In classTritonScatteredStoreOpConversion, funcmatchAndRewritebufferization::ToTensorOp not set restrict and writable, after convert to linalg mlir, this bufferization.to_tensor Op is illegal when using one-shot bufferization.
Wonder why ScatteredLoadOp has the restric and writable attribute but ScatteredStoreOp hasn't ?
In class
TritonScatteredStoreOpConversion
, funcmatchAndRewrite
bufferization::ToTensorOp
not set restrict and writable, after convert to linalg mlir, thisbufferization.to_tensor
Op is illegal when using one-shot bufferization. Wonder why ScatteredLoadOp has the restric and writable attribute but ScatteredStoreOp hasn't ?