> The following committed patch solves
>
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94185
>
> The patch was successfully bootstrapped and tested on x86-64.
This patch creates unoptimal sequence in reload:
(insn 64 63 248 10 (set (reg:DI 0 ax [161])
(zero_extend:DI (mem/c:SI (plus:DI (reg/f:DI 6 bp)
(const_int -2152 [0xfffffffffffff798])) [7
%sfp+-2104 S4 A64]))) "pr94185.C":25:7 114 {*zero_extendsidi2}
(nil))
(insn 248 64 65 10 (set (mem/c:DI (plus:DI (reg/f:DI 6 bp)
(const_int -2152 [0xfffffffffffff798])) [7 %sfp+-2104 S8 A64])
(reg:DI 0 ax [161])) "pr94185.C":25:7 66 {*movdi_internal}
(nil))
using a temporary reg, resulting in:
movl -2152(%rbp), %eax # 64 [c=9 l=6] *zero_extendsidi2/3
movq %rax, -2152(%rbp) # 248 [c=4 l=7] *movdi_internal/5
Please note that the insn pattern could handle matched offsetable
memory in alternative 4:
(define_insn "*zero_extendsidi2"
[(set (match_operand:DI 0 "nonimmediate_operand"
"=r,?r,?o,r ,o,?*y,?!*y,$r,$v,$x,*x,*v,*r,*k")
(zero_extend:DI
(match_operand:SI 1 "x86_64_zext_operand"
"0 ,rm,r ,rmWz,0,r ,m ,v ,r ,m ,*x,*v,*k,*km")))]
memory operand is later split with:
(define_split
[(set (match_operand:DI 0 "memory_operand")
(zero_extend:DI (match_operand:SI 1 "memory_operand")))]
"reload_completed"
[(set (match_dup 4) (const_int 0))]
"split_double_mode (DImode, &operands[0], 1, &operands[3], &operands[4]);")
And would result in a single SImode move of 0 to the high part of the
memory location.
Uros.
