Re: [patch] SPARC VIS instructions or PR1820

[Eric Botcazou <ebotcazou at libertysurf dot fr>]

>  Really?  I haven't figured out how these are any different than SImode.
> I've updated the comment anyway.

You're probably right, SImode will work just fine.  It would be nice to write 
testcases for fpadd32s and fpsub32s then.

>  I didn't know of them, I haven't found extensive VIS documentation
> anywhere.

http://www.sun.com/processors/vis

>  At some point I was having trouble using "b", so I switch to using "e" and
> things stopped failing.

Which kind of trouble?  "e" and  "f" are probably fine anyway.

>  As long as vectors can be packed into integer registers for function
> passing we shouldn't have any trouble.

We chose to be compatible with the Sun VIS SDK.  Which means that the vector 
integers are passed/returned like floats (int/FP regs in 32-bit mode, FP/FP 
regs in 64-bit mode).  We also chose to have only one set of calling 
conventions, independently of whether -mvis is specified.  The problem is 
that I implemented these calling conventions when vector types were tied to 
vector modes; this is not true anymore, so we have to adjust the 
implementation.

> Perhaps vector loads and stores should be available with or without -mvis.

I think this would go against the new approach, where vector modes are only 
available if the hardware supports them.  What do other ports implement?

-- 
Eric Botcazou