1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
|
function gbdemo2 (bnz)
%GBDEMO2 Extreme performance differences: GraphBLAS vs built-in methods
%
% Usage:
%
% gbdemo2 % uses a default bnz = 6000
% gbdemo2 (20000) % uses bnz = 20000
%
% Many of the GraphBLAS operations used in gbdemo are perhaps 3x to
% 50x faster than the corresponding MATLAB operations, depending on how
% many cores your computer has. Here's an example where GraphBLAS is
% asymptotically far faster than MATLAB R2019a: a simple assignment
% for a large matrix C:
%
% C(I,J) = A
%
% The matrix C is constructed via C = kron (B,B) where nnz (B) is
% roughly the bnz provided on input (with a default of bnz = 6000),
% so that C will have about bnz^2 entries, or 36 million by default.
% I and J are chosen randomly, and A is 5000-by-5000.
%
% When the problem becomes large, MATLAB will take a very long time.
% If you have enough memory, and want to see higher speedups in
% GraphBLAS, increase bnz (and be prepared to wait even longer).
% With the default bnz = 6000, this test takes about 4GB of RAM.
%
% On my Dell XPS 4-core laptop (Intel(R) Core(TM) i7-8565U, 16GB
% RAM), using MATLAB R2019a, when C becomes 9 million by 9 million,
% the computation C(I,J)=A for MATLAB matrices C, I, J, and A takes
% several minutes, whereas GraphBLAS takes less than a second, or
% about 500x faster than MATLAB. On a desktop with an Intel(R)
% Xeon(R) CPU E5-2698 v4 @ 2.20GHz with 20 hardware cores, the
% speedup over MATLAB is even more dramatic (up to 2,660x has been
% observed).
%
% See also GrB.assign, subsasgn.
% SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
% SPDX-License-Identifier: Apache-2.0
% reset to the default number of threads
ncores = demo_nproc ;
GrB.clear ;
nthreads = GrB.threads ;
help gbdemo2
fprintf ('\n# of threads used in GraphBLAS: %d\n\n', nthreads) ;
if (nargin < 1)
bnz = 6000 ;
end
k = 5000 ;
anz = 50000 ;
A = sprandn (k, k, anz / k^2) ;
for n = 1000:1000:6000
% reset the random number generator for repeatable results
rng ('default') ;
tic
B = sprandn (n, n, bnz / n^2) ;
C = kron (B, B) ;
cn = size (C,1) ;
I = randperm (cn, k) ;
J = randperm (cn, k) ;
G = GrB (C) ;
t_setup = toc ;
fprintf ('\nC(I,J)=A where C is %g million -by- %g million\n', ...
cn /1e6, cn /1e6) ;
fprintf ('with %g million entries:\n\n', nnz (C) / 1e6) ;
fprintf (' A is %d-by-%d with %d entries\n', k, k, nnz (A)) ;
fprintf (' setup time: %g sec\n', t_setup) ;
% do the assignment in GraphBLAS
tic
G (I,J) = A ;
gb_time = toc ;
fprintf (' GraphBLAS time: %g sec\n', gb_time) ;
fprintf (' Starting MATLAB ... please wait ... \n') ;
% do the same assignment in pure MATLAB
tic
C (I,J) = A ;
builtin_time = toc ;
fprintf (' MATLAB time: %g sec\n', builtin_time) ;
fprintf (' Speedup of GraphBLAS over MATLAB: %g\n', ...
builtin_time / gb_time) ;
% check the result
tic
assert (isequal (C, double (G))) ;
t_check = toc ;
fprintf (' check time: %g sec\n', t_check) ;
fprintf (' all tests passed\n') ;
end
|
