akaashrp opened a new pull request, #20133:
URL: https://github.com/apache/tvm/pull/20133

   This PR adds a WebGPU correctness fallback for inclusive cumsum along a 
non-innermost axis. For tensors of known rank, the lowering reshapes the input 
into [outer, scan, inner]. Each GPU thread handles one (outer, inner) pair and 
scans the middle dimension serially before the result is reshaped back to its 
original shape. The existing parallel kernel remains in use for innermost-axis 
scans. The new kernel exposes parallelism across outer * inner, but does not 
parallelize work along the scan dimension.
   
   This PR:
   - Support positive and negative non-innermost axis indices
   - Support symbolic dimensions and distinct input and output dtypes
   - Preserve the tensor’s virtual device across intermediate reshapes and 
`call_tir`


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to