-
Notifications
You must be signed in to change notification settings - Fork 319
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
CI: Build with OpenBLAS instead of Apple Accelerate on macOS (Homebrew). #515
Conversation
It seems that with macos-13 all issues were resolved and with macos-14 there are two remaining (unrelated?) issues. So I guess this could be merged. I am still trying to locate whether I found where things go bad exactly. |
It looks like that helped indeed. The macOS runner on Intel hardware passes all tests now. And for the runner on Apple Silicon only these two tests are still failing:
|
Unrelated to the changes, there seems to be still (at least) one test that fails sporadically on MinGW:
It only failed for one of the runners and passed when it was re-run... |
Running that test with valgrind shows a couple of warnings:
If that is indeed the case, the behavior might depend on "random" values at non-initialized memory addresses. Edit: I still had local changes in |
The problematic BLAS routine is probably "zdotc" which takes care of the complex dot product by setting pointer of dotprodfun to this. It is called in the Krylov methods like "dotprodfun(n, work(1:n,rr), 1, work(1:n,r+k-1), 1). Maybe the Mac native routine assumes different ordering in complex that what we provide. Difficult to get any closer than this call so I will accept this PR and we may study this later. This seems to have occured before too: The mac routines are probably highly optimized so this is a pity. |
Fix now in "devel".
From: "Markus Mützel" ***@***.***>
To: "ElmerCSC" ***@***.***>
Cc: "Subscribed" ***@***.***>
Sent: Monday, 5 August, 2024 18:46:38
Subject: Re: [ElmerCSC/elmerfem] CI: Build with OpenBLAS instead of Apple Accelerate on macOS (Homebrew). (PR #515)
Unrelated to the changes, there seems to be still (at least) one test that fails sporadically on MinGW:
The following tests FAILED:
324 - MonolithicSlave2 (Failed)
It only failed for one of the runners and passed when it was re-run...
—
Reply to this email directly, [ #515 (comment) | view it on GitHub ] , or [ https://github.com/notifications/unsubscribe-auth/ACTOMSV3WS5LIPFAVEZOWXDZP6M55AVCNFSM6AAAAABMAPHPBGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENRZGM4DIMRZGI | unsubscribe ] .
You are receiving this because you are subscribed to this thread. Message ID: ***@***.***>
|
Could you maybe share the output from these ? Does valgrind report anything ?
From: "Markus Mützel" ***@***.***>
To: "ElmerCSC" ***@***.***>
Cc: "Subscribed" ***@***.***>
Sent: Monday, 5 August, 2024 18:43:25
Subject: Re: [ElmerCSC/elmerfem] CI: Build with OpenBLAS instead of Apple Accelerate on macOS (Homebrew). (PR #515)
It looks like that helped indeed. The macOS runner on Intel hardware passes all tests now. And for the runner on Apple Silicon only this two tests are still failing:
Errors while running CTest
150 - EMWaveBoxHexasEigen (Failed)
214 - FixTangentVelo (Failed)
—
Reply to this email directly, [ #515 (comment) | view it on GitHub ] , or [ https://github.com/notifications/unsubscribe-auth/ACTOMSURP3ZDFLRWC4TG4MTZP6MR3AVCNFSM6AAAAABMAPHPBGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENRZGM3TOOBTHE | unsubscribe ] .
You are receiving this because you are subscribed to this thread. Message ID: ***@***.***>
|
No sweat, I found the logs from github. These give wrong answers (or at least different from other
platforms) and need to be investigated, I guess. On my redhat, valgrind is clean...
From: "Juha Ruokolainen" ***@***.***>
To: "ElmerCSC" ***@***.***>
Cc: "ElmerCSC" ***@***.***>, "Subscribed" ***@***.***>
Sent: Tuesday, 6 August, 2024 10:11:05
Subject: Re: [ElmerCSC/elmerfem] CI: Build with OpenBLAS instead of Apple Accelerate on macOS (Homebrew). (PR #515)
Could you maybe share the output from these ? Does valgrind report anything ?
From: "Markus Mützel" ***@***.***>
To: "ElmerCSC" ***@***.***>
Cc: "Subscribed" ***@***.***>
Sent: Monday, 5 August, 2024 18:43:25
Subject: Re: [ElmerCSC/elmerfem] CI: Build with OpenBLAS instead of Apple Accelerate on macOS (Homebrew). (PR #515)
It looks like that helped indeed. The macOS runner on Intel hardware passes all tests now. And for the runner on Apple Silicon only this two tests are still failing:
Errors while running CTest
150 - EMWaveBoxHexasEigen (Failed)
214 - FixTangentVelo (Failed)
—
Reply to this email directly, [ #515 (comment) | view it on GitHub ] , or [ https://github.com/notifications/unsubscribe-auth/ACTOMSURP3ZDFLRWC4TG4MTZP6MR3AVCNFSM6AAAAABMAPHPBGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENRZGM3TOOBTHE | unsubscribe ] .
You are receiving this because you are subscribed to this thread. Message ID: ***@***.***>
|
Check if using OpenBLAS makes a difference compared to Apple Accelerate when it comes to the failing tests involving complex numbers.
See: #512 (comment)