[Bug c/53353] New: Missing error/warning when using __int128_t with incorrect asm register constraints

[Bug c/53353] New: Missing error/warning when using __int128_t with incorrect asm register constraints

[svfuerst at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53353

             Bug #: 53353
           Summary: Missing error/warning when using __int128_t with
                    incorrect asm register constraints
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: svfuerst@gmail.com


The following (nonsensical) code should give an error (or at least a warning)
about the 128 bit type not fitting into a register:

void foo(__int128_t x)
{
    asm("mov %0, bar":: "r" (x));
}

Instead, the above compiles into an instruction that uses the 16 bit register
%ax.

(Compiled with -Wall -W)

[Bug c/53353] Missing error/warning when using __int128_t with incorrect asm register constraints

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53353

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-05-15 01:40:11 UTC ---
It fits into two registers, rax and rbx.

[Bug c/53353] Missing error/warning when using __int128_t with incorrect asm register constraints

[svfuerst at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53353

--- Comment #2 from Steven Fuerst <svfuerst at gmail dot com> 2012-05-15 01:54:46 UTC ---
Obviously a __int128_t fits in two registers.  The bug is that gcc doesn't
warn/error about code mistakenly trying to fit it into one.  Instead, gcc
generates a weird 16-bit operation, causing silent bad code generation.

[Bug c/53353] Missing error/warning when using __int128_t with incorrect asm register constraints

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53353

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-05-15 02:02:23 UTC ---
(In reply to comment #2)
> Obviously a __int128_t fits in two registers.  The bug is that gcc doesn't
> warn/error about code mistakenly trying to fit it into one.  Instead, gcc
> generates a weird 16-bit operation, causing silent bad code generation.

It does not know what you want to do?  Really it is the "ax:dx" pair most
likely.
Also the inline-asm uses internal GCC representation.
How does GCC not know the inline-asm did the correct thing of what the user
wanted it to do?  Hint GCC cannot read your mind to say you made a mistake
here.

[Bug c/53353] Missing error/warning when using __int128_t with incorrect asm register constraints

[svfuerst at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53353

--- Comment #4 from Steven Fuerst <svfuerst at gmail dot com> 2012-05-15 02:09:05 UTC ---
Actually, it is the rax:rdx pair that is most likely.  ax:dx only has 32 bits. 
rax:rdx is specified by the 'A' constraint, the only gpr option that is 128
bits.

(Yes, it would be nice to pass a 128 bit integer in some other pair of
registers as an enhancement, but currently only the 'A' constraint works.)

[Bug c/53353] Missing error/warning when using __int128_t with incorrect asm register constraints

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53353

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-05-15 06:19:51 UTC ---
You are wrong.  "A" class contains the ax and dx registers, which means that
for say 64-bit value it can give you either %rax, or %rdx, for 128-bit value
obviously only %rax:%rdx pair.  But "r" class contains all the non-fixed
general purpose registers, so it can give you any pair of consecutive
registers.
The only classes that can't hold 128-bit values are the single register
classes, like "a", "b", ..., "D", "S", etc.