On Wed, Dec 17, 2014 at 8:48 PM, Jason Ekstrand wrote:
>
>
> On Wed, Dec 17, 2014 at 2:23 PM, Connor Abbott wrote:
>>
>> On Wed, Dec 17, 2014 at 5:11 PM, Jason Ekstrand
>> wrote:
>> > On Wed, Dec 17, 2014 at 1:52 PM, Connor Abbott
>> > wrote:
>> >>
>> >> I'm sure you're already aware, but there
On Wed, Dec 17, 2014 at 2:23 PM, Connor Abbott wrote:
>
> On Wed, Dec 17, 2014 at 5:11 PM, Jason Ekstrand
> wrote:
> > On Wed, Dec 17, 2014 at 1:52 PM, Connor Abbott
> wrote:
> >>
> >> I'm sure you're already aware, but there are two things we could do to
> >> speed this up:
> >>
> >> 1. Pre-com
On Wed, Dec 17, 2014 at 5:11 PM, Jason Ekstrand wrote:
> On Wed, Dec 17, 2014 at 1:52 PM, Connor Abbott wrote:
>>
>> I'm sure you're already aware, but there are two things we could do to
>> speed this up:
>>
>> 1. Pre-compute def/kill sets for each block similar to what i965 does.
>
>
> Sure, bu
On Wed, Dec 17, 2014 at 1:52 PM, Connor Abbott wrote:
>
> I'm sure you're already aware, but there are two things we could do to
> speed this up:
>
> 1. Pre-compute def/kill sets for each block similar to what i965 does.
>
Sure, but we walk the instructions at most deepest block depth + 1 and the
On Wed, Dec 17, 2014 at 1:52 PM, Connor Abbott wrote:
>
> I'm sure you're already aware, but there are two things we could do to
> speed this up:
>
> 1. Pre-compute def/kill sets for each block similar to what i965 does.
> 2. Use a worklist + an array of flags for "this block is in the
> worklist"
I'm sure you're already aware, but there are two things we could do to
speed this up:
1. Pre-compute def/kill sets for each block similar to what i965 does.
2. Use a worklist + an array of flags for "this block is in the
worklist" rather than walking all the basic blocks in reverse to find
the few
---
src/glsl/Makefile.sources | 1 +
src/glsl/nir/nir.h| 13 ++
src/glsl/nir/nir_live_variables.c | 282 ++
src/glsl/nir/nir_metadata.c | 2 +
src/mesa/main/bitset.h| 1 +
5 files changed, 299 insertions(+)
creat
