Skip to content

Commit 32c847d

Browse files
committed
benchmark/hpmv&hbmv: add benchmark/hpmv.c and benchmark/hbmv.c
Signed-off-by: Xuqiang Chen chenxuqiang3@hisilicon.com
1 parent ad9e531 commit 32c847d

3 files changed

Lines changed: 510 additions & 0 deletions

File tree

benchmark/Makefile

Lines changed: 93 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -69,6 +69,8 @@ goto :: slinpack.goto dlinpack.goto clinpack.goto zlinpack.goto \
6969
sasum.goto dasum.goto casum.goto zasum.goto \
7070
ssymv.goto dsymv.goto csymv.goto zsymv.goto \
7171
chemv.goto zhemv.goto \
72+
chbmv.goto zhbmv.goto \
73+
chpmv.goto zhpmv.goto \
7274
chemm.goto zhemm.goto \
7375
cherk.goto zherk.goto \
7476
cher2k.goto zher2k.goto \
@@ -97,6 +99,8 @@ acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
9799
sasum.acml dasum.acml casum.acml zasum.acml \
98100
ssymv.acml dsymv.acml csymv.acml zsymv.acml \
99101
chemv.acml zhemv.acml \
102+
chbmv.acml zhbmv.acml \
103+
chpmv.acml zhpmv.acml \
100104
chemm.acml zhemm.acml \
101105
cherk.acml zherk.acml \
102106
cher2k.acml zher2k.acml \
@@ -125,6 +129,8 @@ atlas :: slinpack.atlas dlinpack.atlas clinpack.atlas zlinpack.atlas \
125129
sasum.atlas dasum.atlas casum.atlas zasum.atlas \
126130
ssymv.atlas dsymv.atlas csymv.atlas zsymv.atlas \
127131
chemv.atlas zhemv.atlas \
132+
chbmv.atlas zhbmv.atlas \
133+
chpmv.atlas zhpmv.atlas \
128134
chemm.acml zhemm.acml \
129135
chemm.atlas zhemm.atlas \
130136
cherk.atlas zherk.atlas \
@@ -154,6 +160,8 @@ mkl :: slinpack.mkl dlinpack.mkl clinpack.mkl zlinpack.mkl \
154160
sasum.mkl dasum.mkl casum.mkl zasum.mkl \
155161
ssymv.mkl dsymv.mkl csymv.mkl zsymv.mkl \
156162
chemv.mkl zhemv.mkl \
163+
chbmv.mkl zhbmv.mkl \
164+
chpmv.mkl zhpmv.mkl \
157165
chemm.mkl zhemm.mkl \
158166
cherk.mkl zherk.mkl \
159167
cher2k.mkl zher2k.mkl \
@@ -183,6 +191,8 @@ goto :: sgemm.goto dgemm.goto cgemm.goto zgemm.goto \
183191
sasum.goto dasum.goto casum.goto zasum.goto \
184192
ssymv.goto dsymv.goto \
185193
chemv.goto zhemv.goto \
194+
chbmv.goto zhbmv.goto \
195+
chpmv.goto zhpmv.goto \
186196
chemm.goto zhemm.goto \
187197
cherk.goto zherk.goto \
188198
cher2k.goto zher2k.goto \
@@ -210,6 +220,8 @@ acml :: slinpack.acml dlinpack.acml clinpack.acml zlinpack.acml \
210220
sasum.acml dasum.acml casum.acml zasum.acml \
211221
ssymv.acml dsymv.acml csymv.acml zsymv.acml \
212222
chemv.acml zhemv.acml \
223+
chbmv.acml zhbmv.acml \
224+
chpmv.acml zhpmv.acml \
213225
chemm.acml zhemm.acml \
214226
cherk.acml zherk.acml \
215227
cher2k.acml zher2k.acml \
@@ -238,6 +250,8 @@ atlas :: slinpack.atlas dlinpack.atlas clinpack.atlas zlinpack.atlas \
238250
sasum.atlas dasum.atlas casum.atlas zasum.atlas \
239251
ssymv.atlas dsymv.atlas csymv.atlas zsymv.atlas \
240252
chemv.atlas zhemv.atlas \
253+
chbmv.atlas zhbmv.atlas \
254+
chpmv.atlas zhpmv.atlas \
241255
chemm.acml zhemm.acml \
242256
chemm.atlas zhemm.atlas \
243257
cherk.atlas zherk.atlas \
@@ -269,6 +283,8 @@ mkl :: slinpack.mkl dlinpack.mkl clinpack.mkl zlinpack.mkl \
269283
sasum.mkl dasum.mkl casum.mkl zasum.mkl \
270284
ssymv.mkl dsymv.mkl csymv.mkl zsymv.mkl \
271285
chemv.mkl zhemv.mkl \
286+
chbmv.mkl zhbmv.mkl \
287+
chpmv.mkl zhpmv.mkl \
272288
chemm.mkl zhemm.mkl \
273289
cherk.mkl zherk.mkl \
274290
cher2k.mkl zher2k.mkl \
@@ -308,6 +324,8 @@ veclib :: slinpack.veclib dlinpack.veclib clinpack.veclib zlinpack.veclib \
308324
sasum.veclib dasum.veclib casum.veclib zasum.veclib \
309325
ssymv.veclib dsymv.veclib csymv.veclib zsymv.veclib \
310326
chemv.veclib zhemv.veclib \
327+
chbmv.veclib zhbmv.veclib \
328+
chpmv.veclib zhpmv.veclib \
311329
chemm.veclib zhemm.veclib \
312330
cherk.veclib zherk.veclib \
313331
cher2k.veclib zher2k.veclib \
@@ -1542,7 +1560,70 @@ zhemv.mkl : zhemv.$(SUFFIX)
15421560

