[FFmpeg-devel] C99 or GCC extensions

[Axel Holzinger]

Axel Holzinger <aholzinger at gmx.de> writes:

<...>

> > Please also post some benchmarks that show that the uses of
> > FASTDIV didnt become slower. (see START/STOP_TIMER)
> 
> Will do so, although I won't be able to do this for ARM V4, which is
> also concerned.
> 
> Of course I checked the disassembly on Intel C compiler and I
> couldn't see a difference in the code.

Hi again, I looked at the START/STOP TIMER code and come up with the
attached main.c file which implements loops of FASTDIVs with the
existing and the proposed implementation together with the definition
and table of FASTDIV, to keep everything in a single file.

I hope I did get this right, otherwise the following output on my dual
Core 2 Duo machine is obsolete:

65297359600 dezicycles in FASTDIV, 1 runs, 0 skips
65274361920 dezicycles in FASTDIV, 2 runs, 0 skips
65277230340 dezicycles in FASTDIV, 4 runs, 0 skips
65280300170 dezicycles in FASTDIV, 8 runs, 0 skips
65303594555 dezicycles in FASTDIV, 16 runs, 0 skips
65325987667 dezicycles in FASTDIV, 32 runs, 0 skips
65323255820 dezicycles in FASTDIV, 64 runs, 0 skips
65406804228 dezicycles in FASTDIV, 128 runs, 0 skips

65233954960 dezicycles in FASTDIV_NEW, 1 runs, 0 skips
65271136560 dezicycles in FASTDIV_NEW, 2 runs, 0 skips
65343969580 dezicycles in FASTDIV_NEW, 4 runs, 0 skips
65718616670 dezicycles in FASTDIV_NEW, 8 runs, 0 skips
65879762285 dezicycles in FASTDIV_NEW, 16 runs, 0 skips
65961462195 dezicycles in FASTDIV_NEW, 32 runs, 0 skips
65757263897 dezicycles in FASTDIV_NEW, 64 runs, 0 skips
65566616031 dezicycles in FASTDIV_NEW, 128 runs, 0 skips

To me the differences look marginal and all are in the sub percent
range.

HTH
Axel
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: main.c
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080720/52c88ebd/attachment.txt>