NEON optimizations for GrayAlpha -> RGBA/BGRA Premul/Unpremul

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 291 matching lines @@
         const SkPMColor ctable[]) {
 
     src += offset;
     SkPMColor* dst32 = (SkPMColor*) dst;
     for (int x = 0; x < width; x++) {
         dst32[x] = SkPackARGB32NoCheck(src[1], src[0], src[0], src[0]);
         src += deltaSrc;
     }
 }
 
+static void fast_swizzle_grayalpha_to_n32_unpremul(
+        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::grayA_to_RGBA((uint32_t*) dst, src + offset, width);
+}
+
 static void swizzle_grayalpha_to_n32_premul(
         void* dst, const uint8_t* src, int width, int bpp, int deltaSrc, int offset,
         const SkPMColor ctable[]) {
 
     src += offset;
     SkPMColor* dst32 = (SkPMColor*) dst;
     for (int x = 0; x < width; x++) {
         uint8_t pmgray = SkMulDiv255Round(src[1], src[0]);
         dst32[x] = SkPackARGB32NoCheck(src[1], pmgray, pmgray, pmgray);
         src += deltaSrc;
     }
 }
 
+static void fast_swizzle_grayalpha_to_n32_premul(
+        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::grayA_to_rgbA((uint32_t*) dst, src + offset, width);
+}
+
 // kBGRX
 
 static void swizzle_bgrx_to_n32(
         void* SK_RESTRICT dstRow, const uint8_t* SK_RESTRICT src, int dstWidth,
         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[2], src[1], src[0]);
@@ skipping 376 matching lines @@
                     break;
             }
             break;
         case kGrayAlpha:
             switch (dstInfo.colorType()) {
                 case kN32_SkColorType:
                     if (dstInfo.alphaType() == kUnpremul_SkAlphaType) {
                         if (SkCodec::kYes_ZeroInitialized == zeroInit) {
                             proc = &SkipLeadingGrayAlphaZerosThen
                                     <swizzle_grayalpha_to_n32_unpremul>;
+                            fastProc = &SkipLeadingGrayAlphaZerosThen
+                                    <fast_swizzle_grayalpha_to_n32_unpremul>;
                         } else {
                             proc = &swizzle_grayalpha_to_n32_unpremul;
+                            fastProc = &fast_swizzle_grayalpha_to_n32_unpremul;
                         }
                     } else {
                         if (SkCodec::kYes_ZeroInitialized == zeroInit) {
                             proc = &SkipLeadingGrayAlphaZerosThen<swizzle_grayalpha_to_n32_premul>;
+                            fastProc = &SkipLeadingGrayAlphaZerosThen
+                                    <fast_swizzle_grayalpha_to_n32_premul>;
                         } else {
                             proc = &swizzle_grayalpha_to_n32_premul;
+                            fastProc = &fast_swizzle_grayalpha_to_n32_premul;
                         }
                     }
                     break;
                 default:
                     break;
             }
             break;
         case kBGR:
         case kBGRX:
             switch (dstInfo.colorType()) {
@@ skipping 157 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);
 }