Search found 912 matches

by mgates3
Wed May 30, 2012 10:11 am
Forum: User discussion
Topic: OpenCL GEMM routines
Replies: 1
Views: 1813

Re: OpenCL GEMM routines

We've worked on auto-tuned CUDA GEMM routines, which are available in the CUDA MAGMA 1.2 distribution. For OpenCL, we use AMD's BLAS.

-mark
by mgates3
Wed May 30, 2012 10:04 am
Forum: User discussion
Topic: Using MAGMA with CUDA/PGI
Replies: 7
Views: 6789

Re: Using MAGMA with CUDA/PGI

How are you allocating memory on the GPU? To use the GPU interface, magma_dgesv_gpu, at some point you need to call cudaMalloc to allocate GPU memory, which I don't see in your code snippets. If you use the CPU interface, magma_dgesv, then it expects matrices in host memory (and internally copies th...
by mgates3
Wed May 30, 2012 9:49 am
Forum: User discussion
Topic: how to make it work with python
Replies: 1
Views: 2881

Re: how to make it work with python

Unfortunately, we don't have Python bindings for MAGMA. However, the CPU interface is straight-forward C code, so you should be able to build wrappers without too much difficulty, e.g., using SWIG (http://www.swig.org/). Just be aware that matrices are stored in column-wise order. If you use numpy t...
by mgates3
Wed May 30, 2012 9:45 am
Forum: User discussion
Topic: zheevd performance: GPU == CPU??
Replies: 3
Views: 2269

Re: zheevd performance: GPU == CPU??

The GT-430 is a consumer level card primarily intended for graphics applications like games. It's performance is fairly low. Consider that a Tesla 2050 achieves 340 Gflop/s on a zgemm, compared to the 20 Gflop/s you are reporting. In a quick test, I get 14 Gflop/s with a zgemm on 2 CPU cores (depend...
by mgates3
Mon May 21, 2012 12:24 pm
Forum: User discussion
Topic: MAGMA SGEMM Results
Replies: 3
Views: 2722

Re: MAGMA SGEMM Results

Yes, the SGEMM test should work. You can try a small test case and print out the results to see what is going on. E.g., add: printf( "Magma C=" ); magma_sprint( M, N, h_C, ldc ); printf( "Cublas C=" ); magma_sprint( M, N, h_C2, ldc ); before the Error computation in testing_sgemm.cpp, then run with ...
by mgates3
Mon May 21, 2012 11:30 am
Forum: User discussion
Topic: Problem with testing_zgesv
Replies: 12
Views: 9245

Re: Problem with testing_zgesv

Memory mapping errors usually happen when you pass a device pointer where it expects a CPU pointer, or vice-versa, or any other time the pointer is invalid. They are the generic seg-fault for GPU code. What is your make.inc file? I noticed in an earlier post that you used Intel icc. Perhaps try with...
by mgates3
Wed May 16, 2012 2:08 pm
Forum: User discussion
Topic: Incorrect LAPACK/BLAS uppercase symbols
Replies: 4
Views: 2425

Re: Incorrect LAPACK/BLAS uppercase symbols

Yes, we code most functions in double complex and automatically convert to the other precisions. The inconsistencies for uppercase names are an oversight; I will fix our automated script to convert these.

-mark
by mgates3
Tue May 15, 2012 6:58 pm
Forum: User discussion
Topic: clMAGMA problems with CUDA
Replies: 4
Views: 2262

Re: clMAGMA problems with CUDA

I don't understand what you mean by "the CUDA version of OpenCL". CUDA and OpenCL are two different languages to program GPUs. Do you mean the NVIDIA version of OpenCL? If so, then it would be using OpenCL, not CUDA, on an NVIDIA card. NVIDIA does not release BLAS for OpenCL. If you want to use CUBL...
by mgates3
Tue May 15, 2012 6:39 pm
Forum: User discussion
Topic: MAGMA 1.2 Released
Replies: 7
Views: 14816

Re: MAGMA 1.2 Released

I uploaded a new tar file for the MAGMA 1.2 release. This corrects the inclusion of some files in the tar file. This resolves issues with zhetrd_mgpu, zlatrd_mgpu, dgetmatrix, dtrsm, etc. noted previously.

http://icl.cs.utk.edu/magma/software/

-mark
by mgates3
Tue May 15, 2012 3:49 pm
Forum: User discussion
Topic: MAGMA 1.2 Released
Replies: 7
Views: 14816

Re: MAGMA 1.2 Released

Yes, a few issues with the MAGMA 1.2 release have been corrected. I will post a new tar file hopefully later today. The [sdcz]hetrd_mgpu.cpp and [sdcz]latrd_mgpu.cpp files were not intended to be released yet. They should have been removed from the Makefile as well. The [sdc] versions of wrappers we...