yeah.. I am not quite sure myself about it. But skipping on the div
emulation seems like a good idea in general. But it's also not common
enough to actually care all that much about it.
On Tue, Jul 23, 2019 at 5:18 PM Ilia Mirkin wrote:
>
> On Tue, Jul 23, 2019 at 11:15 AM Karol Herbst wrote:
>
On Tue, Jul 23, 2019 at 11:15 AM Karol Herbst wrote:
>
> On Tue, Jul 23, 2019 at 4:50 PM Ilia Mirkin wrote:
> >
> > You handle 1/n but not 1%n? TBH, the 1/n code isn't 100% obvious to
> > me... 1/n = |n|-1 > 0 ? i forget how SLCT works, but I can't
> > think of a way to finish that
On Tue, Jul 23, 2019 at 4:50 PM Ilia Mirkin wrote:
>
> You handle 1/n but not 1%n? TBH, the 1/n code isn't 100% obvious to
> me... 1/n = |n|-1 > 0 ? i forget how SLCT works, but I can't
> think of a way to finish that expression in terms of |n|-1 and n. And
> what about n == 0. I'd just as
You handle 1/n but not 1%n? TBH, the 1/n code isn't 100% obvious to
me... 1/n = |n|-1 > 0 ? i forget how SLCT works, but I can't
think of a way to finish that expression in terms of |n|-1 and n. And
what about n == 0. I'd just as soon drop that case.
On Tue, Jul 23, 2019 at 10:20 AM Mark
Optimizations for 0/n, 1/n and 0%n.
No changes in shader db tests, because it is never used here, but it
should become handy.
Signed-off-by: Mark Menzynski
---
.../nouveau/codegen/nv50_ir_peephole.cpp | 30 +--
1 file changed, 28 insertions(+), 2 deletions(-)
diff --git