yolo-v5 nnie 推理

#include <stdio.h> #include <stdlib.h> #include <string.h> #include <errno.h> #include <fcntl.h> #include <sys/mman.h> #include <sys/stat.h> #include <sys/types.h> #include <unistd.h> #include <signal.h> #include <pthread.h> #include <sys/prctl.h> #include <math.h> #include "config.h" #include "hi_common.h" #include "hi_comm_sys.h" #include "hi_comm_svp.h" #include "sample_comm.h" #include "sample_comm_svp.h" #include "sample_comm_nnie.h" #include "nnie_face_api.h" #include "sample_svp_nnie_software.h" #include "sample_comm_ive.h" #include "hi_type.h" //#include"list.h" /*cnn para*/ static SAMPLE_SVP_NNIE_MODEL_S s_stDetModel = {0}; static SAMPLE_SVP_NNIE_MODEL_S s_stExtModel = {0}; static SAMPLE_SVP_NNIE_PARAM_S s_stDetNnieParam = {0}; static SAMPLE_SVP_NNIE_PARAM_S s_stExtNnieParam = {0}; //static anchor_generator_t* anc_gen = NULL; int IsDebugLog = 0; SAMPLE_SVP_NNIE_CFG_S stNnieCfg = {0}; HI_S32 as32ResultDet[200 * 15] = { 0 }; HI_S32 u32ResultDetCnt = 0; int IndexBuffer[512] = { 0 }; #ifdef SAMPLE_SVP_NNIE_PERF_STAT #define SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_CLREAR() memset(&s_stOpForwardPerfTmp,0,sizeof(s_stOpForwardPerfTmp)); #define SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_SRC_FLUSH_TIME() SAMPLE_SVP_NNIE_PERF_STAT_GET_DIFF_TIME(s_stOpForwardPerfTmp.u64SrcFlushTime) #define SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_PRE_DST_FLUSH_TIME() SAMPLE_SVP_NNIE_PERF_STAT_GET_DIFF_TIME(s_stOpForwardPerfTmp.u64PreDstFulshTime) #define SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_AFTER_DST_FLUSH_TIME() SAMPLE_SVP_NNIE_PERF_STAT_GET_DIFF_TIME(s_stOpForwardPerfTmp.u64AferDstFulshTime) #define SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_OP_TIME() SAMPLE_SVP_NNIE_PERF_STAT_GET_DIFF_TIME(s_stOpForwardPerfTmp.u64OPTime) static SAMPLE_SVP_NNIE_OP_PERF_STAT_S s_stOpForwardPerfTmp = {0}; #else #define SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_CLREAR() #define SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_SRC_FLUSH_TIME() #define SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_PRE_DST_FLUSH_TIME() #define SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_AFTER_DST_FLUSH_TIME() #define SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_OP_TIME() #endif float sigmoid(float x){ return (1.0f / ((float)exp((double)(-x)) + 1.0f)); } /****************************************************************************** * function : NNIE Forward ******************************************************************************/ static HI_S32 SAMPLE_SVP_NNIE_Forward(SAMPLE_SVP_NNIE_PARAM_S *pstNnieParam, SAMPLE_SVP_NNIE_INPUT_DATA_INDEX_S* pstInputDataIdx, SAMPLE_SVP_NNIE_PROCESS_SEG_INDEX_S* pstProcSegIdx,HI_BOOL bInstant) { HI_S32 s32Ret = HI_SUCCESS; HI_U32 i = 0, j = 0; HI_BOOL bFinish = HI_FALSE; SVP_NNIE_HANDLE hSvpNnieHandle = 0; HI_U32 u32TotalStepNum = 0; SAMPLE_SVP_NIE_PERF_STAT_DEF_VAR(); SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_CLREAR(); SAMPLE_COMM_SVP_FlushCache(pstNnieParam->astForwardCtrl[pstProcSegIdx->u32SegIdx].stTskBuf.u64PhyAddr, SAMPLE_SVP_NNIE_CONVERT_64BIT_ADDR(HI_VOID,pstNnieParam->astForwardCtrl[pstProcSegIdx->u32SegIdx].stTskBuf.u64VirAddr), pstNnieParam->astForwardCtrl[pstProcSegIdx->u32SegIdx].stTskBuf.u32Size); SAMPLE_SVP_NNIE_PERF_STAT_BEGIN(); for(i = 0; i < pstNnieParam->astForwardCtrl[pstProcSegIdx->u32SegIdx].u32DstNum; i++) { if(SVP_BLOB_TYPE_SEQ_S32 == pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].enType) { for(j = 0; j < pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].u32Num; j++) { u32TotalStepNum += *(SAMPLE_SVP_NNIE_CONVERT_64BIT_ADDR(HI_U32,pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].unShape.stSeq.u64VirAddrStep)+j); } SAMPLE_COMM_SVP_FlushCache(pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].u64PhyAddr, SAMPLE_SVP_NNIE_CONVERT_64BIT_ADDR(HI_VOID,pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].u64VirAddr), u32TotalStepNum*pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].u32Stride); } else { SAMPLE_COMM_SVP_FlushCache(pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].u64PhyAddr, SAMPLE_SVP_NNIE_CONVERT_64BIT_ADDR(HI_VOID,pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].u64VirAddr), pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].u32Num* pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].unShape.stWhc.u32Chn* pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].unShape.stWhc.u32Height* pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].u32Stride); } } SAMPLE_SVP_NNIE_PERF_STAT_END(); SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_PRE_DST_FLUSH_TIME(); /*set input blob according to node name*/ if(pstInputDataIdx->u32SegIdx != pstProcSegIdx->u32SegIdx) { for(i = 0; i < pstNnieParam->pstModel->astSeg[pstProcSegIdx->u32SegIdx].u16SrcNum; i++) { for(j = 0; j < pstNnieParam->pstModel->astSeg[pstInputDataIdx->u32SegIdx].u16DstNum; j++) { if(0 == strncmp(pstNnieParam->pstModel->astSeg[pstInputDataIdx->u32SegIdx].astDstNode[j].szName, pstNnieParam->pstModel->astSeg[pstProcSegIdx->u32SegIdx].astSrcNode[i].szName, SVP_NNIE_NODE_NAME_LEN)) { pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astSrc[i] = pstNnieParam->astSegData[pstInputDataIdx->u32SegIdx].astDst[j]; break; } } SAMPLE_SVP_CHECK_EXPR_RET((j == pstNnieParam->pstModel->astSeg[pstInputDataIdx->u32SegIdx].u16DstNum), HI_FAILURE,SAMPLE_SVP_ERR_LEVEL_ERROR,"Error,can't find %d-th seg's %d-th src blob!\n", pstProcSegIdx->u32SegIdx,i); } } /*NNIE_Forward*/ SAMPLE_SVP_NNIE_PERF_STAT_BEGIN(); s32Ret = HI_MPI_SVP_NNIE_Forward(&hSvpNnieHandle, // 官方api, 将图片输入NNIE做前向处理 。 pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astSrc, pstNnieParam->pstModel, pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst, // astDst 包含 最终结果输出 以及 中间层输出（需要设置上报） 。 &pstNnieParam->astForwardCtrl[pstProcSegIdx->u32SegIdx], bInstant); SAMPLE_SVP_CHECK_EXPR_RET(HI_SUCCESS != s32Ret,s32Ret,SAMPLE_SVP_ERR_LEVEL_ERROR, "Error,HI_MPI_SVP_NNIE_Forward failed!\n"); if(bInstant) // 该值为 HI_MPI_SVP_NNIE_Forward 输出结果标志 。 { /*Wait NNIE finish*/ while(HI_ERR_SVP_NNIE_QUERY_TIMEOUT == (s32Ret = HI_MPI_SVP_NNIE_Query(pstNnieParam->astForwardCtrl[pstProcSegIdx->u32SegIdx].enNnieId, hSvpNnieHandle, &bFinish, HI_TRUE))) { usleep(100); SAMPLE_SVP_TRACE(SAMPLE_SVP_ERR_LEVEL_INFO, "HI_MPI_SVP_NNIE_Query Query timeout!\n"); } } SAMPLE_SVP_NNIE_PERF_STAT_END(); SAMPLE_SVP_NNIE_PERF_STAT_OP_FORWARD_OP_TIME(); u32TotalStepNum = 0; SAMPLE_SVP_NNIE_PERF_STAT_BEGIN(); for(i = 0; i < pstNnieParam->astForwardCtrl[pstProcSegIdx->u32SegIdx].u32DstNum; i++) { if(SVP_BLOB_TYPE_SEQ_S32 == pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].enType) { for(j = 0; j < pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].u32Num; j++) { u32TotalStepNum += *(SAMPLE_SVP_NNIE_CONVERT_64BIT_ADDR(HI_U32,pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].unShape.stSeq.u64VirAddrStep)+j); } SAMPLE_COMM_SVP_FlushCache(pstNnieParam->astSegData[pstProcSegIdx->u32SegIdx].astDst[i].u64PhyAddr,