Fix compilation for Windows

.github/workflows/build-cpu-windows.yml

@@ -21,7 +21,7 @@ jobs:
       run: |
         mkdir build
         cd build
-        cmake -G "Visual Studio 16 2019" -A x64 -T ClangCL -DBUILD_SUPERBUILD=ON -DBUILD_SHARED_LIBS=OFF -DBUILD_TARGET=CPU ..
+        cmake -G "Visual Studio 16 2019" -A x64 -DBUILD_SUPERBUILD=ON -DBUILD_SHARED_LIBS=OFF -DBUILD_TARGET=CPU ..
       shell: cmd
     - name: Build
       run: cmake --build build --config Release

include/eddl/profiling.h

@@ -1,14 +1,33 @@
-#ifndef _PROFILING
+/*
+* EDDL Library - European Distributed Deep Learning Library.
+* Version: 0.7
+* copyright (c) 2020, Universidad Polit�cnica de Valencia (UPV), PRHLT Research Centre
+* Date: April 2020
+* Author: PRHLT Research Centre, UPV, ([email protected]), ([email protected])
+* All rights reserved
+*/
 
+#ifndef _PROFILING
 #define _PROFILING
 
+#include "eddl/system_info.h"
+
+#ifdef EDDL_WINDOWS
+
+#include <chrono>
+#include <windows.h>
+#include <winsock.h>
+
+int gettimeofday(struct timeval* tp, struct timezone* tzp);
+
+#else
 #include <sys/time.h>
+#endif
 
 #define PROFILING_HEADER(fn) \
     struct timeval prof_t1; \
     gettimeofday(&prof_t1, NULL);
 
-
 #define PROFILING_ENABLE(fn) \
     unsigned long long prof_##fn##_time; \
     unsigned long long prof_##fn##_calls;
@@ -44,13 +63,10 @@
             100.0 * prof_##fn##_time / acc, (float) prof_##fn##_time / (float) prof_##fn##_calls);
 #endif
 
-
-
 //CxHxW
 //
 //HxWxC
 //
 //GxHxWxC (C=4)   Reshape + Permute
 //
 //32xHxW -> Reshape -> 8x4xHxW -> Permute(0, 2, 3, 1) -> 8xHxWx4   // hay capas y funciones
-

src/CMakeLists.txt

@@ -244,10 +244,10 @@ if(BUILD_OPENMP)
 #            endif()
 #        endif()
 
-        find_package(OpenMP 4.5)
-        if (OpenMP_CXX_FOUND)
+        find_package(OpenMP)
+        if(OpenMP_CXX_FOUND)
             target_link_libraries(${PROJECT_NAME} PUBLIC OpenMP::OpenMP_CXX)
-            if (${OpenMP_CXX_VERSION_MAJOR})
+            if(${OpenMP_CXX_VERSION_MAJOR})
                 set(OpenMP_VERSION_MAJOR ${OpenMP_CXX_VERSION_MAJOR} CACHE INTERNAL "" FORCE)
             endif()
             target_compile_definitions(${PROJECT_NAME} PUBLIC OpenMP_VERSION_MAJOR=${OpenMP_VERSION_MAJOR})

src/hardware/cpu/cpu_comparison.cpp

@@ -9,6 +9,7 @@
 
 
 #include "eddl/hardware/cpu/cpu_tensor.h"
+#include "eddl/system_info.h"
 #include <limits>
 
 // CPU: Logic functions: Truth value testing
@@ -24,7 +25,9 @@ bool cpu_all(Tensor *A){
             {
                 res = false;
             }
+#if defined(EDDL_LINUX) || defined(EDDL_UNIX) || defined(EDDL_APPLE)
             #pragma omp cancel for
+#endif
         }
     }
     _profile(_CPU_ALL, 1);
@@ -44,8 +47,9 @@ bool cpu_any(Tensor *A){
             {
                 res = true;
             }
+#if defined(EDDL_LINUX) || defined(EDDL_UNIX) || defined(EDDL_APPLE)
             #pragma omp cancel for
-
+#endif
         }
     }
     _profile(_CPU_ANY, 1);
@@ -155,7 +159,9 @@ bool cpu_allclose(Tensor *A, Tensor *B, float rtol, float atol, bool equal_nan){
                 if(first_idx < 0) { first_idx=i; }
 
             }
+#if defined(EDDL_LINUX) || defined(EDDL_UNIX) || defined(EDDL_APPLE)
             #pragma omp cancel for
+#endif
         }
     }
     _profile(_CPU_ALLCLOSE, 1);

src/hardware/cpu/cpu_core.cpp

@@ -8,29 +8,10 @@
 */
 
 #include "eddl/hardware/cpu/cpu_tensor.h"
+#include "eddl/profiling.h"
 #include <algorithm>
 #include <numeric>
 
