Use ScalePlaneDown2_16To8 for avoiding the 2 step process

Bug: libyuv:950
Change-Id: I5a77bca9a0230fe00abd810939e217833a14683f
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/4134524
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>

1 files changed, 81 insertions, 0 deletions

diff --git a/source/scale_common.cc b/source/scale_common.cc
index b02bdafd..398f3f65 100644
--- a/source/scale_common.cc
+++ b/source/scale_common.cc
@@ -23,6 +23,25 @@ namespace libyuv {
 extern "C" {
 #endif
 
+#ifdef __cplusplus
+#define STATIC_CAST(type, expr) static_cast<type>(expr)
+#else
+#define STATIC_CAST(type, expr) (type)(expr)
+#endif
+
+// TODO(fbarchard): make clamp255 preserve negative values.
+static __inline int32_t clamp255(int32_t v) {
+  return (-(v >= 255) | v) & 255;
+}
+
+// Use scale to convert lsb formats to msb, depending how many bits there are:
+// 32768 = 9 bits
+// 16384 = 10 bits
+// 4096 = 12 bits
+// 256 = 16 bits
+// TODO(fbarchard): change scale to bits
+#define C16TO8(v, scale) clamp255(((v) * (scale)) >> 16)
+
 static __inline int Abs(int v) {
   return v >= 0 ? v : -v;
 }
@@ -62,6 +81,26 @@ void ScaleRowDown2_16_C(const uint16_t* src_ptr,
   }
 }
 
+void ScaleRowDown2_16To8_C(const uint16_t* src_ptr,
+                           ptrdiff_t src_stride,
+                           uint8_t* dst,
+                           int dst_width,
+                           int scale) {
+  int x;
+  (void)src_stride;
+  assert(scale >= 256);
+  assert(scale <= 32768);
+  for (x = 0; x < dst_width - 1; x += 2) {
+    dst[0] = STATIC_CAST(uint8_t, C16TO8(src_ptr[1], scale));
+    dst[1] = STATIC_CAST(uint8_t, C16TO8(src_ptr[3], scale));
+    dst += 2;
+    src_ptr += 4;
+  }
+  if (dst_width & 1) {
+    dst[0] = STATIC_CAST(uint8_t, C16TO8(src_ptr[1], scale));
+  }
+}
+
 void ScaleRowDown2Linear_C(const uint8_t* src_ptr,
                            ptrdiff_t src_stride,
                            uint8_t* dst,
@@ -98,6 +137,25 @@ void ScaleRowDown2Linear_16_C(const uint16_t* src_ptr,
   }
 }
 
+void ScaleRowDown2Linear_16To8_C(const uint16_t* src_ptr,
+                                 ptrdiff_t src_stride,
+                                 uint8_t* dst,
+                                 int dst_width,
+                                 int scale) {
+  const uint16_t* s = src_ptr;
+  int x;
+  (void)src_stride;
+  for (x = 0; x < dst_width - 1; x += 2) {
+    dst[0] = STATIC_CAST(uint8_t, C16TO8((s[0] + s[1] + 1) >> 1, scale));
+    dst[1] = STATIC_CAST(uint8_t, C16TO8((s[2] + s[3] + 1) >> 1, scale));
+    dst += 2;
+    s += 4;
+  }
+  if (dst_width & 1) {
+    dst[0] = STATIC_CAST(uint8_t, C16TO8((s[0] + s[1] + 1) >> 1, scale));
+  }
+}
+
 void ScaleRowDown2Box_C(const uint8_t* src_ptr,
                         ptrdiff_t src_stride,
                         uint8_t* dst,
@@ -160,6 +218,29 @@ void ScaleRowDown2Box_16_C(const uint16_t* src_ptr,
   }
 }
 
+void ScaleRowDown2Box_16To8_C(const uint16_t* src_ptr,
+                              ptrdiff_t src_stride,
+                              uint8_t* dst,
+                              int dst_width,
+                              int scale) {
+  const uint16_t* s = src_ptr;
+  const uint16_t* t = src_ptr + src_stride;
+  int x;
+  for (x = 0; x < dst_width - 1; x += 2) {
+    dst[0] = STATIC_CAST(uint8_t,
+                         C16TO8((s[0] + s[1] + t[0] + t[1] + 2) >> 2, scale));
+    dst[1] = STATIC_CAST(uint8_t,
+                         C16TO8((s[2] + s[3] + t[2] + t[3] + 2) >> 2, scale));
+    dst += 2;
+    s += 4;
+    t += 4;
+  }
+  if (dst_width & 1) {
+    dst[0] = STATIC_CAST(uint8_t,
+                         C16TO8((s[0] + s[1] + t[0] + t[1] + 2) >> 2, scale));
+  }
+}
+
 void ScaleRowDown4_C(const uint8_t* src_ptr,
                      ptrdiff_t src_stride,
                      uint8_t* dst,