diff options
| author | Richard Biener <rguenther@suse.de> | 2020-10-15 09:10:40 +0200 |
|---|---|---|
| committer | Richard Biener <rguenther@suse.de> | 2020-10-15 10:54:24 +0200 |
| commit | 28290cb50c7dbf87458befeb3e295b5cb13560b5 (patch) | |
| tree | 5ddc99caf2358cf5ec76d0136e54f00256b62430 /gcc/tree-data-ref.c | |
| parent | ee21a5f3979adab95069f363b078812a54c5b59f (diff) | |
| download | gcc-28290cb50c7dbf87458befeb3e295b5cb13560b5.zip gcc-28290cb50c7dbf87458befeb3e295b5cb13560b5.tar.gz gcc-28290cb50c7dbf87458befeb3e295b5cb13560b5.tar.bz2 | |
tree-optimization/97482 - fix split_constant_offset of nop-conversions
split_constant_offset is confused about a nop-conversion from
unsigned long to sizetype and tries to prove non-overflowing
of the inner operation. Obviously the conversion could have been
elided so make sure split_constant_offset handles this properly.
It also makes sure that convert_to_ptrofftype does not introduce
conversions not necessary which in this case is the source for
the unnecessary conversion.
2020-10-15 Richard Biener <rguenther@suse.de>
PR tree-optimization/97482
* tree-data-ref.c (split_constant_offset_1): Handle
trivial conversions better.
* fold-const.c (convert_to_ptrofftype_loc): Elide conversion
if the offset is already ptrofftype_p.
* gcc.dg/vect/pr97428.c: New testcase.
Diffstat (limited to 'gcc/tree-data-ref.c')
| -rw-r--r-- | gcc/tree-data-ref.c | 4 |
1 files changed, 3 insertions, 1 deletions
diff --git a/gcc/tree-data-ref.c b/gcc/tree-data-ref.c index 5505ba4..3bf460c 100644 --- a/gcc/tree-data-ref.c +++ b/gcc/tree-data-ref.c @@ -754,7 +754,9 @@ split_constant_offset_1 (tree type, tree op0, enum tree_code code, tree op1, && TYPE_PRECISION (type) >= TYPE_PRECISION (itype) && (POINTER_TYPE_P (type) || INTEGRAL_TYPE_P (type))) { - if (INTEGRAL_TYPE_P (itype) && TYPE_OVERFLOW_WRAPS (itype)) + if (INTEGRAL_TYPE_P (itype) && TYPE_OVERFLOW_WRAPS (itype) + && (TYPE_PRECISION (type) > TYPE_PRECISION (itype) + || TYPE_UNSIGNED (itype) != TYPE_UNSIGNED (type))) { /* Split the unconverted operand and try to prove that wrapping isn't a problem. */ |