[Release 2.9] Inductor perf cherry picks#2723

Merged

pruthvistony merged 23 commits intoROCm:release/2.9from

jataylo:jack-29-perfs

Oct 17, 2025

Commits on Oct 16, 2025

Naive foreach autotune support
jataylo
committed
Update triton_heuristics.py
jataylo
committed
Update triton_heuristics.py
jataylo
committed
Linting
jataylo
committed
Add max_autotune_enabled boolean.

naromero77amd
authored and
jataylo
committed
Conditionalize xblock values for HIP.

naromero77amd
authored and
jataylo
committed
Create tiny_config.

naromero77amd
authored and
jataylo
committed
Do not filter configs when tuning.

naromero77amd
authored and
jataylo
committed
Add tiny_config when autotune enabled.

naromero77amd
authored and
jataylo
committed
Fix persistent reduction conflicts
jataylo
committed
pointwise autotuning returnz (#2636)

AmdSampsa
authored and
jataylo
committed
a few more configs for 2d grids (#2649)

AmdSampsa
authored and
jataylo
committed
[ROCm][inductor] Additional pointwise tunings (#2642)

naromero77amd
authored and
jataylo
committed
pointwise config with MAX_BLOCK.

naromero77amd
authored and
jataylo
committed
Update triton_config method.

naromero77amd
authored and
jataylo
committed
Increase TRITON_MAX_BLOCK X value.

naromero77amd
authored and
jataylo
committed
even more configs

AmdSampsa
authored and
jataylo
committed
More ROCm conditionalisation
jataylo
committed
Lint.

naromero77amd
authored and
jataylo
committed
Reduction heursitics improvements for ROCm

naromero77amd
authored and
jataylo
committed
Add PropagateNan argument to minimum and maximum function.

naromero77amd
authored and
jataylo
committed
Update autotune_lookup_tunable UT.

naromero77amd
authored and
jataylo
committed
New WRT configs for autotuning (#2708)

AmdSampsa
authored and
jataylo
committed