文档
注册
评分
提单
论坛
小AI

Copy

函数功能

VECIN, VECCALC, VECOUT之间的搬运指令,支持mask操作和datablock间隔操作。

函数原型

  • mask逐bit模式
    1
    2
    template <typename T, bool isSetMask = true>
    __aicore__ inline void Copy(const LocalTensor<T>& dstLocal, const LocalTensor<T>& srcLocal, const uint64_t mask[2], const uint8_t repeatTimes, const CopyRepeatParams& repeatParams)
    
  • mask连续模式
    1
    2
    template <typename T, bool isSetMask = true>
    __aicore__ inline void Copy(const LocalTensor<T>& dstLocal, const LocalTensor<T>& srcLocal, const uint64_t mask, const uint8_t repeatTimes, const CopyRepeatParams& repeatParams)
    

参数说明

表1 模板参数说明

参数名

描述

T

操作数数据类型。

isSetMask

是否在接口内部设置mask。

  • true,表示在接口内部设置mask。
  • false,表示在接口外部设置mask,开发者需要使用SetVectorMask接口设置mask值。这种模式下,本接口入参中的mask值必须设置为MASK_PLACEHOLDER。
表2 参数说明

参数名

输入/输出

描述

dstLocal

输出

目的操作数。

类型为LocalTensor,支持的TPosition为VECIN/VECCALC/VECOUT。起始地址需要保证32字节对齐。

Atlas A2训练系列产品/Atlas 800I A2推理产品,支持的数据类型为:uint16_t/int16_t/half/uint32_t/int32_t/float

Atlas 200/500 A2推理产品,支持的数据类型为:uint16_t/int16_t/bfloat16_t/half/uint32_t/int32_t/float

srcLocal

输入

源操作数。

类型为LocalTensor,支持的TPosition为VECIN/VECCALC/VECOUT。起始地址需要保证32字节对齐。

源操作数的数据类型需要与目的操作数保持一致。

Atlas A2训练系列产品/Atlas 800I A2推理产品,支持的数据类型为:uint16_t/int16_t/half/uint32_t/int32_t/float

Atlas 200/500 A2推理产品,支持的数据类型为:uint16_t/int16_t/bfloat16_t/half/uint32_t/int32_t/float

mask

输入

mask用于控制每次迭代内参与计算的元素。

  • 连续模式:表示前面连续的多少个元素参与计算。取值范围和操作数的数据类型有关,数据类型不同,每次迭代内能够处理的元素个数最大值不同。当操作数为16位时,mask∈[1, 128];当操作数为32位时,mask∈[1, 64];当操作数为64位时,mask∈[1, 32]。
  • 逐bit模式:可以按位控制哪些元素参与计算,bit位的值为1表示参与计算,0表示不参与。参数类型为长度为2的uint64_t类型数组。

    例如,mask=[8, 0],8=0b1000,表示仅第4个元素参与计算。

    参数取值范围和操作数的数据类型有关,数据类型不同,每次迭代内能够处理的元素个数最大值不同。当操作数为16位时,mask[0]、mask[1]∈[0, 264-1]并且不同时为0;当操作数为32位时,mask[1]为0,mask[0]∈(0, 264-1];当操作数为64位时,mask[1]为0,mask[0]∈(0, 232-1]。

repeatTimes

输入

重复迭代次数。矢量计算单元,每次读取连续的256 Bytes数据进行计算,为完成对输入数据的处理,必须通过多次迭代(repeat)才能完成所有数据的读取与计算。repeatTimes表示迭代的次数。

关于该参数的具体描述请参考通用参数说明

CopyRepeatParams

输入

控制操作数地址步长的数据结构。CopyRepeatParams类型,定义如下:

1
2
3
4
5
6
struct CopyRepeatParams {
    dstStride = kDefaultDataCopyStride;  
    srcStride = kDefaultDataCopyStride;
    dstRepeatSize = kDefaultRepStride;
    srcRepeatSize = kDefaultRepStride;
};

dstRepeatSize 、srcRepeatSize用于设置相邻迭代间的地址步长,取值范围为[0,4095];dstStride 、srcStride用于设置同一迭代内datablock的地址步长,取值范围为[0,65535]。

