timing.c 25.9 KB
Newer Older
1
2
/**
 *
3
4
 * @copyright (c) 2009-2014 The University of Tennessee and The University
 *                          of Tennessee Research Foundation.
5
 *                          All rights reserved.
6
 * @copyright (c) 2012-2016 Inria. All rights reserved.
7
 * @copyright (c) 2012-2015 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria, Univ. Bordeaux. All rights reserved.
8
9
10
11
12
13
14
15
16
17
18
 *
 **/

/**
 *
 * @file timing.c
 *
 *  MORSE auxiliary routines
 *  MORSE is a software package provided by Univ. of Tennessee,
 *  Univ. of California Berkeley and Univ. of Colorado Denver
 *
19
 * @version 0.9.0
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
 * @author Mathieu Faverge
 * @author Dulceneia Becker
 * @author Cedric Castagnede
 * @date 2010-11-15
 *
 **/

#if defined( _WIN32 ) || defined( _WIN64 )
#define int64_t __int64
#endif

/* Define these so that the Microsoft VC compiler stops complaining
   about scanf and friends */
#define _CRT_SECURE_NO_DEPRECATE
#define _CRT_SECURE_NO_WARNINGS

#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#if defined( _WIN32 ) || defined( _WIN64 )
#include <windows.h>
#else  /* Non-Windows */
#include <unistd.h>
#include <sys/resource.h>
#endif

48
#include <coreblas/include/cblas.h>
49
#include <coreblas/include/lapacke.h>
50
#include <morse.h>
51
#include <coreblas/include/coreblas.h>
52
53
#include "flops.h"
#include "timing.h"
54
#include <control/auxiliary.h>
55

56
#if defined(CHAMELEON_USE_MPI)
57
58
59
#include <mpi.h>
#endif

60
61
62
63
#if defined (CHAMELEON_SCHED_STARPU)
#include <starpu.h>
#endif

64
static int RunTest(int *iparam, _PREC *dparam, double *t_);
65
66
67
68
static void* morse_getaddr_null(const MORSE_desc_t *A, int m, int n)
{
    return (void*)( NULL );
}
69
70
71
72
73
74
75
76
77

int ISEED[4] = {0,0,0,1};   /* initial seed for zlarnv() */

static int
Test(int64_t n, int *iparam) {
    int      i, j, iter;
    int      thrdnbr, niter;
    int64_t  M, N, K, NRHS;
    double  *t;
78
#if defined(CHAMELEON_SIMULATION)
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
    _PREC    eps = 0.;
#else
    _PREC    eps = _LAMCH( 'e' );
#endif
    _PREC    dparam[IPARAM_DNBPARAM];
    double   fmuls, fadds, fp_per_mul, fp_per_add;
    double   sumgf, sumgf2, sumt, sd, flops, gflops;
    char    *s;
    char    *env[] = {
        "OMP_NUM_THREADS",
        "MKL_NUM_THREADS",
        "GOTO_NUM_THREADS",
        "ACML_NUM_THREADS",
        "ATLAS_NUM_THREADS",
        "BLAS_NUM_THREADS", ""
    };
    int gnuplot = 0;

/*
 * if hres = 0 then the test succeed
 * if hres = n then the test failed n times
 */
    int hres = 0;

    memset( &dparam, 0, IPARAM_DNBPARAM * sizeof(_PREC) );
    dparam[IPARAM_THRESHOLD_CHECK] = 100.0;

    thrdnbr = iparam[IPARAM_THRDNBR];
    niter   = iparam[IPARAM_NITER];

