OpenCores
URL https://opencores.org/ocsvn/openrisc/openrisc/trunk

Subversion Repositories openrisc

[/] [openrisc/] [trunk/] [gnu-stable/] [gcc-4.5.1/] [gcc/] [ada/] [s-gerela.adb] - Blame information for rev 826

Details | Compare with Previous | View Log

Line No. Rev Author Line
1 281 jeremybenn
------------------------------------------------------------------------------
2
--                                                                          --
3
--                         GNAT RUN-TIME COMPONENTS                         --
4
--                                                                          --
5
--                        SYSTEM.GENERIC_REAL_LAPACK                        --
6
--                                                                          --
7
--                                 B o d y                                  --
8
--                                                                          --
9
--            Copyright (C) 2009, Free Software Foundation, Inc.            --
10
--                                                                          --
11
-- GNAT is free software;  you can  redistribute it  and/or modify it under --
12
-- terms of the  GNU General Public License as published  by the Free Soft- --
13
-- ware  Foundation;  either version 3,  or (at your option) any later ver- --
14
-- sion.  GNAT is distributed in the hope that it will be useful, but WITH- --
15
-- OUT ANY WARRANTY;  without even the  implied warranty of MERCHANTABILITY --
16
-- or FITNESS FOR A PARTICULAR PURPOSE.                                     --
17
--                                                                          --
18
-- As a special exception under Section 7 of GPL version 3, you are granted --
19
-- additional permissions described in the GCC Runtime Library Exception,   --
20
-- version 3.1, as published by the Free Software Foundation.               --
21
--                                                                          --
22
-- You should have received a copy of the GNU General Public License and    --
23
-- a copy of the GCC Runtime Library Exception along with this program;     --
24
-- see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see    --
25
-- <http://www.gnu.org/licenses/>.                                          --
26
--                                                                          --
27
-- GNAT was originally developed  by the GNAT team at  New York University. --
28
-- Extensive contributions were provided by Ada Core Technologies Inc.      --
29
--                                                                          --
30
------------------------------------------------------------------------------
31
 
32
with Ada.Unchecked_Conversion;        use Ada;
33
with Interfaces;                      use Interfaces;
34
with Interfaces.Fortran;              use Interfaces.Fortran;
35
with Interfaces.Fortran.BLAS;         use Interfaces.Fortran.BLAS;
36
with Interfaces.Fortran.LAPACK;       use Interfaces.Fortran.LAPACK;
37
with System.Generic_Array_Operations; use System.Generic_Array_Operations;
38
 
39
package body System.Generic_Real_LAPACK is
40
 
41
   Is_Real : constant Boolean :=
42
               Real'Machine_Mantissa = Fortran.Real'Machine_Mantissa
43
                and then Fortran.Real (Real'First) = Fortran.Real'First
44
                and then Fortran.Real (Real'Last) = Fortran.Real'Last;
45
 
46
   Is_Double_Precision : constant Boolean :=
47
                           Real'Machine_Mantissa =
48
                                            Double_Precision'Machine_Mantissa
49
                            and then
50
                              Double_Precision (Real'First) =
51
                                            Double_Precision'First
52
                            and then
53
                              Double_Precision (Real'Last) =
54
                                            Double_Precision'Last;
55
 
56
   --  Local subprograms
57
 
58
   function To_Double_Precision (X : Real) return Double_Precision;
59
   pragma Inline_Always (To_Double_Precision);
60
 
61
   function To_Real (X : Double_Precision) return Real;
62
   pragma Inline_Always (To_Real);
63
 
64
   --  Instantiations
65
 
66
   function To_Double_Precision is new
67
     Vector_Elementwise_Operation
68
       (X_Scalar      => Real,
69
        Result_Scalar => Double_Precision,
70
        X_Vector      => Real_Vector,
71
        Result_Vector => Double_Precision_Vector,
72
        Operation     => To_Double_Precision);
73
 
