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

able to build kernels as shared object file(dll),and minor changes to lstm.cpp and utils.cpp to overcome minor msvc build errors

Browse Source
This commit is contained in:
perseusdg
2021-01-21 00:22:15 +04:00
parent 4fa008bd85
commit b994eb2a4e
7 changed files with 49 additions and 8 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/utils.cpp
+6 -1
View File
@@ -170,6 +170,7 @@ void getMemUsage(double& vm_usage_kb, double& resident_set_kb){
using std::ios_base;
using std::ifstream;
using std::string;
SYSTEM_INFO sysInfo;
vm_usage_kb = 0.0;
resident_set_kb = 0.0;
@@ -191,8 +192,12 @@ 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 = sysInfo.dwPageSize/1024;
#endif
vm_usage_kb = vsize / 1024.0;
resident_set_kb = rss * page_size_kb;
}