Add missing OMP5.0 features.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
clang/docs/OpenMPSupport.rst | ||
---|---|---|
194 | Let's keep the explicit (unified shared memory) -> done line and add one for the others as not done. |
clang/docs/OpenMPSupport.rst | ||
---|---|---|
242 | This is a clarification. The spec add restrictions to declare new type on iterators, declare reduction and declare mapper [49:11; 308:17; 327:26] |
clang/docs/OpenMPSupport.rst | ||
---|---|---|
242 | Would be good to put these links to the doc to make it clear |
clang/docs/OpenMPSupport.rst | ||
---|---|---|
194 | @ABataev It makes sense to make it partial. @jdoerfert Keeping that line can be confusing. Line 196 is clear to indicate that the unified_address and unified_shared_memory parts of the requires directive is done. |
clang/docs/OpenMPSupport.rst | ||
---|---|---|
244 | We add five _memory-order-clause_s in the atomic directive - seq_cst, acq_rel, release, acquire and relaxed to support the memory model. |
clang/docs/OpenMPSupport.rst | ||
---|---|---|
244 | I would add as much as possible info about expected features to the doc, if possible |
@kkwli0 I propose you can merge the parts where discussion has reached a consensus while the other parts are resolved. I'm generally fine with this, we can always improve on it further.
clang/docs/OpenMPSupport.rst | ||
---|---|---|
194 | Fair point. thx. | |
218 | No also C/C++. | |
242 |
Agreed. We have the HTML version of the standard online so we can do this "easily" but it will cost someone time and require to change the table layout. Let's postpone it for now until someone find some spare minutes. | |
244 | I guess if they are unclaimed or done we can just list them in the description: `memory model update (seq_cst, acq_rel, ...)` and if we ever support a subset only we can make new rows for them. |
clang/docs/OpenMPSupport.rst | ||
---|---|---|
212 | I think the description here should be something like "support close modifier on map clause". There is no "local" option for the map clause, and also the use of "close" is not dependent on the specification of the unified_shared_memory requirement. |
clang/docs/OpenMPSupport.rst | ||
---|---|---|
218 | Yep, it is not Fortran only. We clarify some pointer attachment behavior in 5.0. | |
242 | Yes, it involves a significant change in the table if we include the corresponding text change in the table. In some cases, it is not clear from the original tickets. I think it is better to leave it as-is. If change the description can help, I welcome any suggestions. |
clang/docs/OpenMPSupport.rst | ||
---|---|---|
218 | I am not sure if this and another feature should be treated as special cases. If it is not, we will need to add references for all the features. |
LG, but still would be good to add some references to the standard. Some of the features use special wording, which very hard to find in the standard.
We have support only for unified memory, so must be partial