ARGBGrayTo takes source and destination pointers so caller does not need to make…

ARGBGrayTo takes source and destination pointers so caller does not need to make a copy of the image.
BUG=none
TEST=none
Review URL: https://webrtc-codereview.appspot.com/696004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@308 16f28f9a-4ce2-e073-06de-1de4eb20be90

include/libyuv/planar_functions.h

@@ -116,6 +116,11 @@ int I420Rect(uint8* dst_y, int dst_stride_y,
 int ARGBRect(uint8* dst_argb, int dst_stride_argb,
              int x, int y, int width, int height, uint32 value);
 
+// Convert ARGB to gray scale ARGB.
+int ARGBGrayTo(const uint8* src_argb, int src_stride_argb,
+               uint8* dst_argb, int dst_stride_argb,
+               int width, int height);
+
 // Make a rectangle of ARGB gray scale.
 int ARGBGray(uint8* dst_argb, int dst_stride_argb,
              int x, int y, int width, int height);

source/planar_functions.cc

@@ -802,7 +802,7 @@ int ARGBAttenuate(const uint8* src_argb, int src_stride_argb,
     ARGBAttenuateRow = ARGBAttenuateRow_SSE2;
   }
 #endif
-#if defined(HAS_ARGBATTENUATE_SSSE3)
+#if defined(HAS_ARGBATTENUATEROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 4) &&
       IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16) &&
       IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
@@ -832,7 +832,7 @@ int ARGBUnattenuate(const uint8* src_argb, int src_stride_argb,
   }
   void (*ARGBUnattenuateRow)(const uint8* src_argb, uint8* dst_argb,
                              int width) = ARGBUnattenuateRow_C;
-#if defined(HAS_ARGBUNATTENUATE_SSE2)
+#if defined(HAS_ARGBUNATTENUATEROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 4) &&
       IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16) &&
       IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
@@ -848,6 +848,36 @@ int ARGBUnattenuate(const uint8* src_argb, int src_stride_argb,
   return 0;
 }
 
+// Convert ARGB to Grayed ARGB.
+int ARGBGrayTo(const uint8* src_argb, int src_stride_argb,
+               uint8* dst_argb, int dst_stride_argb,
+               int width, int height) {
+  if (!src_argb || !dst_argb || width <= 0 || height == 0) {
+    return -1;
+  }
+  if (height < 0) {
+    height = -height;
+    src_argb = src_argb + (height - 1) * src_stride_argb;
+    src_stride_argb = -src_stride_argb;
+  }
+  void (*ARGBGrayRow)(const uint8* src_argb, uint8* dst_argb,
+                      int width) = ARGBGrayRow_C;
+#if defined(HAS_ARGBGRAYROW_SSSE3)
+  if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 8) &&
+      IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16) &&
+      IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
+    ARGBGrayRow = ARGBGrayRow_SSSE3;
+  }
+#endif
+
+  for (int y = 0; y < height; ++y) {
+    ARGBGrayRow(src_argb, dst_argb, width);
+    src_argb += src_stride_argb;
+    dst_argb += dst_stride_argb;
+  }
+  return 0;
+}
+
 // Make a rectangle of ARGB gray scale.
 int ARGBGray(uint8* dst_argb, int dst_stride_argb,
              int dst_x, int dst_y,
@@ -855,7 +885,8 @@ int ARGBGray(uint8* dst_argb, int dst_stride_argb,
   if (!dst_argb || width <= 0 || height <= 0 || dst_x < 0 || dst_y < 0) {
     return -1;
   }
-  void (*ARGBGrayRow)(uint8* dst_argb, int width) = ARGBGrayRow_C;
+  void (*ARGBGrayRow)(const uint8* src_argb, uint8* dst_argb,
+                      int width) = ARGBGrayRow_C;
 #if defined(HAS_ARGBGRAYROW_SSSE3)
   if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 8) &&
       IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
@@ -864,7 +895,7 @@ int ARGBGray(uint8* dst_argb, int dst_stride_argb,
 #endif
   uint8* dst = dst_argb + dst_y * dst_stride_argb + dst_x * 4;
   for (int y = 0; y < height; ++y) {
-    ARGBGrayRow(dst, width);
+    ARGBGrayRow(dst, dst, width);
     dst += dst_stride_argb;
   }
   return 0;

