Checkasm result (osx) for your last patch : hflip_byte_c: 28.5 hflip_byte_ssse3: 29.0 hflip_short_c: 277.7 hflip_short_ssse3: 65.0
if you add a "cmp xq, wq" after the simd loop you can be faster than c (clang), if width is multiple of mmsize*2 hflip_byte_c: 28.5 hflip_byte_ssse3: 27.5 see below otherwise looks ok (i will send later a much cleaner patch for the checkasm, and a patch to use one macro for both func) + > +pb_flip_byte: db 15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0 > +pb_flip_short: db 14,15,12,13,10,11,8,9,6,7,4,5,2,3,0,1 > + > +SECTION .text > + > +INIT_XMM ssse3 > +cglobal hflip_byte, 3, 6, 3, src, dst, w, x, v, r > + mova m0, [pb_flip_byte] > + mov xq, 0 > + mov wd, dword wm > + mov rq, wq > + and rq, 2 * mmsize - 1 > + cmp wq, 2 * mmsize > + jl .loop1 > + sub wq, rq > + > + .loop0: > + neg xq > + movu m1, [srcq + xq - mmsize + 1] > + movu m2, [srcq + xq - 2 * mmsize + 1] > + pshufb m1, m0 > + pshufb m2, m0 > + neg xq > + movu [dstq + xq ], m1 > + movu [dstq + xq + mmsize], m2 > + add xq, mmsize * 2 > + cmp xq, wq > + jl .loop0 > cmp xq, wq je .end > + > + add wq, rq > + > + .loop1: > + neg xq > + mov vb, [srcq + xq] > + neg xq > + mov [dstq + xq], vb > + add xq, 1 > + cmp xq, wq > + jl .loop1 > .end: > +RET > _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
