| 703 | |
| 704 | |
| 705 | void comm_combineSubArrays(qcomp* recv, vector<qindex> recvInds, vector<qindex> sendInds, vector<qindex> numSend) { |
| 706 | |
| 707 | // recv has already been overwritten with local contributions, which enables |
| 708 | // direct GPU-to-recv writing (avoiding a superfluous CPU-CPU copy) |
| 709 | qcomp* send = nullptr; |
| 710 | bool areGpuPtrs = false; |
| 711 | globallyCombineNonUniformSubArrays(recv, send, recvInds, sendInds, numSend, areGpuPtrs); |
| 712 | } |
| 713 | |
| 714 | |
| 715 |
no test coverage detected