View previous topic :: View next topic |
Author |
Message |
Xithix Apprentice
Joined: 31 Dec 2004 Posts: 228
|
Posted: Sat Jan 15, 2005 1:57 pm Post subject: |
|
|
To fix permission denied, run as root. |
|
Back to top |
|
|
shadowteller n00b
Joined: 01 Oct 2004 Posts: 27 Location: Logan, UT
|
Posted: Fri Mar 11, 2005 10:28 pm Post subject: Acovea assistance |
|
|
I would like to say these are some pretty wild scripts. I have just run the scripts for about 26 hours now. Treebench.c got errors (*** glibc detected *** double free or corruption (fasttop): 0x0804e020 ***) so I just ^c out of it.
I got enough data though so that the cool perl script was able to do its deed. My question is now what do I do with it? I have read through this forum and am still quite confused. Code: | Score | So? | Switch (annotation)
------------------------------------------------------------------------------
36.2 | Yes | -freorder-blocks (-O2)
32.8 | Yes | -falign-loops (-O2 GCC 3.3)
32.5 | Yes | -fexpensive-optimizations (-O2)
31.6 | Yes | -falign-labels (-O2 GCC 3.3)
31.6 | Yes | -mieee-fp
31.1 | Yes | -fno-defer-pop (! -O1)
30.9 | Yes | -fpeephole2 (-O2)
30.6 | Yes | -fno-signaling-nans (fast math)
30.2 | Yes | -maccumulate-outgoing-args
30.1 | Yes | -mno-align-stringops
30.1 | Yes | -fcaller-saves (-O2)
29.9 | Yes | -fdelete-null-pointer-checks (-O2)
29.6 | Yes | -minline-all-stringops
28.7 | Yes | -mno-push-args
28.4 | Yes | -fno-crossjumping (! -O1)
28.4 | Yes | -fno-merge-constants (! -O1)
28.2 | Yes | -fsched-interblock (-O2 GCC 3.3)
27.9 | Yes | -fno-thread-jumps (! -O1)
27.3 | Maybe | -funsafe-math-optimizations (fast math)
27.0 | Maybe | -fcse-follow-jumps (-O2)
26.9 | Maybe | -fmove-all-movables
26.1 | Maybe | -ffinite-math-only (fast math)
26.1 | Maybe | -ftracer
25.9 | Maybe | -fno-trapping-math (fast math)
25.6 | Maybe | -fno-math-errno (fast math)
25.6 | Maybe | -fregmove (-O2)
25.5 | Maybe | -malign-double
25.4 | Yes | -fprefetch-loop-arrays
25.4 | Yes | -freorder-functions (-O2 GCC 3.3)
25.1 | Yes | -falign-jumps (-O2 GCC 3.3)
25.0 | Yes | -fno-if-conversion2 (! -O1)
24.9 | Yes | -fno-cprop-registers (! -O1)
24.6 | Yes | -finline-limit
24.5 | Yes | -fno-delayed-branch (! -O1)
24.2 | Maybe | -frerun-cse-after-loop (-O2)
24.0 | No | -fschedule-insns2 (-O2)
22.4 | Maybe | -finline-functions (-O3)
21.7 | No | -fforce-mem (-O2)
21.4 | Maybe | -frename-registers (-O3)
21.3 | Yes | -fno-omit-frame-pointer (! -O1)
21.2 | No | -fstrict-aliasing (-O2)
20.9 | Maybe | -fcse-skip-blocks (-O2)
20.7 | No | -fno-loop-optimize (! -O1)
19.4 | No | -fno-if-conversion (! -O1)
19.2 | Maybe | -frerun-loop-opt (-O2)
19.1 | Maybe | -freduce-all-givs
19.0 | Maybe | -fno-inline
18.1 | No | -fstrength-reduce (-O2)
17.2 | Maybe | -fsched-spec (-O2 GCC 3.3)
14.9 | No | -fnew-ra
14.8 | No | -foptimize-sibling-calls (-O2)
13.9 | No | -fgcse (-O2)
13.4 | No | -funroll-all-loops
11.2 | No | -fno-guess-branch-probability (! -O1)
8.7 | No | -funroll-loops
7.6 | No | -momit-leaf-frame-pointer
6.1 | No | -fomit-frame-pointer
6.1 | No | -fschedule-insns (-O2)
5.3 | No | -mfpmath=387
0.0 | No | -ffloat-store
0.0 | No | -mfpmath=sse
0.0 | No | -mfpmath=sse,387
|
My current cflags are "CFLAGS="-march=pentium4 -O3 -mtune=pentium4 -fomit-frame-pointer -pipe -ffast-math -ftracer"" What should I do? Once I set whatever the new flags are should I recompile my kernel?
Thanx all again. And thanx for such uber scripts. |
|
Back to top |
|
|
scaramanga Apprentice
Joined: 31 May 2005 Posts: 171
|
Posted: Tue May 31, 2005 6:02 pm Post subject: |
|
|
I have the same prob, running Acovea.
The post is; https://forums.gentoo.org/viewtopic-t-343680-highlight-acovea.html
I was wondering if I could use the results as below anyway??. They nmight not be sooperreliable, but very useful nonetheless?.
