fortran: Correct named constants and a datatype name #1927
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@kawashima-fj Down in the datatype engine I prefer to keep the full name of the type, ompi_mpi_c_float_complex instead of ompi_mpi_c_complex (to keep the consistency with the other similar types such as ompi_mpi_c_double_complex). |
|
@bosilca Do you mean |
|
@kawashima-fj this is indeed what I meant. thanks. |
This commit add the following Fortran named constants which are defined in the MPI standard but are missing in Open MPI. - `MPI_LONG_LONG` (defined as a synonym of `MPI_LONG_LONG_INT`) - `MPI_CXX_FLOAT_COMPLEX` - `MPI_C_BOOL` And this commit also changes the value of the following Fortran named constant for consistency. - `MPI_C_COMPLEX` `(MPI_C_FLOAT_COMPLEX` is defined as a synonym of this) Each needs a different solution described below. For `MPI_LONG_LONG`: The value of `MPI_LONG_LONG` is defined to have a same value as `MPI_LONG_LONG_INT` because of the following reasons. 1. It is defined as a synonym of `MPI_LONG_LONG_INT` in the MPI standard. 2. `MPI_LONG_LONG_INT` and `MPI_LONG_LONG` has a same value for C in `mpi.h`. 3. `ompi_mpi_long_long` is not defined in `ompi/datatype/ompi_datatype_module.c`. For `MPI_CXX_FLOAT_COMPLEX`: Existing `MPI_CXX_COMPLEX` is replaced with `MPI_CXX_FLOAT_COMPLEX` bacause `MPI_CXX_FLOAT_COMPLEX` is the right name defined in MPI-3.1 and `MPI_CXX_COMPLEX` is not defined in MPI-3.1 (nor older). But for compatibility, `MPI_CXX_COMPLEX` is treated as a synonym of `MPI_CXX_FLOAT_COMPLEX` on Open MPI. For `MPI_C_BOOL`: `MPI_C_BOOL` is newly added. The value which `MPI_C_COMPLEX` had used (68) is assinged for it because the value becomes no longer in use (described later) and it is a suited position as a datatype added on MPI-2.2. For `MPI_C_COMPLEX`: Existing `MPI_C_FLOAT_COMPLEX` is replaced with `MPI_C_COMPLEX` and `MPI_C_FLOAT_COMPLEX` is changed to have the same value. In other words, make `MPI_C_COMPLEX` the canonical name and make `MPI_C_FLOAT_COMPLEX` an alias of it. This is bacause the relation of these datatypes is same as the relation of `MPI_LONG_LONG_INT` and `MPI_LONG_LONG`, and `MPI_LONG_LONG_INT` and `MPI_LONG_LONG` are implemented like that. But in the datatype engine, we use `ompi_mpi_c_float_complex` instead of `ompi_mpi_c_complex` as a variable name to keep the consistency with the other similar types such as `ompi_mpi_c_double_complex` (see George's comment in open-mpi#1927). We don't delete `ompi_mpi_c_complex` now because it is used in some other places in Open MPI code. It may be cleand up in the future. In addition, `MPI_CXX_COMPLEX`, which was defined only in the Open MPI Fortran binding, is added to `mpi.h` for the C binding. This commit breaks binary compatibility of Fortran `MPI_C_COMPLEX`. When this commit is merged into v2.x branch, the change of `MPI_C_COMPLEX` should be excluded.
The name of `MPI_INTEGER16` obtained using `MPI_TYPE_GET_NAME` from Fortran program was incorrect (`MPI_INTEGER8` was obtained) when `INTEGER*16` is not supported by a compiler. This bug affects only the Fortran binding because `MPI_INTEGER16` is not defined in `mpi.h` if a compiler does not support it.
kawashima-fj
force-pushed
the
pr/fortran-named-constants
branch
from
1128b39 to
722e898
Compare
|
@bosilca I updated the PR. I didn't delete |
|
👍 |
bosilca
pushed a commit
to bosilca/ompi
that referenced
this pull request
Oct 3, 2016
This commit add the following Fortran named constants which are defined in the MPI standard but are missing in Open MPI. - `MPI_LONG_LONG` (defined as a synonym of `MPI_LONG_LONG_INT`) - `MPI_CXX_FLOAT_COMPLEX` - `MPI_C_BOOL` And this commit also changes the value of the following Fortran named constant for consistency. - `MPI_C_COMPLEX` `(MPI_C_FLOAT_COMPLEX` is defined as a synonym of this) Each needs a different solution described below. For `MPI_LONG_LONG`: The value of `MPI_LONG_LONG` is defined to have a same value as `MPI_LONG_LONG_INT` because of the following reasons. 1. It is defined as a synonym of `MPI_LONG_LONG_INT` in the MPI standard. 2. `MPI_LONG_LONG_INT` and `MPI_LONG_LONG` has a same value for C in `mpi.h`. 3. `ompi_mpi_long_long` is not defined in `ompi/datatype/ompi_datatype_module.c`. For `MPI_CXX_FLOAT_COMPLEX`: Existing `MPI_CXX_COMPLEX` is replaced with `MPI_CXX_FLOAT_COMPLEX` bacause `MPI_CXX_FLOAT_COMPLEX` is the right name defined in MPI-3.1 and `MPI_CXX_COMPLEX` is not defined in MPI-3.1 (nor older). But for compatibility, `MPI_CXX_COMPLEX` is treated as a synonym of `MPI_CXX_FLOAT_COMPLEX` on Open MPI. For `MPI_C_BOOL`: `MPI_C_BOOL` is newly added. The value which `MPI_C_COMPLEX` had used (68) is assinged for it because the value becomes no longer in use (described later) and it is a suited position as a datatype added on MPI-2.2. For `MPI_C_COMPLEX`: Existing `MPI_C_FLOAT_COMPLEX` is replaced with `MPI_C_COMPLEX` and `MPI_C_FLOAT_COMPLEX` is changed to have the same value. In other words, make `MPI_C_COMPLEX` the canonical name and make `MPI_C_FLOAT_COMPLEX` an alias of it. This is bacause the relation of these datatypes is same as the relation of `MPI_LONG_LONG_INT` and `MPI_LONG_LONG`, and `MPI_LONG_LONG_INT` and `MPI_LONG_LONG` are implemented like that. But in the datatype engine, we use `ompi_mpi_c_float_complex` instead of `ompi_mpi_c_complex` as a variable name to keep the consistency with the other similar types such as `ompi_mpi_c_double_complex` (see George's comment in open-mpi#1927). We don't delete `ompi_mpi_c_complex` now because it is used in some other places in Open MPI code. It may be cleand up in the future. In addition, `MPI_CXX_COMPLEX`, which was defined only in the Open MPI Fortran binding, is added to `mpi.h` for the C binding. This commit breaks binary compatibility of Fortran `MPI_C_COMPLEX`. When this commit is merged into v2.x branch, the change of `MPI_C_COMPLEX` should be excluded.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR consists of four commits and three of them are trivial. The third commit (b5c34c1) may require discussion. Please read my lengthy commit message. This issue was previously discussed in #1753 but I forgot it for a while so the chance to keep the compatibility between master and v2.x was lost. Sorry.
If this PR is pulled to master, I'll make a PR for v2.x but modify the commit to keep the binary compatibility of
MPI_C_COMPLEXbetween v2.0.0 and v2.0.x.@bosilca please review
Closes #1753