Merge pull request #111 from eyalroz/development

Merge work on the development branch

.travis.yml

@@ -0,0 +1,42 @@
+language: cpp
+
+sudo: enabled
+
+compiler:
+  - gcc
+
+matrix:
+  include:
+    - name: CUDA 9
+      env:
+      - CUDA=9.2.148-1
+      - CUDA_SHORT=9.2
+      - CUDA_APT=9-2
+      - UBUNTU_VERSION=ubuntu1604
+      dist: xenial
+    - name: CUDA 10
+      env:
+      - CUDA=10.1.105-1
+      - CUDA_APT=10-1
+      - CUDA_SHORT=10.1
+      - UBUNTU_VERSION=ubuntu1804
+      dist: bionic
+
+before_install:
+  - INSTALLER=cuda-repo-${UBUNTU_VERSION}_${CUDA}_amd64.deb
+  - wget http://developer.download.nvidia.com/compute/cuda/repos/${UBUNTU_VERSION}/x86_64/${INSTALLER}
+  - sudo dpkg -i ${INSTALLER}
+  - wget https://developer.download.nvidia.com/compute/cuda/repos/${UBUNTU_VERSION}/x86_64/7fa2af80.pub
+  - sudo apt-key add 7fa2af80.pub
+  - sudo apt update -qq
+  - sudo apt install -y cuda-core-${CUDA_APT} cuda-cudart-dev-${CUDA_APT} cuda-nvtx-${CUDA_APT}
+  - sudo apt clean
+  - CUDA_HOME=/usr/local/cuda-${CUDA_SHORT}
+  - LD_LIBRARY_PATH=${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}
+  - PATH=${CUDA_HOME}/bin:${PATH}
+
+before_script:
+    - cmake .
+
+script: 
+    - VERBOSE=1 make examples

CMakeLists.txt

@@ -19,7 +19,7 @@
 
 # We need version 3.8 for native CUDA support in CMake
 cmake_minimum_required(VERSION 3.8 FATAL_ERROR)
-set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/cmake/Modules/" "${CMAKE_CURRENT_SOURCE_DIR}/cmake/Modules")
+list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/Modules")
 
 # -----------------------------------
 # Project name, version & build type
@@ -68,10 +68,22 @@ option(EXPORT_BUILD_DIR "Enables external use without install" OFF)
 # CUDA
 # -------------
 
-#find_package(CUDA 7.0 REQUIRED) # Why do I to do this damn it ?!
-#include_directories( "${CUDA_TOOLKIT_INCLUDE}" )
+# While this should not be necessary with CMake 3.8 and later,
+# it apparently _is_ necessary to achieve the following:
+# 
+# 1. Allow non-CUDA C++ code access to the CUDA libraries
+# 2. Determine the gencode/arch/code flags for the GPUs on the host (= target) machine
+#
+# so...
+include(FindCUDA) 
+
 include_directories( ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES} )
-include(HandleCUDAComputeCapability)
+
+cuda_select_nvcc_arch_flags(CUDA_ARCH_FLAGS_TMP Auto)
+set(CUDA_ARCH_FLAGS ${CUDA_ARCH_FLAGS_TMP} CACHE STRING "CUDA -gencode parameters")
+string(REPLACE ";" " " CUDA_ARCH_FLAGS_STR "${CUDA_ARCH_FLAGS}")
+set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} ${CUDA_ARCH_FLAGS_STR}")
+
 
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler -Wall" )
 set(CMAKE_CUDA_STANDARD 11)
@@ -81,6 +93,7 @@ set(CMAKE_CUDA_EXTENSIONS ON)
 set(CUDA_SEPARABLE_COMPILATION ON) # Does this work with native CUDA support?
 set(CUDA_PROPAGATE_HOST_FLAGS OFF) # Does this work with native CUDA support?
 
+
 # -----------------------
 # Main target(s)
 # -----------------------
@@ -102,17 +115,23 @@ target_include_directories(
 		"${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES}"
 )
 
-if(WIN32)
-	# Windows users report that CMake has trouble 
-	set(CUDA_LIBRARIES "cudadevrt.lib;cudart.lib")
-	target_link_libraries(${PROJECT_NAME} PUBLIC ${CUDA_LIBRARIES})
-endif()
+target_link_libraries(${PROJECT_NAME} ${CUDA_LIBRARIES})
 
 # -----------------------
 # Examples / Tests
 # -----------------------
 
-link_libraries(${CMAKE_CUDA_IMPLICIT_LINK_LIBRARIES} cuda-api-wrappers)
+# This next line should have been enough to make the example programs
+# link against the CUDA runtime library. However, users have reported
+# that doesn't actually happen in some cases / on some platforms; see
+# the project page for details and specifically issue #106. It has
+# been removed in favor of relying on the find_package(CUDA) line
+# above, despite the deprecation of that method of locating CUDA
+# libraries.
+#
+#link_libraries(${CMAKE_CUDA_IMPLICIT_LINK_LIBRARIES})
+
+link_libraries(cuda-api-wrappers)
 
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY "examples/bin")
 add_executable(vectorAdd EXCLUDE_FROM_ALL examples/modified_cuda_samples/vectorAdd/vectorAdd.cu)

README.md

@@ -1,5 +1,7 @@
 # cuda-api-wrappers:<br> Thin C++-flavored wrappers for the CUDA runtime API
 
+Branch Build Status: Master [![Master Build Status](https://travis-ci.org/eyalroz/cuda-api-wrappers.svg?branch=master)](https://travis-ci.org/eyalroz/cuda-api-wrappers) | Development: [![Development Build Status](https://travis-ci.org/eyalroz/cuda-api-wrappers.svg?branch=development)](https://travis-ci.org/eyalroz/cuda-api-wrappers)
+
 nVIDIA's [Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html) for [CUDA](http://www.nvidia.com/object/cuda_home_new.html) is intended for use both in C and C++ code. As such, it uses a C-style API, the lower common denominator (with a few [notable exceptions](https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__HIGHLEVEL.html) of templated function overloads).
 
 This library of wrappers around the Runtime API is intended to allow us to embrace many of the features of C++ (including some C++11) for using the runtime API - but without reducing expressivity or increasing the level of abstraction (as in, e.g., the [Thrust](https://thrust.github.io/) library). Using cuda-api-wrappers, you still have your devices, streams, events and so on - but they will be more convenient to work with in more C++-idiomatic ways.

cmake/Modules/FindCUDAAPIWrappers.cmake

@@ -17,7 +17,7 @@ find_package(PkgConfig) # will this even help us at all?
 
 find_path(
 	CUDA_API_WRAPPERS_INCLUDE_DIR 
-	cuda/api_wrappers.h
+	cuda/api_wrappers.hpp
 	HINTS
 		${CUDA_INCLUDE_DIRS}
 		${CMAKE_CURRENT_SOURCE_DIR}/cuda-api-wrappers