    M    = iparam[IPARAM_M];
    N    = iparam[IPARAM_N];
    K    = iparam[IPARAM_K];
    NRHS = K;
    (void)M;(void)N;(void)K;(void)NRHS;

    if ( (n < 0) || (thrdnbr < 0 ) ) {
        if (gnuplot && (MORSE_My_Mpi_Rank() == 0) ) {
            printf( "set title '%d_NUM_THREADS: ", thrdnbr );
            for (i = 0; env[i][0]; ++i) {
                s = getenv( env[i] );

                if (i) printf( " " ); /* separating space */

                for (j = 0; j < 5 && env[i][j] && env[i][j] != '_'; ++j)
                    printf( "%c", env[i][j] );

                if (s)
                    printf( "=%s", s );
                else
                    printf( "->%s", "?" );
            }
            printf( "'\n" );
            printf( "%s\n%s\n%s\n%s\n%s%s%s\n",
                    "set xlabel 'Matrix size'",
                    "set ylabel 'Gflop/s'",
                    "set key bottom",
                    gnuplot > 1 ? "set terminal png giant\nset output 'timeplot.png'" : "",
                    "plot '-' using 1:5 title '", _NAME, "' with linespoints" );
        }
        return 0;
    }

    if ( MORSE_My_Mpi_Rank() == 0)
        printf( "%7d %7d %7d ", iparam[IPARAM_M], iparam[IPARAM_N], iparam[IPARAM_K] );
    fflush( stdout );

    t = (double*)malloc(niter*sizeof(double));
    memset(t, 0, niter*sizeof(double));

    if (sizeof(_TYPE) == sizeof(_PREC)) {
        fp_per_mul = 1;
        fp_per_add = 1;
    } else {
        fp_per_mul = 6;
        fp_per_add = 2;
    }

    fadds = (double)(_FADDS);
    fmuls = (double)(_FMULS);
    flops = 1e-9 * (fmuls * fp_per_mul + fadds * fp_per_add);
    gflops = 0.0;

    if ( iparam[IPARAM_WARMUP] ) {
163
164
      int status = RunTest( iparam, dparam, &(t[0]));
      if (status != MORSE_SUCCESS) return status;
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
    }

    sumgf  = 0.0;
    double sumgf_upper  = 0.0;
    sumgf2 = 0.0;
    sumt   = 0.0;

    for (iter = 0; iter < niter; iter++)
    {
        if( iter == 0 ) {
          if ( iparam[IPARAM_TRACE] )
            iparam[IPARAM_TRACE] = 2;
          if ( iparam[IPARAM_DAG] )
            iparam[IPARAM_DAG] = 2;
          if ( iparam[IPARAM_PROFILE] )
            iparam[IPARAM_PROFILE] = 2;

182
183
          int status = RunTest( iparam, dparam, &(t[iter]));
          if (status != MORSE_SUCCESS) return status;
184
185
186
187
188

          iparam[IPARAM_TRACE] = 0;
          iparam[IPARAM_DAG] = 0;
          iparam[IPARAM_PROFILE] = 0;
        }
189
190
191
192
        else {
          int status = RunTest( iparam, dparam, &(t[iter]));
          if (status != MORSE_SUCCESS) return status;
        }
193
194
        gflops = flops / t[iter];

195
#if defined (CHAMELEON_SCHED_STARPU)
196
197
        if (iparam[IPARAM_BOUND])
        {
198
199
200
            double upper_gflops = 0.0;
            double tmin = 0.0;
            double integer_tmin = 0.0;
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
#if 0
            if (iparam[IPARAM_BOUNDDEPS]) {
                FILE *out = fopen("bounddeps.pl", "w");
                starpu_bound_print_lp(out);
                fclose(out);
                out = fopen("bound.dot", "w");
                starpu_bound_print_dot(out);
                fclose(out);
            } else {
                FILE *out = fopen("bound.pl", "w");
                starpu_bound_print_lp(out);
                fclose(out);
#else
            {
#endif
                starpu_bound_compute(&tmin, &integer_tmin, 0);
                upper_gflops  = (flops / (tmin / 1000.0));
                sumgf_upper += upper_gflops;
            }
        }
#endif
        sumt   += t[iter];
        sumgf  += gflops;
        sumgf2 += gflops*gflops;
    }

    gflops = sumgf / niter;
    sd = sqrt((sumgf2 - (sumgf*sumgf)/niter)/niter);

    if ( MORSE_My_Mpi_Rank() == 0) {
        printf( "%9.3f %9.2f +-%7.2f  ", sumt/niter, gflops, sd);

