-
Notifications
You must be signed in to change notification settings - Fork 76
/
trt.cpp
216 lines (194 loc) · 9.13 KB
/
trt.cpp
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
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
//
// Created by yoloCao on 20-5-1.
//
#include <trt.h>
#include <NvOnnxParser.h>
#include <NvOnnxParserRuntime.h>
#include <fstream>
#include <algorithm>
#include <cuda_runtime_api.h>
#include <cuda.h>
#include <NvOnnxConfig.h>
#include "resize.h"
#include <chrono>
#include "utils.h"
#include <numeric>
static Logger gLogger;
#define CHECK(status) \
do \
{ \
auto ret = (status); \
if (ret != 0) \
{ \
std::cerr << "Cuda failure: " << ret << std::endl; \
abort(); \
} \
} while (0)
#define CUDA_CHECK(callstr) \
{ \
cudaError_t error_code = callstr; \
if (error_code != cudaSuccess) { \
std::cerr << "CUDA error " << cudaGetErrorString(error_code) << " at " << __FILE__ << ":" << __LINE__; \
assert(0); \
} \
}
namespace yolodet{
inline unsigned int getElementSize(nvinfer1::DataType t)
{
switch (t)
{
case nvinfer1::DataType::kINT32: return 4;
case nvinfer1::DataType::kFLOAT: return 4;
case nvinfer1::DataType::kHALF: return 2;
case nvinfer1::DataType::kINT8: return 1;
}
throw std::runtime_error("Invalid DataType.");
return 0;
}
inline int64_t volume(const nvinfer1::Dims& d)
{
return accumulate(d.d, d.d + d.nbDims, 1, std::multiplies<int64_t>());
}
inline void* safeCudaMalloc(size_t memSize)
{
void* deviceMem;
CHECK(cudaMalloc(&deviceMem, memSize));
if (deviceMem == nullptr)
{
std::cerr << "Out of memory" << std::endl;
exit(1);
}
return deviceMem;
}
yoloNet::yoloNet(const std::string &onnxFile, const std::string &calibFile, int maxBatchSzie,yolodet::RUN_MODE mode) {
cudaSetDevice(0);
auto builder = nvUniquePtr<nvinfer1::IBuilder>(nvinfer1::createInferBuilder(gLogger));
assert(builder!= nullptr);
auto network = nvUniquePtr<nvinfer1::INetworkDefinition>(builder->createNetwork());
assert(network!= nullptr);
mPlugin =nvonnxparser::createPluginFactory(gLogger);
auto parser = nvUniquePtr<nvonnxparser::IParser>(nvonnxparser::createParser(*network, gLogger));
assert(parser!= nullptr);
if (!parser->parseFromFile(onnxFile.c_str(), 2))
{
std::string msg("failed to parse onnx file");
gLogger.log(nvinfer1::ILogger::Severity::kERROR, msg.c_str());
exit(EXIT_FAILURE);
}
builder->setMaxBatchSize(maxBatchSzie);
builder->setMaxWorkspaceSize(1 << 30);// 1G
if (mode==RUN_MODE::FLOAT16)
{
std::cout <<"setFp16Mode"<<std::endl;
if (!builder->platformHasFastFp16())
std::cout << "Notice: the platform do not has fast for fp16" << std::endl;
builder->setFp16Mode(true);
}
if (mode==RUN_MODE::INT8)
{
std::cout <<"setInt8Mode"<<std::endl;
if (!builder->platformHasFastInt8())
std::cout << "Notice: the platform do not has fast for int8" << std::endl;
builder->setInt8Mode(true);
builder->setInt8Calibrator(nullptr);
}
std::cout << "Begin building engine..." << std::endl;
mEngine = shared_ptr<nvinfer1::ICudaEngine>(
builder->buildCudaEngine(*network), InferDeleter());
if (!mEngine){
std::string error_message ="Unable to create engine";
gLogger.log(nvinfer1::ILogger::Severity::kERROR, error_message.c_str());
exit(-1);
}
std::cout << "End building engine..." << std::endl;
shared_ptr<nvinfer1::IHostMemory> modelStream(mEngine->serialize(),InferDeleter());
assert(modelStream != nullptr);
shared_ptr<nvinfer1::IRuntime> mRunTime(nvinfer1::createInferRuntime(gLogger),InferDeleter());
assert(mRunTime != nullptr);
mEngine= shared_ptr<nvinfer1::ICudaEngine>(mRunTime->deserializeCudaEngine(modelStream->data(), modelStream->size(),mPlugin),InferDeleter());
assert(mEngine != nullptr);
assert(initEngine());
}
yoloNet::yoloNet(const std::string &engineFile)
{
cudaSetDevice(0);
using namespace std;
fstream file;
file.open(engineFile,ios::binary | ios::in);
if(!file.is_open())
{
cout << "read engine file" << engineFile <<" failed" << endl;
return;
}
file.seekg(0, ios::end);
int length = file.tellg();
file.seekg(0, ios::beg);
std::unique_ptr<char[]> data(new char[length]);
file.read(data.get(), length);
file.close();
shared_ptr<nvinfer1::IRuntime> mRunTime(nvinfer1::createInferRuntime(gLogger),InferDeleter());
assert(mRunTime != nullptr);
mPlugin = nvonnxparser::createPluginFactory(gLogger);
mEngine = shared_ptr<nvinfer1::ICudaEngine>(mRunTime->deserializeCudaEngine(data.get(), length, mPlugin),InferDeleter());
assert(mEngine != nullptr);
assert(initEngine());
}
bool yoloNet::initEngine(){
const int maxBatchSize= mEngine->getMaxBatchSize();
mContext = shared_ptr<nvinfer1::IExecutionContext>(mEngine->createExecutionContext(),InferDeleter());
assert(mContext != nullptr);
int nbBindings = mEngine->getNbBindings();
mCudaBuffers.resize(nbBindings);
mBindBufferSizes.resize(2);
for(int i = 0; i < nbBindings; ++i){
nvinfer1::DataType dtype = mEngine->getBindingDataType(i);
int totalSize = maxBatchSize*volume(mEngine->getBindingDimensions(i)) * getElementSize(dtype);
if(i == 0){
mBindBufferSizes[0] = totalSize;
mCudaBuffers[0] = safeCudaMalloc(totalSize);
mBindBufferSizes[1] = 0;
} else mBindBufferSizes[1] += totalSize;
}
mCudaBuffers[1] = safeCudaMalloc(mBindBufferSizes[1]);
for (int i = 2; i < nbBindings; ++i) mCudaBuffers[i]=mCudaBuffers[1];
mCudaImg = safeCudaMalloc(4096*4096*3* sizeof(uchar)); // max input image shape
CUDA_CHECK(cudaStreamCreate(&mCudaStream));
inputDim = mEngine->getBindingDimensions(0);
return 1;
}
yoloNet::~yoloNet() {
cudaStreamSynchronize(mCudaStream);
cudaStreamDestroy(mCudaStream);
if(mCudaBuffers[0])CUDA_CHECK(cudaFree(mCudaBuffers[0]));
if(mCudaBuffers[1])CUDA_CHECK(cudaFree(mCudaBuffers[1]));
if(mCudaImg)CUDA_CHECK(cudaFree(mCudaImg));
}
bool yoloNet::saveEngine(const std::string &fileName){
if(mEngine)
{
shared_ptr<nvinfer1::IHostMemory> data(mEngine->serialize(),InferDeleter());
std::ofstream file;
file.open(fileName,std::ios::binary | std::ios::out);
if(!file.is_open())
{
std::cout << "read create engine file" << fileName <<" failed" << std::endl;
return 0;
}
file.write((const char*)data->data(), data->size());
file.close();
}
return 1;
}
bool yoloNet::infer(const cv::Mat &img, void *outputData) {
bool keepRation = 1 ,keepCenter= 1;
CUDA_CHECK(cudaMemcpy(mCudaImg,img.data,img.step[0]*img.rows,cudaMemcpyHostToDevice));
resizeAndNorm(mCudaImg,(float*)mCudaBuffers[0],img.cols,img.rows,inputDim.d[2],inputDim.d[1],keepRation,keepCenter,0);
CUDA_CHECK(cudaMemset(mCudaBuffers[1],0, sizeof(int)));
mContext->execute(1,&mCudaBuffers[0]);
float det=0;
CUDA_CHECK(cudaMemcpy(&det,mCudaBuffers[1], sizeof(float),cudaMemcpyDeviceToHost));
CUDA_CHECK(cudaMemcpy(outputData,mCudaBuffers[1], sizeof(float) + int(det)*7* sizeof(float),cudaMemcpyDeviceToHost));
doNms(outputData,img.cols,img.rows,inputDim.d[2],inputDim.d[1],0.45,keepRation,keepCenter);
return 1;
}
}