On Friday, September 2, 2016 7:28:09 PM PDT Connor Abbott wrote:
> In 144cbf8 ("nir: Make nir_opt_remove_phis see through moves."), Ken
> made nir_opt_remove_phis able to coalesce phi nodes whose sources are
> all moves with the same swizzle. However, he didn't add the logic
> necessary for handlin
On Fri, Sep 2, 2016 at 8:48 PM, Jason Ekstrand wrote:
> I think this is conceptually correct but I haven't reviewed it for complete
> correctness yet or tested it.
>
> That said, I think this is just a special case of the phi distribution pass
> that we talked about on IRC a bit today. Maybe we s
I think this is conceptually correct but I haven't reviewed it for complete
correctness yet or tested it.
That said, I think this is just a special case of the phi distribution pass
that we talked about on IRC a bit today. Maybe we should just go ahead and
implement that? For unary ALU operation
In 144cbf8 ("nir: Make nir_opt_remove_phis see through moves."), Ken
made nir_opt_remove_phis able to coalesce phi nodes whose sources are
all moves with the same swizzle. However, he didn't add the logic
necessary for handling the fact that the phi may now have multiple
different sources, even tho
