mirror of
https://github.com/gnif/LookingGlass.git
synced 2026-11-12 14:57:20 +00:00
7f63dc986e
The dequant shader returned before workgroup barriers for missing blocks. Mask those blocks to zero so ANGLE's D3D compiler accepts the shader while every invocation reaches each barrier. The vcpkg ANGLE build also omitted upstream shared-context locking. Enable both locking definitions so decode dispatches cannot race the render thread through ANGLE's shared D3D11 state manager.
342 lines
15 KiB
C
342 lines
15 KiB
C
/**
|
|
* Looking Glass
|
|
* Copyright © 2017-2026 The Looking Glass Authors
|
|
* https://looking-glass.io
|
|
*
|
|
* This program is free software; you can redistribute it and/or modify it
|
|
* under the terms of the GNU General Public License as published by the Free
|
|
* Software Foundation; either version 2 of the License, or (at your option)
|
|
* any later version.
|
|
*
|
|
* This program is distributed in the hope that it will be useful, but WITHOUT
|
|
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
|
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
|
* more details.
|
|
*
|
|
* You should have received a copy of the GNU General Public License along
|
|
* with this program; if not, write to the Free Software Foundation, Inc., 59
|
|
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
*/
|
|
|
|
/*
|
|
* The coefficient and inverse-wavelet shaders are OpenGL ES adaptations of
|
|
* PyroWave's Vulkan shaders at commit
|
|
* 89f7e47d4abbf650c91fae766728af866c5e32a0.
|
|
*
|
|
* Copyright (c) 2025 Hans-Kristian Arntzen
|
|
*
|
|
* 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.
|
|
*/
|
|
|
|
#pragma once
|
|
|
|
static const char NETWORK_VIDEO_DEQUANT_SHADER_0[] =
|
|
"#version 310 es\n"
|
|
"precision highp float;\n"
|
|
"precision highp int;\n"
|
|
"layout(local_size_x=128) in;\n"
|
|
"layout(r32f,binding=0) writeonly uniform highp image2DArray outputImage;\n"
|
|
"layout(std430,binding=0) readonly buffer OffsetBuffer { uint offsets[]; };\n"
|
|
"layout(std430,binding=1) readonly buffer PayloadBuffer { uint payload[]; };\n"
|
|
"uniform ivec2 resolution;\n"
|
|
"uniform int outputLayer;\n"
|
|
"uniform int blockOffset;\n"
|
|
"uniform int blockStride;\n"
|
|
"shared uint signOffset;\n"
|
|
"shared uint planeOffset[16];\n"
|
|
"shared uint significantPrefix[128];\n"
|
|
"uint readU8(uint byteOffset) {\n"
|
|
" return (payload[byteOffset >> 2u] >> (8u * (byteOffset & 3u))) & 255u;\n"
|
|
"}\n"
|
|
"uint readU16(uint byteOffset) {\n"
|
|
" return readU8(byteOffset) | (readU8(byteOffset + 1u) << 8u);\n"
|
|
"}\n"
|
|
"ivec2 unswizzle8x8(uint index) {\n"
|
|
" uint y = bitfieldExtract(index, 0, 1);\n"
|
|
" uint x = bitfieldExtract(index, 1, 2);\n"
|
|
" y |= bitfieldExtract(index, 3, 2) << 1;\n"
|
|
" x |= bitfieldExtract(index, 5, 1) << 2;\n"
|
|
" return ivec2(x, y);\n"
|
|
"}\n"
|
|
"float quantScale(uint code) { return float(code) / 8.0 + 0.25; }\n"
|
|
"float blockScale(uint code) {\n"
|
|
" int exponent = 4 - int(code >> 3u);\n"
|
|
