Fixed gpu perlin noise on Windows

src/effects/SkPerlinNoiseShader.cpp

Index: src/effects/SkPerlinNoiseShader.cpp
===================================================================
--- src/effects/SkPerlinNoiseShader.cpp	(revision 9474)
+++ src/effects/SkPerlinNoiseShader.cpp	(working copy)
@@ -494,7 +494,7 @@
 
 /////////////////////////////////////////////////////////////////////
 
-#if SK_SUPPORT_GPU && !defined(SK_BUILD_FOR_ANDROID) && !defined(SK_BUILD_FOR_WIN)
+#if SK_SUPPORT_GPU && !defined(SK_BUILD_FOR_ANDROID)
 // CPU noise is faster on Android, so the GPU implementation is only for desktop
 
 #include "GrTBackendEffectFactory.h"
@@ -986,13 +986,16 @@
         stitchDataUni = builder->getUniformCStr(fStitchDataUni);
     }
 
-    const char* chanCoords  = "chanCoords";
+    // There are 4 lines, so the center of each line is 1/8, 3/8, 5/8 and 7/8
+    const char* chanCoordR  = "0.125";
+    const char* chanCoordG  = "0.375";
+    const char* chanCoordB  = "0.625";
+    const char* chanCoordA  = "0.875";
+    const char* chanCoord   = "chanCoord";
     const char* stitchData  = "stitchData";
     const char* ratio       = "ratio";
-    const char* noise       = "noise";
     const char* noiseXY     = "noiseXY";
     const char* noiseVec    = "noiseVec";
-    const char* noiseVecIni = "noiseVecIni";
     const char* noiseSmooth = "noiseSmooth";
     const char* fractVal    = "fractVal";
     const char* uv          = "uv";
@@ -1005,61 +1008,52 @@
     // [-1,1] vector and perform a dot product between that vector and the provided vector.
     const char* dotLattice  = "dot(((%s.ga + %s.rb * vec2(%s)) * vec2(2.0) - vec2(1.0)), %s);";
 
-    // There are 4 lines, so the center of each line is 1/8, 3/8, 5/8 and 7/8
-    builder->fsCodeAppendf("\t\tconst vec4 %s = vec4(0.125, 0.375, 0.625, 0.875);", chanCoords);
+    // Add noise function
+    static const GrGLShaderVar perlinnoise_Args[] =  {

[bsalomon, 2013/06/10 13:39:25]

We already have two conventions for static const globals, I don't think we need
a third :)

In some places kWeNameThemWithALowerCaseKRegardlessOfTheType
gAndInOthersWithAGIfTheyreNotIntegral.

+        GrGLShaderVar(chanCoord, kFloat_GrSLType),
+        GrGLShaderVar(noiseVec, kVec2f_GrSLType)
+    };
 
-    // There are rounding errors if the floor operation is not performed here
-    builder->fsCodeAppendf("\t\tvec2 %s = floor((%s*vec3(%s, 1.0)).xy) * %s;",
-                           noiseVecIni, invMatrixUni, vCoords, baseFrequencyUni);
+    static const GrGLShaderVar perlinnoise_stitch_Args[] =  {
+        GrGLShaderVar(chanCoord, kFloat_GrSLType),
+        GrGLShaderVar(noiseVec, kVec2f_GrSLType),
+        GrGLShaderVar(stitchData, kVec4f_GrSLType)
+    };
 
-    // Loop over the 4 channels
-    builder->fsCodeAppend("\t\tfor (int channel = 3; channel >= 0; --channel) {");
+    SkString noise_code;

[bsalomon, 2013/06/10 13:39:25]

localVarsAreLikeThis.

 
-    if (fStitchTiles) {
-        // Set up TurbulenceInitial stitch values.
-        builder->fsCodeAppendf("\t\tvec4 %s = %s;", stitchData, stitchDataUni);
-    }
+    noise_code.appendf(
+        "\tvec4 %s = vec4(floor(%s) + vec2(%s), fract(%s));",
+        noiseXY, noiseVec, perlinNoise, noiseVec);
 
