shadps4-emu/ext-SDL
1
0
Fork 0
mirror of https://github.com/shadps4-emu/ext-SDL.git synced 2025-01-07 10:50:21 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Added support for HDR10 video on Apple platforms

Browse Source
This commit is contained in:
Sam Lantinga 2024-02-06 15:25:05 -08:00
parent 2039c46d2c
commit 8afba41aef
9 changed files with 12433 additions and 10240 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
src/render/metal/SDL_render_metal.m
View File

@ -76,7 +76,9 @@ static const size_t CONSTANTS_OFFSET_DECODE_BT601_LIMITED = ALIGN_CONSTANTS(16,
static const size_t CONSTANTS_OFFSET_DECODE_BT601_FULL = ALIGN_CONSTANTS(16, CONSTANTS_OFFSET_DECODE_BT601_LIMITED + sizeof(float) * 4 * 4);
static const size_t CONSTANTS_OFFSET_DECODE_BT709_LIMITED = ALIGN_CONSTANTS(16, CONSTANTS_OFFSET_DECODE_BT601_FULL + sizeof(float) * 4 * 4);
static const size_t CONSTANTS_OFFSET_DECODE_BT709_FULL = ALIGN_CONSTANTS(16, CONSTANTS_OFFSET_DECODE_BT709_LIMITED + sizeof(float) * 4 * 4);
static const size_t CONSTANTS_LENGTH = CONSTANTS_OFFSET_DECODE_BT709_FULL + sizeof(float) * 4 * 4;
static const size_t CONSTANTS_OFFSET_DECODE_BT2020_LIMITED = ALIGN_CONSTANTS(16, CONSTANTS_OFFSET_DECODE_BT709_FULL + sizeof(float) * 4 * 4);
static const size_t CONSTANTS_OFFSET_DECODE_BT2020_FULL = ALIGN_CONSTANTS(16, CONSTANTS_OFFSET_DECODE_BT2020_LIMITED + sizeof(float) * 4 * 4);
static const size_t CONSTANTS_LENGTH = CONSTANTS_OFFSET_DECODE_BT2020_FULL + sizeof(float) * 4 * 4;
typedef enum SDL_MetalVertexFunction
{
@ -91,6 +93,7 @@ typedef enum SDL_MetalFragmentFunction
SDL_METAL_FRAGMENT_YUV,
SDL_METAL_FRAGMENT_NV12,
SDL_METAL_FRAGMENT_NV21,
SDL_METAL_FRAGMENT_HDR10,
SDL_METAL_FRAGMENT_COUNT,
} SDL_MetalFragmentFunction;
@ -248,6 +251,8 @@ static NSString *GetFragmentFunctionName(SDL_MetalFragmentFunction function)
return @"SDL_NV12_fragment";
case SDL_METAL_FRAGMENT_NV21:
return @"SDL_NV21_fragment";
case SDL_METAL_FRAGMENT_HDR10:
return @"SDL_HDR10_fragment";
default:
return nil;
}
@ -387,6 +392,7 @@ void MakeShaderPipelines(METAL_RenderData *data, METAL_ShaderPipelines *pipeline
MakePipelineCache(data, &pipelines->caches[SDL_METAL_FRAGMENT_YUV], "SDL YUV pipeline", rtformat, SDL_METAL_VERTEX_COPY, SDL_METAL_FRAGMENT_YUV);
MakePipelineCache(data, &pipelines->caches[SDL_METAL_FRAGMENT_NV12], "SDL NV12 pipeline", rtformat, SDL_METAL_VERTEX_COPY, SDL_METAL_FRAGMENT_NV12);
