bug fix, PetscSubcommDestroy() did not free created MPI_comms

Hg-commit: 828e29ca337d32662a2f8b6c53deb3f905222da1

changed script to not unlink libraries and changed typo in pinit

Hg-commit: df0bd7aea8b022bed8b619527f22b31cb037bd73

make getaggregate work in 2d

Hg-commit: 5fd3e236ec3f75dee197436ee840d3c8eb19c3d4

accidently had removed initialization of flag to false in PetscOptionsTruthGroupBegin

Hg-commit: db1e573d10666cbb7ae5ee6e25306033714507d7

commit after merge

Hg-commit: b8ceb50b98ab67bf8cb250cdd95a4dda3c00f489

more work on AMS option setter

Hg-commit: 97faeb7c34710086b2f0d1c396bbe63d20f1c644

More functionality to perform scaling on GPU; converted some functions to use VecGetArray instead of direct access

Hg-commit: 715db1b2b6e77bb81d32d568d043e81d231bb386

fixed amsoptions.java to use default localhost
amsoptions code now handles all options and snes ex19 runs to completion
still some glitches, cannot handle changing type.

Hg-commit: 3a2988280069fa75d

fixed amsoptions.java to use default localhost
amsoptions code now handles all options and snes ex19 runs to completion
still some glitches, cannot handle changing type.

Hg-commit: 3a2988280069fa75d8bacfad2ac6478e41eb2d3d

show more ...

Added functions to copy to and from GPU, added variables necessary to CUDA to seq vectors

Hg-commit: b3a9b9e32c753dc82ed94216ea373ce2fb6b2b32

Added CUDA initialization.

Hg-commit: 0fc75b9e09451e0c64fbd8970cbc3b5a71f68c1d

more work on options database with Java GUI

Hg-commit: b0ad49ac4b2396d9fd7952b06ab0e7d87c1d6f2e

Arguments in VecStrideScatter were inverted

Hg-commit: 11aa67349d30121b0d7f3892f00581aae9b91043

AMS memory needs address to address of char

Hg-commit: 6a11b74fc40982411964151ed20ba8d71050aa95

updates for AMS

Hg-commit: 6e129ff35682d18b845e5944042caa1f3614694c

added AMSBean, amsoptions.java ams viewer

Hg-commit: 111ac237c359799b5be34e07489680f7acbf029c

cleanup of source for zope

Hg-commit: eb13da4a52adb9db635a651c575f3d0e42ecb70d

removed use of PetscGlobalXXX() from source where it should not have been used
removed use of MPI_IN_PLACE where it should not have been used
added text to developers.tex about not using > 1.1 MPI wi

removed use of PetscGlobalXXX() from source where it should not have been used
removed use of MPI_IN_PLACE where it should not have been used
added text to developers.tex about not using > 1.1 MPI without checks and not using PetscGlobalXXX()

Hg-commit: caac1bef6b4be19d1c0d5e507683ea2000c7a62a

show more ...

bugfix: PetscInitializePackage -> PetscSysInitializePackage

Hg-commit: 022025d933cb02105ddadc33ec4acea8aa106e59

changed PetscGetVersion() to not use unneeded and dangerous PetscStrcat()
added missing \n in PETSc help message
added initial xcode project that builds a simple iphone app

Hg-commit: 66d7e91cc737e8

changed PetscGetVersion() to not use unneeded and dangerous PetscStrcat()
added missing \n in PETSc help message
added initial xcode project that builds a simple iphone app

Hg-commit: 66d7e91cc737e8a060fee138585f89c5bc417239

show more ...

On May 13, 2010, at 10:48 AM, Jed Brown wrote:

I'm concerned by the following which appears in
PetscAttachDebuggerErrorHandler() and Petsc_MPI_DebuggerOnError():

ierr = PetscAttachDebugger();
if

On May 13, 2010, at 10:48 AM, Jed Brown wrote:

I'm concerned by the following which appears in
PetscAttachDebuggerErrorHandler() and Petsc_MPI_DebuggerOnError():

ierr = PetscAttachDebugger();
if (ierr) { /* hopeless so get out */
MPI_Finalize();
exit(*flag);
}

Since the error handler is not guaranteed to be called collectively, it
seems completely unreasonable to call a collective function, especially
not one with broader scope than PETSC_COMM_WORLD. It seems to me that
the hopeless case should instead call MPI_Abort() on the relevant
communicator.

You are probably right; this code fragment is so old it may have been written before I knew what MPI_Abort() was. I am changing it.


Similarly, why does Petsc_MPI_AbortOnError() call abort()
instead of MPI_Abort()?

I think this is clearly documented in the comment above it. Here is the comment:

This is so MPI errors in the debugger will leave all the stack
frames. The default abort cleans up and exits.

You use this error handler if you are running in the debugger (for example with -start_in_debugger). If you call MPI_Abort() it cleans up and you get no useful stack information in the debugger. I will change the comment to:

This is so MPI errors in the debugger will leave all the stack
frames. The default MP_Abort() cleans up and exits thus providing no useful information
in the debugger hence we call abort() instead of MPI_Abort().

Is it clear now, or am I missing something?

Barry

Hg-commit: 4663027817c3c85b7bd6d5229eb182ec49e8cf67

show more ...

Make error macros nest in cascading if statements, revert Hong's reversion of Barry's breakage

The

#define MACRO(x,y) do { multiple(x); statements(y); } while (0)

construct requires the user to

Make error macros nest in cascading if statements, revert Hong's reversion of Barry's breakage

The

#define MACRO(x,y) do { multiple(x); statements(y); } while (0)

construct requires the user to provide a closing semicolon (this patch
fixes this problem in many places), and does the correct thing in
cascading if statements. Note that the alternative

#define MACRO(x,y) { multiple(x); statements(y); }

expands

if (cond) MACRO(x);
else other();

as

if (cond) { multiple(x); statements; }
;
else other();

and the final statement is an else with no matching if. I suggest that
PETSc adopt the do {} while (0) construct anywhere that it currently
uses compound statements within unguarded braces (note that most PETSc
macros return a value and thus do not have this problem).

Hg-commit: ecf653b2f268465b7e7cb7569f7b80897acb163a

show more ...

finished formating of SETERRQ() calls and changed a few PETSC_COMM_SELF to appropriate comm

Hg-commit: 099173676d5157f4b09175cd1f5104bf5e549784

changed many calls to SETERRQ() to use correct (non-PETSC_COMM_SELF) communicator
reformated many badly formatted SETERRQ(); there was nonsense like {} around them or they were wrongly on a seperate

changed many calls to SETERRQ() to use correct (non-PETSC_COMM_SELF) communicator
reformated many badly formatted SETERRQ(); there was nonsense like {} around them or they were wrongly on a seperate line

Hg-commit: 7550f3779f18eaf7ce68bae52b964532aa97c1d5

show more ...

added MPI_Comm as first argument to PetscError() and SETERRQX()

Hg-commit: fe5a2ff050abc00ca2c2979c0af22b117b62874e

Move processing of -fp_trap to come after pushing other signal handlers

This prevents the trapping signal handler for SIGFPE from being overwritten
by the default handler which can't provide a usefu

Move processing of -fp_trap to come after pushing other signal handlers

This prevents the trapping signal handler for SIGFPE from being overwritten
by the default handler which can't provide a useful error message.

Note that the lack of interoperability between PetscPushSignalHandler
and floating point handlers is a deficiency in the interface for the
former. All the solutions I can think of would add nontrivial
complexity so I'm not implementing them now.

Hg-commit: 97a23cb8dc137879bc69e35ee6a7c964fcfff5ee

show more ...