Pack float array uniforms into vec4 arrays
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
diff --git a/src/gpu/effects/GrConvolutionEffect.cpp b/src/gpu/effects/GrConvolutionEffect.cpp
index 1fb2e95..4f09138 100644
--- a/src/gpu/effects/GrConvolutionEffect.cpp
+++ b/src/gpu/effects/GrConvolutionEffect.cpp
@@ -46,9 +46,12 @@
int width = Gr1DKernelEffect::WidthFromRadius(ce.radius());
+ int arrayCount = (width + 3) / 4;
+ SkASSERT(4 * arrayCount >= width);
+
fKernelUni = uniformHandler->addUniformArray(kFragment_GrShaderFlag,
- kFloat_GrSLType, kDefault_GrSLPrecision,
- "Kernel", width);
+ kVec4f_GrSLType, kDefault_GrSLPrecision,
+ "Kernel", arrayCount);
GrGLSLFPFragmentBuilder* fragBuilder = args.fFragBuilder;
SkString coords2D = fragBuilder->ensureFSCoords2D(args.fCoords, 0);
@@ -61,11 +64,13 @@
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.
+ const char* kVecSuffix[4] = { ".x", ".y", ".z", ".w" };
for (int i = 0; i < width; i++) {
SkString index;
SkString kernelIndex;
- index.appendS32(i);
+ index.appendS32(i/4);
kernel.appendArrayAccess(index.c_str(), &kernelIndex);
+ kernelIndex.append(kVecSuffix[i & 0x3]);
if (ce.useBounds()) {
// We used to compute a bool indicating whether we're in bounds or not, cast it to a
@@ -119,7 +124,9 @@
}
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&,