Compare (Results Stored in a Register)

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	x
Atlas inference product 's AI Core	√
Atlas inference product 's Vector Core	x
Atlas training products	x

Functions

Compares the sizes of two tensors element by element. If the comparison result is true, the corresponding bit of the output result is 1. Otherwise, the bit is 0. This interface can be used when the mask parameter is required. The computation result is stored in the register.

The following comparison modes are supported:

LT: less than
GT: greater than

GE: greater than or equal to (greater than or equal to)
EQ: equal to
NE: not equal to
LE: less than or equal to (less than or equal to)

Prototype

Bitwise mask mode

        
             template <typename T, bool isSetMask = true>
__aicore__ inline void Compare(const LocalTensor<T>& src0, const LocalTensor<T>& src1, CMPMODE cmpMode, const uint64_t mask[], const BinaryRepeatParams& repeatParams)

Contiguous mask mode

        
             template <typename T, bool isSetMask = true>
__aicore__ inline void Compare(const LocalTensor<T>& src0, const LocalTensor<T>& src1, CMPMODE cmpMode, const uint64_t mask, const BinaryRepeatParams& repeatParams)

Parameters

**Table 1** Parameters in the template
Parameter	Description
T	Data type of the source operand. 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 inference product 's AI Core, 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** API parameters
Parameter	Input/Output	Meaning
src0, src1	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.
cmpMode	Input	Comparison mode, including EQ, NE, GE, LE, GT, and LT. LT: src0 < src1 GT: src0 > src1 GE: src0 is greater than or equal to (greater than or equal to) src1. EQ: src0 is equal to src1. NE: src0 is not equal to src1. LE: src0 is less than or equal to (less than or equal to) src1.
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].
repeatParams	Input	Parameters that control the operand address strides. They are of the BinaryRepeatParams 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.

Returns

None

Restrictions

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

This API does not have the repeat input. The default repeat value is 1, that is, 256 bytes of data are computed by one instruction.
This API writes the result to the 128-bit cmpMask register. You can use the GetCmpMask API to obtain the data stored in the register.

Examples

In this example, the source operands src0Local and src1Local both store 64 pieces of data of the float type. The example compares the data in src0Local and src1Local element by element. If the element in src0Local is smaller than that in src1Local, the corresponding bit in the dstLocal result is set to 1. Otherwise, the bit is set to 0. The dstLocal result is stored in uint8_t format.

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 some code of the Compute function in Template Sample.

Contiguous mask mode

        
             uint64_t mask = 256 / sizeof(float); // 256 indicates the number of bytes processed in each iteration.
AscendC::BinaryRepeatParams repeatParams = { 1, 1, 1, 8, 8, 8 };
// dstBlkStride, src0BlkStride, src1BlkStride = 1, no gap between blocks in one repeat
// dstRepStride, src0RepStride, src1RepStride = 8, no gap between repeats
AscendC::Compare(src0Local, src1Local, AscendC::CMPMODE::LT, mask, repeatParams);

Bitwise mask mode

        
             uint64_t mask[2] = { UINT64_MAX, 0};
AscendC::BinaryRepeatParams repeatParams = { 1, 1, 1, 8, 8, 8 };
// srcBlkStride, = 1, no gap between blocks in one repeat
// dstRepStride, srcRepStride = 8, no gap between repeats
AscendC::Compare(src0Local, src1Local, AscendC::CMPMODE::LT, mask, repeatParams);

Result example:

Input (src0_gm):
[ 86.72287     9.413112   17.033222  -64.10005   -66.2691    -65.57659
  15.898049   94.61241   -68.920685  -36.16883    15.62852    68.078514
 -59.724575   -9.4302225 -64.770935   66.55523   -84.60122    57.331
  60.42026   -86.78856    37.25265     8.356797  -48.544407   16.73616
  15.28083   -21.889254  -67.93181   -41.01825   -68.79465    20.169441
  44.11346   -27.419518   30.452742  -89.30283   -18.590672   32.45831
   8.392082  -57.198048   98.76846   -81.73067   -38.274437  -83.84363
  64.30617     6.028703  -20.77164    93.71867    54.190437   94.98172
 -47.447758  -65.77461    82.21715    59.953922   23.599781  -77.29708
  26.963976  -63.468987   79.97712   -70.47842    39.00433    52.36555
 -63.94925   -65.77033    26.17237   -71.904884 ]
Input (src1_gm):
[  2.2989323  51.8879    -81.49718    41.189415    6.4081917  92.566666
  53.205498  -94.47063   -75.38387    36.464787   85.60772   -28.70681
  42.58504   -76.15293    38.723816   10.006577   74.53035   -78.38537
  71.945404   -4.060528  -14.501523   28.229202   96.87876    41.558033
 -92.623215   43.318684   35.387154  -16.029816   61.544827    3.3527017
  55.806778  -93.242096   22.86275   -87.506584   35.29523     8.405956
  91.03445   -85.29485    34.30078    -3.8019252  93.40503    15.459968
 -57.99712   -74.39948   -59.900818  -43.132637  -13.123036   41.246174
 -93.01083    75.476875  -45.437893  -99.19293    13.543604   76.23386
  46.192528  -39.23934    75.9787    -38.38979     9.807722  -60.610104
 -23.062874   48.1669     89.913376   73.78631  ]
Output (dst_gm):
[122  86 237  94 150   3 226 242]

