[pd-hung]

Summary

Pin Triton below 3.9. Triton 3.9.0 (released Oct 27) changed interpreter internals TileLens hooks into, such as the load/store builder signatures, the removal of block pointers, and a new tl.condition patch that breaks Gluon.

Test Plan

With Triton 3.8.0: uv run pytest tests/ (357 passed), and pytest tests -m nki (254 passed).

Related Issues

Fixes the CI failures on previous PR

Breaking Changes

Triton 3.9 is no longer supported until the follow-up COMPAT PR.

Checklist

• I added tests to all new functionality I added/bugs I fixed.
• I verified that a human has reviewed all code in this PR.
• I ran npm run build:frontend if the PR modified any TypeScript code.
• I made sure that my code is well documented (comments explaining strange code, docstrings for functions, website modified if new functionality added).

[github-actions]

Performance Benchmark

Benchmark	main (min)	PR (min)	Change	Samples
gemm	N/A	0.096s	N/A	0 / 20
gemm_oob	N/A	0.107s	N/A	0 / 20
indirect_load	N/A	0.019s	N/A	0 / 20
nested_loop	N/A	0.201s	N/A	0 / 20
block_pointer_loop_advance	N/A	0.109s	N/A	0 / 20
liger_jsd	N/A	0.129s	N/A	0 / 20
flaggems_layernorm	N/A	0.341s	N/A	0 / 20
swiglu	0.153s	0.156s	+2.0%	20 / 20
cross_entropy	N/A	0.875s	N/A	0 / 20
fused_linear_jsd	N/A	0.194s	N/A	0 / 20
Total	N/A	2.230s	N/A	N/A

Iterations: 1 warmup + 20 measured
Samples are shown as main / PR; long pytest benchmarks may use fewer samples.