diff options
author | Andreas Müller <andreas.mueller@ost.ch> | 2021-08-12 09:20:10 +0200 |
---|---|---|
committer | GitHub <noreply@github.com> | 2021-08-12 09:20:10 +0200 |
commit | 650bc249248c5f2ce6ca6288766a35a76437ed19 (patch) | |
tree | 573728c63e80a25884128c247770a214f1f481a7 /buch | |
parent | Merge pull request #83 from rfritsche/mceliece (diff) | |
parent | typo (diff) | |
download | SeminarMatrizen-650bc249248c5f2ce6ca6288766a35a76437ed19.tar.gz SeminarMatrizen-650bc249248c5f2ce6ca6288766a35a76437ed19.zip |
Merge pull request #84 from Nunigan/master
Multiplikation #4
Diffstat (limited to '')
37 files changed, 1152 insertions, 336 deletions
diff --git a/buch/papers/multiplikation/code/MM b/buch/papers/multiplikation/code/MM Binary files differdeleted file mode 100755 index d52dda4..0000000 --- a/buch/papers/multiplikation/code/MM +++ /dev/null diff --git a/buch/papers/multiplikation/code/MM.c b/buch/papers/multiplikation/code/MM.c index a897d4f..2588262 100755 --- a/buch/papers/multiplikation/code/MM.c +++ b/buch/papers/multiplikation/code/MM.c @@ -28,11 +28,12 @@ int main() { // omp_set_num_threads(4);
// run_algo(openMP_MM, "openMP_MM",0);
run_algo(MM_dc, "MM_dc",0);
+
run_algo(strassen, "strassen",0);
run_algo(MM, "MM", 0);
- run_algo(winograd, "winograd", 0);
- run_algo_cblas(0);
+ run_algo(winograd, "winograd", 0);
+ run_algo_cblas(0);
return 0;
}
@@ -414,12 +415,12 @@ void run_algo(void (*algo)(), char alog_name[], int print) for(int i=0; i<n_arrays; ++i)
{
- for(int j = 0; j<1; ++j)
+ for(int j = 0; j<10; ++j)
{
- int *C = (int*) malloc(n[i] * n[i] * sizeof(int));
- double dtime = omp_get_wtime();
- algo(Ap[i], Bp[i], (int*) C, n[i]);
- dtime = omp_get_wtime() - dtime;
+ int *C = (int*) malloc(n[i] * n[i] * sizeof(int));
+ double dtime = omp_get_wtime();
+ algo(Ap[i], Bp[i], (int*) C, n[i]);
+ dtime = omp_get_wtime() - dtime;
// printf("The %s program took %f seconds to execute \n", alog_name, dtime);
fprintf(fptr, "%f,%d\n", dtime, n[i]);
@@ -428,7 +429,7 @@ void run_algo(void (*algo)(), char alog_name[], int print) printMatrix((int*)C, n[i]);
}
free(C);
- }
+ }
}
fclose(fptr);
@@ -442,7 +443,7 @@ void run_algo_cblas(int print) fptr = fopen("meas/blas.txt", "w");
for(int i=0; i<n_arrays; ++i)
{
- for(int j = 0; j<1; ++j)
+ for(int j = 0; j<10; ++j)
{
double *dC = (double*) malloc(n[i] * n[i] * sizeof(double));
double dtime = omp_get_wtime();
diff --git a/buch/papers/multiplikation/code/MM.py b/buch/papers/multiplikation/code/MM.py index 7220ae1..8057850 100644 --- a/buch/papers/multiplikation/code/MM.py +++ b/buch/papers/multiplikation/code/MM.py @@ -5,6 +5,7 @@ Created on Fri Mar 19 07:31:29 2021 @author: nunigan """ +import scipy.stats import numpy as np import time import matplotlib.pyplot as plt @@ -133,9 +134,6 @@ def winograd2(A, B): def test_perfomance(n): - import mkl - mkl.set_num_threads(1) - t_mm = [] t_mm_dc = [] t_mm_strassen = [] @@ -148,21 +146,21 @@ def test_perfomance(n): # A = np.random.randint(-100, 100,(i, i)) # B = np.random.randint(-100, 100,(i, i)) - # start = time.time() - # C3 = strassen(A, B) - # t_mm_strassen.append(time.time() - start) + start = time.time() + C3 = strassen(A, B) + t_mm_strassen.append(time.time() - start) - # start = time.time() - # C1 = MM(A, B) - # t_mm.append(time.time() - start) + start = time.time() + C1 = MM(A, B) + t_mm.append(time.time() - start) - # start = time.time() - # C2 = MM_dc(A, B) - # t_mm_dc.append(time.time() - start) + start = time.time() + C2 = MM_dc(A, B) + t_mm_dc.append(time.time() - start) - # start = time.time() - # C4 = winograd2(A, B) - # t_wino.append(time.time() - start) + start = time.time() + C4 = winograd2(A, B) + t_wino.append(time.time() - start) start = time.time() C = A@B @@ -173,10 +171,10 @@ def test_perfomance(n): plt.rc('axes', labelsize=23) plt.rc('xtick', labelsize=23) plt.rc('ytick', labelsize=23) - # plt.plot(n, t_mm, label='Standard', lw=5) - # plt.plot(n, t_mm_dc, label='Divide and conquer', lw=5) - # plt.plot(n, t_mm_strassen, label='Strassen', lw=5) - # plt.plot(n, t_wino, label='Winograd', lw=5) + plt.plot(n, t_mm, label='Standard', lw=5) + plt.plot(n, t_mm_dc, label='Divide and conquer', lw=5) + plt.plot(n, t_mm_strassen, label='Strassen', lw=5) + plt.plot(n, t_wino, label='Winograd', lw=5) plt.plot(n, t_np, label='NumPy A@B', lw=5) # plt.xscale('log', base=2) plt.legend() @@ -186,9 +184,9 @@ def test_perfomance(n): plt.tight_layout() # plt.yscale('log') plt.legend(fontsize=19) - # plt.savefig('meas_' + str(max(n))+ '.pdf') - # arr = np.array([n, t_mm, t_mm_dc, t_mm_strassen, t_wino, t_np]) - # np.savetxt('meas_' + str(max(n))+ '.txt',arr) + plt.savefig('meas_' + str(max(n))+ '.pdf') + arr = np.array([n, t_mm, t_mm_dc, t_mm_strassen, t_wino, t_np]) + np.savetxt('meas_' + str(max(n))+ '.txt',arr) return t_np @@ -249,6 +247,8 @@ def plot_c_res(ave, num): # blas_t = np.mean(blas_t.reshape(-1,ave),axis=1) # blas_n = np.mean(blas_n.reshape(-1,ave),axis=1) + + def func(x, a,b): return b*x**a @@ -261,11 +261,11 @@ def plot_c_res(ave, num): plt.rc('axes', labelsize=23) plt.rc('xtick', labelsize=23) plt.rc('ytick', labelsize=23) - plt.loglog(MM_n, MM_t, label='3 For Loops', lw=5) - plt.loglog(winograd_n, winograd_t, label='Winograd MM', lw=5) - plt.loglog(blas_n, blas_t, label='Blas', lw=5) - plt.loglog(strassen_n, strassen_t, label='Strassen', lw=5) - plt.loglog(MM_dc_n, MM_dc_t, label='Divide and Conquer', lw=5) + plt.loglog(MM_n, MM_t, '.', label='3 For Loops', lw=5) + plt.loglog(winograd_n, winograd_t, '.', label='Winograd MM', lw=5) + plt.loglog(blas_n, blas_t, '.', label='Blas', lw=5) + plt.loglog(strassen_n, strassen_t, '.', label='Strassen', lw=5) + plt.loglog(MM_dc_n, MM_dc_t, '.', label='Divide and Conquer', lw=5) plt.xlabel("n") # plt.yscale('log', base=10) # plt.xscale('log', base=2) @@ -281,16 +281,33 @@ def plot_c_res(ave, num): plt.legend() # return [MM_n,winograd_n,blas_n,strassen_n,MM_dc_n] + + return [MM_t,winograd_t,blas_t,strassen_t,MM_dc_t] +def mean_confidence_interval(data, confidence=0.95): + a = 1.0 * np.array(data) + n = len(a) + m, se = np.mean(a), scipy.stats.sem(a) + h = se * scipy.stats.t.ppf((1 + confidence) / 2., n-1) + return m, h + # test%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if __name__ == '__main__': - # A = plot_c_res(1, 4096) - - - arr = plot(1024) + # A = plot_c_res(10, 4096) + # name = ['MM', 'Wino', 'blas', 'strassen', 'dc'] + # for i in range(5): + # ci_inner = [] + # print(name[i]) + # for j in range(11): + # m,h=mean_confidence_interval(A[i][j*10:(j+1)*10]) + # print("({},{})".format(2**(j+1),m)) + # np.savetxt('meas/ci/' + name[i]+'.txt',ci_inner) + + arr = plot(4096) # n = np.logspace(1,12,12,base=2,dtype=(np.int)) + # n=[2048,4096] # n = np.arange(1,50,2) # A = np.random.randint(-10, 6, (5,3)) # B = np.random.randint(-10, 6, (3,5)) diff --git a/buch/papers/multiplikation/code/c_matrix.h b/buch/papers/multiplikation/code/c_matrix.h