[Mesa-dev] [PATCH] i965: Bind null render targets for shadow sampling + color.

[Pohjolainen, Topi]

On Thu, Jan 18, 2018 at 07:08:57AM -0800, Jason Ekstrand wrote:
> Rb but feel free to get a third opinion if you'd like

Here is another (I was about to send that earlier anyway):

Reviewed-by: Topi Pohjolainen <topi.pohjolainen at intel.com>

> 
> 
> On January 18, 2018 01:51:03 Kenneth Graunke <kenneth at whitecape.org> wrote:
> 
> > Portal 2 appears to bind RGBA8888_UNORM textures to a sampler2DShadow,
> > and calls shadow2D() on it.  This causes undefined behavior in OpenGL.
> > 
> > Unfortunately, our sampler appears to hang in this scenario, which is
> > not acceptable.  Just give them a null surface instead, which returns
> > all zeroes.
> > 
> > Fixes GPU hangs in Portal 2 on Kabylake.
> > 
> > Huge thanks to Jason Ekstrand for noticing this crazy behavior while
> > sifting through crash dumps.
> > 
> > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104487
> > ---
> >  src/mesa/drivers/dri/i965/brw_wm_surface_state.c | 33 +++++++++++++++++++++++-
> >  1 file changed, 32 insertions(+), 1 deletion(-)
> > 
> > diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> > b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> > index adf60a840b0..38af6bc0dea 100644
> > --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> > +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
> > @@ -1095,6 +1095,14 @@ const struct brw_tracked_state
> > brw_renderbuffer_read_surfaces = {
> >     .emit = update_renderbuffer_read_surfaces,
> >  };
> > 
> > +static bool
> > +is_depth_texture(struct intel_texture_object *iobj)
> > +{
> > +   GLenum base_format = _mesa_get_format_base_format(iobj->_Format);
> > +   return base_format == GL_DEPTH_COMPONENT ||
> > +          (base_format == GL_DEPTH_STENCIL && !iobj->base.StencilSampling);
> > +}
> > +
> >  static void
> >  update_stage_texture_surfaces(struct brw_context *brw,
> >                                const struct gl_program *prog,
> > @@ -1121,9 +1129,32 @@ update_stage_texture_surfaces(struct brw_context *brw,
> >        if (prog->SamplersUsed & (1 << s)) {
> >           const unsigned unit = prog->SamplerUnits[s];
> >           const bool used_by_txf = prog->info.textures_used_by_txf & (1 << s);
> > +         struct gl_texture_object *obj = ctx->Texture.Unit[unit]._Current;
> > +         struct intel_texture_object *iobj = intel_texture_object(obj);
> > 
> >           /* _NEW_TEXTURE */
> > -         if (ctx->Texture.Unit[unit]._Current) {
> > +         if (!obj)
> > +            continue;
> > +
> > +         if ((prog->ShadowSamplers & (1 << s)) && !is_depth_texture(iobj)) {
> > +            /* A programming note for the sample_c message says:
> > +             *
> > +             *    "The Surface Format of the associated surface must be
> > +             *     indicated as supporting shadow mapping as indicated in the
> > +             *     surface format table."
> > +             *
> > +             * Accessing non-depth textures via a sampler*Shadow type is
> > +             * undefined.  GLSL 4.50 page 162 says:
> > +             *
> > +             *    "If a shadow texture call is made to a sampler that does not
> > +             *     represent a depth texture, then results are undefined."
> > +             *
> > +             * We give them a null surface (zeros) for undefined.  We've seen
> > +             * GPU hangs with color buffers and sample_c, so we try and avoid
> > +             * those with this hack.
> > +             */
> > +            emit_null_surface_state(brw, NULL, surf_offset + s);
> > +         } else {
> >              brw_update_texture_surface(ctx, unit, surf_offset + s, for_gather,
> >                                         used_by_txf, plane);
> >           }
> > --
> > 2.15.1
> > 
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> 
> 
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev