ARM Skia NEON patches - 21 - new NEON S32_D565_Opaque

src/opts/SkBlitRow_opts_arm_neon.cpp

 /*
  * Copyright 2012 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_opts_arm.h"
 
 #include "SkBlitMask.h"
 #include "SkBlitRow.h"
 #include "SkColorPriv.h"
 #include "SkDither.h"
 #include "SkMathPriv.h"
 #include "SkUtils.h"
 
 #include "SkCachePreload_arm.h"
+#include "SkColor_opts_neon.h"
+#include <arm_neon.h>
 
-#include <arm_neon.h>
+void S32_D565_Opaque_neon(uint16_t* SK_RESTRICT dst,
+                           const SkPMColor* SK_RESTRICT src, int count,
+                           U8CPU alpha, int /*x*/, int /*y*/) {
+    SkASSERT(255 == alpha);
+
+    while (count >= 8) {
+        uint8x8x4_t vsrc;
+        uint16x8_t vdst;
+
+        // Load
+        vsrc = vld4_u8((uint8_t*)src);
+
+        // Convert src to 565
+        vdst = vshll_n_u8(vsrc.val[NEON_R], 8);
+        vdst = vsriq_n_u16(vdst, vshll_n_u8(vsrc.val[NEON_G], 8), 5);
+        vdst = vsriq_n_u16(vdst, vshll_n_u8(vsrc.val[NEON_B], 8), 5+6);
+
+        // Store
+        vst1q_u16(dst, vdst);
+
+        // Prepare next iteration
+        dst += 8;
+        src += 8;
+        count -= 8;
+    };
+
+    // Leftovers
+    while (count > 0) {
+        SkPMColor c = *src++;
+        SkPMColorAssert(c);
+        *dst = SkPixel32ToPixel16_ToU16(c);
+        dst += 1;

[mtklein, 2013/09/20 13:36:08]

For symmetry I'd go with a ++ here.

[kevin.petit.not.used.account, 2013/09/20 14:41:56]

Done.

+        count--;
+    };
+}
 
 void S32A_D565_Opaque_neon(uint16_t* SK_RESTRICT dst,
                            const SkPMColor* SK_RESTRICT src, int count,
                            U8CPU alpha, int /*x*/, int /*y*/) {
     SkASSERT(255 == alpha);
 
     if (count >= 8) {
         uint16_t* SK_RESTRICT keep_dst = 0;
 
         asm volatile (
@@ skipping 1203 matching lines @@
             dst += 1;
             count--;
         }
     }
 }
 
 ///////////////////////////////////////////////////////////////////////////////
 
 const SkBlitRow::Proc sk_blitrow_platform_565_procs_arm_neon[] = {
     // no dither
-    // NOTE: For the two functions below, we don't have a special version
-    //       that assumes that each source pixel is opaque. But our S32A is
-    //       still faster than the default, so use it.
-    S32A_D565_Opaque_neon,  // really S32_D565_Opaque
+    // NOTE: For the S32_D565_Blend function below, we don't have a special
+    //       version that assumes that each source pixel is opaque. But our
+    //       S32A is still faster than the default, so use it.
+    S32_D565_Opaque_neon,
     S32A_D565_Blend_neon,   // really S32_D565_Blend
     S32A_D565_Opaque_neon,
     S32A_D565_Blend_neon,
 
     // dither
     S32_D565_Opaque_Dither_neon,
     S32_D565_Blend_Dither_neon,
     S32A_D565_Opaque_Dither_neon,
     NULL,   // S32A_D565_Blend_Dither
 };
@@ skipping 11 matching lines @@
      * case where we do not inspect the src alpha.
      */
 #if SK_A32_SHIFT == 24
     // This proc assumes the alpha value occupies bits 24-32 of each SkPMColor
     S32A_Opaque_BlitRow32_neon_src_alpha,   // S32A_Opaque,
 #else
     S32A_Opaque_BlitRow32_neon,     // S32A_Opaque,
 #endif
     S32A_Blend_BlitRow32_arm        // S32A_Blend
 };