[COMMITTED] PR tree-optimization/106556 - Evaluate condition arguments with the correct type.
Commit Message
when evaluating a COND_EXPR, we need to evaluate both operands. With the
recent changes to floating point, we missed that we are accidentally
using the LHS range type for the operands.. that was fine when
everything was an irange... but no so any more.
This patch simply uses the right range type for the temporary ranges.
Bootstrapped on x86_64-pc-linux-gnu with no regressions. Pushed.
Andrew
commit 6d57764711d3b4441d4d99876d9d3370bdec6284
Author: Andrew MacLeod <amacleod@redhat.com>
Date: Mon Aug 8 15:13:51 2022 -0400
Evaluate condition arguments with the correct type.
Processing of a cond_expr requires that a range of the correct type for the
operands of the cond_expr is passed in.
PR tree-optimization/106556
gcc/
* gimple-range-gori.cc (gori_compute::condexpr_adjust): Use the
type of the cond_expr operands being evaluted.
gcc/testsuite/
* gfortran.dg/pr106556.f90: New.
@@ -1398,16 +1398,17 @@ gori_compute::condexpr_adjust (vrange &r1, vrange &r2, gimple *, tree cond,
}
// Now solve for SSA1 or SSA2 if they are in the dependency chain.
- Value_Range tmp (type);
if (ssa1 && in_chain_p (ssa1, cond_name))
{
- if (compute_operand_range (tmp, def_stmt, cond_true, ssa1, src))
- r1.intersect (tmp);
+ Value_Range tmp1 (TREE_TYPE (ssa1));
+ if (compute_operand_range (tmp1, def_stmt, cond_true, ssa1, src))
+ r1.intersect (tmp1);
}
if (ssa2 && in_chain_p (ssa2, cond_name))
{
- if (compute_operand_range (tmp, def_stmt, cond_false, ssa2, src))
- r2.intersect (tmp);
+ Value_Range tmp2 (TREE_TYPE (ssa2));
+ if (compute_operand_range (tmp2, def_stmt, cond_false, ssa2, src))
+ r2.intersect (tmp2);
}
if (idx)
{
new file mode 100644
@@ -0,0 +1,10 @@
+! { dg-do compile }
+! { dg-options "-O1 -fnon-call-exceptions -ftree-loop-if-convert" }
+
+
+program p
+ real :: a(2)
+
+ a(:) = 1.0
+ if (minloc (a, dim = 1).ne.1) STOP 1
+end