Register use by wmma

[half-0]

I need to use wmma m8n32k16 api for some calculation, so I just wander how many register will be used by wmma operation per thread?

[Curefab]

You can look up the number of registers here for your data type: PTX ISA 8.5

Please be aware that PTX is an intermediate format; probably MMA instructions are used internally.

[half-0]

Thanks！