Ex8 and minor improvements

This commit is contained in:
Markus Schmidt 2025-11-12 02:04:18 +01:00
commit 77bc8c6aa3
50 changed files with 214845 additions and 43 deletions

View file

@ -19,8 +19,8 @@ Each kernel will be executed 20 times.
will be used to compute the reported bandwidth. will be used to compute the reported bandwidth.
------------------------------------------------------------- -------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds. Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 46252 microseconds. Each test below will take on the order of 59858 microseconds.
(= 46252 clock ticks) (= 59858 clock ticks)
Increase the size of the arrays if this shows that Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test. you are not getting at least 20 clock ticks per test.
------------------------------------------------------------- -------------------------------------------------------------
@ -29,10 +29,10 @@ For best results, please be sure you know the
precision of your system timer. precision of your system timer.
------------------------------------------------------------- -------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time Function Best Rate MB/s Avg time Min time Max time
Copy: 28478.6 0.047858 0.044946 0.054333 Copy: 23508.2 0.063228 0.054449 0.074427
Scale: 20551.4 0.066044 0.062283 0.077807 Scale: 18323.7 0.089940 0.069855 0.116932
Add: 22534.2 0.089671 0.085204 0.099586 Add: 19762.6 0.276166 0.097153 3.054857
Triad: 22709.5 0.088864 0.084546 0.098536 Triad: 19559.9 0.123390 0.098160 0.156530
------------------------------------------------------------- -------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays Solution Validates: avg error less than 1.000000e-13 on all three arrays
------------------------------------------------------------- -------------------------------------------------------------
@ -42,19 +42,19 @@ Solution Validates: avg error less than 1.000000e-13 on all three arrays
Module Error RunTime MFLOPS Module Error RunTime MFLOPS
(usec) (usec)
1 4.0146e-13 0.0021 6622.7552 1 4.0146e-13 0.0029 4831.3737
2 -1.4166e-13 0.0006 12723.3419 2 -1.4166e-13 0.0006 11258.2969
3 4.7184e-14 0.0027 6253.2599 3 4.7184e-14 0.0031 5448.3769
4 -1.2557e-13 0.0026 5758.6323 4 -1.2557e-13 0.0030 5042.5895
5 -1.3800e-13 0.0051 5740.4851 5 -1.3800e-13 0.0060 4867.7339
6 3.2380e-13 0.0051 5674.2511 6 3.2380e-13 0.0054 5322.4399
7 -8.4583e-11 0.0031 3827.0478 7 -8.4583e-11 0.0031 3907.7854
8 3.4867e-13 0.0053 5610.0203 8 3.4867e-13 0.0056 5323.6214
Iterations = 512000000 Iterations = 512000000
NullTime (usec) = 0.0000 NullTime (usec) = 0.0000
MFLOPS(1) = 9507.3864 MFLOPS(1) = 8348.0311
MFLOPS(2) = 5042.7572 MFLOPS(2) = 4650.8807
MFLOPS(3) = 5597.4972 MFLOPS(3) = 5016.3434
MFLOPS(4) = 5766.1547 MFLOPS(4) = 5297.2428

View file

@ -82,8 +82,12 @@ vector<double> benchmark_D(const vector<double>& coeff, const vector<double>& x)
double norm2(const vector<double>& x) double norm2(const vector<double>& x)
{ {
double s = 0.0; double s = 0.0;
for (unsigned int i = 0; i < x.size(); ++i) double xi;
s += x[i]*x[i]; for (unsigned int i = 0; i < x.size(); ++i){
xi = x[i];
s += xi*xi;
}
return sqrt(s); return sqrt(s);
} }
@ -116,7 +120,7 @@ vector<double> matrixMultColumnWise(const vector<double> &A, const vector<double
double sum = 0.0; double sum = 0.0;
for (unsigned int k = 0; k < L; k++) for (unsigned int k = 0; k < L; k++)
{ {
sum += A[k*L+i]*B[k*N+j]; sum += A[k*M+i]*B[k*N+j];
} }
C[i*N+j] = sum; C[i*N+j] = sum;
} }

View file

@ -131,6 +131,7 @@ int main(int argc, char **argv)
double bytesC = (MC * LC + LC * NC + MC * NC)* sizeof(double); double bytesC = (MC * LC + LC * NC + MC * NC)* sizeof(double);
cout << "\n===== Benchmark C =====\n"; cout << "\n===== Benchmark C =====\n";
cout << guardC << endl;
cout << "bytes: " << bytesC << endl; cout << "bytes: " << bytesC << endl;
cout << "Timing in sec. : " << tC << endl; cout << "Timing in sec. : " << tC << endl;
cout << "GFLOPS : " << flopsC / tC / 1024 / 1024 / 1024 << endl; cout << "GFLOPS : " << flopsC / tC / 1024 / 1024 / 1024 << endl;
@ -172,6 +173,7 @@ int main(int argc, char **argv)
double bytesD = (p + 2 * ND)*sizeof(double); double bytesD = (p + 2 * ND)*sizeof(double);
cout << "\n===== Benchmark D =====\n"; cout << "\n===== Benchmark D =====\n";
cout << guardD << endl;
cout << "bytes: " << bytesD << endl; cout << "bytes: " << bytesD << endl;
cout << "Timing in sec. : " << tD << endl; cout << "Timing in sec. : " << tD << endl;
cout << "GFLOPS : " << flopsD / tD / 1024 / 1024 / 1024 << endl; cout << "GFLOPS : " << flopsD / tD / 1024 / 1024 / 1024 << endl;
@ -208,6 +210,8 @@ int main(int argc, char **argv)
cout << "GFLOPS : " << 2.0 * NA / tA / 1024 / 1024 / 1024 << endl; cout << "GFLOPS : " << 2.0 * NA / tA / 1024 / 1024 / 1024 << endl;
cout << "GiByte/s : " cout << "GiByte/s : "
<< NA * sizeof(xA[0]) / tA / 1024 / 1024 / 1024 << endl; << NA * sizeof(xA[0]) / tA / 1024 / 1024 / 1024 << endl;
//a bit faster due to only accessing one vector
} }
@ -240,6 +244,8 @@ int main(int argc, char **argv)
cout << "GFLOPS : " << 5.0 * NA / tA / 1024 / 1024 / 1024 << endl; cout << "GFLOPS : " << 5.0 * NA / tA / 1024 / 1024 / 1024 << endl;
cout << "GiByte/s : " cout << "GiByte/s : "
<< 2.0 * NA * sizeof(xA[0]) / tA / 1024 / 1024 / 1024 << endl; << 2.0 * NA * sizeof(xA[0]) / tA / 1024 / 1024 / 1024 << endl;
//in comparison to benchmark A: a bit slower runtime but more than double the amount of FLOPS therefor also more GFLOPS
} }
@ -276,11 +282,16 @@ int main(int argc, char **argv)
double bytesC = (MC * LC + LC * NC + MC * NC)* sizeof(double); double bytesC = (MC * LC + LC * NC + MC * NC)* sizeof(double);
cout << "\n===== Benchmark 5C =====\n"; cout << "\n===== Benchmark 5C =====\n";
cout << guardC << endl;
cout << "bytes: " << bytesC << endl; cout << "bytes: " << bytesC << endl;
cout << "Timing in sec. : " << tC << endl; cout << "Timing in sec. : " << tC << endl;
cout << "GFLOPS : " << flopsC / tC / 1024 / 1024 / 1024 << endl; cout << "GFLOPS : " << flopsC / tC / 1024 / 1024 / 1024 << endl;
cout << "GiByte/s : " << bytesC / tC / 1024 / 1024 / 1024 << endl; cout << "GiByte/s : " << bytesC / tC / 1024 / 1024 / 1024 << endl;
//slower than rowwise access, due to incoherent acces in the vector memory of A
//Transpose matrix, the it is also row wise-access or reorder loops
} }
return 0; return 0;
} // memory for x and y will be deallocated by their destructors }

51
sheet3/345/output.txt Normal file
View file

@ -0,0 +1,51 @@
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o main.o main.cpp
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o mylib.o mylib.cpp
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o benchmark.o benchmark.cpp
g++ main.o mylib.o benchmark.o -g -O0 -llapack -lblas -flto -o main.GCC_
./main.GCC_
===== Benchmark A =====
<xA,yA> = 1.4e+06
Timing in sec. : 0.00893637
GFLOPS : 0.291808
GiByte/s : 2.33446
===== Benchmark B =====
340000
bytes: 2.31472e+07
Timing in sec. : 0.0133897
GFLOPS : 0.402029
GiByte/s : 1.61001
===== Benchmark C =====
7.37196e+07
bytes: 2.4e+07
Timing in sec. : 8.67235
GFLOPS : 0.21478
GiByte/s : 0.00257736
===== Benchmark D =====
10500
bytes: 3.20001e+07
Timing in sec. : 0.101087
GFLOPS : 0.515935
GiByte/s : 0.294821
===== Benchmark 5A =====
NORM = 150114
Timing in sec. : 0.00703533
GFLOPS : 0.370658
GiByte/s : 1.48263
===== Benchmark 5B =====
<xA,yA> = 1.4e+06
Timing in sec. : 0.0108377
GFLOPS : 0.601533
GiByte/s : 1.92491
===== Benchmark 5C =====
7.37196e+07
bytes: 2.4e+07
Timing in sec. : 15.2407
GFLOPS : 0.122215
GiByte/s : 0.00146658

View file

@ -139,4 +139,4 @@ int main(int argc, char **argv)
return 0; return 0;
} // memory for x and y will be deallocated by their destructors }

24
sheet3/6/output.txt Normal file
View file

@ -0,0 +1,24 @@
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o main.o main.cpp
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o mylib.o mylib.cpp
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o benchmark.o benchmark.cpp
g++ main.o mylib.o benchmark.o -g -O0 -llapack -lblas -flto -o main.GCC_
./main.GCC_
===== Benchmark A =====
<xA,yA> = 1.4e+06
Timing in sec. : 0.000900578
GFLOPS : 2.89559
GiByte/s : 23.1647
===== Benchmark B =====
1.7e+07
bytes: 2.31472e+07
Timing in sec. : 0.000687268
GFLOPS : 7.83252
GiByte/s : 31.3669
===== Benchmark C =====
bytes: 2.4e+07
Timing in sec. : 0.0151789
GFLOPS : 122.713
GiByte/s : 1.47255

View file

@ -13,8 +13,7 @@ COMPILER=GCC_
# COMPILER=PGI_ # COMPILER=PGI_
SOURCES = main.cpp benchmark.cpp
SOURCES = main.cpp
OBJECTS = $(SOURCES:.cpp=.o) OBJECTS = $(SOURCES:.cpp=.o)
PROGRAM = main.${COMPILER} PROGRAM = main.${COMPILER}

43
sheet3/7/benchmark.cpp Normal file
View file

@ -0,0 +1,43 @@
#include <iostream>
#include <vector>
#include <cmath>
using namespace std;
#include <cblas.h>
// Inner product
double benchmark_A(const vector<double> &x, const vector<double> &y)
{
return cblas_ddot(x.size(),x.data(),1,y.data(),1);
}
//Matrix-vector product
vector<double> benchmark_B(const vector<double> &A, const vector<double> &x)
{
unsigned int N = x.size();
unsigned int M = A.size() / N;
vector<double> b(M, 0.0);
cblas_dgemv(CblasRowMajor,CblasNoTrans,M,N,1,A.data(),N,x.data(),1,0.0,b.data(),1);
return b;
}
//Matrix-Matrix product
vector<double> benchmark_C(const vector<double> &A, const vector<double> &B, unsigned int M)
{
unsigned int L = A.size()/M;
unsigned int N = B.size()/L;
vector<double> C(M*N,0.0);
cblas_dgemm(CblasRowMajor,CblasNoTrans,CblasNoTrans,M,N,L,1.0,A.data(),L,B.data(),N,0.0,C.data(),N);
return C;
}

21
sheet3/7/benchmark.h Normal file
View file

@ -0,0 +1,21 @@
#ifndef BENCHMARK_H
#define BENCHMARK_H
#include <vector>
using namespace std;
double benchmark_A(const vector<double> &x,
const vector<double> &y);
vector<double> benchmark_B(const vector<double> &A,
const vector<double> &x);
vector<double> benchmark_C(const vector<double> &A,
const vector<double> &B,
unsigned int M);
#endif

Binary file not shown.

