Axpy

Applicability

Product	Supported/Unsupported
Atlas A3 training products / Atlas A3 inference products	√
Atlas A2 training products / Atlas A2 inference products	√
Atlas 200I/500 A2 inference products	√
Atlas inference product 's AI Core	√
Atlas inference product 's Vector Core	x
Atlas training products	√

Functions

The product of each element in the source operand src and the scalar is added to the corresponding element in the destination operand dst. The calculation formula is as follows:

$\text{[math]}$

Prototype

Computation of the first n pieces of data of a tensor

        
             template <typename T, typename U>
__aicore__ inline void Axpy(const LocalTensor<T>& dst, const LocalTensor<U>& src, const U& scalarValue, const int32_t& count)

High-dimensional tensor sharding computation

Bitwise mask mode

          
               template <typename T, typename U, bool isSetMask = true>
__aicore__ inline void Axpy(const LocalTensor<T>& dst, const LocalTensor<U>& src, const U& scalarValue, uint64_t mask[], const uint8_t repeatTime, const UnaryRepeatParams& repeatParams)

Contiguous mask mode

          
               template <typename T, typename U, bool isSetMask = true>
__aicore__ inline void Axpy(const LocalTensor<T>& dst, const LocalTensor<U>& src, const U& scalarValue, uint64_t mask, const uint8_t repeatTime, const UnaryRepeatParams& repeatParams)

Parameters

**Table 1** Parameters in the template
Parameter	Description
T	Data type of the destination operand. For details about the data type constraints of the destination and source operands, see Table 3. For the Atlas training products , the supported data types are half and float. For the Atlas inference product 's AI Core, the supported data types are half and float. For the Atlas A2 training products / Atlas A2 inference products , the supported data types are half and float. For the Atlas A3 training products / Atlas A3 inference products , the supported data types are half and float. For the Atlas 200I/500 A2 inference products , the supported data types are half and float.
U	Data type of the source operand. For the Atlas training products , the supported data types are half and float. For the Atlas inference product 's AI Core, the supported data types are half and float. For the Atlas A2 training products / Atlas A2 inference products , the supported data types are half and float. For the Atlas A3 training products / Atlas A3 inference products , the supported data types are half and float. For the Atlas 200I/500 A2 inference products , the supported data types are half and float.
isSetMask	Indicates whether to set mask inside the API. true: sets mask inside the API. false: sets mask outside the API. Developers need to use the SetVectorMask API to set the mask value. In this mode, the mask value in the input parameter of this API must be set to the placeholder MASK_PLACEHOLDER.

**Table 2** Parameters
Parameter	Input/Output	Meaning
dst	Output	Destination operand. The type is LocalTensor, and the supported TPosition is VECIN, VECCALC, or VECOUT. The start address of the LocalTensor must be 32-byte aligned.
src	Input	Source operand. The type is LocalTensor, and the supported TPosition is VECIN, VECCALC, or VECOUT. The start address of the LocalTensor must be 32-byte aligned.
scalarValue	Input	Source operand (scalar). The data type of scalarValue must be the same as that of src.
count	Input	Number of elements involved in the computation.
mask/mask[]	Input	The mask parameter is used to control the elements involved in computation in each iteration. Bitwise mode: controls the elements that participate in computation by bit. If a bit is set to 1, the corresponding element participates in the computation. If a bit is set to 0, the corresponding element is masked in the computation. The mask is in array form. The array length and the value range of the array elements are related to the data type of the operand. When the operand is 16-bit, the array length is 2. In this case, mask[0] and mask[1] must be in the range of [0, 2⁶⁴ – 1] and cannot be 0 at the same time. When the operand is 32-bit, the array length is 1. In this case, mask[0] must be in the range of (0, 2⁶⁴ – 1]. When the operand is 64-bit, the array length is 1. In this case, mask[0] must be in the range of (0, 2³² – 1]. For example, if mask = [0, 8] and 8 = 0b1000, only the fourth element participates in computation. Contiguous mode: indicates the number of contiguous elements that participate in computation. The value range is related to the operand data type. The maximum number of elements that can be processed in each repeat varies according to the data type. When the operand is 16-bit, mask ∈ [1, 128]. When the operand is 32-bit, mask ∈ [1, 64]. When the operand is 64-bit, mask ∈ [1, 32].
repeatTime	Input	Number of iteration repeats. The Vector Unit reads 256 bytes of contiguous data for computation each time. To read the complete data for processing, the unit needs to read the input data in multiple repeats. repeatTime indicates the number of iterations. For details about this parameter, see High-dimensional Sharding APIs.
repeatParams	Input	Parameters that control the operand address strides. They are of the UnaryRepeatParams type, and contain such parameters as those that specify the address stride of the operand for the same data block between adjacent iterations and address stride of the operand between different data blocks in a single iteration. For details about the address stride parameters between adjacent iterations, see repeatStride. For details about the address stride parameters of DataBlock in the same iteration, see dataBlockStride.

