Mojo module

dispatch

RDNA dispatch for 2-D convolution.

Provides two paths for 2-D convolution on RDNA 3+:

1. Implicit GEMM (preferred): Fuses im2col into the WMMA matmul kernel's A-tile loader, eliminating the large intermediate buffer. Requires C_in % BLOCK_K == 0 for vectorized im2col loads.

2. Explicit im2col + matmul (fallback): Materializes the im2col buffer then calls the standard RDNA matmul. Used when C_in alignment requirements aren't met.

Functions

• ​dispatch_rdna_conv2d: Try to dispatch Conv2D on RDNA via implicit GEMM (im2col fused into WMMA).

Was this page helpful?

Thank you! We'll create more content like this.

Thank you for helping us improve!

😔 What went wrong?

Some code doesn’t work

It includes inaccurate information

It's missing information I need

It was difficult to understand

Other

Submit