Use thread-local storage for global program cache

Author: kylelutz

.travis.yml

@@ -4,12 +4,12 @@ compiler:
     - clang
 before_install:
     - sudo apt-get update -qq
-    - sudo apt-get install -qq fglrx=2:8.960-0ubuntu1 opencl-headers libboost-chrono1.48-dev libboost-date-time1.48-dev libboost-test1.48-dev libboost-system1.48-dev libboost-filesystem1.48-dev libboost-timer1.48-dev libboost-program-options1.48-dev python-yaml lcov libopencv-dev
+    - sudo apt-get install -qq fglrx=2:8.960-0ubuntu1 opencl-headers libboost-chrono1.48-dev libboost-date-time1.48-dev libboost-test1.48-dev libboost-system1.48-dev libboost-filesystem1.48-dev libboost-timer1.48-dev libboost-program-options1.48-dev libboost-thread1.48-dev python-yaml lcov libopencv-dev
     - gem install coveralls-lcov
 script:
     - mkdir -p build
     - cd build
-    - cmake -DBOOST_COMPUTE_BUILD_TESTS=ON -DBOOST_COMPUTE_BUILD_EXAMPLES=ON -DBOOST_COMPUTE_BUILD_BENCHMARKS=ON -DBOOST_COMPUTE_USE_OFFLINE_CACHE=ON -DBOOST_COMPUTE_ENABLE_COVERAGE=ON -DBOOST_COMPUTE_HAVE_OPENCV=ON -DCMAKE_CXX_FLAGS="-Wall -pedantic -Werror -Wno-variadic-macros -Wno-long-long -Wno-shadow" ..
+    - cmake -DBOOST_COMPUTE_BUILD_TESTS=ON -DBOOST_COMPUTE_BUILD_EXAMPLES=ON -DBOOST_COMPUTE_BUILD_BENCHMARKS=ON -DBOOST_COMPUTE_USE_OFFLINE_CACHE=ON -DBOOST_COMPUTE_ENABLE_COVERAGE=ON -DBOOST_COMPUTE_HAVE_OPENCV=ON -DBOOST_COMPUTE_THREAD_SAFE=ON -DCMAKE_CXX_FLAGS="-Wall -pedantic -Werror -Wno-variadic-macros -Wno-long-long -Wno-shadow" ..
     - make -j8
     - ./example/list_devices
     - ctest --output-on-failure

CMakeLists.txt

@@ -18,6 +18,15 @@ if(${BOOST_COMPUTE_USE_CPP11})
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
 endif()
 
+# thread-safety options
+option(BOOST_COMPUTE_THREAD_SAFE "Compile with BOOST_COMPUTE_THREAD_SAFE defined" OFF)
+if(${BOOST_COMPUTE_THREAD_SAFE})
+  add_definitions(-DBOOST_COMPUTE_THREAD_SAFE)
+  if(${BOOST_COMPUTE_USE_CPP11})
+    add_definitions(-DBOOST_COMPUTE_HAVE_THREAD_LOCAL)
+  endif()
+endif()
+
 # optional third-party libraries
 option(BOOST_COMPUTE_HAVE_EIGEN "Have Eigen" OFF)
 option(BOOST_COMPUTE_HAVE_OPENCV "Have OpenCV" OFF)

doc/faq.qbk

@@ -129,6 +129,18 @@ which provides a std::vector-like interface to a region of host-memory and can
 be used directly with all of the Boost.Compute algorithms.
 
 
+[h3 Is Boost.Compute thread-safe?]
+
+The low-level Boost.Compute APIs offer the same thread-safety guarantees as
+the underyling OpenCL library implementation. However, the high-level APIs
+make use of a few global static objects for features such as automatic program
+caching which makes them not thread-safe by default.
+
+To compile Boost.Compute in thread-safe mode define `BOOST_COMPUTE_THREAD_SAFE`
+before including any of the Boost.Compute headers. By default this will require
+linking your application/library with the Boost.Thread library.
+
+
 [h3 How can I contribute?]
 
