[Bug target/59803] [4.8 Regression] s390x -march=z10 reload ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59803 --- Comment #4 from Jakub Jelinek jakub at gcc dot gnu.org --- Author: jakub Date: Wed Jan 15 09:39:11 2014 New Revision: 206625 URL: http://gcc.gnu.org/viewcvs?rev=206625root=gccview=rev Log: Add PR target/59803 reference to ChangeLog entries. Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog
[Bug target/59803] [4.8 Regression] s390x -march=z10 reload ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59803 --- Comment #5 from Jakub Jelinek jakub at gcc dot gnu.org --- Author: jakub Date: Wed Jan 15 09:40:25 2014 New Revision: 206626 URL: http://gcc.gnu.org/viewcvs?rev=206626root=gccview=rev Log: Add PR target/59803 to ChangeLog entries. Modified: branches/gcc-4_8-branch/gcc/ChangeLog branches/gcc-4_8-branch/gcc/testsuite/ChangeLog
[Bug target/59803] [4.8 Regression] s390x -march=z10 reload ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59803 Jakub Jelinek jakub at gcc dot gnu.org changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution|--- |FIXED --- Comment #6 from Jakub Jelinek jakub at gcc dot gnu.org --- Author: krebbel Date: Wed Jan 15 08:36:44 2014 New Revision: 206623 URL: http://gcc.gnu.org/viewcvs?rev=206623root=gccview=rev Log: 2014-01-15 Andreas Krebbel andreas.kreb...@de.ibm.com * config/s390/s390.c (s390_preferred_reload_class): Don't return ADDR_REGS for invalid symrefs in non-PIC code. 2014-01-15 Andreas Krebbel andreas.kreb...@de.ibm.com * gcc.c-torture/compile/pr59803.c: New testcase. Added: trunk/gcc/testsuite/gcc.c-torture/compile/pr59803.c Modified: trunk/gcc/ChangeLog trunk/gcc/config/s390/s390.c trunk/gcc/testsuite/ChangeLog Author: krebbel Date: Wed Jan 15 08:34:46 2014 New Revision: 206622 URL: http://gcc.gnu.org/viewcvs?rev=206622root=gccview=rev Log: 2014-01-15 Andreas Krebbel andreas.kreb...@de.ibm.com * config/s390/s390.c (s390_preferred_reload_class): Don't return ADDR_REGS for invalid symrefs in non-PIC code. 2014-01-15 Andreas Krebbel andreas.kreb...@de.ibm.com * gcc.c-torture/compile/pr59803.c: New testcase. Added: branches/gcc-4_8-branch/gcc/testsuite/gcc.c-torture/compile/pr59803.c Modified: branches/gcc-4_8-branch/gcc/ChangeLog branches/gcc-4_8-branch/gcc/config/s390/s390.c branches/gcc-4_8-branch/gcc/testsuite/ChangeLog
[Bug target/59803] [4.8 Regression] s390x -march=z10 reload ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59803 Jakub Jelinek jakub at gcc dot gnu.org changed: What|Removed |Added CC||krebbel at gcc dot gnu.org Target Milestone|--- |4.8.3
[Bug target/59803] [4.8 Regression] s390x -march=z10 reload ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59803 Andreas Krebbel krebbel at gcc dot gnu.org changed: What|Removed |Added Status|UNCONFIRMED |ASSIGNED Last reconfirmed||2014-01-14 Assignee|unassigned at gcc dot gnu.org |krebbel at gcc dot gnu.org Ever confirmed|0 |1 --- Comment #1 from Andreas Krebbel krebbel at gcc dot gnu.org --- There is a secondary reload which is supposed to handle that case. I'll try to figure out what went wrong here. Due to LRA being enabled by default on S/390 the situation on trunk is probably not comparable.
[Bug target/59803] [4.8 Regression] s390x -march=z10 reload ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59803 --- Comment #2 from Andreas Krebbel krebbel at gcc dot gnu.org --- Created attachment 31832 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=31832action=edit Experimental fix This patch fixes the problem. I'll post/commit it when it passed regtesting.
[Bug target/59803] [4.8 Regression] s390x -march=z10 reload ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59803 --- Comment #3 from Jakub Jelinek jakub at gcc dot gnu.org --- Thanks, looks good to me in my testing (s390-linux and s390x-linux --enable-checking=release 4.8 branch --with-arch=z196 --with-tune=zEC12 bootstraps/regtests), and it even fixed one FAIL in the testsuite: -FAIL: gcc.dg/torture/vshuf-v32qi.c -O2 (internal compiler error) -FAIL: gcc.dg/torture/vshuf-v32qi.c -O2 (test for excess errors) -UNRESOLVED: gcc.dg/torture/vshuf-v32qi.c -O2 compilation failed to produce executable which apparently ICEd the same way with e.g. -O2 -march=z10 or -O2 -march=z196 without the patch and doesn't with the patch. Though I'd say the #c0 testcase doesn't hurt, can you please add it to say gcc.c-torture/compile/ (or gcc.dg/torture/)?