PaStiX issueshttps://gitlab.inria.fr/solverstack/pastix/-/issues2022-04-12T11:38:41Zhttps://gitlab.inria.fr/solverstack/pastix/-/issues/57pkg-config --static (mingw/msys2)2022-04-12T11:38:41ZRafal brzegowypkg-config --static (mingw/msys2)If I use the command `pkg-config --libs --static pastix` then I have this output:
`-LC:/msys64/mingw64/lib -lpastix -lpastix_kernels -lpastix_starpu -lC:/msys64/mingw64/lib/libm.a -lC:/msys64/mingw64/lib/libopenblas.dll.a -lspm C:/msys6...If I use the command `pkg-config --libs --static pastix` then I have this output:
`-LC:/msys64/mingw64/lib -lpastix -lpastix_kernels -lpastix_starpu -lC:/msys64/mingw64/lib/libm.a -lC:/msys64/mingw64/lib/libopenblas.dll.a -lspm C:/msys64/mingw64/lib/libopenblas.dll.a\;C:/msys64/mingw64/lib/libm.a\;C:/msys64/mingw64/lib/libopenblas.dll.a -lstarpu-1.3 -lpthread -lssp -g0 -lws2_32 -lpthread -LD:/a/msys64/mingw64/lib -lm -lgdi32 -lltdl -lpthread -lhwloc -lm -lgdi32 -lltdl -lpthread -LC:/msys64/mingw64/lib -lspm C:/msys64/mingw64/lib/libopenblas.dll.a\;C:/msys64/mingw64/lib/libm.a\;C:/msys64/mingw64/lib/libopenblas.dll.a`
Why is the `libopenblas.dll.a` version referenced for static linking rather than the usual `-lopenblas` (i.e. the libopenblas.a full static link)?https://gitlab.inria.fr/solverstack/pastix/-/issues/56Solve with L / Mutiplication with L (Cholesky factorization)2021-05-27T14:00:13ZDURUFLE MarcSolve with L / Mutiplication with L (Cholesky factorization)Hi,
in the previous releases of pastix (5.xx), it was possible to solve independently
```
L x = b
or
L^T x = b
```
for the Cholesky factorisation. It was also possible to perform a multiplication by the factor L (or L^T) with the followi...Hi,
in the previous releases of pastix (5.xx), it was possible to solve independently
```
L x = b
or
L^T x = b
```
for the Cholesky factorisation. It was also possible to perform a multiplication by the factor L (or L^T) with the following parameters :
```c
iparm[IPARM_TRANSPOSE_SOLVE] = API_SOLVE_BACKWARD_ONLY;
iparm[IPARM_TRANSPOSE_SOLVE] = API_SOLVE_FORWARD_ONLY;
iparm[IPARM_TRANSPOSE_SOLVE] = API_SOLVE_UTRMV;
iparm[IPARM_TRANSPOSE_SOLVE] = API_SOLVE_LTRMV;
```
How is it possible to perform these tasks with the new version of pastix (6.xx) ?
Thanks for your help.https://gitlab.inria.fr/solverstack/pastix/-/issues/55Free csc user for the distributed case2021-05-26T09:02:52ZDURUFLE MarcFree csc user for the distributed caseHi,
I would like to free the csc matrix during the factorization, I write :
iparm[IPARM_FREE_CSCUSER] = 1;
It works fine for the sequential version. However, for the distributed version, it does not work
because pastix_task_solve needs...Hi,
I would like to free the csc matrix during the factorization, I write :
iparm[IPARM_FREE_CSCUSER] = 1;
It works fine for the sequential version. However, for the distributed version, it does not work
because pastix_task_solve needs the array loc2glob which has been freed by the previous tasks. Is there a way
to avoid freeing this array such that the solve can be completed correctly ?
In terms of memory storage, when I have a moderate fill-in (for instance around 2), this option is interesting ? I was wondering if the arrays are freed before allocating more space for L factor because of fill-in.https://gitlab.inria.fr/solverstack/pastix/-/issues/46Performance model and memory problems on GPU2019-11-25T13:11:25ZGhost UserPerformance model and memory problems on GPUDear developers,
first of all, thank you very much developing such a great solver.
In my current project I am using PastiX' hybrid GPU mode through PaRSEC. It works great and the speedup is really showing. However, there are currently ...Dear developers,
first of all, thank you very much developing such a great solver.
In my current project I am using PastiX' hybrid GPU mode through PaRSEC. It works great and the speedup is really showing. However, there are currently two problems I am facing:
1. A default performance model and cost function is provided for CPU and GPU (common/model.c) but I wonder how to reasonably determine coefficients for my target hardware? I found the CMake-option "PASTIX_GENERATE_MODEL", but can't figure out whether i can turn that data into coefficients.
2. When factorizing some of my larger matrices and choosing a larger block size at the same time, PaRSEC returns illegal memory access errors during factorization. I assume this is happens because more memory than available is requested from the GPU. Do you possibly know a solution for this problem or are you aware of this problem?
Best regards,
Peter