Reoptimization of OpenMP and other improvements on the configuration

AAAlvesJr · May 14, 2016 · b0cd98b · b0cd98b
1 parent 2020314
commit b0cd98b
Show file tree

Hide file tree

Showing 23 changed files with 104 additions and 59 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -121,7 +121,7 @@ if(BUILD_CUDA_TARGETS)
 
     cuda_add_executable(MCBooster_Example_OpenMP_CompareWithRoot
     ${CMAKE_CURRENT_SOURCE_DIR}/src/CompareWithTGenPhaseSpace.cu
-    OPTIONS -Xcompiler -fopenmp -DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_OMP -lgomp
+    OPTIONS -Xcompiler -fopenmp -DMCBOOSTER_BACKEND=OMP -lgomp
     )
 
     target_link_libraries(MCBooster_Example_OpenMP_CompareWithRoot
@@ -150,7 +150,7 @@ if(BUILD_CUDA_TARGETS)
 
 	cuda_add_executable(MCBooster_Example_OpenMP_NVCC_B2KPiJpsi
 	${CMAKE_CURRENT_SOURCE_DIR}/src/Generate.cu
-	OPTIONS -Xcompiler -fopenmp -DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_OMP  -lgomp
+	OPTIONS -Xcompiler -fopenmp -DMCBOOSTER_BACKEND=OMP  -lgomp
 	)
 
 	target_link_libraries(MCBooster_Example_OpenMP_NVCC_B2KPiJpsi
@@ -169,7 +169,7 @@ if(BUILD_CUDA_TARGETS)
 
 	cuda_add_executable(MCBooster_Example_OpenMP_NVCC_PerformanceTest
 	${CMAKE_CURRENT_SOURCE_DIR}/src/PerformanceTest.cu
-	OPTIONS -Xcompiler -fopenmp -DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_OMP -lgomp
+	OPTIONS -Xcompiler -fopenmp -DMCBOOSTER_BACKEND=OMP -lgomp
 	)
 
 	target_link_libraries(MCBooster_Example_OpenMP_NVCC_PerformanceTest
@@ -185,7 +185,7 @@ add_executable(MCBooster_Example_OpenMP_GCC_B2KPiJpsi
 ${CMAKE_CURRENT_SOURCE_DIR}/src/Generate.cpp
 )
 
-set_target_properties(MCBooster_Example_OpenMP_GCC_B2KPiJpsi PROPERTIES COMPILE_FLAGS  "-std=c++11 -fopenmp -DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_OMP -lgomp" )
+set_target_properties(MCBooster_Example_OpenMP_GCC_B2KPiJpsi PROPERTIES COMPILE_FLAGS  "-std=c++11 -fopenmp -DMCBOOSTER_BACKEND=OMP -lgomp" )
 
 target_link_libraries(MCBooster_Example_OpenMP_GCC_B2KPiJpsi
 ${ROOT_LIBRARIES}
@@ -198,7 +198,7 @@ add_executable(MCBooster_Example_OpenMP_GCC_PerformanceTest
 ${CMAKE_CURRENT_SOURCE_DIR}/src/PerformanceTest.cpp
 )
 
-set_target_properties(MCBooster_Example_OpenMP_GCC_PerformanceTest PROPERTIES COMPILE_FLAGS "-std=c++11 -fopenmp -DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_OMP -lgomp" )
+set_target_properties(MCBooster_Example_OpenMP_GCC_PerformanceTest PROPERTIES COMPILE_FLAGS "-std=c++11 -fopenmp -DMCBOOSTER_BACKEND=OMP -lgomp" )
 
 target_link_libraries(MCBooster_Example_OpenMP_GCC_PerformanceTest
 ${ROOT_LIBRARIES}

diff --git a/mcbooster/Config.h b/mcbooster/Config.h
@@ -25,19 +25,51 @@
 #ifndef CONFIG_H_
 #define CONFIG_H_
 
-#define CUDA_API_PER_THREAD_DEFAULT_STREAM
+#include <mcbooster/MCBooster.h>
+#include <iostream>
 
-#include <thrust/detail/config/host_device.h>
+#define CUDA 1
+#define OMP 2
+#define TBB 3
 
