From e6ea606fb92cd611b965806cb005f87495b261f2 Mon Sep 17 00:00:00 2001
From: "reed@google.com" <reed@google.com@2bbb7eff-a529-9590-31e7-b0007b416f81>
Date: Thu, 7 Jul 2011 19:12:50 +0000
Subject: [PATCH] re-enable SSE2 blitmask procs, only excluding if we're black
 (in which case the protable version is still faster)

git-svn-id: http://skia.googlecode.com/svn/trunk@1819 2bbb7eff-a529-9590-31e7-b0007b416f81
---
 src/core/SkBlitRow_D32.cpp    |  1 -
 src/core/SkBlitter_ARGB32.cpp | 25 +++++--------------------
 src/opts/opts_check_SSE2.cpp  | 11 +++++------
 3 files changed, 10 insertions(+), 27 deletions(-)

diff --git a/src/core/SkBlitRow_D32.cpp b/src/core/SkBlitRow_D32.cpp
index 642fe7f8e..72df59c75 100644
--- a/src/core/SkBlitRow_D32.cpp
+++ b/src/core/SkBlitRow_D32.cpp
@@ -230,7 +230,6 @@ static void D32_Mask_Black(void* dst, size_t dstRB, SkBitmap::Config,
 
 SkBlitMask::Proc SkBlitMask::Factory(SkBitmap::Config config, SkColor color) {
     SkBlitMask::Proc proc = PlatformProcs(config, color);
-    proc = NULL;
     if (NULL == proc) {
         switch (config) {
             case SkBitmap::kARGB_8888_Config:
diff --git a/src/core/SkBlitter_ARGB32.cpp b/src/core/SkBlitter_ARGB32.cpp
index f54bd5a27..073757ac0 100644
--- a/src/core/SkBlitter_ARGB32.cpp
+++ b/src/core/SkBlitter_ARGB32.cpp
@@ -409,26 +409,11 @@ void SkARGB32_Black_Blitter::blitMask(const SkMask& mask, const SkIRect& clip) {
     } else if (SkMask::kLCD32_Format == mask.fFormat) {
         blitmask_lcd32(fDevice, mask, clip, fPMColor);
     } else {
-        unsigned width = clip.width();
-        unsigned height = clip.height();
-
-        SkASSERT((int)height > 0);
-        SkASSERT((int)width > 0);
-
-        uint32_t*      device = fDevice.getAddr32(clip.fLeft, clip.fTop);
-        unsigned       maskRB = mask.fRowBytes - width;
-        unsigned       deviceRB = fDevice.rowBytes() - (width << 2);
-        const uint8_t* alpha = mask.getAddr(clip.fLeft, clip.fTop);
-        do {
-            unsigned w = width;
-            do {
-                unsigned aa = *alpha++;
-                *device = (aa << SK_A32_SHIFT) + SkAlphaMulQ(*device, SkAlpha255To256(255 - aa));
-                device += 1;
-            } while (--w != 0);
-            device = (uint32_t*)((char*)device + deviceRB);
-            alpha += maskRB;
-        } while (--height != 0);
+        fBlitMaskProc(fDevice.getAddr32(clip.fLeft, clip.fTop),
+                      fDevice.rowBytes(),
+                      SkBitmap::kARGB_8888_Config,
+                      mask.getAddr(clip.fLeft, clip.fTop), mask.fRowBytes,
+                      0xFF000000, clip.width(), clip.height());
     }
 }
 
diff --git a/src/opts/opts_check_SSE2.cpp b/src/opts/opts_check_SSE2.cpp
index 749117a1a..09d99e9e7 100644
--- a/src/opts/opts_check_SSE2.cpp
+++ b/src/opts/opts_check_SSE2.cpp
@@ -107,17 +107,16 @@ SkBlitRow::Proc32 SkBlitRow::PlatformProcs32(unsigned flags) {
 
 
 SkBlitMask::Proc SkBlitMask::PlatformProcs(SkBitmap::Config dstConfig,
-                                           SkColor color)
-{
-
+                                           SkColor color) {
     SkBlitMask::Proc proc = NULL;
     if (hasSSE2()) {
         switch (dstConfig) {
             case SkBitmap::kARGB_8888_Config:
-                // TODO: is our current SSE2 faster than the portable, even in
-                // the case of black or opaque? If so, no need for this check.
-                if ( SK_ColorBLACK != color && 0xFF != SkColorGetA(color))
+                // The SSE2 version is not (yet) faster for black, so we check
+                // for that.
+                if (SK_ColorBLACK != color) {
                     proc = SkARGB32_BlitMask_SSE2;
+                }
                 break;
             default:
                  break;