MakePipelineCache(data, &pipelines->caches[SDL_METAL_FRAGMENT_NV21], "SDL NV21 pipeline", rtformat, SDL_METAL_VERTEX_COPY, SDL_METAL_FRAGMENT_NV21);
MakePipelineCache(data, &pipelines->caches[SDL_METAL_FRAGMENT_HDR10], "SDL HDR10 pipeline", rtformat, SDL_METAL_VERTEX_COPY, SDL_METAL_FRAGMENT_HDR10);
}
static METAL_ShaderPipelines *ChooseShaderPipelines(METAL_RenderData *data, MTLPixelFormat rtformat)
@ -551,10 +557,8 @@ size_t GetBT601ConversionMatrix( SDL_Colorspace colorspace )
case SDL_COLOR_RANGE_LIMITED:
case SDL_COLOR_RANGE_UNKNOWN:
return CONSTANTS_OFFSET_DECODE_BT601_LIMITED;
break;
case SDL_COLOR_RANGE_FULL:
return CONSTANTS_OFFSET_DECODE_BT601_FULL;
break;
default:
break;
}
@ -567,10 +571,8 @@ size_t GetBT709ConversionMatrix(SDL_Colorspace colorspace)
case SDL_COLOR_RANGE_LIMITED:
case SDL_COLOR_RANGE_UNKNOWN:
return CONSTANTS_OFFSET_DECODE_BT709_LIMITED;
break;
case SDL_COLOR_RANGE_FULL:
return CONSTANTS_OFFSET_DECODE_BT709_FULL;
break;
default:
break;
}
@ -582,11 +584,9 @@ size_t GetBT2020ConversionMatrix(SDL_Colorspace colorspace)
switch (SDL_COLORSPACERANGE(colorspace)) {
case SDL_COLOR_RANGE_LIMITED:
case SDL_COLOR_RANGE_UNKNOWN:
return 0;
break;
return CONSTANTS_OFFSET_DECODE_BT2020_LIMITED;
case SDL_COLOR_RANGE_FULL:
return 0;
break;
return CONSTANTS_OFFSET_DECODE_BT2020_FULL;
default:
break;
}
@ -671,6 +671,9 @@ static int METAL_CreateTexture(SDL_Renderer *renderer, SDL_Texture *texture, SDL
case SDL_PIXELFORMAT_NV21:
pixfmt = MTLPixelFormatR8Unorm;
break;
case SDL_PIXELFORMAT_P010:
pixfmt = MTLPixelFormatR16Unorm;
break;
case SDL_PIXELFORMAT_RGBA64_FLOAT:
pixfmt = MTLPixelFormatRGBA16Float;
break;
@ -706,8 +709,8 @@ static int METAL_CreateTexture(SDL_Renderer *renderer, SDL_Texture *texture, SDL
mtltextureUv = nil;
#if SDL_HAVE_YUV
yuv = (texture->format == SDL_PIXELFORMAT_IYUV) || (texture->format == SDL_PIXELFORMAT_YV12);
nv12 = (texture->format == SDL_PIXELFORMAT_NV12) || (texture->format == SDL_PIXELFORMAT_NV21);
yuv = (texture->format == SDL_PIXELFORMAT_IYUV || texture->format == SDL_PIXELFORMAT_YV12);
nv12 = (texture->format == SDL_PIXELFORMAT_NV12 || texture->format == SDL_PIXELFORMAT_NV21 || texture->format == SDL_PIXELFORMAT_P010);
if (yuv) {
mtltexdesc.pixelFormat = MTLPixelFormatR8Unorm;
