Philip Müller

Results 21 comments of Philip Müller

As far as I can tell there is still a bug in the array branch.

@edopao I have now started to look into this issue a little bit. However, for me this issue only appears on the GPU, if I translate the same code to...

I have now generated a reproducer: ```python import dace dtype = dace.float64 shape = (10, 10) def make_nsdfg(): """ Generates a nested SDFG with that runs the following calculation: ```python...

It is most likely the Scalar to Symbol Transformation.

I think we agree about that. Here is also a short test example, that can be used ad basis of a test: ```python import numpy as np import dace #=================...

A deterministic test can be found [here](https://github.com/philip-paul-mueller/dace/commit/8e4ca8573d34547639f9d7e3e483d33169715390), but it should be fixed in [PR1603](https://github.com/spcl/dace/pull/1603)

This issue seems to be solved by [PR1603](https://github.com/spcl/dace/pull/1603), but be aware of [issue 1644](https://github.com/spcl/dace/issues/1644) which is similar but happens in a different context.

Yes. This problem should be solved by the new map fusion.

The error you have looks familiar, I saw it when I rewrote MapFusion. Back then I had the feeling that subgraph fusion, the component that fails, highly depends on how...

Here is a second reproducer, that allocates the array outside. - If the memory order of `A` is switched from `F` to `C` then the code works as expected (did...