https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120629
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |amacleod at redhat dot com
--- Comment #17 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
In #c16 testcase I see in optimized dump with alias:
# RANGE [irange] unsigned int [1, 9] MASK 0xf VALUE 0x0
_46 = (unsigned int) ivtmp.24_52;
_I_lsm.20_47 = __gcov0._ZN13ipa_predicate9stream_inEP15lto_input_block[3];
# RANGE [irange] unsigned int [0, 7]
_54 = 8 - _46;
# RANGE [irange] sizetype [0, 7] MASK 0x7 VALUE 0x0
_55 = (sizetype) _54;
# RANGE [irange] sizetype [1, 8] MASK 0xf VALUE 0x0
_56 = _55 + 1;
# RANGE [irange] sizetype [4, 32] MASK 0x3c VALUE 0x0
_57 = _56 * 4;
# RANGE [irange] sizetype [1, 8] MASK 0xf VALUE 0x0
_58 = (sizetype) k_10;
# RANGE [irange] sizetype [4, 32] MASK 0x3c VALUE 0x0
_59 = _58 * 4;
# PT = anything
_60 = this_11(D) + _59;
# USE = anything
# CLB = anything
__builtin_memset (_60, 0, _57);
and during RTL expansion of __builtin_memset we reach determine_block_size
which does
int_range_max r;
tree tmin, tmax;
gimple *cg = currently_expanding_gimple_stmt;
get_range_query (cfun)->range_of_expr (r, len, cg);
range_type = get_legacy_range (r, tmin, tmax);
But, somehow instead of returning the [4, 32] range it returns [32, 32].
