[committed] Fix !$omp do translation (PR fortran/46753)
Jakub Jelinek
jakub@redhat.com
Thu Dec 2 14:46:00 GMT 2010
Hi!
During 4.4 development the creation of conditions were incorrectly changed
from build2 to fold_build2. This is wrong, because comparisons with
the highest or lowest type's value can be folded into != test, which is not
valid OpenMP condition (the middle-end code wants to know if the condition
is </<= or >/>=).
Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux,
committed (so far) on the trunk.
2010-12-02 Jakub Jelinek <jakub@redhat.com>
PR fortran/46753
* trans-openmp.c (gfc_trans_omp_do): Use build2_loc instead of
fold_build2_loc for OMP_FOR conditions.
* libgomp.fortran/pr46753.f90: New test.
--- gcc/fortran/trans-openmp.c.jj 2010-09-14 15:24:43.000000000 +0200
+++ gcc/fortran/trans-openmp.c 2010-12-02 11:17:55.000000000 +0100
@@ -1262,10 +1262,10 @@ gfc_trans_omp_do (gfc_code *code, stmtbl
if (simple)
{
TREE_VEC_ELT (init, i) = build2_v (MODIFY_EXPR, dovar, from);
- TREE_VEC_ELT (cond, i) = fold_build2_loc (input_location, simple > 0
- ? LE_EXPR : GE_EXPR,
- boolean_type_node, dovar,
- to);
+ /* The condition should not be folded. */
+ TREE_VEC_ELT (cond, i) = build2_loc (input_location, simple > 0
+ ? LE_EXPR : GE_EXPR,
+ boolean_type_node, dovar, to);
TREE_VEC_ELT (incr, i) = fold_build2_loc (input_location, PLUS_EXPR,
type, dovar, step);
TREE_VEC_ELT (incr, i) = fold_build2_loc (input_location,
@@ -1290,9 +1290,10 @@ gfc_trans_omp_do (gfc_code *code, stmtbl
count = gfc_create_var (type, "count");
TREE_VEC_ELT (init, i) = build2_v (MODIFY_EXPR, count,
build_int_cst (type, 0));
- TREE_VEC_ELT (cond, i) = fold_build2_loc (input_location, LT_EXPR,
- boolean_type_node,
- count, tmp);
+ /* The condition should not be folded. */
+ TREE_VEC_ELT (cond, i) = build2_loc (input_location, LT_EXPR,
+ boolean_type_node,
+ count, tmp);
TREE_VEC_ELT (incr, i) = fold_build2_loc (input_location, PLUS_EXPR,
type, count,
build_int_cst (type, 1));
--- libgomp/testsuite/libgomp.fortran/pr46753.f90.jj 2010-11-19 19:58:10.083000000 +0100
+++ libgomp/testsuite/libgomp.fortran/pr46753.f90 2010-12-02 11:37:48.000000000 +0100
@@ -0,0 +1,17 @@
+! PR fortran/46753
+! { dg-do run }
+
+ integer :: i, j
+ j = 0
+!$omp parallel do reduction(+:j)
+ do i = 2147483636, 2147483646
+ j = j + 1
+ end do
+ if (j.ne.11) call abort
+ j = 0
+!$omp parallel do reduction(+:j)
+ do i = -2147483637, -2147483647, -1
+ j = j + 1
+ end do
+ if (j.ne.11) call abort
+end
Jakub
More information about the Gcc-patches
mailing list