[Mesa-dev] [RFC PATCH] nir/algebraic: Simplify max(abs(a), 0.0) -> abs(a)

[Ian Romanick]

On 5/23/19 7:54 PM, Ilia Mirkin wrote:
> How does max(NaN, 0) work? IIRC there's some provision that this
> becomes 0, while abs(NaN) = NaN.

That is correct.  There are a couple other algebraic patterns that have
the same potential problem (e.g., the min(max()) -> sat() patterns), and
we just make those as imprecise with ~.  I /think/ that should be
adequate here too.

> On Thu, May 23, 2019 at 10:47 PM Alyssa Rosenzweig <alyssa at rosenzweig.io> wrote:
>>
>> I noticed this pattern in glmark's jellyfish scene.
>>
>> Assuming this is correct (it should be...?), could someone do a
>> shader-db run? Thank you!
>>
>> Signed-off-by: Alyssa Rosenzweig <alyssa at rosenzweig.io>
>> Cc: Ian Romanick <ian.d.romanick at intel.com>
>> ---
>>  src/compiler/nir/nir_opt_algebraic.py | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
>> index 89d07aa1261..abd0b6591ce 100644
>> --- a/src/compiler/nir/nir_opt_algebraic.py
>> +++ b/src/compiler/nir/nir_opt_algebraic.py
>> @@ -377,6 +377,7 @@ optimizations = [
>>     (('imax', a, a), a),
>>     (('umin', a, a), a),
>>     (('umax', a, a), a),
>> +   (('fmax', ('fabs', a), 0.0), ('fabs', a)),
>>     (('fmax', ('fmax', a, b), b), ('fmax', a, b)),
>>     (('umax', ('umax', a, b), b), ('umax', a, b)),
>>     (('imax', ('imax', a, b), b), ('imax', a, b)),
>> --
>> 2.20.1
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>