This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: [PATCH] Lexra binutils
On Fri, 19 Sep 2008, Adam Nemet wrote:
> > Then if better alternatives are available as some sort of an extension,
> > then it is up to GCC to emit them as appropriate and possibly GAS to use
> > in the expansion of the unaligned transfer macros as instructed by the
> > -march= flag. No need to tweak base ISA flags.
>
> I am not 100% sure that I understand but I think what you're saying is that
> you want GCC to emit let's say uld and then gas would expand to ldr/ldl or the
> Octeon-specific instruction.
I mean GCC would emit either ldr/ldl, etc. or the CPU-specific
alternatives based on the -march= setting. Handcoded assembly would have
to be updated manually if it was to benefit.
> First, this is not good because you want GCC to schedule ldr and ldl. Second,
> just like Lexar, I'd like to signal an error if someone uses ldl when
> Octeon-specific unaligned instructions are on because these instructions
> replace the standard MIPS unaligned instructions.
So this is in fact not a compliant MIPS processor... Hmm, what can I
say? No idea off the head -- I would have to think about it. In any
case, the approach to enable either set of instructions at a time, but not
both seems a bit unfortunate.
Maciej