74
   function To_Real is new
75
     Vector_Elementwise_Operation
76
       (X_Scalar      => Double_Precision,
77
        Result_Scalar => Real,
78
        X_Vector      => Double_Precision_Vector,
79
        Result_Vector => Real_Vector,
80
        Operation     => To_Real);
81
 
82
   function To_Double_Precision is new
83
     Matrix_Elementwise_Operation
84
       (X_Scalar      => Real,
85
        Result_Scalar => Double_Precision,
86
        X_Matrix      => Real_Matrix,
87
        Result_Matrix => Double_Precision_Matrix,
88
        Operation     => To_Double_Precision);
89
 
90
   function To_Real is new
91
     Matrix_Elementwise_Operation
92
       (X_Scalar      => Double_Precision,
93
        Result_Scalar => Real,
94
        X_Matrix      => Double_Precision_Matrix,
95
        Result_Matrix => Real_Matrix,
96
        Operation     => To_Real);
97
 
98
   function To_Double_Precision (X : Real) return Double_Precision is
99
   begin
100
      return Double_Precision (X);
101
   end To_Double_Precision;
102
 
103
   function To_Real (X : Double_Precision) return Real is
104
   begin
105
      return Real (X);
106
   end To_Real;
107
 
108
   -----------
109
   -- getrf --
110
   -----------
111
 
112
   procedure getrf
113
     (M     : Natural;
114
      N     : Natural;
115
      A     : in out Real_Matrix;
116
      Ld_A  : Positive;
117
      I_Piv : out Integer_Vector;
118
      Info  : access Integer)
119
   is
120
   begin
121
      if Is_Real then
122
         declare
123
            type A_Ptr is
124
               access all BLAS.Real_Matrix (A'Range (1), A'Range (2));
125
            function Conv_A is new Unchecked_Conversion (Address, A_Ptr);
126
         begin
127
            sgetrf (M, N, Conv_A (A'Address).all, Ld_A,
128
                    LAPACK.Integer_Vector (I_Piv), Info);
129
         end;
130
 
131
      elsif Is_Double_Precision then
132
         declare
133
            type A_Ptr is
134
               access all Double_Precision_Matrix (A'Range (1), A'Range (2));
135
            function Conv_A is new Unchecked_Conversion (Address, A_Ptr);
136
         begin
137
            dgetrf (M, N, Conv_A (A'Address).all, Ld_A,
138
                    LAPACK.Integer_Vector (I_Piv), Info);
139
         end;
140
 
141
      else
142
         declare
143
            DP_A : Double_Precision_Matrix (A'Range (1), A'Range (2));
144
         begin
145
            DP_A := To_Double_Precision (A);
146
            dgetrf (M, N, DP_A, Ld_A, LAPACK.Integer_Vector (I_Piv), Info);
147
            A := To_Real (DP_A);
148
         end;
149
      end if;
150
   end getrf;
151
 
152
   -----------
153
   -- getri --
154
   -----------
155
 
156
   procedure getri
157
     (N      : Natural;
158
      A      : in out Real_Matrix;
159
      Ld_A   : Positive;
160
      I_Piv  : Integer_Vector;
161
      Work   : in out Real_Vector;
162
      L_Work : Integer;
163
      Info   : access Integer)
164
   is
165
   begin
166
      if Is_Real then
167
         declare
168
            type A_Ptr is
169
               access all BLAS.Real_Matrix (A'Range (1), A'Range (2));
170
            type Work_Ptr is
171
               access all BLAS.Real_Vector (Work'Range);
172
            function Conv_A is new Unchecked_Conversion (Address, A_Ptr);
173
            function Conv_Work is new Unchecked_Conversion (Address, Work_Ptr);
174
         begin
175
            sgetri (N, Conv_A (A'Address).all, Ld_A,
176
                    LAPACK.Integer_Vector (I_Piv),
177
                    Conv_Work (Work'Address).all, L_Work,
178
                    Info);
179
         end;
180
 
