Add 10 bit rotate methods.

This initial implementation is based on current unoptimized code in webrtc using just plain for loops.

Bug: libyuv:949
Change-Id: Ic87ee49c3a0b62edbaaa4255c263c1f7be4ea02b
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4110782
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>

3 files changed, 380 insertions, 0 deletions

diff --git a/unit_test/rotate_argb_test.cc b/unit_test/rotate_argb_test.cc
index 01ed69ca..74952c4e 100644
--- a/unit_test/rotate_argb_test.cc
+++ b/unit_test/rotate_argb_test.cc
@@ -225,4 +225,110 @@ TEST_F(LibYUVRotateTest, RotatePlane90_TestStride) {
   free_aligned_buffer_page_end(src_argb);
 }
 
+static void TestRotatePlane_16(int src_width,
+                               int src_height,
+                               int dst_width,
+                               int dst_height,
+                               libyuv::RotationMode mode,
+                               int benchmark_iterations,
+                               int disable_cpu_flags,
+                               int benchmark_cpu_info) {
+  if (src_width < 1) {
+    src_width = 1;
+  }
+  if (src_height < 1) {
+    src_height = 1;
+  }
+  if (dst_width < 1) {
+    dst_width = 1;
+  }
+  if (dst_height < 1) {
+    dst_height = 1;
+  }
+  int src_stride = src_width;
+  int src_plane_size = src_stride * abs(src_height);
+  align_buffer_page_end_16(src, src_plane_size);
+  for (int i = 0; i < src_plane_size; ++i) {
+    src[i] = fastrand() & 0xff;
+  }
+
+  int dst_stride = dst_width;
+  int dst_plane_size = dst_stride * dst_height;
+  align_buffer_page_end_16(dst_c, dst_plane_size);
+  align_buffer_page_end_16(dst_opt, dst_plane_size);
+  memset(dst_c, 2, dst_plane_size);
+  memset(dst_opt, 3, dst_plane_size);
+
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  RotatePlane_16(src, src_stride, dst_c, dst_stride, src_width, src_height,
+                 mode);
+
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  for (int i = 0; i < benchmark_iterations; ++i) {
+    RotatePlane_16(src, src_stride, dst_opt, dst_stride, src_width, src_height,
+                   mode);
+  }
+
+  // Rotation should be exact.
+  for (int i = 0; i < dst_plane_size; ++i) {
+    EXPECT_EQ(dst_c[i], dst_opt[i]);
+  }
+
+  free_aligned_buffer_page_end_16(dst_c);
+  free_aligned_buffer_page_end_16(dst_opt);
+  free_aligned_buffer_page_end_16(src);
+}
+
+TEST_F(LibYUVRotateTest, RotatePlane0_16_Opt) {
+  TestRotatePlane_16(benchmark_width_, benchmark_height_, benchmark_width_,
+                     benchmark_height_, kRotate0, benchmark_iterations_,
+                     disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, RotatePlane90_16_Opt) {
+  TestRotatePlane_16(benchmark_width_, benchmark_height_, benchmark_height_,
+                     benchmark_width_, kRotate90, benchmark_iterations_,
+                     disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, RotatePlane180_16_Opt) {
+  TestRotatePlane_16(benchmark_width_, benchmark_height_, benchmark_width_,
+                     benchmark_height_, kRotate180, benchmark_iterations_,
+                     disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, RotatePlane270_16_Opt) {
+  TestRotatePlane_16(benchmark_width_, benchmark_height_, benchmark_height_,
+                     benchmark_width_, kRotate270, benchmark_iterations_,
+                     disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, RotatePlane0_16_Odd) {
+  TestRotatePlane_16(benchmark_width_ + 1, benchmark_height_ + 1,
+                     benchmark_width_ + 1, benchmark_height_ + 1, kRotate0,
+                     benchmark_iterations_, disable_cpu_flags_,
+                     benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, RotatePlane90_16_Odd) {
+  TestRotatePlane_16(benchmark_width_ + 1, benchmark_height_ + 1,
+                     benchmark_height_ + 1, benchmark_width_ + 1, kRotate90,
+                     benchmark_iterations_, disable_cpu_flags_,
+                     benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, RotatePlane180_16_Odd) {
+  TestRotatePlane_16(benchmark_width_ + 1, benchmark_height_ + 1,
+                     benchmark_width_ + 1, benchmark_height_ + 1, kRotate180,
+                     benchmark_iterations_, disable_cpu_flags_,
+                     benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, RotatePlane270_16_Odd) {
+  TestRotatePlane_16(benchmark_width_ + 1, benchmark_height_ + 1,
+                     benchmark_height_ + 1, benchmark_width_ + 1, kRotate270,
+                     benchmark_iterations_, disable_cpu_flags_,
+                     benchmark_cpu_info_);
+}
+
 }  // namespace libyuv
diff --git a/unit_test/rotate_test.cc b/unit_test/rotate_test.cc
index d3887414..9dec7811 100644
--- a/unit_test/rotate_test.cc
+++ b/unit_test/rotate_test.cc
@@ -596,4 +596,266 @@ TESTAPLANARTOP(Android420, NV21, 2, 1, 0, 2, 2, I420, 2, 2)
 #undef TESTAPLANARTOP
 #undef TESTAPLANARTOPI
 
