automatically detect CUDA and use it, without any possible manual intervention

6 years ago · f950085a45
parent 9c64f885a8
commit f950085a45
5 changed files with 48 additions and 138 deletions
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@ -1,5 +1,4 @@
 # To use your libs, launch cmake with -DUSE_INTEGRATED_LIBS:BOOL=FALSE
-# To enable CUDA, launch cmake with -DENABLE_CUDA. Also it is advisable to set the highest compute model with -DCOMPUTE_MODEL=..., depending on your specific card capabilities. If CUDA is not required, this setting is completely ignored.

 cmake_minimum_required(VERSION 3.8 FATAL_ERROR)

@ -17,12 +16,13 @@ else()
  set(USE_INTEGRATED_LIBS "FALSE" CACHE BOOL "Use libs distributed with this repo")
 endif()

-set(ENABLE_CUDA "FALSE" CACHE BOOL "Enable building with CUDA")
-
 enable_language(C)
 enable_language(CXX)
-if(ENABLE_CUDA)
-  enable_language(CUDA) # do not enable after searching for OpenCV package!
+
+include(CheckLanguage)
+check_language(CUDA)
+if(CMAKE_CUDA_COMPILER)
+  enable_language(CUDA)
 endif()

 set(CMAKE_CXX_STANDARD 11)
@ -30,24 +30,24 @@ set(CMAKE_MODULE_PATH "${CMAKE_CURRENT_LIST_DIR}/cmake/Modules/" ${CMAKE_MODULE_
 include_directories(${CMAKE_CURRENT_LIST_DIR}/include)
 include_directories(${CMAKE_CURRENT_LIST_DIR}/src)

-set(COMPUTE_MODEL_75 "75") #Compute capability for Tesla T4, RTX 2080
-set(COMPUTE_MODEL_72 "72") #Compute capability for Jetson Xavier
-set(COMPUTE_MODEL_70 "70") #Compute capability for Tesla V100
-set(COMPUTE_MODEL_62 "62") #Compute capability for Jetson TX2
-set(COMPUTE_MODEL_61 "61") #Compute capability for Tesla P40
-set(COMPUTE_MODEL_60 "60") #Compute capability for Tesla P100
-set(COMPUTE_MODEL_53 "53") #Compute capability for Jetson TX1
-set(COMPUTE_MODEL_52 "52") #Compute capability for Tesla M40/M60
-set(COMPUTE_MODEL_37 "37") #Compute capability for Tesla K80
-set(COMPUTE_MODEL_35 "35") #Compute capability for Tesla K20/K40
-set(COMPUTE_MODEL_30 "30") #Compute capability for Tesla K10, Quadro K4000
-
-set(default_compute_model ${COMPUTE_MODEL_30})
-if(NOT COMPUTE_MODEL)
-  message(STATUS "Setting the compute model to '${default_compute_model}' as none was specified.")
-  set(COMPUTE_MODEL "${default_compute_model}" CACHE
+set(CUDA_COMPUTE_MODEL_75 "75") #Compute capability for Tesla T4, RTX 2080
+set(CUDA_COMPUTE_MODEL_72 "72") #Compute capability for Jetson Xavier
+set(CUDA_COMPUTE_MODEL_70 "70") #Compute capability for Tesla V100
+set(CUDA_COMPUTE_MODEL_62 "62") #Compute capability for Jetson TX2
+set(CUDA_COMPUTE_MODEL_61 "61") #Compute capability for Tesla P40
+set(CUDA_COMPUTE_MODEL_60 "60") #Compute capability for Tesla P100
+set(CUDA_COMPUTE_MODEL_53 "53") #Compute capability for Jetson TX1
+set(CUDA_COMPUTE_MODEL_52 "52") #Compute capability for Tesla M40/M60
+set(CUDA_COMPUTE_MODEL_37 "37") #Compute capability for Tesla K80
+set(CUDA_COMPUTE_MODEL_35 "35") #Compute capability for Tesla K20/K40
+set(CUDA_COMPUTE_MODEL_30 "30") #Compute capability for Tesla K10, Quadro K4000
+
+set(default_cuda_compute_model ${CUDA_COMPUTE_MODEL_30})
+if(CMAKE_CUDA_COMPILER AND NOT CUDA_COMPUTE_MODEL)
+  message(STATUS "Setting the compute model to ${default_cuda_compute_model} as none was specified.")
+  set(CUDA_COMPUTE_MODEL "${default_cuda_compute_model}" CACHE
      STRING "Choose the best compute model supported by your GPU" FORCE)
-  set_property(CACHE COMPUTE_MODEL PROPERTY STRINGS ${COMPUTE_MODEL_75} ${COMPUTE_MODEL_72} ${COMPUTE_MODEL_70} ${COMPUTE_MODEL_62} ${COMPUTE_MODEL_61} ${COMPUTE_MODEL_60} ${COMPUTE_MODEL_53} ${COMPUTE_MODEL_52} ${COMPUTE_MODEL_37} ${COMPUTE_MODEL_35} ${COMPUTE_MODEL_30})
+  set_property(CACHE CUDA_COMPUTE_MODEL PROPERTY STRINGS ${CUDA_COMPUTE_MODEL_75} ${CUDA_COMPUTE_MODEL_72} ${CUDA_COMPUTE_MODEL_70} ${CUDA_COMPUTE_MODEL_62} ${CUDA_COMPUTE_MODEL_61} ${CUDA_COMPUTE_MODEL_60} ${CUDA_COMPUTE_MODEL_53} ${CUDA_COMPUTE_MODEL_52} ${CUDA_COMPUTE_MODEL_37} ${CUDA_COMPUTE_MODEL_35} ${CUDA_COMPUTE_MODEL_30})
 endif()

