From 1428b37c2b747dc668f6fb08b844b7e9b750dd0e Mon Sep 17 00:00:00 2001
From: "fbarchard@google.com"
 <fbarchard@google.com@16f28f9a-4ce2-e073-06de-1de4eb20be90>
Date: Thu, 7 Nov 2013 19:26:31 +0000
Subject: [PATCH] Scale up by 2 unfiltered BUG=none TESTED=manual test with
 width/height set to half. R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/3569004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@845 16f28f9a-4ce2-e073-06de-1de4eb20be90
---
 README.chromium          |  2 +-
 include/libyuv/version.h |  2 +-
 source/scale_argb.cc     | 36 ++++++++++++++++++++++++++++++++++++
 3 files changed, 38 insertions(+), 2 deletions(-)

diff --git a/README.chromium b/README.chromium
index b545c337e..06c1c70d1 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 844
+Version: 845
 License: BSD
 License File: LICENSE
 
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index f166649d0..436574664 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 844
+#define LIBYUV_VERSION 845
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT
diff --git a/source/scale_argb.cc b/source/scale_argb.cc
index 83351b6b1..f224761fb 100644
--- a/source/scale_argb.cc
+++ b/source/scale_argb.cc
@@ -369,6 +369,34 @@ static void ScaleARGBFilterCols_SSSE3(uint8* dst_argb, const uint8* src_argb,
   }
 }
 
+// Reads 4 pixels, duplicates them and writes 8 pixels.
+// Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned.
+#define HAS_SCALEARGBCOLSUP2_SSE2
+__declspec(naked) __declspec(align(16))
+void ScaleARGBColsUp2_SSE2(uint8* dst_argb, const uint8* src_argb,
+                           int dst_width, int /* x */, int /* dx */) {
+  __asm {
+    mov        edx, [esp + 4]    // dst_argb
+    mov        eax, [esp + 8]    // src_argb
+    mov        ecx, [esp + 12]   // dst_width
+
+    align      16
+  wloop:
+    movdqa     xmm0, [eax]
+    lea        eax,  [eax + 16]
+    movdqa     xmm1, xmm0
+    punpckldq  xmm0, xmm0
+    punpckhdq  xmm1, xmm1
+    sub        ecx, 8
+    movdqa     [edx], xmm0
+    movdqa     [edx + 16], xmm1
+    lea        edx, [edx + 32]
+    jg         wloop
+
+    ret
+  }
+}
+
 #elif !defined(LIBYUV_DISABLE_X86) && \
     (defined(__x86_64__) || defined(__i386__))
 
@@ -1335,6 +1363,14 @@ static void ScaleARGBSimple(int src_width, int src_height,
 #if defined(HAS_SCALEARGBCOLS_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     ScaleARGBCols = ScaleARGBCols_SSE2;
+#if defined(HAS_SCALEARGBCOLS_SSE2)
+    if (src_width * 2 == dst_width && IS_ALIGNED(dst_width, 8) &&
+        (x >> 16) == 0 &&
+        IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride, 16) &&
+        IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) {
+      ScaleARGBCols = ScaleARGBColsUp2_SSE2;
+    }
+#endif
   }
 #endif