There's another case missed, SLP reductions still use the wrong
sign for the epilogue operations. I also noticed that we generate
somewhat ugly code for the epilogue since we fall back to scalar
extracts but do not consider to reduce a larger vector to a smaller
one first (and neither multiple
On Mon, 11 Nov 2019 at 14:36, Richard Biener wrote:
>
> On Mon, 11 Nov 2019, Christophe Lyon wrote:
>
> > On Fri, 8 Nov 2019 at 09:57, Richard Biener wrote:
> > >
> > >
> > > I've been sitting on this for a few days since I'm not 100% happy
> > > with how the code looks like. There's possibly
On Mon, 11 Nov 2019, Christophe Lyon wrote:
> On Fri, 8 Nov 2019 at 09:57, Richard Biener wrote:
> >
> >
> > I've been sitting on this for a few days since I'm not 100% happy
> > with how the code looks like. There's possibly still holes in it
> > (chains with mixed signed/unsigned adds for
On Fri, 8 Nov 2019 at 09:57, Richard Biener wrote:
>
>
> I've been sitting on this for a few days since I'm not 100% happy
> with how the code looks like. There's possibly still holes in it
> (chains with mixed signed/unsigned adds for example might pick
> up signed adds in the epilogue), but
Richard Biener writes:
> On Fri, 8 Nov 2019, Richard Sandiford wrote:
>
>> Richard Biener writes:
>> > I've been sitting on this for a few days since I'm not 100% happy
>> > with how the code looks like. There's possibly still holes in it
>> > (chains with mixed signed/unsigned adds for example
On Fri, 8 Nov 2019, Richard Sandiford wrote:
> Richard Biener writes:
> > I've been sitting on this for a few days since I'm not 100% happy
> > with how the code looks like. There's possibly still holes in it
> > (chains with mixed signed/unsigned adds for example might pick
> > up signed adds
Richard Biener writes:
> I've been sitting on this for a few days since I'm not 100% happy
> with how the code looks like. There's possibly still holes in it
> (chains with mixed signed/unsigned adds for example might pick
> up signed adds in the epilogue), but the wrong-code cases should
> work
I've been sitting on this for a few days since I'm not 100% happy
with how the code looks like. There's possibly still holes in it
(chains with mixed signed/unsigned adds for example might pick
up signed adds in the epilogue), but the wrong-code cases should
work fine now. I'm probably going