index 14389fc..63d5390 100644 --- a/buch/papers/multiplikation/code/c_matrix.h +++ b/buch/papers/multiplikation/code/c_matrix.h @@ -1,101 +1,177 @@ -/* Seminar Matrizen, autogenerated File, Michael Schmid, 02/08/2021, 22:48:43 */ +/* Seminar Matrizen, autogenerated File, Michael Schmid, 10/08/2021, 05:46:32 */ #include <stdint.h> const int A0[][2] = { - {75,47}, - {-41,-24} + {60,-84}, + {-66,-1} }; const int B0[][2] = { - {-53,-95}, - {-93,30} + {-45,87}, + {-38,-73} }; const double dB0[][2] = { - {-53,-95}, - {-93,30} + {-45,87}, + {-38,-73} }; const double dA0[][2] = { - {75,47}, - {-41,-24} + {60,-84}, + {-66,-1} }; const int A1[][4] = { - {47,11,-66,8}, - {36,98,39,82}, - {-32,12,40,-79}, - {61,-20,-85,-98} + {-72,-19,-91,62}, + {-36,-74,-44,-47}, + {-39,-31,50,-93}, + {-81,2,-17,-86} }; const int B1[][4] = { - {37,75,-53,9}, - {37,-33,-67,38}, - {70,39,-93,43}, - {43,41,23,-4} + {-66,39,-23,52}, + {-88,-13,13,-13}, + {-45,-70,28,-20}, + {96,5,88,96} }; const double dB1[][4] = { - {37,75,-53,9}, - {37,-33,-67,38}, - {70,39,-93,43}, - {43,41,23,-4} + {-66,39,-23,52}, + {-88,-13,13,-13}, + {-45,-70,28,-20}, + {96,5,88,96} }; const double dA1[][4] = { - {47,11,-66,8}, - {36,98,39,82}, - {-32,12,40,-79}, - {61,-20,-85,-98} + {-72,-19,-91,62}, + {-36,-74,-44,-47}, + {-39,-31,50,-93}, + {-81,2,-17,-86} }; const int A2[][8] = { - {-54,-87,87,69,52,-21,-86,55}, - {19,-75,-61,-50,-55,-23,66,-92}, - {-73,-67,-36,19,84,-11,24,46}, - {-98,62,-76,57,-100,6,-23,-51}, - {62,46,1,-64,42,-9,85,-12}, - {35,-59,-17,-47,78,86,-50,74}, - {-15,45,33,-59,-9,-81,49,96}, - {-57,22,-43,7,-30,-45,-5,13} + {-36,-2,-58,-32,34,-89,49,-55}, + {-68,-73,52,-3,-51,-37,-31,70}, + {73,-90,-21,-79,-15,96,-99,12}, + {68,-25,38,-73,-60,35,-99,72}, + {-43,-87,48,-84,-100,37,80,53}, + {-27,88,-5,-82,-57,-27,20,10}, + {-91,-47,54,-90,-99,-76,50,-18}, + {69,-36,76,5,-67,-38,-95,91} }; const int B2[][8] = { - {-71,-82,-80,-78,83,-97,48,-24}, - {15,75,15,-60,-63,-53,1,-50}, - {-84,63,67,-2,78,93,-13,95}, - {61,-26,-88,56,56,27,26,1}, - {2,54,21,36,9,-41,53,53}, - {85,-11,42,-51,-6,3,27,97}, - {10,-2,90,-76,-75,0,8,-37}, - {10,-64,47,-69,66,-50,89,-66} + {-84,22,-13,-66,-42,51,66,0}, + {37,-65,66,-85,-10,-23,77,5}, + {1,41,-79,0,63,-37,-10,29}, + {72,66,-99,92,-28,65,25,-40}, + {69,-49,65,-18,64,-97,-47,30}, + {36,86,66,-12,-17,89,1,-37}, + {-100,11,27,23,-75,-23,96,-9}, + {68,90,-87,-99,-70,-28,98,-76} }; const double dB2[][8] = { - {-71,-82,-80,-78,83,-97,48,-24}, - {15,75,15,-60,-63,-53,1,-50}, - {-84,63,67,-2,78,93,-13,95}, - {61,-26,-88,56,56,27,26,1}, - {2,54,21,36,9,-41,53,53}, - {85,-11,42,-51,-6,3,27,97}, - {10,-2,90,-76,-75,0,8,-37}, - {10,-64,47,-69,66,-50,89,-66} + {-84,22,-13,-66,-42,51,66,0}, + {37,-65,66,-85,-10,-23,77,5}, + {1,41,-79,0,63,-37,-10,29}, + {72,66,-99,92,-28,65,25,-40}, + {69,-49,65,-18,64,-97,-47,30}, + {36,86,66,-12,-17,89,1,-37}, + {-100,11,27,23,-75,-23,96,-9}, + {68,90,-87,-99,-70,-28,98,-76} }; const double dA2[][8] = { - {-54,-87,87,69,52,-21,-86,55}, - {19,-75,-61,-50,-55,-23,66,-92}, - {-73,-67,-36,19,84,-11,24,46}, - {-98,62,-76,57,-100,6,-23,-51}, - {62,46,1,-64,42,-9,85,-12}, - {35,-59,-17,-47,78,86,-50,74}, - {-15,45,33,-59,-9,-81,49,96}, - {-57,22,-43,7,-30,-45,-5,13} - }; -const int *Ap[3] = {(int*) A0,(int*) A1,(int*) A2}; -const int *Bp[3] = {(int*) B0,(int*) B1,(int*) B2}; -const double *dAp[3] = {(double*) dA0,(double*) dA1,(double*) dA2}; -const double *dBp[3] = {(double*) dB0,(double*) dB1,(double*) dB2}; -int n[3] = {2,4,8}; -int n_arrays = 3; + {-36,-2,-58,-32,34,-89,49,-55}, + {-68,-73,52,-3,-51,-37,-31,70}, + {73,-90,-21,-79,-15,96,-99,12}, + {68,-25,38,-73,-60,35,-99,72}, + {-43,-87,48,-84,-100,37,80,53}, + {-27,88,-5,-82,-57,-27,20,10}, + {-91,-47,54,-90,-99,-76,50,-18}, + {69,-36,76,5,-67,-38,-95,91} + }; +const int A3[][16] = + { + {-24,65,21,19,94,70,-90,-81,53,-41,-23,-1,58,-80,-54,59}, + {-42,76,-19,98,29,-56,92,14,45,11,82,83,48,-13,81,66}, + {43,-57,-67,95,5,72,11,0,-47,55,-24,36,84,54,-31,-54}, + {-39,-40,19,97,-82,-56,27,95,81,-21,-50,-74,-35,-87,-28,-26}, + {-74,-98,79,92,-24,-48,99,94,55,-83,70,98,-24,18,-67,14}, + {20,76,11,-23,-56,21,0,42,64,86,-74,44,93,-76,-30,97}, + {13,20,-73,-11,-30,80,53,-8,60,21,17,-42,82,-72,-6,-80}, + {36,-93,-64,-21,20,-85,15,24,99,81,-52,64,71,-56,52,63}, + {32,9,-2,-85,17,62,-98,-35,75,-58,-44,-20,-47,89,-95,52}, + {93,-43,86,68,-6,-25,90,57,60,-10,65,-97,43,46,-60,-41}, + {43,-33,0,50,-100,26,-60,95,39,-70,-61,-81,9,-23,-99,-4}, + {20,61,15,43,-96,93,-55,38,-29,-1,-10,26,-87,18,64,6}, + {-98,-84,51,16,-14,86,52,59,44,-39,-2,10,82,-66,54,19}, + {89,-49,-37,-6,-53,40,-11,46,-51,-56,86,34,11,13,-20,-49}, + {-90,14,28,-45,-25,-56,-51,-61,28,-8,51,91,95,-10,-85,58}, + {8,-44,88,-71,-27,11,89,37,86,-78,-44,-56,-87,0,-42,-61} + }; +const int B3[][16] = + { + {62,-30,62,92,29,-93,-95,44,-33,-88,-29,9,-88,-42,-90,-70}, + {60,37,-44,-93,-87,6,-53,2,-29,53,-49,59,6,83,-15,50}, + {-19,85,-49,-14,84,-4,12,88,-83,-81,-24,-16,-12,-42,-63,-71}, + {-42,-78,-58,-61,-29,67,-28,-46,64,7,6,-13,88,-42,95,-24}, + {-90,-56,8,-30,-89,70,37,-29,24,-8,-10,-2,-25,-63,-95,-91}, + {10,-81,42,-28,-13,-68,-72,-20,-22,5,-79,-50,-88,62,57,69}, + {-67,24,-71,-43,11,48,33,-93,-82,-65,-4,5,-15,25,-54,-45}, + {-49,19,-29,90,-97,-87,78,-39,-75,-85,-79,-35,54,3,-73,7}, + {-7,39,70,-42,32,-100,56,4,-24,-57,38,-49,-50,-44,79,-42}, + {37,-65,-55,22,-97,-42,-76,95,97,-27,38,11,0,-81,-23,35}, + {26,-70,10,-29,47,-70,-52,29,-13,-18,5,34,18,32,87,91}, + {-84,41,-19,96,-51,-19,81,75,81,92,2,-40,-42,-69,-10,-61}, + {-30,98,71,-51,91,-59,58,86,86,-22,-84,7,66,-55,-52,23}, + {-71,-44,-9,90,26,18,26,-10,-85,64,-47,3,72,81,74,-8}, + {52,-59,-91,22,8,-63,84,9,-11,-54,-78,-71,-98,42,96,57}, + {18,-39,34,-50,-62,-96,-2,-78,52,94,-33,2,-19,-9,-86,-75} + }; +const double dB3[][16] = + { + {62,-30,62,92,29,-93,-95,44,-33,-88,-29,9,-88,-42,-90,-70}, + {60,37,-44,-93,-87,6,-53,2,-29,53,-49,59,6,83,-15,50}, + {-19,85,-49,-14,84,-4,12,88,-83,-81,-24,-16,-12,-42,-63,-71}, + {-42,-78,-58,-61,-29,67,-28,-46,64,7,6,-13,88,-42,95,-24}, + {-90,-56,8,-30,-89,70,37,-29,24,-8,-10,-2,-25,-63,-95,-91}, + {10,-81,42,-28,-13,-68,-72,-20,-22,5,-79,-50,-88,62,57,69}, + {-67,24,-71,-43,11,48,33,-93,-82,-65,-4,5,-15,25,-54,-45}, + {-49,19,-29,90,-97,-87,78,-39,-75,-85,-79,-35,54,3,-73,7}, + {-7,39,70,-42,32,-100,56,4,-24,-57,38,-49,-50,-44,79,-42}, + {37,-65,-55,22,-97,-42,-76,95,97,-27,38,11,0,-81,-23,35}, + {26,-70,10,-29,47,-70,-52,29,-13,-18,5,34,18,32,87,91}, + {-84,41,-19,96,-51,-19,81,75,81,92,2,-40,-42,-69,-10,-61}, + {-30,98,71,-51,91,-59,58,86,86,-22,-84,7,66,-55,-52,23}, + {-71,-44,-9,90,26,18,26,-10,-85,64,-47,3,72,81,74,-8}, + {52,-59,-91,22,8,-63,84,9,-11,-54,-78,-71,-98,42,96,57}, + {18,-39,34,-50,-62,-96,-2,-78,52,94,-33,2,-19,-9,-86,-75} + }; +const double