- Intel-nda - mailman3.cels.anl.gov

[JLSE-Intel-NDA] Fwd: [Early-users-discuss] KNL book is available as PDF online
by Kumaran, Kalyan 16 Aug '16

16 Aug '16

Begin forwarded message: From: "Hammond, Jeff R" <jeff.r.hammond(a)intel.com<mailto:[email protected]>> Subject: Re: [Early-users-discuss] KNL book is available as PDF online Date: August 16, 2016 at 10:49:37 AM CDT To: "early-users-discuss(a)lists.alcf.anl.gov<mailto:[email protected]>" <early-users-discuss(a)lists.alcf.anl.gov<mailto:[email protected]>> --------------------------------------------- No NDA material may be included in messages to early-users-discuss! _______________________________________________ I attached the two chapters that everyone should read this week. Jeff From: Jeff R Hammond <jeff.r.hammond(a)intel.com<mailto:[email protected]>> Date: Tuesday, August 16, 2016 at 10:46 AM To: "early-users-discuss(a)lists.alcf.anl.gov<mailto:[email protected]>" <early-users-discuss(a)lists.alcf.anl.gov<mailto:[email protected]>> Subject: KNL book is available as PDF online http://www.sciencedirect.com/science/book/9780128091944 This book has a number of important architectural details that I can’t find anywhere else online right now. I had no trouble downloading, but I don’t know if this is because I have an Intel IP address. Please let me know if you cannot download. Jeff _______________________________________________ No NDA material may be included in messages to early-users-discuss! early-users-discuss mailing list early-users-discuss(a)lists.alcf.anl.gov<mailto:[email protected]> https://lists.alcf.anl.gov/mailman/listinfo/early-users-discuss

1 0

[JLSE-Intel-NDA] knl03 in trouble?
by Tom LeCompte 16 Aug '16

16 Aug '16

Hi everyone- My jobs on knl03 immediately exit. The same jobs run OK on 0-2 and 6-8. Is this a known problem? Cobalt doesn't say much: Tue Aug 16 15:20:40 2016 +0000 (UTC) submitted with cwd set to: /home/lecompte/g4/Geant4HepExpMTBenchmark/run jobid 520 submitted from terminal /dev/pts/5 Tue Aug 16 15:23:55 2016 +0000 (UTC) Info: resources released; starting job epilogue scripts Cheers, Tom

1 0

Re: [JLSE-Intel-NDA] GCC Compilers on KNL
by Allen, Benjamin S. 15 Aug '16

15 Aug '16

Tom, sorry. Typo'ed your name. Ben > On Aug 15, 2016, at 10:48 AM, Allen, Benjamin S. <bsallen(a)alcf.anl.gov> wrote: > > Hi Tim, > > No we only have RHEL7's GCC 4.8 installed. Feel free to use one of the higher end nodes to speed up the compile instead of the logins. I also suggest using local storage (i.e /tmp) for the compile if possible. > > Ben > >> On Aug 15, 2016, at 10:40 AM, Tom LeCompte <lecompte(a)anl.gov> wrote: >> >> Hi everyone- >> >> I'm doing a bit of yak shaving here. >> >> I want to compile a piece of code (Geant4) under ICC/ICPC, and get an error in /usr/include/c++/4.8.5/bits/hashtable.h This turns out to be a known problem, and the authors tell me the fix is to use the headers for GCC 4.9+ (ICC uses the GCC headers). Before running around and installing my own GCC 4.9, is there one installed on the cluster? If not, are there any plans to install it? >> >> Cheers, >> Tom >> _______________________________________________ >> Intel-nda mailing list >> Intel-nda(a)lists.jlse.anl.gov >> https://lists.jlse.anl.gov/mailman/listinfo/intel-nda > > _______________________________________________ > Intel-nda mailing list > Intel-nda(a)lists.jlse.anl.gov > https://lists.jlse.anl.gov/mailman/listinfo/intel-nda

1 0

Re: [JLSE-Intel-NDA] GCC Compilers on KNL
by Allen, Benjamin S. 15 Aug '16

15 Aug '16

Hi Tim, No we only have RHEL7's GCC 4.8 installed. Feel free to use one of the higher end nodes to speed up the compile instead of the logins. I also suggest using local storage (i.e /tmp) for the compile if possible. Ben > On Aug 15, 2016, at 10:40 AM, Tom LeCompte <lecompte(a)anl.gov> wrote: > > Hi everyone- > > I'm doing a bit of yak shaving here. > > I want to compile a piece of code (Geant4) under ICC/ICPC, and get an error in /usr/include/c++/4.8.5/bits/hashtable.h This turns out to be a known problem, and the authors tell me the fix is to use the headers for GCC 4.9+ (ICC uses the GCC headers). Before running around and installing my own GCC 4.9, is there one installed on the cluster? If not, are there any plans to install it? > > Cheers, > Tom > _______________________________________________ > Intel-nda mailing list > Intel-nda(a)lists.jlse.anl.gov > https://lists.jlse.anl.gov/mailman/listinfo/intel-nda

