[AMDGPU] Use shared memory in multi_mma ukernel #17172
Job | Run time |
---|---|
7s | |
11m 36s | |
11m 50s | |
4m 14s | |
25m 37s | |
1m 58s | |
16m 51s | |
6m 13s | |
16m 1s | |
6m 32s | |
3m 5s | |
4m 32s | |
9m 9s | |
9m 10s | |
3m 0s | |
1m 51s | |
9m 30s | |
34s | |
46s | |
1m 8s | |
2s | |
2h 23m 46s |
Job | Run time |
---|---|
7s | |
11m 36s | |
11m 50s | |
4m 14s | |
25m 37s | |
1m 58s | |
16m 51s | |
6m 13s | |
16m 1s | |
6m 32s | |
3m 5s | |
4m 32s | |
9m 9s | |
9m 10s | |
3m 0s | |
1m 51s | |
9m 30s | |
34s | |
46s | |
1m 8s | |
2s | |
2h 23m 46s |