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

merge from gitlab

Browse Source
This commit is contained in:
perseusdg
2021-04-08 00:28:00 +05:30
parent 2e92944f1d
commit cc594f09ef
4 changed files with 9 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/tkDNN/NetworkRT.h
+4 -6
View File
@@ -55,16 +55,14 @@ class NetworkRT {
public: public:
nvinfer1::DataType dtRT; nvinfer1::DataType dtRT;
//nvinfer1::IBuilder *builderRT; nvinfer1::IBuilder *builderRT;
std::unique_ptr<nvinfer1::IBuilder,InferDeleter> builderRT; nvinfer1::IRuntime *runtimeRT;
//nvinfer1::IRuntime *runtimeRT;
std::unique_ptr<nvinfer1::IRuntime,InferDeleter> runtimeRT;
nvinfer1::INetworkDefinition *networkRT; nvinfer1::INetworkDefinition *networkRT;
#if NV_TENSORRT_MAJOR >= 6 #if NV_TENSORRT_MAJOR >= 6
nvinfer1::IBuilderConfig *configRT; nvinfer1::IBuilderConfig *configRT;
#endif #endif
std::shared_ptr<nvinfer1::ICudaEngine> engineRT;
//nvinfer1::ICudaEngine *engineRT; nvinfer1::ICudaEngine *engineRT;
nvinfer1::IExecutionContext *contextRT; nvinfer1::IExecutionContext *contextRT;
const static int MAX_BUFFERS_RT = 10; const static int MAX_BUFFERS_RT = 10;
include/tkDNN/utils.h
-14
View File
@@ -14,9 +14,6 @@
#ifdef __linux__ #ifdef __linux__
#include <unistd.h> #include <unistd.h>
#elif _WIN32
#define NOMINMAX
#include <windows.h>
#endif #endif
#include <ios> #include <ios>
@@ -110,17 +107,6 @@ double t_ns = time_ms.count();
FatalError(_error.str()); \ FatalError(_error.str()); \
} \ } \
} }
struct InferDeleter
{
template <typename T>
void operator()(T* obj) const
{
if (obj)
{
obj->destroy();
}
}
};
typedef enum { typedef enum {
ERROR_CUDNN = 2, ERROR_CUDNN = 2,
src/NetworkRT.cpp
+4 -33
View File
@@ -33,8 +33,7 @@ NetworkRT::NetworkRT(Network *net, const char *name) {
float(NV_TENSORRT_PATCH)/100; float(NV_TENSORRT_PATCH)/100;
std::cout<<"New NetworkRT (TensorRT v"<<rt_ver<<")\n"; std::cout<<"New NetworkRT (TensorRT v"<<rt_ver<<")\n";
builderRT = std::unique_ptr<nvinfer1::IBuilder,InferDeleter>(createInferBuilder(loggerRT)); builderRT = createInferBuilder(loggerRT);
//builderRT = createInferBuilder(loggerRT);
std::cout<<"Float16 support: "<<builderRT->platformHasFastFp16()<<"\n"; std::cout<<"Float16 support: "<<builderRT->platformHasFastFp16()<<"\n";
std::cout<<"Int8 support: "<<builderRT->platformHasFastInt8()<<"\n"; std::cout<<"Int8 support: "<<builderRT->platformHasFastInt8()<<"\n";
#if NV_TENSORRT_MAJOR >= 5 #if NV_TENSORRT_MAJOR >= 5
@@ -138,8 +137,7 @@ NetworkRT::NetworkRT(Network *net, const char *name) {
printCudaMemUsage(); printCudaMemUsage();
std::cout<<"Building tensorRT cuda engine...\n"; std::cout<<"Building tensorRT cuda engine...\n";
#if NV_TENSORRT_MAJOR >= 6 #if NV_TENSORRT_MAJOR >= 6
//engineRT = builderRT->buildEngineWithConfig(*networkRT, *configRT); engineRT = builderRT->buildEngineWithConfig(*networkRT, *configRT);
engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(builderRT->buildEngineWithConfig(*networkRT,*configRT),InferDeleter());
#else #else
//engineRT = builderRT->buildCudaEngine(*networkRT); //engineRT = builderRT->buildCudaEngine(*networkRT);
engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(builderRT->buildCudaEngine(*networkRT)); engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(builderRT->buildCudaEngine(*networkRT));
@@ -637,10 +635,8 @@ bool NetworkRT::deserialize(const char *filename) {
} }
pluginFactory = new PluginFactory(); pluginFactory = new PluginFactory();
//runtimeRT = createInferRuntime(loggerRT); runtimeRT = createInferRuntime(loggerRT);
runtimeRT = std::unique_ptr<nvinfer1::IRuntime,InferDeleter>(createInferRuntime(loggerRT)); engineRT = runtimeRT->deserializeCudaEngine(gieModelStream, size, (IPluginFactory *) pluginFactory);
//engineRT = runtimeRT->deserializeCudaEngine(gieModelStream, size, (IPluginFactory *) pluginFactory);
engineRT = std::shared_ptr<nvinfer1::ICudaEngine>(runtimeRT->deserializeCudaEngine(gieModelStream,size,(IPluginFactory*)pluginFactory),InferDeleter());
//if (gieModelStream) delete [] gieModelStream; //if (gieModelStream) delete [] gieModelStream;
