Skip to content

Commit 651dd22

Browse files
committed
added benchmark program for lapack ?getri functions
1 parent 7c611a2 commit 651dd22

2 files changed

Lines changed: 306 additions & 0 deletions

File tree

benchmark/Makefile

Lines changed: 72 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -41,6 +41,7 @@ goto :: slinpack.goto dlinpack.goto clinpack.goto zlinpack.goto \
4141
cher2k.goto zher2k.goto \
4242
sgemv.goto dgemv.goto cgemv.goto zgemv.goto \
4343
sgeev.goto dgeev.goto cgeev.goto zgeev.goto \
44+
sgetri.goto dgetri.goto cgetri.goto zgetri.goto \
4445
ssymm.goto dsymm.goto csymm.goto zsymm.goto
4546

4647
acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
@@ -57,6 +58,7 @@ acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
5758
cher2k.acml zher2k.acml \
5859
sgemv.acml dgemv.acml cgemv.acml zgemv.acml \
5960
sgeev.acml dgeev.acml cgeev.acml zgeev.acml \
61+
sgetri.acml dgetri.acml cgetri.acml zgetri.acml \
6062
ssymm.acml dsymm.acml csymm.acml zsymm.acml
6163

6264
atlas :: slinpack.atlas dlinpack.atlas clinpack.atlas zlinpack.atlas \
@@ -74,6 +76,7 @@ atlas :: slinpack.atlas dlinpack.atlas clinpack.atlas zlinpack.atlas \
7476
cher2k.atlas zher2k.atlas \
7577
sgemv.atlas dgemv.atlas cgemv.atlas zgemv.atlas \
7678
sgeev.atlas dgeev.atlas cgeev.atlas zgeev.atlas \
79+
sgetri.atlas dgetri.atlas cgetri.atlas zgetri.atlas \
7780
ssymm.atlas dsymm.atlas csymm.atlas zsymm.atlas
7881

7982
mkl :: slinpack.mkl dlinpack.mkl clinpack.mkl zlinpack.mkl \
@@ -90,6 +93,7 @@ mkl :: slinpack.mkl dlinpack.mkl clinpack.mkl zlinpack.mkl \
9093
cher2k.mkl zher2k.mkl \
9194
sgemv.mkl dgemv.mkl cgemv.mkl zgemv.mkl \
9295
sgeev.mkl dgeev.mkl cgeev.mkl zgeev.mkl \
96+
sgetri.mkl dgetri.mkl cgetri.mkl zgetri.mkl \
9397
ssymm.mkl dsymm.mkl csymm.mkl zsymm.mkl
9498

9599
all :: goto atlas acml mkl
@@ -779,6 +783,61 @@ zgeev.mkl : zgeev.$(SUFFIX)
779783
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
780784

781785

786+
##################################### Sgetri ####################################################
787+
sgetri.goto : sgetri.$(SUFFIX) ../$(LIBNAME)
788+
$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) -lm
789+
790+
sgetri.acml : sgetri.$(SUFFIX)
791+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
792+
793+
sgetri.atlas : sgetri.$(SUFFIX)
794+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
795+
796+
sgetri.mkl : sgetri.$(SUFFIX)
797+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
798+
799+
##################################### Dgetri ####################################################
800+
dgetri.goto : dgetri.$(SUFFIX) ../$(LIBNAME)
801+
$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) -lm
802+
803+
dgetri.acml : dgetri.$(SUFFIX)
804+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
805+
806+
dgetri.atlas : dgetri.$(SUFFIX)
807+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
808+
809+
dgetri.mkl : dgetri.$(SUFFIX)
810+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
811+
812+
##################################### Cgetri ####################################################
813+
814+
cgetri.goto : cgetri.$(SUFFIX) ../$(LIBNAME)
815+
$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) -lm
816+
817+
cgetri.acml : cgetri.$(SUFFIX)
818+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
819+
820+
cgetri.atlas : cgetri.$(SUFFIX)
821+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
822+
823+
cgetri.mkl : cgetri.$(SUFFIX)
824+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
825+
826+
##################################### Zgetri ####################################################
827+
828+
zgetri.goto : zgetri.$(SUFFIX) ../$(LIBNAME)
829+
$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) -lm
830+
831+
zgetri.acml : zgetri.$(SUFFIX)
832+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
833+
834+
zgetri.atlas : zgetri.$(SUFFIX)
835+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
836+
837+
zgetri.mkl : zgetri.$(SUFFIX)
838+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
839+
840+
782841

