Revert of Fast blurred rectangles on GPU (https://codereview.chromium.org/119343003/)

Reason for revert:
Two problems:

1. broke compile on Build-Mac10.6-* bots
2. broke GM expectations on many bots

We'll need to examine the GM impact before re-landing.  I can work with you to do this on Tuesday if you like; just IM or email me.

Original issue's description:
> Perform the same analytic blur calculation on the GPU that we do on the CPU.  Results in significant performance gains when using Ganesh to render drop shadows in Chrome.
>
> BUG=
>
> Committed: http://code.google.com/p/skia/source/detail?r=13210

R=bsalomon@google.com, reed@google.com, humper@google.com, rmistry@google.com
TBR=bsalomon@google.com, humper@google.com, reed@google.com
NOTREECHECKS=true
NOTRY=true
BUG=

Author: epoger@google.com

Review URL: https://codereview.chromium.org/140853008

git-svn-id: http://skia.googlecode.com/svn/trunk@13213 2bbb7eff-a529-9590-31e7-b0007b416f81

1 files changed, 34 insertions, 36 deletions

diff --git a/src/effects/SkBlurMask.cpp b/src/effects/SkBlurMask.cpp
index f61536e65a..32a3d20d31 100644
--- a/src/effects/SkBlurMask.cpp
+++ b/src/effects/SkBlurMask.cpp
@@ -671,7 +671,7 @@ static float gaussianIntegral(float x) {
     return 0.4375f + (-x3 / 6.0f - 3.0f * x2 * 0.25f - 1.125f * x);
 }
 
-/*  ComputeBlurProfile allocates and fills in an array of floating
+/*  compute_profile allocates and fills in an array of floating
     point values between 0 and 255 for the profile signature of
     a blurred half-plane with the given blur radius.  Since we're
     going to be doing screened multiplications (i.e., 1 - (1-x)(1-y))
@@ -682,11 +682,11 @@ static float gaussianIntegral(float x) {
     memory returned in profile_out.
 */
 
-void SkBlurMask::ComputeBlurProfile(SkScalar sigma, uint8_t **profile_out) {
+static void compute_profile(SkScalar sigma, unsigned int **profile_out) {
     int size = SkScalarCeilToInt(6*sigma);
 
     int center = size >> 1;
-    uint8_t *profile = SkNEW_ARRAY(uint8_t, size);
+    unsigned int *profile = SkNEW_ARRAY(unsigned int, size);
 
     float invr = 1.f/(2*sigma);
 
@@ -707,7 +707,7 @@ void SkBlurMask::ComputeBlurProfile(SkScalar sigma, uint8_t **profile_out) {
 // Implementation adapted from Michael Herf's approach:
 // http://stereopsis.com/shadowrect/
 
-uint8_t SkBlurMask::ProfileLookup(const uint8_t *profile, int loc, int blurred_width, int sharp_width) {
+static inline unsigned int profile_lookup( unsigned int *profile, int loc, int blurred_width, int sharp_width ) {
     int dx = SkAbs32(((loc << 1) + 1) - blurred_width) - sharp_width; // how far are we from the original edge?
     int ox = dx >> 1;
     if (ox < 0) {
@@ -717,30 +717,6 @@ uint8_t SkBlurMask::ProfileLookup(const uint8_t *profile, int loc, int blurred_w
     return profile[ox];
 }
 
-void SkBlurMask::ComputeBlurredScanline(uint8_t *pixels, const uint8_t *profile, 
-                                        unsigned int width, SkScalar sigma) {
-
-    unsigned int profile_size = SkScalarCeilToInt(6*sigma);
-    SkAutoTMalloc<uint8_t> horizontalScanline(width);
-
-    unsigned int sw = width - profile_size;
-    // nearest odd number less than the profile size represents the center
-    // of the (2x scaled) profile
-    int center = ( profile_size & ~1 ) - 1;
-
-    int w = sw - center;
-
-    for (unsigned int x = 0 ; x < width ; ++x) {
-       if (profile_size <= sw) {
-           pixels[x] = ProfileLookup(profile, x, width, w);
-       } else {
-           float span = float(sw)/(2*sigma);
-           float giX = 1.5f - (x+.5f)/(2*sigma);
-           pixels[x] = (uint8_t) (255 * (gaussianIntegral(giX) - gaussianIntegral(giX + span)));
-       }
-    }                                       
-}
-
 bool SkBlurMask::BlurRect(SkMask *dst, const SkRect &src,
                           SkScalar radius, Style style,
                           SkIPoint *margin, SkMask::CreateMode createMode) {
@@ -781,10 +757,10 @@ bool SkBlurMask::BlurRect(SkScalar sigma, SkMask *dst,
         }
         return true;
     }
-    uint8_t *profile = NULL;
+    unsigned int *profile = NULL;
 
-    ComputeBlurProfile(sigma, &profile);
-    SkAutoTDeleteArray<uint8_t> ada(profile);
+    compute_profile(sigma, &profile);
+    SkAutoTDeleteArray<unsigned int> ada(profile);
 
     size_t dstSize = dst->computeImageSize();
     if (0 == dstSize) {
@@ -798,17 +774,39 @@ bool SkBlurMask::BlurRect(SkScalar sigma, SkMask *dst,
     int dstHeight = dst->fBounds.height();
     int dstWidth = dst->fBounds.width();
 
+    // nearest odd number less than the profile size represents the center
+    // of the (2x scaled) profile
+    int center = ( profile_size & ~1 ) - 1;
+
+    int w = sw - center;
+    int h = sh - center;
+
     uint8_t *outptr = dp;
 
     SkAutoTMalloc<uint8_t> horizontalScanline(dstWidth);
-    SkAutoTMalloc<uint8_t> verticalScanline(dstHeight);
-    
-    ComputeBlurredScanline(horizontalScanline, profile, dstWidth, sigma);
-    ComputeBlurredScanline(verticalScanline, profile, dstHeight, sigma);
+
+    for (int x = 0 ; x < dstWidth ; ++x) {
+        if (profile_size <= sw) {
+            horizontalScanline[x] = profile_lookup(profile, x, dstWidth, w);
+        } else {
+            float span = float(sw)/(2*sigma);
+            float giX = 1.5f - (x+.5f)/(2*sigma);
+            horizontalScanline[x] = (uint8_t) (255 * (gaussianIntegral(giX) - gaussianIntegral(giX + span)));
+        }
+    }
 
     for (int y = 0 ; y < dstHeight ; ++y) {
+        unsigned int profile_y;
+        if (profile_size <= sh) {
+            profile_y = profile_lookup(profile, y, dstHeight, h);
+        } else {
+            float span = float(sh)/(2*sigma);
+            float giY = 1.5f - (y+.5f)/(2*sigma);
+            profile_y = (uint8_t) (255 * (gaussianIntegral(giY) - gaussianIntegral(giY + span)));
+        }
+
         for (int x = 0 ; x < dstWidth ; x++) {
-            unsigned int maskval = SkMulDiv255Round(horizontalScanline[x], verticalScanline[y]);
+            unsigned int maskval = SkMulDiv255Round(horizontalScanline[x], profile_y);
             *(outptr++) = maskval;
         }
     }