-    builder->fsCodeAppendf("\t\t%s[channel] = 0.0;", outputColor);
-
-    builder->fsCodeAppendf("\t\tfloat %s = 1.0;", ratio);
-    builder->fsCodeAppendf("\t\tvec2 %s = %s;", noiseVec, noiseVecIni);
-
-    // Loop over all octaves
-    builder->fsCodeAppendf("\t\tfor (int octave = 0; octave < %d; ++octave) {", fNumOctaves);
-
-    builder->fsCodeAppendf("\t\tvec4 %s = vec4(floor(%s) + vec2(%s), fract(%s));",
-                  noiseXY, noiseVec, perlinNoise, noiseVec);
-
     // smooth curve : t * t * (3 - 2 * t)
-    builder->fsCodeAppendf("\t\tvec2 %s = %s.zw * %s.zw * (vec2(3.0) - vec2(2.0) * %s.zw);",
-                  noiseSmooth, noiseXY, noiseXY, noiseXY);
+    noise_code.appendf("\n\tvec2 %s = %s.zw * %s.zw * (vec2(3.0) - vec2(2.0) * %s.zw);",
+        noiseSmooth, noiseXY, noiseXY, noiseXY);
 
     // Adjust frequencies if we're stitching tiles
     if (fStitchTiles) {
-        builder->fsCodeAppendf("\t\tif(%s.x >= %s.y) { %s.x -= %s.x; }",
-                      noiseXY, stitchData, noiseXY, stitchData);
-        builder->fsCodeAppendf("\t\tif(%s.x >= (%s.y - 1.0)) { %s.x -= (%s.x - 1.0); }",
-                      noiseXY, stitchData, noiseXY, stitchData);
-        builder->fsCodeAppendf("\t\tif(%s.y >= %s.w) { %s.y -= %s.z; }",
-                      noiseXY, stitchData, noiseXY, stitchData);
-        builder->fsCodeAppendf("\t\tif(%s.y >= (%s.w - 1.0)) { %s.y -= (%s.z - 1.0); }",
-                      noiseXY, stitchData, noiseXY, stitchData);
+        noise_code.appendf("\n\tif(%s.x >= %s.y) { %s.x -= %s.x; }",
+            noiseXY, stitchData, noiseXY, stitchData);
+        noise_code.appendf("\n\tif(%s.x >= (%s.y - 1.0)) { %s.x -= (%s.x - 1.0); }",
+            noiseXY, stitchData, noiseXY, stitchData);
+        noise_code.appendf("\n\tif(%s.y >= %s.w) { %s.y -= %s.z; }",
+            noiseXY, stitchData, noiseXY, stitchData);
+        noise_code.appendf("\n\tif(%s.y >= (%s.w - 1.0)) { %s.y -= (%s.z - 1.0); }",
+            noiseXY, stitchData, noiseXY, stitchData);
     }
 
     // Get texture coordinates and normalize
-    builder->fsCodeAppendf("\t\t%s.xy = fract(floor(mod(%s.xy, 256.0)) / vec2(256.0));",
-                  noiseXY, noiseXY);
+    noise_code.appendf("\n\t%s.xy = fract(floor(mod(%s.xy, 256.0)) / vec2(256.0));\n",
+        noiseXY, noiseXY);
 
     // Get permutation for x
     {
         SkString xCoords("");
         xCoords.appendf("vec2(%s.x, 0.5)", noiseXY);
 
-        builder->fsCodeAppendf("\t\tvec2 %s;\t\t%s.x = ", latticeIdx, latticeIdx);
-        builder->appendTextureLookup(GrGLShaderBuilder::kFragment_ShaderType,
-                                     samplers[0], xCoords.c_str(), kVec2f_GrSLType);
-        builder->fsCodeAppend(".r;\n");
+        noise_code.appendf("\n\tvec2 %s;\n\t%s.x = ", latticeIdx, latticeIdx);
+        builder->appendTextureLookup(&noise_code, samplers[0], xCoords.c_str(), kVec2f_GrSLType);
+        noise_code.append(".r;");
     }
 
     // Get permutation for x + 1
@@ -1067,104 +1061,145 @@
         SkString xCoords("");
         xCoords.appendf("vec2(fract(%s.x + %s), 0.5)", noiseXY, inc8bit);
 
-        builder->fsCodeAppendf("\t\t%s.y = ", latticeIdx);
-        builder->appendTextureLookup(GrGLShaderBuilder::kFragment_ShaderType,
-                                     samplers[0], xCoords.c_str(), kVec2f_GrSLType);
-        builder->fsCodeAppend(".r;\n");
+        noise_code.appendf("\n\t%s.y = ", latticeIdx);
+        builder->appendTextureLookup(&noise_code, samplers[0], xCoords.c_str(), kVec2f_GrSLType);
+        noise_code.append(".r;");
     }
 
