July 2012 - petsc-dev - mailman3.cels.anl.gov

Re: [petsc-dev] Is there a reason for short arch and host names?
by Sean Farley 27 Jul '12

27 Jul '12

On Fri, Jul 27, 2012 at 6:20 PM, Jed Brown <five9a2(a)gmail.com> wrote: > By that logic, you should write +(a,b) instead of a+b. I'm looking forward > to if (!(rank)) {...}. Not really, just for functions whose names are more than one-character long. Duh.

1 0

Re: [petsc-dev] Is there a reason for short arch and host names?
by Jed Brown 27 Jul '12

27 Jul '12

By that logic, you should write +(a,b) instead of a+b. I'm looking forward to if (!(rank)) {...}. On Jul 27, 2012 6:13 PM, "Barry Smith" <bsmith(a)mcs.anl.gov> wrote: > > On Jul 27, 2012, at 6:08 PM, Jed Brown wrote: > > > (((MOAR) PARENTHESES)) > > > > One would think you were a Lisper. > > No because lisp puts the parenthesis in the wrong places. > > Actually I can handle only one concept: calling a function and I like > the syntax to be the same for doing that all the time. > > Barry > > Just because it may be a pre-processor or compile time function and > not a "real" runtime function doesn't make any different to me: I still > need function(args). > > > > > On Jul 27, 2012 6:05 PM, "Barry Smith" <bsmith(a)mcs.anl.gov> wrote: > > > > I absolutely hate sizeof arch. That is one of the most perverse > things people can do in C. What is wrong with good-old sizeof(arch)? > > > > Needs to go in the PETSc style guide :-) > > > > > > Barry > > > > > > On Jul 27, 2012, at 2:35 PM, Jed Brown wrote: > > > > > I'm sick of truncated PETSC_ARCH in log_summary. Does anyone object to > my pushing the following to petsc-3.3? Should we just make all the fields > huge so we're guaranteed to get everything? Since pname can be as long as > PETSC_MAX_PATH_LEN, this is not a line we can reasonably limit/align. > > > > > > --- a/src/sys/plog/plog.c > > > +++ b/src/sys/plog/plog.c > > > @@ -1269,7 +1269,7 @@ > > > PetscStageInfo *stageInfo = PETSC_NULL; > > > PetscEventPerfInfo *eventInfo = PETSC_NULL; > > > PetscClassPerfInfo *classInfo; > > > - char arch[10], hostname[64], username[16], > pname[PETSC_MAX_PATH_LEN], date[64]; > > > + char arch[64], hostname[128], username[16], > pname[PETSC_MAX_PATH_LEN], date[64]; > > > const char *name; > > > PetscLogDouble locTotalTime, TotalTime, TotalFlops; > > > PetscLogDouble numMessages, messageLength, avgMessLen, > numReductions; > > > @@ -1311,12 +1311,12 @@ > > > ierr = PetscFPrintf(comm, fd, "*** WIDEN YOUR WINDOW TO > 120 CHARACTERS. Use 'enscript -r -fCourier9' to print this document > ***\n");CHKERRQ(ierr); > > > ierr = PetscFPrintf(comm, fd, > "************************************************************************************************************************\n");CHKERRQ(ierr); > > > ierr = PetscFPrintf(comm, fd, > "\n---------------------------------------------- PETSc Performance > Summary: ----------------------------------------------\n\n");CHKERRQ(ierr); > > > - ierr = PetscGetArchType(arch, 10);CHKERRQ(ierr); > > > - ierr = PetscGetHostName(hostname, 64);CHKERRQ(ierr); > > > - ierr = PetscGetUserName(username, 16);CHKERRQ(ierr); > > > + ierr = PetscGetArchType(arch,sizeof arch);CHKERRQ(ierr); > > > + ierr = PetscGetHostName(hostname,sizeof hostname);CHKERRQ(ierr); > > > + ierr = PetscGetUserName(username,sizeof username);CHKERRQ(ierr); > > > ierr = PetscGetProgramName(pname, PETSC_MAX_PATH_LEN);CHKERRQ(ierr); > > > - ierr = PetscGetDate(date, 64);CHKERRQ(ierr); > > > - ierr = PetscGetVersion(version,256);CHKERRQ(ierr); > > > + ierr = PetscGetDate(date,sizeof date);CHKERRQ(ierr); > > > + ierr = PetscGetVersion(version,sizeof version);CHKERRQ(ierr); > > > if (size == 1) { > > > ierr = PetscFPrintf(comm,fd,"%s on a %s named %s with %d > processor, by %s %s\n", pname, arch, hostname, size, username, > date);CHKERRQ(ierr); > > > } else { > > > > > > >

