[Tags: v2.1] Error behaviorfor 'flag_gems.masked_fill(inp, mask, value)'

[Dayrker]

Describe the bug

位置：
FlagGems/tests/test_unary_pointwise_ops.py::test_accuracy_masked_fill
FlagGems/src/flag_gems/ops/masked_fill.py

现象1：
在masked_fill.py中加打印， with flag_gems.use_gems():使用时不会出现打印，flag_gems.masked_fill(inp, mask, value)使用时才会走打印引处
怀疑原因：
算子没替换好

现象2：
flag_gems.masked_fill(inp, mask, value)使用时，‘line61: value = value.contiguous()’出错
怀疑原因：
value一般输入是int，不需要.contiguous()

[StrongSpoon]

@GwokHiujin please take a look. I think we need another kernel to deal with scalar-type value.

[GwokHiujin]

@GwokHiujin please take a look. I think we need another kernel to deal with scalar-type value.

Indeed, PyTorch includes the aten::masked_fill.Scalar(Tensor self, Tensor mask, Scalar value) -> (Tensor) operation. I will proceed with this task.