Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
-
MigMuc
- Posts: 2
- Joined: Tue Oct 25, 2016 3:03 pm
Post
by MigMuc » Tue Oct 25, 2016 3:11 pm
I successfully compiled clmagma-1.3.0 and can run the examples. But I noticed that some of them
throw an error message
Code: Select all
OpenCL runtime error: CL_INVALID_COMMAND_QUEUE (-36) in magma_queue_sync at interface.cpp:255
.
This error message applies to the testing_[sdcz]swap and testing_[sdcz]transpose examples.
-
mgates3
- Posts: 918
- Joined: Fri Jan 06, 2012 2:13 pm
Post
by mgates3 » Tue Oct 25, 2016 5:28 pm
We're working on a new release of clMAGMA, and will look into this as part of that. It could be there's a bug in the tester.
-mark
-
MigMuc
- Posts: 2
- Joined: Tue Oct 25, 2016 3:03 pm
Post
by MigMuc » Tue Oct 25, 2016 6:19 pm
Thank you for your quick response. I am looking forward to work with an improved clmagma release. I tested already some functions in MAGMA for CUDA using the
Fortran interfaces. I would really appreciate if you could also implement the out-of-core (hybrid) solvers magma_[sdcz]getrf since the matrices are rather large and GPU
memory is a bottleneck.