     // Get (x,y) coordinates with the permutated x
-    builder->fsCodeAppendf("\t\t%s = fract(%s + %s.yy);", latticeIdx, latticeIdx, noiseXY);
+    noise_code.appendf("\n\t%s = fract(%s + %s.yy);", latticeIdx, latticeIdx, noiseXY);
 
-    builder->fsCodeAppendf("\t\tvec2 %s = %s.zw;", fractVal, noiseXY);
+    noise_code.appendf("\n\tvec2 %s = %s.zw;", fractVal, noiseXY);
 
-    builder->fsCodeAppendf("\t\tvec2 %s;", uv);
+    noise_code.appendf("\n\n\tvec2 %s;", uv);
     // Compute u, at offset (0,0)
     {
         SkString latticeCoords("");
-        latticeCoords.appendf("vec2(%s.x, %s[channel])", latticeIdx, chanCoords);
-        builder->fsCodeAppendf("vec4 %s = ", lattice);
-        builder->appendTextureLookup(GrGLShaderBuilder::kFragment_ShaderType,
-                                     samplers[1], latticeCoords.c_str(), kVec2f_GrSLType);
-        builder->fsCodeAppendf(".bgra;\n\t\t%s.x = ", uv);
-        builder->fsCodeAppendf(dotLattice, lattice, lattice, inc8bit, fractVal);
+        latticeCoords.appendf("vec2(%s.x, %s)", latticeIdx, chanCoord);
+        noise_code.appendf("\n\tvec4 %s = ", lattice);
+        builder->appendTextureLookup(&noise_code, samplers[1], latticeCoords.c_str(),
+            kVec2f_GrSLType);
+        noise_code.appendf(".bgra;\n\t%s.x = ", uv);
+        noise_code.appendf(dotLattice, lattice, lattice, inc8bit, fractVal);
     }
 
-    builder->fsCodeAppendf("\t\t%s.x -= 1.0;", fractVal);
+    noise_code.appendf("\n\t%s.x -= 1.0;", fractVal);
     // Compute v, at offset (-1,0)
     {
         SkString latticeCoords("");
-        latticeCoords.appendf("vec2(%s.y, %s[channel])", latticeIdx, chanCoords);
-        builder->fsCodeAppend("lattice = ");
-        builder->appendTextureLookup(GrGLShaderBuilder::kFragment_ShaderType,
-                                     samplers[1], latticeCoords.c_str(), kVec2f_GrSLType);
-        builder->fsCodeAppendf(".bgra;\n\t\t%s.y = ", uv);
-        builder->fsCodeAppendf(dotLattice, lattice, lattice, inc8bit, fractVal);
+        latticeCoords.appendf("vec2(%s.y, %s)", latticeIdx, chanCoord);
+        noise_code.append("lattice = ");
+        builder->appendTextureLookup(&noise_code, samplers[1], latticeCoords.c_str(),
+            kVec2f_GrSLType);
+        noise_code.appendf(".bgra;\n\t%s.y = ", uv);
+        noise_code.appendf(dotLattice, lattice, lattice, inc8bit, fractVal);
     }
 
     // Compute 'a' as a linear interpolation of 'u' and 'v'
-    builder->fsCodeAppendf("\t\tvec2 %s;", ab);
-    builder->fsCodeAppendf("\t\t%s.x = mix(%s.x, %s.y, %s.x);", ab, uv, uv, noiseSmooth);
+    noise_code.appendf("\n\tvec2 %s;", ab);
+    noise_code.appendf("\n\t%s.x = mix(%s.x, %s.y, %s.x);", ab, uv, uv, noiseSmooth);
 
-    builder->fsCodeAppendf("\t\t%s.y -= 1.0;", fractVal);
+    noise_code.appendf("\n\t%s.y -= 1.0;", fractVal);
     // Compute v, at offset (-1,-1)
     {
         SkString latticeCoords("");
-        latticeCoords.appendf("vec2(fract(%s.y + %s), %s[channel])",
-            latticeIdx, inc8bit, chanCoords);
-        builder->fsCodeAppend("lattice = ");
-        builder->appendTextureLookup(GrGLShaderBuilder::kFragment_ShaderType,
-                                     samplers[1], latticeCoords.c_str(), kVec2f_GrSLType);
-        builder->fsCodeAppendf(".bgra;\n\t\t%s.y = ", uv);
-        builder->fsCodeAppendf(dotLattice, lattice, lattice, inc8bit, fractVal);
+        latticeCoords.appendf("vec2(fract(%s.y + %s), %s)", latticeIdx, inc8bit, chanCoord);
+        noise_code.append("lattice = ");
+        builder->appendTextureLookup(&noise_code, samplers[1], latticeCoords.c_str(),
+            kVec2f_GrSLType);
+        noise_code.appendf(".bgra;\n\t%s.y = ", uv);
+        noise_code.appendf(dotLattice, lattice, lattice, inc8bit, fractVal);
     }
 
-    builder->fsCodeAppendf("\t\t%s.x += 1.0;", fractVal);
+    noise_code.appendf("\n\t%s.x += 1.0;", fractVal);
     // Compute u, at offset (0,-1)
     {
         SkString latticeCoords("");
-        latticeCoords.appendf("vec2(fract(%s.x + %s), %s[channel])",
-            latticeIdx, inc8bit, chanCoords);
-        builder->fsCodeAppend("lattice = ");
-        builder->appendTextureLookup(GrGLShaderBuilder::kFragment_ShaderType,
-                                     samplers[1], latticeCoords.c_str(), kVec2f_GrSLType);
-        builder->fsCodeAppendf(".bgra;\n\t\t%s.x = ", uv);
-        builder->fsCodeAppendf(dotLattice, lattice, lattice, inc8bit, fractVal);
+        latticeCoords.appendf("vec2(fract(%s.x + %s), %s)", latticeIdx, inc8bit, chanCoord);
+        noise_code.append("lattice = ");
+        builder->appendTextureLookup(&noise_code, samplers[1], latticeCoords.c_str(),
+            kVec2f_GrSLType);
+        noise_code.appendf(".bgra;\n\t%s.x = ", uv);
+        noise_code.appendf(dotLattice, lattice, lattice, inc8bit, fractVal);
     }
 
     // Compute 'b' as a linear interpolation of 'u' and 'v'
-    builder->fsCodeAppendf("\t\t%s.y = mix(%s.x, %s.y, %s.x);", ab, uv, uv, noiseSmooth);
+    noise_code.appendf("\n\t%s.y = mix(%s.x, %s.y, %s.x);", ab, uv, uv, noiseSmooth);
     // Compute the noise as a linear interpolation of 'a' and 'b'
-    builder->fsCodeAppendf("\t\tfloat %s = mix(%s.x, %s.y, %s.y);", noise, ab, ab, noiseSmooth);
+    noise_code.appendf("\n\treturn mix(%s.x, %s.y, %s.y);\n", ab, ab, noiseSmooth);
 
-    builder->fsCodeAppendf("\t\t%s[channel] += ", outputColor);
-    builder->fsCodeAppendf((fType == SkPerlinNoiseShader::kFractalNoise_Type) ?
-                  "%s / %s;" : "abs(%s) / %s;", noise, ratio);
+    SkString noise_funcName;
+    builder->emitFunction(GrGLShaderBuilder::kFragment_ShaderType, kFloat_GrSLType, "perlinnoise",

[bsalomon, 2013/06/10 13:39:25]

Would this be easier to read if the there was just a single if?

if (fStitchTiles) {
 builder->emitFunction()
} else
 builder->emitFunction()
}

+                          fStitchTiles ? SK_ARRAY_COUNT(perlinnoise_stitch_Args) :
+                          SK_ARRAY_COUNT(perlinnoise_Args),
+                          fStitchTiles ? perlinnoise_stitch_Args : perlinnoise_Args,
+                          noise_code.c_str(), &noise_funcName);
 
-    builder->fsCodeAppendf("\t\t%s *= vec2(2.0);", noiseVec);
-    builder->fsCodeAppendf("\t\t%s *= 2.0;", ratio);
+    // There are rounding errors if the floor operation is not performed here
+    builder->fsCodeAppendf("\n\t\tvec2 %s = floor((%s * vec3(%s, 1.0)).xy) * %s;",
+                           noiseVec, invMatrixUni, vCoords, baseFrequencyUni);
 
+    // Clear the color accumulator
+    builder->fsCodeAppendf("\n\t\t%s = vec4(0.0);", outputColor);
+
     if (fStitchTiles) {
-        builder->fsCodeAppendf("\t\t%s.xz *= vec2(2.0);", stitchData);
-        builder->fsCodeAppendf("\t\t%s.yw = %s.xz + vec2(%s);", stitchData, stitchData, perlinNoise);
+        // Set up TurbulenceInitial stitch values.
+        builder->fsCodeAppendf("\n\t\tvec4 %s = %s;", stitchData, stitchDataUni);
     }
-    builder->fsCodeAppend("\t\t}"); // end of the for loop on octaves
 
