Common simd instructions and the gcc vector extensions

[Hendrik Siedelmann]

Hello everybody,

I'm trying to use the vector extensions, but I'm unable to find common
functions like min/max or saturated arithmetic. Given that those are
pretty essential simd instructions I expected something like
_builtin_min(). Are there other ways to express those instructions?

Cheers,
Hendrik

[Marc Glisse]

Post by Hendrik Siedelmann
I'm trying to use the vector extensions, but I'm unable to find common
functions like min/max or saturated arithmetic. Given that those are pretty
essential simd instructions I expected something like _builtin_min(). Are
there other ways to express those instructions?

Did you try (a<b)?a:b ?

[Hendrik Siedelmann]

Post by Marc Glisse

Post by Hendrik Siedelmann
I'm trying to use the vector extensions, but I'm unable to find common
functions like min/max or saturated arithmetic. Given that those are
pretty essential simd instructions I expected something like
_builtin_min(). Are there other ways to express those instructions?

Did you try (a<b)?a:b ?

You mean element-wise? That doesn't get vectorized for some reason. Or
do you mean with whole vectors? That complains with:
error: used vector type where scalar is required
Are there tricks like this for saturated arithmetics?

Thanks
Hendrik

[Marc Glisse]

Post by Hendrik Siedelmann

Post by Marc Glisse

Post by Hendrik Siedelmann
I'm trying to use the vector extensions, but I'm unable to find common
functions like min/max or saturated arithmetic. Given that those are
pretty essential simd instructions I expected something like
_builtin_min(). Are there other ways to express those instructions?

Did you try (a<b)?a:b ?

You mean element-wise?

No.

Post by Hendrik Siedelmann
That doesn't get vectorized for some reason.

Some versions are (one of the vectors involved has to be non-local).
I hope this limitation will be lifted some day.

Post by Hendrik Siedelmann
Or do you mean with whole vectors? That complains with: error: used
vector type where scalar is required

That's only supported in C++, sorry.

Post by Hendrik Siedelmann
Are there tricks like this for saturated arithmetics?

I am not familiar with that so I cannot answer.