1 0

Re: [petsc-dev] Is there a reason for short arch and host names?
by Barry Smith 27 Jul '12

27 Jul '12

On Jul 27, 2012, at 6:08 PM, Jed Brown wrote: > (((MOAR) PARENTHESES)) > > One would think you were a Lisper. No because lisp puts the parenthesis in the wrong places. Actually I can handle only one concept: calling a function and I like the syntax to be the same for doing that all the time. Barry Just because it may be a pre-processor or compile time function and not a "real" runtime function doesn't make any different to me: I still need function(args). > > On Jul 27, 2012 6:05 PM, "Barry Smith" <bsmith(a)mcs.anl.gov> wrote: > > I absolutely hate sizeof arch. That is one of the most perverse things people can do in C. What is wrong with good-old sizeof(arch)? > > Needs to go in the PETSc style guide :-) > > > Barry > > > On Jul 27, 2012, at 2:35 PM, Jed Brown wrote: > > > I'm sick of truncated PETSC_ARCH in log_summary. Does anyone object to my pushing the following to petsc-3.3? Should we just make all the fields huge so we're guaranteed to get everything? Since pname can be as long as PETSC_MAX_PATH_LEN, this is not a line we can reasonably limit/align. > > > > --- a/src/sys/plog/plog.c > > +++ b/src/sys/plog/plog.c > > @@ -1269,7 +1269,7 @@ > > PetscStageInfo *stageInfo = PETSC_NULL; > > PetscEventPerfInfo *eventInfo = PETSC_NULL; > > PetscClassPerfInfo *classInfo; > > - char arch[10], hostname[64], username[16], pname[PETSC_MAX_PATH_LEN], date[64]; > > + char arch[64], hostname[128], username[16], pname[PETSC_MAX_PATH_LEN], date[64]; > > const char *name; > > PetscLogDouble locTotalTime, TotalTime, TotalFlops; > > PetscLogDouble numMessages, messageLength, avgMessLen, numReductions; > > @@ -1311,12 +1311,12 @@ > > ierr = PetscFPrintf(comm, fd, "*** WIDEN YOUR WINDOW TO 120 CHARACTERS. Use 'enscript -r -fCourier9' to print this document ***\n");CHKERRQ(ierr); > > ierr = PetscFPrintf(comm, fd, "************************************************************************************************************************\n");CHKERRQ(ierr); > > ierr = PetscFPrintf(comm, fd, "\n---------------------------------------------- PETSc Performance Summary: ----------------------------------------------\n\n");CHKERRQ(ierr); > > - ierr = PetscGetArchType(arch, 10);CHKERRQ(ierr); > > - ierr = PetscGetHostName(hostname, 64);CHKERRQ(ierr); > > - ierr = PetscGetUserName(username, 16);CHKERRQ(ierr); > > + ierr = PetscGetArchType(arch,sizeof arch);CHKERRQ(ierr); > > + ierr = PetscGetHostName(hostname,sizeof hostname);CHKERRQ(ierr); > > + ierr = PetscGetUserName(username,sizeof username);CHKERRQ(ierr); > > ierr = PetscGetProgramName(pname, PETSC_MAX_PATH_LEN);CHKERRQ(ierr); > > - ierr = PetscGetDate(date, 64);CHKERRQ(ierr); > > - ierr = PetscGetVersion(version,256);CHKERRQ(ierr); > > + ierr = PetscGetDate(date,sizeof date);CHKERRQ(ierr); > > + ierr = PetscGetVersion(version,sizeof version);CHKERRQ(ierr); > > if (size == 1) { > > ierr = PetscFPrintf(comm,fd,"%s on a %s named %s with %d processor, by %s %s\n", pname, arch, hostname, size, username, date);CHKERRQ(ierr); > > } else { > > >

