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
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
|
!> \brief \b DLASSQ updates a sum of squares represented in scaled form.
!
! =========== DOCUMENTATION ===========
!
! Online html documentation available at
! http://www.netlib.org/lapack/explore-html/
!
!> \htmlonly
!> Download DLASSQ + dependencies
!> <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlassq.f90">
!> [TGZ]</a>
!> <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlassq.f90">
!> [ZIP]</a>
!> <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlassq.f90">
!> [TXT]</a>
!> \endhtmlonly
!
! Definition:
! ===========
!
! SUBROUTINE DLASSQ( N, X, INCX, SCALE, SUMSQ )
!
! .. Scalar Arguments ..
! INTEGER INCX, N
! DOUBLE PRECISION SCALE, SUMSQ
! ..
! .. Array Arguments ..
! DOUBLE PRECISION X( * )
! ..
!
!
!> \par Purpose:
! =============
!>
!> \verbatim
!>
!> DLASSQ returns the values scale_out and sumsq_out such that
!>
!> (scale_out**2)*sumsq_out = x( 1 )**2 +...+ x( n )**2 + (scale**2)*sumsq,
!>
!> where x( i ) = X( 1 + ( i - 1 )*INCX ). The value of sumsq is
!> assumed to be non-negative.
!>
!> scale and sumsq must be supplied in SCALE and SUMSQ and
!> scale_out and sumsq_out are overwritten on SCALE and SUMSQ respectively.
!>
!> \endverbatim
!
! Arguments:
! ==========
!
!> \param[in] N
!> \verbatim
!> N is INTEGER
!> The number of elements to be used from the vector x.
!> \endverbatim
!>
!> \param[in] X
!> \verbatim
!> X is DOUBLE PRECISION array, dimension (1+(N-1)*abs(INCX))
!> The vector for which a scaled sum of squares is computed.
!> x( i ) = X( 1 + ( i - 1 )*INCX ), 1 <= i <= n.
!> \endverbatim
!>
!> \param[in] INCX
!> \verbatim
!> INCX is INTEGER
!> The increment between successive values of the vector x.
!> If INCX > 0, X(1+(i-1)*INCX) = x(i) for 1 <= i <= n
!> If INCX < 0, X(1-(n-i)*INCX) = x(i) for 1 <= i <= n
!> If INCX = 0, x isn't a vector so there is no need to call
!> this subroutine. If you call it anyway, it will count x(1)
!> in the vector norm N times.
!> \endverbatim
!>
!> \param[in,out] SCALE
!> \verbatim
!> SCALE is DOUBLE PRECISION
!> On entry, the value scale in the equation above.
!> On exit, SCALE is overwritten by scale_out, the scaling factor
!> for the sum of squares.
!> \endverbatim
!>
!> \param[in,out] SUMSQ
!> \verbatim
!> SUMSQ is DOUBLE PRECISION
!> On entry, the value sumsq in the equation above.
!> On exit, SUMSQ is overwritten by sumsq_out, the basic sum of
!> squares from which scale_out has been factored out.
!> \endverbatim
!
! Authors:
! ========
!
!> \author Edward Anderson, Lockheed Martin
!
!> \par Contributors:
! ==================
!>
!> Weslley Pereira, University of Colorado Denver, USA
!> Nick Papior, Technical University of Denmark, DK
!
!> \par Further Details:
! =====================
!>
!> \verbatim
!>
!> Anderson E. (2017)
!> Algorithm 978: Safe Scaling in the Level 1 BLAS
!> ACM Trans Math Softw 44:1--28
!> https://doi.org/10.1145/3061665
!>
!> Blue, James L. (1978)
!> A Portable Fortran Program to Find the Euclidean Norm of a Vector
!> ACM Trans Math Softw 4:15--23
!> https://doi.org/10.1145/355769.355771
!>
!> \endverbatim
!
!> \ingroup lassq
!
! =====================================================================
subroutine DLASSQ( n, x, incx, scale, sumsq )
use LA_CONSTANTS, &
only: wp=>dp, zero=>dzero, one=>done, &
sbig=>dsbig, ssml=>dssml, tbig=>dtbig, tsml=>dtsml
use LA_XISNAN
!
! -- LAPACK auxiliary routine --
! -- LAPACK is a software package provided by Univ. of Tennessee, --
! -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
!
! .. Scalar Arguments ..
integer :: incx, n
real(wp) :: scale, sumsq
! ..
! .. Array Arguments ..
real(wp) :: x(*)
! ..
! .. Local Scalars ..
integer :: i, ix
logical :: notbig
real(wp) :: abig, amed, asml, ax, ymax, ymin
! ..
!
! Quick return if possible
!
if( LA_ISNAN(scale) .or. LA_ISNAN(sumsq) ) return
if( sumsq == zero ) scale = one
if( scale == zero ) then
scale = one
sumsq = zero
end if
if (n <= 0) then
return
end if
!
! Compute the sum of squares in 3 accumulators:
! abig -- sums of squares scaled down to avoid overflow
! asml -- sums of squares scaled up to avoid underflow
! amed -- sums of squares that do not require scaling
! The thresholds and multipliers are
! tbig -- values bigger than this are scaled down by sbig
! tsml -- values smaller than this are scaled up by ssml
!
notbig = .true.
asml = zero
amed = zero
abig = zero
ix = 1
if( incx < 0 ) ix = 1 - (n-1)*incx
do i = 1, n
ax = abs(x(ix))
if (ax > tbig) then
abig = abig + (ax*sbig)**2
notbig = .false.
else if (ax < tsml) then
if (notbig) asml = asml + (ax*ssml)**2
else
amed = amed + ax**2
end if
ix = ix + incx
end do
!
! Put the existing sum of squares into one of the accumulators
!
if( sumsq > zero ) then
ax = scale*sqrt( sumsq )
if (ax > tbig) then
if (scale > one) then
scale = scale * sbig
abig = abig + scale * (scale * sumsq)
else
! sumsq > tbig^2 => (sbig * (sbig * sumsq)) is representable
abig = abig + scale * (scale * (sbig * (sbig * sumsq)))
end if
else if (ax < tsml) then
if (notbig) then
if (scale < one) then
scale = scale * ssml
asml = asml + scale * (scale * sumsq)
else
! sumsq < tsml^2 => (ssml * (ssml * sumsq)) is representable
asml = asml + scale * (scale * (ssml * (ssml * sumsq)))
end if
end if
else
amed = amed + scale * (scale * sumsq)
end if
end if
!
! Combine abig and amed or amed and asml if more than one
! accumulator was used.
!
if (abig > zero) then
!
! Combine abig and amed if abig > 0.
!
if (amed > zero .or. LA_ISNAN(amed)) then
abig = abig + (amed*sbig)*sbig
end if
scale = one / sbig
sumsq = abig
else if (asml > zero) then
!
! Combine amed and asml if asml > 0.
!
if (amed > zero .or. LA_ISNAN(amed)) then
amed = sqrt(amed)
asml = sqrt(asml) / ssml
if (asml > amed) then
ymin = amed
ymax = asml
else
ymin = asml
ymax = amed
end if
scale = one
sumsq = ymax**2*( one + (ymin/ymax)**2 )
else
scale = one / ssml
sumsq = asml
end if
else
!
! Otherwise all values are mid-range or zero
!
scale = one
sumsq = amed
end if
return
end subroutine
|