View file

@ -8,14 +8,14 @@
#include <sstream> #include <sstream>
#include <vector> #include <vector>
#include <lapacke.h> #include <lapacke.h>
#include "timing.h"
#include "benchmark.h"
using namespace std; using namespace std;
using namespace std::chrono; // timing using namespace std::chrono; // timing
int main() int main()
{ {
unsigned int n= 10; unsigned int n= 32;
unsigned int nhrs = 1;
vector<double> M(n*n,4.0); vector<double> M(n*n,4.0);
@ -32,7 +32,7 @@ int main()
} }
vector<double> M2 = M;
@ -40,22 +40,69 @@ int main()
LAPACKE_dgetrf(LAPACK_ROW_MAJOR,n,n, M.data(),n,ipiv.data()); //M=PLU LAPACKE_dgetrf(LAPACK_ROW_MAJOR,n,n, M.data(),n,ipiv.data()); //M=PLU
unsigned int runtimes[] = {1,2,4,8,16,32};
for(unsigned int i=0; i < 6;i++)
double time;
unsigned int nhrsmax = 1000000;
for(unsigned int i=nhrsmax/10; i < nhrsmax;i+=nhrsmax/10)
{ {
nhrs = runtimes[i];
vector<double> b(n*nhrs,0.0);
for (unsigned int j=0; j<n; j++)
{
for (unsigned int k=0; k<nhrs; k++)
{
b[j*nhrs+k] = j*nhrs+k;
}
}
LAPACKE_dgetrs(LAPACK_ROW_MAJOR,'N',n,nhrs,M.data(),n,ipiv.data(),b.data(),nhrs);
}
unsigned int nhrs = i;
//FOR CHECKING
vector<double> X(n*nhrs,1.0);
vector<double> b = benchmark_C(M2,X,n);
tic();
LAPACKE_dgetrs(LAPACK_ROW_MAJOR,'N',n,nhrs,M.data(),n,ipiv.data(),b.data(),nhrs);
time = toc();
cout << "Time for nhrs=" << nhrs << ": " << time << endl;
double max_err = 0.0;
for (unsigned int j = 0; j < n * nhrs; j++)
{
double err = b[j] - X[j];
err *= err;
if (err > max_err) max_err = err;
}
cout <<"max err^2:" << max_err <<endl;
cout <<endl;
}
/*
Time for nhrs=100000: 0.0605495
max err^2:4.93038e-32
Time for nhrs=200000: 0.127608
max err^2:4.93038e-32
Time for nhrs=300000: 0.182197
max err^2:4.93038e-32
Time for nhrs=400000: 0.202608
max err^2:4.93038e-32
Time for nhrs=500000: 0.24484
max err^2:4.93038e-32
Time for nhrs=600000: 0.298055
max err^2:4.93038e-32
Time for nhrs=700000: 0.362414
max err^2:4.93038e-32
Time for nhrs=800000: 0.410004
max err^2:4.93038e-32
Time for nhrs=900000: 0.492339
max err^2:4.93038e-32
Time grows slow (linearly)
*/

Binary file not shown.

51
sheet3/7/timing.h Normal file
View file

@ -0,0 +1,51 @@
//
// Gundolf Haase, Oct 18 2024
//
#pragma once
#include <chrono> // timing
#include <stack>
//using Clock = std::chrono::system_clock; //!< The wall clock timer chosen
using Clock = std::chrono::high_resolution_clock;
using TPoint= std::chrono::time_point<Clock>;
// [Galowicz, C++17 STL Cookbook, p. 29]
std::stack<TPoint> MyStopWatch; //!< starting time of stopwatch
/** Starts stopwatch timer.
* Use as @code tic(); myfunction(...) ; double tsec = toc(); @endcode
*
* The timining can be nested and the recent time point is stored on top of the stack.
*
* @return recent time point
* @see toc
*/
auto tic()
{
MyStopWatch.push(Clock::now());
return MyStopWatch.top();
}
/** Returns the elapsed time from stopwatch.
*
* The time point from top of the stack is used
* if time point @p t_b is not passed as input parameter.
* Use as @code tic(); myfunction(...) ; double tsec = toc(); @endcode
* or as @code auto t_b = tic(); myfunction(...) ; double tsec = toc(t_b); @endcode
* The last option is to be used in the case of
* non-nested but overlapping time measurements.
*
* @param[in] t_b start time of some stop watch
* @return elapsed time in seconds.
*
*/
double toc(TPoint const &t_b = MyStopWatch.top())
{
// https://en.cppreference.com/w/cpp/chrono/treat_as_floating_point
using Unit = std::chrono::seconds;
using FpSeconds = std::chrono::duration<double, Unit::period>;
auto t_e = Clock::now();
MyStopWatch.pop();
return FpSeconds(t_e-t_b).count();
}

2877
sheet3/8/Doxyfile Normal file

File diff suppressed because it is too large Load diff

54
sheet3/8/Makefile Normal file
View file

@ -0,0 +1,54 @@
#
# use GNU-Compiler tools
COMPILER=GCC_
# COMPILER=GCC_SEQ_
# alternatively from the shell
# export COMPILER=GCC_
# or, alternatively from the shell
# make COMPILER=GCC_
MAIN = main
SOURCES = ${MAIN}.cpp vdop.cpp geom.cpp\
getmatrix.cpp jacsolve.cpp userset.cpp
# dexx.cpp debugd.cpp skalar.cpp vecaccu.cpp accudiag.cpp
OBJECTS = $(SOURCES:.cpp=.o)
PROGRAM = ${MAIN}.${COMPILER}
# uncomment the next to lines for debugging and detailed performance analysis
CXXFLAGS += -g
# -pg slows down the code on my laptop when using CLANG_
#LINKFLAGS += -pg
#CXXFLAGS += -Q --help=optimizers
#CXXFLAGS += -fopt-info
include ../${COMPILER}default.mk
#############################################################################
# additional specific cleaning in this directory
clean_all::
@rm -f t.dat*
#############################################################################
# special testing
# NPROCS = 4
#
TFILE = t.dat
# TTMP = t.tmp
#
graph: $(PROGRAM)
# @rm -f $(TFILE).*
# next two lines only sequentially
./$(PROGRAM)
@mv $(TFILE).000 $(TFILE)
# $(MPIRUN) $(MPIFLAGS) -np $(NPROCS) $(PROGRAM)
# @echo " "; echo "Manipulate data for graphics."; echo " "
# @cat $(TFILE).* > $(TTMP)
# @sort -b -k 2 $(TTMP) -o $(TTMP).1
# @sort -b -k 1 $(TTMP).1 -o $(TTMP).2
# @awk -f nl.awk $(TTMP).2 > $(TFILE)
# @rm -f $(TTMP).* $(TTMP) $(TFILE).*
#
-gnuplot jac.dem

5
sheet3/8/ToDo.txt Normal file
View file

@ -0,0 +1,5 @@
// Jan 15, 2019
geom.h:75 void SetValues(std::vector<double> &v) const; // GH: TODO with functor
Set vector values using a functor ff(x,y).
See solution in Progs/cds

View file

@ -0,0 +1,43 @@
function [ xc, ia, v ] = ascii_read_meshvector( fname )
%
% Loads the 2D triangular mesh (coordinates, vertex connectivity)
% together with values on its vertices from an ASCII file.
% Matlab indexing is stored (starts with 1).
%
% The input file format is compatible
% with Mesh_2d_3_matlab:Write_ascii_matlab(..) in jacobi_oo_stl/geom.h
%
%
% IN: fname - filename
% OUT: xc - coordinates
% ia - mesh connectivity
% v - solution vector
DELIMETER = ' ';
fprintf('Read file %s\n',fname)
% Read mesh constants
nn = dlmread(fname,DELIMETER,[0 0 0 3]); %% row_1, col_1, row_2, col_2 in C indexing!!!
nnode = nn(1);
ndim = nn(2);
nelem = nn(3);
nvert = nn(4);
% Read coordinates
row_start = 0+1;
row_end = 0+nnode;
xc = dlmread(fname,DELIMETER,[row_start 0 row_end ndim-1]);
% Read connectivity
row_start = row_end+1;
row_end = row_end+nelem;
ia = dlmread(fname,DELIMETER,[row_start 0 row_end nvert-1]);
% Read solution
row_start = row_end+1;
row_end = row_end+nnode;
v = dlmread(fname,DELIMETER,[row_start 0 row_end 0]);
end

