Repository navigation
Commit 1be4895
committed
GPU Metal: fence only threadgroup memory in the work-group collectives
The Metal work-group scan, reduction and broadcast exchange data through
threadgroup memory only, so their barriers need not order device memory.
GPUbarrier() fences both, which on Apple GPUs makes each barrier wait for
the outstanding device-memory accesses. With threadgroup-only fences the
stream-compaction scanStart takes 12% less GPU time (30 kHz Pb-Pb, 1/4 TF,
no --debug, 3 interleaved rounds); results unchanged.1 parent 7f86b5e commit 1be4895
1 file changed
Lines changed: 8 additions & 6 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
508 | 508 | | |
509 | 509 | | |
510 | 510 | | |
511 | | - | |
| 511 | + | |
| 512 | + | |
| 513 | + | |
512 | 514 | | |
513 | 515 | | |
514 | 516 | | |
| |||
567 | 569 | | |
568 | 570 | | |
569 | 571 | | |
570 | | - | |
| 572 | + | |
571 | 573 | | |
572 | 574 | | |
573 | 575 | | |
574 | 576 | | |
575 | 577 | | |
576 | 578 | | |
577 | | - | |
| 579 | + | |
578 | 580 | | |
579 | 581 | | |
580 | 582 | | |
581 | | - | |
| 583 | + | |
582 | 584 | | |
583 | 585 | | |
584 | 586 | | |
| |||
589 | 591 | | |
590 | 592 | | |
591 | 593 | | |
592 | | - | |
| 594 | + | |
593 | 595 | | |
594 | | - | |
| 596 | + | |
595 | 597 | | |
596 | 598 | | |
597 | 599 | | |
| |||
0 commit comments