" return (float(8u + (code & 7u)) / 8.0) * exp2(float(exponent));\n"
|
|
"}\n"
|
|
"void storeValues(ivec2 coord, mat2x4 values) {\n"
|
|
" for (int y=0; y<2; ++y) for (int x=0; x<4; ++x) {\n"
|
|
" ivec2 target = coord + ivec2(x,y);\n"
|
|
" if (all(lessThan(target,resolution)))\n"
|
|
" imageStore(outputImage,ivec3(target,outputLayer),vec4(values[y][x]));\n"
|
|
" }\n"
|
|
"}\n"
|
|
"void main() {\n"
|
|
" uint localIndex = gl_LocalInvocationIndex;\n"
|
|
" int blockIndex = blockOffset + int(gl_WorkGroupID.y) * blockStride +\n"
|
|
" int(gl_WorkGroupID.x);\n"
|
|
" uint sourceWord = offsets[blockIndex];\n"
|
|
" uint blockPresent = uint(sourceWord != 0xffffffffu);\n"
|
|
" // Missing blocks must still reach every workgroup barrier.\n"
|
|
" sourceWord *= blockPresent;\n"
|
|
" uint subblock = bitfieldExtract(localIndex,0,3);\n"
|
|
" uint blockX = bitfieldExtract(localIndex,3,2);\n"
|
|
" uint blockY = bitfieldExtract(localIndex,5,2);\n"
|
|
" uint group = blockY * 4u + blockX;\n"
|
|
" ivec2 coord = ivec2(gl_WorkGroupID.xy) * 32 +\n"
|
|
" 8 * ivec2(blockX,blockY) + unswizzle8x8(subblock << 3u);\n"
|
|
" uint sourceByte = sourceWord * 4u;\n"
|
|
" uint ballot = (payload[sourceWord] & 65535u) * blockPresent;\n"
|
|
" if (localIndex == 0u) {\n"
|
|
" uint cursor = sourceByte + 8u + 3u * uint(bitCount(ballot));\n"
|
|
" for (uint index=0u; index<16u; ++index) {\n"
|
|
" planeOffset[index] = cursor;\n"
|
|
" if (bitfieldExtract(ballot,int(index),1) != 0u) {\n"
|
|
" uint compact = uint(bitCount(bitfieldExtract(ballot,0,int(index))));\n"
|
|
" uint control = readU16(sourceByte + 8u + 2u * compact);\n"
|
|
" uint qbits = readU8(sourceByte + 8u +\n"
|
|
" 2u * uint(bitCount(ballot)) + compact) & 15u;\n"
|
|
" uint low = control & 0x5555u;\n"
|
|
" uint high = control & 0xaaaau;\n"
|
|
" high |= high >> 1u;\n"
|
|
" cursor += uint(bitCount(low) + bitCount(high)) + 8u * qbits;\n"
|
|
" }\n"
|
|
" }\n"
|
|
" signOffset = cursor * 8u * blockPresent;\n"
|
|
" }\n"
|
|
" barrier();\n";
|
|
|
|
static const char NETWORK_VIDEO_DEQUANT_SHADER_1[] =
|
|
" mat2x4 values = mat2x4(vec4(0.0),vec4(0.0));\n"
|
|
" uint significant = 0u;\n"
|
|
" if (bitfieldExtract(ballot,int(group),1) != 0u) {\n"
|
|
" uint compact = uint(bitCount(bitfieldExtract(ballot,0,int(group))));\n"
|
|
" uint control = readU16(sourceByte + 8u + 2u * compact);\n"
|
|
" uint scaleCode = readU8(sourceByte + 8u +\n"
|
|
" 2u * uint(bitCount(ballot)) + compact);\n"
|
|
" uint planes = (scaleCode & 15u) +\n"
|
|
" bitfieldExtract(control,int(2u * subblock),2);\n"
|
|
" int magnitude[8];\n"
|
|
" for (int i=0; i<8; ++i) magnitude[i]=0;\n"
|
|
" uint priorBits = 2u * subblock;\n"
|
|
" uint priorMask = (1u << priorBits) - 1u;\n"
|
|
" uint low = control & 0x5555u & priorMask;\n"
|
|
" uint high = control & 0xaaaau & priorMask;\n"
|
|
" high |= high >> 1u;\n"
|
|
" uint cursor = planeOffset[group] + uint(bitCount(low) +\n"
|
|
" bitCount(high)) + (scaleCode & 15u) * subblock;\n"
|
|
" for (int plane=int(planes)-1; plane>=0; --plane) {\n"
|
