[RFC] Move linalg::PadTensorOp to tensor::PadOp
linalg::PadTensorOp
operation allows to pad the source
tensor with given low
and high
padding config.
%pad_value = ... : f32
%padded_tensor = linalg.pad_tensor %source low[0, 0] high[%ub0, %ub1] {
^bb0(%arg1: index, %arg2: index):
linalg.yield %pad_value : f32
} : tensor<2x3xf32> to tensor<?x?xf32>
I think this operation is useful not only for Linalg users and should be moved to tensor
dialect and become tensor.pad
.