dA3[][16] = + { + {-24,65,21,19,94,70,-90,-81,53,-41,-23,-1,58,-80,-54,59}, + {-42,76,-19,98,29,-56,92,14,45,11,82,83,48,-13,81,66}, + {43,-57,-67,95,5,72,11,0,-47,55,-24,36,84,54,-31,-54}, + {-39,-40,19,97,-82,-56,27,95,81,-21,-50,-74,-35,-87,-28,-26}, + {-74,-98,79,92,-24,-48,99,94,55,-83,70,98,-24,18,-67,14}, + {20,76,11,-23,-56,21,0,42,64,86,-74,44,93,-76,-30,97}, + {13,20,-73,-11,-30,80,53,-8,60,21,17,-42,82,-72,-6,-80}, + {36,-93,-64,-21,20,-85,15,24,99,81,-52,64,71,-56,52,63}, + {32,9,-2,-85,17,62,-98,-35,75,-58,-44,-20,-47,89,-95,52}, + {93,-43,86,68,-6,-25,90,57,60,-10,65,-97,43,46,-60,-41}, + {43,-33,0,50,-100,26,-60,95,39,-70,-61,-81,9,-23,-99,-4}, + {20,61,15,43,-96,93,-55,38,-29,-1,-10,26,-87,18,64,6}, + {-98,-84,51,16,-14,86,52,59,44,-39,-2,10,82,-66,54,19}, + {89,-49,-37,-6,-53,40,-11,46,-51,-56,86,34,11,13,-20,-49}, + {-90,14,28,-45,-25,-56,-51,-61,28,-8,51,91,95,-10,-85,58}, + {8,-44,88,-71,-27,11,89,37,86,-78,-44,-56,-87,0,-42,-61} + }; +const int *Ap[4] = {(int*) A0,(int*) A1,(int*) A2,(int*) A3}; +const int *Bp[4] = {(int*) B0,(int*) B1,(int*) B2,(int*) B3}; +const double *dAp[4] = {(double*) dA0,(double*) dA1,(double*) dA2,(double*) dA3}; +const double *dBp[4] = {(double*) dB0,(double*) dB1,(double*) dB2,(double*) dB3}; +int n[4] = {2,4,8,16}; +int n_arrays = 4; diff --git a/buch/papers/multiplikation/code/c_meas_4096.pdf b/buch/papers/multiplikation/code/c_meas_4096.pdf Binary files differindex 5236afb..f637ae4 100644 --- a/buch/papers/multiplikation/code/c_meas_4096.pdf +++ b/buch/papers/multiplikation/code/c_meas_4096.pdf diff --git a/buch/papers/multiplikation/code/ci.txt b/buch/papers/multiplikation/code/ci.txt new file mode 100644 index 0000000..e69de29 --- /dev/null +++ b/buch/papers/multiplikation/code/ci.txt diff --git a/buch/papers/multiplikation/code/helper_class.py b/buch/papers/multiplikation/code/helper_class.py index 485fa76..3b74f67 100755 --- a/buch/papers/multiplikation/code/helper_class.py +++ b/buch/papers/multiplikation/code/helper_class.py @@ -101,5 +101,6 @@ if __name__ == '__main__': helper = Helper() # n = np.arange(2,10) - n = np.logspace(1,3,3,base=2,dtype=(np.int)) - C = helper.write_c_matrix(n) + n = np.logspace(1,11,11,base=2,dtype=(np.int)) + # n=[8192] + # C = helper.write_c_matrix(n) diff --git a/buch/papers/multiplikation/code/meas/MM.txt b/buch/papers/multiplikation/code/meas/MM.txt index e296dd7..7bffb6e 100644 --- a/buch/papers/multiplikation/code/meas/MM.txt +++ b/buch/papers/multiplikation/code/meas/MM.txt @@ -1,12 +1,110 @@ -0.000001,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 0.000001,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000001,4 +0.000001,4 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000001,8 0.000001,8 -0.000010,16 -0.000081,32 -0.000654,64 -0.005556,128 -0.054253,256 -0.487317,512 -4.162845,1024 -125.909034,2048 -1111.312696,4096 +0.000011,16 +0.000011,16 +0.000011,16 +0.000011,16 +0.000011,16 +0.000021,16 +0.000011,16 +0.000011,16 +0.000011,16 +0.000011,16 +0.000090,32 +0.000093,32 +0.000083,32 +0.000082,32 +0.000090,32 +0.000080,32 +0.000080,32 +0.000080,32 +0.000089,32 +0.000126,32 +0.000771,64 +0.000651,64 +0.000651,64 +0.000651,64 +0.000731,64 +0.000673,64 +0.000745,64 +0.000672,64 +0.000671,64 +0.000707,64 +0.005642,128 +0.005579,128 +0.005768,128 +0.005745,128 +0.005518,128 +0.005877,128 +0.005513,128 +0.005850,128 +0.005769,128 +0.005581,128 +0.052188,256 +0.051988,256 +0.051888,256 +0.051518,256 +0.051709,256 +0.051543,256 +0.051707,256 +0.051845,256 +0.051495,256 +0.051834,256 +0.507020,512 +0.504111,512 +0.502049,512 +0.529743,512 +0.501028,512 +0.502097,512 +0.503490,512 +0.502079,512 +0.506688,512 +0.504163,512 +4.538722,1024 +4.291473,1024 +4.516302,1024 +4.374630,1024 +4.719557,1024 +4.438999,1024 +4.641680,1024 +4.407959,1024 +4.441451,1024 +4.677313,1024 +129.433279,2048 +129.277802,2048 +129.284817,2048 +129.086884,2048 +129.197444,2048 +129.350999,2048 +129.264250,2048 +129.295723,2048 +129.402601,2048 +129.300820,2048 diff --git a/buch/papers/multiplikation/code/meas/MM_dc.txt b/buch/papers/multiplikation/code/meas/MM_dc.txt index f6be928..b78b925 100644 --- a/buch/papers/multiplikation/code/meas/MM_dc.txt +++ b/buch/papers/multiplikation/code/meas/MM_dc.txt @@ -1,12 +1,110 @@ 0.000003,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 0.000002,4 -0.000010,8 -0.000068,16 -0.000594,32 -0.004264,64 -0.036289,128 -0.324645,256 -2.612010,512 -19.928951,1024 -159.333884,2048 -1147.106865,4096 +0.000001,4 +0.000001,4 +0.000001,4 +0.000001,4 +0.000001,4 +0.000001,4 +0.000001,4 +0.000001,4 +0.000001,4 +0.000008,8 +0.000008,8 +0.000008,8 +0.000008,8 +0.000007,8 +0.000007,8 +0.000007,8 +0.000007,8 +0.000018,8 +0.000008,8 +0.000075,16 +0.000063,16 +0.000088,16 +0.000062,16 +0.000086,16 +0.000092,16 +0.000081,16 +0.000080,16 +0.000070,16 +0.000085,16 +0.000581,32 +0.000659,32 +0.000584,32 +0.000714,32 +0.000666,32 +0.000574,32 +0.000616,32 +0.000534,32 +0.000506,32 +0.000506,32 +0.004567,64 +0.004502,64 +0.004332,64 +0.004578,64 +0.004543,64 +0.004426,64 +0.004497,64 +0.004329,64 +0.004288,64 +0.004277,64 +0.036456,128 +0.034901,128 +0.034545,128 +0.034283,128 +0.035150,128 +0.034663,128 +0.034901,128 +0.034022,128 +0.034368,128 +0.035154,128 +0.296292,256 +0.297592,256 +0.302464,256 +0.299557,256 +0.299367,256 +0.306394,256 +0.287616,256 +0.292630,256 +0.289542,256 +0.277019,256 +2.331956,512 +2.224501,512 +2.203910,512 +2.198937,512 +2.206083,512 +2.199477,512 +2.199847,512 +2.225379,512 +2.202491,512 +2.235926,512 +17.649432,1024 +17.636769,1024 +17.639024,1024 +17.625402,1024 +17.722286,1024 +17.611777,1024 +17.653120,1024 +17.748270,1024 +17.691817,1024 +17.614448,1024 +141.943689,2048 +141.580812,2048 +141.882050,2048 +141.516253,2048 +141.351237,2048 +141.641167,2048 +141.596407,2048 +141.607048,2048 +141.469723,2048 +141.515550,2048 diff --git a/buch/papers/multiplikation/code/meas/blas.txt b/buch/papers/multiplikation/code/meas/blas.txt index 92a61b9..9414d8f 100644 --- a/buch/papers/multiplikation/code/meas/blas.txt +++ b/buch/papers/multiplikation/code/meas/blas.txt @@ -1,12 +1,110 @@ 0.000001,2 -0.000001,4 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 0.000001,8 +0.000000,8 +0.000000,8 +0.000000,8 +0.000000,8 +0.000000,8 +0.000000,8 +0.000000,8 +0.000000,8 +0.000000,8 0.000003,16 -0.000022,32 -0.000179,64 +0.000003,16 +0.000003,16 +0.000003,16 +0.000003,16 +0.000003,16 +0.000012,16 +0.000003,16 +0.000003,16 +0.000003,16 +0.000021,32 +0.000019,32 +0.000030,32 +0.000020,32 +0.000020,32 +0.000020,32 +0.000020,32 +0.000020,32 +0.000020,32 +0.000020,32 +0.000180,64 +0.000192,64 +0.000163,64 +0.000153,64 +0.000153,64 +0.000197,64 +0.000163,64 +0.000267,64 +0.000226,64 +0.000164,64 +0.001216,128 +0.001233,128 +0.001364,128 0.001278,128 -0.010165,256 -0.074739,512 -0.704748,1024 -6.845095,2048 -55.845038,4096 +0.001211,128 +0.001295,128 +0.001206,128 +0.001371,128 +0.001225,128 +0.001250,128 +0.009733,256 +0.009497,256 +0.009586,256 +0.009600,256 +0.009768,256 +0.009566,256 +0.009731,256 +0.009550,256 +0.009664,256 +0.009794,256 +0.077453,512 +0.076616,512 +0.088812,512 +0.075990,512 +0.076925,512 +0.076303,512 +0.075915,512 +0.075600,512 +0.075122,512 +0.075029,512 +0.769186,1024 +0.775780,1024 +0.753906,1024 +0.757834,1024 +0.772001,1024 +0.770950,1024 +0.791317,1024 +0.753319,1024 +0.747228,1024 +0.752347,1024 +7.625205,2048 +7.652278,2048 +7.640682,2048 +7.649428,2048 +7.632806,2048 +7.579347,2048 +7.612317,2048 +7.676742,2048 +7.632979,2048 +7.619210,2048 diff --git a/buch/papers/multiplikation/code/meas/ci/MM.txt b/buch/papers/multiplikation/code/meas/ci/MM.txt new file mode 100644 index 0000000..e69de29 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/ci/MM.txt diff --git a/buch/papers/multiplikation/code/meas/ci/Wino.txt b/buch/papers/multiplikation/code/meas/ci/Wino.txt new file mode 100644 index 0000000..e69de29 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/ci/Wino.txt diff --git a/buch/papers/multiplikation/code/meas/ci/blas.txt b/buch/papers/multiplikation/code/meas/ci/blas.txt new file mode 100644 index 0000000..e69de29 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/ci/blas.txt diff --git a/buch/papers/multiplikation/code/meas/ci/dc.txt b/buch/papers/multiplikation/code/meas/ci/dc.txt new file mode 100644 index 0000000..e69de29 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/ci/dc.txt diff --git a/buch/papers/multiplikation/code/meas/ci/strassen.txt b/buch/papers/multiplikation/code/meas/ci/strassen.txt new file mode 100644 index 0000000..e69de29 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/ci/strassen.txt diff --git a/buch/papers/multiplikation/code/meas/old/8196/MM.txt b/buch/papers/multiplikation/code/meas/old/8196/MM.txt new file mode 100644 index 0000000..0edf9f6 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/old/8196/MM.txt @@ -0,0 +1 @@ +9376.173434,8192 diff --git a/buch/papers/multiplikation/code/meas/old/8196/MM_dc.txt b/buch/papers/multiplikation/code/meas/old/8196/MM_dc.txt new file mode 100644 index 0000000..36f6ff0 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/old/8196/MM_dc.txt @@ -0,0 +1 @@ +9606.402522,8192 diff --git a/buch/papers/multiplikation/code/meas/old/8196/blas.txt b/buch/papers/multiplikation/code/meas/old/8196/blas.txt new file mode 100644 index 0000000..b5989fb --- /dev/null +++ b/buch/papers/multiplikation/code/meas/old/8196/blas.txt @@ -0,0 +1 @@ +478.429957,8192 diff --git a/buch/papers/multiplikation/code/meas/old/8196/strassen.txt b/buch/papers/multiplikation/code/meas/old/8196/strassen.txt new file mode 100644 index 0000000..ca06e97 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/old/8196/strassen.txt @@ -0,0 +1 @@ +3014.235467,8192 diff --git a/buch/papers/multiplikation/code/meas/old/8196/winograd.txt b/buch/papers/multiplikation/code/meas/old/8196/winograd.txt new file mode 100644 index 0000000..2a529c4 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/old/8196/winograd.txt @@ -0,0 +1 @@ +10071.512655,8192 diff --git a/buch/papers/multiplikation/code/meas/old/MM.txt b/buch/papers/multiplikation/code/meas/old/MM.txt new file mode 100644 index 0000000..e296dd7 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/old/MM.txt @@ -0,0 +1,12 @@ +0.000001,2 +0.000001,4 +0.000001,8 +0.000010,16 +0.000081,32 +0.000654,64 +0.005556,128 +0.054253,256 +0.487317,512 +4.162845,1024 +125.909034,2048 +1111.312696,4096 diff --git a/buch/papers/multiplikation/code/meas/old/MM_dc.txt b/buch/papers/multiplikation/code/meas/old/MM_dc.txt new file mode 100644 index 0000000..f6be928 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/old/MM_dc.txt @@ -0,0 +1,12 @@ +0.000003,2 +0.000002,4 +0.000010,8 +0.000068,16 +0.000594,32 +0.004264,64 +0.036289,128 +0.324645,256 +2.612010,512 +19.928951,1024 +159.333884,2048 +1147.106865,4096 diff --git a/buch/papers/multiplikation/code/meas/old/blas.txt b/buch/papers/multiplikation/code/meas/old/blas.txt new file mode 100644 index 0000000..92a61b9 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/old/blas.txt @@ -0,0 +1,12 @@ +0.000001,2 +0.000001,4 +0.000001,8 +0.000003,16 +0.000022,32 +0.000179,64 +0.001278,128 +0.010165,256 +0.074739,512 +0.704748,1024 +6.845095,2048 +55.845038,4096 diff --git a/buch/papers/multiplikation/code/meas/old/strassen.txt b/buch/papers/multiplikation/code/meas/old/strassen.txt new file mode 100644 index 0000000..fdfbf2b --- /dev/null +++ b/buch/papers/multiplikation/code/meas/old/strassen.txt @@ -0,0 +1,12 @@ +0.000001,2 +0.000003,4 +0.000010,8 +0.000066,16 +0.000470,32 +0.003368,64 +0.024232,128 +0.172000,256 +1.209262,512 +8.457472,1024 +59.267256,2048 +414.648901,4096 diff --git a/buch/papers/multiplikation/code/meas/old/winograd.txt b/buch/papers/multiplikation/code/meas/old/winograd.txt new file mode 100644 index 0000000..d185906 --- /dev/null +++ b/buch/papers/multiplikation/code/meas/old/winograd.txt @@ -0,0 +1,12 @@ +0.000001,2 +0.000001,4 +0.000002,8 +0.000011,16 +0.000100,32 +0.000654,64 +0.005229,128 +0.057440,256 +0.517850,512 +4.539413,1024 +130.627663,2048 +1179.261048,4096 diff --git a/buch/papers/multiplikation/code/meas/strassen.txt b/buch/papers/multiplikation/code/meas/strassen.txt index fdfbf2b..d6e040e 100644 --- a/buch/papers/multiplikation/code/meas/strassen.txt +++ b/buch/papers/multiplikation/code/meas/strassen.txt @@ -1,12 +1,110 @@ -0.000001,2 -0.000003,4 -0.000010,8 -0.000066,16 -0.000470,32 -0.003368,64 -0.024232,128 -0.172000,256 -1.209262,512 -8.457472,1024 -59.267256,2048 -414.648901,4096 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000004,4 +0.000002,4 +0.000002,4 +0.000002,4 +0.000002,4 +0.000002,4 +0.000002,4 +0.000002,4 +0.000002,4 +0.000001,4 +0.000020,8 +0.000018,8 +0.000008,8 +0.000008,8 +0.000008,8 +0.000008,8 +0.000008,8 +0.000008,8 +0.000008,8 +0.000019,8 +0.000080,16 +0.000075,16 +0.000078,16 +0.000085,16 +0.000065,16 +0.000065,16 +0.000065,16 +0.000064,16 +0.000065,16 +0.000065,16 +0.000546,32 +0.000480,32 +0.000563,32 +0.000551,32 +0.000502,32 +0.000504,32 +0.000463,32 +0.000462,32 +0.000508,32 +0.000462,32 +0.003675,64 +0.003665,64 +0.003493,64 +0.003708,64 +0.003465,64 +0.003502,64 +0.003710,64 +0.003537,64 +0.003637,64 +0.003568,64 +0.025342,128 +0.025179,128 +0.026475,128 +0.025758,128 +0.025333,128 +0.024988,128 +0.025727,128 +0.025298,128 +0.025283,128 +0.025098,128 +0.181311,256 +0.178432,256 +0.177075,256 +0.177474,256 +0.177025,256 +0.177805,256 +0.177944,256 +0.178151,256 +0.177858,256 +0.178742,256 +1.283374,512 +1.246682,512 +1.245898,512 +1.251547,512 +1.250288,512 +1.250495,512 +1.257037,512 +1.255247,512 +1.255382,512 +1.259050,512 +8.784102,1024 +8.845725,1024 +8.771100,1024 +8.770184,1024 +8.955977,1024 +8.849161,1024 +8.806902,1024 +8.808937,1024 +8.848900,1024 +8.861383,1024 +61.787123,2048 +61.972599,2048 +61.822434,2048 +62.051331,2048 +61.946171,2048 +61.911404,2048 +61.872671,2048 +61.791260,2048 +61.818110,2048 +62.045588,2048 