add not_equal NPU op #34560
add not_equal NPU op #34560
Conversation
|
Thanks for your contribution! |
|
|
| ops::LessEqualNPUKernel<plat::NPUDeviceContext, int8_t>, | ||
| ops::LessEqualNPUKernel<plat::NPUDeviceContext, uint8_t>, | ||
| ops::LessEqualNPUKernel<plat::NPUDeviceContext, int16_t>, | ||
| // ops::LessEqualNPUKernel<plat::NPUDeviceContext, uint16_t>, |
There was a problem hiding this comment.
这里的数据类型对应CANN的以下数据类型，都可以支持
static TensorType RealNumberType() {
return TensorType{DT_DOUBLE, DT_FLOAT, DT_FLOAT16, DT_INT16, DT_INT32, DT_INT64,
DT_INT8, DT_UINT16, DT_UINT32, DT_UINT64, DT_UINT8, DT_BF16};
}
There was a problem hiding this comment.
已注释的数据类型paddle不支持，取消注释编译会报错
There was a problem hiding this comment.
注释掉的数据类型都从代码里面删除一下，Paddle尽量不包含冗余代码，如果需要有需要说明的地方可以卸载注释里面
| ops::LessEqualNPUKernel<plat::NPUDeviceContext, int8_t>, | ||
| ops::LessEqualNPUKernel<plat::NPUDeviceContext, uint8_t>, | ||
| ops::LessEqualNPUKernel<plat::NPUDeviceContext, int16_t>, | ||
| // ops::LessEqualNPUKernel<plat::NPUDeviceContext, uint16_t>, |
There was a problem hiding this comment.
这里的数据类型对应CANN的以下数据类型，都可以支持
static TensorType RealNumberType() {
return TensorType{DT_DOUBLE, DT_FLOAT, DT_FLOAT16, DT_INT16, DT_INT32, DT_INT64,
DT_INT8, DT_UINT16, DT_UINT32, DT_UINT64, DT_UINT8, DT_BF16};
}
| @@ -142,11 +142,12 @@ def test_attr_name(self): | |||
| globals()[cls_name] = Cls | |||
|
|
|||
|
|
|||
| for _type_name in {'float16', 'float32', 'int32'}: | |||
| for _type_name in {'float16', 'float32', 'int32', 'int64'}: | |||
There was a problem hiding this comment.
在C++算子中支持的所有数据类型都需要在单测里面跑一下验证一下
There was a problem hiding this comment.
单测这边只支持bool、float32、float64、int32、int64，因此只添加了equal算子支持的bool类型
| ops::LessEqualNPUKernel<plat::NPUDeviceContext, int8_t>, | ||
| ops::LessEqualNPUKernel<plat::NPUDeviceContext, uint8_t>, | ||
| ops::LessEqualNPUKernel<plat::NPUDeviceContext, int16_t>, | ||
| // ops::LessEqualNPUKernel<plat::NPUDeviceContext, uint16_t>, |
There was a problem hiding this comment.
注释掉的数据类型都从代码里面删除一下，Paddle尽量不包含冗余代码，如果需要有需要说明的地方可以卸载注释里面
PR types
New features
PR changes
OPs
Describe
add not_equal NPU op
