search

nod-ai / iree-amd-aie

IREE plugin repository for the AMD AIE accelerator
Apache License 2.0
69 stars 30 forks source link

Fix `AIEX::NpuDmaMemcpyNdOp::getOffsetInBytes` #765

Closed makslevental closed 2 months ago

makslevental commented 2 months ago

Current version doesn't take into account striding (effectively assumes unit stride for all dims). Not sure how this hasn't come up before...

Thanks @erwei-xilinx

yzhang93 commented 2 months ago

Thanks @makslevental and @erwei-xilinx! This should fix the numeric issue we had for batch matmul size 2.

newling commented 2 months ago

This also fixes https://github.com/nod-ai/iree-amd-aie/issues/762