xref: /phasta/phSolver/compressible/au2mfg.f (revision 595995161822a203c8467e0e4a253d7bd7d6df32)
1        subroutine Au2MFG (ypre,      y,    ac,     x,
2     &                     rmes,      res,       Dy,
3     &                     uBrg,      BDiag,     iBC,       BC,
4     &                     iper,      ilwork,
5     &                     shp,       shgl,
6     &                     shpb,      shglb)
7c
8c----------------------------------------------------------------------
9c
10c This routine performs a matrix-vector product for the Matrix-Free
11c  Implicit/Iterative solver using a two-sided scheme and subtracts
12c  it from the residual.
13c
14c input:
15c  y      (nshg,ndof)           : Y-variables
16c  ypre   (nshg,ndof)           : preconditioned Y-variables
17c  x      (numnp,nsd)            : node coordinates
18c  rmes   (nshg,nflow)           : modified residual
19c  res    (nshg,nflow)           : residual
20c  Dy     (nshg,nflow)           : solution step
21c  BDiag  (nshg,nflow,nflow)         : block-diagonal preconditioner
22c  iBC    (nshg)                : BC codes
23c  BC     (nshg,ndofBC)         : BC constraint parameters
24c
25c output:
26c  uBrg   (nshg,nflow)           : Krylov vector ( R - A x )
27c
28c
29c Zdenek Johan, Winter 1991.  (Fortran 90)
30c----------------------------------------------------------------------
31c
32        include "common.h"
33c
34        dimension y(nshg,ndof),               ypre(nshg,nflow),
35     &            x(numnp,nsd),               ac(nshg,ndof),
36     &            rmes(nshg,nflow),            ytmp(nshg,nflow),
37     &            res(nshg,nflow),             Dy(nshg,nflow),
38     &            uBrg(nshg,nflow),            BDiag(nshg,nflow,nflow),
39     &            iBC(nshg),                  BC(nshg,ndofBC),
40     &            iper(nshg),                 Dy2(nshg,nflow)
41c
42        dimension uBtmp1(nshg,nflow),          uBtmp2(nshg,nflow),
43     &            tmpBC(nshg),                ilwork(nlwork)
44c
45        dimension shp(MAXTOP,maxsh,MAXQPT),
46     &            shgl(MAXTOP,nsd,maxsh,MAXQPT),
47     &            shpb(MAXTOP,maxsh,MAXQPT),
48     &            shglb(MAXTOP,nsd,maxsh,MAXQPT)
49
50c$$$        dimension shp(nshl,ngauss),    shgl(nsd,nshl,ngauss),
51c$$$     &            shpb(nshl,ngaussb), shglb(nsd,nshl,ngaussb)
52c
53c.... compute the finite difference interval
54c
55        Dy2 = Dy**2
56	call sumgat (Dy2, nflow, summed, ilwork)
57        eps = epsM**pt66 / sqrt(summed)
58c
59c.... calculate R(y + eps x)
60c
61        uBtmp1 = zero
62c
63c        call yshuffle(ypre, 'new2old ')
64c
65        uBrg = ypre + eps * Dy
66c
67        call i3LU (BDiag,  uBrg,  'backward')
68c
69        call yshuffle(uBrg, 'old2new ')
70c
71        call itrBC (uBrg, uBrg, iBC,  BC,   iper, ilwork)
72c
73        call itrRes (uBrg,            y,
74     &               x,               shp,
75     &               shgl,            iBC,
76     &               BC,              shpb,
77     &               shglb,           uBtmp1,
78     &               iper,            ilwork)
79c
80        call i3LU (BDiag,  uBtmp1,  'forward ')
81c
82c.... calculate R(y - eps x)
83c
84        uBtmp2 = zero
85c
86c        call yshuffle(ypre, 'new2old ')
87c
88        uBrg = ypre - eps * Dy
89c
90        call i3LU (BDiag,  uBrg,  'backward')
91c
92        call yshuffle(uBrg, 'old2new ')
93
94        call itrBC (uBrg, uBrg, iBC,  BC,   iper, ilwork)
95c
96        call itrRes (uBrg,            y,
97     &               x,               shp,
98     &               shgl,            iBC,
99     &               BC,              shpb,
100     &               shglb,           uBtmp2,
101     &               iper,            ilwork, ac)
102c
103        call i3LU (BDiag,  uBtmp2,  'forward ')
104c
105c.... compute  R(y) - ( R(y + eps x) - R(y - eps x) ) / 2 eps
106c
107        uBrg = res - ( uBtmp1 - uBtmp2 ) / (two * eps)
108c
109c.... flop count
110c
111        flops = flops + 8*nflow*nshg+nshg
112c
113c.... end
114c
115        return
116        end
117