Search found 878 matches

by mgates3
Fri Aug 16, 2019 7:56 am
Forum: User discussion
Topic: Using sgemm for rectangular (non-square) matrix multiply
Replies: 7
Views: 6735

Re: Using sgemm for rectangular (non-square) matrix multiply

First, in the future, please start a new topic, rather than replying to a 7 year old topic. Your variable names seem mixed up: there are M rows and N cols in C. You are adding beta*C multiple times. Normally I use i for rows, j for cols. Unfortunately k is used for dimension, so we need something el...
by mgates3
Mon Aug 12, 2019 12:26 pm
Forum: User discussion
Topic: ILP64 name-mangling
Replies: 4
Views: 54

Re: ILP64 name-mangling

On second glance, it doesn't appear that CFLAGS is propagated to CXXFLAGS by CMake, and it won't propagate to Fortran or NVCC for CUDA. Even worse, it doesn't seem that CMake picks up NVCCFLAGS from the environment, so there doesn't appear to be a way to override or append to it, without editing CMa...
by mgates3
Sun Aug 11, 2019 9:13 am
Forum: User discussion
Topic: ILP64 name-mangling
Replies: 4
Views: 54

Re: ILP64 name-mangling

That should work. Ideally, CMake would detect 64-bit BLAS, which we do in some other projects (BLAS++) but not yet in MAGMA.
-mark
by mgates3
Fri Aug 09, 2019 4:55 pm
Forum: User discussion
Topic: ILP64 name-mangling
Replies: 4
Views: 54

Re: ILP64 name-mangling

If you compiled without CMake, then editing magma_mangling.h as you describe should work. But as you noticed, CMake #defines its own mangling macro, called MAGMA_GLOBAL, in the file magma_mangling_cmake.h. The quickest hack to do would be comment out magma_mangling_cmake (or #undef MAGMA_GLOBAL), an...
by mgates3
Mon Aug 05, 2019 12:55 pm
Forum: User discussion
Topic: Fault injection
Replies: 0
Views: 31

Re: Fault injection

That seems fine for testing. MAGMA doesn't do anything to detect or prevent faults from bit flips. If the bit happens to be low-order (like 0.00000001), then the difference will be negligible, but if the bit happens to be high-order (like 1.) or an exponent bit, the error is likely to be significant...
by mgates3
Fri Jul 19, 2019 9:56 pm
Forum: User discussion
Topic: matrices i.e. zgbsv and dsbgv?
Replies: 1
Views: 59

Re: matrices i.e. zgbsv and dsbgv?

We don’t have any plans for band matrices. Since they are closer to being memory bandwidth limited, they would not accelerate well on the GPU.
-Mark
by mgates3
Wed Jul 17, 2019 11:14 am
Forum: User discussion
Topic: What could be wrong ?
Replies: 1
Views: 85

Re: What could be wrong ?

This looks like portions of the really old MAGMA testers. I suggest starting from the code in the current MAGMA testers (though there's a lot you can strip out of those as being testing code). Please be more explicit about what went wrong. Do you mean the matrix A was all zeros on output? What was i...
by mgates3
Wed Jul 17, 2019 11:04 am
Forum: User discussion
Topic: *_gpu versions of the gesvd routines
Replies: 1
Views: 63

Re: *_gpu versions of the gesvd routines

No, there isn't any particular reason, svd_gpu routines just haven't been developed. Thanks for the request; we'll consider them for future developments.
-mark
by mgates3
Sun Jul 07, 2019 1:52 pm
Forum: User discussion
Topic: installation problems Magma 2.5
Replies: 2
Views: 274

Re: installation problems Magma 2.5

Have you tried the latest MAGMA 2.5.1 alpha 1? It should resolve many compiling issues from 2.5.
https://icl.utk.edu/magma/software/index.html
Mark
by mgates3
Wed Jul 03, 2019 12:30 am
Forum: User discussion
Topic: dense eigensolvers?
Replies: 1
Views: 111

Re: dense eigensolvers?

For complex matrices: magma_[cz]heevd* uses divide-and-conquer magma_[cz]heevr* uses MRRR magma_[cz]heevx* uses QR iteration if MagmaRangeAll, or bisection if MagmaRangeV or MagmaRangeI. For real matrices, only divide-and-conquer is available: magma_[sd]syevd* BTW, MAGMA is at version 2.5, not 2.0. ...