Compare (Results Stored in a Register)

Function Usage

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 result is stored in a register.

The following comparison modes are supported:

LT: lower than
GT: greater than

GE: greater than or equal to
EQ: equal
NE: not equal to
LE: lower than or equal to

Prototype

Bitwise mask mode

        
             template <typename T, bool isSetMask = true>
__aicore__ inline void Compare(const LocalTensor<T>& src0Local, const LocalTensor<T>& src1Local, 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>& src0Local, const LocalTensor<T>& src1Local, 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. The value can be half or 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 MASK_PLACEHOLDER.

**Table 2** API parameters
Parameter	Input/Output	Meaning
src0Local and src1Local	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 ≥ src1 EQ: src0 = src1 NE: src0 ≠ src1 LE: src0 ≤ src1
mask	Input	mask is used to control the elements that participate in computation in each iteration. 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 iteration 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]. 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 parameter type is a uint64_t array whose length is 2. For example, if mask = [0, 8] and 8 = 0b1000, only the fourth element participates in computation. The parameter value range is related to the operand data type. The maximum number of elements that can be processed in each iteration varies according to the data type. When the operand is 16-bit, mask[0] and mask[1] ∈ [0, 2⁶⁴ -1] and cannot be 0 at the same time. When the operand is 32-bit, mask[1] is 0 and mask[0] ∈ (0, 2⁶⁴ – 1]. When the operand is 64-bit, mask[1] is 0 and mask[0] ∈ (0, 2³² – 1].
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 of the operand between adjacent iterations, see repeatStride. For details about the address stride of the operand between different data blocks in a single iteration, see dataBlockStride.

Returns

None

Availability

Precautions

For details about the alignment requirements of the operand address offset, see General Restrictions.

This instruction does not have repeat input. The default value of repeat is 1, that is, 256-byte data is calculated in an instruction.
This command writes the result to the 128-bit cmpMask register. The data saved in the register can be obtained by using the GetCmpMask API.

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::QuePosition::VECIN, 1> inQueueSrc0, inQueueSrc1;
    AscendC::TQue<AscendC::QuePosition::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 Instruction