        if (iparam[IPARAM_BOUND] && !iparam[IPARAM_BOUNDDEPS])
            printf(" %9.2f",  sumgf_upper/niter);

        if ( iparam[IPARAM_PEAK] )
        {
            if (dparam[IPARAM_ESTIMATED_PEAK]<0.0f)
                printf("  n/a    n/a   ");
            else
                printf("  %5.2f%%  %9.2f ", 100.0f*(gflops/dparam[IPARAM_ESTIMATED_PEAK]), dparam[IPARAM_ESTIMATED_PEAK]);
        }

        if ( iparam[IPARAM_CHECK] ){
            hres = ( dparam[IPARAM_RES] / n / eps / (dparam[IPARAM_ANORM] * dparam[IPARAM_XNORM] + dparam[IPARAM_BNORM] ) > dparam[IPARAM_THRESHOLD_CHECK] );

            if (hres)
                printf( "%8.5e %8.5e %8.5e %8.5e                       %8.5e FAILURE",
                    dparam[IPARAM_RES], dparam[IPARAM_ANORM], dparam[IPARAM_XNORM], dparam[IPARAM_BNORM],
                    dparam[IPARAM_RES] / n / eps / (dparam[IPARAM_ANORM] * dparam[IPARAM_XNORM] + dparam[IPARAM_BNORM] ));
            else
                printf( "%8.5e %8.5e %8.5e %8.5e                       %8.5e SUCCESS",
                    dparam[IPARAM_RES], dparam[IPARAM_ANORM], dparam[IPARAM_XNORM], dparam[IPARAM_BNORM],
                    dparam[IPARAM_RES] / n / eps / (dparam[IPARAM_ANORM] * dparam[IPARAM_XNORM] + dparam[IPARAM_BNORM] ));
        }

        if ( iparam[IPARAM_INVERSE] )
            printf( " %8.5e %8.5e %8.5e     %8.5e",
                    dparam[IPARAM_RNORM], dparam[IPARAM_ANORM], dparam[IPARAM_AinvNORM],
                    dparam[IPARAM_RNORM] /((dparam[IPARAM_ANORM] * dparam[IPARAM_AinvNORM])*n*eps));

        printf("\n");

        fflush( stdout );
    }
    free(t);

    return hres;
}

static int
startswith(const char *s, const char *prefix) {
    size_t n = strlen( prefix );
    if (strncmp( s, prefix, n ))
        return 0;
    return 1;
}

static int
get_range(char *range, int *start_p, int *stop_p, int *step_p) {
    char *s, *s1, buf[21];
    int colon_count, copy_len, nbuf=20, n;
    int start=1000, stop=10000, step=1000;

    colon_count = 0;
    for (s = strchr( range, ':'); s; s = strchr( s+1, ':'))
        colon_count++;

    if (colon_count == 0) { /* No colon in range. */
        if (sscanf( range, "%d", &start ) < 1 || start < 1)
            return -1;
        step = start / 10;
        if (step < 1) step = 1;
        stop = start + 10 * step;

    } else if (colon_count == 1) { /* One colon in range.*/
        /* First, get the second number (after colon): the stop value. */
        s = strchr( range, ':' );
        if (sscanf( s+1, "%d", &stop ) < 1 || stop < 1)
            return -1;

        /* Next, get the first number (before colon): the start value. */
        n = s - range;
        copy_len = n > nbuf ? nbuf : n;
        strncpy( buf, range, copy_len );
        buf[copy_len] = 0;
        if (sscanf( buf, "%d", &start ) < 1 || start > stop || start < 1)
            return -1;

        /* Let's have 10 steps or less. */
        step = (stop - start) / 10;
        if (step < 1)
            step = 1;
    } else if (colon_count == 2) { /* Two colons in range. */
        /* First, get the first number (before the first colon): the start value. */
        s = strchr( range, ':' );
        n = s - range;
        copy_len = n > nbuf ? nbuf : n;
        strncpy( buf, range, copy_len );
        buf[copy_len] = 0;
        if (sscanf( buf, "%d", &start ) < 1 || start < 1)
            return -1;

