https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93231
Bug ID: 93231
Summary: [10 Regression] ICEs since rr280132
Product: gcc
Version: 10.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: jakub at gcc dot gnu.org
Target Milestone: ---
int ctz2 (int x)
{
static const char table[32] =
{
0, 1, 28, 2, 29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4, 8,
31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6, 11, 5, 10, 9
};
return table[((int)((x & -x) * -0x077CB531)) >> 27];
}
ICEs because
unsigned HOST_WIDE_INT val = tree_to_uhwi (mulc);
is used without checking that the INTEGER_CST fits into uhwi.
If we don't want to support signed x, we should start the function by
verification that inner_type is INTEGRAL_TYPE_P which is TYPE_UNSIGNED, if we
do want to support even signed values, it needs to be tweaked differently.
Similarly,
int ctz3 (unsigned x)
{
static const char table[32] =
{
0, 1, 28, 2, 29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4, 8,
31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6, 11, 5, 10, 9
};
return table[((unsigned)((x & -x) * 0x077CB531U)) >> -27];
}
ICEs because -27 doesn't fit into uhwi. We should just punt if it doesn't.
I'm also surprised by
/* Check the array is not wider than integer type and the input is a 32-bit
or 64-bit type. */
if (TYPE_PRECISION (type) > 32)
return false;
because the comment doesn't match what the check is doing, either you want an
array with 32-bit or smaller elts, then the comment should match that, or
you care about integers and then you should compare against TYPE_PRECISION
(integer_type_node).
