Use NEON optimizations for RGB -> RGB(FF) or BGR(FF) in SkSwizzler

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"
 #include "SkSwizzler.h"
 #include "SkTemplates.h"
 
 static void copy(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

[msarett, 2016/01/21 22:09:07]

I could be convinced to add these comments back in, if we feel they are useful.

[scroggo, 2016/01/21 23:19:39]

I see no reason to take them out.

[msarett, 2016/01/21 23:29:47]

Added back in.

-    // sampling, deltaSrc should equal bpp.
+
     SkASSERT(deltaSrc == bpp);
 
     memcpy(dst, src + offset, width * bpp);
 }
 
 static void sample1(void* dst, const uint8_t* src, int width, int bpp, int deltaSrc, int offset,
         const SkPMColor ctable[]) {
     src += offset;
     uint8_t* dst8 = (uint8_t*) dst;
     for (int x = 0; x < width; x++) {
@@ skipping 292 matching lines @@
         uint8_t alpha = src[3];
         dst[x] = SkPackARGB32NoCheck(alpha, src[2], src[1], src[0]);
         src += deltaSrc;
     }
 }
 
 static void fast_swizzle_bgra_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);
 
-    // These swizzles trust that the alpha value is already 0xFF.
 #ifdef SK_PMCOLOR_IS_RGBA
     SkOpts::swaprb_xxxa((uint32_t*) dst, (const uint32_t*) (src + offset), width);
 #else
     memcpy(dst, src + offset, width * bpp);
 #endif
 }
 
 static void swizzle_bgra_to_n32_premul(
         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++) {
         uint8_t alpha = src[3];
         dst[x] = SkPremultiplyARGBInline(alpha, src[2], src[1], src[0]);
         src += deltaSrc;
     }
 }
 
 static void fast_swizzle_bgra_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);
 
 #ifdef SK_PMCOLOR_IS_RGBA
     SkOpts::premul_swaprb_xxxa((uint32_t*) dst, (const uint32_t*) (src + offset), width);
 #else
     SkOpts::premul_xxxa((uint32_t*) dst, (const uint32_t*) (src + offset), width);
 #endif
 }
 
 // kRGB
 
 static void swizzle_rgb_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] = SkPackARGB32(0xFF, src[0], src[1], src[2]);
+        dst[x] = SkPackARGB32NoCheck(0xFF, src[0], src[1], src[2]);
         src += deltaSrc;
     }
 }
 
+static void fast_swizzle_rgb_to_n32(
+        void* dst, const uint8_t* src, int width, int bpp, int deltaSrc,
+        int offset, const SkPMColor ctable[]) {
 
+    SkASSERT(deltaSrc == bpp);
+
+#ifdef SK_PMCOLOR_IS_RGBA
+    SkOpts::xxx_xxxa((uint32_t*) dst, (const uint32_t*) (src + offset), width);
+#else
+    SkOpts::xxx_swaprb_xxxa((uint32_t*) dst, (const uint32_t*) (src + offset), width);
+#endif
+}
 
 static void swizzle_rgb_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[1], src[2]);
         src += deltaSrc;
@@ skipping 12 matching lines @@
         unsigned alpha = src[3];
         dst[x] = SkPremultiplyARGBInline(alpha, src[0], src[1], src[2]);
         src += deltaSrc;
     }
 }
 
 static void fast_swizzle_rgba_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);
 
 #ifdef SK_PMCOLOR_IS_RGBA
     SkOpts::premul_xxxa((uint32_t*) dst, (const uint32_t*) (src + offset), width);
 #else
     SkOpts::premul_swaprb_xxxa((uint32_t*) dst, (const uint32_t*) (src + offset), width);
 #endif
 }
 
 static void swizzle_rgba_to_n32_unpremul(
         void* SK_RESTRICT dstRow, const uint8_t* SK_RESTRICT src, int dstWidth,
         int bpp, int deltaSrc, int offset, const SkPMColor ctable[]) {
 
     src += offset;
     uint32_t* SK_RESTRICT dst = reinterpret_cast<uint32_t*>(dstRow);
     for (int x = 0; x < dstWidth; x++) {
         unsigned alpha = src[3];
         dst[x] = SkPackARGB32NoCheck(alpha, src[0], src[1], src[2]);
         src += deltaSrc;
     }
 }
 
 static void fast_swizzle_rgba_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);
 
-    // These swizzles trust that the alpha value is already 0xFF.
 #ifdef SK_PMCOLOR_IS_RGBA
     memcpy(dst, src + offset, width * bpp);
 #else
     SkOpts::swaprb_xxxa((uint32_t*) dst, (const uint32_t*) (src + offset), width);
 #endif
 }
 
 // kCMYK
 //
 // CMYK is stored as four bytes per pixel.
@@ skipping 215 matching lines @@
                     }
                     break;
                 default:
                     break;
             }
             break;
         case kRGB:
             switch (dstInfo.colorType()) {
                 case kN32_SkColorType:
                     proc = &swizzle_rgb_to_n32;
+                    fastProc = &fast_swizzle_rgb_to_n32;
                     break;
                 case kRGB_565_SkColorType:
                     proc = &swizzle_rgb_to_565;
                 default:
                     break;
             }
             break;
         case kRGBA:
             switch (dstInfo.colorType()) {
                 case kN32_SkColorType:
@@ skipping 103 matching lines @@
 
     return fAllocatedWidth;
 }
 
 void SkSwizzler::swizzle(void* dst, const uint8_t* SK_RESTRICT src) {
     SkASSERT(nullptr != dst && nullptr != src);
     RowProc proc = fFastProc ? fFastProc : fProc;
     proc(SkTAddOffset<void>(dst, fDstOffsetBytes), src, fSwizzleWidth, fSrcBPP,
             fSampleX * fSrcBPP, fSrcOffsetUnits, fColorTable);
 }