New bitmap filter checkin; this time with less build breakage

BUG=

Review URL: https://codereview.chromium.org/18942002

git-svn-id: http://skia.googlecode.com/svn/trunk@9944 2bbb7eff-a529-9590-31e7-b0007b416f81

18 files changed, 1048 insertions, 219 deletions

diff --git a/bench/BitmapBench.cpp b/bench/BitmapBench.cpp
index 3bf6b779cb..d8363fcb9d 100644
--- a/bench/BitmapBench.cpp
+++ b/bench/BitmapBench.cpp
@@ -149,7 +149,7 @@ protected:
         int count = N;
 #ifdef SK_RELEASE
         // in DEBUG, N is always 1
-        if (paint.getFlags() & SkPaint::kBicubicFilterBitmap_Flag) {
+        if (paint.getFlags() & SkPaint::kHighQualityFilterBitmap_Flag) {
             count /= BICUBIC_DUR_SCALE;
         }
 #endif
@@ -170,7 +170,7 @@ protected:
 #ifdef SK_DEBUG
         return 1;
 #else
-        return (paint.getFlags() & SkPaint::kBicubicFilterBitmap_Flag) ?
+        return (paint.getFlags() & SkPaint::kHighQualityFilterBitmap_Flag) ?
                 (float)BICUBIC_DUR_SCALE : 1;
 #endif
     }
@@ -266,12 +266,12 @@ protected:
         }
 
         uint32_t orMask = 0;
-        uint32_t clearMask = SkPaint::kFilterBitmap_Flag | SkPaint::kBicubicFilterBitmap_Flag;
+        uint32_t clearMask = SkPaint::kFilterBitmap_Flag | SkPaint::kHighQualityFilterBitmap_Flag;
         if (fFlags & kBilerp_Flag) {
             orMask |= SkPaint::kFilterBitmap_Flag;
         }
         if (fFlags & kBicubic_Flag) {
-            orMask |= SkPaint::kBicubicFilterBitmap_Flag;
+            orMask |= SkPaint::kHighQualityFilterBitmap_Flag;
         }
         this->setPaintMasks(orMask, clearMask);
 
diff --git a/bench/BitmapScaleBench.cpp b/bench/BitmapScaleBench.cpp
new file mode 100644
index 0000000000..45ea1b46ef
--- /dev/null
+++ b/bench/BitmapScaleBench.cpp
@@ -0,0 +1,135 @@
+/*
+ * 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 "SkBenchmark.h"
+#include "SkCanvas.h"
+#include "SkPaint.h"
+#include "SkRandom.h"
+#include "SkShader.h"
+#include "SkString.h"
+#include "SkBlurMask.h"
+
+class BitmapScaleBench: public SkBenchmark {
+    int         fLoopCount;
+    int         fInputSize;
+    int         fOutputSize;
+    SkString    fName;
+
+public:
+    BitmapScaleBench(void *param, int is, int os) : INHERITED(param) {
+        fInputSize = is;
+        fOutputSize = os;
+
+        fLoopCount = 100;
+    }
+
+protected:
+    
+    SkBitmap fInputBitmap, fOutputBitmap;
+    SkMatrix fMatrix;
+    
+    virtual const char* onGetName() {
+        return fName.c_str();
+    }
+
+    int inputSize() const {
+        return fInputSize;
+    }
+    
+    int outputSize() const {
+        return fOutputSize;
+    }
+    
+    float scale() const {
+        return float(outputSize())/inputSize();
+    }
+    
+    SkIPoint onGetSize() SK_OVERRIDE {
+        return SkIPoint::Make( fOutputSize, fOutputSize );
+    }
+
+    void setName(const char * name) {
+        fName.printf( "bitmap_scale_%s_%d_%d", name, fInputSize, fOutputSize );
+    }
+    
+    virtual void onPreDraw() {
+        fInputBitmap.setConfig(SkBitmap::kARGB_8888_Config, fInputSize, fInputSize);
+        fInputBitmap.allocPixels();
+        fInputBitmap.eraseColor(SK_ColorWHITE);
+        fInputBitmap.setIsOpaque(true);
+        
+        fOutputBitmap.setConfig(SkBitmap::kARGB_8888_Config, fOutputSize, fOutputSize);
+        fOutputBitmap.allocPixels();
+        fOutputBitmap.setIsOpaque(true);
+        
+        fMatrix.setScale( scale(), scale() );
+    }
+
+    virtual void onDraw(SkCanvas*) {
+        SkPaint paint;
+        this->setupPaint(&paint);
+
+        preBenchSetup();
+
+        for (int i = 0; i < SkBENCHLOOP(fLoopCount); i++) {
+            doScaleImage();
+        }
+    }
+
+    virtual void doScaleImage() = 0;
+    virtual void preBenchSetup() {}
+private:
+    typedef SkBenchmark INHERITED;
+};
+
+class BitmapFilterScaleBench: public BitmapScaleBench {
+ public:
+    BitmapFilterScaleBench(void *param, int is, int os) : INHERITED(param, is, os) {
+        setName( "filter" );
+    }
+protected:
+    virtual void doScaleImage() SK_OVERRIDE {
+        SkCanvas canvas( fOutputBitmap );
+        SkPaint paint;
+        
+        paint.setFlags( SkPaint::kHighQualityFilterBitmap_Flag | SkPaint::kFilterBitmap_Flag );
+        
+        canvas.drawBitmapMatrix( fInputBitmap, fMatrix, &paint );
+    }
+private:
+    typedef BitmapScaleBench INHERITED;
+};
+
+class BitmapDirectScaleBench: public BitmapScaleBench {
+ public:
+    BitmapDirectScaleBench(void *param, int is, int os) : INHERITED(param, is, os) {
+        setName( "direct" );
+    }
+protected:
+    virtual void doScaleImage() SK_OVERRIDE {
+        fInputBitmap.scale( &fOutputBitmap );
+    }
+private:
+    typedef BitmapScaleBench INHERITED;
+};
+
+
+DEF_BENCH(return new BitmapFilterScaleBench(p, 10, 90);)
+DEF_BENCH(return new BitmapFilterScaleBench(p, 30, 90);)
+DEF_BENCH(return new BitmapFilterScaleBench(p, 80, 90);)
+// DEF_BENCH(return new BitmapFilterScaleBench(p, 90, 90);)
+// DEF_BENCH(return new BitmapFilterScaleBench(p, 90, 80);)
+// DEF_BENCH(return new BitmapFilterScaleBench(p, 90, 30);)
+// DEF_BENCH(return new BitmapFilterScaleBench(p, 90, 10);)
+
+DEF_BENCH(return new BitmapDirectScaleBench(p, 10, 90);)
+DEF_BENCH(return new BitmapDirectScaleBench(p, 30, 90);)
+DEF_BENCH(return new BitmapDirectScaleBench(p, 80, 90);)
+// DEF_BENCH(return new BitmapDirectScaleBench(p, 90, 90);)
+// DEF_BENCH(return new BitmapDirectScaleBench(p, 90, 80);)
+// DEF_BENCH(return new BitmapDirectScaleBench(p, 90, 30);)
+// DEF_BENCH(return new BitmapDirectScaleBench(p, 90, 10);)
diff --git a/gm/filterbitmap.cpp b/gm/filterbitmap.cpp
index 830790c03c..0afcb95101 100644
--- a/gm/filterbitmap.cpp
+++ b/gm/filterbitmap.cpp
@@ -35,7 +35,7 @@ static void draw_col(SkCanvas* canvas, const SkBitmap& bm, const SkMatrix& mat,
     canvas->translate(dx, 0);
     canvas->drawBitmapMatrix(bm, mat, &paint);
 
-    paint.setFlags(paint.getFlags() | SkPaint::kBicubicFilterBitmap_Flag);
+    paint.setFlags(paint.getFlags() | SkPaint::kHighQualityFilterBitmap_Flag);
     canvas->translate(dx, 0);
     canvas->drawBitmapMatrix(bm, mat, &paint);
 }
@@ -64,7 +64,7 @@ public:
         this->setBGColor(0xFFDDDDDD);
     }
 
-    void setName( const char name[] ) {
+    void setName(const char name[]) {
         fName.set(name);
     }
 