source/row.h

@@ -39,7 +39,7 @@ extern "C" {
 #define HAS_ABGRTOYROW_SSSE3
 #define HAS_ARGB1555TOARGBROW_SSE2
 #define HAS_ARGB4444TOARGBROW_SSE2
-#define HAS_ARGBATTENUATE_SSSE3
+#define HAS_ARGBATTENUATEROW_SSSE3
 #define HAS_ARGBBLENDROW_SSSE3
 #define HAS_ARGBTOARGB1555ROW_SSE2
 #define HAS_ARGBTOARGB4444ROW_SSE2
@@ -48,7 +48,7 @@ extern "C" {
 #define HAS_ARGBTORGB565ROW_SSE2
 #define HAS_ARGBTOUVROW_SSSE3
 #define HAS_ARGBTOYROW_SSSE3
-#define HAS_ARGBUNATTENUATE_SSE2
+#define HAS_ARGBUNATTENUATEROW_SSE2
 #define HAS_BGRATOARGBROW_SSSE3
 #define HAS_BGRATOUVROW_SSSE3
 #define HAS_BGRATOYROW_SSSE3
@@ -489,8 +489,8 @@ extern uint32 fixed_invtbl8[256];
 void ARGBUnattenuateRow_C(const uint8* src_argb, uint8* dst_argb, int width);
 void ARGBUnattenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb, int width);
 
-void ARGBGrayRow_C(uint8* dst_argb, int width);
-void ARGBGrayRow_SSSE3(uint8* dst_argb, int width);
+void ARGBGrayRow_C(const uint8* src_argb, uint8* dst_argb, int width);
+void ARGBGrayRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width);
 
 void ARGBSepiaRow_C(uint8* dst_argb, int width);
 void ARGBSepiaRow_SSSE3(uint8* dst_argb, int width);

source/row_common.cc

@@ -279,11 +279,13 @@ static __inline int RGBToGray(uint8 r, uint8 g, uint8 b) {
   return (( 76 * r + 152 * g +  28 * b) >> 8);
 }
 
-void ARGBGrayRow_C(uint8* dst_argb, int width) {
+void ARGBGrayRow_C(const uint8* src_argb, uint8* dst_argb, int width) {
   for (int x = 0; x < width; ++x) {
-    uint8 y = RGBToGray(dst_argb[2], dst_argb[1], dst_argb[0]);
+    uint8 y = RGBToGray(src_argb[2], src_argb[1], src_argb[0]);
     dst_argb[2] = dst_argb[1] = dst_argb[0] = y;
+    dst_argb[3] = src_argb[3];
     dst_argb += 4;
+    src_argb += 4;
   }
 }
 

source/row_posix.cc

@@ -2656,7 +2656,7 @@ void ARGBAttenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb, int width) {
 }
 #endif  // HAS_ARGBATTENUATE_SSE2
 
-#ifdef HAS_ARGBATTENUATE_SSSE3
+#ifdef HAS_ARGBATTENUATEROW_SSSE3
 // Shuffle table duplicating alpha
 CONST uvec8 kShuffleAlpha0 = {
   3u, 3u, 3u, 3u, 3u, 3u, 128u, 128u, 7u, 7u, 7u, 7u, 7u, 7u, 128u, 128u,
@@ -2709,9 +2709,9 @@ void ARGBAttenuateRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width) {
 #endif
   );
 }
-#endif  // HAS_ARGBATTENUATE_SSSE3
+#endif  // HAS_ARGBATTENUATEROW_SSSE3
 
