Re: [PATCH]: AltiVec patch, was: Update SSE5 vector multiplication, shift, rotate

[Paolo Bonzini <bonzini at gnu dot org>]

> > But you need to check if the returned optab is really suitable for
> > the vecvec case, no?
>
> I don't think so, because v{rot,lsh,ash}{l,r}_optab is actually only for 
> suitable optabs.  Hmmm probably this would not be backwards compatible 
> for AltiVec.  But is bringing AltiVec up to date much harder than the 
> attached patch?

(Patch at http://permalink.gmane.org/gmane.comp.gcc.patches/161247 )

David, is it ok to apply the attached patch together with the rest of 
the SSE5 patches?  It basically enforces throughout all ports that 
"vashl", "vashr", "vlshr" are the optabs for vector-vector shifts, and 
"ashl", "ashr", "lshr" are used for vector-scalar shifts (where all the 
elements are shifted by the same amount).

Since I don't have the possibility to test on powerpc hardware, the 
patch was tested by comparing the testsuite assembly output with no 
differences.

Paolo