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

fix and enable vp9_dc_128_predictor_16x16

Browse Source 
widen the loads and stores to 128-bit.

this was added, but not enabled in:
493a857 Add some sse2 code for intra prediction.

Change-Id: I277d7db608a7db7d75cc0bde86f48fa66ad487e4
This commit is contained in:
James Zern
2015-05-05 11:40:13 -07:00
parent e47811ef8f
commit ccae5d99d2
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vp9/common/vp9_rtcd_defs.pl
View File

@@ -171,7 +171,7 @@ add_proto qw/void vp9_dc_left_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_strid
specialize qw/vp9_dc_left_predictor_16x16/; specialize qw/vp9_dc_left_predictor_16x16/;
add_proto qw/void vp9_dc_128_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; add_proto qw/void vp9_dc_128_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_dc_128_predictor_16x16/; specialize qw/vp9_dc_128_predictor_16x16/, "$sse2_x86inc";
add_proto qw/void vp9_d207_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; add_proto qw/void vp9_d207_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vp9_d207_predictor_32x32/, "$ssse3_x86inc"; specialize qw/vp9_d207_predictor_32x32/, "$ssse3_x86inc";

4
vp9/common/x86/vp9_intrapred_sse2.asm
View File

@@ -263,14 +263,14 @@ cglobal dc_top_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
REP_RET REP_RET
INIT_MMX sse INIT_XMM sse2
cglobal dc_128_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset cglobal dc_128_predictor_16x16, 4, 5, 3, dst, stride, above, left, goffset
GET_GOT goffsetq GET_GOT goffsetq
DEFINE_ARGS dst, stride, stride3, lines4 DEFINE_ARGS dst, stride, stride3, lines4
lea stride3q, [strideq*3] lea stride3q, [strideq*3]
mov lines4d, 4 mov lines4d, 4
movq m0, [GLOBAL(dc_128)] mova m0, [GLOBAL(dc_128)]
.loop: .loop:
mova [dstq ], m0 mova [dstq ], m0
mova [dstq+strideq ], m0 mova [dstq+strideq ], m0