ref: 850ebe40729a96d4d6af7c14970e3e12ae385425
parent: bde1bd997e6dfb99ac0dcf2bd69a6700532d1bdc
parent: bb2041d8de5e1b23e86bbe5e94484f03239e3123
author: Jim Bankoski <jimbankoski@google.com>
date: Mon May 19 09:39:17 EDT 2014
Merge "Fix valgrind read out of bounds error."
--- a/vp8/common/x86/variance_impl_mmx.asm
+++ b/vp8/common/x86/variance_impl_mmx.asm
@@ -342,8 +342,8 @@
movsxd rdx, dword ptr arg(3) ;[recon_stride]
; Row 1
- movq mm0, [rax] ; Copy eight bytes to mm0
- movq mm1, [rbx] ; Copy eight bytes to mm1
+ movd mm0, [rax] ; Copy four bytes to mm0
+ movd mm1, [rbx] ; Copy four bytes to mm1
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
@@ -351,12 +351,12 @@
pmaddwd mm0, mm0 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
- movq mm1, [rbx] ; Copy eight bytes to mm1
+ movd mm1, [rbx] ; Copy four bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
; Row 2
- movq mm0, [rax] ; Copy eight bytes to mm0
+ movd mm0, [rax] ; Copy four bytes to mm0
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
@@ -365,12 +365,12 @@
pmaddwd mm0, mm0 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
- movq mm1, [rbx] ; Copy eight bytes to mm1
+ movd mm1, [rbx] ; Copy four bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
; Row 3
- movq mm0, [rax] ; Copy eight bytes to mm0
- punpcklbw mm0, mm6 ; unpack to higher prrcision
+ movd mm0, [rax] ; Copy four bytes to mm0
+ punpcklbw mm0, mm6 ; unpack to higher precision
punpcklbw mm1, mm6
psubsw mm0, mm1 ; A-B (low order) to MM0
paddw mm5, mm0 ; accumulate differences in mm5
@@ -378,11 +378,11 @@
pmaddwd mm0, mm0 ; square and accumulate
add rbx,rdx ; Inc pointer into ref data
add rax,rcx ; Inc pointer into the new data
- movq mm1, [rbx] ; Copy eight bytes to mm1
+ movd mm1, [rbx] ; Copy four bytes to mm1
paddd mm7, mm0 ; accumulate in mm7
; Row 4
- movq mm0, [rax] ; Copy eight bytes to mm0
+ movd mm0, [rax] ; Copy four bytes to mm0
punpcklbw mm0, mm6 ; unpack to higher prrcision
punpcklbw mm1, mm6