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
|
===============
Changes: 2.0.17
===============
.. rubric:: General:
- Added support for Windows NT/95 using the Microsoft Developers
Studio Visual C++. See the file 'Installation' for details.
- Other new machines on which this release has been tested: Cray
T3E, SGI Origin See the file petsc/Installation for a complete
list.
.. rubric:: TS (Timestepping Solvers):
- Modified the pseudo-transient continuation updates:
- Changed the default update to dt =
current_dt*previous_fnorm/current_fnorm.
- Added the routine TSPseudoIncrementDtFromInitialDt() and the
corresponding option -ts_pseudo_increment_dt_from_initial_dt to
indicate use of the alternative update formula dt =
initial_dt*initial_fnorm/current_fnorm.
- Changed the calling sequence of TSRegister().
.. rubric:: SNES (Nonlinear Solvers):
- Added support for computing large, sparse Jacobians efficiently
via finite differences, see Section 5.6 (Finite Difference
Jacobian Approximations) of the users manual.
- Added the routines SNESGetNumberLinearIterations() and
SNESSetConvergenceHistory(). See man pages for details.
- Activated a counter of function evaluations, which is used in
convergence tests to terminate solver if the number of function
evaluations exceeds a given tolerance. Note: Users of matrix-free
Newton-Krylov methods may need to reset the default allowable
maximum (1000), via SNESSetTolerances() or -snes_max_func
<maxfunc>.
- Changed the calling sequence of SNESRegister().
.. rubric:: SLES (Linear Solvers):
- See PC and KSP
.. rubric:: KSP (Krylov Subspace Methods):
- Changed the calling sequence of KSPRegister().
.. rubric:: PC (Preconditioners):
- Changed the calling sequence of PCRegister().
- New Additive Schwarz variants (preconditioner type PCASM)
- Added the routine PCASMSetType() (and the corresponding option
-pc_asm_type [basic,restrict,interpolate,none]) for setting the
variant of the additive Schwarz method. See the man page and users
manual for details.
- Changed the default variant of PCASM from full restriction and
interpolation to full restriction only, since this version
requires less communication and for many problems converges faster
than the basic variant that uses full restriction and
interpolation. Users can still employ the basic ASM by calling
PCASMSetType(pc,PC_ASM_BASIC) or by using the option -pc_asm_type
basic.
- Added an interface to the SPAI preconditioner implementation of
Steven Bernard; see src/contrib/spai. This has undergone little
testing and optimization; it is intended mainly for "hackers".
.. rubric:: MAT (Matrices):
- Added the matrix option,
MatSetOption(mat,MAT_NEW_NONZERO_LOCATION_ERROR), that will cause
an error if a new nonzero is generated in a sparse matrix.
(currently implemented for AIJ and BAIJ matrices only). This is a
useful flag when using SAME_NONZERO_PATTERN in calling
SLESSetOperators() to ensure that the nonzero pattern truly does
remain unchanged. For examples, see the programs
petsc/src/snes/examples/tutorials/[ex5.c,ex5f.F].
- Added the routine MatSetUnfactored(), intended primarily for use
with in-place ILU(0) factorization as a preconditioner for
matrix-free Krylov methods. See the manual page for details.
- Added the routines MatConvertRegisterAll() and
MatLoadRegisterAll() to allow the restriction of the matrix
routines linked into an application code. This can decrease the
size of your executable and the time it takes to link your
program. For details, see the manual page and
petsc/src/snes/examples/tutorials/ex5.c
- Added the routine MatSetValuesBlocked(), for more efficient
assembly of block AIJ formatted matrices (MATSEQBAIJ and
MATMPIBAIJ).
- Changed the calling sequence of MatReorderingRegister();
.. rubric:: DA (Distributed Arrays):
- Added additional arguments to DACreate1d(), DACreate2d(), and
DACreate3d() to allow the user to set the distribution of nodes on
each processor; set these arguments to PETSC_NULL for the standard
default distribution.
- Modified DAGetInfo() to return the type of periodicity.
