[Bug target/98737] Atomic operation on x86 no optimized to use flags

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98737

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2021-01-19
           Keywords|                            |missed-optimization
     Ever confirmed|0                           |1
             Target|x86                         |x86_64-*-* i?86-*-*

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  Probably difficult to achieve since the atomics map to
parallel/UNSPEC:

(insn 7 6 8 (parallel [
            (set (reg:DI 83 [ _2 ])
                (unspec_volatile:DI [
                        (mem/v:DI (symbol_ref:DI ("a") [flags 0x2]  <var_decl
0x7ffff7ff5b40 a>) [-1  S8 A64])
                        (const_int 3 [0x3])
                    ] UNSPECV_XCHG))
            (set (mem/v:DI (symbol_ref:DI ("a") [flags 0x2]  <var_decl
0x7ffff7ff5b40 a>) [-1  S8 A64])
                (plus:DI (mem/v:DI (symbol_ref:DI ("a") [flags 0x2]  <var_decl
0x7ffff7ff5b40 a>) [-1  S8 A64])
                    (reg:DI 86)))
            (clobber (reg:CC 17 flags))
        ]) "t.c":5:10 -1
     (nil))