From 90a36b29d3ffac9ecaa239191c7106c9993dc093 Mon Sep 17 00:00:00 2001
From: "fbarchard@google.com"
 <fbarchard@google.com@16f28f9a-4ce2-e073-06de-1de4eb20be90>
Date: Fri, 3 Jan 2014 00:34:55 +0000
Subject: [PATCH] Use 64 bit fixed point for scaling columns if source is 32k
 or wider. BUG=302 TESTED=out\release\libyuv_unittest --gtest_filter=*I*ToI*
 R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/6509004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@942 16f28f9a-4ce2-e073-06de-1de4eb20be90
---
 README.chromium            |  2 +-
 include/libyuv/scale_row.h |  2 ++
 include/libyuv/version.h   |  2 +-
 source/scale.cc            | 11 ++++++++---
 source/scale_common.cc     | 24 ++++++++++++++++++++++++
 5 files changed, 36 insertions(+), 5 deletions(-)

diff --git a/README.chromium b/README.chromium
index d677519fb..499366e12 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 941
+Version: 942
 License: BSD
 License File: LICENSE
 
diff --git a/include/libyuv/scale_row.h b/include/libyuv/scale_row.h
index fddb15215..e1b0333cd 100644
--- a/include/libyuv/scale_row.h
+++ b/include/libyuv/scale_row.h
@@ -112,6 +112,8 @@ void ScaleColsUp2_C(uint8* dst_ptr, const uint8* src_ptr,
                     int dst_width, int, int);
 void ScaleFilterCols_C(uint8* dst_ptr, const uint8* src_ptr,
                        int dst_width, int x, int dx);
+void ScaleFilterCols64_C(uint8* dst_ptr, const uint8* src_ptr,
+                         int dst_width, int x, int dx);
 void ScaleRowDown38_C(const uint8* src_ptr, ptrdiff_t /* src_stride */,
                       uint8* dst, int dst_width);
 void ScaleRowDown38_3_Box_C(const uint8* src_ptr,
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index 7a2eabfd9..4e6894ed4 100644
--- a/include/libyuv/version.h
+++ b/include/libyuv/version.h
@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 941
+#define LIBYUV_VERSION 942
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT
diff --git a/source/scale.cc b/source/scale.cc
index 5bc6c9b7c..1f135aee3 100644
--- a/source/scale.cc
+++ b/source/scale.cc
@@ -511,9 +511,11 @@ void ScalePlaneBilinearDown(int src_width, int src_height,
 #endif
 
   void (*ScaleFilterCols)(uint8* dst_ptr, const uint8* src_ptr,
-                          int dst_width, int x, int dx) = ScaleFilterCols_C;
+      int dst_width, int x, int dx) =
+      (src_width >= 32768) ? ScaleFilterCols64_C : ScaleFilterCols_C;
+
 #if defined(HAS_SCALEFILTERCOLS_SSSE3)
-  if (TestCpuFlag(kCpuHasSSSE3)) {
+  if (TestCpuFlag(kCpuHasSSSE3) && src_width < 32768) {
     ScaleFilterCols = ScaleFilterCols_SSSE3;
   }
 #endif
@@ -614,8 +616,11 @@ void ScalePlaneBilinearUp(int src_width, int src_height,
   void (*ScaleFilterCols)(uint8* dst_ptr, const uint8* src_ptr,
        int dst_width, int x, int dx) =
        filtering ? ScaleFilterCols_C : ScaleCols_C;
+  if (filtering && src_width >= 32768) {
+    ScaleFilterCols = ScaleFilterCols64_C;
+  }
 #if defined(HAS_SCALEFILTERCOLS_SSSE3)
-  if (filtering && TestCpuFlag(kCpuHasSSSE3)) {
+  if (filtering && TestCpuFlag(kCpuHasSSSE3) && src_width < 32768) {
     ScaleFilterCols = ScaleFilterCols_SSSE3;
   }
 #endif
diff --git a/source/scale_common.cc b/source/scale_common.cc
index 5d0fcb8b9..6cc5ff8cb 100644
--- a/source/scale_common.cc
+++ b/source/scale_common.cc
@@ -229,6 +229,30 @@ void ScaleFilterCols_C(uint8* dst_ptr, const uint8* src_ptr,
     dst_ptr[0] = BLENDER(a, b, x & 0xffff);
   }
 }
+
+void ScaleFilterCols64_C(uint8* dst_ptr, const uint8* src_ptr,
+                         int dst_width, int x32, int dx) {
+  int64 x = static_cast<int64>(x32);
+  for (int j = 0; j < dst_width - 1; j += 2) {
+    int64 xi = x >> 16;
+    int a = src_ptr[xi];
+    int b = src_ptr[xi + 1];
+    dst_ptr[0] = BLENDER(a, b, x & 0xffff);
+    x += dx;
+    xi = x >> 16;
+    a = src_ptr[xi];
+    b = src_ptr[xi + 1];
+    dst_ptr[1] = BLENDER(a, b, x & 0xffff);
+    x += dx;
+    dst_ptr += 2;
+  }
+  if (dst_width & 1) {
+    int64 xi = x >> 16;
+    int a = src_ptr[xi];
+    int b = src_ptr[xi + 1];
+    dst_ptr[0] = BLENDER(a, b, x & 0xffff);
+  }
+}
 #undef BLENDER
 
 void ScaleRowDown38_C(const uint8* src_ptr, ptrdiff_t /* src_stride */,