 if(USE_INTEGRATED_LIBS)
@ -108,17 +108,17 @@ if(OpenCV_FOUND)
  include_directories(${OpenCV_INCLUDE_DIRS})
 endif()

-if(ENABLE_CUDA)
+if(CMAKE_CUDA_COMPILER)
  add_definitions(-DGPU)
-  if(COMPUTE_MODEL VERSION_GREATER_EQUAL 75 AND CMAKE_CUDA_COMPILER_VERSION VERSION_LESS_EQUAL 9)
+  if(CUDA_COMPUTE_MODEL VERSION_GREATER_EQUAL 75 AND CMAKE_CUDA_COMPILER_VERSION VERSION_LESS_EQUAL 9)
    message(STATUS "CUDA 10 is required for Compute Model 7.5 - Lowering settings to 7.0")
-    set(COMPUTE_MODEL "70")
+    set(CUDA_COMPUTE_MODEL ${CUDA_COMPUTE_MODEL_70})
  endif()
  find_package(CUDNN)
  if(CUDNN_FOUND)
    add_definitions(-DCUDNN)
    include_directories(${CUDNN_INCLUDE_DIR})
-    if(COMPUTE_MODEL VERSION_GREATER_EQUAL 53)
+    if(CUDA_COMPUTE_MODEL VERSION_GREATER_EQUAL 53)
      add_definitions(-DCUDNN_HALF)
      message(STATUS "Enabled CUDNN_HALF")
    endif()
@ -126,29 +126,29 @@ if(ENABLE_CUDA)
  if (MSVC)
    if(CUDNN_FOUND)
      if(OpenCV_FOUND)
-        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${COMPUTE_MODEL},code=[sm_${COMPUTE_MODEL},compute_${COMPUTE_MODEL}] -Wno-deprecated-declarations -Xcompiler=\"/wd4028,/wd4190,/wd4244,/wd4267,/wd4305,/wd4477,/wd4996,/DGPU,/DCUDNN,/DOPENCV\" ${CMAKE_CUDA_FLAGS}")
+        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${CUDA_COMPUTE_MODEL},code=[sm_${CUDA_COMPUTE_MODEL},compute_${CUDA_COMPUTE_MODEL}] -Wno-deprecated-declarations -Xcompiler=\"/wd4028,/wd4190,/wd4244,/wd4267,/wd4305,/wd4477,/wd4996,/DGPU,/DCUDNN,/DOPENCV\" ${CMAKE_CUDA_FLAGS}")
      else()
-        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${COMPUTE_MODEL},code=[sm_${COMPUTE_MODEL},compute_${COMPUTE_MODEL}] -Wno-deprecated-declarations -Xcompiler=\"/wd4028,/wd4190,/wd4244,/wd4267,/wd4305,/wd4477,/wd4996,/DGPU,/DCUDNN\" ${CMAKE_CUDA_FLAGS}")
+        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${CUDA_COMPUTE_MODEL},code=[sm_${CUDA_COMPUTE_MODEL},compute_${CUDA_COMPUTE_MODEL}] -Wno-deprecated-declarations -Xcompiler=\"/wd4028,/wd4190,/wd4244,/wd4267,/wd4305,/wd4477,/wd4996,/DGPU,/DCUDNN\" ${CMAKE_CUDA_FLAGS}")
      endif()
    else()
      if(OpenCV_FOUND)
-        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${COMPUTE_MODEL},code=[sm_${COMPUTE_MODEL},compute_${COMPUTE_MODEL}] -Wno-deprecated-declarations -Xcompiler=\"/wd4028,/wd4190,/wd4244,/wd4267,/wd4305,/wd4477,/wd4996,/DGPU,/DOPENCV\" ${CMAKE_CUDA_FLAGS}")
+        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${CUDA_COMPUTE_MODEL},code=[sm_${CUDA_COMPUTE_MODEL},compute_${CUDA_COMPUTE_MODEL}] -Wno-deprecated-declarations -Xcompiler=\"/wd4028,/wd4190,/wd4244,/wd4267,/wd4305,/wd4477,/wd4996,/DGPU,/DOPENCV\" ${CMAKE_CUDA_FLAGS}")
      else()
-        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${COMPUTE_MODEL},code=[sm_${COMPUTE_MODEL},compute_${COMPUTE_MODEL}] -Wno-deprecated-declarations -Xcompiler=\"/wd4028,/wd4190,/wd4244,/wd4267,/wd4305,/wd4477,/wd4996,/DGPU\" ${CMAKE_CUDA_FLAGS}")
