From: Dan Bonachea (bonachea_at_cs_dot_berkeley_dot_edu)
Date: Sun Sep 24 2006 - 16:48:12 PDT
At 04:14 PM 9/24/2006, Konstantin Kleisouris wrote: >Dan, > > Thank you for all the advice. The patch you e-mailed me solved the >problem that I mentioned in my previous e-mail. As a result I was able >to generate a upcc binary. > However, I still can not compile my program. I am not sure whether >this has to do with upc or the fact that the OS on the machine is >really old. When I try to compile a program this is what I get: > >upcc -o M1IntraInter rand1.c util.c M1IntraInter.c slice.c >train_data.c radixb_par.c radix_local.c -DRAND_FILE=\"`pwd`/randfile\" >-lm -pthreads=4 >upcc: Error building pthread objects with thread-local data: >sort: option requires an argument -- k >sort [-bcdfiMmnru] [-o output] [-T directory] [-ykmem] [-t char] > [+pos1 [-pos2]] [-k field_start[type][,field_end[type]] [file...] >make: *** [global.tld] Error 2 > > >Do you happen to have an idea as to why I get this message? Hi Konstantin - It appears you've discovered another bug.. We should put you on staff... ;-) Looks like the Solaris 7 'sort' command is pickier about argument formatting than ones we've seen on all other systems. Attached is another patch that should hopefully fix this issue.. just apply it to your source directory: $ cd $SRCDIR $ patch -p0 < sort-patch patching file upcc.pl then do another 'gmake all install' from the builddir. Hope this helps, and thanks again for the useful feedback... Dan >Thank you, >Konstantinos Kleisouris > > > > > > >On Sat, 23 Sep 2006 18:02:26 -0700 > Dan Bonachea <bonachea_at_cs_dot_berkeley_dot_edu> wrote: > > Hi Konstantin - Apologies for the continued trouble. > > > > If the mmap.c fix that I suggested got you past the gmake phase, then > > I think that's the right fix to use for that Solaris 7 problem (we'll > > merge something similar into our dev tree). > > > > The remaining issue (with upcc) is actually an orthogonal problem - > > it's a known bug in the 2.2.2 release which affects very recent > > versions of gmake (3.81 and newer, and you appear to be using 3.81): > > > > http://upc-bugs.lbl.gov/bugzilla/show_bug.cgi?id=1645 > > > > There are several possible workarounds: > > > > * Apply the attached patch to upcc.mak in your source tree, ie: > > > > > > $ cd $SRCDIR > > $ patch -p0 < bug1645.patch > > patching file upcc.mak.in > > > > you'll then need to do another top-level 'gmake all install' to > > propagate the fix. > > > > -or- > > > > * This bug is already fixed in our 2.3.16 beta, so you could also > > just download that newer snapshot and use it instead (although you'd > > need to re-apply the same Solaris7 fix to gasnet_mmap.c) > > > > Hope this gets you up and running - thanks for your patience! > > > > Dan > > > > At 05:24 PM 9/23/2006, Konstantin Kleisouris wrote: > > >Hi everyone, > > > > > > I tried the approached proposed by Dan and Paul. > > > So, Dan proposed that I should add the following two lines to > > the > > >top of the gasnet_mmap.c file. > > > > > >#define GASNETI_MMAP_FLAGS (MAP_PRIVATE | MAP_NORESERVE) > > >#define GASNETI_MMAP_FILE "/dev/zero" > > > > > > gmake succeeded, but when I type "upcc --version" I get the > > >following messages: > > > UPC-to-C translator | <unable to determine> > > > Runtime interface # | Runtime supports 3.0 -> 3.6: Translator > > uses > > ><unable to | determine> > > > > > > check the versiondan.txt file that I have attached to the > > current > > >e-mail. Also, I tried to compile one of my upc programs with the > > upcc > > >that was generated after doing "make install". Check below what I > > get: > > > > > >../UPC/bin/upcc -o M1IntraInter rand1.c util.c M1IntraInter.c > > slice.c > > >train_data.c radixb_par.c radix_local.c > > -DRAND_FILE=\"`pwd`/randfile\" > > >-lm -pthreads=4 > > >upcc: error during preprocessing: > > >syntax error at -e line 2, near "if" > > >Backslash found where operator expected at -e line 2, near ") \" > > > (Missing operator before \?) > > >syntax error at -e line 3, near ";}" > > >Execution of -e aborted due to compilation errors. > > >make: *** [rand1.ipre2] Error 255 > > > > > > > > > I also tried Paul's approach which was to replace MAP_ANON with > > >MAP_ANONYMOUS on line 43 of gasnet/gasnet_mmap.c > > > > > > gmake failed to build libgasnet-smp-seq.a. I have attached a > > >gmakepaul.txt file to the current e-mail so that you see what > > messages > > >I get. > > > > > >Kosta > > > > > > > > > > > >On Fri, 22 Sep 2006 17:52:45 -0700 > > > "Paul H. Hargrove" <PHHargrove_at_lbl_dot_gov> wrote: > > > > Konstantinos, > > > > > > > > I'd like you to try changing MAP_ANON to MAP_ANONYMOUS on line > > 43 > > > > of > > > > gasnet/gasnet_mmap.c of the Berkeley UPC distribution. If that > > fixes > > > > the problem, let us know and our next release will include a > > > > configure-time check to pick between MAP_ANON and MAP_ANONYMOUS. > > > > > > > > -Paul > > > > > > > > Konstantin Kleisouris wrote: > > > > > Hi all, > > > > > > > > > > I am experiencing difficulty in building libgasnet-smp-seq.a > > on a > > > > > sparc-sun-solaris2.7. The version of gcc on the machine is > > 2.95.2 > > > > > However, when gmake tries to build libgasnet-smp-seq.a I get: > > > > > `MAP_ANON' undeclared. > > > > > I have attached to the current e-mail the messages that I get > > > > when I > > > > > run the configure program (conflog file attachment) and the > > > > messages I > > > > > get when I run gmake (gmakelog attachement). > > > > > When I run configure I use -disable-mpi -disable-udp since I > > > > don't > > > > > need the udp or mpi conduits. > > > > > > > > > > Sincerely, > > > > > Konstantinos Kleisouris > > > > > > > > > > > > -- > > > > Paul H. Hargrove PHHargrove_at_lbl_dot_gov > > > > Future Technologies Group > > > > HPC Research Department Tel: +1-510-495-2352 > > > > Lawrence Berkeley National Laboratory Fax: +1-510-486-6900 > > > > > >