181
      elsif Is_Double_Precision then
182
         declare
183
            type A_Ptr is
184
               access all Double_Precision_Matrix (A'Range (1), A'Range (2));
185
            type Work_Ptr is
186
               access all Double_Precision_Vector (Work'Range);
187
            function Conv_A is new Unchecked_Conversion (Address, A_Ptr);
188
            function Conv_Work is new Unchecked_Conversion (Address, Work_Ptr);
189
         begin
190
            dgetri (N, Conv_A (A'Address).all, Ld_A,
191
                    LAPACK.Integer_Vector (I_Piv),
192
                    Conv_Work (Work'Address).all, L_Work,
193
                    Info);
194
         end;
195
 
196
      else
197
         declare
198
            DP_A : Double_Precision_Matrix (A'Range (1), A'Range (2));
199
            DP_Work : Double_Precision_Vector (Work'Range);
200
         begin
201
            DP_A := To_Double_Precision (A);
202
            dgetri (N, DP_A, Ld_A, LAPACK.Integer_Vector (I_Piv),
203
                    DP_Work, L_Work, Info);
204
            A := To_Real (DP_A);
205
            Work (1) := To_Real (DP_Work (1));
206
         end;
207
      end if;
208
   end getri;
209
 
210
   -----------
211
   -- getrs --
212
   -----------
213
 
214
   procedure getrs
215
     (Trans  : access constant Character;
216
      N      : Natural;
217
      N_Rhs  : Natural;
218
      A      : Real_Matrix;
219
      Ld_A   : Positive;
220
      I_Piv  : Integer_Vector;
221
      B      : in out Real_Matrix;
222
      Ld_B   : Positive;
223
      Info   : access Integer)
224
   is
225
   begin
226
      if Is_Real then
227
         declare
228
            subtype A_Type is BLAS.Real_Matrix (A'Range (1), A'Range (2));
229
            type B_Ptr is
230
               access all BLAS.Real_Matrix (B'Range (1), B'Range (2));
231
            function Conv_A is new Unchecked_Conversion (Real_Matrix, A_Type);
232
            function Conv_B is new Unchecked_Conversion (Address, B_Ptr);
233
         begin
234
            sgetrs (Trans, N, N_Rhs,
235
                    Conv_A (A), Ld_A,
236
                    LAPACK.Integer_Vector (I_Piv),
237
                    Conv_B (B'Address).all, Ld_B,
238
                    Info);
239
         end;
240
 
241
      elsif Is_Double_Precision then
242
         declare
243
            subtype A_Type is
244
               Double_Precision_Matrix (A'Range (1), A'Range (2));
245
            type B_Ptr is
246
               access all Double_Precision_Matrix (B'Range (1), B'Range (2));
247
            function Conv_A is new Unchecked_Conversion (Real_Matrix, A_Type);
248
            function Conv_B is new Unchecked_Conversion (Address, B_Ptr);
249
         begin
250
            dgetrs (Trans, N, N_Rhs,
251
                    Conv_A (A), Ld_A,
252
                    LAPACK.Integer_Vector (I_Piv),
253
                    Conv_B (B'Address).all, Ld_B,
254
                    Info);
255
         end;
256
 
257
      else
258
         declare
259
            DP_A : Double_Precision_Matrix (A'Range (1), A'Range (2));
260
            DP_B : Double_Precision_Matrix (B'Range (1), B'Range (2));
261
         begin
262
            DP_A := To_Double_Precision (A);
263
            DP_B := To_Double_Precision (B);
264
            dgetrs (Trans, N, N_Rhs,
265
                    DP_A, Ld_A,
266
                    LAPACK.Integer_Vector (I_Piv),
267
                    DP_B, Ld_B,
268
                    Info);
269
            B := To_Real (DP_B);
270
         end;
271
      end if;
272
   end getrs;
273
 
