| View previous topic :: View next topic |
| Author |
Message |
Angel666
n00b
Joined: 14 Nov 2003
Posts: 45
Location: Palo Alto, CA
|
 Posted: Sun Apr 11, 2004 6:25 am Post subject: |
 |
|
First of all, i would like to say thanks to Hypnos for actually writing this brilliant little script!
Now, my results:
| Code: | Score | So? | Switch (annotation)
------------------------------------------------------------------------------
35.0 | Yes | -freorder-blocks (-O2)
31.8 | Yes | -falign-loops (-O2 GCC 3.3)
31.5 | Yes | -fdelete-null-pointer-checks (-O2)
31.3 | Yes | -fno-if-conversion2 (! -O1)
31.3 | Maybe | -finline-functions (-O3)
30.4 | Yes | -malign-double
29.5 | Yes | -funit-at-a-time
29.5 | Yes | -fno-trapping-math (fast math)
29.2 | Yes | -fpeephole2 (-O2)
29.1 | Maybe | -fexpensive-optimizations (-O2)
28.6 | Yes | -ffinite-math-only (fast math)
28.4 | Maybe | -fpeel-loops
28.2 | Maybe | -fforce-mem (-O2)
28.0 | Yes | -mno-push-args
27.7 | Maybe | -fsched-interblock (-O2 GCC 3.3)
27.6 | Yes | -fregmove (-O2)
27.6 | Maybe | -fgcse (-O2)
27.5 | Maybe | -funsafe-math-optimizations (fast math)
27.2 | Yes | -maccumulate-outgoing-args
27.2 | Yes | -falign-functions
27.1 | Yes | -finline-limit
26.8 | Yes | -fno-crossjumping (! -O1)
26.7 | Yes | -fsched-spec (-O2 GCC 3.3)
26.5 | Yes | -falign-jumps (-O2 GCC 3.3)
26.5 | Maybe | -fstrict-aliasing (-O2)
26.4 | Yes | -fcse-skip-blocks (-O2)
26.0 | Yes | -fno-signaling-nans (fast math)
25.8 | Yes | -fno-omit-frame-pointer (! -O1)
25.5 | Yes | -minline-all-stringops
25.4 | Yes | -ftracer
25.1 | Yes | -funswitch-loops
24.9 | Maybe | -fweb
24.7 | Maybe | -frerun-loop-opt (-O2)
24.6 | Maybe | -fcaller-saves (-O2)
24.5 | Maybe | -falign-labels (-O2 GCC 3.3)
24.4 | Maybe | -frerun-cse-after-loop (-O2)
24.4 | Yes | -fmove-all-movables
24.2 | Maybe | -fcse-follow-jumps (-O2)
24.2 | Maybe | -mieee-fp
23.9 | Yes | -fno-defer-pop (! -O1)
23.8 | Yes | -frename-registers (-O3)
23.5 | Maybe | -fno-thread-jumps (! -O1)
23.3 | Maybe | -fno-cprop-registers (! -O1)
22.9 | Yes | -freorder-functions (-O2 GCC 3.3)
22.7 | Yes | -fno-delayed-branch (! -O1)
22.1 | Yes | -freduce-all-givs
21.8 | Maybe | -fstrength-reduce (-O2)
21.0 | Maybe | -fno-math-errno (fast math)
20.3 | No | -fschedule-insns2 (-O2)
19.5 | Maybe | -mno-align-stringops
18.6 | Maybe | -foptimize-sibling-calls (-O2)
18.1 | Maybe | -fno-merge-constants (! -O1)
17.3 | No | -fno-if-conversion (! -O1)
16.9 | Maybe | -fbranch-target-load-optimize
15.8 | No | -fprefetch-loop-arrays
14.9 | No | -fschedule-insns (-O2)
11.1 | No | -fnew-ra
10.8 | No | -fno-loop-optimize (! -O1)
9.7 | No | -ffloat-store
9.4 | No | -fomit-frame-pointer
8.5 | No | -funroll-loops
8.0 | No | -momit-leaf-frame-pointer
6.7 | No | -fno-guess-branch-probability (! -O1)
6.6 | No | -funroll-all-loops
6.1 | No | -fno-inline
0.0 | No | -fbranch-target-load-optimize2
0.0 | No | -mfpmath=387
0.0 | No | -mfpmath=sse
0.0 | No | -mfpmath=sse,387 |
And my new CFLAGS (chosen with -O2 already used and only "Yes" options):
| Code: | | CFLAGS="-O2 -Wall -pipe -march=pentium4 -fno-if-conversion2 -finline-functions -funit-at-a-time -fpeel-loops -mno-push-args -maccumulate-outgoing-args -falign-functions -finline-limit=600 -fno-crossjumping -fno-omit-frame-pointer -minline-all-stringops -ftracer -funswitch-loops -fmove-all-movables -fno-defer-pop -frename-registers -fno-delayed-branch -freduce-all-givs" |
I have to say, that is much better than it was before! Just a few notes - the link that you gave me, for optimizations, is for GCC 3.3, so it does not contain options like -fnew-ra, which are exactly the ones that I am interested in. If any documentation is availible, then it would be helpful.
