Ok, yes I had understood that the private statement produces local copies
of the subsequent variables for each thread, but do not need that indeed
here.
It seems to work now... thanks!
What does re-entrancy in case of MPIR correspond to (I understand this
usually means that a program can be interrupted and continued after another
program has been executed) - like if not all variables are cleared, and the
C program is ended and started up again?
I did not use MSVC for the build. From the documentation it seems that I
have built in re-entrant mode - how could I check this?
Thanks!
On Thursday, November 12, 2015 at 12:06:20 PM UTC+1, Bill Hart wrote:
> Ah, you are using Windows. I don't really know how to build in reentrant
> mode using MSVC. Hopefully Brian can help you with that. It might be the
> default, I don't know.
>
>
> On 12 November 2015 at 11:53, <highcal...@gmail.com <javascript:>> wrote:
>
>> Ok; how can I build it in re-entrant mode and check whether I might have
>> done so already (since the successful build took a long time...)?
>> Regarding the separation of memory regions: What is wrong with the
>> following code:
>>
>> int i, N = 10;
>> mpf_t a[N], b;
>> for (i=0; i<N; i++) { mpf_init(a[i]); }
>> mpf_init(b);
>>
>> #pragma omp parallel private(a, b)
>> {
>> for (i=0; i<N; i++) { mpf_set_d(a[i], (double) i); }
>> mpf_set_d(b, 1.0);
>>
>> #pragma omp for
>> for (i=0; i<N; i++) {
>>
>> mpf_add(a[i], a[i], b);
>> }
>> }
>>
>>
> I'm not sure what the private(a, b) does. Does it make a copy of a and b
> per thread. I'm not sure if that is what you want to do. I'm not certain it
> would make any difference anyway, unless it copies the whole of the array
> a, which it might.
>
> The program looks ok to me otherwise.
>
> Probably the issue is with not building in reentrant mode. The memory
> manager in MPIR needs to be reentrant.
>
> Bill.
>
> which again lets the program crash at runtime only.
>> And where is a good introduction to using pthreads?
>> Thanks a lot.
>>
>>
>> On Thursday, November 12, 2015 at 11:22:46 AM UTC+1, Bill Hart wrote:
>>>
>>> Hi,
>>>
>>> You need to build MPIR in reentrant mode.
>>>
>>> Also, you need to ensure that no two threads can write to the same MPIR
>>> variable at the same time. It's just like writing any other parallel
>>> program which uses data structures. There must be a separation of memory
>>> regions used by the different threads. This is harder to do with OpenMP
>>> than with pthreads directly.
>>>
>>> Bill.
>>>
>>> On 12 November 2015 at 11:07, <highcal...@gmail.com> wrote:
>>>
>>>> Dear Bill, Dear All,
>>>>
>>>> after using MPIR successfully for some time, I am looking for a faster
>>>> execution of my code and though of a parallel version. Is there a way to
>>>> do
>>>> it?
>>>>
>>>> I tried OpenMP by doing the following: In a toy example, I included
>>>> <omp.h>, gcc’ed with "–fopenmp", and along the following non-MPIR code:
>>>>
>>>>
>>>>
>>>> int i, nloops, thread_id=0, N = 1000000;
>>>>
>>>>
>>>>
>>>> #pragma omp parallel private(thread_id, nloops)
>>>>
>>>>
>>>>
>>>> {
>>>>
>>>> nloops=0;
>>>>
>>>>
>>>>
>>>> #pragma omp for
>>>>
>>>> for (i=0; i<N; i++) {
>>>>
>>>> nloops++;
>>>>
>>>>
>>>>
>>>> }
>>>>
>>>> thread_id = omp_get_thread_num();
>>>>
>>>> printf("Thread %d performed %d
>>>> iterations of the loop.\n", thread_id, nloops );
>>>>
>>>> }
>>>>
>>>>
>>>>
>>>> which resulted in the output:
>>>>
>>>> Thread 3 performed 250000 iterations of the loop.
>>>>
>>>> Thread 0 performed 250000 iterations of the loop.
>>>>
>>>> Thread 2 performed 250000 iterations of the loop.
>>>>
>>>> Thread 1 performed 250000 iterations of the loop.
>>>>
>>>>
>>>>
>>>> I then tried an MPIR version:
>>>>
>>>> int i, nloops, thread_id=0, N = 1000000;
>>>>
>>>> mpf_t a, b; mpf_inits(a, b, 0);
>>>>
>>>>
>>>>
>>>> #pragma omp parallel private(thread_id, nloops, a, b)
>>>>
>>>>
>>>>
>>>> {
>>>>
>>>> mpf_set_d(a, 0.0);
>>>>
>>>> mpf_set_d(b, 1.0);
>>>>
>>>>
>>>>
>>>> #pragma omp for
>>>>
>>>> for (i=0; i<N; i++) {
>>>>
>>>> mpf_add(a, a, b);
>>>>
>>>> }
>>>>
>>>> thread_id = omp_get_thread_num();
>>>>
>>>> gmp_printf("Thread %d performed %.*Ff
>>>> iterations of the loop.\n", thread_id, 10, a );
>>>>
>>>> }
>>>>
>>>> mpf_clears(a, b, 0);
>>>>
>>>>
>>>>
>>>> which compiled without errors but resulted in an error (“program …
>>>> stopped working during execution”) at runtime.
>>>>
>>>> Is it at all possible to run it in parallel with MPIR – and is there a
>>>> mistake in the implementation?
>>>>
>>>> Tests: It runs through without error if the “mpf_add” inside the
>>>> “#pragma for” loop is erased and b is set not after but before the
>>>> “#pragma
>>>> omp parallel” statement (but not if the “mpf_add” command is erased and b
>>>> is declared as above).
>>>>
>>>> I am using gcc and MinGW on a 64-bit (2 cores) Windows 7 machine.
>>>>
>>>>
>>>> Thanks and best regards!
>>>>
>>>>
>>>>
>>>> --
>>>> You received this message because you are subscribed to the Google
>>>> Groups "mpir-devel" group.
>>>> To unsubscribe from this group and stop receiving emails from it, send
>>>> an email to mpir-devel+...@googlegroups.com.
>>>> To post to this group, send email to mpir-...@googlegroups.com.
>>>> Visit this group at http://groups.google.com/group/mpir-devel.
>>>> For more options, visit https://groups.google.com/d/optout.
>>>>
>>>
>>> --
>> You received this message because you are subscribed to the Google Groups
>> "mpir-devel" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to mpir-devel+...@googlegroups.com <javascript:>.
>> To post to this group, send email to mpir-...@googlegroups.com
>> <javascript:>.
>> Visit this group at http://groups.google.com/group/mpir-devel.
>> For more options, visit https://groups.google.com/d/optout.
>>
>
>
--
You received this message because you are subscribed to the Google Groups
"mpir-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to mpir-devel+unsubscr...@googlegroups.com.
To post to this group, send email to mpir-devel@googlegroups.com.
Visit this group at http://groups.google.com/group/mpir-devel.
For more options, visit https://groups.google.com/d/optout.