1 0

Re: [petsc-dev] Is there a reason for short arch and host names?
by Jed Brown 27 Jul '12

27 Jul '12

(((MOAR) PARENTHESES)) One would think you were a Lisper. On Jul 27, 2012 6:05 PM, "Barry Smith" <bsmith(a)mcs.anl.gov> wrote: > > I absolutely hate sizeof arch. That is one of the most perverse > things people can do in C. What is wrong with good-old sizeof(arch)? > > Needs to go in the PETSc style guide :-) > > > Barry > > > On Jul 27, 2012, at 2:35 PM, Jed Brown wrote: > > > I'm sick of truncated PETSC_ARCH in log_summary. Does anyone object to > my pushing the following to petsc-3.3? Should we just make all the fields > huge so we're guaranteed to get everything? Since pname can be as long as > PETSC_MAX_PATH_LEN, this is not a line we can reasonably limit/align. > > > > --- a/src/sys/plog/plog.c > > +++ b/src/sys/plog/plog.c > > @@ -1269,7 +1269,7 @@ > > PetscStageInfo *stageInfo = PETSC_NULL; > > PetscEventPerfInfo *eventInfo = PETSC_NULL; > > PetscClassPerfInfo *classInfo; > > - char arch[10], hostname[64], username[16], > pname[PETSC_MAX_PATH_LEN], date[64]; > > + char arch[64], hostname[128], username[16], > pname[PETSC_MAX_PATH_LEN], date[64]; > > const char *name; > > PetscLogDouble locTotalTime, TotalTime, TotalFlops; > > PetscLogDouble numMessages, messageLength, avgMessLen, > numReductions; > > @@ -1311,12 +1311,12 @@ > > ierr = PetscFPrintf(comm, fd, "*** WIDEN YOUR WINDOW TO > 120 CHARACTERS. Use 'enscript -r -fCourier9' to print this document > ***\n");CHKERRQ(ierr); > > ierr = PetscFPrintf(comm, fd, > "************************************************************************************************************************\n");CHKERRQ(ierr); > > ierr = PetscFPrintf(comm, fd, > "\n---------------------------------------------- PETSc Performance > Summary: ----------------------------------------------\n\n");CHKERRQ(ierr); > > - ierr = PetscGetArchType(arch, 10);CHKERRQ(ierr); > > - ierr = PetscGetHostName(hostname, 64);CHKERRQ(ierr); > > - ierr = PetscGetUserName(username, 16);CHKERRQ(ierr); > > + ierr = PetscGetArchType(arch,sizeof arch);CHKERRQ(ierr); > > + ierr = PetscGetHostName(hostname,sizeof hostname);CHKERRQ(ierr); > > + ierr = PetscGetUserName(username,sizeof username);CHKERRQ(ierr); > > ierr = PetscGetProgramName(pname, PETSC_MAX_PATH_LEN);CHKERRQ(ierr); > > - ierr = PetscGetDate(date, 64);CHKERRQ(ierr); > > - ierr = PetscGetVersion(version,256);CHKERRQ(ierr); > > + ierr = PetscGetDate(date,sizeof date);CHKERRQ(ierr); > > + ierr = PetscGetVersion(version,sizeof version);CHKERRQ(ierr); > > if (size == 1) { > > ierr = PetscFPrintf(comm,fd,"%s on a %s named %s with %d processor, > by %s %s\n", pname, arch, hostname, size, username, date);CHKERRQ(ierr); > > } else { > > > >

