On Wed, Jan 13, 2016 at 4:35 PM, Axel Davy wrote:
>
>>
>> Axel Davy benchmarked this briefly. We may need more benchmarks though.
>>
>> Marek
>>
>
> I confirm setting this register helps get a few % with heaven.
>
> There was also another register to kill color exports early when doing
> depth onl
>
> Axel Davy benchmarked this briefly. We may need more benchmarks though.
>
> Marek
>
I confirm setting this register helps get a few % with heaven.
There was also another register to kill color exports early when doing
depth only pass that helped a few % (but less).
Axel
___
On 12.01.2016 18:55, Marek Olšák wrote:
Hi Nicolai,
I only tested piglit on Bonaire.
The scratch buffer can deadlock too, though I guess we have enough
scratch space that we should never run out of it.
a) The lockup is prevented by the change to the value of
SHADER_PGM_RSRC3_VS. It means that
Hi Nicolai,
I only tested piglit on Bonaire.
The scratch buffer can deadlock too, though I guess we have enough
scratch space that we should never run out of it.
a) The lockup is prevented by the change to the value of
SHADER_PGM_RSRC3_VS. It means that 1 compute unit is reserved for PS
only. Th
How well-tested is this? I'm a bit confused because I thought the
scratch_waves is about the scratch buffer ring (which should always be
allocated immediately when the shader starts), while the late_alloc is
about allocating the space where exports go.
So late_alloc is basically bounded by
(a
From: Marek Olšák
This allows more VS waves to run on CIK and VI.
---
src/gallium/drivers/radeonsi/si_pipe.c | 4 +++-
src/gallium/drivers/radeonsi/si_pipe.h | 4
src/gallium/drivers/radeonsi/si_state.c | 5 +++--
3 files changed, 10 insertions(+), 3 deletions(-)
diff --git a/src/gallium
