odd width support for ARGBSobel functions. Improves performance for images that…

odd width support for ARGBSobel functions.  Improves performance for images that are not a multiple of 8 pixels.
BUG=444
TESTED=libyuvTest.ARGBSobel_Opt
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/54589004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1415 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1416
+Version: 1417
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -1819,6 +1819,18 @@ void SobelXYRow_SSE2(const uint8* src_sobelx, const uint8* src_sobely,
                      uint8* dst_argb, int width);
 void SobelXYRow_NEON(const uint8* src_sobelx, const uint8* src_sobely,
                      uint8* dst_argb, int width);
+void SobelRow_Any_SSE2(const uint8* src_sobelx, const uint8* src_sobely,
+                       uint8* dst_argb, int width);
+void SobelRow_Any_NEON(const uint8* src_sobelx, const uint8* src_sobely,
+                       uint8* dst_argb, int width);
+void SobelToPlaneRow_Any_SSE2(const uint8* src_sobelx, const uint8* src_sobely,
+                              uint8* dst_y, int width);
+void SobelToPlaneRow_Any_NEON(const uint8* src_sobelx, const uint8* src_sobely,
+                              uint8* dst_y, int width);
+void SobelXYRow_Any_SSE2(const uint8* src_sobelx, const uint8* src_sobely,
+                         uint8* dst_argb, int width);
+void SobelXYRow_Any_NEON(const uint8* src_sobelx, const uint8* src_sobely,
+                         uint8* dst_argb, int width);
 
 void ARGBPolynomialRow_C(const uint8* src_argb,
                          uint8* dst_argb, const float* poly,

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1416
+#define LIBYUV_VERSION 1417
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/planar_functions.cc

@@ -2095,14 +2095,20 @@ int ARGBSobel(const uint8* src_argb, int src_stride_argb,
   void (*SobelRow)(const uint8* src_sobelx, const uint8* src_sobely,
                    uint8* dst_argb, int width) = SobelRow_C;
 #if defined(HAS_SOBELROW_SSE2)
-  if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 16)) {
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    SobelRow = SobelRow_Any_SSE2;
+    if (IS_ALIGNED(width, 16)) {
       SobelRow = SobelRow_SSE2;
     }
+  }
 #endif
 #if defined(HAS_SOBELROW_NEON)
-  if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) {
+  if (TestCpuFlag(kCpuHasNEON)) {
+    SobelRow = SobelRow_Any_NEON;
+    if (IS_ALIGNED(width, 8)) {
       SobelRow = SobelRow_NEON;
     }
+  }
 #endif
   return ARGBSobelize(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
                       width, height, SobelRow);
@@ -2116,14 +2122,20 @@ int ARGBSobelToPlane(const uint8* src_argb, int src_stride_argb,
   void (*SobelToPlaneRow)(const uint8* src_sobelx, const uint8* src_sobely,
                           uint8* dst_, int width) = SobelToPlaneRow_C;
 #if defined(HAS_SOBELTOPLANEROW_SSE2)
-  if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 16)) {
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    SobelToPlaneRow = SobelToPlaneRow_Any_SSE2;
+    if (IS_ALIGNED(width, 16)) {
       SobelToPlaneRow = SobelToPlaneRow_SSE2;
     }
+  }
 #endif
 #if defined(HAS_SOBELTOPLANEROW_NEON)
-  if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 16)) {
+  if (TestCpuFlag(kCpuHasNEON)) {
+    SobelToPlaneRow = SobelToPlaneRow_Any_NEON;
+    if (IS_ALIGNED(width, 16)) {
       SobelToPlaneRow = SobelToPlaneRow_NEON;
     }
+  }
 #endif
   return ARGBSobelize(src_argb, src_stride_argb, dst_y, dst_stride_y,
                       width, height, SobelToPlaneRow);
@@ -2138,14 +2150,20 @@ int ARGBSobelXY(const uint8* src_argb, int src_stride_argb,
   void (*SobelXYRow)(const uint8* src_sobelx, const uint8* src_sobely,
                      uint8* dst_argb, int width) = SobelXYRow_C;
 #if defined(HAS_SOBELXYROW_SSE2)
-  if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 16)) {
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    SobelXYRow = SobelXYRow_Any_SSE2;
+    if (IS_ALIGNED(width, 16)) {
       SobelXYRow = SobelXYRow_SSE2;
     }
+  }
 #endif
 #if defined(HAS_SOBELXYROW_NEON)
-  if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) {
+  if (TestCpuFlag(kCpuHasNEON)) {
+    SobelXYRow = SobelXYRow_Any_NEON;
+    if (IS_ALIGNED(width, 8)) {
       SobelXYRow = SobelXYRow_NEON;
     }
+  }
 #endif
   return ARGBSobelize(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
                       width, height, SobelXYRow);

source/row_any.cc

@@ -558,51 +558,71 @@ MERGEUVROW_ANY(MergeUVRow_Any_NEON, MergeUVRow_NEON, MergeUVRow_C, 15)
 #endif
 #undef MERGEUVROW_ANY
 