Template Sample

      
       
         
         
           #include "kernel_operator.h"
template <typename T> class KernelCmpCmpmask {
public:
    __aicore__ inline KernelCmpCmpmask() {}
    __aicore__ inline void Init(__gm__ uint8_t* src0Gm, __gm__ uint8_t* src1Gm, __gm__ uint8_t* dstGm,
        uint32_t dataSize, AscendC::CMPMODE mode)
    {
        srcDataSize = dataSize;
        dstDataSize = 32;
        cmpMode = mode;
        src0Global.SetGlobalBuffer((__gm__ T*)src0Gm);
        src1Global.SetGlobalBuffer((__gm__ T*)src1Gm);
        dstGlobal.SetGlobalBuffer((__gm__ uint8_t*)dstGm);
        pipe.InitBuffer(inQueueSrc0, 1, srcDataSize * sizeof(T));
        pipe.InitBuffer(inQueueSrc1, 1, srcDataSize * sizeof(T));
        pipe.InitBuffer(outQueueDst, 1, dstDataSize * sizeof(uint8_t));
    }
    __aicore__ inline void Process()
    {
        CopyIn();
        Compute();
        CopyOut();
    }
private:
    __aicore__ inline void CopyIn()
    {
        AscendC::LocalTensor<T> src0Local = inQueueSrc0.AllocTensor<T>();
        AscendC::LocalTensor<T> src1Local = inQueueSrc1.AllocTensor<T>();
        AscendC::DataCopy(src0Local, src0Global, srcDataSize);
        AscendC::DataCopy(src1Local, src1Global, srcDataSize);
        inQueueSrc0.EnQue(src0Local);
        inQueueSrc1.EnQue(src1Local);
    }
    __aicore__ inline void Compute()
    {
        AscendC::LocalTensor<T> src0Local = inQueueSrc0.DeQue<T>();
        AscendC::LocalTensor<T> src1Local = inQueueSrc1.DeQue<T>();
        AscendC::LocalTensor<uint8_t> dstLocal = outQueueDst.AllocTensor<uint8_t>();
        AscendC::Duplicate(dstLocal.ReinterpretCast<float>(), static_cast<float>(0), 8);
        AscendC::BinaryRepeatParams repeatParams;
        uint32_t mask = 256 / sizeof(T);
        AscendC::Compare(src0Local, src1Local, cmpMode, mask, repeatParams);
        AscendC::PipeBarrier<PIPE_V>();
        AscendC::GetCmpMask(dstLocal);
        outQueueDst.EnQue<uint8_t>(dstLocal);
        inQueueSrc0.FreeTensor(src0Local);
        inQueueSrc1.FreeTensor(src1Local);
    }
    __aicore__ inline void CopyOut()
    {
        AscendC::LocalTensor<uint8_t> dstLocal = outQueueDst.DeQue<uint8_t>();
        AscendC::DataCopy(dstGlobal, dstLocal, dstDataSize);
        outQueueDst.FreeTensor(dstLocal);
    }
private:
    AscendC::TPipe pipe;
    AscendC::TQue<AscendC::TPosition::VECIN, 1> inQueueSrc0, inQueueSrc1;
    AscendC::TQue<AscendC::TPosition::VECOUT, 1> outQueueDst;
    AscendC::GlobalTensor<T> src0Global, src1Global;
    AscendC::GlobalTensor<uint8_t> dstGlobal;
    uint32_t srcDataSize = 0;
    uint32_t dstDataSize = 0;
    AscendC::CMPMODE cmpMode;
};
template <typename T>
__aicore__ void main_cpu_cmp_cmpmask_demo(__gm__ uint8_t* src0Gm, __gm__ uint8_t* src1Gm, __gm__ uint8_t* dstGm, uint32_t dataSize, AscendC::CMPMODE mode)
{
    KernelCmpCmpmask<T> op;
    op.Init(src0Gm, src1Gm, dstGm, dataSize, mode);
    op.Process();
}
extern "C" __global__ __aicore__ void kernel_vec_compare_cmpmask_64_LT_float(GM_ADDR src0_gm, GM_ADDR src1_gm, GM_ADDR dst_gm)
{
    main_cpu_cmp_cmpmask_demo<float>(src0_gm, src1_gm, dst_gm, 64, AscendC::CMPMODE::LT);
}

          

        

      
     

Parent topic: Comparison and selection