CreateVecIndex

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	x

Function Usage

Creates a vector index with a specified start value.

Prototype

Computation of the first n data elements of a tensor

        
             template <typename T>
__aicore__ inline void CreateVecIndex(LocalTensor<T> dst, const T &firstValue, uint32_t count)

High-dimensional tensor sharding computation

Bitwise mask mode

          
               template <typename T>
__aicore__ inline void CreateVecIndex(LocalTensor<T> &dst, const T &firstValue, uint64_t mask[], uint8_t repeatTime, uint16_t dstBlkStride, uint8_t dstRepStride)

Contiguous mask mode

          
               template <typename T>
__aicore__ inline void CreateVecIndex(LocalTensor<T> &dst, const T &firstValue, uint64_t mask, uint8_t repeatTime, uint16_t dstBlkStride, uint8_t dstRepStride)

Parameters

**Table 1** Parameters in the template
Parameter	Description
T	Operand data type. For the Atlas inference product 's AI Core, the supported data types are int16_t, half, int32_t, and float. For the Atlas A2 training products / Atlas A2 inference products , the supported data types are int16_t, half, int32_t, and float. For the Atlas A3 training products / Atlas A3 inference products , the supported data types are int16_t, half, int32_t, and float. For the Atlas 200I/500 A2 inference products , the supported data types are int16_t, half, int32_t, and float.

**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.
firstValue	Input	The first value of the index. The data type must be the same as that of the element in dstLocal.
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 repeats. For details about this parameter, see Common Parameters.
dstBlkStride	Input	Address stride of the destination operand between different data blocks in a single repeat
dstRepStride	Input	Address stride of the destination operand for the same data block between adjacent repeats

Returns

None

Restrictions

For details about the operand address alignment requirements, see General Address Alignment Restrictions.
Ensure that the value of firstValue is within the value range corresponding to the data type of the elements in dst.

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 related code snippets of the Compute function in Template Samples.

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

        
             uint64_t mask = 128;
// repeatTime = 1
// dstBlkStride = 1. Data is continuously written in a single repeat.
// dstRepStride = 8, continuous data write between adjacent repeats.
AscendC::CreateVecIndex(dstLocal, (T)0, mask, repeatTime, dstBlkStride, dstRepStride);

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

        
             uint64_t mask[2] = { UINT64_MAX, UINT64_MAX };
// repeatTime = 1
// dstBlkStride = 1. Data is continuously written in a single repeat.
// dstRepStride = 8, continuous data write between adjacent repeats.
AscendC::CreateVecIndex(dstLocal, (T)0, mask, repeatTime, dstBlkStride, dstRepStride);

Example of computing the first n data elements of a tensor

        
             AscendC::CreateVecIndex(dstLocal, (T)0, 128);

Result example:

Input data (firstValue): 0
Output data (dstLocal): [0 1 2... 127]

Template Samples

      
       
         
         
           #include "kernel_operator.h"
template <typename T>
class CreateVecIndexTest {
public:
    __aicore__ inline CreateVecIndexTest() {}
    __aicore__ inline void Init(GM_ADDR dstGm, uint64_t mask, uint8_t repeatTime,
        uint16_t dstBlkStride, uint8_t dstRepStride)
    {
        m_mask = mask;
        m_repeatTime = repeatTime;
        m_dstBlkStride = dstBlkStride;
        m_dstRepStride = dstRepStride;
        m_elementCount = m_dstBlkStride * m_dstRepStride * 32 * m_repeatTime / sizeof(T);
        m_dstGlobal.SetGlobalBuffer((__gm__ T*)dstGm);
        m_pipe.InitBuffer(m_queOut, 1, m_dstBlkStride * m_dstRepStride * 32 * m_repeatTime);
        m_pipe.InitBuffer(m_queTmp, 1, 1024);
    }
    __aicore__ inline void Process()
    {
        CopyIn();
        Compute();
        CopyOut();
    }
private:
    __aicore__ inline void CopyIn()
    {
        ;
    }
    __aicore__ inline void Compute()
    {
        AscendC::LocalTensor<T> dstLocal = m_queOut.AllocTensor<T>();
        AscendC::LocalTensor<uint8_t> tmpLocal = m_queTmp.AllocTensor<uint8_t>();
        AscendC::Duplicate(dstLocal, (T)0, m_elementCount);
        AscendC::PipeBarrier<PIPE_ALL>();
        AscendC::CreateVecIndex(dstLocal, (T)0, m_repeatTime * 256 / sizeof(T));
        m_queOut.EnQue(dstLocal);
        m_queTmp.FreeTensor(tmpLocal);
    }
    __aicore__ inline void CopyOut()
    {
        AscendC::LocalTensor<T> dstLocal = m_queOut.DeQue<T>();

        AscendC::DataCopy(m_dstGlobal, dstLocal, m_elementCount);
        m_queOut.FreeTensor(dstLocal);
    }
private:
    AscendC::TPipe m_pipe;
    uint32_t m_elementCount;
    uint32_t m_mask;
    uint32_t m_repeatTime;
    uint32_t m_dstBlkStride;
    uint32_t m_dstRepStride;
    AscendC::GlobalTensor<T> m_dstGlobal;
    AscendC::TQue<AscendC::TPosition::VECOUT, 1> m_queOut;
    AscendC::TQue<AscendC::TPosition::VECIN, 1> m_queTmp;
}; // class CreateVecIndexTest
template <typename T>
__global__ __aicore__ void testCreateVecIndex(GM_ADDR dstGm, uint64_t mask, uint8_t repeatTime,
        uint16_t dstBlkStride, uint8_t dstRepStride)
{
    CreateVecIndexTest<T> op;
    op.Init(dstGm, mask, repeatTime, dstBlkStride, dstRepStride);
    op.Process();
}

          

        

      
     

Parent topic: Data Padding