On Sep 7, 2016 12:23 PM, "Chad Versace" <<a href="mailto:chadversary@chromium.org">chadversary@chromium.org</a>> wrote: > > On Fri 02 Sep 2016, Jason Ekstrand wrote: > > --- > >Â src/intel/isl/isl.c | 60 +++++++++++++++++++++++++++++++++++++++++++++++++++-- > >Â 1 file changed, 58 insertions(+), 2 deletions(-) > > > > diff --git a/src/intel/isl/isl.c b/src/intel/isl/isl.c > > index 33e83b1..cb9eb23 100644 > > --- a/src/intel/isl/isl.c > > +++ b/src/intel/isl/isl.c > > @@ -1284,6 +1284,63 @@ isl_surf_get_hiz_surf(const struct isl_device *dev, > >Â Â Â assert(surf->msaa_layout == ISL_MSAA_LAYOUT_NONE || > >Â Â Â Â Â Â surf->msaa_layout == ISL_MSAA_LAYOUT_INTERLEAVED); > > > > +Â Â /* From the Broadwell PRM Vol. 7, "Hierarchical Depth Buffer": > > +Â Â * > > +Â Â *Â Â "The Surface Type, Height, Width, Depth, Minimum Array Element, Render > > +Â Â *Â Â Target View Extent, and Depth Coordinate Offset X/Y of the > > +Â Â *Â Â hierarchical depth buffer are inherited from the depth buffer. The > > +Â Â *Â Â height and width of the hierarchical depth buffer that must be > > +Â Â *Â Â allocated are computed by the following formulas, where HZ is the > > +Â Â *Â Â hierarchical depth buffer and Z is the depth buffer. The Z_Height, > > +Â Â *Â Â Z_Width, and Z_Depth values given in these formulas are those present > > +Â Â *Â Â in 3DSTATE_DEPTH_BUFFER incremented by one. > > +Â Â * > > +Â Â *Â Â "The value of Z_Height and Z_Width must each be multiplied by 2 before > > +Â Â *Â Â being applied to the table below if Number of Multisamples is set to > > +Â Â *Â Â NUMSAMPLES_4. The value of Z_Height must be multiplied by 2 and > > +Â Â *Â Â Z_Width must be multiplied by 4 before being applied to the table > > +Â Â *Â Â below if Number of Multisamples is set to NUMSAMPLES_8." > > +Â Â * > > +Â Â * In the Sky Lake PRM, the second paragraph is replaced with this: > > +Â Â * > > +Â Â *Â Â "The Z_Height and Z_Width values must equal those present in > > +Â Â *Â Â 3DSTATE_DEPTH_BUFFER incremented by one." > > +Â Â * > > +Â Â * In other words, on Sandy Bridge through Broadwell, each 128-bit HiX > > TypoÂ s/HiX/HiZ/ > > > +Â Â * block corresponds to a region of 8x4 samples in the primary depth > > +Â Â * surface.Â On Sky Lake, on the other hand, each HiZ block corresponds to > > +Â Â * a region of 8x4 pixels in the primary depth surface regarldess of the > > TypoÂ Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â ^^^^^^^^^^ > > > +Â Â * number of samples.