[x265] [PATCH] intra-sse3.cpp: Replace PredIntraAng4_2 vector class function with intrinsic

[dnyaneshwar at multicorewareinc.com]

# HG changeset patch
# User Dnyaneshwar Gorade <dnyaneshwar at multicorewareinc.com>
# Date 1381481671 -19800
#      Fri Oct 11 14:24:31 2013 +0530
# Node ID 9415a602a6c454a05f0866550c659a507567cdd7
# Parent  7ecc6ca251677985398fdf00738c139278359293
intra-sse3.cpp: Replace PredIntraAng4_2 vector class function with intrinsic.

diff -r 7ecc6ca25167 -r 9415a602a6c4 source/common/vec/intra-sse3.cpp
--- a/source/common/vec/intra-sse3.cpp	Fri Oct 11 14:22:21 2013 +0530
+++ b/source/common/vec/intra-sse3.cpp	Fri Oct 11 14:24:31 2013 +0530
@@ -1656,25 +1656,27 @@
 
 void PredIntraAng4_2(pixel* dst, int dstStride, pixel *refMain, int dirMode)
 {
-    PRED_INTRA_ANG4_START
-
-        tmp16_1 = (Vec16uc)load_partial(const_int(8), refMain + 1);
-
-    row11 = extend_low(tmp16_1);    //offsets(0,1,2,3)
-    tmp2uq = reinterpret_i(tmp16_1);
-    tmp2uq = tmp2uq >> 8;
-    tmp16_2 = reinterpret_i(tmp2uq);
-    row12 = extend_low(tmp16_2);    //offsets(1,2,3,4)
-    row21 = row11;                  //offsets(0,1,2,3)
+    PRED_INTRA_ANGLE_4_START
+
+    tmp16_1 = _mm_loadl_epi64((__m128i*)(refMain + 1));
+    row11   =_mm_unpacklo_epi8(tmp16_1, _mm_setzero_si128());
+
+    tmp2 = tmp16_1;
+    tmp2 = _mm_srl_epi64(tmp2,_mm_cvtsi32_si128(8));
+
+    tmp16_2 = tmp2;
+    row12 = _mm_unpacklo_epi8(tmp16_2, _mm_setzero_si128());
+
+    row21 = row11;
     row22 = row12;
     row31 = row11;
     row32 = row12;
     row41 = row11;
     row42 = row12;
 
-    v_deltaPos = v_ipAngle = 2;
-
-    PRED_INTRA_ANG4_END
+    deltaPos = ipAngle = _mm_set1_epi16(2);
+
+    PRED_INTRA_ANGLE_4_END
 }
 
 void PredIntraAng4_m_2(pixel* dst, int dstStride, pixel *refMain, int dirMode)