* Some SPECfp results for the simplify-rtx problem
@ 2001-09-08 23:49 Andreas Jaeger
2001-09-09 1:28 ` Jan Hubicka
2001-09-09 5:35 ` Andreas Jaeger
0 siblings, 2 replies; 4+ messages in thread
From: Andreas Jaeger @ 2001-09-08 23:49 UTC (permalink / raw)
To: gcc
Here're the results of SPECfp2000 on an Athlon 1.2 Ghz MP system
(different machine to last time, you can't compare the numbers) as
asked for by Richard Henderson.
Base is an old GCC version without RTH's simplifiy-rtx.patch and Peak
uses the current GCC CVS that includes the simplifiy-rtx patch set:
Flags are: -O2 -march=athlon -malign-double
Estimated Estimated
Base Base Base Peak Peak Peak
Benchmarks Ref Time Run Time Ratio Ref Time Run Time Ratio
168.wupwise 1600 340 471* 1600 336 477*
171.swim 3100 805 385* 3100 791 392*
172.mgrid 1800 518 348* 1800 517 348*
173.applu 2100 685 306* 2100 693 303*
179.art 2600 1915 136* 2600 1907 136*
183.equake 1300 336 387* 1300 338 385*
200.sixtrack 1100 545 202* 1100 522 211*
Fortran 77 tests: 168.wupwise, 171.swim, 172.mgrid, 173.applu,
200.sixtrack, 301.apsi.
C tests (for reference): 177.mesa, 179.art, 183.equake, 188.ammp
Unfortunatly both compilers have errors and 177.mesa, 188.ammp and
301.apsi fail with an "Output miscompare".
I now run on the same machine tests with GCC with RTH's "address_cost
tweak" installed:
Base flags: -O2 -march=athlon -malign-double
Peak flags: -O2 -march=athlon -malign-double -fno-reduce-all-givs
Estimated Estimated
Base Base Base Peak Peak Peak
Benchmarks Ref Time Run Time Ratio Ref Time Run Time Ratio
------------ -------- -------- -------- -------- -------- --------
168.wupwise 1600 336 477* 1600 339 471*
171.swim 3100 980 316* 3100 783 396*
172.mgrid 1800 518 348* 1800 533 338*
173.applu 2100 694 303* 2100 654 321*
179.art 2600 1730 150* 2600 1709 152*
183.equake 1300 339 384* 1300 340 383*
200.sixtrack 1100 520 211* 1100 491 224*
And the same with the CVS version (without the "address_cost tweak"):
Base Base Base Peak Peak Peak
Benchmarks Ref Time Run Time Ratio Ref Time Run Time Ratio
------------ -------- -------- -------- -------- -------- --------
168.wupwise 1600 342 468* 1600 333 480*
171.swim 3100 1155 268* 3100 1151 269*
172.mgrid 1800 517 348* 1800 535 336*
173.applu 2100 693 303* 2100 655 321*
179.art 2600 1957 133* 2600 1906 136*
183.equake 1300 333 390* 1300 336 387*
200.sixtrack 1100 521 211* 1100 501 219*
Btw. Honza just committed a fix for "fcmov" which might fix the
failures I see.
Andreas
--
Andreas Jaeger
SuSE Labs aj@suse.de
private aj@arthur.inka.de
http://www.suse.de/~aj
^ permalink raw reply [flat|nested] 4+ messages in thread* Re: Some SPECfp results for the simplify-rtx problem
2001-09-08 23:49 Some SPECfp results for the simplify-rtx problem Andreas Jaeger
@ 2001-09-09 1:28 ` Jan Hubicka
2001-09-09 23:48 ` Andreas Jaeger
2001-09-09 5:35 ` Andreas Jaeger
1 sibling, 1 reply; 4+ messages in thread
From: Jan Hubicka @ 2001-09-09 1:28 UTC (permalink / raw)
To: Andreas Jaeger; +Cc: gcc
>
> Fortran 77 tests: 168.wupwise, 171.swim, 172.mgrid, 173.applu,
> 200.sixtrack, 301.apsi.
> C tests (for reference): 177.mesa, 179.art, 183.equake, 188.ammp
>
> Unfortunatly both compilers have errors and 177.mesa, 188.ammp and
> 301.apsi fail with an "Output miscompare".
>
> I now run on the same machine tests with GCC with RTH's "address_cost
> tweak" installed:
Please can you run it with the strength-reduction tweek I pointed out
in the reply to Richard's email? I plan to install it today, but I want
automated checker to come with results before I do so. (and the patch
needs re-bootstrapping as it aged somewhat)
It should track the same problem from other direction and thus I would
like to see effect of Richard's change on the updated code.
Also please run the integer tests too. The effect of Richard's patch can
be increase of register pressure in the integer code that can negatively affect
the performance.
>
> Base flags: -O2 -march=athlon -malign-double
> Peak flags: -O2 -march=athlon -malign-double -fno-reduce-all-givs
> Estimated Estimated
> Base Base Base Peak Peak Peak
> Benchmarks Ref Time Run Time Ratio Ref Time Run Time Ratio
> ------------ -------- -------- -------- -------- -------- --------
> 168.wupwise 1600 336 477* 1600 339 471*
> 171.swim 3100 980 316* 3100 783 396*
> 172.mgrid 1800 518 348* 1800 533 338*
> 173.applu 2100 694 303* 2100 654 321*
> 179.art 2600 1730 150* 2600 1709 152*
> 183.equake 1300 339 384* 1300 340 383*
> 200.sixtrack 1100 520 211* 1100 491 224*
Hmm, looks like we are at something...
