search for: _nowait

Hi Johannes, I've been thinking on how to implement a solution for the problem and I have some questions: 1. You mention in the problem description that the current asynchronous functions can be used, and probably modified. The problem is that those runtime functions with _nowait at the end are not asynchronous. Looking at their implementations, they just wait for other tasks is there are pending tasks, and then call its synchronous versions. So, is it a good idea to wrap them in functions that call them as asynchronous (using std::async), then return an object to wait...

...to create the testcases with the FIXMEs showing how it should look after the transformation, I've been trying to come up with a way of splitting the runtime call __tgt_target_teams based on Silei's recent patch D77005. I like the idea of using its asynchronous version __tgt_target_teams_nowait, but I think we would have to modify it, either to return the async_info struct: __tgt_async_info __tgt_target_teams_nowait(...) // This would break the current canonical way of returning a status code. or to receive it as a pointer and modify its content, then waiting on that async_info objec...

...bonm%40eafit.edu.co%7C1c12e490f3804f6a772d08d7cf570a3a%7C99f7b55e9cbe467b8143919782918afb%7C0&sdata=XIo6OoK0Xnhat9bZbTNz9Rq9hYvBQhknTxULoA8Ma%2Fk%3D&reserved=0 , the mapping request, > > which may or may not cause data-transfers, are part of the call to > __tgt_target_teams[_nowait] and __tgt_target_data_update[_nowait]. Start > by reading the impl. of those two functions in the openmp runtime > (llvm-project/openmp/...). > > Let me know if this helps and once you have new questions. > > Cheers, > Johannes > > P.S. I merged your first patch. You...

Hi Johannes, Yes, I'm highly interested, I already started. It would be nice if you post the details again. I didn't take note of them, I was always looking at the issue in GH. The problem I'm facing right now is this: The first task is identifying OpenMP API calls referred to device memory allocation. I first understood how the compiler driver orchestrates the process of offloading