/**
* Copyright (c) 2025 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
/* !
* \file select.asc
* \brief
*/
#include "acl/acl.h"
#include "data_utils.h"
#include "kernel_operator.h"
class KernelSelect {
public:
__aicore__ inline KernelSelect() {}
__aicore__ inline void Init(__gm__ uint8_t* src0Gm, __gm__ uint8_t* src1Gm, __gm__ uint8_t* selGm,
__gm__ uint8_t* dstGm, AscendC::TPipe* pipeIn)
{
pipe = pipeIn;
src0Global.SetGlobalBuffer((__gm__ float*)src0Gm);
src1Global.SetGlobalBuffer((__gm__ float*)src1Gm);
selMaskGlobal.SetGlobalBuffer((__gm__ uint8_t*)selGm);
dstGlobal.SetGlobalBuffer((__gm__ float*)dstGm);
pipe->InitBuffer(inQueueSrc0, 1, dataSize * sizeof(float));
pipe->InitBuffer(inQueueSrc1, 1, dataSize * sizeof(float));
pipe->InitBuffer(inQueueSelMask, 1, selDataSize * sizeof(uint8_t));
pipe->InitBuffer(outQueueDst, 1, dataSize * sizeof(float));
}
__aicore__ inline void Process()
{
CopyIn();
Compute();
CopyOut();
}
private:
__aicore__ inline void CopyIn()
{
AscendC::LocalTensor<float> src0Local = inQueueSrc0.AllocTensor<float>();
AscendC::LocalTensor<float> src1Local = inQueueSrc1.AllocTensor<float>();
AscendC::LocalTensor<uint8_t> selMaskLocal = inQueueSelMask.AllocTensor<uint8_t>();
AscendC::DataCopy(src0Local, src0Global, dataSize);
AscendC::DataCopy(src1Local, src1Global, dataSize);
AscendC::DataCopy(selMaskLocal, selMaskGlobal, selDataSize);
inQueueSrc0.EnQue(src0Local);
inQueueSrc1.EnQue(src1Local);
inQueueSelMask.EnQue(selMaskLocal);
}
__aicore__ inline void Compute()
{
AscendC::LocalTensor<float> src0Local = inQueueSrc0.DeQue<float>();
AscendC::LocalTensor<float> src1Local = inQueueSrc1.DeQue<float>();
AscendC::LocalTensor<uint8_t> maskLocal = inQueueSelMask.DeQue<uint8_t>();
AscendC::LocalTensor<float> dstLocal = outQueueDst.AllocTensor<float>();
AscendC::Select(dstLocal, maskLocal, src0Local, src1Local, AscendC::SELMODE::VSEL_CMPMASK_SPR, dataSize);
outQueueDst.EnQue<float>(dstLocal);
inQueueSrc0.FreeTensor(src0Local);
inQueueSrc1.FreeTensor(src1Local);
inQueueSelMask.FreeTensor(maskLocal);
}
__aicore__ inline void CopyOut()
{
AscendC::LocalTensor<float> dstLocal = outQueueDst.DeQue<float>();
AscendC::DataCopy(dstGlobal, dstLocal, dataSize);
outQueueDst.FreeTensor(dstLocal);
}
private:
AscendC::TPipe* pipe;
AscendC::TQue<AscendC::TPosition::VECIN, 1> inQueueSrc0, inQueueSrc1, inQueueSelMask;
AscendC::TQue<AscendC::TPosition::VECOUT, 1> outQueueDst;
AscendC::GlobalTensor<float> src0Global, src1Global, dstGlobal;
AscendC::GlobalTensor<uint8_t> selMaskGlobal;
uint32_t dataSize = 256;
uint32_t oneSelectDataSize = 256/ sizeof(float);
uint32_t selDataSize = dataSize / oneSelectDataSize * 32;
// (模式1和模式2时,uint32_t selDataSize = dataSize / 8;)
};
__global__ __aicore__ void select_custom(GM_ADDR x, GM_ADDR y, GM_ADDR sel, GM_ADDR z)
{
KERNEL_TASK_TYPE_DEFAULT(KERNEL_TYPE_AIV_ONLY);
AscendC::TPipe pipe;
KernelSelect op;
op.Init(x, y, sel, z, &pipe);
op.Process();
}
int32_t main(int32_t argc, char* argv[])
{
uint32_t numBlocks = 1;
uint32_t dataSize = 256;
uint32_t oneSelectDataSize = 256/ sizeof(float);
uint32_t selDataSize = dataSize / oneSelectDataSize * 32;
size_t inputByteSize = dataSize * sizeof(float);
size_t outputByteSize = inputByteSize;
size_t selByteSize = selDataSize * sizeof(uint8_t);
int32_t deviceId = 0;
aclrtSetDevice(deviceId);
aclrtStream stream = nullptr;
aclrtCreateStream(&stream);
uint8_t *xHost, *yHost, *zHost, *selHost;
uint8_t *xDevice, *yDevice, *zDevice, *selDevice;
aclrtMallocHost((void **)(&xHost), inputByteSize);
aclrtMallocHost((void **)(&yHost), inputByteSize);
aclrtMallocHost((void **)(&zHost), outputByteSize);
aclrtMallocHost((void **)(&selHost), selByteSize);
aclrtMalloc((void **)&xDevice, inputByteSize, ACL_MEM_MALLOC_HUGE_FIRST);
aclrtMalloc((void **)&yDevice, inputByteSize, ACL_MEM_MALLOC_HUGE_FIRST);
aclrtMalloc((void **)&zDevice, outputByteSize, ACL_MEM_MALLOC_HUGE_FIRST);
aclrtMalloc((void **)&selDevice, selByteSize, ACL_MEM_MALLOC_HUGE_FIRST);
ReadFile("./input/input_x.bin", inputByteSize, xHost, inputByteSize);
ReadFile("./input/input_y.bin", inputByteSize, yHost, inputByteSize);
ReadFile("./input/input_sel.bin", selByteSize, selHost, selByteSize);
aclrtMemcpy(xDevice, inputByteSize, xHost, inputByteSize,
ACL_MEMCPY_HOST_TO_DEVICE);
aclrtMemcpy(yDevice, inputByteSize, yHost, inputByteSize,
ACL_MEMCPY_HOST_TO_DEVICE);
aclrtMemcpy(selDevice, selByteSize, selHost, selByteSize,
ACL_MEMCPY_HOST_TO_DEVICE);
select_custom<<<numBlocks, nullptr, stream>>>(xDevice, yDevice, selDevice, zDevice);
aclrtSynchronizeStream(stream);
aclrtMemcpy(zHost, outputByteSize, zDevice, outputByteSize,
ACL_MEMCPY_DEVICE_TO_HOST);
WriteFile("./output/output.bin", zHost, outputByteSize);
aclrtFree(xDevice);
aclrtFree(yDevice);
aclrtFree(zDevice);
aclrtFree(selDevice);
aclrtFreeHost(xHost);
aclrtFreeHost(yHost);
aclrtFreeHost(zHost);
aclrtFreeHost(selHost);
aclrtDestroyStream(stream);
aclrtResetDevice(deviceId);
aclFinalize();
return 0;
}