first of all, I have been using MAGMA for the last year and I'd like to thanks the team for the great library they have developped.
My company has recently invested in a dedicated server with GPU capabilities and we have decided to go for 2 AMD Radeon 7970 and start using clMagma.
After a few issues with setting up the Radeon cards, I have finally managed to get them to respond, compile clMagma using ATLAS and AMD's OpenCL.
Then moving on to testing clMagma I was unable to run any of the test programs (from example testing_dgesvd) has clMagma was unable to load any of the Opencl Kernels.
I ended up in CL_MAGMA_RT.cpp and in the function BuildFromBinaries I had to tweek the call to clCreateProgramWithBinary which was catching both the error status and error code in the variable.
I don't know whether this is a known bug.
testing_dgesvd also assumed that only 1 GPU device is present. Is there any fundamental reason why clMagma should not be used on a machine with several GPUs?
My last question is now that testing_dgesvd is running, it is increadibly slow. Has anybody come accross this kind of problem on the Radeon 7970?