[BACKEND] Refactor the TritonGPU dialect utils to call the APIs of the TritonGPU layout attributes.

[chengjunlu]

Refactor the TritonGPU dialect utils to call the APIs of the TritonGPU attributes and its inheritance.

[chengjunlu]

Hi @ThomasRaoux ,
The CI test failed because the H100 cases fails.

FAILED hopper/test_persistent_warp_specialized_gemm.py::test_full_static_persistent_matmul_kernel[256-64-16-4-2-256-256-64-False-True-add-rows-float16-False-3-True]

I am looking for a H100 platform to debug this case.
I cannot find the helpful information about the failure in the test log. Do you know where it has more detail information about the crash?

[ptillet]

judging from the log in the crash

Thread 0x00007f782dc00700 (most recent call first):
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 474 in read
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 507 in from_io
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 1049 in _thread_receiver
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 296 in run
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 361 in _perform_spawn
Current thread 0x00007f782e5dd740 (most recent call first):
  File "/home/ptillet/.local/lib/python3.8/site-packages/triton/compiler/compiler.py", line 123 in optimize_ttgir
  File "/home/ptillet/.local/lib/python3.8/site-packages/triton/compiler/compiler.py", line 437 in <lambda>
  File "/home/ptillet/.local/lib/python3.8/site-packages/triton/compiler/compiler.py", line 543 in compile
  File "/home/ptillet/.local/lib/python3.8/site-packages/triton/runtime/jit.py", line 532 in run
  File "/home/ptillet/actions-runner/_work/triton/triton/python/test/unit/hopper/test_persistent_warp_specialized_gemm.py", line 889 in test_full_static_persistent_matmul_kernel
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/python.py", line 194 in pytest_pyfunc_call
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_callers.py", line 77 in _multicall
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_manager.py", line 115 in _hookexec
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_hooks.py", line 493 in __call__
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/python.py", line 1792 in runtest
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/runner.py", line 169 in pytest_runtest_call
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_callers.py", line 77 in _multicall
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_manager.py", line 115 in _hookexec
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_hooks.py", line 493 in __call__
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/runner.py", line 262 in <lambda>
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/runner.py", line 341 in from_call
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/runner.py", line 261 in call_runtest_hook
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/runner.py", line 222 in call_and_report
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/runner.py", line 133 in runtestprotocol
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/runner.py", line 114 in pytest_runtest_protocol
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_callers.py", line 77 in _multicall
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_manager.py", line 115 in _hookexec
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_hooks.py", line 493 in __call__
  File "/home/ptillet/.local/lib/python3.8/site-packages/xdist/remote.py", line 174 in run_one_test
  File "/home/ptillet/.local/lib/python3.8/site-packages/xdist/remote.py", line 157 in pytest_runtestloop
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_callers.py", line 77 in _multicall
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_manager.py", line 115 in _hookexec
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_hooks.py", line 493 in __call__
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/main.py", line 325 in _main
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/main.py", line 271 in wrap_session
  File "/home/ptillet/.local/lib/python3.8/site-packages/_pytest/main.py", line 318 in pytest_cmdline_main
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_callers.py", line 77 in _multicall
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_manager.py", line 115 in _hookexec
  File "/home/ptillet/.local/lib/python3.8/site-packages/pluggy/_hooks.py", line 493 in __call__
  File "/home/ptillet/.local/lib/python3.8/site-packages/xdist/remote.py", line 355 in <module>
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 1157 in executetask
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 296 in run
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 361 in _perform_spawn
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 343 in integrate_as_primary_thread
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 1142 in serve
  File "/home/ptillet/.local/lib/python3.8/site-packages/execnet/gateway_base.py", line 1640 in serve
  File "<string>", line 8 in <module>
  File "<string>", line 1 in <module>
................................................[gw4] node down: Not properly terminated
F
replacing crashed worker gw4

it seems like a segfault in one of the ttgir optimization passes

[ThomasRaoux]

Nice cleanup, thanks!
I think the crash may be related to the splitNum problem I pointed out. Looking at which tests fail it is definitely related to cases with multiple CTAs.
Let me know if you are not able to get a H100 to debug and I can try to help.

[chengjunlu]

Nice cleanup, thanks! I think the crash may be related to the splitNum problem I pointed out. Looking at which tests fail it is definitely related to cases with multiple CTAs. Let me know if you are not able to get a H100 to debug and I can try to help.

@ThomasRaoux Very appreciate you reviewed the code carefully. I think the code you pointed out is the root cause of the bug.
I just get the accessing to a H100. I will fix the issue soon.

[chengjunlu]

Hi @ThomasRaoux ,
I have fixed the issue and changed the code based on your comments.
Would you like to help to review again?

[ThomasRaoux]

LGTM