+static void I010TestRotate(int src_width,
+                           int src_height,
+                           int dst_width,
+                           int dst_height,
+                           libyuv::RotationMode mode,
+                           int benchmark_iterations,
+                           int disable_cpu_flags,
+                           int benchmark_cpu_info) {
+  if (src_width < 1) {
+    src_width = 1;
+  }
+  if (src_height == 0) {
+    src_height = 1;
+  }
+  if (dst_width < 1) {
+    dst_width = 1;
+  }
+  if (dst_height < 1) {
+    dst_height = 1;
+  }
+  int src_i010_y_size = src_width * Abs(src_height);
+  int src_i010_uv_size = ((src_width + 1) / 2) * ((Abs(src_height) + 1) / 2);
+  int src_i010_size = src_i010_y_size + src_i010_uv_size * 2;
+  align_buffer_page_end_16(src_i010, src_i010_size);
+  for (int i = 0; i < src_i010_size; ++i) {
+    src_i010[i] = fastrand() & 0x3ff;
+  }
+
+  int dst_i010_y_size = dst_width * dst_height;
+  int dst_i010_uv_size = ((dst_width + 1) / 2) * ((dst_height + 1) / 2);
+  int dst_i010_size = dst_i010_y_size + dst_i010_uv_size * 2;
+  align_buffer_page_end_16(dst_i010_c, dst_i010_size);
+  align_buffer_page_end_16(dst_i010_opt, dst_i010_size);
+  memset(dst_i010_c, 2, dst_i010_size * 2);
+  memset(dst_i010_opt, 3, dst_i010_size * 2);
+
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  I010Rotate(src_i010, src_width, src_i010 + src_i010_y_size,
+             (src_width + 1) / 2, src_i010 + src_i010_y_size + src_i010_uv_size,
+             (src_width + 1) / 2, dst_i010_c, dst_width,
+             dst_i010_c + dst_i010_y_size, (dst_width + 1) / 2,
+             dst_i010_c + dst_i010_y_size + dst_i010_uv_size,
+             (dst_width + 1) / 2, src_width, src_height, mode);
+
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  for (int i = 0; i < benchmark_iterations; ++i) {
+    I010Rotate(
+        src_i010, src_width, src_i010 + src_i010_y_size, (src_width + 1) / 2,
+        src_i010 + src_i010_y_size + src_i010_uv_size, (src_width + 1) / 2,
+        dst_i010_opt, dst_width, dst_i010_opt + dst_i010_y_size,
+        (dst_width + 1) / 2, dst_i010_opt + dst_i010_y_size + dst_i010_uv_size,
+        (dst_width + 1) / 2, src_width, src_height, mode);
+  }
+
+  // Rotation should be exact.
+  for (int i = 0; i < dst_i010_size; ++i) {
+    EXPECT_EQ(dst_i010_c[i], dst_i010_opt[i]);
+  }
+
+  free_aligned_buffer_page_end_16(dst_i010_c);
+  free_aligned_buffer_page_end_16(dst_i010_opt);
+  free_aligned_buffer_page_end_16(src_i010);
+}
+
+TEST_F(LibYUVRotateTest, I010Rotate0_Opt) {
+  I010TestRotate(benchmark_width_, benchmark_height_, benchmark_width_,
+                 benchmark_height_, kRotate0, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I010Rotate90_Opt) {
+  I010TestRotate(benchmark_width_, benchmark_height_, benchmark_height_,
+                 benchmark_width_, kRotate90, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I010Rotate180_Opt) {
+  I010TestRotate(benchmark_width_, benchmark_height_, benchmark_width_,
+                 benchmark_height_, kRotate180, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I010Rotate270_Opt) {
+  I010TestRotate(benchmark_width_, benchmark_height_, benchmark_height_,
+                 benchmark_width_, kRotate270, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+static void I210TestRotate(int src_width,
+                           int src_height,
+                           int dst_width,
+                           int dst_height,
+                           libyuv::RotationMode mode,
+                           int benchmark_iterations,