        /* Next, get the second number (after the first colon): the stop value. */
        s1 = strchr( s+1, ':' );
        n = s1 - (s + 1);
        copy_len = n > nbuf ? nbuf : n;
        strncpy( buf, s+1, copy_len );
        buf[copy_len] = 0;
        if (sscanf( buf, "%d", &stop ) < 1 || stop < start)
            return -1;

        /* Finally, get the third number (after the second colon): the step value. */
        if (sscanf( s1+1, "%d", &step ) < 1 || step < 1)
            return -1;
    } else

        return -1;

    *start_p = start;
    *stop_p = stop;
    *step_p = step;

    return 0;
}

static void
show_help(char *prog_name) {
    printf( "Usage:\n%s [options]\n\n", prog_name );
    printf( "Options are:\n"
            "  --help           Show this help\n"
            "\n"
            "  --threads=X      Number of CPU workers (default: _SC_NPROCESSORS_ONLN)\n"
            "  --gpus=X         Number of GPU workers (default: 0)\n"
            "\n"
            "  --[a]sync        Enable/Disable synchronous calls in wrapper function such as POTRI. (default: async)\n"
357
            "  --[no]bigmat     Allocating one big mat or plenty of small (default: bigmat)\n"
358
            "  --[no]check      Check result (default: nocheck)\n"
359
            "  --[no]progress   Display progress indicator (default: noprogress)\n"
360
            "  --[no]gemm3m     Use gemm3m complex method (default: nogemm3m)\n"
361
            "  --[no]inv        Check on inverse (default: noinv)\n"
362
            "  --[no]warmup     Perform a warmup run to pre-load libraries (default: warmup)\n"
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
            "  --[no]trace      Enable/Disable trace generation (default: notrace)\n"
            "  --[no]dag        Enable/Disable DAG generation (default: nodag)\n"
            "                   Generates a dot_dag_file.dot.\n"
             "  --[no]profile   Print profiling informations (default: noprofile)\n"
             "  --nocpu         All GPU kernels are exclusively executed on GPUs (default: 0)\n"
/*             "  --inplace        Enable layout conversion inplace for lapack interface timers (default: enable)\n" */
/*             "  --outplace       Enable layout conversion out of place for lapack interface timers (default: disable)\n" */
/*             "  --[no]atun       Activate autotuning (default: noatun)\n" */
            "\n"
            "  --n_range=R      Range of N values\n"
            "                   with R=Start:Stop:Step (default: 500:5000:500)\n"
            "  --m=X            dimension (M) of the matrices (default: N)\n"
            "  --k=X            dimension (K) of the matrices (default: 1)\n"
            "  --nrhs=X         Number of right-hand size (default: 1)\n"
            "  --nb=N           Nb size. (default: 128)\n"
            "  --ib=N           IB size. (default: 32)\n"
            "\n"
            "  --niter=N        Number of iterations performed for each test (default: 1)\n"
            "\n"
            "  --rhblk=N        If N > 0, enable Householder mode for QR and LQ factorization\n"
            "                   N is the size of each subdomain (default: 0)\n"
/*             "\n" */
/*             " Options specific to the conversion format timings xgetri and xgecfi:\n" */
/*             "  --ifmt           Input format. (default: 0)\n" */
/*             "  --ofmt           Output format. (default: 1)\n" */
/*             "                   The possible values are:\n" */
/*             "                     0 - MorseCM, Column major\n" */
/*             "                     1 - MorseCCRB, Column-Colum rectangular block\n" */
/*             "                     2 - MorseCRRB, Column-Row rectangular block\n" */
/*             "                     3 - MorseRCRB, Row-Colum rectangular block\n" */
/*             "                     4 - MorseRRRB, Row-Row rectangular block\n" */
/*             "                     5 - MorseRM, Row Major\n" */
/*             "  --thrdbypb       Number of threads per subproblem for inplace transformation (default: 1)\n" */
            "\n");
}


