Bug 62224 – ebuild patch for musepack-tools => 30% speedup in encoding

• Status: RESOLVED FIXED
• Product: Gentoo Linux

[Description Joël 2004-08-30 05:44:48 UTC]

Here's a patch for musepack-tools-1.15r-r1.ebuild

With a pentium-m I get 30% speedup on encoding. Please test on Pentium 4, Athlon XP.

The patch simply adds a filter for SSE flags. If this is a win for Pentium 4 too, then we should definitely make a -r2 with this :-)

PS: Yes, the patch is safe. It only disables stuff.

[Comment 1 Joël 2004-08-30 05:46:41 UTC]

Created attachment 38498 [details, diff]
Patch for musepack-tools-1.15r-r1.ebuild

Only useful for people who have -mfpmath=sse or -mfpmath=sse,387 in their
CFLAGS. Please confirm the speed improvement in this thread :-)

[Comment 2 Joël 2004-08-30 05:52:59 UTC]

Oh, I forgot to mention..  the speedup was present with:

sys-devel/gcc-3.3.4-r1
sys-devel/gcc-3.4.1-r2

[Comment 3 Jeremy Huddleston (RETIRED) 2004-09-01 00:57:38 UTC]

I don't have access to a P4 to test this out... can someone else in sound grab this...

[Comment 4 Patrick Dawson (RETIRED) 2004-09-05 16:39:21 UTC]

I'll test on P4 soon.

[Comment 5 Patrick Dawson (RETIRED) 2004-09-05 17:00:27 UTC]

My system:
Pentium 4 2.40GHz (HyperThreading enabled)
gcc-3.3.4-r1

Test file: an 80MB WAV

Methodology: rm -f test.mpc && time mppenc test.wav

Data:
CFLAGS="-O2 -march=pentium4 -fomit-frame-pointer -ftracer -pipe"
0m26.378s
0m26.177s
0m26.314s

CFLAGS="-O2 -march=pentium4 -fomit-frame-pointer -ftracer -pipe -mfpmath=sse"
0m26.500s
0m26.557s
0m26.397s

CFLAGS="-O2 -march=pentium4 -fomit-frame-pointer -ftracer -pipe -mfpmath=sse,387"
0m30.726s
0m30.593s
0m31.023s

Summary:
-mfpmath=sse makes little or no difference
-mfpmath=sse,387 slows it down

Conclusion: It's safe to filter these flags on pentium4.

[Comment 6 Patrick Dawson (RETIRED) 2004-09-05 17:28:08 UTC]

In portage. I bumped the revision because this is a pretty quick compile.