-#if !(THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB)
+#ifndef __linux__
+#error "MCBooster: Platform not supported. MCBooster is designed to run only on LIXUX systems."
+#endif
 
- #include <cuda.h>
- #include <cuda_runtime.h>
- #include <cuda_runtime_api.h>
+#if (__cplusplus < 201103L)
+#error "This library needs a C++11 compliant compiler"
+#endif
 
 
+#ifndef MCBOOSTER_BACKEND
+  #define MCBOOSTER_BACKEND CUDA
 #endif
 
 
+#if MCBOOSTER_BACKEND!=CUDA && MCBOOSTER_BACKEND!=OMP && MCBOOSTER_BACKEND!=TBB
+
+#error "MCBooster: Backend not supported. MCBOOSTER_BACKEND = CUDA, OMP or TBB "
+
+#endif
+
+
+
+#if MCBOOSTER_BACKEND==CUDA
+	#define CUDA_API_PER_THREAD_DEFAULT_STREAM
+	#define THRUST_DEVICE_SYSTEM CUDA//THRUST_DEVICE_SYSTEM_CUDA
+	#define THRUST_HOST_SYSTEM OMP
+	#include <cuda.h>
+	#include <cuda_runtime.h>
+	#include <cuda_runtime_api.h>
+#elif MCBOOSTER_BACKEND==OMP
+    #define THRUST_DEVICE_SYSTEM OMP
+	#define THRUST_HOST_SYSTEM OMP
+#elif MCBOOSTER_BACKEND==TBB
+    #define THRUST_DEVICE_SYSTEM TBB
+	#define THRUST_HOST_SYSTEM TBB
+#endif
+
+
+#include <thrust/detail/config/host_device.h>
 
-#endif /* CUDA_H_ */
+#endif /* CONFIG_H_ */
diff --git a/mcbooster/Evaluate.h b/mcbooster/Evaluate.h
@@ -39,7 +39,7 @@
 #include <mcbooster/GTypes.h>
 #include <mcbooster/functors/Calculate.h>
 
-namespace MCBooster
+namespace mcbooster
 {
 /** Template functor for evaluate an arbitrary function object.
  * Template functor for evaluate an arbitrary function object over the a set of particles stored

diff --git a/mcbooster/EvaluateArray.h b/mcbooster/EvaluateArray.h
@@ -38,7 +38,7 @@
 #include <mcbooster/GTypes.h>
 #include <mcbooster/functors/Calculate.h>
 
-namespace MCBooster
+namespace mcbooster
 {
 
 /** Template functor for calculate an array of variables over a given set of particles.
@@ -280,7 +280,7 @@ void EvaluateArray(const CUSTOMFUNC funcObj, ParticlesSet_d &pset,
 
 	}
 
-#if THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB
+#if MCBOOSTER_BACKEND!=CUDA
 
 #pragma omp parallel num_threads(  arrayWidth )
 	{
@@ -319,7 +319,7 @@ void EvaluateArray(const CUSTOMFUNC funcObj, ParticlesSet_d &pset,
 	return;
 }
 
-#if !(THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB)
+#if MCBOOSTER_BACKEND==CUDA
 
 /** Template functor for calculate an array of variables over a given set of particles.
  * Template functor for evaluate an arbitrary function object over the a set of particles stored
@@ -560,8 +560,8 @@ void EvaluateArray(const CUSTOMFUNC funcObj, ParticlesSet_d &pset,
 				Calculate3<CUSTOMFUNC>(funcObj));
 
 	}
-
-#if THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB
+/*
+//#if THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB
 
 #pragma omp parallel num_threads(  arrayWidth )
 	{
@@ -571,7 +571,8 @@ void EvaluateArray(const CUSTOMFUNC funcObj, ParticlesSet_d &pset,
 		thrust::copy(it_array.begin(),it_array.end(),
 				varset[omp_get_thread_num()]->begin());
 	}
-#else
+
+#else*/
 	cudaStream_t s[arrayWidth];
 
 	for (GInt_t d = 0; d < arrayWidth; d++)
@@ -596,7 +597,7 @@ void EvaluateArray(const CUSTOMFUNC funcObj, ParticlesSet_d &pset,
 	for (GInt_t d = 0; d < arrayWidth; d++)
 		delete it[d];
 
