[x265] [PATCH] asm: improvement on chroma_hpp{2, 4} by reduce memory operator

[Min Chen]

# HG changeset patch
# User Min Chen <chenm003 at 163.com>
# Date 1383051361 -28800
# Node ID 7ee3e65d3d1b9f765741b519ea77111ce83a04e6
# Parent  7416f8a54b710f0e0eb482eb6d07109ba8574e91
asm: improvement on chroma_hpp{2,4} by reduce memory operator

diff -r 7416f8a54b71 -r 7ee3e65d3d1b source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm	Tue Oct 29 16:35:24 2013 +0800
+++ b/source/common/x86/ipfilter8.asm	Tue Oct 29 20:56:01 2013 +0800
@@ -76,17 +76,19 @@
 SECTION .text
 
 %macro FILTER_H4_w2_2 3
-    movu        %2, [srcq - 1]
+    movh        %2, [srcq - 1]
     pshufb      %2, %2, Tm0
+    movh        %1, [srcq + srcstrideq - 1]
+    pshufb      %1, %1, Tm0
+    punpcklqdq  %2, %1
     pmaddubsw   %2, coef2
-    movu        %1, [srcq + srcstrideq - 1]
-    pshufb      %1, %1, Tm0
-    pmaddubsw   %1, coef2
-    phaddw      %2, %1
+    phaddw      %2, %2
     pmulhrsw    %2, %3
     packuswb    %2, %2
-    pextrw      [dstq], %2, 0
-    pextrw      [dstq + dststrideq], %2, 2
+    movd        r4, %2
+    mov         [dstq], r4w
+    shr         r4, 16
+    mov         [dstq + dststrideq], r4w
 %endmacro
 
 ;-----------------------------------------------------------------------------
@@ -158,17 +160,18 @@
 RET
 
 %macro FILTER_H4_w4_2 3
-    movu        %2, [srcq - 1]
+    movh        %2, [srcq - 1]
     pshufb      %2, %2, Tm0
     pmaddubsw   %2, coef2
-    movu        %1, [srcq + srcstrideq - 1]
+    movh        %1, [srcq + srcstrideq - 1]
     pshufb      %1, %1, Tm0
     pmaddubsw   %1, coef2
     phaddw      %2, %1
     pmulhrsw    %2, %3
     packuswb    %2, %2
-    movd        [dstq],      %2
-    pextrd      [dstq + dststrideq], %2,  1
+    movd        [dstq], %2
+    palignr     %2, %2, 4
+    movd        [dstq + dststrideq], %2
 %endmacro
 
 ;-----------------------------------------------------------------------------