Maybe i have a different ideal for this than anyone else, but here is my idea: (i wish i knew perl/python  ) Some options are very good for all benchmarks, or are at least positive for all of them. And some flags are good and bad for different operations. In my opinion, these should somehow be separated, for example additional options like "Stable" and "Unstable" so the user would know if that particular flag was good across the board or very specific. Maybe it would be redundant, but it would certainly make things simpler (in my opinion) if someone wanted to just choose optimistic options without having to compromize any programs, or for the person who wants full optimization power at the expense of a few slow apps.
_________________
"One World, One web, One program" - Microsoft Promo ad.
"Ein Volk, Ein Reich, Ein Fuhrer" - Adolf Hitler |
|
| Back to top |
|
 |
aethyr
Veteran
Joined: 06 Apr 2003
Posts: 1085
Location: NYC
|
| Posted: Sun Apr 11, 2004 7:07 am Post subject: |
|
|
| Angel666 wrote: | | the link that you gave me, for optimizations, is for GCC 3.3, so it does not contain options like -fnew-ra, which are exactly the ones that I am interested in. If any documentation is availible, then it would be helpful. |
I don't know if you're talking about the link I provided:
http://gcc.gnu.org/onlinedocs/gcc-3.3/gcc/Optimize-Options.html
http://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html
AFAIK -fnew-ra is a 3.3 flag. As found on those provided links:
| Code: | -fnew-ra
Use a graph coloring register allocator. Currently this option is meant for testing, so we are interested to hear about miscompilations with -fnew-ra.
-ftracer
Perform tail duplication to enlarge superblock size. This transformation simplifies the control flow of the function allowing other optimizations to do better job. |
|
|
| Back to top |
|
|
Angel666
n00b
Joined: 14 Nov 2003
Posts: 45
Location: Palo Alto, CA
|
| Posted: Sun Apr 11, 2004 7:28 am Post subject: |
|
|
Ooops... sorry. I don't know how i missed that, i swear i searched both those pages for those flags... 
_________________
"One World, One web, One program" - Microsoft Promo ad.
"Ein Volk, Ein Reich, Ein Fuhrer" - Adolf Hitler |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Sun Apr 11, 2004 8:08 am Post subject: |
|
|
| Angel666 wrote: | | Maybe i have a different ideal for this than anyone else, but here is my idea: (i wish i knew perl/python ) Some options are very good for all benchmarks, or are at least positive for all of them. |
Yes, they have high scores and are recommended based on statistical confidence.
| Quote: | | And some flags are good and bad for different operations. |
The opposite.
| Quote: | | In my opinion, these should somehow be separated, for example additional options like "Stable" and "Unstable" so the user would know if that particular flag was good across the board or very specific. Maybe it would be redundant, but it would certainly make things simpler (in my opinion) if someone wanted to just choose optimistic options without having to compromize any programs, or for the person who wants full optimization power at the expense of a few slow apps. |
"Score" tells you how well you make out in the end after compiling a running a gazillion apps. "So?" is a recommendation of whether or not it's worth it, given the volatility in the positive performance of the flag.