diff --git a/buch/papers/multiplikation/code/meas/winograd.txt b/buch/papers/multiplikation/code/meas/winograd.txt index d185906..970a3f4 100644 --- a/buch/papers/multiplikation/code/meas/winograd.txt +++ b/buch/papers/multiplikation/code/meas/winograd.txt @@ -1,12 +1,110 @@ 0.000001,2 -0.000001,4 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,2 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 +0.000000,4 0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000002,8 +0.000011,16 +0.000011,16 +0.000011,16 +0.000011,16 +0.000011,16 +0.000011,16 +0.000011,16 +0.000021,16 +0.000011,16 0.000011,16 -0.000100,32 -0.000654,64 -0.005229,128 -0.057440,256 -0.517850,512 -4.539413,1024 -130.627663,2048 -1179.261048,4096 +0.000092,32 +0.000092,32 +0.000081,32 +0.000081,32 +0.000081,32 +0.000081,32 +0.000088,32 +0.000079,32 +0.000079,32 +0.000079,32 +0.000670,64 +0.000739,64 +0.000609,64 +0.000609,64 +0.000700,64 +0.000648,64 +0.000626,64 +0.000626,64 +0.000626,64 +0.000626,64 +0.005321,128 +0.005286,128 +0.005180,128 +0.005223,128 +0.005249,128 +0.005299,128 +0.005205,128 +0.005268,128 +0.005464,128 +0.005378,128 +0.053123,256 +0.052325,256 +0.052729,256 +0.052930,256 +0.052207,256 +0.053178,256 +0.052122,256 +0.052681,256 +0.052965,256 +0.052486,256 +0.527028,512 +0.525201,512 +0.521822,512 +0.525147,512 +0.525241,512 +0.527725,512 +0.526321,512 +0.526479,512 +0.524020,512 +0.520768,512 +4.732299,1024 +4.617253,1024 +4.647425,1024 +4.519233,1024 +4.917471,1024 +4.564929,1024 +4.870771,1024 +4.555407,1024 +4.727473,1024 +4.559349,1024 +136.409028,2048 +136.390557,2048 +136.541672,2048 +136.598491,2048 +137.720790,2048 +136.825926,2048 +136.367686,2048 +136.650627,2048 +136.642195,2048 +136.622805,2048 diff --git a/buch/papers/multiplikation/code/meas_4096.pdf b/buch/papers/multiplikation/code/meas_4096.pdf Binary files differindex e889d17..ecf2cff 100644 --- a/buch/papers/multiplikation/code/meas_4096.pdf +++ b/buch/papers/multiplikation/code/meas_4096.pdf diff --git a/buch/papers/multiplikation/code/meas_4096.txt b/buch/papers/multiplikation/code/meas_4096.txt index e69de29..cae1bc6 100644 --- a/buch/papers/multiplikation/code/meas_4096.txt +++ b/buch/papers/multiplikation/code/meas_4096.txt @@ -0,0 +1,6 @@ +2.048000000000000000e+03 4.096000000000000000e+03 +6.154183513402938843e+03 4.681333474493026733e+04 +7.375929301261901855e+03 5.846600176072120667e+04 +3.860573610544204712e+03 2.290433094644546509e+04 +4.884613198995590210e+03 4.359707747149467468e+04 +2.157390117645263672e-01 1.491588830947875977e+00 diff --git a/buch/papers/multiplikation/images/algo_tab.pdf b/buch/papers/multiplikation/images/algo_tab.pdf Binary files differnew file mode 100644 index 0000000..7f2bb4f --- /dev/null +++ b/buch/papers/multiplikation/images/algo_tab.pdf diff --git a/buch/papers/multiplikation/images/algo_tab.tex b/buch/papers/multiplikation/images/algo_tab.tex new file mode 100644 index 0000000..50ce392 --- /dev/null +++ b/buch/papers/multiplikation/images/algo_tab.tex @@ -0,0 +1,122 @@ +\documentclass{article} +\usepackage[left=25mm,right=25mm,top=25mm,bottom=25mm]{geometry} +\usepackage[utf8]{inputenc} +\usepackage[T1]{fontenc} +\usepackage{times} +\usepackage{geometry} +\usepackage{amsmath} +\usepackage{amssymb} +\usepackage{algorithm} +\usepackage{algpseudocode} +\usepackage{mathrsfs} +\usepackage{amsfonts} +\usepackage{amsthm} +\usepackage{lipsum} +\usepackage{amscd} +\usepackage{graphicx} +\usepackage{fancyhdr} +\usepackage{textcomp} +\usepackage{pgfplots} +\usepackage{txfonts} +\usepackage[all]{xy} +\usepackage{paralist} +\usepackage[colorlinks=true]{hyperref} +\usepackage{array} +\usepackage{tikz} +\usepackage{slashed} +\usepackage{pdfpages} +\usepackage{multicol} +\usepackage{cite} +\usepackage{url} +\usepackage{amsmath,amsfonts,amssymb} +\usepackage{tikz} +\usetikzlibrary{arrows,matrix,positioning} +\usetikzlibrary{overlay-beamer-styles} +\usetikzlibrary{matrix.skeleton} +\usetikzlibrary{automata,positioning} +\usetikzlibrary{decorations.text} +\usepackage{listings} +\usepackage{multirow} +\usepackage{color} + +\begin{document} + + + +\begin{table}[t] + \begin{tabular}{ll} + \begin{minipage}{0.4\textwidth} + \begin{algorithm}[H]\footnotesize\caption{} + \label{multiplikation:alg:b1} + \setlength{\lineskip}{7pt} + \begin{algorithmic} + \Function{B1}{$a, b$} + \State \textbf{return} $a+b$ + \EndFunction + \State + \State + \end{algorithmic} + \end{algorithm} + \end{minipage} + & + \begin{minipage}{0.4\textwidth} + \begin{algorithm}[H]\footnotesize\caption{} + \label{multiplikation:alg:b2} + \setlength{\lineskip}{7pt} + \begin{algorithmic} + \Function{B2}{$a, b$} + \State $ x \gets a+b $ + \State $ y \gets a \cdot b $ + \State \textbf{return} $x+y$ + \EndFunction + \end{algorithmic} +\end{algorithm} + + \end{minipage} + \end{tabular} +\end{table} + +\begin{table} + \begin{tabular}[t]{ll} + \begin{minipage}{0.4\textwidth} + \begin{algorithm}[H]\footnotesize\caption{} + \setlength{\lineskip}{7pt} + \begin{algorithmic} + \label{multiplikation:alg:linear} + \Function{L}{$\mathbf{a}, \mathbf{b}$,n} + \State $ sum \gets 0$ + \For{$i = 0,1,2 \dots,n$} + \State $ sum \gets sum + A[i] \cdot B[i] $ + \EndFor + + \State \textbf{return} $sum$ + + \EndFunction + \State + \State + \end{algorithmic} + \end{algorithm} + \end{minipage} + & + \begin{minipage}{0.4\textwidth} + \begin{algorithm}[H]\footnotesize\caption{} + \label{multiplikation:alg:q1} + \setlength{\lineskip}{7pt} + \begin{algorithmic} + \Function{Q}{$\mathbf{A}, \mathbf{B}$,n} + \State $ sum \gets 0$ + \For{$i = 0,1,2 \dots,n$} + \For{$j = 0,1,2 \dots,n$} + \State $ sum \gets sum + A[i] \cdot B[j] $ + \EndFor + \EndFor + \State \textbf{return} $sum$ + \EndFunction + \end{algorithmic} + \end{algorithm} + \end{minipage} + \end{tabular} +\end{table} + +dhdfh +\end{document} diff --git a/buch/papers/multiplikation/images/meas_c.pdf b/buch/papers/multiplikation/images/meas_c.pdf Binary files differindex 3a4cfd8..faf347e 100644 --- a/buch/papers/multiplikation/images/meas_c.pdf +++ b/buch/papers/multiplikation/images/meas_c.pdf diff --git a/buch/papers/multiplikation/images/meas_c.tex b/buch/papers/multiplikation/images/meas_c.tex index 818a7e6..fe2bd2f 100644 --- a/buch/papers/multiplikation/images/meas_c.tex +++ b/buch/papers/multiplikation/images/meas_c.tex @@ -43,8 +43,8 @@ \begin{tikzpicture} \begin{axis}[ xmode=log, ymode=log, -xmin=60, xmax=5000, -ymin=1e-4, ymax=2e3, +xmin=30, xmax=10000, +ymin=1e-5, ymax=2e4, grid=both, major grid style={black!50}, xlabel = data Input ($n$), @@ -57,85 +57,91 @@ width=12cm, height=8cm, ] \addlegendentry{Winograd} \addplot[ color=purple, + error bars/.cd, y dir=both, y explicit, ] coordinates { -% (2, 0.000001) -% (4, 0.000001) -% (8, 0.000002) -% (16, 0.000011) -% (32, 0.000100) -(64, 0.000654) -(128, 0.005229) -(256, 0.057440) -(512, 