[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 Andi Kleen andi-gcc at firstfloor dot org changed: What|Removed |Added Status|NEW |RESOLVED Resolution||FIXED --- Comment #20 from Andi Kleen andi-gcc at firstfloor dot org 2013-03-15 13:55:28 UTC --- Fixed for some time
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #19 from eraman at gcc dot gnu.org 2012-08-22 21:07:40 UTC --- Author: eraman Date: Wed Aug 22 21:07:30 2012 New Revision: 190601 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=190601 Log: 2012-08-22 Easwaran Raman era...@google.com Backport r187387 and r187477 from trunk. r187387: 2012-05-11 Uros Bizjak ubiz...@gmail.com PR target/53291 * config/i386/i386.md (xtest): Use NE condition in ix86_expand_setcc. r187477: 2012-05-14 Andrew Pinski apin...@cavium.com H.J. Lu hongjiu...@intel.com Jakub Jelinek ja...@redhat.com PR target/53315 * config/i386/i386.md (xbegin_1): Use + in constraint and match_dup. (xbegin): Updated. gcc/testsuite/ChangeLog.google-4_7: 2012-08-22 Easwaran Raman era...@google.com Backport r187477 from trunk: 2012-05-14 Andi Kleen a...@linux.intel.com Jakub Jelinek ja...@redhat.com PR target/53315 * gcc.target/i386/i386.exp (check_effective_target_rtm): New. * gcc.target/i386/rtm-check.h: New file. * gcc.target/i386/pr53315.c: New test. Added: branches/google/gcc-4_7/gcc/testsuite/gcc.target/i386/pr53315.c - copied unchanged from r187477, trunk/gcc/testsuite/gcc.target/i386/pr53315.c branches/google/gcc-4_7/gcc/testsuite/gcc.target/i386/rtm-check.h - copied unchanged from r187477, trunk/gcc/testsuite/gcc.target/i386/rtm-check.h Modified: branches/google/gcc-4_7/gcc/ChangeLog.google-4_7 branches/google/gcc-4_7/gcc/config/i386/i386.md branches/google/gcc-4_7/gcc/testsuite/ChangeLog.google-4_7 branches/google/gcc-4_7/gcc/testsuite/gcc.target/i386/i386.exp
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #18 from Jakub Jelinek jakub at gcc dot gnu.org 2012-05-14 18:47:09 UTC --- Author: jakub Date: Mon May 14 18:47:05 2012 New Revision: 187477 URL: http://gcc.gnu.org/viewcvs?root=gccview=revrev=187477 Log: 2012-05-14 Andrew Pinski apin...@cavium.com H.J. Lu hongjiu...@intel.com Jakub Jelinek ja...@redhat.com PR target/53315 * config/i386/i386.md (xbegin_1): Use + in constraint and match_dup. (xbegin): Updated. 2012-05-14 Andi Kleen a...@linux.intel.com Jakub Jelinek ja...@redhat.com PR target/53315 * gcc.target/i386/i386.exp (check_effective_target_rtm): New. * gcc.target/i386/rtm-check.h: New file. * gcc.target/i386/pr53315.c: New test. Added: trunk/gcc/testsuite/gcc.target/i386/pr53315.c trunk/gcc/testsuite/gcc.target/i386/rtm-check.h Modified: trunk/gcc/ChangeLog trunk/gcc/config/i386/i386.md trunk/gcc/testsuite/ChangeLog trunk/gcc/testsuite/gcc.target/i386/i386.exp
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #11 from Jakub Jelinek jakub at gcc dot gnu.org 2012-05-12 09:14:53 UTC --- Created attachment 27385 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=27385 gcc48-pr53315.patch That is because the patch is buggy. Fixed thusly, though haven't tested it on Haswell (obviously) nor sim. Note, it would be nice to have a peephole or something similar (guess peepholes won't do anything across multiple bbs, perhaps machine reorg) to optimize that movl$-1, %eax xbegin .L2 .L2: cmpl$-1, %eax jne .L3 xorl%eax, %eax into say movl$-1, %eax xbegin .L3 xorl%eax, %eax or even xbegin .L3 xorl%eax, %eax
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #12 from Uros Bizjak ubizjak at gmail dot com 2012-05-12 11:29:53 UTC --- (In reply to comment #11) Created attachment 27385 [details] gcc48-pr53315.patch Please introduce check-rtm.h header for use in runtime testcases, as is the case with i.e. check-sse2.h and other feature check headers.