274
   -----------
275
   -- orgtr --
276
   -----------
277
 
278
   procedure orgtr
279
     (Uplo   : access constant Character;
280
      N      : Natural;
281
      A      : in out Real_Matrix;
282
      Ld_A   : Positive;
283
      Tau    : Real_Vector;
284
      Work   : out Real_Vector;
285
      L_Work : Integer;
286
      Info   : access Integer)
287
   is
288
   begin
289
      if Is_Real then
290
         declare
291
            type A_Ptr is
292
               access all BLAS.Real_Matrix (A'Range (1), A'Range (2));
293
            subtype Tau_Type is BLAS.Real_Vector (Tau'Range);
294
            type Work_Ptr is
295
               access all BLAS.Real_Vector (Work'Range);
296
            function Conv_A is new Unchecked_Conversion (Address, A_Ptr);
297
            function Conv_Tau is
298
               new Unchecked_Conversion (Real_Vector, Tau_Type);
299
            function Conv_Work is new Unchecked_Conversion (Address, Work_Ptr);
300
         begin
301
            sorgtr (Uplo, N,
302
                    Conv_A (A'Address).all, Ld_A,
303
                    Conv_Tau (Tau),
304
                    Conv_Work (Work'Address).all, L_Work,
305
                    Info);
306
         end;
307
 
308
      elsif Is_Double_Precision then
309
         declare
310
            type A_Ptr is
311
               access all Double_Precision_Matrix (A'Range (1), A'Range (2));
312
            subtype Tau_Type is Double_Precision_Vector (Tau'Range);
313
            type Work_Ptr is
314
               access all Double_Precision_Vector (Work'Range);
315
            function Conv_A is new Unchecked_Conversion (Address, A_Ptr);
316
            function Conv_Tau is
317
               new Unchecked_Conversion (Real_Vector, Tau_Type);
318
            function Conv_Work is new Unchecked_Conversion (Address, Work_Ptr);
319
         begin
320
            dorgtr (Uplo, N,
321
                    Conv_A (A'Address).all, Ld_A,
322
                    Conv_Tau (Tau),
323
                    Conv_Work (Work'Address).all, L_Work,
324
                    Info);
325
         end;
326
 
327
      else
328
         declare
329
            DP_A : Double_Precision_Matrix (A'Range (1), A'Range (2));
330
            DP_Work : Double_Precision_Vector (Work'Range);
331
            DP_Tau  : Double_Precision_Vector (Tau'Range);
332
         begin
333
            DP_A := To_Double_Precision (A);
334
            DP_Tau := To_Double_Precision (Tau);
335
            dorgtr (Uplo, N, DP_A, Ld_A, DP_Tau, DP_Work, L_Work, Info);
336
            A := To_Real (DP_A);
337
            Work (1) := To_Real (DP_Work (1));
338
         end;
339
      end if;
340
   end orgtr;
341
 
342
   -----------
343
   -- steqr --
344
   -----------
345
 
346
   procedure steqr
347
     (Comp_Z : access constant Character;
348
      N      : Natural;
349
      D      : in out Real_Vector;
350
      E      : in out Real_Vector;
351
      Z      : in out Real_Matrix;
352
      Ld_Z   : Positive;
353
      Work   : out Real_Vector;
354
      Info   : access Integer)
355
   is
356
   begin
357
      if Is_Real then
358
         declare
359
            type D_Ptr is access all BLAS.Real_Vector (D'Range);
360
            type E_Ptr is access all BLAS.Real_Vector (E'Range);
361
            type Z_Ptr is
362
               access all BLAS.Real_Matrix (Z'Range (1), Z'Range (2));
363
            type Work_Ptr is
364
               access all BLAS.Real_Vector (Work'Range);
365
            function Conv_D is new Unchecked_Conversion (Address, D_Ptr);
366
            function Conv_E is new Unchecked_Conversion (Address, E_Ptr);
367
            function Conv_Z is new Unchecked_Conversion (Address, Z_Ptr);
368
            function Conv_Work is new Unchecked_Conversion (Address, Work_Ptr);
369
         begin
370
            ssteqr (Comp_Z, N,
371
                    Conv_D (D'Address).all,
372
                    Conv_E (E'Address).all,
373
                    Conv_Z (Z'Address).all,
374
                    Ld_Z,
375
                    Conv_Work (Work'Address).all,
376
                    Info);
377
         end;
378
 
