[Bug middle-end/38126] suboptimal code for (a && b || !a && !b)

[Bug middle-end/38126] suboptimal code for (a && b || !a && !b)

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38126

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-02-01
     Ever Confirmed|0                           |1

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-02-01 22:37:56 UTC ---
Confirmed.

[Bug middle-end/38126] suboptimal code for (a && b || !a && !b)

[svfuerst at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38126

Steven Fuerst <svfuerst at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |svfuerst at gmail dot com

--- Comment #3 from Steven Fuerst <svfuerst at gmail dot com> 2012-02-02 05:25:34 UTC ---
For the first pair of functions, this is even faster:

    neg %rdi
    sbb %eax, %eax
    neg %rsi
    adc $1, %eax
    and $1, %eax
    retq

[Bug middle-end/38126] suboptimal code for (a && b || !a && !b)

[svfuerst at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38126

--- Comment #4 from Steven Fuerst <svfuerst at gmail dot com> 2012-02-02 06:11:27 UTC ---
Two more cases for simple boolean logic optimizations.

gcc-4.7 produces with -O3 for

int test_and(long long x, long long y)
{
    return x && y;
}

    test   %rsi, %rsi
    setne  %dl
    xor    %eax, %eax
    test   %rdi, %rdi
    setne  %al
    and    %edx, %eax
    retq

Whereas this is faster:

    neg %rdi
    sbb %rdi, %rdi
    xor %eax, %eax
    and %rsi, %rdi
    setne %al
    retq

Also

int test_other(long long x, long long y)
{
    return !x && y; /* or !(x || !y) */
}

gives

    test   %rsi,%rsi
    setne  %dl
    xor    %eax,%eax
    test   %rdi,%rdi
    sete   %al
    and    %edx,%eax
    retq

when
    sub $1, %rsi
    sbb %rsi, %rsi
    xor %eax, %eax
    or %rdi, %rsi
    sete %al
    retq
is faster.

[Bug middle-end/38126] suboptimal code for (a && b || !a && !b)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=38126

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
For the original testcase with GCC 7, we get the same(similar enough) code gen
for both functions now.
foo:
.LFB0:
        .cfi_startproc
        testq   %rdi, %rdi
        setne   %al
        testq   %rsi, %rsi
        sete    %dl
        xorl    %edx, %eax
        movzbl  %al, %eax
        ret
bar:
.LFB1:
        .cfi_startproc
        testq   %rdi, %rdi
        sete    %al
        testq   %rsi, %rsi
        setne   %dl
        xorl    %edx, %eax
        movzbl  %al, %eax
        ret

[Bug target/38126] suboptimal code for (a && b || !a && !b)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=38126

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64
          Component|middle-end                  |target
   Target Milestone|---                         |13.0

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Martin Sebor from comment #1)
> Code involving bool variables is similarly suboptimal:

The bool part was fixed for GCC 13 by r13-1779-g375668e0508fbe .

The rest is a target issue dealing with using sbb more.

[Bug target/38126] suboptimal code for (a && b || !a && !b)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=38126

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|13.0                        |---