static void
print_header(char *prog_name, int * iparam) {
    const char *bound_header   = iparam[IPARAM_BOUND]   ? "   thGflop/s" : "";
    const char *check_header   = iparam[IPARAM_CHECK]   ? "     ||Ax-b||       ||A||       ||x||       ||b|| ||Ax-b||/N/eps/(||A||||x||+||b||)  RETURN" : "";
    const char *inverse_header = iparam[IPARAM_INVERSE] ? " ||I-A*Ainv||       ||A||    ||Ainv||       ||Id - A*Ainv||/((||A|| ||Ainv||).N.eps)" : "";
    const char *peak_header    = iparam[IPARAM_PEAK]    ? "  (% of peak)  peak" : "";
406
#if defined(CHAMELEON_SIMULATION)
407
408
409
410
411
412
    _PREC    eps = 0.;
#else
    _PREC    eps = _LAMCH( 'e' );
#endif

    printf( "#\n"
413
            "# CHAMELEON %d.%d.%d, %s\n"
414
415
            "# Nb threads: %d\n"
            "# Nb GPUs:    %d\n"
416
417
418
419
#if defined(CHAMELEON_USE_MPI)
            "# Nb mpi:     %d\n"
            "# PxQ:        %dx%d\n"
#endif
420
421
422
423
            "# NB:         %d\n"
            "# IB:         %d\n"
            "# eps:        %e\n"
            "#\n",
424
425
426
            CHAMELEON_VERSION_MAJOR,
            CHAMELEON_VERSION_MINOR,
            CHAMELEON_VERSION_MICRO,
427
428
429
            prog_name,
            iparam[IPARAM_THRDNBR],
            iparam[IPARAM_NCUDAS],
430
431
432
433
#if defined(CHAMELEON_USE_MPI)
            iparam[IPARAM_NMPI],
            iparam[IPARAM_P], iparam[IPARAM_Q],
#endif
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
            iparam[IPARAM_NB],
            iparam[IPARAM_IB],
            eps );

    printf( "#     M       N  K/NRHS   seconds   Gflop/s Deviation%s%s%s\n",
            bound_header, peak_header, iparam[IPARAM_INVERSE] ? inverse_header : check_header);
    return;
}

int
main(int argc, char *argv[]) {
    int i, m, mx, nx;
    int nbnode = 1;
    int start =  500;
    int stop  = 5000;
    int step  =  500;
    int iparam[IPARAM_SIZEOF];
    int success = 0;

    memset(iparam, 0, IPARAM_SIZEOF*sizeof(int));

    iparam[IPARAM_THRDNBR       ] = -1;
    iparam[IPARAM_THRDNBR_SUBGRP] = 1;
    iparam[IPARAM_SCHEDULER     ] = 0;
    iparam[IPARAM_M             ] = -1;
    iparam[IPARAM_N             ] = 500;
    iparam[IPARAM_K             ] = 1;
    iparam[IPARAM_LDA           ] = -1;
    iparam[IPARAM_LDB           ] = -1;
    iparam[IPARAM_LDC           ] = -1;
    iparam[IPARAM_MB            ] = 128;
    iparam[IPARAM_NB            ] = 128;
    iparam[IPARAM_IB            ] = 32;
    iparam[IPARAM_NITER         ] = 1;
468
    iparam[IPARAM_WARMUP        ] = 1;
469
    iparam[IPARAM_CHECK         ] = 0;
470
    iparam[IPARAM_BIGMAT        ] = 1;
471
472
473
474
475
476
477
478
479
480
481
482
483
484
    iparam[IPARAM_VERBOSE       ] = 0;
    iparam[IPARAM_AUTOTUNING    ] = 0;
    iparam[IPARAM_INPUTFMT      ] = 0;
    iparam[IPARAM_OUTPUTFMT     ] = 0;
    iparam[IPARAM_TRACE         ] = 0;
    iparam[IPARAM_DAG           ] = 0;
    iparam[IPARAM_ASYNC         ] = 1;
    iparam[IPARAM_MX            ] = -1;
    iparam[IPARAM_NX            ] = -1;
    iparam[IPARAM_RHBLK         ] = 0;
    iparam[IPARAM_MX            ] = -1;
    iparam[IPARAM_NX            ] = -1;
    iparam[IPARAM_RHBLK         ] = 0;
    iparam[IPARAM_INPLACE       ] = MORSE_OUTOFPLACE;
485
    iparam[IPARAM_MODE          ] = 0;
486
487
488