@@ -99,12 +99,12 @@ private:
 
 class FilterBitmapTextGM: public FilterBitmapGM {
   public:
-      FilterBitmapTextGM( float textSize )
-      : fTextSize( textSize )
+      FilterBitmapTextGM(float textSize)
+      : fTextSize(textSize)
         {
             char name[1024];
-            sprintf( name, "filterbitmap_text_%.2fpt", fTextSize );
-            setName( name );
+            sprintf(name, "filterbitmap_text_%.2fpt", fTextSize);
+            setName(name);
         }
 
   protected:
@@ -140,12 +140,12 @@ class FilterBitmapTextGM: public FilterBitmapGM {
 
 class FilterBitmapCheckerboardGM: public FilterBitmapGM {
   public:
-      FilterBitmapCheckerboardGM( int size, int num_checks )
-      : fSize( size ), fNumChecks( num_checks )
+      FilterBitmapCheckerboardGM(int size, int num_checks)
+      : fSize(size), fNumChecks(num_checks)
         {
             char name[1024];
-            sprintf( name, "filterbitmap_checkerboard_%d_%d", fSize, fNumChecks );
-            setName( name );
+            sprintf(name, "filterbitmap_checkerboard_%d_%d", fSize, fNumChecks);
+            setName(name);
         }
 
   protected:
@@ -179,12 +179,12 @@ class FilterBitmapCheckerboardGM: public FilterBitmapGM {
 
 class FilterBitmapImageGM: public FilterBitmapGM {
   public:
-      FilterBitmapImageGM( const char filename[] )
-      : fFilename( filename )
+      FilterBitmapImageGM(const char filename[])
+      : fFilename(filename)
         {
             char name[1024];
-            sprintf( name, "filterbitmap_image_%s", filename );
-            setName( name );
+            sprintf(name, "filterbitmap_image_%s", filename);
+            setName(name);
         }
 
   protected:
@@ -197,7 +197,7 @@ class FilterBitmapImageGM: public FilterBitmapGM {
 
       void make_bitmap() SK_OVERRIDE {
           SkString path(skiagm::GM::gResourcePath);
-          path.append( "/" );
+          path.append("/");
           path.append(fFilename);
 
           SkImageDecoder *codec = NULL;
diff --git a/gm/scalebitmap.cpp b/gm/scalebitmap.cpp
new file mode 100644
index 0000000000..5d08ba85de
--- /dev/null
+++ b/gm/scalebitmap.cpp
@@ -0,0 +1,114 @@
+/*
+ * 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 "gm.h"
+
+#include "SkImageDecoder.h"
+#include "SkStream.h"
+
+class ScaleBitmapGM : public skiagm::GM {
+    
+public:
+    
+    ScaleBitmapGM(const char filename[], float scale)
+        : fFilename(filename), fScale(scale)
+    {
+        this->setBGColor(0xFFDDDDDD);
+        fName.printf("scalebitmap_%s_%f", filename, scale);
+        
+        SkString path(skiagm::GM::gResourcePath);
+        path.append("/");
+        path.append(fFilename);
+
+        SkImageDecoder *codec = NULL;
+        SkFILEStream stream(path.c_str());
+        if (stream.isValid()) {
+            codec = SkImageDecoder::Factory(&stream);
+        }
+        if (codec) {
+            stream.rewind();
+            codec->decode(&stream, &fBM, SkBitmap::kARGB_8888_Config,
+                SkImageDecoder::kDecodePixels_Mode);
+            SkDELETE(codec);
+        } else {
+            fBM.setConfig(SkBitmap::kARGB_8888_Config, 1, 1);
+            fBM.allocPixels();
+            *(fBM.getAddr32(0,0)) = 0xFF0000FF; // red == bad
+        }
+        fSize = fBM.height();
+    }
+
+protected:
+    
+    
+    SkBitmap    fBM;
+    SkString    fName;
+    SkString    fFilename;
+    int         fSize;
+    float       fScale;
+    
+    
+    virtual SkString onShortName() SK_OVERRIDE {
+        return fName;
+    }
+
+    virtual SkISize onISize() SK_OVERRIDE {
+        return SkISize::Make(fBM.width() * fScale, fBM.height() * fScale);
+    }
+
+    virtual void onDraw(SkCanvas* canvas) SK_OVERRIDE {
+        SkBitmap dst;
+        dst.setConfig(SkBitmap::kARGB_8888_Config, fBM.width() * fScale, fBM.height() * fScale);
+        dst.allocPixels();
+        fBM.scale(&dst);
+        
+        canvas->drawBitmap(dst, 0, 0);
+    }
+
+private:
+    typedef skiagm::GM INHERITED;
+};
+
+class ScaleBitmapMipmapGM: public ScaleBitmapGM {
+    SkMatrix fMatrix;
+    
+public:
+    ScaleBitmapMipmapGM(const char filename[], float scale)
+        : INHERITED(filename, scale)
+    {
+        fName.printf("scalebitmap_mipmap_%s_%f", filename, scale);
+        fBM.buildMipMap();
+        fMatrix.setScale(scale, scale);
+    }
+protected:
+    virtual void onDraw(SkCanvas *canvas) SK_OVERRIDE {
+        SkPaint paint;
+        
+        paint.setFilterBitmap(true);
+        canvas->drawBitmapMatrix(fBM, fMatrix, &paint);
+    }
+private:
+     typedef ScaleBitmapGM INHERITED;  
+};
+
+//////////////////////////////////////////////////////////////////////////////
+
+DEF_GM( return new ScaleBitmapGM("mandrill_128.png", 2); )
+DEF_GM( return new ScaleBitmapGM("mandrill_64.png", 4); )
+DEF_GM( return new ScaleBitmapGM("mandrill_32.png", 8); )
+DEF_GM( return new ScaleBitmapGM("mandrill_16.png", 16); )
+
+DEF_GM( return new ScaleBitmapGM("nature.jpg", 0.5f); )
+DEF_GM( return new ScaleBitmapGM("nature.jpg", 0.25f); )
+DEF_GM( return new ScaleBitmapGM("nature.jpg", 0.125f); )
+DEF_GM( return new ScaleBitmapGM("nature.jpg", 0.0625f); )
+
+DEF_GM( return new ScaleBitmapMipmapGM("nature.jpg", 0.5f); )
+DEF_GM( return new ScaleBitmapMipmapGM("nature.jpg", 0.25f); )
+DEF_GM( return new ScaleBitmapMipmapGM("nature.jpg", 0.125f); )
+DEF_GM( return new ScaleBitmapMipmapGM("nature.jpg", 0.0625f); )
+
diff --git a/gyp/bench.gypi b/gyp/bench.gypi
index 37e65526a7..956fec9d83 100644
--- a/gyp/bench.gypi
+++ b/gyp/bench.gypi
@@ -5,7 +5,6 @@
     '../bench/benchmain.cpp',
     '../bench/SkBenchmark.h',
     '../bench/SkBenchmark.cpp',
-
     '../bench/AAClipBench.cpp',
     '../bench/BicubicBench.cpp',
     '../bench/BitmapBench.cpp',
diff --git a/gyp/core.gypi b/gyp/core.gypi
index 270323150c..e87c0f4a0e 100644
--- a/gyp/core.gypi
+++ b/gyp/core.gypi
@@ -21,13 +21,14 @@
         '<(skia_src_path)/core/SkBBoxHierarchyRecord.cpp',
         '<(skia_src_path)/core/SkBBoxHierarchyRecord.h',
         '<(skia_src_path)/core/SkBitmap.cpp',
+        '<(skia_src_path)/core/SkBitmapFilter.h',
+        '<(skia_src_path)/core/SkBitmapFilter.cpp',
         '<(skia_src_path)/core/SkBitmapHeap.cpp',
         '<(skia_src_path)/core/SkBitmapHeap.h',
         '<(skia_src_path)/core/SkBitmapProcShader.cpp',
         '<(skia_src_path)/core/SkBitmapProcShader.h',
         '<(skia_src_path)/core/SkBitmapProcState.cpp',
         '<(skia_src_path)/core/SkBitmapProcState.h',
-        '<(skia_src_path)/core/SkBitmapProcBicubic.cpp',
         '<(skia_src_path)/core/SkBitmapProcState_matrix.h',
         '<(skia_src_path)/core/SkBitmapProcState_matrixProcs.cpp',
         '<(skia_src_path)/core/SkBitmapProcState_sample.h',
diff --git a/gyp/gmslides.gypi b/gyp/gmslides.gypi
index 480279a7f4..35d326eb84 100644
--- a/gyp/gmslides.gypi
+++ b/gyp/gmslides.gypi
@@ -90,6 +90,7 @@
     '../gm/rrects.cpp',
     '../gm/roundrects.cpp',
     '../gm/samplerstress.cpp',
+    # '../gm/scalebitmap.cpp',
     '../gm/shaderbounds.cpp',
     '../gm/selftest.cpp',
     '../gm/shadertext.cpp',
diff --git a/gyp/opts.gyp b/gyp/opts.gyp
index 2c87b81fc1..bcc85a9bae 100644
--- a/gyp/opts.gyp
+++ b/gyp/opts.gyp
@@ -43,9 +43,13 @@
               ],
             }],
           ],
+          'include_dirs': [
+            '../include/utils',
+          ],
           'sources': [
             '../src/opts/opts_check_SSE2.cpp',
             '../src/opts/SkBitmapProcState_opts_SSE2.cpp',
+            '../src/opts/SkBitmapFilter_opts_SSE2.cpp',
             '../src/opts/SkBlitRow_opts_SSE2.cpp',
             '../src/opts/SkBlitRect_opts_SSE2.cpp',
             '../src/opts/SkUtils_opts_SSE2.cpp',
diff --git a/include/core/SkBitmap.h b/include/core/SkBitmap.h
index 887169ccb5..d7f43274c6 100644
--- a/include/core/SkBitmap.h
+++ b/include/core/SkBitmap.h
@@ -538,7 +538,7 @@ public:
         by calling copyTo().
      */
     bool canCopyTo(Config newConfig) const;
-
+    
     /**
      *  DEPRECATED -- will be replaced with API on SkPaint
      */
@@ -702,6 +702,18 @@ private:
     int extractMipLevel(SkBitmap* dst, SkFixed sx, SkFixed sy);
     bool hasMipMap() const;
     void freeMipMap();
+    
+    /** Make a scaled copy of this bitmap into the provided destination.
+      * The caller is responsible for having set the width and height of the
+      * provided destination bitmap, and also having allocated its pixel 
+      * memory.
+      * 
+      * This function is temporary and for testing purposes only; it will 
+      * likely move once it has been properly plumbed into the bitmap 
+      * shader infrastructure.
+      */
+      
+    void scale(SkBitmap *dst) const;
 
