Like the check that was added in
r16-6374-g2680785a6d14d7d78550edb8361f3b47eb5c4cb6,
except this is for the new code that was added with
r16-6436-g92f2a86794ad88d773b102.
Basically the code only expects to work with scalar integer modes but forgot
to check that.
Pushed as obvious after bootstrap/test on x86_64-linux-gnu.
PR rtl-optimization/123491
gcc/ChangeLog:
* ifcvt.cc (noce_try_cond_zero_arith): Reject non-scalar
integer modes for the inner mode of the subreg.
Signed-off-by: Andrew Pinski <[email protected]>
---
gcc/ifcvt.cc | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/gcc/ifcvt.cc b/gcc/ifcvt.cc
index e27d9cfc131..c029a583e84 100644
--- a/gcc/ifcvt.cc
+++ b/gcc/ifcvt.cc
@@ -3213,9 +3213,13 @@ noce_try_cond_zero_arith (struct noce_if_info *if_info)
afterwards. */
if (GET_CODE (tmp) != SUBREG || !subreg_lowpart_p (tmp))
return false;
- start_sequence ();
-
tmp = SUBREG_REG (tmp);
+ /* Only handle integer scalar modes for the inner mode of
+ the subreg. */
+ if (!SCALAR_INT_MODE_P (GET_MODE (tmp)))
+ return false;
+
+ start_sequence ();
target = gen_reg_rtx (GET_MODE (tmp));
target = noce_emit_cmove (if_info, target, code,
XEXP (cond, 0), XEXP (cond, 1),
--
2.43.0