View file

@ -0,0 +1,49 @@
function ascii_write_mesh( xc, ia, e, basename)
%
% Saves the 2D triangular mesh in the minimal way (only coordinates, vertex connectivity, minimal boundary edge info)
% in an ASCII file.
% Matlab indexing is stored (starts with 1).
%
% The output file format is compatible with Mesh_2d_3_matlab:Mesh_2d_3_matlab(std::string const &fname) in jacobi_oo_stl/geom.h
%
% IN:
% coordinates xc: [2][nnode]
% connectivity ia: [4][nelem] with t(4,:) are the subdomain numbers
% edges e: [7][nedges] boundary edges
% e([1,2],:) - start/end vertex of edge
% e([3,4],:) - start/end values
% e(5,:) - segment number
% e([6,7],:) - left/right subdomain
% basename: file name without extension
%
% Data have been generated via <https://de.mathworks.com/help/pde/ug/initmesh.html initmesh>.
%
fname = [basename, '.txt'];
nnode = int32(size(xc,2));
ndim = int32(size(xc,1));
nelem = int32(size(ia,2));
nvert_e = int32(3);
dlmwrite(fname,nnode,'delimiter','\t','precision',16) % number of nodes
dlmwrite(fname,ndim,'-append','delimiter','\t','precision',16) % space dimension
dlmwrite(fname,nelem,'-append','delimiter','\t','precision',16) % number of elements
dlmwrite(fname,nvert_e,'-append','delimiter','\t','precision',16) % number of vertices per element
% dlmwrite(fname,xc(:),'-append','delimiter','\t','precision',16) % coordinates
dlmwrite(fname,xc([1,2],:).','-append','delimiter','\t','precision',16) % coordinates
% no subdomain info transferred
tmp=int32(ia(1:3,:));
% dlmwrite(fname,tmp(:),'-append','delimiter','\t','precision',16) % connectivity in Matlab indexing
dlmwrite(fname,tmp(:,:).','-append','delimiter','\t','precision',16) % connectivity in Matlab indexing
% store only start and end point of boundary edges,
nbedges = size(e,2);
dlmwrite(fname,nbedges,'-append','delimiter','\t','precision',16) % number boundary edges
tmp=int32(e(1:2,:));
% dlmwrite(fname,tmp(:),'-append','delimiter','\t','precision',16) % boundary edges in Matlab indexing
dlmwrite(fname,tmp(:,:).','-append','delimiter','\t','precision',16) % boundary edges in Matlab indexing
end

522
sheet3/8/geom.cpp Normal file
View file

@ -0,0 +1,522 @@
// see: http://llvm.org/docs/CodingStandards.html#include-style
#include "geom.h"
#include <algorithm>
#include <cassert>
#include <fstream>
#include <iostream>
#include <list>
#include <string>
#include <vector>
using namespace std;
Mesh::Mesh(int ndim, int nvert_e, int ndof_e)
: _nelem(0), _nvert_e(nvert_e), _ndof_e(ndof_e), _nnode(0), _ndim(ndim), _ia(0), _xc(0)
{
}
Mesh::~Mesh()
{}
void Mesh::SetValues(std::vector<double> &v, const std::function<double(double, double)> &func) const
{
int const nnode = Nnodes(); // number of vertices in mesh
assert( nnode == static_cast<int>(v.size()) );
for (int k = 0; k < nnode; ++k)
{
v[k] = func( _xc[2 * k], _xc[2 * k + 1] );
}
}
void Mesh::Debug() const
{
cout << "\n ############### Debug M E S H ###################\n";
cout << "\n ............... Coordinates ...................\n";
for (int k = 0; k < _nnode; ++k)
{
cout << k << " : " ;
for (int i = 0; i < _ndof_e; ++i )
{
cout << _xc[2*k+i] << " ";
}
cout << endl;
}
cout << "\n ............... Elements ...................\n";
for (int k = 0; k < _nelem; ++k)
{
cout << k << " : ";
for (int i = 0; i < _ndof_e; ++i )
cout << _ia[_ndof_e * k + i] << " ";
cout << endl;
}
return;
}
void Mesh::Write_ascii_matlab(std::string const &fname, std::vector<double> const &v) const
{
assert(Nnodes() == static_cast<int>(v.size())); // fits vector length to mesh information?
ofstream fout(fname); // open file ASCII mode
if ( !fout.is_open() )
{
cout << "\nFile " << fname << " has not been opened.\n\n" ;
assert( fout.is_open() && "File not opened." );
}
string const DELIMETER(" "); // define the same delimeter as in matlab/ascii_read*.m
int const OFFSET(1); // convert C-indexing to matlab
// Write data: #nodes, #space dimensions, #elements, #vertices per element
fout << Nnodes() << DELIMETER << Ndims() << DELIMETER << Nelems() << DELIMETER << NverticesElements() << endl;
// Write cordinates: x_k, y_k in seperate lines
assert( Nnodes()*Ndims() == static_cast<int>(_xc.size()));
for (int k = 0, kj = 0; k < Nnodes(); ++k)
{
for (int j = 0; j < Ndims(); ++j, ++kj)
{
fout << _xc[kj] << DELIMETER;
}
fout << endl;
}
// Write connectivity: ia_k,0, ia_k,1 etc in seperate lines
assert( Nelems()*NverticesElements() == static_cast<int>(_ia.size()));
for (int k = 0, kj = 0; k < Nelems(); ++k)
{
for (int j = 0; j < NverticesElements(); ++j, ++kj)
{
fout << _ia[kj] + OFFSET << DELIMETER; // C to matlab
}
fout << endl;
}
// Write vector
for (int k = 0; k < Nnodes(); ++k)
{
fout << v[k] << endl;
}
fout.close();
return;
}
void Mesh::Visualize(std::vector<double> const &v) const
{
// define external command
const string exec_m("matlab -nosplash < visualize_results.m"); // Matlab
//const string exec_m("octave --no-window-system --no-gui visualize_results.m"); // Octave
//const string exec_m("flatpak run org.octave.Octave visualize_results.m"); // Octave (flatpak): desktop GH
const string fname("uv.txt");
Write_ascii_matlab(fname, v);
int ierror = system(exec_m.c_str()); // call external command
if (ierror != 0)
{
cout << endl << "Check path to Matlab/octave on your system" << endl;
}
cout << endl;
return;
}
// #####################################################################
Mesh_2d_3_square::Mesh_2d_3_square(int nx, int ny, int myid, int procx, int procy)
: Mesh(2, 3, 3), // two dimensions, 3 vertices, 3 dofs
_myid(myid), _procx(procx), _procy(procy), _neigh{{-1, -1, -1, -1}}, _color(0),
_xl(0.0), _xr(1.0), _yb(0.0), _yt(1.0), _nx(nx), _ny(ny)
{
//void IniGeom(int const myid, int const procx, int const procy, int neigh[], int &color)
int const ky = _myid / _procx;
int const kx = _myid % _procy; // MOD(myid,procx)
// Determine the neighbors of domain/rank myid
_neigh[0] = (ky == 0) ? -1 : _myid - _procx; // South
_neigh[1] = (kx == _procx - 1) ? -1 : _myid + 1; // East
_neigh[2] = (ky == _procy - 1) ? -1 : _myid + _procx; // North
_neigh[3] = (kx == 0) ? -1 : _myid - 1; // West
_color = (kx + ky) & 1 ;
// subdomain is part of unit square
double const hx = 1. / _procx;
double const hy = 1. / _procy;
_xl = kx * hx; // left
_xr = (kx + 1) * hx; // right
_yb = ky * hy; // bottom
_yt = (ky + 1) * hy; // top
// Calculate coordinates
int const nnode = (_nx + 1) * (_ny + 1); // number of nodes
Resize_Coords(nnode, 2); // coordinates in 2D [nnode][ndim]
GetCoordsInRectangle(_nx, _ny, _xl, _xr, _yb, _yt, GetCoords().data());
// Calculate element connectivity (linear triangles)
int const nelem = 2 * _nx * _ny; // number of elements
Resize_Connectivity(nelem, 3); // connectivity matrix [nelem][3]
GetConnectivityInRectangle(_nx, _ny, GetConnectivity().data());
return;
}
void Mesh_2d_3_square::SetU(std::vector<double> &u) const
{
int dx = _nx + 1;
for (int j = 0; j <= _ny; ++j)
{
int k = j * dx;
for (int i = 0; i <= _nx; ++i, ++k)
{
u[k] = 0.0;
}
}
}
void Mesh_2d_3_square::SetF(std::vector<double> &f) const
{
int dx = _nx + 1;
for (int j = 0; j <= _ny; ++j)
{
int k = j * dx;
for (int i = 0; i <= _nx; ++i, ++k)
{
f[k] = 1.0;
}
}
}
std::vector<int> Mesh_2d_3_square::Index_DirichletNodes() const
{
int const dx = 1,
dy = _nx + 1,
bl = 0,
br = _nx,
tl = _ny * (_nx + 1),
tr = (_ny + 1) * (_nx + 1) - 1;
int const start[4] = { bl, br, tl, bl},
end[4] = { br, tr, tr, tl},
step[4] = { dx, dy, dx, dy};
vector<int> idx(0);
for (int j = 0; j < 4; j++)
{
if (_neigh[j] < 0)
{
for (int i = start[j]; i <= end[j]; i += step[j])
{
idx.push_back(i); // node i is Dirichlet node
}
}
}
// remove multiple elements
sort(idx.begin(), idx.end()); // sort
idx.erase( unique(idx.begin(), idx.end()), idx.end() ); // remove duplicate data
return idx;
}
void Mesh_2d_3_square::SaveVectorP(std::string const &name, vector<double> const &u) const
{
// construct the file name for subdomain myid
const string tmp( std::to_string(_myid / 100) + to_string((_myid % 100) / 10) + to_string(_myid % 10) );
const string namep = name + "." + tmp;
ofstream ff(namep.c_str());
ff.precision(6);
ff.setf(ios::fixed, ios::floatfield);
// assumes tensor product grid in unit square; rowise numbered (as generated in class constructor)
// output is provided for tensor product grid visualization ( similar to Matlab-surf() )
auto const &xc = GetCoords();
int k = 0;
for (int j = 0; j <= _ny; ++j)
{
for (int i = 0; i <= _nx; ++i, ++k)
ff << xc[2 * k + 0] << " " << xc[2 * k + 1] << " " << u[k] << endl;
ff << endl;
}
ff.close();
return;
}
void Mesh_2d_3_square::GetCoordsInRectangle(int const nx, int const ny,
double const xl, double const xr, double const yb, double const yt,
double xc[])
{
const double hx = (xr - xl) / nx,
hy = (yt - yb) / ny;
int k = 0;
for (int j = 0; j <= ny; ++j)
{
const double y0 = yb + j * hy;
for (int i = 0; i <= nx; ++i, k += 2)
{
xc[k ] = xl + i * hx;
xc[k + 1] = y0;
}
}
return;
}
void Mesh_2d_3_square::GetConnectivityInRectangle(int const nx, int const ny, int ia[])
{
const int dx = nx + 1;
int k = 0;
int l = 0;
for (int j = 0; j < ny; ++j, ++k)
{
for (int i = 0; i < nx; ++i, ++k)
{
ia[l ] = k;
ia[l + 1] = k + 1;
ia[l + 2] = k + dx + 1;
l += 3;
ia[l ] = k;
ia[l + 1] = k + dx;
ia[l + 2] = k + dx + 1;
l += 3;
}
}
return;
}
// #################### still some old code (--> MPI) ############################
// Copies the values of w corresponding to the boundary
// South (ib==1), East (ib==2), North (ib==3), West (ib==4)
void GetBound(int const ib, int const nx, int const ny, double const w[], double s[])
{
const int //dx = 1,
dy = nx + 1,
bl = 0,
br = nx,
tl = ny * (nx + 1),
tr = (ny + 1) * (nx + 1) - 1;
switch (ib)
{
case 1:
{
for (int i = bl, j = 0; i <= br; ++i, ++j)
s[j] = w[i];
break;
}
case 3:
{
for (int i = tl, j = 0; i <= tr; ++i, ++j)
s[j] = w[i];
break;
}
case 4:
{
for (int i = bl, j = 0; i <= tl; i += dy, ++j)
s[j] = w[i];
break;
}
case 2:
{
for (int i = br, j = 0; i <= tr; i += dy, ++j)
s[j] = w[i];
break;
}
default:
{
cout << endl << "Wrong parameter ib in " << __FILE__ << ":" << __LINE__ << endl;
}
}
return;
}
// ----------------------------------------------------------------------------------------------------------
// Computes w: = w + s at nodes on the boundary
// South (ib == 1), East (ib == 2), North (ib == 3), West (ib == 4)
void AddBound(int const ib, int const nx, int const ny, double w[], double const s[])
{
int const dy = nx + 1,
bl = 0,
br = nx,
tl = ny * (nx + 1),
tr = (ny + 1) * (nx + 1) - 1;
switch (ib)
{
case 1:
{
for (int i = bl, j = 0; i <= br; ++i, ++j)
w[i] += s[j];
break;
}
case 3:
{
for (int i = tl, j = 0; i <= tr; ++i, ++j)
w[i] += s[j];
break;
}
case 4:
{
for (int i = bl, j = 0; i <= tl; i += dy, ++j)
w[i] += s[j];
break;
}
case 2:
{
for (int i = br, j = 0; i <= tr; i += dy, ++j)
w[i] += s[j];
break;
}
default:
{
cout << endl << "Wrong parameter ib in " << __FILE__ << ":" << __LINE__ << endl;
}
}
return;
}
// ####################################################################
Mesh_2d_3_matlab::Mesh_2d_3_matlab(string const &fname)
: Mesh(2, 3, 3), // two dimensions, 3 vertices, 3 dofs
bedges(0)
{
ifstream ifs(fname);
if (!(ifs.is_open() && ifs.good()))
{
cerr << "Mesh_2d_3_matlab: Error cannot open file " << fname << endl;
assert(ifs.is_open());
}
int const OFFSET(1); // Matlab to C indexing
cout << "ASCI file " << fname << " opened" << endl;
// Read some mesh constants
int nnode, ndim, nelem, nvert_e;
ifs >> nnode >> ndim >> nelem >> nvert_e;
cout << nnode << " " << ndim << " " << nelem << " " << nvert_e << endl;
assert(ndim == 2 && nvert_e == 3);
// Allocate memory
Resize_Coords(nnode, ndim); // coordinates in 2D [nnode][ndim]
Resize_Connectivity(nelem, nvert_e); // connectivity matrix [nelem][nvert]
// Read ccordinates
auto &xc = GetCoords();
for (int k = 0; k < nnode * ndim; ++k)
{
ifs >> xc[k];
}
// Read connectivity
auto &ia = GetConnectivity();
for (int k = 0; k < nelem * nvert_e; ++k)
{
ifs >> ia[k];
ia[k] -= OFFSET; // Matlab to C indexing
}
// additional read of boundary information (only start/end point)
int nbedges;
ifs >> nbedges;
bedges.resize(nbedges * 2);
for (int k = 0; k < nbedges * 2; ++k)
{
ifs >> bedges[k];
bedges[k] -= OFFSET; // Matlab to C indexing
}
return;
}
// binary
//{
//ifstream ifs(fname, ios_base::in | ios_base::binary);
//if(!(ifs.is_open() && ifs.good()))
//{
//cerr << "ReadBinMatrix: Error cannot open file " << file << endl;
//assert(ifs.is_open());
//}
//cout << "ReadBinMatrix: file opened" << file << endl;
//}
// binaryIO.cpp
//void read_binMatrix(const string& file, vector<int> &cnt, vector<int> &col, vector<double> &ele)
//{
//ifstream ifs(file, ios_base::in | ios_base::binary);
//if(!(ifs.is_open() && ifs.good()))
//{
//cerr << "ReadBinMatrix: Error cannot open file " << file << endl;
//assert(ifs.is_open());
//}
//cout << "ReadBinMatrix: Opened file " << file << endl;
//int _size;
//ifs.read(reinterpret_cast<char*>(&_size), sizeof(int)); // old: ifs.read((char*)&_size, sizeof(int));
//cnt.resize(_size);
//cout << "ReadBinMatrix: cnt size: " << _size << endl;
//ifs.read((char*)&_size, sizeof(int));
//col.resize(_size);
//cout << "ReadBinMatrix: col size: " << _size << endl;
//ifs.read((char*)&_size, sizeof(int));
//ele.resize(_size);
//cout << "ReadBinMatrix: ele size: " << _size << endl;
//ifs.read((char*)cnt.data(), cnt.size() * sizeof(int));
//ifs.read((char*)col.data(), col.size() * sizeof(int));
//ifs.read((char*)ele.data(), ele.size() * sizeof(double));
//ifs.close();
//cout << "ReadBinMatrix: Finished reading matrix.." << endl;
//}
std::vector<int> Mesh_2d_3_matlab::Index_DirichletNodes() const
{
vector<int> idx(bedges); // copy
sort(idx.begin(), idx.end()); // sort
idx.erase( unique(idx.begin(), idx.end()), idx.end() ); // remove duplicate data
return idx;
}

