[x265] [PATCH] Enable the new dct4 asm routine

[nabajit at multicorewareinc.com]

# HG changeset patch
# User Nabajit Deka
# Date 1385475027 -19800
#      Tue Nov 26 19:40:27 2013 +0530
# Node ID 4f0a40d94cc6a9fa8940dfb3f18149771b4db70b
# Parent  713e6b21099e37136a1778b8c24e251951f46fd2
Enable the new dct4 asm routine.

diff -r 713e6b21099e -r 4f0a40d94cc6 source/common/vec/dct-sse3.cpp
--- a/source/common/vec/dct-sse3.cpp	Tue Nov 26 19:37:16 2013 +0530
+++ b/source/common/vec/dct-sse3.cpp	Tue Nov 26 19:40:27 2013 +0530
@@ -1737,7 +1737,7 @@
 void Setup_Vec_DCTPrimitives_sse3(EncoderPrimitives &p)
 {
 #if !HIGH_BIT_DEPTH
-    p.dct[DCT_4x4] = dct4;
+    //p.dct[DCT_4x4] = dct4;
     p.idct[IDCT_4x4] = idct4;
     p.idct[IDCT_8x8] = idct8;
     p.idct[IDCT_16x16] = idct16;
diff -r 713e6b21099e -r 4f0a40d94cc6 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Tue Nov 26 19:37:16 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp	Tue Nov 26 19:40:27 2013 +0530
@@ -33,6 +33,7 @@
 #include "ipfilter8.h"
 #include "blockcopy8.h"
 #include "intrapred.h"
+#include "dct8.h"
 }
 
 #define INIT2_NAME(name1, name2, cpu) \
@@ -568,6 +569,7 @@
         p.transpose[BLOCK_16x16] = x265_transpose16_sse2;
         p.transpose[BLOCK_32x32] = x265_transpose32_sse2;
         p.transpose[BLOCK_64x64] = x265_transpose64_sse2;
+        p.dct[DCT_4x4] = x265_dct4_sse2;
     }
     if (cpuMask & X265_CPU_SSSE3)
     {