On Friday, August 10, 2018 8:34:12 AM PDT Jason Ekstrand wrote:
> On August 10, 2018 03:19:18 Kenneth Graunke <[email protected]> wrote:
> 
> > The hardware doesn't support rendering to R24_UNORM_X8_TYPELESS, so
> > Jason decided to fake it with a bit of shader math and R32_UNORM RTs.
> >
> > The only problem is that R32_UNORM isn't renderable either...so we've
> > just traded one bad format for another.
> >
> > This patch makes us use R32_UINT instead.
> > ---
> > src/intel/blorp/blorp_blit.c | 11 ++++++-----
> > 1 file changed, 6 insertions(+), 5 deletions(-)
> >
> > diff --git a/src/intel/blorp/blorp_blit.c b/src/intel/blorp/blorp_blit.c
> > index c85ec8543a9..7b49f9afa35 100644
> > --- a/src/intel/blorp/blorp_blit.c
> > +++ b/src/intel/blorp/blorp_blit.c
> > @@ -984,14 +984,15 @@ convert_color(struct nir_builder *b, nir_ssa_def 
> > *color,
> >    nir_ssa_def *value;
> >
> >    if (key->dst_format == ISL_FORMAT_R24_UNORM_X8_TYPELESS) {
> > -      /* The destination image is bound as R32_UNORM but the data needs to 
> > be
> > +      /* The destination image is bound as R32_UINT but the data needs to 
> > be
> >        * in R24_UNORM_X8_TYPELESS.  The bottom 24 are the actual data and 
> > the
> >        * top 8 need to be zero.  We can accomplish this by simply 
> > multiplying
> >        * by a factor to scale things down.
> >        */
> > -      float factor = (float)((1 << 24) - 1) / (float)UINT32_MAX;
> > -      value = nir_fmul(b, nir_fsat(b, nir_channel(b, color, 0)),
> > -                          nir_imm_float(b, factor));
> > +      unsigned factor = (1 << 24) - 1;
> > +      value = nir_fsat(b, nir_channel(b, color, 0));
> > +      value = nir_fmul(b, value, nir_imm_float(b, factor));
> > +      value = nir_iand(b, nir_f2i32(b, value), nir_imm_int(b, factor));
> 
> Do we really need the and?

Probably not - without it, we'll just leave garbage in the X8 bits
instead of zeroing them.  But that's probably fine, unless it affects
HiZ compression somehow?  I was mostly borrowing from u_format_zs.c
code to do this without thinking too hard.

Attachment: signature.asc
Description: This is a digitally signed message part.

_______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Reply via email to