Closed AshburnLee closed 1 week ago
Status update on Rolling:
There are 19 pytest.skip
that were added by our team, they are in a single test file: test_core.py
. There are 166 FAILED cases, 6 of them have diff issue:
test_core.py::test_inline_asm[1]
Arrays are not equaltest_core.py::test_inline_asm_multiple_outputs
Arrays are not equaltest_core.py::test_inline_asm_packed[1]
Arrays are not equaltest_core.py::test_inline_asm_packed_multiple_outputs
Arrays are not equaltest_core.py::test_inline_asm_with_pointers[1]
Arrays are not equalThe above 5 cases are supposed to be skipped, checkout issue#223
test_core.py::test_precise_math[1-tl.math.sqrt_rn(x)-tl.math.sqrt(x.to(tl.float64)).to(tl.float32)]
Full diff issues#847 is trackingtest_dot3d
, all core dumped. issue#500 is trackingTodo: I will put those cases in a skiplist and remove pytest.skip
from code
Update:
Compared with OpenAI Triton and found 8 pytest.skip
in test_core.py
that were added by us. Removed and put FAILED in deselect files.
Below FAILED is being tracked:
test_precise_math[1-tl.math.sqrt_rn(x)-tl.math.sqrt(x.to(tl.float64)).to(tl.float32)]
has been tracked by https://github.com/intel/intel-xpu-backend-for-triton/issues/847
test_dot3d[*]
has been tracked by https://github.com/intel/intel-xpu-backend-for-triton/issues/500
test_reduce[4-32-1-xor_sum-bool-shape173-1-False]
test_reduce[4-32-1-xor_sum-bool-shape175-1-False]
test_reduce[4-32-1-xor_sum-bool-shape177-1-False]
test_reduce[64-16-1-xor_sum-bool-shape173-1-False]
test_reduce[64-16-1-xor_sum-bool-shape175-1-False]
test_reduce[64-16-1-xor_sum-bool-shape177-1-False]
has been tracked by https://github.com/intel/intel-xpu-backend-for-triton/issues/1535
test_store_constant_default_dtype[1]
has been tracked by https://github.com/intel/intel-xpu-backend-for-triton/issues/1534
test_dot_without_load[float16]
has been tracked by https://github.com/intel/intel-xpu-backend-for-triton/issues/671
test_masked_load_shared_memory[dtype1]
has been tracked by https://github.com/intel/intel-xpu-backend-for-triton/issues/671
test_tl_range
has been tracked by https://github.com/intel/intel-xpu-backend-for-triton/issues/1541
test_dot_max_num_imprecise_acc[*]
has been tracked by https://github.com/intel/intel-xpu-backend-for-triton/issues/1541
Update:
test_reduce[4-32-1-xor_sum-bool-shape173-1-False]
test_reduce[4-32-1-xor_sum-bool-shape175-1-False]
test_reduce[4-32-1-xor_sum-bool-shape177-1-False]
test_reduce[64-16-1-xor_sum-bool-shape173-1-False]
test_reduce[64-16-1-xor_sum-bool-shape175-1-False]
test_reduce[64-16-1-xor_sum-bool-shape177-1-False]
has been tracked by https://github.com/intel/intel-xpu-backend-for-triton/issues/1535
Those 6 cases now passed. That issue has been closed 2 days ago
Update:
conda basekit build and test
as there is a import error from 7/4/2024. Tracked by issue: https://github.com/intel/intel-xpu-backend-for-triton/issues/1562As discussed, skiplist for conda basekit build and test
will not be created.