[Bug c/96632] New: missed-optimization with conditionally unsetting bits in memory.

[Bug c/96632] New: missed-optimization with conditionally unsetting bits in memory.

[goldstein.w.n at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96632

            Bug ID: 96632
           Summary: missed-optimization with conditionally unsetting bits
                    in memory.
           Product: gcc
           Version: 10.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: goldstein.w.n at gmail dot com
  Target Milestone: ---

https://stackoverflow.com/questions/63432157/why-does-gcc-use-btq-in-conjunction-with-btcq-when-conditionally-set-a-bit-in-a

Basically:

void __attribute__((noinline))
cond_unset_bit(uint64_t * v, uint32_t b) {
    if(__builtin_expect(!!(*v & ((1UL) << b)), 1)) {
        *v ^= ((1UL) << b);
    }
}

Compiles to:

cond_unset_bit(unsigned long*, unsigned int):
        movq    (%rdi), %rax
        btq     %rsi, %rax
        jnc     .L6
        btcq    %rsi, %rax
        movq    %rax, (%rdi)
.L6:
        ret

The btq instruction is unnecessary.

cond_unset_bit(unsigned long*, unsigned int):
        movq    (%rdi), %rax
        btcq    %rsi, %rax
        jnc     .L6
        movq    %rax, (%rdi)
.L6:
        ret


Accomplishes the same thing without the btq instruction.

[Bug rtl-optimization/96632] missed-optimization with conditionally unsetting bits in memory.

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96632

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2020-08-25
          Component|target                      |rtl-optimization

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  A bit tricky to do I guess.  There's no representation on
GIMPLE we could use so it's left for RTL (unless we add bit
set/clear/flip-and-test internal functions).

[Bug rtl-optimization/96632] missed-optimization with conditionally unsetting bits in memory.

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96632

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2020-08-25 00:00:00         |2021-8-14
           Severity|normal                      |enhancement

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I notice only ICC can do this.