clMAGMA 0.1 Beta Released

[admin]

clMAGMA 0.1 Beta is now available. This release provides OpenCL implementations for MAGMA's one-sided dense matrix factorizations (LU, QR, and Cholesky), and thus extending MAGMA's support to include AMD GPUs. Sample performance results on AMD Tahiti GPU are illustrated on the figures below.








Download the source here:
http://icl.cs.utk.edu/projectsfiles/magma/cl/clMAGMA_0.0.1.tar.gz

[fletchjp]

Is this only single precision in this release? Is there a plan for a release in double precision?

John

[wazoo42]

Looking at their overview page it would seem that double precision is available as well.

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

[wazoo42]

Looking at their overview page it seems that double precision is available as well. I'm interested to see what those benchmark results.

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

[mgates3]

All four precisions (single, double, single complex, double complex) are available in clMAGMA for QR and Cholesky. However, LU factorization is currently only in single precision. We intend to add support for other precisions but don't have a definite time table yet.
-mark

[Stan Tomov]

All four precisions for LU are now available in clMAGMA 0.2.
Stan

[adensmore]

Amazing what these MAGMA guys have done by producing this heterogeneous code, and porting it to OpenCL.

I succeeded in building clMAGMA v0.3 using VS2008, Windows SDK v6.0A, gfortran (GNU 4.4) and MKL 11, on WinXPHE 32bit using an AMD HD5850 GPU. The unix makefile commands were revised to suit DOS syntax. In magmawinthreads.h I added the lines "#define __midl 600", "#define __int3264 int", and "typedef unsigned char byte". In common_magma.h, '#include <limits.h>' changed to '#include "limits.h"'. In operators.h and magma_types.h the ".x" were all changed to ".s[0]", and ".y" changed to ".s[1]". In Makefile.internal: 1) ifeq / else ifeq / else ifeq / else / endif changed to ifeq / else ifeq / else ifeq / else / endif / endif / endif, and 2) PTRSIZE = $(shell $(CC) -nologo $(PTRFILE) > nul && $(PTREXEC) && del $(PTREXEC) > nul ), and 3) #commenting out the section regarding Plasma. In magma_types.h: "#if HAVE_CUBLAS" changed to "#ifdef HAVE_CUBLAS", and "#elif HAVE_clAmdBlas" to "#elif defined(HAVE_clAmdBlas)". In interface.cpp "__func__" changed to "__FUNCTION__". At the top of every .cpp testing code the line "[module(name="magma_<>")];" was added, where <> = sfortran, dfortran, cfortran, zfortran, or param. One instance of pseudo-code "or" in each testing code replaced with the c logical operator "||". To overcome a linker error (zdotc and cdotc both being defined in libmagma.a and mkl_intel_c.lib) I simply deleted the zdotc and cdotc module entry from the top of the zhet21.f and chet21.f files in the lin dir. In magma_lapack.h, lines 59-60, "*small" replaced with "*smal". Before building I installed AMD StreamSDK v2.3 and clAmdBlas v1.8, and got Stream and Blas to compile and run their respective samples. Building clMAGMA was done in a VS2008 command prompt window, which automatically sets most of the required path entries.

To get the "d" and "z" precision .cl codes to build I had to add the #pragma line (#pragma OPENCL EXTENSION cl_amd_fp64 : enable) as cited in the June 13, '12 posting at viewtopic.php?f=2&t=501&p=1613&hilit=clmagma#p1613 to the top of each "d" and "z" .cl source file in the interface_opencl dir. Some of the testing codes, in each precision type except "s" (e.g., dgesv, cgesv and zgesv), have to be edited to define the appropriate type of PRECISION in the #if defined(PRECISION_x) line that encases all the test code, or they won't run. The xgemm test codes -- Ifound them all to work with the argument line -M 1024 -N 1024 -K 1024.

I've posted more detail concerning my build here: viewtopic.php?f=2&t=727

If anyone would like a copy of my clMAGMA v1.0 build on WinXP (make cleanall), send me an email at adensmore@ucla.edu with subject line "Requesting clMAGMA source".

[admin]

All precisions are supported now. You can remove/ignore those print statements. Thanks for pointing it out. We will have clMAGMA 1.0 release on October 24, and this will be fixed.

[matusi143]

It has been about a year since the release of v1.0 of clMAGMA. Is there any development going on and if so how can I get involved?