From 7bee90a62a0e5fec52a4b8a6f0b1d88c175d63f8 Mon Sep 17 00:00:00 2001 From: Benoit Jacob Date: Fri, 11 Apr 2008 08:18:47 +0000 Subject: [PATCH] Merge Gael's experimental OpenMP parallelization support into Assign.h. --- Eigen/Core | 7 ++++ Eigen/src/Core/Assign.h | 81 +++++++++++++++++++++++++++++++++++------ Eigen/src/Core/Matrix.h | 3 +- bench/benchmarkX.cpp | 24 +++++++++--- bench/benchmarkXL.cpp | 34 +++++++++++++++++ 5 files changed, 131 insertions(+), 18 deletions(-) create mode 100644 bench/benchmarkXL.cpp diff --git a/Eigen/Core b/Eigen/Core index 24dc37145..743e21bd7 100644 --- a/Eigen/Core +++ b/Eigen/Core @@ -10,6 +10,13 @@ #endif #endif +#ifndef EIGEN_DONT_USE_OPENMP +#ifdef _OPENMP +#define EIGEN_USE_OPENMP +#include +#endif +#endif + #include #include #include diff --git a/Eigen/src/Core/Assign.h b/Eigen/src/Core/Assign.h index f33f491bf..53d6250f3 100644 --- a/Eigen/src/Core/Assign.h +++ b/Eigen/src/Core/Assign.h @@ -3,6 +3,7 @@ // // Copyright (C) 2007 Michael Olbrich // Copyright (C) 2006-2008 Benoit Jacob +// Copyright (C) 2008 Gael Guennebaud // // Eigen is free software; you can redistribute it and/or // modify it under the terms of the GNU Lesser General Public @@ -175,8 +176,28 @@ struct ei_operator_equals_impl unroll ? Derived::SizeAtCompileTime : Dynamic >::run(dst.derived(), src.derived()); else - for(int i = 0; i < dst.size(); i++) - dst.coeffRef(i) = src.coeff(i); + { + #ifdef EIGEN_USE_OPENMPf + if(Derived::Flags & OtherDerived::Flags & LargeBit) + { + #ifdef __INTEL_COMPILER + #pragma omp parallel default(none) shared(other) + #else + #pragma omp parallel default(none) + #endif + { + #pragma omp for + for(int i = 0; i < dst.size(); i++) + dst.coeffRef(i) = src.coeff(i); + } + } + else + #endif // EIGEN_USE_OPENMP + { + for(int i = 0; i < dst.size(); i++) + dst.coeffRef(i) = src.coeff(i); + } + } } else // copying a matrix expression into a matrix { @@ -192,18 +213,56 @@ struct ei_operator_equals_impl { if(Derived::ColsAtCompileTime == Dynamic || Derived::RowsAtCompileTime != Dynamic) { - // traverse in column-major order - for(int j = 0; j < dst.cols(); j++) - for(int i = 0; i < dst.rows(); i++) - dst.coeffRef(i, j) = src.coeff(i, j); + #ifdef EIGEN_USE_OPENMP + if(Derived::Flags & OtherDerived::Flags & LargeBit) + { + #ifdef __INTEL_COMPILER + #pragma omp parallel default(none) shared(other) + #else + #pragma omp parallel default(none) + #endif + { + #pragma omp for + for(int j = 0; j < dst.cols(); j++) + for(int i = 0; i < dst.rows(); i++) + dst.coeffRef(i, j) = src.coeff(i, j); + } + } + else + #endif // EIGEN_USE_OPENMP + { + // traverse in column-major order + for(int j = 0; j < dst.cols(); j++) + for(int i = 0; i < dst.rows(); i++) + dst.coeffRef(i, j) = src.coeff(i, j); + } } else { - // traverse in row-major order - // in order to allow the compiler to unroll the inner loop - for(int i = 0; i < dst.rows(); i++) - for(int j = 0; j < dst.cols(); j++) - dst.coeffRef(i, j) = src.coeff(i, j); + #ifdef