https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90530
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|middle-end |target
--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> ---
I wonder what the difference to PA is between the reloaded
+(insn 20 181 182 2 (set (reg:DI 50 %fr22 [95])
+ (mult:DI (zero_extend:DI (reg:SI 52 %fr24))
+ (reg:DI 53 %fr25))) 179 {*pa.md:5338}
and the original
-(insn 20 10 21 2 (set (reg:DI 95)
- (mult:DI (zero_extend:DI (subreg:SI (reg/f:DI 146) 4))
- (const_int 4294967295 [0xffffffff]))) 179 {*pa.md:5338}
most definitely a reload as
+(insn 177 176 178 2 (set (reg:SI 52 %fr24)
+ (subreg:SI (reg:DI 51 %fr23) 4)) -1
+ (nil))
which isn't recognized isn't good. I don't know PA on what it can
and what not but these kind of sets from subregs appear during RTL
expansion for
;; _26 = _27 * 18446744073709551613;
and RTL opts push them into the mult.
To me it looks like a testcase like
int a;
__SIZE_TYPE__ foo ()
{
return (__SIZE_TYPE__)&a * 18446744073709551613ull;
}
should trigger the same issue. It creates the same RTL with
-O2 -fno-tree-vrp -fdisable-tree-evrp but the doesn't ICE beause
we reload
(insn 16 10 19 2 (set (reg:DI 77)
(mult:DI (zero_extend:DI (subreg:SI (reg:DI 81) 4))
(const_int 4294967293 [0xfffffffd]))) "t.c":6:14 179 {*pa.md:5338}
via memory as
(insn 47 52 48 2 (set (mem/c:DI (plus:DI (reg/f:DI 30 %r30)
(const_int -56 [0xffffffffffffffc8])) [0 S8 A64])
(reg:DI 19 %r19)) "t.c":6:14 128 {*pa.md:4178}
(nil))
(insn 48 47 16 2 (set (reg:DI 51 %fr23)
(mem/c:DI (plus:DI (reg/f:DI 30 %r30)
(const_int -56 [0xffffffffffffffc8])) [0 S8 A64])) "t.c":6:14
128 {*pa.md:4178}
(nil))
(insn 16 48 53 2 (set (reg:DI 52 %fr24)
(mult:DI (zero_extend:DI (reg:SI 51 %fr23))
(reg:DI 52 %fr24))) "t.c":6:14 179 {*pa.md:5338}
(nil))
not sure if we're just lucky here or what.
Anyhow this is IMHO clearly a target issue.
