Cast (Multiple Types to float)

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product/Atlas A3 inference product	√
Atlas A2 training product/Atlas A2 inference product	√
Atlas 200I/500 A2 inference product	x
Atlas inference product AI Core	x
Atlas inference product Vector Core	x
Atlas training product	x

Function Usage

Converts the input data to the float type.

Prototype

Converts bfloat16_t to float.

__aicore__ inline float Cast(const bfloat16_t& bVal)

Converts multiple data types to float.

template <typename T, typename U = float,
          typename = Std::enable_if_t<
          (Std::is_same<T, bfloat16_t>::value || Std::is_same<T, hifloat8_t>::value ||
           Std::is_same<T, fp8_e5m2_t>::value || Std::is_same<T, fp8_e4m3fn_t>::value ||
           Std::is_same<T, fp4x2_e1m2_t>::value || Std::is_same<T, fp4x2_e2m1_t>::value), 
          void>>
__aicore__ constexpr inline U Cast(T bVal)

Parameters

**Table 1** Template parameters
Parameter	Description
T	Operand data type. For the Atlas 350 Accelerator Card, the supported data types are fp4x2_e2m1_t, fp4x2_e1m2_t, hifloat8_t, fp8_e5m2_t, fp8_e4m3fn_t, and bfloat16_t. To meet the 1-byte data size requirement, the fp4x2_e1m2_t and fp4x2_e2m1_t types are constructed by concatenating two 4-bit scalar data elements. During conversion, the lower 4 bits of the fp4x2_e1m2_t and fp4x2_e2m1_t scalar data are converted, that is, bits [0:3].
U	Data type of the return value. Only float is supported.

**Table 2** API parameters
Parameter	Input/Output	Meaning
bVal	Input	Scalar data to be converted.

Returns

Converted scalar data of the float type.

Constraints

None

Example

// Convert the first element of srcLocal to the float type using Cast and assign the result to fValue.
float fValue = AscendC::Cast(srcLocal.GetValue(0));
// Fill the scalar fValue into the first 32 positions of dstLocal.
AscendC::Duplicate(dstLocal, fValue, 32);

Result example:

Input (srcLocal):
[21.3750, 74.0000, 57.2500,  ...,  2.4062, 72.5000, 67.5000]
Output (dstLocal):
[21.375 21.375 21.375 ... 21.375 21.375 21.375]

// The following is an example of the input type fp4x2_e1m2_t:
float fValue = AscendC::Cast<T, float>(srcLocal.GetValue(0));
AscendC::Duplicate(dstLocal, fValue, bufferSize);

// Input (srcLocal) (binary expression): [0b10001111 0b10001111 0b10001111 ...]
// Output (dstLocal): [-1.75 -1.75 -1.75 ...]

Parent topic: Scalar Computation