numeric-linalg

Educational material on the SciPy implementation of numerical linear algebra algorithms

NameSizeMode
..
lapack/BLAS/SRC/zrotg.f90 7809B -rw-r--r--
001
002
003
004
005
006
007
008
009
010
011
012
013
014
015
016
017
018
019
020
021
022
023
024
025
026
027
028
029
030
031
032
033
034
035
036
037
038
039
040
041
042
043
044
045
046
047
048
049
050
051
052
053
054
055
056
057
058
059
060
061
062
063
064
065
066
067
068
069
070
071
072
073
074
075
076
077
078
079
080
081
082
083
084
085
086
087
088
089
090
091
092
093
094
095
096
097
098
099
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
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
!> \brief \b ZROTG  generates a Givens rotation with real cosine and complex sine.
!
!  =========== DOCUMENTATION ===========
!
! Online html documentation available at
!            http://www.netlib.org/lapack/explore-html/
!
!> \par Purpose:
!  =============
!>
!> \verbatim
!>
!> ZROTG constructs a plane rotation
!>    [  c         s ] [ a ] = [ r ]
!>    [ -conjg(s)  c ] [ b ]   [ 0 ]
!> where c is real, s is complex, and c**2 + conjg(s)*s = 1.
!>
!> The computation uses the formulas
!>    |x| = sqrt( Re(x)**2 + Im(x)**2 )
!>    sgn(x) = x / |x|  if x /= 0
!>           = 1        if x  = 0
!>    c = |a| / sqrt(|a|**2 + |b|**2)
!>    s = sgn(a) * conjg(b) / sqrt(|a|**2 + |b|**2)
!>    r = sgn(a)*sqrt(|a|**2 + |b|**2)
!> When a and b are real and r /= 0, the formulas simplify to
!>    c = a / r
!>    s = b / r
!> the same as in DROTG when |a| > |b|.  When |b| >= |a|, the
!> sign of c and s will be different from those computed by DROTG
!> if the signs of a and b are not the same.
!>
!> \endverbatim
!>
!> @see lartg, @see lartgp
!
!  Arguments:
!  ==========
!
!> \param[in,out] A
!> \verbatim
!>          A is DOUBLE COMPLEX
!>          On entry, the scalar a.
!>          On exit, the scalar r.
!> \endverbatim
!>
!> \param[in] B
!> \verbatim
!>          B is DOUBLE COMPLEX
!>          The scalar b.
!> \endverbatim
!>
!> \param[out] C
!> \verbatim
!>          C is DOUBLE PRECISION
!>          The scalar c.
!> \endverbatim
!>
!> \param[out] S
!> \verbatim
!>          S is DOUBLE COMPLEX
!>          The scalar s.
!> \endverbatim
!
!  Authors:
!  ========
!
!> \author Weslley Pereira, University of Colorado Denver, USA
!
!> \date December 2021
!
!> \ingroup rotg
!
!> \par Further Details:
!  =====================
!>
!> \verbatim
!>
!> Based on the algorithm from
!>
!>  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
!>
!> \endverbatim
!
!  =====================================================================
subroutine ZROTG( a, b, c, s )
   integer, parameter :: wp = kind(1.d0)
!
!  -- Reference BLAS level1 routine --
!  -- Reference BLAS is a software package provided by Univ. of Tennessee,    --
!  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
!
!  .. Constants ..
   real(wp), parameter :: zero = 0.0_wp
   real(wp), parameter :: one  = 1.0_wp
   complex(wp), parameter :: czero  = 0.0_wp
!  ..
!  .. Scaling constants ..
   real(wp), parameter :: safmin = real(radix(0._wp),wp)**max( &
      minexponent(0._wp)-1, &
      1-maxexponent(0._wp) &
   )
   real(wp), parameter :: safmax = real(radix(0._wp),wp)**max( &
      1-minexponent(0._wp), &
      maxexponent(0._wp)-1 &
   )
   real(wp), parameter :: rtmin = sqrt( safmin )
!  ..
!  .. Scalar Arguments ..
   real(wp) :: c
   complex(wp) :: a, b, s
!  ..
!  .. Local Scalars ..
   real(wp) :: d, f1, f2, g1, g2, h2, u, v, w, rtmax
   complex(wp) :: f, fs, g, gs, r, t
!  ..
!  .. Intrinsic Functions ..
   intrinsic :: abs, aimag, conjg, max, min, real, sqrt
!  ..
!  .. Statement Functions ..
   real(wp) :: ABSSQ
!  ..
!  .. Statement Function definitions ..
   ABSSQ( t ) = real( t )**2 + aimag( t )**2
!  ..
!  .. Executable Statements ..
!
   f = a
   g = b
   if( g == czero ) then
      c = one
      s = czero
      r = f
   else if( f == czero ) then
      c = zero
      if( real(g) == zero ) then
         r = abs(aimag(g))
         s = conjg( g ) / r
      elseif( aimag(g) == zero ) then
         r = abs(real(g))
         s = conjg( g ) / r
      else
         g1 = max( abs(real(g)), abs(aimag(g)) )
         rtmax = sqrt( safmax/2 )
         if( g1 > rtmin .and. g1 < rtmax ) then
!
