Marek Kaluba

I can't use the workaround. This is the akonadi log: ``` [2022-11-03T17:17:19.678Z] akonadi_ews_resource default Please register the custom scheme 'urn' via QWebEngineUrlScheme::registerScheme() before installing the custom scheme handler. [2022-11-03T17:17:19.678Z] akonadi_ews_resource...

ok, I thought that AA was supposed to be (guaranteed) faster than linear, but I've just fact-checked myself ;)

I think you should just follow the advice given by the error message ;)

I'm on `master` as of 5be0e1684d12c4cfd4d22c5fba236a84a092ab5b I have `CUDA_PATH=/opt/cuda` in my env pointing to `cuda-11.4.2`. I compiled scs with ```bash make purge make test_gpu ``` as advised and then test...

I disabled AA but it changed just the numerical values in the log, not the behaviour; here's valgrind log: https://gist.github.com/kalmarek/adb225c93de2bb8d9a7032caec42eea9 I think the problem is somewhere in problem generation (before...

> That's just the output of the first test which is testing data validation and is working correctly. You will see the same if you run the non gpu tests...

@bodono: so I set `VERBOSITY=2` and it seems that cg is never run succesfully. those cuda errors ``` linsys/gpu/indirect/private.c:506:scs_solve_lin_sys ERROR_CUDA (#): invalid argument ``` seem to go away if i...

Even with `VERBOSITY=4` I don't see other output, since `cg_gpu_norm(cublas_handle, r, n) < tol` is satisfied in `https://github.com/cvxgrp/scs/blob/77c86c89bc8d75dce0e8475c364f805fdb62cef0/linsys/gpu/indirect/private.c#L399` If I put the `printf` statement above I get the old ```...

I set it to 1 but I get a similar behavior (though no errors). I also checked that `nrm` is always `0` in `cg_gpu_norm`, though `&r[1]` prints as `1.000000`...

Thanks! I asked for the access to a nvidia gpu at my institution; If I can reproduce it there I'll get back to you!