Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
openmp/libomptarget/DeviceRTL/src/exports | ||
---|---|---|
5 | I'd prefer our ompx extension to be C interfaces instead of C++. |
openmp/libomptarget/DeviceRTL/src/exports | ||
---|---|---|
5 | Why should they be one or the other? ompx_ can (and will) coexist, no? |
openmp/libomptarget/DeviceRTL/src/exports | ||
---|---|---|
5 | Could we do *ompx* instead to avoid the mangled name in the list? |
openmp/libomptarget/DeviceRTL/src/exports | ||
---|---|---|
5 |
I think that is even worse. If the interface function calls will be generated by the front end, non-mangled interface functions are better. After all, name mangling is part of ABI. |
openmp/libomptarget/DeviceRTL/src/exports | ||
---|---|---|
5 | we could, but I'd honestly prefer to restrict it to the prefix and namespace. Though, given that this is our runtime, we might be able to avoid using ompx in the middle of a symbol name. |
If the interface function calls will be generated by the front end, non-mangled interface functions are better. After all, name mangling is part of ABI.
But they are not (always). The user will be able to call ompx::mapping::getThreadIdInBlock (or a renamed version of it).
openmp/libomptarget/DeviceRTL/src/exports | ||
---|---|---|
5 | If the uses are for users, that would be fine. Wonder how LLVM libraries expose their C++ interfaces. |
I'd prefer our ompx extension to be C interfaces instead of C++.