mmr/tkDNN
Archived
1
0
Fork 0
Code Issues 46 Pull Requests 12 Releases 4 Activity

merge perseusdg windows10 support

Browse Source
This commit was merged in pull request #218.
This commit is contained in:
Francesco Gatti
2021-04-24 18:55:26 +02:00
parent adac8576b0 39323ca8d3
commit 6611a91201
36 changed files with 421 additions and 149 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/LSTM.cpp
+9 -4
View File
@@ -87,17 +87,22 @@ LSTM::LSTM( Network *net, int hiddensize, bool returnSeq, std::string fname_weig
checkCUDNN(cudnnCreateRNNDescriptor(&rnnDesc));
#if CUDNN_MAJOR > 7
checkCUDNN(cudnnSetRNNDescriptor_v6(net->cudnnHandle,
checkCUDNN(cudnnSetRNNDescriptor_v6(net->cudnnHandle,rnnDesc, stateSize, numLayers, dropoutDesc,
cudnnRNNInputMode_t::CUDNN_LINEAR_INPUT,
//(bidirectional ? cudnnDirectionMode_t::CUDNN_BIDIRECTIONAL : cudnnDirectionMode_t::CUDNN_UNIDIRECTIONAL),
cudnnDirectionMode_t::CUDNN_UNIDIRECTIONAL,
cudnnRNNMode_t::CUDNN_LSTM,
cudnnRNNAlgo_t::CUDNN_RNN_ALGO_STANDARD,
net->dataType));
#else
checkCUDNN(cudnnSetRNNDescriptor(net->cudnnHandle,
#endif
rnnDesc, stateSize, numLayers, dropoutDesc,
checkCUDNN(cudnnSetRNNDescriptor(net->cudnnHandle,rnnDesc, stateSize, numLayers, dropoutDesc,
cudnnRNNInputMode_t::CUDNN_LINEAR_INPUT,
//(bidirectional ? cudnnDirectionMode_t::CUDNN_BIDIRECTIONAL : cudnnDirectionMode_t::CUDNN_UNIDIRECTIONAL),
cudnnDirectionMode_t::CUDNN_UNIDIRECTIONAL,
cudnnRNNMode_t::CUDNN_LSTM,
cudnnRNNAlgo_t::CUDNN_RNN_ALGO_STANDARD,
net->dataType));
#endif
// Get temp space sizes
src/NetworkRT.cpp
+82 -36
View File
@@ -139,7 +139,8 @@ NetworkRT::NetworkRT(Network *net, const char *name) {
#if NV_TENSORRT_MAJOR >= 6
engineRT = builderRT->buildEngineWithConfig(*networkRT, *configRT);
#else
engineRT = builderRT->buildCudaEngine(*networkRT);
//engineRT = builderRT->buildCudaEngine(*networkRT);
engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(builderRT->buildCudaEngine(*networkRT));
#endif
if(engineRT == nullptr)
FatalError("cloud not build cuda engine")
@@ -567,7 +568,7 @@ ILayer* NetworkRT::convert_layer(ITensor *input, DeformConv2d *l) {
IPluginLayer *lRT = networkRT->addPlugin(inputs, 2, *plugin);
checkNULL(lRT);
lRT->setName( ("Deformable" + std::to_string(l->id)).c_str() );
delete(inputs);
delete[](inputs);
// batchnorm
void *bias_b, *power_b, *mean_b, *variance_b, *scales_b;
if(dtRT == DataType::kHALF) {
@@ -644,7 +645,7 @@ bool NetworkRT::deserialize(const char *filename) {
IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialData, size_t serialLength) {
const char * buf = reinterpret_cast<const char*>(serialData);
const char * buf = reinterpret_cast<const char*>(serialData),*bufCheck = buf;
std::string name(layerName);
//std::cout<<name<<std::endl;
@@ -652,11 +653,18 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
if(name.find("ActivationLeaky") == 0) {
ActivationLeakyRT *a = new ActivationLeakyRT();
a->size = readBUF<int>(buf);
assert(buf == bufCheck + serialLength);
return a;
}
if(name.find("ActivationMish") == 0) {
ActivationMishRT *a = new ActivationMishRT();
a->size = readBUF<int>(buf);
assert(buf == bufCheck + serialLength);
return a;
}
if(name.find("ActivationLogistic") == 0) {
ActivationLogisticRT *a = new ActivationLogisticRT();
a->size = readBUF<int>(buf);
return a;
}
