AuroraMiddleware/skia2
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Pack float array uniforms into vec4 arrays

Browse Source 
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1756583002

Review URL: https://codereview.chromium.org/1756583002
This commit is contained in:
jvanverth 2016-03-02 13:21:16 -08:00 committed by Commit bot
parent 03509eafa3
commit 78d6eb0775
2 changed files with 25 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
src/gpu/effects/GrConvolutionEffect.cpp
View File

@ -46,9 +46,12 @@ void GrGLConvolutionEffect::emitCode(EmitArgs& args) {
int width = Gr1DKernelEffect::WidthFromRadius(ce.radius()); int width = Gr1DKernelEffect::WidthFromRadius(ce.radius());
int arrayCount = (width + 3) / 4;
SkASSERT(4 * arrayCount >= width);
fKernelUni = uniformHandler->addUniformArray(kFragment_GrShaderFlag, fKernelUni = uniformHandler->addUniformArray(kFragment_GrShaderFlag,
kFloat_GrSLType, kDefault_GrSLPrecision, kVec4f_GrSLType, kDefault_GrSLPrecision,
"Kernel", width); "Kernel", arrayCount);
GrGLSLFPFragmentBuilder* fragBuilder = args.fFragBuilder; GrGLSLFPFragmentBuilder* fragBuilder = args.fFragBuilder;
SkString coords2D = fragBuilder->ensureFSCoords2D(args.fCoords, 0); SkString coords2D = fragBuilder->ensureFSCoords2D(args.fCoords, 0);
@ -61,11 +64,13 @@ void GrGLConvolutionEffect::emitCode(EmitArgs& args) {
fragBuilder->codeAppendf("vec2 coord = %s - %d.0 * %s;", coords2D.c_str(), ce.radius(), imgInc); fragBuilder->codeAppendf("vec2 coord = %s - %d.0 * %s;", coords2D.c_str(), ce.radius(), imgInc);
// Manually unroll loop because some drivers don't; yields 20-30% speedup. // Manually unroll loop because some drivers don't; yields 20-30% speedup.
const char* kVecSuffix[4] = { ".x", ".y", ".z", ".w" };
for (int i = 0; i < width; i++) { for (int i = 0; i < width; i++) {
SkString index; SkString index;
SkString kernelIndex; SkString kernelIndex;
index.appendS32(i); index.appendS32(i/4);
kernel.appendArrayAccess(index.c_str(), &kernelIndex); kernel.appendArrayAccess(index.c_str(), &kernelIndex);
kernelIndex.append(kVecSuffix[i & 0x3]);
if (ce.useBounds()) { if (ce.useBounds()) {
// We used to compute a bool indicating whether we're in bounds or not, cast it to a // We used to compute a bool indicating whether we're in bounds or not, cast it to a
@ -119,7 +124,9 @@ void GrGLConvolutionEffect::onSetData(const GrGLSLProgramDataManager& pdman,
} }
int width = Gr1DKernelEffect::WidthFromRadius(conv.radius()); int width = Gr1DKernelEffect::WidthFromRadius(conv.radius());
pdman.set1fv(fKernelUni, width, conv.kernel()); int arrayCount = (width + 3) / 4;
SkASSERT(4 * arrayCount >= width);
pdman.set4fv(fKernelUni, arrayCount, conv.kernel());
} }
void GrGLConvolutionEffect::GenKey(const GrProcessor& processor, const GrGLSLCaps&, void GrGLConvolutionEffect::GenKey(const GrProcessor& processor, const GrGLSLCaps&,