-#ifdef HAS_ARGBUNATTENUATE_SSE2
+#ifdef HAS_ARGBUNATTENUATEROW_SSE2
 // Unattenuate 4 pixels at a time.
 // aligned to 16 bytes
 void ARGBUnattenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb,
@@ -2764,7 +2764,7 @@ void ARGBUnattenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb,
 #endif
   );
 }
-#endif  // HAS_ARGBUNATTENUATE_SSE2
+#endif  // HAS_ARGBUNATTENUATEROW_SSE2
 
 #ifdef HAS_ARGBGRAYROW_SSSE3
 // Constant for ARGB color to gray scale.  0.11 * B + 0.59 * G + 0.30 * R
@@ -2773,9 +2773,10 @@ CONST vec8 kARGBToGray = {
 };
 
 // Convert 8 ARGB pixels (64 bytes) to 8 Gray ARGB pixels
-void ARGBGrayRow_SSSE3(uint8* dst_argb, int width) {
+void ARGBGrayRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width) {
   asm volatile (
-    "movdqa    %2,%%xmm4                       \n"
+    "movdqa    %3,%%xmm4                       \n"
+    "sub       %0,%1                           \n"
 
     // 8 pixel loop.
     ".p2align  4                               \n"
@@ -2799,14 +2800,15 @@ void ARGBGrayRow_SSSE3(uint8* dst_argb, int width) {
     "movdqa    %%xmm0,%%xmm1                   \n"
     "punpcklwd %%xmm3,%%xmm0                   \n"
     "punpckhwd %%xmm3,%%xmm1                   \n"
-    "sub       $0x8,%1                         \n"
-    "movdqa    %%xmm0,(%0)                     \n"
-    "movdqa    %%xmm1,0x10(%0)                 \n"
+    "sub       $0x8,%2                         \n"
+    "movdqa    %%xmm0,(%0,%1,1)                \n"
+    "movdqa    %%xmm1,0x10(%0,%1,1)            \n"
     "lea       0x20(%0),%0                     \n"
     "jg        1b                              \n"
-  : "+r"(dst_argb),   // %0
-    "+r"(width)       // %1
-  : "m"(kARGBToGray)  // %2
+  : "+r"(src_argb),   // %0
+    "+r"(dst_argb),   // %1
+    "+r"(width)       // %2
+  : "m"(kARGBToGray)  // %3
   : "memory", "cc"
 #if defined(__SSE2__)
     , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4"

source/row_win.cc

@@ -2741,7 +2741,7 @@ void ARGBAttenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb, int width) {
 }
 #endif  // HAS_ARGBATTENUATE_SSE2
 
-#ifdef HAS_ARGBATTENUATE_SSSE3
+#ifdef HAS_ARGBATTENUATEROW_SSSE3
 // Shuffle table duplicating alpha.
 static const uvec8 kShuffleAlpha0 = {
   3u, 3u, 3u, 3u, 3u, 3u, 128u, 128u, 7u, 7u, 7u, 7u, 7u, 7u, 128u, 128u,
@@ -2788,9 +2788,9 @@ void ARGBAttenuateRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width) {
     ret
   }
 }
-#endif  // HAS_ARGBATTENUATE_SSSE3
+#endif  // HAS_ARGBATTENUATEROW_SSSE3
 
-#ifdef HAS_ARGBUNATTENUATE_SSE2
+#ifdef HAS_ARGBUNATTENUATEROW_SSE2
 // Unattenuate 4 pixels at a time.
 // Aligned to 16 bytes.
 __declspec(naked) __declspec(align(16))
@@ -2843,7 +2843,7 @@ void ARGBUnattenuateRow_SSE2(const uint8* src_argb, uint8* dst_argb,
     ret
   }
 }
-#endif  // HAS_ARGBUNATTENUATE_SSE2
+#endif  // HAS_ARGBUNATTENUATEROW_SSE2
 
 #ifdef HAS_ARGBGRAYROW_SSSE3
 // Constant for ARGB color to gray scale: 0.11 * B + 0.59 * G + 0.30 * R
@@ -2853,11 +2853,13 @@ static const vec8 kARGBToGray = {
 
