Re: [Mesa-dev] [PATCH 11/22 v2] nir: Recognize some more open-coded fmin / fmax

2018-03-05 Thread Jason Ekstrand 
On Mon, Mar 5, 2018 at 2:25 PM, Ian Romanick  wrote:

> On 03/05/2018 02:07 PM, Jason Ekstrand wrote:
> > On Mon, Mar 5, 2018 at 11:11 AM, Ian Romanick  > > wrote:
> >
> > On 02/28/2018 12:48 PM, Jason Ekstrand wrote:
> > > Rb
> >
> > For both or just this one?
> >
> >
> > If you put the ~ on there, both.
>
> I'm not sure what you mean.  That's all the other patch ([PATCH 11.1/22]
> nir: Mark bcsel-to-fmin (or fmax) transformations as inexact) does.
>

Sorry, I got confused.  Yes, RB to both.


> > > On Wed, Feb 28, 2018 at 12:18 PM, Ian Romanick <
> i...@freedesktop.org 
> > > >> wrote:
> > >
> > > From: Ian Romanick 
> > >  > >>
> > >
> > > This transformation is inexact because section 4.7.1 (Range and
> > > Precision) says:
> > >
> > > Operations and built-in functions that operate on a NaN
> > are not
> > > required to return a NaN as the result.
> > >
> > > The fmin or fmax might not return NaN in cases where the
> original
> > > expression would be required to return NaN.
> > >
> > > v2: Reorder operands and mark as inexact.  The latter
> suggested by
> > > Jason.
> > >
> > > shader-db results:
> > >
> > > Haswell, Broadwell, and Skylake had similar results. (Skylake
> > shown)
> > > total instructions in shared programs: 14514817 -> 14514808
> > (<.01%)
> > > instructions in affected programs: 229 -> 220 (-3.93%)
> > > helped: 3
> > > HURT: 0
> > > helped stats (abs) min: 1 max: 4 x̄: 3.00 x̃: 4
> > > helped stats (rel) min: 2.86% max: 4.12% x̄: 3.70% x̃: 4.12%
> > >
> > > total cycles in shared programs: 533145211 -> 533144939 (<.01%)
> > > cycles in affected programs: 37268 -> 36996 (-0.73%)
> > > helped: 8
> > > HURT: 0
> > > helped stats (abs) min: 2 max: 134 x̄: 34.00 x̃: 2
> > > helped stats (rel) min: 0.02% max: 14.22% x̄: 3.53% x̃: 0.05%
> > >
> > > Sandy Bridge and Ivy Bridge had similar results. (Ivy Bridge
> > shown)
> > > total cycles in shared programs: 257618409 -> 257618403 (<.01%)
> > > cycles in affected programs: 12582 -> 12576 (-0.05%)
> > > helped: 3
> > > HURT: 0
> > > helped stats (abs) min: 2 max: 2 x̄: 2.00 x̃: 2
> > > helped stats (rel) min: 0.05% max: 0.05% x̄: 0.05% x̃: 0.05%
> > >
> > > No changes on Iron Lake or GM45.
> > >
> > > Signed-off-by: Ian Romanick  > 
> > >  > >>
> > > ---
> > >  src/compiler/nir/nir_opt_algebraic.py | 2 ++
> > >  1 file changed, 2 insertions(+)
> > >
> > > diff --git a/src/compiler/nir/nir_opt_algebraic.py
> > > b/src/compiler/nir/nir_opt_algebraic.py
> > > index d40d59b..17f4d9c 100644
> > > --- a/src/compiler/nir/nir_opt_algebraic.py
> > > +++ b/src/compiler/nir/nir_opt_algebraic.py
> > > @@ -170,6 +170,8 @@ optimizations = [
> > > (('fge', ('fneg', ('fabs', a)), 0.0), ('feq', a, 0.0)),
> > > (('bcsel', ('flt', b, a), b, a), ('fmin', a, b)),
> > > (('bcsel', ('flt', a, b), b, a), ('fmax', a, b)),
> > > +   (('~bcsel', ('fge', a, b), b, a), ('fmin', a, b)),
> > > +   (('~bcsel', ('fge', b, a), b, a), ('fmax', a, b)),
> > > (('bcsel', ('inot', a), b, c), ('bcsel', a, c, b)),
> > > (('bcsel', a, ('bcsel', a, b, c), d), ('bcsel', a, b, d)),
> > > (('bcsel', a, True, 'b@bool'), ('ior', a, b)),
> > > --
> > > 2.9.5
> > >
> > > ___
> > > mesa-dev mailing list
> > > mesa-dev@lists.freedesktop.org
> > 
> >  > >
> > > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> > 
> > >  > >
> > >
> > >
> >
> >
>
>
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Re: [Mesa-dev] [PATCH 11/22 v2] nir: Recognize some more open-coded fmin / fmax