15431561
zhemv.veclib : zhemv.$(SUFFIX)
15441562
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1563+
##################################### Chbmv ####################################################
15451564

1565+
chbmv.goto : chbmv.$(SUFFIX) ../$(LIBNAME)
1566+
$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
1567+
1568+
chbmv.acml : chbmv.$(SUFFIX)
1569+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1570+
1571+
chbmv.atlas : chbmv.$(SUFFIX)
1572+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1573+
1574+
chbmv.mkl : chbmv.$(SUFFIX)
1575+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1576+
1577+
chbmv.veclib : chbmv.$(SUFFIX)
1578+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1579+
##################################### Zhbmv ####################################################
1580+
1581+
zhbmv.goto : zhbmv.$(SUFFIX) ../$(LIBNAME)
1582+
$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
1583+
1584+
zhbmv.acml : zhbmv.$(SUFFIX)
1585+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1586+
1587+
zhbmv.atlas : zhbmv.$(SUFFIX)
1588+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1589+
1590+
zhbmv.mkl : zhbmv.$(SUFFIX)
1591+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1592+
1593+
zhbmv.veclib : zhbmv.$(SUFFIX)
1594+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1595+
##################################### Chpmv ####################################################
1596+
1597+
chpmv.goto : chpmv.$(SUFFIX) ../$(LIBNAME)
1598+
$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
1599+
1600+
chpmv.acml : chpmv.$(SUFFIX)
1601+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1602+
1603+
chpmv.atlas : chpmv.$(SUFFIX)
1604+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1605+
1606+
chpmv.mkl : chpmv.$(SUFFIX)
1607+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1608+
1609+
chpmv.veclib : chpmv.$(SUFFIX)
1610+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1611+
##################################### Zhpmv ####################################################
1612+
1613+
zhpmv.goto : zhpmv.$(SUFFIX) ../$(LIBNAME)
1614+
$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
1615+
1616+
zhpmv.acml : zhpmv.$(SUFFIX)
1617+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBACML) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1618+
1619+
zhpmv.atlas : zhpmv.$(SUFFIX)
1620+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBATLAS) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1621+
1622+
zhpmv.mkl : zhpmv.$(SUFFIX)
1623+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBMKL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
1624+
1625+
zhpmv.veclib : zhpmv.$(SUFFIX)
1626+
-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
15461627
##################################### Sdot ####################################################
15471628
sdot.goto : sdot.$(SUFFIX) ../$(LIBNAME)
15481629
$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
@@ -2331,6 +2412,18 @@ chemv.$(SUFFIX) : hemv.c
23312412
zhemv.$(SUFFIX) : hemv.c
23322413
$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
23332414