@ -715,6 +718,10 @@ static int METAL_CreateTexture(SDL_Renderer *renderer, SDL_Texture *texture, SDL
mtltexdesc.height = (texture->h + 1) / 2;
mtltexdesc.textureType = MTLTextureType2DArray;
mtltexdesc.arrayLength = 2;
} else if (texture->format == SDL_PIXELFORMAT_P010) {
mtltexdesc.pixelFormat = MTLPixelFormatRG16Unorm;
mtltexdesc.width = (texture->w + 1) / 2;
mtltexdesc.height = (texture->h + 1) / 2;
} else if (nv12) {
mtltexdesc.pixelFormat = MTLPixelFormatRG8Unorm;
mtltexdesc.width = (texture->w + 1) / 2;
@ -750,6 +757,8 @@ static int METAL_CreateTexture(SDL_Renderer *renderer, SDL_Texture *texture, SDL
texturedata.fragmentFunction = SDL_METAL_FRAGMENT_NV12;
} else if (texture->format == SDL_PIXELFORMAT_NV21) {
texturedata.fragmentFunction = SDL_METAL_FRAGMENT_NV21;
} else if (texture->format == SDL_PIXELFORMAT_P010) {
texturedata.fragmentFunction = SDL_METAL_FRAGMENT_HDR10;
} else
#endif
{
@ -2011,6 +2020,8 @@ static SDL_Renderer *METAL_CreateRenderer(SDL_Window *window, SDL_PropertiesID c
SDL_memcpy(constantdata + CONSTANTS_OFFSET_DECODE_BT601_FULL, SDL_GetYCbCRtoRGBConversionMatrix(SDL_COLORSPACE_BT601_FULL, 0, 0, 8), YCbCr_shader_matrix_size);
SDL_memcpy(constantdata + CONSTANTS_OFFSET_DECODE_BT709_LIMITED, SDL_GetYCbCRtoRGBConversionMatrix(SDL_COLORSPACE_BT709_LIMITED, 0, 0, 8), YCbCr_shader_matrix_size);
SDL_memcpy(constantdata + CONSTANTS_OFFSET_DECODE_BT709_FULL, SDL_GetYCbCRtoRGBConversionMatrix(SDL_COLORSPACE_BT709_FULL, 0, 0, 8), YCbCr_shader_matrix_size);
SDL_memcpy(constantdata + CONSTANTS_OFFSET_DECODE_BT2020_LIMITED, SDL_GetYCbCRtoRGBConversionMatrix(SDL_COLORSPACE_BT2020_LIMITED, 0, 0, 10), YCbCr_shader_matrix_size);
SDL_memcpy(constantdata + CONSTANTS_OFFSET_DECODE_BT2020_FULL, SDL_GetYCbCRtoRGBConversionMatrix(SDL_COLORSPACE_BT2020_FULL, 0, 0, 10), YCbCr_shader_matrix_size);
mtlbufquadindicesstaging = [data.mtldevice newBufferWithLength:indicessize options:MTLResourceStorageModeShared];
@ -2139,7 +2150,7 @@ SDL_RenderDriver METAL_RenderDriver = {
{
"metal",
(SDL_RENDERER_ACCELERATED | SDL_RENDERER_PRESENTVSYNC),
8,
9,
{ SDL_PIXELFORMAT_ARGB8888,
SDL_PIXELFORMAT_ABGR8888,
SDL_PIXELFORMAT_RGBA64_FLOAT,
@ -2147,7 +2158,8 @@ SDL_RenderDriver METAL_RenderDriver = {
SDL_PIXELFORMAT_YV12,
SDL_PIXELFORMAT_IYUV,
SDL_PIXELFORMAT_NV12,
SDL_PIXELFORMAT_NV21 },
SDL_PIXELFORMAT_NV21,
SDL_PIXELFORMAT_P010 },
0,
0,
}

