https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78037
Uroš Bizjak <ubizjak at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|ASSIGNED |NEW
CC| |rguenth at gcc dot gnu.org
Component|target |tree-optimization
Assignee|ubizjak at gmail dot com |unassigned at gcc dot
gnu.org
--- Comment #10 from Uroš Bizjak <ubizjak at gmail dot com> ---
Actually, the problem already starts with tree Early-VRP pass. Consider
following testcase:
int foo (int x)
{
return __builtin_ctz (x) & 0x1f;
}
And these definitions in i386.h:
/* The value at zero is only defined for the BMI instructions
LZCNT and TZCNT, not the BSR/BSF insns in the original isa. */
#define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
((VALUE) = GET_MODE_BITSIZE (MODE), TARGET_BMI ? 1 : 0)
#define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
((VALUE) = GET_MODE_BITSIZE (MODE), TARGET_LZCNT ? 1 : 0)
compile with gcc -O2 -mbmi, _.evrp dump shows:
Value ranges after Early VRP:
x.0_1: [0, +INF]
_2: [0, 31]
x_3(D): VARYING
_4: [0, 31]
foo (int x)
{
unsigned int x.0_1;
int _2;
int _4;
<bb 2>:
x.0_1 = (unsigned int) x_3(D);
_2 = __builtin_ctz (x.0_1);
_4 = _2;
return _4;
}
It looks that VRP pass doesn't account for CTZ_DEFINED_VALUE_AT_ZERO and
blindly assumes undefined value at zero.
Unassign and recategorize PR as tree optimization problem.
