring_buffer, int begin_index, int first_scanline, int ring_buffer_num_entries, int ring_buffer_length) +{ + int ring_buffer_index = (begin_index + (get_scanline - first_scanline)) % ring_buffer_num_entries; + return stbir__get_ring_buffer_entry(ring_buffer, ring_buffer_index, ring_buffer_length); +} + + +static void stbir__encode_scanline(stbir__info* stbir_info, int num_pixels, void *output_buffer, float *encode_buffer, int channels, int alpha_channel, int decode) +{ + int x; + int n; + int num_nonalpha; + stbir_uint16 nonalpha[STBIR_MAX_CHANNELS]; + + if (!(stbir_info->flags&STBIR_FLAG_ALPHA_PREMULTIPLIED)) + { + for (x=0; x < num_pixels; ++x) + { + int pixel_index = x*channels; + + float alpha = encode_buffer[pixel_index + alpha_channel]; + float reciprocal_alpha = alpha ? 1.0f / alpha : 0; + + // unrolling this produced a 1% slowdown upscaling a large RGBA linear-space image on my machine - stb + for (n = 0; n < channels; n++) + if (n != alpha_channel) + encode_buffer[pixel_index + n] *= reciprocal_alpha; + + // We added in a small epsilon to prevent the color channel from being deleted with zero alpha. + // Because we only add it for integer types, it will automatically be discarded on integer + // conversion, so we don't need to subtract it back out (which would be problematic for + // numeric precision reasons). + } + } + + // build a table of all channels that need colorspace correction, so + // we don't perform colorspace correction on channels that don't need it. + for (x = 0, num_nonalpha = 0; x < channels; ++x) + { + if (x != alpha_channel || (stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE)) + { + nonalpha[num_nonalpha++] = (stbir_uint16)x; + } + } + + #define STBIR__ROUND_INT(f) ((int) ((f)+0.5)) + #define STBIR__ROUND_UINT(f) ((stbir_uint32) ((f)+0.5)) + + #ifdef STBIR__SATURATE_INT + #define STBIR__ENCODE_LINEAR8(f) stbir__saturate8 (STBIR__ROUND_INT((f) * stbir__max_uint8_as_float )) + #define STBIR__ENCODE_LINEAR16(f) stbir__saturate16(STBIR__ROUND_INT((f) * stbir__max_uint16_as_float)) + #else + #define STBIR__ENCODE_LINEAR8(f) (unsigned char ) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint8_as_float ) + #define STBIR__ENCODE_LINEAR16(f) (unsigned short) STBIR__ROUND_INT(stbir__saturate(f) * stbir__max_uint16_as_float) + #endif + + switch (decode) + { + case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_LINEAR): + for (x=0; x < num_pixels; ++x) + { + int pixel_index = x*channels; + + for (n = 0; n < channels; n++) + { + int index = pixel_index + n; + ((unsigned char*)output_buffer)[index] = STBIR__ENCODE_LINEAR8(encode_buffer[index]); + } + } + break; + + case STBIR__DECODE(STBIR_TYPE_UINT8, STBIR_COLORSPACE_SRGB): + for (x=0; x < num_pixels; ++x) + { + int pixel_index = x*channels; + + for (n = 0; n < num_nonalpha; n++) + { + int index = pixel_index + nonalpha[n]; + ((unsigned char*)output_buffer)[index] = stbir__linear_to_srgb_uchar(encode_buffer[index]); + } + + if (!(stbir_info->flags & STBIR_FLAG_ALPHA_USES_COLORSPACE)) + ((unsigned char *)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR8(encode_buffer[pixel_index+alpha_channel]); + } + break; + + case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_LINEAR): + for (x=0; x < num_pixels; ++x) + { + int pixel_index = x*channels; + + for (n = 0; n < channels; n++) + { + int index = pixel_index + n; + ((unsigned short*)output_buffer)[index] = STBIR__ENCODE_LINEAR16(encode_buffer[index]); + } + } + break; + + case STBIR__DECODE(STBIR_TYPE_UINT16, STBIR_COLORSPACE_SRGB): + for (x=0; x < num_pixels; ++x) + { + int pixel_index = x*channels; + + for (n = 0; n < num_nonalpha; n++) + { + int index = pixel_index + nonalpha[n]; + ((unsigned short*)output_buffer)[index] = (unsigned short)STBIR__ROUND_INT(stbir__linear_to_srgb(stbir__saturate(encode_buffer[index])) * stbir__max_uint16_as_float); + } + + if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE)) + ((unsigned short*)output_buffer)[pixel_index + alpha_channel] = STBIR__ENCODE_LINEAR16(encode_buffer[pixel_index + alpha_channel]); + } + + break; + + case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_LINEAR): + for (x=0; x < num_pixels; ++x) + { + int pixel_index = x*channels; + + for (n = 0; n < channels; n++) + { + int index = pixel_index + n; + ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__saturate(encode_buffer[index])) * stbir__max_uint32_as_float); + } + } + break; + + case STBIR__DECODE(STBIR_TYPE_UINT32, STBIR_COLORSPACE_SRGB): + for (x=0; x < num_pixels; ++x) + { + int pixel_index = x*channels; + + for (n = 0; n < num_nonalpha; n++) + { + int index = pixel_index + nonalpha[n]; + ((unsigned int*)output_buffer)[index] = (unsigned int)STBIR__ROUND_UINT(((double)stbir__linear_to_srgb(stbir__saturate(encode_buffer[index]))) * stbir__max_uint32_as_float); + } + + if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE)) + ((unsigned int*)output_buffer)[pixel_index + alpha_channel] = (unsigned int)STBIR__ROUND_INT(((double)stbir__saturate(encode_buffer[pixel_index + alpha_channel])) * stbir__max_uint32_as_float); + } + break; + + case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_LINEAR): + for (x=0; x < num_pixels; ++x) + { + int pixel_index = x*channels; + + for (n = 0; n < channels; n++) + { + int index = pixel_index + n; + ((float*)output_buffer)[index] = encode_buffer[index]; + } + } + break; + + case STBIR__DECODE(STBIR_TYPE_FLOAT, STBIR_COLORSPACE_SRGB): + for (x=0; x < num_pixels; ++x) + { + int pixel_index = x*channels; + + for (n = 0; n < num_nonalpha; n++) + { + int index = pixel_index + nonalpha[n]; + ((float*)output_buffer)[index] = stbir__linear_to_srgb(encode_buffer[index]); + } + + if (!(stbir_info->flags&STBIR_FLAG_ALPHA_USES_COLORSPACE)) + ((float*)output_buffer)[pixel_index + alpha_channel] = encode_buffer[pixel_index + alpha_channel]; + } + break; + + default: + STBIR_ASSERT(!"Unknown type/colorspace/channels combination."); + break; + } +} + +static void stbir__resample_vertical_upsample(stbir__info* stbir_info, int n) +{ + int x, k; + int output_w = stbir_info->output_w; + stbir__contributors* vertical_contributors = stbir_info->vertical_contributors; + float* vertical_coefficients = stbir_info->vertical_coefficients; + int channels = stbir_info->channels; + int alpha_channel = stbir_info->alpha_channel; + int type = stbir_info->type; + int colorspace = stbir_info->colorspace; + int ring_buffer_entries = stbir_info->ring_buffer_num_entries; + void* output_data = stbir_info->output_data; + float* encode_buffer = stbir_info->encode_buffer; + int decode = STBIR__DECODE(type, colorspace); + int coefficient_width = stbir_info->vertical_coefficient_width; + int coefficient_counter; + int contributor = n; + + float* ring_buffer = stbir_info->ring_buffer; + int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index; + int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline; + int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float); + + int n0,n1, output_row_start; + int coefficient_group = coefficient_width * contributor; + + n0 = vertical_contributors[contributor].n0; + n1 = vertical_contributors[contributor].n1; + + output_row_start = n * stbir_info->output_stride_bytes; + + STBIR_ASSERT(stbir__use_height_upsampling(stbir_info)); + + memset(encode_buffer, 0, output_w * sizeof(float) * channels); + + // I tried reblocking this for better cache usage of encode_buffer + // (using x_outer, k, x_inner), but it lost speed. -- stb + + coefficient_counter = 0; + switch (channels) { + case 1: + for (k = n0; k <= n1; k++) + { + int coefficient_index = coefficient_counter++; + float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length); + float coefficient = vertical_coefficients[coefficient_group + coefficient_index]; + for (x = 0; x < output_w; ++x) + { + int in_pixel_index = x * 1; + encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient; + } + } + break; + case 2: + for (k = n0; k <= n1; k++) + { + int coefficient_index = coefficient_counter++; + float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length); + float coefficient = vertical_coefficients[coefficient_group + coefficient_index]; + for (x = 0; x < output_w; ++x) + { + int in_pixel_index = x * 2; + encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient; + encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient; + } + } + break; + case 3: + for (k = n0; k <= n1; k++) + { + int coefficient_index = coefficient_counter++; + float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length); + float coefficient = vertical_coefficients[coefficient_group + coefficient_index]; + for (x = 0; x < output_w; ++x) + { + int in_pixel_index = x * 3; + encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient; + encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient; + encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient; + } + } + break; + case 4: + for (k = n0; k <= n1; k++) + { + int coefficient_index = coefficient_counter++; + float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length); + float coefficient = vertical_coefficients[coefficient_group + coefficient_index]; + for (x = 0; x < output_w; ++x) + { + int in_pixel_index = x * 4; + encode_buffer[in_pixel_index + 0] += ring_buffer_entry[in_pixel_index + 0] * coefficient; + encode_buffer[in_pixel_index + 1] += ring_buffer_entry[in_pixel_index + 1] * coefficient; + encode_buffer[in_pixel_index + 2] += ring_buffer_entry[in_pixel_index + 2] * coefficient; + encode_buffer[in_pixel_index + 3] += ring_buffer_entry[in_pixel_index + 3] * coefficient; + } + } + break; + default: + for (k = n0; k <= n1; k++) + { + int coefficient_index = coefficient_counter++; + float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length); + float coefficient = vertical_coefficients[coefficient_group + coefficient_index]; + for (x = 0; x < output_w; ++x) + { + int in_pixel_index = x * channels; + int c; + for (c = 0; c < channels; c++) + encode_buffer[in_pixel_index + c] += ring_buffer_entry[in_pixel_index + c] * coefficient; + } + } + break; + } + stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, encode_buffer, channels, alpha_channel, decode); +} + +static void stbir__resample_vertical_downsample(stbir__info* stbir_info, int n) +{ + int x, k; + int output_w = stbir_info->output_w; + stbir__contributors* vertical_contributors = stbir_info->vertical_contributors; + float* vertical_coefficients = stbir_info->vertical_coefficients; + int channels = stbir_info->channels; + int ring_buffer_entries = stbir_info->ring_buffer_num_entries; + float* horizontal_buffer = stbir_info->horizontal_buffer; + int coefficient_width = stbir_info->vertical_coefficient_width; + int contributor = n + stbir_info->vertical_filter_pixel_margin; + + float* ring_buffer = stbir_info->ring_buffer; + int ring_buffer_begin_index = stbir_info->ring_buffer_begin_index; + int ring_buffer_first_scanline = stbir_info->ring_buffer_first_scanline; + int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float); + int n0,n1; + + n0 = vertical_contributors[contributor].n0; + n1 = vertical_contributors[contributor].n1; + + STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info)); + + for (k = n0; k <= n1; k++) + { + int coefficient_index = k - n0; + int coefficient_group = coefficient_width * contributor; + float coefficient = vertical_coefficients[coefficient_group + coefficient_index]; + + float* ring_buffer_entry = stbir__get_ring_buffer_scanline(k, ring_buffer, ring_buffer_begin_index, ring_buffer_first_scanline, ring_buffer_entries, ring_buffer_length); + + switch (channels) { + case 1: + for (x = 0; x < output_w; x++) + { + int in_pixel_index = x * 1; + ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient; + } + break; + case 2: + for (x = 0; x < output_w; x++) + { + int in_pixel_index = x * 2; + ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient; + ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient; + } + break; + case 3: + for (x = 0; x < output_w; x++) + { + int in_pixel_index = x * 3; + ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient; + ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient; + ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient; + } + break; + case 4: + for (x = 0; x < output_w; x++) + { + int in_pixel_index = x * 4; + ring_buffer_entry[in_pixel_index + 0] += horizontal_buffer[in_pixel_index + 0] * coefficient; + ring_buffer_entry[in_pixel_index + 1] += horizontal_buffer[in_pixel_index + 1] * coefficient; + ring_buffer_entry[in_pixel_index + 2] += horizontal_buffer[in_pixel_index + 2] * coefficient; + ring_buffer_entry[in_pixel_index + 3] += horizontal_buffer[in_pixel_index + 3] * coefficient; + } + break; + default: + for (x = 0; x < output_w; x++) + { + int in_pixel_index = x * channels; + + int c; + for (c = 0; c < channels; c++) + ring_buffer_entry[in_pixel_index + c] += horizontal_buffer[in_pixel_index + c] * coefficient; + } + break; + } + } +} + +static void stbir__buffer_loop_upsample(stbir__info* stbir_info) +{ + int y; + float scale_ratio = stbir_info->vertical_scale; + float out_scanlines_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(1/scale_ratio) * scale_ratio; + + STBIR_ASSERT(stbir__use_height_upsampling(stbir_info)); + + for (y = 0; y < stbir_info->output_h; y++) + { + float in_center_of_out = 0; // Center of the current out scanline in the in scanline space + int in_first_scanline = 0, in_last_scanline = 0; + + stbir__calculate_sample_range_upsample(y, out_scanlines_radius, scale_ratio, stbir_info->vertical_shift, &in_first_scanline, &in_last_scanline, &in_center_of_out); + + STBIR_ASSERT(in_last_scanline - in_first_scanline + 1 <= stbir_info->ring_buffer_num_entries); + + if (stbir_info->ring_buffer_begin_index >= 0) + { + // Get rid of whatever we don't need anymore. + while (in_first_scanline > stbir_info->ring_buffer_first_scanline) + { + if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline) + { + // We just popped the last scanline off the ring buffer. + // Reset it to the empty state. + stbir_info->ring_buffer_begin_index = -1; + stbir_info->ring_buffer_first_scanline = 0; + stbir_info->ring_buffer_last_scanline = 0; + break; + } + else + { + stbir_info->ring_buffer_first_scanline++; + stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries; + } + } + } + + // Load in new ones. + if (stbir_info->ring_buffer_begin_index < 0) + stbir__decode_and_resample_upsample(stbir_info, in_first_scanline); + + while (in_last_scanline > stbir_info->ring_buffer_last_scanline) + stbir__decode_and_resample_upsample(stbir_info, stbir_info->ring_buffer_last_scanline + 1); + + // Now all buffers should be ready to write a row of vertical sampling. + stbir__resample_vertical_upsample(stbir_info, y); + + STBIR_PROGRESS_REPORT((float)y / stbir_info->output_h); + } +} + +static void stbir__empty_ring_buffer(stbir__info* stbir_info, int first_necessary_scanline) +{ + int output_stride_bytes = stbir_info->output_stride_bytes; + int channels = stbir_info->channels; + int alpha_channel = stbir_info->alpha_channel; + int type = stbir_info->type; + int colorspace = stbir_info->colorspace; + int output_w = stbir_info->output_w; + void* output_data = stbir_info->output_data; + int decode = STBIR__DECODE(type, colorspace); + + float* ring_buffer = stbir_info->ring_buffer; + int ring_buffer_length = stbir_info->ring_buffer_length_bytes/sizeof(float); + + if (stbir_info->ring_buffer_begin_index >= 0) + { + // Get rid of whatever we don't need anymore. + while (first_necessary_scanline > stbir_info->ring_buffer_first_scanline) + { + if (stbir_info->ring_buffer_first_scanline >= 0 && stbir_info->ring_buffer_first_scanline < stbir_info->output_h) + { + int output_row_start = stbir_info->ring_buffer_first_scanline * output_stride_bytes; + float* ring_buffer_entry = stbir__get_ring_buffer_entry(ring_buffer, stbir_info->ring_buffer_begin_index, ring_buffer_length); + stbir__encode_scanline(stbir_info, output_w, (char *) output_data + output_row_start, ring_buffer_entry, channels, alpha_channel, decode); + STBIR_PROGRESS_REPORT((float)stbir_info->ring_buffer_first_scanline / stbir_info->output_h); + } + + if (stbir_info->ring_buffer_first_scanline == stbir_info->ring_buffer_last_scanline) + { + // We just popped the last scanline off the ring buffer. + // Reset it to the empty state. + stbir_info->ring_buffer_begin_index = -1; + stbir_info->ring_buffer_first_scanline = 0; + stbir_info->ring_buffer_last_scanline = 0; + break; + } + else + { + stbir_info->ring_buffer_first_scanline++; + stbir_info->ring_buffer_begin_index = (stbir_info->ring_buffer_begin_index + 1) % stbir_info->ring_buffer_num_entries; + } + } + } +} + +static void stbir__buffer_loop_downsample(stbir__info* stbir_info) +{ + int y; + float scale_ratio = stbir_info->vertical_scale; + int output_h = stbir_info->output_h; + float in_pixels_radius = stbir__filter_info_table[stbir_info->vertical_filter].support(scale_ratio) / scale_ratio; + int pixel_margin = stbir_info->vertical_filter_pixel_margin; + int max_y = stbir_info->input_h + pixel_margin; + + STBIR_ASSERT(!stbir__use_height_upsampling(stbir_info)); + + for (y = -pixel_margin; y < max_y; y++) + { + float out_center_of_in; // Center of the current out scanline in the in scanline space + int out_first_scanline, out_last_scanline; + + stbir__calculate_sample_range_downsample(y, in_pixels_radius, scale_ratio, stbir_info->vertical_shift, &out_first_scanline, &out_last_scanline, &out_center_of_in); + + STBIR_ASSERT(out_last_scanline - out_first_scanline + 1 <= stbir_info->ring_buffer_num_entries); + + if (out_last_scanline < 0 || out_first_scanline >= output_h) + continue; + + stbir__empty_ring_buffer(stbir_info, out_first_scanline); + + stbir__decode_and_resample_downsample(stbir_info, y); + + // Load in new ones. + if (stbir_info->ring_buffer_begin_index < 0) + stbir__add_empty_ring_buffer_entry(stbir_info, out_first_scanline); + + while (out_last_scanline > stbir_info->ring_buffer_last_scanline) + stbir__add_empty_ring_buffer_entry(stbir_info, stbir_info->ring_buffer_last_scanline + 1); + + // Now the horizontal buffer is ready to write to all ring buffer rows. + stbir__resample_vertical_downsample(stbir_info, y); + } + + stbir__empty_ring_buffer(stbir_info, stbir_info->output_h); +} + +static void stbir__setup(stbir__info *info, int input_w, int input_h, int output_w, int output_h, int channels) +{ + info->input_w = input_w; + info->input_h = input_h; + info->output_w = output_w; + info->output_h = output_h; + info->channels = channels; +} + +static void stbir__calculate_transform(stbir__info *info, float s0, float t0, float s1, float t1, float *transform) +{ + info->s0 = s0; + info->t0 = t0; + info->s1 = s1; + info->t1 = t1; + + if (transform) + { + info->horizontal_scale = transform[0]; + info->vertical_scale = transform[1]; + info->horizontal_shift = transform[2]; + info->vertical_shift = transform[3]; + } + else + { + info->horizontal_scale = ((float)info->output_w / info->input_w) / (s1 - s0); + info->vertical_scale = ((float)info->output_h / info->input_h) / (t1 - t0); + + info->horizontal_shift = s0 * info->output_w / (s1 - s0); + info->vertical_shift = t0 * info->output_h / (t1 - t0); + } +} + +static void stbir__choose_filter(stbir__info *info, stbir_filter h_filter, stbir_filter v_filter) +{ + if (h_filter == 0) + h_filter = stbir__use_upsampling(info->horizontal_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE; + if (v_filter == 0) + v_filter = stbir__use_upsampling(info->vertical_scale) ? STBIR_DEFAULT_FILTER_UPSAMPLE : STBIR_DEFAULT_FILTER_DOWNSAMPLE; + info->horizontal_filter = h_filter; + info->vertical_filter = v_filter; +} + +static stbir_uint32 stbir__calculate_memory(stbir__info *info) +{ + int pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale); + int filter_height = stbir__get_filter_pixel_width(info->vertical_filter, info->vertical_scale); + + info->horizontal_num_contributors = stbir__get_contributors(info->horizontal_scale, info->horizontal_filter, info->input_w, info->output_w); + info->vertical_num_contributors = stbir__get_contributors(info->vertical_scale , info->vertical_filter , info->input_h, info->output_h); + + // One extra entry because floating point precision problems sometimes cause an extra to be necessary. + info->ring_buffer_num_entries = filter_height + 1; + + info->horizontal_contributors_size = info->horizontal_num_contributors * sizeof(stbir__contributors); + info->horizontal_coefficients_size = stbir__get_total_horizontal_coefficients(info) * sizeof(float); + info->vertical_contributors_size = info->vertical_num_contributors * sizeof(stbir__contributors); + info->vertical_coefficients_size = stbir__get_total_vertical_coefficients(info) * sizeof(float); + info->decode_buffer_size = (info->input_w + pixel_margin * 2) * info->channels * sizeof(float); + info->horizontal_buffer_size = info->output_w * info->channels * sizeof(float); + info->ring_buffer_size = info->output_w * info->channels * info->ring_buffer_num_entries * sizeof(float); + info->encode_buffer_size = info->output_w * info->channels * sizeof(float); + + STBIR_ASSERT(info->horizontal_filter != 0); + STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late + STBIR_ASSERT(info->vertical_filter != 0); + STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); // this now happens too late + + if (stbir__use_height_upsampling(info)) + // The horizontal buffer is for when we're downsampling the height and we + // can't output the result of sampling the decode buffer directly into the + // ring buffers. + info->horizontal_buffer_size = 0; + else + // The encode buffer is to retain precision in the height upsampling method + // and isn't used when height downsampling. + info->encode_buffer_size = 0; + + return info->horizontal_contributors_size + info->horizontal_coefficients_size + + info->vertical_contributors_size + info->vertical_coefficients_size + + info->decode_buffer_size + info->horizontal_buffer_size + + info->ring_buffer_size + info->encode_buffer_size; +} + +static int stbir__resize_allocated(stbir__info *info, + const void* input_data, int input_stride_in_bytes, + void* output_data, int output_stride_in_bytes, + int alpha_channel, stbir_uint32 flags, stbir_datatype type, + stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace, + void* tempmem, size_t tempmem_size_in_bytes) +{ + size_t memory_required = stbir__calculate_memory(info); + + int width_stride_input = input_stride_in_bytes ? input_stride_in_bytes : info->channels * info->input_w * stbir__type_size[type]; + int width_stride_output = output_stride_in_bytes ? output_stride_in_bytes : info->channels * info->output_w * stbir__type_size[type]; + +#ifdef STBIR_DEBUG_OVERWRITE_TEST +#define OVERWRITE_ARRAY_SIZE 8 + unsigned char overwrite_output_before_pre[OVERWRITE_ARRAY_SIZE]; + unsigned char overwrite_tempmem_before_pre[OVERWRITE_ARRAY_SIZE]; + unsigned char overwrite_output_after_pre[OVERWRITE_ARRAY_SIZE]; + unsigned char overwrite_tempmem_after_pre[OVERWRITE_ARRAY_SIZE]; + + size_t begin_forbidden = width_stride_output * (info->output_h - 1) + info->output_w * info->channels * stbir__type_size[type]; + memcpy(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE); + memcpy(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE); + memcpy(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE); + memcpy(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE); +#endif + + STBIR_ASSERT(info->channels >= 0); + STBIR_ASSERT(info->channels <= STBIR_MAX_CHANNELS); + + if (info->channels < 0 || info->channels > STBIR_MAX_CHANNELS) + return 0; + + STBIR_ASSERT(info->horizontal_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); + STBIR_ASSERT(info->vertical_filter < STBIR__ARRAY_SIZE(stbir__filter_info_table)); + + if (info->horizontal_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table)) + return 0; + if (info->vertical_filter >= STBIR__ARRAY_SIZE(stbir__filter_info_table)) + return 0; + + if (alpha_channel < 0) + flags |= STBIR_FLAG_ALPHA_USES_COLORSPACE | STBIR_FLAG_ALPHA_PREMULTIPLIED; + + if (!(flags&STBIR_FLAG_ALPHA_USES_COLORSPACE) || !(flags&STBIR_FLAG_ALPHA_PREMULTIPLIED)) { + STBIR_ASSERT(alpha_channel >= 0 && alpha_channel < info->channels); + } + + if (alpha_channel >= info->channels) + return 0; + + STBIR_ASSERT(tempmem); + + if (!tempmem) + return 0; + + STBIR_ASSERT(tempmem_size_in_bytes >= memory_required); + + if (tempmem_size_in_bytes < memory_required) + return 0; + + memset(tempmem, 0, tempmem_size_in_bytes); + + info->input_data = input_data; + info->input_stride_bytes = width_stride_input; + + info->output_data = output_data; + info->output_stride_bytes = width_stride_output; + + info->alpha_channel = alpha_channel; + info->flags = flags; + info->type = type; + info->edge_horizontal = edge_horizontal; + info->edge_vertical = edge_vertical; + info->colorspace = colorspace; + + info->horizontal_coefficient_width = stbir__get_coefficient_width (info->horizontal_filter, info->horizontal_scale); + info->vertical_coefficient_width = stbir__get_coefficient_width (info->vertical_filter , info->vertical_scale ); + info->horizontal_filter_pixel_width = stbir__get_filter_pixel_width (info->horizontal_filter, info->horizontal_scale); + info->vertical_filter_pixel_width = stbir__get_filter_pixel_width (info->vertical_filter , info->vertical_scale ); + info->horizontal_filter_pixel_margin = stbir__get_filter_pixel_margin(info->horizontal_filter, info->horizontal_scale); + info->vertical_filter_pixel_margin = stbir__get_filter_pixel_margin(info->vertical_filter , info->vertical_scale ); + + info->ring_buffer_length_bytes = info->output_w * info->channels * sizeof(float); + info->decode_buffer_pixels = info->input_w + info->horizontal_filter_pixel_margin * 2; + +#define STBIR__NEXT_MEMPTR(current, newtype) (newtype*)(((unsigned char*)current) + current##_size) + + info->horizontal_contributors = (stbir__contributors *) tempmem; + info->horizontal_coefficients = STBIR__NEXT_MEMPTR(info->horizontal_contributors, float); + info->vertical_contributors = STBIR__NEXT_MEMPTR(info->horizontal_coefficients, stbir__contributors); + info->vertical_coefficients = STBIR__NEXT_MEMPTR(info->vertical_contributors, float); + info->decode_buffer = STBIR__NEXT_MEMPTR(info->vertical_coefficients, float); + + if (stbir__use_height_upsampling(info)) + { + info->horizontal_buffer = NULL; + info->ring_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float); + info->encode_buffer = STBIR__NEXT_MEMPTR(info->ring_buffer, float); + + STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->encode_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes); + } + else + { + info->horizontal_buffer = STBIR__NEXT_MEMPTR(info->decode_buffer, float); + info->ring_buffer = STBIR__NEXT_MEMPTR(info->horizontal_buffer, float); + info->encode_buffer = NULL; + + STBIR_ASSERT((size_t)STBIR__NEXT_MEMPTR(info->ring_buffer, unsigned char) == (size_t)tempmem + tempmem_size_in_bytes); + } + +#undef STBIR__NEXT_MEMPTR + + // This signals that the ring buffer is empty + info->ring_buffer_begin_index = -1; + + stbir__calculate_filters(info->horizontal_contributors, info->horizontal_coefficients, info->horizontal_filter, info->horizontal_scale, info->horizontal_shift, info->input_w, info->output_w); + stbir__calculate_filters(info->vertical_contributors, info->vertical_coefficients, info->vertical_filter, info->vertical_scale, info->vertical_shift, info->input_h, info->output_h); + + STBIR_PROGRESS_REPORT(0); + + if (stbir__use_height_upsampling(info)) + stbir__buffer_loop_upsample(info); + else + stbir__buffer_loop_downsample(info); + + STBIR_PROGRESS_REPORT(1); + +#ifdef STBIR_DEBUG_OVERWRITE_TEST + STBIR_ASSERT(memcmp(overwrite_output_before_pre, &((unsigned char*)output_data)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0); + STBIR_ASSERT(memcmp(overwrite_output_after_pre, &((unsigned char*)output_data)[begin_forbidden], OVERWRITE_ARRAY_SIZE) == 0); + STBIR_ASSERT(memcmp(overwrite_tempmem_before_pre, &((unsigned char*)tempmem)[-OVERWRITE_ARRAY_SIZE], OVERWRITE_ARRAY_SIZE) == 0); + STBIR_ASSERT(memcmp(overwrite_tempmem_after_pre, &((unsigned