I think my script does what you describe, but perhaps I misunderstand ...
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
ikaro
Advocate
Joined: 14 Jul 2003
Posts: 2527
Location: Denmark
|
| Posted: Sun Apr 11, 2004 8:11 am Post subject: |
|
|
should i buy another cpu ?
_________________
linux: #232767
Last edited by ikaro on Fri May 14, 2004 9:05 pm; edited 1 time in total |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Sun Apr 11, 2004 8:15 am Post subject: |
|
|
| ikaro wrote: | | should i buy another cpu ? |
Goodness, no!  All this says is the gcc isn't quite sure what to do with your CPU, but that doesn't mean it's slow, even in comparison to optimized code running on Intel CPU.
I think the only combination that would win hands down is icc+Intel.
----
Perhaps it's placebo, but after recompiling Mozilla and X.org-X11 with my custom CFLAGS forced, my desktop is a bit snappier. We'll see how stable it runs ...
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
ikaro
Advocate
Joined: 14 Jul 2003
Posts: 2527
Location: Denmark
|
| Posted: Sun Apr 11, 2004 8:53 am Post subject: |
|
|
Edited after recomendation :
| Code: |
CFLAGS="
-march=athlon-xp
-O3
-pipe
-fno-cprop-registers
-fno-thread-jumps
-fno-defer-pop
-maccumulate-outgoing-args
-fno-if-conversion2
-fno-delayed-branch
-fno-crossjumping
-fno-merge-constants
-fno-omit-frame-pointer
-ftracer
-finline-limit=600
-minline-all-stringops
-mno-push-args
-fmove-all-movables
-mno-align-stringops"
|
_________________
linux: #232767
Last edited by ikaro on Sun Apr 11, 2004 12:46 pm; edited 3 times in total |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Sun Apr 11, 2004 9:07 am Post subject: |
|
|
| ikaro wrote: | so how does this looks like ?
| Code: |
CFLAGS="
-march=athlon-xp
-O3
-pipe
-maccumulate-outgoing-args
-ftracer
-finline-limit=600
-fno-cprop-registers
-fno-crossjumping
-fno-thread-jumps
-fno-defer-pop
-fno-signaling-nans
-fno-if-conversion2
-fno-omit-frame-pointer
-minline-all-stringops
-mno-push-args
-fmove-all-movables
-malign-double"
|
Thanks. |
It generally looks fine; but:
* I don't understand your logic for choosing the "maybe" switches, but it probably doesn't matter much (that's why they're "maybe"'s) ...
* I recommend against "-malign-double" (which is considered unsafe) and "-fno-signaling-nans" which can confuse programs which rely on standards-compliant floating point math.
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
ikaro
Advocate
Joined: 14 Jul 2003
Posts: 2527
Location: Denmark
|
| Posted: Sun Apr 11, 2004 9:38 am Post subject: |
|
|
I will take your advice and update the cflags. im only posting the results and let people decide whats best.
Thanks again.
_________________
linux: #232767
Last edited by ikaro on Wed Apr 14, 2004 4:22 pm; edited 1 time in total |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Sun Apr 11, 2004 10:10 am Post subject: |
|
|
I am pleased -- my speed boost is not placebo: Q3A is playable now at max settings, and UT2004 is more responsive (texture loading is still an obvious problem with my OSS Radeon driver).
Overall, it's comparable to the boost you get moving from 2.4->2.6 (I'm stuck on 2.4 since suspend-to-disk doesn't work for me on 2.6).
