NEON optimizations for gray -> RGBA (or BGRA) conversions

src/codec/SkSwizzler.cpp

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "SkCodecPriv.h"
 #include "SkColorPriv.h"
 #include "SkOpts.h"
@@ skipping 252 matching lines @@
         int bpp, int deltaSrc, int offset, const SkPMColor ctable[]) {
 
     src += offset;
     SkPMColor* SK_RESTRICT dst = (SkPMColor*)dstRow;
     for (int x = 0; x < dstWidth; x++) {
         dst[x] = SkPackARGB32NoCheck(0xFF, *src, *src, *src);
         src += deltaSrc;
     }
 }
 
+static void fast_swizzle_gray_to_n32(
+        void* dst, const uint8_t* src, int width, int bpp, int deltaSrc, int offset,
+        const SkPMColor ctable[]) {
+
+    // This function must not be called if we are sampling.  If we are not
+    // sampling, deltaSrc should equal bpp.
+    SkASSERT(deltaSrc == bpp);
+
+    // Note that there is no need to distinguish between RGB and BGR.
+    // Each color channel will get the same value.
+    SkOpts::gray_to_RGB1((uint32_t*) dst, src + offset, width);
+}
+
 static void swizzle_gray_to_565(
         void* SK_RESTRICT dstRow, const uint8_t* SK_RESTRICT src, int dstWidth,
         int bytesPerPixel, int deltaSrc, int offset, const SkPMColor ctable[]) {
 
     src += offset;
     uint16_t* SK_RESTRICT dst = (uint16_t*)dstRow;
     for (int x = 0; x < dstWidth; x++) {
         dst[x] = SkPack888ToRGB16(src[0], src[0], src[0]);
         src += deltaSrc;
     }
@@ skipping 349 matching lines @@
                     fastProc = &copy;
                     break;
                 default:
                     break;
             }
             break;
         case kGray:
             switch (dstInfo.colorType()) {
                 case kN32_SkColorType:
                     proc = &swizzle_gray_to_n32;
+                    fastProc = &fast_swizzle_gray_to_n32;
                     break;
                 case kGray_8_SkColorType:
                     proc = &sample1;
                     fastProc = &copy;
                     break;
                 case kRGB_565_SkColorType:
                     proc = &swizzle_gray_to_565;
                     break;
                 default:
                     break;
@@ skipping 162 matching lines @@
     }
 
     return fAllocatedWidth;
 }
 
 void SkSwizzler::swizzle(void* dst, const uint8_t* SK_RESTRICT src) {
     SkASSERT(nullptr != dst && nullptr != src);
     fActualProc(SkTAddOffset<void>(dst, fDstOffsetBytes), src, fSwizzleWidth, fSrcBPP,
             fSampleX * fSrcBPP, fSrcOffsetUnits, fColorTable);
 }