-#endif
+//#endif*/
 	return;
 }
 #endif

diff --git a/mcbooster/GContainers.h b/mcbooster/GContainers.h
@@ -42,22 +42,22 @@
 #include <thrust/host_vector.h>
 #include <thrust/complex.h>
 
-#if !(THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB)
+#if MCBOOSTER_BACKEND==CUDA
 #include <thrust/system/cuda/experimental/pinned_allocator.h>
 #endif
 
 using namespace std;
 
-namespace MCBooster
+namespace mcbooster
 {
 
-#if (THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB)
+#if (MCBOOSTER_BACKEND==OMP)
 /*!
  * Generic template typedef for thrust::host_vector. Use it instead of Thrust implementation
  * in order to avoid problems to compile OpenMP based applications using gcc and without a cuda runtime installation.
  */
 	template <typename T>
-		using  mc_device_vector = thrust::device_vector<T>;
+		using  mc_device_vector = thrust::host_vector<T>;
 /*!
  * Generic template typedef for thrust::host_vector. Use it instead of Thrust implementation
  * in order to avoid problems to compile OpenMP based applications using gcc and without a cuda runtime installation.
@@ -67,7 +67,7 @@ namespace MCBooster
 	template <typename T>
 		using  mc_host_vector   = thrust::host_vector<T>;
 
-#else
+#elif(MCBOOSTER_BACKEND==CUDA)
 	/*!
 	 * Generic template typedef for thrust::host_vector. Use it instead of Thrust implementation
 	 * in order to avoid problems to compile OpenMP based applications using gcc and without a cuda runtime installation.
@@ -84,6 +84,21 @@ namespace MCBooster
 		using  mc_host_vector = thrust::host_vector<T,
 				thrust::cuda::experimental::pinned_allocator<T>>;
 
+#elif(MCBOOSTER_BACKEND==TBB)
+	/*!
+	 * Generic template typedef for thrust::host_vector. Use it instead of Thrust implementation
+	 * in order to avoid problems to compile OpenMP based applications using gcc and without a cuda runtime installation.
+	 */
+	template <typename T>
+		using  mc_device_vector = thrust::device_vector<T>;
+	/*!
+	 * Generic template typedef for thrust::host_vector. Use it instead of Thrust implementation
+	 * in order to avoid problems to compile OpenMP based applications using gcc and without a cuda runtime installation.
+	 * mc_host_vectot will always allocate page locked memory on CUDA SYSTEMs in order to maximize speed in memory transfers
+	 * to the device.
+	 */
+	template <typename T>
+		using  mc_host_vector = thrust::host_vector<T>;
 #endif
 
 

diff --git a/mcbooster/GFunctional.h b/mcbooster/GFunctional.h
@@ -34,7 +34,7 @@
 #include <mcbooster/Vector3R.h>
 #include <mcbooster/Vector4R.h>
 
-namespace MCBooster
+namespace mcbooster
 {
 /** \struct  IFunction
  *  IFunction is the base class for arbitrary functions return any type suported by the framwork.

diff --git a/mcbooster/GTypes.h b/mcbooster/GTypes.h
@@ -26,7 +26,7 @@
 #ifndef GTYPES_H_
 #define GTYPES_H_
 
-namespace MCBooster
+namespace mcbooster
 {
 //---- types -------------------------------------------------------------------
 

diff --git a/mcbooster/Generate.h b/mcbooster/Generate.h
@@ -67,11 +67,13 @@
 #include <thrust/sort.h>
 #include <thrust/iterator/counting_iterator.h>
 
-#if !(THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB)
+#if MCBOOSTER_BACKEND==CUDA
 #include <thrust/system/cuda/execution_policy.h>
-#endif
-
+#elif MCBOOSTER_BACKEND==OPENMP
 #include <thrust/system/omp/execution_policy.h>
+#elif MCBOOSTER_BACKEND==TBB
+#include <thrust/system/tbb/execution_policy.h>
+#endif
 
 #define TIMER CLOCK_REALTIME
 
@@ -85,7 +87,7 @@
 
 using namespace std;
 
-namespace MCBooster {
+namespace mcbooster {
 /*!
  * Function to calculate time intervals in seconds.
  */
@@ -404,8 +406,7 @@ void PhaseSpace::ExportUnweighted(Events *_Events) {
 
