Neon optimized RAW (RGB big endian)

BUG=68
TEST=unittest on try bot
Review URL: https://webrtc-codereview.appspot.com/819005

git-svn-id: http://libyuv.googlecode.com/svn/trunk@368 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 367
+Version: 368
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -117,6 +117,7 @@ extern "C" {
 #define HAS_I422TORGBAROW_NEON
 #define HAS_ARGBTORGBAROW_NEON
 #define HAS_ARGBTORGB24ROW_NEON
+#define HAS_ARGBTORAWROW_NEON
 #endif
 
 #if defined(_MSC_VER) && !defined(__CLR_VER)
@@ -260,6 +261,7 @@ void ARGBToARGB4444Row_SSE2(const uint8* src_argb, uint8* dst_rgb, int pix);
 
 void ARGBToRGBARow_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
 void ARGBToRGB24Row_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
+void ARGBToRAWRow_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
 
 void ARGBToRGBARow_C(const uint8* src_argb, uint8* dst_rgb, int pix);
 void ARGBToRGB24Row_C(const uint8* src_argb, uint8* dst_rgb, int pix);
@@ -478,6 +480,7 @@ void ARGBToARGB1555Row_Any_SSE2(const uint8* src_argb, uint8* dst_rgb, int pix);
 void ARGBToARGB4444Row_Any_SSE2(const uint8* src_argb, uint8* dst_rgb, int pix);
 
 void ARGBToRGB24Row_Any_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
+void ARGBToRAWRow_Any_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
 
 void ARGBToYRow_Any_SSSE3(const uint8* src_argb, uint8* dst_y, int pix);
 void BGRAToYRow_Any_SSSE3(const uint8* src_argb, 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 367
+#define LIBYUV_VERSION 368
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/convert.cc

@@ -745,24 +745,7 @@ int YUY2ToI420(const uint8* src_yuy2, int src_stride_yuy2,
     }
   }
 #endif
-#if defined(HAS_YUY2TOYROW_NEON)
-  if (TestCpuFlag(kCpuHasNEON)) {
-    if (width > 16) {
-      YUY2ToUVRow = YUY2ToUVRow_Any_NEON;
-      YUY2ToYRow = YUY2ToYRow_Any_NEON;
-    }
-    if (IS_ALIGNED(width, 16)) {
-      YUY2ToUVRow = YUY2ToUVRow_Unaligned_NEON;
-      YUY2ToYRow = YUY2ToYRow_Unaligned_NEON;
-      if (IS_ALIGNED(src_yuy2, 16) && IS_ALIGNED(src_stride_yuy2, 16)) {
-        YUY2ToUVRow = YUY2ToUVRow_NEON;
-        if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
-          YUY2ToYRow = YUY2ToYRow_NEON;
-        }
-      }
-    }
-  }
-#endif
+
   for (int y = 0; y < height - 1; y += 2) {
     YUY2ToUVRow(src_yuy2, src_stride_yuy2, dst_u, dst_v, width);
     YUY2ToYRow(src_yuy2, dst_y, width);

source/convert_from.cc

@@ -882,6 +882,7 @@ int I420ToRGBA(const uint8* src_y, int src_stride_y,
 }
 
 // Convert I420 to RGB24.
+// TODO(fbarchard): One step I420ToRGB24Row_NEON.
 int I420ToRGB24(const uint8* src_y, int src_stride_y,
                 const uint8* src_u, int src_stride_u,
                 const uint8* src_v, int src_stride_v,
@@ -928,7 +929,6 @@ int I420ToRGB24(const uint8* src_y, int src_stride_y,
   }
 #endif
 #if defined(HAS_ARGBTORGB24ROW_NEON)
-  // TODO(fbarchard): One step I420ToRGB24Row_NEON.
   if (TestCpuFlag(kCpuHasNEON)) {
     if (width * 3 <= kMaxStride) {
       ARGBToRGB24Row = ARGBToRGB24Row_Any_NEON;
@@ -953,6 +953,7 @@ int I420ToRGB24(const uint8* src_y, int src_stride_y,
 }
 
 // Convert I420 to RAW.
+// TODO(fbarchard): One step I420ToRAWRow_NEON.
 int I420ToRAW(const uint8* src_y, int src_stride_y,
               const uint8* src_u, int src_stride_u,
               const uint8* src_v, int src_stride_v,
@@ -998,6 +999,16 @@ int I420ToRAW(const uint8* src_y, int src_stride_y,
     }
   }
 #endif
+#if defined(HAS_ARGBTORAWROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON)) {
+    if (width * 3 <= kMaxStride) {
+      ARGBToRAWRow = ARGBToRAWRow_Any_NEON;
+    }
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToRAWRow = ARGBToRAWRow_NEON;
+    }
+  }
+#endif
 
   for (int y = 0; y < height; ++y) {
     I422ToARGBRow(src_y, src_u, src_v, row, width);

source/planar_functions.cc

@@ -406,6 +406,16 @@ int ARGBToRAW(const uint8* src_argb, int src_stride_argb,
     }
   }
 #endif
+#if defined(HAS_ARGBTORAWROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON)) {
+    if (width * 3 <= kMaxStride) {
+      ARGBToRAWRow = ARGBToRAWRow_Any_NEON;
+    }
+    if (IS_ALIGNED(width, 16)) {
+      ARGBToRAWRow = ARGBToRAWRow_NEON;
+    }
+  }
+#endif
 
   for (int y = 0; y < height; ++y) {
     ARGBToRAWRow(src_argb, dst_raw, width);

source/row_common.cc

@@ -981,6 +981,7 @@ RGBANY(ARGBToARGB4444Row_Any_SSE2, ARGBToARGB4444Row_SSE2, 2)
 #endif
 #if defined(HAS_ARGBTORGB24ROW_NEON)
 RGBANY(ARGBToRGB24Row_Any_NEON, ARGBToRGB24Row_NEON, 3)
+RGBANY(ARGBToRAWRow_Any_NEON, ARGBToRAWRow_NEON, 3)
 #endif
 #undef RGBANY
 

source/row_neon.cc

@@ -395,6 +395,24 @@ void ARGBToRGB24Row_NEON(const uint8* src_argb, uint8* dst_rgb24, int pix) {
 }
 #endif  // HAS_ARGBTORGB24ROW_NEON
 
+#ifdef HAS_ARGBTORAWROW_NEON
+void ARGBToRAWRow_NEON(const uint8* src_argb, uint8* dst_raw, int pix) {
+  asm volatile (
+  "1:                                          \n"
+    "vld4.u8    {q1,q2,q3,q4}, [%0]!           \n"  // load 16 pixels of ARGB.
+    "vswp.u8    q1, q3                         \n"  // swap R, B
+    "subs       %2, %2, #16                    \n"  // 16 processed per loop.
+    "vst3.u8    {q1,q2,q3}, [%1]!              \n"  // store 16 pixels of RAW.
+    "bgt        1b                             \n"
+  : "+r"(src_argb),  // %0
+    "+r"(dst_raw),   // %1
+    "+r"(pix)        // %2
+  :
+  : "memory", "cc", "q1", "q2", "q3", "q4" // Clobber List
+  );
+}
+#endif  // HAS_ARGBTORAWROW_NEON
+
 #endif  // __ARM_NEON__
 
 #ifdef __cplusplus