"maxrreccount" with differents values, one for each kernel. It's is possible?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello.

I’d like to know if it’s possible to limit the number of registers per kernel using a different number for each different kernel.
Currently I’m “making a mess” using launch_bounds, but it doesn’t solve the generic situation.
Any idea?

Thank you.

2

What’s wrong with launch_bounds?

3

:"> Nothing! It was my fault!

Sorry! External Media

4

Would using launch_bounds yield similar performance compared to compiling it with maxrregcount directives?