Review Mojo's priorities

• I have read the roadmap and priorities and I believe this request falls within the priorities.

What is your request?

I am working on a pure mojo local inference engine Millrace. Depending on the operation Millrace is about 3x-10x times slower than MLX and Ollama, on my M4 mini. If the compiler had intrinsic support for simdgroup_matrix on M4 Claude could close the gap. There is already a similar intrinsic for M5.

Many thanks for your help,

Marius

What is your motivation for this change?

I would like to run local models on my M4 mac mini, some use cases are possible once the tokens/second reach the hundred of tokens/sec.

Any other details?

No response