Moved the CUDA stuff into its own folder.

dlib/CMakeLists.txt

@@ -254,8 +254,8 @@ if (NOT TARGET dlib)
          threads/async.cpp
          timer/timer.cpp
          stack_trace.cpp
-         dnn/cpu_dlib.cpp
-         dnn/tensor_tools.cpp
+         cuda/cpu_dlib.cpp
+         cuda/tensor_tools.cpp
          )
 
       if(UNIX)
@@ -657,13 +657,13 @@ if (NOT TARGET dlib)
 
          if (CUDA_FOUND AND cudnn AND (NOT USING_OLD_VISUAL_STUDIO_COMPILER) AND cuda_test_compile_worked AND cudnn_test_compile_worked AND cudnn_include)
             set(source_files ${source_files} 
-               dnn/cuda_dlib.cu 
-               dnn/cudnn_dlibapi.cpp
-               dnn/cublas_dlibapi.cpp
-               dnn/cusolver_dlibapi.cu
-               dnn/curand_dlibapi.cpp
-               dnn/cuda_data_ptr.cpp
-               dnn/gpu_data.cpp
+               cuda/cuda_dlib.cu 
+               cuda/cudnn_dlibapi.cpp
+               cuda/cublas_dlibapi.cpp
+               cuda/cusolver_dlibapi.cu
+               cuda/curand_dlibapi.cpp
+               cuda/cuda_data_ptr.cpp
+               cuda/gpu_data.cpp
                )
             set(dlib_needed_libraries ${dlib_needed_libraries} 
                ${CUDA_CUBLAS_LIBRARIES} 

dlib/all/source.cpp

@@ -79,8 +79,8 @@
 #include "../gui_core/gui_core_kernel_2.cpp"
 #endif // DLIB_NO_GUI_SUPPORT
 
-#include "../dnn/cpu_dlib.cpp"
-#include "../dnn/tensor_tools.cpp"
+#include "../cuda/cpu_dlib.cpp"
+#include "../cuda/tensor_tools.cpp"
 
 #endif // DLIB_ISO_CPP_ONLY
 

dlib/cmake_utils/test_for_cuda/CMakeLists.txt

@@ -2,7 +2,7 @@
 cmake_minimum_required(VERSION 2.8.12)
 project(cuda_test)
 
-include_directories(../../dnn)
+include_directories(../../cuda)
 add_definitions(-DDLIB_USE_CUDA)
 
 # Override the FindCUDA.cmake setting to avoid duplication of host flags if using a toolchain:

dlib/cmake_utils/test_for_cudnn/CMakeLists.txt

@@ -14,6 +14,6 @@ include(find_cudnn.txt)
 
 if (cudnn_include AND cudnn)
    include_directories(${cudnn_include})
-   cuda_add_library(cudnn_test STATIC ../../dnn/cudnn_dlibapi.cpp ${cudnn} )
+   cuda_add_library(cudnn_test STATIC ../../cuda/cudnn_dlibapi.cpp ${cudnn} )
    enable_cpp11_for_target(cudnn_test)
 endif()

dlib/dnn.h

@@ -9,7 +9,7 @@
 #   pragma warning( disable: 4503 )
 #endif
 
-#include "dnn/tensor.h"
+#include "cuda/tensor.h"
 #include "dnn/input.h"
 
 // Problem:    Visual Studio's vcpkgsrv.exe constantly uses a single CPU core,
@@ -27,8 +27,8 @@
 #include "dnn/core.h"
 #include "dnn/solvers.h"
 #include "dnn/trainer.h"
-#include "dnn/cpu_dlib.h"
-#include "dnn/tensor_tools.h"
+#include "cuda/cpu_dlib.h"
+#include "cuda/tensor_tools.h"
 #include "dnn/utilities.h"
 #include "dnn/validation.h"
 

dlib/dnn/core.h

@@ -4,7 +4,7 @@
 #define DLIB_DNn_CORE_H_
 
 #include "core_abstract.h"
-#include "tensor.h"
+#include "../cuda/tensor.h"
 #include <iterator>
 #include <memory>
 #include <sstream>
@@ -16,7 +16,7 @@
 #include <tuple>
 #include <cmath>
 #include <vector>
-#include "tensor_tools.h"
+#include "../cuda/tensor_tools.h"
 #include <type_traits>
 #include "../metaprogramming.h"
 

dlib/dnn/core_abstract.h

@@ -3,7 +3,7 @@
 #undef DLIB_DNn_CORE_ABSTRACT_H_
 #ifdef DLIB_DNn_CORE_ABSTRACT_H_
 
-#include "tensor_abstract.h"
+#include "../cuda/tensor_abstract.h"
 #include <memory>
 #include <type_traits>
 #include <tuple>

dlib/dnn/input.h

@@ -10,7 +10,7 @@
 #include "../image_processing.h"
 #include <sstream>
 #include <array>
-#include "tensor_tools.h"
+#include "../cuda/tensor_tools.h"
 
 
 namespace dlib

dlib/dnn/layers.h

@@ -4,13 +4,13 @@
 #define DLIB_DNn_LAYERS_H_
 
 #include "layers_abstract.h"
-#include "tensor.h"
+#include "../cuda/tensor.h"
 #include "core.h"
 #include <iostream>
 #include <string>
 #include "../rand.h"
 #include "../string.h"
-#include "tensor_tools.h"
+#include "../cuda/tensor_tools.h"
 #include "../vectorstream.h"
 #include "utilities.h"
 #include <sstream>

dlib/dnn/layers_abstract.h

@@ -3,7 +3,7 @@
 #undef DLIB_DNn_LAYERS_ABSTRACT_H_
 #ifdef DLIB_DNn_LAYERS_ABSTRACT_H_
 
-#include "tensor_abstract.h"
+#include "../cuda/tensor_abstract.h"
 #include "core_abstract.h"
 
 

dlib/dnn/loss.h

@@ -6,7 +6,7 @@
 #include "loss_abstract.h"
 #include "core.h"
 #include "../matrix.h"
-#include "tensor_tools.h"
+#include "../cuda/tensor_tools.h"
 #include "../geometry.h"
 #include "../image_processing/box_overlap_testing.h"
 #include "../image_processing/full_object_detection.h"

dlib/dnn/solvers.h

@@ -4,7 +4,7 @@
 #define DLIB_DNn_SOLVERS_H_
 
 #include "solvers_abstract.h"
-#include "tensor.h"
+#include "../cuda/tensor.h"
 #include <iostream>
 #include "layers.h"
 

dlib/dnn/solvers_abstract.h

@@ -3,7 +3,7 @@
 #undef DLIB_DNn_SOLVERS_ABSTRACT_H_
 #ifdef DLIB_DNn_SOLVERS_ABSTRACT_H_
 
-#include "tensor_abstract.h"
+#include "../cuda/tensor_abstract.h"
 #include <iostream>
 
 namespace dlib

dlib/dnn/trainer.h

@@ -14,7 +14,7 @@
 
 #include "../pipe.h"
 #include "../threads.h"
-#include "cuda_dlib.h"
+#include "../cuda/cuda_dlib.h"
 #include "../statistics/running_gradient.h"
 #include <atomic>
 #include <cstdio>

dlib/test/cublas.cpp

@@ -8,7 +8,7 @@
 #include <cstdlib>
 #include <ctime>
 #include <vector>
-#include "../dnn/tensor_tools.h"
+#include "../cuda/tensor_tools.h"
 
 #include "tester.h"