[PATCH] Fix up WIDEN_MULT_EXPR (PR middle-end/57251)
Jakub Jelinek
jakub@redhat.com
Mon May 13 16:40:00 GMT 2013
Hi!
On the following testcase, the widen multiplies etc. is introduced so late
that no forwprop or ccp follows it. The WIDEN_MULT_EXPR tries hard not to
use widening multiply if both arguments are INTEGER_CSTs, but if they
are e.g. SSA_NAMEs that expand_normal into CONST_INT, we can still ICE.
The following patch attempts to handle those cases. Bootstrapped/regtested
on x86_64-linux and i686-linux, ok for trunk/4.8?
2013-05-13 Jakub Jelinek <jakub@redhat.com>
PR middle-end/57251
* expr.c (expand_expr_real_2) <case WIDEN_MULT_EXPR>: Handle
the case when both op0 and op1 have VOIDmode.
* gcc.dg/torture/pr57251.c: New test.
--- gcc/expr.c.jj 2013-05-07 10:27:07.000000000 +0200
+++ gcc/expr.c 2013-05-13 12:01:49.339087536 +0200
@@ -8390,6 +8390,15 @@ expand_expr_real_2 (sepops ops, rtx targ
else
expand_operands (treeop0, treeop1, NULL_RTX, &op1, &op0,
EXPAND_NORMAL);
+ /* op0 and op1 might still be constant, despite the above
+ != INTEGER_CST check. Handle it. */
+ if (GET_MODE (op0) == VOIDmode && GET_MODE (op1) == VOIDmode)
+ {
+ op0 = convert_modes (innermode, mode, op0, true);
+ op1 = convert_modes (innermode, mode, op1, false);
+ return REDUCE_BIT_FIELD (expand_mult (mode, op0, op1,
+ target, unsignedp));
+ }
goto binop3;
}
}
@@ -8412,6 +8421,19 @@ expand_expr_real_2 (sepops ops, rtx targ
{
expand_operands (treeop0, treeop1, NULL_RTX, &op0, &op1,
EXPAND_NORMAL);
+ /* op0 and op1 might still be constant, despite the above
+ != INTEGER_CST check. Handle it. */
+ if (GET_MODE (op0) == VOIDmode && GET_MODE (op1) == VOIDmode)
+ {
+ widen_mult_const:
+ op0 = convert_modes (innermode, mode, op0, zextend_p);
+ op1
+ = convert_modes (innermode, mode, op1,
+ TYPE_UNSIGNED (TREE_TYPE (treeop1)));
+ return REDUCE_BIT_FIELD (expand_mult (mode, op0, op1,
+ target,
+ unsignedp));
+ }
temp = expand_widening_mult (mode, op0, op1, target,
unsignedp, this_optab);
return REDUCE_BIT_FIELD (temp);
@@ -8424,9 +8446,14 @@ expand_expr_real_2 (sepops ops, rtx targ
op0 = expand_normal (treeop0);
if (TREE_CODE (treeop1) == INTEGER_CST)
op1 = convert_modes (innermode, mode,
- expand_normal (treeop1), unsignedp);
+ expand_normal (treeop1),
+ TYPE_UNSIGNED (TREE_TYPE (treeop1)));
else
op1 = expand_normal (treeop1);
+ /* op0 and op1 might still be constant, despite the above
+ != INTEGER_CST check. Handle it. */
+ if (GET_MODE (op0) == VOIDmode && GET_MODE (op1) == VOIDmode)
+ goto widen_mult_const;
temp = expand_binop (mode, other_optab, op0, op1, target,
unsignedp, OPTAB_LIB_WIDEN);
hipart = gen_highpart (innermode, temp);
--- gcc/testsuite/gcc.dg/torture/pr57251.c.jj 2013-05-13 11:46:34.937151578 +0200
+++ gcc/testsuite/gcc.dg/torture/pr57251.c 2013-05-13 11:54:23.070414122 +0200
@@ -0,0 +1,12 @@
+/* PR middle-end/57251 */
+/* { dg-do compile } */
+/* { dg-options "-ftracer" } */
+
+short a, b;
+int
+f (void)
+{
+ long long i = 2;
+ a ? f () ? : 0 : b--;
+ b &= i *= a |= 0;
+}
Jakub
More information about the Gcc-patches
mailing list