public inbox for gcc-bugs@sourceware.org help / color / mirror / Atom feed
* [Bug tree-optimization/39141] New: overzealous unrolling (peeling) destroys code locality @ 2009-02-09 16:18 amylaar at gcc dot gnu dot org 2009-03-04 19:09 ` [Bug tree-optimization/39141] " amylaar at gcc dot gnu dot org ` (2 more replies) 0 siblings, 3 replies; 4+ messages in thread From: amylaar at gcc dot gnu dot org @ 2009-02-09 16:18 UTC (permalink / raw) To: gcc-bugs I see a 50% cycle increase for EEMBC idctrn01 going from gcc 4.2.1 to gcc 4.4 . There are two issues, overzealous unrolling, and constant propagation in the unrolled loops. While both issues can be avoided by reducing the parameter "max-completely-peeled-insns" to 200, this is not really satisfactory, since that is a rather fragile parameter setting and is not really related to the problem, and it holds other code back, like EEMBC viterb00 benchmark, which looses 5% performance with that setting. There are a number of loops which are completely unrolled, thus pushing their containing loop size above the size of the instruction cache. There is no point in doing such unrolling, since it is more expensive to fill the instruction cache with an unrolled loop than to execute a rolled loop. I have implemented a heuristic to estimate the size of the outer loop (or function in absence of an outer loop) assuming that its inner loops will be unrolled in accordance with PARAM_MAX_UNROLL_TIMES, PARAM_MAX_UNROLLED_INSNS and PARAM_MAX_COMPLETELY_PEEL_TIMES, and if that size exceeds a threshold (a new parameter), complete unrolling is inhibited. This change has reduced the idctrn01 regression to 13.7% while leaving the other EEMBC benchmarks alone. I expect that I can address the remaining performance regression by inhibiting inappropriate constant propagation - there are 798 addresses of the form absolute address+offset in the main benchmark assembly file, each of which translates into one instruction word too many, in total 20% of the text size of that module. I can post the unroller heuristic patch as soon as we have the confirmation from the FSF that they have filed the Copyright Assignment we gave them last year. -- Summary: overzealous unrolling (peeling) destroys code locality Product: gcc Version: 4.4.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: tree-optimization AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: amylaar at gcc dot gnu dot org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39141 ^ permalink raw reply [flat|nested] 4+ messages in thread
* [Bug tree-optimization/39141] overzealous unrolling (peeling) destroys code locality 2009-02-09 16:18 [Bug tree-optimization/39141] New: overzealous unrolling (peeling) destroys code locality amylaar at gcc dot gnu dot org @ 2009-03-04 19:09 ` amylaar at gcc dot gnu dot org 2009-03-05 0:34 ` amylaar at gcc dot gnu dot org 2010-01-27 20:04 ` amylaar at gcc dot gnu dot org 2 siblings, 0 replies; 4+ messages in thread From: amylaar at gcc dot gnu dot org @ 2009-03-04 19:09 UTC (permalink / raw) To: gcc-bugs ------- Comment #1 from amylaar at gcc dot gnu dot org 2009-03-04 19:08 ------- patch submitted -- amylaar at gcc dot gnu dot org changed: What |Removed |Added ---------------------------------------------------------------------------- BugsThisDependsOn|39302 | AssignedTo|unassigned at gcc dot gnu |amylaar at gcc dot gnu dot |dot org |org Status|UNCONFIRMED |ASSIGNED Ever Confirmed|0 |1 Keywords| |patch Last reconfirmed|0000-00-00 00:00:00 |2009-03-04 19:08:52 date| | http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39141 ^ permalink raw reply [flat|nested] 4+ messages in thread
* [Bug tree-optimization/39141] overzealous unrolling (peeling) destroys code locality 2009-02-09 16:18 [Bug tree-optimization/39141] New: overzealous unrolling (peeling) destroys code locality amylaar at gcc dot gnu dot org 2009-03-04 19:09 ` [Bug tree-optimization/39141] " amylaar at gcc dot gnu dot org @ 2009-03-05 0:34 ` amylaar at gcc dot gnu dot org 2010-01-27 20:04 ` amylaar at gcc dot gnu dot org 2 siblings, 0 replies; 4+ messages in thread From: amylaar at gcc dot gnu dot org @ 2009-03-05 0:34 UTC (permalink / raw) To: gcc-bugs ------- Comment #2 from amylaar at gcc dot gnu dot org 2009-03-05 00:33 ------- The patch is here: http://gcc.gnu.org/ml/gcc-patches/2009-03/msg00231.html -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39141 ^ permalink raw reply [flat|nested] 4+ messages in thread
* [Bug tree-optimization/39141] overzealous unrolling (peeling) destroys code locality 2009-02-09 16:18 [Bug tree-optimization/39141] New: overzealous unrolling (peeling) destroys code locality amylaar at gcc dot gnu dot org 2009-03-04 19:09 ` [Bug tree-optimization/39141] " amylaar at gcc dot gnu dot org 2009-03-05 0:34 ` amylaar at gcc dot gnu dot org @ 2010-01-27 20:04 ` amylaar at gcc dot gnu dot org 2 siblings, 0 replies; 4+ messages in thread From: amylaar at gcc dot gnu dot org @ 2010-01-27 20:04 UTC (permalink / raw) To: gcc-bugs ------- Comment #3 from amylaar at gcc dot gnu dot org 2010-01-27 20:04 ------- Subject: Bug 39141 Author: amylaar Date: Wed Jan 27 20:03:57 2010 New Revision: 156301 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=156301 Log: PR tree-optimization/39141 * tree-ssa-loop-manip.c (gimple_can_duplicate_loop_to_header_edge): New function. * tree-ssa-loop-ivcanon.c (enum unroll_level): New value UL_ESTIMATE_GROWTH. (try_unroll_loop_completely): Handle UL_ESTIMATE_GROWTH. (canonicalize_loop_induction_variables): Likewise. (tree_unroll_loops_completely): Don't completely unroll loops where the outer loop/function is larger than PARAM_MAX_COMPLETELY_PEELED_OUTER_INSNS, or will/would become thus due to unrolling. * cfgloop.h (enum li_flags): New value LI_REALLY_FROM_INNERMOST. (fel_init): Handle LI_REALLY_FROM_INNERMOST. * tree-flow.h (gimple_can_duplicate_loop_to_header_edge): Declare. * params.def (PARAM_MAX_COMPLETELY_PEELED_OUTER_INSNS): New parameter. Added: branches/mpost-opt-imp-20100127/gcc/ChangeLog.mpost Modified: branches/mpost-opt-imp-20100127/gcc/cfgloop.h branches/mpost-opt-imp-20100127/gcc/params.def branches/mpost-opt-imp-20100127/gcc/tree-flow.h branches/mpost-opt-imp-20100127/gcc/tree-ssa-loop-ivcanon.c branches/mpost-opt-imp-20100127/gcc/tree-ssa-loop-manip.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39141 ^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2010-01-27 20:04 UTC | newest] Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2009-02-09 16:18 [Bug tree-optimization/39141] New: overzealous unrolling (peeling) destroys code locality amylaar at gcc dot gnu dot org 2009-03-04 19:09 ` [Bug tree-optimization/39141] " amylaar at gcc dot gnu dot org 2009-03-05 0:34 ` amylaar at gcc dot gnu dot org 2010-01-27 20:04 ` amylaar at gcc dot gnu dot org
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).