2018-03-05 Thread Ian Romanick 
On 03/05/2018 02:07 PM, Jason Ekstrand wrote:
> On Mon, Mar 5, 2018 at 11:11 AM, Ian Romanick  > wrote:
> 
> On 02/28/2018 12:48 PM, Jason Ekstrand wrote:
> > Rb
> 
> For both or just this one?
> 
> 
> If you put the ~ on there, both.

I'm not sure what you mean.  That's all the other patch ([PATCH 11.1/22]
nir: Mark bcsel-to-fmin (or fmax) transformations as inexact) does.

> > On Wed, Feb 28, 2018 at 12:18 PM, Ian Romanick  
> > >> wrote:
> >
> >     From: Ian Romanick  
> >      >>
> >
> >     This transformation is inexact because section 4.7.1 (Range and
> >     Precision) says:
> >
> >         Operations and built-in functions that operate on a NaN
> are not
> >         required to return a NaN as the result.
> >
> >     The fmin or fmax might not return NaN in cases where the original
> >     expression would be required to return NaN.
> >
> >     v2: Reorder operands and mark as inexact.  The latter suggested by
> >     Jason.
> >
> >     shader-db results:
> >
> >     Haswell, Broadwell, and Skylake had similar results. (Skylake
> shown)
> >     total instructions in shared programs: 14514817 -> 14514808
> (<.01%)
> >     instructions in affected programs: 229 -> 220 (-3.93%)
> >     helped: 3
> >     HURT: 0
> >     helped stats (abs) min: 1 max: 4 x̄: 3.00 x̃: 4
> >     helped stats (rel) min: 2.86% max: 4.12% x̄: 3.70% x̃: 4.12%
> >
> >     total cycles in shared programs: 533145211 -> 533144939 (<.01%)
> >     cycles in affected programs: 37268 -> 36996 (-0.73%)
> >     helped: 8
> >     HURT: 0
> >     helped stats (abs) min: 2 max: 134 x̄: 34.00 x̃: 2
> >     helped stats (rel) min: 0.02% max: 14.22% x̄: 3.53% x̃: 0.05%
> >
> >     Sandy Bridge and Ivy Bridge had similar results. (Ivy Bridge
> shown)
> >     total cycles in shared programs: 257618409 -> 257618403 (<.01%)
> >     cycles in affected programs: 12582 -> 12576 (-0.05%)
> >     helped: 3
> >     HURT: 0
> >     helped stats (abs) min: 2 max: 2 x̄: 2.00 x̃: 2
> >     helped stats (rel) min: 0.05% max: 0.05% x̄: 0.05% x̃: 0.05%
> >
> >     No changes on Iron Lake or GM45.
> >
> >     Signed-off-by: Ian Romanick  
> >      >>
> >     ---
> >      src/compiler/nir/nir_opt_algebraic.py | 2 ++
> >      1 file changed, 2 insertions(+)
> >
> >     diff --git a/src/compiler/nir/nir_opt_algebraic.py
> >     b/src/compiler/nir/nir_opt_algebraic.py
> >     index d40d59b..17f4d9c 100644
> >     --- a/src/compiler/nir/nir_opt_algebraic.py
> >     +++ b/src/compiler/nir/nir_opt_algebraic.py
> >     @@ -170,6 +170,8 @@ optimizations = [
> >         (('fge', ('fneg', ('fabs', a)), 0.0), ('feq', a, 0.0)),
> >         (('bcsel', ('flt', b, a), b, a), ('fmin', a, b)),
> >         (('bcsel', ('flt', a, b), b, a), ('fmax', a, b)),
> >     +   (('~bcsel', ('fge', a, b), b, a), ('fmin', a, b)),
> >     +   (('~bcsel', ('fge', b, a), b, a), ('fmax', a, b)),
> >         (('bcsel', ('inot', a), b, c), ('bcsel', a, c, b)),
> >         (('bcsel', a, ('bcsel', a, b, c), d), ('bcsel', a, b, d)),
> >         (('bcsel', a, True, 'b@bool'), ('ior', a, b)),
> >     --
> >     2.9.5
> >
> >     ___
> >     mesa-dev mailing list
> >     mesa-dev@lists.freedesktop.org
> 
>  >
> >     https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> 
> >      >
> >
> >
> 
> 

___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Re: [Mesa-dev] [PATCH 11/22 v2] nir: Recognize some more open-coded fmin / fmax

2018-03-05 Thread Jason Ekstrand 
On Mon, Mar 5, 2018 at 11:11 AM, Ian Romanick  wrote:

> On 02/28/2018 12:48 PM, Jason Ekstrand wrote:
> > Rb
>
> For both or just this one?
>

If you put the ~ on there, both.


> > On Wed, Feb 28, 2018 at 12:18 PM, Ian Romanick  > > wrote:
> >
> > From: Ian Romanick  > >
> >
> > This transformation is inexact because section 4.7.1 (Range and
> > Precision) says:
> >
> > Operations and built-in functions that operate on a NaN are not
> > required to return a NaN as the result.
> >
> > The fmin or fmax might not return NaN in cases where the original
> > expression would be required to return NaN.
> >
> > v2: Reorder operands and mark as inexact.  The latter suggested by
> > Jason.
> >
> > shader-db results:
> >
> > Haswell, Broadwell, and Skylake had similar results. (Skylake shown)
> > total instructions in shared programs: 14514817 -> 14514808 (<.01%)
> > instructions in affected programs: 229 -> 220 (-3.93%)
> > helped: 3
> > HURT: 0
> > helped stats (abs) min: 1 max: 4 x̄: 3.00 x̃: 4
> > helped stats (rel) min: 2.86% max: 4.12% x̄: 3.70% x̃: 4.12%
> >
> > total cycles in shared programs: 533145211 -> 533144939 (<.01%)
> > cycles in affected programs: 37268 -> 36996 (-0.73%)
> > helped: 8
> > HURT: 0
> > helped stats (abs) min: 2 max: 134 x̄: 34.00 x̃: 2
> > helped stats (rel) min: 0.02% max: 14.22% x̄: 3.53% x̃: 0.05%
> >
> > Sandy Bridge and Ivy Bridge had similar results. (Ivy Bridge shown)
> > total cycles in shared programs: 257618409 -> 257618403 (<.01%)
> > cycles in affected programs: 12582 -> 12576 (-0.05%)
> > helped: 3
> > HURT: 0
> > helped stats (abs) min: 2 max: 2 x̄: 2.00 x̃: 2
> > helped stats (rel) min: 0.05% max: 0.05% x̄: 0.05% x̃: 0.05%
> >
> > No changes on Iron Lake or GM45.
> >
> > Signed-off-by: Ian Romanick  > >
> > ---
> >  src/compiler/nir/nir_opt_algebraic.py | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/src/compiler/nir/nir_opt_algebraic.py
> > b/src/compiler/nir/nir_opt_algebraic.py
> > index d40d59b..17f4d9c 100644
> > --- a/src/compiler/nir/nir_opt_algebraic.py
> > +++ b/src/compiler/nir/nir_opt_algebraic.py
> > @@ -170,6 +170,8 @@ optimizations = [
> > (('fge', ('fneg', ('fabs', a)), 0.0), ('feq', a, 0.0)),
> > (('bcsel', ('flt', b, a), b, a), ('fmin', a, b)),
> > (('bcsel', ('flt', a, b), b, a), ('fmax', a, b)),
> > +   (('~bcsel', ('fge', a, b), b, a), ('fmin', a, b)),
> > +   (('~bcsel', ('fge', b, a), b, a), ('fmax', a, b)),
> > (('bcsel', ('inot', a), b, c), ('bcsel', a, c, b)),
> > (('bcsel', a, ('bcsel', a, b, c), d), ('bcsel', a, b, d)),
> > (('bcsel', a, True, 'b@bool'), ('ior', a, b)),
> > --
> > 2.9.5
> >
> > ___
> > mesa-dev mailing list
> > mesa-dev@lists.freedesktop.org  freedesktop.org>
> > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> > 
> >
> >
>
>
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Re: [Mesa-dev] [PATCH 11/22 v2] nir: Recognize some more open-coded fmin / fmax

