-
|
Hello, we're currently trying to update the PETSc build instructions. They currently work for 5.13, but it seems that many things have been improved/simplified since then. |
Beta Was this translation helpful? Give feedback.
Replies: 8 comments 21 replies
-
|
I don't claim to understand what is done in that Python script. But it looks like you are building all of SuiteSparse with the root Makefile and then picking the headers and libraries for AMD, COLAMD, BTF, CAMD, CCOLAMD, CHOLMOD, UMFPACK, KLU, SPQR, and SuiteSparse_config. That might become easier once #415 is ready and merged.
PS: The latest release is SuiteSparse 7.3.0 currently. |
Beta Was this translation helpful? Give feedback.
-
|
PS: SuiteSparse 7.3.0 is unusable because of #478. |
Beta Was this translation helpful? Give feedback.
-
|
I've posted a SuiteSparse 7.3.1.beta1 that fixes issue #478. |
Beta Was this translation helpful? Give feedback.
-
|
Sadly, we still can't use that in PETSc because we don't know how to turn-off building of shared libraries (probably related to #424?). |
Beta Was this translation helpful? Give feedback.
-
|
Yeah. Currently, it's only possible to disable building of static libraries. Shared libraries are always built by the current build system. |
Beta Was this translation helpful? Give feedback.
-
|
I would then have a look at the SuiteSparse build system once the related pull requests have been merged and try to propose something that follows common CMake conventions. |
Beta Was this translation helpful? Give feedback.
-
|
@MarDiehl: I opened #522 which should allow building SuiteSparse without shared libraries. |
Beta Was this translation helpful? Give feedback.
-
|
The new build system is now very stable, and appears in a prerelease: SuiteSparse 7.4.0.beta1. |
Beta Was this translation helpful? Give feedback.
-
|
I currently try with beta2. Currently, I'm struggling on forcing SuiteSparse to use exactly the LAPACK/BLAS that we want to use: https://gitlab.com/petsc/petsc/-/merge_requests/710 |
Beta Was this translation helpful? Give feedback.
-
|
Correct MR is https://gitlab.com/petsc/petsc/-/merge_requests/7104 (missing |
Beta Was this translation helpful? Give feedback.
-
|
Also, how do we pass our own METIS? |
Beta Was this translation helpful? Give feedback.
-
|
I don't have a gitlab login so I'll need to reply here. To set a specific BLAS use something like
or
which says "use the libblas.so". It reports: that's just one example. You should be able to use any If you want to skip the cmake FindBLAS.cmake entirely, you'd need to modify the SuiteSparse_config/cmake_modules/SuiteSparseBLAS.cmake to instead
etc, then if I see Regarding METIS: you can't. I can't use the default METIS library because it's not configured to work with SuiteSparse. I need to patch it in various (minor) ways, like to add my own memory manager, and to #define the integer sizes to use. METIS allows for that but it requires that you edit METIS, and the new functions have the same name as the editted METIS. That means they could collide with some -libmetis in the system that assumes a different integer size for its indices inside its graphs / sparse matrices. To avoid that, I rename all METIS functions, in my copy, using #define's. See CHOLMOD/SuiteSparse_metis, so I do not collide with another copy of METIS (like the one you use). So unless you have a very special copy of METIS, my copy should be fine; it doesn't conflict with another copy of METIS linked in alongside SuiteSparse. It's all bundled inside CHOLMOD: |
Beta Was this translation helpful? Give feedback.
-
|
For METIS: see https://github.com/DrTimothyAldenDavis/SuiteSparse/blob/dev/CHOLMOD/SuiteSparse_metis/README.txt |
Beta Was this translation helpful? Give feedback.
-
|
Let me know if setting BLA_VENDOR is sufficient for you. If it isn't then I could see a simple solution like the following:
That would require me to change SuiteSparse_config/cmake_modules/SuiteSparseBLAS.cmake, so that it would check if BLAS_LIBRARIES is already defined, just at the very top of that file. I don't think it would be hard to add. But the simplest solution would be to see if BLA_VENDOR works for PETsc. |
Beta Was this translation helpful? Give feedback.
-
|
One issue that can occur: sometimes you can get stale headers for suitesparse. They can be there from Linux distros, spack, homebrew, etc. Linking with one version of libumfpack.so but with another version of umfpack.h can cause problems. That's why I added library routines like umfpack_version to all of my libraries. The headers also #define the version numbers. So you can then check. That's what my Example package does. If you build your package using cmake and the latest *Config.cmake files for each of my libraries then this problem is avoided. So if you're not getting umfpack.h but are getting cholmod.h then it might in fact be a stale cholmod.h header. |
Beta Was this translation helpful? Give feedback.
-
|
It looks like PETSc expect headers in plain |
Beta Was this translation helpful? Give feedback.
-
|
Yes, that has changed. It was to help avoid the problem of getting stale headers from older installations of suitesparse. |
Beta Was this translation helpful? Give feedback.
-
|
quick update from GitLab: |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the update. I can definitely add the update to the changelog. Yes I did see that Debian and maybe Ubuntu have used /usr/local/include/suitesparse for a while. Probably predating the existence of any cmake build system of mine. I had plain Makefiles before. The new cmake system should be much easier to use going forward but it is an adjustment in the short term. I'm tied up the next few days (my son is getting married 😀). But I can try to reply here. Let me know if there's anything I can do to help, or any changes you'd like that could be helpful. I can wait to post a stable SuiteSparse 7.4.0 until after we resolve these issues you're seeing in PETsc with the new SuiteSparse build system. |
Beta Was this translation helpful? Give feedback.
-
|
I just posted 7.4.0.beta4 that might address some of these issues. See https://github.com/DrTimothyAldenDavis/SuiteSparse/releases/tag/v7.4.0.beta4 |
Beta Was this translation helpful? Give feedback.
-
|
I'd like to post a stable SuiteSparse 7.5.0 soon, with major additions to the GraphBLAS library. Are these issues now resolved in the current 7.4.0, or perhaps 7.5.0.betaX? |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
It's hard for me to follow the discussion there. Is 7.5.0 ready to go, as far as PETsc is concerned? I'd like to release a stable 7.5.0 in the next day or two, but if there are any pending updates you need in SuiteSparse that could resolve anything for the PETsc integration, I'd like to add those. |
Beta Was this translation helpful? Give feedback.
-
|
I'm pretty sure everything is fine now. All tests pass and the remaining open questions are related to details of the PETSc build system. From my side, green lights for a 7.5.0 release. |
Beta Was this translation helpful? Give feedback.
I'm pretty sure everything is fine now. All tests pass and the remaining open questions are related to details of the PETSc build system. From my side, green lights for a 7.5.0 release.