CUDA: fix misaligned shared memory read #13290
| Job | Run time |
|---|---|
| 2m 21s | |
| 8m 24s | |
| 2m 46s | |
| 1m 37s | |
| 2m 34s | |
| 4m 16s | |
| 11m 11s | |
| 13m 33s | |
| 4m 53s | |
| 2m 14s | |
| 8m 30s | |
| 2m 54s | |
| 2m 5s | |
| 5m 56s | |
| 1m 56s | |
| 3m 33s | |
| 1m 49s | |
| 11m 15s | |
| 6m 8s | |
| 17m 50s | |
| 1m 54s | |
| 1m 36s | |
| 8m 7s | |
| 7m 36s | |
| 36m 58s | |
| 8m 33s | |
| 6m 45s | |
| 12m 14s | |
| 5m 31s | |
| 29m 24s | |
| 6m 33s | |
| 11m 28s | |
| 9m 12s | |
| 6m 31s | |
| 8m 49s | |
| 22m 15s | |
| 6m 27s | |
| 7m 32s | |
| 7m 20s | |
| 4m 15s | |
| 3m 18s | |
| 0s | |
| 5h 28m 3s |