Closed Abhishek-Varma closed 17 hours ago
Hi @jtuyls - I've addressed your changes.
Note: For the size/stride check only one lit test would suffice because staticSize/staticStride size is same - so we only need to check on shapeArr.size() being greater than that at the beginning of static metadata formation.
Hi @jtuyls - I've addressed your changes.
Note: For the size/stride check only one lit test would suffice because staticSize/staticStride size is same - so we only need to check on shapeArr.size() being greater than that at the beginning of static metadata formation.
That's not really guaranteed from the function's perspective right? Someone could change the sizes of the size and stride vectors so they're not the same anymore? You could add an assert to ensure the sizes are the same.
Hi @jtuyls - I've addressed your changes. Note: For the size/stride check only one lit test would suffice because staticSize/staticStride size is same - so we only need to check on shapeArr.size() being greater than that at the beginning of static metadata formation.
That's not really guaranteed from the function's perspective right? Someone could change the sizes of the size and stride vectors so they're not the same anymore? You could add an assert to ensure the sizes are the same.
Sure. I've added an assert now within the function.
-- This commit adds a fix for supporting implicit source/target addressing in amdaie.npu.dma_cpy_nd ops as part of
--iree-amdaie-lower-to-aie
pass.Signed-off-by: Abhishek Varma abhvarma@amd.com