|
" uint bits = readU8(cursor++);\n"
|
|
" for (int i=0; i<8; ++i)\n"
|
|
" magnitude[i] = (magnitude[i] << 1) | int((bits >> uint(i)) & 1u);\n"
|
|
" }\n"
|
|
" float scale = blockScale(payload[sourceWord+1u] & 255u) *\n"
|
|
" quantScale(scaleCode >> 4u);\n"
|
|
" for (int x=0; x<4; ++x) for (int y=0; y<2; ++y) {\n"
|
|
" int index = x * 2 + y;\n"
|
|
" float value = float(magnitude[index]);\n"
|
|
" if (value != 0.0) { value += 0.5; ++significant; }\n"
|
|
" values[y][x] = value * scale;\n"
|
|
" }\n"
|
|
" }\n"
|
|
" significantPrefix[localIndex] = significant;\n"
|
|
" barrier();\n"
|
|
" if (localIndex == 0u) {\n"
|
|
" uint prefix = 0u;\n"
|
|
" for (uint index=0u; index<128u; ++index) {\n"
|
|
" uint count = significantPrefix[index];\n"
|
|
" significantPrefix[index] = prefix;\n"
|
|
" prefix += count;\n"
|
|
" }\n"
|
|
" }\n"
|
|
" barrier();\n"
|
|
" uint bitCursor = signOffset + significantPrefix[localIndex];\n"
|
|
" uint word = payload[bitCursor >> 5u];\n"
|
|
" uint upper = payload[(bitCursor >> 5u) + 1u];\n"
|
|
" uint shift = bitCursor & 31u;\n"
|
|
" if (shift != 0u) word = (word >> shift) | (upper << (32u-shift));\n"
|
|
" uint signIndex = 0u;\n"
|
|
" for (int x=0; x<4; ++x) for (int y=0; y<2; ++y) {\n"
|
|
" if (values[y][x] != 0.0) {\n"
|
|
" if (((word >> signIndex) & 1u) != 0u) values[y][x] = -values[y][x];\n"
|
|
" ++signIndex;\n"
|
|
" }\n"
|
|
" }\n"
|
|
" storeValues(coord,values);\n"
|
|
"}\n";
|
|
|
|
static const char NETWORK_VIDEO_IDWT_SHADER_0[] =
|
|
"#version 310 es\n"
|
|
"precision highp float;\n"
|
|
"precision highp int;\n"
|
|
"layout(local_size_x=64) in;\n"
|
|
"layout(binding=0) uniform highp sampler2DArray coefficients;\n"
|
|
"layout(r32f,binding=1) writeonly uniform highp image2D outputImage;\n"
|
|
"uniform ivec2 resolution;\n"
|
|
"uniform vec2 invResolution;\n"
|
|
"uniform int baseLayer;\n"
|
|
"uniform bool dcShift;\n"
|
|
"const int APRON=4, HALF_APRON=2, BLOCK=32, HALF_BLOCK=16;\n"
|
|
"const float ALPHA=-1.586134342059924;\n"
|
|
"const float BETA=-0.052980118572961;\n"
|
|
"const float GAMMA=0.882911075530934;\n"
|
|
"const float DELTA=0.443506852043971;\n"
|
|
"const float K=1.230174104914001;\n"
|
|
"const float INV_K=1.0/1.230174104914001;\n"
|
|
"shared vec2 sharedBlock[20][41];\n"
|
|
"ivec2 unswizzle8x8(uint index) {\n"
|
|
" uint y=bitfieldExtract(index,0,1), x=bitfieldExtract(index,1,2);\n"
|
|
" y|=bitfieldExtract(index,3,2)<<1; x|=bitfieldExtract(index,5,1)<<2;\n"
|
|
" return ivec2(x,y);\n"
|
|
"}\n"
|
|
"bvec2 and2(bvec2 a,bvec2 b) { return bvec2(a.x&&b.x,a.y&&b.y); }\n"
|
|
"vec2 mirrorUV(ivec2 coord,bool evenX,bool evenY) {\n"
|
|
" coord-=ivec2(and2(bvec2(evenX,evenY),lessThan(coord,ivec2(0))));\n"
|
|
" coord+=1;\n"
|
|
" coord+=ivec2(and2(bvec2(!evenX,!evenY),greaterThanEqual(coord,resolution)));\n"
|
|
" return (vec2(coord)*invResolution).yx;\n"
|
|
"}\n"
|
|
"vec4 gatherBand(ivec2 coord,bool evenX,bool evenY,int band) {\n"
|
|
" return textureGather(coefficients,vec3(mirrorUV(coord,evenX,evenY),\n"
|
|
