Fixed bug for 32-bit compilation without GPU.

build/darknet/darknet_no_gpu.vcxproj

@@ -78,9 +78,15 @@
       <WarningLevel>Level3</WarningLevel>
       <Optimization>Disabled</Optimization>
       <SDLCheck>true</SDLCheck>
+      <AdditionalIncludeDirectories>C:\opencv_3.0\opencv\build\include;..\..\3rdparty\include;%(AdditionalIncludeDirectories);</AdditionalIncludeDirectories>
+      <PreprocessorDefinitions>_CRTDBG_MAP_ALLOC;_MBCS;_TIMESPEC_DEFINED;_SCL_SECURE_NO_WARNINGS;_CRT_SECURE_NO_WARNINGS;_CRT_RAND_S;WIN32;NDEBUG;_CONSOLE;_LIB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <UndefinePreprocessorDefinitions>OPENCV;</UndefinePreprocessorDefinitions>
+      <ForcedIncludeFiles>stdlib.h;crtdbg.h;%(ForcedIncludeFiles)</ForcedIncludeFiles>
     </ClCompile>
     <Link>
       <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalLibraryDirectories>C:\opencv_3.0\opencv\build\x64\vc14\lib;C:\opencv_2.4.13\opencv\build\x64\vc12\lib;%(AdditionalLibraryDirectories)</AdditionalLibraryDirectories>
+      <AdditionalDependencies>..\..\3rdparty\lib\x86\pthreadVC2.lib;%(AdditionalDependencies)</AdditionalDependencies>
     </Link>
   </ItemDefinitionGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
@@ -89,9 +95,10 @@
       <Optimization>Disabled</Optimization>
       <SDLCheck>true</SDLCheck>
       <AdditionalIncludeDirectories>C:\opencv_3.0\opencv\build\include;..\..\3rdparty\include;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
-      <PreprocessorDefinitions>_MBCS;OPENCV;_TIMESPEC_DEFINED;_SCL_SECURE_NO_WARNINGS;_CRT_SECURE_NO_WARNINGS;_CRT_RAND_S;WIN32;NDEBUG;_CONSOLE;_LIB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <PreprocessorDefinitions>_MBCS;_TIMESPEC_DEFINED;_SCL_SECURE_NO_WARNINGS;_CRT_SECURE_NO_WARNINGS;_CRT_RAND_S;WIN32;NDEBUG;_CONSOLE;_LIB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
-      <UndefinePreprocessorDefinitions>CUDNN</UndefinePreprocessorDefinitions>
+      <UndefinePreprocessorDefinitions>OPENCV;CUDNN</UndefinePreprocessorDefinitions>
       <OpenMPSupport>true</OpenMPSupport>
+      <ForcedIncludeFiles>stdlib.h;crtdbg.h;%(ForcedIncludeFiles)</ForcedIncludeFiles>
     </ClCompile>
     <Link>
       <GenerateDebugInformation>true</GenerateDebugInformation>
@@ -111,15 +118,17 @@
       <FunctionLevelLinking>true</FunctionLevelLinking>
       <IntrinsicFunctions>true</IntrinsicFunctions>
       <SDLCheck>true</SDLCheck>
-      <AdditionalIncludeDirectories>C:\opencv_2.4.9\opencv\build\include;..\..\..\3rdparty\include;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir);$(cudnn)\include</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>C:\opencv_3.0\opencv\build\include;..\..\3rdparty\include;%(AdditionalIncludeDirectories);</AdditionalIncludeDirectories>
-      <PreprocessorDefinitions>OPENCV;_CRT_SECURE_NO_WARNINGS;GPU;WIN32;NDEBUG;_CONSOLE;_LIB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <PreprocessorDefinitions>_TIMESPEC_DEFINED;_CRT_SECURE_NO_WARNINGS;WIN32;NDEBUG;_CONSOLE;_LIB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <UndefinePreprocessorDefinitions>
+      </UndefinePreprocessorDefinitions>
     </ClCompile>
     <Link>
       <GenerateDebugInformation>true</GenerateDebugInformation>
       <EnableCOMDATFolding>true</EnableCOMDATFolding>
       <OptimizeReferences>true</OptimizeReferences>
-      <AdditionalLibraryDirectories>C:\opencv_2.4.9\opencv\build\x64\vc12\lib;$(CUDA_PATH)lib\$(PlatformName);$(cudnn)\lib\x64;%(AdditionalLibraryDirectories)</AdditionalLibraryDirectories>
+      <AdditionalLibraryDirectories>C:\opencv_3.0\opencv\build\x64\vc14\lib;C:\opencv_2.4.13\opencv\build\x64\vc12\lib;%(AdditionalLibraryDirectories)</AdditionalLibraryDirectories>
-      <AdditionalDependencies>..\..\..\3rdparty\lib\x64\pthreadVC2.lib;cublas.lib;curand.lib;cudart.lib;cudnn.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>..\..\3rdparty\lib\x86\pthreadVC2.lib;%(AdditionalDependencies)</AdditionalDependencies>
     </Link>
   </ItemDefinitionGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">

src/convolutional_layer.c

@@ -615,7 +615,7 @@ void binary_align_weights(convolutional_layer *l)
     binarize_weights(l->weights, m, k, l->binary_weights);
 
     size_t align_weights_size = new_lda * m;
-    l->align_bit_weights_size = align_weights_size / 8;// +1;
+    l->align_bit_weights_size = align_weights_size / 8 + 1;
     float *align_weights = calloc(align_weights_size, sizeof(float));
     l->align_bit_weights = calloc(l->align_bit_weights_size, sizeof(char));
 

src/gemm.c

@@ -1719,6 +1719,25 @@ void convolution_2d(int w, int h, int ksize, int n, int c, int pad, int stride,
     }
 }
 