if(name.find("ActivationLogistic") == 0) {
@@ -665,27 +673,33 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
return a;
}
if(name.find("ActivationCReLU") == 0) {
ActivationReLUCeiling *a = new ActivationReLUCeiling(readBUF<float>(buf));
float activationReluTemp = readBUF<float>(buf);
ActivationReLUCeiling* a = new ActivationReLUCeiling(activationReluTemp);
a->size = readBUF<int>(buf);
assert(buf == bufCheck + serialLength);
return a;
}
if(name.find("Region") == 0) {
RegionRT *r = new RegionRT(readBUF<int>(buf), //classes
readBUF<int>(buf), //coords
readBUF<int>(buf)); //num
int classesTemp = readBUF<int>(buf);
int coordsTemp = readBUF<int>(buf);
int numTemp = readBUF<int>(buf);
RegionRT* r = new RegionRT(classesTemp, coordsTemp, numTemp);
r->c = readBUF<int>(buf);
r->h = readBUF<int>(buf);
r->w = readBUF<int>(buf);
assert(buf == bufCheck + serialLength);
return r;
}
if(name.find("Reorg") == 0) {
ReorgRT *r = new ReorgRT(readBUF<int>(buf)); //stride
int strideTemp = readBUF<int>(buf);
ReorgRT *r = new ReorgRT(strideTemp);
r->c = readBUF<int>(buf);
r->h = readBUF<int>(buf);
r->w = readBUF<int>(buf);
assert(buf == bufCheck + serialLength);
return r;
}
@@ -701,27 +715,34 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
r->h = readBUF<int>(buf);
r->w = readBUF<int>(buf);
return r;
assert(buf == bufCheck + serialLength);
}
if(name.find("Pooling") == 0) {
MaxPoolFixedSizeRT *r = new MaxPoolFixedSizeRT( readBUF<int>(buf), //c
readBUF<int>(buf), //h
readBUF<int>(buf), //w
readBUF<int>(buf), //n
readBUF<int>(buf), //strideH
readBUF<int>(buf), //strideW
readBUF<int>(buf), //winSize
readBUF<int>(buf)); //padding
int cTemp = readBUF<int>(buf);
int hTemp = readBUF<int>(buf);
int wTemp = readBUF<int>(buf);
int nTemp = readBUF<int>(buf);
int strideHTemp = readBUF<int>(buf);
int strideWTemp = readBUF<int>(buf);
int winSizeTemp = readBUF<int>(buf);
int paddingTemp = readBUF<int>(buf);
MaxPoolFixedSizeRT* r = new MaxPoolFixedSizeRT(cTemp, hTemp, wTemp, nTemp, strideHTemp, strideWTemp, winSizeTemp, paddingTemp);
assert(buf == bufCheck + serialLength);
return r;
}
if(name.find("Resize") == 0) {
ResizeLayerRT *r = new ResizeLayerRT(readBUF<int>(buf), //o_c
readBUF<int>(buf), //o_h
readBUF<int>(buf)); //o_w
int o_cTemp = readBUF<int>(buf);
int o_hTemp = readBUF<int>(buf);
int o_wTemp = readBUF<int>(buf);
ResizeLayerRT* r = new ResizeLayerRT(o_cTemp, o_hTemp, o_wTemp);
r->i_c = readBUF<int>(buf);
r->i_h = readBUF<int>(buf);
r->i_w = readBUF<int>(buf);
assert(buf == bufCheck + serialLength);
return r;
}
@@ -732,6 +753,7 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
r->w = readBUF<int>(buf);
r->rows = readBUF<int>(buf);
r->cols = readBUF<int>(buf);
assert(buf == bufCheck + serialLength);
return r;
}
@@ -743,20 +765,25 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
new_dim.h = readBUF<int>(buf);
new_dim.w = readBUF<int>(buf);
ReshapeRT *r = new ReshapeRT(new_dim);
assert(buf == bufCheck + serialLength);
return r;
}
if(name.find("Yolo") == 0) {
YoloRT *r = new YoloRT(readBUF<int>(buf), //classes
readBUF<int>(buf), //num
nullptr, //yolo
readBUF<int>(buf), //n_masks
readBUF<float>(buf), //scale_xy
readBUF<float>(buf), //nms_thresh
readBUF<int>(buf), //nms_kind
readBUF<int>(buf) //new_coords
);
int classes_temp = readBUF<int>(buf);
int num_temp = readBUF<int>(buf);
int n_masks_temp = readBUF<int>(buf);
float scale_xy_temp = readBUF<float>(buf);
float nms_thresh_temp = readBUF<float>(buf);
int nms_kind_temp = readBUF<int>(buf);
int new_coords_temp = readBUF<int>(buf);
YoloRT *r = new YoloRT(classes_temp,num_temp,nullptr,n_masks_temp,scale_xy_temp,nms_thresh_temp,nms_kind_temp,new_coords_temp);
r->c = readBUF<int>(buf);
r->h = readBUF<int>(buf);
r->w = readBUF<int>(buf);
@@ -773,36 +800,54 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
tmp[j] = readBUF<char>(buf);
r->classesNames[i] = std::string(tmp);
}
assert(buf == bufCheck + serialLength);
yolos[n_yolos++] = r;
return r;
}
if(name.find("Upsample") == 0) {
UpsampleRT *r = new UpsampleRT(readBUF<int>(buf)); //stride
int strideTemp = readBUF<int>(buf);
UpsampleRT* r = new UpsampleRT(strideTemp);
r->c = readBUF<int>(buf);
r->h = readBUF<int>(buf);
r->w = readBUF<int>(buf);
assert(buf == bufCheck + serialLength);
return r;
}
if(name.find("Route") == 0) {
RouteRT *r = new RouteRT(readBUF<int>(buf),readBUF<int>(buf));
int groupsTemp = readBUF<int>(buf);
int group_idTemp = readBUF<int>(buf);
RouteRT* r = new RouteRT(groupsTemp, group_idTemp);
r->in = readBUF<int>(buf);
for(int i=0; i<RouteRT::MAX_INPUTS; i++)
r->c_in[i] = readBUF<int>(buf);
r->c = readBUF<int>(buf);
r->h = readBUF<int>(buf);
r->w = readBUF<int>(buf);
assert(buf == bufCheck + serialLength);
return r;
}
if(name.find("Deformable") == 0) {
DeformableConvRT *r = new DeformableConvRT(readBUF<int>(buf), readBUF<int>(buf), readBUF<int>(buf),
readBUF<int>(buf), readBUF<int>(buf), readBUF<int>(buf),
readBUF<int>(buf), readBUF<int>(buf),
readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),
readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),readBUF<int>(buf),
nullptr);
int chuck_dimTemp = readBUF<int>(buf);
int khTemp = readBUF<int>(buf);
int kwTemp = readBUF<int>(buf);
int shTemp = readBUF<int>(buf);
int swTemp = readBUF<int>(buf);
int phTemp = readBUF<int>(buf);
int pwTemp = readBUF<int>(buf);
int deformableGroupTemp = readBUF<int>(buf);
int i_nTemp = readBUF<int>(buf);
int i_cTemp = readBUF<int>(buf);
int i_hTemp = readBUF<int>(buf);
int i_wTemp = readBUF<int>(buf);
int o_nTemp = readBUF<int>(buf);
int o_cTemp = readBUF<int>(buf);
int o_hTemp = readBUF<int>(buf);
int o_wTemp = readBUF<int>(buf);
DeformableConvRT* r = new DeformableConvRT(chuck_dimTemp, khTemp, kwTemp, shTemp, swTemp, phTemp, pwTemp, deformableGroupTemp, i_nTemp, i_cTemp, i_hTemp, i_wTemp, o_nTemp, o_cTemp, o_hTemp, o_wTemp, nullptr);
dnnType *aus = new dnnType[r->chunk_dim*2];
for(int i=0; i<r->chunk_dim*2; i++)
aus[i] = readBUF<dnnType>(buf);
@@ -833,6 +878,7 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
aus[i] = readBUF<dnnType>(buf);
checkCuda( cudaMemcpy(r->ones_d2, aus, sizeof(dnnType)*r->dim_ones, cudaMemcpyHostToDevice) );
free(aus);
assert(buf == bufCheck + serialLength);
return r;
}
src/Yolo.cpp
+5 -5
View File
@@ -9,6 +9,7 @@
#include "Layer.h"
#include "kernels.h"
namespace tk { namespace dnn {
Yolo::Yolo(Network *net, int classes, int num, std::string fname_weights, int n_masks, float scale_xy, double nms_thresh, nmsKind_t nsm_kind, int new_coords) :
@@ -95,7 +96,6 @@ dnnType* Yolo::infer(dataDim_t &dim, dnnType* srcData) {
activationLOGISTICForward(srcData + index, dstData + index, 2*dim.w*dim.h);
if (this->scaleXY != 1) scalAdd(dstData + index, 2 * dim.w*dim.h, this->scaleXY, -0.5*(this->scaleXY - 1), 1);
index = entry_index(b, n*dim.w*dim.h, 4, classes, input_dim, output_dim);
activationLOGISTICForward(srcData + index, dstData + index, (1+classes)*dim.w*dim.h);
}
@@ -212,10 +212,10 @@ float yolo_box_iou(Yolo::box a, Yolo::box b)
}
void box_c(const Yolo::box a, const Yolo::box b, float& top, float& bot, float& left, float& right) {
top = std::min(a.y - a.h / 2, b.y - b.h / 2);
bot = std::max(a.y + a.h / 2, b.y + b.h / 2);
left = std::min(a.x - a.w / 2, b.x - b.w / 2);
right = std::max(a.x + a.w / 2, b.x + b.w / 2);
top = (std::min)(a.y - a.h / 2, b.y - b.h / 2);
bot = (std::max)(a.y + a.h / 2, b.y + b.h / 2);
left = (std::min)(a.x - a.w / 2, b.x - b.w / 2);
right = (std::max)(a.x + a.w / 2, b.x + b.w / 2);
}
// https://github.com/Zzh-tju/DIoU-darknet
src/Yolo3Detection.cpp
+4 -3
View File
@@ -94,9 +94,10 @@ void Yolo3Detection::preprocess(cv::Mat &frame, const int bi){
void Yolo3Detection::postprocess(const int bi, const bool mAP){
//get yolo outputs
dnnType *rt_out[netRT->pluginFactory->n_yolos];
for(int i=0; i<netRT->pluginFactory->n_yolos; i++)
rt_out[i] = (dnnType*)netRT->buffersRT[i+1] + netRT->buffersDIM[i+1].tot()*bi;
std::vector<float *> rt_out;
//dnnType *rt_out[netRT->pluginFactory->n_yolos];
for(int i=0; i<netRT->pluginFactory->n_yolos; i++)
rt_out.push_back((dnnType*)netRT->buffersRT[i+1] + netRT->buffersDIM[i+1].tot()*bi);
float x_ratio = float(originalSize[bi].width) / float(netRT->input_dim.w);
float y_ratio = float(originalSize[bi].height) / float(netRT->input_dim.h);
src/kernels/deformable_conv.cu
+1 -1
View File
@@ -18,7 +18,7 @@ inline int GET_BLOCKS(const int N)
}
__device__ float dmcn_im2col_bilinear(const float *bottom_data, const int data_width,
__device__ __host__ float dmcn_im2col_bilinear(const float *bottom_data, const int data_width,
const int height, const int width, float h, float w) {
int h_low = floor(h);
int w_low = floor(w);
src/utils.cpp
+15 -2
View File
@@ -23,14 +23,23 @@ bool fileExist(const char *fname) {
void downloadWeightsifDoNotExist(const std::string& input_bin, const std::string& test_folder, const std::string& weights_url){
if(!fileExist(input_bin.c_str())){
std::string mkdir_cmd = "mkdir " + test_folder;
std::string wget_cmd = "wget " + weights_url + " -O " + test_folder + "/weights.zip";
std::string wget_cmd = "curl " + weights_url + " --output " + test_folder + "/weights.zip";
#ifdef __linux__
std::string unzip_cmd = "unzip " + test_folder + "/weights.zip -d" + test_folder;
std::string rm_cmd = "rm " + test_folder + "/weights.zip";
#elif _WIN32
std::string unzip_cmd = "7z x " + test_folder + "/weights.zip -o" + test_folder;
#endif
int err = 0;
err = system(mkdir_cmd.c_str());
err = system(wget_cmd.c_str());
err = system(unzip_cmd.c_str());
#ifdef __linux__
err = system(rm_cmd.c_str());
#endif
}
}
@@ -191,8 +200,12 @@ void getMemUsage(double& vm_usage_kb, double& resident_set_kb){
>> O >> itrealvalue >> starttime >> vsize >> rss;
stat_stream.close();
#ifdef __linux__
long page_size_kb = sysconf(_SC_PAGE_SIZE) / 1024; // in case x86-64 is configured to use 2MB pages
#elif _WIN32
long page_size_kb = 4096/1024;
#endif
vm_usage_kb = vsize / 1024.0;
resident_set_kb = rss * page_size_kb;
}