-
Notifications
You must be signed in to change notification settings - Fork 75
[NO CP][release/2.7][ROCm][inductor] Inductor heuristic upstream backports #2807
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[NO CP][release/2.7][ROCm][inductor] Inductor heuristic upstream backports #2807
Conversation
|
Jenkins build for 7850a9c97813ff2687769efd9a6c4ff5ff749187 commit finished as FAILURE |
|
Jenkins build for dbdb5542c2ae0f09415495c33bfd7d5d0f77bc53 commit finished as FAILURE |
Added a check that includes autotune configs for 2D POI only if their size is big enough. (cherry picked from commit a2b0fd7)
|
Jenkins build for d235a1504f6702249dd72deef1a8f68ce991320a commit finished as FAILURE |
|
Jenkins build for 627a5718c93f8c54fca6787f3167b2b454717226 commit finished as FAILURE |
|
Jenkins build for b1cdd5584626c1f0c2c6bad6b58272da6901e619 commit finished as FAILURE |
|
Jenkins build for b1cdd5584626c1f0c2c6bad6b58272da6901e619 commit finished as FAILURE |
|
Jenkins build for d356b844b19b6dfb588b2f5815ebbefca0bba579 commit finished as FAILURE |
|
Tested with TORCHINDUCTOR_MAX_AUTOTUNE_POINTWISE=1 to confirm we are getting the extra configs (note that some of them are getting filtered/scaled out as expected). For triton_red_fused_sum_view_22.py: triton_poi_fused_threshold_backward_36 (1D) triton_poi_fused_slice_13 (2D) triton_poi_fused__to_copy_index_add_new_zeros_4 (contans the atomic add config) triton_per_fused_sum_view_23 |
|
Ran linter several times to clean the file up. |
a5d6423 to
badfab0
Compare
|
Jenkins build for badfab0d09d48b0a580339e5119455ce0f30fcc7 commit finished as FAILURE |
|
Ran the following test suites as follows: No new regressions reported: |
jataylo
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
These are backports based on these upstream PRs. Cherrypicks were performed when they where possible.
pytorch#163908 (persistent reduction autotune)
pytorch#161280 (reduction)
pytorch#162053 (foreach)
pytorch#163197 (pointwise)
pytorch#166470 (pointwise config for atomic add)
Also included are some additional customer-specific configs which were not upstreamed but are in this backport to 2.9
#2723
Did not backport filter functions such as
_maybe_filter_configs_for_tma_restrictionshttps://github.com/ROCm/pytorch/blob/release/2.9/torch/_inductor/runtime/triton_heuristics.py#L2614