Re: a strange problem when openmp with glib ghash

[Lex Trotman <elextr gmail com>]

Happy to help, but please ensure that you reply to the mailing list in
future, not to the person who answers your questions, so others can
also learn or help.

Cheers
Lex

On 22 September 2010 12:45, Qingpeng Niu <niuqingpeng gmail com> wrote:
> Hi, Lex
> Great you are right. If change to A it will works fine. B will not work.
>
> [A]
> #pragma omp parallel
>
> {
> #pragma omp for
>  for(1...100)
>
>  {
>  }
> }
>
> On Tue, Sep 21, 2010 at 10:33 PM, Lex Trotman <elextr gmail com> wrote:
>>
>> As I understand it, but thats a question for the omp mailing list.
>>
>> Cheers
>> Lex
>>
>> On 22 September 2010 12:18, Qingpeng Niu <niuqingpeng gmail com> wrote:
>> > Thank you for quick reponse. But I am a little confused about you
>> > mentioned.
>> > Do you mean the following two loop has difference and i should use the A
>> > but
>> > not B?
>> > [A]
>> > #pragma omp parallel
>> >
>> > {
>> > #pragma omp for
>> >  for(1...100)
>> >
>> >  {
>> >  }
>> > }
>> > [B]
>> > #pragma omp parallel for
>> >
>> > for(1..100)
>> > {
>> >
>> > }
>> >
>> >
>> >
>> > On Tue, Sep 21, 2010 at 10:08 PM, Lex Trotman <elextr gmail com> wrote:
>> >>
>> >> On 22 September 2010 11:39, Qingpeng Niu <niuqingpeng gmail com> wrote:
>> >> > Hi, Lex
>> >> > Thank you for responding me second time.  I am touched. Sorry i
>> >> > forgot
>> >> > one
>> >> > line.
>> >> > They are not shared by each thread.
>> >> > There is a
>> >> > gh=pdt_a[i].gh;
>> >> > in the second parallel for.
>> >> > Therefore they are seperated. But  only add with g hash table it will
>> >> > stop
>> >> > there. If i comment the g hash table code in second loop, it will not
>> >> > stop.
>> >> > I think it should be some problem with glib.
>> >>
>> >> omp parallel runs the whole block in a thread, ie each thread loops
>> >> through all the hashes, so they are shared.  Use omp for to make each
>> >> loop iteration a separate thread.
>> >>
>> >> Cheers
>> >> Lex
>> >>
>> >> > Qingpeng Niu
>> >> > On Tue, Sep 21, 2010 at 6:45 PM, Lex Trotman <elextr gmail com>
>> >> > wrote:
>> >> >>
>> >> >> On 22 September 2010 02:31, Qingpeng Niu <niuqingpeng gmail com>
>> >> >> wrote:
>> >> >> > Hi,
>> >> >> >
>> >> >> > I do a parallel open mp for which used shared data of
>> >> >> > g_hash_table.
>> >> >> > Program
>> >> >> > do fine with the first parallel for loop but the third parallel
>> >> >> > for
>> >> >> > loop
>> >> >> > program will stop there and never be entered.
>> >> >> > g_thread_init(NULL);
>> >> >> > ..............
>> >> >> > #pragma omp parallel for private(i) shared(pdt_a)
>> >> >> >         for(i=0;i<psize;i++)
>> >> >> >          {
>> >> >> >                .............
>> >> >> >                 pdt_a[i].gh   = g_hash_table_new(g_str_hash,
>> >> >> > compare_strings);
>> >> >> >                ............
>> >> >> >          }
>> >> >> > #pragma omp parallel for private(i)
>> >> >> > shared(pdt_a,pit_a,lines,psize,nbuckets)
>> >> >> >         for(i=0;i<psize;i++)
>> >> >> >         {
>> >> >> >                for....
>> >> >> >                    lookup = g_hash_table_lookup(gh, input);
>> >> >> >               .....................
>> >> >> >                     g_hash_table_insert(gh, data, p_data);
>> >> >> >               .....................
>> >> >> >
>> >> >> > g_hash_table_replace(gh,strdup(input),p_data);
>> >> >> >         }
>> >> >> >
>> >> >>
>> >> >> It is not clear from this code if the hash tables are shared between
>> >> >> threads or if each thread has its own.
>> >> >>
>> >> >> If they are shared then you need to do the locking, from Gthread
>> >> >> documentation : "After calling g_thread_init(), GLib is completely
>> >> >> thread safe (all global data is automatically locked), but
>> >> >> individual
>> >> >> data structure instances are not automatically locked for
>> >> >> performance
>> >> >> reasons. So, for example you must coordinate accesses to the same
>> >> >> GHashTable from multiple threads."
>> >> >>
>> >> >> If they are not shared then this is a question for the openmp ML not
>> >> >> the glib ML.
>> >> >>
>> >> >> Cheers
>> >> >> Lex
>> >> >>
>> >> >>
>> >> >> > #pragma omp parallel for private(i)   // this parallel for will
>> >> >> > not
>> >> >> > be
>> >> >> > entered. Program will stop here and halt.
>> >> >> >         for(i=0;i<psize;i++)
>> >> >> > --
>> >> >> > Qingpeng Niu
>> >> >> > Department of Computer Science and Engineering at OSU
>> >> >> >
>> >> >> > _______________________________________________
>> >> >> > gtk-list mailing list
>> >> >> > gtk-list gnome org
>> >> >> > http://mail.gnome.org/mailman/listinfo/gtk-list
>> >> >> >
>> >> >> >
>> >> >> _______________________________________________
>> >> >> gtk-list mailing list
>> >> >> gtk-list gnome org
>> >> >> http://mail.gnome.org/mailman/listinfo/gtk-list
>> >> >
>> >> >
>> >> >
>> >> > --
>> >> > Qingpeng Niu
>> >> > Department of Computer Science and Engineering at OSU
>> >> >
>> >
>> >
>> >
>> > --
>> > Qingpeng Niu
>> > Department of Computer Science and Engineering at OSU
>> >
>
>
>
> --
> Qingpeng Niu
> Department of Computer Science and Engineering at OSU
>