https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126646
Reshma Roy <Reshma.Roy at amd dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |Reshma.Roy at amd dot com
--- Comment #1 from Reshma Roy <Reshma.Roy at amd dot com> ---
Confirmed on the gcc trunk gcc (GCC) 17.0.0 20260811 (experimental).
The gimple IR for the first case with -O2 is as follows:
unsigned int f (unsigned int a, unsigned int b)
{
unsigned int _5;
<bb 2> [local count: 1073741824]:
a_2 = MIN_EXPR <a_1(D), 1>;
b_4 = MIN_EXPR <b_3(D), 1>;
_5 = a_2 | b_4;
return _5;
}
Here there are two MIN_EXPR.
The gimple IR for the second snippet has only one MIN_EXPR:
unsigned int g (unsigned int a, unsigned int b)
{
unsigned int _4;
<bb 2> [local count: 1073741824]:
a_3 = a_1(D) | b_2(D);
_4 = MIN_EXPR <a_3, 1>;
return _4;
}
The MIN_EXPR survives into tree-ssa so this needs a match.pd pattern.
Can this be assigned to me?
