* [PATCH] PR opt/7291: bzero problems on x86_64
@ 2002-07-23 19:32 Roger Sayle
2002-07-23 22:33 ` Mark Mitchell
2002-07-23 22:58 ` Andreas Jaeger
0 siblings, 2 replies; 4+ messages in thread
From: Roger Sayle @ 2002-07-23 19:32 UTC (permalink / raw)
To: gcc-patches; +Cc: fvdl
This patch is essentially Frank van der Linden's fix as proposed
in the GNATS bug report for PR optimization/7291. This slips
through the correctness checks in the testsuite: memset-1.c tests
variable alignment and variable length and memset-2.c tests
variable alignment and constant length. The new test case below
adds memset-3.c which tests constant alignment and variable length.
Frank has confirmed that the new testcase currently fails on x86_64
but passses with the patch below applied.
Ok for mainline?
2002-07-23 Frank van der Linden <fvdl@wasabisystems.com>
PR optimization/7291
* config/i386/i386.c (ix86_expand_clrstr): Fix bzero alignment
problem on x86_64.
2002-07-23 Roger Sayle <roger@eyesopen.com>
* gcc.c-torture/execute/memset-3.c: New testcase.
Index: config/i386/i386.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.c,v
retrieving revision 1.438
diff -c -3 -p -r1.438 i386.c
*** config/i386/i386.c 23 Jul 2002 06:26:28 -0000 1.438
--- config/i386/i386.c 24 Jul 2002 01:51:47 -0000
*************** ix86_expand_clrstr (src, count_exp, alig
*** 10193,10199 ****
gen_rtx_SUBREG (SImode, zeroreg, 0)));
if (TARGET_64BIT && (align <= 4 || count == 0))
{
! rtx label = ix86_expand_aligntest (countreg, 2);
emit_insn (gen_strsetsi (destreg,
gen_rtx_SUBREG (SImode, zeroreg, 0)));
emit_label (label);
--- 10193,10199 ----
gen_rtx_SUBREG (SImode, zeroreg, 0)));
if (TARGET_64BIT && (align <= 4 || count == 0))
{
! rtx label = ix86_expand_aligntest (countreg, 4);
emit_insn (gen_strsetsi (destreg,
gen_rtx_SUBREG (SImode, zeroreg, 0)));
emit_label (label);
*** /dev/null Thu Aug 30 14:30:55 2001
--- gcc.c-torture/execute/memset-3.c Tue Jul 23 19:40:21 2002
***************
*** 0 ****
--- 1,208 ----
+ /* Copyright (C) 2002 Free Software Foundation.
+
+ Test memset with various combinations of constant pointer alignments and
+ lengths to make sure any optimizations in the compiler are correct.
+
+ Written by Roger Sayle, July 22, 2002. */
+
+ #ifndef MAX_OFFSET
+ #define MAX_OFFSET (sizeof (long long))
+ #endif
+
+ #ifndef MAX_COPY
+ #define MAX_COPY 15
+ #endif
+
+ #ifndef MAX_EXTRA
+ #define MAX_EXTRA (sizeof (long long))
+ #endif
+
+ #define MAX_LENGTH (MAX_OFFSET + MAX_COPY + MAX_EXTRA)
+
+ static union {
+ char buf[MAX_LENGTH];
+ long long align_int;
+ long double align_fp;
+ } u;
+
+ char A = 'A';
+
+ void reset ()
+ {
