[MPlayer-dev-eng] [PATCH] -fno-strict-aliasing on gcc > 3

[Uoti A Urpala]

Reimar Döffinger wrote:
> Anyway I'd say the values are well in the error margin.

> compiled with gcc 3.4.4 (Debian)


gcc 3.4 does less with strict-aliasing than gcc 4 (and most people
didn't seem to hit problems with non-strict-aliasing-safe code before
gcc 4).

Here are some benchmarks with gcc 4.0.1. The video is a 1280x544 h264
.mov trailer decoded with FFmpeg H.264.

mplayer -nosound -vo null -benchmark serenity_m720p.mov

with default optimization:

BENCHMARKs: VC:  82.437s VO:   0.015s A:   0.000s Sys:   1.883s =   84.335s
BENCHMARK%: VC: 97.7495% VO:  0.0179% A:  0.0000% Sys:  2.2326% = 100.0000%

BENCHMARKs: VC:  82.791s VO:   0.014s A:   0.000s Sys:   1.301s =   84.106s
BENCHMARK%: VC: 98.4365% VO:  0.0164% A:  0.0000% Sys:  1.5471% = 100.0000%

BENCHMARKs: VC:  82.633s VO:   0.013s A:   0.000s Sys:   1.286s =   83.932s
BENCHMARK%: VC: 98.4522% VO:  0.0161% A:  0.0000% Sys:  1.5317% = 100.0000%


with -fno-strict-aliasing:

BENCHMARKs: VC:  86.441s VO:   0.014s A:   0.000s Sys:   1.318s =   87.772s
BENCHMARK%: VC: 98.4836% VO:  0.0154% A:  0.0000% Sys:  1.5010% = 100.0000%

BENCHMARKs: VC:  86.486s VO:   0.013s A:   0.000s Sys:   1.286s =   87.785s
BENCHMARK%: VC: 98.5202% VO:  0.0148% A:  0.0000% Sys:  1.4650% = 100.0000%

BENCHMARKs: VC:  86.573s VO:   0.013s A:   0.000s Sys:   1.311s =   87.897s
BENCHMARK%: VC: 98.4936% VO:  0.0153% A:  0.0000% Sys:  1.4910% = 100.0000%


I'd say that's at least clearly above the error margin.