https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119474
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
So one reason is TYPE_ADDR_SPACE isn't prevalent where it should be - ARRAY_REF
results are one common issue (and we lack verification here, IIRC):
<array_ref 0x7ffff6fa43b8
type <integer_type 0x7ffff6e255e8 int sizes-gimplified public type_6 SI
size <integer_cst 0x7ffff6e19a98 constant 32>
unit-size <integer_cst 0x7ffff6e19ab0 constant 4>
align:32 warn_if_not_align:0 symtab:0 alias-set 1 canonical-type
0x7ffff6e255e8 precision:32 min <integer_cst 0x7ffff6e19a50 -2147483648> max
<integer_cst 0x7ffff6e19a68 2147483647>
pointer_to_this <pointer_type 0x7ffff6e2c888>>
arg:0 <component_ref 0x7ffff6fa3870
type <array_type 0x7ffff6fa8498 type <integer_type 0x7ffff6fa83f0 int
address-space-4>
is that address-space somehow introduced "late" via some OpenACC magic?
As said, we're lacking IL checking here, if only because it would trip over all
over the place.