+static inline int popcnt_64(uint64_t val64) {
+#ifdef WIN32  // Windows
+#ifdef _WIN64 // Windows 64-bit
+    int tmp_count = __popcnt64(val64);
+#else         // Windows 32-bit
+    int tmp_count = __popcnt(val64);
+    tmp_count += __popcnt(val64 >> 32);
+#endif
+#else   // Linux
+#ifdef __x86_64__  // Linux 64-bit
+    int tmp_count = __builtin_popcountll(val64);
+#else  // Linux 32-bit
+    int tmp_count = __builtin_popcount(val64);
+    tmp_count += __builtin_popcount(val64);
+#endif
+#endif
+    return tmp_count;
+}
+
 void gemm_nn_custom_bin_mean_transposed(int M, int N, int K, float ALPHA_UNUSED,
     unsigned char *A, int lda,
     unsigned char *B, int ldb,
@@ -1739,11 +1758,7 @@ void gemm_nn_custom_bin_mean_transposed(int M, int N, int K, float ALPHA_UNUSED,
                 uint64_t b_bit64 = *((uint64_t *)(B + (j*ldb + k) / 8));
                 uint64_t c_bit64 = xnor_int64(a_bit64, b_bit64);
 
-#ifdef WIN32
+                int tmp_count = popcnt_64(c_bit64);
-                int tmp_count = __popcnt64(c_bit64);
-#else
-                int tmp_count = __builtin_popcountll(c_bit64);
-#endif
 
                 if (K - k < 64)  tmp_count = tmp_count - (64 - (K - k));    // remove extra bits
                 count += tmp_count;

src/layer.c

@@ -33,8 +33,8 @@ void free_layer(layer l)
 	if (l.scale_updates)      free(l.scale_updates);
 	if (l.weights)            free(l.weights);
 	if (l.weight_updates)     free(l.weight_updates);
-    if (l.weights)            free(l.align_bit_weights);
+    if (l.align_bit_weights)  free(l.align_bit_weights);
-    if (l.weights)            free(l.mean_arr);
+    if (l.mean_arr)           free(l.mean_arr);
 	if (l.delta)              free(l.delta);
 	if (l.output)             free(l.output);
 	if (l.squared)            free(l.squared);
@@ -84,6 +84,12 @@ void free_layer(layer l)
 	if (l.mean_delta_gpu)          cuda_free(l.mean_delta_gpu);
 	if (l.x_gpu)                   cuda_free(l.x_gpu);
 	if (l.x_norm_gpu)              cuda_free(l.x_norm_gpu);
+
+    if (l.align_bit_weights_gpu)   cuda_free(l.align_bit_weights_gpu);
+    if (l.mean_arr_gpu)            cuda_free(l.mean_arr_gpu);
+    if (l.align_workspace_gpu)     cuda_free(l.align_workspace_gpu);
+    if (l.transposed_align_workspace_gpu) cuda_free(l.transposed_align_workspace_gpu);
+
 	if (l.weights_gpu)             cuda_free(l.weights_gpu);
 	if (l.weight_updates_gpu)      cuda_free(l.weight_updates_gpu);
 	if (l.weights_gpu16)           cuda_free(l.weights_gpu16);

src/yolo_layer.c

@@ -374,17 +374,19 @@ int get_yolo_detections(layer l, int w, int h, int netw, int neth, float thresh,
         for(n = 0; n < l.n; ++n){
             int obj_index  = entry_index(l, 0, n*l.w*l.h + i, 4);
             float objectness = predictions[obj_index];
-            if(objectness <= thresh) continue;
+            //if(objectness <= thresh) continue;
-            int box_index  = entry_index(l, 0, n*l.w*l.h + i, 0);
+            if (objectness > thresh) {
-            dets[count].bbox = get_yolo_box(predictions, l.biases, l.mask[n], box_index, col, row, l.w, l.h, netw, neth, l.w*l.h);
+                int box_index = entry_index(l, 0, n*l.w*l.h + i, 0);
-            dets[count].objectness = objectness;
+                dets[count].bbox = get_yolo_box(predictions, l.biases, l.mask[n], box_index, col, row, l.w, l.h, netw, neth, l.w*l.h);
-            dets[count].classes = l.classes;
+                dets[count].objectness = objectness;
-            for(j = 0; j < l.classes; ++j){
+                dets[count].classes = l.classes;
-                int class_index = entry_index(l, 0, n*l.w*l.h + i, 4 + 1 + j);
+                for (j = 0; j < l.classes; ++j) {
-                float prob = objectness*predictions[class_index];
+                    int class_index = entry_index(l, 0, n*l.w*l.h + i, 4 + 1 + j);
-                dets[count].prob[j] = (prob > thresh) ? prob : 0;
+                    float prob = objectness*predictions[class_index];
+                    dets[count].prob[j] = (prob > thresh) ? prob : 0;
+                }
+                ++count;
             }
-            ++count;
         }
     }
     correct_yolo_boxes(dets, count, w, h, netw, neth, relative, letter);