[Bug target/44199] ppc64 glibc miscompilation
--- Comment #26 from bergner at gcc dot gnu dot org 2010-06-02 15:40 --- Subject: Bug 44199 Author: bergner Date: Wed Jun 2 15:40:09 2010 New Revision: 160160 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=160160 Log: Backport from GCC 4.4: 2010-05-26 Jakub Jelinek ja...@redhat.com PR target/44199 * config/rs6000/rs6000.c (rs6000_emit_epilogue): Fix up a backport glitch. Modified: branches/ibm/gcc-4_4-branch/gcc/ChangeLog.ibm branches/ibm/gcc-4_4-branch/gcc/config/rs6000/rs6000.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #25 from bergner at gcc dot gnu dot org 2010-05-27 16:31 --- Subject: Bug 44199 Author: bergner Date: Thu May 27 16:31:05 2010 New Revision: 159930 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=159930 Log: Backport from mainline: 2010-05-26 Jakub Jelinek ja...@redhat.com PR target/44199 * config/rs6000/rs6000.c (rs6000_emit_epilogue): If cfun-calls_alloca or total_size is larger than red zone size for non-V4 ABI, emit a stack_tie resp. frame_tie insn before stack pointer restore. * config/rs6000/rs6000.md (frame_tie): New insn. Modified: branches/ibm/gcc-4_4-branch/gcc/ChangeLog.ibm branches/ibm/gcc-4_4-branch/gcc/config/rs6000/rs6000.c branches/ibm/gcc-4_4-branch/gcc/config/rs6000/rs6000.md -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #18 from jakub at gcc dot gnu dot org 2010-05-26 06:01 --- Subject: Bug 44199 Author: jakub Date: Wed May 26 06:00:44 2010 New Revision: 159853 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=159853 Log: PR target/44199 * config/rs6000/rs6000.c (rs6000_emit_epilogue): If cfun-calls_alloca or total_size is larger than red zone size for non-V4 ABI, emit a stack_tie resp. frame_tie insn before stack pointer restore. * config/rs6000/rs6000.md (frame_tie): New insn. Modified: trunk/gcc/ChangeLog trunk/gcc/config/rs6000/rs6000.c trunk/gcc/config/rs6000/rs6000.md -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #19 from jakub at gcc dot gnu dot org 2010-05-26 06:02 --- Subject: Bug 44199 Author: jakub Date: Wed May 26 06:02:30 2010 New Revision: 159854 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=159854 Log: PR target/44199 * config/rs6000/rs6000.c (rs6000_emit_epilogue): If cfun-calls_alloca or total_size is larger than red zone size for non-V4 ABI, emit a stack_tie resp. frame_tie insn before stack pointer restore. * config/rs6000/rs6000.md (frame_tie): New insn. Modified: branches/gcc-4_5-branch/gcc/ChangeLog branches/gcc-4_5-branch/gcc/config/rs6000/rs6000.c branches/gcc-4_5-branch/gcc/config/rs6000/rs6000.md -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #20 from jakub at gcc dot gnu dot org 2010-05-26 06:05 --- Subject: Bug 44199 Author: jakub Date: Wed May 26 06:05:29 2010 New Revision: 159855 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=159855 Log: PR target/44199 * config/rs6000/rs6000.c (rs6000_emit_epilogue): If cfun-calls_alloca or total_size is larger than red zone size for non-V4 ABI, emit a stack_tie resp. frame_tie insn before stack pointer restore. * config/rs6000/rs6000.md (frame_tie): New insn. Modified: branches/gcc-4_4-branch/gcc/ChangeLog branches/gcc-4_4-branch/gcc/config/rs6000/rs6000.c branches/gcc-4_4-branch/gcc/config/rs6000/rs6000.md -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #21 from jakub at gcc dot gnu dot org 2010-05-26 06:07 --- Should be fixed now. -- jakub at gcc dot gnu dot org changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED Resolution||FIXED http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #22 from pthaugen at gcc dot gnu dot org 2010-05-26 15:51 --- The 4.4 patch isn't complete. /home/gccbuild/gcc_4.4_anonsvn/gcc/gcc/config/rs6000/rs6000.c:17166: undefined reference to `offset_below_red_zone_p' /home/gccbuild/gcc_4.4_anonsvn/gcc/gcc/config/rs6000/rs6000.c:17188: undefined reference to `offset_below_red_zone_p' -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #23 from jakub at gcc dot gnu dot org 2010-05-26 16:09 --- Subject: Bug 44199 Author: jakub Date: Wed May 26 16:09:25 2010 New Revision: 159878 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=159878 Log: PR target/44199 * config/rs6000/rs6000.c (rs6000_emit_epilogue): Fix up a backport glitch. Modified: branches/gcc-4_4-branch/gcc/config/rs6000/rs6000.c -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #24 from jakub at gcc dot gnu dot org 2010-05-26 16:10 --- Oops sorry, forgot redhat/gcc-4_4-branch has this function backported. Fixed now. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #17 from dominiq at lps dot ens dot fr 2010-05-22 13:31 --- is : http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44229 potentially a similar problem? It does not look like: the patch in comment #6 does not fix pr44229. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #14 from jakub at gcc dot gnu dot org 2010-05-21 08:08 --- For me it bootstrapped/regtested on 4.4 branch without any testsuite changes (both 32-bit and 64-bit). To see the unsupported tests difference, you can grep ^UNSUPPORTED gcc/testsuite/gcc/gcc.log | sort between the unpatched/patched builds and see what the differences are I guess. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #15 from bergner at gcc dot gnu dot org 2010-05-21 19:14 --- I also did a powerpc64-linux bootstrap and regtest (both 32-bit and 64-bit) and I didn't see any new failures and I also did not see any extra UNSUPPORTED tests. The only time UNSUPPORTED showed up in the test_summary output (UNRESOLVED: one_time_plugin.c compilation,...) was due to different source directory paths. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #16 from iains at gcc dot gnu dot org 2010-05-21 19:24 --- is : http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44229 potentially a similar problem? -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #9 from pthaugen at gcc dot gnu dot org 2010-05-20 16:23 --- Spec testing went fine, differences in the noise range. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #10 from jakub at gcc dot gnu dot org 2010-05-20 16:31 --- Have you also bootstrapped/regtested the patch? I can do so (easily for 4.4 branch, with more effort for the trunk), but haven't done that yet. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #11 from pthaugen at gcc dot gnu dot org 2010-05-20 17:59 --- No I didn't bootstrap/regtest. I can take care of trunk if you want to do 4.4. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #12 from jakub at gcc dot gnu dot org 2010-05-20 18:23 --- If you could, it would be very much appreciated. Starting 4.4 bootstrap/regtest now. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #13 from pthaugen at gcc dot gnu dot org 2010-05-21 02:32 --- Bootstrap of trunk went fine. Regression test (contrib/test_summary) showed the following difference between base/patched versions, but didn't have any testcases show up in the diff, so not sure what to make of that. This is for 32-bit gcc testsuite. # of expected passes 59078 --- # of expected passes 59075 98c98 # of unsupported tests872 --- # of unsupported tests875 -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #1 from pinskia at gcc dot gnu dot org 2010-05-19 18:30 --- Looks related to PR 30282. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #2 from jakub at gcc dot gnu dot org 2010-05-19 19:09 --- *** Bug 44200 has been marked as a duplicate of this bug. *** -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #3 from jakub at gcc dot gnu dot org 2010-05-19 19:09 --- *** Bug 44201 has been marked as a duplicate of this bug. *** -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #4 from jakub at gcc dot gnu dot org 2010-05-19 19:23 ---
Yes, it is related, but solving it in the scheduler in generic way isn't going
to be trivial.
E.g. x86_64 emits memory_blockage early in ix86_expand_epilogue:
/* See the comment about red zone and frame
pointer usage in ix86_expand_prologue. */
if (frame_pointer_needed frame.red_zone_size)
emit_insn (gen_memory_blockage ());
Another testcase:
extern void *alloca (__SIZE_TYPE__);
__attribute__((noinline, noclone))
long *bar (long *p)
{
asm volatile ( : : r (p) : memory);
return p;
}
long
foo (long x)
{
long *p = (long *) alloca (x * sizeof (long));
long *q = bar (p);
return q[0];
}
which shows that some blockage or preventing scheduling over the stack release
is needed even when the frame size is smaller than the red zone size and that
the memory address doesn't need to be obviously based on stack pointer (it
would be just safe to allow moving over stack accesses that are provably in the
red zone or above.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #5 from jakub at gcc dot gnu dot org 2010-05-19 19:27 --- rs6000.c already has rs6000_emit_stack_tie and calls it in several places in the epilogue generation, I guess we just should add another call to this. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #6 from jakub at gcc dot gnu dot org 2010-05-19 20:31 --- Created an attachment (id=20705) -- (http://gcc.gnu.org/bugzilla/attachment.cgi?id=20705action=view) gcc46-pr44199.patch Untested patch. Unfortunately, rs6000_emit_stack_tie isn't good enough. 1) it uses frame alias set, but the stack block can have arbitrary alias sets in it and we need to avoid moving all of them over 2) for the frame_pointer_needed case using sp based BLK mem isn't sufficient, as then fp based mem can still be scheduled over it. Using stack_tie insn just with hard frame pointer based BLK mem doesn't work either, then the sp restore insn can be moved over it. Could anyone please test this on SPEC to see if it causes meassurable differences? -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #7 from bergner at gcc dot gnu dot org 2010-05-19 21:52 --- Pat is going to SPEC test the patch and will report back here with his results. -- bergner at gcc dot gnu dot org changed: What|Removed |Added CC||pthaugen at gcc dot gnu dot ||org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
[Bug target/44199] ppc64 glibc miscompilation
--- Comment #8 from amodra at gmail dot com 2010-05-20 04:31 --- FWIW, Jakub's patch looks a reasonable fix to me. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44199