18
src/gpu/effects/GrMatrixConvolutionEffect.cpp
View File

@ -39,14 +39,17 @@ void GrGLMatrixConvolutionEffect::emitCode(EmitArgs& args) {
int kWidth = mce.kernelSize().width(); int kWidth = mce.kernelSize().width();
int kHeight = mce.kernelSize().height(); int kHeight = mce.kernelSize().height();
int arrayCount = (kWidth + 3) / 4;
SkASSERT(4 * arrayCount >= kWidth);
GrGLSLUniformHandler* uniformHandler = args.fUniformHandler; GrGLSLUniformHandler* uniformHandler = args.fUniformHandler;
fImageIncrementUni = uniformHandler->addUniform(kFragment_GrShaderFlag, fImageIncrementUni = uniformHandler->addUniform(kFragment_GrShaderFlag,
kVec2f_GrSLType, kDefault_GrSLPrecision, kVec2f_GrSLType, kDefault_GrSLPrecision,
"ImageIncrement"); "ImageIncrement");
fKernelUni = uniformHandler->addUniformArray(kFragment_GrShaderFlag, fKernelUni = uniformHandler->addUniformArray(kFragment_GrShaderFlag,
kFloat_GrSLType, kDefault_GrSLPrecision, kVec4f_GrSLType, kDefault_GrSLPrecision,
"Kernel", "Kernel",
kWidth * kHeight); arrayCount);
fKernelOffsetUni = uniformHandler->addUniform(kFragment_GrShaderFlag, fKernelOffsetUni = uniformHandler->addUniform(kFragment_GrShaderFlag,
kVec2f_GrSLType, kDefault_GrSLPrecision, kVec2f_GrSLType, kDefault_GrSLPrecision,
"KernelOffset"); "KernelOffset");
@ -67,10 +70,14 @@ void GrGLMatrixConvolutionEffect::emitCode(EmitArgs& args) {
fragBuilder->codeAppendf("vec2 coord = %s - %s * %s;", coords2D.c_str(), kernelOffset, imgInc); fragBuilder->codeAppendf("vec2 coord = %s - %s * %s;", coords2D.c_str(), kernelOffset, imgInc);
fragBuilder->codeAppend("vec4 c;"); fragBuilder->codeAppend("vec4 c;");
const char* kVecSuffix[4] = { ".x", ".y", ".z", ".w" };
for (int y = 0; y < kHeight; y++) { for (int y = 0; y < kHeight; y++) {
for (int x = 0; x < kWidth; x++) { for (int x = 0; x < kWidth; x++) {
GrGLSLShaderBuilder::ShaderBlock block(fragBuilder); GrGLSLShaderBuilder::ShaderBlock block(fragBuilder);
fragBuilder->codeAppendf("float k = %s[%d * %d + %d];", kernel, y, kWidth, x); int offset = y*kWidth + x;
fragBuilder->codeAppendf("float k = %s[%d]%s;", kernel, offset / 4,
kVecSuffix[offset & 0x3]);
SkString coord; SkString coord;
coord.printf("coord + vec2(%d, %d) * %s", x, y, imgInc); coord.printf("coord + vec2(%d, %d) * %s", x, y, imgInc);
fDomain.sampleTexture(fragBuilder, fDomain.sampleTexture(fragBuilder,
@ -130,7 +137,10 @@ void GrGLMatrixConvolutionEffect::onSetData(const GrGLSLProgramDataManager& pdma
imageIncrement[1] = ySign / texture.height(); imageIncrement[1] = ySign / texture.height();
pdman.set2fv(fImageIncrementUni, 1, imageIncrement); pdman.set2fv(fImageIncrementUni, 1, imageIncrement);
pdman.set2fv(fKernelOffsetUni, 1, conv.kernelOffset()); pdman.set2fv(fKernelOffsetUni, 1, conv.kernelOffset());
pdman.set1fv(fKernelUni, conv.kernelSize().width() * conv.kernelSize().height(), conv.kernel()); int kernelCount = conv.kernelSize().width() * conv.kernelSize().height();
int arrayCount = (kernelCount + 3) / 4;
SkASSERT(4 * arrayCount >= kernelCount);
pdman.set4fv(fKernelUni, arrayCount, conv.kernel());
pdman.set1f(fGainUni, conv.gain()); pdman.set1f(fGainUni, conv.gain());
pdman.set1f(fBiasUni, conv.bias()); pdman.set1f(fBiasUni, conv.bias());
fDomain.setData(pdman, conv.domain(), texture.origin()); fDomain.setData(pdman, conv.domain(), texture.origin());