[Ffmpeg-devel] benchmark of different CABAC routines

[Guillaume Poirier]

Hi,
Guillaume Poirier wrote:

>Hi,
>
>P4 Northwood 2.40GHz gcc (GCC) 3.3.5 (Debian 1:3.3.5-13) --march=pentium4
>
>I shall test on Prescott/Smithfield later if I have some time
>  
>
svn r6654, PentiumD 3Ghz, x86-64 mode, gcc-4.0, -march=nocona:

BRANCHLESS_CABAC_DECODER, CMOV_IS_FAST 1
7491 dezicycles in decode_residual, 2095551 runs, 1601 skipsbits/s dup=0 drop=0

BRANCHLESS_CABAC_DECODER, CMOV_IS_FAST 0
7489 dezicycles in decode_residual, 2095555 runs, 1597 skipsbits/s dup=0 drop=0

undef BRANCHLESS_CABAC_DECODER, CMOV_IS_FAST 1
7409 dezicycles in decode_residual, 2095486 runs, 1666 skipsbits/s dup=0 drop=0

undef BRANCHLESS_CABAC_DECODER, CMOV_IS_FAST 0
7408 dezicycles in decode_residual, 2095488 runs, 1664 skipsbits/s dup=0 drop=0

ARCH_X86_DISABLED, BRANCHLESS_CABAC_DECODER
7475 dezicycles in decode_residual, 2095542 runs, 1610 skipsbits/s dup=0 drop=0

ARCH_X86_DISABLED, undef BRANCHLESS_CABAC_DECODER
7662 dezicycles in decode_residual, 2095480 runs, 1672 skipsbits/s dup=0 drop=0

So this Pentium4 still doesn't like branchless code, and doesn't seem to
be fast at executing cmov either.
However, this time it likes assembly code.

It's really one-of-a-kind processor.

Guillaume