     friend struct SkBitmapProcState;
 };
diff --git a/include/core/SkPaint.h b/include/core/SkPaint.h
index 9785213e88..2d7b3fa542 100644
--- a/include/core/SkPaint.h
+++ b/include/core/SkPaint.h
@@ -43,11 +43,14 @@ typedef const SkGlyph& (*SkDrawCacheProc)(SkGlyphCache*, const char**,
 
 typedef const SkGlyph& (*SkMeasureCacheProc)(SkGlyphCache*, const char**);
 
+#define kBicubicFilterBitmap_Flag kHighQualityFilterBitmap_Flag
+
 /** \class SkPaint
 
     The SkPaint class holds the style and color information about how to draw
     geometries, text and bitmaps.
 */
+
 class SK_API SkPaint {
 public:
     SkPaint();
@@ -108,7 +111,7 @@ public:
         kAutoHinting_Flag     = 0x800,  //!< mask to force Freetype's autohinter
         kVerticalText_Flag    = 0x1000,
         kGenA8FromLCD_Flag    = 0x2000, // hack for GDI -- do not use if you can help it
-        kBicubicFilterBitmap_Flag = 0x4000, // temporary flag
+        kHighQualityFilterBitmap_Flag = 0x4000, // temporary flag
         kHighQualityDownsampleBitmap_Flag = 0x8000, // temporary flag
 