-#define MATHROW_ANY(NAMEANY, ARGBMATH_SIMD, ARGBMATH_C, MASK)                  \
+#define MATHROW_ANY(NAMEANY, ARGBMATH_SIMD, ARGBMATH_C, SBPP, DBPP, MASK)      \
     void NAMEANY(const uint8* src_argb0, const uint8* src_argb1,               \
                  uint8* dst_argb, int width) {                                 \
       int n = width & ~MASK;                                                   \
       if (n > 0) {                                                             \
         ARGBMATH_SIMD(src_argb0, src_argb1, dst_argb, n);                      \
       }                                                                        \
-      ARGBMATH_C(src_argb0 + n * 4,                                            \
-                 src_argb1 + n * 4,                                            \
-                 dst_argb + n * 4,                                             \
+      ARGBMATH_C(src_argb0 + n * SBPP,                                         \
+                 src_argb1 + n * SBPP,                                         \
+                 dst_argb + n * DBPP,                                          \
                  width & MASK);                                                \
     }
 
 #ifdef HAS_ARGBMULTIPLYROW_SSE2
 MATHROW_ANY(ARGBMultiplyRow_Any_SSE2, ARGBMultiplyRow_SSE2, ARGBMultiplyRow_C,
-            3)
+            4, 4, 3)
 #endif
 #ifdef HAS_ARGBADDROW_SSE2
-MATHROW_ANY(ARGBAddRow_Any_SSE2, ARGBAddRow_SSE2, ARGBAddRow_C, 3)
+MATHROW_ANY(ARGBAddRow_Any_SSE2, ARGBAddRow_SSE2, ARGBAddRow_C, 4, 4, 3)
 #endif
 #ifdef HAS_ARGBSUBTRACTROW_SSE2
 MATHROW_ANY(ARGBSubtractRow_Any_SSE2, ARGBSubtractRow_SSE2, ARGBSubtractRow_C,
-            3)
+            4, 4, 3)
 #endif
 #ifdef HAS_ARGBMULTIPLYROW_AVX2
 MATHROW_ANY(ARGBMultiplyRow_Any_AVX2, ARGBMultiplyRow_AVX2, ARGBMultiplyRow_C,
-            7)
+            4, 4, 7)
 #endif
 #ifdef HAS_ARGBADDROW_AVX2
-MATHROW_ANY(ARGBAddRow_Any_AVX2, ARGBAddRow_AVX2, ARGBAddRow_C, 7)
+MATHROW_ANY(ARGBAddRow_Any_AVX2, ARGBAddRow_AVX2, ARGBAddRow_C, 4, 4, 7)
 #endif
 #ifdef HAS_ARGBSUBTRACTROW_AVX2
 MATHROW_ANY(ARGBSubtractRow_Any_AVX2, ARGBSubtractRow_AVX2, ARGBSubtractRow_C,
-            7)
+            4, 4, 7)
 #endif
 #ifdef HAS_ARGBMULTIPLYROW_NEON
 MATHROW_ANY(ARGBMultiplyRow_Any_NEON, ARGBMultiplyRow_NEON, ARGBMultiplyRow_C,
-            7)
+            4, 4, 7)
 #endif
 #ifdef HAS_ARGBADDROW_NEON
-MATHROW_ANY(ARGBAddRow_Any_NEON, ARGBAddRow_NEON, ARGBAddRow_C, 7)
+MATHROW_ANY(ARGBAddRow_Any_NEON, ARGBAddRow_NEON, ARGBAddRow_C, 4, 4, 7)
 #endif
 #ifdef HAS_ARGBSUBTRACTROW_NEON
 MATHROW_ANY(ARGBSubtractRow_Any_NEON, ARGBSubtractRow_NEON, ARGBSubtractRow_C,
-            7)
+            4, 4, 7)
+#endif
+#ifdef HAS_SOBELROW_SSE2
+MATHROW_ANY(SobelRow_Any_SSE2, SobelRow_SSE2, SobelRow_C, 1, 4, 15)
+#endif
+#ifdef HAS_SOBELROW_NEON
+MATHROW_ANY(SobelRow_Any_NEON, SobelRow_NEON, SobelRow_C, 1, 4, 7)
+#endif
+#ifdef HAS_SOBELTOPLANEROW_SSE2
+MATHROW_ANY(SobelToPlaneRow_Any_SSE2, SobelToPlaneRow_SSE2, SobelToPlaneRow_C,
+            1, 1, 15)
+#endif
+#ifdef HAS_SOBELTOPLANEROW_NEON
+MATHROW_ANY(SobelToPlaneRow_Any_NEON, SobelToPlaneRow_NEON, SobelToPlaneRow_C,
+            1, 1, 7)
+#endif
+#ifdef HAS_SOBELXYROW_SSE2
+MATHROW_ANY(SobelXYRow_Any_SSE2, SobelXYRow_SSE2, SobelXYRow_C, 1, 4, 15)
+#endif
+#ifdef HAS_SOBELXYROW_NEON
+MATHROW_ANY(SobelXYRow_Any_NEON, SobelXYRow_NEON, SobelXYRow_C, 1, 4, 7)
 #endif
 #undef MATHROW_ANY