1 0

Re: [petsc-dev] Is there a reason for short arch and host names?
by Barry Smith 27 Jul '12

27 Jul '12

I absolutely hate sizeof arch. That is one of the most perverse things people can do in C. What is wrong with good-old sizeof(arch)? Needs to go in the PETSc style guide :-) Barry On Jul 27, 2012, at 2:35 PM, Jed Brown wrote: > I'm sick of truncated PETSC_ARCH in log_summary. Does anyone object to my pushing the following to petsc-3.3? Should we just make all the fields huge so we're guaranteed to get everything? Since pname can be as long as PETSC_MAX_PATH_LEN, this is not a line we can reasonably limit/align. > > --- a/src/sys/plog/plog.c > +++ b/src/sys/plog/plog.c > @@ -1269,7 +1269,7 @@ > PetscStageInfo *stageInfo = PETSC_NULL; > PetscEventPerfInfo *eventInfo = PETSC_NULL; > PetscClassPerfInfo *classInfo; > - char arch[10], hostname[64], username[16], pname[PETSC_MAX_PATH_LEN], date[64]; > + char arch[64], hostname[128], username[16], pname[PETSC_MAX_PATH_LEN], date[64]; > const char *name; > PetscLogDouble locTotalTime, TotalTime, TotalFlops; > PetscLogDouble numMessages, messageLength, avgMessLen, numReductions; > @@ -1311,12 +1311,12 @@ > ierr = PetscFPrintf(comm, fd, "*** WIDEN YOUR WINDOW TO 120 CHARACTERS. Use 'enscript -r -fCourier9' to print this document ***\n");CHKERRQ(ierr); > ierr = PetscFPrintf(comm, fd, "************************************************************************************************************************\n");CHKERRQ(ierr); > ierr = PetscFPrintf(comm, fd, "\n---------------------------------------------- PETSc Performance Summary: ----------------------------------------------\n\n");CHKERRQ(ierr); > - ierr = PetscGetArchType(arch, 10);CHKERRQ(ierr); > - ierr = PetscGetHostName(hostname, 64);CHKERRQ(ierr); > - ierr = PetscGetUserName(username, 16);CHKERRQ(ierr); > + ierr = PetscGetArchType(arch,sizeof arch);CHKERRQ(ierr); > + ierr = PetscGetHostName(hostname,sizeof hostname);CHKERRQ(ierr); > + ierr = PetscGetUserName(username,sizeof username);CHKERRQ(ierr); > ierr = PetscGetProgramName(pname, PETSC_MAX_PATH_LEN);CHKERRQ(ierr); > - ierr = PetscGetDate(date, 64);CHKERRQ(ierr); > - ierr = PetscGetVersion(version,256);CHKERRQ(ierr); > + ierr = PetscGetDate(date,sizeof date);CHKERRQ(ierr); > + ierr = PetscGetVersion(version,sizeof version);CHKERRQ(ierr); > if (size == 1) { > ierr = PetscFPrintf(comm,fd,"%s on a %s named %s with %d processor, by %s %s\n", pname, arch, hostname, size, username, date);CHKERRQ(ierr); > } else { >

1 0

Re: [petsc-dev] Is there a reason for short arch and host names?
by Matthew Knepley 27 Jul '12

