Autows Testing

AutoWS Correctness Testing

Do NOT run third_party/tlx/tutorials/testing/test_correctness.py for autoWS. Those tests cover manual warp specialization via TLX, which is a separate system.

The canonical test list lives in third_party/nvidia/hopper/run_all.sh — check that file if the list below seems out of date.

Python tests

# GEMM autoWS Python test
pytest python/test/unit/language/test_tutorial09_warp_specialization.py

# Addmm autoWS Python test
pytest python/test/unit/language/test_autows_addmm.py

# AutoWS FA correctness (replaces test_autows_flash_attention.py).
# Covers both fused_attention_ws_device_tma (compiler DP) and
# fused_attention_ws_device_tma_dp (explicit DP) tutorial kernels.
pytest third_party/tlx/tutorials/testing/test_correctness_autows.py

# FA autoWS tutorial kernels
TRITON_ALWAYS_COMPILE=1 pytest python/tutorials/fused-attention-ws-device-tma.py
TRITON_ALWAYS_COMPILE=1 python python/tutorials/test_tlx_bwd_from_fused_attention.py

# FA autoWS Hopper tutorial kernel
TRITON_ALWAYS_COMPILE=1 TRITON_USE_META_WS=1 pytest python/tutorials/fused-attention-ws-device-tma-hopper.py

LIT tests

Run all WarpSpecialization LIT tests:

lit test/Hopper/WarpSpecialization/

If tests hang

Run third_party/tlx/killgpu.sh to kill GPU processes that have been running too long.