[Mesa-stable] [Mesa-dev] [PATCH] nir: Use double-precision pow() when bit_size is 64, powf() otherwise

[Jason Ekstrand]

On Mon, Mar 28, 2016 at 11:31 AM, Ian Romanick <idr at freedesktop.org> wrote:

> On 03/23/2016 08:09 PM, Jason Ekstrand wrote:
> > Is there a 64-bit pow in GLSL?  If so, this is the right implementation.
>
> There isn't.  But there also aren't double versions of sin or cos, and
> 9076c4e2 added those.  However, it didn't add double versions of fexp2
> or flog2 (both of which are not supported in GLSL).
>
> Ugh... it also did not add double versions of frsq or fsqrt... which are
> supported in GLSL.
>
> So, there are some extra functions and some missing functions.  My gut
> tells me we should just do all of them so that people don't have to
> think about which ones should / shouldn't be implemented.
>

That seems reasonable.  Either that or flag those opcodes as only
supporting 32-bit types for now.  I think I'd rather just implement them if
it's not too much work.  You know someone's going to write a an fp64_pow
extension some day...


> Looking more closely... I think fabs, fnot, fsign, and fsat should have
> explicit double support as well... it seems the existing implementations
> of both of these will cause some small values to be incorrectly
> "flushed" to zero.  What are the rules in C when you compare a double
> variable with a single constant?
>
> void foo(double d)
> {
>     /* Does d get converted to single, or does 0.0f get converted to
>      * double?
>      */
>     if (d == 0.0f)
>         printf("zero\n");
> }
>
> > Reviewed-by: Jason Ekstrand <jason at jlekstrand.net
> > <mailto:jason at jlekstrand.net>>
> >
> > On Mar 23, 2016 7:42 PM, "Ian Romanick" <idr at freedesktop.org
> > <mailto:idr at freedesktop.org>> wrote:
> >
> >     From: Ian Romanick <ian.d.romanick at intel.com
> >     <mailto:ian.d.romanick at intel.com>>
> >
> >     Found (randomly) by inspection.  Looking at the rest of the changes
> in
> >     this file in commit 9076c4e2, I'm certain this is what was intended.
> >
> >     Signed-off-by: Ian Romanick <ian.d.romanick at intel.com
> >     <mailto:ian.d.romanick at intel.com>>
> >     Cc: Connor Abbott <cwabbott0 at gmail.com <mailto:cwabbott0 at gmail.com>>
> >     Cc: mesa-stable at lists.freedesktop.org
> >     <mailto:mesa-stable at lists.freedesktop.org>
> >     ---
> >      src/compiler/nir/nir_opcodes.py | 2 +-
> >      1 file changed, 1 insertion(+), 1 deletion(-)
> >
> >     diff --git a/src/compiler/nir/nir_opcodes.py
> >     b/src/compiler/nir/nir_opcodes.py
> >     index 553f924..ac59130 100644
> >     --- a/src/compiler/nir/nir_opcodes.py
> >     +++ b/src/compiler/nir/nir_opcodes.py
> >     @@ -520,7 +520,7 @@ for (int i = 0; i < 32; i += 8) {
> >      }
> >      """)
> >
> >     -binop("fpow", tfloat, "", "bit_size == 64 ? powf(src0, src1) :
> >     pow(src0, src1)")
> >     +binop("fpow", tfloat, "", "bit_size == 64 ? pow(src0, src1) :
> >     powf(src0, src1)")
> >
> >      binop_horiz("pack_half_2x16_split", 1, tuint32, 1, tfloat32, 1,
> >     tfloat32,
> >                  "pack_half_1x16(src0.x) | (pack_half_1x16(src1.x) <<
> 16)")
> >     --
> >     2.5.5
> >
> >     _______________________________________________
> >     mesa-dev mailing list
> >     mesa-dev at lists.freedesktop.org <mailto:
> 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-stable/attachments/20160328/4f3ab40c/attachment.html>