[Mesa-dev] [PATCH 08/29] nir/format_convert: Add linear <-> sRGB helpers

[Pohjolainen, Topi]

On Tue, Mar 06, 2018 at 09:53:03AM -0800, Jason Ekstrand wrote:
> On Sat, Mar 3, 2018 at 11:57 PM, Pohjolainen, Topi <
> topi.pohjolainen at gmail.com> wrote:
> 
> > On Fri, Jan 26, 2018 at 05:59:37PM -0800, Jason Ekstrand wrote:
> > > ---
> > >  src/compiler/nir/nir_format_convert.h | 26 ++++++++++++++++++++++++++
> > >  1 file changed, 26 insertions(+)
> >
> > I don't what is the official reference for the formulas and I just took
> > what
> > google gave me first. There was discussion about the cutoff point between
> > linear and curved. I got the impression that what you chose below
> > (open-ended: 0 <= linear < 0.0031308f and 0 <= linear < 0.04045f instead of
> > close-ended) was more "correct". Just thought I check that anyway (that you
> > chose that on purpose).
> >
> 
> I did some digging.  Everyone agrees that it's supposed to be <= 0.04045
> for sRGB -> linear.  For linear -> sRGB, it's a bit more confused.  OpenGL
> 4.5, OpenGL ES 3.2, and the WSI chapter of Vulkan 1.0 all say < 0.0031308
> but the Khronos data format spec, the W3C spec it references, and Wikipedia
> all say <= 0.0031308.  I've sent an e-mail off to Andrew Garrard, the
> author of the data format spec, to see if he can shed some light on it.  In
> any case, no 8-bit UNORM value will ever hit this corner case, so I don't
> think it matters in practice.  In the mean time, I've adjusted the  NIR
> helpers to match the OpenGL [ES] convention for both formulas.

Great, thanks for checking it!

> 
> 
> > Reviewed-by: Topi Pohjolainen <topi.pohjolainen at intel.com>
> >
> > >
> > > diff --git a/src/compiler/nir/nir_format_convert.h
> > b/src/compiler/nir/nir_format_convert.h
> > > index e09c955..07618dc 100644
> > > --- a/src/compiler/nir/nir_format_convert.h
> > > +++ b/src/compiler/nir/nir_format_convert.h
> > > @@ -104,3 +104,29 @@ nir_format_pack_uint(nir_builder *b, nir_ssa_def
> > *color,
> > >     return nir_format_pack_uint_unmasked(b, nir_iand(b, color,
> > mask_imm),
> > >                                          bits, num_components);
> > >  }
> > > +
> > > +static inline nir_ssa_def *
> > > +nir_format_linear_to_srgb(nir_builder *b, nir_ssa_def *c)
> > > +{
> > > +   nir_ssa_def *linear = nir_fmul(b, c, nir_imm_float(b, 12.92f));
> > > +   nir_ssa_def *curved =
> > > +      nir_fsub(b, nir_fmul(b, nir_imm_float(b, 1.055f),
> > > +                              nir_fpow(b, c, nir_imm_float(b, 1.0 /
> > 2.4))),
> > > +                  nir_imm_float(b, 0.055f));
> > > +
> > > +   return nir_fsat(b, nir_bcsel(b, nir_flt(b, c, nir_imm_float(b,
> > 0.0031308f)),
> > > +                                   linear, curved));
> > > +}
> > > +
> > > +static inline nir_ssa_def *
> > > +nir_format_srgb_to_linear(nir_builder *b, nir_ssa_def *c)
> > > +{
> > > +   nir_ssa_def *linear = nir_fdiv(b, c, nir_imm_float(b, 12.92f));
> > > +   nir_ssa_def *curved =
> > > +      nir_fpow(b, nir_fdiv(b, nir_fadd(b, c, nir_imm_float(b, 0.055f)),
> > > +                              nir_imm_float(b, 1.055f)),
> > > +                  nir_imm_float(b, 2.4f));
> > > +
> > > +   return nir_fsat(b, nir_bcsel(b, nir_flt(b, c, nir_imm_float(b,
> > 0.04045f)),
> > > +                                   linear, curved));
> > > +}
> > > --
> > > 2.5.0.400.gff86faf
> > >
> > > _______________________________________________
> > > mesa-dev mailing list
> > > mesa-dev at lists.freedesktop.org
> > > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> >