>
> Btw. Honza just committed a fix for "fcmov" which might fix the
> failures I see.
Hope so. The specint misscomparisons are gone.
Honza
>
> Andreas
> --
> Andreas Jaeger
> SuSE Labs aj@suse.de
> private aj@arthur.inka.de
> http://www.suse.de/~aj
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: Some SPECfp results for the simplify-rtx problem
2001-09-09 1:28 ` Jan Hubicka
@ 2001-09-09 23:48 ` Andreas Jaeger
0 siblings, 0 replies; 4+ messages in thread
From: Andreas Jaeger @ 2001-09-09 23:48 UTC (permalink / raw)
To: Jan Hubicka; +Cc: gcc
Jan Hubicka <jh@suse.cz> writes:
>>
>> Fortran 77 tests: 168.wupwise, 171.swim, 172.mgrid, 173.applu,
>> 200.sixtrack, 301.apsi.
>> C tests (for reference): 177.mesa, 179.art, 183.equake, 188.ammp
>>
>> Unfortunatly both compilers have errors and 177.mesa, 188.ammp and
>> 301.apsi fail with an "Output miscompare".
>>
>> I now run on the same machine tests with GCC with RTH's "address_cost
>> tweak" installed:
> Please can you run it with the strength-reduction tweek I pointed out
> in the reply to Richard's email? I plan to install it today, but I want
> automated checker to come with results before I do so. (and the patch
> needs re-bootstrapping as it aged somewhat)
>
> It should track the same problem from other direction and thus I would
> like to see effect of Richard's change on the updated code.
>
> Also please run the integer tests too. The effect of Richard's patch can
> be increase of register pressure in the integer code that can negatively affect
> the performance.
Here're the results:
Base flags: -O2 -march=athlon -malign-double
Compiler is reference (CVS version)
Peak flags: -O2 -march=athlon -malign-double
Compiler is CVS + RT's "address_cost tweak"
CFP 2000
Estimated Estimated
Base Base Base Peak Peak Peak
Benchmarks Ref Time Run Time Ratio Ref Time Run Time Ratio
------------ -------- -------- -------- -------- -------- --------
168.wupwise 1600 337 474* 1600 336 476*
171.swim 3100 1085 286* 3100 1118 277*
172.mgrid 1800 521 345* 1800 516 349*
173.applu 2100 696 302* 2100 694 303*
177.mesa 1400 288 485* 1400 289 484*
179.art 2600 1906 136* 2600 1743 149*
183.equake 1300 341 382* 1300 340 382*
188.ammp 2200 840 262* 2200 853 258*
200.sixtrack 1100 520 212* 1100 520 212*
301.apsi 2600 946 275* 2600 949 274*
CINT2000
Estimated Estimated
Base Base Base Peak Peak Peak
Benchmarks Ref Time Run Time Ratio Ref Time Run Time Ratio
------------ -------- -------- -------- -------- -------- --------
164.gzip 1400 289 484* 1400 291 482*
175.vpr 1400 485 288* 1400 486 288*
176.gcc 1100 289 380* 1100 294 375*
181.mcf 1800 908 198* 1800 914 197*
186.crafty 1000 171 585* 1000 170 588*
197.parser 1800 506 356* 1800 506 356*
252.eon 1300 227 572* 1300 228 571*
253.perlbmk 1800 341 527* 1800 342 527*
254.gap 1100 246 447* 1100 246 447*
255.vortex 1900 440 432* 1900 437 435*
256.bzip2 1500 399 376* 1500 404 371*
300.twolf 3000 916 327* 3000 916 328*
Andreas
--
Andreas Jaeger
SuSE Labs aj@suse.de
private aj@arthur.inka.de
http://www.suse.de/~aj
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: Some SPECfp results for the simplify-rtx problem
2001-09-08 23:49 Some SPECfp results for the simplify-rtx problem Andreas Jaeger
2001-09-09 1:28 ` Jan Hubicka
@ 2001-09-09 5:35 ` Andreas Jaeger
1 sibling, 0 replies; 4+ messages in thread
From: Andreas Jaeger @ 2001-09-09 5:35 UTC (permalink / raw)
To: gcc
One more run with the current CVS version (including Honza's fcom
patch):
Base flags: -O2 -march=athlon -malign-double
Peak flags: -O2 -march=athlon -malign-double -fno-reduce-all-givs
Estimated Estimated
Base Base Base Peak Peak Peak
Benchmarks Ref Time Run Time Ratio Ref Time Run Time Ratio
------------ -------- -------- -------- -------- -------- --------
168.wupwise 1600 336 476* 1600 335 477*
171.swim 3100 1169 265* 3100 1143 271*
172.mgrid 1800 522 345* 1800 534 337*
173.applu 2100 697 301* 2100 657 320*
177.mesa 1400 289 484* 1400 293 477*
179.art 2600 1928 135* 2600 1904 137*
183.equake 1300 337 385* 1300 332 391*
188.ammp 2200 811 271* 2200 823 267*
200.sixtrack 1100 520 211* 1100 502 219*
301.apsi 2600 1053 247* 2600 1066 244*
Andreas
--
Andreas Jaeger
SuSE Labs aj@suse.de
private aj@arthur.inka.de
http://www.suse.de/~aj
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2001-09-09 23:48 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2001-09-08 23:49 Some SPECfp results for the simplify-rtx problem Andreas Jaeger
2001-09-09 1:28 ` Jan Hubicka
2001-09-09 23:48 ` Andreas Jaeger
2001-09-09 5:35 ` Andreas Jaeger
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).