Some minor fixes for yuy2, r12 register, mask on any function.

BUG=none
TEST=none
Review URL: https://webrtc-codereview.appspot.com/936023

git-svn-id: http://libyuv.googlecode.com/svn/trunk@489 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 488
+Version: 489
 License: BSD
 License File: LICENSE
 

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 488
+#define LIBYUV_VERSION 489
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/convert.cc

@@ -579,59 +579,6 @@ int Q420ToI420(const uint8* src_y, int src_stride_y,
   return 0;
 }
 
-// Test if over reading on source is safe.
-// TODO(fbarchard): Find more efficient solution to safely do odd sizes.
-// Macros to control read policy, from slowest to fastest:
-// READSAFE_NEVER - disables read ahead on systems with strict memory reads
-// READSAFE_ODDHEIGHT - last row of odd height done with C.
-//   This policy assumes that the caller handles the last row of an odd height
-//   image using C.
-// READSAFE_PAGE - enable read ahead within same page.
-//   A page is 4096 bytes. When reading ahead, if the last pixel is near the
-//   end the page, and a read spans the page into the next page, a memory
-//   exception can occur if that page has not been allocated, or is a guard
-//   page. This setting ensures the overread is within the same page.
-// READSAFE_ALWAYS - enables read ahead on systems without memory exceptions
-//   or where buffers are padded by 64 bytes.
-
-#if defined(HAS_RGB24TOARGBROW_SSSE3) || \
-    defined(HAS_RGB24TOARGBROW_SSSE3) || \
-    defined(HAS_RAWTOARGBROW_SSSE3) || \
-    defined(HAS_RGB565TOARGBROW_SSE2) || \
-    defined(HAS_ARGB1555TOARGBROW_SSE2) || \
-    defined(HAS_ARGB4444TOARGBROW_SSE2)
-
-#define READSAFE_ODDHEIGHT
-
-static bool TestReadSafe(const uint8* src_yuy2, int src_stride_yuy2,
-                        int width, int height, int bpp, int overread) {
-  if (width > kMaxStride) {
-    return false;
-  }
-#if defined(READSAFE_ALWAYS)
-  return true;
-#elif defined(READSAFE_NEVER)
-  return false;
-#elif defined(READSAFE_ODDHEIGHT)
-  if (!(width & 15) ||
-      (src_stride_yuy2 >= 0 && (height & 1) && width * bpp >= overread)) {
-    return true;
-  }
-  return false;
-#elif defined(READSAFE_PAGE)
-  if (src_stride_yuy2 >= 0) {
-    src_yuy2 += (height - 1) * src_stride_yuy2;
-  }
-  uintptr_t last_adr = (uintptr_t)(src_yuy2) + width * bpp - 1;
-  uintptr_t last_read_adr = last_adr + overread - 1;
-  if (((last_adr ^ last_read_adr) & ~4095) == 0) {
-    return true;
-  }
-  return false;
-#endif
-}
-#endif
-
 // Convert YUY2 to I420.
 LIBYUV_API
 int YUY2ToI420(const uint8* src_yuy2, int src_stride_yuy2,

source/row_any.cc

@@ -282,7 +282,7 @@ UV422ANY(UYVYToUV422Row_Any_SSE2, UYVYToUV422Row_Unaligned_SSE2,
 #endif
 #ifdef HAS_YUY2TOUV422ROW_NEON
 UV422ANY(ARGBToUV444Row_Any_NEON, ARGBToUV444Row_NEON,
-         ARGBToUV444Row_C, 4, 8, 0)
+         ARGBToUV444Row_C, 4, 7, 0)
 UV422ANY(ARGBToUV422Row_Any_NEON, ARGBToUV422Row_NEON,
          ARGBToUV422Row_C, 4, 15, 1)
 UV422ANY(ARGBToUV411Row_Any_NEON, ARGBToUV411Row_NEON,

source/row_common.cc

@@ -517,7 +517,7 @@ void ARGBToUV422Row_C(const uint8* src_argb,
     dst_u += 1;
     dst_v += 1;
   }
-  if ((width & 3) == 1) {
+  if (width & 1) {
     uint8 ab = src_argb[0];
     uint8 ag = src_argb[1];
     uint8 ar = src_argb[2];

source/row_neon.cc

@@ -906,16 +906,17 @@ void MirrorRow_NEON(const uint8* src, uint8* dst, int width) {
 #endif  // HAS_MIRRORROW_NEON
 
 #ifdef HAS_MIRRORUVROW_NEON
-void MirrorUVRow_NEON(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int width) {
+void MirrorUVRow_NEON(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
+                      int width) {
   asm volatile (
     // Start at end of source row.
-    "mov        r3, #-16                       \n"
+    "mov        r12, #-16                      \n"
     "add        %0, %0, %3, lsl #1             \n"
     "sub        %0, #16                        \n"
 
     ".p2align  2                               \n"
   "1:                                          \n"
-    "vld2.8     {d0, d1}, [%0], r3             \n"  // src -= 16
+    "vld2.8     {d0, d1}, [%0], r12            \n"  // src -= 16
     "subs       %3, #8                         \n"  // 8 pixels per loop.
     "vrev64.8   q0, q0                         \n"
     "vst1.8     {d0}, [%1]!                    \n"  // dst += 8
@@ -926,7 +927,7 @@ void MirrorUVRow_NEON(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int width
     "+r"(dst_v),   // %2
     "+r"(width)    // %3
   :
-  : "memory", "cc", "r3", "q0"
+  : "memory", "cc", "r12", "q0"
   );
 }
 #endif  // HAS_MIRRORUVROW_NEON

source/scale_argb.cc

@@ -710,7 +710,6 @@ void ScaleARGBFilterRows_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
     , "xmm0", "xmm1", "xmm2", "xmm5"
 #endif
   );
-
 }
 #endif  // defined(__x86_64__) || defined(__i386__)
 

unit_test/convert_test.cc

@@ -823,8 +823,8 @@ TESTATOB(RGB24, 3, 3, 1, ARGB, 4, 4, 1, 0)
 TESTATOB(RGB565, 2, 2, 1, ARGB, 4, 4, 1, 0)
 TESTATOB(ARGB1555, 2, 2, 1, ARGB, 4, 4, 1, 0)
 TESTATOB(ARGB4444, 2, 2, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(YUY2, 2, 4, 1, ARGB, 4, 4, 1, 14)
-TESTATOB(UYVY, 2, 4, 1, ARGB, 4, 4, 1, 14)
+TESTATOB(YUY2, 2, 4, 1, ARGB, 4, 4, 1, 4)
+TESTATOB(UYVY, 2, 4, 1, ARGB, 4, 4, 1, 4)
 TESTATOB(BayerBGGR, 1, 2, 2, ARGB, 4, 4, 1, 0)
 TESTATOB(BayerRGGB, 1, 2, 2, ARGB, 4, 4, 1, 0)
 TESTATOB(BayerGBRG, 1, 2, 2, ARGB, 4, 4, 1, 0)