Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

TensorRT8.6 C++量化bug修改 #80

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
Open

TensorRT8.6 C++量化bug修改 #80

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d89b4f9
save float32
FelixFu520 Jul 14, 2023
d20911a
fix malloc pData
FelixFu520 Jul 14, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
46 changes: 35 additions & 11 deletions cookbook/04-BuildEngineByONNXParser/pyTorch-ONNX-TensorRT/C++/calibrator.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,46 +2,64 @@

using namespace nvinfer1;

MyCalibrator::MyCalibrator(const std::string &calibrationDataFile, const int nCalibration, const Dims32 dim, const std::string &cacheFile):
MyCalibrator::MyCalibrator(const std::string& calibrationDataFile, const int nCalibration, const Dims32 dim, const std::string& cacheFile) :
nCalibration(nCalibration), dim(dim), cacheFile(cacheFile), iBatch(0)
{
#ifdef DEBUG
std::cout << "[MyCalibrator::MyCalibrator]" << std::endl;
#endif
cnpy::npz_t npzFile = cnpy::npz_load(calibrationDataFile);
cnpy::NpyArray array = npzFile[std::string("calibrationData")];
pData = array.data<float>();
cnpy::NpyArray array = npzFile[std::string("calibrationData")];
auto pDataTemp = array.data<float>();
pData = (float*)malloc(array.num_bytes());
memcpy(pData, pDataTemp, array.num_bytes());

if (pData == nullptr)
{
std::cout << "Failed getting calibration data!" << std::endl;
return;
}

nBatch = array.num_bytes() / bufferSize;
nElement = 1;
for (int i = 0; i < dim.nbDims; ++i)
{
nElement *= dim.d[i];
}
bufferSize = sizeof(float) * nElement;
cudaMalloc((void **)&bufferD, bufferSize);
nBatch = array.num_bytes() / bufferSize;
cudaMalloc((void**)&bufferD, bufferSize);

return;
}

MyCalibrator::~MyCalibrator() noexcept
{
#ifdef DEBUG
std::cout << "[MyCalibrator::~MyCalibrator]" << std::endl;
#endif
if (bufferD != nullptr)
{
cudaFree(bufferD);
}
if (pData != nullptr) {
free(pData);
}
return;
}

int32_t MyCalibrator::getBatchSize() const noexcept
{
#ifdef DEBUG
std::cout << "[MyCalibrator::getBatchSize]" << std::endl;
#endif
return dim.d[0];
}

bool MyCalibrator::getBatch(void *bindings[], char const *names[], int32_t nbBindings) noexcept
bool MyCalibrator::getBatch(void* bindings[], char const* names[], int32_t nbBindings) noexcept
{
#ifdef DEBUG
std::cout << "[MyCalibrator::getBatch]" << std::endl;
#endif
if (iBatch < nBatch)
{
cudaMemcpy(bufferD, &pData[iBatch * nElement], bufferSize, cudaMemcpyHostToDevice);
Expand All @@ -55,36 +73,42 @@ bool MyCalibrator::getBatch(void *bindings[], char const *names[], int32_t nbBin
}
}

void const *MyCalibrator::readCalibrationCache(std::size_t &length) noexcept
void const* MyCalibrator::readCalibrationCache(std::size_t& length) noexcept
{
#ifdef DEBUG
std::cout << "[MyCalibrator::readCalibrationCache]" << std::endl;
#endif
std::fstream f;
f.open(cacheFile, std::fstream::in);
if (f.fail())
{
std::cout << "Failed finding cache file!" << std::endl;
return nullptr;
}
char *ptr = new char[length];
char* ptr = new char[length];
if (f.is_open())
{
f >> ptr;
}
return ptr;
}

void MyCalibrator::writeCalibrationCache(void const *ptr, std::size_t length) noexcept
void MyCalibrator::writeCalibrationCache(void const* ptr, std::size_t length) noexcept
{
#ifdef DEBUG
std::cout << "[MyCalibrator::writeCalibrationCache]" << std::endl;
#endif
std::ofstream f(cacheFile, std::ios::binary);
if (f.fail())
{
std::cout << "Failed opening cache file to write!" << std::endl;
return;
}
f.write(static_cast<char const *>(ptr), length);
f.write(static_cast<char const*>(ptr), length);
if (f.fail())
{
std::cout << "Failed saving cache file!" << std::endl;
return;
}
f.close();
}
}
4 changes: 2 additions & 2 deletions ...04-BuildEngineByONNXParser/pyTorch-ONNX-TensorRT/C++/createCalibrationAndInferenceData.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,12 +28,12 @@
inferenceData = cv2.imread(inferenceDataFile, cv2.IMREAD_GRAYSCALE).astype(np.float32)

calibrationDataFileList = sorted(glob(calibrationDataPath + "*.jpg"))[:nCalibrationData]
calibrationData = np.empty([nCalibrationData, 1, nHeight, nWidth])
calibrationData = np.empty([nCalibrationData, 1, nHeight, nWidth], dtype=np.float32)
for i in range(nCalibrationData):
calibrationData[i, 0] = cv2.imread(calibrationDataFileList[i], cv2.IMREAD_GRAYSCALE).astype(np.float32)

dataDictionary = {}
dataDictionary["inferenceData"] = inferenceData
dataDictionary["calibrationData"] = calibrationData
np.savez("data.npz", **dataDictionary)
print("Succeeded creating data for calibration and inference!")
print("Succeeded creating data for calibration and inference!")