Hi, Michael On Thu, Jun 4, 2015 at 4:28 PM, Michael Ferguson <[email protected]> wrote:
> Hi Hui - > > qthreads is the default now in most configurations. > thanks > > I don't know what 'bound threads' are... maybe you're asking > if they're allocated to particular cores (like with numactl)? > If that's your question - I think so - with qthreads anyway. > sorry actually I'm not sure what that means,either. I just found this instruction on PAPI, so I just want to know if Chapel threads satisfy this requirement: "PAPI only supports thread level measurements with *kernel or bound threads*, which are threads that have a scheduling entity known and handled by the operating system’s kernel. In most cases, such as with SMP or OpenMP complier directives, *bound* threads will be the default. Each thread is responsible for the creation, start, stop, and read of its own counters. When a thread is created, it inherits no PAPI information from the calling thread. " what do you think ? > > -michael > > On 6/4/15, 4:20 PM, "Hui Zhang" <[email protected]> wrote: > > >Hello, Michael > > > > > >1. I know Chapel 1.10 used fifo by default, how about Chapel 1.11 ? > > > >2. Are the threads in Chapel 'bound threads' ? > > > > > >thanks > > > > > > > >On Thu, Jun 4, 2015 at 1:31 PM, Michael Ferguson > ><[email protected]> wrote: > > > >Hi Hui - > > > >You might look at runtime/include/chpl-tasks.h, in particular > >chpl_task_startMovedTask and chpl_task_addToTaskList and friends. > > > >If you're using CHPL_TASKS=fifo, a method that works with > >pthreads *should* work... > > > >-michael > > > > > >On 6/4/15, 1:25 PM, "Hui Zhang" <[email protected]> wrote: > > > >>Hello, Michael > >> > >> > >>No, I'm actually trying to do that with my own code...I've had success > >>with explicitly-created threads like pthread, but for Chapel, I don't > >>know exactly the way threads get spawned > >> since it's implicit... > >> > >> > >> > >>On Thu, Jun 4, 2015 at 12:39 PM, Michael Ferguson > >><[email protected]> wrote: > >> > >>Hi Hui - > >> > >>Are you trying to use gprof? How are you monitoring the performance? > >> > >>AFAIK, gprof in particular just doesn't work with multiple threads... > >>I've heard that the profiler with gperftools works better, > >>but I don't have any experience with it myself. > >> > >>Cheers, > >> > >>-michael > >> > >>On 6/4/15, 12:13 PM, "Hui Zhang" <[email protected]> wrote: > >> > >>>Hello, > >>> > >>> > >>> > >>>I was trying to monitor the performance of a simple Chapel code, but > >>>even > >>>I don't use parallelism in the code(like forall, coforall...), just as > >>>simple as a single-thread C program, > >>> I found Chapel still creates a worker thread for the user code, and the > >>>master thread to initialize and finalize the program. Therefore, I want > >>>to know : > >>> > >>>1. When and how the worker thread is spawned ? > >>>2. What's your suggestion to monitor the worker thread since all I'm > >>>getting now is from the master thread, which isn't useful to me. > >>> > >>>3. Further, if parallelism is used(like forall,etc), is there any way to > >>>monitor all the threads ? > >>> > >>> > >>> > >>>Thanks > >>> > >>> > >>> > >>>-- > >>>Best regards > >>> > >>> > >>>Hui Zhang > >>> > >>> > >> > >> > >> > >> > >> > >> > >> > >> > >> > >>-- > >>Best regards > >> > >> > >>Hui Zhang > >> > >> > > > > > > > > > > > > > > > > > > > >-- > >Best regards > > > > > >Hui Zhang > > > > > > -- Best regards Hui Zhang
------------------------------------------------------------------------------
_______________________________________________ Chapel-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/chapel-developers