1 0

[JLSE-Intel-NDA] GCC Compilers on KNL
by Tom LeCompte 15 Aug '16

15 Aug '16

Hi everyone- I'm doing a bit of yak shaving here. I want to compile a piece of code (Geant4) under ICC/ICPC, and get an error in /usr/include/c++/4.8.5/bits/hashtable.h This turns out to be a known problem, and the authors tell me the fix is to use the headers for GCC 4.9+ (ICC uses the GCC headers). Before running around and installing my own GCC 4.9, is there one installed on the cluster? If not, are there any plans to install it? Cheers, Tom

1 0

[JLSE-Intel-NDA] Fwd: [catalysts] KNL in Architecture Optimization Manual
by Kumaran, Kalyan 14 Aug '16

14 Aug '16

> Begin forwarded message: > > > Hi everyone, > > FYI: The KNL is now covered in Intel's Architecture Optimization Manual (http://www.intel.com/content/www/us/en/architecture-and-technology/64-ia-32…) specifically chapter 16. > > -Hal > > -- > Hal Finkel > Assistant Computational Scientist > Leadership Computing Facility > Argonne National Laboratory > _______________________________________________ > catalysts mailing list > catalysts(a)lists.alcf.anl.gov > https://lists.alcf.anl.gov/cgi-bin/mailman/listinfo/catalysts

1 0

[JLSE-Intel-NDA] bus error?
by Steven C. Pieper 12 Aug '16

12 Aug '16

On the KNL, what is a bus error (signal 7) as distinguished from a seg fault (11)? I am getting these with (very) low probability in OpenMP regions of my Fortran code. The same code has worked fine for years on mira. Also on KNL the same subroutines have been successfully invoked up to 100,000 times. But attempts to scale up to bigger runs are being hit now. Removing the !$OMP statements for the given loop seems to get rid of the bus error, but of course this slows the program down -- we have to be able to use multithreading on this machine. Steve -- Steven C. Pieper, Senior Physicist Emeritus: spieper(a)anl.gov Argonne National Laboratory, Physics Division, Bldg. 203, Argonne, IL 60439 Phone: 630-252-4232 Fax -6008 Secretary, Debra Beres, -4100 http://www.phy.anl.gov/theory/staff/scp.html

1 0

[JLSE-Intel-NDA] Cobalt Fixes
by Allen, Benjamin S. 11 Aug '16

11 Aug '16

There were some value normalization fixes that needed to happen between /var/run/hwloc/knl_memoryside_cache, and what the BIOS values use that was causing Cobalt to not set knl_cluster_mode or knl_memory_mode in certain cases. Largely when SNC-2, SNC-4, or Hybrid were specified. This is now fixed. Please let me know if you run into any further instances where the values in /var/run/hwloc/knl_memoryside_cache don't match the values you selected. I also removed Auto from the possible values for both knl_memory_mode and knl_cluster_mode as Auto sets to Cached and Quadrant respectively. Thanks, Ben

1 0

Re: [JLSE-Intel-NDA] the mystery of COBALT_JOBSIZE
by Paul Hargrove 11 Aug '16

11 Aug '16

And why is cat even needed? Is there a reason not to use $(wc -l $COBALT_NODEFILE) ? -Paul On Thu, Aug 11, 2016 at 11:35 AM, Allen, Benjamin S. <bsallen(a)alcf.anl.gov> wrote: > This is fixed. > > [bsallen@knl04 ~]$ cat $COBALT_NODEFILE > knl04 > knl08 > [bsallen@knl04 ~]$ cat $COBALT_NODEFILE | wc -l > 2 > > > > On Aug 11, 2016, at 1:03 PM, Allen, Benjamin S. <bsallen(a)alcf.anl.gov> > wrote: > > > > It's because wc -l counts newlines, and the node file is missing the > trailing newline. I'll fix this this afternoon. > > > > Ben > > > > > > From: "Steven C. Pieper" <spieper(a)anl.gov> > > Sent: Aug 11, 2016 12:49 PM > > To: intel-nda(a)lists.jlse.anl.gov > > Subject: Re: [JLSE-Intel-NDA] the mystery of COBALT_JOBSIZE > > > > The second possibility doesn't work for some reason I don't > > understand. I had observed that cat does not list two lines > > for a 2-node node file: > > > > [knl02 ~] echo $COBALT_NODEFILE > > /var/tmp/cobalt.327 > > [knl02 ~] cat /var/tmp/cobalt.327 > > knl07 > > knl02[knl02 ~] NODECOUNT=$(cat $COBALT_NODEFILE | wc -l) > > [knl02 ~] echo $NODECOUNT > > 1 > > > > however vi does show the two lines: > > > > knl07 > > knl02 > > ~ > > > > Seems very strange. > > > > Steve > > Steven C. Pieper, Senior Physicist Emeritus: > spieper(a)anl.gov > > > > Argonne National Laboratory, Physics Division, Bldg. 203, Argonne, IL > 60439 > > Phone: 630-252-4232 Fax -6008 Secretary, Debra Beres, > -4100 > > > > http://www.phy.anl.gov/theory/staff/scp.html > > On 08/11/16 11:43, Allen, Benjamin S. wrote: > >> Couple ways to deal with this: > >> > >> mpirun -f $COBALT_NODEFILE -ppn 64 > >> > >> or > >> > >> NODECOUNT=$(cat $COBALT_NODEFILE | wc -l) > >> > >> COBALT_NODEFILE and COBALT_JOBID are the only two environmental > variables passed from Cobalt, similar to Cooley. > >> > >> Ben > >> > >> > >>> On Aug 10, 2016, at 6:08 PM, Alvaro <vama(a)alcf.anl.gov> > >>> wrote: > >>> > >>> hi. > >>> > >>> I am not getting any value for COBALT_JOBSIZE. > >>> > >>> So I cannot do > >>> > >>> proc=64 > >>> > >>> mpirun -n $((COBALT_JOBSIZE*proc)) > >>> > >>> How I can get the number of nodes in cobalt? > >>> > >>> > >>> -- > >>> -Alvaro > >>> http://web.alcf.anl.gov/~vama/ > >>> > >>> _______________________________________________ > >>> Intel-nda mailing list > >>> > >>> Intel-nda(a)lists.jlse.anl.gov > >>> https://lists.jlse.anl.gov/mailman/listinfo/intel-nda > >> _______________________________________________ > >> Intel-nda mailing list > >> > >> Intel-nda(a)lists.jlse.anl.gov > >> https://lists.jlse.anl.gov/mailman/listinfo/intel-nda > > > > > > _______________________________________________ > Intel-nda mailing list > Intel-nda(a)lists.jlse.anl.gov > https://lists.jlse.anl.gov/mailman/listinfo/intel-nda > -- Paul H. Hargrove PHHargrove(a)lbl.gov Computer Languages & Systems Software (CLaSS) Group Computer Science Department Tel: +1-510-495-2352 Lawrence Berkeley National Laboratory Fax: +1-510-486-6900