379
      elsif Is_Double_Precision then
380
         declare
381
            type D_Ptr is access all Double_Precision_Vector (D'Range);
382
            type E_Ptr is access all Double_Precision_Vector (E'Range);
383
            type Z_Ptr is
384
               access all Double_Precision_Matrix (Z'Range (1), Z'Range (2));
385
            type Work_Ptr is
386
               access all Double_Precision_Vector (Work'Range);
387
            function Conv_D is new Unchecked_Conversion (Address, D_Ptr);
388
            function Conv_E is new Unchecked_Conversion (Address, E_Ptr);
389
            function Conv_Z is new Unchecked_Conversion (Address, Z_Ptr);
390
            function Conv_Work is new Unchecked_Conversion (Address, Work_Ptr);
391
         begin
392
            dsteqr (Comp_Z, N,
393
                    Conv_D (D'Address).all,
394
                    Conv_E (E'Address).all,
395
                    Conv_Z (Z'Address).all,
396
                    Ld_Z,
397
                    Conv_Work (Work'Address).all,
398
                    Info);
399
         end;
400
 
401
      else
402
         declare
403
            DP_D    : Double_Precision_Vector (D'Range);
404
            DP_E    : Double_Precision_Vector (E'Range);
405
            DP_Z    : Double_Precision_Matrix (Z'Range (1), Z'Range (2));
406
            DP_Work : Double_Precision_Vector (Work'Range);
407
         begin
408
            DP_D := To_Double_Precision (D);
409
            DP_E := To_Double_Precision (E);
410
 
411
            if Comp_Z.all = 'V' then
412
               DP_Z := To_Double_Precision (Z);
413
            end if;
414
 
415
            dsteqr (Comp_Z, N, DP_D, DP_E, DP_Z, Ld_Z, DP_Work, Info);
416
 
417
            D := To_Real (DP_D);
418
            E := To_Real (DP_E);
419
            Z := To_Real (DP_Z);
420
         end;
421
      end if;
422
   end steqr;
423
 
424
   -----------
425
   -- sterf --
426
   -----------
427
 
428
   procedure sterf
429
     (N    : Natural;
430
      D    : in out Real_Vector;
431
      E    : in out Real_Vector;
432
      Info : access Integer)
433
   is
434
   begin
435
      if Is_Real then
436
         declare
437
            type D_Ptr is access all BLAS.Real_Vector (D'Range);
438
            type E_Ptr is access all BLAS.Real_Vector (E'Range);
439
            function Conv_D is new Unchecked_Conversion (Address, D_Ptr);
440
            function Conv_E is new Unchecked_Conversion (Address, E_Ptr);
441
         begin
442
            ssterf (N, Conv_D (D'Address).all, Conv_E (E'Address).all, Info);
443
         end;
444
 
445
      elsif Is_Double_Precision then
446
         declare
447
            type D_Ptr is access all Double_Precision_Vector (D'Range);
448
            type E_Ptr is access all Double_Precision_Vector (E'Range);
449
            function Conv_D is new Unchecked_Conversion (Address, D_Ptr);
450
            function Conv_E is new Unchecked_Conversion (Address, E_Ptr);
451
         begin
452
            dsterf (N, Conv_D (D'Address).all, Conv_E (E'Address).all, Info);
453
         end;
454
 
455
      else
456
         declare
457
            DP_D    : Double_Precision_Vector (D'Range);
458
            DP_E    : Double_Precision_Vector (E'Range);
459
 
460
         begin
461
            DP_D := To_Double_Precision (D);
462
            DP_E := To_Double_Precision (E);
463
 
464
            dsterf (N, DP_D, DP_E, Info);
465
 
466
            D := To_Real (DP_D);
467
            E := To_Real (DP_E);
468
         end;
469
      end if;
470
   end sterf;
471
 
