This fixes an oversight in my GIMPLE store-merging patch, whereby masking out
the bits of a signed type obviously fails to turn it into an unsigned type...
And since output_merged_store will now convert to the unsigned bitfield type,
an existing conversion from a smaller type can be bypassed in process_store.
Tested on x86-64/Linux, applied on the mainline as obvious.
2018-06-03 Eric Botcazou <ebotca...@adacore.com>
PR tree-optimization/86034
* gimple-ssa-store-merging.c (output_merged_store): Convert the RHS to
the unsigned bitfield type in a bit insertion sequence if it does not
have a larger precision than the bitfield size.
(process_store): Also bypass widening conversions for BIT_INSERT_EXPR.
2018-06-03 Eric Botcazou <ebotca...@adacore.com>
* gcc.dg/torture/pr86034.c: New test.
--
Eric Botcazou
Index: gimple-ssa-store-merging.c
===================================================================
--- gimple-ssa-store-merging.c (revision 261101)
+++ gimple-ssa-store-merging.c (working copy)
@@ -3778,7 +3778,14 @@ imm_store_chain_info::output_merged_stor
const HOST_WIDE_INT end_gap
= (try_bitpos + try_size) - (info->bitpos + info->bitsize);
tree tem = info->ops[0].val;
- if ((BYTES_BIG_ENDIAN ? start_gap : end_gap) > 0)
+ if (TYPE_PRECISION (TREE_TYPE (tem)) <= info->bitsize)
+ {
+ tree bitfield_type
+ = build_nonstandard_integer_type (info->bitsize,
+ UNSIGNED);
+ tem = gimple_convert (&seq, loc, bitfield_type, tem);
+ }
+ else if ((BYTES_BIG_ENDIAN ? start_gap : end_gap) > 0)
{
const unsigned HOST_WIDE_INT imask
= (HOST_WIDE_INT_1U << info->bitsize) - 1;
@@ -4270,13 +4277,12 @@ pass_store_merging::process_store (gimpl
|| !multiple_p (bitpos, BITS_PER_UNIT))
&& const_bitsize <= 64)
{
- /* Bypass a truncating conversion to the bit-field type. */
+ /* Bypass a conversion to the bit-field type. */
if (is_gimple_assign (def_stmt) && CONVERT_EXPR_CODE_P (rhs_code))
{
tree rhs1 = gimple_assign_rhs1 (def_stmt);
if (TREE_CODE (rhs1) == SSA_NAME
- && INTEGRAL_TYPE_P (TREE_TYPE (rhs1))
- && const_bitsize <= TYPE_PRECISION (TREE_TYPE (rhs1)))
+ && INTEGRAL_TYPE_P (TREE_TYPE (rhs1)))
rhs = rhs1;
}
rhs_code = BIT_INSERT_EXPR;
/* PR tree-optimization/86034 */
/* Testcase by Zhendong Su <s...@cs.ucdavis.edu> */
/* { dg-do run } */
struct A
{
int b;
int c:24;
int d:10;
int e;
} f;
int g;
void h ()
{
struct A i = { 0, 0, -1, 0 };
L:
f = i;
i.d = 0;
if (g < 0)
goto L;
}
int main (void)
{
h ();
if (f.e != 0)
__builtin_abort ();
return 0;
}