1 0

Re: [JLSE-Intel-NDA] the mystery of COBALT_JOBSIZE
by Allen, Benjamin S. 11 Aug '16

11 Aug '16

This is fixed. [bsallen@knl04 ~]$ cat $COBALT_NODEFILE knl04 knl08 [bsallen@knl04 ~]$ cat $COBALT_NODEFILE | wc -l 2 > On Aug 11, 2016, at 1:03 PM, Allen, Benjamin S. <bsallen(a)alcf.anl.gov> wrote: > > It's because wc -l counts newlines, and the node file is missing the trailing newline. I'll fix this this afternoon. > > Ben > > > From: "Steven C. Pieper" <spieper(a)anl.gov> > Sent: Aug 11, 2016 12:49 PM > To: intel-nda(a)lists.jlse.anl.gov > Subject: Re: [JLSE-Intel-NDA] the mystery of COBALT_JOBSIZE > > The second possibility doesn't work for some reason I don't > understand. I had observed that cat does not list two lines > for a 2-node node file: > > [knl02 ~] echo $COBALT_NODEFILE > /var/tmp/cobalt.327 > [knl02 ~] cat /var/tmp/cobalt.327 > knl07 > knl02[knl02 ~] NODECOUNT=$(cat $COBALT_NODEFILE | wc -l) > [knl02 ~] echo $NODECOUNT > 1 > > however vi does show the two lines: > > knl07 > knl02 > ~ > > Seems very strange. > > Steve > Steven C. Pieper, Senior Physicist Emeritus: spieper(a)anl.gov > > Argonne National Laboratory, Physics Division, Bldg. 203, Argonne, IL 60439 > Phone: 630-252-4232 Fax -6008 Secretary, Debra Beres, -4100 > > http://www.phy.anl.gov/theory/staff/scp.html > On 08/11/16 11:43, Allen, Benjamin S. wrote: >> Couple ways to deal with this: >> >> mpirun -f $COBALT_NODEFILE -ppn 64 >> >> or >> >> NODECOUNT=$(cat $COBALT_NODEFILE | wc -l) >> >> COBALT_NODEFILE and COBALT_JOBID are the only two environmental variables passed from Cobalt, similar to Cooley. >> >> Ben >> >> >>> On Aug 10, 2016, at 6:08 PM, Alvaro <vama(a)alcf.anl.gov> >>> wrote: >>> >>> hi. >>> >>> I am not getting any value for COBALT_JOBSIZE. >>> >>> So I cannot do >>> >>> proc=64 >>> >>> mpirun -n $((COBALT_JOBSIZE*proc)) >>> >>> How I can get the number of nodes in cobalt? >>> >>> >>> -- >>> -Alvaro >>> http://web.alcf.anl.gov/~vama/ >>> >>> _______________________________________________ >>> Intel-nda mailing list >>> >>> Intel-nda(a)lists.jlse.anl.gov >>> https://lists.jlse.anl.gov/mailman/listinfo/intel-nda >> _______________________________________________ >> Intel-nda mailing list >> >> Intel-nda(a)lists.jlse.anl.gov >> https://lists.jlse.anl.gov/mailman/listinfo/intel-nda > >

1 0