lyblsgo
2023-04-11 6e37612b61094e0ab08b38d35f588c2b17eae267
c++ grpc&onnxruntime support multi threads
13个文件已修改
2个文件已删除
446 ■■■■ 已修改文件
funasr/runtime/grpc/CMakeLists.txt 2 ●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/grpc/paraformer_server.cc 11 ●●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/grpc/paraformer_server.h 6 ●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/include/librapidasrapi.h 77 ●●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/src/CMakeLists.txt 28 ●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/src/FeatureExtract.cpp 22 ●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/src/FeatureExtract.h 10 ●●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/src/commonfunc.h 4 ●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/src/librapidasrapi.cpp 182 ●●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/src/paraformer_onnx.cpp 15 ●●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/src/paraformer_onnx.h 4 ●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/src/precomp.h 2 ●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/tester/CMakeLists.txt 2 ●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/tester/tester.cpp 16 ●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/onnxruntime/tester/tester_rtf.cpp 65 ●●●●● 补丁 | 查看 | 原始文档 | blame | 历史
funasr/runtime/grpc/CMakeLists.txt
@@ -74,7 +74,7 @@
    "${_target}.cc")
  target_link_libraries(${_target}
    rg_grpc_proto
    rapidasr
    funasr
    ${EXTRA_LIBS}
    ${_REFLECTION}
    ${_GRPC_GRPCPP}
funasr/runtime/grpc/paraformer_server.cc
@@ -15,7 +15,6 @@
#include "paraformer.grpc.pb.h"
#include "paraformer_server.h"
using grpc::Server;
using grpc::ServerBuilder;
using grpc::ServerContext;
@@ -24,13 +23,12 @@
using grpc::ServerWriter;
using grpc::Status;
using paraformer::Request;
using paraformer::Response;
using paraformer::ASR;
ASRServicer::ASRServicer(const char* model_path, int thread_num, bool quantize) {
    AsrHanlde=RapidAsrInit(model_path, thread_num, quantize);
    AsrHanlde=FunASRInit(model_path, thread_num, quantize);
    std::cout << "ASRServicer init" << std::endl;
    init_flag = 0;
}
@@ -42,7 +40,7 @@
    Request req;
    std::unordered_map<std::string, std::string> client_buffers;
    std::unordered_map<std::string, std::string> client_transcription;
    while (stream->Read(&req)) {
        if (req.isend()) {
            std::cout << "asr end" << std::endl;
@@ -130,8 +128,8 @@
                    stream->Write(res);
                }
                else {
                    RPASR_RESULT Result= RapidAsrRecogPCMBuffer(AsrHanlde, tmp_data.c_str(), data_len_int, RASR_NONE, NULL);
                    std::string asr_result = ((RPASR_RECOG_RESULT*)Result)->msg;
                    FUNASR_RESULT Result= FunASRRecogPCMBuffer(AsrHanlde, tmp_data.c_str(), data_len_int, RASR_NONE, NULL);
                    std::string asr_result = ((FUNASR_RECOG_RESULT*)Result)->msg;
                    auto end_time = std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::system_clock::now().time_since_epoch()).count();
                    std::string delay_str = std::to_string(end_time - begin_time);
@@ -162,7 +160,6 @@
    }
    return Status::OK;
}
void RunServer(const std::string& port, int thread_num, const char* model_path, bool quantize) {
    std::string server_address;
funasr/runtime/grpc/paraformer_server.h
@@ -15,7 +15,7 @@
#include <chrono>
#include "paraformer.grpc.pb.h"
#include "librapidasrapi.h"
#include "libfunasrapi.h"
using grpc::Server;
@@ -35,7 +35,7 @@
{
    std::string msg;
    float  snippet_time;
}RPASR_RECOG_RESULT;
}FUNASR_RECOG_RESULT;
class ASRServicer final : public ASR::Service {
@@ -45,6 +45,6 @@
  public:
    ASRServicer(const char* model_path, int thread_num, bool quantize);
