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

CMakeLists.txt

@@ -2,7 +2,13 @@ cmake_minimum_required(VERSION 3.5)
 
 project (tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
+if(LINUX)
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declarations -Wno-unused-variable")
+endif()
+if(WIN32)
+set(CMAKE_CXX_STANDARD 14)
+set(CMAKE_CXX_FLAGS "/O2 /FS ")
+endif()
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include/tkDNN)
 
 # project specific flags
@@ -18,7 +24,7 @@ add_definitions(-DTKDNN_PATH="${CMAKE_CURRENT_SOURCE_DIR}")
 find_package(CUDA 9.0 REQUIRED)
 SET(CUDA_SEPARABLE_COMPILATION ON)
 #set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -arch=sm_30 --compiler-options '-fPIC'")
-set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32)
+set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --maxrregcount=32 -arch=sm_61 )
 
 find_package(CUDNN REQUIRED)
 include_directories(${CUDNN_INCLUDE_DIR})
@@ -28,6 +34,7 @@ include_directories(${CUDNN_INCLUDE_DIR})
 file(GLOB tkdnn_CUSRC "src/kernels/*.cu" "src/sorting.cu")
 cuda_include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include ${CUDA_INCLUDE_DIRS} ${CUDNN_INCLUDE_DIRS})
 cuda_add_library(kernels SHARED ${tkdnn_CUSRC})
+target_link_libraries(kernels ${CUDA_CUBLAS_LIBRARIES})
 
 
 #-------------------------------------------------------------------------------
@@ -48,7 +55,7 @@ set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DOPENCV")
 file(GLOB tkdnn_SRC "src/*.cpp")
 set(tkdnn_LIBS kernels ${CUDA_LIBRARIES} ${CUDA_CUBLAS_LIBRARIES} ${CUDNN_LIBRARIES} ${OpenCV_LIBS} yaml-cpp)
 
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
 include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include ${CUDA_INCLUDE_DIRS} ${OPENCV_INCLUDE_DIRS} ${NVINFER_INCLUDES})
 add_library(tkDNN SHARED ${tkdnn_SRC})
 target_link_libraries(tkDNN ${tkdnn_LIBS})

include/tkDNN/DetectionNN.h

@@ -4,7 +4,12 @@
 #include <iostream>
 #include <signal.h>
 #include <stdlib.h>    
+#ifdef __linux__
 #include <unistd.h>
+#elif _WIN32
+#include <Windows.h>
+#endif 
+
 #include <mutex>
 #include "utils.h"
 

include/tkDNN/ImuOdom.h

@@ -1,7 +1,13 @@
 #include <iostream>
 #include <signal.h>
 #include <stdlib.h>     /* srand, rand */
+
+#ifdef __linux__
 #include <unistd.h>
+#elif _WIN32
+#include <Windows.h>
+#endif 
+
 #include <mutex>
 #include <Eigen/Dense>
 #include "utils.h"

include/tkDNN/Int8BatchStream.h

@@ -12,7 +12,12 @@
 #include <iomanip>
 #include <signal.h>
 #include <stdlib.h>
+#ifdef __linux__    
 #include <unistd.h>
+#elif _WIN32
+#include <Windows.h>
+#endif
+
 #include <mutex>
 
 #include "NvInfer.h"

include/tkDNN/utils.h

@@ -12,7 +12,12 @@
 #include <cublas_v2.h>
 #include <cudnn.h>
 
+#ifdef __linux__
 #include <unistd.h>
+#elif _WIN32
+#include <Windows.h>
+#endif 
+
 #include <ios>
 
 
@@ -39,6 +44,7 @@
 #define TKDNN_VERBOSE 0
 
 // Simple Timer 
+#ifdef __linux__
 #define TKDNN_TSTART timespec start, end;                               \
                     clock_gettime(CLOCK_MONOTONIC, &start);            
 
@@ -48,6 +54,8 @@
     if(show) std::cout<<col<<"Time:"<<std::setw(16)<<t_ns<<" ms\n"<<COL_END; 
 
 #define TKDNN_TSTOP TKDNN_TSTOP_C(COL_CYANB, TKDNN_VERBOSE)
+#elif _WIN32
+#endif 
 
 /********************************************************
  * Prints the error message, and exits

src/LSTM.cpp

@@ -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,
+    checkCUDNN(cudnnSetRNNDescriptor(net->cudnnHandle,rnnDesc, stateSize, numLayers, dropoutDesc,
-#endif
-        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

@@ -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;
 }