rgb565 and argb1555 neon

BUG=none
TEST=none
Review URL: https://webrtc-codereview.appspot.com/881004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@420 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 418
+Version: 420
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -150,6 +150,8 @@ extern "C" {
 #define HAS_YUY2TOYROW_NEON
 #define HAS_I422TOYUY2ROW_NEON
 #define HAS_I422TOUYVYROW_NEON
+#define HAS_ARGBTORGB565ROW_NEON
+#define HAS_ARGBTOARGB1555ROW_NEON
 #define HAS_ARGBTOARGB4444ROW_NEON
 #endif
 
@@ -308,8 +310,8 @@ void ABGRToARGBRow_SSSE3(const uint8* src_abgr, uint8* dst_argb, int pix);
 void RGBAToARGBRow_SSSE3(const uint8* src_rgba, uint8* dst_argb, int pix);
 void RGB24ToARGBRow_SSSE3(const uint8* src_rgb24, uint8* dst_argb, int pix);
 void RAWToARGBRow_SSSE3(const uint8* src_rgb24, uint8* dst_argb, int pix);
-void ARGB1555ToARGBRow_SSE2(const uint8* src_argb, uint8* dst_argb, int pix);
 void RGB565ToARGBRow_SSE2(const uint8* src_argb, uint8* dst_argb, int pix);
+void ARGB1555ToARGBRow_SSE2(const uint8* src_argb, uint8* dst_argb, int pix);
 void ARGB4444ToARGBRow_SSE2(const uint8* src_argb, uint8* dst_argb, int pix);
 
 void BGRAToARGBRow_NEON(const uint8* src_bgra, uint8* dst_argb, int pix);
@@ -337,6 +339,8 @@ 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 ARGBToRGB565Row_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
+void ARGBToARGB1555Row_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
 void ARGBToARGB4444Row_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
 
 void ARGBToRGBARow_C(const uint8* src_argb, uint8* dst_rgb, int pix);
@@ -593,6 +597,8 @@ 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 ARGBToRGB565Row_Any_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
+void ARGBToARGB1555Row_Any_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
 void ARGBToARGB4444Row_Any_NEON(const uint8* src_argb, uint8* dst_rgb, int pix);
 
 void ARGBToYRow_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 418
+#define LIBYUV_VERSION 420
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/convert_from_argb.cc

@@ -245,6 +245,13 @@ int ARGBToRGB565(const uint8* src_argb, int src_stride_argb,
       ARGBToRGB565Row = ARGBToRGB565Row_SSE2;
     }
   }
+#elif defined(HAS_ARGBTORGB565ROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
+    ARGBToRGB565Row = ARGBToRGB565Row_Any_NEON;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToRGB565Row = ARGBToRGB565Row_NEON;
+    }
+  }
 #endif
 
   for (int y = 0; y < height; ++y) {
@@ -278,6 +285,13 @@ int ARGBToARGB1555(const uint8* src_argb, int src_stride_argb,
       ARGBToARGB1555Row = ARGBToARGB1555Row_SSE2;
     }
   }
+#elif defined(HAS_ARGBTOARGB1555ROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
+    ARGBToARGB1555Row = ARGBToARGB1555Row_Any_NEON;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToARGB1555Row = ARGBToARGB1555Row_NEON;
+    }
+  }
 #endif
 
   for (int y = 0; y < height; ++y) {

source/row_common.cc

@@ -1073,6 +1073,10 @@ RGBANY(ARGBToARGB4444Row_Any_SSE2, ARGBToARGB4444Row_SSE2, ARGBToARGB4444Row_C,
 #if defined(HAS_ARGBTORGB24ROW_NEON)
 RGBANY(ARGBToRGB24Row_Any_NEON, ARGBToRGB24Row_NEON, ARGBToRGB24Row_C, 7, 4, 3)
 RGBANY(ARGBToRAWRow_Any_NEON, ARGBToRAWRow_NEON, ARGBToRAWRow_C, 7, 4, 3)
+RGBANY(ARGBToRGB565Row_Any_NEON, ARGBToRGB565Row_NEON, ARGBToRGB565Row_C,
+       7, 4, 2)
+RGBANY(ARGBToARGB1555Row_Any_NEON, ARGBToARGB1555Row_NEON, ARGBToARGB1555Row_C,
+       7, 4, 2)
 RGBANY(ARGBToARGB4444Row_Any_NEON, ARGBToARGB4444Row_NEON, ARGBToARGB4444Row_C,
        7, 4, 2)
 #endif

