AccelerateHS / AccelerateHS/accelerate-llvm

Concurrent tests of native backend fail with: lost signal due to full pipe: 11

Aberta
#53 0 comentários 0 reações 0 responsáveis Ver no GitHub
bug
Linguagem predominante
Haskell
Estrelas
171
Forks
65
Métricas de merge de PRs
Nenhum PR com merge em 30d

Descrição

I am submitting a...
- [x] bug report
- [ ] feature request
- [ ] support request => you might also like to ask your question on the [mailing list](mailto:accelerate-haskell@googlegroups.com) or [gitter chat](https://gitter.im/AccelerateHS/Lobby).

## Description

The tests of the native backend fail on multicore architectures. The problem appears to occur when multiple test cases are ran in parallel. If you let the test suite execute the tests sequentially, they do pass. Note that the different tests are then sequential, but each test case may be executed on multiple threads.

I also tried to revert the recent changes

## Expected behaviour

Tests pass. When multiple computations are given to accelerate at a time, it will let the set of worker threads work on both computations at the same time. I think that this cooperation has some bug.

## Current behaviour

The threads fail by printing the following the console many times:
```
nofib-llvm-native: lost signal due to full pipe: 11
```
I think some test case fails, which causes that the process becomes in some invalid state, causing that the later tests will also fail with this error.

It may also happen the test suite stops at some test. It looks like the same error, but the test case will then not run the other tests any more.
```
DIM3
sum: nofib-llvm-native: SignalException 11
```

I did verify that the recent change in the scheduler did not influence this. Also, the refactoring of tuples seems unrelated.

## Possible solution (optional)

## Steps to reproduce (for bugs)

1. Run the tests on a machine with enough cores

## Your environment

I ran the tests on our shared machine.

Guia de contribuição

Nenhum guia de contribuição indexado para este repositório

Avaliação

Esta issue ainda não foi avaliada.

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.