783842
###################################################################################################
784843

@@ -932,6 +991,19 @@ cgeev.$(SUFFIX) : geev.c
932991
zgeev.$(SUFFIX) : geev.c
933992
$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
934993

994+
sgetri.$(SUFFIX) : getri.c
995+
$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
996+
997+
dgetri.$(SUFFIX) : getri.c
998+
$(CC) $(CFLAGS) -c -UCOMPLEX -DDOUBLE -o $(@F) $^
999+
1000+
cgetri.$(SUFFIX) : getri.c
1001+
$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
1002+
1003+
zgetri.$(SUFFIX) : getri.c
1004+
$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
1005+
1006+
9351007

9361008

9371009
clean ::

benchmark/getri.c

Lines changed: 234 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,234 @@
1+
/*********************************************************************/
2+
/* Copyright 2009, 2010 The University of Texas at Austin. */
3+
/* All rights reserved. */
4+
/* */
5+
/* Redistribution and use in source and binary forms, with or */
6+
/* without modification, are permitted provided that the following */
7+
/* conditions are met: */
8+
/* */
9+
/* 1. Redistributions of source code must retain the above */
10+
/* copyright notice, this list of conditions and the following */
11+
/* disclaimer. */
12+
/* */
13+
/* 2. Redistributions in binary form must reproduce the above */
14+
/* copyright notice, this list of conditions and the following */
15+
/* disclaimer in the documentation and/or other materials */
16+
/* provided with the distribution. */
17+
/* */
18+
/* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
19+
/* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
20+
/* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
21+
/* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
22+
/* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
23+
/* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
24+
/* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
25+
/* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
26+
/* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
27+
/* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
28+
/* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
29+
/* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
30+
/* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
31+
/* POSSIBILITY OF SUCH DAMAGE. */
32+
/* */
33+
/* The views and conclusions contained in the software and */
34+
/* documentation are those of the authors and should not be */
35+
/* interpreted as representing official policies, either expressed */
36+
/* or implied, of The University of Texas at Austin. */
37+
/*********************************************************************/
38+
39+
#include <stdio.h>
40+
#include <stdlib.h>
41+
#ifdef __CYGWIN32__
42+
#include <sys/time.h>
43+
#endif
44+
#include "common.h"
45+
46+
#undef GETRF
47+
#undef GETRI
48+
49+
#ifndef COMPLEX
50+
#ifdef XDOUBLE
51+
#define GETRF BLASFUNC(qgetrf)
52+
#define GETRI BLASFUNC(qgetri)
53+
#elif defined(DOUBLE)
54+
#define GETRF BLASFUNC(dgetrf)
55+
#define GETRI BLASFUNC(dgetri)
56+
#else
57+
#define GETRF BLASFUNC(sgetrf)
58+
#define GETRI BLASFUNC(sgetri)
59+
#endif
60+
#else
61+
#ifdef XDOUBLE
62+
#define GETRF BLASFUNC(xgetrf)
63+
#define GETRI BLASFUNC(xgetri)
64+
#elif defined(DOUBLE)
65+
#define GETRF BLASFUNC(zgetrf)
66+
#define GETRI BLASFUNC(zgetri)
67+
#else
68+
#define GETRF BLASFUNC(cgetrf)
69+
#define GETRI BLASFUNC(cgetri)
70+
#endif
71+
#endif
72+
73+
extern void GETRI(blasint *m, FLOAT *a, blasint *lda, blasint *ipiv, FLOAT *work, blasint *lwork, blasint *info);
74+
75+
#if defined(__WIN32__) || defined(__WIN64__)
76+
77+
#ifndef DELTA_EPOCH_IN_MICROSECS
78+
#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