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
lookitsme
n00b
Joined: 06 Nov 2003
Posts: 48
Location: Kuala Lumpur, Malaysia
|
| Posted: Mon Apr 12, 2004 12:06 pm Post subject: |
|
|
I'm a bit confused here. I'm running the latest version of the run script at the moment. Its finished the 1st 4 test already. And if I run the digest perl script it gives me all no's:
| Code: |
Score | So? | Switch (annotation)
------------------------------------------------------------------------------
14.6 | No | -fno-thread-jumps (! -O1)
14.3 | No | -fschedule-insns2 (-O2)
13.6 | No | -frerun-loop-opt (-O2)
12.6 | No | -finline-limit
12.5 | No | -minline-all-stringops
12.5 | No | -funsafe-math-optimizations (fast math)
12.2 | No | -fsched-spec (-O2 GCC 3.3)
12.0 | No | -fdelete-null-pointer-checks (-O2)
11.9 | No | -fcaller-saves (-O2)
11.7 | No | -mieee-fp
11.6 | No | -falign-labels (-O2 GCC 3.3)
11.5 | No | -fno-omit-frame-pointer (! -O1)
11.3 | No | -falign-jumps (-O2 GCC 3.3)
11.1 | No | -fpeephole2 (-O2)
11.1 | No | -mno-align-stringops
10.9 | No | -falign-loops (-O2 GCC 3.3)
10.7 | No | -fno-defer-pop (! -O1)
10.6 | No | -frename-registers (-O3)
10.5 | No | -mno-push-args
10.3 | No | -fno-if-conversion2 (! -O1)
10.2 | No | -frerun-cse-after-loop (-O2)
10.2 | No | -fforce-mem (-O2)
10.0 | No | -fexpensive-optimizations (-O2)
9.9 | No | -fno-delayed-branch (! -O1)
9.7 | No | -freorder-functions (-O2 GCC 3.3)
9.6 | No | -fcse-skip-blocks (-O2)
9.5 | No | -malign-double
9.3 | No | -fno-merge-constants (! -O1)
9.1 | No | -fno-signaling-nans (fast math)
9.1 | No | -fno-loop-optimize (! -O1)
9.1 | No | -fsched-interblock (-O2 GCC 3.3)
8.9 | No | -freduce-all-givs
8.9 | No | -fmove-all-movables
8.7 | No | -foptimize-sibling-calls (-O2)
8.5 | No | -fcse-follow-jumps (-O2)
8.3 | No | -fgcse (-O2)
8.3 | No | -finline-functions (-O3)
8.1 | No | -fstrength-reduce (-O2)
7.5 | No | -ffinite-math-only (fast math)
7.5 | No | -freorder-blocks (-O2)
7.2 | No | -ftracer
7.2 | No | -fprefetch-loop-arrays
6.9 | No | -fno-crossjumping (! -O1)
6.1 | No | -fno-math-errno (fast math)
6.0 | No | -fregmove (-O2)
5.4 | No | -fstrict-aliasing (-O2)
5.4 | No | -fno-cprop-registers (! -O1)
5.1 | No | -fno-if-conversion (! -O1)
5.0 | No | -fno-inline
4.6 | No | -fschedule-insns (-O2)
0.0 | No | -fno-guess-branch-probability (! -O1)
0.0 | No | -ffloat-store
0.0 | No | -fnew-ra
0.0 | No | -funroll-loops
0.0 | No | -funroll-all-loops
0.0 | No | -maccumulate-outgoing-args
0.0 | No | -mfpmath=387
0.0 | No | -mfpmath=sse
0.0 | No | -mfpmath=sse,387
0.0 | No | -fomit-frame-pointer
0.0 | No | -momit-leaf-frame-pointer
0.0 | No | -fno-trapping-math (fast math)
|
Am I just impatient and should I wait for all 7 tests to finish??? |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Mon Apr 12, 2004 1:23 pm Post subject: |
|
|
| lookitsme wrote: | | Am I just impatient and should I wait for all 7 tests to finish??? |
Yes. What's happening in all likelihood is that the standard deviation is too big for the means you're getting.
