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
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
|
SUBROUTINE STIMQP( LINE, NM, MVAL, NVAL, NLDA, LDAVAL, TIMMIN, A,
$ COPYA, TAU, WORK, IWORK, RESLTS, LDR1, LDR2,
$ NOUT )
*
* -- LAPACK timing routine (version 3.0) --
* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
* Courant Institute, Argonne National Lab, and Rice University
* June 30, 1999
*
* .. Scalar Arguments ..
CHARACTER*80 LINE
INTEGER LDR1, LDR2, NLDA, NM, NOUT
REAL TIMMIN
* ..
* .. Array Arguments ..
INTEGER IWORK( * ), LDAVAL( * ), MVAL( * ), NVAL( * )
REAL A( * ), COPYA( * ), RESLTS( LDR1, LDR2, * ),
$ TAU( * ), WORK( * )
* ..
*
* Purpose
* =======
*
* STIMQP times the LAPACK routines to perform the QR factorization with
* column pivoting of a REAL general matrix.
*
* Two matrix types may be used for timing. The number of types is
* set in the parameter NMODE and the matrix types are set in the vector
* MODES, using the following key:
* 2. BREAK1 D(1:N-1)=1 and D(N)=1.0/COND in SLATMS
* 3. GEOM D(I)=COND**(-(I-1)/(N-1)) in SLATMS
* These numbers are chosen to correspond with the matrix types in the
* test code.
*
* Arguments
* =========
*
* LINE (input) CHARACTER*80
* The input line that requested this routine. The first six
* characters contain either the name of a subroutine or a
* generic path name. The remaining characters may be used to
* specify the individual routines to be timed. See ATIMIN for
* a full description of the format of the input line.
*
* NM (input) INTEGER
* The number of values of M and N contained in the vectors
* MVAL and NVAL. The matrix sizes are used in pairs (M,N).
*
* MVAL (input) INTEGER array, dimension (NM)
* The values of the matrix row dimension M.
*
* NVAL (input) INTEGER array, dimension (NM)
* The values of the matrix column dimension N.
*
* NLDA (input) INTEGER
* The number of values of LDA contained in the vector LDAVAL.
*
* LDAVAL (input) INTEGER array, dimension (NLDA)
* The values of the leading dimension of the array A.
*
* TIMMIN (input) REAL
* The minimum time a subroutine will be timed.
*
* A (workspace) REAL array, dimension (LDAMAX*NMAX)
* where LDAMAX and NMAX are the maximum values of LDA and N.
*
* COPYA (workspace) REAL array, dimension (LDAMAX*NMAX)
*
* TAU (workspace) REAL array, dimension (min(M,N))
*
* WORK (workspace) REAL array, dimension (3*NMAX)
*
* IWORK (workspace) INTEGER array, dimension (2*NMAX)
*
* RESLTS (workspace) REAL array, dimension
* (LDR1,LDR2,NLDA)
* The timing results for each subroutine over the relevant
* values of MODE, (M,N), and LDA.
*
* LDR1 (input) INTEGER
* The first dimension of RESLTS. LDR1 >= max(1,NM).
*
* LDR2 (input) INTEGER
* The second dimension of RESLTS. LDR2 >= max(1,NM).
*
* NOUT (input) INTEGER
* The unit number for output.
*
* =====================================================================
*
* .. Parameters ..
INTEGER NSUBS, NMODE
PARAMETER ( NSUBS = 1, NMODE = 2 )
REAL ONE
PARAMETER ( ONE = 1.0E+0 )
* ..
* .. Local Scalars ..
CHARACTER*3 PATH
CHARACTER*6 CNAME
INTEGER I, IC, ICL, ILDA, IM, IMODE, INFO, LDA, M,
$ MINMN, MODE, N
REAL COND, DMAX, OPS, S1, S2, TIME, UNTIME
* ..
* .. Local Arrays ..
LOGICAL TIMSUB( NSUBS )
CHARACTER*6 SUBNAM( NSUBS )
INTEGER ISEED( 4 ), MODES( NMODE )
* ..
* .. External Functions ..
REAL SECOND, SLAMCH, SMFLOP, SOPLA
EXTERNAL SECOND, SLAMCH, SMFLOP, SOPLA
* ..
* .. External Subroutines ..
EXTERNAL ATIMCK, ATIMIN, ICOPY, SGEQPF, SLACPY, SLATMS,
$ SPRTB5
* ..
