[patch] Vectorize loops containing function calls

[Richard Guenther]

On 3/29/06, Paolo Bonzini <paolo.bonzini@lu.unisi.ch> wrote:
>
> > The vectorizer part of the patch is straightforward (basically mirrors
> > the way ordinary arithmetics operators are handled, since there is not
> > that much difference).  To find the appropriate library function,
> > a new target hook (builtin_vectorized_function) is used; I am not
> > really sure whether this is the right way.
>
> Sorry if this is a dumb question: why not use the sin_optab and friends?

Do we have those for vector modes?  Can we handle emitting
__m128d,__m128d __vrd4_log(__m128d x1, __m128d x2); this way (I guess not,
I also didn't look if Zdenek uses those yet)?

I wonder anyway how's the defined semantics of all the *_optab stuff - are they
supposed to be used even if math-errno and snans and friends are in effect, or
are they merely required to conform to ieee754 or even -ffast-math?

Richard.