Mikael Simberg

Results 232 issues of Mikael Simberg

Currently the address sanitizer CI job added in #972 explicitly disables `detect_stack_use_after_return` (default is also off, but this is to protect from a change to the default). It seems like...

effort: 3
type: bug
category: CI

Currently our custom signal handlers may allocate, which thread sanitizer reports as unsafe (since it may cause further signals). Check if we can somehow clean this up. However, this is...

effort: 3
effort: 4
priority: low
type: bug

It's unclear if these are false positives, or real. E.g. (in nearly any test using the pika schedulers, both `deque` and `MoodyCamel` queue): ``` WARNING: ThreadSanitizer: data race (pid=3631945) Atomic...

effort: 2
effort: 3
effort: 4
effort: 5
priority: medium
type: bug

As reported by thread sanitizer (e.g. in `channel_mpsc_shift` test): ``` ================== WARNING: ThreadSanitizer: data race (pid=3631340) Read of size 4 at 0x7b04000028c0 by thread T1: #0 pika::experimental::base_channel_mpsc::get(int*) const /home/mjs/src/pika/libs/pika/synchronization/include/pika/synchronization/channel_mpsc.hpp:112 (channel_mpsc_shift_test+0x10fbb9)...

effort: 2
effort: 3
effort: 4
effort: 5
priority: medium
type: bug

In addition to the current performance benchmarks, it would make sense to monitor the number of instructions using valgrind's cachegrind tool. This should have significantly less variation than timing benchmarks...

effort: 3
priority: low
type: feature
category: CI

#891 adds a special pipeline configuration for pushing performance data to elastic. Once we've reduced duplication in the pipelines and added release configurations to CSCS CI configurations (#862), we should...

effort: 2
effort: 3
priority: medium
type: feature
category: CI
type: cleanup

The initial number 128 is currently hardcoded. This should be a configurable number.

effort: 2
priority: low
type: feature
category: CUDA

We currently disable the `schedule_from` customization for `thread_pool_scheduler` with stdexec: https://github.com/pika-org/pika/blob/cb8ebe0667c903cd3155201b2a356f0e15b19855/libs/pika/executors/include/pika/executors/thread_pool_scheduler.hpp#L269-L287. We should make sure the customization is used even with stdexec enabled. In the case of this example, `transfer`...

effort: 3
priority: medium
type: refactoring
category: senders/receivers
M1: P2300 reference implementation

`any_receiver` should be possible to convert into more or less a reference. The non-type-erased receiver should be possible to store in the operation state, with `any_receiver` simply pointing to the...

effort: 3
priority: low
type: feature
type: refactoring
category: senders/receivers