* .. Intrinsic Functions ..
INTRINSIC MIN, REAL
* ..
* .. Data statements ..
DATA SUBNAM / 'SGEQPF' /
DATA MODES / 2, 3 /
DATA ISEED / 0, 0, 0, 1 /
* ..
* .. Executable Statements ..
*
* Extract the timing request from the input line.
*
PATH( 1: 1 ) = 'Single precision'
PATH( 2: 3 ) = 'QP'
CALL ATIMIN( PATH, LINE, NSUBS, SUBNAM, TIMSUB, NOUT, INFO )
IF( .NOT.TIMSUB( 1 ) .OR. INFO.NE.0 )
$ GO TO 80
*
* Check that M <= LDA for the input values.
*
CNAME = LINE( 1: 6 )
CALL ATIMCK( 1, CNAME, NM, MVAL, NLDA, LDAVAL, NOUT, INFO )
IF( INFO.GT.0 ) THEN
WRITE( NOUT, FMT = 9999 )CNAME
GO TO 80
END IF
*
* Set the condition number and scaling factor for the matrices
* to be generated.
*
DMAX = ONE
COND = ONE / SLAMCH( 'Precision' )
*
* Do for each pair of values (M,N):
*
DO 60 IM = 1, NM
M = MVAL( IM )
N = NVAL( IM )
MINMN = MIN( M, N )
*
* Do for each value of LDA:
*
DO 50 ILDA = 1, NLDA
LDA = LDAVAL( ILDA )
DO 40 IMODE = 1, NMODE
MODE = MODES( IMODE )
*
* Generate a test matrix of size m by n using the
* singular value distribution indicated by MODE.
*
DO 10 I = 1, N
IWORK( N+I ) = 0
10 CONTINUE
CALL SLATMS( M, N, 'Uniform', ISEED, 'Nonsymm', TAU,
$ MODE, COND, DMAX, M, N, 'No packing', COPYA,
$ LDA, WORK, INFO )
*
* SGEQPF: QR factorization with column pivoting
*
CALL SLACPY( 'All', M, N, COPYA, LDA, A, LDA )
CALL ICOPY( N, IWORK( N+1 ), 1, IWORK, 1 )
IC = 0
S1 = SECOND( )
20 CONTINUE
CALL SGEQPF( M, N, A, LDA, IWORK, TAU, WORK, INFO )
S2 = SECOND( )
TIME = S2 - S1
IC = IC + 1
IF( TIME.LT.TIMMIN ) THEN
CALL SLACPY( 'All', M, N, COPYA, LDA, A, LDA )
CALL ICOPY( N, IWORK( N+1 ), 1, IWORK, 1 )
GO TO 20
END IF
*
* Subtract the time used in SLACPY and ICOPY.
*
ICL = 1
S1 = SECOND( )
30 CONTINUE
S2 = SECOND( )
UNTIME = S2 - S1
ICL = ICL + 1
IF( ICL.LE.IC ) THEN
CALL SLACPY( 'All', M, N, COPYA, LDA, A, LDA )
CALL ICOPY( N, IWORK( N+1 ), 1, IWORK, 1 )
GO TO 30
END IF
*
TIME = ( TIME-UNTIME ) / REAL( IC )
OPS = SOPLA( 'SGEQPF', M, N, 0, 0, 1 )
RESLTS( IMODE, IM, ILDA ) = SMFLOP( OPS, TIME, INFO )
*
40 CONTINUE
50 CONTINUE
60 CONTINUE
*
* Print tables of results
*
WRITE( NOUT, FMT = 9998 )SUBNAM( 1 )
IF( NLDA.GT.1 ) THEN
DO 70 I = 1, NLDA
WRITE( NOUT, FMT = 9997 )I, LDAVAL( I )
70 CONTINUE
END IF
WRITE( NOUT, FMT = * )
CALL SPRTB5( 'Type', 'M', 'N', NMODE, MODES, NM, MVAL, NVAL, NLDA,
$ RESLTS, LDR1, LDR2, NOUT )
80 CONTINUE
9999 FORMAT( 1X, A6, ' timing run not attempted', / )
9998 FORMAT( / ' *** Speed of ', A6, ' in megaflops ***' )
9997 FORMAT( 5X, 'line ', I2, ' with LDA = ', I5 )
RETURN
*
* End of STIMQP
*
END
|
