[x264-devel] X264 6 tap filter MMX Question

[Jason Garrett-Glaser]

2011/4/15 peter <bo3861926 at sina.com>:
> Hi,there.
>
> Recently I'm interested in the 6 tap filter MMX code in mc-a2.asm of
> X264.But I'm confused by such code (THE BLUE LINE)as below;
>
> ;-----------------------------------------------------------------------------
> ; void hpel_filter_h( uint8_t *dst, uint8_t *src, int width );
> ;-----------------------------------------------------------------------------
> cglobal hpel_filter_h_mmxext, 3,3
> add r0, r2
> add r1, r2
> neg r2
> %define src r1+r2
> pxor m0, m0
> .loop:
> movd m1, [src-2]
> movd m2, [src-1]
> movd m3, [src ]
> movd m6, [src+1]
> movd m4, [src+2]
> movd m5, [src+3]
> punpcklbw m1, m0
> punpcklbw m2, m0
> punpcklbw m3, m0
> punpcklbw m6, m0
> punpcklbw m4, m0
> punpcklbw m5, m0
> paddw m3, m6 ; c0
> paddw m2, m4 ; b0
> paddw m1, m5 ; a0
> movd m7, [src+7]
> movd m6, [src+6]
> punpcklbw m7, m0
> punpcklbw m6, m0
> paddw m4, m7 ; c1
> paddw m5, m6 ; b1
> movd m7, [src+5]
> movd m6, [src+4]
> punpcklbw m7, m0
> punpcklbw m6, m0
> paddw m6, m7 ; a1
> movq m7, [pw_1];
> FILT_H2 m1, m2, m3, m4, m5, m6
> FILT_PACK m1, m4, 1, m7;Here it is!
> movntq [r0+r2], m1
> add r2, 8
> jl .loop
> REP_RET
>
>
>
> FILT_PACK m1,m4,1,m7,This macro computes m1= m1+m7;m4=m4+m7; Before this
> step,m1 and m4 conatin 4 outcomes like this (A+F)/16-(B+E)*5/16+(C+D)*5/4,it
> almost complete the 6 tap filter.
>
>
>
> %macro FILT_PACK 4-6 b
> paddw %1, %4
> paddw %2, %4
> %if %0 == 6
> psubusw %1, %6
> psubusw %2, %6
> psrlw %1, %3
> psrlw %2, %3
> %else
> psraw %1, %3
> psraw %2, %3
> %endif
> %ifnidn w, %5
> packuswb %1, %2
> %endif
> %endmacro
>
> So my question : Why do the 8 points add the m7? What does contain in m7?
>
> Thanks anyone who reply this mail in advance~~~
>
> Best,
>
> Peter

Read the C code for the function.

(A + B*-5 + C*20 + D*20 + E*-5 + F + 32) >> 6

Note the 32.

Jason