[Mesa-dev] [RFC PATCH 13/16] i965: Support array_spacing_lod0 for multiple miplevels

[Pohjolainen, Topi]

On Mon, Jun 02, 2014 at 09:44:16AM -0700, Jordan Justen wrote:
> On Mon, Jun 2, 2014 at 2:36 AM, Pohjolainen, Topi
> <topi.pohjolainen at intel.com> wrote:
> > On Thu, May 29, 2014 at 01:53:52PM -0700, Jordan Justen wrote:
> >> Previously array spacing lod0 was only used with a single mip level.
> >>
> >> It indicated that no mip level spacing should be used between array
> >> slices.
> >>
> >> gen6 stencil only support LOD0, so we need to allocate the miptree
> >> similar to array spacing lod0, except we also need space for more
> >> than just one mip level.
> >>
> >> So, the miptree is allocated with tightly packed array slice spacing,
> >> but we still also pack the miplevels into the region similar to a
> >> normal multi mip level packing.
> >>
> >> Essentially, the miptree is set up as if the texture slices formed
> >> one tall 2D texture, rather than a 2D array.
> >>
> >> Signed-off-by: Jordan Justen <jordan.l.justen at intel.com>
> >> ---
> >>  src/mesa/drivers/dri/i965/brw_tex_layout.c | 21 +++++++++++++++++++--
> >>  1 file changed, 19 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/src/mesa/drivers/dri/i965/brw_tex_layout.c b/src/mesa/drivers/dri/i965/brw_tex_layout.c
> >> index 76044b2..8a0912d 100644
> >> --- a/src/mesa/drivers/dri/i965/brw_tex_layout.c
> >> +++ b/src/mesa/drivers/dri/i965/brw_tex_layout.c
> >> @@ -203,6 +203,11 @@ brw_miptree_layout_2d(struct intel_mipmap_tree *mt)
> >>        if (mt->compressed)
> >>        img_height /= mt->align_h;
> >>
> >> +      if (mt->array_spacing_lod0) {
> >> +         /* Compact arrays with separated miplevels */
> >> +      img_height *= depth;
> >
> > I always thought arrays are only handled by brw_miptree_layout_texture_array().
> > When do we end up here to setup an array?
> 
> brw_miptree_layout_texture_array calls brw_miptree_layout_2d.

Ah, right, it was too obvious, my bad.

> 
> >> +      }
> >> +
> >>        /* Because the images are packed better, the final offset
> >>         * might not be the maximal one:
> >>         */
> >> @@ -238,6 +243,7 @@ brw_miptree_layout_texture_array(struct brw_context *brw,
> >>                                struct intel_mipmap_tree *mt)
> >>  {
> >>     int h0, h1;
> >> +   unsigned height = mt->physical_height0;
> >>
> >>     h0 = ALIGN(mt->physical_height0, mt->align_h);
> >>     h1 = ALIGN(minify(mt->physical_height0, 1), mt->align_h);
> >> @@ -251,11 +257,22 @@ brw_miptree_layout_texture_array(struct brw_context *brw,
> >>     brw_miptree_layout_2d(mt);
> >>
> >>     for (unsigned level = mt->first_level; level <= mt->last_level; level++) {
> >> +      unsigned img_height;
> >> +      img_height = ALIGN(height, mt->align_h);
> >> +      if (mt->compressed)
> >> +         img_height /= mt->align_h;
> >
> > This is confusing me also. You are introducing new layout for layered stencil
> > only. Do we really allow compression of stencil buffers?
> 
> I guess I was just trying to generically extend array_spacing_lod0 to
> multiple miplevels. It does seem like array_spacing_lod0 is no longer
> the correct name in this case. How about something like
> compact_arrays_separate_lods?

I guess I was looking something specific to stencil. In fact wouldn't
intel_mipmap_tree::format be always MESA_FORMAT_S_UINT8 in this case? Couldn't
we just use that and gen == 6 as the trigger for the special layout?

Or are you thinking some other usecase besides stencil for this?

> 
> This is definitely one of the RFC patches. :)
> 
> -Jordan
> 
> >
> >> +
> >>        for (int q = 0; q < mt->physical_depth0; q++) {
> >> -      intel_miptree_set_image_offset(mt, level, q, 0, q * physical_qpitch);
> >> +         if (mt->array_spacing_lod0) {
> >> +            intel_miptree_set_image_offset(mt, level, q, 0, q * img_height);
> >> +         } else {
> >> +            intel_miptree_set_image_offset(mt, level, q, 0, q * physical_qpitch);
> >> +         }
> >>        }
> >> +      height = minify(height, 1);
> >>     }
> >> -   mt->total_height = physical_qpitch * mt->physical_depth0;
> >> +   if (!mt->array_spacing_lod0)
> >> +      mt->total_height = physical_qpitch * mt->physical_depth0;
> >>
> >>     align_cube(mt);
> >>  }
> >> --
> >> 2.0.0.rc4
> >>
> >> _______________________________________________
> >> mesa-dev mailing list
> >> mesa-dev at lists.freedesktop.org
> >> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> > http://lists.freedesktop.org/mailman/listinfo/mesa-dev