2018-03-05 Thread Ian Romanick 
On 02/28/2018 12:48 PM, Jason Ekstrand wrote:
> Rb

For both or just this one?

> On Wed, Feb 28, 2018 at 12:18 PM, Ian Romanick  > wrote:
> 
> From: Ian Romanick  >
> 
> This transformation is inexact because section 4.7.1 (Range and
> Precision) says:
> 
>     Operations and built-in functions that operate on a NaN are not
>     required to return a NaN as the result.
> 
> The fmin or fmax might not return NaN in cases where the original
> expression would be required to return NaN.
> 
> v2: Reorder operands and mark as inexact.  The latter suggested by
> Jason.
> 
> shader-db results:
> 
> Haswell, Broadwell, and Skylake had similar results. (Skylake shown)
> total instructions in shared programs: 14514817 -> 14514808 (<.01%)
> instructions in affected programs: 229 -> 220 (-3.93%)
> helped: 3
> HURT: 0
> helped stats (abs) min: 1 max: 4 x̄: 3.00 x̃: 4
> helped stats (rel) min: 2.86% max: 4.12% x̄: 3.70% x̃: 4.12%
> 
> total cycles in shared programs: 533145211 -> 533144939 (<.01%)
> cycles in affected programs: 37268 -> 36996 (-0.73%)
> helped: 8
> HURT: 0
> helped stats (abs) min: 2 max: 134 x̄: 34.00 x̃: 2
> helped stats (rel) min: 0.02% max: 14.22% x̄: 3.53% x̃: 0.05%
> 
> Sandy Bridge and Ivy Bridge had similar results. (Ivy Bridge shown)
> total cycles in shared programs: 257618409 -> 257618403 (<.01%)
> cycles in affected programs: 12582 -> 12576 (-0.05%)
> helped: 3
> HURT: 0
> helped stats (abs) min: 2 max: 2 x̄: 2.00 x̃: 2
> helped stats (rel) min: 0.05% max: 0.05% x̄: 0.05% x̃: 0.05%
> 
> No changes on Iron Lake or GM45.
> 
> Signed-off-by: Ian Romanick  >
> ---
>  src/compiler/nir/nir_opt_algebraic.py | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/src/compiler/nir/nir_opt_algebraic.py
> b/src/compiler/nir/nir_opt_algebraic.py
> index d40d59b..17f4d9c 100644
> --- a/src/compiler/nir/nir_opt_algebraic.py
> +++ b/src/compiler/nir/nir_opt_algebraic.py
> @@ -170,6 +170,8 @@ optimizations = [
>     (('fge', ('fneg', ('fabs', a)), 0.0), ('feq', a, 0.0)),
>     (('bcsel', ('flt', b, a), b, a), ('fmin', a, b)),
>     (('bcsel', ('flt', a, b), b, a), ('fmax', a, b)),
> +   (('~bcsel', ('fge', a, b), b, a), ('fmin', a, b)),
> +   (('~bcsel', ('fge', b, a), b, a), ('fmax', a, b)),
>     (('bcsel', ('inot', a), b, c), ('bcsel', a, c, b)),
>     (('bcsel', a, ('bcsel', a, b, c), d), ('bcsel', a, b, d)),
>     (('bcsel', a, True, 'b@bool'), ('ior', a, b)),
> --
> 2.9.5
> 
> ___
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org 
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> 
> 
> 

