Barry and Jose,
I have gotten my compile issues down to the following:
usolve.F:84:46:
84 | MatInfo info(MAT_INFO_SIZE)
| 1
Error: Symbol 'mat_info_size' at (1) has no IMPLICIT type; did you
mean 'mpi_win_size'?
I take on item back...I was a failure at using the debugger. Here is
the backtrace. MatCreate seems to have valid data :/
* thread #1, queue = 'com.apple.main-thread', stop reason = signal
SIGSTOP
* frame #0: 0x7fff69d92746
libsystem_kernel.dylib`__semwait_signal + 10
Problem resolved. MatInfo is now a derived type with % access to the
components
MatInfo info
info%nz_allocated ! defined in ftn/mat/petscmat.h
etc.
-sanjay
-
On 3/23/25 3:22 PM, Sanjay Govindjee wrote:
Barry and Jose,
I have gotten my compile issues down to the following:
u
Jose,
Using the pointer construction to set up the arrays has resolved the
issue with finding the preallocation interfaces.
I'll look at the other issues tomorrow. Thanks for the help.
-sanjay
-
On 3/23/25 1:43 AM, Sanjay Govindjee wrote:
Complicated history of a 40+ year old code. But in s
Complicated history of a 40+ year old code. But in short
mr( np(246) ) is the first location of an allocated block of memory.
If need be I can explicitly set up a pointer to the array, something like
integer, pointer :: arname( : )
arname(1:arrlen) => mr(np(246):np(246)+arrlen-1)
I'
What is mr(np(246))? It should be an array, not a single entry of an array. As
indicated in the list of changes, in some cases you can use the syntax [z]
instead of z to represent an array of single value z.
> El 23 mar 2025, a las 9:24, Sanjay Govindjee escribió:
>
> Jose,
>I've tried lo
Jose,
I've tried lots of combinations but I still get the same error. I
think the signatures are all correct. I've attached the routine in case
you see something obvious. If not I will try to make a standalone
program that generates the same compile error.
upremas.F:66:72:
66
"use petscmat" will use those definitions.
As I said, you probably have mismatching arguments. For instance
call MatSeqAIJSetPreallocation(Mmat,
& PETSC_NULL_INTEGER_ARRAY,mr(np(246)),
& ierr)
The second argument is a Pe
For what its worth I have encountered a few more compile errors (and if
I exclude the failing-to-compile routines some link errors).
The link errors are:
Undefined symbols for architecture x86_64:
"_vecgetarrayreadf90_"
"_vecrestorearrayreadf90_",
The other compile errors are in
Jose,
What module should I be using to load petscmat.h90?
-sanjay
--
On 3/23/25 12:54 AM, Jose E. Roman wrote:
Have a look at the list of changes - it is currently here
https://urldefense.us/v3/__https://petsc.org/main/changes/dev/__;!!G_uCfscf7eWS!dTThTluS6q8pSm6Y7utH4DnjwWwV_R6ZlHfMvrTadm2
Follow-up. I notice in the documentation that there is a comment about
setting files to be *.F90
Right now my files are *.F when they contain PETSc functions and data.
I can also confirm that I have
$PETSC_ARCH/ftn/mat/petscmat.h90
and it has the interface as you note.
--
On 3/23/25 12
Have a look at the list of changes - it is currently here
https://urldefense.us/v3/__https://petsc.org/main/changes/dev/__;!!G_uCfscf7eWS!ZGeiat-1aoVqZo1IPI1kAEiFCl1GTP4Z65w0m3KJrZCopfOFNwbrmOPmjLGwC4J7Tw79C-8ozaNsXI05qkXz8Y18$
until the new version is released. See the last section "Fortran".
Here is what I have right now:
call MatMPIBAIJSetPreallocation(Mmat,nsbk,
& PETSC_NULL_INTEGER,mr(np(246)),
& PETSC_NULL_INTEGER,mr(np(247)),
& ierr)
call MatSeqBAIJSetPreallocation(Mmat,nsbk,
& PETS
The Fortran interfaces for those functions are generated correctly, see
$PETSC_ARCH/ftn/mat/petscmat.h90
For instance:
interface MatMPIBAIJSetPreallocation
subroutine MatMPIBAIJSetPreallocation(a,b,c,d,e,f, z)
import tMat
Mat :: a
PetscInt :: b
PetscInt :: c
PetscInt :: d(*)
Pets
Small update. I managed to eliminate all the errors associated with
PetscViewer and below (it had to do with the fact that I had not yet
built a module that was needed). The errors related to the
preallocation routines still persists.
-sanjay
On 3/23/25 12:19 AM, Sanjay Govindjee wrote:
Hi
Hi Barry,
I have moved to main and rebuilt the PETSc libraries etc. Right now
I am having trouble just getting my source code to compile. Plenty of
subroutines with PETSc calls compile but a few are throwing errors and
killing my compile. I suspect there will be more but if I can figure
the
Ok. Thanks for the recommendation.
---
Sanjay Govindjee, PhD, PE
Horace, Dorothy, and Katherine Johnson Professor in Engineering
Distinguished Professor of Civil and Environmental Engineering
779 Davis Hall
University of California
B
On Fri, 21 Mar 2025, Sanjay Govindjee via petsc-users wrote:
> Thanks. I will give that a try. To be clear, I should do
> > > git checkout origin/main
> > > make all
> > > make check
> Or do I also need to run my (long) .\configure before the make all?
I would start a clean build from a c
Thanks. I will give that a try. To be clear, I should do
> git checkout origin/main
> make all
> make check
Or do I also need to run my (long) .\configure before the make all?
-sanjay
On 3/21/25 7:17 AM, Barry Smith wrote:
I have just pushed a major update to the Fortran interface to
Some notes on this change are at
https://urldefense.us/v3/__https://gitlab.com/petsc/petsc/-/merge_requests/7517__;!!G_uCfscf7eWS!aQQ-TMsusc4QHRyMCbC8OwFN-znAEEo80ngP1kAlhvQJ_3zegcumVUvSqcBuE6lTbRZ0LWeAKXEXHEBXPmWRk-6NNBs$
Satish
On Fri, 21 Mar 2025, Barry Smith wrote:
>
> I have just pu
I have just pushed a major update to the Fortran interface to the main
PETSc git branch. Could you please try to work with main (to become release in
a couple of weeks) with your Fortran code as we debug the problem? This will
save you a lot of work and hopefully make the debugging more str
I am trying to upgrade my code to PETSc 3.22.4 (the code was last
updated to 3.19.4 or perhaps 3.18.1, I've lost track). I've been using
this code with PETSc for over 20 years.
To get my code to compile and link during this update, I only need to
make two changes; one was to use PetscViewerPus
22 matches
Mail list logo