+ int i;
+
+ for (i = 0; i < MAX_LENGTH; i++)
+ u.buf[i] = 'a';
+ }
+
+ void check (int off, int len, int ch)
+ {
+ char *q;
+ int i;
+
+ q = u.buf;
+ for (i = 0; i < off; i++, q++)
+ if (*q != 'a')
+ abort ();
+
+ for (i = 0; i < len; i++, q++)
+ if (*q != ch)
+ abort ();
+
+ for (i = 0; i < MAX_EXTRA; i++, q++)
+ if (*q != 'a')
+ abort ();
+ }
+
+ int main ()
+ {
+ int len;
+ char *p;
+
+ /* off == 0 */
+ for (len = 0; len < MAX_COPY; len++)
+ {
+ reset ();
+
+ p = memset (u.buf, '\0', len);
+ if (p != u.buf) abort ();
+ check (0, len, '\0');
+
+ p = memset (u.buf, A, len);
+ if (p != u.buf) abort ();
+ check (0, len, 'A');
+
+ p = memset (u.buf, 'B', len);
+ if (p != u.buf) abort ();
+ check (0, len, 'B');
+ }
+
+ /* off == 1 */
+ for (len = 0; len < MAX_COPY; len++)
+ {
+ reset ();
+
+ p = memset (u.buf+1, '\0', len);
+ if (p != u.buf+1) abort ();
+ check (1, len, '\0');
+
+ p = memset (u.buf+1, A, len);
+ if (p != u.buf+1) abort ();
+ check (1, len, 'A');
+
+ p = memset (u.buf+1, 'B', len);
+ if (p != u.buf+1) abort ();
+ check (1, len, 'B');
+ }
+
+ /* off == 2 */
+ for (len = 0; len < MAX_COPY; len++)
+ {
+ reset ();
+
+ p = memset (u.buf+2, '\0', len);
+ if (p != u.buf+2) abort ();
+ check (2, len, '\0');
+
+ p = memset (u.buf+2, A, len);
+ if (p != u.buf+2) abort ();
+ check (2, len, 'A');
+
+ p = memset (u.buf+2, 'B', len);
+ if (p != u.buf+2) abort ();
+ check (2, len, 'B');
+ }
+
+ /* off == 3 */
+ for (len = 0; len < MAX_COPY; len++)
+ {
+ reset ();
+
+ p = memset (u.buf+3, '\0', len);
+ if (p != u.buf+3) abort ();
+ check (3, len, '\0');
+
+ p = memset (u.buf+3, A, len);
+ if (p != u.buf+3) abort ();
+ check (3, len, 'A');
+
+ p = memset (u.buf+3, 'B', len);
+ if (p != u.buf+3) abort ();
+ check (3, len, 'B');
+ }
+
+ /* off == 4 */
+ for (len = 0; len < MAX_COPY; len++)
+ {
+ reset ();
+
+ p = memset (u.buf+4, '\0', len);
+ if (p != u.buf+4) abort ();
+ check (4, len, '\0');
+
+ p = memset (u.buf+4, A, len);
+ if (p != u.buf+4) abort ();
+ check (4, len, 'A');
+
+ p = memset (u.buf+4, 'B', len);
+ if (p != u.buf+4) abort ();
+ check (4, len, 'B');
+ }
+
+ /* off == 5 */
+ for (len = 0; len < MAX_COPY; len++)
+ {
+ reset ();
+
+ p = memset (u.buf+5, '\0', len);
+ if (p != u.buf+5) abort ();
+ check (5, len, '\0');
+
+ p = memset (u.buf+5, A, len);
+ if (p != u.buf+5) abort ();
+ check (5, len, 'A');
+
+ p = memset (u.buf+5, 'B', len);
+ if (p != u.buf+5) abort ();
+ check (5, len, 'B');
+ }
+
+ /* off == 6 */
+ for (len = 0; len < MAX_COPY; len++)
+ {
+ reset ();
+
+ p = memset (u.buf+6, '\0', len);
+ if (p != u.buf+6) abort ();
+ check (6, len, '\0');
+
+ p = memset (u.buf+6, A, len);
