[Mlir-commits] [mlir] [mlir][nvgpu] Mark TMA descriptor as MemWriteAt in `tma.async.store` (PR #79427)

[Mehdi Amini]

joker-eph wrote:

Seems to me that this document is misleading: https://github.com/llvm/llvm-project/blob/b40d5b1b08564d23d5e0769892ebbc32447b2987/mlir/include/mlir/Dialect/NVGPU/IR/NVGPU.td#L689C1-L691C1

> The Op creates a tensor map descriptor object representing tiled memory region.  To do that it calls CUDA Driver's `cuTensorMapEncodeTiled`. 

Taken as-is, it would mean this initialize a host-side struct.

Can you follow-up with another PR to document the type as a device pointer, and check all these ops as well, thanks :)


https://github.com/llvm/llvm-project/pull/79427