Launch_bounds support for GPU code

Mojo
1

Hello, I am porting some CUDA code to Mojo.
Does Mojo support a mechanism like launch_bounds to provide the compiler with more info on how to manage SM resources?
Are there plans to support it in the future?
Thank you

2

Hi, you can check out this file for reference.

1 Like
3

Hi @massimim, thank you for sharing this question! We do support this, via the @__llvm_metadata( MAX_THREADS_PER_BLOCK_METADATA =StaticTuple[Int32, 1](256)) decorator on the kernel function - here is an example: https://github.com/modularml/modular/blob/main/max/examples/custom_ops/kernels/histogram.mojo#L46-L50

1 Like
4

Thank you Emil

2 Likes
5

Thank you Caroline.

1 Like
6

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.