[llvm-commits] [llvm] r77940 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp lib/Target/X86/X86InstrMMX.td test/CodeGen/X86/2009-08-02-mmx-scalar-to-vector.ll test/CodeGen/X86/mmx-bitcast-to-i64.ll

[David Greene]

On Tuesday 04 August 2009 08:50, David Greene wrote:
> On Monday 03 August 2009 00:08, Eli Friedman wrote:
> > On Sun, Aug 2, 2009 at 8:33 PM, Rafael Espindola<espindola at google.com>
>
> wrote:
> > >> I don't see any connection between the way it's written and for the
> > >> assembler and the way the JIT deals with it.  And you still haven't
> > >> addressed the issue, which is that I'm pretty sure it was done for
> > >> compatibility reasons.  If you don't know anything about that, I would
> > >> suggest switching it back to movd.
> > >
> > > Do you know what we are trying to be compatible with?
> >
> > I looked it up; we are trying to be compatible with OS X.  See
> > http://lists.cs.uiuc.edu/pipermail/llvmdev/2007-December/011849.html .
>
> This *still* isn't fixed?
>
> This is a real problem.  movq is the right thing to use for rr
> instructions.

Actually, the problem is with mr and rm variants because the assembler
can't know to move 64 bits without a rex prefix.

                                  -Dave