Open whitneywhtsang opened 2 days ago
Upstream commit https://github.com/triton-lang/triton/commit/1b35f118fd532ea4d032883408ea5729118b7767 implemented implicit tensor conversion for assignment operators. The feature is tested with test_core.py::test_store_constant_default_dtype.
test_core.py::test_store_constant_default_dtype
This upstream commit causes interpreter UT failures, a workaround is added in https://github.com/intel/intel-xpu-backend-for-triton/pull/1533: python/triton/runtime/interpreter.py:
python/triton/runtime/interpreter.py
# FIXME: This is a temporary workaround to avoid compilation failures. self.rewritted_fn[self.fn] = self.fn return self.fn
UT test_core.py::test_store_constant_default_dtype fails due to the workaround:
if is_interpreter(): pytest.skip("FIXME: Incorrect result on XPU")
This issue is to remove the workaround, and reenabled test_core.py::test_store_constant_default_dtype.
Upstream commit https://github.com/triton-lang/triton/commit/1b35f118fd532ea4d032883408ea5729118b7767 implemented implicit tensor conversion for assignment operators. The feature is tested with
test_core.py::test_store_constant_default_dtype
.This upstream commit causes interpreter UT failures, a workaround is added in https://github.com/intel/intel-xpu-backend-for-triton/pull/1533:
python/triton/runtime/interpreter.py
:UT
test_core.py::test_store_constant_default_dtype
fails due to the workaround:This issue is to remove the workaround, and reenabled
test_core.py::test_store_constant_default_dtype
.