2415+
chbmv.$(SUFFIX) : hbmv.c
2416+
$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
2417+
2418+
zhbmv.$(SUFFIX) : hbmv.c
2419+
$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
2420+
2421+
chpmv.$(SUFFIX) : hpmv.c
2422+
$(CC) $(CFLAGS) -c -DCOMPLEX -UDOUBLE -o $(@F) $^
2423+
2424+
zhpmv.$(SUFFIX) : hpmv.c
2425+
$(CC) $(CFLAGS) -c -DCOMPLEX -DDOUBLE -o $(@F) $^
2426+
23342427
sdot.$(SUFFIX) : dot.c
23352428
$(CC) $(CFLAGS) -c -UCOMPLEX -UDOUBLE -o $(@F) $^
23362429

benchmark/hbmv.c

Lines changed: 210 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,210 @@
1+
/***************************************************************************
2+
Copyright (c) 2014, The OpenBLAS Project
3+
All rights reserved.
4+
Redistribution and use in source and binary forms, with or without
5+
modification, are permitted provided that the following conditions are
6+
met:
7+
1. Redistributions of source code must retain the above copyright
8+
notice, this list of conditions and the following disclaimer.
9+
2. Redistributions in binary form must reproduce the above copyright
10+
notice, this list of conditions and the following disclaimer in
11+
the documentation and/or other materials provided with the
12+
distribution.
13+
3. Neither the name of the OpenBLAS project nor the names of
14+
its contributors may be used to endorse or promote products
15+
derived from this software without specific prior written permission.
16+
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
17+
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18+
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19+
ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
20+
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21+
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
22+
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
23+
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
24+
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
25+
USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26+
*****************************************************************************/
27+
28+
#include <stdio.h>
29+
#include <stdlib.h>
30+
#ifdef __CYGWIN32__
31+
#include <sys/time.h>
32+
#endif
33+
#include "common.h"
34+
35+
36+
#undef HBMV
37+
38+
39+
#ifdef DOUBLE
40+
#define HBMV BLASFUNC(zhbmv)
41+
#else
42+
#define HBMV BLASFUNC(chbmv)
43+
#endif
44+
45+
46+
#if defined(__WIN32__) || defined(__WIN64__)
47+
48+
#ifndef DELTA_EPOCH_IN_MICROSECS
49+
#define DELTA_EPOCH_IN_MICROSECS 11644473600000000ULL
50+
#endif
51+
52+
int gettimeofday(struct timeval *tv, void *tz) {
53+
54+
FILETIME ft;
55+
unsigned __int64 tmpres = 0;
56+
static int tzflag;
57+
58+
if (NULL != tv)
59+
{
60+
GetSystemTimeAsFileTime(&ft);
61+
62+
tmpres |= ft.dwHighDateTime;
63+
tmpres <<= 32;
64+
tmpres |= ft.dwLowDateTime;
65+
66+
/*converting file time to unix epoch*/
67+
tmpres /= 10; /*convert into microseconds*/
68+
tmpres -= DELTA_EPOCH_IN_MICROSECS;
69+
tv->tv_sec = (long)(tmpres / 1000000UL);
70+
tv->tv_usec = (long)(tmpres % 1000000UL);
71+
}
72+
73+
return 0;
74+
}
75+
76+
#endif
77+
78+
#if !defined(__WIN32__) && !defined(__WIN64__) && !defined(__CYGWIN32__) && 0
79+
80+
static void *huge_malloc(BLASLONG size) {
81+
int shmid;
82+
void *address;
83+
84+
#ifndef SHM_HUGETLB
85+
