From: Costin Iancu (cciancu_at_lbl_dot_gov)
Date: Thu Dec 13 2007 - 10:20:51 PST
This man page should answer your question, look for -shared-heap http://upc.lbl.gov/docs/user/upcrun.html Regards, Costin > Hi all, > And dear Hargrove I am thankful for your reply . I did not > recieve > the reply for my previous problem regarding native compiler flags on > my email address, instead I read the reply on upc mailing archive. > That problem is solved, and also I want to say that we are using > myrinet with myrinet express (mx) library, although we have a giga > bit ethernet as well. We are using MPI conduit because UPC does not > support mx library. > Now I am facing another problem of "Out of shared memory" run > time > error. Our cluster consists of 4 v890 servers connected via myrinet > and each server has 32 GB of main memory. I am writing the code > given below, it works for 2048x2048 grid size. It also works well > with 4096x4096 and 8 threads accros the cluster nodes. The same code > gives error for 16 threads using the grid size 4096x4096. I dont > know what am I doing wrong here, suggestions will be appreciated. > The code is gives below: > > #define size_x 4096 > #define size_x 4096 > > shared [size_x*size_y/THREADS] double *ez; > shared [size_x*size_y/THREADS] double *hy; > shared [size_x*size_y/THREADS] double *hx; > > ez=(shared [size_x*size_y/THREADS] double*) > upc_all_alloc(THREADS,size_x*size_y*sizeof(double)); > > ez=(shared [size_x*size_y/THREADS] double*) > upc_all_alloc(THREADS,size_x*size_y*sizeof(double)); > > ez=(shared [size_x*size_y/THREADS] double*) > upc_all_alloc(THREADS,size_x*size_y*sizeof(double)); > > The code is compiled and run with the following commands: > > /home/aftab/berkeley_upc-2.6.0/upcc "-Wc,-xO4" -T16 --network=mpi > -pthreads=4 -shared-heap=256MB 2dfdtd-pri.upc -o 2DFDTD > > /home/aftab/berkeley_upc-2.6.0/upcrun -n16 2DFDTD > > The error message is given as: > > UPC Runtime error: out of shared memory > Local shared memory in use: 0 MB per-thread, 0 MB total > Global shared memory in use: 128 MB per-thread, 2060 MB total > Total shared memory limit: 256 MB per-thread, 0 MB total > upc_all_alloc unable to service request from thread 0 for 134225920 > more > bytes > > > > > > -- > This message has been scanned for viruses and > dangerous content by MailScanner, and is > believed to be clean. > -- Costin C. Iancu cciancu_at_lbl_dot_gov Future Technologies Group Phone: 510-495-2122 Lawrence Berkeley National Laboratory Fax: 510-486-6900