相邻迭代间的地址步长参数说明请参考repeatStride(相邻迭代间相同datablock的地址步长);同一迭代内datablock的地址步长参数说明请参考dataBlockStride(同一迭代内不同datablock的地址步长)

返回值

支持的型号

Atlas A2训练系列产品/Atlas 800I A2推理产品

Atlas 200/500 A2推理产品

约束说明

源操作数和目的操作数的起始地址需要保证32字节对齐。

调用示例

本样例中只展示Compute流程中的部分代码。如果您需要运行样例代码,请将该代码段拷贝并替换样例模板中Compute函数的部分代码即可。

  • mask连续模式
    1
    2
    3
    4
    5
    uint64_t mask = 128;
    // repeatTimes = 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::Copy(dstLocal, srcLocal, mask, 4, { 1, 1, 8, 8 });
    

    结果示例如下:

    输入数据(srcLocal): [9 -2 8 ... 9 0]
    输出数据(dstLocal): 
    [9 -2 8 ... 9]
  • mask逐bit模式
    1
    2
    3
    4
    5
    uint64_t mask[2] = { UINT64_MAX, UINT64_MAX };
    // repeatTimes = 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::Copy(dstLocal, srcLocal, mask, 4, { 1, 1, 8, 8 });
    

    结果示例如下:

    输入数据(srcLocal): [9 -2 8 ... 9 0]
    输出数据(dstLocal): 
    [9 -2 8 ... 9]

样例模板

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
#include "kernel_operator.h"
class KernelCopy {
public:
    __aicore__ inline KernelCopy() {}
    __aicore__ inline void Init(__gm__ uint8_t* srcGm, __gm__ uint8_t* dstGm)
    {
        srcGlobal.SetGlobalBuffer((__gm__ int32_t*)srcGm);
        dstGlobal.SetGlobalBuffer((__gm__ int32_t*)dstGm);
        pipe.InitBuffer(inQueueSrc, 1, 512 * sizeof(int32_t));
        pipe.InitBuffer(outQueueDst, 1, 512 * sizeof(int32_t));
    }
    __aicore__ inline void Process()
    {
        CopyIn();
        Compute();
        CopyOut();
    }
private:
    __aicore__ inline void CopyIn()
    {
        AscendC::LocalTensor<int32_t> srcLocal = inQueueSrc.AllocTensor<int32_t>();
        AscendC::DataCopy(srcLocal, srcGlobal, 512);
        inQueueSrc.EnQue(srcLocal);
    }
    __aicore__ inline void Compute()
    {
        AscendC::LocalTensor<int32_t> srcLocal = inQueueSrc.DeQue<int32_t>();
        AscendC::LocalTensor<int32_t> dstLocal = outQueueDst.AllocTensor<int32_t>();
        uint64_t mask = 64;
        AscendC::Copy(dstLocal, srcLocal, mask, 4, { 1, 1, 8, 8 });
        outQueueDst.EnQue<int32_t>(dstLocal);
        inQueueSrc.FreeTensor(srcLocal);
    }
    __aicore__ inline void CopyOut()
    {
        AscendC::LocalTensor<int32_t> dstLocal = outQueueDst.DeQue<int32_t>();
        AscendC::DataCopy(dstGlobal, dstLocal, 512);
        outQueueDst.FreeTensor(dstLocal);
    }
private:
    AscendC::TPipe pipe;
    AscendC::TQue<AscendC::QuePosition::VECIN, 1> inQueueSrc;
    AscendC::TQue<AscendC::QuePosition::VECOUT, 1> outQueueDst;
    AscendC::GlobalTensor<int32_t> srcGlobal, dstGlobal;
};
extern "C" __global__ __aicore__ void copy_simple_kernel(__gm__ uint8_t* srcGm, __gm__ uint8_t* dstGm)
{
    KernelCopy op;
    op.Init(srcGm, dstGm);
    op.Process();
}
搜索结果
找到“0”个结果

当前产品无相关内容

未找到相关内容,请尝试其他搜索词