" float(baseLayer+band)),0);\n"
|
|
"}\n"
|
|
"void store4(ivec2 coord,vec4 a,vec4 b,vec4 c,vec4 d) {\n"
|
|
" sharedBlock[coord.y][2*coord.x+0]=vec2(a.x,c.x);\n"
|
|
" sharedBlock[coord.y][2*coord.x+1]=vec2(b.x,d.x);\n"
|
|
" sharedBlock[coord.y][2*coord.x+2]=vec2(a.y,c.y);\n"
|
|
" sharedBlock[coord.y][2*coord.x+3]=vec2(b.y,d.y);\n"
|
|
" sharedBlock[coord.y+1][2*coord.x+0]=vec2(a.z,c.z);\n"
|
|
" sharedBlock[coord.y+1][2*coord.x+1]=vec2(b.z,d.z);\n"
|
|
" sharedBlock[coord.y+1][2*coord.x+2]=vec2(a.w,c.w);\n"
|
|
" sharedBlock[coord.y+1][2*coord.x+3]=vec2(b.w,d.w);\n"
|
|
"}\n"
|
|
"void loadApron(uint index) {\n"
|
|
" ivec2 base=ivec2(gl_WorkGroupID.xy)*HALF_BLOCK-HALF_APRON;\n"
|
|
" ivec2 local=2*unswizzle8x8(index); ivec2 coord=base+local;\n"
|
|
" store4(local,gatherBand(coord,true,true,0).wxzy,\n"
|
|
" gatherBand(coord,false,true,2).wxzy,\n"
|
|
" gatherBand(coord,true,false,1).wxzy,\n"
|
|
" gatherBand(coord,false,false,3).wxzy);\n"
|
|
" ivec2 horizontal=ivec2(HALF_BLOCK+2*int(index%2u),2*int(index/2u));\n"
|
|
" if (horizontal.y<HALF_BLOCK+2*HALF_APRON) { coord=base+horizontal;\n"
|
|
" store4(horizontal,gatherBand(coord,true,true,0).wxzy,\n"
|
|
" gatherBand(coord,false,true,2).wxzy,\n"
|
|
" gatherBand(coord,true,false,1).wxzy,\n"
|
|
" gatherBand(coord,false,false,3).wxzy); }\n"
|
|
" ivec2 vertical=horizontal.yx;\n"
|
|
" if (vertical.x<HALF_BLOCK) { coord=base+vertical;\n"
|
|
" store4(vertical,gatherBand(coord,true,true,0).wxzy,\n"
|
|
" gatherBand(coord,false,true,2).wxzy,\n"
|
|
" gatherBand(coord,true,false,1).wxzy,\n"
|
|
" gatherBand(coord,false,false,3).wxzy); }\n"
|
|
" barrier();\n"
|
|
"}\n";
|
|
|
|
static const char NETWORK_VIDEO_IDWT_SHADER_1[] =
|
|
"void inverse8x2(uint index) {\n"
|
|
" vec2 values[16]; ivec2 local=ivec2(8*int(index%4u),int(index/4u));\n"
|
|
" for(int i=0;i<16;i+=2){values[i]=sharedBlock[local.y][local.x+i]*K;\n"
|
|
" values[i+1]=sharedBlock[local.y][local.x+i+1]*INV_K;}\n"
|
|
" for(int i=2;i<15;i+=2)values[i]-=DELTA*(values[i-1]+values[i+1]);\n"
|
|
" for(int i=3;i<14;i+=2)values[i]-=GAMMA*(values[i-1]+values[i+1]);\n"
|
|
" for(int i=4;i<13;i+=2)values[i]-=BETA*(values[i-1]+values[i+1]);\n"
|
|
" for(int i=5;i<12;i+=2)values[i]-=ALPHA*(values[i-1]+values[i+1]);\n"
|
|
" barrier();\n"
|
|
" for(int i=2;i<6;++i){vec2 a=values[2*i],b=values[2*i+1];\n"
|
|
" int y=(local.x>>1)+(i-2); sharedBlock[y][2*local.y]=vec2(a.x,b.x);\n"
|
|
" sharedBlock[y][2*local.y+1]=vec2(a.y,b.y);}\n"
|
|
"}\n"
|
|
"void inverse4x2(uint index,bool activeLane,int yOffset) {\n"
|
|
" vec2 values[12]; ivec2 local=ivec2(4*int(index%8u),int(index/8u)+yOffset);\n"
|
|
" if(activeLane){for(int i=0;i<12;i+=2){values[i]=sharedBlock[local.y][local.x+i]*K;\n"
|
|
" values[i+1]=sharedBlock[local.y][local.x+i+1]*INV_K;}\n"
|
|
" for(int i=2;i<11;i+=2)values[i]-=DELTA*(values[i-1]+values[i+1]);\n"
|
|
" for(int i=3;i<10;i+=2)values[i]-=GAMMA*(values[i-1]+values[i+1]);\n"
|
|
" for(int i=4;i<9;i+=2)values[i]-=BETA*(values[i-1]+values[i+1]);\n"
|
|
