public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug tree-optimization/55590] New: SRA still produces unnecessarily unaligned memory accesses
@ 2012-12-04 10:35 jamborm at gcc dot gnu.org
2012-12-05 15:18 ` [Bug tree-optimization/55590] " jamborm at gcc dot gnu.org
` (2 more replies)
0 siblings, 3 replies; 4+ messages in thread
From: jamborm at gcc dot gnu.org @ 2012-12-04 10:35 UTC (permalink / raw)
To: gcc-bugs
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55590
Bug #: 55590
Summary: SRA still produces unnecessarily unaligned memory
accesses
Classification: Unclassified
Product: gcc
Version: 4.8.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
AssignedTo: jamborm@gcc.gnu.org
ReportedBy: jamborm@gcc.gnu.org
SRA can still produce unaligned memory accesses which should be
aligned when it's basing its new scalar access on a MEM_REF buried
below COMPONENT_REFs or ARRAY_REFs.
Testcase 1:
/* { dg-do compile } */
/* { dg-options "-O2 -mavx" } */
#include <immintrin.h>
struct S
{
__m128 a, b;
};
struct T
{
int a;
struct S s;
};
void foo (struct T *p, __m128 v)
{
struct S s;
s = p->s;
s.b = _mm_add_ps(s.b, v);
p->s = s;
}
/* { dg-final { scan-assembler-not "vmovups" } } */
on x86_64 compiles to
vmovups 32(%rdi), %xmm1
vaddps %xmm0, %xmm1, %xmm0
vmovups %xmm0, 32(%rdi)
even though it should really be
vaddps 32(%rdi), %xmm0, %xmm0
vmovaps %xmm0, 32(%rdi)
ret
Testcase 2 (which describes why this should be fixed differently from
the recent IPA-SRA patch because of the variable array index):
/* { dg-do compile } */
/* { dg-options "-O2 -mavx" } */
#include <immintrin.h>
struct S
{
__m128 a, b;
};
struct T
{
int a;
struct S s[8];
};
void foo (struct T *p, int i, __m128 v)
{
struct S s;
s = p->s[i];
s.b = _mm_add_ps(s.b, v);
p->s[i] = s;
}
/* { dg-final { scan-assembler-not "vmovups" } } */
Compiles to
movslq %esi, %rsi
salq $5, %rsi
leaq 16(%rdi,%rsi), %rax
vmovups 16(%rax), %xmm1
vaddps %xmm0, %xmm1, %xmm0
vmovups %xmm0, 16(%rax)
ret
when it should produce
movslq %esi, %rsi
salq $5, %rsi
leaq 16(%rdi,%rsi), %rax
vaddps 16(%rax), %xmm0, %xmm0
vmovaps %xmm0, 16(%rax)
ret
I'm testing a patch.
^ permalink raw reply [flat|nested] 4+ messages in thread
* [Bug tree-optimization/55590] SRA still produces unnecessarily unaligned memory accesses
2012-12-04 10:35 [Bug tree-optimization/55590] New: SRA still produces unnecessarily unaligned memory accesses jamborm at gcc dot gnu.org
@ 2012-12-05 15:18 ` jamborm at gcc dot gnu.org
2012-12-07 12:51 ` jamborm at gcc dot gnu.org
2012-12-07 13:08 ` jamborm at gcc dot gnu.org
2 siblings, 0 replies; 4+ messages in thread
From: jamborm at gcc dot gnu.org @ 2012-12-05 15:18 UTC (permalink / raw)
To: gcc-bugs
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55590
Martin Jambor <jamborm at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
URL| |http://gcc.gnu.org/ml/gcc-p
| |atches/2012-12/msg00237.htm
| |l
--- Comment #1 from Martin Jambor <jamborm at gcc dot gnu.org> 2012-12-05 15:17:29 UTC ---
I have submitted the patch to the mailing list:
http://gcc.gnu.org/ml/gcc-patches/2012-12/msg00237.html
^ permalink raw reply [flat|nested] 4+ messages in thread
* [Bug tree-optimization/55590] SRA still produces unnecessarily unaligned memory accesses
2012-12-04 10:35 [Bug tree-optimization/55590] New: SRA still produces unnecessarily unaligned memory accesses jamborm at gcc dot gnu.org
2012-12-05 15:18 ` [Bug tree-optimization/55590] " jamborm at gcc dot gnu.org
@ 2012-12-07 12:51 ` jamborm at gcc dot gnu.org
2012-12-07 13:08 ` jamborm at gcc dot gnu.org
2 siblings, 0 replies; 4+ messages in thread
From: jamborm at gcc dot gnu.org @ 2012-12-07 12:51 UTC (permalink / raw)
To: gcc-bugs
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55590
--- Comment #2 from Martin Jambor <jamborm at gcc dot gnu.org> 2012-12-07 12:50:46 UTC ---
Author: jamborm
Date: Fri Dec 7 12:50:43 2012
New Revision: 194300
URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=194300
Log:
2012-12-07 Martin Jambor <mjambor@suse.cz>
PR tree-optimization/55590
* tree-sra.c (build_ref_for_offset): Use get_object_alignment_1 to
get base alignment.
* testsuite/gcc.target/i386/pr55590-1.c: New test.
* testsuite/gcc.target/i386/pr55590-2.c: Likewise.
Added:
trunk/gcc/testsuite/gcc.target/i386/pr55590-1.c
trunk/gcc/testsuite/gcc.target/i386/pr55590-2.c
Modified:
trunk/gcc/ChangeLog
trunk/gcc/testsuite/ChangeLog
trunk/gcc/tree-sra.c
^ permalink raw reply [flat|nested] 4+ messages in thread
* [Bug tree-optimization/55590] SRA still produces unnecessarily unaligned memory accesses
2012-12-04 10:35 [Bug tree-optimization/55590] New: SRA still produces unnecessarily unaligned memory accesses jamborm at gcc dot gnu.org
2012-12-05 15:18 ` [Bug tree-optimization/55590] " jamborm at gcc dot gnu.org
2012-12-07 12:51 ` jamborm at gcc dot gnu.org
@ 2012-12-07 13:08 ` jamborm at gcc dot gnu.org
2 siblings, 0 replies; 4+ messages in thread
From: jamborm at gcc dot gnu.org @ 2012-12-07 13:08 UTC (permalink / raw)
To: gcc-bugs
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55590
Martin Jambor <jamborm at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |RESOLVED
Resolution| |FIXED
--- Comment #3 from Martin Jambor <jamborm at gcc dot gnu.org> 2012-12-07 13:07:56 UTC ---
Fixed.
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2012-12-07 13:08 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2012-12-04 10:35 [Bug tree-optimization/55590] New: SRA still produces unnecessarily unaligned memory accesses jamborm at gcc dot gnu.org
2012-12-05 15:18 ` [Bug tree-optimization/55590] " jamborm at gcc dot gnu.org
2012-12-07 12:51 ` jamborm at gcc dot gnu.org
2012-12-07 13:08 ` jamborm at gcc dot gnu.org
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).