    grpc::Status Recognize(grpc::ServerContext* context, grpc::ServerReaderWriter<Response, Request>* stream);
    RPASR_HANDLE AsrHanlde;
    FUNASR_HANDLE AsrHanlde;
    
};
funasr/runtime/onnxruntime/include/librapidasrapi.h
File was deleted
funasr/runtime/onnxruntime/src/CMakeLists.txt
@@ -6,38 +6,38 @@
# message("${files}")
add_library(rapidasr ${files})
add_library(funasr ${files})
if(WIN32)
        set(EXTRA_LIBS libfftw3f-3 yaml-cpp)
        if(CMAKE_CL_64)
            target_link_directories(rapidasr PUBLIC ${CMAKE_SOURCE_DIR}/win/lib/x64)
            target_link_directories(funasr PUBLIC ${CMAKE_SOURCE_DIR}/win/lib/x64)
        else()
            target_link_directories(rapidasr PUBLIC ${CMAKE_SOURCE_DIR}/win/lib/x86)
            target_link_directories(funasr PUBLIC ${CMAKE_SOURCE_DIR}/win/lib/x86)
        endif()
        target_include_directories(rapidasr PUBLIC ${CMAKE_SOURCE_DIR}/win/include )
        target_include_directories(funasr PUBLIC ${CMAKE_SOURCE_DIR}/win/include )
        
        target_compile_definitions(rapidasr PUBLIC -D_RPASR_API_EXPORT)
        target_compile_definitions(funasr PUBLIC -D_FUNASR_API_EXPORT)
else()
    set(EXTRA_LIBS fftw3f pthread yaml-cpp)
    target_include_directories(rapidasr PUBLIC "/usr/local/opt/fftw/include")
    target_link_directories(rapidasr PUBLIC "/usr/local/opt/fftw/lib")
    target_include_directories(funasr PUBLIC "/usr/local/opt/fftw/include")
    target_link_directories(funasr PUBLIC "/usr/local/opt/fftw/lib")
    target_include_directories(rapidasr PUBLIC "/usr/local/opt/openblas/include")
    target_link_directories(rapidasr PUBLIC "/usr/local/opt/openblas/lib")
    target_include_directories(funasr PUBLIC "/usr/local/opt/openblas/include")
    target_link_directories(funasr PUBLIC "/usr/local/opt/openblas/lib")
    target_include_directories(rapidasr PUBLIC "/usr/include")
    target_link_directories(rapidasr PUBLIC "/usr/lib64")
    target_include_directories(funasr PUBLIC "/usr/include")
    target_link_directories(funasr PUBLIC "/usr/lib64")
    target_include_directories(rapidasr PUBLIC  ${FFTW3F_INCLUDE_DIR})
    target_link_directories(rapidasr PUBLIC ${FFTW3F_LIBRARY_DIR})
    target_include_directories(funasr PUBLIC  ${FFTW3F_INCLUDE_DIR})
    target_link_directories(funasr PUBLIC ${FFTW3F_LIBRARY_DIR})
    include_directories(${ONNXRUNTIME_DIR}/include)    
endif()
include_directories(${CMAKE_SOURCE_DIR}/include)
target_link_libraries(rapidasr PUBLIC onnxruntime ${EXTRA_LIBS})
target_link_libraries(funasr PUBLIC onnxruntime ${EXTRA_LIBS})
funasr/runtime/onnxruntime/src/FeatureExtract.cpp
@@ -5,16 +5,10 @@
FeatureExtract::FeatureExtract(int mode) : mode(mode)
{
    fftw_init();
}
FeatureExtract::~FeatureExtract()
{
    fftwf_free(fft_input);
    fftwf_free(fft_out);
    if(p){
        fftwf_destroy_plan(p);
    }
}
void FeatureExtract::reset()
@@ -28,23 +22,18 @@
    return fqueue.size();
}
void FeatureExtract::fftw_init()
void FeatureExtract::insert(fftwf_plan plan, float *din, int len, int flag)
{
    int fft_size = 512;
    fft_input = (float *)fftwf_malloc(sizeof(float) * fft_size);
    fft_out = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * fft_size);
    float* fft_input = (float *)fftwf_malloc(sizeof(float) * fft_size);
    fftwf_complex* fft_out = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * fft_size);
    memset(fft_input, 0, sizeof(float) * fft_size);
    p = fftwf_plan_dft_r2c_1d(fft_size, fft_input, fft_out, FFTW_ESTIMATE);
}
void FeatureExtract::insert(float *din, int len, int flag)
{
    const float *window = (const float *)&window_hex;
    if (mode == 3)
        window = (const float *)&window_hamm_hex;
    int window_size = 400;
    int fft_size = 512;
    int window_shift = 160;
    speech.load(din, len);
@@ -72,7 +61,7 @@
