[Bug rtl-optimization/50440] New: 128 bit unsigned int subtraction generates too many register moves

[Bug rtl-optimization/50440] New: 128 bit unsigned int subtraction generates too many register moves

[schnetter at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50440

             Bug #: 50440
           Summary: 128 bit unsigned int subtraction generates too many
                    register moves
    Classification: Unclassified
           Product: gcc
           Version: 4.6.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: schnetter@gmail.com


I want to perform 128 bit integer arithmetic, and I am declaring my type like
this:

{{{
typedef unsigned int uint128_t __attribute__((mode(TI)));
uint128_t add (uint128_t x, uint128_t y) { return x+y; }
uint128_t sub (uint128_t x, uint128_t y) { return x-y; }
}}}

This is on an Intel Xeon processor in x86_64 mode. I build with the command

gcc-4.6.1 -O3 -march=native -S sub128.c

and I find that, while the "add" routine looks optimal, the "sub" routine has
several unnecessary register moves:

{{{
add:
    movq    %rdx, %rax
    movq    %rcx, %rdx
    addq    %rdi, %rax
    adcq    %rsi, %rdx
    ret
sub:
    movq    %rsi, %r10
    movq    %rdi, %rsi
    subq    %rdx, %rsi
    movq    %r10, %rdi
    sbbq    %rcx, %rdi
    movq    %rsi, %rax
    movq    %rdi, %rdx
    ret
}}}

[Bug target/50440] 128 bit unsigned int subtraction generates too many register moves

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50440

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-*-*
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |missed-optimization
   Last reconfirmed|                            |2011-09-25
          Component|rtl-optimization            |target
     Ever Confirmed|0                           |1

--- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-09-25 10:20:19 UTC ---
Seems to be an artifact of the *subti3_doubleword splitter.

[Bug target/50440] 128 bit unsigned int subtraction generates too many register moves

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=50440

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |sayle at gcc dot gnu.org

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The trunk we get:
```
sub:
        xchgq   %rdi, %rsi
        movq    %rdx, %r8
        movq    %rsi, %rax
        movq    %rdi, %rdx
        subq    %r8, %rax
        sbbq    %rcx, %rdx
        ret
```

There still seems to be too many movq going on compared to LLVM even.