#define SHM_HUGETLB 04000
86+
#endif
87+
88+
if ((shmid =shmget(IPC_PRIVATE,
89+
(size + HUGE_PAGESIZE) & ~(HUGE_PAGESIZE - 1),
90+
SHM_HUGETLB | IPC_CREAT |0600)) < 0) {
91+
printf( "Memory allocation failed(shmget).\n");
92+
exit(1);
93+
}
94+
95+
address = shmat(shmid, NULL, SHM_RND);
96+
97+
if ((BLASLONG)address == -1){
98+
printf( "Memory allocation failed(shmat).\n");
99+
exit(1);
100+
}
101+
102+
shmctl(shmid, IPC_RMID, 0);
103+
104+
return address;
105+
}
106+
107+
#define malloc huge_malloc
108+
109+
#endif
110+
111+
int main(int argc, char *argv[]){
112+
113+
FLOAT *a, *x, *y;
114+
FLOAT alpha[] = {1.0, 1.0};
115+
FLOAT beta [] = {0.0, 0.0};
116+
blasint k = 1;
117+
char uplo='L';
118+
blasint m, i, j;
119+
blasint inc_x=1, inc_y=1;
120+
int loops = 1;
121+
int l;
122+
char *p;
123+
124+
int from = 1;
125+
int to = 200;
126+
int step = 1;
127+
128+
struct timeval start, stop;
129+
double time1,timeg;
130+
131+
argc--;argv++;
132+
133+
if (argc > 0) { from = atol(*argv); argc--; argv++;}
134+
if (argc > 0) { to = MAX(atol(*argv), from); argc--; argv++;}
135+
if (argc > 0) { step = atol(*argv); argc--; argv++;}
136+
137+
if ((p = getenv("OPENBLAS_LOOPS"))) loops = atoi(p);
138+
if ((p = getenv("OPENBLAS_INCX"))) inc_x = atoi(p);
139+
if ((p = getenv("OPENBLAS_INCY"))) inc_y = atoi(p);
140+
if ((p = getenv("OPENBLAS_UPLO"))) uplo=*p;
141+
if ((p = getenv("OPENBLAS_K"))) k = atoi(p);
142+
143+
fprintf(stderr, "From : %3d To : %3d Step = %3d Uplo = '%c' k = %d Inc_x = %d Inc_y = %d Loops = %d\n",
144+
from, to, step, uplo, k, inc_x, inc_y, loops);
145+
146+
if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL) {
147+
fprintf(stderr,"Out of Memory!!\n");
148+
exit(1);
149+
}
150+
151+
if (( x = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_x) * COMPSIZE)) == NULL) {
152+
fprintf(stderr,"Out of Memory!!\n");
153+
exit(1);
154+
}
155+
156+
if (( y = (FLOAT *)malloc(sizeof(FLOAT) * to * abs(inc_y) * COMPSIZE)) == NULL) {
157+
fprintf(stderr,"Out of Memory!!\n");
158+
exit(1);
159+
}
160+
161+
#ifdef linux
162+
srandom(getpid());
163+
#endif
164+
165+
fprintf(stderr, " SIZE Flops\n");
166+
167+
for(m = from; m <= to; m += step) {
168+
169+
timeg=0;
170+
171+
fprintf(stderr, " %6dx%d : ", (int)m, (int)m);
172+
173+
for(j = 0; j < m; j++) {
174+
for(i = 0; i < m * COMPSIZE; i++) {
175+
a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
176+
}
177+
}
178+
179+
for (l = 0; l < loops; l++) {
180+
181+
for (i = 0; i < m * COMPSIZE * abs(inc_x); i++) {
182+
x[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
183+
}
184+
185+
for (i = 0; i < m * COMPSIZE * abs(inc_y); i++) {
186+
y[i] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
187+
}
188+
189+
gettimeofday( &start, (struct timezone *)0);
190+
191+
HBMV (&uplo, &m, &k, alpha, a, &m, x, &inc_x, beta, y, &inc_y );
192+
193+
gettimeofday( &stop, (struct timezone *)0);
194+
195+
time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
196+
197+
timeg += time1;
198+
199+
}
200+
201+
timeg /= loops;
202+
203+
fprintf(stderr, " %10.2f MFlops\n",
204+
COMPSIZE * COMPSIZE * 2. * (double)(2 * k + 1) * (double)m / timeg * 1.e-6);
205+
}
206+
207+
return 0;
208+
}
209+
210+
// void main(int argc, char *argv[]) __attribute__((weak, alias("MAIN__")));

0 commit comments

Comments
 (0)