ARGBLumaColorTable function.

BUG=267
TEST=Luma*
R=thorcarpenter@google.com

Review URL: https://webrtc-codereview.appspot.com/2202004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@783 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 782
+Version: 783
 License: BSD
 License File: LICENSE
 

include/libyuv/planar_functions.h

@@ -210,6 +210,15 @@ int RGBColorTable(uint8* dst_argb, int dst_stride_argb,
                   const uint8* table_argb,
                   int x, int y, int width, int height);
 
+// Apply a luma/color table each ARGB pixel but preserve destination alpha.
+// Table contains 32768 values indexed by [Y][C] where 7 it 7 bit luma from
+// RGB (YJ style) and C is an 8 bit color component (R, G or B).
+LIBYUV_API
+int ARGBLumaColorTable(const uint8* src_argb, int src_stride_argb,
+                       uint8* dst_argb, int dst_stride_argb,
+                       const uint8* luma_rgb_table,
+                       int width, int height);
+
 // Apply a 3 term polynomial to ARGB values.
 // poly points to a 4x4 matrix.  The first row is constants.  The 2nd row is
 // coefficients for b, g, r and a.  The 3rd row is coefficients for b squared,

include/libyuv/row.h

@@ -144,6 +144,7 @@ extern "C" {
 #define HAS_ARGBCOLORTABLEROW_X86
 #define HAS_ARGBPOLYNOMIALROW_SSE2
 #define HAS_RGBCOLORTABLEROW_X86
+#define HAS_ARGBLUMACOLORTABLEROW_SSE2
 
 // Caveat: Visual C 2012 required for AVX2.
 #if _MSC_VER >= 1700
@@ -1554,8 +1555,15 @@ void ARGBPolynomialRow_SSE2(const uint8* src_argb,
                             uint8* dst_argb, const float* poly,
                             int width);
 void ARGBPolynomialRow_AVX2(const uint8* src_argb,
-                           uint8* dst_argb, const float* poly,
-                           int width);
+                            uint8* dst_argb, const float* poly,
+                            int width);
+
+void ARGBLumaColorTableRow_C(const uint8* src_argb,
+                             uint8* dst_argb, const uint8* luma,
+                             int width);
+void ARGBLumaColorTableRow_SSSE3(const uint8* src_argb,
+                                 uint8* dst_argb, const uint8* luma,
+                                 int width);
 
 // Divide num by div and return as 16.16 fixed point result.
 int FixedDiv_C(int num, int div);

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 782
+#define LIBYUV_VERSION 783
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/planar_functions.cc

@@ -2069,6 +2069,38 @@ int ARGBPolynomial(const uint8* src_argb, int src_stride_argb,
   return 0;
 }
 
+// Apply a lumacolortable to each ARGB pixel.
+LIBYUV_API
+int ARGBLumaColorTable(const uint8* src_argb, int src_stride_argb,
+                       uint8* dst_argb, int dst_stride_argb,
+                       const uint8* luma,
+                       int width, int height) {
+  if (!src_argb || !dst_argb || !luma || width <= 0 || height <= 0) {
+    return -1;
+  }
+  // Coalesce contiguous rows.
+  if (src_stride_argb == width * 4 && dst_stride_argb == width * 4) {
+    return ARGBLumaColorTable(src_argb, 0,
+                              dst_argb, 0,
+                              luma,
+                              width * height, 1);
+  }
+  void (*ARGBLumaColorTableRow)(const uint8* src_argb,
+                                uint8* dst_argb, const uint8* luma,
+                                int width) = ARGBLumaColorTableRow_C;
+#if defined(HAS_ARGBLUMACOLORTABLEROW_SSSE3)
+  if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 2)) {
+    ARGBLumaColorTableRow = ARGBLumaColorTableRow_SSSE3;
+  }
+#endif
+  for (int y = 0; y < height; ++y) {
+    ARGBLumaColorTableRow(src_argb, dst_argb, luma, width);
+    src_argb += src_stride_argb;
+    dst_argb += dst_stride_argb;
+  }
+  return 0;
+}
+
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv

source/row_common.cc

@@ -2051,6 +2051,39 @@ void ARGBPolynomialRow_C(const uint8* src_argb,
   }
 }
 
+// RGB to Luminance.
+// Leverage the fact that we want shifted left by 8 by the caller.
+//
+// Borrowed from libyuv/files/source/row_common.cc.
+// JPeg 7 bit Y:
+// b 0.11400 * 128 = 14.592 = 15
+// g 0.58700 * 128 = 75.136 = 75
+// r 0.29900 * 128 = 38.272 = 38
+static __inline unsigned int RGBToYJx256(uint8 r, uint8 g, uint8 b) {
+  return (38u * r + 75u * g + 15u * b) & 0x7F00u;
+}
+void ARGBLumaColorTableRow_C(const uint8* src_argb,
+                             uint8* dst_argb, const uint8* luma,
+                             int width) {
+  for (int i = 0; i < width - 1; i += 2) {
+    // Luminance in rows, color values in columns.
+    const uint8* luma0 = RGBToYJx256(src_argb[2], src_argb[1], src_argb[0]) +
+      luma;
+    dst_argb[0] = luma0[src_argb[0]];
+    dst_argb[1] = luma0[src_argb[1]];
+    dst_argb[2] = luma0[src_argb[2]];
+    dst_argb[3] = src_argb[3];
+    const uint8* luma1 = RGBToYJx256(src_argb[6], src_argb[5], src_argb[4]) +
+      luma;
+    dst_argb[4] = luma0[src_argb[4]];
+    dst_argb[5] = luma0[src_argb[5]];
+    dst_argb[6] = luma0[src_argb[6]];
+    dst_argb[7] = src_argb[7];
+    src_argb += 8;
+    dst_argb += 8;
+  }
+}
+
 #undef clamp0
 #undef clamp255
 

unit_test/planar_test.cc

@@ -1719,5 +1719,67 @@ TEST_F(libyuvTest, TestARGBPolynomial) {
   }
 }
 
+TEST_F(libyuvTest, TestARGBLumaColorTable) {
+  SIMD_ALIGNED(uint8 orig_pixels[1280][4]);
+  SIMD_ALIGNED(uint8 dst_pixels[1280][4]);
+
+  SIMD_ALIGNED(uint8 kLumaColorTable[32768]);
+  int v = 0;
+  for (int i = 0; i < 32768; ++i) {
+    kLumaColorTable[i] = v;
+    v += 3;
+  }
+  // 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.
+  ARGBLumaColorTable(&orig_pixels[0][0], 0, &dst_pixels[0][0], 0,
+                     &kLumaColorTable[0], 16, 1);
+  EXPECT_EQ(253u, dst_pixels[0][0]);
+  EXPECT_EQ(0u, dst_pixels[0][1]);
+  EXPECT_EQ(0u, dst_pixels[0][2]);
+  EXPECT_EQ(128u, dst_pixels[0][3]);
+  EXPECT_EQ(0u, dst_pixels[1][0]);
+  EXPECT_EQ(253u, dst_pixels[1][1]);
+  EXPECT_EQ(0u, dst_pixels[1][2]);
+  EXPECT_EQ(0u, dst_pixels[1][3]);
+  EXPECT_EQ(0u, dst_pixels[2][0]);
+  EXPECT_EQ(0u, dst_pixels[2][1]);
+  EXPECT_EQ(253u, dst_pixels[2][2]);
+  EXPECT_EQ(255u, dst_pixels[2][3]);
+  EXPECT_EQ(48u, dst_pixels[3][0]);
+  EXPECT_EQ(192u, dst_pixels[3][1]);
+  EXPECT_EQ(64u, dst_pixels[3][2]);
+  EXPECT_EQ(224u, dst_pixels[3][3]);
+
+  for (int i = 0; i < 1280; ++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 < benchmark_pixels_div1280_; ++i) {
+    ARGBLumaColorTable(&orig_pixels[0][0], 0, &dst_pixels[0][0], 0,
+                       &kLumaColorTable[0], 1280, 1);
+  }
+}
+
 
 }  // namespace libyuv