Â The dimensions of a HiZ block in both pixes and > > TypoÂ Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â ^^^^^ > > > +Â Â * samples are given in the table below: > > +Â Â * > > +Â Â *Â Â Â Â Â Â Â Â Â Â | SNB - BDW |Â Â Â SKL+ > > +Â Â *Â Â Â Â Â Â Â ------------------+------------- > > +Â Â *Â Â Â Â Â Â Â Â 1xÂ |Â 8 x 4 sa |Â Â 8 x 4 sa > > +Â Â *Â Â Â Â Â Â Â Â MSAA |Â 8 x 4 px |Â Â 8 x 4 px > > +Â Â *Â Â Â Â Â Â Â ------------------+------------- > > +Â Â *Â Â Â Â Â Â Â Â 2xÂ |Â 8 x 4 sa |Â 16 x 4 sa > > +Â Â *Â Â Â Â Â Â Â Â MSAA |Â 4 x 4 px |Â Â 8 x 4 px > > +Â Â *Â Â Â Â Â Â Â ------------------+------------- > > +Â Â *Â Â Â Â Â Â Â Â 4xÂ |Â 8 x 4 sa |Â 16 x 8 sa > > +Â Â *Â Â Â Â Â Â Â Â MSAA |Â 4 x 2 px |Â Â 8 x 4 px > > +Â Â *Â Â Â Â Â Â Â ------------------+------------- > > +Â Â *Â Â Â Â Â Â Â Â 8xÂ |Â 8 x 4 sa |Â 32 x 8 sa > > +Â Â *Â Â Â Â Â Â Â Â MSAA |Â 2 x 2 px |Â Â 8 x 4 px > > +Â Â *Â Â Â Â Â Â Â ------------------+------------- > > +Â Â *Â Â Â Â Â Â Â Â 16xÂ |Â Â N/AÂ Â | 32 x 16 sa > > +Â Â *Â Â Â Â Â Â Â Â MSAA |Â Â N/AÂ Â |Â 8 xÂ 4 px > > +Â Â *Â Â Â Â Â Â Â ------------------+------------- > > THANK YOU THANK YOU THANK YOU. > You win all the cat gifs [<a href="https://s-media-cache-ak0.pinimg.com/originals/34/71/dc/3471dc5d6443b2af5476ad3dbbca9d7c.jpg">https://s-media-cache-ak0.pinimg.com/originals/34/71/dc/3471dc5d6443b2af5476ad3dbbca9d7c.jpg</a>] Yeah... This stuff has proven surprisingly difficult to reason about.Â The table gives us a central spot to write down "the facts" so that future generations of ISL hackers understand what's going on. > > +Â Â * > > +Â Â * There are a number of different ways that this discrepency could be > > +Â Â * handled.Â The way we have chosen is to simply make MSAA HiZ have the > > +Â Â * same number of samples as the parent surface pre-Sky Lake and always be > > +Â Â * single-sampled on Sky Lake and above.Â Since the block sizes of > > +Â Â * compressed formats are given in samples, this neatly handles everything > > +Â Â * without the need for additional HiZ formats with different block sizes > > +Â Â * on SKL+. > > +Â Â */ > > This makes perfect sense. > > > +Â Â const unsigned samples = ISL_DEV_GEN(dev) >= 9 ? 1 : surf->samples; > > + > >Â Â Â isl_surf_init(dev, hiz_surf, > >Â Â Â Â Â Â Â Â Â Â .dim = ISL_SURF_DIM_2D, > >Â Â Â Â Â Â Â Â Â Â .format = ISL_FORMAT_HIZ, > > @@ -1292,8 +1349,7 @@ isl_surf_get_hiz_surf(const struct isl_device *dev, > >Â Â Â Â Â Â Â Â Â Â .depth = 1, > >Â Â Â Â Â Â Â Â Â Â .levels = surf->levels, > >Â Â Â Â Â Â Â Â Â Â .array_len = surf->logical_level0_px.array_len, > > -Â Â Â Â Â Â Â Â Â /* On SKL+, HiZ is always single-sampled */ > > -Â Â Â Â Â Â Â Â Â .samples = ISL_DEV_GEN(dev) >= 9 ? 1 : surf->samples, > > +Â Â Â Â Â Â Â Â Â .samples = samples, > >Â Â Â Â Â Â Â Â Â Â .usage = ISL_SURF_USAGE_HIZ_BIT, > >Â Â Â Â Â Â Â Â Â Â .tiling_flags = ISL_TILING_HIZ_BIT); > >Â } > > -- > > Reviewed-by: Chad Versace <<a href="mailto:chadversary@chromium.org">chadversary@chromium.org</a>> Thanks. I'll get the convent typos fixed when I get back to my computer. --Jason