-    builder->fsCodeAppend("\t\t}"); // end of the for loop on channels
+    builder->fsCodeAppendf("\n\t\tfloat %s = 1.0;", ratio);
 
+    // Loop over all octaves
+    builder->fsCodeAppendf("\n\t\tfor (int octave = 0; octave < %d; ++octave) {", fNumOctaves);
+
+    builder->fsCodeAppendf("\n\t\t\t%s += ", outputColor);
+    if (fType != SkPerlinNoiseShader::kFractalNoise_Type) {
+        builder->fsCodeAppend("abs(");
+    }
+    if (fStitchTiles) {
+        builder->fsCodeAppendf(
+            "vec4(\n\t\t\t\t%s(%s, %s, %s),\n\t\t\t\t%s(%s, %s, %s),"
+                 "\n\t\t\t\t%s(%s, %s, %s),\n\t\t\t\t%s(%s, %s, %s))",
+            noise_funcName.c_str(), chanCoordR, noiseVec, stitchData,
+            noise_funcName.c_str(), chanCoordG, noiseVec, stitchData,
+            noise_funcName.c_str(), chanCoordB, noiseVec, stitchData,
+            noise_funcName.c_str(), chanCoordA, noiseVec, stitchData);
+    } else {
+        builder->fsCodeAppendf(
+            "vec4(\n\t\t\t\t%s(%s, %s),\n\t\t\t\t%s(%s, %s),"
+                 "\n\t\t\t\t%s(%s, %s),\n\t\t\t\t%s(%s, %s))",
+            noise_funcName.c_str(), chanCoordR, noiseVec,
+            noise_funcName.c_str(), chanCoordG, noiseVec,
+            noise_funcName.c_str(), chanCoordB, noiseVec,
+            noise_funcName.c_str(), chanCoordA, noiseVec);
+    }
+    if (fType != SkPerlinNoiseShader::kFractalNoise_Type) {
+        builder->fsCodeAppendf(")"); // end of "abs("
+    }
+    builder->fsCodeAppendf(" * %s;", ratio);
+
+    builder->fsCodeAppendf("\n\t\t\t%s *= vec2(2.0);", noiseVec);
+    builder->fsCodeAppendf("\n\t\t\t%s *= 0.5;", ratio);
+
+    if (fStitchTiles) {
+        builder->fsCodeAppendf("\n\t\t\t%s.xz *= vec2(2.0);", stitchData);
+        builder->fsCodeAppendf("\n\t\t\t%s.yw = %s.xz + vec2(%s);", stitchData, stitchData, perlinNoise);
+    }
+    builder->fsCodeAppend("\n\t\t}"); // end of the for loop on octaves
+
     if (fType == SkPerlinNoiseShader::kFractalNoise_Type) {
         // The value of turbulenceFunctionResult comes from ((turbulenceFunctionResult) + 1) / 2
         // by fractalNoise and (turbulenceFunctionResult) by turbulence.
-        builder->fsCodeAppendf("\t\t%s = %s * vec4(0.5) + vec4(0.5);", outputColor, outputColor);
+        builder->fsCodeAppendf("\n\t\t%s = %s * vec4(0.5) + vec4(0.5);", outputColor, outputColor);
     }
 
-    builder->fsCodeAppendf("\t\t%s.a *= %s;", outputColor, alphaUni);
+    builder->fsCodeAppendf("\n\t\t%s.a *= %s;", outputColor, alphaUni);
 
     // Clamp values
-    builder->fsCodeAppendf("\t\t%s = clamp(%s, 0.0, 1.0);", outputColor, outputColor);
+    builder->fsCodeAppendf("\n\t\t%s = clamp(%s, 0.0, 1.0);", outputColor, outputColor);
 
     // Pre-multiply the result
-    builder->fsCodeAppendf("\t\t%s = vec4(%s.rgb * %s.aaa, %s.a);\n",
+    builder->fsCodeAppendf("\n\t\t%s = vec4(%s.rgb * %s.aaa, %s.a);\n",
                   outputColor, outputColor, outputColor, outputColor);
 }
 
@@ -1295,7 +1330,7 @@
 #else
 
 GrEffectRef* SkPerlinNoiseShader::asNewEffect(GrContext*, const SkPaint&) const {
-#if !defined(SK_BUILD_FOR_ANDROID) && !defined(SK_BUILD_FOR_WIN)
+#if !defined(SK_BUILD_FOR_ANDROID)
     SkDEBUGFAIL("Should not call in GPU-less build");
 #endif
     return NULL;