Re: rfc + [autovect] [patch] vectorization of type conversions

[Richard Henderson <rth at redhat dot com>]

On Mon, Oct 17, 2005 at 02:47:59PM +0200, Dorit Naishlos wrote:
> For the special case of "unordered" widening-multiplication I added target
> hooks, cause I think only Altivec supports this. These are
> targetm.vectorize.builtin_mul_widen_even, and
> targetm.vectorize.builtin_mul_widen_odd.
> These are generated in testcases vect-reduc-dot-[s8,u8].c.

I thought you were just going to have a DOT_PROD_EXPR or something.
Recall that SSE, for example, doesn't have anything except pmaddwd,
which includes a partial summation.

> VEC_UNPACK_LO_EXPR}, both of which perform type-promotion. However I hope
> that the VEC_WIDEN_MULT_HI_EXPR will not be confused with the MULT_HI_EXPR
> tree-code we want to add in the future (as also discussed in
> http://gcc.gnu.org/ml/gcc/2004-08/msg00652.html).

Almost certainly they will be, since I've just gone back and re-read
the post, and I'm still not sure what you're wanting.



r~