ARGBToI444 for Neon

BUG=none
TEST=libyuvTest.ARGBToI444_Opt
Review URL: https://webrtc-codereview.appspot.com/932013

git-svn-id: http://libyuv.googlecode.com/svn/trunk@475 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 474
+Version: 475
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -193,6 +193,7 @@ extern "C" {
 #define HAS_I444TOARGBROW_NEON
 #define HAS_I411TOARGBROW_NEON
 #define HAS_ARGBTOYROW_NEON
+#define HAS_ARGBTOUV444ROW_NEON
 #define HAS_BGRATOYROW_NEON
 #define HAS_ABGRTOYROW_NEON
 #define HAS_RGBATOYROW_NEON
@@ -342,6 +343,8 @@ void RGBAToYRow_Unaligned_SSSE3(const uint8* src_rgba, uint8* dst_y, int pix);
 void RGB24ToYRow_Unaligned_SSSE3(const uint8* src_rgb24, uint8* dst_y, int pix);
 void RAWToYRow_Unaligned_SSSE3(const uint8* src_raw, uint8* dst_y, int pix);
 void ARGBToYRow_NEON(const uint8* src_argb, uint8* dst_y, int pix);
+void ARGBToUV444Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v,
+                         int pix);
 void BGRAToYRow_NEON(const uint8* src_bgra, uint8* dst_y, int pix);
 void ABGRToYRow_NEON(const uint8* src_abgr, uint8* dst_y, int pix);
 void RGBAToYRow_NEON(const uint8* src_rgba, uint8* dst_y, int pix);

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 474
+#define LIBYUV_VERSION 475
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/convert_from_argb.cc

@@ -38,6 +38,8 @@ int ARGBToI444(const uint8* src_argb, int src_stride_argb,
   }
   void (*ARGBToYRow)(const uint8* src_argb, uint8* dst_y, int pix) =
       ARGBToYRow_C;
+  void (*ARGBToUV444Row)(const uint8* src_argb, uint8* dst_u, uint8* dst_v,
+                         int pix) = ARGBToUV444Row_C;
 #if defined(HAS_ARGBTOYROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
     ARGBToYRow = ARGBToYRow_Any_SSSE3;
@@ -54,12 +56,13 @@ int ARGBToI444(const uint8* src_argb, int src_stride_argb,
     ARGBToYRow = ARGBToYRow_Any_NEON;
     if (IS_ALIGNED(width, 8)) {
       ARGBToYRow = ARGBToYRow_NEON;
+      ARGBToUV444Row = ARGBToUV444Row_NEON;
     }
   }
 #endif
 
   for (int y = 0; y < height; ++y) {
-    ARGBToUV444Row_C(src_argb, dst_u, dst_v, width);
+    ARGBToUV444Row(src_argb, dst_u, dst_v, width);
     ARGBToYRow(src_argb, dst_y, width);
     src_argb += src_stride_argb;
     dst_y += dst_stride_y;

source/row_neon.cc

@@ -1572,6 +1572,46 @@ void ARGBToYRow_NEON(const uint8* src_argb, uint8* dst_y, int pix) {
 }
 #endif  // HAS_ARGBTOYROW_NEON
 
+#ifdef HAS_ARGBTOYROW_NEON
+void ARGBToUV444Row_NEON(const uint8* src_argb, uint8* dst_u, uint8* dst_v,
+                         int pix) {
+  asm volatile (
+    "vmov.u8    d24, #112 / 2                  \n"  // UB / VR  coefficient
+    "vmov.u8    d25, #74 / 2                   \n"  // UG -coefficient
+    "vmov.u8    d26, #38 / 2                   \n"  // UR -coefficient
+    "vmov.u8    d27, #18 / 2                   \n"  // VB -coefficient
+    "vmov.u8    d28, #94 / 2                   \n"  // VG -coefficient
+    "vmov.u16   q15, #0x8080 / 2               \n"  // 128
+    ".p2align  2                               \n"
+  "1:                                          \n"
+    "vld4.8     {d0, d1, d2, d3}, [%0]!        \n"  // load 8 ARGB pixels.
+    "subs       %3, %3, #8                     \n"  // 8 processed per loop.
+    "vmull.u8   q2, d0, d24                    \n"  // B
+    "vmlsl.u8   q2, d1, d25                    \n"  // G
+    "vmlsl.u8   q2, d2, d26                    \n"  // R
+    "vadd.u16   q2, q2, q15                    \n"  // +128 -> unsigned
+
+    "vmull.u8   q3, d2, d24                    \n"  // R
+    "vmlsl.u8   q3, d1, d28                    \n"  // G
+    "vmlsl.u8   q3, d0, d27                    \n"  // B
+    "vadd.u16   q3, q3, q15                    \n"  // +128 -> unsigned
+
+    "vqshrn.u16  d0, q2, #7                    \n"  // 16 bit to 8 bit U
+    "vqshrn.u16  d1, q3, #7                    \n"  // 16 bit to 8 bit V
+
+    "vst1.8     {d0}, [%1]!                    \n"  // store 8 pixels U.
+    "vst1.8     {d1}, [%2]!                    \n"  // store 8 pixels U.
+    "bgt        1b                             \n"
+  : "+r"(src_argb),  // %0
+    "+r"(dst_u),     // %1
+    "+r"(dst_v),     // %2
+    "+r"(pix)        // %3
+  :
+  : "memory", "cc", "q0", "q1", "q2", "q3", "q4"
+  );
+}
+#endif  // HAS_ARGBTOYROW_NEON
+
 #ifdef HAS_RGB565TOYROW_NEON
 void RGB565ToYRow_NEON(const uint8* src_rgb565, uint8* dst_y, int pix) {
   asm volatile (