Raffaele Solcà

Strange gaps in trace when a thread is scheduling tasks

2

Using the default scheduler blank gaps appear in the trace when one of the threads is scheduling the tasks. The gaps disappear immediately after the scheduling task finished. ![Screenshot 2023-02-21...

Unreproducible test failures

21

@albestro @teonnik @msimberg I created this issue to collect non-reproducible test failures, i.e. failures that occur so rarely that it is almost impossible to debug them. Please add any case...

band to tridiagonal scaling improvement: First attemp

2

Duplicate work on columns to reduce HHR communication complexity. - [ ] Check performance

Currently only `A x = D x` is checked in the miniapp. Note: it needs gemm (NT or TN), so it conflicts with #969 and #1022.

Tune parameters should be improved: 4 types of parameters should be available: a) single parameter for all cases (as it is now) b) parameter that store 2 values (MC /...

Create submatrix view (reference and pipelined)

2

The following is needed: - [x] sub-pipeline of matrix (#898) - [x] arbitrary offset for distribution (#901) - [x] sub-distribution from distribution (#934) - [x] sub-matrix reference of matrix (#934)...

- rename methods - [x] add new `snake_case` methods (#1007) - [ ] deprecate old methods - constructors: - [ ] evaluate which methods are needed - [ ] evaluate...

https://github.com/eth-cscs/DLA-Future/pull/594#issuecomment-1188954697

The following steps are needed: - [ ] permute the eigenvectors to have the optimal shape (#914) - [ ] Implement the gemm with submatrices and make sure that gemm...

Currently we assert band_size >= 2.