         // when adding extra flags, note that the fFlags member is specified
diff --git a/src/core/SkBitmapFilter.cpp b/src/core/SkBitmapFilter.cpp
new file mode 100644
index 0000000000..79c0a98781
--- /dev/null
+++ b/src/core/SkBitmapFilter.cpp
@@ -0,0 +1,374 @@
+/*
+ * 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 "SkBitmapProcState.h"
+#include "SkBitmap.h"
+#include "SkColor.h"
+#include "SkColorPriv.h"
+#include "SkUnPreMultiply.h"
+#include "SkShader.h"
+#include "SkRTConf.h"
+#include "SkMath.h"
+
+void highQualityFilter(const SkBitmapProcState& s, int x, int y,
+                   SkPMColor* SK_RESTRICT colors, int count) {
+
+    const int maxX = s.fBitmap->width() - 1;
+    const int maxY = s.fBitmap->height() - 1;
+
+    while (count-- > 0) {
+        SkPoint srcPt;
+        s.fInvProc(*s.fInvMatrix, SkIntToScalar(x),
+                    SkIntToScalar(y), &srcPt);
+        srcPt.fX -= SK_ScalarHalf;
+        srcPt.fY -= SK_ScalarHalf;
+
+        int sx = SkScalarFloorToInt(srcPt.fX);
+        int sy = SkScalarFloorToInt(srcPt.fY);
+        
+        SkFixed weight = 0;
+        SkFixed fr = 0, fg = 0, fb = 0, fa = 0;
+        
+        int y0 = SkClampMax(int(ceil(sy-s.getBitmapFilter()->width() + 0.5f)), maxY);
+        int y1 = SkClampMax(int(floor(sy+s.getBitmapFilter()->width() + 0.5f)), maxY);
+        int x0 = SkClampMax(int(ceil(sx-s.getBitmapFilter()->width() + 0.5f)), maxX);
+        int x1 = SkClampMax(int(floor(sx+s.getBitmapFilter()->width() + 0.5f)), maxX);
+        
+        for (int src_y = y0; src_y <= y1; src_y++) {
+            SkFixed yweight = s.getBitmapFilter()->lookup((srcPt.fY - src_y));
+            
+            for (int src_x = x0; src_x <= x1 ; src_x++) {
+                SkFixed xweight = s.getBitmapFilter()->lookup((srcPt.fX - src_x));
+                
+                SkFixed combined_weight = SkFixedMul(xweight, yweight);
+                
+                SkPMColor c = *s.fBitmap->getAddr32(src_x, src_y);
+                fr += combined_weight * SkGetPackedR32(c);
+                fg += combined_weight * SkGetPackedG32(c);
+                fb += combined_weight * SkGetPackedB32(c);
+                fa += combined_weight * SkGetPackedA32(c);
+                weight += combined_weight;
+            }
+        }
+
+        fr = SkFixedDiv(fr, weight);
+        fg = SkFixedDiv(fg, weight);
+        fb = SkFixedDiv(fb, weight);
+        fa = SkFixedDiv(fa, weight);
+
+        int a = SkClampMax(SkFixedRoundToInt(fa), 255);
+        int r = SkClampMax(SkFixedRoundToInt(fr), a);
+        int g = SkClampMax(SkFixedRoundToInt(fg), a);
+        int b = SkClampMax(SkFixedRoundToInt(fb), a);
+
+        *colors++ = SkPackARGB32(a, r, g, b);
+
+        x++;
+    }
+}
+
+void highQualityFilter_ScaleOnly(const SkBitmapProcState &s, int x, int y,
+                             SkPMColor *SK_RESTRICT colors, int count) {
+     const int maxX = s.fBitmap->width() - 1;
+     const int maxY = s.fBitmap->height() - 1;
+     
+     SkPoint srcPt;
+     
+     s.fInvProc(*s.fInvMatrix, SkIntToScalar(x),
+                 SkIntToScalar(y), &srcPt);
+     srcPt.fY -= SK_ScalarHalf;
+     int sy = SkScalarFloorToInt(srcPt.fY);
+     int y0 = SkClampMax(int(ceil(sy-s.getBitmapFilter()->width() + 0.5f)), maxY);
+     int y1 = SkClampMax(int(floor(sy+s.getBitmapFilter()->width() + 0.5f)), maxY);
+
+     while (count-- > 0) {
+         s.fInvProc(*s.fInvMatrix, SkIntToScalar(x),
+                     SkIntToScalar(y), &srcPt);
+         srcPt.fX -= SK_ScalarHalf;
+         srcPt.fY -= SK_ScalarHalf;
+
+         int sx = SkScalarFloorToInt(srcPt.fX);
+
+         SkFixed weight = 0;
+         SkFixed fr = 0, fg = 0, fb = 0, fa = 0;
+
+         int x0 = SkClampMax(int(ceil(sx-s.getBitmapFilter()->width() + 0.5f)), maxX);
+         int x1 = SkClampMax(int(floor(sx+s.getBitmapFilter()->width() + 0.5f)), maxX);
+
+         for (int src_y = y0; src_y <= y1; src_y++) {
+             SkFixed yweight = s.getBitmapFilter()->lookup((srcPt.fY - src_y));
+
+             for (int src_x = x0; src_x <= x1 ; src_x++) {
+                 SkFixed xweight = s.getBitmapFilter()->lookup((srcPt.fX - src_x));
+
+                 SkFixed combined_weight = SkFixedMul(xweight, yweight);
+
+                 SkPMColor c = *s.fBitmap->getAddr32(src_x, src_y);
+                 fr += combined_weight * SkGetPackedR32(c);
+                 fg += combined_weight * SkGetPackedG32(c);
+                 fb += combined_weight * SkGetPackedB32(c);
+                 fa += combined_weight * SkGetPackedA32(c);
+                 weight += combined_weight;
+             }
+         }
+
+         fr = SkFixedDiv(fr, weight);
+         fg = SkFixedDiv(fg, weight);
+         fb = SkFixedDiv(fb, weight);
+         fa = SkFixedDiv(fa, weight);
+
+         int a = SkClampMax(SkFixedRoundToInt(fa), 255);
+         int r = SkClampMax(SkFixedRoundToInt(fr), a);
+         int g = SkClampMax(SkFixedRoundToInt(fg), a);
+         int b = SkClampMax(SkFixedRoundToInt(fb), a);
+
+         *colors++ = SkPackARGB32(a, r, g, b);
+
+         x++;
+     }                                 
+}
+
+SK_CONF_DECLARE(const char *, c_bitmapFilter, "bitmap.filter", "mitchell", "Which bitmap filter to use [mitchell, sinc, gaussian, triangle, box]");
+
+static SkBitmapFilter *allocateBitmapFilter() {
+    if (!strcmp(c_bitmapFilter, "mitchell")) {
+        return SkNEW_ARGS(SkMitchellFilter,(1.f/3.f,1.f/3.f));
+    } else if (!strcmp(c_bitmapFilter, "sinc")) {
+        return SkNEW_ARGS(SkSincFilter,(3));
+    } else if (!strcmp(c_bitmapFilter, "gaussian")) {
+        return SkNEW_ARGS(SkGaussianFilter,(2));
+    } else if (!strcmp(c_bitmapFilter, "triangle")) {
+        return SkNEW(SkTriangleFilter);
+    } else if (!strcmp(c_bitmapFilter, "box")) {
+        return SkNEW(SkBoxFilter);
+    } else {
+        SkASSERT(!!!"Unknown filter type");
+    }
+    
+    return NULL;
+}
+
+SkBitmapProcState::ShaderProc32
+SkBitmapProcState::chooseBitmapFilterProc(const SkPaint& paint) {
+    // we need to be requested
+    uint32_t mask = SkPaint::kFilterBitmap_Flag
+                  | SkPaint::kHighQualityFilterBitmap_Flag
+                  ;
+    if ((paint.getFlags() & mask) != mask) {
+        return NULL;
+    }
+
+    // TODO: consider supporting other configs (e.g. 565, A8)
+    if (fBitmap->config() != SkBitmap::kARGB_8888_Config) {
+        return NULL;
+    }
+
+    // TODO: consider supporting repeat and mirror
+    if (SkShader::kClamp_TileMode != fTileModeX || SkShader::kClamp_TileMode != fTileModeY) {
+        return NULL;
+    }
+
+    // TODO: support blending inside our procs
+    if (0xFF != paint.getAlpha()) {
+        return NULL;
+    }
+    
+    if (fInvType & (SkMatrix::kAffine_Mask | SkMatrix::kScale_Mask)) {
+        fBitmapFilter = allocateBitmapFilter();
+    }
+    
+    if (fInvType & SkMatrix::kAffine_Mask) {
+        return highQualityFilter;
+    } else if (fInvType & SkMatrix::kScale_Mask) {
+        return highQualityFilter_ScaleOnly;
+    } else {
+        return NULL;
+    }
+}
+
+static void divideByWeights(SkFixed *sums, SkFixed *weights, SkBitmap *dst) {
+    for (int y = 0 ; y < dst->height() ; y++) {
+        for (int x = 0 ; x < dst->width() ; x++) {
+            SkFixed fr = SkFixedDiv(sums[4*(y*dst->width() + x) + 0], weights[y*dst->width() + x]);
+            SkFixed fg = SkFixedDiv(sums[4*(y*dst->width() + x) + 1], weights[y*dst->width() + x]);
+            SkFixed fb = SkFixedDiv(sums[4*(y*dst->width() + x) + 2], weights[y*dst->width() + x]);
+            SkFixed fa = SkFixedDiv(sums[4*(y*dst->width() + x) + 3], weights[y*dst->width() + x]);
+            int a = SkClampMax(SkFixedRoundToInt(fa), 255);
+            int r = SkClampMax(SkFixedRoundToInt(fr), a);
+            int g = SkClampMax(SkFixedRoundToInt(fg), a);
+            int b = SkClampMax(SkFixedRoundToInt(fb), a);
+
+            *dst->getAddr32(x,y) = SkPackARGB32(a, r, g, b);
+        }
+    }
+}
+
+static void upScaleHoriz(const SkBitmap *src, SkBitmap *dst, float scale, SkBitmapFilter *filter) {
+    for (int y = 0 ; y < src->height() ; y++) {
+        for (int x = 0 ; x < dst->width() ; x++) {
+            float sx = x / scale - 0.5f;
+            int x0 = SkClampMax(int(ceil(sx-filter->width() + 0.5f)), src->width()-1);
+            int x1 = SkClampMax(int(floor(sx+filter->width() + 0.5f)), src->width()-1);
+            
+            SkFixed total_weight = 0;
+            SkFixed fr = 0, fg = 0, fb = 0, fa = 0;
+            
+            for (int src_x = x0 ; src_x <= x1 ; src_x++) {
+                SkFixed weight = filter->lookup(sx - src_x);
+                SkPMColor c = *src->getAddr32(src_x,y);
+                fr += weight * SkGetPackedR32(c);
+                fg += weight * SkGetPackedG32(c);
+                fb += weight * SkGetPackedB32(c);
+                fa += weight * SkGetPackedA32(c);
+                total_weight += weight;
+            }
+            fr = SkFixedDiv(fr, total_weight);
+            fg = SkFixedDiv(fg, total_weight);
+            fb = SkFixedDiv(fb, total_weight);
