[Mesa-dev] [PATCH 23/29] nir: add i2d and u2d opcodes

[Jason Ekstrand]

On Mar 22, 2016 8:19 AM, "Samuel Iglesias Gonsálvez" <siglesias at igalia.com>
wrote:
>
>
>
> On 21/03/16 23:56, Jason Ekstrand wrote:
> > On Mon, Mar 21, 2016 at 5:06 AM, Samuel Iglesias Gonsálvez <
> > siglesias at igalia.com> wrote:
> >
> >> From: Iago Toral Quiroga <itoral at igalia.com>
> >>
> >> ---
> >>  src/compiler/nir/glsl_to_nir.cpp | 6 ++++++
> >>  src/compiler/nir/nir_opcodes.py  | 2 ++
> >>  2 files changed, 8 insertions(+)
> >>
> >> diff --git a/src/compiler/nir/glsl_to_nir.cpp
> >> b/src/compiler/nir/glsl_to_nir.cpp
> >> index 952d787..d087a77 100644
> >> --- a/src/compiler/nir/glsl_to_nir.cpp
> >> +++ b/src/compiler/nir/glsl_to_nir.cpp
> >> @@ -1357,6 +1357,12 @@ nir_visitor::visit(ir_expression *ir)
> >>     case ir_unop_d2i:  result = nir_d2i(&b, srcs[0]);   break;
> >>     case ir_unop_d2u:  result = nir_d2u(&b, srcs[0]);   break;
> >>     case ir_unop_d2b:  result = nir_d2b(&b, srcs[0]);   break;
> >> +   case ir_unop_i2d:
> >> +      result = supports_ints ? nir_i2d(&b, srcs[0]) : nir_fmov(&b,
> >> srcs[0]);
> >> +      break;
> >> +   case ir_unop_u2d:
> >> +      result = supports_ints ? nir_u2d(&b, srcs[0]) : nir_fmov(&b,
> >> srcs[0]);
> >>
> >
> > If you're going to be using the u2d opcode, you'd better support
integers.
> >
>
> We did the same than integer to float conversions to keep this code
> aligned with what they do.

Right.  I don't think that would be correct for hardware that doesn't have
integers anyway.  You would want an ftrunc in the non-integer case with an
abs for f2u.  NIR has yet to be used on any platforms that don't support
native integers so all those cases de-paths are dead anyway.

> We can add an assert for support_ints here and only call to nir_u2d but,
> to be consistent, we would need to do similar changes to i2d, u2f, i2f
> too in a separate patch.

Feel free to add an assert. I don't think updating the others is needed.
It's not that f2u is invalid without integers so much as no hardware that
supports doubles won't have native integers.

> What do you think?
>
> Sam
>
> >
> >> +      break;
> >>     case ir_unop_i2u:
> >>     case ir_unop_u2i:
> >>     case ir_unop_bitcast_i2f:
> >> diff --git a/src/compiler/nir/nir_opcodes.py
> >> b/src/compiler/nir/nir_opcodes.py
> >> index a161ac1..cf6ce83 100644
> >> --- a/src/compiler/nir/nir_opcodes.py
> >> +++ b/src/compiler/nir/nir_opcodes.py
> >> @@ -164,6 +164,7 @@ unop_convert("f2u", tuint32, tfloat32, "src0") #
> >> Float-to-unsigned conversion
> >>  unop_convert("d2i", tint32, tfloat64, "src0") # Double-to-integer
> >> conversion.
> >>  unop_convert("d2u", tuint32, tfloat64, "src0") # Double-to-unsigned
> >> conversion.
> >>  unop_convert("i2f", tfloat32, tint32, "src0") # Integer-to-float
> >> conversion.
> >> +unop_convert("i2d", tfloat64, tint32, "src0") # Integer-to-double
> >> conversion.
> >>  # Float-to-boolean conversion
> >>  unop_convert("f2b", tbool, tfloat32, "src0 != 0.0f")
> >>  unop_convert("d2b", tbool, tfloat64, "src0 != 0.0")
> >> @@ -173,6 +174,7 @@ unop_convert("b2f", tfloat32, tbool, "src0 ? 1.0f :
> >> 0.0f")
> >>  unop_convert("i2b", tbool, tint32, "src0 != 0")
> >>  unop_convert("b2i", tint32, tbool, "src0 ? 1 : 0") # Boolean-to-int
> >> conversion
> >>  unop_convert("u2f", tfloat32, tuint32, "src0") # Unsigned-to-float
> >> conversion.
> >> +unop_convert("u2d", tfloat64, tuint32, "src0") # Unsigned-to-double
> >> conversion.
> >>  # double-to-float conversion
> >>  unop_convert("d2f", tfloat32, tfloat64, "src0") # Single to double
> >> precision
> >>  unop_convert("f2d", tfloat64, tfloat32, "src0") # Double to single
> >> precision
> >> --
> >> 2.5.0
> >>
> >> _______________________________________________
> >> mesa-dev mailing list
> >> mesa-dev at lists.freedesktop.org
> >> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> >>
> >
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160322/324079ad/attachment.html>