 We are actively seeking additional C++ developers with experience in

example/CMakeLists.txt

@@ -31,6 +31,10 @@ if (${BOOST_COMPUTE_USE_OFFLINE_CACHE})
   set(EXAMPLE_BOOST_COMPONENTS ${EXAMPLE_BOOST_COMPONENTS} system filesystem)
 endif()
 
+if(${BOOST_COMPUTE_THREAD_SAFE} AND NOT ${BOOST_COMPUTE_USE_CPP11})
+  set(EXAMPLE_BOOST_COMPONENTS ${EXAMPLE_BOOST_COMPONENTS} thread)
+endif()
+
 find_package(Boost 1.48 REQUIRED COMPONENTS ${EXAMPLE_BOOST_COMPONENTS})
 include_directories(SYSTEM ${Boost_INCLUDE_DIRS})
 

include/boost/compute/detail/global_static.hpp

@@ -0,0 +1,37 @@
+//---------------------------------------------------------------------------//
+// Copyright (c) 2013-2014 Kyle Lutz <[email protected]>
+//
+// Distributed under the Boost Software License, Version 1.0
+// See accompanying file LICENSE_1_0.txt or copy at
+// http://www.boost.org/LICENSE_1_0.txt
+//
+// See http://kylelutz.github.com/compute for more information.
+//---------------------------------------------------------------------------//
+
+#ifndef BOOST_COMPUTE_DETAIL_GLOBAL_STATIC_HPP
+#define BOOST_COMPUTE_DETAIL_GLOBAL_STATIC_HPP
+
+#include <boost/compute/config.hpp>
+
+#ifdef BOOST_COMPUTE_THREAD_SAFE
+#  ifdef BOOST_COMPUTE_HAVE_THREAD_LOCAL
+     // use c++11 thread local storage
+#    define BOOST_COMPUTE_DETAIL_GLOBAL_STATIC(type, name, ctor) \
+       thread_local type name ctor;
+#  else
+      // use thread_specific_ptr from boost.thread
+#     include <boost/thread/tss.hpp>
+#     define BOOST_COMPUTE_DETAIL_GLOBAL_STATIC(type, name, ctor) \
+        static ::boost::thread_specific_ptr< type > BOOST_PP_CAT(name, _tls_ptr_); \
+        if(!BOOST_PP_CAT(name, _tls_ptr_).get()){ \
+            BOOST_PP_CAT(name, _tls_ptr_).reset(new type ctor); \
+        } \
+        static type &name = *BOOST_PP_CAT(name, _tls_ptr_);
+#  endif
+#else
+   // no thread-safety, just use static
+#  define BOOST_COMPUTE_DETAIL_GLOBAL_STATIC(type, name, ctor) \
+     static type name ctor;
+#endif
+
+#endif // BOOST_COMPUTE_DETAIL_GLOBAL_STATIC_HPP

include/boost/compute/detail/program_cache.hpp

@@ -20,6 +20,7 @@
 #include <boost/compute/context.hpp>
 #include <boost/compute/program.hpp>
 #include <boost/compute/detail/lru_cache.hpp>
+#include <boost/compute/detail/global_static.hpp>
 
 namespace boost {
 namespace compute {
@@ -66,7 +67,7 @@ inline boost::shared_ptr<program_cache> get_program_cache(const context &context
 {
     typedef lru_cache<cl_context, boost::shared_ptr<program_cache> > cache_map;
 
-    static cache_map caches(8);
+    BOOST_COMPUTE_DETAIL_GLOBAL_STATIC(cache_map, caches, (8));
 
     boost::shared_ptr<program_cache> cache = caches.get(context.get());
     if(!cache){

perf/CMakeLists.txt

@@ -6,6 +6,10 @@ if (${BOOST_COMPUTE_USE_OFFLINE_CACHE})
   set(PERF_BOOST_COMPONENTS ${PERF_BOOST_COMPONENTS} filesystem)
 endif()
 
+if(${BOOST_COMPUTE_THREAD_SAFE} AND NOT ${BOOST_COMPUTE_USE_CPP11})
+  set(PERF_BOOST_COMPONENTS ${PERF_BOOST_COMPONENTS} thread)
+endif()
+
 find_package(Boost 1.48 REQUIRED COMPONENTS ${PERF_BOOST_COMPONENTS})
 include_directories(SYSTEM ${Boost_INCLUDE_DIRS})
 

test/CMakeLists.txt

@@ -3,8 +3,12 @@ include_directories(../include)
 set(BOOST_COMPONENTS unit_test_framework)
 
 if (${BOOST_COMPUTE_USE_OFFLINE_CACHE})
-    set(BOOST_COMPONENTS ${BOOST_COMPONENTS} system filesystem)
-    add_definitions(-DBOOST_COMPUTE_USE_OFFLINE_CACHE)
+  set(BOOST_COMPONENTS ${BOOST_COMPONENTS} system filesystem)
+  add_definitions(-DBOOST_COMPUTE_USE_OFFLINE_CACHE)
+endif()
+
+if(${BOOST_COMPUTE_THREAD_SAFE} AND NOT ${BOOST_COMPUTE_USE_CPP11})
+  set(BOOST_COMPONENTS ${BOOST_COMPONENTS} thread)
 endif()
 
 find_package(Boost 1.48 REQUIRED COMPONENTS ${BOOST_COMPONENTS})