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
aethyr
Veteran
Joined: 06 Apr 2003
Posts: 1085
Location: NYC
|
| Posted: Mon Apr 12, 2004 3:17 pm Post subject: |
|
|
Results! 850Mhz Celeron2 (Coppermine) gcc 3.3: | Code: |
Score | So? | Switch (annotation)
------------------------------------------------------------------------------
31.0 | Maybe | -fgcse (-O2)
30.1 | Yes | -frerun-cse-after-loop (-O2)
29.9 | Maybe | -freorder-functions (-O2 GCC 3.3)
29.8 | Yes | -finline-functions (-O3)
29.8 | Yes | -mno-align-stringops
29.4 | Yes | -fmove-all-movables
29.1 | Maybe | -fforce-mem (-O2)
28.4 | Maybe | -mieee-fp
28.3 | Yes | -fno-defer-pop (! -O1)
28.0 | Yes | -maccumulate-outgoing-args
28.0 | Yes | -fno-cprop-registers (! -O1)
27.8 | Yes | -falign-labels (-O2 GCC 3.3)
27.5 | Yes | -finline-limit
27.1 | Maybe | -fschedule-insns2 (-O2)
26.9 | Yes | -fcse-follow-jumps (-O2)
26.9 | Yes | -fsched-interblock (-O2 GCC 3.3)
26.6 | Yes | -foptimize-sibling-calls (-O2)
26.6 | Yes | -fno-math-errno (fast math)
26.4 | Yes | -fno-merge-constants (! -O1)
26.3 | Maybe | -fno-trapping-math (fast math)
26.1 | Yes | -fcaller-saves (-O2)
25.9 | Yes | -fsched-spec (-O2 GCC 3.3)
25.8 | Yes | -fprefetch-loop-arrays
25.7 | Yes | -fdelete-null-pointer-checks (-O2)
25.3 | Yes | -ffinite-math-only (fast math)
24.8 | Yes | -fno-thread-jumps (! -O1)
24.7 | Yes | -fno-crossjumping (! -O1)
24.7 | Yes | -fno-signaling-nans (fast math)
24.5 | No | -fstrict-aliasing (-O2)
24.3 | Maybe | -fno-if-conversion2 (! -O1)
24.1 | Maybe | -minline-all-stringops
23.7 | Maybe | -frerun-loop-opt (-O2)
23.6 | Yes | -fcse-skip-blocks (-O2)
23.3 | Maybe | -ftracer
23.2 | Maybe | -funsafe-math-optimizations (fast math)
23.0 | Maybe | -frename-registers (-O3)
22.7 | Yes | -malign-double
22.7 | Maybe | -mno-push-args
21.7 | Maybe | -falign-loops (-O2 GCC 3.3)
21.4 | Maybe | -freorder-blocks (-O2)
20.5 | Maybe | -fexpensive-optimizations (-O2)
20.2 | Maybe | -fpeephole2 (-O2)
19.8 | Maybe | -fstrength-reduce (-O2)
19.6 | Maybe | -falign-jumps (-O2 GCC 3.3)
19.4 | Maybe | -fno-omit-frame-pointer (! -O1)
19.4 | Maybe | -fno-delayed-branch (! -O1)
19.2 | Maybe | -freduce-all-givs
17.3 | No | -fno-inline
13.8 | No | -funroll-all-loops
13.2 | No | -fno-if-conversion (! -O1)
12.3 | No | -fregmove (-O2)
11.7 | No | -fomit-frame-pointer
11.5 | No | -fnew-ra
11.2 | No | -fno-guess-branch-probability (! -O1)
9.3 | No | -fschedule-insns (-O2)
8.4 | No | -ffloat-store
6.9 | No | -fno-loop-optimize (! -O1)
0.0 | No | -funroll-loops
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 |
|
|
lookitsme
n00b
Joined: 06 Nov 2003
Posts: 48
Location: Kuala Lumpur, Malaysia
|
| Posted: Tue Apr 13, 2004 5:56 am Post subject: |
|
|
| Hypnos wrote: | | lookitsme wrote: | | Am I just impatient and should I wait for all 7 tests to finish??? |
Yes. What's happening in all likelihood is that the standard deviation is too big for the means you're getting. |
Thanks Hypnos, I showed a little more patience and the last test is just finished. Results for my Intel Pentium 4 with GCC 3.3:
| Code: |
Score | So? | Switch (annotation)
------------------------------------------------------------------------------
32.6 | Maybe | -fschedule-insns2 (-O2)
31.5 | Yes | -fno-defer-pop (! -O1)
31.0 | Yes | -fno-thread-jumps (! -O1)
30.8 | Yes | -falign-jumps (-O2 GCC 3.3)
30.8 | Yes | -fsched-spec (-O2 GCC 3.3)
29.9 | Yes | -fcaller-saves (-O2)