char*)tempmem)[tempmem_size_in_bytes], OVERWRITE_ARRAY_SIZE) == 0); +#endif + + return 1; +} + + +static int stbir__resize_arbitrary( + void *alloc_context, + const void* input_data, int input_w, int input_h, int input_stride_in_bytes, + void* output_data, int output_w, int output_h, int output_stride_in_bytes, + float s0, float t0, float s1, float t1, float *transform, + int channels, int alpha_channel, stbir_uint32 flags, stbir_datatype type, + stbir_filter h_filter, stbir_filter v_filter, + stbir_edge edge_horizontal, stbir_edge edge_vertical, stbir_colorspace colorspace) +{ + stbir__info info; + int result; + size_t memory_required; + void* extra_memory; + + stbir__setup(&info, input_w, input_h, output_w, output_h, channels); + stbir__calculate_transform(&info, s0,t0,s1,t1,transform); + stbir__choose_filter(&info, h_filter, v_filter); + memory_required = stbir__calculate_memory(&info); + extra_memory = STBIR_MALLOC(memory_required, alloc_context); + + if (!extra_memory) + return 0; + + result = stbir__resize_allocated(&info, input_data, input_stride_in_bytes, + output_data, output_stride_in_bytes, + alpha_channel, flags, type, + edge_horizontal, edge_vertical, + colorspace, extra_memory, memory_required); + + STBIR_FREE(extra_memory, alloc_context); + + return result; +} + +STBIRDEF int stbir_resize_uint8( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + int num_channels) +{ + return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes, + output_pixels, output_w, output_h, output_stride_in_bytes, + 0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT, + STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR); +} + +STBIRDEF int stbir_resize_float( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + float *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + int num_channels) +{ + return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes, + output_pixels, output_w, output_h, output_stride_in_bytes, + 0,0,1,1,NULL,num_channels,-1,0, STBIR_TYPE_FLOAT, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT, + STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_LINEAR); +} + +STBIRDEF int stbir_resize_uint8_srgb(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + int num_channels, int alpha_channel, int flags) +{ + return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes, + output_pixels, output_w, output_h, output_stride_in_bytes, + 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT, + STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP, STBIR_COLORSPACE_SRGB); +} + +STBIRDEF int stbir_resize_uint8_srgb_edgemode(const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_wrap_mode) +{ + return stbir__resize_arbitrary(NULL, input_pixels, input_w, input_h, input_stride_in_bytes, + output_pixels, output_w, output_h, output_stride_in_bytes, + 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, STBIR_FILTER_DEFAULT, STBIR_FILTER_DEFAULT, + edge_wrap_mode, edge_wrap_mode, STBIR_COLORSPACE_SRGB); +} + +STBIRDEF int stbir_resize_uint8_generic( const unsigned char *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + unsigned char *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space, + void *alloc_context) +{ + return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes, + output_pixels, output_w, output_h, output_stride_in_bytes, + 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT8, filter, filter, + edge_wrap_mode, edge_wrap_mode, space); +} + +STBIRDEF int stbir_resize_uint16_generic(const stbir_uint16 *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + stbir_uint16 *output_pixels , int output_w, int output_h, int output_stride_in_bytes, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space, + void *alloc_context) +{ + return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes, + output_pixels, output_w, output_h, output_stride_in_bytes, + 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_UINT16, filter, filter, + edge_wrap_mode, edge_wrap_mode, space); +} + + +STBIRDEF int stbir_resize_float_generic( const float *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + float *output_pixels , int output_w, int output_h, int output_stride_in_bytes, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_wrap_mode, stbir_filter filter, stbir_colorspace space, + void *alloc_context) +{ + return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes, + output_pixels, output_w, output_h, output_stride_in_bytes, + 0,0,1,1,NULL,num_channels,alpha_channel,flags, STBIR_TYPE_FLOAT, filter, filter, + edge_wrap_mode, edge_wrap_mode, space); +} + + +STBIRDEF int stbir_resize( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + void *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + stbir_datatype datatype, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical, + stbir_filter filter_horizontal, stbir_filter filter_vertical, + stbir_colorspace space, void *alloc_context) +{ + return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes, + output_pixels, output_w, output_h, output_stride_in_bytes, + 0,0,1,1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical, + edge_mode_horizontal, edge_mode_vertical, space); +} + + +STBIRDEF int stbir_resize_subpixel(const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + void *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + stbir_datatype datatype, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical, + stbir_filter filter_horizontal, stbir_filter filter_vertical, + stbir_colorspace space, void *alloc_context, + float x_scale, float y_scale, + float x_offset, float y_offset) +{ + float transform[4]; + transform[0] = x_scale; + transform[1] = y_scale; + transform[2] = x_offset; + transform[3] = y_offset; + return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes, + output_pixels, output_w, output_h, output_stride_in_bytes, + 0,0,1,1,transform,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical, + edge_mode_horizontal, edge_mode_vertical, space); +} + +STBIRDEF int stbir_resize_region( const void *input_pixels , int input_w , int input_h , int input_stride_in_bytes, + void *output_pixels, int output_w, int output_h, int output_stride_in_bytes, + stbir_datatype datatype, + int num_channels, int alpha_channel, int flags, + stbir_edge edge_mode_horizontal, stbir_edge edge_mode_vertical, + stbir_filter filter_horizontal, stbir_filter filter_vertical, + stbir_colorspace space, void *alloc_context, + float s0, float t0, float s1, float t1) +{ + return stbir__resize_arbitrary(alloc_context, input_pixels, input_w, input_h, input_stride_in_bytes, + output_pixels, output_w, output_h, output_stride_in_bytes, + s0,t0,s1,t1,NULL,num_channels,alpha_channel,flags, datatype, filter_horizontal, filter_vertical, + edge_mode_horizontal, edge_mode_vertical, space); +} + +#endif // STB_IMAGE_RESIZE_IMPLEMENTATION + +/* +------------------------------------------------------------------------------ +This software is available under 2 licenses -- choose whichever you prefer. +------------------------------------------------------------------------------ +ALTERNATIVE A - MIT License +Copyright (c) 2017 Sean Barrett +Permission is hereby granted, free of charge, to any person obtaining a copy of +this software and associated documentation files (the "Software"), to deal in +the Software without restriction, including without limitation the rights to +use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies +of the Software, and to permit persons to whom the Software is furnished to do +so, subject to the following conditions: +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +------------------------------------------------------------------------------ +ALTERNATIVE B - Public Domain (www.unlicense.org) +This is free and unencumbered software released into the public domain. +Anyone is free to copy, modify, publish, use, compile, sell, or distribute this +software, either in source code form or as a compiled binary, for any purpose, +commercial or non-commercial, and by any means. +In jurisdictions that recognize copyright laws, the author or authors of this +software dedicate any and all copyright interest in the software to the public +domain. We make this dedication for the benefit of the public at large and to +the detriment of our heirs and successors. We intend this dedication to be an +overt act of relinquishment in perpetuity of all present and future rights to +this software under copyright law. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN +ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. +------------------------------------------------------------------------------ +*/ diff --git a/Source/vk2dfd.cpp b/Source/vk2dfd.cpp new file mode 100644 index 0000000..46c6532 --- /dev/null +++ b/Source/vk2dfd.cpp @@ -0,0 +1,33 @@ +/* -*- tab-width: 4; -*- */ +/* vi: set sw=2 ts=4 expandtab: */ + +/* Copyright 2019-2020 Mark Callow + * SPDX-License-Identifier: Apache-2.0 + */ + +/** + * @file + * @~English + * @brief Create a DFD for a VkFormat. + */ + +#include "dfd.h" + +/** + * @~English + * @brief Create a DFD matching a VkFormat. + * + * @param[in] format VkFormat for which to create a DFD. + * + * @return pointer to the created DFD or 0 if format not supported or + * unrecognized. Caller is responsible for freeing the created + * DFD. + */ +uint32_t * vk2dfd(enum VkFormat format) + { + switch (format) { +#include "vk2dfd.inl" + default: return 0; + } + } + diff --git a/Source/vk2dfd.inl b/Source/vk2dfd.inl new file mode 100644 index 0000000..7ceb2fd --- /dev/null +++ b/Source/vk2dfd.inl @@ -0,0 +1,294 @@ +/* Copyright 2019-2020 The Khronos Group Inc. */ +/* SPDX-License-Identifier: Apache-2.0 */ + +/***************************** Do not edit. ***************************** + Automatically generated by makevk2dfd.pl. + *************************************************************************/ + +/* Vulkan combined depth & stencil formats are not included here + * because they do not exist outside a Vulkan device. + */ +case VK_FORMAT_R4G4_UNORM_PACK8: { + int channels[] = {1,0}; int bits[] = {4,4}; + return createDFDPacked(0, 2, bits, channels, s_UNORM); +} +case VK_FORMAT_R4G4B4A4_UNORM_PACK16: { + int channels[] = {3,2,1,0}; int bits[] = {4,4,4,4}; + return createDFDPacked(0, 4, bits, channels, s_UNORM); +} +case VK_FORMAT_B4G4R4A4_UNORM_PACK16: { + int channels[] = {3,0,1,2}; int bits[] = {4,4,4,4}; + return createDFDPacked(0, 4, bits, channels, s_UNORM); +} +case VK_FORMAT_R5G6B5_UNORM_PACK16: { + int channels[] = {2,1,0}; int bits[] = {5,6,5}; + return createDFDPacked(0, 3, bits, channels, s_UNORM); +} +case VK_FORMAT_B5G6R5_UNORM_PACK16: { + int channels[] = {0,1,2}; int bits[] = {5,6,5}; + return createDFDPacked(0, 3, bits, channels, s_UNORM); +} +case VK_FORMAT_R5G5B5A1_UNORM_PACK16: { + int channels[] = {3,2,1,0}; int bits[] = {1,5,5,5}; + return createDFDPacked(0, 4, bits, channels, s_UNORM); +} +case VK_FORMAT_B5G5R5A1_UNORM_PACK16: { + int channels[] = {3,0,1,2}; int bits[] = {1,5,5,5}; + return createDFDPacked(0, 4, bits, channels, s_UNORM); +} +case VK_FORMAT_A1R5G5B5_UNORM_PACK16: { + int channels[] = {2,1,0,3}; int bits[] = {5,5,5,1}; + return createDFDPacked(0, 4, bits, channels, s_UNORM); +} +case VK_FORMAT_R8_UNORM: return createDFDUnpacked(0, 1, 1, 0, s_UNORM); +case VK_FORMAT_R8_SNORM: return createDFDUnpacked(0, 1, 1, 0, s_SNORM); +case VK_FORMAT_R8_USCALED: return createDFDUnpacked(0, 1, 1, 0, s_USCALED); +case VK_FORMAT_R8_SSCALED: return createDFDUnpacked(0, 1, 1, 0, s_SSCALED); +case VK_FORMAT_R8_UINT: return createDFDUnpacked(0, 1, 1, 0, s_UINT); +case VK_FORMAT_R8_SINT: return createDFDUnpacked(0, 1, 1, 0, s_SINT); +case VK_FORMAT_R8_SRGB: return createDFDUnpacked(0, 1, 1, 0, s_SRGB); +case VK_FORMAT_R8G8_UNORM: return createDFDUnpacked(0, 2, 1, 0, s_UNORM); +case VK_FORMAT_R8G8_SNORM: return createDFDUnpacked(0, 2, 1, 0, s_SNORM); +case VK_FORMAT_R8G8_USCALED: return createDFDUnpacked(0, 2, 1, 0, s_USCALED); +case VK_FORMAT_R8G8_SSCALED: return createDFDUnpacked(0, 2, 1, 0, s_SSCALED); +case VK_FORMAT_R8G8_UINT: return createDFDUnpacked(0, 2, 1, 0, s_UINT); +case VK_FORMAT_R8G8_SINT: return createDFDUnpacked(0, 2, 1, 0, s_SINT); +case VK_FORMAT_R8G8_SRGB: return createDFDUnpacked(0, 2, 1, 0, s_SRGB); +case VK_FORMAT_R8G8B8_UNORM: return createDFDUnpacked(0, 3, 1, 0, s_UNORM); +case VK_FORMAT_R8G8B8_SNORM: return createDFDUnpacked(0, 3, 1, 0, s_SNORM); +case VK_FORMAT_R8G8B8_USCALED: return createDFDUnpacked(0, 3, 1, 0, s_USCALED); +case VK_FORMAT_R8G8B8_SSCALED: return createDFDUnpacked(0, 3, 1, 0, s_SSCALED); +case VK_FORMAT_R8G8B8_UINT: return createDFDUnpacked(0, 3, 1, 0, s_UINT); +case VK_FORMAT_R8G8B8_SINT: return createDFDUnpacked(0, 3, 1, 0, s_SINT); +case VK_FORMAT_R8G8B8_SRGB: return createDFDUnpacked(0, 3, 1, 0, s_SRGB); +case VK_FORMAT_B8G8R8_UNORM: return createDFDUnpacked(0, 3, 1, 1, s_UNORM); +case VK_FORMAT_B8G8R8_SNORM: return createDFDUnpacked(0, 3, 1, 1, s_SNORM); +case VK_FORMAT_B8G8R8_USCALED: return createDFDUnpacked(0, 3, 1, 1, s_USCALED); +case VK_FORMAT_B8G8R8_SSCALED: return createDFDUnpacked(0, 3, 1, 1, s_SSCALED); +case VK_FORMAT_B8G8R8_UINT: return createDFDUnpacked(0, 3, 1, 1, s_UINT); +case VK_FORMAT_B8G8R8_SINT: return createDFDUnpacked(0, 3, 1, 1, s_SINT); +case VK_FORMAT_B8G8R8_SRGB: return createDFDUnpacked(0, 3, 1, 1, s_SRGB); +case VK_FORMAT_R8G8B8A8_UNORM: return createDFDUnpacked(0, 4, 1, 0, s_UNORM); +case VK_FORMAT_R8G8B8A8_SNORM: return createDFDUnpacked(0, 4, 1, 0, s_SNORM); +case VK_FORMAT_R8G8B8A8_USCALED: return createDFDUnpacked(0, 4, 1, 0, s_USCALED); +case VK_FORMAT_R8G8B8A8_SSCALED: return createDFDUnpacked(0, 4, 1, 0, s_SSCALED); +case VK_FORMAT_R8G8B8A8_UINT: return createDFDUnpacked(0, 4, 1, 0, s_UINT); +case VK_FORMAT_R8G8B8A8_SINT: return createDFDUnpacked(0, 4, 1, 0, s_SINT); +case VK_FORMAT_R8G8B8A8_SRGB: return createDFDUnpacked(0, 4, 1, 0, s_SRGB); +case VK_FORMAT_B8G8R8A8_UNORM: return createDFDUnpacked(0, 4, 1, 1, s_UNORM); +case VK_FORMAT_B8G8R8A8_SNORM: return createDFDUnpacked(0, 4, 1, 1, s_SNORM); +case VK_FORMAT_B8G8R8A8_USCALED: return createDFDUnpacked(0, 4, 1, 