From: Thomas Koenig <tkoenig@netcologne.de>
To: "fortran@gcc.gnu.org" <fortran@gcc.gnu.org>,
gcc-patches <gcc-patches@gcc.gnu.org>
Subject: [patch, fortran] [4.6/4.7 Regression] Fix PR 48955
Date: Sun, 22 May 2011 18:59:00 -0000 [thread overview]
Message-ID: <4DD92C77.5090107@netcologne.de> (raw)
[-- Attachment #1: Type: text/plain, Size: 1027 bytes --]
Hello world,
the attached patch fixes PR 48955, a wrong-code regression for 4.6 and
4.7, including the test case from
http://gcc.gnu.org/ml/fortran/2011-05/msg00093.html . It follows the
outline from http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48955#c7 .
Regression-tested. OK for trunk?
Thomas
2011-05-22 Thomas Koenig <tkoenig@gcc.gnu.org>
PR fortran/48955
* gfortran.h (gfc_reverse): Change to struct with two boolean
variables, forward_ok and backward_ok.
* trans-expr.c (gfc_trans_assignment_1): Initialize using
new gfc_reverse struct.
* trans-array.c (gfc_init_loopinfo): Likewise.
(gfc_trans_scalarized_loop): Use new gfc_reverse.
* dependency.c (gfc_dep_resovler): Use new gfc_reverse.
If a forward dependency is found, a backward dependency is
no longer possible and vice versa.
2011-05-22 Thomas Koenig <tkoenig@gcc.gnu.org>
PR fortran/48955
* gfortran.dg/dependency_40.f90: New test.
[-- Attachment #2: reverse-2.diff --]
[-- Type: text/x-patch, Size: 4184 bytes --]
Index: trans-expr.c
===================================================================
--- trans-expr.c (Revision 173754)
+++ trans-expr.c (Arbeitskopie)
@@ -6068,9 +6068,14 @@ gfc_trans_assignment_1 (gfc_expr * expr1, gfc_expr
/* Calculate the bounds of the scalarization. */
gfc_conv_ss_startstride (&loop);
+
/* Enable loop reversal. */
for (n = 0; n < loop.dimen; n++)
- loop.reverse[n] = GFC_REVERSE_NOT_SET;
+ {
+ loop.reverse[n].forward_ok = true;
+ loop.reverse[n].backward_ok = true;
+ }
+
/* Resolve any data dependencies in the statement. */
gfc_conv_resolve_dependencies (&loop, lss, rss);
/* Setup the scalarizing loops. */
Index: trans-array.c
===================================================================
--- trans-array.c (Revision 173754)
+++ trans-array.c (Arbeitskopie)
@@ -2244,7 +2244,8 @@ gfc_init_loopinfo (gfc_loopinfo * loop)
for (n = 0; n < GFC_MAX_DIMENSIONS; n++)
{
loop->order[n] = n;
- loop->reverse[n] = GFC_CANNOT_REVERSE;
+ loop->reverse[n].forward_ok = true;
+ loop->reverse[n].backward_ok = false;
}
loop->ss = gfc_ss_terminator;
@@ -2925,7 +2926,8 @@ gfc_trans_scalarized_loop_end (gfc_loopinfo * loop
}
else
{
- bool reverse_loop = (loop->reverse[n] == GFC_REVERSE_SET)
+ bool reverse_loop = loop->reverse[n].backward_ok
+ && !loop->reverse[n].forward_ok
&& (loop->temp_ss == NULL);
loopbody = gfc_finish_block (pbody);
Index: gfortran.h
===================================================================
--- gfortran.h (Revision 173754)
+++ gfortran.h (Arbeitskopie)
@@ -576,12 +576,10 @@ typedef enum
}
gfc_fcoarray;
-typedef enum
+typedef struct
{
- GFC_REVERSE_NOT_SET,
- GFC_REVERSE_SET,
- GFC_CAN_REVERSE,
- GFC_CANNOT_REVERSE
+ bool forward_ok;
+ bool backward_ok;
}
gfc_reverse;
Index: dependency.c
===================================================================
--- dependency.c (Revision 173754)
+++ dependency.c (Arbeitskopie)
@@ -1807,34 +1807,35 @@ gfc_dep_resolver (gfc_ref *lref, gfc_ref *rref, gf
/* Now deal with the loop reversal logic: This only works on
ranges and is activated by setting
- reverse[n] == GFC_CAN_REVERSE
+ reverse[n].backward_ok = true
The ability to reverse or not is set by previous conditions
in this dimension. If reversal is not activated, the
value GFC_DEP_BACKWARD is reset to GFC_DEP_OVERLAP. */
if (rref->u.ar.dimen_type[n] == DIMEN_RANGE
- && lref->u.ar.dimen_type[n] == DIMEN_RANGE)
+ && lref->u.ar.dimen_type[n] == DIMEN_RANGE)
{
- /* Set reverse if backward dependence and not inhibited. */
- if (reverse && reverse[n] != GFC_CANNOT_REVERSE)
- reverse[n] = (this_dep == GFC_DEP_BACKWARD) ?
