[PATCH 1/5] benchtests: Add benchmarks for frexp functions

[Osama Abdelkader]

On Wed, Oct 22, 2025 at 11:06:45AM +0200, Paul Zimmermann wrote:
>        Dear Osama,
> 
> on my computer I get for frexpf with your data set:
> 
>   "frexpf": {
>    "workload-random-m20-p20": {
>     "duration": 3.19203e+09,
>     "iterations": 2.616e+08,
>     "reciprocal-throughput": 10.5231,
>     "latency": 13.8808,
>     "max-throughput": 9.50294e+07,
>     "min-throughput": 7.20417e+07
>    }
>   }
> 
> If I use the larger data set from lgammaf_r-inputs, I get:
> 
>   "frexpf": {
>    "workload-random-m20-p20": {
>     "duration": 3.22859e+09,
>     "iterations": 2.84e+08,
>     "reciprocal-throughput": 9.01051,
>     "latency": 13.726,
>     "max-throughput": 1.10982e+08,
>     "min-throughput": 7.28542e+07
>    }
>   }
> 
> There is a significant difference for the reciprocal-throughput.
> Can you reproduce that difference?
> 
> Usually I use data sets of 1000 values. I see there are very small files,
> for example sqrt-inputs. Is there any recommendation for the size of the
> data sets?

No idea to be honest.

> 
> Paul
> 

Dear Paul,

Thanks for the test, I'm going to re-try that with data set from lgammaf_r-inputs to be on the same page.

Best regards,
Osama