The compiler couldn't identify the loop listed before about allo

[chhzh123]

It seems this is not an Allo bug, but the misusage of the schedule primitivies. There are two problems here:

1. Since you have already split the j loop, there is no j loop when you call the buffer_at function, but only j.outer and j.inner.
2. There is no need to create an intermediate buffer for force_x as force_x is already outside the j loop.

Please let us know if you need further guidance on optimizing this kernel.

[zhangzhiru]

We should print out proper error messages instead of letting the compiler crash.

[chhzh123]

We have already printed out the error message in this case:

  File "/Users/rhodama/CORNELL/Design_project/bin_sp24/allo/allo/ir/transform.py", line 104, in find_loop_in_bands
    raise RuntimeError(f"Cannot find the band of loop {axis_name}")
RuntimeError: Cannot find the band of loop j in "kernel_md"

We can further provide a list of available loop axes for users to choose from, but we cannot instruct users where to insert the buffers as it may need additional analysis (which is costly).

[zhangzhiru]

Let's keep this issue open for now. I have some suggestions on the error messaging mechanism.