Pack float array uniforms into vec4 arrays

src/gpu/effects/GrMatrixConvolutionEffect.cpp

 /*
  * Copyright 2014 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 #include "GrMatrixConvolutionEffect.h"
 #include "glsl/GrGLSLFragmentProcessor.h"
 #include "glsl/GrGLSLFragmentShaderBuilder.h"
 #include "glsl/GrGLSLProgramDataManager.h"
@@ skipping 21 matching lines @@
     typedef GrGLSLFragmentProcessor INHERITED;
 };
 
 void GrGLMatrixConvolutionEffect::emitCode(EmitArgs& args) {
     const GrMatrixConvolutionEffect& mce = args.fFp.cast<GrMatrixConvolutionEffect>();
     const GrTextureDomain& domain = mce.domain();
 
     int kWidth = mce.kernelSize().width();
     int kHeight = mce.kernelSize().height();
 
+    int arrayCount = (kWidth + 3) / 4;
+    SkASSERT(4 * arrayCount >= kWidth);
+
     GrGLSLUniformHandler* uniformHandler = args.fUniformHandler;
     fImageIncrementUni = uniformHandler->addUniform(kFragment_GrShaderFlag,
                                                     kVec2f_GrSLType, kDefault_GrSLPrecision,
                                                     "ImageIncrement");
     fKernelUni = uniformHandler->addUniformArray(kFragment_GrShaderFlag,
-                                                 kFloat_GrSLType, kDefault_GrSLPrecision,
+                                                 kVec4f_GrSLType, kDefault_GrSLPrecision,
                                                  "Kernel",
-                                                 kWidth * kHeight);
+                                                 arrayCount);
     fKernelOffsetUni = uniformHandler->addUniform(kFragment_GrShaderFlag,
                                                   kVec2f_GrSLType, kDefault_GrSLPrecision,
                                                   "KernelOffset");
     fGainUni = uniformHandler->addUniform(kFragment_GrShaderFlag,
                                           kFloat_GrSLType, kDefault_GrSLPrecision, "Gain");
     fBiasUni = uniformHandler->addUniform(kFragment_GrShaderFlag,
                                           kFloat_GrSLType, kDefault_GrSLPrecision, "Bias");
 
     const char* kernelOffset = uniformHandler->getUniformCStr(fKernelOffsetUni);
     const char* imgInc = uniformHandler->getUniformCStr(fImageIncrementUni);
     const char* kernel = uniformHandler->getUniformCStr(fKernelUni);
     const char* gain = uniformHandler->getUniformCStr(fGainUni);
     const char* bias = uniformHandler->getUniformCStr(fBiasUni);
 
     GrGLSLFPFragmentBuilder* fragBuilder = args.fFragBuilder;
     SkString coords2D = fragBuilder->ensureFSCoords2D(args.fCoords, 0);
     fragBuilder->codeAppend("vec4 sum = vec4(0, 0, 0, 0);");
     fragBuilder->codeAppendf("vec2 coord = %s - %s * %s;", coords2D.c_str(), kernelOffset, imgInc);
     fragBuilder->codeAppend("vec4 c;");
 
+    const char* kVecSuffix[4] = { ".x", ".y", ".z", ".w" };
     for (int y = 0; y < kHeight; y++) {
         for (int x = 0; x < kWidth; x++) {
             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;
             coord.printf("coord + vec2(%d, %d) * %s", x, y, imgInc);
             fDomain.sampleTexture(fragBuilder,
                                   uniformHandler,
                                   args.fGLSLCaps,
                                   domain,
                                   "c",
                                   coord,
                                   args.fSamplers[0]);
             if (!mce.convolveAlpha()) {
@@ skipping 39 matching lines @@
                                             const GrProcessor& processor) {
     const GrMatrixConvolutionEffect& conv = processor.cast<GrMatrixConvolutionEffect>();
     GrTexture& texture = *conv.texture(0);
 
     float imageIncrement[2];
     float ySign = texture.origin() == kTopLeft_GrSurfaceOrigin ? 1.0f : -1.0f;
     imageIncrement[0] = 1.0f / texture.width();
     imageIncrement[1] = ySign / texture.height();
     pdman.set2fv(fImageIncrementUni, 1, imageIncrement);
     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(fBiasUni, conv.bias());
     fDomain.setData(pdman, conv.domain(), texture.origin());
 }
 
 GrMatrixConvolutionEffect::GrMatrixConvolutionEffect(GrTexture* texture,
                                                      const SkIRect& bounds,
                                                      const SkISize& kernelSize,
                                                      const SkScalar* kernel,
                                                      SkScalar gain,
@@ skipping 104 matching lines @@
     return GrMatrixConvolutionEffect::Create(d->fTextures[texIdx],
                                              bounds,
                                              kernelSize,
                                              kernel.get(),
                                              gain,
                                              bias,
                                              kernelOffset,
                                              tileMode,
                                              convolveAlpha);
 }