Improve performance of highQualityFilter

src/core/SkBitmapFilter.cpp

 /*
  * Copyright 2013 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "SkErrorInternals.h"
 #include "SkConvolver.h"
 #include "SkBitmapProcState.h"
@@ skipping 10 matching lines @@
 // resampling an image as it is blitted.  Typically these are used only when
 // the image is rotated or has some other complex transformation applied.
 // Scaled images will usually be rescaled directly before rasterization.
 
 namespace {
 
 template <typename Color, typename ColorPacker>
 void highQualityFilter(ColorPacker pack, const SkBitmapProcState& s, int x, int y, Color* SK_RESTRICT colors, int count) {
     const int maxX = s.fBitmap->width();
     const int maxY = s.fBitmap->height();
+    SkAutoTMalloc<SkScalar> xWeights(maxX);
 
     while (count-- > 0) {
         SkPoint srcPt;
         s.fInvProc(s.fInvMatrix, x + 0.5f,
                     y + 0.5f, &srcPt);
         srcPt.fX -= SK_ScalarHalf;
         srcPt.fY -= SK_ScalarHalf;
 
         SkScalar weight = 0;
         SkScalar fr = 0, fg = 0, fb = 0, fa = 0;
 
         int y0 = SkClampMax(SkScalarCeilToInt(srcPt.fY-s.getBitmapFilter()->width()), maxY);
         int y1 = SkClampMax(SkScalarFloorToInt(srcPt.fY+s.getBitmapFilter()->width()+1), maxY);
         int x0 = SkClampMax(SkScalarCeilToInt(srcPt.fX-s.getBitmapFilter()->width()), maxX);
         int x1 = SkClampMax(SkScalarFloorToInt(srcPt.fX+s.getBitmapFilter()->width())+1, maxX);
 
+        for (int srcX = x0; srcX < x1 ; srcX++) {
+            // Looking these up once instead of each loop is a ~15% speedup.
+            xWeights[srcX - x0] = s.getBitmapFilter()->lookupScalar((srcPt.fX - srcX));
+        }
+
         for (int srcY = y0; srcY < y1; srcY++) {
             SkScalar yWeight = s.getBitmapFilter()->lookupScalar((srcPt.fY - srcY));
 
             for (int srcX = x0; srcX < x1 ; srcX++) {
-                SkScalar xWeight = s.getBitmapFilter()->lookupScalar((srcPt.fX - srcX));
+                SkScalar xWeight = xWeights[srcX - x0];
 
                 SkScalar combined_weight = SkScalarMul(xWeight, yWeight);
 
                 SkPMColor c = *s.fBitmap->getAddr32(srcX, srcY);
                 fr += combined_weight * SkGetPackedR32(c);
                 fg += combined_weight * SkGetPackedG32(c);
                 fb += combined_weight * SkGetPackedB32(c);
                 fa += combined_weight * SkGetPackedA32(c);
                 weight += combined_weight;
             }
@@ skipping 77 matching lines @@
     }
 
     if (fInvType & SkMatrix::kScale_Mask) {
         fShaderProc32 = highQualityFilter32;
         fShaderProc16 = highQualityFilter16;
         return true;
     } else {
         return false;
     }
 }