Where

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product/Atlas A3 inference product	x
Atlas A2 training product/Atlas A2 inference product	x
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

Selects elements from two source operands based on specified conditions to generate a target operand. Both source operands can be LocalTensor or scalar.

$\text{[math]}$

Prototype

template <typename T, typename U, typename S, typename V>
__aicore__ inline void Where(const LocalTensor<T>& dst, const U& src0, const S& src1, const LocalTensor<V>& condition, const uint32_t count)

Parameters

**Table 1** Template parameters
Parameter	Description
T	Data type of the destination operand. For the Atlas 350 Accelerator Card, the supported data types are bool, int8_t, uint8_t, int16_t, uint16_t, half, bfloat16_t, int32_t, uint32_t, float, int64_t, and uint64_t.
U	LocalTensor or Scalar type. The type is automatically inferred based on the input parameter src0. Developers do not need to configure this parameter. Ensure that the data type of src0 is the same as that of the destination operand.
S	LocalTensor or Scalar type. The type is automatically inferred based on the input parameter src1. Developers do not need to configure this parameter. Ensure that the data type of src1 is the same as that of the destination operand.
V	Condition data type. Currently, only the bool data type is supported.

**Table 2** Parameters
Parameter	Type	Description
dst	Output	Destination operand. The type is LocalTensor, and TPosition can be VECIN, VECCALC, or VECOUT.
src0, src1	Input	Source operand. The type can be Scalar or LocalTensor. When the type is LocalTensor, the supported TPosition is VECIN, VECCALC, or VECOUT. The source and destination operands must have the same data type.
condition	Input	Condition operand. The type is LocalTensor, and TPosition can be VECIN, VECCALC, or VECOUT.
count	Input	Number of elements involved in the computation.

Returns

None

Constraints

The source operand address must not overlap the destination operand address.
For details about the alignment requirements of the operand address offset, see General Description and Restrictions.

Examples

For details about a complete example, see where operator sample.

AscendC::LocalTensor<half> dst, src0, src1;
AscendC::LocalTensor<bool> condition;
uint32_t count = 512; // Number of elements involved in computation
AscendC::Where(dst, src0, src1,  condition, count);

Result example:

Input (src0):
[1, 2, 3, ... 511, 512]
Input (src1):
[-1, -2, -3, ... -511, -512]
Condition input data (condition). If the value is 0, src1 is selected. If the value is 1, src0 is selected.
[0, 1, 0, ... 0, 1]
Output (dst):
[-1, 2, -3, ... -511, 512]

Parent topic: Where APIs