+            fa = SkFixedDiv(fa, total_weight);
+
+            int a = SkClampMax(SkFixedRoundToInt(fa), 255);
+            int r = SkClampMax(SkFixedRoundToInt(fr), a);
+            int g = SkClampMax(SkFixedRoundToInt(fg), a);
+            int b = SkClampMax(SkFixedRoundToInt(fb), a);
+
+            *dst->getAddr32(x,y) = SkPackARGB32(a, r, g, b);
+        }
+    }
+}
+
+static void downScaleHoriz(const SkBitmap *src, SkBitmap *dst, float scale, SkBitmapFilter *filter) {
+    SkFixed *sums = SkNEW_ARRAY(SkFixed, dst->width() * dst->height() * 4);
+    SkFixed *weights = SkNEW_ARRAY(SkFixed, dst->width() * dst->height());
+    
+    SkAutoTDeleteArray<SkFixed> ada1(sums);
+    SkAutoTDeleteArray<SkFixed> ada2(weights);
+
+    memset(sums, 0, dst->width() * dst->height() * sizeof(SkFixed) * 4);
+    memset(weights, 0, dst->width() * dst->height() * sizeof(SkFixed));
+    
+    for (int y = 0 ; y < src->height() ; y++) {
+        for (int x = 0 ; x < src->width() ; x++) {
+            // splat each source pixel into the destination image
+            float dx = (x + 0.5f) * scale;
+            int x0 = SkClampMax(int(ceil(dx-filter->width() + 0.5f)), dst->width()-1);
+            int x1 = SkClampMax(int(floor(dx+filter->width() + 0.5f)), dst->width()-1);
+            
+            SkPMColor c = *src->getAddr32(x,y);
+            
+            for (int dst_x = x0 ; dst_x <= x1 ; dst_x++) {
+                SkFixed weight = filter->lookup(dx - dst_x);
+                sums[4*(y*dst->width() + dst_x) + 0] += weight*SkGetPackedR32(c);
+                sums[4*(y*dst->width() + dst_x) + 1] += weight*SkGetPackedG32(c);
+                sums[4*(y*dst->width() + dst_x) + 2] += weight*SkGetPackedB32(c);
+                sums[4*(y*dst->width() + dst_x) + 3] += weight*SkGetPackedA32(c);
+                weights[y*dst->width() + dst_x] += weight;
+            }
+        }
+    }
+    
+    divideByWeights(sums, weights, dst);
+}
+
+static void upScaleVert(const SkBitmap *src, SkBitmap *dst, float scale, SkBitmapFilter *filter) {
+    for (int y = 0 ; y < dst->height() ; y++) {
+        for (int x = 0 ; x < dst->width() ; x++) {
+            float sy = y / scale - 0.5f;
+            int y0 = SkClampMax(int(ceil(sy-filter->width() + 0.5f)), src->height()-1);
+            int y1 = SkClampMax(int(floor(sy+filter->width() + 0.5f)), src->height()-1);
+            
+            SkFixed total_weight = 0;
+            SkFixed fr = 0, fg = 0, fb = 0, fa = 0;
+            
+            for (int src_y = y0 ; src_y <= y1 ; src_y++) {
+                SkFixed weight = filter->lookup(sy - src_y);
+                SkPMColor c = *src->getAddr32(x,src_y);
+                fr += weight * SkGetPackedR32(c);
+                fg += weight * SkGetPackedG32(c);
+                fb += weight * SkGetPackedB32(c);
+                fa += weight * SkGetPackedA32(c);
+                total_weight += weight;
+            }
+            fr = SkFixedDiv(fr, total_weight);
+            fg = SkFixedDiv(fg, total_weight);
+            fb = SkFixedDiv(fb, total_weight);
+            fa = SkFixedDiv(fa, total_weight);
+
+            int a = SkClampMax(SkFixedRoundToInt(fa), 255);
+            int r = SkClampMax(SkFixedRoundToInt(fr), a);
+            int g = SkClampMax(SkFixedRoundToInt(fg), a);
+            int b = SkClampMax(SkFixedRoundToInt(fb), a);
+
+            *dst->getAddr32(x,y) = SkPackARGB32(a, r, g, b);
+        }
+    }
+}
+
+static void downScaleVert(const SkBitmap *src, SkBitmap *dst, float scale, SkBitmapFilter *filter) {
+    SkFixed *sums = SkNEW_ARRAY(SkFixed, dst->width() * dst->height() * 4);
+    SkFixed *weights = SkNEW_ARRAY(SkFixed, dst->width() * dst->height());
+    
+    SkAutoTDeleteArray<SkFixed> ada1(sums);
+    SkAutoTDeleteArray<SkFixed> ada2(weights);
+
+    memset(sums, 0, dst->width() * dst->height() * sizeof(SkFixed) * 4);
+    memset(weights, 0, dst->width() * dst->height() * sizeof(SkFixed));
+
+    for (int y = 0 ; y < src->height() ; y++) {
+        for (int x = 0 ; x < src->width() ; x++) {
+            // splat each source pixel into the destination image
+            float dy = (y + 0.5f) * scale;
+            int y0 = SkClampMax(int(ceil(dy-filter->width() + 0.5f)), dst->height()-1);
+            int y1 = SkClampMax(int(floor(dy+filter->width() + 0.5f)), dst->height()-1);
+
+            SkPMColor c = *src->getAddr32(x,y);
+
+            for (int dst_y = y0 ; dst_y <= y1 ; dst_y++) {
+                SkFixed weight = filter->lookup(dy - dst_y);
+                sums[4*(dst_y*dst->width() + x) + 0] += weight*SkGetPackedR32(c);
+                sums[4*(dst_y*dst->width() + x) + 1] += weight*SkGetPackedG32(c);
+                sums[4*(dst_y*dst->width() + x) + 2] += weight*SkGetPackedB32(c);
+                sums[4*(dst_y*dst->width() + x) + 3] += weight*SkGetPackedA32(c);
+                weights[dst_y*dst->width() + x] += weight;
+            }
+        }
+    }
+    
+    divideByWeights(sums, weights, dst);
+}
+
+void SkBitmap::scale(SkBitmap *dst) const {
+    
+    SkBitmap horiz_temp;
+    
+    horiz_temp.setConfig(SkBitmap::kARGB_8888_Config, dst->width(), height());
+    horiz_temp.allocPixels();
+    
+    SkBitmapFilter *filter = allocateBitmapFilter();
+    
+    float horiz_scale = float(dst->width()) / width();
+    
+    if (horiz_scale == 1) {
+        this->copyPixelsTo(horiz_temp.getPixels(), getSize());
+    } else if (horiz_scale > 1) {
+        upScaleHoriz(this, &horiz_temp, horiz_scale, filter);
+    } else if (horiz_scale < 1) {
+        downScaleHoriz(this, &horiz_temp, horiz_scale, filter);
+    }
+    
+    float vert_scale = float(dst->height()) / height();
+    
+    if (vert_scale == 1) {
+        horiz_temp.copyPixelsTo(dst->getPixels(), dst->getSize());
+    } else if (vert_scale > 1) {
+        upScaleVert(&horiz_temp, dst, vert_scale, filter);
+    } else if (vert_scale < 1) {
+        downScaleVert(&horiz_temp, dst, vert_scale, filter);
+    }
+    
+    SkDELETE(filter);
+}
diff --git a/src/core/SkBitmapFilter.h b/src/core/SkBitmapFilter.h
new file mode 100644
index 0000000000..4a9f78a4a7
--- /dev/null
+++ b/src/core/SkBitmapFilter.h
@@ -0,0 +1,150 @@
+
+/*
+ * Copyright 2013 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+
+#ifndef SkBitmapFilter_DEFINED
+#define SkBitmapFilter_DEFINED
+
+#include "SkMath.h"
+
+// size of the precomputed bitmap filter tables for high quality filtering.
+// Used to precompute the shape of the filter kernel.
+// Table size chosen from experiments to see where I could start to see a difference.
+
+#define SKBITMAP_FILTER_TABLE_SIZE 32
+
+class SkBitmapFilter {
+  public:
+      SkBitmapFilter(float width)
+      : fWidth(width), fInvWidth(1.f/width) {
+          precomputed = false;
+      }
+      
+      SkFixed lookup( float x ) const {
+          if (!precomputed) {
+              precomputeTable();
+          }
+          int filter_idx = int(fabsf(x * invWidth() * SKBITMAP_FILTER_TABLE_SIZE));
+          return fFilterTable[ SkTMin(filter_idx, SKBITMAP_FILTER_TABLE_SIZE-1) ];
+      }      
+      
+      float lookupFloat( float x ) const {
+          if (!precomputed) {
+              precomputeTable();
+          }
+          int filter_idx = int(fabsf(x * invWidth() * SKBITMAP_FILTER_TABLE_SIZE));
+          return fFilterTableFloat[ SkTMin(filter_idx, SKBITMAP_FILTER_TABLE_SIZE-1) ];
+      }      
+      
+      float width() const { return fWidth; }
+      float invWidth() const { return fInvWidth; }
+      virtual float evaluate(float x) const = 0;
+      virtual ~SkBitmapFilter() {}
+  protected:
+      float fWidth;
+      float fInvWidth;
+      
+      mutable bool precomputed;
+      mutable SkFixed fFilterTable[SKBITMAP_FILTER_TABLE_SIZE];
+      mutable float fFilterTableFloat[SKBITMAP_FILTER_TABLE_SIZE];
+  private:
+      void precomputeTable() const {
+          precomputed = true;
+          SkFixed *ftp = fFilterTable;
+          float *ftp_float = fFilterTableFloat;
+          for (int x = 0; x < SKBITMAP_FILTER_TABLE_SIZE; ++x) {
+              float fx = ((float)x + .5f) * this->width() / SKBITMAP_FILTER_TABLE_SIZE;
+              float filter_value = evaluate(fx);
+              *ftp_float++ = filter_value;
+              *ftp++ = SkFloatToFixed(filter_value);
+          }
+      }
+};
+
+class SkMitchellFilter: public SkBitmapFilter {
+  public:
+      SkMitchellFilter(float b, float c, float width=2.0f)
+      : SkBitmapFilter(width), B(b), C(c) { 
+      }
+      
+      virtual float evaluate(float x) const SK_OVERRIDE {
+          x = fabsf(x);
+          if (x > 2.f) {
+              return 0;
+          } else if (x > 1.f) {
+              return ((-B - 6*C) * x*x*x + (6*B + 30*C) * x*x +
+                      (-12*B - 48*C) * x + (8*B + 24*C)) * (1.f/6.f);
+          } else {
+              return ((12 - 9*B - 6*C) * x*x*x +
+                      (-18 + 12*B + 6*C) * x*x +
+                      (6 - 2*B)) * (1.f/6.f);
+          }
+      }
+  protected:
+      float B, C;
+};
+
+class SkGaussianFilter: public SkBitmapFilter {
+  public:
