https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109960
Daniel Henrique Barboza <daniel.barboza at oss dot qualcomm.com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |daniel.barboza at oss dot
qualcomm
| |.com
--- Comment #13 from Daniel Henrique Barboza <daniel.barboza at oss dot
qualcomm.com> ---
I tried this one a little bit with match.pd with a simplified example that
doesn't have mem pointers (since match.pd can't deal with them):
```
static inline bool b1(unsigned a)
{
return (a&1);
}
static inline bool b2(unsigned a)
{
return (a&2);
}
static inline bool b5(unsigned a)
{
return (a&5);
}
static inline bool b9(unsigned a)
{
return (a&9);
}
bool bar3 (unsigned a)
{
return b1 (a) || b2(a) || b5(a) || b9 (a) ;
}
```
I tried to convince match.pd that this ifcond format can be converted (from
optimized):
;; basic block 2
_6 = (_BoolD.4579) a_3(D);
if (_6 != 0)
goto <bb 4>; [50.00%]
else
goto <bb 3>; [50.00%]
;; basic block 3
# RANGE [irange] unsigned int [0, 0][2, 14] MASK 0xe VALUE 0x0
_10 = a_3(D) & 15;
_11 = _10 != 0;
;; basic block 4
# prephitmp_5 = PHI <_6(2), _11(3)>
# VUSE <.MEM_2(D)>
return prephitmp_5;
But even with this very crude an unsafe pattern I couldn't convince it:
(simplify
(cond (ne (convert@0 @A) integer_zerop)
@0 (ne (bit_and @A INTEGER_CST@1) integer_zerop))
(ne (bit_and @A (bit_ior @1 { build_one_cst (TREE_TYPE (@A)); }))
{ build_zero_cst (TREE_TYPE (@A)); } ))
If this happens to be a hot code I suppose we can throw stuff at
tree-ssa-phiopt and friends until it works. Unless there's something more
fundamental preventing this from happening ...