source/row_neon.cc

@@ -908,6 +908,67 @@ void I422ToUYVYRow_NEON(const uint8* src_y,
   );
 }
 
+#ifdef HAS_ARGBTORGB565ROW_NEON
+void ARGBToRGB565Row_NEON(const uint8* src_argb, uint8* dst_rgb565, int pix) {
+  asm volatile (
+    ".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.
+    "vshr.u8    d0, d0, #3                     \n"  // B
+    "vshr.u8    d1, d1, #2                     \n"  // G
+    "vshr.u8    d2, d2, #3                     \n"  // R
+    "vmovl.u8   q8, d0                         \n"  // B
+    "vmovl.u8   q9, d1                         \n"  // G
+    "vmovl.u8   q10, d2                        \n"  // R
+    "vshl.u16   q9, q9, #5                     \n"  // G
+    "vshl.u16   q10, q10, #11                  \n"  // R
+    "vorr       q0, q8, q9                     \n"  // BG
+    "vorr       q0, q0, q10                    \n"  // BGR
+    "vst1.8     {q0}, [%1]!                    \n"  // store 8 pixels RGB565.
+    "bgt        1b                             \n"
+  : "+r"(src_argb),  // %0
+    "+r"(dst_rgb565),  // %1
+    "+r"(pix)        // %2
+  :
+  : "memory", "cc", "d0", "d1", "d2", "d3", "d4", "q8", "q9", "q10"
+  );
+}
+#endif  // HAS_ARGBTORGB565ROW_NEON
+
+#ifdef HAS_ARGBTOARGB1555ROW_NEON
+void ARGBToARGB1555Row_NEON(const uint8* src_argb, uint8* dst_argb1555,
+                            int pix) {
+  asm volatile (
+    ".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.
+    "vshr.u8    d0, d0, #3                     \n"  // B
+    "vshr.u8    d1, d1, #3                     \n"  // G
+    "vshr.u8    d2, d2, #3                     \n"  // R
+    "vshr.u8    d3, d3, #7                     \n"  // A
+    "vmovl.u8   q8, d0                         \n"  // B
+    "vmovl.u8   q9, d1                         \n"  // G
+    "vmovl.u8   q10, d2                        \n"  // R
+    "vmovl.u8   q11, d3                        \n"  // A
+    "vshl.u16   q9, q9, #5                     \n"  // G
+    "vshl.u16   q10, q10, #10                  \n"  // R
+    "vshl.u16   q11, q11, #15                  \n"  // A
+    "vorr       q0, q8, q9                     \n"  // BG
+    "vorr       q1, q10, q11                   \n"  // RA
+    "vorr       q0, q0, q1                     \n"  // BGRA
+    "vst1.8     {q0}, [%1]!                    \n"  // store 8 pixels ARGB1555.
+    "bgt        1b                             \n"
+  : "+r"(src_argb),  // %0
+    "+r"(dst_argb1555),  // %1
+    "+r"(pix)        // %2
+  :
+  : "memory", "cc", "d0", "d1", "d2", "d3", "q8", "q9", "q10", "q11"
+  );
+}
+#endif  // HAS_ARGBTOARGB1555ROW_NEON
+
 #ifdef HAS_ARGBTOARGB4444ROW_NEON
 void ARGBToARGB4444Row_NEON(const uint8* src_argb, uint8* dst_argb4444,
                             int pix) {