From 6c72d5740231f47c664a8e765a8df05cd124c88c Mon Sep 17 00:00:00 2001
From: "yang.zhang" <yang.zhang@linaro.org>
Date: Thu, 30 Jul 2015 05:37:06 -0700
Subject: Optimize RGB16 blitH functions with NEON for ARM platform.

Here are some performance resultsi on Nexus 9:
SkRGB16BlitterBlitH_neon:
    +--------+-----------+
    |height  |  C/NEON   |
    +--------+-----------+
    |1       | 0.888531  |
    +--------+-----------+
    |8       | 1.231800  |
    +--------+-----------+
    |18      | 1.073327  |
    +--------+-----------+
    |32      | 1.136991  |
    +--------+-----------+
    |76      | 1.174638  |
    +--------+-----------+
    |85      | 1.188551  |
    +--------+-----------+
    |120     | 1.180261  |
    +--------+-----------+
    |128     | 1.183726  |
    +--------+-----------+
    |512     | 1.220806  |
    +--------+-----------+

BUG=skia:

Review URL: https://codereview.chromium.org/1229673008
---
 src/core/SkBlitter_RGB16.cpp | 12 ++++++++++++
 1 file changed, 12 insertions(+)

(limited to 'src/core')

diff --git a/src/core/SkBlitter_RGB16.cpp b/src/core/SkBlitter_RGB16.cpp
index 8caf908312..b481b27b57 100644
--- a/src/core/SkBlitter_RGB16.cpp
+++ b/src/core/SkBlitter_RGB16.cpp
@@ -27,6 +27,10 @@ extern void SkRGB16BlitterBlitV_neon(uint16_t* device,
                                      size_t deviceRB,
                                      unsigned scale,
                                      uint32_t src32);
+extern void SkRGB16BlitterBlitH_neon(uint16_t* device,
+                                     int width,
+                                     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
@@ -338,10 +342,14 @@ void SkRGB16_Opaque_Blitter::blitAntiH(int x, int y,
                 uint32_t src32 = srcExpanded * scale5;
                 scale5 = 32 - scale5; // now we can use it on the device
                 int n = count;
+#if SK_ARM_NEON_IS_ALWAYS && defined(SK_CPU_LENDIAN)
+                SkRGB16BlitterBlitH_neon(device, n, scale5, src32);
+#else
                 do {
                     uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
                     *device++ = SkCompact_rgb_16((src32 + dst32) >> 5);
                 } while (--n != 0);
+#endif
                 goto DONE;
             }
         }
@@ -601,10 +609,14 @@ void SkRGB16_Blitter::blitAntiH(int x, int y,
             unsigned scale5 = SkAlpha255To256(aa) * scale >> (8 + 3);
             uint32_t src32 =  srcExpanded * scale5;
             scale5 = 32 - scale5;
+#if SK_ARM_NEON_IS_ALWAYS && defined(SK_CPU_LENDIAN)
+            SkRGB16BlitterBlitH_neon(device, count, scale5, src32);
+#else
             do {
                 uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
                 *device++ = SkCompact_rgb_16((src32 + dst32) >> 5);
             } while (--count != 0);
+#endif
             continue;
         }
         device += count;
-- 
cgit v1.2.3