+        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${CUDA_COMPUTE_MODEL},code=[sm_${CUDA_COMPUTE_MODEL},compute_${CUDA_COMPUTE_MODEL}] -Wno-deprecated-declarations -Xcompiler=\"/wd4028,/wd4190,/wd4244,/wd4267,/wd4305,/wd4477,/wd4996,/DGPU\" ${CMAKE_CUDA_FLAGS}")
      endif()
    endif()
  else()
    if(CUDNN_FOUND)
      if(OpenCV_FOUND)
-        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${COMPUTE_MODEL},code=[sm_${COMPUTE_MODEL},compute_${COMPUTE_MODEL}] --compiler-options \"-Wall -Wno-unused-result -Wno-unknown-pragmas -Wfatal-errors -Wno-deprecated-declarations -fPIC -fopenmp -Ofast -DOPENCV -DGPU -DCUDNN \" ${CMAKE_CUDA_FLAGS}")
+        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${CUDA_COMPUTE_MODEL},code=[sm_${CUDA_COMPUTE_MODEL},compute_${CUDA_COMPUTE_MODEL}] --compiler-options \"-Wall -Wno-unused-result -Wno-unknown-pragmas -Wfatal-errors -Wno-deprecated-declarations -fPIC -fopenmp -Ofast -DOPENCV -DGPU -DCUDNN \" ${CMAKE_CUDA_FLAGS}")
      else()
-        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${COMPUTE_MODEL},code=[sm_${COMPUTE_MODEL},compute_${COMPUTE_MODEL}] --compiler-options \"-Wall -Wno-unused-result -Wno-unknown-pragmas -Wfatal-errors -Wno-deprecated-declarations -fPIC -fopenmp -Ofast -DGPU -DCUDNN \" ${CMAKE_CUDA_FLAGS}")
+        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${CUDA_COMPUTE_MODEL},code=[sm_${CUDA_COMPUTE_MODEL},compute_${CUDA_COMPUTE_MODEL}] --compiler-options \"-Wall -Wno-unused-result -Wno-unknown-pragmas -Wfatal-errors -Wno-deprecated-declarations -fPIC -fopenmp -Ofast -DGPU -DCUDNN \" ${CMAKE_CUDA_FLAGS}")
      endif()
    else()
      if(OpenCV_FOUND)
-        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${COMPUTE_MODEL},code=[sm_${COMPUTE_MODEL},compute_${COMPUTE_MODEL}] --compiler-options \"-Wall -Wno-unused-result -Wno-unknown-pragmas -Wfatal-errors -Wno-deprecated-declarations -fPIC -fopenmp -Ofast -DOPENCV -DGPU \" ${CMAKE_CUDA_FLAGS}")
+        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${CUDA_COMPUTE_MODEL},code=[sm_${CUDA_COMPUTE_MODEL},compute_${CUDA_COMPUTE_MODEL}] --compiler-options \"-Wall -Wno-unused-result -Wno-unknown-pragmas -Wfatal-errors -Wno-deprecated-declarations -fPIC -fopenmp -Ofast -DOPENCV -DGPU \" ${CMAKE_CUDA_FLAGS}")
      else()
-        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${COMPUTE_MODEL},code=[sm_${COMPUTE_MODEL},compute_${COMPUTE_MODEL}] --compiler-options \"-Wall -Wno-unused-result -Wno-unknown-pragmas -Wfatal-errors -Wno-deprecated-declarations -fPIC -fopenmp -Ofast -DGPU \" ${CMAKE_CUDA_FLAGS}")
+        set(CMAKE_CUDA_FLAGS "-gencode arch=compute_${CUDA_COMPUTE_MODEL},code=[sm_${CUDA_COMPUTE_MODEL},compute_${CUDA_COMPUTE_MODEL}] --compiler-options \"-Wall -Wno-unused-result -Wno-unknown-pragmas -Wfatal-errors -Wno-deprecated-declarations -fPIC -fopenmp -Ofast -DGPU \" ${CMAKE_CUDA_FLAGS}")
      endif()
    endif()
  endif()
@ -195,7 +195,7 @@ list(REMOVE_ITEM sources
  ${CMAKE_CURRENT_LIST_DIR}/src/getopt.c
 )

