Properly enable S32_D16_filter_DX_SSE2 optimization.

src/opts/opts_check_SSE2.cpp

 /*
  * Copyright 2009 The Android Open Source Project
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "SkBitmapProcState_opts_SSE2.h"
 #include "SkBitmapProcState_opts_SSSE3.h"
 #include "SkBitmapFilter_opts_SSE2.h"
@@ skipping 104 matching lines @@
 void SkBitmapProcState::platformConvolutionProcs(SkConvolutionProcs* procs) {
     if (cachedHasSSE2()) {
         procs->fExtraHorizontalReads = 3;
         procs->fConvolveVertically = &convolveVertically_SSE2;
         procs->fConvolve4RowsHorizontally = &convolve4RowsHorizontally_SSE2;
         procs->fConvolveHorizontally = &convolveHorizontally_SSE2;
         procs->fApplySIMDPadding = &applySIMDPadding_SSE2;
     }
 }
 
 void SkBitmapProcState::platformProcs() {

[mtklein, 2014/04/16 16:40:39]

I can see how easy this was to mess up before.  Can we make it even clearer? 
I'm still having trouble identifying all the procs we want to set and following
through all the paths they can take.

Maybe something like

if (hasSSE2) {
   fill in all SSE2 procs here
}

if (hasSSSE3) {
  fill in all SSSE3 procs here (overwriting our previous write of fSampleProc32)
}

?

     if (cachedHasSSSE3()) {
         if (fSampleProc32 == S32_opaque_D32_filter_DX) {
             fSampleProc32 = S32_opaque_D32_filter_DX_SSSE3;
         } else if (fSampleProc32 == S32_alpha_D32_filter_DX) {
             fSampleProc32 = S32_alpha_D32_filter_DX_SSSE3;
         }
 
         if (fSampleProc32 == S32_opaque_D32_filter_DXDY) {
             fSampleProc32 = S32_opaque_D32_filter_DXDY_SSSE3;
         } else if (fSampleProc32 == S32_alpha_D32_filter_DXDY) {
             fSampleProc32 = S32_alpha_D32_filter_DXDY_SSSE3;
         }
     } else if (cachedHasSSE2()) {
         if (fSampleProc32 == S32_opaque_D32_filter_DX) {
             fSampleProc32 = S32_opaque_D32_filter_DX_SSE2;
         } else if (fSampleProc32 == S32_alpha_D32_filter_DX) {
             fSampleProc32 = S32_alpha_D32_filter_DX_SSE2;
         }
+    }
 
+    if (cachedHasSSE2()) {
         if (fSampleProc16 == S32_D16_filter_DX) {
             fSampleProc16 = S32_D16_filter_DX_SSE2;
         }
-    }
 
-    if (cachedHasSSSE3() || cachedHasSSE2()) {
         if (fMatrixProc == ClampX_ClampY_filter_scale) {
             fMatrixProc = ClampX_ClampY_filter_scale_SSE2;
         } else if (fMatrixProc == ClampX_ClampY_nofilter_scale) {
             fMatrixProc = ClampX_ClampY_nofilter_scale_SSE2;
         }
 
         if (fMatrixProc == ClampX_ClampY_filter_affine) {
             fMatrixProc = ClampX_ClampY_filter_affine_SSE2;
         } else if (fMatrixProc == ClampX_ClampY_nofilter_affine) {
             fMatrixProc = ClampX_ClampY_nofilter_affine_SSE2;
@@ skipping 170 matching lines @@
     } else {
         return SkPlatformXfermodeFactory_impl(rec, mode);
     }
 }
 
 SkXfermodeProc SkPlatformXfermodeProcFactory(SkXfermode::Mode mode);
 
 SkXfermodeProc SkPlatformXfermodeProcFactory(SkXfermode::Mode mode) {
     return NULL;
 }