0.517850) -(1024,4.539413) -(2048,130.627663) +%(2,1e-07) +%(4,5e-07) +%(8,2.0000000000000003e-06) +%(16,1.1999999999999999e-05) +(32,8.329999999999999e-05) +(64,0.0006479) +(128,0.0052873) +(256,0.052674599999999995) +(512,0.5249752000000001) +(1024,4.671161) +(2048,136.6769777) (4096,1179.261048) +(8192,10071.512655) }; \addlegendentry{Strassen} \addplot [ color=black, ]coordinates { - % (2,0.000001 ) - % (4,0.000003 ) - % (8,0.000010 ) - % (16,0.000066 ) - % (32,0.000470 ) - (64,0.003368 ) - (128,0.024232 ) - (256,0.172000 ) - (512,1.209262 ) -(1024,8.457472 ) -(2048,59.267256) +%(2,1e-07) +%(4,2.1e-06) +%(8,1.13e-05) +%(16,7.07e-05) +(32,0.0005041) +(64,0.003596) +(128,0.0254481) +(256,0.1781817) +(512,1.2555) +(1024,8.8302371) +(2048,61.9018691) (4096,414.648901) +(8192,3014.235467) }; \addlegendentry{MM div and conq} \addplot[ color=green, ] coordinates { - % (2,0.000003 ) - % (4,0.000002 ) - % (8,0.000010 ) - % (16,0.000068 ) - % (32,0.000594 ) - (64,0.004264 ) - (128,0.036289 ) - (256,0.324645 ) - (512,2.612010 ) -(1024,19.928951 ) -(2048,159.333884 ) +%(2,3e-07) +%(4,1.1e-06) +%(8,8.6e-06) +%(16,7.819999999999999e-05) +(32,0.0005940000000000001) +(64,0.0044339) +(128,0.0348443) +(256,0.29484730000000003) +(512,2.2228507) +(1024,17.659234500000004) +(2048,141.6103936) (4096,1147.106865) +(8192,9606.402522) }; \addlegendentry{MM} \addplot [ color=red, ]coordinates { - % (2,0.000001 ) - % (4,0.000001 ) - % (8,0.000001 ) - % (16,0.000010 ) - % (32,0.000081 ) - (64,0.000654 ) - (128,0.005556 ) - (256,0.054253 ) - (512,0.487317 ) -(1024,4.162845 ) -(2048,125.909034 ) +%(2,0.0) +%(4,3e-07) +%(8,1.8000000000000001e-06) +%(16,1.1999999999999999e-05) +(32,8.93e-05) +(64,0.0006923) +(128,0.0056842) +(256,0.051771500000000005) +(512,0.5062468000000001) +(1024,4.5048086) +(2048,129.2894619) (4096,1111.312696) +(8192,9376.173434) }; \addlegendentry{BLAS} \addplot[ color=blue, ] coordinates { - % (2,0.000001 ) - % (4,0.000001 ) - % (8,0.000001 ) - % (16,0.000003 ) - % (32,0.000022 ) - (64,0.000179 ) - (128,0.001278 ) - (256,0.010165 ) - (512,0.074739 ) -(1024,0.704748 ) -(2048,6.845095 ) +%(2,1e-07) +%(4,0.0) +%(8,1e-07) +%(16,3.9e-06) +(32,2.1000000000000002e-05) +(64,0.00018580000000000002) +(128,0.0012649) +(256,0.0096489) +(512,0.0773765) +(1024,0.7643868) +(2048,7.6320993999999995) (4096,55.845038) +(8192,478.429957) }; \end{axis} \end{tikzpicture} diff --git a/buch/papers/multiplikation/images/meas_python.pdf b/buch/papers/multiplikation/images/meas_python.pdf Binary files differindex cea2232..ab3b14b 100644 --- a/buch/papers/multiplikation/images/meas_python.pdf +++ b/buch/papers/multiplikation/images/meas_python.pdf diff --git a/buch/papers/multiplikation/images/meas_python.tex b/buch/papers/multiplikation/images/meas_python.tex index ee4db43..d942f46 100644 --- a/buch/papers/multiplikation/images/meas_python.tex +++ b/buch/papers/multiplikation/images/meas_python.tex @@ -43,8 +43,8 @@ \begin{tikzpicture} \begin{axis}[ xmode=log, ymode=log, -xmin=30, xmax=1050, -ymin=0.01, ymax=900, +xmin=30, xmax=4200, +ymin=0.01, ymax=70000, grid=both, major grid style={black!50}, xlabel = data input ($n$), @@ -68,7 +68,8 @@ width=12cm, height=8cm, (256, 8.29899 ) (512, 68.3699 ) (1024,537.374 ) - +(2046,4884.61) +(4096,43597.1) }; \addlegendentry{Strassen} \addplot [ color=black, @@ -79,10 +80,12 @@ width=12cm, height=8cm, % (16,0.00475407 ) (32,0.0485256 ) (64,0.220414 ) - (128,1.44718 2 ) - (256,9.93866 0 ) - (512,63.961 2 ) -(1024,461.494 2 ) + (128,1.44718 ) + (256,9.93866 ) + (512,63.961 ) +(1024,461.494 ) +(2046,3860.57) +(4096,22904.3) }; \addlegendentry{MM div and conq} @@ -98,6 +101,8 @@ width=12cm, height=8cm, (256,13.27 ) (512,105.397 ) (1024,847.321 ) +(2046,7375.93) +(4096,58466) }; \addlegendentry{MM} @@ -113,25 +118,27 @@ width=12cm, height=8cm, (256, 11.0062 ) (512, 85.4768) (1024,750.757 ) +(2046,6154.18) +(4096,46813.3) }; -% \addlegendentry{NumPy} -% \addplot[ color=blue, -% ] coordinates { -% (2,1.83582e-05 ) -% (4,7.86781e-06) -% (8,1.00136e-05) -% (16,5.4121e-05 ) -% (32,4.26769e-05) -% (64,0.000118494) -% (128,0.000244141 ) -% (256,0.000695705 ) -% (512,0.00221705 ) -% (1024,0.0188088 ) -% }; +% \addlegendentry{NumPy} +% \addplot[ color=blue, +% ] coordinates { +% % (2,1.83582e-05 ) +% % (4,7.86781e-06) +% % (8,1.00136e-05) +% % (16,5.4121e-05 ) +% (32,4.26769e-05) +% (64,0.000118494) +% (128,0.000244141 ) +% (256,0.000695705 ) +% (512,0.00221705 ) +% (1024,0.0188088 ) +% (2046,0.215739) +% (4096,1.49159) +% }; + \end{axis} \end{tikzpicture} \end{document} - - - diff --git a/buch/papers/multiplikation/loesungsmethoden.tex b/buch/papers/multiplikation/loesungsmethoden.tex index a7612e1..0760719 100755 --- a/buch/papers/multiplikation/loesungsmethoden.tex +++ b/buch/papers/multiplikation/loesungsmethoden.tex @@ -39,13 +39,13 @@ Die \texttt{for i} Schleife iteriert \"uber alle Zeilen der $\mathbf{A}$ Matrix, \end{algorithmic} \end{algorithm} -Die Laufzeit dieser Struktur mit drei \texttt{For} Schleifen ist $\mathcal{O}\left(n^3\right)$ +Die Laufzeit dieser Struktur mit drei \texttt{For} Schleifen ist $\mathcal{O} (n^3)$ \subsubsection{Divide and Conquer Methode} F\"ur gewisse Algorithmen f\"uhren \textit{Divide and Conquer} Ans\"atze \cite{multiplikation:DAC} zu markant besseren Laufzeiten. Die Grundidee ist, dass ein Problem in mehrere, meist simplere und kleinere Teilprobleme aufgeteilt wird. -Das bekannteste Beispiel ist wohl die \textit{Fast Fourier Transform} wobei die Laufzeit von $\mathcal{O}\left(n^2\right)$ zu $\mathcal{O}(n \log n)$ verbessert werden kann. +Das bekannteste Beispiel ist wohl die \textit{Fast Fourier Transform} wobei die Laufzeit von $\mathcal{O} (n^2)$ zu $\mathcal{O}(n \log n)$ verbessert werden kann. Die Matrizenmultiplikation kann ebenfalls mit solch einem Ansatz berechnet werden. Zur vereinfachten Veranschaulichung kann die Situation mit $\mathbf{A}$ und $\mathbf{B}$ der Gr\"osse $2^n \times 2^n$ verwendet werden. @@ -68,7 +68,7 @@ Das Matrizen Produkt \end{bmatrix}, \end{equation} \begin{equation} -\mathbf{C}_{ij} = \sum_{k=1}2n \mathbf{A}_{ik} \mathbf{B}_{kj} +\mathbf{C}_{ij} = \sum_{k=1}^{2n} \mathbf{A}_{ik} \mathbf{B}_{kj} \label{multiplikation:eq:MM_block} \end{equation} ist identisch zu der Gleichung \eqref{multiplikation:eq:MM}, f\"ur die Multiplikation der Untermatrize $\mathbf{A}_{ik}$ und $\mathbf{B}_{kj}$ wird die Matrizenmultiplikation verwendet. @@ -109,7 +109,7 @@ Die Laufzeit dieser rekursiven Funktion kann mit dem \textit{Master Theorem} \ci Ohne auf dieses vertieft einzugehen, bestimmt die Anzahl rekursiver Aufrufe $\mathcal{T} $ der Funktion die Laufzeit. In diesem Fall wird die Funktion pro Durchlauf acht mal rekursiv aufgerufen, dies f\"uhrt \begin{equation} \label{multiplikation:eq:laufzeitdac} - \mathcal{T}(n) = 8 \cdot \mathcal{T}\left (\frac{n}{2}\right ) + n^2 = \mathcal{O}(n^{\log_2 8}) = \mathcal{O}\left (n^{3} \right ) + \mathcal{T}(n) = 8 \cdot \mathcal{T} \left(\frac{n}{2}\right ) + n^2 = \mathcal{O}(n^{\log_2 8}) = \mathcal{O} (n^{3} ) \end{equation} zu einer kubischen Laufzeit. Die Addition zweier Matrizen $\mathbf{A} + \mathbf{B} = \mathbf{C}$ hat eine Laufzeit von $\mathcal{O}(n^{2})$ und kann neben dem dominierendem Anteil von $\mathcal{O}(n^{3})$ ignoriert werden. @@ -202,7 +202,7 @@ Die Funktion wird sieben mal rekursiv aufgerufen. Dies f\"uhrt nach dem \textit{Master Theorem} zu einer Laufzeit von \begin{equation} \label{multiplikation:eq:laufzeitstrassen} \mathcal{T}(n) = -7 \cdot \mathcal{T}(\frac{n}{2}) + n^2 = \mathcal{O}\left(n^{\log_2 7}\right ) = \mathcal{O}\left(n^{2.8074} \right ) +7 \cdot \mathcal{T}\left(\frac{n}{2}\right) + n^2 = \mathcal{O}(n^{\log_2 7} ) = \mathcal{O}(n^{2.8074} ) \end{equation} und ist somit schneller als die Standardmethode. Man beachte, dass die Anzahl von Additionen und Subtraktionen gr\"osser und die Anzahl der Multiplikationen kleiner wurde. @@ -265,9 +265,9 @@ N=2n, \quad T = n^2 \\ \end{equation} sein, damit man etwas einspart. Die Implementation kann Algorithmus \ref{multiplikation:alg:winograd} entnommen werden. -Falls $m=n=p$ werden $\frac{n^3}/{2}$ Multiplikationen benötigt. +Falls $m=n=p$ werden $\frac{n^3}{2}$ Multiplikationen benötigt. Im Abschnitt \ref{muliplikation:sec:bigo} wurde bereits erläutert: falls $n \rightarrow \infty$ können Konstanten vernachlässigt werden und - somit entsteht für diesen Algorithmus wieder die Ursprüngliche Laufzeit von $\mathcal{O}\left(n^3 \right)$. + somit entsteht für diesen Algorithmus wieder die Ursprüngliche Laufzeit von $\mathcal{O}(n^3 )$. \begin{algorithm}\footnotesize\caption{Winograds Matrizenmultiplikation} \setlength{\lineskip}{7pt} \label{multiplikation:alg:winograd} @@ -336,33 +336,33 @@ Die meisten Numerischen Bibliotheken von High-Level Skriptsprachen wie \texttt{M \item Level 2 \begin{itemize} \item Operationen der Art: $\mathbf{y} \leftarrow \alpha \mathbf{A}\mathbf{x}+\beta \mathbf{y}$ - \item Dieses Level hat $\mathcal{O}\left(n^2\right)$ Charakteristik + \item Dieses Level hat $\mathcal{O}(n^2)$ Charakteristik \end{itemize} \item Level 3 \begin{itemize} \item Operationen der Art: $\mathbf{C} \leftarrow \alpha \mathbf{A}\mathbf{B}+\beta\mathbf{C}$ - \item Dieses Level hat $\mathcal{O}\left(n^3\right)$ Charakteristik + \item Dieses Level hat $\mathcal{O}(n^3)$ Charakteristik \end{itemize} \end{itemize} Die \textit{BLAS} sind auf die modernen Computer Prozessoren optimiert und k\"onnen dank einer ausgeklügelter Verwendung der Speicherarchitektur zu erheblichen Leistungsoptimierungen f\"uhren. -\subsubsection{General Matrix Multiplication (GEMM)} - -Die \textit{Double-GEMM} \cite{multiplikation:DGEMM} ist definiert als: - -\textit{DGEMM performs one of the matrix-matrix operations} -$$ - C := \alpha \cdot op( A )\cdot op( B ) + \beta \cdot C, - $$ - \textit{where op( X ) is one of} -$$ -op( X ) = X \quad \text{ or } \quad op( X ) = X^T, -$$ - \textit{alpha and beta are scalars, and A, B and C are matrices, with op( A ) - an m by k matrix, op( B ) a k by n matrix and C an m by n matrix. - } +%\subsubsection{General Matrix Multiplication (GEMM)} +% +%Die \textit{Double-GEMM} \cite{multiplikation:DGEMM} ist definiert als: +% +%\textit{DGEMM performs one of the matrix-matrix operations} +%$$ +% C := \alpha \cdot op( A )\cdot op( B ) + \beta \cdot C, +% $$ +% \textit{where op( X ) is one of} +%$$ +%op( X ) = X \quad \text{ or } \quad op( X ) = X^T, +%$$ +% \textit{alpha and beta are scalars, and A, B and C are matrices, with op( A ) +% an m by k matrix, op( B ) a k by n matrix and C an m by n matrix. +% } %Die Implementation von $\alpha\mathbf{A}\mathbf{B} + \beta \mathbf{C} = \mathbf{C}$, wobei $\alpha = 1.0$ und $\beta = 0.0$ in der \texttt{C}-Version von \textit{BLAS}, ist als %\begin{lstlisting}[style=multiplikationC] @@ -379,7 +379,7 @@ $$ Folgende Algorithmen wurden jeweils in \texttt{C} und \texttt{Python} implementiert. \begin{itemize} \item Standard Matrizenmultiplikation - \item \textit{Devide and Conquer} Matrizenmultiplikation + \item \textit{Divide and Conquer} Matrizenmultiplikation \item Strassens Matrizenmultiplikation \item Winograds Matrizenmultiplikation \item \texttt{BLAS} Matrizenmultiplikation in \texttt{C} @@ -389,6 +389,14 @@ Folgende Algorithmen wurden jeweils in \texttt{C} und \texttt{Python} implementi Der Code kann im zum Buch gehörigem \textit{GitHub} \footnote{\url{https://github.com/AndreasFMueller/SeminarMatrizen.git}} Repository gefunden werden. Anzumerken ist, dass die Matrizenmultiplikation von \texttt{NumPy} als einzige Implementation Multiprocessing und Multithreading verwendet, dies f\"uhrt zu den tiefen Messzeiten. In Abbildung \ref{multiplikation:fig:python} und Abbildung \ref{multiplikation:fig:c_meas_4096} sind de Messresultate grafisch dargestellt. Die selben Messresultate sind tabellarisch in Tabelle \ref{multiplikation:tab:messung_Python} und Tabelle \ref{multiplikation:tab:messung_C} ersichtlich. + +In der Messung mit der Programmiersprache \texttt{C}, kann ein typischer Cache-Effekt beobachtet wer- +den. +Bei den Algorithmen von Winograd und der Standardmethode hat bei einer Matrizengrösse von $n = 2048$ wohl eine Zeile der Matrize nicht an einer Cache Speicherstelle platzt. +Diese beiden Algorithmen sind die Einzigen, welche \texttt{for}-Schleifen über die ganze Breite der Matrizen verwenden. +Dies führt dazu, dass ganze Zeilen zwischengespeichert werden müssen. +Bei den anderen Algorithmen ist dies nicht der Fall. + Die Hardwareinformationen des verwendeten Computers sind in der Tabelle \ref{multiplikation:tab:pc_config} aufgelistet. @@ -400,14 +408,15 @@ Die Hardwareinformationen des verwendeten Computers sind in der Tabelle \ref{mul \textbf{n} & \textbf{MM (\textit{s})} & \textbf{MM DC (\textit{s})} & \textbf{Strassen (\textit{s})} & \textbf{Winograd (\textit{s})} & \textbf{BLAS (\textit{s})} \\ \hline \multicolumn{6}{c}{} \\ - \textbf{32} & 0.000081 &0.000594 & 0.00047& 0.00010 & 0.000022 \\ - \textbf{64} & 0.00065 & 0.0042& 0.0033& 0.00065& 0.00017 \\ - \textbf{128} & 0.0055 & 0.036& 0.024& 0.0052 & 0.0012 \\ - \textbf{256} & 0.054 & 0.32 & 0.17 & 0.057& 0.010 \\ - \textbf{512} & 0.48 & 2.61 & 1.20 & 0.51 & 0.074\\ - \textbf{1024} & 4.16 & 19.92& 8.45 & 4.53 & 0.704 \\ - \textbf{2048} & 125.90 & 159.33& 59.26 & 130.62 & 6.84 \\ - \textbf{4096} & 1111.31 & 1147.10& 414.64 & 1179.26 & 55.84\\ + \textbf{32} & 0.000089 & 0.000594 & 0.0005 & 0.00008 & 0.000021 \\ + \textbf{64} & 0.00069 & 0.0044 & 0.0036 & 0.00064 & 0.00018 \\ + \textbf{128} & 0.0057 & 0.035 & 0.025 & 0.0052 & 0.0012 \\ + \textbf{256} & 0.052 & 0.29 & 0.178 & 0.053 & 0.0096 \\ + \textbf{512} & 0.51 & 2.22 & 1.25 & 0.55 & 0.077 \\ + \textbf{1024} & 4.50 & 17.65 & 8.83 & 4.67 & 0.764 \\ + \textbf{2048} & 129.28 & 141.61 & 61.901 & 136.67 & 7.63 \\ + \textbf{4096} & 1111.31 & 1147.10 & 414.64 & 1179.26 & 55.84 \\ + \textbf{8192} & 9376.17 & 9606.40 & 3014.23 & 