EIGEN_USE_OPENMP + if(Derived::Flags & OtherDerived::Flags & LargeBit) + { + #ifdef __INTEL_COMPILER + #pragma omp parallel default(none) shared(other) + #else + #pragma omp parallel default(none) + #endif + { + #pragma omp for + for(int i = 0; i < dst.rows(); i++) + for(int j = 0; j < dst.cols(); j++) + dst.coeffRef(i, j) = src.coeff(i, j); + } + } + else + #endif // EIGEN_USE_OPENMP + { + // traverse in row-major order + // in order to allow the compiler to unroll the inner loop + for(int i = 0; i < dst.rows(); i++) + for(int j = 0; j < dst.cols(); j++) + dst.coeffRef(i, j) = src.coeff(i, j); + } } } } diff --git a/Eigen/src/Core/Matrix.h b/Eigen/src/Core/Matrix.h index 7c2c5afba..e79cbcbfb 100644 --- a/Eigen/src/Core/Matrix.h +++ b/Eigen/src/Core/Matrix.h @@ -358,7 +358,8 @@ using Eigen::RowVector##SizeSuffix##TypeSuffix; EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 2) \ EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 3) \ EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 4) \ -EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, X) +EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, X) \ +EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, XL) #define EIGEN_USING_MATRIX_TYPEDEFS \ EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(i) \ diff --git a/bench/benchmarkX.cpp b/bench/benchmarkX.cpp index 09173e1ed..590f2636b 100644 --- a/bench/benchmarkX.cpp +++ b/bench/benchmarkX.cpp @@ -5,18 +5,30 @@ using namespace std; USING_PART_OF_NAMESPACE_EIGEN +#ifndef MATTYPE +#define MATTYPE MatrixXd +#endif + +#ifndef MATSIZE +#define MATSIZE 20 +#endif + +#ifndef REPEAT +#define REPEAT 100000 +#endif + int main(int argc, char *argv[]) { - MatrixXd I = MatrixXd::identity(20,20); - MatrixXd m(20,20); - for(int i = 0; i < 20; i++) for(int j = 0; j < 20; j++) + MATTYPE I = MATTYPE::identity(MATSIZE,MATSIZE); + MATTYPE m(MATSIZE,MATSIZE); + for(int i = 0; i < MATSIZE; i++) for(int j = 0; j < MATSIZE; j++) { - m(i,j) = 0.1 * (i+20*j); + m(i,j) = 0.1 * (i+MATSIZE*j)/MATSIZE; } - for(int a = 0; a < 100000; a++) + for(int a = 0; a < REPEAT; a++) { m = I + 0.00005 * (m + m*m); } - cout << m << endl; + cout << m(0,0) << endl; return 0; } diff --git a/bench/benchmarkXL.cpp b/bench/benchmarkXL.cpp new file mode 100644 index 000000000..ff128da06 --- /dev/null +++ b/bench/benchmarkXL.cpp @@ -0,0 +1,34 @@ +// g++ -O3 -DNDEBUG benchmarkX.cpp -o benchmarkX && time ./benchmarkX + +#include + +using namespace std; +USING_PART_OF_NAMESPACE_EIGEN + +#ifndef MATTYPE +#define MATTYPE MatrixXLd +#endif + +#ifndef MATSIZE +#define MATSIZE 400 +#endif + +#ifndef REPEAT +#define REPEAT 10000 +#endif + +int main(int argc, char *argv[]) +{ + MATTYPE I = MATTYPE::identity(MATSIZE,MATSIZE); + MATTYPE m(MATSIZE,MATSIZE); + for(int i = 0; i < MATSIZE; i++) for(int j = 0; j < MATSIZE; j++) + { + m(i,j) = 0.1 * (i+MATSIZE*j)/MATSIZE; + } + for(int a = 0; a < REPEAT; a++) + { + m = I + 0.00005 * (m + m/4); + } + cout << m(0,0) << endl; + return 0; +}