Dither from I420 to RGB565 in 2 steps - I420ToARGB then ARGBToRGB565.

BUG=407
TESTED=untested
R=brucedawson@google.com, tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/48429004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1315 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1314
+Version: 1315
 License: BSD
 License File: LICENSE
 

include/libyuv/convert_from.h

@@ -137,6 +137,17 @@ int I420ToRGB565(const uint8* src_y, int src_stride_y,
                  uint8* dst_frame, int dst_stride_frame,
                  int width, int height);
 
+// Convert I420 To RGB565 with 4x4 dither matrix (16 bytes).
+// Values in dither matrix from 0 to 7 recommended.
+// The order of the dither matrix is first byte is upper left.
+
+LIBYUV_API
+int I420ToRGB565Dither(const uint8* src_y, int src_stride_y,
+                       const uint8* src_u, int src_stride_u,
+                       const uint8* src_v, int src_stride_v,
+                       uint8* dst_frame, int dst_stride_frame,
+                       const uint8* dither4x4, int width, int height);
+
 LIBYUV_API
 int I420ToARGB1555(const uint8* src_y, int src_stride_y,
                    const uint8* src_u, int src_stride_u,

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1314
+#define LIBYUV_VERSION 1315
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/convert_from.cc

@@ -993,6 +993,114 @@ int I420ToRGB565(const uint8* src_y, int src_stride_y,
   return 0;
 }
 
+// Ordered 8x8 dither for 888 to 565.  Values from 0 to 7.
+static const uint8 kDither565_4x4[16] = {
+  0, 4, 1, 5,
+  6, 2, 7, 3,
+  1, 5, 0, 4,
+  7, 3, 6, 2,
+};
+
+// Convert I420 to RGB565 with dithering.
+LIBYUV_API
+int I420ToRGB565Dither(const uint8* src_y, int src_stride_y,
+                       const uint8* src_u, int src_stride_u,
+                       const uint8* src_v, int src_stride_v,
+                       uint8* dst_rgb565, int dst_stride_rgb565,
+                       const uint8* dither4x4, int width, int height) {
+  int y;
+  void (*I422ToRGB565Row)(const uint8* y_buf,
+                          const uint8* u_buf,
+                          const uint8* v_buf,
+                          uint8* rgb_buf,
+                          int width) = I422ToRGB565Row_C;
+  void (*I422ToARGBRow)(const uint8* y_buf,
+                        const uint8* u_buf,
+                        const uint8* v_buf,
+                        uint8* rgb_buf,
+                        int width) = I422ToARGBRow_C;
+  void (*ARGBToRGB565DitherRow)(const uint8* src_argb, uint8* dst_rgb,
+      const uint32 dither4, int pix) = ARGBToRGB565DitherRow_C;
+  if (!src_y || !src_u || !src_v || !dst_rgb565 ||
+      width <= 0 || height == 0) {
+    return -1;
+  }
+  // Negative height means invert the image.
+  if (height < 0) {
+    height = -height;
+    dst_rgb565 = dst_rgb565 + (height - 1) * dst_stride_rgb565;
+    dst_stride_rgb565 = -dst_stride_rgb565;
+  }
+  if (!dither4x4) {
+    dither4x4 = kDither565_4x4;
+  }
+#if defined(HAS_I422TOARGBROW_SSSE3)
+  if (TestCpuFlag(kCpuHasSSSE3)) {
+    I422ToARGBRow = I422ToARGBRow_Any_SSSE3;
+    if (IS_ALIGNED(width, 8)) {
+      I422ToARGBRow = I422ToARGBRow_SSSE3;
+    }
+  }
+#endif
+#if defined(HAS_I422TOARGBROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    I422ToARGBRow = I422ToARGBRow_Any_AVX2;
+    if (IS_ALIGNED(width, 16)) {
+      I422ToARGBRow = I422ToARGBRow_AVX2;
+    }
+  }
+#endif
+#if defined(HAS_I422TOARGBROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON)) {
+    I422ToARGBRow = I422ToARGBRow_Any_NEON;
+    if (IS_ALIGNED(width, 8)) {
+      I422ToARGBRow = I422ToARGBRow_NEON;
+    }
+  }
+#endif
+#if defined(HAS_I422TOARGBROW_MIPS_DSPR2)
+  if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(width, 4) &&
+      IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) &&
+      IS_ALIGNED(src_u, 2) && IS_ALIGNED(src_stride_u, 2) &&
+      IS_ALIGNED(src_v, 2) && IS_ALIGNED(src_stride_v, 2)) {
+    I422ToARGBRow = I422ToARGBRow_MIPS_DSPR2;
+  }
+#endif
+#if defined(HAS_ARGBTORGB565DITHERROW_SSE2)
+  if (TestCpuFlag(kCpuHasSSE2)) {
+    ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_Any_SSE2;
+    if (IS_ALIGNED(width, 4)) {
+      ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_SSE2;
+    }
+  }
+#endif
+#if defined(HAS_ARGBTORGB565DITHERROW_AVX2)
+  if (TestCpuFlag(kCpuHasAVX2)) {
+    ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_Any_AVX2;
+    if (IS_ALIGNED(width, 8)) {
+      ARGBToRGB565DitherRow = ARGBToRGB565DitherRow_AVX2;
+    }
+  }
+#endif
+  {
+    // Allocate a row of argb.
+    align_buffer_64(row_argb, width * 4);
+    for (y = 0; y < height; ++y) {
+      I422ToARGBRow(src_y, src_u, src_v, row_argb, width);
+      ARGBToRGB565DitherRow(row_argb, dst_rgb565,
+                            *(uint32*)(dither4x4 + ((y & 3) << 2)), width);
+      dst_rgb565 += dst_stride_rgb565;
+      src_y += src_stride_y;
+      if (y & 1) {
+        src_u += src_stride_u;
+        src_v += src_stride_v;
+      }
+    }
+    free_aligned_buffer_64(row_argb);
+  }
+  return 0;
+}
+
 // Convert I420 to specified format
 LIBYUV_API
 int ConvertFromI420(const uint8* y, int y_stride,

