Is there an example of coprthr MPI spanning multiple hosts?

[capnrob97]

Can someone point me in the right direction of MPI code using libcoprthr that runs on more than one parallella?

Thanks!

[aolofsson]

I could be wrong, but I think you would need to use a hybrid approach for that. Use something like OpenMPI for multiple parallellas and then the coprthr one within a node.
Andreas

[capnrob97]

[capnrob97]

I was able to get that to work

Now to get all 64 cores in the cluster doing some real work.

17: Message from Host parallella2 eCore 0x003 - Rank 2123 ( 1, 3): "Hello World from core 0x88a!"
17: Message from Host parallella3 eCore 0x005 - Rank 2123 ( 1, 3): "Hello World from core 0x88a!"
17: Message from Host parallella2 eCore 0x002 - Rank 2123 ( 1, 3): "Hello World from core 0x88a!"
17: Message from Host parallella3 eCore 0x004 - Rank 2123 ( 1, 3): "Hello World from core 0x88a!"
17: Message from Host parallella1 eCore 0x000 - Rank 2123 ( 1, 3): "Hello World from core 0x88a!"
17: Message from Host parallella4 eCore 0x006 - Rank 2123 ( 1, 3): "Hello World from core 0x88a!"
17: Message from Host parallella4 eCore 0x007 - Rank 2123 ( 1, 3): "Hello World from core 0x84b!"
18: Message from Host parallella1 eCore 0x001 - Rank 2120 ( 1, 0): "Hello World from core 0x84b!"
18: Message from Host parallella2 eCore 0x003 - Rank 2120 ( 1, 0): "Hello World from core 0x84b!"
18: Message from Host parallella3 eCore 0x005 - Rank 2120 ( 1, 0): "Hello World from core 0x84b!"
18: Message from Host parallella2 eCore 0x002 - Rank 2120 ( 1, 0): "Hello World from core 0x84b!"
18: Message from Host parallella3 eCore 0x004 - Rank 2120 ( 1, 0): "Hello World from core 0x84b!"
18: Message from Host parallella1 eCore 0x000 - Rank 2120 ( 1, 0): "Hello World from core 0x84b!"
18: Message from Host parallella4 eCore 0x007 - Rank 2120 ( 1, 0): "Hello World from core 0x84b!"
18: Message from Host parallella4 eCore 0x006 - Rank 2120 ( 1, 0): "Hello World from core 0x848!"
19: Message from Host parallella1 eCore 0x001 - Rank 2250 ( 3, 2): "Hello World from core 0x848!"
19: Message from Host parallella2 eCore 0x003 - Rank 2250 ( 3, 2): "Hello World from core 0x848!"
19: Message from Host parallella3 eCore 0x005 - Rank 2250 ( 3, 2): "Hello World from core 0x848!"
19: Message from Host parallella2 eCore 0x002 - Rank 2250 ( 3, 2): "Hello World from core 0x848!"
19: Message from Host parallella3 eCore 0x004 - Rank 2250 ( 3, 2): "Hello World from core 0x848!"
19: Message from Host parallella1 eCore 0x000 - Rank 2250 ( 3, 2): "Hello World from core 0x848!"

[aolofsson]

Very nice!!

[MiguelTasende]

[aolofsson]

Nice! Can't wait to see what you come up with for BLAS and linpack!

[capnrob97]

Calculating Pi to 250,000 digits.

Currently just using openMPI and the 8 ARM cores in the cluster.

Next step is to get all 64 Epiphany cores involved as well.

Still running, started it last night, it gets slower the farther you move to the right of the decimal.

Digits are in hex.

Each core does every 8th hex digit, using MPI send / receive to make sure they display in proper order.

[capnrob97]

Finished about 14 hours later.

Will be interested in seeing the speedup with the Epiphany cores and if the results change using floats instead of doubles that this code uses.

[aolofsson]

Very cool! What method are you using to calculate pi?