sheet 5
This commit is contained in:
parent
4141f31f0c
commit
71c9d9b858
4 changed files with 1434 additions and 0 deletions
30
Sheet_5/bsp_5_4/Makefile
Normal file
30
Sheet_5/bsp_5_4/Makefile
Normal file
|
|
@ -0,0 +1,30 @@
|
||||||
|
#
|
||||||
|
# use GNU-Compiler tools
|
||||||
|
COMPILER=GCC_
|
||||||
|
# alternatively from the shell
|
||||||
|
# export COMPILER=GCC_
|
||||||
|
# or, alternatively from the shell
|
||||||
|
# make COMPILER=GCC_
|
||||||
|
|
||||||
|
# use Intel compilers
|
||||||
|
#COMPILER=ICC_
|
||||||
|
|
||||||
|
# use PGI compilers
|
||||||
|
# COMPILER=PGI_
|
||||||
|
|
||||||
|
|
||||||
|
SOURCES = main.cpp bsp_3_lib_bench_par.cpp
|
||||||
|
OBJECTS = $(SOURCES:.cpp=.o)
|
||||||
|
|
||||||
|
PROGRAM = main.${COMPILER}
|
||||||
|
|
||||||
|
# uncomment the next to lines for debugging and detailed performance analysis
|
||||||
|
CXXFLAGS += -g
|
||||||
|
LINKFLAGS += -g
|
||||||
|
# do not use -pg with PGI compilers
|
||||||
|
|
||||||
|
ifndef COMPILER
|
||||||
|
COMPILER=GCC_
|
||||||
|
endif
|
||||||
|
|
||||||
|
include ../${COMPILER}default.mk
|
||||||
1069
Sheet_5/bsp_5_4/bsp_3_lib_bench_par.cpp
Normal file
1069
Sheet_5/bsp_5_4/bsp_3_lib_bench_par.cpp
Normal file
File diff suppressed because it is too large
Load diff
288
Sheet_5/bsp_5_4/bsp_3_lib_bench_par.h
Normal file
288
Sheet_5/bsp_5_4/bsp_3_lib_bench_par.h
Normal file
|
|
@ -0,0 +1,288 @@
|
||||||
|
#pragma once
|
||||||
|
|
||||||
|
#include <omp.h>
|
||||||
|
#include <vector>
|
||||||
|
|
||||||
|
/** Inner product
|
||||||
|
@param[in] x vector
|
||||||
|
@param[in] y vector
|
||||||
|
@return resulting Euclidean inner product <x,y>
|
||||||
|
*/
|
||||||
|
double scalar(std::vector<double> const &x, std::vector<double> const &y);
|
||||||
|
|
||||||
|
/** Inner product - parallel
|
||||||
|
@param[in] x vector
|
||||||
|
@param[in] y vector
|
||||||
|
@return resulting Euclidean inner product <x,y>
|
||||||
|
*/
|
||||||
|
double scalar_par(std::vector<double> const &x, std::vector<double> const &y);
|
||||||
|
|
||||||
|
|
||||||
|
/** Inner product with cblas
|
||||||
|
@param[in] x vector
|
||||||
|
@param[in] y vector
|
||||||
|
@return resulting Euclidean inner product <x,y>
|
||||||
|
*/
|
||||||
|
double scalar_cblas(std::vector<double> const &x, std::vector<double> const &y);
|
||||||
|
|
||||||
|
|
||||||
|
/** Inner product with Kahan summation
|
||||||
|
@param[in] x vector
|
||||||
|
@param[in] y vector
|
||||||
|
@return resulting Euclidean inner product <x,y>
|
||||||
|
*/
|
||||||
|
double scalar_kahan(std::vector<double> const &x, std::vector<double> const &y);
|
||||||
|
|
||||||
|
|
||||||
|
/** euclidean norm
|
||||||
|
@param[in] x vector
|
||||||
|
@return resulting Euclidean norm
|
||||||
|
*/
|
||||||
|
double norm_eucl(std::vector<double> const &x);
|
||||||
|
|
||||||
|
|
||||||
|
/** sum of vector
|
||||||
|
@param[in] x vector
|
||||||
|
@return sum of the vector elements
|
||||||
|
*/
|
||||||
|
double sum(std::vector<double> const &x);
|
||||||
|
|
||||||
|
|
||||||
|
/** sum of vector - parallel
|
||||||
|
@param[in] x vector
|
||||||
|
@return sum of the vector elements
|
||||||
|
*/
|
||||||
|
double sum_par(std::vector<double> const &x);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Matrix-Vektor-Multiplikation (row-wise access)
|
||||||
|
*
|
||||||
|
* \param[in] a Matrix with row wise access
|
||||||
|
* \param[in] x vector which gets multiplied
|
||||||
|
* \return resulting product a*x (vector)
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
std::vector<double> MatVec(std::vector<double> const & a, std::vector<double> const & x);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Matrix-Vektor-Multiplikation (row-wise access) - parallel
|
||||||
|
*
|
||||||
|
* \param[in] a Matrix with row wise access
|
||||||
|
* \param[in] x vector which gets multiplied
|
||||||
|
* \return resulting product a*x (vector)
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
std::vector<double> MatVec_par(std::vector<double> const & a, std::vector<double> const & x);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Matrix-Vektor-Multiplikation mit cblas (row-wise access)
|
||||||
|
*
|
||||||
|
* \param[in] a Matrix with row wise access
|
||||||
|
* \param[in] x vector which gets multiplied
|
||||||
|
* \return resulting product a*x (vector)
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
