From: Dorit Naishlos <DORIT@il.ibm.com>
To: Olga Golovanevsky <OLGA@il.ibm.com>
Cc: Devang Patel <dpatel@apple._com>, gcc-patches@gcc.gnu.org
Subject: Re: [lno] [patch] "while" with unknown loop bound
Date: Wed, 09 Jun 2004 15:59:00 -0000 [thread overview]
Message-ID: <OF23FB40B2.3F011075-ONC2256EAE.004E1E71-C2256EAE.004E7139@il.ibm.com> (raw)
[-- Attachment #1: Type: text/plain, Size: 2096 bytes --]
committed, along with the test-case below.
dorit
(See attached file: tree-ssa-vect-30.c)
Olga Golovanevsky
To: Dorit Naishlos/Haifa/IBM@IBMIL
06/06/2004 15:35 cc: Zdenek Dvorak <rakdver@atrey.karlin.mff.cuni.cz>, Sebastian Pop
<sebastian.pop@cri.ensmp.fr>, Devang Patel <dpatel@apple. com>,
gcc-patches@gcc.gnu.org
From: Olga Golovanevsky/Haifa/IBM@IBMIL
Subject: [lno] [patch] "while" with unknown loop bound
This patch allows vectorizing of "while" loop with unknown loop bound.
Thus, for example, the following loop will vectorize now:
while (n--) {
a[i] = b[i];
i++;
}
The patch also contains correction in cfg construction for
loop duplication in function tree_duplicate_loop_to_exit_cfg.
Bootstrapped on powerpc-apple-darwin7.0.0.
ChangeLog:
* tree-vctorizer.c
(vect_update_initial_conditions_of_duplicated_loop):
Handle general form of "init" and "step" of access function.
* tree-ssa-loop-manip.c (tree_duplicate_loop_to_exit_cfg):
The exit from loop is taken from bb previous to latch instead
of latch itself.
#### diffJ6_2.txt has been removed from this note on June 09, 2004 by Dorit
Naishlos
[-- Attachment #2: tree-ssa-vect-30.c --]
[-- Type: application/octet-stream, Size: 1295 bytes --]
/* { dg-do run { target powerpc*-*-* } } */
/* { dg-do compile { target i?86-*-* } } */
/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-stats -maltivec" { target powerpc*-*-* } } */
/* { dg-options "-O2 -ftree-vectorize -fdump-tree-vect-stats -msse2" { target i?86-*-* } } */
#include <stdarg.h>
#include <signal.h>
#define N 16
float b[N] = {0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30};
float a[N];
float c[N];
int main1 (int n)
{
int i=0;
/* Vectorized: unknown loop bound. */
while (n--) {
a[i] = b[i];
i++;
}
/* check results: */
for (i = 0; i < n; i++)
{
if (a[i] != b[i])
abort ();
}
return 0;
}
int main2 (unsigned int n)
{
int i=0;
/* Vectorized: unknown loop bound. */
while (n--) {
c[i] = b[i];
i++;
}
/* check results: */
for (i = 0; i < n; i++)
{
if (c[i] != b[i])
abort ();
}
return 0;
}
void
sig_ill_handler (int sig)
{
exit(0);
}
int main (void)
{
/* Exit on systems without altivec. */
signal (SIGILL, sig_ill_handler);
/* Altivec instruction, 'vor %v0,%v0,%v0'. */
asm volatile (".long 0x10000484");
signal (SIGILL, SIG_DFL);
return main1 (N);
return main2 (N);
}
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 2 "vect" } } */
next reply other threads:[~2004-06-09 14:07 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2004-06-09 15:59 Dorit Naishlos [this message]
-- strict thread matches above, loose matches on Subject: below --
2004-06-06 13:57 Olga Golovanevsky
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=OF23FB40B2.3F011075-ONC2256EAE.004E1E71-C2256EAE.004E7139@il.ibm.com \
--to=dorit@il.ibm.com \
--cc=OLGA@il.ibm.com \
--cc=dpatel@apple._com \
--cc=gcc-patches@gcc.gnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).