Optimize RGB16 blitV functions with NEON for ARM platform.

src/core/SkBlitter_RGB16.cpp

 /*
  * Copyright 2006 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 "SkBlitRow.h"
 #include "SkCoreBlitters.h"
 #include "SkColorPriv.h"
 #include "SkDither.h"
 #include "SkShader.h"
 #include "SkUtils.h"
 #include "SkUtilsArm.h"
 #include "SkXfermode.h"
 
 #if SK_MIPS_HAS_DSP
 extern void blitmask_d565_opaque_mips(int width, int height, uint16_t* device,
                                       unsigned deviceRB, const uint8_t* alpha,
                                       uint32_t expanded32, unsigned maskRB);
 #endif
 
 #if SK_ARM_NEON_IS_ALWAYS && defined(SK_CPU_LENDIAN)
     #include <arm_neon.h>
+extern void SkRGB16BlitterBlitV_neon(uint16_t* device,
+                                     int height,
+                                     size_t deviceRB,
+                                     unsigned scale,
+                                     uint32_t src32);
 #else
     // if we don't have neon, then our black blitter is worth the extra code
     #define USE_BLACK_BLITTER
 #endif
 
 void sk_dither_memset16(uint16_t dst[], uint16_t value, uint16_t other,
                         int count) {
     if (count > 0) {
         // see if we need to write one short before we can cast to an 4byte ptr
         // (we do this subtract rather than (unsigned)dst so we don't get warnings
@@ skipping 442 matching lines @@
 }
 
 void SkRGB16_Opaque_Blitter::blitV(int x, int y, int height, SkAlpha alpha) {
     uint16_t* SK_RESTRICT device = fDevice.writable_addr16(x, y);
     size_t    deviceRB = fDevice.rowBytes();
 
     // TODO: respect fDoDither
     unsigned scale5 = SkAlpha255To256(alpha) >> 3;
     uint32_t src32 =  fExpandedRaw16 * scale5;
     scale5 = 32 - scale5;
+#if SK_ARM_NEON_IS_ALWAYS && defined(SK_CPU_LENDIAN)
+    SkRGB16BlitterBlitV_neon(device, height, deviceRB, scale5, src32);
+#else
     do {
         uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
         *device = SkCompact_rgb_16((src32 + dst32) >> 5);
         device = (uint16_t*)((char*)device + deviceRB);
     } while (--height != 0);
+#endif
 }
 
 void SkRGB16_Opaque_Blitter::blitRect(int x, int y, int width, int height) {
     SkASSERT(x + width <= fDevice.width() && y + height <= fDevice.height());
     uint16_t* SK_RESTRICT device = fDevice.writable_addr16(x, y);
     size_t      deviceRB = fDevice.rowBytes();
     uint16_t    color16 = fColor16;
 
     if (fDoDither) {
         uint16_t ditherColor = fRawDither16;
@@ skipping 150 matching lines @@
 }
 
 void SkRGB16_Blitter::blitV(int x, int y, int height, SkAlpha alpha) {
     uint16_t* SK_RESTRICT device = fDevice.writable_addr16(x, y);
     size_t    deviceRB = fDevice.rowBytes();
 
     // TODO: respect fDoDither
     unsigned scale5 = SkAlpha255To256(alpha) * fScale >> (8 + 3);
     uint32_t src32 =  fExpandedRaw16 * scale5;
     scale5 = 32 - scale5;
+#if SK_ARM_NEON_IS_ALWAYS && defined(SK_CPU_LENDIAN)
+    SkRGB16BlitterBlitV_neon(device, height, deviceRB, scale5, src32);
+#else
     do {
         uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
         *device = SkCompact_rgb_16((src32 + dst32) >> 5);
         device = (uint16_t*)((char*)device + deviceRB);
     } while (--height != 0);
+#endif
 }
 
 void SkRGB16_Blitter::blitRect(int x, int y, int width, int height) {
     SkASSERT(x + width <= fDevice.width() && y + height <= fDevice.height());
     uint16_t* SK_RESTRICT device = fDevice.writable_addr16(x, y);
     size_t    deviceRB = fDevice.rowBytes();
 
     while (--height >= 0) {
         fColorProc16(device, fSrcColor32, width, x, y);
         device = (uint16_t*)((char*)device + deviceRB);
@@ skipping 373 matching lines @@
 #endif
         } else if (0xFF == SkColorGetA(color)) {
             blitter = allocator->createT<SkRGB16_Opaque_Blitter>(device, paint);
         } else {
             blitter = allocator->createT<SkRGB16_Blitter>(device, paint);
         }
     }
 
     return blitter;
 }