Check if all conversions are performed correctly

[stijnh]

Casting between different floating-point type is a common task. Some of the conversions seems rather slow (such double to bfloat16). We need to check each combination of type and see if we use the fastest method available.

• double to float
• double to half
• double to bfloat16
• float to double
• float to half
• float to bfloat16
• half to double
• half to float
• half to bfloat16
• bfloat16 to double
• bfloat16 to float
• bfloat16 to half

FP8:

• double to fp8
• float to fp8
• half to fp8
• bfloat16 to fp8
• fp8 to double
• fp8 to float
• fp8 to half
• fp8 to bfloat16