+      SkGaussianFilter(float a, float width=2.0f)
+      : SkBitmapFilter(width), alpha(a), expWidth(expf(-alpha * width * width)) { 
+      }
+      
+      virtual float evaluate(float x) const SK_OVERRIDE {
+          return SkTMax(0.f, float(expf(-alpha*x*x) - expWidth));
+      }
+  protected:
+      float alpha, expWidth;
+};
+
+class SkTriangleFilter: public SkBitmapFilter {
+  public:
+      SkTriangleFilter(float width=1)
+      : SkBitmapFilter(width) { 
+      }
+      
+      virtual float evaluate(float x) const SK_OVERRIDE {
+          return SkTMax(0.f, fWidth - fabsf(x));
+      }
+  protected:
+};
+
+class SkBoxFilter: public SkBitmapFilter {
+  public:
+      SkBoxFilter(float width=0.5f)
+      : SkBitmapFilter(width) { 
+      }
+      
+      virtual float evaluate(float x) const SK_OVERRIDE {
+          return 1;
+      }
+  protected:
+};
+
+
+class SkSincFilter: public SkBitmapFilter {
+  public:
+      SkSincFilter(float t, float width=3.f)
+      : SkBitmapFilter(width), tau(t) { 
+      }
+      
+      virtual float evaluate(float x) const SK_OVERRIDE {
+          x = sk_float_abs(x * fInvWidth);
+          if (x < 1e-5f) return 1.f;
+          if (x > 1.f)   return 0.f;
+          x *= (float) M_PI;
+          float sinc = sk_float_sin(x) / x;
+          float lanczos = sk_float_sin(x * tau) / (x * tau);
+          return sinc * lanczos;
+      }
+  protected:
+      float tau;
+};
+
+
+#endif
diff --git a/src/core/SkBitmapProcBicubic.cpp b/src/core/SkBitmapProcBicubic.cpp
deleted file mode 100644
index f2cd0e30e7..0000000000
--- a/src/core/SkBitmapProcBicubic.cpp
+++ /dev/null
@@ -1,187 +0,0 @@
-/*
- * 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 "SkBitmapProcState.h"
-#include "SkBitmap.h"
-#include "SkColor.h"
-#include "SkColorPriv.h"
-#include "SkUnPreMultiply.h"
-#include "SkRTConf.h"
-#include "SkShader.h"
-
-#define DS(x) SkDoubleToScalar(x)
-
-#define MUL(a, b)   ((a) * (b))
-
-static inline SkPMColor cubicBlend(const SkFixed cc[4], SkPMColor c0, SkPMColor c1, SkPMColor c2, SkPMColor c3) {
-    SkFixed fa = MUL(cc[0], SkGetPackedA32(c0)) + MUL(cc[1], SkGetPackedA32(c1)) + MUL(cc[2], SkGetPackedA32(c2)) + MUL(cc[3], SkGetPackedA32(c3));
-    SkFixed fr = MUL(cc[0], SkGetPackedR32(c0)) + MUL(cc[1], SkGetPackedR32(c1)) + MUL(cc[2], SkGetPackedR32(c2)) + MUL(cc[3], SkGetPackedR32(c3));
-    SkFixed fg = MUL(cc[0], SkGetPackedG32(c0)) + MUL(cc[1], SkGetPackedG32(c1)) + MUL(cc[2], SkGetPackedG32(c2)) + MUL(cc[3], SkGetPackedG32(c3));
-    SkFixed fb = MUL(cc[0], SkGetPackedB32(c0)) + MUL(cc[1], SkGetPackedB32(c1)) + MUL(cc[2], SkGetPackedB32(c2)) + MUL(cc[3], SkGetPackedB32(c3));
-
-    int a = SkClampMax(SkFixedRoundToInt(fa), 255);
-    int r = SkClampMax(SkFixedRoundToInt(fr), a);
-    int g = SkClampMax(SkFixedRoundToInt(fg), a);
-    int b = SkClampMax(SkFixedRoundToInt(fb), a);
-
-    return SkPackARGB32(a, r, g, b);
-}
-
-static float poly_eval(const float cc[4], float t) {
-    return cc[0] + t * (cc[1] + t * (cc[2] + t * cc[3]));
-}
-
-static void build_coeff4(SkFixed dst[4], float t) {
-    static const SkScalar coefficients[16] = {
-        DS( 1.0 / 18.0), DS(-9.0 / 18.0), DS( 15.0 / 18.0), DS( -7.0 / 18.0),
-        DS(16.0 / 18.0), DS( 0.0 / 18.0), DS(-36.0 / 18.0), DS( 21.0 / 18.0),
-        DS( 1.0 / 18.0), DS( 9.0 / 18.0), DS( 27.0 / 18.0), DS(-21.0 / 18.0),
-        DS( 0.0 / 18.0), DS( 0.0 / 18.0), DS( -6.0 / 18.0), DS(  7.0 / 18.0),
-    };
-
-    dst[0] = SkFloatToFixed(poly_eval(&coefficients[ 0], t));
-    dst[1] = SkFloatToFixed(poly_eval(&coefficients[ 4], t));
-    dst[2] = SkFloatToFixed(poly_eval(&coefficients[ 8], t));
-    dst[3] = SkFloatToFixed(poly_eval(&coefficients[12], t));
-}
-
-static SkPMColor doBicubicFilter(const SkBitmap *bm, SkFixed coeffX[4], SkFixed coeffY[4],
-                            int x0, int x1, int x2, int x3,
-                            int y0, int y1, int y2, int y3 )
-{
-    SkPMColor s00 = *bm->getAddr32(x0, y0);
-    SkPMColor s10 = *bm->getAddr32(x1, y0);
-    SkPMColor s20 = *bm->getAddr32(x2, y0);
-    SkPMColor s30 = *bm->getAddr32(x3, y0);
-    SkPMColor s0 = cubicBlend(coeffX, s00, s10, s20, s30);
-    SkPMColor s01 = *bm->getAddr32(x0, y1);
-    SkPMColor s11 = *bm->getAddr32(x1, y1);
-    SkPMColor s21 = *bm->getAddr32(x2, y1);
-    SkPMColor s31 = *bm->getAddr32(x3, y1);
-    SkPMColor s1 = cubicBlend(coeffX, s01, s11, s21, s31);
-    SkPMColor s02 = *bm->getAddr32(x0, y2);
-    SkPMColor s12 = *bm->getAddr32(x1, y2);
-    SkPMColor s22 = *bm->getAddr32(x2, y2);
-    SkPMColor s32 = *bm->getAddr32(x3, y2);
-    SkPMColor s2 = cubicBlend(coeffX, s02, s12, s22, s32);
-    SkPMColor s03 = *bm->getAddr32(x0, y3);
-    SkPMColor s13 = *bm->getAddr32(x1, y3);
-    SkPMColor s23 = *bm->getAddr32(x2, y3);
-    SkPMColor s33 = *bm->getAddr32(x3, y3);
-    SkPMColor s3 = cubicBlend(coeffX, s03, s13, s23, s33);
-    return cubicBlend(coeffY, s0, s1, s2, s3);
-}
-
-static void bicubicFilter(const SkBitmapProcState& s, int x, int y,
-                          SkPMColor* SK_RESTRICT colors, int count) {
-
-    const int maxX = s.fBitmap->width() - 1;
-    const int maxY = s.fBitmap->height() - 1;
-
-    while (count-- > 0) {
-        SkPoint srcPt;
-        s.fInvProc(*s.fInvMatrix, SkIntToScalar(x),
-                    SkIntToScalar(y), &srcPt);
-        srcPt.fX -= SK_ScalarHalf;
-        srcPt.fY -= SK_ScalarHalf;
-        SkScalar fractx = srcPt.fX - SkScalarFloorToScalar(srcPt.fX);
-        SkScalar fracty = srcPt.fY - SkScalarFloorToScalar(srcPt.fY);
-
-        SkFixed coeffX[4], coeffY[4];
-        build_coeff4(coeffX, fractx);
-        build_coeff4(coeffY, fracty);
-
-        int sx = SkScalarFloorToInt(srcPt.fX);
-        int sy = SkScalarFloorToInt(srcPt.fY);
-
-        // Here is where we can support other tile modes (e.g. repeat or mirror)
-        int x0 = SkClampMax(sx - 1, maxX);
-        int x1 = SkClampMax(sx    , maxX);
-        int x2 = SkClampMax(sx + 1, maxX);
-        int x3 = SkClampMax(sx + 2, maxX);
-        int y0 = SkClampMax(sy - 1, maxY);
-        int y1 = SkClampMax(sy    , maxY);
-        int y2 = SkClampMax(sy + 1, maxY);
-        int y3 = SkClampMax(sy + 2, maxY);
-
-        *colors++ = doBicubicFilter( s.fBitmap, coeffX, coeffY, x0, x1, x2, x3, y0, y1, y2, y3 );
-
-        x++;
-    }
-}
-
-static void bicubicFilter_ScaleOnly(const SkBitmapProcState &s, int x, int y,
-                                    SkPMColor *SK_RESTRICT colors, int count) {
-    const int maxX = s.fBitmap->width() - 1;
-    const int maxY = s.fBitmap->height() - 1;
-
-    SkPoint srcPt;
-    s.fInvProc(*s.fInvMatrix, SkIntToScalar(x), SkIntToScalar(y), &srcPt);
-    srcPt.fY -= SK_ScalarHalf;
-    SkScalar fracty = srcPt.fY - SkScalarFloorToScalar(srcPt.fY);
-    SkFixed coeffX[4], coeffY[4];
-    build_coeff4(coeffY, fracty);
-    int sy = SkScalarFloorToInt(srcPt.fY);
-    int y0 = SkClampMax(sy - 1, maxY);
-    int y1 = SkClampMax(sy    , maxY);
-    int y2 = SkClampMax(sy + 1, maxY);
-    int y3 = SkClampMax(sy + 2, maxY);
-
-    while (count-- > 0) {
-        s.fInvProc(*s.fInvMatrix, SkIntToScalar(x), SkIntToScalar(y), &srcPt);
-        srcPt.fX -= SK_ScalarHalf;
-        SkScalar fractx = srcPt.fX - SkScalarFloorToScalar(srcPt.fX);
-
-        build_coeff4(coeffX, fractx);
-
-        int sx = SkScalarFloorToInt(srcPt.fX);
-
-        // Here is where we can support other tile modes (e.g. repeat or mirror)
-        int x0 = SkClampMax(sx - 1, maxX);
-        int x1 = SkClampMax(sx    , maxX);
-        int x2 = SkClampMax(sx + 1, maxX);
-        int x3 = SkClampMax(sx + 2, maxX);
-
-        *colors++ = doBicubicFilter( s.fBitmap, coeffX, coeffY, x0, x1, x2, x3, y0, y1, y2, y3 );
-
-        x++;
-    }
-}
-
-SkBitmapProcState::ShaderProc32
-SkBitmapProcState::chooseBicubicFilterProc(const SkPaint& paint) {
-    // we need to be requested
-    uint32_t mask = SkPaint::kFilterBitmap_Flag
-                  | SkPaint::kBicubicFilterBitmap_Flag
-                  ;
-    if ((paint.getFlags() & mask) != mask) {
-        return NULL;
-    }
-
-    // TODO: consider supporting other configs (e.g. 565, A8)
-    if (fBitmap->config() != SkBitmap::kARGB_8888_Config) {
-        return NULL;
-    }
-
-    // TODO: consider supporting repeat and mirror
-    if (SkShader::kClamp_TileMode != fTileModeX || SkShader::kClamp_TileMode != fTileModeY) {
-        return NULL;
-    }
-
-    // TODO: support blending inside our procs
-    if (0xFF != paint.getAlpha()) {
-        return NULL;
-    }
-
-    if (fInvType & SkMatrix::kAffine_Mask) {
-        return bicubicFilter;
-    } else if (fInvType & SkMatrix::kScale_Mask) {
-        return bicubicFilter_ScaleOnly;
-    } else {
-        return NULL;
-    }
-}
diff --git a/src/core/SkBitmapProcState.cpp b/src/core/SkBitmapProcState.cpp
index 2a79f2f490..a2c7f990c2 100644
--- a/src/core/SkBitmapProcState.cpp
+++ b/src/core/SkBitmapProcState.cpp
@@ -302,13 +302,13 @@ bool SkBitmapProcState::chooseProcs(const SkMatrix& inv, const SkPaint& paint) {
         fShaderProc32 = this->chooseShaderProc32();
     }
 
