[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 --- Comment #13 from GCC Commits --- The master branch has been updated by Richard Biener : https://gcc.gnu.org/g:d782ec8362eadc3169286eb1e39c631effd02323 commit r14-6550-gd782ec8362eadc3169286eb1e39c631effd02323 Author: Richard Biener Date: Wed Dec 13 14:23:31 2023 +0100 tree-optimization/112793 - SLP of constant/external code-generated twice The following makes the attempt at code-generating a constant/external SLP node twice well-formed as that can happen when partitioning BB vectorization attempts where we keep constants/externals unpartitioned. PR tree-optimization/112793 * tree-vect-slp.cc (vect_schedule_slp_node): Already code-generated constant/external nodes are OK. * g++.dg/vect/pr112793.cc: New testcase.
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 Richard Biener changed: What|Removed |Added Status|NEW |ASSIGNED Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot gnu.org Target Milestone|--- |14.0 --- Comment #12 from Richard Biener --- Confirmed, mine. We're scheduling the same SLP node twice from different SLP instances. We're supposed to merge SLP instances which share nodes though, but we're only doing that for internal defs. We need to either duplicate the constant/external nodes in different subgraphs or merge subgraphs with common constants/externals as well. I'll think about it. This isn't new in GCC 14, it should be a problem since GCC 11.
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 --- Comment #11 from Sam James --- Created attachment 56747 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56747=edit another cvised testcase
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 --- Comment #10 from Andrew Pinski --- Created attachment 56746 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56746=edit Delta reduced testcase This one is slightly less reduced than the other one but this is a delta based reduced testcase rather than a cvise based one. I didn't manually reduce it further though.
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 Andrew Pinski changed: What|Removed |Added Last reconfirmed||2023-12-01 Ever confirmed|0 |1 Status|UNCONFIRMED |NEW --- Comment #9 from Andrew Pinski --- .
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 Andrew Pinski changed: What|Removed |Added Attachment #56744|0 |1 is obsolete|| --- Comment #8 from Andrew Pinski --- Created attachment 56745 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56745=edit testcase without any undefinedness to it Further reduced and without any undefined/uninitialized variables in it (as far as I can tell): Note I added a `#if 0` section which produces almost the same IR (except ssa names are different) from the other but does not produce an ICE. I don't understand why though.
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 --- Comment #7 from Andrew Pinski --- e.g. r[8 * 2] is r[16] and 16 >= 16 :).
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 --- Comment #6 from Andrew Pinski --- (In reply to Sam James from comment #5) > Created attachment 56744 [details] > reduced.ii Note this reduced testcase has undefined code in it ...
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 --- Comment #5 from Sam James --- Created attachment 56744 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56744=edit reduced.ii
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 --- Comment #4 from Andrew Pinski --- With s/__builtin_ctzs/__builtin_ctzg/ and `-O3 -march=skylake-avx512`, I can reproduce it with an unmodified trunk GCC.
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 --- Comment #3 from Sam James --- (In reply to Andrew Pinski from comment #2) > Does gentoo has a patch which adds __builtin_ctzs ? No, nothing like that. Here's what this one was built with: https://gitweb.gentoo.org/proj/gcc-patches.git/tree/14.0.0/gentoo
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 --- Comment #2 from Andrew Pinski --- Does gentoo has a patch which adds __builtin_ctzs ?
[Bug tree-optimization/112793] [14 regression] ICE when building stellarium (internal compiler error: in vect_schedule_slp_node, at tree-vect-slp.cc:9062)
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112793 --- Comment #1 from Andrew Pinski --- Hmm __builtin_ctzs