generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge "Add zero offset checking in SSE2 sub-pixel filter function"

Browse Source
This commit is contained in:
Yunqing Wang 2011-03-08 12:26:58 -08:00 committed by Code Review
commit 859abd6b5d
1 changed files with 33 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
vp8/encoder/x86/variance_impl_sse2.asm
View File

@ -627,6 +627,10 @@ filter_block2d_bil_var_sse2_loop:
filter_block2d_bil_var_sse2_sp_only:
movsxd rdx, dword ptr arg(6) ; yoffset
cmp rdx, 0 ; skip all if both xoffset=0 and yoffset=0
je filter_block2d_bil_var_sse2_full_pixel
shl rdx, 5
lea rdx, [rdx + rcx] ; VFilter
@ -671,6 +675,35 @@ filter_block2d_bil_sp_only_loop:
jmp filter_block2d_bil_variance
filter_block2d_bil_var_sse2_full_pixel:
mov rsi, arg(0) ;ref_ptr
mov rdi, arg(2) ;src_ptr
movsxd rcx, dword ptr arg(4) ;Height
movsxd rax, dword ptr arg(1) ;ref_pixels_per_line
movsxd rbx, dword ptr arg(3) ;src_pixels_per_line
pxor xmm0, xmm0 ;
filter_block2d_bil_full_pixel_loop:
movq xmm1, QWORD PTR [rsi] ;
punpcklbw xmm1, xmm0 ;
movq xmm2, QWORD PTR [rdi] ;
punpcklbw xmm2, xmm0 ;
psubw xmm1, xmm2 ;
paddw xmm6, xmm1 ;
pmaddwd xmm1, xmm1 ;
paddd xmm7, xmm1 ;
lea rsi, [rsi + rax] ;ref_pixels_per_line
lea rdi, [rdi + rbx] ;src_pixels_per_line
sub rcx, 1 ;
jnz filter_block2d_bil_full_pixel_loop ;
jmp filter_block2d_bil_variance
filter_block2d_bil_var_sse2_fp_only:
mov rsi, arg(0) ;ref_ptr
mov rdi, arg(2) ;src_ptr