-    // see if our platform has any accelerated overrides
-    this->platformProcs();
-
     if (NULL == fShaderProc32) {
-        fShaderProc32 = this->chooseBicubicFilterProc(paint);
+        fShaderProc32 = this->chooseBitmapFilterProc(paint);
     }
 
+    // see if our platform has any accelerated overrides
+    this->platformProcs();
+
     return true;
 }
 
diff --git a/src/core/SkBitmapProcState.h b/src/core/SkBitmapProcState.h
index 7d754fe509..69de2ca86a 100644
--- a/src/core/SkBitmapProcState.h
+++ b/src/core/SkBitmapProcState.h
@@ -11,6 +11,7 @@
 #define SkBitmapProcState_DEFINED
 
 #include "SkBitmap.h"
+#include "SkBitmapFilter.h"
 #include "SkMatrix.h"
 
 #define FractionalInt_IS_64BIT
@@ -113,6 +114,8 @@ struct SkBitmapProcState {
     // are ignored
     ShaderProc32 getShaderProc32() const { return fShaderProc32; }
     ShaderProc16 getShaderProc16() const { return fShaderProc16; }
+    
+    SkBitmapFilter* getBitmapFilter() const { return fBitmapFilter; }
 
 #ifdef SK_DEBUG
     MatrixProc getMatrixProc() const;
@@ -139,12 +142,11 @@ private:
     MatrixProc chooseMatrixProc(bool trivial_matrix);
     bool chooseProcs(const SkMatrix& inv, const SkPaint&);
     ShaderProc32 chooseShaderProc32();
+    
+    void buildFilterCoefficients(SkFixed dst[4], float t) const;
+    SkBitmapFilter *fBitmapFilter;
 
-
-    /** test method for choosing a bicubic shading filter
-      */
-
-    ShaderProc32 chooseBicubicFilterProc(const SkPaint &paint);
+    ShaderProc32 chooseBitmapFilterProc(const SkPaint &paint);
 
     // Return false if we failed to setup for fast translate (e.g. overflow)
     bool setupForTranslate();
@@ -200,4 +202,10 @@ void ClampX_ClampY_nofilter_affine(const SkBitmapProcState& s,
 void S32_D16_filter_DX(const SkBitmapProcState& s,
                                    const uint32_t* xy, int count, uint16_t* colors);
 
