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
|
function test120
%TEST120 performance tests for GrB_assign
% SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
% SPDX-License-Identifier: Apache-2.0
fprintf ('test120:------------------- C(I,J)<!M> += scalar:\n') ;
[save save_chunk] = nthreads_get ;
chunk = 4096 ;
rng ('default') ;
n = 4000 ; ;
k = 3000 ;
% I0.begin = 0 ;
% I0.inc = 1 ;
% I0.end = k-1 ;
I1 = randperm (k) ;
I0 = uint64 (I1) - 1 ;
d.mask = 'complement' ;
ncores = feature_numcores ;
for dc = [2 0 1e-6 1e-5 1e-4 1e-3 1e-2 0.1 1]
if (dc == 2)
C0 = sparse (rand (n)) ;
else
C0 = sprand (n, n, dc) ;
end
for dm = [2 0 1e-6 1e-5 1e-4 1e-3 1e-2 0.1 1]
if (dm == 2)
M = sparse (ones (k)) ;
else
M = spones (sprand (k, k, dm)) ;
end
Mbool = logical (M) ;
fprintf ('\n--------------------------------------\n') ;
fprintf ('dc = %g nnz(C) %8.4f million\n', dc, nnz(C0)/1e6) ;
fprintf ('dm = %g nnz(M) %8.4f million\n', dm, nnz(M)/1e6) ;
tm = inf ;
if (n < 500)
% built-in methods are exceedingly slow for this case
tic
C1 = C0 ;
% Csub = C1 (1:k, 1:k) ;
Csub = C1 (I1, I1) ;
Csub (~Mbool) = Csub (~Mbool) + pi ;
% C1 (1:k, 1:k) = Csub ;
C1 (I1, I1) = Csub ;
tm = toc ;
end
scalar = sparse (pi) ;
for nthreads = [1 2 4 8 16 20 32 40 64]
if (nthreads > 2*ncores)
break ;
end
if (nthreads > 1 && t1 < 0.01)
break ;
end
nthreads_set (nthreads, chunk) ;
C2 = GB_mex_subassign (C0, M, 'plus', scalar, I0, I0, d) ;
tic
C2 = GB_mex_subassign (C0, M, 'plus', scalar, I0, I0, d) ;
tg = toc ;
if (n < 500)
assert (isequal (C1, C2.matrix)) ;
end
if (nthreads == 1)
t1 = tg ;
end
fprintf ('%3d : builtin: %10.4f GB: %8.4f ', nthreads, tm, tg) ;
fprintf (' speedup %10.4f %10.4f\n', tm / tg, t1/tg) ;
end
end
end
nthreads_set (save, save_chunk) ;
|
