[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 --- Comment #12 from paolo at gcc dot gnu.org paolo at gcc dot gnu.org 2011-08-18 11:30:50 UTC --- Author: paolo Date: Thu Aug 18 11:30:42 2011 New Revision: 177848 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=177848 Log: 2011-08-18 Paolo Carlini paolo.carl...@oracle.com Joseph Myers jos...@codesourcery.com PR tree-optimization/49963 * hwint.c (absu_hwi): Define. * hwint.h (absu_hwi): Declare. * fold-const.c (fold_plusminus_mult_expr): Use absu_hwi instead of abs_hwi. * tree-ssa-math-opts.c (gimple_expand_builtin_pow): Likewise. * tree-ssa-loop-prefetch.c (prune_ref_by_group_reuse): Likewise. Modified: trunk/gcc/ChangeLog trunk/gcc/fold-const.c trunk/gcc/hwint.c trunk/gcc/hwint.h trunk/gcc/tree-ssa-loop-prefetch.c trunk/gcc/tree-ssa-math-opts.c
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 Paolo Carlini paolo.carlini at oracle dot com changed: What|Removed |Added Status|NEW |RESOLVED Resolution||FIXED --- Comment #13 from Paolo Carlini paolo.carlini at oracle dot com 2011-08-18 11:33:54 UTC --- Fixed.
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 Paolo Carlini paolo.carlini at oracle dot com changed: What|Removed |Added CC||rguenther at suse dot de --- Comment #9 from Paolo Carlini paolo.carlini at oracle dot com 2011-08-17 07:32:46 UTC --- So, what do people think? Shall we go ahead with something like the patchlet I attached?
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 --- Comment #10 from rguenther at suse dot de rguenther at suse dot de 2011-08-17 11:25:02 UTC --- On Wed, 17 Aug 2011, paolo.carlini at oracle dot com wrote: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 Paolo Carlini paolo.carlini at oracle dot com changed: What|Removed |Added CC||rguenther at suse dot de --- Comment #9 from Paolo Carlini paolo.carlini at oracle dot com 2011-08-17 07:32:46 UTC --- So, what do people think? Shall we go ahead with something like the patchlet I attached? Sure, please post patches to gcc-patches@, bugzilla isn't the place to get attention to them
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 --- Comment #11 from Paolo Carlini paolo.carlini at oracle dot com 2011-08-17 17:22:16 UTC --- Ok, thanks, will do later today.
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 --- Comment #7 from Georg-Johann Lay gjl at gcc dot gnu.org 2011-08-10 09:45:12 UTC --- Is this related to PR42146, an ICE on array size of 0x8000?
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 John David Anglin danglin at gcc dot gnu.org changed: What|Removed |Added CC||danglin at gcc dot gnu.org --- Comment #8 from John David Anglin danglin at gcc dot gnu.org 2011-08-10 14:13:06 UTC --- Also occurs on hppa2.0w-hp-hpux11.11.
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 --- Comment #6 from Georg-Johann Lay gjl at gcc dot gnu.org 2011-08-08 08:28:46 UTC --- (In reply to comment #5) Created attachment 24909 [details] Draft patch I'm attaching a patch which tries to implement what Joseph suggests (thanks!). Essentially, I'm replacing all uses of abs_hwi outside hwint.c itself, besides three in tree-ssa-math-opts.c which seem correct anyway: the argument is an HOST_WIDE_INT / 2 or / 3 and the result is passed to a function taking a plain HOST_WIDE_INT. Type-wise the patch makes a lot of sense to me. Maybe submitter could test it? I'm sanity checking it on x86_64. That works for me, i.e. gcc.c-torture/execute/divconst-2.c passes with your patch.
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 Richard Guenther rguenth at gcc dot gnu.org changed: What|Removed |Added Status|UNCONFIRMED |NEW Last reconfirmed||2011.08.03 15:26:26 CC||paolo.carlini at oracle dot ||com, rth at gcc dot ||gnu.org, spop at gcc dot ||gnu.org Target Milestone|--- |4.7.0 Summary|ICE: in abs_hwi, at |[4.7 Regression] ICE: in |hwint.c:108 |abs_hwi, at hwint.c:108 Ever Confirmed|0 |1 --- Comment #1 from Richard Guenther rguenth at gcc dot gnu.org 2011-08-03 15:26:26 UTC --- Well - as I said. Here we go. Please fix :P
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 --- Comment #2 from Richard Guenther rguenth at gcc dot gnu.org 2011-08-03 15:27:36 UTC --- Probably fails on any 32bit HWI platform.
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 --- Comment #3 from Paolo Carlini paolo.carlini at oracle dot com 2011-08-03 15:45:06 UTC --- Eh, my way of fixing it would be removing the assert ;) Seriously, too bad, I can try to look a bit into it but help is welcome of course, I don't think the project wants to rely on my judgment about this vs 49914...
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 --- Comment #4 from joseph at codesourcery dot com joseph at codesourcery dot com 2011-08-03 18:59:33 UTC --- I think this is a case for a function absu_hwi or similar that returns an unsigned HOST_WIDE_INT value. (Actually it's a case for operating properly on the INTEGER_CSTs or on double_ints without being conditional on whether the values pass host_integerp, but that's a more complicated fix.)
[Bug tree-optimization/49963] [4.7 Regression] ICE: in abs_hwi, at hwint.c:108
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49963 --- Comment #5 from Paolo Carlini paolo.carlini at oracle dot com 2011-08-03 20:29:51 UTC --- Created attachment 24909 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=24909 Draft patch I'm attaching a patch which tries to implement what Joseph suggests (thanks!). Essentially, I'm replacing all uses of abs_hwi outside hwint.c itself, besides three in tree-ssa-math-opts.c which seem correct anyway: the argument is an HOST_WIDE_INT / 2 or / 3 and the result is passed to a function taking a plain HOST_WIDE_INT. Type-wise the patch makes a lot of sense to me. Maybe submitter could test it? I'm sanity checking it on x86_64.
