[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 Andrew Pinski changed: What|Removed |Added Keywords||wrong-code Status|NEW |RESOLVED Resolution|--- |FIXED --- Comment #10 from Andrew Pinski --- Fixed.
[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 --- Comment #9 from Jan Hubicka --- Author: hubicka Date: Wed Dec 16 04:58:13 2015 New Revision: 231671 URL: https://gcc.gnu.org/viewcvs?rev=231671=gcc=rev Log: PR lto/68878 * lto-symtab.c (lto_symtab_prevailing_virtual_decl): New function. * lto-symtab.h (lto_symtab_prevailing_virtual_decl): Declare. (lto_symtab_prevailing_decl): Use it. Modified: trunk/gcc/lto/ChangeLog trunk/gcc/lto/lto-symtab.c trunk/gcc/lto/lto-symtab.h
[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 Richard Biener changed: What|Removed |Added Target Milestone|--- |6.0
[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 --- Comment #8 from Jan Hubicka --- OK, the bug is caused by fact that new decl merging won't merge decls without symbol table entries. Devirtualization uses them and it means we lose a ctor. I have fix for the merging, but it seems to be a latent bug in ipa-cp or ipa-devirt I need to fix first - it should only cause a drop in number of devirtualizations, not a wrong code issue.
[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 Jan Hubicka changed: What|Removed |Added CC||hubicka at gcc dot gnu.org --- Comment #1 from Jan Hubicka --- HJ, can you please check if the earlier revision is good? I do not have omnetpp sources from spec2k6 and wonder how this particular revision can lead to a wrong code... It only slightly changes the way we merge builtin decls that should be a no-op because we already merge them at the streaming time. You can build with -fdump-ipa-cgraph and look if there are any transparent aliases introduced. (just grep the dumps for transparent_alias)
[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 Mikhail Maltsev changed: What|Removed |Added Status|UNCONFIRMED |NEW Last reconfirmed||2015-12-13 CC||miyuki at gcc dot gnu.org Ever confirmed|0 |1 --- Comment #2 from Mikhail Maltsev --- Confirmed. I checked r231437, and it seems to be OK. "transparent_alias" is not present in dumps (neither in bad revision, nor in good one). Here are some dumps, which, I think, show the problem.
[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 --- Comment #3 from Mikhail Maltsev --- Created attachment 37013 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37013=edit cSimpleModule ctor before transformation - good revision
[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 --- Comment #4 from Mikhail Maltsev --- Created attachment 37014 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37014=edit cSimpleModule ctor after transformation - good revision
[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 --- Comment #5 from Mikhail Maltsev --- Created attachment 37015 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37015=edit cSimpleModule ctor before transformation - bad revision Almost identical to "good revision"
[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 --- Comment #6 from Mikhail Maltsev --- Created attachment 37016 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37016=edit cSimpleModule ctor after transformation - bad revision Note that lots of code is removed and we fall through BB4 to an EH landing pad (hence, _Unwind_Resume in the backtrace).
[Bug middle-end/68878] [6 Regression] 471.omnetpp in SPEC CPU 2006 is miscompiled with LTO
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68878 --- Comment #7 from Jan Hubicka --- Looks like a result of redirection to builtin_unreachable. Can you check the WPA dumps if there is difference in redirections, please? Honza