Anderson2021 autoscheduler should ignore multitarget

[steven-johnson]

If you are building with multitarget and specify the Anderson2021 autoscheduler, the schedule file produced will contain multiple identical copies of the same schedule; this is a waste of time (since we compute the same schedule multiple times) and confusion (since the schedule is the same). We should improve this, but it requires some additional plumbing to do it cleanly.

[antonysigma]

Hi @steven-johnson , the ASAN reports a segfault when I use Anderson2021 autoscheduler with multi-output generators. Is multi-output equivalent to multi-target in Halide?

Is the segfault (link below) a feature or a bug? The stack trace contains a function compile_multitarget(), but I am not sure if it is related.
comp-imaging/ProxImaL#75

[steven-johnson]

This might be a duplicate of #7606 -- can you offer a repro case of what you are seeing?

Is multi-output equivalent to multi-target in Halide?

Yes.

[antonysigma]

This might be a duplicate of #7606 -- can you offer a repro case of what you are seeing?

Sure. I will submit a draft PR featuring the repro case. Moving the discussion to #7606 .