381
sheet3/8/geom.h Normal file
View file

@ -0,0 +1,381 @@
#ifndef GEOM_FILE
#define GEOM_FILE
#include <array>
#include <functional> // function; C++11
#include <string>
#include <vector>
/**
* Basis class for finite element meshes.
*/
class Mesh
{
public:
/**
* Constructor initializing the members with default values.
*
* @param[in] ndim space dimensions (dimension for coordinates)
* @param[in] nvert_e number of vertices per element (dimension for connectivity)
* @param[in] ndof_e degrees of freedom per element (= @p nvert_e for linear elements)
*/
explicit Mesh(int ndim, int nvert_e = 0, int ndof_e = 0);
/**
* Destructor.
*
* See clang warning on
* <a href="https://stackoverflow.com/questions/28786473/clang-no-out-of-line-virtual-method-definitions-pure-abstract-c-class/40550578">weak-vtables</a>.
*/
virtual ~Mesh();
/**
* Number of finite elements in (sub)domain.
* @return number of elements.
*/
int Nelems() const
{
return _nelem;
}
/**
* Global number of vertices for each finite element.
* @return number of vertices per element.
*/
int NverticesElements() const
{
return _nvert_e;
}
/**
* Global number of degrees of freedom (dof) for each finite element.
* @return degrees of freedom per element.
*/
int NdofsElement() const
{
return _ndof_e;
}
/**
* Number of vertices in mesh.
* @return number of vertices.
*/
int Nnodes() const
{
return _nnode;
}
/**
* Space dimension.
* @return number of dimensions.
*/
int Ndims() const
{
return _ndim;
}
/**
* (Re-)Allocates memory for the element connectivity and redefines the appropriate dimensions.
*
* @param[in] nelem number of elements
* @param[in] nvert_e number of vertices per element
*/
void Resize_Connectivity(int nelem, int nvert_e)
{
SetNelem(nelem); // number of elements
SetNverticesElement(nvert_e); // vertices per element
_ia.resize(nelem * nvert_e);
}
/**
* Read connectivity information (g1,g2,g3)_i.
* @return convectivity vector [nelems*ndofs].
*/
const std::vector<int> &GetConnectivity() const
{
return _ia;
}
/**
* Access/Change connectivity information (g1,g2,g3)_i.
* @return convectivity vector [nelems*ndofs].
*/
std::vector<int> &GetConnectivity()
{
return _ia;
}
/**
* (Re-)Allocates memory for the element connectivity and redefines the appropriate dimensions.
*
* @param[in] nnodes number of nodes
* @param[in] ndim space dimension
*/
void Resize_Coords(int nnodes, int ndim)
{
SetNnode(nnodes); // number of nodes
SetNdim(ndim); // space dimension
_xc.resize(nnodes * ndim);
}
/**
* Read coordinates of vertices (x,y)_i.
* @return coordinates vector [nnodes*2].
*/
const std::vector<double> &GetCoords() const
{
return _xc;
}
/**
* Access/Change coordinates of vertices (x,y)_i.
* @return coordinates vector [nnodes*2].
*/
std::vector<double> &GetCoords()
{
return _xc;
}
/**
* Calculate values in vector @p v via function @p func(x,y)
* @param[in] v vector
* @param[in] func function of (x,y) returning a double value.
*/
void SetValues(std::vector<double> &v, const std::function<double(double, double)> &func) const;
/**
* Prints the information for a finite element mesh
*/
void Debug() const;
/**
* Determines the indices of those vertices with Dirichlet boundary conditions
* @return index vector.
*/
virtual std::vector<int> Index_DirichletNodes() const = 0;
/**
* Write vector @p v toghether with its mesh information to an ASCii file @p fname.
*
* The data are written in C-style.
*
* @param[in] fname file name
* @param[in] v vector
*/
void Write_ascii_matlab(std::string const &fname, std::vector<double> const &v) const;
/**
* Visualize @p v together with its mesh information via matlab or octave.
*
* Comment/uncomment those code lines in method Mesh:Visualize (geom.cpp)
* that are supported on your system.
*
* @param[in] v vector
*
* @warning matlab files ascii_read_meshvector.m visualize_results.m
* must be in the executing directory.
*/
void Visualize(std::vector<double> const &v) const;
protected:
void SetNelem(int nelem)
{
_nelem = nelem;
}
void SetNverticesElement(int nvert)
{
_nvert_e = nvert;
}
void SetNdofsElement(int ndof)
{
_ndof_e = ndof;
}
void SetNnode(int nnode)
{
_nnode = nnode;
}
void SetNdim(int ndim)
{
_ndim = ndim;
}
private:
int _nelem; //!< number elements
int _nvert_e; //!< number of vertices per element
int _ndof_e; //!< degrees of freedom (d.o.f.) per element
int _nnode; //!< number nodes/vertices
int _ndim; //!< space dimension of the problem (1, 2, or 3)
std::vector<int> _ia; //!< element connectivity
std::vector<double> _xc; //!< coordinates
};
/**
* 2D finite element mesh of the square consiting of linear triangular elements.
*/
class Mesh_2d_3_square: public Mesh
{
public:
/**
* Generates the f.e. mesh for the unit square.
*
* @param[in] nx number of discretization intervals in x-direction
* @param[in] ny number of discretization intervals in y-direction
* @param[in] myid my MPI-rank / subdomain
* @param[in] procx number of ranks/subdomains in x-direction
* @param[in] procy number of processes in y-direction
*/
Mesh_2d_3_square(int nx, int ny, int myid = 0, int procx = 1, int procy = 1);
/**
* Destructor
*/
~Mesh_2d_3_square() override
{}
/**
* Set solution vector based on a tensor product grid in the rectangle.
* @param[in] u solution vector
*/
void SetU(std::vector<double> &u) const;
/**
* Set right hand side (rhs) vector on a tensor product grid in the rectangle.
* @param[in] f rhs vector
*/
void SetF(std::vector<double> &f) const;
/**
* Determines the indices of those vertices with Dirichlet boundary conditions
* @return index vector.
*/
std::vector<int> Index_DirichletNodes() const override;
/**
* Stores the values of vector @p u of (sub)domain into a file @p name for further processing in gnuplot.
* The file stores rowise the x- and y- coordinates together with the value from @p u .
* The domain [@p xl, @p xr] x [@p yb, @p yt] is discretized into @p nx x @p ny intervals.
*
* @param[in] name basename of file name (file name will be extended by the rank number)
* @param[in] u local vector
*
* @warning Assumes tensor product grid in unit square; rowise numbered
* (as generated in class constructor).
* The output is provided for tensor product grid visualization
* ( similar to Matlab-surf() ).
*
* @see Mesh_2d_3_square
*/
void SaveVectorP(std::string const &name, std::vector<double> const &u) const;
// here will still need to implement in the class
// GetBound(), AddBound()
// or better a generalized way with indices and their appropriate ranks for MPI communication
private:
/**
* Determines the coordinates of the dicretization nodes of the domain [@p xl, @p xr] x [@p yb, @p yt]
* which is discretized into @p nx x @p ny intervals.
*
* @param[in] ny number of discretization intervals in y-direction
* @param[in] xl x-coordinate of left boundary
* @param[in] xr x-coordinate of right boundary
* @param[in] yb y-coordinate of lower boundary
* @param[in] yt y-coordinate of upper boundary
* @param[out] xc coordinate vector of length 2n with x(2*k,2*k+1) as coodinates of node k
*/
void GetCoordsInRectangle(int nx, int ny, double xl, double xr, double yb, double yt,
double xc[]);
/**
* Determines the element connectivity of linear triangular elements of a FEM discretization
* of a rectangle using @p nx x @p ny equidistant intervals for discretization.
* @param[in] nx number of discretization intervals in x-direction
* @param[in] ny number of discretization intervals in y-direction
* @param[out] ia element connectivity matrix with ia(3*s,3*s+1,3*s+2) as node numbers od element s
*/
void GetConnectivityInRectangle(int nx, int ny, int ia[]);
private:
int _myid; //!< my MPI rank
int _procx; //!< number of MPI ranks in x-direction
int _procy; //!< number of MPI ranks in y-direction
std::array<int, 4> _neigh; //!< MPI ranks of neighbors (negative: no neighbor but b.c.)
int _color; //!< red/black coloring (checker board) of subdomains
double _xl; //!< x coordinate of lower left corner of square
double _xr; //!< x coordinate of lower right corner of square
double _yb; //!< y coordinate or lower left corner of square
double _yt; //!< y coordinate of upper right corner of square
int _nx; //!< number of intervals in x-direction
int _ny; //!< number of intervals in y-direction
};
// #################### still some old code (--> MPI) ############################
/**
* Copies the values of @p w corresponding to boundary @p ib
* onto vector s. South (ib==1), East (ib==2), North (ib==3), West (ib==4).
* The vector @p s has to be long enough!!
* @param[in] ib my local boundary
* @param[in] nx number of discretization intervals in x-direction
* @param[in] ny number of discretization intervals in y-direction
* @param[in] w vector for all nodes of local discretization
* @param[out] s short vector with values on boundary @p ib
*/
// GH_NOTE: Absicherung bei s !!
void GetBound(int ib, int nx, int ny, double const w[], double s[]);
/**
* Computes @p w := @p w + @p s at the interface/boundary nodes on the
* boundary @p ib . South (ib==1), East (ib==2), North (ib==3), West (ib==4)
* @param[in] ib my local boundary
* @param[in] nx number of discretization intervals in x-direction
* @param[in] ny number of discretization intervals in y-direction
* @param[in,out] w vector for all nodes of local discretization
* @param[in] s short vector with values on boundary @p ib
*/
void AddBound(int ib, int nx, int ny, double w[], double const s[]);
// #################### Mesh from Matlab ############################
/**
* 2D finite element mesh of the square consiting of linear triangular elements.
*/
class Mesh_2d_3_matlab: public Mesh
{
public:
/**
* Reads mesh data from a binary file.
*
* File format, see ascii_write_mesh.m
*
* @param[in] fname file name
*/
explicit Mesh_2d_3_matlab(std::string const &fname);
/**
* Determines the indices of those vertices with Dirichlet boundary conditions.
* @return index vector.
*
* @warning All boundary nodes are considered as Dirchlet nodes.
*/
std::vector<int> Index_DirichletNodes() const override;
private:
/**
* Determines the indices of those vertices with Dirichlet boundary conditions
* @return index vector.
*/
int Nnbedges() const
{
return static_cast<int>(bedges.size());
}
std::vector<int> bedges; //!< boundary edges [nbedges][2] storing start/end vertex
};
#endif

