LeakyRelu

Applicability

Product	Supported
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

Performs element-wise leaky rectified linear unit (Leaky ReLU) operation. The formula is as follows:

$\text{[math]}$

Leaky ReLU is a commonly used activation function in artificial neural networks. Its mathematical expression is as follows:

As shown in the following figures, the difference between ReLU and Leaky ReLU is that ReLU sets all negative values to zero, while Leaky ReLU applies a slope to negative values.

Prototype

Computation of the first n data elements of a tensor

        
             template <typename T, bool isSetMask = true>
__aicore__ inline void LeakyRelu(const LocalTensor<T>& dst, const LocalTensor<T>& src, const T& scalarValue, const int32_t& count)

High-dimensional tensor sharding computation

Bitwise mask mode

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

Contiguous mask mode

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

When dst and src use the TensorTrait type, the data type of scalarValue (corresponding to the LiteType in TensorTrait) may differ from that of TensorTrait. To address this, the template type U is introduced to represent the data type of scalarValue. std::enable_if is used to check whether LiteType extracted from T is the same as U. If they match, the API compiles successfully; otherwise, the compilation fails. The API prototype is defined as follows:

Computation of the first n data elements of a tensor

        
             template <typename T, typename U, bool isSetMask = true, typename Std::enable_if<Std::is_same<PrimT<T>, U>::value, bool>::type = true>
__aicore__ inline void LeakyRelu(const LocalTensor<T>& dst, const LocalTensor<T>& src, const U& scalarValue, const int32_t& count)

High-dimensional tensor sharding computation

Bitwise mask mode

          
               template <typename T, typename U, bool isSetMask = true, typename Std::enable_if<Std::is_same<PrimT<T>, U>::value, bool>::type = true>
__aicore__ inline void LeakyRelu(const LocalTensor<T>& dst, const LocalTensor<T>& 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, typename Std::enable_if<Std::is_same<PrimT<T>, U>::value, bool>::type = true>
__aicore__ inline void LeakyRelu(const LocalTensor<T>& dst, const LocalTensor<T>& src, const U& scalarValue, uint64_t mask, const uint8_t repeatTime, const UnaryRepeatParams& repeatParams)

Parameters

**Table 1** Template parameters
Parameter	Description
T	Operand data type. For the Atlas inference product 's AI Core, the supported data types are half and float. For Atlas A2 training products / Atlas A2 inference products , the supported data types are half and float. For Atlas A3 training products / Atlas A3 inference products , the supported data types are half and float. For Atlas 200I/500 A2 inference products , the supported data types are half and float.
U	Data type of scalarValue. For the Atlas inference product 's AI Core, the supported data types are half and float. For Atlas A2 training products / Atlas A2 inference products , the supported data types are half and float. For Atlas A3 training products / Atlas A3 inference products , the supported data types are half and float. For Atlas 200I/500 A2 inference products , the supported data types are half and float.
isSetMask	Whether to set the mask mode and mask value inside the API. true: Set inside the API. The APIs for high-dimensional tensor sharding and the APIs for computing the first n data elements of a tensor internally use the normal mode or counter mode of the mask. In general, keep isSetMask as the default value, which allows the API to automatically set the mask mode and mask value based on the mask or count parameter passed in by developers. false: Set outside the API. For high-dimensional tensor sharding computation APIs, in scenarios with high performance requirements, you need to use SetMaskNorm/SetMaskCount to set the mask mode and use SetVectorMask to set the mask value. The mask input parameter of this API must be set to MASK_PLACEHOLDER. For APIs that compute the first n data elements of a tensor, in scenarios with high performance requirements, you need to set the mask mode to counter mode using SetMaskCount and set the mask value using SetVectorMask. The count parameter in this API does not take effect. You are advised to set it to 1. For the following models, the isSetMask parameter in the API for computing the first n data elements of a tensor does not take effect. Keep it as the default value. Atlas 200I/500 A2 inference products

**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. Its data type must match that of the destination operand.
scalarValue	Input	Source operand, and its data type must be the same as the element type of the tensor in the destination operand.
count	Input	Number of elements involved in the computation.
mask/mask[]	Input	mask is used to control the elements that participate in computation in each iteration. Bitwise mode: controls which elements are involved in computation bit by bit. A bit value of 1 means the corresponding element participates in computation, while 0 means it does not. The mask value is an array. The array length and the value range of the array elements are related to the operand data type. When the operand is 16-bit, the array length is 2, mask[0] and mask[1] ∈ [0, 2⁶⁴ -1] and cannot be 0 at the same time. When the operand is 32-bit, the array length is 1 and mask[0] ∈ (0, 2⁶⁴ – 1]. When the operand is 64-bit, the array length is 1 and mask[0] ∈ (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 repeat iterations. The vector compute unit reads 256 bytes of contiguous data for computation each time. To process the input data, the data needs to be read and computed over multiple repeats. repeatTime indicates the number of repeats. For details about this parameter, see High-dimensional Sharding APIs.
repeatParams	Input	Structure for controlling element-wise operations. For details, see UnaryRepeatParams.

Returns

None

Restrictions

For details about the operand address alignment requirements, see General Address Alignment Restrictions.
For details about the constraints on operand address overlapping, see General Address Overlapping Restrictions.

Example

For more examples, see LINK.

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

        
         
           
           
             #include "kernel_operator.h"
class KernelBinaryScalar {
public:
    __aicore__ inline KernelBinaryScalar() {}
    __aicore__ inline void Init(__gm__ uint8_t* src, __gm__ uint8_t* dstGm)
    {
        srcGlobal.SetGlobalBuffer((__gm__ half*)src);
        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>();

       uint64_t mask = 128;
        half scalar = 2;
        // repeatTime = 4, 128 elements one repeat, 512 elements total
        // dstBlkStride, srcBlkStride = 1, no gap between blocks in one repeat
        // dstRepStride, srcRepStride =8, no gap between repeats
        AscendC::LeakyRelu(dstLocal, srcLocal, scalar, mask, 4, {1, 1, 8, 8});

        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 binary_scalar_simple_kernel(__gm__ uint8_t* src, __gm__ uint8_t* dstGm)
{
    KernelBinaryScalar op;
    op.Init(src, dstGm);
    op.Process();
}

            

          

        
       

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

        
             uint64_t mask[2] = { UINT64_MAX, UINT64_MAX };
half scalar = 2;
// repeatTime = 4. 128 elements are processed in a single iteration. To compute 512 elements, four iterations are required.
// dstBlkStride, srcBlkStride = 1. The interval between addresses of src0 data involved in computation within each iteration is one data block, indicating that data is read and written continuously within a single iteration.
// dstRepStride, srcRepStride = 8. The address interval between adjacent iterations is eight data blocks, indicating that data is read and written continuously between adjacent iterations.
AscendC::LeakyRelu(dstLocal, srcLocal, scalar, mask, 4, {1, 1, 8, 8});

Example of computing the first n data elements of a tensor

        
             half scalar = 2;
AscendC::LeakyRelu(dstLocal, srcLocal, scalar, 512);

Result example:

Input (src0Local): [1. 2. 3. ... 512.]
Input (scalar): 2
Output (dstLocal): [1. 2. 3. ... 512.]

Parent topic: Basic Arithmetic