ARGBToY_NEON for faster screencasts

BUG=107
TEST=convert_test
Review URL: https://webrtc-codereview.appspot.com/892004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@423 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 421
+Version: 423
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -153,6 +153,7 @@ extern "C" {
 #define HAS_ARGBTORGB565ROW_NEON
 #define HAS_ARGBTOARGB1555ROW_NEON
 #define HAS_ARGBTOARGB4444ROW_NEON
+#define HAS_ARGBTOYROW_NEON
 #endif
 
 // The following are available on Mips platforms
@@ -259,6 +260,8 @@ void ABGRToUVRow_Unaligned_SSSE3(const uint8* src_argb0, int src_stride_argb,
 void RGBAToUVRow_Unaligned_SSSE3(const uint8* src_argb0, int src_stride_argb,
                                  uint8* dst_u, uint8* dst_v, int width);
 
+void ARGBToYRow_NEON(const uint8* src_argb, uint8* dst_y, int pix);
+
 void MirrorRow_SSSE3(const uint8* src, uint8* dst, int width);
 void MirrorRow_SSE2(const uint8* src, uint8* dst, int width);
 void MirrorRow_NEON(const uint8* src, uint8* dst, int width);
@@ -613,6 +616,8 @@ void ABGRToUVRow_Any_SSSE3(const uint8* src_argb0, int src_stride_argb,
                            uint8* dst_u, uint8* dst_v, int width);
 void RGBAToUVRow_Any_SSSE3(const uint8* src_argb0, int src_stride_argb,
                            uint8* dst_u, uint8* dst_v, int width);
+void ARGBToYRow_Any_NEON(const uint8* src_argb, uint8* dst_y, int pix);
+
 void I422ToARGBRow_Any_NEON(const uint8* y_buf,
                             const uint8* u_buf,
                             const uint8* v_buf,

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 421
+#define LIBYUV_VERSION 423
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/convert.cc

@@ -909,6 +909,15 @@ int ARGBToI420(const uint8* src_argb, int src_stride_argb,
       }
     }
   }
+#elif defined(HAS_ARGBTOYROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON)) {
+    if (width > 8) {
+      ARGBToYRow = ARGBToYRow_Any_NEON;
+    }
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToYRow = ARGBToYRow_NEON;
+    }
+  }
 #endif
 
   for (int y = 0; y < height - 1; y += 2) {

source/row_common.cc

@@ -1083,27 +1083,30 @@ RGBANY(ARGBToARGB4444Row_Any_NEON, ARGBToARGB4444Row_NEON, ARGBToARGB4444Row_C,
 #undef RGBANY
 
 // RGB/YUV to Y does multiple of 16 with SIMD and last 16 with SIMD.
-#define YANY(NAMEANY, ARGBTOY_SIMD, BPP)                                       \
+#define YANY(NAMEANY, ARGBTOY_SIMD, BPP, NUM)                                  \
     void NAMEANY(const uint8* src_argb, uint8* dst_y, int width) {             \
-      ARGBTOY_SIMD(src_argb, dst_y, width - 16);                               \
-      ARGBTOY_SIMD(src_argb + (width - 16) * BPP, dst_y + (width - 16), 16);   \
+      ARGBTOY_SIMD(src_argb, dst_y, width - NUM);                              \
+      ARGBTOY_SIMD(src_argb + (width - NUM) * BPP, dst_y + (width - NUM), NUM);\
     }
 
 #ifdef HAS_ARGBTOYROW_SSSE3
-YANY(ARGBToYRow_Any_SSSE3, ARGBToYRow_Unaligned_SSSE3, 4)
-YANY(BGRAToYRow_Any_SSSE3, BGRAToYRow_Unaligned_SSSE3, 4)
-YANY(ABGRToYRow_Any_SSSE3, ABGRToYRow_Unaligned_SSSE3, 4)
+YANY(ARGBToYRow_Any_SSSE3, ARGBToYRow_Unaligned_SSSE3, 4, 16)
+YANY(BGRAToYRow_Any_SSSE3, BGRAToYRow_Unaligned_SSSE3, 4, 16)
+YANY(ABGRToYRow_Any_SSSE3, ABGRToYRow_Unaligned_SSSE3, 4, 16)
 #endif
 #ifdef HAS_RGBATOYROW_SSSE3
-YANY(RGBAToYRow_Any_SSSE3, RGBAToYRow_Unaligned_SSSE3, 4)
+YANY(RGBAToYRow_Any_SSSE3, RGBAToYRow_Unaligned_SSSE3, 4, 16)
+#endif
+#ifdef HAS_ARGBTOYROW_NEON
+YANY(ARGBToYRow_Any_NEON, ARGBToYRow_NEON, 4, 8)
 #endif
 #ifdef HAS_YUY2TOYROW_SSE2
-YANY(YUY2ToYRow_Any_SSE2, YUY2ToYRow_Unaligned_SSE2, 2)
-YANY(UYVYToYRow_Any_SSE2, UYVYToYRow_Unaligned_SSE2, 2)
+YANY(YUY2ToYRow_Any_SSE2, YUY2ToYRow_Unaligned_SSE2, 2, 16)
+YANY(UYVYToYRow_Any_SSE2, UYVYToYRow_Unaligned_SSE2, 2, 16)
 #endif
 #ifdef HAS_YUY2TOYROW_NEON
-YANY(YUY2ToYRow_Any_NEON, YUY2ToYRow_NEON, 2)
-YANY(UYVYToYRow_Any_NEON, UYVYToYRow_NEON, 2)
+YANY(YUY2ToYRow_Any_NEON, YUY2ToYRow_NEON, 2, 16)
+YANY(UYVYToYRow_Any_NEON, UYVYToYRow_NEON, 2, 16)
 #endif
 #undef YANY
 

source/row_neon.cc

@@ -995,6 +995,34 @@ void ARGBToARGB4444Row_NEON(const uint8* src_argb, uint8* dst_argb4444,
   );
 }
 #endif  // HAS_ARGBTOARGB4444ROW_NEON
+
+#ifdef HAS_ARGBTOYROW_NEON
+void ARGBToYRow_NEON(const uint8* src_argb, uint8* dst_y, int pix) {
+  asm volatile (
+    "vmov.u8    d4, #13                        \n"  // B * 0.1016 coefficient
+    "vmov.u8    d5, #65                        \n"  // G * 0.5078 coefficient
+    "vmov.u8    d6, #33                        \n"  // R * 0.2578 coefficient
+    "vmov.u8    d7, #16                        \n"  // Add 16 constant
+    ".p2align  2                               \n"
+  "1:                                          \n"
+    "vld4.8     {d0, d1, d2, d3}, [%0]!        \n"  // load 8 pixels of ARGB.
+    "subs       %2, %2, #8                     \n"  // 8 processed per loop.
+    "vmull.u8   q8, d0, d4                     \n"  // B
+    "vmlal.u8   q8, d1, d5                     \n"  // G
+    "vmlal.u8   q8, d2, d6                     \n"  // R
+    "vqrshrun.s16 d0, q8, #7                   \n"  // 16 bit to 8 bit Y
+    "vqadd.u8   d0, d7                         \n"
+    "vst1.8     {d0}, [%1]!                    \n"  // store 8 pixels Y.
+    "bgt        1b                             \n"
+  : "+r"(src_argb),  // %0
+    "+r"(dst_y),  // %1
+    "+r"(pix)        // %2
+  :
+  : "memory", "cc", "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "q8"
+  );
+}
+#endif  // HAS_ARGBTOYROW_NEON
+
 #endif  // __ARM_NEON__
 
 #ifdef __cplusplus