Hi,
MPI_Bsend() is buffered send depends a lot on the system providing
the buffer space needed for the call to complete and it fails if memory
was not available.
A blocking call works better as it blocks until the send
completes. In most cases blocking send will work at the same time giving
the same results as a buffered send.
You might want to try using blocking send for your program.
Nihar
On Thu, 18 Sep 2003, poknam wrote:
- Thank you very much!! It works now.
- Now I want to follow the logic of the program and write my own one. Here's
- the code:
-
- program main
- include 'mpif.h'
- parameter (BUFSIZE = 2048)
- integer ierr, rank
- integer tag, status(MPI_STATUS_SIZE)
- integer size
- real message(BUFSIZE)
- real BUFFER(2048)
- call MPI_INIT(ierr)
- call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
-
- tag = 0
- if (rank == 0) then
- print *, 'Hello world! I am proc ', rank,
- 1 ', sending to proc 1..'
- size = BUFSIZE * 4 + MPI_BSEND_OVERHEAD
- call MPI_BUFFER_ATTACH(BUFFER, size, ierr)
- call MPI_BSEND(message, BUFSIZE, MPI_REAL,
- 1 1, tag, MPI_COMM_WORLD, ierr)
- call MPI_BUFFER_DETACH(BUFFER, size, ierr)
- else if (rank == 1) then
- call SLEEP(3)
- print *, 'Hello world! I am proc ', rank,
- 1 ', just wake up!!'
- call MPI_RECV(message, BUFSIZE, MPI_REAL,
- 1 0, tag, MPI_COMM_WORLD, status, ierr)
- print *, 'Hello world! Received 1 message ',
- 1 'proc 1!!'
- endif
- print *, 'Proc ', rank, ' finished!!'
- call MPI_FINALIZE(ierr)
- end
-
- Runtime error occur, I've tried changing the value of "size" or the
- dimension of "BUFFER", but the result is the same. Here's the output:
-
- Hello world! I am proc 0, sending to proc 1..
- MPI_Bsend: unclassified: No buffer space available (rank 0, MPI_COMM_WORLD)
- Rank (0, MPI_COMM_WORLD): Call stack within LAM: Rank (0, MPI_COMM_WORLD):
- - MPI_Bsend() Rank (0, MPI_COMM_WORLD): - main()
-
- Can anyone tell me why?? Million thanks.
-
- P.N. Lai
-
- -----Original Message-----
- From: lam-bounces_at_[hidden] [mailto:lam-bounces_at_[hidden]] On Behalf Of
- Nick Nevin
- Sent: Thursday, September 18, 2003 8:19 PM
- To: General LAM/MPI mailing list
- Subject: Re: LAM: SIGSEGV in MPI_Buffer_detach
-
- On Thu, Sep 18, 2003 at 06:40:29PM +0800, poknam wrote:
- > This is a very simple program downloaded from web,
- >
- > But there's runtime error in MPI_BUFFER_DETACH,
- >
- > I've tried other fortran programs using MPI_BUFFER_DETACH, still a
- > similar error occurs.
- >
- > But the error ONLY occur in Fortran, C has no problem.
- >
- > Also, I've tried comment the line with MPI_BUFFER_DETACH, then it runs
- > smoothly
- >
- > Below is the code:
- >
- >
- [snip]
- >
- > call MPI_BUFFER_DETACH(buffer,528,ierr)
-
- You need to pass a variable not a constant as the second argument. For
- example
-
- call MPI_BUFFER_DETACH(buffer,k,ierr)
-
- -nick
- _______________________________________________
- This list is archived at http://www.lam-mpi.org/MailArchives/lam/
-
-
- _______________________________________________
- This list is archived at http://www.lam-mpi.org/MailArchives/lam/
-
|