1, s_USCALED); +case VK_FORMAT_B8G8R8A8_SSCALED: return createDFDUnpacked(0, 4, 1, 1, s_SSCALED); +case VK_FORMAT_B8G8R8A8_UINT: return createDFDUnpacked(0, 4, 1, 1, s_UINT); +case VK_FORMAT_B8G8R8A8_SINT: return createDFDUnpacked(0, 4, 1, 1, s_SINT); +case VK_FORMAT_B8G8R8A8_SRGB: return createDFDUnpacked(0, 4, 1, 1, s_SRGB); +case VK_FORMAT_A8B8G8R8_UNORM_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {8,8,8,8}; + return createDFDPacked(0, 4, bits, channels, s_UNORM); +} +case VK_FORMAT_A8B8G8R8_SNORM_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {8,8,8,8}; + return createDFDPacked(0, 4, bits, channels, s_SNORM); +} +case VK_FORMAT_A8B8G8R8_USCALED_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {8,8,8,8}; + return createDFDPacked(0, 4, bits, channels, s_USCALED); +} +case VK_FORMAT_A8B8G8R8_SSCALED_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {8,8,8,8}; + return createDFDPacked(0, 4, bits, channels, s_SSCALED); +} +case VK_FORMAT_A8B8G8R8_UINT_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {8,8,8,8}; + return createDFDPacked(0, 4, bits, channels, s_UINT); +} +case VK_FORMAT_A8B8G8R8_SINT_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {8,8,8,8}; + return createDFDPacked(0, 4, bits, channels, s_SINT); +} +case VK_FORMAT_A8B8G8R8_SRGB_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {8,8,8,8}; + return createDFDPacked(0, 4, bits, channels, s_SRGB); +} +case VK_FORMAT_A2R10G10B10_UNORM_PACK32: { + int channels[] = {2,1,0,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_UNORM); +} +case VK_FORMAT_A2R10G10B10_SNORM_PACK32: { + int channels[] = {2,1,0,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_SNORM); +} +case VK_FORMAT_A2R10G10B10_USCALED_PACK32: { + int channels[] = {2,1,0,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_USCALED); +} +case VK_FORMAT_A2R10G10B10_SSCALED_PACK32: { + int channels[] = {2,1,0,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_SSCALED); +} +case VK_FORMAT_A2R10G10B10_UINT_PACK32: { + int channels[] = {2,1,0,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_UINT); +} +case VK_FORMAT_A2R10G10B10_SINT_PACK32: { + int channels[] = {2,1,0,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_SINT); +} +case VK_FORMAT_A2B10G10R10_UNORM_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_UNORM); +} +case VK_FORMAT_A2B10G10R10_SNORM_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_SNORM); +} +case VK_FORMAT_A2B10G10R10_USCALED_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_USCALED); +} +case VK_FORMAT_A2B10G10R10_SSCALED_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_SSCALED); +} +case VK_FORMAT_A2B10G10R10_UINT_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_UINT); +} +case VK_FORMAT_A2B10G10R10_SINT_PACK32: { + int channels[] = {0,1,2,3}; int bits[] = {10,10,10,2}; + return createDFDPacked(0, 4, bits, channels, s_SINT); +} +case VK_FORMAT_R16_UNORM: return createDFDUnpacked(0, 1, 2, 0, s_UNORM); +case VK_FORMAT_R16_SNORM: return createDFDUnpacked(0, 1, 2, 0, s_SNORM); +case VK_FORMAT_R16_USCALED: return createDFDUnpacked(0, 1, 2, 0, s_USCALED); +case VK_FORMAT_R16_SSCALED: return createDFDUnpacked(0, 1, 2, 0, s_SSCALED); +case VK_FORMAT_R16_UINT: return createDFDUnpacked(0, 1, 2, 0, s_UINT); +case VK_FORMAT_R16_SINT: return createDFDUnpacked(0, 1, 2, 0, s_SINT); +case VK_FORMAT_R16_SFLOAT: return createDFDUnpacked(0, 1, 2, 0, s_SFLOAT); +case VK_FORMAT_R16G16_UNORM: return createDFDUnpacked(0, 2, 2, 0, s_UNORM); +case VK_FORMAT_R16G16_SNORM: return createDFDUnpacked(0, 2, 2, 0, s_SNORM); +case VK_FORMAT_R16G16_USCALED: return createDFDUnpacked(0, 2, 2, 0, s_USCALED); +case VK_FORMAT_R16G16_SSCALED: return createDFDUnpacked(0, 2, 2, 0, s_SSCALED); +case VK_FORMAT_R16G16_UINT: return createDFDUnpacked(0, 2, 2, 0, s_UINT); +case VK_FORMAT_R16G16_SINT: return createDFDUnpacked(0, 2, 2, 0, s_SINT); +case VK_FORMAT_R16G16_SFLOAT: return createDFDUnpacked(0, 2, 2, 0, s_SFLOAT); +case VK_FORMAT_R16G16B16_UNORM: return createDFDUnpacked(0, 3, 2, 0, s_UNORM); +case VK_FORMAT_R16G16B16_SNORM: return createDFDUnpacked(0, 3, 2, 0, s_SNORM); +case VK_FORMAT_R16G16B16_USCALED: return createDFDUnpacked(0, 3, 2, 0, s_USCALED); +case VK_FORMAT_R16G16B16_SSCALED: return createDFDUnpacked(0, 3, 2, 0, s_SSCALED); +case VK_FORMAT_R16G16B16_UINT: return createDFDUnpacked(0, 3, 2, 0, s_UINT); +case VK_FORMAT_R16G16B16_SINT: return createDFDUnpacked(0, 3, 2, 0, s_SINT); +case VK_FORMAT_R16G16B16_SFLOAT: return createDFDUnpacked(0, 3, 2, 0, s_SFLOAT); +case VK_FORMAT_R16G16B16A16_UNORM: return createDFDUnpacked(0, 4, 2, 0, s_UNORM); +case VK_FORMAT_R16G16B16A16_SNORM: return createDFDUnpacked(0, 4, 2, 0, s_SNORM); +case VK_FORMAT_R16G16B16A16_USCALED: return createDFDUnpacked(0, 4, 2, 0, s_USCALED); +case VK_FORMAT_R16G16B16A16_SSCALED: return createDFDUnpacked(0, 4, 2, 0, s_SSCALED); +case VK_FORMAT_R16G16B16A16_UINT: return createDFDUnpacked(0, 4, 2, 0, s_UINT); +case VK_FORMAT_R16G16B16A16_SINT: return createDFDUnpacked(0, 4, 2, 0, s_SINT); +case VK_FORMAT_R16G16B16A16_SFLOAT: return createDFDUnpacked(0, 4, 2, 0, s_SFLOAT); +case VK_FORMAT_R32_UINT: return createDFDUnpacked(0, 1, 4, 0, s_UINT); +case VK_FORMAT_R32_SINT: return createDFDUnpacked(0, 1, 4, 0, s_SINT); +case VK_FORMAT_R32_SFLOAT: return createDFDUnpacked(0, 1, 4, 0, s_SFLOAT); +case VK_FORMAT_R32G32_UINT: return createDFDUnpacked(0, 2, 4, 0, s_UINT); +case VK_FORMAT_R32G32_SINT: return createDFDUnpacked(0, 2, 4, 0, s_SINT); +case VK_FORMAT_R32G32_SFLOAT: return createDFDUnpacked(0, 2, 4, 0, s_SFLOAT); +case VK_FORMAT_R32G32B32_UINT: return createDFDUnpacked(0, 3, 4, 0, s_UINT); +case VK_FORMAT_R32G32B32_SINT: return createDFDUnpacked(0, 3, 4, 0, s_SINT); +case VK_FORMAT_R32G32B32_SFLOAT: return createDFDUnpacked(0, 3, 4, 0, s_SFLOAT); +case VK_FORMAT_R32G32B32A32_UINT: return createDFDUnpacked(0, 4, 4, 0, s_UINT); +case VK_FORMAT_R32G32B32A32_SINT: return createDFDUnpacked(0, 4, 4, 0, s_SINT); +case VK_FORMAT_R32G32B32A32_SFLOAT: return createDFDUnpacked(0, 4, 4, 0, s_SFLOAT); +case VK_FORMAT_R64_UINT: return createDFDUnpacked(0, 1, 8, 0, s_UINT); +case VK_FORMAT_R64_SINT: return createDFDUnpacked(0, 1, 8, 0, s_SINT); +case VK_FORMAT_R64_SFLOAT: return createDFDUnpacked(0, 1, 8, 0, s_SFLOAT); +case VK_FORMAT_R64G64_UINT: return createDFDUnpacked(0, 2, 8, 0, s_UINT); +case VK_FORMAT_R64G64_SINT: return createDFDUnpacked(0, 2, 8, 0, s_SINT); +case VK_FORMAT_R64G64_SFLOAT: return createDFDUnpacked(0, 2, 8, 0, s_SFLOAT); +case VK_FORMAT_R64G64B64_UINT: return createDFDUnpacked(0, 3, 8, 0, s_UINT); +case VK_FORMAT_R64G64B64_SINT: return createDFDUnpacked(0, 3, 8, 0, s_SINT); +case VK_FORMAT_R64G64B64_SFLOAT: return createDFDUnpacked(0, 3, 8, 0, s_SFLOAT); +case VK_FORMAT_R64G64B64A64_UINT: return createDFDUnpacked(0, 4, 8, 0, s_UINT); +case VK_FORMAT_R64G64B64A64_SINT: return createDFDUnpacked(0, 4, 8, 0, s_SINT); +case VK_FORMAT_R64G64B64A64_SFLOAT: return createDFDUnpacked(0, 4, 8, 0, s_SFLOAT); +case VK_FORMAT_B10G11R11_UFLOAT_PACK32: { + int channels[] = {0,1,2}; int