___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Re: [Mesa-dev] [PATCH 11/22 v2] nir: Recognize some more open-coded fmin / fmax

2018-02-28 Thread Jason Ekstrand 
Rb

On Wed, Feb 28, 2018 at 12:18 PM, Ian Romanick  wrote:

> From: Ian Romanick 
>
> This transformation is inexact because section 4.7.1 (Range and
> Precision) says:
>
> Operations and built-in functions that operate on a NaN are not
> required to return a NaN as the result.
>
> The fmin or fmax might not return NaN in cases where the original
> expression would be required to return NaN.
>
> v2: Reorder operands and mark as inexact.  The latter suggested by
> Jason.
>
> shader-db results:
>
> Haswell, Broadwell, and Skylake had similar results. (Skylake shown)
> total instructions in shared programs: 14514817 -> 14514808 (<.01%)
> instructions in affected programs: 229 -> 220 (-3.93%)
> helped: 3
> HURT: 0
> helped stats (abs) min: 1 max: 4 x̄: 3.00 x̃: 4
> helped stats (rel) min: 2.86% max: 4.12% x̄: 3.70% x̃: 4.12%
>
> total cycles in shared programs: 533145211 -> 533144939 (<.01%)
> cycles in affected programs: 37268 -> 36996 (-0.73%)
> helped: 8
> HURT: 0
> helped stats (abs) min: 2 max: 134 x̄: 34.00 x̃: 2
> helped stats (rel) min: 0.02% max: 14.22% x̄: 3.53% x̃: 0.05%
>
> Sandy Bridge and Ivy Bridge had similar results. (Ivy Bridge shown)
> total cycles in shared programs: 257618409 -> 257618403 (<.01%)
> cycles in affected programs: 12582 -> 12576 (-0.05%)
> helped: 3
> HURT: 0
> helped stats (abs) min: 2 max: 2 x̄: 2.00 x̃: 2
> helped stats (rel) min: 0.05% max: 0.05% x̄: 0.05% x̃: 0.05%
>
> No changes on Iron Lake or GM45.
>
> Signed-off-by: Ian Romanick 
> ---
>  src/compiler/nir/nir_opt_algebraic.py | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/src/compiler/nir/nir_opt_algebraic.py
> b/src/compiler/nir/nir_opt_algebraic.py
> index d40d59b..17f4d9c 100644
> --- a/src/compiler/nir/nir_opt_algebraic.py
> +++ b/src/compiler/nir/nir_opt_algebraic.py
> @@ -170,6 +170,8 @@ optimizations = [
> (('fge', ('fneg', ('fabs', a)), 0.0), ('feq', a, 0.0)),
> (('bcsel', ('flt', b, a), b, a), ('fmin', a, b)),
> (('bcsel', ('flt', a, b), b, a), ('fmax', a, b)),
> +   (('~bcsel', ('fge', a, b), b, a), ('fmin', a, b)),
> +   (('~bcsel', ('fge', b, a), b, a), ('fmax', a, b)),
> (('bcsel', ('inot', a), b, c), ('bcsel', a, c, b)),
> (('bcsel', a, ('bcsel', a, b, c), d), ('bcsel', a, b, d)),
> (('bcsel', a, True, 'b@bool'), ('ior', a, b)),
> --
> 2.9.5
>
> ___
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
___
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev