[Mesa-dev] [PATCH v2 1/2] anv/cmd_buffer: Don't temporarily enable CCS_E within a render pass

[Jason Ekstrand]

On Thu, Feb 2, 2017 at 3:49 PM, Nanley Chery <nanleychery at gmail.com> wrote:

> On Thu, Feb 02, 2017 at 01:38:45PM -0800, Jason Ekstrand wrote:
> > Perf numbers in the commit message would be nice
> >
>
> How's the following?
>
>    Reclaims about 1.95% FPS for Dota 2 on some configurations.
>

Sounds good!


> > Reviewed-by: Jason Ekstrand <jason at jlekstrand.net>
>
> Thanks!
>
> >
> > On Thu, Feb 2, 2017 at 8:15 AM, Nanley Chery <nanleychery at gmail.com>
> wrote:
> >
> > > Compressing a render target and decompressing it in the same
> > > single-subpass render pass may waste bandwidth. While this may be
> > > beneficial in some circumstances, it does not help in all.
> > >
> > > v2 (Jason Ekstrand):
> > > - Provide a more thorough comment
> > > - Enable CCS_D for input attachments
> > >
> > > Signed-off-by: Nanley Chery <nanley.g.chery at intel.com>
> > > ---
> > >  src/intel/vulkan/genX_cmd_buffer.c | 17 ++++++++++++++---
> > >  1 file changed, 14 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/src/intel/vulkan/genX_cmd_buffer.c
> > > b/src/intel/vulkan/genX_cmd_buffer.c
> > > index 63911f2c8e..cba876b58e 100644
> > > --- a/src/intel/vulkan/genX_cmd_buffer.c
> > > +++ b/src/intel/vulkan/genX_cmd_buffer.c
> > > @@ -266,13 +266,24 @@ color_attachment_compute_aux_usage(struct
> > > anv_device *device,
> > >        att_state->fast_clear = false;
> > >     }
> > >
> > > -   if (isl_format_supports_lossless_compression(&device->info,
> > > -                                                iview->isl.format)) {
> > > +   /**
> > > +    * TODO: Consider using a heuristic to determine if temporarily
> > > enabling
> > > +    * CCS_E for this image view would be beneficial.
> > > +    *
> > > +    * While fast-clear resolves and partial resolves are fairly cheap
> in
> > > the
> > > +    * case where you render to most of the pixels, full resolves are
> not
> > > +    * because they potentially involve reading and writing the entire
> > > +    * framebuffer.  If we can't texture with CCS_E, we should leave it
> > > off and
> > > +    * limit ourselves to fast clears.
> > > +    */
> > > +   if (iview->image->aux_usage == ISL_AUX_USAGE_CCS_E) {
> > >        att_state->aux_usage = ISL_AUX_USAGE_CCS_E;
> > >        att_state->input_aux_usage = ISL_AUX_USAGE_CCS_E;
> > >     } else if (att_state->fast_clear) {
> > >        att_state->aux_usage = ISL_AUX_USAGE_CCS_D;
> > > -      if (GEN_GEN >= 9) {
> > > +      if (GEN_GEN >= 9 &&
> > > +          !isl_format_supports_lossless_compression(&device->info,
> > > +
> iview->isl.format)) {
> > >           /* From the Sky Lake PRM, RENDER_SURFACE_STATE::
> > > AuxiliarySurfaceMode:
> > >            *
> > >            *    "If Number of Multisamples is MULTISAMPLECOUNT_1,
> AUX_CCS_D
> > > --
> > > 2.11.0
> > >
> > > _______________________________________________
> > > mesa-dev mailing list
> > > mesa-dev at lists.freedesktop.org
> > > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> > >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170202/711a989b/attachment.html>