`Math.Max(float)` and `Math.Min(float)` are 3x slower than the `double` overloads

[CoryCharlton]

Library/API/IoT binding

System.Math

Visual Studio version

No response

.NET nanoFramework extension version

No response

Target name(s)

No response

Firmware version

No response

Device capabilities

No response

Description

While benchmarking the System.Math function I noticed that Math.Max(float) and Math.Min(float) are 3x slower than Math.Max(double) and Math.Min(double).

While reviewing the code I see that both of the float overloads just delegate down to the double overload and cast the result to a float. I'm assuming this is the root cause of the performance difference.

How to reproduce

Run the benchmarks at https://github.com/CoryCharlton/CCSWE.nanoFramework.Math

Expected behaviour

Similar, if not faster, performance.

Screenshots

No response

Sample project or code

No response

Aditional information

Console export: MaxBenchmarks benchmark class.
| MethodName        | ItterationCount | Mean                  | Min   | Max   |
| --------------------------------------------------------------------------- |
| Max_Double_System | 250             | 13.12 ms              | 10 ms | 20 ms |
| Max_Float_System  | 250             | 29.32 ms              | 20 ms | 40 ms |

Console export: MinBenchmarks benchmark class.
| MethodName        | ItterationCount | Mean                  | Min   | Max   |
| --------------------------------------------------------------------------- |
| Min_Double_System | 250             | 11.2 ms               | 0 ms  | 20 ms |
| Min_Float_System  | 250             | 35.64 ms              | 20 ms | 50 ms |

[CoryCharlton]

I have no problem taking a stab at fixing/implementing this (well aside from getting a build environment setup) but I wanted to make sure I wasn't missing something regarding why the float overloads were not already implemented on the native side.

[Ellerbach]

Yes, native can be slower than pure manage in some cases. And yes, if you can make the PR, even better! Thanks a lot for the investigation. I guess it would be good to implement all in manages for both float and double and see when native is faster and when it's slower! And then make the best out of all this!

[nfbot]

🤨 Well... we've told you that this issue type was reserved to project Team Members. And we meant that. So, we're closing it now. Please choose an appropriate one..