[Compilation] Add _NO_HALF2 flags to be consistent with Pytorch

https://github.com/pytorch/pytorch/blob/eb7b89771e989f832147b8f28f62c0b43d5d3c12/cmake/Dependencies.cmake#L1693

setup.py

@@ -136,6 +136,8 @@ ext_modules.append(
                     "-std=c++17",
                     "-U__CUDA_NO_HALF_OPERATORS__",
                     "-U__CUDA_NO_HALF_CONVERSIONS__",
+                    "-U__CUDA_NO_HALF2_OPERATORS__",
+                    "-U__CUDA_NO_BFLOAT16_CONVERSIONS__",
                     "--expt-relaxed-constexpr",
                     "--expt-extended-lambda",
                     "--use_fast_math",