29.3 | Yes | -fpeephole2 (-O2)
29.2 | Yes | -finline-limit
29.1 | Yes | -falign-loops (-O2 GCC 3.3)
28.5 | Yes | -fdelete-null-pointer-checks (-O2)
28.3 | Maybe | -fcse-skip-blocks (-O2)
27.9 | Yes | -frerun-loop-opt (-O2)
27.9 | Yes | -frerun-cse-after-loop (-O2)
27.7 | Yes | -fno-omit-frame-pointer (! -O1)
27.7 | Maybe | -funsafe-math-optimizations (fast math)
27.5 | Yes | -mno-align-stringops
27.3 | Yes | -freorder-functions (-O2 GCC 3.3)
27.2 | Yes | -malign-double
26.6 | Maybe | -fforce-mem (-O2)
26.6 | Maybe | -fmove-all-movables
26.4 | Maybe | -frename-registers (-O3)
26.3 | Maybe | -fcse-follow-jumps (-O2)
25.8 | Maybe | -foptimize-sibling-calls (-O2)
24.3 | Maybe | -mno-push-args
24.3 | Maybe | -finline-functions (-O3)
24.3 | Maybe | -fno-signaling-nans (fast math)
24.1 | Maybe | -minline-all-stringops
24.1 | Yes | -freduce-all-givs
24.0 | Maybe | -fsched-interblock (-O2 GCC 3.3)
23.9 | Yes | -fno-if-conversion2 (! -O1)
23.2 | Maybe | -mieee-fp
22.8 | Maybe | -ftracer
22.7 | Maybe | -fno-merge-constants (! -O1)
22.6 | Yes | -freorder-blocks (-O2)
22.5 | Maybe | -ffinite-math-only (fast math)
22.0 | Maybe | -falign-labels (-O2 GCC 3.3)
21.9 | Maybe | -fno-crossjumping (! -O1)
21.3 | Maybe | -fstrength-reduce (-O2)
21.3 | Maybe | -fno-math-errno (fast math)
20.7 | Maybe | -maccumulate-outgoing-args
20.0 | Maybe | -fno-delayed-branch (! -O1)
20.0 | No | -fexpensive-optimizations (-O2)
18.7 | No | -fgcse (-O2)
17.8 | Maybe | -fno-cprop-registers (! -O1)
17.3 | No | -fregmove (-O2)
16.9 | Maybe | -fprefetch-loop-arrays
16.8 | No | -fno-if-conversion (! -O1)
16.8 | No | -fstrict-aliasing (-O2)
15.2 | Maybe | -fno-trapping-math (fast math)
13.9 | No | -fschedule-insns (-O2)
13.3 | No | -fomit-frame-pointer
11.7 | No | -fno-guess-branch-probability (! -O1)
11.2 | No | -fno-loop-optimize (! -O1)
10.8 | Maybe | -momit-leaf-frame-pointer
10.6 | No | -fno-inline
7.8 | No | -ffloat-store
7.4 | No | -fnew-ra
7.1 | No | -funroll-loops
6.7 | No | -mfpmath=387
0.0 | No | -funroll-all-loops
0.0 | No | -mfpmath=sse
0.0 | No | -mfpmath=sse,387
|
I'm recompiling some programs at the moment (Mozilla, Evolution and what else I can find) with O2 and all Yes options from the results. But since I'm at work accessing my machine over SSH I cant tell the difference until tonight. Although I'm already under the impression that the compiling process itself speeded up. |
|
| Back to top |
|
|
nmcsween
Guru
Joined: 12 Nov 2003
Posts: 381
|
| Posted: Tue Apr 13, 2004 11:02 am Post subject: |
|
|
| Can someone tell me where exactly the scripts go? and what they should be called? and also how to run them? |
|
| Back to top |
|
|
aethyr
Veteran
Joined: 06 Apr 2003
Posts: 1085
Location: NYC
|
| Posted: Tue Apr 13, 2004 1:13 pm Post subject: |
|
|
| Ultraoctane.com wrote: | | Can someone tell me where exactly the scripts go? and what they should be called? and also how to run them? |
Scripts should probably go in the same directory that you generate your acovea data from, they can be called anything (just copy and paste them from this topic into a file). You can run them by the following command:
|
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Tue Apr 13, 2004 2:30 pm Post subject: |
|
|
| lookitsme wrote: | | [...] Although I'm already under the impression that the compiling process itself speeded up. |
Yes -- turning off a lot of the -O1 stuff seems to be a real time saver.
