Improve min max performance and IEEE compliance

[CoryCharlton]

Description

• Reduce IsNan checks to improve performance
• Fix Min not returning NaN in all cases it should
• Implement Min(float) and Max(float)

Motivation and Context

• Fixes Math.Max(float) and Math.Min(float) are 3x slower than the double overloads Home#1395
• Fixes Math.Min(double/float) do not handle NaN correctly Home#1396

How Has This Been Tested?

• On device
• Unit tests

Screenshots

Types of changes

• Improvement (non-breaking change that improves a feature, code or algorithm)
• Bug fix (non-breaking change which fixes an issue with code or algorithm)
• New feature (non-breaking change which adds functionality to code)
• Breaking change (fix or feature that would cause existing functionality to change)
• Config and build (change in the configuration and build system, has no impact on code or features)
• Dependencies (update dependencies and changes associated, has no impact on code or features)
• Unit Tests (add new Unit Test(s) or improved existing one(s), has no impact on code or features)
• Documentation (changes or updates in the documentation, has no impact on code or features)

Checklist:

• My code follows the code style of this project (only if there are changes in source code).
• My changes require an update to the documentation (there are changes that require the docs website to be updated).
• I have updated the documentation accordingly (the changes require an update on the docs in this repo).
• I have read the CONTRIBUTING document.
• I have tested everything locally and all new and existing tests passed (only if there are changes in source code).
• I have added new tests to cover my changes.

[CoryCharlton]

Look like the build failure is because the firmware changes need to be merged first.

[CoryCharlton]

For reference:

Before...

| MethodName | ItterationCount | Mean                  | Min   | Max   |
| -------------------------------------------------------------------- |
| Max_Double | 250             | 8.7599999999999998 ms | 0 ms  | 10 ms |
| Max_Float  | 250             | 43.52 ms              | 30 ms | 50 ms |
| Max_Int    | 250             | 22.199999999999998 ms | 20 ms | 30 ms |
| Min_Double | 250             | 13.32 ms              | 10 ms | 20 ms |
| Min_Float  | 250             | 28.959999999999999 ms | 20 ms | 40 ms |
| Min_Int    | 250             | 13.72 ms              | 10 ms | 20 ms |

After...

| MethodName | ItterationCount | Mean     | Min   | Max   |
| ------------------------------------------------------- |
| Max_Double | 250             | 6.48 ms  | 0 ms  | 10 ms |
| Max_Float  | 250             | 7.04 ms  | 0 ms  | 10 ms |
| Max_Int    | 250             | 7.48 ms  | 0 ms  | 10 ms |
| Min_Double | 250             | 7 ms     | 0 ms  | 10 ms |
| Min_Float  | 250             | 7.8 ms   | 0 ms  | 10 ms |
| Min_Int    | 250             | 10.44 ms | 10 ms | 20 ms |

[josesimoes]

Look like the build failure is because the firmware changes need to be merged first.

That is correct. Currently a PR on another repo that depends on nanoCLR library has to have that one merged before it can be consumed in the pipeline.

[josesimoes]

Impressive! Thanks for all the work on this.
Great usage of the benchmark library! 😉

Overall all good. We're better moving the Max and Min to mscorlib.
Even if that's not release rigth away. It won't break anything.