348
sheet3/8/getmatrix.cpp Normal file
View file

@ -0,0 +1,348 @@
#include "getmatrix.h"
#include "userset.h"
#include <algorithm>
#include <cassert>
#include <cmath>
#include <iomanip>
#include <iostream>
#include <list>
#include <vector>
using namespace std;
// general routine for lin. triangular elements
void CalcElem(int const ial[3], double const xc[], double ske[3][3], double fe[3])
//void CalcElem(const int* __restrict__ ial, const double* __restrict__ xc, double* __restrict__ ske[3], double* __restrict__ fe)
{
const int i1 = 2 * ial[0], i2 = 2 * ial[1], i3 = 2 * ial[2];
const double x13 = xc[i3 + 0] - xc[i1 + 0], y13 = xc[i3 + 1] - xc[i1 + 1],
x21 = xc[i1 + 0] - xc[i2 + 0], y21 = xc[i1 + 1] - xc[i2 + 1],
x32 = xc[i2 + 0] - xc[i3 + 0], y32 = xc[i2 + 1] - xc[i3 + 1];
const double jac = fabs(x21 * y13 - x13 * y21);
ske[0][0] = 0.5 / jac * (y32 * y32 + x32 * x32);
ske[0][1] = 0.5 / jac * (y13 * y32 + x13 * x32);
ske[0][2] = 0.5 / jac * (y21 * y32 + x21 * x32);
ske[1][0] = ske[0][1];
ske[1][1] = 0.5 / jac * (y13 * y13 + x13 * x13);
ske[1][2] = 0.5 / jac * (y21 * y13 + x21 * x13);
ske[2][0] = ske[0][2];
ske[2][1] = ske[1][2];
ske[2][2] = 0.5 / jac * (y21 * y21 + x21 * x21);
const double xm = (xc[i1 + 0] + xc[i2 + 0] + xc[i3 + 0]) / 3.0,
ym = (xc[i1 + 1] + xc[i2 + 1] + xc[i3 + 1]) / 3.0;
//fe[0] = fe[1] = fe[2] = 0.5 * jac * FunctF(xm, ym) / 3.0;
fe[0] = fe[1] = fe[2] = 0.5 * jac * fNice(xm, ym) / 3.0;
}
// general routine for lin. triangular elements,
// non-symm. matrix
// node numbering in element: a s c e n d i n g indices !!
[[deprecated("Use CRS_Matrix::AddElem_3(...) instead.")]]
void AddElem(int const ial[3], double const ske[3][3], double const fe[3],
int const id[], int const ik[], double sk[], double f[])
{
for (int i = 0; i < 3; ++i)
{
const int ii = ial[i], // row ii (global index)
id1 = id[ii], // start and
id2 = id[ii + 1]; // end of row ii in matrix
int ip = id1;
for (int j = 0; j < 3; ++j) // no symmetry assumed
{
const int jj = ial[j];
bool not_found = true;
do // find entry jj (global index) in row ii
{
not_found = (ik[ip] != jj);
++ip;
}
while (not_found && ip < id2);
#ifndef NDEBUG // compiler option -DNDEBUG switches off the check
if (not_found) // no entry found !!
{
cout << "Error in AddElem: (" << ii << "," << jj << ") ["
<< ial[0] << "," << ial[1] << "," << ial[2] << "]\n";
assert(!not_found);
}
#endif
sk[ip - 1] += ske[i][j];
}
f[ii] += fe[i];
}
}
// ----------------------------------------------------------------------------
// ####################################################################
CRS_Matrix::CRS_Matrix(Mesh const &mesh)
: _mesh(mesh), _nrows(0), _nnz(0), _id(0), _ik(0), _sk(0)
{
Derive_Matrix_Pattern();
return;
}
void CRS_Matrix::Derive_Matrix_Pattern()
{
int const nelem(_mesh.Nelems());
int const ndof_e(_mesh.NdofsElement());
auto const &ia(_mesh.GetConnectivity());
// Determine the number of matrix rows
_nrows = *max_element(ia.cbegin(), ia.cbegin() + ndof_e * nelem);
++_nrows; // node numberng: 0 ... nnode-1
assert(*min_element(ia.cbegin(), ia.cbegin() + ndof_e * nelem) == 0); // numbering starts with 0 ?
// Collect for each node those nodes it is connected to (multiple entries)
// Detect the neighboring nodes
vector< list<int> > cc(_nrows); // cc[i] is the list of nodes a node i is connected to
for (int i = 0; i < nelem; ++i)
{
int const idx = ndof_e * i;
for (int k = 0; k < ndof_e; ++k)
{
list<int> &cck = cc.at(ia[idx + k]);
cck.insert( cck.end(), ia.cbegin() + idx, ia.cbegin() + idx + ndof_e );
}
}
// Delete the multiple entries
_nnz = 0;
for (auto &it : cc)
{
it.sort();
it.unique();
_nnz += static_cast<int>(it.size());
// cout << it.size() << " :: "; copy(it->begin(),it->end(), ostream_iterator<int,char>(cout," ")); cout << endl;
}
// CSR data allocation
_id.resize(_nrows + 1); // Allocate memory for CSR row pointer
_ik.resize(_nnz); // Allocate memory for CSR column index vector
// copy CSR data
_id[0] = 0; // begin of first row
for (size_t i = 0; i < cc.size(); ++i)
{
//cout << i << " " << nid.at(i) << endl;;
const list<int> &ci = cc.at(i);
const auto nci = static_cast<int>(ci.size());
_id[i + 1] = _id[i] + nci; // begin of next line
copy(ci.begin(), ci.end(), _ik.begin() + _id[i] );
}
assert(_nnz == _id[_nrows]);
_sk.resize(_nnz); // Allocate memory for CSR column index vector
return;
}
void CRS_Matrix::Debug() const
{
// ID points to first entry of row
// no symmetry assumed
cout << "\nMatrix (nnz = " << _id[_nrows] << ")\n";
for (int row = 0; row < _nrows; ++row)
{
cout << "Row " << row << " : ";
int const id1 = _id[row];
int const id2 = _id[row + 1];
for (int j = id1; j < id2; ++j)
{
cout.setf(ios::right, ios::adjustfield);
cout << "[" << setw(2) << _ik[j] << "] " << setw(4) << _sk[j] << " ";
}
cout << endl;
}
return;
}
void CRS_Matrix::CalculateLaplace(vector<double> &f)
{
assert(_mesh.NdofsElement() == 3); // only for triangular, linear elements
//cout << _nnz << " vs. " << _id[_nrows] << " " << _nrows<< endl;
assert(_nnz == _id[_nrows]);
for (int k = 0; k < _nrows; ++k)
{
_sk[k] = 0.0;
}
for (int k = 0; k < _nrows; ++k)
{
f[k] = 0.0;
}
double ske[3][3], fe[3];
// Loop over all elements
auto const nelem = _mesh.Nelems();
auto const &ia = _mesh.GetConnectivity();
auto const &xc = _mesh.GetCoords();
for (int i = 0; i < nelem; ++i)
{
CalcElem(ia.data() + 3 * i, xc.data(), ske, fe);
AddElem_3(ia.data() + 3 * i, ske, fe, f);
}
//Debug();
return;
}
void CRS_Matrix::ApplyDirichletBC(std::vector<double> const &u, std::vector<double> &f)
{
double const PENALTY = 1e6;
auto const idx = _mesh.Index_DirichletNodes();
int const nidx = static_cast<int>(idx.size());
for (int row = 0; row < nidx; ++row)
{
int const k = idx[row];
int const id1 = fetch(k, k); // Find diagonal entry of row
assert(id1 >= 0);
_sk[id1] += PENALTY; // matrix weighted scaling feasible
f[k] += PENALTY * u[k];
}
return;
}
void CRS_Matrix::GetDiag(vector<double> &d) const
{
assert( _nrows == static_cast<int>(d.size()) );
for (int row = 0; row < _nrows; ++row)
{
const int ia = fetch(row, row); // Find diagonal entry of row
assert(ia >= 0);
d[row] = _sk[ia];
}
return;
}
bool CRS_Matrix::Compare2Old(int nnode, int const id[], int const ik[], double const sk[]) const
{
bool bn = (nnode == _nrows); // number of rows
if (!bn)
{
cout << "######### Error: " << "number of rows" << endl;
}
bool bz = (id[nnode] == _nnz); // number of non zero elements
if (!bz)
{
cout << "######### Error: " << "number of non zero elements" << endl;
}
bool bd = equal(id, id + nnode + 1, _id.cbegin()); // row starts
if (!bd)
{
cout << "######### Error: " << "row starts" << endl;
}
bool bk = equal(ik, ik + id[nnode], _ik.cbegin()); // column indices
if (!bk)
{
cout << "######### Error: " << "column indices" << endl;
}
bool bv = equal(sk, sk + id[nnode], _sk.cbegin()); // values
if (!bv)
{
cout << "######### Error: " << "values" << endl;
}
return bn && bz && bd && bk && bv;
}
void CRS_Matrix::Mult(vector<double> &w, vector<double> const &u) const
{
assert( _nrows == static_cast<int>(w.size()) );
assert( w.size() == u.size() );
for (int row = 0; row < _nrows; ++row)
{
double wi = 0.0;
for (int ij = _id[row]; ij < _id[row + 1]; ++ij)
{
wi += _sk[ij] * u[ _ik[ij] ];
}
w[row] = wi;
}
return;
}
void CRS_Matrix::Defect(vector<double> &w,
vector<double> const &f, vector<double> const &u) const
{
assert( _nrows == static_cast<int>(w.size()) );
assert( w.size() == u.size() && u.size() == f.size() );
for (int row = 0; row < _nrows; ++row)
{
double wi = f[row];
for (int ij = _id[row]; ij < _id[row + 1]; ++ij)
{
wi -= _sk[ij] * u[ _ik[ij] ];
}
w[row] = wi;
}
return;
}
int CRS_Matrix::fetch(int const row, int const col) const
{
int const id2 = _id[row + 1]; // end and
int ip = _id[row]; // start of recent row (global index)
while (ip < id2 && _ik[ip] != col) // find index col (global index)
{
++ip;
}
if (ip >= id2)
{
ip = -1;
#ifndef NDEBUG // compiler option -DNDEBUG switches off the check
cout << "No column " << col << " in row " << row << endl;
assert(ip >= id2);
#endif
}
return ip;
}
// general routine for lin. triangular elements,
// non-symm. matrix
// node numbering in element: a s c e n d i n g indices !!
void CRS_Matrix::AddElem_3(int const ial[3], double const ske[3][3], double const fe[3], vector<double> &f)
{
for (int i = 0; i < 3; ++i)
{
const int ii = ial[i]; // row ii (global index)
for (int j = 0; j < 3; ++j) // no symmetry assumed
{
const int jj = ial[j]; // column jj (global index)
int ip = fetch(ii, jj); // find column entry jj in row ii
#ifndef NDEBUG // compiler option -DNDEBUG switches off the check
if (ip < 0) // no entry found !!
{
cout << "Error in AddElem: (" << ii << "," << jj << ") ["
<< ial[0] << "," << ial[1] << "," << ial[2] << "]\n";
assert(ip >= 0);
}
#endif
_sk[ip] += ske[i][j];
}
f[ii] += fe[i];
}
}