-if(ENABLE_CUDA)
+if(CMAKE_CUDA_COMPILER)
  file(GLOB cuda_sources "${CMAKE_CURRENT_LIST_DIR}/src/*.cu")
 endif()

@ -215,7 +215,7 @@ set_source_files_properties(${sources} PROPERTIES LANGUAGE CXX)

 add_library(darklib SHARED ${CMAKE_CURRENT_LIST_DIR}/include/yolo_v2_class.hpp ${CMAKE_CURRENT_LIST_DIR}/src/yolo_v2_class.cpp ${sources} ${headers} ${cuda_sources})
 set_target_properties(darklib PROPERTIES POSITION_INDEPENDENT_CODE ON)
-if(ENABLE_CUDA)
+if(CMAKE_CUDA_COMPILER)
  set_target_properties(darklib PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
 endif()
 target_compile_definitions(darklib PRIVATE LIB_EXPORTS=1)
@ -264,7 +264,7 @@ endif()
 target_link_libraries(darknet Threads::Threads)
 target_link_libraries(darklib Threads::Threads)

-if(ENABLE_CUDA)
+if(CMAKE_CUDA_COMPILER)
  target_link_libraries(darknet curand cublas)
  target_link_libraries(darklib curand cublas)
 endif()
--- a/CMakeSettings.json
+++ b/CMakeSettings.json
@ -6,7 +6,7 @@
      "cmakeCommandArgs": "",
      "configurationType": "Release",
      "generator": "Visual Studio 15 2017 Win64",
-      "name": "x64-Release CPU - int. libs",
+      "name": "x64-Release-int-libs",
      "variables": [
        {
          "name": "USE_INTEGRATED_LIBS",
@ -20,7 +20,7 @@
      "cmakeCommandArgs": "",
      "configurationType": "Debug",
      "generator": "Visual Studio 15 2017 Win64",
-      "name": "x64-Debug CPU",
+      "name": "x64-Debug",
      "variables": [
        {
          "name": "CMAKE_TOOLCHAIN_FILE",
@ -42,7 +42,7 @@
      "cmakeCommandArgs": "",
      "configurationType": "Release",
      "generator": "Visual Studio 15 2017 Win64",
-      "name": "x64-Release CPU",
+      "name": "x64-Release",
      "variables": [
        {
          "name": "CMAKE_TOOLCHAIN_FILE",
@ -57,58 +57,6 @@
          "value": "${env.VCPKG_DEFAULT_TRIPLET}"
        }
      ]
-    },
-    {
-      "buildCommandArgs": "-m -v:minimal",
-      "buildRoot": "${env.USERPROFILE}\\CMakeBuilds\\${workspaceHash}\\build\\${name}",
-      "cmakeCommandArgs": "",
-      "configurationType": "Debug",
-      "generator": "Visual Studio 15 2017 Win64",
-      "name": "x64-Debug GPU",
-      "variables": [
-        {
-          "name": "CMAKE_TOOLCHAIN_FILE",
