Reland Diff for fbgemm::all_to_one to reduce CUDAEvents costs #2568
Job | Run time |
---|---|
14m 30s | |
14m 26s | |
14m 33s | |
14m 5s | |
14m 40s | |
14m 8s | |
9m 29s | |
7m 33s | |
6m 0s | |
7m 14s | |
6m 50s | |
7m 3s | |
6m 27s | |
6m 47s | |
6m 52s | |
6m 51s | |
11m 54s | |
12m 18s | |
11m 42s | |
14m 31s | |
8m 13s | |
11m 1s | |
13m 5s | |
12m 13s | |
4h 12m 25s |