+ if (p != u.buf+6) abort ();
+ check (6, len, 'A');
+
+ p = memset (u.buf+6, 'B', len);
+ if (p != u.buf+6) abort ();
+ check (6, len, 'B');
+ }
+
+ /* off == 7 */
+ for (len = 0; len < MAX_COPY; len++)
+ {
+ reset ();
+
+ p = memset (u.buf+7, '\0', len);
+ if (p != u.buf+7) abort ();
+ check (7, len, '\0');
+
+ p = memset (u.buf+7, A, len);
+ if (p != u.buf+7) abort ();
+ check (7, len, 'A');
+
+ p = memset (u.buf+7, 'B', len);
+ if (p != u.buf+7) abort ();
+ check (7, len, 'B');
+ }
+
+ exit (0);
+ }
+
Roger
--
Roger Sayle, E-mail: roger@eyesopen.com
OpenEye Scientific Software, WWW: http://www.eyesopen.com/
Suite 1107, 3600 Cerrillos Road, Tel: (+1) 505-473-7385
Santa Fe, New Mexico, 87507. Fax: (+1) 505-473-0833
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] PR opt/7291: bzero problems on x86_64
2002-07-23 19:32 [PATCH] PR opt/7291: bzero problems on x86_64 Roger Sayle
@ 2002-07-23 22:33 ` Mark Mitchell
2002-07-23 22:58 ` Andreas Jaeger
1 sibling, 0 replies; 4+ messages in thread
From: Mark Mitchell @ 2002-07-23 22:33 UTC (permalink / raw)
To: Roger Sayle, gcc-patches; +Cc: fvdl
> Ok for mainline?
Yes, thanks.
--
Mark Mitchell mark@codesourcery.com
CodeSourcery, LLC http://www.codesourcery.com
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] PR opt/7291: bzero problems on x86_64
2002-07-23 19:32 [PATCH] PR opt/7291: bzero problems on x86_64 Roger Sayle
2002-07-23 22:33 ` Mark Mitchell
@ 2002-07-23 22:58 ` Andreas Jaeger
2002-07-25 10:11 ` Mark Mitchell
1 sibling, 1 reply; 4+ messages in thread
From: Andreas Jaeger @ 2002-07-23 22:58 UTC (permalink / raw)
To: Roger Sayle; +Cc: gcc-patches, fvdl
Roger Sayle <roger@eyesopen.com> writes:
> This patch is essentially Frank van der Linden's fix as proposed
> in the GNATS bug report for PR optimization/7291. This slips
> through the correctness checks in the testsuite: memset-1.c tests
> variable alignment and variable length and memset-2.c tests
> variable alignment and constant length. The new test case below
> adds memset-3.c which tests constant alignment and variable length.
>
> Frank has confirmed that the new testcase currently fails on x86_64
> but passses with the patch below applied.
>
>
> Ok for mainline?
Exactly the same patch is needed for the 3.1 branch. Ok to add for 3.2
after it has been branched?
Thanks Roger and Frank!
Andreas
>
>
> 2002-07-23 Frank van der Linden <fvdl@wasabisystems.com>
>
> PR optimization/7291
> * config/i386/i386.c (ix86_expand_clrstr): Fix bzero alignment
> problem on x86_64.
>
> 2002-07-23 Roger Sayle <roger@eyesopen.com>
>
> * gcc.c-torture/execute/memset-3.c: New testcase.