+                           int disable_cpu_flags,
+                           int benchmark_cpu_info) {
+  if (src_width < 1) {
+    src_width = 1;
+  }
+  if (src_height == 0) {
+    src_height = 1;
+  }
+  if (dst_width < 1) {
+    dst_width = 1;
+  }
+  if (dst_height < 1) {
+    dst_height = 1;
+  }
+  int src_i210_y_size = src_width * Abs(src_height);
+  int src_i210_uv_size = ((src_width + 1) / 2) * Abs(src_height);
+  int src_i210_size = src_i210_y_size + src_i210_uv_size * 2;
+  align_buffer_page_end_16(src_i210, src_i210_size);
+  for (int i = 0; i < src_i210_size; ++i) {
+    src_i210[i] = fastrand() & 0x3ff;
+  }
+
+  int dst_i210_y_size = dst_width * dst_height;
+  int dst_i210_uv_size = ((dst_width + 1) / 2) * dst_height;
+  int dst_i210_size = dst_i210_y_size + dst_i210_uv_size * 2;
+  align_buffer_page_end_16(dst_i210_c, dst_i210_size);
+  align_buffer_page_end_16(dst_i210_opt, dst_i210_size);
+  memset(dst_i210_c, 2, dst_i210_size * 2);
+  memset(dst_i210_opt, 3, dst_i210_size * 2);
+
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  I210Rotate(src_i210, src_width, src_i210 + src_i210_y_size,
+             (src_width + 1) / 2, src_i210 + src_i210_y_size + src_i210_uv_size,
+             (src_width + 1) / 2, dst_i210_c, dst_width,
+             dst_i210_c + dst_i210_y_size, (dst_width + 1) / 2,
+             dst_i210_c + dst_i210_y_size + dst_i210_uv_size,
+             (dst_width + 1) / 2, src_width, src_height, mode);
+
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  for (int i = 0; i < benchmark_iterations; ++i) {
+    I210Rotate(
+        src_i210, src_width, src_i210 + src_i210_y_size, (src_width + 1) / 2,
+        src_i210 + src_i210_y_size + src_i210_uv_size, (src_width + 1) / 2,
+        dst_i210_opt, dst_width, dst_i210_opt + dst_i210_y_size,
+        (dst_width + 1) / 2, dst_i210_opt + dst_i210_y_size + dst_i210_uv_size,
+        (dst_width + 1) / 2, src_width, src_height, mode);
+  }
+
+  // Rotation should be exact.
+  for (int i = 0; i < dst_i210_size; ++i) {
+    EXPECT_EQ(dst_i210_c[i], dst_i210_opt[i]);
+  }
+
+  free_aligned_buffer_page_end_16(dst_i210_c);
+  free_aligned_buffer_page_end_16(dst_i210_opt);
+  free_aligned_buffer_page_end_16(src_i210);
+}
+
+TEST_F(LibYUVRotateTest, I210Rotate0_Opt) {
+  I210TestRotate(benchmark_width_, benchmark_height_, benchmark_width_,
+                 benchmark_height_, kRotate0, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I210Rotate90_Opt) {
+  I210TestRotate(benchmark_width_, benchmark_height_, benchmark_height_,
+                 benchmark_width_, kRotate90, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I210Rotate180_Opt) {
+  I210TestRotate(benchmark_width_, benchmark_height_, benchmark_width_,
+                 benchmark_height_, kRotate180, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I210Rotate270_Opt) {
+  I210TestRotate(benchmark_width_, benchmark_height_, benchmark_height_,
+                 benchmark_width_, kRotate270, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+static void I410TestRotate(int src_width,
+                           int src_height,
+                           int dst_width,
+                           int dst_height,
+                           libyuv::RotationMode mode,
+                           int benchmark_iterations,
+                           int disable_cpu_flags,
+                           int benchmark_cpu_info) {
+  if (src_width < 1) {
+    src_width = 1;
+  }
+  if (src_height == 0) {
+    src_height = 1;
+  }
+  if (dst_width < 1) {
+    dst_width = 1;
+  }
+  if (dst_height < 1) {