            pre_val = cur_val;
        }
        fftwf_execute(p);
        fftwf_execute_dft_r2c(plan, fft_input, fft_out);
        melspect((float *)fft_out, tmp_feature);
        int tmp_flag = S_MIDDLE;
@@ -82,6 +71,8 @@
        fqueue.push(tmp_feature, tmp_flag);
    }
    speech.update(i);
    fftwf_free(fft_input);
    fftwf_free(fft_out);
}
bool FeatureExtract::fetch(Tensor<float> *&dout)
@@ -130,7 +121,6 @@
void FeatureExtract::melspect(float *din, float *dout)
{
    float fftmag[256];
//    float tmp;
    const float *melcoe = (const float *)melcoe_hex;
    int i;
    for (i = 0; i < 256; i++) {
funasr/runtime/onnxruntime/src/FeatureExtract.h
@@ -15,11 +15,7 @@
    FeatureQueue fqueue;
    int mode;
    float *fft_input;
    fftwf_complex *fft_out;
    fftwf_plan p;
    void fftw_init();
    //void fftw_init();
    void melspect(float *din, float *dout);
    void global_cmvn(float *din);
@@ -27,9 +23,9 @@
    FeatureExtract(int mode);
    ~FeatureExtract();
    int size();
    int status();
    //int status();
    void reset();
    void insert(float *din, int len, int flag);
    void insert(fftwf_plan plan, float *din, int len, int flag);
    bool fetch(Tensor<float> *&dout);
};
funasr/runtime/onnxruntime/src/commonfunc.h
@@ -5,7 +5,7 @@
{
    std::string msg;
    float  snippet_time;
}RPASR_RECOG_RESULT;
}FUNASR_RECOG_RESULT;
#ifdef _WIN32
@@ -53,4 +53,4 @@
        }
    }
}
}
funasr/runtime/onnxruntime/src/librapidasrapi.cpp
File was deleted
funasr/runtime/onnxruntime/src/paraformer_onnx.cpp
@@ -18,7 +18,11 @@
    cmvn_path = pathAppend(path, "am.mvn");
    config_path = pathAppend(path, "config.yaml");
    //fe = new FeatureExtract(3);
    int fft_size = 512;
    fft_input = (float *)fftwf_malloc(sizeof(float) * fft_size);
    fft_out = (fftwf_complex *)fftwf_malloc(sizeof(fftwf_complex) * fft_size);
    memset(fft_input, 0, sizeof(float) * fft_size);
    plan = fftwf_plan_dft_r2c_1d(fft_size, fft_input, fft_out, FFTW_ESTIMATE);
    //sessionOptions.SetInterOpNumThreads(1);
    sessionOptions.SetIntraOpNumThreads(nNumThread);
@@ -52,8 +56,6 @@
ModelImp::~ModelImp()
{
    //if(fe)
    //    delete fe;
    if (m_session)
    {
        delete m_session;
@@ -61,11 +63,14 @@
    }
    if(vocab)
        delete vocab;
    fftwf_free(fft_input);
    fftwf_free(fft_out);
    fftwf_destroy_plan(plan);
    fftwf_cleanup();
}
void ModelImp::reset()
{
    //fe->reset();
    printf("Not Imp!!!!!!\n");
}
@@ -163,7 +168,7 @@
    Tensor<float>* in;
    FeatureExtract* fe = new FeatureExtract(3);
    fe->reset();
    fe->insert(din, len, flag);
    fe->insert(plan, din, len, flag);
    fe->fetch(in);
    apply_lfr(in);
    apply_cmvn(in);
funasr/runtime/onnxruntime/src/paraformer_onnx.h
@@ -8,7 +8,9 @@
    class ModelImp : public Model {
    private:
        //FeatureExtract* fe;
        float *fft_input;
        fftwf_complex *fft_out;
        fftwf_plan plan;
        Vocab* vocab;
        vector<float> means_list;
funasr/runtime/onnxruntime/src/precomp.h
@@ -46,7 +46,7 @@
#include <Audio.h>
#include "Model.h"
#include "paraformer_onnx.h"
#include "librapidasrapi.h"
#include "libfunasrapi.h"
using namespace paraformer;
funasr/runtime/onnxruntime/tester/CMakeLists.txt
@@ -8,7 +8,7 @@
    endif()
endif()
set(EXTRA_LIBS rapidasr)
set(EXTRA_LIBS funasr)
include_directories(${CMAKE_SOURCE_DIR}/include)
funasr/runtime/onnxruntime/tester/tester.cpp
@@ -5,7 +5,7 @@
#include <win_func.h>
#endif
#include "librapidasrapi.h"
#include "libfunasrapi.h"
#include <iostream>
#include <fstream>
@@ -26,7 +26,7 @@
    // is quantize
    bool quantize = false;
    istringstream(argv[3]) >> boolalpha >> quantize;
    RPASR_HANDLE AsrHanlde=RapidAsrInit(argv[1], nThreadNum, quantize);
    FUNASR_HANDLE AsrHanlde=FunASRInit(argv[1], nThreadNum, quantize);