45
src/render/metal/SDL_shaders_metal.metal
View File

@ -34,6 +34,19 @@ float sRGBfromLinear(float v)
return v;
}
float3 PQtoNits(float3 v)
{
const float c1 = 0.8359375;
const float c2 = 18.8515625;
const float c3 = 18.6875;
const float oo_m1 = 1.0 / 0.1593017578125;
const float oo_m2 = 1.0 / 78.84375;
float3 num = max(pow(abs(v), oo_m2) - c1, 0.0);
float3 den = c2 - c3 * pow(abs(v), oo_m2);
return 10000.0 * pow(abs(num / den), oo_m1);
}
float4 GetOutputColor(float4 rgba, float color_scale)
{
float4 output;
@ -215,3 +228,35 @@ fragment float4 SDL_NV21_fragment(CopyVertexOutput vert [[stage_in]],
return GetOutputColorFromSRGB(rgb, c.scRGB_output, c.color_scale) * vert.color;
}
fragment float4 SDL_HDR10_fragment(CopyVertexOutput vert [[stage_in]],
constant ShaderConstants &c [[buffer(0)]],
constant YUVDecode &decode [[buffer(1)]],
texture2d<float> texY [[texture(0)]],
texture2d<float> texUV [[texture(1)]],
sampler s [[sampler(0)]])
{
const float3x3 mat2020to709 = {
{ 1.660496, -0.587656, -0.072840 },
{ -0.124547, 1.132895, -0.008348 },
{ -0.018154, -0.100597, 1.118751 }
};
float3 yuv;
yuv.x = texY.sample(s, vert.texcoord).r;
yuv.yz = texUV.sample(s, vert.texcoord).rg;
float3 rgb;
yuv += decode.offset;
rgb.r = dot(yuv, decode.Rcoeff);
rgb.g = dot(yuv, decode.Gcoeff);
rgb.b = dot(yuv, decode.Bcoeff);
rgb = PQtoNits(rgb);
rgb = rgb * mat2020to709;
rgb = scRGBfromNits(rgb);
return GetOutputColorFromSCRGB(rgb, c.scRGB_output, c.color_scale) * vert.color;
}

4708
src/render/metal/SDL_shaders_metal_ios.h
View File

File diff suppressed because it is too large Load Diff

4808
src/render/metal/SDL_shaders_metal_iphonesimulator.h
View File

File diff suppressed because it is too large Load Diff

4508
src/render/metal/SDL_shaders_metal_macos.h
View File

File diff suppressed because it is too large Load Diff

4708
src/render/metal/SDL_shaders_metal_tvos.h
View File

File diff suppressed because it is too large Load Diff

3847
src/render/metal/SDL_shaders_metal_tvsimulator.h
View File

File diff suppressed because it is too large Load Diff

6
src/video/SDL_pixels.c
View File

@ -815,10 +815,8 @@ static const float *SDL_GetBT601ConversionMatrix( SDL_Colorspace colorspace )
case SDL_COLOR_RANGE_LIMITED:
case SDL_COLOR_RANGE_UNKNOWN:
return mat_BT601_Limited_8bit;
break;
case SDL_COLOR_RANGE_FULL:
return mat_BT601_Full_8bit;
break;
default:
break;
}
@ -831,10 +829,8 @@ static const float *SDL_GetBT709ConversionMatrix(SDL_Colorspace colorspace)
case SDL_COLOR_RANGE_LIMITED:
case SDL_COLOR_RANGE_UNKNOWN:
return mat_BT709_Limited_8bit;
break;
case SDL_COLOR_RANGE_FULL:
return mat_BT709_Full_8bit;
break;
default:
break;
}
@ -847,10 +843,8 @@ static const float *SDL_GetBT2020ConversionMatrix(SDL_Colorspace colorspace)
case SDL_COLOR_RANGE_LIMITED:
case SDL_COLOR_RANGE_UNKNOWN:
return mat_BT2020_Limited_10bit;
break;
case SDL_COLOR_RANGE_FULL:
return mat_BT2020_Full_10bit;
break;
default:
break;
}

5
test/testffmpeg.c
View File

@ -759,6 +759,11 @@ static SDL_bool GetTextureForVideoToolboxFrame(AVFrame *frame, SDL_Texture **tex
case kCVPixelFormatType_420YpCbCr8BiPlanarFullRange:
format = SDL_PIXELFORMAT_NV12;
break;
case kCVPixelFormatType_420YpCbCr10BiPlanarVideoRange:
case kCVPixelFormatType_420YpCbCr10BiPlanarFullRange:
format = SDL_PIXELFORMAT_P010;
HDR_video = SDL_TRUE;
break;
default:
SDL_SetError("Unsupported texture format %c%c%c%c",
(char)((nPixelBufferType >> 24) & 0xFF),