+    dst_height = 1;
+  }
+  int src_i410_y_size = src_width * Abs(src_height);
+  int src_i410_uv_size = src_width * Abs(src_height);
+  int src_i410_size = src_i410_y_size + src_i410_uv_size * 2;
+  align_buffer_page_end_16(src_i410, src_i410_size);
+  for (int i = 0; i < src_i410_size; ++i) {
+    src_i410[i] = fastrand() & 0x3ff;
+  }
+
+  int dst_i410_y_size = dst_width * dst_height;
+  int dst_i410_uv_size = dst_width * dst_height;
+  int dst_i410_size = dst_i410_y_size + dst_i410_uv_size * 2;
+  align_buffer_page_end_16(dst_i410_c, dst_i410_size);
+  align_buffer_page_end_16(dst_i410_opt, dst_i410_size);
+  memset(dst_i410_c, 2, dst_i410_size * 2);
+  memset(dst_i410_opt, 3, dst_i410_size * 2);
+
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  I410Rotate(src_i410, src_width, src_i410 + src_i410_y_size, src_width,
+             src_i410 + src_i410_y_size + src_i410_uv_size, src_width,
+             dst_i410_c, dst_width, dst_i410_c + dst_i410_y_size, dst_width,
+             dst_i410_c + dst_i410_y_size + dst_i410_uv_size, dst_width,
+             src_width, src_height, mode);
+
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  for (int i = 0; i < benchmark_iterations; ++i) {
+    I410Rotate(src_i410, src_width, src_i410 + src_i410_y_size, src_width,
+               src_i410 + src_i410_y_size + src_i410_uv_size, src_width,
+               dst_i410_opt, dst_width, dst_i410_opt + dst_i410_y_size,
+               dst_width, dst_i410_opt + dst_i410_y_size + dst_i410_uv_size,
+               dst_width, src_width, src_height, mode);
+  }
+
+  // Rotation should be exact.
+  for (int i = 0; i < dst_i410_size; ++i) {
+    EXPECT_EQ(dst_i410_c[i], dst_i410_opt[i]);
+  }
+
+  free_aligned_buffer_page_end_16(dst_i410_c);
+  free_aligned_buffer_page_end_16(dst_i410_opt);
+  free_aligned_buffer_page_end_16(src_i410);
+}
+
+TEST_F(LibYUVRotateTest, I410Rotate0_Opt) {
+  I410TestRotate(benchmark_width_, benchmark_height_, benchmark_width_,
+                 benchmark_height_, kRotate0, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I410Rotate90_Opt) {
+  I410TestRotate(benchmark_width_, benchmark_height_, benchmark_height_,
+                 benchmark_width_, kRotate90, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I410Rotate180_Opt) {
+  I410TestRotate(benchmark_width_, benchmark_height_, benchmark_width_,
+                 benchmark_height_, kRotate180, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I410Rotate270_Opt) {
+  I410TestRotate(benchmark_width_, benchmark_height_, benchmark_height_,
+                 benchmark_width_, kRotate270, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
 }  // namespace libyuv
diff --git a/unit_test/unit_test.h b/unit_test/unit_test.h
index 6cc99a65..42e40d12 100644
--- a/unit_test/unit_test.h
+++ b/unit_test/unit_test.h
@@ -78,6 +78,18 @@ static inline bool SizeValid(int src_width,
   free(var##_mem);                        \
   var = 0
 
+#define align_buffer_page_end_16(var, size)                                 \
+  uint8_t* var##_mem =                                                      \
+      reinterpret_cast<uint8_t*>(malloc(((size * 2) + 4095 + 63) & ~4095)); \
+  uint16_t* var = reinterpret_cast<uint16_t*>(                              \
+      (intptr_t)(var##_mem + (((size * 2) + 4095 + 63) & ~4095) -           \
+                 (size * 2)) &                                              \
+      ~63)
+
+#define free_aligned_buffer_page_end_16(var) \
+  free(var##_mem);                           \
+  var = 0
+
 #ifdef WIN32
 static inline double get_time() {
   LARGE_INTEGER t, f;