27 Jul '12

I like it. Matt On Fri, Jul 27, 2012 at 2:35 PM, Jed Brown <jedbrown(a)mcs.anl.gov> wrote: > I'm sick of truncated PETSC_ARCH in log_summary. Does anyone object to my > pushing the following to petsc-3.3? Should we just make all the fields huge > so we're guaranteed to get everything? Since pname can be as long as > PETSC_MAX_PATH_LEN, this is not a line we can reasonably limit/align. > > --- a/src/sys/plog/plog.c > +++ b/src/sys/plog/plog.c > @@ -1269,7 +1269,7 @@ > PetscStageInfo *stageInfo = PETSC_NULL; > PetscEventPerfInfo *eventInfo = PETSC_NULL; > PetscClassPerfInfo *classInfo; > - char arch[10], hostname[64], username[16], > pname[PETSC_MAX_PATH_LEN], date[64]; > + char arch[64], hostname[128], username[16], > pname[PETSC_MAX_PATH_LEN], date[64]; > const char *name; > PetscLogDouble locTotalTime, TotalTime, TotalFlops; > PetscLogDouble numMessages, messageLength, avgMessLen, > numReductions; > @@ -1311,12 +1311,12 @@ > ierr = PetscFPrintf(comm, fd, "*** WIDEN YOUR WINDOW TO 120 > CHARACTERS. Use 'enscript -r -fCourier9' to print this document > ***\n");CHKERRQ(ierr); > ierr = PetscFPrintf(comm, fd, > "************************************************************************************************************************\n");CHKERRQ(ierr); > ierr = PetscFPrintf(comm, fd, > "\n---------------------------------------------- PETSc Performance > Summary: ----------------------------------------------\n\n");CHKERRQ(ierr); > - ierr = PetscGetArchType(arch, 10);CHKERRQ(ierr); > - ierr = PetscGetHostName(hostname, 64);CHKERRQ(ierr); > - ierr = PetscGetUserName(username, 16);CHKERRQ(ierr); > + ierr = PetscGetArchType(arch,sizeof arch);CHKERRQ(ierr); > + ierr = PetscGetHostName(hostname,sizeof hostname);CHKERRQ(ierr); > + ierr = PetscGetUserName(username,sizeof username);CHKERRQ(ierr); > ierr = PetscGetProgramName(pname, PETSC_MAX_PATH_LEN);CHKERRQ(ierr); > - ierr = PetscGetDate(date, 64);CHKERRQ(ierr); > - ierr = PetscGetVersion(version,256);CHKERRQ(ierr); > + ierr = PetscGetDate(date,sizeof date);CHKERRQ(ierr); > + ierr = PetscGetVersion(version,sizeof version);CHKERRQ(ierr); > if (size == 1) { > ierr = PetscFPrintf(comm,fd,"%s on a %s named %s with %d processor, > by %s %s\n", pname, arch, hostname, size, username, date);CHKERRQ(ierr); > } else { > > -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener

1 0

Is there a reason for short arch and host names?
by Jed Brown 27 Jul '12

27 Jul '12

I'm sick of truncated PETSC_ARCH in log_summary. Does anyone object to my pushing the following to petsc-3.3? Should we just make all the fields huge so we're guaranteed to get everything? Since pname can be as long as PETSC_MAX_PATH_LEN, this is not a line we can reasonably limit/align. --- a/src/sys/plog/plog.c +++ b/src/sys/plog/plog.c @@ -1269,7 +1269,7 @@ PetscStageInfo *stageInfo = PETSC_NULL; PetscEventPerfInfo *eventInfo = PETSC_NULL; PetscClassPerfInfo *classInfo; - char arch[10], hostname[64], username[16], pname[PETSC_MAX_PATH_LEN], date[64]; + char arch[64], hostname[128], username[16], pname[PETSC_MAX_PATH_LEN], date[64]; const char *name; PetscLogDouble locTotalTime, TotalTime, TotalFlops; PetscLogDouble numMessages, messageLength, avgMessLen, numReductions; @@ -1311,12 +1311,12 @@ ierr = PetscFPrintf(comm, fd, "*** WIDEN YOUR WINDOW TO 120 CHARACTERS. Use 'enscript -r -fCourier9' to print this document ***\n");CHKERRQ(ierr); ierr = PetscFPrintf(comm, fd, "************************************************************************************************************************\n");CHKERRQ(ierr); ierr = PetscFPrintf(comm, fd, "\n---------------------------------------------- PETSc Performance Summary: ----------------------------------------------\n\n");CHKERRQ(ierr); - ierr = PetscGetArchType(arch, 10);CHKERRQ(ierr); - ierr = PetscGetHostName(hostname, 64);CHKERRQ(ierr); - ierr = PetscGetUserName(username, 16);CHKERRQ(ierr); + ierr = PetscGetArchType(arch,sizeof arch);CHKERRQ(ierr); + ierr = PetscGetHostName(hostname,sizeof hostname);CHKERRQ(ierr); + ierr = PetscGetUserName(username,sizeof username);CHKERRQ(ierr); ierr = PetscGetProgramName(pname, PETSC_MAX_PATH_LEN);CHKERRQ(ierr); - ierr = PetscGetDate(date, 64);CHKERRQ(ierr); - ierr = PetscGetVersion(version,256);CHKERRQ(ierr); + ierr = PetscGetDate(date,sizeof date);CHKERRQ(ierr); + ierr = PetscGetVersion(version,sizeof version);CHKERRQ(ierr); if (size == 1) { ierr = PetscFPrintf(comm,fd,"%s on a %s named %s with %d processor, by %s %s\n", pname, arch, hostname, size, username, date);CHKERRQ(ierr); } else {