    iparam[IPARAM_INVERSE       ] = 0;
    iparam[IPARAM_NCUDAS        ] = 0;
489
    iparam[IPARAM_NMPI          ] = 1;
490
491
    iparam[IPARAM_P             ] = 1;
    iparam[IPARAM_Q             ] = 1;
492
    iparam[IPARAM_GEMM3M        ] = 0;
493
    iparam[IPARAM_PROGRESS      ] = 0;
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
    iparam[IPARAM_PROFILE       ] = 0;
    iparam[IPARAM_PRINT_ERRORS  ] = 0;
    iparam[IPARAM_PEAK          ] = 0;
    iparam[IPARAM_PARALLEL_TASKS] = 0;
    iparam[IPARAM_NO_CPU        ] = 0;
    iparam[IPARAM_BOUND         ] = 0;
    iparam[IPARAM_BOUNDDEPS     ] = 0;
    iparam[IPARAM_BOUNDDEPSPRIO ] = 0;

    for (i = 1; i < argc && argv[i]; ++i) {
        if ( startswith( argv[i], "--help") || startswith( argv[i], "-help") ||
             startswith( argv[i], "--h") || startswith( argv[i], "-h") ) {
            show_help( argv[0] );
            return EXIT_SUCCESS;
        } else if (startswith( argv[i], "--threads=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_THRDNBR]) );
        } else if (startswith( argv[i], "--gpus=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_NCUDAS]) );
        } else if (startswith( argv[i], "--check" )) {
            iparam[IPARAM_CHECK] = 1;
        } else if (startswith( argv[i], "--nocheck" )) {
            iparam[IPARAM_CHECK] = 0;
516
517
518
519
        } else if (startswith( argv[i], "--bigmat" )) {
            iparam[IPARAM_BIGMAT] = 1;
        } else if (startswith( argv[i], "--nobigmat" )) {
            iparam[IPARAM_BIGMAT] = 0;
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
        } else if (startswith( argv[i], "--inv" )) {
            iparam[IPARAM_INVERSE] = 1;
        } else if (startswith( argv[i], "--noinv" )) {
            iparam[IPARAM_INVERSE] = 0;
        } else if (startswith( argv[i], "--warmup" )) {
            iparam[IPARAM_WARMUP] = 1;
        } else if (startswith( argv[i], "--nowarmup" )) {
            iparam[IPARAM_WARMUP] = 0;
/*         } else if (startswith( argv[i], "--atun" )) { */
/*             iparam[IPARAM_AUTOTUNING] = 1; */
/*         } else if (startswith( argv[i], "--noatun" )) { */
/*             iparam[IPARAM_AUTOTUNING] = 0; */
        } else if (startswith( argv[i], "--trace" )) {
            iparam[IPARAM_TRACE] = 1;
        } else if (startswith( argv[i], "--notrace" )) {
            iparam[IPARAM_TRACE] = 0;
536
537
538
539
        } else if (startswith( argv[i], "--gemm3m" )) {
            iparam[IPARAM_GEMM3M] = 1;
        } else if (startswith( argv[i], "--nogemm3m" )) {
            iparam[IPARAM_GEMM3M] = 0;
540
541
542
543
        } else if (startswith( argv[i], "--progress" )) {
            iparam[IPARAM_PROGRESS] = 1;
        } else if (startswith( argv[i], "--noprogress" )) {
            iparam[IPARAM_PROGRESS] = 0;
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
        } else if (startswith( argv[i], "--dag" )) {
            iparam[IPARAM_DAG] = 1;
        } else if (startswith( argv[i], "--nodag" )) {
            iparam[IPARAM_DAG] = 0;
        } else if (startswith( argv[i], "--sync" )) {
            iparam[IPARAM_ASYNC] = 0;
        } else if (startswith( argv[i], "--async" )) {
            iparam[IPARAM_ASYNC] = 1;
        } else if (startswith( argv[i], "--n_range=" )) {
            get_range( strchr( argv[i], '=' ) + 1, &start, &stop, &step );
        } else if (startswith( argv[i], "--m=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_M]) );
        } else if (startswith( argv[i], "--nb=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_NB]) );
            iparam[IPARAM_MB] = iparam[IPARAM_NB];
