Skip to content

[WIP] Merge inductor perf updates from perf branch #2633

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.

Sign up for GitHub

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

Jump to bottom
Draft
wants to merge 15 commits into
base: rocm7.1_internal_testing
Choose a base branch
from
Draft

[WIP] Merge inductor perf updates from perf branch #2633

wants to merge 15 commits into from
+114 −64

Conversation

jataylo
Copy link

@jataylo jataylo commented Sep 11, 2025

Change list:

  1. Use tl.minimum, tl.maximum instead of libdevice implementations
  2. Enable pipelining in reduction codegen
  3. Increase max block size in tuning
  4. Increase spill thresholding to avoid filtering out good configs
  5. Add waves_per_eu tune support for reduction configs
  6. Add new pointwise tuning configuration
  7. Add new reduction tuning configuration

iupaikov-amd and others added 14 commits September 11, 2025 12:37
@iupaikov-amd @jataylo
Added triton perf improvement changes
da10605 
(cherry picked from commit 47078b5)
@iupaikov-amd @jataylo
Added a place to put more reduction configs
beebe3a 
(cherry picked from commit 44e07d7)
@iupaikov-amd @jataylo
added a dummy config for reduction kernels
30fe066 
(cherry picked from commit 1f90e3f)
@iupaikov-amd @jataylo
Increased the number of spils for kernel configs
95e6f88 
(cherry picked from commit b41f0e4)
@iupaikov-amd @jataylo
added a better reduction config
a5b5084 
(cherry picked from commit 5515cea)
@xiaohuguo2023 @jataylo
enable pipeline
1a0da8c 
(cherry picked from commit 4b62333)
@xiaohuguo2023 @jataylo
add one more reduction config to avoid large vgpr spills
b9d3752 
(cherry picked from commit 2d423a3)
@jataylo
Bug fix and optimisation for persistent reduction kernel tuning
3b5f047 
Original PR had incorrect indentation. Updated PR such that autotune will always add tiny configs, otherwise use the hinted configs only.

(cherry picked from commit 8c58805)
@jataylo
Duplicate config block
8f670db 
(cherry picked from commit c8f6b02)
@jataylo
Fix 1D config being applied to 2D configs
f1d1d1d 
(cherry picked from commit 3b996ff)
@jataylo
Remove outdated condition
7f1c35f 
(cherry picked from commit a6e5fe6)
@jataylo
Initialise result configs to fix bug
6c068ee 
(cherry picked from commit 7c7ad78)
@jataylo
[Perf branch] Add support for 2d reduction and bug fix (ROCm#2629)
d417cd0 
(cherry picked from commit d81b7e9)
@jataylo
Update triton_heuristics.py
57669c4 
(cherry picked from commit 4377d1f)
@rocm-repo-management-api
Copy link

rocm-repo-management-api bot commented Sep 11, 2025 •
edited
Loading

Jenkins build for 5e8eb7823acd695ae930a8bc5174b3892a6a5d11 commit finished as FAILURE
Links: Blue Ocean view / Build artifacts

@rocm-repo-management-api
Copy link

rocm-repo-management-api bot commented Sep 11, 2025 •
edited
Loading

Jenkins build for 5e8eb7823acd695ae930a8bc5174b3892a6a5d11 commit finished as FAILURE
Links: Blue Ocean view / Build artifacts

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@jataylo @iupaikov-amd @xiaohuguo2023