-#if defined(WIN32) || defined(_WIN32) || defined(__WIN32__) || defined(__NT__)
-
-    #include <chrono>
-    #include <windows.h>
-    #include <winsock.h>
-
-    int gettimeofday(struct timeval* tp, struct timezone* tzp)
-    {
-        namespace sc = std::chrono;
-        sc::system_clock::duration d = sc::system_clock::now().time_since_epoch();
-        sc::seconds s = sc::duration_cast<sc::seconds>(d);
-        tp->tv_sec = s.count();
-        tp->tv_usec = sc::duration_cast<sc::microseconds>(d - s).count();
-
-        return 0;
-    }
-#else
-    #include <sys/time.h>
-#endif
-
 int num_instances[_NUM_CPU_FUNCS];
 float mb_memory_needed;
 

src/hardware/cpu/nn/cpu_losses.cpp

@@ -31,7 +31,7 @@ float cpu_categorical_cross_entropy(Tensor* y_true, Tensor* y_pred){
     float eps = 10e-8;
 
     #pragma omp parallel for reduction(+:sum)
-    for (unsigned int bi = 0; bi<y_true->shape[0]; bi++) {  // Batches
+    for (long long int bi = 0; bi<y_true->shape[0]; bi++) {  // Batches
         unsigned int step_i = bi * y_true->stride[0];
 
         // Compute cross-entropy
@@ -51,7 +51,7 @@ void cpu_d_categorical_cross_entropy(Tensor* y_true, Tensor* y_pred, Tensor* del
     float eps = 10e-8;
 
     #pragma omp parallel for
-    for (unsigned int i = 0; i<y_true->size; i++) {
+    for (long long int i = 0; i<y_true->size; i++) {
         delta->ptr[i] = -y_true->ptr[i] * (1.0f/ (y_pred->ptr[i]+eps) );
     }
 }
@@ -61,7 +61,7 @@ float cpu_binary_cross_entropy(Tensor* y_true, Tensor* y_pred){
     float eps = 10e-8;
 
     #pragma omp parallel for reduction(+:sum)
-    for (unsigned int i = 0; i<y_true->size; i++) {
+    for (long long int i = 0; i < y_true->size; i++) {
         sum += y_true->ptr[i] * ::logf(y_pred->ptr[i]+eps) + (1.0-y_true->ptr[i]) * ::logf(1.0f-y_pred->ptr[i]+eps);
     }
 
@@ -74,7 +74,7 @@ void cpu_d_binary_cross_entropy(Tensor* y_true, Tensor* y_pred, Tensor* delta){
     float eps = 10e-8;
 
     #pragma omp parallel for
-    for (unsigned int i = 0; i<y_true->size; i++) {
+    for (long long int i = 0; i<y_true->size; i++) {
         delta->ptr[i] = -( y_true->ptr[i] * 1.0f/(y_pred->ptr[i]+eps) + (1.0-y_true->ptr[i]) * 1.0f/(1.0f-y_pred->ptr[i]+eps) * -1.0f );
     }
 }

src/net/net_api.cpp

@@ -15,10 +15,11 @@
 #include <string>
 #include <chrono>
 #include <stdexcept>
+#include "eddl/layers/core/layer_core.h"
 #include "eddl/net/net.h"
-#include "eddl/utils.h"
 #include "eddl/random.h"
-#include "eddl/layers/core/layer_core.h"
+#include "eddl/system_info.h"
+#include "eddl/utils.h"
 
 
 #ifdef cFPGA
@@ -151,9 +152,13 @@ void Net::run_snets(void *(*F)(void *t))
   int rc;
   struct tdata td[100];
 
-  int comp=snets.size();
+  int comp = snets.size();
 
+#ifdef EDDL_WINDOWS
+  #pragma omp parallel for
+#else
   #pragma omp taskloop num_tasks(comp)
+#endif
   for (int i = 0; i < comp; i++) {
     // Thread params
     td[i].net = snets[i];

src/profiling.cpp

@@ -0,0 +1,23 @@
+/*
+* EDDL Library - European Distributed Deep Learning Library.
+* Version: 0.7
+* copyright (c) 2020, Universidad Politécnica de Valencia (UPV), PRHLT Research Centre
+* Date: April 2020
+* Author: PRHLT Research Centre, UPV, ([email protected]), ([email protected])
+* All rights reserved
+*/
+
+#include "eddl/profiling.h"
+
+#ifdef EDDL_WINDOWS
+int gettimeofday(struct timeval* tp, struct timezone* tzp)
+{
+    namespace sc = std::chrono;
+    sc::system_clock::duration d = sc::system_clock::now().time_since_epoch();
+    sc::seconds s = sc::duration_cast<sc::seconds>(d);
+    tp->tv_sec = s.count();
+    tp->tv_usec = sc::duration_cast<sc::microseconds>(d - s).count();
+
+    return 0;
+}
+#endif