diff options
author | James Almer <jamrial@gmail.com> | 2014-12-20 03:18:57 -0300 |
---|---|---|
committer | Michael Niedermayer <michaelni@gmx.at> | 2014-12-20 13:25:43 +0100 |
commit | 7696e429c7608c2959570a7bc448a76bbfa70dd4 (patch) | |
tree | d916c0d2da832e64851f2b1475fd31e7c94a9648 /libavcodec/x86/vp3dsp.asm | |
parent | 4ae87554f3c8bc54db572873f5049427a7e6cb31 (diff) | |
download | ffmpeg-7696e429c7608c2959570a7bc448a76bbfa70dd4.tar.gz |
x86/vp3dsp: port put_vp_no_rnd_pixels8_l2_mmx to yasm
Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Diffstat (limited to 'libavcodec/x86/vp3dsp.asm')
-rw-r--r-- | libavcodec/x86/vp3dsp.asm | 44 |
1 files changed, 44 insertions, 0 deletions
diff --git a/libavcodec/x86/vp3dsp.asm b/libavcodec/x86/vp3dsp.asm index 24496ae8cf..7ad4c3179b 100644 --- a/libavcodec/x86/vp3dsp.asm +++ b/libavcodec/x86/vp3dsp.asm @@ -36,6 +36,7 @@ vp3_idct_data: times 8 dw 64277 pb_7: times 8 db 0x07 pb_1F: times 8 db 0x1f pb_81: times 8 db 0x81 +pb_FE: times 8 db 0xFE cextern pb_1 cextern pb_3 @@ -147,6 +148,49 @@ cglobal vp3_h_loop_filter, 3, 4 STORE_4_WORDS m3 RET +%macro PAVGB_NO_RND 0 + mova m4, m0 + mova m5, m2 + pand m4, m1 + pand m5, m3 + pxor m1, m0 + pxor m3, m2 + pand m1, m6 + pand m3, m6 + psrlq m1, 1 + psrlq m3, 1 + paddb m4, m1 + paddb m5, m3 +%endmacro + +INIT_MMX mmx +cglobal put_vp_no_rnd_pixels8_l2, 5, 6, 0, dst, src1, src2, stride, h, stride3 + mova m6, [pb_FE] + lea stride3q,[strideq+strideq*2] +.loop + mova m0, [src1q] + mova m1, [src2q] + mova m2, [src1q+strideq] + mova m3, [src2q+strideq] + PAVGB_NO_RND + mova [dstq], m4 + mova [dstq+strideq], m5 + + mova m0, [src1q+strideq*2] + mova m1, [src2q+strideq*2] + mova m2, [src1q+stride3q] + mova m3, [src2q+stride3q] + PAVGB_NO_RND + mova [dstq+strideq*2], m4 + mova [dstq+stride3q], m5 + + lea src1q, [src1q+strideq*4] + lea src2q, [src2q+strideq*4] + lea dstq, [dstq+strideq*4] + sub hd, 4 + jnz .loop + RET + ; from original comments: The Macro does IDct on 4 1-D Dcts %macro BeginIDCT 0 movq m2, I(3) |