In tir level, we can use tir.decl_buffer to control stride of buffer to generate kernel code with special stride. But it seems we can’t specifiy stride in Relay IR level.
Do you think it is a feasible solution to add stride to TensorType?
In tir level, we can use tir.decl_buffer to control stride of buffer to generate kernel code with special stride. But it seems we can’t specifiy stride in Relay IR level.
Do you think it is a feasible solution to add stride to TensorType?