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 Jakub Jelinek jakub at gcc dot gnu.org changed: What|Removed |Added Attachment #27385|0 |1 is obsolete|| --- Comment #13 from Jakub Jelinek jakub at gcc dot gnu.org 2012-05-12 14:05:07 UTC --- Created attachment 27387 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=27387 gcc48-pr53315.patch Like this?
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #14 from Andi Kleen andi-gcc at firstfloor dot org 2012-05-12 16:04:27 UTC --- I can confirm the simple test program works correctly with Jakub's patch. I'll leave full bootstrap to HJ.
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #15 from Andi Kleen andi-gcc at firstfloor dot org 2012-05-12 16:06:00 UTC --- Oh yes and it would be really nice to have those peepholes for xbegin Jakub. I normally use my own macros with asm goto to avoid the ugly code. Do you think machine reorg could be done without slowing down the compiler?
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #16 from Uros Bizjak ubizjak at gmail dot com 2012-05-12 16:31:21 UTC --- (In reply to comment #13) Like this? Yes, this is OK, after someone confirms that the testcase works as expected on HW or simulator.
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #17 from Uros Bizjak ubizjak at gmail dot com 2012-05-12 16:36:39 UTC --- (In reply to comment #15) Do you think machine reorg could be done without slowing down the compiler? Yes, xbegin RTM pattern can raise a flag that triggers machine reorg (so it won't fire for functions that don't emit xbegin).
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #6 from Uros Bizjak ubizjak at gmail dot com 2012-05-11 09:41:15 UTC --- (In reply to comment #5) Created attachment 27370 [details] A patch Patch looks OK to me, but please let Andi play with this a bit, so we are sure we won't hit some other reload limitation with this pattern.
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 Uros Bizjak ubizjak at gmail dot com changed: What|Removed |Added Keywords||patch Status|UNCONFIRMED |NEW Last reconfirmed||2012-05-11 Version|unknown |4.8.0 Target Milestone|--- |4.8.0 Ever Confirmed|0 |1 --- Comment #7 from Uros Bizjak ubizjak at gmail dot com 2012-05-11 09:51:46 UTC --- As discussed in PR 53291, please also add a runtime testcase that will cover PR 53291 as well as this PR.
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #8 from Andi Kleen andi-gcc at firstfloor dot org 2012-05-11 18:02:43 UTC --- I tested HJs fix on the test case and also on a more complex program and it all works as expected. Please commit.
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315
--- Comment #9 from Andi Kleen andi-gcc at firstfloor dot org 2012-05-11
21:35:47 UTC ---
Sorry I was wrong earlier. Retested now fully with a full test case and HJs
patch and i always get aborts
The xbegin gets miscompiled now, the in transaction branch disappears.
400460: 48 83 ec 08 sub$0x8,%rsp
400464: b8 ff ff ff ff mov$0x,%eax
400469: c7 f8 00 00 00 00 xbeginq 40046f main+0xf
40046f: bf d8 06 40 00 mov$0x4006d8,%edi
400474: 31 f6 xor%esi,%esi
400476: 31 c0 xor%eax,%eax
400478: e8 b3 ff ff ff callq 400430 printf@plt
40047d: 31 ff xor%edi,%edi
40047f: e8 bc ff ff ff callq 400440 exit@plt
/* PR53315 and PR53291 */
/* { dg-do run } */
/* { dg-options -O2 -mrtm } */
#include immintrin.h
#include cpuid.h
#include stdlib.h
#include stdio.h
static int cpu_has_rtm(void)
{
if (__get_cpuid_max(0, NULL) = 7) {
unsigned a, b, c, d;
__cpuid_count(7, 0, a, b, c, d);
return !!(b bit_RTM);
}
return 0;
}
int main(void)
{
int flag = -1;
unsigned status;
if (!cpu_has_rtm) {
printf(no tsx support. untested\n);
exit(0);
}
if ((status = _xbegin()) == _XBEGIN_STARTED) {
flag = _xtest();
_xend();
} else {
/* Note this is legal according to the TSX spec */
printf(unexpected abort %x. untested\n, status);
exit(0);
}
if (flag != 1)
abort();
if (_xtest() != 0)
abort();
return 0;
}
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315
--- Comment #10 from phpbbaid at gmail dot com 2012-05-11 23:02:22 UTC ---
please unsubscribe
-Original Message-
From: andi-gcc at firstfloor dot org
Sent: Friday, May 11, 2012 11:35 PM
To: gcc-bugs@gcc.gnu.org
Subject: [Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315
--- Comment #9 from Andi Kleen andi-gcc at firstfloor dot org 2012-05-11
21:35:47 UTC ---
Sorry I was wrong earlier. Retested now fully with a full test case and HJs
patch and i always get aborts
The xbegin gets miscompiled now, the in transaction branch disappears.