std::vector<double> MatVec_cblas(std::vector<double> const & a, std::vector<double> const & x);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Matrix-Vektor-Multiplikation (column-wise access)
|
||||||
|
*
|
||||||
|
* \param[in] a Matrix with row wise access
|
||||||
|
* \param[in] x vector which gets multiplied
|
||||||
|
* \return resulting product a*x (vector)
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
std::vector<double> MatVec_column(std::vector<double> const & a, std::vector<double> const & x);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Matrix-Matrix-Multiplikation (row-wise access)
|
||||||
|
*
|
||||||
|
* \param[in] a matrix with row wise access (M*L)
|
||||||
|
* \param[in] b matrix with row wise access (L*N)
|
||||||
|
* \param[in] L inner dimension of the matrix product
|
||||||
|
* \return resulting product a*b
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
std::vector<double> MatMatProd(std::vector<double> const & a, std::vector<double> const & b, int const & L);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Matrix-Matrix-Multiplikation (row-wise access) - parallel
|
||||||
|
*
|
||||||
|
* \param[in] a matrix with row wise access (M*L)
|
||||||
|
* \param[in] b matrix with row wise access (L*N)
|
||||||
|
* \param[in] L inner dimension of the matrix product
|
||||||
|
* \return resulting product a*b
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
std::vector<double> MatMatProd_par(std::vector<double> const & a, std::vector<double> const & b, int const & L);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Matrix-Matrix-Multiplikation mit cblas (row-wise access)
|
||||||
|
*
|
||||||
|
* \param[in] a matrix with row wise access (M*L)
|
||||||
|
* \param[in] b matrix with row wise access (L*N)
|
||||||
|
* \param[in] L inner dimension of the matrix product
|
||||||
|
* \return resulting product a*b
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
std::vector<double> MatMatProd_cblas(std::vector<double> const & a, std::vector<double> const & b, int const & L);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Polynomauswertung an Stelle x
|
||||||
|
*
|
||||||
|
* \param[in] a Vekor mit den Koeffizienten des Polynoms a=[a0,a1,a2,...]
|
||||||
|
* \param[in] x Vektor, für welchen das Polynom ausgewertet werden soll
|
||||||
|
* \return resulting vector p(x)
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
std::vector<double> PolynomEval(std::vector<double> const & a, std::vector<double> const & x);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Polynomauswertung an Stelle x - parallel
|
||||||
|
*
|
||||||
|
* \param[in] a Vekor mit den Koeffizienten des Polynoms a=[a0,a1,a2,...]
|
||||||
|
* \param[in] x Vektor, für welchen das Polynom ausgewertet werden soll
|
||||||
|
* \return resulting vector p(x)
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
std::vector<double> PolynomEval_par(std::vector<double> const & a, std::vector<double> const & x);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking A - the scalar product
|
||||||
|
*
|
||||||
|
* \param N size of the vector
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_A(int const & N, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking A - the scalar product with cblas
|
||||||
|
*
|
||||||
|
* \param N size of the vector
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_A_cblas(int const & N, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking A - the scalar product with Kahan summation
|
||||||
|
*
|
||||||
|
* \param N size of the vector
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_A_kahan(int const & N, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking A - norm
|
||||||
|
*
|
||||||
|
* \param N size of the vector
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_A_norm(int const & N, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking B - matrix-vector product Ax=b (row wise access)
|
||||||
|
*
|
||||||
|
* \param N size of vector x
|
||||||
|
* \param M size of vector b (=> A: M*N)
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_B(int const & N, int const & M, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking B - matrix-vector product Ax=b (row wise access) [parallel]
|
||||||
|
*
|
||||||
|
* \param N size of vector x
|
||||||
|
* \param M size of vector b (=> A: M*N)
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_B_par(int const & N, int const & M, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking B - matrix-vector product Ax=b with cblas (row wise access)