    if (!AsrHanlde)
    {
@@ -42,17 +42,17 @@
    gettimeofday(&start, NULL);
    float snippet_time = 0.0f;
    RPASR_RESULT Result=RapidAsrRecogFile(AsrHanlde, argv[2], RASR_NONE, NULL);
    FUNASR_RESULT Result=FunASRRecogFile(AsrHanlde, argv[2], RASR_NONE, NULL);
    gettimeofday(&end, NULL);
   
    if (Result)
    {
        string msg = RapidAsrGetResult(Result, 0);
        string msg = FunASRGetResult(Result, 0);
        setbuf(stdout, NULL);
        printf("Result: %s \n", msg.c_str());
        snippet_time = RapidAsrGetRetSnippetTime(Result);
        RapidAsrFreeResult(Result);
        snippet_time = FunASRGetRetSnippetTime(Result);
        FunASRFreeResult(Result);
    }
    else
    {
@@ -65,9 +65,9 @@
    printf("Model inference takes %lfs.\n", (double)taking_micros / 1000000);
    printf("Model inference RTF: %04lf.\n", (double)taking_micros/ (snippet_time*1000000));
    RapidAsrUninit(AsrHanlde);
    FunASRUninit(AsrHanlde);
    return 0;
}
funasr/runtime/onnxruntime/tester/tester_rtf.cpp
@@ -5,30 +5,13 @@
#include <win_func.h>
#endif
#include "librapidasrapi.h"
#include "libfunasrapi.h"
#include <iostream>
#include <fstream>
#include <sstream>
#include <vector>
#include <thread>
using namespace std;
void runReg(vector<string> wav_list, RPASR_HANDLE AsrHanlde)
{
    for (size_t i = 0; i < wav_list.size(); i++)
    {
        RPASR_RESULT Result=RapidAsrRecogFile(AsrHanlde, wav_list[i].c_str(), RASR_NONE, NULL);
        if(Result){
            string msg = RapidAsrGetResult(Result, 0);
            printf("Result: %s \n", msg.c_str());
            RapidAsrFreeResult(Result);
        }else{
            cout <<"No return data!";
        }
    }
}
int main(int argc, char *argv[])
{
@@ -64,21 +47,53 @@
    bool quantize = false;
    istringstream(argv[3]) >> boolalpha >> quantize;
    RPASR_HANDLE AsrHanlde=RapidAsrInit(argv[1], nThreadNum, quantize);
    FUNASR_HANDLE AsrHanlde=FunASRInit(argv[1], nThreadNum, quantize);
    if (!AsrHanlde)
    {
        printf("Cannot load ASR Model from: %s, there must be files model.onnx and vocab.txt", argv[1]);
        exit(-1);
    }
    gettimeofday(&end, NULL);
    long seconds = (end.tv_sec - start.tv_sec);
    long modle_init_micros = ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);
    printf("Model initialization takes %lfs.\n", (double)modle_init_micros / 1000000);
    // warm up
    for (size_t i = 0; i < 30; i++)
    {
        FUNASR_RESULT Result=FunASRRecogFile(AsrHanlde, wav_list[0].c_str(), RASR_NONE, NULL);
    }
    // forward
    float snippet_time = 0.0f;
    float total_length = 0.0f;
    long total_time = 0.0f;
    
    std::thread t1(runReg, wav_list, AsrHanlde);
    std::thread t2(runReg, wav_list, AsrHanlde);
    for (size_t i = 0; i < wav_list.size(); i++)
    {
        gettimeofday(&start, NULL);
        FUNASR_RESULT Result=FunASRRecogFile(AsrHanlde, wav_list[i].c_str(), RASR_NONE, NULL);
        gettimeofday(&end, NULL);
        seconds = (end.tv_sec - start.tv_sec);
        long taking_micros = ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);
        total_time += taking_micros;
    t1.join();
    t2.join();
        if(Result){
            string msg = FunASRGetResult(Result, 0);
            printf("Result: %s \n", msg);
    //runReg(wav_list, AsrHanlde);
            snippet_time = FunASRGetRetSnippetTime(Result);
            total_length += snippet_time;
            FunASRFreeResult(Result);
        }else{
            cout <<"No return data!";
        }
    }
    RapidAsrUninit(AsrHanlde);
    printf("total_time_wav %ld ms.\n", (long)(total_length * 1000));
    printf("total_time_comput %ld ms.\n", total_time / 1000);
    printf("total_rtf %05lf .\n", (double)total_time/ (total_length*1000000));
    FunASRUninit(AsrHanlde);
    return 0;
}