Score | So? | Switch (annotation)
------------------------------------------------------------------------------
30.4 | Yes | -fno-crossjumping (! -O1)
30.2 | Yes | -fno-if-conversion2 (! -O1)
29.8 | Yes | -fprefetch-loop-arrays
29.5 | Yes | -falign-labels (-O2 GCC 3.3)
29.0 | Yes | -freorder-functions (-O2 GCC 3.3)
28.5 | Maybe | -funsafe-math-optimizations (fast math)
27.4 | Yes | -minline-all-stringops
27.2 | Yes | -frename-registers (-O3)
26.8 | Yes | -fno-delayed-branch (! -O1)
26.3 | Yes | -fsched-interblock (-O2 GCC 3.3)
26.1 | Yes | -fcse-follow-jumps (-O2)
25.7 | Yes | -falign-loops (-O2 GCC 3.3)
25.6 | Yes | -frerun-cse-after-loop (-O2)
25.6 | Yes | -fno-omit-frame-pointer (! -O1)
25.5 | Yes | -fno-signaling-nans (fast math)
25.4 | Yes | -finline-functions (-O3)
25.3 | Yes | -maccumulate-outgoing-args
25.3 | Yes | -freorder-blocks (-O2)
25.1 | Yes | -fno-thread-jumps (! -O1)
24.3 | Maybe | -fno-math-errno (fast math)
24.2 | Yes | -fcaller-saves (-O2)
23.8 | Yes | -fmove-all-movables
23.1 | Yes | -foptimize-sibling-calls (-O2)
22.9 | Yes | -fdelete-null-pointer-checks (-O2)
22.7 | Yes | -fstrength-reduce (-O2)
22.0 | Yes | -mno-align-stringops
21.8 | Maybe | -fno-defer-pop (! -O1)
21.8 | Maybe | -fgcse (-O2)
21.4 | Maybe | -falign-jumps (-O2 GCC 3.3)
21.2 | Maybe | -fpeephole2 (-O2)
20.6 | Maybe | -ffinite-math-only (fast math)
20.6 | No | -frerun-loop-opt (-O2)
20.1 | Maybe | -fno-trapping-math (fast math)
20.0 | Maybe | -malign-double
20.0 | Maybe | -ftracer
19.8 | Yes | -freduce-all-givs
19.5 | Maybe | -mieee-fp
19.4 | Maybe | -fsched-spec (-O2 GCC 3.3)
19.1 | Yes | -funroll-loops
19.0 | Maybe | -fno-merge-constants (! -O1)
19.0 | No | -fforce-mem (-O2)
18.8 | Maybe | -fexpensive-optimizations (-O2)
18.6 | Maybe | -mno-push-args
18.0 | Maybe | -finline-limit
17.8 | Maybe | -fcse-skip-blocks (-O2)
17.7 | No | -fstrict-aliasing (-O2)
16.5 | Yes | -fno-cprop-registers (! -O1)
15.3 | No | -fno-guess-branch-probability (! -O1)
15.0 | Yes | -funroll-all-loops
13.9 | No | -fschedule-insns2 (-O2)
12.8 | No | -fno-inline
12.2 | No | -fno-if-conversion (! -O1)
11.5 | No | -fregmove (-O2)
11.0 | No | -fnew-ra
7.1 | No | -fschedule-insns (-O2)
5.5 | No | -fomit-frame-pointer
5.0 | No | -fno-loop-optimize (! -O1)
3.2 | No | -ffloat-store
0.0 | No | -mfpmath=387
0.0 | No | -mfpmath=sse
0.0 | No | -mfpmath=sse,387
0.0 | No | -momit-leaf-frame-pointer |
|
Back to top |
|
|
scaramanga Apprentice
Joined: 31 May 2005 Posts: 171
|
Posted: Fri Jun 03, 2005 6:22 pm Post subject: |
|
|
From the above list of suggested Acovea CFLAGS, I use the following;
CFLAGS="-O2 -march=pentium4 -fno-crossjumping -fno-if-conversion2 -fprefetch-loop-arrays -minline-all-stringops -frename-registers -fno-delayed-branch -fno-omit-frame-pointer -finline-functions -maccumulate-outgoing-args -fno-thread-jumps -fmove-all-movables -mno-align-stringops -funroll-loops -fno-cprop-registers -funroll-all-loops -pipe -wall"
Have I deleted all dangerous flags there?. |
|
Back to top |
|
|
bur Apprentice
Joined: 20 Feb 2004 Posts: 229
|
Posted: Fri Jun 17, 2005 1:47 pm Post subject: |
|
|
Sorry if this is already covered somewhere in this thread, but reading 16 pages was a bit too much...
I have a Duron 1000, what config file should I use? The first posts in this thread (from 2004 though) say that noone was able to run the benchmark on an Athlon so far. And after emerging acovea I only had files for pentium and opteron. Is there any place to get configs for other CPUs like a Duron? Using -march=athlon-xp does work for me, so maybe I could also go with an Athlon XP file. |
|
Back to top |
|
|
bur Apprentice
Joined: 20 Feb 2004 Posts: 229
|
Posted: Sat Jul 02, 2005 11:22 pm Post subject: |
|
|
Okay, so I edited the gcc3.3_opteron file. I used -march=athlon-xp and commented out all "specific to Intel-style architecture" and "fast-math" options. The tests do run without problems so far, though I don't have any results yet.
What I do wonder is, if there are specific intel options, are there specific amd options that I should insert in the config file? |
|
Back to top |
|
|
LinuxBlues l33t
Joined: 26 Mar 2005 Posts: 892
|
Posted: Wed Oct 12, 2005 9:12 pm Post subject: |
|
|
Scott Ladd (Acovea's author) wrote: | Results from running Acovea against a few algorithms should NOT be applied across a broad spectrum of applications. For my Gentoo-based systems, I don't set the value of make.conf's CFLAGS based on Acovea results; I build specific, time-critical applications using algorithm-specific options. |
http://www.coyotegulch.com/products/acovea/
Acovea is using only a reduced number of algorithms, it explains the reduced performance of -mfpmath=sse, for instance. |
|
Back to top |
|
|
|