Error: All viewed matrices must belong to the same allocation extent

Errors
1

Hi Everyone!

As the title suggests, I encounter the error “TypeError: All viewed matrices must belong to the same allocation extent”, raised in pyfr/backends/base/types.py, when starting a case.

I have set up a very simple test case in 2D and 3D, consisting of quads and tri elements or quads, tri, hex, pyr and tet elements, respectively. You can find the meshes / .ini files below. I noticed that the error is only raised for certain combinations of polynomial order & solution / flux point sets:

2D, working:

  • Quad elements: GLL
  • Tri elements: Alpha-Opt
  • Line interfaces: GLL

2D, not working:

  • Quad elements: GLL
  • Tri elements: Williams-Shunn
  • Line interfaces: GLL

For 2D, changing polynomial order did not change anything. However, in 3D it is a bit different:

3D, p = 1,2, working:

  • Quad, Pyr & Hex elements: GLL
  • Tri, Tet elements: Alpha-Opt
  • Line & Quad interfaces: GLL
  • Tri interfaces: Alpha-Opt

For p >= 3 the configuration above raised the error.

3D, all p, not working:

  • Quad, Pyr & Hex elements: GLL
  • Tri elements: Williams-Shun
  • Tet elements: Shunn-Ham
  • Line & Quad interfaces: GLL
  • Tri interfaces: Williams-Shun

I mainly used PyFR version 2.1 (in 3.1 it raised the same error, although technically my rocm version is not new enough), with hip backend and rocm/6.4.0. The BCs & ICs are relatively arbitrary, since the issue arises before anything starts anyways.

Here are the meshes and .ini files:
Test_folder

Does anyone know how to solve this issue?

2

This is an edge case which we currently don’t handle, it shouldn’t be to difficult to fix. I’ll put together a patch tomorrow.

Regards, Freddie.

3

A fix should now be in the develop branch.

Regards, Freddie.

4

It worked, great, thanks a lot!