[Bug rtl-optimization/80747] [6/7/8 Regression] gcc.dg/tree-ssa/tailrecursion-4.c fails with ICE when compiled with options "-fprofile-use -freorder-blocks-and-partition"
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80747 --- Comment #8 from Jakub Jelinek --- Author: jakub Date: Thu Dec 21 19:28:10 2017 New Revision: 255954 URL: https://gcc.gnu.org/viewcvs?rev=255954=gcc=rev Log: PR rtl-optimization/80747 PR rtl-optimization/83512 * cfgrtl.c (force_nonfallthru_and_redirect): When splitting succ edge from ENTRY, copy partition from e->dest to the newly created bb. * bb-reorder.c (reorder_basic_blocks_simple): If last_tail is ENTRY, use BB_PARTITION of its successor block as current_partition. Don't copy partition when splitting succ edge from ENTRY. * gcc.dg/pr80747.c: New test. * gcc.dg/pr83512.c: New test. Added: trunk/gcc/testsuite/gcc.dg/pr80747.c trunk/gcc/testsuite/gcc.dg/pr83512.c Modified: trunk/gcc/ChangeLog trunk/gcc/bb-reorder.c trunk/gcc/cfgrtl.c trunk/gcc/testsuite/ChangeLog
[Bug rtl-optimization/80747] [6/7/8 Regression] gcc.dg/tree-ssa/tailrecursion-4.c fails with ICE when compiled with options "-fprofile-use -freorder-blocks-and-partition"
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80747 Jakub Jelinek changed: What|Removed |Added CC||asolokha at gmx dot com --- Comment #7 from Jakub Jelinek --- *** Bug 83512 has been marked as a duplicate of this bug. ***
[Bug rtl-optimization/80747] [6/7/8 Regression] gcc.dg/tree-ssa/tailrecursion-4.c fails with ICE when compiled with options "-fprofile-use -freorder-blocks-and-partition"
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80747 Jakub Jelinek changed: What|Removed |Added Status|NEW |ASSIGNED Assignee|unassigned at gcc dot gnu.org |jakub at gcc dot gnu.org --- Comment #6 from Jakub Jelinek --- Created attachment 42948 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=42948=edit gcc8-pr80747.patch Untested fix. I actually think instead of having some magic default we should, in case we need to ignore the ENTRY succ edge in the greedy algorithm and thus last_tail remains ENTRY, use the partition of the ENTRY successor bb. And there is another bug that we don't copy the partition early enough and thus create EDGE_CROSSING and mark the jump as crossing, then we copy the partition, but nothing undoes the edge flags set to EDGE_CROSSING or CROSSING_JUMP_P on the jump.
[Bug rtl-optimization/80747] [6/7/8 Regression] gcc.dg/tree-ssa/tailrecursion-4.c fails with ICE when compiled with options "-fprofile-use -freorder-blocks-and-partition"
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80747 Jeffrey A. Law changed: What|Removed |Added CC||law at redhat dot com --- Comment #5 from Jeffrey A. Law --- Still a problem on the trunk. The attached patches doesn't seem to actually fix the problem.
[Bug rtl-optimization/80747] [6/7/8 Regression] gcc.dg/tree-ssa/tailrecursion-4.c fails with ICE when compiled with options "-fprofile-use -freorder-blocks-and-partition"
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80747 Richard Biener changed: What|Removed |Added Target|mips-sde-elf|mips-sde-elf, x86_64-*-* Priority|P3 |P2 Last reconfirmed|2017-05-15 00:00:00 |2017-7-25 CC||hubicka at gcc dot gnu.org --- Comment #4 from Richard Biener --- Honza? You were looking into hot/cold partition issues lately. Re-confirmed on trunk x86_64.
[Bug rtl-optimization/80747] [6/7/8 Regression] gcc.dg/tree-ssa/tailrecursion-4.c fails with ICE when compiled with options "-fprofile-use -freorder-blocks-and-partition"
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80747 Richard Biener changed: What|Removed |Added Target Milestone|--- |6.5
[Bug rtl-optimization/80747] [6/7/8 Regression] gcc.dg/tree-ssa/tailrecursion-4.c fails with ICE when compiled with options "-fprofile-use -freorder-blocks-and-partition"
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80747
--- Comment #3 from Segher Boessenkool ---
Hi Zhangwen,
That patch looks correct. The explanation maybe not: the problem happens
if the earliest "end" is the entry block itself, not a real BB (that is,
if we start the function with a newly inserted jump). We could detect
just that, and/or we can add some extra asserts; or do as in your patch,
that looks fine to me as well.
A few minor comments for when you submit the patch:
-- It is spelled "default", not "defualt";
-- We don't use "yoda conditions" in GCC ("0 == x");
-- You shouldn't use 0 but instead use BB_UNPARTITIONED.
Thanks for finding the problem and making a patch!
[Bug rtl-optimization/80747] [6/7/8 Regression] gcc.dg/tree-ssa/tailrecursion-4.c fails with ICE when compiled with options "-fprofile-use -freorder-blocks-and-partition"
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80747 Martin Liška changed: What|Removed |Added Status|UNCONFIRMED |NEW Last reconfirmed||2017-05-15 CC||marxin at gcc dot gnu.org, ||segher at gcc dot gnu.org Summary|gcc.dg/tree-ssa/tailrecursi |[6/7/8 Regression] |on-4.c fails with ICE when |gcc.dg/tree-ssa/tailrecursi |compiled with options |on-4.c fails with ICE when |"-fprofile-use |compiled with options |-freorder-blocks-and-partit |"-fprofile-use |ion"|-freorder-blocks-and-partit ||ion" Ever confirmed|0 |1 --- Comment #2 from Martin Liška --- Confirmed, started with r228318 and can be reproduced on x86_64-linux-gnu.
