~600 of the Triton tests are failing

[aregm]

https://github.com/openai/triton/tree/main/python/test/unit need to pass 100%

In order to complete this task we need to complete the following work items:

• [atomic_rmw]: Support for fp16 #152
• [test_scan2d]: Complete support for cumsum and cumprod #153
• [test_locality]: Fails to run correctly on PVC GPU. #154
• [test_scan_layouts]: Fails to run correctly on PVC #155
• [test_reduce_layouts]: Fails to run correctly on PVC (for DPAS layout) #156
• [test_dot_without_load]: Fails to verify when fdata type is float16 #157
• [test_masked_load_shared_memory]: Fails to run correctly on PVC #158
• [block_pointers]: Enable test test_block_pointer.py on PVC #159
• [tl.random]: Add support for Triton's tt.random.* operations #160
• [test_line_info]: Enable the test in test_line_info.py #161
• [test_chain_reduce]: Fails to run correctly on PVC #164
• Port and run operator tests #167
• Test dot_combine in test_line_info.py fails to compile #191
• Fix and enable tests in test_subprocess.py #205
• Port and run runtime tests #206
• [test_typeconvert_downcast]: Fails to run correctly on PVC #212
• [test_typeconvert_upcast]: Fails to run correctly on PVC #213
• [test_blocksparse.py::test_attention_fwd_bwd] fails to run correctly on PVC #251
• [test_flash_attention.py::test_op] fails to run correctly on PVC #253
• [operators/test_blocksparse.py::test_softmax] fails to run correctly on PVC #255
• [operators/test_inductor.py::test_scan2d_broadcast] fails to run correctly on PVC #256
• [operators/test_blocksparse.py::test_matmul] fails to run correctly on PVC #257
• [test_histogram]: fails to run correctly on PVC #261
• Fix tl.device_print code generation for XPU #291
• [test_subprocess]: test_assert with arg "device_assert", "assert", "no_debug", "double_assert" fails to run correctly #311
• [test_subprocess]: test_assert_nested fails to run correctly #312
• Port and Run tests in the regression directory #377
• Investigate and re-enable two matmul tests #381
• 10 test cases in test_matmul.py failed when set the threads_per_warp to 16. #435
• [test_core.py::test_dot3d] fails to run correctly on PVC #500
• [test_core.py::test_split] fails to run correctly on PVC #506
• Enable test_atomic_rmw core tests #513
• Investigate and enable passing tests for test_dot in test_core.py #514
• Improve success rate of test_core.py with TRITON_INTERPRET=1 #597
• [operators/test_flash_attention.py::test_op] Fix failures with TRITON_INTERPRET=1 #598
• [language/test_core.py::test_precise_math] tl.math.div_rn(x,y) fails on XPU #601

[aregm]

Currently, around 10% of tests fail on the branch https://github.com/intel-sandbox/triton-dse. While it is being merged to the public branch, @tdeng5 can you please start triaging the failing tests into buckets and opening issues on them, so we can start working?

[etiotto]

@aregm I have already opened issues to track the work required for this large work item. I have added to the description links to the various work items that collectively resolve this one. @tdeng5 please feel free to add anything else you think is required.

[tdeng5]

A task was created to collect the UT fails status: #223

[etiotto]

Assigning @whitneywhtsang and myself as lead for this work item. We will distribute subtasks as needed while bringing the rest of the team up to speed.

[vlad-penkin]

@etiotto i suggest to close this ticket and just track the remaining open tickets.