Re: [patch, tree-ssa] PR54295 Incorrect value extension in widening multiply-accumulate

[Andrew Stubbs <ams at codesourcery dot com>]

On 17/08/12 15:31, Richard Earnshaw wrote:
> On 17/08/12 15:22, Andrew Stubbs wrote:
> > On 17/08/12 15:04, Richard Earnshaw wrote:
> > > The fix is to make is_widening_mult_p note that it has been called with
> > > a WIDEN_MULT_EXPR and rather than decompose the operands again, to
> > > simply extract the existing operands, which have already been formulated
> > > correctly for a widening multiply operation.
> >
> > As long as the existing test cases work, I think the only problem with
> > this idea is if some architecture has a wider range of widening
> > multiply-and-accumulate than it does plain widening multiply.
>
> But surely in that case step1 wouldn't have applied and we'd then be
> looking at a MULT_EXPR not a WIDEN_MULT_EXPR.

Not necessarily.

For example, it will use a 32x32->64 signed widening multiply for 16 bit 
unsigned data if there is no unsigned 16x16->64 option. Hypothetically, 
if there happened to be an unsigned 16x16->64 multiply-and-accumulate 
then you'd end up masking it.

Like I said though, cross that bridge if it ever comes to it.

Andrew