472
   -----------
473
   -- sytrd --
474
   -----------
475
 
476
   procedure sytrd
477
     (Uplo   : access constant Character;
478
      N      : Natural;
479
      A      : in out Real_Matrix;
480
      Ld_A   : Positive;
481
      D      : out Real_Vector;
482
      E      : out Real_Vector;
483
      Tau    : out Real_Vector;
484
      Work   : out Real_Vector;
485
      L_Work : Integer;
486
      Info   : access Integer)
487
   is
488
   begin
489
      if Is_Real then
490
         declare
491
            type A_Ptr is
492
               access all BLAS.Real_Matrix (A'Range (1), A'Range (2));
493
            type D_Ptr is access all BLAS.Real_Vector (D'Range);
494
            type E_Ptr is access all BLAS.Real_Vector (E'Range);
495
            type Tau_Ptr is access all BLAS.Real_Vector (Tau'Range);
496
            type Work_Ptr is
497
               access all BLAS.Real_Vector (Work'Range);
498
            function Conv_A is new Unchecked_Conversion (Address, A_Ptr);
499
            function Conv_D is new Unchecked_Conversion (Address, D_Ptr);
500
            function Conv_E is new Unchecked_Conversion (Address, E_Ptr);
501
            function Conv_Tau is new Unchecked_Conversion (Address, Tau_Ptr);
502
            function Conv_Work is new Unchecked_Conversion (Address, Work_Ptr);
503
         begin
504
            ssytrd (Uplo, N,
505
                    Conv_A (A'Address).all, Ld_A,
506
                    Conv_D (D'Address).all,
507
                    Conv_E (E'Address).all,
508
                    Conv_Tau (Tau'Address).all,
509
                    Conv_Work (Work'Address).all,
510
                    L_Work,
511
                    Info);
512
         end;
513
 
514
      elsif Is_Double_Precision then
515
         declare
516
            type A_Ptr is
517
               access all Double_Precision_Matrix (A'Range (1), A'Range (2));
518
            type D_Ptr is access all Double_Precision_Vector (D'Range);
519
            type E_Ptr is access all Double_Precision_Vector (E'Range);
520
            type Tau_Ptr is access all Double_Precision_Vector (Tau'Range);
521
            type Work_Ptr is
522
               access all Double_Precision_Vector (Work'Range);
523
            function Conv_A is new Unchecked_Conversion (Address, A_Ptr);
524
            function Conv_D is new Unchecked_Conversion (Address, D_Ptr);
525
            function Conv_E is new Unchecked_Conversion (Address, E_Ptr);
526
            function Conv_Tau is new Unchecked_Conversion (Address, Tau_Ptr);
527
            function Conv_Work is new Unchecked_Conversion (Address, Work_Ptr);
528
         begin
529
            dsytrd (Uplo, N,
530
                    Conv_A (A'Address).all, Ld_A,
531
                    Conv_D (D'Address).all,
532
                    Conv_E (E'Address).all,
533
                    Conv_Tau (Tau'Address).all,
534
                    Conv_Work (Work'Address).all,
535
                    L_Work,
536
                    Info);
537
         end;
538
 
539
      else
540
         declare
541
            DP_A    : Double_Precision_Matrix (A'Range (1), A'Range (2));
542
            DP_D    : Double_Precision_Vector (D'Range);
543
            DP_E    : Double_Precision_Vector (E'Range);
544
            DP_Tau  : Double_Precision_Vector (Tau'Range);
545
            DP_Work : Double_Precision_Vector (Work'Range);
546
         begin
547
            DP_A := To_Double_Precision (A);
548
 
549
            dsytrd (Uplo, N, DP_A, Ld_A, DP_D, DP_E, DP_Tau,
550
                    DP_Work, L_Work, Info);
551
 
552
            if L_Work /= -1 then
553
               A := To_Real (DP_A);
554
               D := To_Real (DP_D);
555
               E := To_Real (DP_E);
556
               Tau := To_Real (DP_Tau);
557
            end if;
558
 
559
            Work (1) := To_Real (DP_Work (1));
560
         end;
561
      end if;
562
   end sytrd;
563
 
564
end System.Generic_Real_LAPACK;

powered by: WebSVN 2.1.0

© copyright 1999-2024 OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.