COPRTHR 2 Discussion

[nickoppen]

[nickoppen]

[sebraa]

You need to link against the eSDK as well. Try to add "-le-lib" to the linker flags.

[nickoppen]

This was weird! The -le-lib got rid of the e_dma_copy not found problem but not the __corenum_to_coreid not found problem.

Looking back at the coprthr2 examples, all of my switches were the same but the example used MPI_Sendrecv_replace and when I used this call instead of a Send() Recv() combination... everything was fine (even without the -le-lib).

I'll run with MPI_Sendrecv_replace for now.

PS the same problem showed up with MPI_Sendrecv (without the replace)

[nickoppen]

[mjvbhaskar1000]

Hello All,

I am new to parallella. I have downloaded and installed the latest headless installation and installed the required softwares.
When I try to build nbody_mpi or mpi-fft2d, etc i get the following error.
cc -O2 -DMPI_BUF_SIZE=128 -DMPI_BUF_SIZE=128 -I. -I/usr/local/browndeer/include -I/usr/local/browndeer/include/coprthr -c mpi_fft2d_host.c
cc -o mpi_fft2d_host.x mpi_fft2d_host.o -L/usr/local/browndeer/lib -lcoprthr -lcoprthrcc -lm
mpi_fft2d_host.o: In function `coprthr_mpiexec':
mpi_fft2d_host.c:(.text+0x56): undefined reference to `coprthr_ncreate'
collect2: error: ld returned 1 exit status
make: *** [mpi_fft2d_host.x] Error 1
Kindly help me.
Thanks in advance.

Regards,
Bhaskar M

[nickoppen]

One potential problem that I can see is that you are using cc rather than gcc as the compiler/linker.

Check your environment variables. The coprthr install files mostly come with a text file that says what needs to be changed. Everything needs to be exactly right before it all hangs together.

Other than that, it's a bit hard to diagnose problems like this via a forum post. Often it takes a bit of a "can opener" approach to figure it out.

[nickoppen]

Thinking about my post above regarding MPI_Sendrecv_replace()... I implemented a topology where no message had to travel further than to an adjacent core and ALL MESSAGES ARRIVED!

(The sent message was still not overwritten by the incoming message though... no surprises there)

[mjvbhaskar1000]

[mjvbhaskar1000]