1 0

Re: [petsc-dev] Complex Cholesky
by Jose E. Roman 27 Jul '12

27 Jul '12

El 26/07/2012, a las 21:34, Hong Zhang escribió: > Jose : > > Then why not give a PETSC_ERR_SUP error? > > Patched 3.3 and petsc-dev. > http://petsc.cs.iit.edu/petsc/petsc-dev/rev/ff096aff2a39 > > When I find time, I'll implement MatSolve() for seqsbaij_hermitian :-) > > Hong > Thanks. I think the same error should go to MatGetFactor_seqaij_petsc otherwise my second sample run gives wrong results: $ ./ex10 -f $PETSC_DIR/share/petsc/datafiles/matrices/hpd-complex-int32-float64 -ksp_type preonly -pc_type cholesky Number of iterations = 1 Residual norm 1.64361 Jose

1 0

Re: [petsc-dev] Regression tests
by Jed Brown 26 Jul '12

26 Jul '12

On Thu, Jul 26, 2012 at 3:00 PM, Matthew Knepley <knepley(a)gmail.com> wrote: > How can your make.log go away if its in PETSC_ARCH/conf ? Is this a useful > argument? The symlinks point there. The way it has always worked is that the symlinks point at the latest PETSC_ARCH/conf/ and the files there get overwritten (with one level of backup). Are you saying the symlink should sometimes gets removed and sometimes the normal file gets deleted and replaced by a symlink? I think that's confusing.

1 0

Re: [petsc-dev] Regression tests
by Matthew Knepley 26 Jul '12

26 Jul '12

On Thu, Jul 26, 2012 at 2:49 PM, Jed Brown <jedbrown(a)mcs.anl.gov> wrote: > On Thu, Jul 26, 2012 at 2:44 PM, Matthew Knepley <knepley(a)gmail.com>wrote: > >> You do not read it correctly. It operates exactly as everything else. You >> move make.log to make.log.bkp, which is what we do now, and >> write into make.log. It seems to be writing without moving first. >> >> >>> I think it is wrong to reuse the name make.log for both purposes. >>> Regardless, the symlinks are not removed, so make.log output overwrites >>> $PETSC_ARCH/conf/make.log and (the next time you run it) >>> $PETSC_ARCH/conf/make.log.bkp. >>> >> >> In my first message, it says "this is a bug". What is hard about that? >> What necessitates 3 replies missing that point? > > > Sounds like your "correct behavior" is exactly what I'm complaining about. > Since there is only one level of backup, running twice, as in > > ./builder2.py buildExample src/... > ./builder2.py buildExample src/... > > destroys all trace of the make.log for the library. > How can your make.log go away if its in PETSC_ARCH/conf ? Is this a useful argument? Matt -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener

1 0