The main thread does the setup work, including creating the
communicator. Only after this is completed does the worker thread begin
its work, and while the worker thread is working, the main thread does
not make any MPI calls. Therefore, even though I am using 2 threads,
there are never any simultaneous calls of MPI functions. I also read
section 12.4 of the user's manual, but it sheds no light on why
MPI_Irecv becomes a blocking call if I call it from the worker thread
instead of the main thread.
I found references to another threading problem in the archives which
has some similarities to mine, except that the end result is a seg
fault, not the conversion of a nonblocking call to a blocking call.
This is discussed at
http://www.lam-mpi.org/MailArchives/lam/msg06479.php and
http://www.lam-mpi.org/MailArchives/lam/msg06505.php ("This is a LAM
build bug to do with threads and errno") etc. The solution was
recompilation using the Sun compilers with a special CFLAGS setting.
Before going to all this work, I would like to know if it is really
going to help.
I would expect LAM/MPI to work in this scenario because it it very
common in GUI-based applications to have a worker thread do
time-consuming work so that the main thread can keep the GUI refreshed.
For example, it is very common to use a main and worker thread so that
the worker thread can report its progress to a global variable and have
the main thread read the progress value and display it in a progress
bar.
-----Original Message-----
From: Nihar Sanghvi [mailto:nsanghvi_at_[hidden]]
Sent: Thursday, September 18, 2003 7:15 PM
To: General LAM/MPI mailing list
Subject: Re: LAM: MPI_Irecv() blocks when called from a pthread
Hi Richard,
From your email it is not clear if you are trying to use MPI in
different threads at the same time.
Currently, LAM does not support multi-threading. The only
multi-threaded support it provides is that it puts a lock on the MPI
library. So, if your code is expecting to run multiple threads in MPI
simultaneously, it won't work as you would want it to.
Whatever is your first call to MPI Library will be blocking and
MPI_Irecv wont be able to enter MPI library until the first call
returns.
You could find more details about this in the User Manual at
http://www.lam-mpi.org/download/files/7.0-user.pdf in section 12.4.
Nihar
On Thu, 18 Sep 2003, Bailey, Richard T (US SSA) wrote:
- I developed a C++ app w/ a master/slave architecture for splitting an
- array in the master and having the slaves each process part of the
- array. The master sets up the slaves, including the communicator,
when
- the master object is constructed. The Master->ProcessArray() member
- splits the array and, depending on input params either does or does
not
- create a separate thread to send and receive the subarrays from the
- previously-established slaves. In either case, I use nonblocking
- MPI_Irecv() and MPI_Isend() calls for master/slave communication. When
I
- do not use a separate thread, all is well. However, if I do use a
- separate thread (so the client code can do other work while the thread
- spawned by Master->ProcessArray() is running), then MPI_Irecv()
blocks,
- with the effect that only 1 slave ever gets work and all processing is
- sequential. I put 1 second sleeps in the slaves and put fprintfs
around
- the code to verify the it is the MPI_Irecv() that is causing the
- problem:
-
-
-
- fprintf(
-
- stderr,
-
- "V: Starting AsyncRecvSubarrayParamsOutFromSlave %d @T
%8.3f\n",
-
- TileJob->SlaveStatus->Rank,
-
- GetSecs()
-
- );
-
-
-
- MPI_Irecv(
-
- & TileJob->SubarrayParamsOut,
-
- sizeof(SubarrayParamsOutType) / sizeof(int),
-
- MPI_INT,
-
- TileJob->SlaveStatus->Rank, // Source
-
- SubarrayParamsOutTag,
-
- AllComms,
-
- & TileJob->SlaveStatus->MPIRequest
-
- );
-
-
-
- fprintf(
-
- stderr,
-
- "V: Done w/ AsyncRecvSubarrayParamsOutFromSlave %d @T
%8.3f\n",
-
- TileJob->SlaveStatus->Rank,
-
- GetSecs()
-
- );
-
-
-
- Here is the output:
-
- V: Starting AsyncRecvSubarrayParamsOutFromSlave 1 @T 1.755
-
- V: Done w/ AsyncRecvSubarrayParamsOutFromSlave 1 @T 2.763
-
-
-
- The same output from a non-threaded run shows that MPI_Irecv() is
- nonblocking.
-
-
-
- I am running LAM v. 7.0 compiled w/ gcc 2.95.3 on a Sun workstation OS
- 5.7. The thread is created with pthread_create(). I tried
initializing
- MPI 3 different ways with the same result:
-
- MPI_Init(NULL, NULL);
-
- MPI_Init_thread(NULL, NULL, MPI_THREAD_SINGLE, &DontCare);
-
- MPI_Init_thread(NULL, NULL, MPI_THREAD_SERIALIZED, &DontCare);
-
-
-
- What do I need to do to get this to work in the multithreaded mode?
-
-
-
-
-
- R. T. Bailey
-
-
---------------------------------------
Nihar Sanghvi
LAM-MPI Team
Graduate Student (Indiana University)
http://www.lam-mpi.org
--------------------------------------
_______________________________________________
This list is archived at http://www.lam-mpi.org/MailArchives/lam/
|