        } else if (startswith( argv[i], "--nrhs=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_K]) );
        } else if (startswith( argv[i], "--k=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_K]) );
        } else if (startswith( argv[i], "--ib=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_IB]) );
        } else if (startswith( argv[i], "--niter=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &iparam[IPARAM_NITER] );
        } else if (startswith( argv[i], "--mx=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_MX]) );
        } else if (startswith( argv[i], "--nx=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_NX]) );
        } else if (startswith( argv[i], "--rhblk=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_RHBLK]) );
/*         } else if (startswith( argv[i], "--inplace" )) { */
/*             iparam[IPARAM_INPLACE] = MORSE_INPLACE; */
/*         } else if (startswith( argv[i], "--outplace" )) { */
/*             iparam[IPARAM_INPLACE] = MORSE_OUTOFPLACE; */
/*         } else if (startswith( argv[i], "--ifmt=" )) { */
/*             sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_INPUTFMT]) ); */
/*         } else if (startswith( argv[i], "--ofmt=" )) { */
/*             sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_OUTPUTFMT]) ); */
/*         } else if (startswith( argv[i], "--thrdbypb=" )) { */
/*             sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_THRDNBR_SUBGRP]) ); */
        } else if (startswith( argv[i], "--profile" )) {
            iparam[IPARAM_PROFILE] = 1;
        } else if (startswith( argv[i], "--peak" )) {
            iparam[IPARAM_PEAK] = 1;
        } else if (startswith( argv[i], "--noprofile" )) {
            iparam[IPARAM_PROFILE] = 0;
        } else if (startswith( argv[i], "--printerrors" )) {
            iparam[IPARAM_PRINT_ERRORS] = 1;
/*         } else if (startswith( argv[i], "--parallel=" )) { */
/*             sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_PARALLEL_TASKS]) ); */
/*         } else if (startswith( argv[i], "--noparallel" )) { */
/*             iparam[IPARAM_PARALLEL_TASKS] = 0; */
        } else if (startswith( argv[i], "--nocpu" )) {
            iparam[IPARAM_NO_CPU] = 1;
        } else if (startswith( argv[i], "--bounddepsprio" )) {
            iparam[IPARAM_BOUND] = 1;
            iparam[IPARAM_BOUNDDEPS] = 1;
            iparam[IPARAM_BOUNDDEPSPRIO] = 1;
        } else if (startswith( argv[i], "--bounddeps" )) {
            iparam[IPARAM_BOUND] = 1;
            iparam[IPARAM_BOUNDDEPS] = 1;
        } else if (startswith( argv[i], "--bound" )) {
            iparam[IPARAM_BOUND] = 1;
        } else if (startswith( argv[i], "--p=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_P]) );
608
609
610
611
612
613
        } else if (startswith( argv[i], "--mode=" )) {
            sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_MODE]) );
            if (iparam[IPARAM_MODE] < 0 || iparam[IPARAM_MODE] > 20){
                fprintf( stderr, "Invalid mode: %s from 0 to 20\n", argv[i] );
                exit(0);
            }
614
615
616
617
        } else {
            fprintf( stderr, "Unknown option: %s\n", argv[i] );
        }
    }
618
619
620
621
622
623
624
#if !defined(CHAMELEON_USE_CUDA)
    if (iparam[IPARAM_NCUDAS] != 0){
    	fprintf(stderr, "ERROR: CHAMELEON_USE_CUDA is not defined. "
    			"The number of CUDA devices must be set to 0 (--gpus=0).\n");
    	return EXIT_FAILURE;
    }
#endif
625
626
627
628
629
630
631
632
633

    m  = iparam[IPARAM_M];
    mx = iparam[IPARAM_MX];
    nx = iparam[IPARAM_NX];