|
||||||
|
*
|
||||||
|
* \param N size of vector x
|
||||||
|
* \param M size of vector b (=> A: M*N)
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_B_cblas(int const & N, int const & M, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking B - matrix-vector product Ax=b (column wise access)
|
||||||
|
*
|
||||||
|
* \param N size of vector x
|
||||||
|
* \param M size of vector b (=> A: M*N)
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_B_column(int const & N, int const & M, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking C - Matrix-Matrix product C=A*B A_M*L, B_L*N
|
||||||
|
*
|
||||||
|
* \param N
|
||||||
|
* \param M
|
||||||
|
* \param L
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_C(int const & N, int const & M, int const & L, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking C - Matrix-Matrix product C=A*B A_M*L, B_L*N [parallel]
|
||||||
|
*
|
||||||
|
* \param N
|
||||||
|
* \param M
|
||||||
|
* \param L
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_C_par(int const & N, int const & M, int const & L, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking C - Matrix-Matrix product with cblas; C=A*B A_M*L, B_L*N
|
||||||
|
*
|
||||||
|
* \param N
|
||||||
|
* \param M
|
||||||
|
* \param L
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_C_cblas(int const & N, int const & M, int const & L, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking D - polynomial evaluation
|
||||||
|
*
|
||||||
|
* \param p the degree of the polynomial
|
||||||
|
* \param N size of the input vector x where p(x)
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_D(int const & p, int const & N, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief Benchmarking D - polynomial evaluation [parallel]
|
||||||
|
*
|
||||||
|
* \param p the degree of the polynomial
|
||||||
|
* \param N size of the input vector x where p(x)
|
||||||
|
* \param Nloops number of iterations we want to do for the measuring
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void benchmark_D_par(int const & p, int const & N, int const & Nloops);
|
||||||
|
|
||||||
|
|
||||||
|
/** \brief solving system of linear equations with time measurement
|
||||||
|
*
|
||||||
|
* \param N size of the system Ax=b with A(nxn)
|
||||||
|
* \param nrhs number of right hand sides b(nxnrhs)
|
||||||
|
*
|
||||||
|
*/
|
||||||
|
void solver(int const & N, int const & nrhs);
|
||||||
|
|
||||||
47
Sheet_5/bsp_5_4/main.cpp
Normal file
47
Sheet_5/bsp_5_4/main.cpp
Normal file
|
|
@ -0,0 +1,47 @@
|
||||||
|
#include "bsp_3_lib_bench_par.h"
|
||||||
|
#include <cmath>
|
||||||
|
#include <iostream>
|
||||||
|
|
||||||
|
using namespace std;
|
||||||
|
|
||||||
|
int main()
|
||||||
|
{
|
||||||
|
benchmark_B(8000,8000,400);
|
||||||
|
benchmark_B_par(8000,8000,400);
|
||||||
|
benchmark_C(4000,4000,4000,1);
|
||||||
|
benchmark_C_par(4000,4000,4000,1);
|
||||||
|
benchmark_D(1e4,1e5,15);
|
||||||
|
benchmark_D_par(1e4,1e5,15);
|
||||||
|
|
||||||
|
// comparing the time for sum and inner product with and without parallelization
|
||||||
|
cout << "\nComparing the runtime (in sec) for inner product and sum with and without parallelization\n\n";
|
||||||
|
for(int k=3; k<=8; ++k)
|
||||||
|
{
|
||||||
|
|
||||||
|
int N = pow(10,k);
|
||||||
|
cout << "k = " << k << " N = " << N << endl;
|
||||||
|
vector<double> v1(N,1.0/N);
|
||||||
|
vector<double> v2(N,N);
|
||||||
|
|
||||||
|
auto tstart = omp_get_wtime();
|
||||||
|
double s = scalar(v1, v2);
|
||||||
|
auto t_diff_scalar = omp_get_wtime() - tstart;
|
||||||
|
|
||||||
|
tstart = omp_get_wtime();
|
||||||
|
double sp = scalar_par(v1, v2);
|
||||||
|
auto t_diff_scalar_par = omp_get_wtime() - tstart;
|
||||||
|
|
||||||
|
tstart = omp_get_wtime();
|
||||||
|
double su = sum(v1);
|
||||||
|
auto t_diff_sum = omp_get_wtime() - tstart;
|
||||||
|
|
||||||
|
tstart = omp_get_wtime();
|
||||||
|
double sup = sum_par(v1);
|
||||||
|
auto t_diff_sum_par = omp_get_wtime() - tstart;
|
||||||
|
|
||||||
|
cout << "sum " << t_diff_sum << " inner_prod " << t_diff_scalar << endl;
|
||||||
|
cout << "sum_par " << t_diff_sum_par << " inner_prod_par " << t_diff_scalar_par << endl << endl;;
|
||||||
|
}
|
||||||
|
|
||||||
|
return 0;
|
||||||
|
}
|
||||||
Loading…
Add table
Add a link
Reference in a new issue