79+
#endif
80+
81+
int gettimeofday(struct timeval *tv, void *tz){
82+
83+
FILETIME ft;
84+
unsigned __int64 tmpres = 0;
85+
static int tzflag;
86+
87+
if (NULL != tv)
88+
{
89+
GetSystemTimeAsFileTime(&ft);
90+
91+
tmpres |= ft.dwHighDateTime;
92+
tmpres <<= 32;
93+
tmpres |= ft.dwLowDateTime;
94+
95+
/*converting file time to unix epoch*/
96+
tmpres /= 10; /*convert into microseconds*/
97+
tmpres -= DELTA_EPOCH_IN_MICROSECS;
98+
tv->tv_sec = (long)(tmpres / 1000000UL);
99+
tv->tv_usec = (long)(tmpres % 1000000UL);
100+
}
101+
102+
return 0;
103+
}
104+
105+
#endif
106+
107+
#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
108+
109+
static void *huge_malloc(BLASLONG size){
110+
int shmid;
111+
void *address;
112+
113+
#ifndef SHM_HUGETLB
114+
#define SHM_HUGETLB 04000
115+
#endif
116+
117+
if ((shmid =shmget(IPC_PRIVATE,
118+
(size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
119+
SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
120+
printf( "Memory allocation failed(shmget).\n");
121+
exit(1);
122+
}
123+
124+
address = shmat(shmid, NULL, SHM_RND);
125+
126+
if ((BLASLONG)address == -1){
127+
printf( "Memory allocation failed(shmat).\n");
128+
exit(1);
129+
}
130+
131+
shmctl(shmid, IPC_RMID, 0);
132+
133+
return address;
134+
}
135+
136+
#define malloc huge_malloc
137+
138+
#endif
139+
140+
int MAIN__(int argc, char *argv[]){
141+
142+
FLOAT *a,*work;
143+
FLOAT wkopt[4];
144+
blasint *ipiv;
145+
blasint m, i, j, info,lwork;
146+
147+
int from = 1;
148+
int to = 200;
149+
int step = 1;
150+
151+
struct timeval start, stop;
152+
double time1;
153+
154+
argc--;argv++;
155+
156+
if (argc > 0) { from = atol(*argv); argc--; argv++;}
157+
if (argc > 0) { to = MAX(atol(*argv), from); argc--; argv++;}
158+
if (argc > 0) { step = atol(*argv); argc--; argv++;}
159+
160+
161+
fprintf(stderr, "From : %3d To : %3d Step = %3d\n", from, to, step);
162+
163+
if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
164+
fprintf(stderr,"Out of Memory!!\n");exit(1);
165+
}
166+
167+
if (( ipiv = (blasint *)malloc(sizeof(blasint) * to * COMPSIZE)) == NULL){
168+
fprintf(stderr,"Out of Memory!!\n");exit(1);
169+
}
170+
171+
172+
173+
for(j = 0; j < to; j++){
174+
for(i = 0; i < to * COMPSIZE; i++){
175+
a[i + j * to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
176+
}
177+
}
178+
179+
180+
lwork = -1;
181+
m=to;
182+
183+
GETRI(&m, a, &m, ipiv, wkopt, &lwork, &info);
184+
185+
lwork = (blasint)wkopt[0];
186+
if (( work = (FLOAT *)malloc(sizeof(FLOAT) * lwork * COMPSIZE)) == NULL){
187+
fprintf(stderr,"Out of Memory!!\n");exit(1);
188+
}
189+
190+
191+
#ifdef linux
192+
srandom(getpid());
193+
#endif
194+
195+
fprintf(stderr, " SIZE FLops Time Lwork\n");
196+
197+
for(m = from; m <= to; m += step){
198+
199+
fprintf(stderr, " %6d : ", (int)m);
200+
201+
GETRF (&m, &m, a, &m, ipiv, &info);
202+
203+
if (info) {
204+
fprintf(stderr, "Matrix is not singular .. %d\n", info);
205+
exit(1);
206+
}
207+
208+
gettimeofday( &start, (struct timezone *)0);
209+
210+
lwork = -1;
211+
GETRI(&m, a, &m, ipiv, wkopt, &lwork, &info);
212+
213+
lwork = (blasint)wkopt[0];
214+
GETRI(&m, a, &m, ipiv, work, &lwork, &info);
215+
gettimeofday( &stop, (struct timezone *)0);
216+
217+
if (info) {
218+
fprintf(stderr, "failed compute inverse matrix .. %d\n", info);
219+
exit(1);
220+
}
221+
222+
time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
223+
224+
fprintf(stderr,
225+
" %10.2f MFlops : %10.2f Sec : %d\n",
226+
COMPSIZE * COMPSIZE * (4.0/3.0 * (double)m * (double)m *(double)m - (double)m *(double)m + 5.0/3.0* (double)m) / time1 * 1.e-6,time1,lwork);
227+
228+
229+
}
230+
231+
return 0;
232+
}
233+
234+
void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));

0 commit comments

Comments
 (0)