bits[] = {11,11,10}; + return createDFDPacked(0, 3, bits, channels, s_UFLOAT); +} +case VK_FORMAT_E5B9G9R9_UFLOAT_PACK32: { + int bits[] = {0}; int channels[] = {0}; + return createDFDPacked(0, 6, bits, channels, s_UFLOAT); +} +case VK_FORMAT_D16_UNORM: return createDFDDepthStencil(16,0,2); +case VK_FORMAT_X8_D24_UNORM_PACK32: return createDFDDepthStencil(24,0,4); +case VK_FORMAT_D32_SFLOAT: return createDFDDepthStencil(32,0,4); +case VK_FORMAT_S8_UINT: return createDFDDepthStencil(0,8,1); +case VK_FORMAT_BC1_RGB_UNORM_BLOCK: return createDFDCompressed(c_BC1_RGB, 4, 4, 1, s_UNORM); +case VK_FORMAT_BC1_RGB_SRGB_BLOCK: return createDFDCompressed(c_BC1_RGB, 4, 4, 1, s_SRGB); +case VK_FORMAT_BC1_RGBA_UNORM_BLOCK: return createDFDCompressed(c_BC1_RGBA, 4, 4, 1, s_UNORM); +case VK_FORMAT_BC1_RGBA_SRGB_BLOCK: return createDFDCompressed(c_BC1_RGBA, 4, 4, 1, s_SRGB); +case VK_FORMAT_BC2_UNORM_BLOCK: return createDFDCompressed(c_BC2, 4, 4, 1, s_UNORM); +case VK_FORMAT_BC2_SRGB_BLOCK: return createDFDCompressed(c_BC2, 4, 4, 1, s_SRGB); +case VK_FORMAT_BC3_UNORM_BLOCK: return createDFDCompressed(c_BC3, 4, 4, 1, s_UNORM); +case VK_FORMAT_BC3_SRGB_BLOCK: return createDFDCompressed(c_BC3, 4, 4, 1, s_SRGB); +case VK_FORMAT_BC4_UNORM_BLOCK: return createDFDCompressed(c_BC4, 4, 4, 1, s_UNORM); +case VK_FORMAT_BC4_SNORM_BLOCK: return createDFDCompressed(c_BC4, 4, 4, 1, s_SNORM); +case VK_FORMAT_BC5_UNORM_BLOCK: return createDFDCompressed(c_BC5, 4, 4, 1, s_UNORM); +case VK_FORMAT_BC5_SNORM_BLOCK: return createDFDCompressed(c_BC5, 4, 4, 1, s_SNORM); +case VK_FORMAT_BC6H_UFLOAT_BLOCK: return createDFDCompressed(c_BC6H, 4, 4, 1, s_UFLOAT); +case VK_FORMAT_BC6H_SFLOAT_BLOCK: return createDFDCompressed(c_BC6H, 4, 4, 1, s_SFLOAT); +case VK_FORMAT_BC7_UNORM_BLOCK: return createDFDCompressed(c_BC7, 4, 4, 1, s_UNORM); +case VK_FORMAT_BC7_SRGB_BLOCK: return createDFDCompressed(c_BC7, 4, 4, 1, s_SRGB); +case VK_FORMAT_ETC2_R8G8B8_UNORM_BLOCK: return createDFDCompressed(c_ETC2_R8G8B8, 4, 4, 1, s_UNORM); +case VK_FORMAT_ETC2_R8G8B8_SRGB_BLOCK: return createDFDCompressed(c_ETC2_R8G8B8, 4, 4, 1, s_SRGB); +case VK_FORMAT_ETC2_R8G8B8A1_UNORM_BLOCK: return createDFDCompressed(c_ETC2_R8G8B8A1, 4, 4, 1, s_UNORM); +case VK_FORMAT_ETC2_R8G8B8A1_SRGB_BLOCK: return createDFDCompressed(c_ETC2_R8G8B8A1, 4, 4, 1, s_SRGB); +case VK_FORMAT_ETC2_R8G8B8A8_UNORM_BLOCK: return createDFDCompressed(c_ETC2_R8G8B8A8, 4, 4, 1, s_UNORM); +case VK_FORMAT_ETC2_R8G8B8A8_SRGB_BLOCK: return createDFDCompressed(c_ETC2_R8G8B8A8, 4, 4, 1, s_SRGB); +case VK_FORMAT_EAC_R11_UNORM_BLOCK: return createDFDCompressed(c_EAC_R11, 4, 4, 1, s_UNORM); +case VK_FORMAT_EAC_R11_SNORM_BLOCK: return createDFDCompressed(c_EAC_R11, 4, 4, 1, s_SNORM); +case VK_FORMAT_EAC_R11G11_UNORM_BLOCK: return createDFDCompressed(c_EAC_R11G11, 4, 4, 1, s_UNORM); +case VK_FORMAT_EAC_R11G11_SNORM_BLOCK: return createDFDCompressed(c_EAC_R11G11, 4, 4, 1, s_SNORM); +case VK_FORMAT_ASTC_4x4_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 4, 4, 1, s_UNORM); +case VK_FORMAT_ASTC_4x4_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 4, 4, 1, s_SRGB); +case VK_FORMAT_ASTC_5x4_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 5, 4, 1, s_UNORM); +case VK_FORMAT_ASTC_5x4_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 5, 4, 1, s_SRGB); +case VK_FORMAT_ASTC_5x5_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 5, 5, 1, s_UNORM); +case VK_FORMAT_ASTC_5x5_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 5, 5, 1, s_SRGB); +case VK_FORMAT_ASTC_6x5_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 6, 5, 1, s_UNORM); +case VK_FORMAT_ASTC_6x5_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 6, 5, 1, s_SRGB); +case VK_FORMAT_ASTC_6x6_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 6, 6, 1, s_UNORM); +case VK_FORMAT_ASTC_6x6_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 6, 6, 1, s_SRGB); +case VK_FORMAT_ASTC_8x5_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 8, 5, 1, s_UNORM); +case VK_FORMAT_ASTC_8x5_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 8, 5, 1, s_SRGB); +case VK_FORMAT_ASTC_8x6_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 8, 6, 1, s_UNORM); +case VK_FORMAT_ASTC_8x6_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 8, 6, 1, s_SRGB); +case VK_FORMAT_ASTC_8x8_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 8, 8, 1, s_UNORM); +case VK_FORMAT_ASTC_8x8_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 8, 8, 1, s_SRGB); +case VK_FORMAT_ASTC_10x5_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 10, 5, 1, s_UNORM); +case VK_FORMAT_ASTC_10x5_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 10, 5, 1, s_SRGB); +case VK_FORMAT_ASTC_10x6_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 10, 6, 1, s_UNORM); +case VK_FORMAT_ASTC_10x6_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 10, 6, 1, s_SRGB); +case VK_FORMAT_ASTC_10x8_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 10, 8, 1, s_UNORM); +case VK_FORMAT_ASTC_10x8_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 10, 8, 1, s_SRGB); +case VK_FORMAT_ASTC_10x10_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 10, 10, 1, s_UNORM); +case VK_FORMAT_ASTC_10x10_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 10, 10, 1, s_SRGB); +case VK_FORMAT_ASTC_12x10_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 12, 10, 1, s_UNORM); +case VK_FORMAT_ASTC_12x10_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 12, 10, 1, s_SRGB); +case VK_FORMAT_ASTC_12x12_UNORM_BLOCK: return createDFDCompressed(c_ASTC, 12, 12, 1, s_UNORM); +case VK_FORMAT_ASTC_12x12_SRGB_BLOCK: return createDFDCompressed(c_ASTC, 12, 12, 1, s_SRGB); +case VK_FORMAT_PVRTC1_2BPP_UNORM_BLOCK_IMG: return createDFDCompressed(c_PVRTC, 8, 4, 1, s_UNORM); +case VK_FORMAT_PVRTC1_4BPP_UNORM_BLOCK_IMG: return createDFDCompressed(c_PVRTC, 4, 4, 1, s_UNORM); +case VK_FORMAT_PVRTC2_2BPP_UNORM_BLOCK_IMG: return createDFDCompressed(c_PVRTC2, 8, 4, 1, s_UNORM); +case VK_FORMAT_PVRTC2_4BPP_UNORM_BLOCK_IMG: return createDFDCompressed(c_PVRTC2, 4, 4, 1, s_UNORM); +case VK_FORMAT_PVRTC1_2BPP_SRGB_BLOCK_IMG: return createDFDCompressed(c_PVRTC, 8, 4, 1, s_SRGB); +case VK_FORMAT_PVRTC1_4BPP_SRGB_BLOCK_IMG: return createDFDCompressed(c_PVRTC, 4, 4, 1, s_SRGB); +case VK_FORMAT_PVRTC2_2BPP_SRGB_BLOCK_IMG: return createDFDCompressed(c_PVRTC2, 8, 4, 1, s_SRGB); +case VK_FORMAT_PVRTC2_4BPP_SRGB_BLOCK_IMG: return createDFDCompressed(c_PVRTC2, 4, 4, 1, s_SRGB); +case VK_FORMAT_A4R4G4B4_UNORM_PACK16_EXT: { + int channels[] = {2,1,0,3}; int bits[] = {4,4,4,4}; + return createDFDPacked(0, 4, bits, channels, s_UNORM); +} +case VK_FORMAT_A4B4G4R4_UNORM_PACK16_EXT: { + int channels[] = {0,1,2,3}; int bits[] = {4,4,4,4}; + return createDFDPacked(0, 4, bits, channels, s_UNORM); +} diff --git a/build/.keep b/build/.keep new file mode 100644 index 0000000..e69de29 diff --git a/meson.build b/meson.build new file mode 100644 index 0000000..b295e4a --- /dev/null +++ b/meson.build @@ -0,0 +1,3 @@ +project('TextureTaffy', 'cpp') + +subdir('Source') \ No newline at end of file