 // Convert 8 ARGB pixels (64 bytes) to 8 Gray ARGB pixels.
 __declspec(naked) __declspec(align(16))
-void ARGBGrayRow_SSSE3(uint8* dst_argb, int width) {
+void ARGBGrayRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width) {
   __asm {
-    mov        eax, [esp + 4]   /* dst_argb */
-    mov        ecx, [esp + 8]   /* width */
+    mov        eax, [esp + 4]   /* src_argb */
+    mov        edx, [esp + 8]   /* dst_argb */
+    mov        ecx, [esp + 12]  /* width */
     movdqa     xmm4, kARGBToGray
+    sub        edx, eax
 
     align      16
  convertloop:
@@ -2881,8 +2883,8 @@ void ARGBGrayRow_SSSE3(uint8* dst_argb, int width) {
     punpcklwd  xmm0, xmm3   // GGGA first 4
     punpckhwd  xmm1, xmm3   // GGGA next 4
     sub        ecx, 8
-    movdqa     [eax], xmm0
-    movdqa     [eax + 16], xmm1
+    movdqa     [eax + edx], xmm0
+    movdqa     [eax + edx + 16], xmm1
     lea        eax, [eax + 32]
     jg         convertloop
     ret

unit_test/planar_test.cc

@@ -486,6 +486,61 @@ TEST_F(libyuvTest, TestARGBGray) {
   }
 }
 
+TEST_F(libyuvTest, TestARGBGrayTo) {
+  SIMD_ALIGNED(uint8 orig_pixels[256][4]);
+  SIMD_ALIGNED(uint8 gray_pixels[256][4]);
+
+  // Test blue
+  orig_pixels[0][0] = 255u;
+  orig_pixels[0][1] = 0u;
+  orig_pixels[0][2] = 0u;
+  orig_pixels[0][3] = 128u;
+  // Test green
+  orig_pixels[1][0] = 0u;
+  orig_pixels[1][1] = 255u;
+  orig_pixels[1][2] = 0u;
+  orig_pixels[1][3] = 0u;
+  // Test red
+  orig_pixels[2][0] = 0u;
+  orig_pixels[2][1] = 0u;
+  orig_pixels[2][2] = 255u;
+  orig_pixels[2][3] = 255u;
+  // Test color
+  orig_pixels[3][0] = 16u;
+  orig_pixels[3][1] = 64u;
+  orig_pixels[3][2] = 192u;
+  orig_pixels[3][3] = 224u;
+  // Do 16 to test asm version.
+  ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 16, 1);
+  EXPECT_EQ(27u, gray_pixels[0][0]);
+  EXPECT_EQ(27u, gray_pixels[0][1]);
+  EXPECT_EQ(27u, gray_pixels[0][2]);
+  EXPECT_EQ(128u, gray_pixels[0][3]);
+  EXPECT_EQ(151u, gray_pixels[1][0]);
+  EXPECT_EQ(151u, gray_pixels[1][1]);
+  EXPECT_EQ(151u, gray_pixels[1][2]);
+  EXPECT_EQ(0u, gray_pixels[1][3]);
+  EXPECT_EQ(75u, gray_pixels[2][0]);
+  EXPECT_EQ(75u, gray_pixels[2][1]);
+  EXPECT_EQ(75u, gray_pixels[2][2]);
+  EXPECT_EQ(255u, gray_pixels[2][3]);
+  EXPECT_EQ(96u, gray_pixels[3][0]);
+  EXPECT_EQ(96u, gray_pixels[3][1]);
+  EXPECT_EQ(96u, gray_pixels[3][2]);
+  EXPECT_EQ(224u, gray_pixels[3][3]);
+
+  for (int i = 0; i < 256; ++i) {
+    orig_pixels[i][0] = i;
+    orig_pixels[i][1] = i / 2;
+    orig_pixels[i][2] = i / 3;
+    orig_pixels[i][3] = i;
+  }
+
+  for (int i = 0; i < 1000 * 1280 * 720 / 256; ++i) {
+    ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 256, 1);
+  }
+}
+
 TEST_F(libyuvTest, TestARGBSepia) {
   SIMD_ALIGNED(uint8 orig_pixels[256][4]);