-          "value": "${env.VCPKG_ROOT}\\scripts\\buildsystems\\vcpkg.cmake"
-        },
-        {
-          "name": "USE_INTEGRATED_LIBS",
-          "value": "FALSE"
-        },
-        {
-          "name": "ENABLE_CUDA",
-          "value": "TRUE"
-        },
-        {
-          "name": "VCPKG_TARGET_TRIPLET",
-          "value": "${env.VCPKG_DEFAULT_TRIPLET}"
-        }
-      ]
-    },
-    {
-      "buildCommandArgs": "-m -v:minimal",
-      "buildRoot": "${env.USERPROFILE}\\CMakeBuilds\\${workspaceHash}\\build\\${name}",
-      "cmakeCommandArgs": "",
-      "configurationType": "Release",
-      "generator": "Visual Studio 15 2017 Win64",
-      "name": "x64-Release GPU",
-      "variables": [
-        {
-          "name": "CMAKE_TOOLCHAIN_FILE",
-          "value": "${env.VCPKG_ROOT}\\scripts\\buildsystems\\vcpkg.cmake"
-        },
-        {
-          "name": "USE_INTEGRATED_LIBS",
-          "value": "FALSE"
-        },
-        {
-          "name": "ENABLE_CUDA",
-          "value": "TRUE"
-        },
-        {
-          "name": "VCPKG_TARGET_TRIPLET",
-          "value": "${env.VCPKG_DEFAULT_TRIPLET}"
-        }
-      ]
    }
  ]
 }
--- a/build.ps1
+++ b/build.ps1
@ -16,36 +16,22 @@ if ((Get-Command "cl.exe" -ErrorAction SilentlyContinue) -eq $null)
  Write-Host "Visual Studio 2017 ${vstype} Command Prompt variables set.`n" -ForegroundColor Yellow
 }

-# CPU ONLY, DEBUG
+# DEBUG
 New-Item -Path .\build_win_debug -ItemType directory -Force
 Set-Location build_win_debug
 cmake -G "Visual Studio 15 2017" -T "host=x64" -A "x64" "-DCMAKE_TOOLCHAIN_FILE=$env:VCPKG_ROOT\scripts\buildsystems\vcpkg.cmake" "-DVCPKG_TARGET_TRIPLET=$env:VCPKG_DEFAULT_TRIPLET" "-DCMAKE_BUILD_TYPE=Debug" ..
 cmake --build . --config Debug --parallel ${number_of_build_workers}
 Set-Location ..

-# CPU ONLY, RELEASE
+# RELEASE
 New-Item -Path .\build_win_release -ItemType directory -Force
 Set-Location build_win_release
 cmake -G "Visual Studio 15 2017" -T "host=x64" -A "x64" "-DCMAKE_TOOLCHAIN_FILE=$env:VCPKG_ROOT\scripts\buildsystems\vcpkg.cmake" "-DVCPKG_TARGET_TRIPLET=$env:VCPKG_DEFAULT_TRIPLET" "-DCMAKE_BUILD_TYPE=Release" ..
 cmake --build . --config Release --parallel ${number_of_build_workers}
 Set-Location ..

