On Mon, Apr 24, 2017 at 11:42 PM, Samuel Pitoiset
wrote:
>
>
> On 04/24/2017 11:22 PM, Rob Clark wrote:
>>
>> On Mon, Apr 24, 2017 at 5:18 PM, Samuel Pitoiset
>> wrote:
>>>
>>>
>>>
>>> On 04/24/2017 11:12 PM, Rob Clark wrote:
so I guess this is likely to hurt pipe drivers that don'
It would kill nv30, I believe.
On Apr 24, 2017 10:30 PM, "Roland Scheidegger" wrote:
> Am 24.04.2017 um 23:12 schrieb Rob Clark:
> > so I guess this is likely to hurt pipe drivers that don't (yet?)
> > have a real compiler backend. (Ie. etnaviv and freedreno/a2xx.) So
> > maybe it should be op
Am 24.04.2017 um 23:12 schrieb Rob Clark:
> so I guess this is likely to hurt pipe drivers that don't (yet?)
> have a real compiler backend. (Ie. etnaviv and freedreno/a2xx.) So
> maybe it should be optional.
I suppose softpipe, too? Though that's fine, noone cares if it gets a
bit slower. Might
On 04/24/2017 11:22 PM, Rob Clark wrote:
On Mon, Apr 24, 2017 at 5:18 PM, Samuel Pitoiset
wrote:
On 04/24/2017 11:12 PM, Rob Clark wrote:
so I guess this is likely to hurt pipe drivers that don't (yet?) have
a real compiler backend. (Ie. etnaviv and freedreno/a2xx.) So maybe
it should b
On Mon, Apr 24, 2017 at 5:18 PM, Samuel Pitoiset
wrote:
>
>
> On 04/24/2017 11:12 PM, Rob Clark wrote:
>>
>> so I guess this is likely to hurt pipe drivers that don't (yet?) have
>> a real compiler backend. (Ie. etnaviv and freedreno/a2xx.) So maybe
>> it should be optional.
>
>
> I can't say fo
On 04/24/2017 11:12 PM, Rob Clark wrote:
so I guess this is likely to hurt pipe drivers that don't (yet?) have
a real compiler backend. (Ie. etnaviv and freedreno/a2xx.) So maybe
it should be optional.
I can't say for these drivers, but it seems safer to add an option if
this can hurt them
so I guess this is likely to hurt pipe drivers that don't (yet?) have
a real compiler backend. (Ie. etnaviv and freedreno/a2xx.) So maybe
it should be optional.
Also I wonder about the pre-llvm radeon gen's, since sb uses the
actual instruction encoding for IR between tgsi->sb and backend opt
pa
The main goal of this pass to merge temporary registers in order
to reduce the total number of registers and also to produce
optimal TGSI code.
In fact, compilers seem to be confused when temporary variables
are already merged, maybe because it's done too early in the
process.
Removing the pass,