    /* Initialize MORSE */
    MORSE_Init( iparam[IPARAM_THRDNBR],
                iparam[IPARAM_NCUDAS] );

Mathieu Faverge's avatar
Mathieu Faverge committed
634
    /* Get the number of threads set by the runtime */
Mathieu Faverge's avatar
Mathieu Faverge committed
635
636
    iparam[IPARAM_THRDNBR] = MORSE_GetThreadNbr();

637
638
639
640
    /* Stops profiling here to avoid profiling uninteresting routines.
       It will be reactivated in the time_*.c routines with the macro START_TIMING() */
    RUNTIME_stop_profiling();

641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
    MORSE_Disable(MORSE_AUTOTUNING);
    MORSE_Set(MORSE_TILE_SIZE,        iparam[IPARAM_NB] );
    MORSE_Set(MORSE_INNER_BLOCK_SIZE, iparam[IPARAM_IB] );

    /* Householder mode */
    if (iparam[IPARAM_RHBLK] < 1) {
        MORSE_Set(MORSE_HOUSEHOLDER_MODE, MORSE_FLAT_HOUSEHOLDER);
    } else {
        MORSE_Set(MORSE_HOUSEHOLDER_MODE, MORSE_TREE_HOUSEHOLDER);
        MORSE_Set(MORSE_HOUSEHOLDER_SIZE, iparam[IPARAM_RHBLK]);
    }

    if (iparam[IPARAM_PROFILE] == 1)
        MORSE_Enable(MORSE_PROFILING_MODE);

    if (iparam[IPARAM_PRINT_ERRORS] == 1)
        MORSE_Enable(MORSE_ERRORS);

659
660
661
    if (iparam[IPARAM_PROGRESS] == 1)
        MORSE_Enable(MORSE_PROGRESS);

662
663
664
    if (iparam[IPARAM_GEMM3M] == 1)
        MORSE_Enable(MORSE_GEMM3M);

665
#if defined(CHAMELEON_USE_MPI)
666
    MORSE_Comm_size( &nbnode );
667
    iparam[IPARAM_NMPI] = nbnode;
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
    /* Check P */
    if ( (iparam[IPARAM_P] > 1) &&
         (nbnode % iparam[IPARAM_P] != 0) ) {
      fprintf(stderr, "ERROR: %d doesn't divide the number of node %d\n",
              iparam[IPARAM_P], nbnode );
      return EXIT_FAILURE;
    }
#endif
    iparam[IPARAM_Q] = nbnode / iparam[IPARAM_P];

    /* Layout conversion */
    MORSE_Set(MORSE_TRANSLATION_MODE, iparam[IPARAM_INPLACE]);

    if ( MORSE_My_Mpi_Rank() == 0 )
        print_header( argv[0], iparam);

    if (step < 1) step = 1;

686
687
    int status = Test( -1, iparam ); /* print header */
    if (status != MORSE_SUCCESS) return status;
688
689
690
691
    for (i = start; i <= stop; i += step)
    {
        if ( nx > 0 ) {
            iparam[IPARAM_M] = i;
692
            iparam[IPARAM_N] = chameleon_max(1, i/nx);
693
        } else if ( mx > 0 ) {
694
            iparam[IPARAM_M] = chameleon_max(1, i/mx);
695
696
697
698
699
700
            iparam[IPARAM_N] = i;
        } else {
            if ( m == -1 )
                iparam[IPARAM_M] = i;
            iparam[IPARAM_N] = i;
        }
701
702
703
        int status = Test( iparam[IPARAM_N], iparam );
        if (status != MORSE_SUCCESS) return status;
        success += status;
704
705
706
707
708
709
    }

    MORSE_Finalize();
    return success;
}