178
sheet3/8/getmatrix.h Normal file
View file

@ -0,0 +1,178 @@
#ifndef GETMATRIX_FILE
#define GETMATRIX_FILE
#include "geom.h"
#include <vector>
/**
* Calculates the element stiffness matrix @p ske and the element load vector @p fe
* of one triangular element with linear shape functions.
* @param[in] ial node indices of the three element vertices
* @param[in] xc vector of node coordinates with x(2*k,2*k+1) as coodinates of node k
* @param[out] ske element stiffness matrix
* @param[out] fe element load vector
*/
void CalcElem(int const ial[3], double const xc[], double ske[3][3], double fe[3]);
/**
* Adds the element stiffness matrix @p ske and the element load vector @p fe
* of one triangular element with linear shape functions to the appropriate positions in
* the symmetric stiffness matrix, stored as CSR matrix K(@p sk,@p id, @p ik)
*
* @param[in] ial node indices of the three element vertices
* @param[in] ske element stiffness matrix
* @param[in] fe element load vector
* @param[out] sk vector non-zero entries of CSR matrix
* @param[in] id index vector containing the first entry in a CSR row
* @param[in] ik column index vector of CSR matrix
* @param[out] f distributed local vector storing the right hand side
*
* @warning Algorithm requires indices in connectivity @p ial in ascending order.
* Currently deprecated.
*/
void AddElem(int const ial[3], double const ske[3][3], double const fe[3],
int const id[], int const ik[], double sk[], double f[]);
// #####################################################################
/**
* Square matrix in CRS format (compressed row storage; also named CSR),
* see an <a href="https://en.wikipedia.org/wiki/Sparse_matrix">introduction</a>.
*/
class CRS_Matrix
{
public:
/**
* Intializes the CRS matrix structure from the given discetization in @p mesh.
*
* The sparse matrix pattern is generated but the values are 0.
*
* @param[in] mesh given discretization
*
* @warning A reference to the discretization @p mesh is stored inside this class.
* Therefore, changing @p mesh outside requires also
* to call method @p Derive_Matrix_Pattern explicitely.
*
* @see Derive_Matrix_Pattern
*/
explicit CRS_Matrix(Mesh const & mesh);
/**
* Destructor.
*/
~CRS_Matrix()
{}
/**
* Generates the sparse matrix pattern and overwrites the existing pattern.
*
* The sparse matrix pattern is generated but the values are 0.
*/
void Derive_Matrix_Pattern();
/**
* Calculates the entries of f.e. stiffness matrix and load/rhs vector @p f for the Laplace operator in 2D.
* No memory is allocated.
*
* @param[in,out] f (preallocated) rhs/load vector
*/
void CalculateLaplace(std::vector<double> &f);
/**
* Applies Dirichlet boundary conditions to stiffness matrix and to load vector @p f.
* The <a href="https://www.jstor.org/stable/2005611?seq=1#metadata_info_tab_contents">penalty method</a>
* is used for incorporating the given values @p u.
*
* @param[in] u (global) vector with Dirichlet data
* @param[in,out] f load vector
*/
void ApplyDirichletBC(std::vector<double> const &u, std::vector<double> &f);
/**
* Extracts the diagonal elemenst of the sparse matrix.
*
* @param[in,out] d (prellocated) vector of diagonal elements
*/
void GetDiag(std::vector<double> &d) const;
/**
* Performs the matrix-vector product w := K*u.
*
* @param[in,out] w resulting vector (preallocated)
* @param[in] u vector
*/
void Mult(std::vector<double> &w, std::vector<double> const &u) const;
/**
* Calculates the defect/residuum w := f - K*u.
*
* @param[in,out] w resulting vector (preallocated)
* @param[in] f load vector
* @param[in] u vector
*/
void Defect(std::vector<double> &w,
std::vector<double> const &f, std::vector<double> const &u) const;
/**
* Number rows in matrix.
* @return number of rows.
*/
int Nrows() const
{return _nrows;}
/**
* Show the matrix entries.
*/
void Debug() const;
/**
* Finds in a CRS matrix the access index for an entry at row @p row and column @p col.
*
* @param[in] row row index
* @param[in] col column index
* @return index for element (@p row, @p col). If no appropriate entry exists then -1 will be returned.
*
* @warning assert() stops the function in case that matrix element (@p row, @p col) doesn't exist.
*/
int fetch(int row, int col) const;
/**
* Adds the element stiffness matrix @p ske and the element load vector @p fe
* of one triangular element with linear shape functions to the appropriate positions in
* the stiffness matrix, stored as CSR matrix K(@p sk,@p id, @p ik).
*
* @param[in] ial node indices of the three element vertices
* @param[in] ske element stiffness matrix
* @param[in] fe element load vector
* @param[in,out] f distributed local vector storing the right hand side
*
* @warning Algorithm assumes linear triangular elements (ndof_e==3).
*/
void AddElem_3(int const ial[3], double const ske[3][3], double const fe[3], std::vector<double> &f);
/**
* Compare @p this CRS matrix with an external CRS matrix stored in C-Style.
*
* The method prints statements on differences found.
*
* @param[in] nnode row number of external matrix
* @param[in] id start indices of matrix rows of external matrix
* @param[in] ik column indices of external matrix
* @param[in] sk non-zero values of external matrix
*
* @return true iff all data are identical.
*/
bool Compare2Old(int nnode, int const id[], int const ik[], double const sk[]) const;
private:
Mesh const & _mesh; //!< reference to discretization
int _nrows; //!< number of rows in matrix
int _nnz; //!< number of non-zero entries
std::vector<int> _id; //!< start indices of matrix rows
std::vector<int> _ik; //!< column indices
std::vector<double> _sk; //!< non-zero values
};
#endif