>
>
> Index: config/i386/i386.c
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.c,v
> retrieving revision 1.438
> diff -c -3 -p -r1.438 i386.c
> *** config/i386/i386.c 23 Jul 2002 06:26:28 -0000 1.438
> --- config/i386/i386.c 24 Jul 2002 01:51:47 -0000
> *************** ix86_expand_clrstr (src, count_exp, alig
> *** 10193,10199 ****
> gen_rtx_SUBREG (SImode, zeroreg, 0)));
> if (TARGET_64BIT && (align <= 4 || count == 0))
> {
> ! rtx label = ix86_expand_aligntest (countreg, 2);
> emit_insn (gen_strsetsi (destreg,
> gen_rtx_SUBREG (SImode, zeroreg, 0)));
> emit_label (label);
> --- 10193,10199 ----
> gen_rtx_SUBREG (SImode, zeroreg, 0)));
> if (TARGET_64BIT && (align <= 4 || count == 0))
> {
> ! rtx label = ix86_expand_aligntest (countreg, 4);
> emit_insn (gen_strsetsi (destreg,
> gen_rtx_SUBREG (SImode, zeroreg, 0)));
> emit_label (label);
> *** /dev/null Thu Aug 30 14:30:55 2001
> --- gcc.c-torture/execute/memset-3.c Tue Jul 23 19:40:21 2002
> ***************
> *** 0 ****
> --- 1,208 ----
> + /* Copyright (C) 2002 Free Software Foundation.
> +
> + Test memset with various combinations of constant pointer alignments and
> + lengths to make sure any optimizations in the compiler are correct.
> +
> + Written by Roger Sayle, July 22, 2002. */
> +
> + #ifndef MAX_OFFSET
> + #define MAX_OFFSET (sizeof (long long))
> + #endif
> +
> + #ifndef MAX_COPY
> + #define MAX_COPY 15
> + #endif
> +
> + #ifndef MAX_EXTRA
> + #define MAX_EXTRA (sizeof (long long))
> + #endif
> +
> + #define MAX_LENGTH (MAX_OFFSET + MAX_COPY + MAX_EXTRA)
> +
> + static union {
> + char buf[MAX_LENGTH];
> + long long align_int;
> + long double align_fp;
> + } u;
> +
> + char A = 'A';
> +
> + void reset ()
> + {
> + int i;
> +
> + for (i = 0; i < MAX_LENGTH; i++)
> + u.buf[i] = 'a';
> + }
> +
> + void check (int off, int len, int ch)
> + {
> + char *q;
> + int i;
> +
> + q = u.buf;
> + for (i = 0; i < off; i++, q++)
> + if (*q != 'a')
> + abort ();
> +
> + for (i = 0; i < len; i++, q++)
> + if (*q != ch)
> + abort ();
> +
> + for (i = 0; i < MAX_EXTRA; i++, q++)
> + if (*q != 'a')
> + abort ();
> + }
> +
> + int main ()
> + {
> + int len;
> + char *p;
> +
> + /* off == 0 */
> + for (len = 0; len < MAX_COPY; len++)
> + {
> + reset ();
> +
> + p = memset (u.buf, '\0', len);
> + if (p != u.buf) abort ();
> + check (0, len, '\0');
> +
> + p = memset (u.buf, A, len);
> + if (p != u.buf) abort ();
> + check (0, len, 'A');
> +
> + p = memset (u.buf, 'B', len);
> + if (p != u.buf) abort ();
> + check (0, len, 'B');
> + }
> +
> + /* off == 1 */
> + for (len = 0; len < MAX_COPY; len++)
> + {
> + reset ();
> +
> + p = memset (u.buf+1, '\0', len);
> + if (p != u.buf+1) abort ();
> + check (1, len, '\0');
> +
> + p = memset (u.buf+1, A, len);
> + if (p != u.buf+1) abort ();
> + check (1, len, 'A');
> +
> + p = memset (u.buf+1, 'B', len);
> + if (p != u.buf+1) abort ();
> + check (1, len, 'B');
> + }
> +
> + /* off == 2 */
> + for (len = 0; len < MAX_COPY; len++)
> + {
> + reset ();
> +
> + p = memset (u.buf+2, '\0', len);
> + if (p != u.buf+2) abort ();
> + check (2, len, '\0');
> +
