Skip to content

Actions: ROCm/iris

Actions

All workflows

Actions

Loading...
Loading

Showing runs from all workflows
2,500+ workflow runs
2,500+ workflow runs

Filter by Workflow

⌥ + click/return to exclude

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
Iris Nightly Triton Test Iris Nightly Triton Test #45: Scheduled
1h 13m 7s main
main
1h 13m 7s
Nightly Docker Build Nightly Docker Build #32: Scheduled
1h 3m 50s main
main
1h 3m 50s
Add graph capture test for one_shot_vllm, rename shmem to ctx Lint & Auto-Fix #4058: Commit 4919e67 pushed by mawad-amd
35s muhaawad/one-shot-vllm
muhaawad/one-shot-vllm
35s
Add one_shot_vllm all-reduce variant Lint & Auto-Fix #4057: Commit b7d8b00 pushed by mawad-amd
39s muhaawad/one-shot-vllm
muhaawad/one-shot-vllm
39s
Revert all_to_all to 512KB threshold (native crashes at larger sizes) Lint & Auto-Fix #4056: Commit 965fceb pushed by mawad-amd
19s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
19s
Reduce all_to_all native threshold to 1MB (4MB caused memory faults) Lint & Auto-Fix #4055: Commit d86d5cb pushed by mawad-amd
41s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
41s
Lower all_gather large threshold to 2MB, extend all_to_all native to 4MB Lint & Auto-Fix #4054: Commit e2e462c pushed by mawad-amd
42s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
42s
Lower reduce NCCL small threshold from 512KB to 256KB Lint & Auto-Fix #4053: Commit 2b102fe pushed by mawad-amd
38s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
38s
Tune thresholds: all_to_all 512KB, scatter 512KB, all_reduce fused 256KB Lint & Auto-Fix #4052: Commit d6194ba pushed by mawad-amd
46s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
46s
Cap one_shot_fused at 512KB, two_shot handles 512KB-8MB Lint & Auto-Fix #4051: Commit 3b8518b pushed by mawad-amd
30s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
30s
Route small all-reduce to one_shot_fused, lower NCCL threshold to 32KB Lint & Auto-Fix #4050: Commit 3ea2337 pushed by mawad-amd
34s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
34s
Replace fused barrier with per-block barrier in one_shot_fused Lint & Auto-Fix #4049: Commit 0ccc064 pushed by mawad-amd
31s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
31s
Raise all_to_all NCCL threshold to 4MB total tensor size Lint & Auto-Fix #4048: Commit e213146 pushed by mawad-amd
57s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
57s
Tune NCCL thresholds based on v16 benchmark data Lint & Auto-Fix #4047: Commit 06a56ad pushed by mawad-amd
20s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
20s
Disable ring broadcast/reduce and adjust thresholds Lint & Auto-Fix #4046: Commit ce7e11e pushed by mawad-amd
49s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
49s
Switch all_gather default to persistent (push) variant Lint & Auto-Fix #4045: Commit c3d778c pushed by mawad-amd
50s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
50s
Add ring broadcast and ring reduce kernels for large messages Lint & Auto-Fix #4044: Commit 52f4488 pushed by mawad-amd
25s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
25s
one_shot_fused: dynamic SM count based on tile count Lint & Auto-Fix #4043: Commit 98bd6c4 pushed by mawad-amd
44s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
44s
one_shot_fused: PID-0-only barrier with sense reversal Lint & Auto-Fix #4042: Commit 9978191 pushed by mawad-amd
37s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
37s
Add one_shot_fused all_reduce variant with fused start+end barriers Lint & Auto-Fix #4041: Commit 12f3877 pushed by mawad-amd
50s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
50s
Revert all_to_all threshold to 64MB — native kernel wins up to 32MB Lint & Auto-Fix #4040: Commit e1480f8 pushed by mawad-amd
17s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
17s
Tune NCCL routing thresholds for all_to_all and scatter Lint & Auto-Fix #4039: Commit 7345cf2 pushed by mawad-amd
33s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
33s
Enable native all_to_all Triton kernel for all message sizes Lint & Auto-Fix #4038: Commit 0be75ac pushed by mawad-amd
15s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
15s
Cache transpose buffers in all_to_all and group info in reduce_scatter Lint & Auto-Fix #4037: Commit 28e9145 pushed by mawad-amd
41s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
41s
Remove dead mask computation in two_shot fast path Lint & Auto-Fix #4036: Commit d8fd626 pushed by mawad-amd
16s muhaawad/new-collectives-v2
muhaawad/new-collectives-v2
16s