Data type naming rule

[ahori]

Problem

Adding new basic data types in the standard is a very time consuming process and may result in losing the chance for MPI users to use new data type, such as FP16.

In another words;
"MPI standard only defines these data types. If you want to add new one, then you can do it by yourself by simply following this rule. Don't bother MPI forum for having new data types any more!"

Proposal

I propose to have 2 kinds of basic data types; one is now named fundamental data types which are defined in the current MPI standard and additional data types which implementors can decided to have in their implementation. By having the fixed fundamental type set, MPI forum will not be bothered for discussing on new data types any more.

In addition to this, new data type naming rule for the additional data types will be defined in the standard so that users can easily identify the data types used in the associated compiler from the MPI data type names. This may result in having new data type names, if the current data type names does not conform the naming rule. In this case, the current data type names will be synonyms of the fundamental data type names.

Further, adding MPI_Type_equivalent() function is also proposed so that users can see if two data types having different names are the same, in terms of binary representation and size, or not.

For more details, see attached PDF.

Changes to the Text

See attached PDF (draft).
FP16-20180131-AH.pdf

Impact on Implementations

New data type names should be added but these are essentially the synonyms of the current ones. This effort will be not that much.

Impact on Users

No impact at all (I hope)

References

#65 MPI standard will not define FP16 types but implementor can decide to implement or not.
#66 Implementors can decide to have them.
#69 External32 must be left unchanged because of compatibility

[ahori]

Updated version is here to be discussed on 31st of Jan.
FP16-20180131-AH-rev1.pdf

[ahori]

Adding slides which will be presented in MPI Forum in Portland, 2018.
FP16-2018-March-MPI-Forum.pdf

[ahori]

Here is the PDF file (including some updates from the plenary discussion)
mpi32-report.pdf