Re: [petsc-dev] Strange "Warnings" When Compiling With -flto
On Wed, Jun 9, 2021 at 5:07 PM Jacob Faibussowitsch <[email protected]> wrote:
Hello All,
I’m getting these strange “warnings” (if you can even call them that), about struct vector alignments all over the shop when compiling with link time optimization enabled on summit. For reference this is with spectrum-mpi wrapping gcc v10.2.0. Here are a few examples:
/autofs/nccs-svm1_home1/jfaibussow/petsc/src/mat/impls/baij/seq/baij.c: In function 'MatDuplicateNoCreate_SeqBAIJ': /autofs/nccs-svm1_home1/jfaibussow/petsc/src/mat/impls/baij/seq/baij.c:3250:21: note: the layout of aggregates containing vectors with 8-byte alignment has changed in GCC 5 3250 | c->bs2 = a->bs2; | ^
autofs/nccs-svm1_home1/jfaibussow/petsc/src/dm/dt/dualspace/impls/lagrange/dspacelagrange.c: In function 'PetscDualSpaceSetUp_Lagrange': /autofs/nccs-svm1_home1/jfaibussow/petsc/src/dm/dt/dualspace/impls/lagrange/dspacelagrange.c:2291:15: note: the layout of aggregates containing vectors with 8-byte alignment has changed in GCC 5 2291 | sp->spdim = sp->spintdim = spdim; | ^
/autofs/nccs-svm1_home1/jfaibussow/petsc/src/mat/impls/sell/mpi/mpisell.c: In function 'MatDuplicate_MPISELL': /autofs/nccs-svm1_home1/jfaibussow/petsc/src/mat/impls/sell/mpi/mpisell.c:1232:27: note: the layout of aggregates containing vectors with 8-byte alignment has changed in GCC 5 1232 | a->size = oldmat->size; | ^
There doesn’t seem to be anything the compiler wants you to do, which is strange. It just tells you about it. Googling it has not yielded useful info either. Is this our own buggy code at fault or just an internal compiler logging so rare that they forgot to turn it off?
There seems to be a discussion of the option to turn these off here: https://stackoverflow.com/questions/52020305/what-exactly-does-gccs-wpsabi-o... Thanks, Matt
Best regards,
Jacob Faibussowitsch (Jacob Fai - booss - oh - vitch)
-- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
participants (1)
-
Matthew Knepley