CUDA: fix MMQ writeback for int8 tensor cores #13222
| Job | Run time |
|---|---|
| 2m 22s | |
| 9m 0s | |
| 2m 47s | |
| 1m 37s | |
| 2m 55s | |
| 1m 47s | |
| 1m 38s | |
| 1m 56s | |
| 13m 26s | |
| 11m 2s | |
| 11m 5s | |
| 2m 16s | |
| 2m 23s | |
| 3m 42s | |
| 17m 45s | |
| 1m 57s | |
| 6m 0s | |
| 8m 23s | |
| 5m 51s | |
| 15s | |
| 1s | |
| 3m 26s | |
| 15s | |
| 6m 28s | |
| 1s | |
| 31m 15s | |
| 7m 0s | |
| 30m 57s | |
| 7m 16s | |
| 12m 37s | |
| 7m 22s | |
| 11m 20s | |
| 7m 25s | |
| 9m 1s | |
| 17m 45s | |
| 8m 26s | |
| 7m 33s | |
| 7m 32s | |
| 6m 15s | |
| 3m 49s | |
| 3m 7s | |
| 0s | |
| 4h 56m 58s |