.. rubric:: VEC (Vectors):
- Added the routine VecCreateGhost() to create vectors that have
ghost padding at the end of the local array. This is useful for
gathering remote values to perform local calculations that involve
off-processor ghost values. This is often appropriate for codes
using unstructured grids. See
petsc/src/vec/examples/tutorials/ex9.c for possible usage.
.. rubric:: IS (Index Sets):
.. rubric:: Draw (Graphics):
- Application codes should not need to use #include "draw.h" anymore
from C/C++, since this file is now included automatically when
"petsc.h" or any other PETSc include file is included.
.. rubric:: Viewers:
- VIEWER_DRAWX_WORLD, VIEWER_DRAWX_SELF, VIEWER_MATLAB_WORLD are now
supported from Fortran.
- Added VIEWER_DRAWX_(MPI_Comm comm) from C. Useful for rapid code
prototyping without having to declare a Viewer.
.. rubric:: System Routines:
- Since memory leaks and uninitialized memory can be serious
problems for large-scale application codes, we've added several
new tools to assist in their diagnosis. These tools are all work
in conjunction with the PETSc memory allocation (the default for
codes that are compiled in debug mode with
BOPT=[g,g_c++,g_complex]).
- Added the runtime option -trmalloc_log, which activates logging of
all calls to malloc via the new routines PetscTrLog() and
PetscTrLogDump().
- Added the routine PetscGetResidentSetSize() to determine the total
memory used by a process (this is activated by -trmalloc_log); see
the man page for details.
- Added the option -trmalloc_nan for tracking down allocated memory
that is used before it has been initialized. This option calls the
new routines PetscInitializeNans() and PetscInitializeLargeInts().
So far these work on the Sun4 system.
.. rubric:: Error Handling:
- The error checking macros SETERRQ() and SETERRA() now have the
calling sequence SETERRQ(int ierr,int pierr,char \*message); where
pierr is an additional integer error code passed to the error
handler. Currently you should just set pierr=1.
- Also, SETERRQ() and SETERRA() now use the macro \__FUNC_\_ to keep
track of routine names. Users need not worry about this in their
application codes, but can take advantage of this feature if
desired by setting this macro before each user-defined routine
that may call SETERRQ(), SETERRA(), CHKERRQ(), or CHKERRA().
\__FUNC_\_ should be set to a string containing the routine name.
For example, #undef \__FUNC_\_ #define \__FUNC_\_ "MyRoutine1" int
MyRoutine1() { /\* code here \*/ return 0; } See
petsc/src/snes/examples/tutorials/ex3.c for an example.
- PETSc error handlers now take two additional arguments. Consult
the man page for PetscPushErrorHandler() for more information.
.. rubric:: Event Logging:
- Changed PLogPrintSummary(MPI_Comm,FILE \*) to
PLogPrintSummary(MPI_Comm,char \*).
- Now the option -log_summary takes [filename] as an optional
argument.
.. rubric:: Fortran Interface:
- Added some limited support for direct use of Fortran90 pointers in
the routines Vec[Get,Restore]ArrayF90(),
Mat[Get,Restore]ArrayF90(), IS[Get,Restore]IndicesF90(),
ISBlock[Get,Restore]IndicesF90(), VecDuplicateVecsF90(),
VecDestroyVecsF90(), DAGetGlobalIndicesF90(). See the man pages
and the section 'Fortran90' in the users manal for details.
Unfortunately, these routines currently work only with the NAG F90
compiler. We hope to support other compilers as well, but we will
need assistance from the vendors since the Fortran90/C interface
is not a defined standard.
- Added the macro PetscDoubleExp(a,b) = a d b (machines where double
precision arithmetic is used) = a e b (machines where single
precision arithmetic is used, e.g., Crays) This macro is intended
for use only if you wish to maintain a Fortran code that is
portable to both the Cray T3d/T3e and other Unix machines.
- For mixed Fortran/C users: added the makefile flag FCONF that may
be used in place of the flag CONF. For an example of usage, see
src/vec/examples/tutorials/makefile
|
