See attached test-case, test.f: > gfortran -c test.f -O1 Compiles OK > gfortran -c test.f -O2 test.f: In function "test": test.f:13: internal compiler error: in emit_swap_insn, at reg-stack.c:827 > gfortran -v Using built-in specs. Target: i586-redhat-linux Configured with: ../configure --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --with-bugurl=http://bugzilla.redhat.com/bugzilla --enable-bootstrap --enable-shared --enable-threads=posix --enable-checking=release --with-system-zlib --enable-__cxa_atexit --disable-libunwind-exceptions --enable-languages=c,c++,objc,obj-c++,java,fortran,ada --enable-java-awt=gtk --disable-dssi --enable-plugin --with-java-home=/usr/lib/jvm/java-1.5.0-gcj-1.5.0.0/jre --enable-libgcj-multifile --enable-java-maintainer-mode --with-ecj-jar=/usr/share/java/eclipse-ecj.jar --disable-libjava-multilib --with-ppl --with-cloog --with-tune=generic --with-arch=i586 --build=i586-redhat-linux Thread model: posix gcc version 4.4.0 20090506 (Red Hat 4.4.0-4) (GCC) > uname -a Linux pmblakely 2.6.29.5-191.fc11.i686.PAE #1 SMP Tue Jun 16 23:19:53 EDT 2009 i686 i686 i386 GNU/Linux This looks similar to the test-case for bug 39607 (fixed), in that the ICE occurs when returned/class variables are only updated if a certain condition holds, and otherwise a separate function is called, resulting in those variables not being updated. Note that the test-case for bug 39607 compiles correctly (i.e. no ICE) for this version of gcc, so it's not a case of that patch not having gone through for this compiler version.
Created attachment 18091 [details] Minimal test-case test.f
Didn't GCC say report the bug to http://bugzilla.redhat.com/bugzilla ?
(In reply to comment #2) > Didn't GCC say report the bug to http://bugzilla.redhat.com/bugzilla ? > Oh, yes. Sorry. I'll post it there.
The test also fails with FSF 4.4.1 and 4.5.0, but not with 4.3.4, which makes for 4.4 regression.
Revision 139590 (IRA): http://gcc.gnu.org/ml/gcc-cvs/2008-08/msg01152.html is very likely the cause.
Revision 139590 is the cause.
Thanks for reporting this. I started to work on the PR.
The problem was in usage of df_get_live_out in ira.c::build_insn_chain instead of DF_LR_OUT. Later contains r58 (assigned to st0 register) and it creates restore insn for st0 after the call and prevents reg-stack crashing. The original IRA patch used DF_LR_OUT. After removing the old RA, accidentally IRA started to use df_get_live_out used by the old RA. I'll send a patch solving the problem soon.
Subject: Bug 40587 Author: vmakarov Date: Fri Jul 3 22:36:31 2009 New Revision: 149212 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=149212 Log: 2009-07-03 Vladimir Makarov <vmakarov@redhat.com> PR target/40587 * ira.c (build_insn_chain): Use DF_LR_OUT instead of df_get_live_out. * testsuite/gfortran.dg/pr40587.f: New test. Added: trunk/gcc/testsuite/gfortran.dg/pr40587.f Modified: trunk/gcc/ChangeLog trunk/gcc/ira.c
Subject: Bug 40587 Author: vmakarov Date: Fri Jul 3 22:46:30 2009 New Revision: 149213 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=149213 Log: 2009-07-03 Vladimir Makarov <vmakarov@redhat.com> PR target/40587 * ira.c (build_insn_chain): Use DF_LR_OUT instead of df_get_live_out. * testsuite/gfortran.dg/pr40587.f: New test. Added: branches/gcc-4_4-branch/gcc/testsuite/gfortran.dg/pr40587.f Modified: branches/gcc-4_4-branch/gcc/ChangeLog branches/gcc-4_4-branch/gcc/ira.c
Fixed.