-# CUDA, DEBUG
-New-Item -Path .\build_win_debug_cuda -ItemType directory -Force
-Set-Location build_win_debug_cuda
-cmake -G "Visual Studio 15 2017" -T "host=x64" -A "x64" "-DENABLE_CUDA:BOOL=TRUE" "-DCMAKE_TOOLCHAIN_FILE=$env:VCPKG_ROOT\scripts\buildsystems\vcpkg.cmake" "-DVCPKG_TARGET_TRIPLET=$env:VCPKG_DEFAULT_TRIPLET" "-DCMAKE_BUILD_TYPE=Debug" ..
-cmake --build . --config Debug --parallel ${number_of_build_workers}
-Set-Location ..
-
-# CUDA, RELEASE
-New-Item -Path .\build_win_release_cuda -ItemType directory -Force
-Set-Location build_win_release_cuda
-cmake -G "Visual Studio 15 2017" -T "host=x64" -A "x64" "-DENABLE_CUDA:BOOL=TRUE" "-DCMAKE_TOOLCHAIN_FILE=$env:VCPKG_ROOT\scripts\buildsystems\vcpkg.cmake" "-DVCPKG_TARGET_TRIPLET=$env:VCPKG_DEFAULT_TRIPLET" "-DCMAKE_BUILD_TYPE=Release" ..
-cmake --build . --config Release --parallel ${number_of_build_workers}
-Set-Location ..
-
-# CPU ONLY, USE LOCAL PTHREAD LIB, NO VCPKG: remember to use "vcpkg.exe integrate remove" in case you had enable user-wide vcpkg integration
-New-Item -Path .\build_win_release_cuda_custom_libs -ItemType directory -Force
+# USE LOCAL PTHREAD LIB, NO VCPKG: remember to use "vcpkg.exe integrate remove" in case you had enable user-wide vcpkg integration
+New-Item -Path .\build_win_release_cuda_int_libs -ItemType directory -Force
 Set-Location build_win_release_cuda_custom_libs
 cmake -G "Visual Studio 15 2017" -T "host=x64" -A "x64" ..
 cmake --build . --config Release --parallel ${number_of_build_workers}
--- a/build.sh
+++ b/build.sh
@ -9,7 +9,7 @@ fi

 rm -f uselib darknet

-# CPU ONLY, DEBUG
+# DEBUG
 mkdir -p build_debug
 cd build_debug
 cmake .. -DCMAKE_BUILD_TYPE=Debug
@ -17,26 +17,10 @@ cmake --build . --target install -- -j${number_of_build_workers}
 #cmake --build . --target install --parallel ${number_of_build_workers}  #valid only for CMake 3.12+
 cd ..

-# CPU ONLY, RELEASE
+# RELEASE
 mkdir -p build_release
 cd build_release
 cmake .. -DCMAKE_BUILD_TYPE=Release
 cmake --build . --target install -- -j${number_of_build_workers}
 #cmake --build . --target install --parallel ${number_of_build_workers}  #valid only for CMake 3.12+
 cd ..
-
-# CUDA, DEBUG
-mkdir -p build_debug_gpu
-cd build_debug_gpu
-cmake .. -DCMAKE_BUILD_TYPE=Debug -DENABLE_CUDA:BOOL=TRUE
-cmake --build . --target install -- -j${number_of_build_workers}
-#cmake --build . --target install --parallel ${number_of_build_workers}  #valid only for CMake 3.12+
-cd ..
-
-# CUDA, RELEASE
-mkdir -p build_release_gpu
-cd build_release_gpu
-cmake .. -DCMAKE_BUILD_TYPE=Release -DENABLE_CUDA:BOOL=TRUE
-cmake --build . --target install -- -j${number_of_build_workers}
-#cmake --build . --target install --parallel ${number_of_build_workers}  #valid only for CMake 3.12+
-cd ..
--- a/cmake/Modules/FindCUDNN.cmake
+++ b/cmake/Modules/FindCUDNN.cmake
@ -19,26 +19,18 @@
 #  ``CUDNN_LIBRARIES``
 #    The CUDNN libraries.
 #
-# Hints
-# ^^^^^
-#
-#  ``CUDNN_ROOT_DIR``
-#    Set this variable to a directory that contains a CUDNN installation.
-#

 include(FindPackageHandleStandardArgs)

-#set(CUDNN_ROOT_DIR "" CACHE PATH "Folder contains NVIDIA cuDNN")
-
 if(NOT CUDNN_INCLUDE_DIR)
  find_path(CUDNN_INCLUDE_DIR cudnn.h
-    HINTS ${CUDNN_ROOT_DIR} ${CUDA_HOME} ${CUDA_TOOLKIT_ROOT_DIR}
+    HINTS ${CUDA_HOME} ${CUDA_TOOLKIT_ROOT_DIR}
    PATH_SUFFIXES cuda/include include)
 endif()

 if(NOT CUDNN_LIBRARY)
  find_library(CUDNN_LIBRARY cudnn
-    HINTS ${CUDNN_ROOT_DIR} ${CUDA_HOME} ${CUDA_TOOLKIT_ROOT_DIR}
+    HINTS ${CUDA_HOME} ${CUDA_TOOLKIT_ROOT_DIR}
    PATH_SUFFIXES lib lib64 cuda/lib cuda/lib64 lib/x64)
 endif()