This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [dataflow] [RFC] Remove many almost useless DCE passes
- From: Vladimir Makarov <vmakarov at redhat dot com>
- To: Steven Bosscher <stevenb dot gcc at gmail dot com>
- Cc: Paolo Bonzini <paolo dot bonzini at lu dot unisi dot ch>, GCC Patches <gcc-patches at gcc dot gnu dot org>, Kenneth Zadeck <zadeck at naturalbridge dot com>
- Date: Thu, 01 Mar 2007 14:40:39 -0500
- Subject: Re: [dataflow] [RFC] Remove many almost useless DCE passes
- References: <45E3D856.1030007@lu.unisi.ch> <571f6b510702262323l65ad866drb110392ca39c1643@mail.gmail.com> <45E3DD5B.4040206@lu.unisi.ch> <45E44626.2000402@redhat.com> <45E446CC.4030106@lu.unisi.ch> <45E44B63.8080609@redhat.com> <571f6b510702270742q318254beha843442d15d37c1@mail.gmail.com> <45E45391.4080705@redhat.com> <45E453C5.2000902@lu.unisi.ch> <45E455AD.5090109@redhat.com> <571f6b510702270808ic064174h7d4787ef888574ec@mail.gmail.com>
Steven Bosscher wrote:
On 2/27/07, Vladimir Makarov <vmakarov@redhat.com> wrote:
Good because SPECFP2000 for ppc64 is 0.5% better (SPECInt2000 is the
same) on the branch than on the mainline at the last merge point. And
this patch could remove this achievement.
Or it might not make a difference at all.
Which is actually more likely IMHO, but I'll let the numbers speak for
me, instead of doing some hand waiving and speculating.
You was right. The patch make the compiler a bit faster and a bit
better with the point of performance at least for ppc64. May it will be
interesting for you. Here is the data for ppc64 (2.5 Ghz PPC970FX).
Base: before the patch (-O2)
Peak: after the patch (-O2)
========================================================================
164.gzip 1400 190 739 * 1400 189 740 *
175.vpr 1400 308 454 * 1400 309 454 *
176.gcc X X
181.mcf 1800 535 337 * 1800 537 335 *
186.crafty 1000 86.6 1155 * 1000 87.2 1147 *
197.parser 1800 343 525 * 1800 342 527 *
252.eon 1300 113 1154 * 1300 115 1132 *
253.perlbmk 1800 243 739 * 1800 245 736 *
254.gap 1100 152 722 * 1100 152 723 *
255.vortex 1900 235 807 * 1900 221 858 *
256.bzip2 1500 253 593 * 1500 253 592 *
300.twolf X X
Est. SPECint_base2000 677
Est. SPECint2000 679
========================================================================
168.wupwise 1600 155 1032* 1600 157 1018*
171.swim 3100 1159 267* 3100 1159 267*
172.mgrid 1800 285 632* 1800 285 632*
173.applu 2100 299 703* 2100 296 711*
177.mesa 1400 142 989* 1400 141 996*
178.galgel 2900 690 421* 2900 689 421*
179.art 2600 396 657* 2600 397 655*
183.equake 1300 136 959* 1300 135 960*
187.facerec 1900 213 892* 1900 203 936*
188.ammp 2200 564 390* 2200 564 390*
189.lucas 2000 221 906* 2000 221 907*
191.fma3d 2100 231 909* 2100 232 904*
200.sixtrack 1100 363 303* 1100 358 307*
301.apsi 2600 448 581* 2600 446 583*
Est. SPECfp_base2000 632
Est. SPECfp2000 635
Compilation time
INT without twolf:
Base:
real 7m32.027s
user 6m40.094s
sys 0m29.926s
peak:
real 7m39.357s
user 6m39.559s
sys 0m29.911s
FP:
base:
real 6m52.848s
user 6m3.951s
sys 0m25.355s
peak:
real 6m44.545s
user 6m3.576s
sys 0m25.095s
Code size change/base/peak (text segment)
----------------CINT2000-----------------
0.000% 46542 46542 164.gzip
-0.049% 163913 163833 175.vpr
-0.007% 1504222 1504110 176.gcc
0.114% 14064 14080 181.mcf
0.016% 195381 195413 186.crafty
-0.063% 126698 126618 197.parser
0.002% 421115 421123 252.eon
0.003% 629279 629295 253.perlbmk
0.000% 579926 579926 254.gap
-0.027% 653406 653230 255.vortex
=0.000% 39422 39422 256.bzip2
Average = -0.000782929%
----------------CFP2000-----------------
0.111% 28723 28755 168.wupwise
0.150% 10678 10694 171.swim
=0.000% 16887 16887 172.mgrid
0.033% 49094 49110 173.applu
0.000% 579469 579469 177.mesa
0.028% 226460 226524 178.galgel
=0.000% 18296 18296 179.art
0.234% 20506 20554 183.equake
-0.024% 66766 66750 187.facerec
0.000% 132960 132960 188.ammp
0.000% 44718 44718 189.lucas
0.072% 996921 997641 191.fma3d
-0.320% 859844 857092 200.sixtrack
-0.023% 137088 137056 301.apsi
Average = 0.0153551%