> + p = memset (u.buf+2, A, len);
> + if (p != u.buf+2) abort ();
> + check (2, len, 'A');
> +
> + p = memset (u.buf+2, 'B', len);
> + if (p != u.buf+2) abort ();
> + check (2, len, 'B');
> + }
> +
> + /* off == 3 */
> + for (len = 0; len < MAX_COPY; len++)
> + {
> + reset ();
> +
> + p = memset (u.buf+3, '\0', len);
> + if (p != u.buf+3) abort ();
> + check (3, len, '\0');
> +
> + p = memset (u.buf+3, A, len);
> + if (p != u.buf+3) abort ();
> + check (3, len, 'A');
> +
> + p = memset (u.buf+3, 'B', len);
> + if (p != u.buf+3) abort ();
> + check (3, len, 'B');
> + }
> +
> + /* off == 4 */
> + for (len = 0; len < MAX_COPY; len++)
> + {
> + reset ();
> +
> + p = memset (u.buf+4, '\0', len);
> + if (p != u.buf+4) abort ();
> + check (4, len, '\0');
> +
> + p = memset (u.buf+4, A, len);
> + if (p != u.buf+4) abort ();
> + check (4, len, 'A');
> +
> + p = memset (u.buf+4, 'B', len);
> + if (p != u.buf+4) abort ();
> + check (4, len, 'B');
> + }
> +
> + /* off == 5 */
> + for (len = 0; len < MAX_COPY; len++)
> + {
> + reset ();
> +
> + p = memset (u.buf+5, '\0', len);
> + if (p != u.buf+5) abort ();
> + check (5, len, '\0');
> +
> + p = memset (u.buf+5, A, len);
> + if (p != u.buf+5) abort ();
> + check (5, len, 'A');
> +
> + p = memset (u.buf+5, 'B', len);
> + if (p != u.buf+5) abort ();
> + check (5, len, 'B');
> + }
> +
> + /* off == 6 */
> + for (len = 0; len < MAX_COPY; len++)
> + {
> + reset ();
> +
> + p = memset (u.buf+6, '\0', len);
> + if (p != u.buf+6) abort ();
> + check (6, len, '\0');
> +
> + p = memset (u.buf+6, A, len);
> + if (p != u.buf+6) abort ();
> + check (6, len, 'A');
> +
> + p = memset (u.buf+6, 'B', len);
> + if (p != u.buf+6) abort ();
> + check (6, len, 'B');
> + }
> +
> + /* off == 7 */
> + for (len = 0; len < MAX_COPY; len++)
> + {
> + reset ();
> +
> + p = memset (u.buf+7, '\0', len);
> + if (p != u.buf+7) abort ();
> + check (7, len, '\0');
> +
> + p = memset (u.buf+7, A, len);
> + if (p != u.buf+7) abort ();
> + check (7, len, 'A');
> +
> + p = memset (u.buf+7, 'B', len);
> + if (p != u.buf+7) abort ();
> + check (7, len, 'B');
> + }
> +
> + exit (0);
> + }
> +
>
> Roger
Andreas
--
Andreas Jaeger
SuSE Labs aj@suse.de
private aj@arthur.inka.de
http://www.suse.de/~aj
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] PR opt/7291: bzero problems on x86_64
2002-07-23 22:58 ` Andreas Jaeger
@ 2002-07-25 10:11 ` Mark Mitchell
0 siblings, 0 replies; 4+ messages in thread
From: Mark Mitchell @ 2002-07-25 10:11 UTC (permalink / raw)
To: Andreas Jaeger, Roger Sayle; +Cc: gcc-patches, fvdl
> Exactly the same patch is needed for the 3.1 branch. Ok to add for 3.2
> after it has been branched?
Yes.
--
Mark Mitchell mark@codesourcery.com
CodeSourcery, LLC http://www.codesourcery.com
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2002-07-25 16:28 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2002-07-23 19:32 [PATCH] PR opt/7291: bzero problems on x86_64 Roger Sayle
2002-07-23 22:33 ` Mark Mitchell
2002-07-23 22:58 ` Andreas Jaeger
2002-07-25 10:11 ` Mark Mitchell
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).