!        Use unscaled algorithm
!
!           The following two lines can be replaced by `d = abs( g )`.
!           This algorithm do not use the intrinsic complex abs.
            g2 = ABSSQ( g )
            d = sqrt( g2 )
            s = conjg( g ) / d
            r = d
         else
!
!        Use scaled algorithm
!
            u = min( safmax, max( safmin, g1 ) )
            gs = g / u
!           The following two lines can be replaced by `d = abs( gs )`.
!           This algorithm do not use the intrinsic complex abs.
            g2 = ABSSQ( gs )
            d = sqrt( g2 )
            s = conjg( gs ) / d
            r = d*u
         end if
      end if
   else
      f1 = max( abs(real(f)), abs(aimag(f)) )
      g1 = max( abs(real(g)), abs(aimag(g)) )
      rtmax = sqrt( safmax/4 )
      if( f1 > rtmin .and. f1 < rtmax .and. &
          g1 > rtmin .and. g1 < rtmax ) then
!
!        Use unscaled algorithm
!
         f2 = ABSSQ( f )
         g2 = ABSSQ( g )
         h2 = f2 + g2
         ! safmin <= f2 <= h2 <= safmax
         if( f2 >= h2 * safmin ) then
            ! safmin <= f2/h2 <= 1, and h2/f2 is finite
            c = sqrt( f2 / h2 )
            r = f / c
            rtmax = rtmax * 2
            if( f2 > rtmin .and. h2 < rtmax ) then
               ! safmin <= sqrt( f2*h2 ) <= safmax
               s = conjg( g ) * ( f / sqrt( f2*h2 ) )
            else
               s = conjg( g ) * ( r / h2 )
            end if
         else
            ! f2/h2 <= safmin may be subnormal, and h2/f2 may overflow.
            ! Moreover,
            !  safmin <= f2*f2 * safmax < f2 * h2 < h2*h2 * safmin <= safmax,
            !  sqrt(safmin) <= sqrt(f2 * h2) <= sqrt(safmax).
            ! Also,
            !  g2 >> f2, which means that h2 = g2.
            d = sqrt( f2 * h2 )
            c = f2 / d
            if( c >= safmin ) then
               r = f / c
            else
               ! f2 / sqrt(f2 * h2) < safmin, then
               !  sqrt(safmin) <= f2 * sqrt(safmax) <= h2 / sqrt(f2 * h2) <= h2 * (safmin / f2) <= h2 <= safmax
               r = f * ( h2 / d )
            end if
            s = conjg( g ) * ( f / d )
         end if
      else
!
!        Use scaled algorithm
!
         u = min( safmax, max( safmin, f1, g1 ) )
         gs = g / u
         g2 = ABSSQ( gs )
         if( f1 / u < rtmin ) then
!
!           f is not well-scaled when scaled by g1.
!           Use a different scaling for f.
!
            v = min( safmax, max( safmin, f1 ) )
            w = v / u
            fs = f / v
            f2 = ABSSQ( fs )
            h2 = f2*w**2 + g2
         else
!
!           Otherwise use the same scaling for f and g.
!
            w = one
            fs = f / u
            f2 = ABSSQ( fs )
            h2 = f2 + g2
         end if
         ! safmin <= f2 <= h2 <= safmax
         if( f2 >= h2 * safmin ) then
            ! safmin <= f2/h2 <= 1, and h2/f2 is finite
            c = sqrt( f2 / h2 )
            r = fs / c
            rtmax = rtmax * 2
            if( f2 > rtmin .and. h2 < rtmax ) then
               ! safmin <= sqrt( f2*h2 ) <= safmax
               s = conjg( gs ) * ( fs / sqrt( f2*h2 ) )
            else
               s = conjg( gs ) * ( r / h2 )
            end if
         else
            ! f2/h2 <= safmin may be subnormal, and h2/f2 may overflow.
            ! Moreover,
            !  safmin <= f2*f2 * safmax < f2 * h2 < h2*h2 * safmin <= safmax,
            !  sqrt(safmin) <= sqrt(f2 * h2) <= sqrt(safmax).
            ! Also,
            !  g2 >> f2, which means that h2 = g2.
            d = sqrt( f2 * h2 )
            c = f2 / d
            if( c >= safmin ) then
               r = fs / c
            else
               ! f2 / sqrt(f2 * h2) < safmin, then
               !  sqrt(safmin) <= f2 * sqrt(safmax) <= h2 / sqrt(f2 * h2) <= h2 * (safmin / f2) <= h2 <= safmax
               r = fs * ( h2 / d )
            end if
            s = conjg( gs ) * ( fs / d )
         end if
         ! Rescale c and r
         c = c * w
         r = r * u
      end if
   end if
   a = r
   return
end subroutine