10071.51& 478.42 \\ \multicolumn{6}{c}{} \\ \hline \hline @@ -427,13 +436,14 @@ Die Hardwareinformationen des verwendeten Computers sind in der Tabelle \ref{mul \textbf{n} & \textbf{MM (\textit{s})} & \textbf{MM DC (\textit{s})} & \textbf{Strassen (\textit{s})} & \textbf{Winograd (\textit{s})} & \textbf{\texttt{NumPy}(\textit{s})} \\ \hline \multicolumn{6}{c}{} \\ - \textbf{32} & 0.0240 &0.0271 & 0.04852& 0.01871 & 4.26e-05 \\ + \textbf{32} & 0.0240 &0.0271 & 0.04852& 0.01871 & 0.0000426 \\ \textbf{64} & 0.186 & 0.265& 0.2204& 0.1530& 0.000118 \\ \textbf{128} & 1.563 & 1.777& 1.447& 1.1947 & 0.000244 \\ \textbf{256} & 11.006 & 13.27 & 9.938 & 8.298& 0.000695 \\ \textbf{512} & 85.476 & 105.397 & 63.961 & 68.36 & 0.00221\\ \textbf{1024} & 750.757 & 847.321& 461.494 & 537.374 & 0.0188 \\ - \textbf{4096} & - & - & - & - & 1.633 \\ + \textbf{2048} & 6154.18 & 7375.93& 3860.57 & 4884.61 & 0.215 \\ + \textbf{4096} & 46813.3 & 58466 & 22904.3 & 43597.1 & 1.49 \\ \multicolumn{6}{c}{} \\ \hline \hline diff --git a/buch/papers/multiplikation/problemstellung.tex b/buch/papers/multiplikation/problemstellung.tex index e53b0de..c8ba274 100755 --- a/buch/papers/multiplikation/problemstellung.tex +++ b/buch/papers/multiplikation/problemstellung.tex @@ -14,87 +14,102 @@ Gezielt wird auf Algorithmen eingegangen, welche das Problem schneller als der S Die Big $\mathcal{O}$ Notation beschreibt die Laufzeitkomplexit\"at eines Algorithmus in Abhängigkeit zur Inputgrösse \cite{multiplikation:bigo}. $f(x) \in \mathcal{O}(g(x))$ besagt, dass die Funktion $f$ nicht wesentlich schneller w\"achst als $g$ wenn $x \rightarrow \infty$. % Es gibt eine Konstante $K$ derart, dass $f(x) \le K g(x)$ für $x\to\infty$ -Als Beispiel: benötigt eine Funktion $g$ $\mathcal{O}\left(n^2 \right)$ Multiplikationen, so wächst $f$ mit $\mathcal{O}\left(n+ n^2 \right)$ nicht wesentlich schneller falls $x\to\infty$. +Als Beispiel: benötigt eine Funktion $g$ $\mathcal{O} (n^2 )$ Multiplikationen, so wächst $f$ mit $\mathcal{O} (n+ n^2 )$ nicht wesentlich schneller falls $x\to\infty$. Vereinfacht werden f\"ur Algorithmen die folgende Notation verwendet: \begin{itemize} \item $f \in \mathcal{O}(1) \rightarrow f$ ist beschr\"ankt \item $f \in \mathcal{O}(n) \rightarrow f$ w\"achst linear - \item $f \in \mathcal{O}\left (n^2 \right ) \rightarrow f$ w\"achst quadratisch + \item $f \in \mathcal{O} (n^2 ) \rightarrow f$ w\"achst quadratisch \item $f \in \mathcal{O}(\log n) \rightarrow f$ w\"achst logarithmisch \item $f \in \mathcal{O}(n \log n) \rightarrow f$ hat super-lineares Wachstum - \item $f \in \mathcal{O}\left (e^n \right ) \rightarrow f$ w\"achst exponentiell + \item $f \in \mathcal{O} (e^n ) \rightarrow f$ w\"achst exponentiell \item usw. \end{itemize} In der Abbildung \ref{multiplikation:fig:bigo} k\"onnen die verschiedenen Laufzeiten miteinander verglichen werden. Bei einer logarithmischen Darstellung werden Polynome der Form $f(x) = x^k$ als Gerade und Exponentialfunktionen der Form $f(x) = a^x$ als nach oben gekr\"ummte Kurven dargestellt. -Sch\"on zu erkennen ist, dass Logarithmische Kurven beschr\"ankt sind. + \subsubsection{Beispiel Algorithmen} Es folgen einige Beispiele von Algorithmen welche zu einer bestimmten Zeitkomplexit\"atsklasse zugeteilt werden k\"onnen. -\begin{minipage}{0.4\textwidth} - \begin{algorithm}[H]\footnotesize\caption{} - \label{multiplikation:alg:b1} - \setlength{\lineskip}{7pt} - \begin{algorithmic} - \Function{B1}{$a, b$} - \State \textbf{return} $a+b$ - \EndFunction - \end{algorithmic} - \end{algorithm} - - \begin{algorithm}[H]\footnotesize\caption{} - \setlength{\lineskip}{7pt} - \begin{algorithmic} - \label{multiplikation:alg:linear} - \Function{L}{$\mathbf{a}, \mathbf{b}$,n} - \State $ sum \gets 0$ - \For{$i = 0,1,2 \dots,n$} - \State $ sum \gets sum + A[i] \cdot B[i] $ - \EndFor - - \State \textbf{return} $sum$ - - \EndFunction - \end{algorithmic} - \end{algorithm} -\end{minipage} -\hspace{2cm} -\begin{minipage}{0.4\textwidth} - - \begin{algorithm}[H]\footnotesize\caption{} - \label{multiplikation:alg:b2} - \setlength{\lineskip}{7pt} - \begin{algorithmic} - \Function{B2}{$a, b$} - \State $ x \gets a+b $ - \State $ y \gets a \cdot b $ - \State \textbf{return} $x+y$ - \EndFunction - \end{algorithmic} - \end{algorithm} - - - \begin{algorithm}[H]\footnotesize\caption{} - \label{multiplikation:alg:q1} - \setlength{\lineskip}{7pt} - \begin{algorithmic} - \Function{Q}{$\mathbf{A}, \mathbf{B}$,n} - \State $ sum \gets 0$ - \For{$i = 0,1,2 \dots,n$} - \For{$j = 0,1,2 \dots,n$} - \State $ sum \gets sum + A[i] \cdot B[j] $ - \EndFor - \EndFor - \State \textbf{return} $sum$ - \EndFunction - \end{algorithmic} - \end{algorithm} - -\end{minipage} + +\begin{table}[t] + \begin{tabular}{ll} + \begin{minipage}{0.48\textwidth} + \begin{algorithm}[H]\footnotesize\caption{} + \label{multiplikation:alg:b1} + \setlength{\lineskip}{7pt} + \begin{algorithmic} + \Function{B1}{$a, b$} + \State \textbf{return} $a+b$ + \EndFunction + \State + \State + \end{algorithmic} + \end{algorithm} + \end{minipage} + & + \begin{minipage}{0.48\textwidth} + \begin{algorithm}[H]\footnotesize\caption{} + \label{multiplikation:alg:b2} + \setlength{\lineskip}{7pt} + \begin{algorithmic} + \Function{B2}{$a, b$} + \State $ x \gets a+b $ + \State $ y \gets a \cdot b $ + \State \textbf{return} $x+y$ + \EndFunction + \end{algorithmic} + \end{algorithm} + + \end{minipage} + \end{tabular} +\end{table} + +\begin{table} + \begin{tabular}[t]{ll} + \begin{minipage}{0.48\textwidth} + \begin{algorithm}[H]\footnotesize\caption{} + \setlength{\lineskip}{7pt} + \begin{algorithmic} + \label{multiplikation:alg:linear} + \Function{L}{$\mathbf{a}, \mathbf{b}$,n} + \State $ sum \gets 0$ + \For{$i = 0,1,2 \dots,n$} + \State $ sum \gets sum + A[i] \cdot B[i] $ + \EndFor + + \State \textbf{return} $sum$ + + \EndFunction + \State + \State + \end{algorithmic} + \end{algorithm} + \end{minipage} + & + \begin{minipage}{0.48\textwidth} + \begin{algorithm}[H]\footnotesize\caption{} + \label{multiplikation:alg:q1} + \setlength{\lineskip}{7pt} + \begin{algorithmic} + \Function{Q}{$\mathbf{A}, \mathbf{B}$,n} + \State $ sum \gets 0$ + \For{$i = 0,1,2 \dots,n$} + \For{$j = 0,1,2 \dots,n$} + \State $ sum \gets sum + A[i] \cdot B[j] $ + \EndFor + \EndFor + \State \textbf{return} $sum$ + \EndFunction + \end{algorithmic} + \end{algorithm} + \end{minipage} + \end{tabular} +\end{table} \paragraph{Beschr\"ankter Algorithmus} @@ -111,7 +126,7 @@ Die \texttt{for}-Schleife wird $n$-mal durchlaufen und f\"uhrt deshalb zu $\math \paragraph{Quadratischer Algorithmus} Der Algorithmus \ref{multiplikation:alg:q1} hat ein quadratisches Verhalten. -Die beiden \texttt{for}-Schleifen werden jeweils $n$-mal durchlaufen und f\"uhrt deshalb zu $\mathcal{O}\left(n^2\right)$. +Die beiden \texttt{for}-Schleifen werden jeweils $n$-mal durchlaufen und f\"uhrt deshalb zu $\mathcal{O} (n^2 )$. \begin{figure} |