5
sheet3/8/gnuplot.rot Normal file
View file

@ -0,0 +1,5 @@
zrot=(zrot+10)%360
xrot=(xrot+17)%180
set view xrot,zrot
replot
reread

7
sheet3/8/gprofng_script1 Normal file
View file

@ -0,0 +1,7 @@
# max 5 lines
limit 5
# define metrics
metrics name:e.llm
# show absolute numbers
compare on
functions

7
sheet3/8/gprofng_script2 Normal file
View file

@ -0,0 +1,7 @@
# max 5 lines
limit 5
# define metrics
metrics name:e.llm
# show absolute numbers
compare ratio
functions

21
sheet3/8/jac.dem Normal file
View file

@ -0,0 +1,21 @@
set style data lines
set parametric
set hidden3d
set nokey
#set xrange [0:1]
#set yrange [-0:1]
#set zrange [-2:2]
set cntrparam levels 15
set contour base
set title "Solution"
xrot=60
zrot=0
splot "t.dat"
#splot "lsg.gnu"
pause -1 "Press ENTER to continue."
#load "gnuplot.rot"
#set title ""
#set autosc
#set nohidden
#set nopara
#set key

75
sheet3/8/jacobi.cbp Normal file
View file

@ -0,0 +1,75 @@
<?xml version="1.0" encoding="UTF-8" standalone="yes" ?>
<CodeBlocks_project_file>
<FileVersion major="1" minor="6" />
<Project>
<Option title="jacobi" />
<Option pch_mode="2" />
<Option compiler="gcc" />
<Build>
<Target title="Debug">
<Option output="bin/Debug/jacobi" prefix_auto="1" extension_auto="1" />
<Option object_output="obj/Debug/" />
<Option type="1" />
<Option compiler="gcc" />
<Compiler>
<Add option="-g" />
</Compiler>
</Target>
<Target title="Release">
<Option output="bin/Release/jacobi" prefix_auto="1" extension_auto="1" />
<Option object_output="obj/Release/" />
<Option type="1" />
<Option compiler="gcc" />
<Compiler>
<Add option="-O2" />
</Compiler>
<Linker>
<Add option="-s" />
</Linker>
</Target>
</Build>
<Compiler>
<Add option="-Wshadow" />
<Add option="-Winit-self" />
<Add option="-Wredundant-decls" />
<Add option="-Wcast-align" />
<Add option="-Wundef" />
<Add option="-Wfloat-equal" />
<Add option="-Wunreachable-code" />
<Add option="-Wmissing-declarations" />
<Add option="-Wswitch-default" />
<Add option="-Weffc++" />
<Add option="-Wmain" />
<Add option="-pedantic" />
<Add option="-Wextra" />
<Add option="-Wall" />
<Add option="-fexceptions" />
</Compiler>
<Unit filename="geom.cpp" />
<Unit filename="geom.h" />
<Unit filename="getmatrix.cpp" />
<Unit filename="getmatrix.h" />
<Unit filename="jacsolve.cpp" />
<Unit filename="jacsolve.h" />
<Unit filename="main.cpp" />
<Unit filename="userset.cpp" />
<Unit filename="userset.h" />
<Unit filename="vdop.cpp" />
<Unit filename="vdop.h" />
<Extensions>
<code_completion />
<envvars />
<lib_finder disable_auto="1" />
<debugger />
<DoxyBlocks>
<comment_style block="0" line="0" />
<doxyfile_project />
<doxyfile_build extract_all="1" />
<doxyfile_warnings />
<doxyfile_output />
<doxyfile_dot class_diagrams="1" have_dot="1" />
<general use_at_in_tags="1" />
</DoxyBlocks>
</Extensions>
</Project>
</CodeBlocks_project_file>

4
sheet3/8/jacobi.layout Normal file
View file

@ -0,0 +1,4 @@
<?xml version="1.0" encoding="UTF-8" standalone="yes" ?>
<CodeBlocks_layout_file>
<ActiveTarget name="Debug" />
</CodeBlocks_layout_file>

61
sheet3/8/jacsolve.cpp Normal file
View file

@ -0,0 +1,61 @@
#include "vdop.h"
#include "getmatrix.h"
#include "jacsolve.h"
#include <cassert>
#include <cmath>
#include <iostream>
#include <vector>
using namespace std;
// #####################################################################
// const int neigh[], const int color,
// const MPI::Intracomm& icomm,
void JacobiSolve(CRS_Matrix const &SK, vector<double> const &f, vector<double> &u)
{
const double omega = 1.0;
const int maxiter = 1000;
const double tol = 1e-5, // tolerance
tol2 = tol * tol; // tolerance^2
int nrows = SK.Nrows(); // number of rows == number of columns
assert( nrows == static_cast<int>(f.size()) && f.size() == u.size() );
cout << endl << " Start Jacobi solver for " << nrows << " d.o.f.s" << endl;
// Choose initial guess
for (int k = 0; k < nrows; ++k)
{
u[k] = 0.0; // u := 0
}
vector<double> dd(nrows); // matrix diagonal
vector<double> r(nrows); // residual
vector<double> w(nrows); // correction
SK.GetDiag(dd); // dd := diag(K)
////DebugVector(dd);{int ijk; cin >> ijk;}
// Initial sweep
SK.Defect(r, f, u); // r := f - K*u
vddiv(w, r, dd); // w := D^{-1}*r
double sigma0 = dscapr(w, r); // s0 := <w,r>
// Iteration sweeps
int iter = 0;
double sigma = sigma0;
while ( sigma > tol2 * sigma0 && maxiter > iter)
{
++iter;
vdaxpy(u, u, omega, w ); // u := u + om*w
SK.Defect(r, f, u); // r := f - K*u
vddiv(w, r, dd); // w := D^{-1}*r
sigma = dscapr(w, r); // s0 := <w,r>
// cout << "Iteration " << iter << " : " << sqrt(sigma/sigma0) << endl;
}
cout << "aver. Jacobi rate : " << exp(log(sqrt(sigma / sigma0)) / iter) << " (" << iter << " iter)" << endl;
cout << "final error: " << sqrt(sigma / sigma0) << " (rel) " << sqrt(sigma) << " (abs)\n";
return;
}

18
sheet3/8/jacsolve.h Normal file
View file

@ -0,0 +1,18 @@
#ifndef JACSOLVE_FILE
#define JACSOLVE_FILE
#include "getmatrix.h"
#include <vector>
/**
* Solves linear system of equations K @p u = @p f via the Jacobi iteration.
* We use a distributed symmetric CSR matrix @p SK and initial guess of the
* solution is set to 0.
* @param[in] SK CSR matrix
* @param[in] f distributed local vector storing the right hand side
* @param[out] u accumulated local vector storing the solution.
*/
void JacobiSolve(CRS_Matrix const &SK, std::vector<double> const &f, std::vector<double> &u);
#endif

129
sheet3/8/main.cpp Normal file
View file

@ -0,0 +1,129 @@
// MPI code in C++.
// See [Gropp/Lusk/Skjellum, "Using MPI", p.33/41 etc.]
// and /opt/mpich/include/mpi2c++/comm.h for details
#include "geom.h"
#include "getmatrix.h"
#include "jacsolve.h"
#include "userset.h"
#include "vdop.h"
#include <chrono> // timing
#include <cmath>
#include <iostream>
using namespace std;
using namespace std::chrono; // timing
int main(int, char ** )
{
const int numprocs = 1;
const int myrank = 0;
if (myrank == 0)
{
cout << "\n There are " << numprocs << " processes running.\n \n";
}
const auto procx = static_cast<int>(sqrt(numprocs + 0.0));
const int procy = procx;
if (procy * procx != numprocs)
{
cout << "\n Wrong number of processors !\n \n";
}
else
{
// #####################################################################
// Here starts the real code
// #####################################################################
//bool ScaleUp = !true;
int nx, ny, NXglob, NYglob; /* number of local intervals on (xl,xr)=:nx, (yb,yt)=:ny */
//nx = 1024;
//ny = 1024;
nx = 100;
ny = 100;
NXglob = nx * procx;
NYglob = ny * procy;
cout << "Intervalls: " << NXglob << " x " << NYglob << endl;
// ##################### STL ###########################################
{
Mesh_2d_3_square const mesh(nx, ny);
//mesh.Debug();
CRS_Matrix SK(mesh); // CRS matrix
//SK.Debug();
vector<double> uv(SK.Nrows(), 0.0); // temperature
vector<double> fv(SK.Nrows(), 0.0); // r.h.s.
SK.CalculateLaplace(fv);
//SK.Debug();
//mesh.SetU(uv); // deprecated
//mesh.SetF(fv); // deprecated
// Two ways to initialize the vector
//mesh.SetValues(uv,f_zero); // functional
mesh.SetValues(uv, [](double x, double y) -> double {return 0.0 * x *y;} ); // lambda function
SK.ApplyDirichletBC(uv, fv);
//SK.Compare2Old(nnode, id, ik, sk);
//SK.Debug();
auto tstart = system_clock::now(); // start timer
JacobiSolve(SK, fv, uv ); // solve the system of equations
auto tend = system_clock::now(); // end timer
auto duration = duration_cast<microseconds>(tend - tstart);
auto t1 = static_cast<double>(duration.count()) / 1e6 ; // t1 in seconds
cout << "JacobiSolve: timing in sec. : " << t1 << endl;
//CompareVectors(uv, nnode, u, 1e-6); // Check correctness
//mesh.SaveVectorP("t.dat", uv);
//mesh.Visualize(uv);
}
// ##################### STL ###########################################
{
//Mesh_2d_3_matlab const mesh("square_tiny.txt");
Mesh_2d_3_matlab const mesh("square_100.txt");
//Mesh_2d_3_matlab const mesh("L_shape.txt");
//mesh.Debug();
CRS_Matrix SK(mesh); // CRS matrix
//SK.Debug();
vector<double> uv(SK.Nrows(), 0.0); // temperature
vector<double> fv(SK.Nrows(), 0.0); // r.h.s.
SK.CalculateLaplace(fv);
//SK.Debug();
//mesh.SetU(uv); // deprecated
// Two ways to initialize the vector
//mesh.SetValues(uv,f_zero); // user function
mesh.SetValues(uv, [](double x, double y) -> double {return 0.0 * x *y;} ); // lambda function
SK.ApplyDirichletBC(uv, fv);
//SK.Compare2Old(nnode, id, ik, sk);
//SK.Debug();
auto tstart = system_clock::now(); // start timer
JacobiSolve(SK, fv, uv ); // solve the system of equations
auto tend = system_clock::now(); // end timer
auto duration = duration_cast<microseconds>(tend - tstart);
auto t1 = static_cast<double>(duration.count()) / 1e6 ; // t1 in seconds
cout << "JacobiSolve: timing in sec. : " << t1 << endl;
//mesh.Write_ascii_matlab("uv.txt", uv);
//mesh.Visualize(uv);
}
return 0;
}
}