**Table 3** Data type constraints:
src Data Type	scalar Data Type	dst Data Type	PAR	Availability
half	half	half	128	Atlas training products Atlas A2 training products / Atlas A2 inference products Atlas A3 training products / Atlas A3 inference products Atlas inference product 's AI Core Atlas 200I/500 A2 inference products
float	float	float	64	Atlas training products Atlas A2 training products / Atlas A2 inference products Atlas A3 training products / Atlas A3 inference products Atlas inference product 's AI Core Atlas 200I/500 A2 inference products
half	half	float	64	Atlas training products Atlas A2 training products / Atlas A2 inference products Atlas A3 training products / Atlas A3 inference products Atlas inference product 's AI Core Atlas 200I/500 A2 inference products

Returns

None

Constraints

For details about the operand address alignment requirements, see General Address Alignment Restrictions.
For details about the operand address overlapping restrictions, see General Address Overlap Restrictions.

When a high-dimensional tensor sharding computation API is used, if the data type of srcLocal and scalar is half and that of dstLocal is float, the number of source operand elements processed in one iteration must be the same as that of destination operand elements. Therefore, the first four data blocks are selected for computation in each iteration. This restriction needs to be considered when you set the repeatStride parameter, mask parameter, and address overlapping.

Examples

This example shows only part of the code used in the computation process (Compute). To run the sample code, copy the code snippet and replace parts of code of the Compute function in the complete sample template in More Examples.

Example of high-dimensional tensor sharding computation (contiguous mask mode)

        
             // repeatTime = 4, mask = 128, 128 elements one repeat, 512 elements total
// The data types of srcLocal, scalar, and dstLocal are half.
// dstBlkStride, srcBlkStride = 1, no gap between blocks in one repeat
// dstRepStride, srcRepStride = 8, no gap between repeats 
AscendC::Axpy(dstLocal, srcLocal, (half)2.0, 128, 4,{ 1, 1, 8, 8 });

// The data types of srcLocal, scalar, and dstLocal are half and float, respectively.
// repeatTime = 8, mask = 64, 64 elements one repeat, 512 elements total
// dstBlkStride, srcBlkStride = 1, no gap between blocks in one repeat
// dstRepStride = 8, srcRepStride = 4, no gap between repeats 
AscendC::Axpy(dstLocal, srcLocal, (half)2.0, 64, 8,{ 1, 1, 8, 4 }); // Select the first four data blocks of the source operand for computation in each iteration.

Example of high-dimensional tensor sharding computation (bitwise mask mode)

        
             uint64_t mask[2] = { 0xFFFFFFFFFFFFFFFF, 0xFFFFFFFFFFFFFFFF };
// repeatTime = 4, 128 elements per repeat, 512 elements in total, half-precision combination
// dstBlkStride, srcBlkStride = 1, no gap between blocks in one repeat
// dstRepStride, srcRepStride = 8, no gap between repeats
AscendC::Axpy(dstLocal, srcLocal, (half)2.0, mask, 4,{ 1, 1, 8, 8 });

Example of computing the first n pieces of data of a tensor

        
             AscendC::Axpy(dstLocal, src0Local, (half)2.0, 512);// half type

More Examples

Complete example 1: The data type of srcLocal, scalar, and dstLocal is half.

        
         
           
           
             #include "kernel_operator.h"
class KernelAxpy {
public:
    __aicore__ inline KernelAxpy() {}
    __aicore__ inline void Init(__gm__ uint8_t* srcGm, __gm__ uint8_t* dstGm)
    {
        srcGlobal.SetGlobalBuffer((__gm__ half*)srcGm);
        dstGlobal.SetGlobalBuffer((__gm__ half*)dstGm);
        pipe.InitBuffer(inQueueSrc, 1, 512 * sizeof(half));
        pipe.InitBuffer(outQueueDst, 1, 512 * sizeof(half));
    }
    __aicore__ inline void Process()
    {
        CopyIn();
        Compute();
        CopyOut();
    }
private:
    __aicore__ inline void CopyIn()
    {
        AscendC::LocalTensor<half> srcLocal = inQueueSrc.AllocTensor<half>();
        AscendC::DataCopy(srcLocal, srcGlobal, 512);
        inQueueSrc.EnQue(srcLocal);
    }
    __aicore__ inline void Compute()
    {
        AscendC::LocalTensor<half> srcLocal = inQueueSrc.DeQue<half>();
        AscendC::LocalTensor<half> dstLocal = outQueueDst.AllocTensor<half>();
 