 	_Events->fMaxWeight = fMaxWeight;
 
-#if THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB
-
+#if MCBOOSTER_BACKEND!=CUDA
 #pragma omp parallel num_threads( fNDaughters + 1 )
 	{
 
@@ -477,7 +478,7 @@ void PhaseSpace::Export(Events *_Events) {
 	 */
 	_Events->fMaxWeight = fMaxWeight;
 
-#if THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB
+#if MCBOOSTER_BACKEND!=CUDA
 
 #pragma omp parallel num_threads( fNDaughters + 1 )
 	{
@@ -544,7 +545,7 @@ void PhaseSpace::Generate(const Vector4R fMother) {
 	 * in any system of reference. The daughters will be generated in this system.
 	 */
 
-#if !(THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB)
+#if MCBOOSTER_BACKEND==CUDA
 	cudaDeviceSetCacheConfig(cudaFuncCachePreferL1);
 #endif
 	/* random number generation */
@@ -671,7 +672,7 @@ void PhaseSpace::Generate(Particles_d fMothers) {
 	 * Run the generator and calculate the maximum weight. It takes as input the device vector with the four-vectors of the mother particle
 	 * in any system of reference. The daughters will be generated in this system.
 	 */
-#if !(THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_OMP || THRUST_DEVICE_SYSTEM==THRUST_DEVICE_SYSTEM_TBB)
+#if MCBOOSTER_BACKEND==CUDA
 	cudaDeviceSetCacheConfig(cudaFuncCachePreferL1);
 #endif
 

diff --git a/mcbooster/Vector3R.h b/mcbooster/Vector3R.h
@@ -36,7 +36,7 @@
 #include <math.h>
 
 using std::ostream;
-namespace MCBooster
+namespace mcbooster
 {
 class Vector3R
 {

diff --git a/mcbooster/Vector4R.h b/mcbooster/Vector4R.h
@@ -33,20 +33,12 @@
 #include <iostream>
 #include <cmath>
 #include "mcbooster/Vector3R.h"
-/*
-#ifndef __host__
-#define __host__
-#endif // __host__
-
-#ifndef __device__
-#define __device__
-#endif // __device_
-*/
+
 using std::ostream;
 
 
 
-namespace MCBooster
+namespace mcbooster
 {
 
 class Vector3R;

diff --git a/mcbooster/functors/Calculate.h b/mcbooster/functors/Calculate.h
@@ -33,7 +33,7 @@
 #include <mcbooster/GContainers.h>
 #include <mcbooster/GTypes.h>
 
-namespace MCBooster
+namespace mcbooster
 {
 
 template<typename FUNCTION, typename RESULT>

diff --git a/mcbooster/functors/DecayMother.h b/mcbooster/functors/DecayMother.h
@@ -38,7 +38,7 @@
 
 using namespace std;
 
-namespace MCBooster
+namespace mcbooster
 {
 
 struct DecayMother

diff --git a/mcbooster/functors/DecayMothers.h b/mcbooster/functors/DecayMothers.h
@@ -38,7 +38,7 @@
 
 using namespace std;
 
-namespace MCBooster
+namespace mcbooster
 {
 
 struct DecayMothers

diff --git a/mcbooster/functors/FlagAcceptReject.h b/mcbooster/functors/FlagAcceptReject.h
@@ -36,7 +36,7 @@
 #include <thrust/random.h>
 #include <mcbooster/GTypes.h>
 
-namespace MCBooster
+namespace mcbooster
 {
 /**\struct FlagAcceptReject
  * Flags generated events as accepted (1) or rejected (0).

diff --git a/mcbooster/functors/IsAccepted.h b/mcbooster/functors/IsAccepted.h
@@ -36,7 +36,7 @@
 #include <mcbooster/Config.h>
 #include <mcbooster/GTypes.h>
 
-namespace MCBooster
+namespace mcbooster
 {
 
 struct isAccepted

diff --git a/mcbooster/functors/RandGen.h b/mcbooster/functors/RandGen.h
@@ -32,7 +32,7 @@
 #include <mcbooster/GContainers.h>
 
 
-namespace MCBooster
+namespace mcbooster
 {
 /**\struct RandGen
  * Fill a given vector with random numbers between 0 and 1.