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
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
|
//------------------------------------------------------------------------------
// GB_mex_context_text: based on Demo/Programcontext_demo
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2025, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
#include "GB_mex.h"
#include "GB_mex_errors.h"
#define MIN(x,y) ((x) < (y)) ? (x) : (y)
#define MAX(x,y) ((x) > (y)) ? (x) : (y)
#define FREE_ALL { GB_mx_put_global (true) ; }
#define GET_DEEP_COPY ;
#define FREE_DEEP_COPY ;
void mexFunction
(
int nargout,
mxArray *pargout [ ],
int nargin,
const mxArray *pargin [ ]
)
{
GrB_Info info ;
bool malloc_debug = GB_mx_get_global (true) ;
//==========================================================================
// variant or context_demo
//==========================================================================
int nthreads_max = 0 ;
OK (GxB_Global_Option_get (GxB_GLOBAL_NTHREADS, &nthreads_max)) ;
nthreads_max = MIN (nthreads_max, 256) ;
printf ("context demo: nthreads_max %d\n", nthreads_max) ;
OK (GxB_print (GxB_CONTEXT_WORLD, 3)) ;
// use only a power of 2 number of threads
int nthreads = 1 ;
while (1)
{
if (2*nthreads > nthreads_max) break ;
nthreads = 2 * nthreads ;
}
nthreads = MIN (nthreads, 8) ;
printf ("\nnthreads to use: %d\n", nthreads) ;
OK (GxB_Global_Option_set (GxB_GLOBAL_NTHREADS, nthreads)) ;
#ifdef _OPENMP
omp_set_max_active_levels (2) ;
#endif
//==========================================================================
// more tests
//==========================================================================
GxB_Context Context = NULL ;
METHOD (GxB_Context_new (&Context)) ;
OK (GxB_Context_set (Context, GxB_NTHREADS, 5)) ;
OK (GxB_Context_set_String (Context, "my context", GrB_NAME)) ;
OK (GxB_Context_fprint (Context, "context", 3, stdout)) ;
int nth = 0 ;
OK (GxB_Context_get (GxB_CONTEXT_WORLD, GxB_NTHREADS, &nth)) ;
printf ("nth %d nthreads %d\n", nth, nthreads) ;
CHECK (nth == nthreads) ;
OK (GxB_Context_engage (Context)) ;
OK (GxB_Context_get (Context, GxB_NTHREADS, &nth)) ;
CHECK (nth == 5) ;
double chunk = 4096 ;
OK (GxB_Context_set (Context, GxB_CHUNK, chunk)) ;
chunk = 0 ;
OK (GxB_Context_get (Context, GxB_CHUNK, &chunk)) ;
CHECK (chunk == 4096) ;
chunk = 0 ;
OK (GxB_Context_get_FP64 (Context, GxB_CHUNK, &chunk)) ;
CHECK (chunk == 4096) ;
OK (GxB_Context_set (Context, GxB_CHUNK, -1)) ;
OK (GxB_Context_get (Context, GxB_CHUNK, &chunk)) ;
OK (chunk == GB_CHUNK_DEFAULT) ;
int gpu = 9 ;
OK (GxB_Context_get (Context, GxB_CONTEXT_GPU_ID, &gpu)) ;
CHECK (gpu == -1) ;
OK (GxB_Context_set (Context, GxB_CONTEXT_GPU_ID, 3)) ;
OK (GxB_Context_get (Context, GxB_CONTEXT_GPU_ID, &gpu)) ;
printf ("gpu now %d\n", gpu) ;
CHECK (gpu == -1) ;
OK (GxB_Context_engage (GxB_CONTEXT_WORLD)) ;
OK (GxB_Context_fprint (Context, "context", 3, stdout)) ;
chunk = -1 ;
GrB_Info expected = GrB_INVALID_VALUE ;
ERR (GxB_Context_get_FP64 (Context, GxB_NTHREADS, &chunk)) ;
ERR (GxB_Context_set_FP64 (Context, GxB_NTHREADS, chunk)) ;