        AscendC::Duplicate(dstLocal, (half)0.0, 512);
        AscendC::Axpy(dstLocal, srcLocal, (half)2.0, 512);
 
        outQueueDst.EnQue<half>(dstLocal);
        inQueueSrc.FreeTensor(srcLocal);
    }
    __aicore__ inline void CopyOut()
    {
        AscendC::LocalTensor<half> dstLocal = outQueueDst.DeQue<half>();
        AscendC::DataCopy(dstGlobal, dstLocal, 512);
        outQueueDst.FreeTensor(dstLocal);
    }
private:
    AscendC::TPipe pipe;
    AscendC::TQue<AscendC::TPosition::VECIN, 1> inQueueSrc;
    AscendC::TQue<AscendC::TPosition::VECOUT, 1> outQueueDst;
    AscendC::GlobalTensor<half> srcGlobal, dstGlobal;
};
extern "C" __global__ __aicore__ void kernel_vec_ternary_scalar_Axpy_half_2_half(__gm__ uint8_t* srcGm, __gm__ uint8_t* dstGm)
{
    KernelAxpy op;
    op.Init(srcGm, dstGm);
    op.Process();
}

            

          

        
       

Result example:

Input (srcGm):
[1. 1. 1. 1. 1. 1. ... 1.]
Output (dstGm):
[2. 2. 2. 2. 2. 2. ... 2.]

Complete example 2: The data type of srcLocal and scalar is half and that of dstLocal is float.

        
         
           
           
             #include "kernel_operator.h"
class KernelAxpy {
public:
    __aicore__ inline KernelAxpy() {}
    __aicore__ inline void Init(__gm__ uint8_t* srcGm, __gm__ uint8_t* dstGm)
    {
        srcGlobal.SetGlobalBuffer((__gm__ half*)srcGm);
        dstGlobal.SetGlobalBuffer((__gm__ float*)dstGm);
        pipe.InitBuffer(outQueueDst, 1, 512 * sizeof(float));
        pipe.InitBuffer(inQueueSrc, 1, 512 * sizeof(half));
    }
    __aicore__ inline void Process()
    {
        CopyIn();
        Compute();
        CopyOut();
    }
private:
    __aicore__ inline void CopyIn()
    {
        AscendC::LocalTensor<half> srcLocal = inQueueSrc.AllocTensor<half>();
        AscendC::DataCopy(srcLocal, srcGlobal, 512);
        inQueueSrc.EnQue(srcLocal);
    }
    __aicore__ inline void Compute()
    {
        AscendC::LocalTensor<half> srcLocal = inQueueSrc.DeQue<half>();
        AscendC::LocalTensor<float> dstLocal = outQueueDst.AllocTensor<float>();
 
        AscendC::Duplicate(dstLocal, 0.0f, 512);
        AscendC::Axpy(dstLocal, srcLocal, (half)2.0, 64, 8, { 1, 1, 8, 4 });
 
        outQueueDst.EnQue<float>(dstLocal);
        inQueueSrc.FreeTensor(srcLocal);
    }
    __aicore__ inline void CopyOut()
    {
        AscendC::LocalTensor<float> dstLocal = outQueueDst.DeQue<float>();
        AscendC::DataCopy(dstGlobal, dstLocal, 512);
        outQueueDst.FreeTensor(dstLocal);
    }
private:
    AscendC::TPipe pipe;
    AscendC::TQue<AscendC::TPosition::VECIN, 1> inQueueSrc;
    AscendC::TQue<AscendC::TPosition::VECOUT, 1> outQueueDst;
    AscendC::GlobalTensor<half> srcGlobal;
    AscendC::GlobalTensor<float> dstGlobal;
};
extern "C" __global__ __aicore__ void kernel_vec_ternary_scalar_Axpy_half_2_float(__gm__ uint8_t* srcGm, __gm__ uint8_t* dstGm)
{
    KernelAxpy op;
    op.Init(srcGm, dstGm);
    op.Process();
}

            

          

        
       

Result example:

Input (srcGm):
[1. 1. 1. 1. 1. 1. ... 1.]
Output (dstGm):
[2. 2. 2. 2. 2. 2. ... 2.]

Parent topic: Complex computation