ompi/attributes: revamp attribute handling. #3891
Conversation
|
from #2940
i set the milestone to |
|
@ggouaillardet, your commit message leaves a lot to be desired in the "why?" category. What was broken, what's fixed, what's still broken? What's your opinion on the ABI changes and why? We can't take an ABI change for 3.0.1 and would prefer not take it for the fall release, but it's unclear where we even are on ABI changes. |
we now have 12 cases to deal (4 writers and 3 readers) : 1. C `void*` is written into the attribute value, and the value is read into a C `void*` (unity) 2. C `void*` is written, Fortran `INTEGER` is read 3. C `void*` is written, Fortran `INTEGER(KIND=MPI_ADDRESS_KIND)` is read 4. Fortran `INTEGER` is written, C `void*` is read 5. Fortran `INTEGER` is written, Fortran `INTEGER` is read (unity) 6. Fortran `INTEGER` is written, Fortran `INTEGER(KIND=MPI_ADDRESS_KIND)` is read 7. Fortran `INTEGER(KIND=MPI_ADDRESS_KIND)` is written, C `void*` is read 8. Fortran `INTEGER(KIND=MPI_ADDRESS_KIND)` is written, Fortran `INTEGER` is read 9. Fortran `INTEGER(KIND=MPI_ADDRESS_KIND)` is written, Fortran `INTEGER(KIND=MPI_ADDRESS_KIND)` is read (unity) 10. Intrinsic is written, C `void*` is read 11. Intrinsic is written, Fortran `INTEGER` is read 12. Intrinsic is written, Fortran `INTEGER(KIND=MPI_ADDRESS_KIND)` is read We previously handled 9 cases ( 3 x 3 types) : a) C `void *` b) Fortran MPI-1 c) Fortran MPI-2 though a) is unchanged and c) is equivalent to Fortran `INTEGER(KIND=MPI_ADDRESS_KIND)`, b) was not correctly handled since a Fortran `INTEGER` is not always equivalent to a C `int`. though it might have happened to work on little endian systems, the change is critical for big endian systems. MPI-2 Fortran "integer representation" has type `INTEGER(KIND=MPI_ADDRESS_KIND)` as clarified at mpiwg-rma/rma-issues#1 Signed-off-by: Gilles Gouaillardet <[email protected]> (cherry picked from commit open-mpi/ompi@72cfbb6)
ggouaillardet
force-pushed
the
topic/v3.0.x/attributes
branch
from
197a3f6 to
3b2df6d
Compare
|
@bwbarrett i rewrote the commit message to explain what this commit is fixing @bosilca i checked again about the ABI changes, and unless i got it wrong (again), there are no ABI changes :-) the type of the callbacks for |
|
@bwbarrett good to go |
we now have 12 cases to deal (4 writers and 3 readers) :
void*is written into the attribute value, and the value is read into a Cvoid*(unity)void*is written, FortranINTEGERis readvoid*is written, FortranINTEGER(KIND=MPI_ADDRESS_KIND)is readINTEGERis written, Cvoid*is readINTEGERis written, FortranINTEGERis read (unity)INTEGERis written, FortranINTEGER(KIND=MPI_ADDRESS_KIND)is readINTEGER(KIND=MPI_ADDRESS_KIND)is written, Cvoid*is readINTEGER(KIND=MPI_ADDRESS_KIND)is written, FortranINTEGERis readINTEGER(KIND=MPI_ADDRESS_KIND)is written, FortranINTEGER(KIND=MPI_ADDRESS_KIND)is read (unity)void*is readINTEGERis readINTEGER(KIND=MPI_ADDRESS_KIND)is readMPI-2 Fortran "integer representation" has type
INTEGER(KIND=MPI_ADDRESS_KIND)as clarifiedat mpiwg-rma/rma-issues#1
Signed-off-by: Gilles Gouaillardet [email protected]
(cherry picked from commit 72cfbb6)