400460: 48 83 ec 08 sub$0x8,%rsp
400464: b8 ff ff ff ff mov$0x,%eax
400469: c7 f8 00 00 00 00 xbeginq 40046f main+0xf
40046f: bf d8 06 40 00 mov$0x4006d8,%edi
400474: 31 f6 xor%esi,%esi
400476: 31 c0 xor%eax,%eax
400478: e8 b3 ff ff ff callq 400430 printf@plt
40047d: 31 ff xor%edi,%edi
40047f: e8 bc ff ff ff callq 400440 exit@plt
/* PR53315 and PR53291 */
/* { dg-do run } */
/* { dg-options -O2 -mrtm } */
#include immintrin.h
#include cpuid.h
#include stdlib.h
#include stdio.h
static int cpu_has_rtm(void)
{
if (__get_cpuid_max(0, NULL) = 7) {
unsigned a, b, c, d;
__cpuid_count(7, 0, a, b, c, d);
return !!(b bit_RTM);
}
return 0;
}
int main(void)
{
int flag = -1;
unsigned status;
if (!cpu_has_rtm) {
printf(no tsx support. untested\n);
exit(0);
}
if ((status = _xbegin()) == _XBEGIN_STARTED) {
flag = _xtest();
_xend();
} else {
/* Note this is legal according to the TSX spec */
printf(unexpected abort %x. untested\n, status);
exit(0);
}
if (flag != 1)
abort();
if (_xtest() != 0)
abort();
return 0;
}
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 Andi Kleen andi-gcc at firstfloor dot org changed: What|Removed |Added CC||kirill.yukhin at intel dot ||com --- Comment #1 from Andi Kleen andi-gcc at firstfloor dot org 2012-05-10 22:51:45 UTC --- Hmm it goes away when i remove the (status = _xbegin) so it may be actually xbegin not xtest
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #2 from Uros Bizjak ubizjak at gmail dot com 2012-05-10 23:42:05 UTC --- (In reply to comment #1) Hmm it goes away when i remove the (status = _xbegin) so it may be actually xbegin not xtest Yes, it is xbegin_1 insn pattern that doesn't get registers allocated. However, the pattern (although a bit unusual), looks correct to me.
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #3 from Andrew Pinski pinskia at gcc dot gnu.org 2012-05-10 23:44:01 UTC --- Reload does not know how to handle reloading into jump instructions.
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315
--- Comment #4 from Andrew Pinski pinskia at gcc dot gnu.org 2012-05-10
23:47:57 UTC ---
(define_expand xbegin
[(set (match_operand:SI 0 register_operand)
(unspec_volatile:SI [(match_dup 1)] UNSPECV_XBEGIN))]
TARGET_RTM
{
rtx label = gen_label_rtx ();
operands[0] = force_reg (SImode, constm1_rtx);
emit_jump_insn (gen_xbegin_1 (operands[0], label));
emit_label (label);
LABEL_NUSES (label) = 1;
DONE;
})
(define_insn xbegin_1
[(set (pc)
(if_then_else (ne (unspec [(const_int 0)] UNSPEC_XBEGIN_ABORT)
(const_int 0))
(label_ref (match_operand 1))
(pc)))
(set (match_operand:SI 0 register_operand +a)
(unspec_volatile:SI [(match_dup 0)]
UNSPECV_XBEGIN))]
TARGET_RTM
xbegin\t%l2
[(set_attr type other)
(set_attr length 6)])
I think this is the one of the few cases where you want to use + in the
constraint and match_dup.
[Bug target/53315] simple xtest program generates ICE
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53315 --- Comment #5 from H.J. Lu hjl.tools at gmail dot com 2012-05-11 03:14:45 UTC --- Created attachment 27370 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=27370 A patch