+void highQualityFilter_ScaleOnly(const SkBitmapProcState &s, int x, int y,
+                             SkPMColor *SK_RESTRICT colors, int count);
+void highQualityFilter(const SkBitmapProcState &s, int x, int y,
+                   SkPMColor *SK_RESTRICT colors, int count);
+                              
+
 #endif
diff --git a/src/opts/SkBitmapFilter_opts_SSE2.cpp b/src/opts/SkBitmapFilter_opts_SSE2.cpp
new file mode 100644
index 0000000000..b1a07f2844
--- /dev/null
+++ b/src/opts/SkBitmapFilter_opts_SSE2.cpp
@@ -0,0 +1,182 @@
+/*
+ * 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 "SkBitmapProcState.h"
+#include "SkBitmap.h"
+#include "SkColor.h"
+#include "SkColorPriv.h"
+#include "SkUnPreMultiply.h"
+#include "SkShader.h"
+
+#include "SkBitmapFilter_opts_SSE2.h"
+
+#include <emmintrin.h>
+
+#if 0
+static inline void print128i(__m128i value) {
+    int *v = (int*) &value;
+    printf("% .11d % .11d % .11d % .11d\n", v[0], v[1], v[2], v[3]);
+}
+
+static inline void print128i_16(__m128i value) {
+    short *v = (short*) &value;
+    printf("% .5d % .5d % .5d % .5d % .5d % .5d % .5d % .5d\n", v[0], v[1], v[2], v[3], v[4], v[5], v[6], v[7]);
+}
+
+static inline void print128i_8(__m128i value) {
+    unsigned char *v = (unsigned char*) &value;
+    printf("%.3u %.3u %.3u %.3u %.3u %.3u %.3u %.3u %.3u %.3u %.3u %.3u %.3u %.3u %.3u %.3u\n", 
+           v[0], v[1], v[2], v[3], v[4], v[5], v[6], v[7],
+           v[8], v[9], v[10], v[11], v[12], v[13], v[14], v[15]
+           );
+}
+
+static inline void print128f(__m128 value) {
+    float *f = (float*) &value;
+    printf("%3.4f %3.4f %3.4f %3.4f\n", f[0], f[1], f[2], f[3]);
+}
+#endif
+
+// because the border is handled specially, this is guaranteed to have all 16 pixels
+// available to it without running off the bitmap's edge.
+
+int debug_x = 20;
+int debug_y = 255;
+
+void highQualityFilter_SSE2(const SkBitmapProcState& s, int x, int y,
+                        SkPMColor* SK_RESTRICT colors, int count) {
+    
+    const int maxX = s.fBitmap->width() - 1;
+    const int maxY = s.fBitmap->height() - 1;
+    
+    while (count-- > 0) {
+        SkPoint srcPt;
+        s.fInvProc(*s.fInvMatrix, SkIntToScalar(x),
+                    SkIntToScalar(y), &srcPt);
+        srcPt.fX -= SK_ScalarHalf;
+        srcPt.fY -= SK_ScalarHalf;
+
+        int sx = SkScalarFloorToInt(srcPt.fX);
+        int sy = SkScalarFloorToInt(srcPt.fY);
+        
+        __m128 weight = _mm_setzero_ps();
+        __m128 accum = _mm_setzero_ps();
+        
+        int y0 = SkTMax(0, int(ceil(sy-s.getBitmapFilter()->width() + 0.5f)));
+        int y1 = SkTMin(maxY, int(floor(sy+s.getBitmapFilter()->width() + 0.5f)));
+        int x0 = SkTMax(0, int(ceil(sx-s.getBitmapFilter()->width() + 0.5f)));
+        int x1 = SkTMin(maxX, int(floor(sx+s.getBitmapFilter()->width() + 0.5f)));
+        
+        for (int src_y = y0; src_y <= y1; src_y++) {
+            float yweight = s.getBitmapFilter()->lookupFloat( (srcPt.fY - src_y) );
+            
+            for (int src_x = x0; src_x <= x1 ; src_x++) {
+                float xweight = s.getBitmapFilter()->lookupFloat( (srcPt.fX - src_x) );
+                
+                float combined_weight = xweight * yweight;
+
+                SkPMColor color = *s.fBitmap->getAddr32(src_x, src_y);
+                
+                __m128i c = _mm_cvtsi32_si128( color );
+                c = _mm_unpacklo_epi8(c, _mm_setzero_si128());
+                c = _mm_unpacklo_epi16(c, _mm_setzero_si128());
+                
+                __m128 cfloat = _mm_cvtepi32_ps( c );
+                
+                __m128 weightVector = _mm_set1_ps(combined_weight);
+                
+                accum = _mm_add_ps(accum, _mm_mul_ps(cfloat, weightVector));
+                weight = _mm_add_ps( weight, weightVector );
+            }
+        }
+
+        accum = _mm_div_ps(accum, weight);
+        accum = _mm_add_ps(accum, _mm_set1_ps(0.5f));
+        
+        __m128i accumInt = _mm_cvtps_epi32( accum );
+        
+        int localResult[4];
+        _mm_storeu_si128((__m128i *) (localResult), accumInt);
+        int a = SkClampMax(localResult[0], 255);
+        int r = SkClampMax(localResult[1], a);
+        int g = SkClampMax(localResult[2], a);
+        int b = SkClampMax(localResult[3], a);
+        
+        *colors++ = SkPackARGB32(a, r, g, b);
+
+        x++;
+    }
+}
+
+void highQualityFilter_ScaleOnly_SSE2(const SkBitmapProcState &s, int x, int y,
+                             SkPMColor *SK_RESTRICT colors, int count) {
+    const int maxX = s.fBitmap->width() - 1;
+    const int maxY = s.fBitmap->height() - 1;
+    
+    SkPoint srcPt;
+    s.fInvProc(*s.fInvMatrix, SkIntToScalar(x),
+                SkIntToScalar(y), &srcPt);
+    srcPt.fY -= SK_ScalarHalf;
+    int sy = SkScalarFloorToInt(srcPt.fY);
+        
+    int y0 = SkTMax(0, int(ceil(sy-s.getBitmapFilter()->width() + 0.5f)));
+    int y1 = SkTMin(maxY, int(floor(sy+s.getBitmapFilter()->width() + 0.5f)));
+    
+    while (count-- > 0) {
+        srcPt.fX -= SK_ScalarHalf;
+        srcPt.fY -= SK_ScalarHalf;
+        
+        int sx = SkScalarFloorToInt(srcPt.fX);
+        
+        float weight = 0;
+        __m128 accum = _mm_setzero_ps();
+        
+        int x0 = SkTMax(0, int(ceil(sx-s.getBitmapFilter()->width() + 0.5f)));
+        int x1 = SkTMin(maxX, int(floor(sx+s.getBitmapFilter()->width() + 0.5f)));
+        
+        for (int src_y = y0; src_y <= y1; src_y++) {
+            float yweight = s.getBitmapFilter()->lookupFloat( (srcPt.fY - src_y) );
+        
+            for (int src_x = x0; src_x <= x1 ; src_x++) {
+                float xweight = s.getBitmapFilter()->lookupFloat( (srcPt.fX - src_x) );
+        
+                float combined_weight = xweight * yweight;
+        
+                SkPMColor color = *s.fBitmap->getAddr32(src_x, src_y);
+        
+                __m128 c = _mm_set_ps(SkGetPackedB32(color), 
+                                      SkGetPackedG32(color),
+                                      SkGetPackedR32(color),
+                                      SkGetPackedA32(color));
+        
+                __m128 weightVector = _mm_set1_ps(combined_weight);
+        
+                accum = _mm_add_ps(accum, _mm_mul_ps(c, weightVector));
+                weight += combined_weight;
+            }
+        }
+        
+        __m128 totalWeightVector = _mm_set1_ps(weight);
+        accum = _mm_div_ps(accum, totalWeightVector);
+        accum = _mm_add_ps(accum, _mm_set1_ps(0.5f));
+        
+        float localResult[4];
+        _mm_storeu_ps(localResult, accum);
+        int a = SkClampMax(int(localResult[0]), 255);
+        int r = SkClampMax(int(localResult[1]), a);
+        int g = SkClampMax(int(localResult[2]), a);
+        int b = SkClampMax(int(localResult[3]), a);
+        
+        *colors++ = SkPackARGB32(a, r, g, b);
+        
+        x++;
+        
+        s.fInvProc(*s.fInvMatrix, SkIntToScalar(x),
+                    SkIntToScalar(y), &srcPt);
+        
+    } 
+}
diff --git a/src/opts/SkBitmapFilter_opts_SSE2.h b/src/opts/SkBitmapFilter_opts_SSE2.h
new file mode 100644
index 0000000000..974b00896c
--- /dev/null
+++ b/src/opts/SkBitmapFilter_opts_SSE2.h
@@ -0,0 +1,20 @@
+
+/*
+ * Copyright 2013 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+
+#ifndef SkBitmapFilter_opts_sse2_DEFINED
+#define SkBitmapFilter_opts_sse2_DEFINED
+
+#include "SkBitmapProcState.h"
+
+void highQualityFilter_ScaleOnly_SSE2(const SkBitmapProcState &s, int x, int y,
+                          SkPMColor *SK_RESTRICT colors, int count);
+void highQualityFilter_SSE2(const SkBitmapProcState &s, int x, int y,
+                SkPMColor *SK_RESTRICT colors, int count);
+                
+#endif
diff --git a/src/opts/opts_check_SSE2.cpp b/src/opts/opts_check_SSE2.cpp
index d23e3844d5..37ce9036ca 100644
--- a/src/opts/opts_check_SSE2.cpp
+++ b/src/opts/opts_check_SSE2.cpp
@@ -7,6 +7,7 @@
 
 #include "SkBitmapProcState_opts_SSE2.h"
 #include "SkBitmapProcState_opts_SSSE3.h"
+#include "SkBitmapFilter_opts_SSE2.h"
 #include "SkBlitMask.h"
 #include "SkBlitRow.h"
 #include "SkBlitRect_opts_SSE2.h"
@@ -14,6 +15,8 @@
 #include "SkUtils_opts_SSE2.h"
 #include "SkUtils.h"
 
+#include "SkRTConf.h"
+
 #if defined(_MSC_VER) && defined(_WIN64)
 #include <intrin.h>
 #endif
@@ -102,6 +105,8 @@ static bool cachedHasSSSE3() {
     return gHasSSSE3;
 }
 
+SK_CONF_DECLARE( bool, c_hqfilter_sse, "bitmap.filter.highQualitySSE", false, "Use SSE optimized version of high quality image filters");
+
 void SkBitmapProcState::platformProcs() {
     if (cachedHasSSSE3()) {
 #if !defined(SK_BUILD_FOR_ANDROID)
@@ -142,6 +147,14 @@ void SkBitmapProcState::platformProcs() {
         } else if (fMatrixProc == ClampX_ClampY_nofilter_affine) {
             fMatrixProc = ClampX_ClampY_nofilter_affine_SSE2;
         }
+        if (c_hqfilter_sse) {
+            if (fShaderProc32 == highQualityFilter) {
+                fShaderProc32 = highQualityFilter_SSE2;
+            }
+            if (fShaderProc32 == highQualityFilter_ScaleOnly) {
+                fShaderProc32 = highQualityFilter_ScaleOnly_SSE2;
+            }
+        }
     }
 }