On Sat, 2025-07-05 at 17:55 +0800, Xi Ruoyao wrote:
> On Sat, 2025-07-05 at 11:20 +0800, Lulu Cheng wrote:
> > For the gcc.target/loongarch/bitwise-shift-reassoc-clobber.c,
> > some extensions are eliminated in ext_dce in commit r16-1835.
> >
> > This will result in the following rtx being generated in the
> > combine pass:
> > (insn 12 10 15 2 (set (reg/v:DI 23 $r23 [ x ])
> > (sign_extend:DI (plus:SI
> > (subreg:SI
> > (and:DI (ashift:DI
> > (reg/v:DI 23 $r23 [ x ])
> > (const_int 3 [0x3]))
> > (const_int 2208 [0x8a0])) 0)
> > (reg:SI 23 $r23 [ x ]))))
> > "test.c":7:7 266 {and_alsl_reversesi_extended}
> > (nil))
> >
> > However, in the template '<optab>_alsl_reversesi_extended',
> > operands[0] and operands[4] cannot be assigned to the same hardware
> > register, which causes ICE during the reload pass.
>
> Oops. We have:
>
> Trying 10 -> 12:
> 10: r91:DI=sign_extend($r23:DI<<0x3&0x8a0#0+$r23:SI)
> REG_DEAD $r23:DI
> 12: $r23:DI=r91:DI
> REG_DEAD r91:DI
> Successfully matched this instruction:
> (set (reg/v:DI 23 $r23 [ x ])
> (sign_extend:DI (plus:SI (subreg:SI (and:DI (ashift:DI (reg/v:DI
> 23 $r23 [ x ])
> (const_int 3 [0x3]))
> (const_int 2208 [0x8a0])) 0)
> (reg:SI 23 $r23 [ x ]))))
> allowing combination of insns 10 and 12
>
> But the instruction description already has earlyclobber:
>
> [(set (match_operand:DI 0 "register_operand" "=&r")
> (sign_extend:DI
> (plus:SI
> (subreg:SI
> (any_bitwise:DI
> (ashift:DI
> (match_operand:DI 1 "register_operand" "r0")
> (match_operand:SI 2 "const_immalsl_operand" ""))
> (match_operand:DI 3 "const_int_operand" "i"))
> 0)
> (match_operand:SI 4 "register_operand" "r"))))]
>
> To me if we need to "fix up" this in our machine description we'd need
> to check REGNO everywhere we have an earlyclobber.
>
> So Jeff & Richard: shouldn't combine simply reject those cases where
> substitute a hard register in would violate the earlyclobber?
Possibly this is https://gcc.gnu.org/PR101882. Specifically comment 5
from Segher:
"The LRA change is correct AFAICS. But combine makes a change that
violates the earlyclobber... I need to do something about that, too."
--
Xi Ruoyao <xry...@xry111.site>
