Can TMA Expect_tx Sync Multiple Transfers to Different SMEM Addresses in One Block?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

I’m trying to use expect_tx with TMA to transfer two separate memory regions:

  • Transfer 1: from global to smem_ptr0.
  • Transfer 2: from global to smem_ptr1.

I want to avoid calling sync after each transfer and instead sync both together. I’ve seen examples for transfers to different blocks’ SMEM addresses, but I’m not sure if it’s possible for different addresses within the same block.

Does anyone know if this is supported?

Thanks in advance!