Modify device unary pointwise ops

[meinie0826]

PR Category

OP Test

Type of Change

Refactor

Description

Issue

Progress

• Change is properly reviewed (1 reviewer required, 2 recommended).
• Change is responded to an issue.
• Change is fully covered by a UT.

Performance

[iclementine]

This file has some error in unit test.
Since the output dtype is not the same.

________________ test_accuracy_dot_tensor_tensor[dtype2-shape6] ________________

shape = (81,), dtype = torch.bfloat16

    @pytest.mark.dot
    @pytest.mark.parametrize("shape", UT_SHAPES_1D)
    @pytest.mark.parametrize("dtype", FLOAT_DTYPES)
    def test_accuracy_dot_tensor_tensor(shape, dtype):
        if flag_gems.vendor_name == "kunlunxin":
            torch.manual_seed(0)
            torch.cuda.manual_seed_all(0)
    
        inp1 = torch.randn(shape, dtype=dtype, device="cpu")
        inp2 = torch.randn(shape, dtype=dtype, device="cpu")
        ref_inp1 = to_reference(inp1, True)
        ref_inp2 = to_reference(inp2, True)
        inp1 = to_reference(inp1, True)
        inp2 = to_reference(inp2, True)
        ref_out = torch.dot(ref_inp1, ref_inp2)
        with flag_gems.use_gems():
            res_out = torch.dot(inp1, inp2)
>       gems_assert_close(res_out, ref_out, dtype, equal_nan=True)

tests/test_blas_ops.py:196: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
tests/accuracy_utils.py:200: in gems_assert_close
    flag_gems.testing.assert_close(