[Mesa-dev] [PATCH 3/4] radeonsi: fix gl_PrimitiveID in tessellation with instanced draws on SI

[Nicolai Hähnle]

On 02.06.2017 15:01, Marek Olšák wrote:
> On Wed, May 3, 2017 at 3:58 PM, Nicolai Hähnle <nhaehnle at gmail.com> wrote:
>> From: Nicolai Hähnle <nicolai.haehnle at amd.com>
>>
>> Cc: mesa-stable at lists.freedesktop.org
>> ---
>>   src/gallium/drivers/radeonsi/si_state_draw.c | 14 ++++++++++++++
>>   1 file changed, 14 insertions(+)
>>
>> diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
>> index e6a9ee0..3d1d1f8 100644
>> --- a/src/gallium/drivers/radeonsi/si_state_draw.c
>> +++ b/src/gallium/drivers/radeonsi/si_state_draw.c
>> @@ -181,20 +181,34 @@ static void si_emit_derived_tess_state(struct si_context *sctx,
>>
>>          /* Not necessary for correctness, but improves performance. The
>>           * specific value is taken from the proprietary driver.
>>           */
>>          *num_patches = MIN2(*num_patches, 40);
>>
>>          /* SI bug workaround - limit LS-HS threadgroups to only one wave. */
>>          if (sctx->b.chip_class == SI) {
>>                  unsigned one_wave = 64 / MAX2(num_tcs_input_cp, num_tcs_output_cp);
>>                  *num_patches = MIN2(*num_patches, one_wave);
>> +
>> +               if (sctx->screen->b.info.max_se == 1) {
>> +                       /* The VGT HS block increments the patch ID unconditionally
>> +                        * within a single threadgroup. This results in incorrect
>> +                        * patch IDs when instanced draws are used.
>> +                        *
>> +                        * The intended solution is to restrict threadgroups to
>> +                        * a single instance by setting SWITCH_ON_EOI, which
>> +                        * should cause IA to split instances up. However, this
>> +                        * doesn't work correctly on SI when there is no other
>> +                        * SE to switch to.
>> +                        */
>> +                       *num_patches = 1;
>> +               }
> 
> Hi Nicolai,
> 
> This commit massively decreases tessellation performance on SI 1-SE
> parts. We need a different solution. Would this work: "Set num_patches
> to the greatest divisor of the the number of patches per instance."

Yes, that should be sufficient. Actually, it'd be even better to only do 
the workaround when PRIM_ID is read, which is probably quite rare...

Do you already have a patch for the first part?

Thanks,
Nicolai
-- 
Lerne, wie die Welt wirklich ist,
Aber vergiss niemals, wie sie sein sollte.