Deduction Relationship

Tensor-Tensor Deduction Relationship

Deduction Rules

When the input aclTensor data types of an API (such as aclnnAdd, aclnnMul, etc.) are inconsistent, the API internally deduces a data type and converts the input data to that data type for computation.

For the data types supported by aclTensor, see Data Types. Some of these types follow the deduction rules below. The deduction principle is similar to Type Promotion in PyTorch.

The type deduction rules are as follows:

NOTE

  • For ease of description, the data types used in the table are in abbreviated form, with the following meanings: ACL_FLOAT(f32), ACL_FLOAT16(f16), ACL_DOUBLE(f64), ACL_BF16(bf16), ACL_INT8(s8), ACL_UINT8(u8), ACL_INT16(s16), ACL_UINT16(u16), ACL_INT32(s32), ACL_UINT32(u32), ACL_INT64(s64), ACL_UINT64(u64), ACL_BOOL(bool), ACL_COMPLEX32(c32), ACL_COMPLEX64(c64), ACL_COMPLEX128(c128).
  • The table header and the leftmost column represent the two input data types to be deduced, and the corresponding cell in the table indicates the deduced data type.
  • A cross (×) in the table indicates that the two types cannot be used for deduction computation.

Table 1 Data type deduction relationship

Data Type f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 bool c32 c64 c128
f32 f32 f32 f64 f32 f32 f32 f32 × f32 × f32 × f32 c64 c64 c128
f16 f32 f16 f64 f32 f16 f16 f16 × f16 × f16 × f16 c32 c64 c128
f64 f64 f64 f64 f64 f64 f64 f64 × f64 × f64 × f64 c128 c128 c128
bf16 f32 f32 f64 bf16 bf16 bf16 bf16 × bf16 × bf16 × bf16 c32 c64 c128
s8 f32 f16 f64 bf16 s8 s16 s16 × s32 × s64 × s8 c32 c64 c128
u8 f32 f16 f64 bf16 s16 u8 s16 × s32 × s64 × u8 c32 c64 c128
s16 f32 f16 f64 bf16 s16 s16 s16 × s32 × s64 × s16 c32 c64 c128
u16 × × × × × × × u16 × × × × × × × ×
s32 f32 f16 f64 bf16 s32 s32 s32 × s32 × s64 × s32 c32 c64 c128
u32 × × × × × × × × × u32 × × × × × ×
s64 f32 f16 f64 bf16 s64 s64 s64 × s64 × s64 × s64 c32 c64 c128
u64 × × × × × × × × × × × u64 × × × ×
bool f32 f16 f64 bf16 s8 u8 s16 × s32 × s64 × bool c32 c64 c128
c32 c64 c32 c128 c32 c32 c32 c32 × c32 × c32 × c32 c32 c64 c128
c64 c64 c64 c128 c64 c64 c64 c64 × c64 × c64 × c64 c64 c64 c128
c128 c128 c128 c128 c128 c128 c128 c128 × c128 × c128 × c128 c128 c128 c128

Deduction Examples

  • When calling the aclnnAdd API, if the input parameters have inconsistent data types — one being float16 and the other float32 — the API internally converts the float16 data type to float32 and then performs the computation.
  • When calling the aclnnAdd API, if the input parameters have inconsistent data types — one being float32 and the other bool — the API internally converts the bool data type to float32 and then performs the computation.

Deduction Relationship between Tensor and Scalar

Deduction Rules

When the input Tensor data type and the input Scalar data type of an API (such as aclnnAdds, aclnnMuls, etc.) are inconsistent, the API internally deduces a data type and converts the input data to that data type for computation.

For the data types supported by aclTensor, see Data Types. Some of these types follow the deduction rules below. The deduction principle is similar to Type Promotion in PyTorch.

The type deduction rules are as follows:

NOTE

  • For ease of description, the data types used in the table are abbreviated. Their meanings are: ACL_FLOAT(f32), ACL_FLOAT16(f16), ACL_DOUBLE(f64), ACL_BF16(bf16), ACL_INT8(s8), ACL_UINT8(u8), ACL_INT16(s16), ACL_UINT16(u16), ACL_INT32(s32), ACL_UINT32(u32), ACL_INT64(s64), ACL_UINT64(u64), ACL_BOOL(bool), ACL_COMPLEX32(c32), ACL_COMPLEX64(c64), ACL_COMPLEX128(c128).
  • In the table, the header row represents the input Tensor data type to be deduced, the leftmost column represents the input Scalar data type to be deduced, and the corresponding cell in the table indicates the deduced data type.
  • A cross mark (x) in the table indicates that the two types cannot undergo deduction computation.

Table 2 Data type deduction relationship

Data Type f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 bool c32 c64 c128
f32 f32 f16 f64 bf16 f32 f32 f32 x f32 x f32 x f32 c32 c64 c128
f16 f32 f16 f64 bf16 f32 f32 f32 x f32 x f32 x f32 c32 c64 c128
f64 f32 f16 f64 bf16 f32 f32 f32 x f32 x f32 x f32 c128 c128 c128
bf16 f32 f16 f64 bf16 f32 f32 f32 x f32 x f32 x f32 c32 c64 c128
s8 f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 s8 c32 c64 c128
u8 f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 u8 c32 c64 c128
s16 f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 s16 c32 c64 c128
u16 f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 x c32 c64 c128
s32 f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 s32 c32 c64 c128
u32 f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 x c32 c64 c128
s64 f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 s64 c32 c64 c128
u64 f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 x c32 c64 c128
bool f32 f16 f64 bf16 s8 u8 s16 u16 s32 u32 s64 u64 bool c32 c64 c128
c32 c64 c32 c128 c64 c64 c64 c64 c64 c64 c64 c64 c64 c64 c32 c64 c128
c64 c64 c32 c128 c64 c64 c64 c64 c64 c64 c64 c64 c64 c64 c32 c64 c128
c128 c64 c32 c128 c64 c64 c64 c64 c64 c64 c64 c64 c64 c64 c32 c64 c128

Deduction Examples

  • If the data type of the input Tensor is float16 and the data type of the input Scalar is float32, the API internally converts the float32 data type of the input Scalar to float16 before performing the computation.
  • If the data type of the input Tensor is bool and the data type of the input Scalar is float32, the API internally converts the bool data type of the input Tensor to float32 before performing the computation.