[Bug target/27855] [4.5/4.7/4.8 regression] reassociation causes the RA to be confused
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27855 --- Comment #46 from Uros Bizjak ubizjak at gmail dot com 2013-01-17 16:36:30 UTC --- (In reply to comment #45) Is there any improvement if you use -fschedule-insns1 -fsched-pressure? Yes, please see the table bellow: ALGORITHM NB REPSTIME MFLOPS = = = == == -DREPS=1 -msse3 -O2 -ffast-math atlasmm 60 1 0.899 4806.07 -msse3 -O2 -ffast-math -fschedule-insns -fsched-pressure atlasmm 60 1 0.697 6198.93 -DREPS=1 -msse3 -O2 -ffast-math -fno-tree-reassoc atlasmm 60 1 0.672 6429.56 -DREPS=1 -msse3 -O2 -ffast-math -fno-tree-reassoc -fschedule-insns -fsched-pressure atlasmm 60 1 0.683 6326.00 -DREPS=1 -msse3 -O2 -ffast-math -ftree-vectorize atlasmm 60 1 0.898 4811.42 -DREPS=1 -msse3 -O2 -ffast-math -ftree-vectorize -fno-tree-reassoc atlasmm 60 1 0.919 4701.47 -DREPS=1 -msse3 -O2 -ffast-math -ftree-vectorize -fno-tree-reassoc -fschedule-insns -fsched-pressure atlasmm 60 1 0.911 4742.77
[Bug target/27855] [4.5/4.7/4.8 regression] reassociation causes the RA to be confused
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27855 Uros Bizjak ubizjak at gmail dot com changed: What|Removed |Added Target|i686-pc-linux-gnu |i686-pc-linux-gnu, ||x86_64-pc-linux-gnu Status|WAITING |NEW --- Comment #44 from Uros Bizjak ubizjak at gmail dot com 2013-01-16 16:40:36 UTC --- Still the same old story... Target: x86_64-unknown-linux-gnu gcc version 4.8.0 20130116 (experimental) [trunk revision 195240] (GCC) corei7 SNB: -DREPS=1 -msse3 -O2 -ffast-math ALGORITHM NB REPSTIME MFLOPS = = = == == atlasmm 60 1 1.636 2640.99 -DREPS=1 -msse3 -O2 -ffast-math -fno-tree-reassoc ALGORITHM NB REPSTIME MFLOPS = = = == == atlasmm 60 1 1.218 3547.34 -DREPS=1 -msse3 -O2 -ffast-math -ftree-vectorize ALGORITHM NB REPSTIME MFLOPS = = = == == atlasmm 60 1 1.654 2612.25 -DREPS=1 -msse3 -O2 -ffast-math -ftree-vectorize -fno-tree-reassoc ALGORITHM NB REPSTIME MFLOPS = = = == == atlasmm 60 1 1.555 2778.56 The testcase is at http://www.cs.utsa.edu/~whaley/mmbench4.tar.gz Reconfirmed with 4.8.0.
[Bug target/27855] [4.5/4.7/4.8 regression] reassociation causes the RA to be confused
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27855 --- Comment #45 from Steven Bosscher steven at gcc dot gnu.org 2013-01-16 23:04:44 UTC --- (In reply to comment #44) Is there any improvement if you use -fschedule-insns1 -fsched-pressure?
[Bug target/27855] [4.5/4.7/4.8 regression] reassociation causes the RA to be confused
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27855 Richard Biener rguenth at gcc dot gnu.org changed: What|Removed |Added Status|NEW |WAITING
[Bug target/27855] [4.5/4.7/4.8 regression] reassociation causes the RA to be confused
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27855 --- Comment #43 from Andrew Pinski pinskia at gcc dot gnu.org 2013-01-01 05:25:15 UTC --- Bug 50557 looks like the same issue or at least a related one and it is mentioned that LRA fixes the register allocation issue. So yes it would be good to get new numbers for the trunk.
[Bug target/27855] [4.5/4.7/4.8 regression] reassociation causes the RA to be confused
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27855 --- Comment #42 from Steven Bosscher steven at gcc dot gnu.org 2012-11-09 23:06:18 UTC --- It'd be interesting to get some new timings, with the new register-pressure aware scheduling and with LRA. Oh, and maybe with the patch to restrain the reassoc pass a bit (http://gcc.gnu.org/ml/gcc-patches/2012-10/msg02122.html).
[Bug target/27855] [4.5/4.7/4.8 regression] reassociation causes the RA to be confused
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27855 Jakub Jelinek jakub at gcc dot gnu.org changed: What|Removed |Added Target Milestone|4.7.2 |4.7.3 --- Comment #41 from Jakub Jelinek jakub at gcc dot gnu.org 2012-09-20 10:15:15 UTC --- GCC 4.7.2 has been released.
[Bug target/27855] [4.5/4.7/4.8 regression] reassociation causes the RA to be confused
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27855 Richard Guenther rguenth at gcc dot gnu.org changed: What|Removed |Added Target Milestone|4.5.4 |4.6.4 --- Comment #39 from Richard Guenther rguenth at gcc dot gnu.org 2012-07-02 11:17:03 UTC --- The 4.5 branch is being closed, adjusting target milestone.
[Bug target/27855] [4.5/4.7/4.8 regression] reassociation causes the RA to be confused
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27855 Paolo Bonzini bonzini at gnu dot org changed: What|Removed |Added Target Milestone|4.6.4 |4.7.2 --- Comment #40 from Paolo Bonzini bonzini at gnu dot org 2012-07-02 12:24:08 UTC --- Would be interesting to look at it again in 4.8 with Richard's new pressure-aware scheduling. I think there was some other patch that could be used to enable -fschedule-insns on register-starved machines, so it would be worthwhile doing a round of SPEC benchmarking with different options, on both x86-32 and x86-64...
[Bug target/27855] [4.5/4.7/4.8 regression] reassociation causes the RA to be confused
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27855 Jakub Jelinek jakub at gcc dot gnu.org changed: What|Removed |Added Target Milestone|4.4.7 |4.5.4 --- Comment #38 from Jakub Jelinek jakub at gcc dot gnu.org 2012-03-13 12:47:17 UTC --- 4.4 branch is being closed, moving to 4.5.4 target.