nth = 99 ;
ERR (GxB_Context_get_INT32 (Context, GxB_CHUNK, &nth)) ;
ERR (GxB_Context_set_INT32 (Context, GxB_CHUNK, nth)) ;
ERR (GxB_Context_get (Context, 999, &nth)) ;
ERR (GxB_Context_set (Context, 999, nth)) ;
OK (GxB_Context_disengage (NULL)) ;
GrB_free (&Context) ;
expected = GrB_NULL_POINTER ;
ERR (GxB_Context_fprint (Context, "context", 3, stdout)) ;
//==========================================================================
// context_demo continued
//==========================================================================
//--------------------------------------------------------------------------
// construct tuples for a decent-sized random matrix
//--------------------------------------------------------------------------
uint64_t n = 1000 ; // 10000 ;
uint64_t nvals = 20000 ; // 2000000 ;
simple_rand_seed (1) ;
uint64_t *I = mxMalloc (nvals * sizeof (uint64_t)) ; // OK
uint64_t *J = mxMalloc (nvals * sizeof (uint64_t)) ; // OK
double *X = mxMalloc (nvals * sizeof (double)) ;
for (int k = 0 ; k < nvals ; k++)
{
I [k] = simple_rand_i ( ) % n ;
J [k] = simple_rand_i ( ) % n ;
X [k] = simple_rand_x ( ) ;
}
GrB_Matrix C = NULL ;
OK (GrB_Matrix_new (&C, GrB_FP64, n, n)) ;
OK (GrB_Matrix_build (C, I, J, X, nvals, GrB_PLUS_FP64)) ;
//--------------------------------------------------------------------------
// create random matrices parallel
//--------------------------------------------------------------------------
int nmats = MIN (2*nthreads, 256) ;
GrB_Matrix G [256] ;
for (int nmat = 1 ; nmat <= nmats ; nmat = 2*nmat)
{
double t1 = 0 ;
// create nmat matrices, each in parallel with varying # of threads
for (int nthreads2 = 1 ; nthreads2 <= nthreads ; nthreads2 *= 2)
{
int nouter = 1 ; // # of user threads in outer loop
int ninner = nthreads2 ; // # of threads each user thread can use
printf ("\nnmat: %4d nthreads2: %4d\n", nmat, nthreads2) ;
while (ninner >= 1)
{
if (nouter <= nmat)
{
double t = GB_omp_get_wtime ( ) ;
#pragma omp parallel for num_threads (nouter) \
schedule (dynamic, 1)
for (int k = 0 ; k < nmat ; k++)
{
// each user thread constructs its own context
GxB_Context Context = NULL ;
OK (GxB_Context_new (&Context)) ;
OK (GxB_Context_set (Context, GxB_NTHREADS, ninner)) ;
OK (GxB_Context_engage (Context)) ;
// kth user thread builds kth matrix with ninner threads
GrB_Matrix A = NULL ;
OK (GrB_Matrix_new (&A, GrB_FP64, n, n)) ;
OK (GrB_Matrix_build (A, I, J, X, nvals,
GrB_PLUS_FP64)) ;
// save the matrix just built
G [k] = A ;
A = NULL ;
// each user thread frees its own context
OK (GxB_Context_disengage (Context)) ;
OK (GxB_Context_free (&Context)) ;
}
t = GB_omp_get_wtime ( ) - t ;
if (nouter == 1 && ninner == 1) t1 = t ;
printf (" threads (%4d,%4d): %4d "
"time: %8.4f sec speedup: %8.3f\n",
nouter, ninner, nouter * ninner, t, t1/t) ;
// check results
for (int k = 0 ; k < nmat ; k++)
{
bool ok = GB_mx_isequal (C, G [k], 1e-14) ;
GrB_free (&(G [k])) ;
CHECK (ok) ;
}
}
nouter = nouter * 2 ;
ninner = ninner / 2 ;
}
}
}
mxFree (I) ;
mxFree (J) ;
mxFree (X) ;
GrB_free (&C) ;
GB_mx_put_global (true) ;
}
|
