[Mesa-dev] [PATCH v3 134/133] nir/opcodes: Add algebraic properties metadata

[Marek Olšák]

Hi Jason,

The DX10 version of min/max is indeed commutative. However, GLSL specs
define min/max using a conditional assignment:

mix(x) = y < x ? y : x;
max(x) = x < y ? y : x;

which means that if either argument is NaN, x is returned., which
means that you can combine open-coded min/max functions into min/max
instructions if your hardware implements this behavior. It also means
that if your hardware doesn't implement this behavior, you should
lower min/max to conditional assignments if you want it to be super
correct.

Marek


On Sat, Dec 20, 2014 at 3:09 AM, Jason Ekstrand <jason at jlekstrand.net> wrote:
>
>
> On Fri, Dec 19, 2014 at 6:02 PM, Matt Turner <mattst88 at gmail.com> wrote:
>>
>> On Fri, Dec 19, 2014 at 5:04 PM, Jason Ekstrand <jason at jlekstrand.net>
>> wrote:
>> > v3: fmin and fmax technically aren't commutative or associative.  Things
>> >     get funny when one of the arguments is a NaN.
>>
>> I have a difficult time believing that the GLSL definitions of min()
>> and max() intentionally have that effect.
>>
>> To recap, the definition of min() in GLSL is
>>
>> > Returns y if y < x, otherwise it returns x.
>>
>> and since comparisons against NaN are always false, min(x, y) where x
>> or y is NaN will always return x, regardless of which argument is NaN.
>> That doesn't seem like a useful behavior.
>>
>> Indeed the GLSL spec says
>>
>> > Operations and built-in functions that operate on a NaN are not required
>> > to return a NaN as the result.
>>
>> I think based on that text alone, min() and max() should be considered
>> commutative and associative.
>
>
> As a datapoint, apparently Merek (cc'd) bumped into this at one point on
> some unreal game.  Perhaps he could shed some light.
>
>>
>> Ian, perhaps we should clarify this with Khronos?
>>
>> An IEEE 754 spec draft [1] defines minNum(x, y) as "y if x<y, x if
>> y<x, the canonicalized floating-point
>> number if one operand is a floating-point number and the other a NaN.
>> Otherwise it is either x
>> or y."
>
>
> That seems to imply that min(x, NaN) = min(NaN, x) = x which would make it
> commutative and associative.
>
>>
>> i965's SEL instruction implements that behavior.
>>
>> [1] http://754r.ucbtest.org/drafts/archive/2006-10-04.pdf