return true; return true;
@@ -673,7 +669,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
} }
if(name.find("ActivationCReLU") == 0) { if(name.find("ActivationCReLU") == 0) {
float activationReluTemp = readBUF<float>(buf); float activationReluTemp = readBUF<float>(buf);
//ActivationReLUCeiling *a = new ActivationReLUCeiling(readBUF<float>(buf));
ActivationReLUCeiling* a = new ActivationReLUCeiling(activationReluTemp); ActivationReLUCeiling* a = new ActivationReLUCeiling(activationReluTemp);
a->size = readBUF<int>(buf); a->size = readBUF<int>(buf);
assert(buf == bufCheck + serialLength); assert(buf == bufCheck + serialLength);
@@ -684,9 +679,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
int classesTemp = readBUF<int>(buf); int classesTemp = readBUF<int>(buf);
int coordsTemp = readBUF<int>(buf); int coordsTemp = readBUF<int>(buf);
int numTemp = readBUF<int>(buf); int numTemp = readBUF<int>(buf);
/*RegionRT *r = new RegionRT(readBUF<int>(buf), //classes
readBUF<int>(buf), //coords
readBUF<int>(buf)); //num8*/
RegionRT* r = new RegionRT(classesTemp, coordsTemp, numTemp); RegionRT* r = new RegionRT(classesTemp, coordsTemp, numTemp);
r->c = readBUF<int>(buf); r->c = readBUF<int>(buf);
@@ -698,7 +690,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
if(name.find("Reorg") == 0) { if(name.find("Reorg") == 0) {
int strideTemp = readBUF<int>(buf); int strideTemp = readBUF<int>(buf);
//ReorgRT *r = new ReorgRT(readBUF<int>(buf)); //stride
ReorgRT *r = new ReorgRT(strideTemp); ReorgRT *r = new ReorgRT(strideTemp);
r->c = readBUF<int>(buf); r->c = readBUF<int>(buf);
r->h = readBUF<int>(buf); r->h = readBUF<int>(buf);
@@ -723,15 +714,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
} }
if(name.find("Pooling") == 0) { 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 cTemp = readBUF<int>(buf);
int hTemp = readBUF<int>(buf); int hTemp = readBUF<int>(buf);
int wTemp = readBUF<int>(buf); int wTemp = readBUF<int>(buf);
@@ -747,9 +729,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
} }
if(name.find("Resize") == 0) { 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_cTemp = readBUF<int>(buf);
int o_hTemp = readBUF<int>(buf); int o_hTemp = readBUF<int>(buf);
int o_wTemp = readBUF<int>(buf); int o_wTemp = readBUF<int>(buf);
@@ -822,7 +801,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
return r; return r;
} }
if(name.find("Upsample") == 0) { if(name.find("Upsample") == 0) {
//UpsampleRT *r = new UpsampleRT(readBUF<int>(buf)); //stride
int strideTemp = readBUF<int>(buf); int strideTemp = readBUF<int>(buf);
UpsampleRT* r = new UpsampleRT(strideTemp); UpsampleRT* r = new UpsampleRT(strideTemp);
r->c = readBUF<int>(buf); r->c = readBUF<int>(buf);
@@ -833,7 +811,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
} }
if(name.find("Route") == 0) { if(name.find("Route") == 0) {
//RouteRT *r = new RouteRT(readBUF<int>(buf),readBUF<int>(buf));
int groupsTemp = readBUF<int>(buf); int groupsTemp = readBUF<int>(buf);
int group_idTemp = readBUF<int>(buf); int group_idTemp = readBUF<int>(buf);
RouteRT* r = new RouteRT(groupsTemp, group_idTemp); RouteRT* r = new RouteRT(groupsTemp, group_idTemp);
@@ -848,12 +825,6 @@ IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialDa
} }
if(name.find("Deformable") == 0) { 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 chuck_dimTemp = readBUF<int>(buf);
int khTemp = readBUF<int>(buf); int khTemp = readBUF<int>(buf);
int kwTemp = readBUF<int>(buf); int kwTemp = readBUF<int>(buf);
src/utils.cpp
+1 -2
View File
@@ -203,8 +203,7 @@ void getMemUsage(double& vm_usage_kb, double& resident_set_kb){
#ifdef __linux__ #ifdef __linux__
long page_size_kb = sysconf(_SC_PAGE_SIZE) / 1024; // in case x86-64 is configured to use 2MB pages long page_size_kb = sysconf(_SC_PAGE_SIZE) / 1024; // in case x86-64 is configured to use 2MB pages
#elif _WIN32 #elif _WIN32
SYSTEM_INFO sysInfo; long page_size_kb = 4096/1024;
long page_size_kb = sysInfo.dwPageSize/1024;
#endif #endif
vm_usage_kb = vsize / 1024.0; vm_usage_kb = vsize / 1024.0;