Re: x86_64 extensions of SSE2 instruction set

[Jan Hubicka <jh at suse dot cz>]

> On Tue, 30 Jan 2001, Jan Hubicka wrote:
> 
> > cvsss2si %xmm0, (%rax)  is ambiguous
> > (this is resolved in the Intel syntax by qword ptr or dwrod ptr keywords)
> > 
> > Using the default scheme by adding 'l' or 'q' suffix don't seems to fit
> > here nicely (imagine cvtsi2ssq).
> > 
> > So I would suggest other naming probably:
> > cvtss2si - always 32bit
> > cvtss2sqi (convert scalar single to signed quad integer) - 64bit version.
> 
> I think it better to stick with the current scheme for the sake of
What do you exactly mean by current scheme?
So you mean that in x86_64 gas should accept
cvtss2siq
cvtsi2ssq
cvtss2sil
cvtsi2ssl
and refuse cvtss2si?
This is rather strange - the converisons in integer looks like
movsbl, but here is missing letter for the first type (scalar float) in our
case. Especially those converting from integer looks quite weird - but
if we will not be able to come with something better (probably AMD officially
naming the instruction differently) I will probably stay with this solution.
> consistency.  It's hard enough to remember these mnemonics as it is
> without having to remember that certain instructions have different rules
> for suffixes.  In any case, there are plenty of other "funny looking"
> instructions when you add a suffix.
> 
> Alan Modra
> -- 
> Linuxcare.  Support for the Revolution.