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
nmcsween
Guru
Joined: 12 Nov 2003
Posts: 381
|
| Posted: Tue Apr 13, 2004 3:33 pm Post subject: |
|
|
| Nice thread hypnos. Now im wondering is there a database of acovea tests anywhere? If not why don't we take the initiative? I could supply a few sql databases and about 1GB of webspace and also some php work. |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Tue Apr 13, 2004 5:57 pm Post subject: |
|
|
| Ultraoctane.com wrote: | | Nice thread hypnos. Now im wondering is there a database of acovea tests anywhere? |
Not that I know of.
| Quote: | | If not why don't we take the initiative? I could supply a few sql databases and about 1GB of webspace and also some php work. |
What do you have in mind exactly? What my script does is find the best "average" CFLAGS across a world-representative set of benchmarks for a given CPU -- that could all fit on one table, albeit a large one, but a database might be overkill ...
EDIT: I guess D>2, since you have CPU, switches and compiler version.
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
nmcsween
Guru
Joined: 12 Nov 2003
Posts: 381
|
| Posted: Tue Apr 13, 2004 8:52 pm Post subject: |
|
|
What I have in mind is to see how every flag preforms.
| Quote: | | GCC supports different switch sets for different languages and processors; in the case of the Pentium 4, for example, more than 60 different switches affect code optimization and generation. Testing every possible combination of those switches is simply impractical. Even at the unattainable theoretical speed of one test per second, it would take over hundreds of billions of years to run the complete set of possibilities! |
That is close to what I have in mind but to make it a reality why not only try the optimistic switches for that proc. Now this would reqiure some kind of distrubuted computing effort but it could be done. Maybe a online database on what switches have been tested on each proc? and a perl script that fetches that information and tests the next switche(s)? |
|
| Back to top |
|
|
nmcsween
Guru
Joined: 12 Nov 2003
Posts: 381
|
| Posted: Tue Apr 13, 2004 11:48 pm Post subject: |
|
|
| Even if it was just the overall best cflags in the a database it would save people a considerable amount of time. |
|
| Back to top |
|
|
wilburpan
l33t
Joined: 21 Jan 2003
Posts: 977
|
| Posted: Wed Apr 14, 2004 3:08 am Post subject: |
|
|
How long does it take to run the two acovea scripts (which I very imaginatively named acovea1 and acovea2, by the way  )? I'm using a 700 MHz P3.
_________________
I'm only hanging out in OTW until I get rid of this stupid l33t ranking.....Crap. That didn't work. |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Wed Apr 14, 2004 3:10 am Post subject: |
|
|
| Ultraoctane.com wrote: | | Even if it was just the overall best cflags in the a database it would save people a considerable amount of time. |
I agree.
So then, maybe a web interface where people enter the CPU specs and benchmark numbers or summary results from my script?
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
Hypnos
Advocate
Joined: 18 Jul 2002
Posts: 2889
Location: Omnipresent
|
| Posted: Wed Apr 14, 2004 3:32 am Post subject: |
|
|
| wilburpan wrote: | | How long does it take to run the two acovea scripts (which I very imaginatively named acovea1 and acovea2, by the way )? I'm using a 700 MHz P3. |
The first bash script, ~48 hours on my 1.6GHz P4; the second Perl script, a few milliseconds.
_________________
Personal overlay | Simple backup scheme |
|
| Back to top |
|
|
nmcsween
Guru
Joined: 12 Nov 2003
Posts: 381
|
| Posted: Wed Apr 14, 2004 4:30 am Post subject: |
|
|
| I think it would be better if it was inputed directly via a upload. A web interface seems like a burden on whoever runs the test.We could load it with a perl socket connection. Although I think I need root access on the server to use perl sockets. If anyone knows of a better way please post. |
|
| Back to top |
|
|
|
Gentoo Chat
