Rewrite the Haswell SROT/DROT kernel tail loop with AVX2 to get consistent FMA rounding #6577
| Job | Run time |
|---|---|
| 7m 50s | |
| 8m 48s | |
| 7m 40s | |
| 9m 55s | |
| 7m 50s | |
| 1h 47m 44s | |
| 1h 47m 45s | |
| 1h 16m 56s | |
| 1h 12m 30s | |
| 1h 16m 57s | |
| 1h 34m 59s | |
| 46m 10s | |
| 23m 3s | |
| 29m 50s | |
| 11m 35s | |
| 18m 20s | |
| 28m 36s | |
| 15m 21s | |
| 21m 58s | |
| 51m 42s | |
| 45m 26s | |
| 16m 36s | |
| 38m 29s | |
| 52m 4s | |
| 38m 58s | |
| 14m 54s | |
| 48m 7s | |
| 43m 21s | |
| 50m 29s | |
| 21m 23s | |
| 43m 16s | |
| 20h 38m 32s |