Duplicate

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

Supports scalar and tensor modes.

Scalar mode: Broadcasts scalarValue to the register and stores it in dstReg. (If mask is available, the value is stored in the position filtered by mask in dstReg.)
Tensor mode: Broadcasts the least significant element of srcReg to the register and stores it in the position filtered by mask in dstReg.

Prototype

Broadcast scalarValue to the register. mask is not supported.

template <typename T = DefaultType, typename U, typename S>
__simd_callee__ inline void Duplicate(S& dstReg, U scalarValue);

Broadcast scalarValue to the register. mask is supported.

template <typename T = DefaultType, MaskMergeMode mode = MaskMergeMode::ZEROING, typename U, typename S>
__simd_callee__ inline void Duplicate(S& dstReg, U scalarValue, MaskReg& mask);

Broadcast the least significant element of srcReg to the register.

template <typename T = DefaultType, HighLowPart pos = HighLowPart::LOWEST, MaskMergeMode mode = MaskMergeMode::ZEROING, typename S>
__simd_callee__ inline void Duplicate(S& dstReg, S& srcReg, MaskReg& mask)

Parameters

**Table 1** Template parameters
Parameter	Description
T	Operand data type. For the Atlas 350 Accelerator Card, the supported data types are bool, int8_t, uint8_t, fp4x2_e2m1_t, fp4x2_e1m2_t, hifloat8_t, fp8_e5m2_t, fp8_e4m3fn_t, fp8_e8m0_t, int16_t, uint16_t, half, bfloat16_t, int32_t, uint32_t, float, complex32, int64_t, uint64_t, and complex64.
U	Data type of a scalar. 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, complex32, int64_t, uint64_t, and complex64.
mode	Set it to MERGING or ZEROING. MERGING: The elements that are not filtered by mask keep their original values in dst. ZEROING: The elements that are not filtered by mask are set to zero in dst.
pos	The default value is LOWEST, indicating that the least significant bit in srcReg is broadcast to dstReg. (Other values are not supported currently.)
S	RegTensor type of the destination operand, for example, RegTensor<half>. It is automatically inferred by the compiler and does not need to be specified.

**Table 2** Function parameters
Parameter	Input/Output	Description
dstReg	Output	Destination operand. The type is RegTensor.
scalarValue	Input	Source operand. The type is scalar.
srcReg	Input	Source operand. The type is RegTensor.
mask	Input	Valid indication of the source operand element operation. For details, see MaskReg.

**Table 3** Data type constraints of source and destination operands
dstReg	Scalar/srcReg
bool	bool
int8_t	int8_t
uint8_t	uint8_t
fp4x2_e2m1_t	int8_t
fp4x2_e1m2_t	int8_t
hifloat8_t	int8_t
fp8_e5m2_t	int8_t
fp8_e4m3fn_t	int8_t
fp8_e8m0_t	int8_t
int16_t	int16_t
uint16_t	uint16_t
half	half
bfloat16_t	bfloat16_t
int32_t	int32_t
uint32_t	uint32_t
float	float
complex32	complex32
int64_t	int64_t
uint64_t	uint64_t
complex64	complex64

Constraints

None

Examples

Example 1

template<typename T>
__simd_vf__ inline void DuplicateVF(__ubuf__ T* dstAddr, T scalarValue, uint32_t oneRepeatSize, uint16_t repeatTimes)
{
    AscendC::Reg::RegTensor<T> dstReg;
    AscendC::Reg::MaskReg mask = AscendC::Reg::CreateMask<T>();    
    for (uint16_t i = 0; i < repeatTimes; i++) {
        AscendC::Reg::Duplicate(dstReg, scalarValue);
        AscendC::Reg::StoreAlign(dstAddr + i * oneRepeatSize, dstReg, mask);
    }
}

Example 2

template<typename T>
__simd_vf__ inline void DuplicateVF(__ubuf__ T* dstAddr, T scalarValue, uint32_t count, uint32_t oneRepeatSize, uint16_t repeatTimes)
{
    AscendC::Reg::RegTensor<T> dstReg;
    AscendC::Reg::MaskReg mask;    
    for (uint16_t i = 0; i < repeatTimes; i++) {
        mask = AscendC::Reg::UpdateMask<T>(count);
        AscendC::Reg::LoadAlign(srcReg, src0Addr + i * oneRepeatSize);
        AscendC::Reg::Duplicate(dstReg, scalarValue, mask);
        AscendC::Reg::StoreAlign(dstAddr + i * oneRepeatSize, dstReg, mask);
    }
}

Example 3

template<typename T>
__simd_vf__ inline void DuplicateVF(__ubuf__ T* dstAddr, __ubuf__ T* srcAddr, uint32_t count, uint32_t oneRepeatSize, uint16_t repeatTimes)
{
    AscendC::Reg::RegTensor<T> srcReg;
    AscendC::Reg::RegTensor<T> dstReg;
    AscendC::Reg::MaskReg mask;
    for (uint16_t i = 0; i < repeatTimes; i++) {
        mask = AscendC::Reg::UpdateMask<T>(count);
        AscendC::Reg::LoadAlign(srcReg, src0Addr + i * oneRepeatSize);
        AscendC::Reg::Duplicate(dstReg, srcReg, mask);
        AscendC::Reg::StoreAlign(dstAddr + i * oneRepeatSize, dstReg, mask);
    }
}

Parent topic: Data Padding