Rotate by 180 in place.

BUG=18
TEST=none
Review URL: https://webrtc-codereview.appspot.com/456010

git-svn-id: http://libyuv.googlecode.com/svn/trunk@227 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 226
+Version: 227
 License: BSD
 License File: LICENSE
 

include/libyuv/version.h

@@ -11,7 +11,7 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define INCLUDE_LIBYUV_VERSION 226
+#define INCLUDE_LIBYUV_VERSION 227
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_
 

source/convert.cc

@@ -1579,10 +1579,14 @@ int ConvertToI420(const uint8* sample, size_t sample_size,
   }
   int r = 0;
 
-  // For formats that need 2 pass rotation, convert into buffer first
-  bool need_rot = rotation && format != FOURCC_NV12 &&
+  // One pass rotation is available for some formats.  For the rest, convert
+  // to I420 (with optional vertical flipping) into a temporary I420 buffer,
+  // and then rotate the I420 to the final destination buffer.
+  // For in-place conversion, if destination y is same as source sample,
+  // also enable temporary buffer.
+  bool need_buf = (rotation && format != FOURCC_NV12 &&
       format != FOURCC_NV21 && format != FOURCC_I420 &&
-      format != FOURCC_YV12;
+      format != FOURCC_YV12) || y == sample;
   uint8* tmp_y = y;
   uint8* tmp_u = u;
   uint8* tmp_v = v;
@@ -1591,7 +1595,7 @@ int ConvertToI420(const uint8* sample, size_t sample_size,
   int tmp_v_stride = v_stride;
   uint8* buf = NULL;
   int abs_dst_height = (dst_height < 0) ? -dst_height : dst_height;
-  if (need_rot) {
+  if (need_buf) {
     int y_size = dst_width * abs_dst_height;
     int uv_size = ((dst_width + 1) / 2) * ((abs_dst_height + 1) / 2);
     buf = new uint8[y_size + uv_size * 2];
@@ -1890,7 +1894,7 @@ int ConvertToI420(const uint8* sample, size_t sample_size,
       r = -1;  // unknown fourcc - return failure code.
   }
 
-  if (need_rot) {
+  if (need_buf) {
     if (!r) {
       r = I420Rotate(y, y_stride,
                      u, u_stride,

source/rotate.cc

@@ -874,14 +874,40 @@ void RotatePlane180(const uint8* src, int src_stride,
   {
     MirrorRow = MirrorRow_C;
   }
-
-  // Rotate by 180 is a mirror and vertical flip
-  src += src_stride * (height - 1);
-
-  for (int y = 0; y < height; ++y) {
-    MirrorRow(src, dst, width);
-    src -= src_stride;
+  void (*CopyRow)(const uint8* src, uint8* dst, int width) = CopyRow_C;
+#if defined(HAS_COPYROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 64)) {
+    CopyRow = CopyRow_NEON;
+  }
+#elif defined(HAS_COPYROW_X86)
+  if (IS_ALIGNED(width, 4)) {
+    CopyRow = CopyRow_X86;
+#if defined(HAS_COPYROW_SSE2)
+    if (TestCpuFlag(kCpuHasSSE2) &&
+        IS_ALIGNED(width, 32) &&
+        IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
+      CopyRow = CopyRow_SSE2;
+    }
+#endif
+  }
+#endif
+  if (width > kMaxStride) {
+    return;
+  }
+  // Swap first and last row and mirror the content.  Uses a temporary row.
+  SIMD_ALIGNED(uint8 row[kMaxStride]);
+  const uint8* src_bot = src + src_stride * (height - 1);
+  uint8* dst_bot = dst + dst_stride * (height - 1);
+  int half_height = (height + 1) >> 1;
+  // Odd height will harmlessly mirror the middle row twice.
+  for (int y = 0; y < half_height; ++y) {
+    MirrorRow(src, row, width);  // Mirror first row into a buffer
+    src += src_stride;
+    MirrorRow(src_bot, dst, width);  // Mirror last row into first row
     dst += dst_stride;
+    CopyRow(row, dst_bot, width);  // Copy first mirrored row into last
+    src_bot -= src_stride;
+    dst_bot -= dst_stride;
   }
 }