unit_test/convert_test.cc

@@ -496,7 +496,6 @@ TEST_F(libyuvTest, FMT_PLANAR##To##FMT_B##N) {                                 \
     TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
         YALIGN, benchmark_width_, DIFF, _Opt, +, 0, FMT_C, BPP_C)
 
-// TODO(fbarchard): Make vertical alignment unnecessary on bayer.
 TESTPLANARTOB(I420, 2, 2, ARGB, 4, 4, 1, 2, ARGB, 4)
 TESTPLANARTOB(J420, 2, 2, ARGB, 4, 4, 1, 2, ARGB, 4)
 TESTPLANARTOB(I420, 2, 2, BGRA, 4, 4, 1, 2, ARGB, 4)
@@ -1451,4 +1450,84 @@ TEST_F(libyuvTest, TestDither) {
   free_aligned_buffer_64(dst_argbdither);
 }
 
+#define TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
+                       YALIGN, W1280, DIFF, N, NEG, OFF, FMT_C, BPP_C)         \
+TEST_F(libyuvTest, FMT_PLANAR##To##FMT_B##Dither##N) {                         \
+  const int kWidth = ((W1280) > 0) ? (W1280) : 1;                              \
+  const int kHeight = ALIGNINT(benchmark_height_, YALIGN);                     \
+  const int kStrideB = ALIGNINT(kWidth * BPP_B, ALIGN);                        \
+  const int kSizeUV =                                                          \
+    SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y);              \
+  align_buffer_64(src_y, kWidth * kHeight + OFF);                              \
+  align_buffer_64(src_u, kSizeUV + OFF);                                       \
+  align_buffer_64(src_v, kSizeUV + OFF);                                       \
+  align_buffer_64(dst_argb_c, kStrideB * kHeight + OFF);                       \
+  align_buffer_64(dst_argb_opt, kStrideB * kHeight + OFF);                     \
+  srandom(time(NULL));                                                         \
+  for (int i = 0; i < kWidth * kHeight; ++i) {                                 \
+    src_y[i + OFF] = (random() & 0xff);                                        \
+  }                                                                            \
+  for (int i = 0; i < kSizeUV; ++i) {                                          \
+    src_u[i + OFF] = (random() & 0xff);                                        \
+    src_v[i + OFF] = (random() & 0xff);                                        \
+  }                                                                            \
+  memset(dst_argb_c + OFF, 1, kStrideB * kHeight);                             \
+  memset(dst_argb_opt + OFF, 101, kStrideB * kHeight);                         \
+  MaskCpuFlags(0);                                                             \
+  FMT_PLANAR##To##FMT_B##Dither(src_y + OFF, kWidth,                           \
+                        src_u + OFF, SUBSAMPLE(kWidth, SUBSAMP_X),             \
+                        src_v + OFF, SUBSAMPLE(kWidth, SUBSAMP_X),             \
+                        dst_argb_c + OFF, kStrideB,                            \
+                        NULL, kWidth, NEG kHeight);                            \
+  MaskCpuFlags(-1);                                                            \
+  for (int i = 0; i < benchmark_iterations_; ++i) {                            \
+    FMT_PLANAR##To##FMT_B##Dither(src_y + OFF, kWidth,                         \
+                          src_u + OFF, SUBSAMPLE(kWidth, SUBSAMP_X),           \
+                          src_v + OFF, SUBSAMPLE(kWidth, SUBSAMP_X),           \