20
sheet3/8/nl.awk Normal file
View file

@ -0,0 +1,20 @@
#
# Have to add a newline for a new row of coordinates
#
BEGIN { OFS=" "; YO=-1.23456789; X=YO; Y=YO; Z=YO }
{
if ($1!="")
{
if ($1!=YO) { print " "; YO=$1 }
if ($1==X && $2==Y)
{
# print $1,$2,($3+Z)/2
}
else
{
print $1,$2,$3
}
X=$1; Y=$2; Z=$3;
}
}
END {}

9
sheet3/8/out_100_GCC.txt Normal file
View file

@ -0,0 +1,9 @@
There are 1 processes running.
Intervalls: 100 x 100
Start Jacobi solver for 10201 d.o.f.s
aver. Jacobi rate : 0.997922 (1000 iter)
final error: 0.124971 (rel) 0.000194029 (abs)
JacobiSolve: timing in sec. : 0.155127

24
sheet3/8/output.txt Normal file
View file

@ -0,0 +1,24 @@
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o main.o main.cpp
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o vdop.o vdop.cpp
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o geom.o geom.cpp
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o getmatrix.o getmatrix.cpp
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o jacsolve.o jacsolve.cpp
g++ -c -g -O0 -funroll-all-loops -std=c++17 -Wall -pedantic -Wextra -Weffc++ -Woverloaded-virtual -Wfloat-equal -Wshadow -Wredundant-decls -Winline -fmax-errors=1 -flto -o userset.o userset.cpp
g++ main.o vdop.o geom.o getmatrix.o jacsolve.o userset.o -O0 -llapack -lblas -flto -o main.GCC_
./main.GCC_
There are 1 processes running.
Intervalls: 100 x 100
Start Jacobi solver for 10201 d.o.f.s
aver. Jacobi rate : 0.997922 (1000 iter)
final error: 0.124971 (rel) 0.000194029 (abs)
JacobiSolve: timing in sec. : 0.799123
ASCI file square_100.txt opened
17361 2 34320 3
Start Jacobi solver for 17361 d.o.f.s
aver. Jacobi rate : 0.998401 (1000 iter)
final error: 0.201744 (rel) 0.000265133 (abs)
JacobiSolve: timing in sec. : 1.54385

1826
sheet3/8/small_Doxyfile Normal file

File diff suppressed because it is too large Load diff

41
sheet3/8/square.m Normal file
View file

@ -0,0 +1,41 @@
% Square:
% flatpak run org.octave.Octave <filename>
% or
% octave --no-window-system --no-gui -qf <filename>
clear all
clc
% %% L-shape
% g=[2 0 2 0 0 1 0; % #vertices,v_1x, v_2x, v_1y, v_2y, subdomain_left, subdomain_right
% 2 2 2 0 1 1 0;
% 2 2 1 1 0.5 1 0;
% 2 1 1 0.5 2 1 0;
% 2 1 0 2 2 1 0;
% 2 0 0 2 0 1 0]';
%% square
g=[2 0 1 0 0 1 0; % #vertices,v_1x, v_2x, v_1y, v_2y, subdomain_left, subdomain_right
2 1 1 0 1 1 0;
2 1 0 1 1 1 0;
2 0 0 1 0 1 0]';
[p,e,t] = initmesh(g,'hmax',0.01);
pdemesh(p,e,t)
%% GH
% output from <https://de.mathworks.com/help/pde/ug/initmesh.html initmesh>
%
% coordinates p: [2][nnode]
% connectivity t: [4][nelem] with t(4,:) are the subdomain numbers
% edges e: [7][nedges] boundary edges
% e([1,2],:) - start/end vertex of edge
% e([3,4],:) - start/end values
% e(5,:) - segment number
% e([6,7],:) - left/right subdomain
ascii_write_mesh( p, t, e, mfilename);
% tmp=t(1:3,:)

138487
sheet3/8/square_100.txt Normal file

File diff suppressed because it is too large Load diff

95
sheet3/8/square_tiny.txt Normal file
View file

@ -0,0 +1,95 @@
13
2
16
3
0
0
1
0
1
1
0
1
0.5
0
1
0.5
0.5
1
0
0.5
0.4999999999999999
0.4999999999999999
0.3333333333333333
0.6666666666666666
0.6666666666666666
0.6666666666666666
0.6666666666666666
0.3333333333333333
0.3333333333333333
0.3333333333333333
8
1
13
5
2
12
6
3
11
7
4
10
1
5
13
10
8
13
2
6
12
3
7
11
4
8
10
12
9
13
10
9
11
7
10
11
11
9
12
6
11
12
9
10
13
5
12
13
8
1
5
5
2
2
6
6
3
3
7
7
4
4
8
8
1

16
sheet3/8/userset.cpp Normal file
View file

@ -0,0 +1,16 @@
#include "userset.h"
#include <cmath>
double FunctF(double const x, double const y)
{
// return std::sin(3.14159*1*x)*std::sin(3.14159*1*y);
// return 16.0*1024. ;
// return (double)1.0 ;
return x * x * std::sin(2.5 * 3.14159 * y);
}
double FunctU(const double /* x */, double const /* y */)
{
return 1.0 ;
}

44
sheet3/8/userset.h Normal file
View file

@ -0,0 +1,44 @@
#ifndef USERSET_FILE
#define USERSET_FILE
#include <cmath>
/**
* User function: f(@p x,@p y)
* @param[in] x x-coordinate of discretization point
* @param[in] y y-coordinate of discretization point
* @return value for right hand side f(@p x,@p y)
*/
double FunctF(double const x, double const y);
/**
* User function: u(@p x,@p y)
* @param[in] x x-coordinate of discretization point
* @param[in] y y-coordinate of discretization point
* @return value for solution vector u(@p x,@p y)
*/
double FunctU(double const x, double const y);
/**
* User function: f(@p x,@p y) = @f$ x^2 \sin(2.5\pi y)@f$.
* @param[in] x x-coordinate of discretization point
* @param[in] y y-coordinate of discretization point
* @return value f(@p x,@p y)
*/
inline double fNice(double const x, double const y)
{
return x * x * std::sin(2.5 * 3.14159 * y);
}
/**
* User function: f(@p x,@p y) = 0$.
* @param[in] x x-coordinate of discretization point
* @param[in] y y-coordinate of discretization point
* @return value 0
*/
inline double f_zero(double const x, double const y)
//double f_zero(double const /*x*/, double const /*y*/)
{
return 0.0 + 0.0*(x+y);
}
#endif

69043
sheet3/8/uv.txt Normal file

File diff suppressed because it is too large Load diff

84
sheet3/8/vdop.cpp Normal file
View file

@ -0,0 +1,84 @@
#include "vdop.h"
#include <cassert> // assert()
#include <cmath>
#include <iostream>
#include <vector>
using namespace std;
void vddiv(vector<double> &x, vector<double> const &y,
vector<double> const &z)
{
assert( x.size() == y.size() && y.size() == z.size() );
size_t n = x.size();
for (size_t k = 0; k < n; ++k)
{
x[k] = y[k] / z[k];
}
return;
}
//******************************************************************************
void vdaxpy(std::vector<double> &x, std::vector<double> const &y,
double alpha, std::vector<double> const &z )
{
assert( x.size() == y.size() && y.size() == z.size() );
size_t n = x.size();
for (size_t k = 0; k < n; ++k)
{
x[k] = y[k] + alpha * z[k];
}
return;
}
//******************************************************************************
double dscapr(std::vector<double> const &x, std::vector<double> const &y)
{
assert( x.size() == y.size());
size_t n = x.size();
double s = 0.0;
for (size_t k = 0; k < n; ++k)
{
s += x[k] * y[k];
}
return s;
}
//******************************************************************************
void DebugVector(vector<double> const &v)
{
cout << "\nVector (nnode = " << v.size() << ")\n";
for (size_t j = 0; j < v.size(); ++j)
{
cout.setf(ios::right, ios::adjustfield);
cout << v[j] << " ";
}
cout << endl;
return;
}
//******************************************************************************
bool CompareVectors(std::vector<double> const &x, int const n, double const y[], double const eps)
{
bool bn = (static_cast<int>(x.size()) == n);
if (!bn)
{
cout << "######### Error: " << "number of elements" << endl;
}
//bool bv = equal(x.cbegin(),x.cend(),y);
bool bv = equal(x.cbegin(), x.cend(), y,
[eps](double a, double b) -> bool
{ return std::abs(a - b) < eps * (1.0 + 0.5 * (std::abs(a) + std::abs(a))); }
);
if (!bv)
{
assert(static_cast<int>(x.size()) == n);
cout << "######### Error: " << "values" << endl;
}
return bn && bv;
}

58
sheet3/8/vdop.h Normal file
View file

@ -0,0 +1,58 @@
#ifndef VDOP_FILE
#define VDOP_FILE
#include <vector>
/** @brief Element-wise vector divison x_k = y_k/z_k.
*
* @param[out] x target vector
* @param[in] y source vector
* @param[in] z source vector
*
*/
void vddiv(std::vector<double> & x, std::vector<double> const& y,
std::vector<double> const& z);
/** @brief Element-wise daxpy operation x(k) = y(k) + alpha*z(k).
*
* @param[out] x target vector
* @param[in] y source vector
* @param[in] alpha scalar
* @param[in] z source vector
*
*/
void vdaxpy(std::vector<double> & x, std::vector<double> const& y,
double alpha, std::vector<double> const& z );
/** @brief Calculates the Euclidian inner product of two vectors.
*
* @param[in] x vector
* @param[in] y vector
* @return Euclidian inner product @f$\langle x,y \rangle@f$
*
*/
double dscapr(std::vector<double> const& x, std::vector<double> const& y);
/**
* Print entries of a vector.
* @param[in] v vector values
*/
void DebugVector(std::vector<double> const &v);
/** @brief Compares an STL vector with POD vector.
*
* The accuracy criteria @f$ |x_k-y_k| < \varepsilon \left({1+0.5(|x_k|+|y_k|)}\right) @f$
* follows the book by
* <a href="https://www.springer.com/la/book/9783319446592">Stoyan/Baran</a>, p.8.
*
* @param[in] x STL vector
* @param[in] n length of POD vector
* @param[in] y POD vector
* @param[in] eps relative accuracy criteria (default := 0.0).
* @return true iff pairwise vector elements are relatively close to each other.
*
*/
bool CompareVectors(std::vector<double> const& x, int n, double const y[], double const eps=0.0);
#endif

View file

@ -0,0 +1,20 @@
%% Visualize results
%
% flatpak run org.octave.Octave <filename>
% or
% octave --no-window-system --no-gui -qf <filename>
%
% or
% matlab -nosplash < <filename>
clear all
clc
%%
fname = 'uv.txt';
[xc,ia,v] = ascii_read_meshvector(fname);
h = trisurf(ia, xc(:,1), xc(:,2), v);
waitfor(h) % wait for closing the figure