- GFC_REVERSE_SET : reverse[n];
+ if (reverse)
+ {
+ if (this_dep == GFC_DEP_FORWARD
+ || this_dep == GFC_DEP_OVERLAP
+ || this_dep == GFC_DEP_ERROR)
+ reverse[n].backward_ok = false;
- /* Inhibit loop reversal if dependence not compatible. */
- if (reverse && reverse[n] != GFC_REVERSE_NOT_SET
- && this_dep != GFC_DEP_EQUAL
- && this_dep != GFC_DEP_BACKWARD
- && this_dep != GFC_DEP_NODEP)
+ if (this_dep == GFC_DEP_BACKWARD
+ || this_dep == GFC_DEP_OVERLAP
+ || this_dep == GFC_DEP_ERROR)
+ reverse[n].forward_ok = false;
+
+ if (!reverse[n].forward_ok && !reverse[n].backward_ok)
+ this_dep = GFC_DEP_OVERLAP;
+ }
+ else
{
- reverse[n] = GFC_CANNOT_REVERSE;
- if (this_dep != GFC_DEP_FORWARD)
+ /* If no intention of reversing, convert backward
+ dependence to overlap. */
+ if (this_dep == GFC_DEP_BACKWARD)
this_dep = GFC_DEP_OVERLAP;
}
-
- /* If no intention of reversing or reversing is explicitly
- inhibited, convert backward dependence to overlap. */
- if (this_dep == GFC_DEP_BACKWARD
- && (reverse == NULL || reverse[n] == GFC_CANNOT_REVERSE))
- this_dep = GFC_DEP_OVERLAP;
}
/* Overlap codes are in order of priority. We only need to
[-- Attachment #3: dependency_40.f90 --]
[-- Type: text/plain, Size: 783 bytes --]
! { dg-do run }
! PR 48955 - missing array temporary when there was both a forward
! and a backward dependency.
! Test case slightly modified from the original one by Kacper Kowalik.
program ala
implicit none
integer, parameter :: n = 6
real, dimension(n), parameter :: result = [1.,10.,30.,90.,270., 243.];
real, dimension(n) :: v0, v1
character(len=80) :: line1, line2
v0 = [1.0, 3.0, 9.0, 27.0, 81.0, 243.0]
v1 = v0
v1(2:n-1) = v1(1:n-2) + v1(3:n)
if (any(v1 /= result)) call abort
v1 = v0
v1(2:n-1) = v0(1:n-2) + v0(3:n)
if (any(v1 /= result)) call abort
v1 = v0
v1(2:n-1) = v1(3:n) + v1(1:n-2)
if (any(v1 /= result)) call abort
v1 = v0
v1(2:n-1) = v0(3:n) + v0(1:n-2)
if (any(v1 /= result)) call abort
end program ala
next reply other threads:[~2011-05-22 15:32 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-05-22 18:59 Thomas Koenig [this message]
2011-05-22 20:56 ` Paul Richard Thomas
2011-05-24 22:06 ` Paul Richard Thomas
2011-05-26 9:22 ` Thomas Koenig
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=4DD92C77.5090107@netcologne.de \
--to=tkoenig@netcologne.de \
--cc=fortran@gcc.gnu.org \
--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).