+                          dst_argb_opt + OFF, kStrideB,                        \
+                          NULL, kWidth, NEG kHeight);                          \
+  }                                                                            \
+  int max_diff = 0;                                                            \
+  /* Convert to ARGB so 565 is expanded to bytes that can be compared. */      \
+  align_buffer_64(dst_argb32_c, kWidth * BPP_C  * kHeight);                    \
+  align_buffer_64(dst_argb32_opt, kWidth * BPP_C  * kHeight);                  \
+  memset(dst_argb32_c, 2, kWidth * BPP_C  * kHeight);                          \
+  memset(dst_argb32_opt, 102, kWidth * BPP_C  * kHeight);                      \
+  FMT_B##To##FMT_C(dst_argb_c + OFF, kStrideB,                                 \
+                   dst_argb32_c, kWidth * BPP_C ,                              \
+                   kWidth, kHeight);                                           \
+  FMT_B##To##FMT_C(dst_argb_opt + OFF, kStrideB,                               \
+                   dst_argb32_opt, kWidth * BPP_C ,                            \
+                   kWidth, kHeight);                                           \
+  for (int i = 0; i < kWidth * BPP_C * kHeight; ++i) {                         \
+    int abs_diff =                                                             \
+        abs(static_cast<int>(dst_argb32_c[i]) -                                \
+            static_cast<int>(dst_argb32_opt[i]));                              \
+    if (abs_diff > max_diff) {                                                 \
+      max_diff = abs_diff;                                                     \
+    }                                                                          \
+  }                                                                            \
+  EXPECT_LE(max_diff, DIFF);                                                   \
+  free_aligned_buffer_64(src_y);                                               \
+  free_aligned_buffer_64(src_u);                                               \
+  free_aligned_buffer_64(src_v);                                               \
+  free_aligned_buffer_64(dst_argb_c);                                          \
+  free_aligned_buffer_64(dst_argb_opt);                                        \
+  free_aligned_buffer_64(dst_argb32_c);                                        \
+  free_aligned_buffer_64(dst_argb32_opt);                                      \
+}
+
+#define TESTPLANARTOBD(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
+                      YALIGN, DIFF, FMT_C, BPP_C)                              \
+    TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,     \
+        YALIGN, benchmark_width_ - 4, DIFF, _Any, +, 0, FMT_C, BPP_C)          \
+    TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,     \
+        YALIGN, benchmark_width_, DIFF, _Unaligned, +, 1, FMT_C, BPP_C)        \
+    TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,     \
+        YALIGN, benchmark_width_, DIFF, _Invert, -, 0, FMT_C, BPP_C)           \
+    TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,     \
+        YALIGN, benchmark_width_, DIFF, _Opt, +, 0, FMT_C, BPP_C)
+
+TESTPLANARTOBD(I420, 2, 2, RGB565, 2, 2, 1, 9, ARGB, 4)
+
 }  // namespace libyuv