" for(int i=5;i<8;i+=2)values[i]-=ALPHA*(values[i-1]+values[i+1]);}\n"
|
|
" barrier();\n"
|
|
" if(activeLane)for(int i=2;i<4;++i){vec2 a=values[2*i],b=values[2*i+1];\n"
|
|
" int y=(local.x>>1)+(i-2); sharedBlock[y][2*local.y]=vec2(a.x,b.x);\n"
|
|
" sharedBlock[y][2*local.y+1]=vec2(a.y,b.y);}\n"
|
|
"}\n"
|
|
"void main(){uint index=gl_LocalInvocationIndex;loadApron(index);\n"
|
|
" inverse8x2(index);inverse4x2(index,index<32u,HALF_BLOCK);barrier();\n"
|
|
" inverse8x2(index);barrier();ivec2 local=unswizzle8x8(index);\n"
|
|
" ivec2 outputSize=imageSize(outputImage);\n"
|
|
" for(int y=local.y;y<HALF_BLOCK;y+=8)for(int x=local.x;x<BLOCK;x+=8){\n"
|
|
" vec2 value=sharedBlock[y][x];if(dcShift)value+=vec2(0.5);\n"
|
|
" ivec2 a=ivec2(2*y,x)+BLOCK*ivec2(gl_WorkGroupID.yx);\n"
|
|
" ivec2 b=a+ivec2(1,0);if(all(lessThan(a,outputSize)))\n"
|
|
" imageStore(outputImage,a,vec4(value.x));\n"
|
|
" if(all(lessThan(b,outputSize)))imageStore(outputImage,b,vec4(value.y));}\n"
|
|
"}\n";
|
|
|
|
static const char NETWORK_VIDEO_YUV_SHADER[] =
|
|
"#version 310 es\n"
|
|
"precision highp float;\n"
|
|
"precision highp int;\n"
|
|
"layout(local_size_x=8,local_size_y=8) in;\n"
|
|
"layout(binding=0) uniform highp sampler2D lumaPlane;\n"
|
|
"layout(binding=1) uniform highp sampler2D cbPlane;\n"
|
|
"layout(binding=2) uniform highp sampler2D crPlane;\n"
|
|
"layout(rgba16f,binding=0) writeonly uniform highp image2D outputImage;\n"
|
|
"uniform ivec2 outputSize;\n"
|
|
"uniform bool bt2020;\n"
|
|
"uniform bool chromaLeft;\n"
|
|
"uniform float lumaOffset;\n"
|
|
"uniform float lumaScale;\n"
|
|
"uniform float chromaMidpoint;\n"
|
|
"uniform float chromaScale;\n"
|
|
"void main(){\n"
|
|
" ivec2 pixel = ivec2(gl_GlobalInvocationID.xy);\n"
|
|
" if(any(greaterThanEqual(pixel,outputSize)))return;\n"
|
|
" ivec2 planeSize = textureSize(lumaPlane,0);\n"
|
|
" vec2 uv = (vec2(pixel)+0.5)/vec2(planeSize);\n"
|
|
" vec2 offset = chromaLeft?\n"
|
|
" vec2(0.5/float(planeSize.x),0.0):vec2(0.0);\n"
|
|
" float y = clamp((texture(lumaPlane,uv).x-lumaOffset)*\n"
|
|
" lumaScale,0.0,1.0);\n"
|
|
" float cb = clamp((texture(cbPlane,uv+offset).x-chromaMidpoint)*\n"
|
|
" chromaScale,-0.5,0.5);\n"
|
|
" float cr = clamp((texture(crPlane,uv+offset).x-chromaMidpoint)*\n"
|
|
" chromaScale,-0.5,0.5);\n"
|
|
" mat3 m = bt2020?\n"
|
|
" mat3(vec3(1.0),vec3(0.0,-0.16455313,1.8814),\n"
|
|
" vec3(1.4746,-0.57135313,0.0)):\n"
|
|
" mat3(vec3(1.0),vec3(0.0,-0.13397432/0.7152,1.8556),\n"
|
|
" vec3(1.5748,-0.33480248/0.7152,0.0));\n"
|
|
" vec3 rgb = clamp(m*vec3(y,cb,cr),vec3(0.0),vec3(1.0));\n"
|
|
" imageStore(outputImage,pixel,vec4(rgb,1.0));}\n";
|
|
|
|
static const char * const NETWORK_VIDEO_DEQUANT_SOURCES[] =
|
|
{
|
|
NETWORK_VIDEO_DEQUANT_SHADER_0,
|
|
NETWORK_VIDEO_DEQUANT_SHADER_1,
|
|
};
|
|
|
|
static const char * const NETWORK_VIDEO_IDWT_SOURCES[] =
|
|
{
|
|
NETWORK_VIDEO_IDWT_SHADER_0,
|
|
NETWORK_VIDEO_IDWT_SHADER_1,
|
|
};
|
|
|
|
static const char * const NETWORK_VIDEO_YUV_SOURCES[] =
|
|
{
|
|
NETWORK_VIDEO_YUV_SHADER,
|
|
};
|