Search found 26 matches

by luiceur
Tue Apr 30, 2013 8:40 am
Forum: User discussion
Topic: Error using magma_dsyevd_gpu when N < 14
Replies: 1
Views: 1348

Error using magma_dsyevd_gpu when N < 14

Hi there, I've been playing around with magma_dsyevd_gpu and I've found out that I get errors when N <14. ./testing_dsyevd_gpu -N 12 -JV MAGMA 1.3.0 device 0: Tesla C2050, 1147.0 MHz clock, 2687.4 MB memory, capability 2.0 testing_dsyevd -N 12 [-JV] [-JN] N ldda CPU Time(s) GPU Time(s) =============...
by luiceur
Tue Apr 02, 2013 5:58 am
Forum: User discussion
Topic: MAGMA weak performance SGEMM
Replies: 2
Views: 1306

Re: MAGMA weak performance SGEMM

Hope MAGMA beats that soon with a new release
by luiceur
Thu Mar 14, 2013 3:58 pm
Forum: User discussion
Topic: Running executable file, compiled with magma, in other machi
Replies: 3
Views: 3729

Re: Running executable file, compiled with magma, in other m

I have been able to statically compile my code, however I've notice the following: 1 - If the code is dynamically compiled to MKL , then both magmablas_dgemm and magma_dpotrf_gpu-magma_dpotri_gpu works fine. 2 - If the code is statically compiled to MKL: 2.1 - magmablas_dgemm works fine 2.2 - magma_...
by luiceur
Wed Mar 13, 2013 11:44 am
Forum: User discussion
Topic: Running executable file, compiled with magma, in other machi
Replies: 3
Views: 3729

Running executable file, compiled with magma, in other machi

Hi, We are trying to run an executable file succesfully compiled with magma. However, if the target machine does not have MKL loaded or installed, I've got shared libraries errors: error while loading shared libraries: libmkl_intel_lp64.so: cannot open shared object file: No such file or directory I...
by luiceur
Mon Feb 25, 2013 6:06 am
Forum: User discussion
Topic: SGEMM when beta=0
Replies: 5
Views: 3450

Re: SGEMM when beta=0

I believe I am experiencing something similar with sgemm. It is extremelly hard to reproduce the effect, but I think I finally spotted it. From time to time, and I've also noticed that frontend machine dependant, part of the matrix result of sgemm when beta = 0 are NaN. Should I zeroing the matrix o...
by luiceur
Tue Nov 20, 2012 6:53 am
Forum: User discussion
Topic: asynchronous magmablas_sgemm calls
Replies: 9
Views: 5103

Re: asynchronous magmablas_sgemm calls

Thanks Mark for your help. Just to clarify, because magmablas_gemm and cublas_gemm are both async, calling them: cudaSetDevice( 0 ); cublasSgemm( handle0, CUBLAS_OP_N, CUBLAS_OP_N, m, n0, k, alpha, A0, lda, B0, ldb, beta, C0, ldc ); cudaSetDevice( 1 ); cublasSgemm( handle1, CUBLAS_OP_N, CUBLAS_OP_N,...
by luiceur
Mon Nov 19, 2012 10:13 am
Forum: User discussion
Topic: asynchronous magmablas_sgemm calls
Replies: 9
Views: 5103

Re: asynchronous magmablas_sgemm calls

What if I use multiple GPUs? In theory using 2 GPUs will divide the time by two, is there any reason I would not take advantage of that? Could magmablas_sgemm access another GPUs addresses as it would happen with a CUDA kernel?
Cheers,
by luiceur
Fri Nov 09, 2012 7:26 am
Forum: User discussion
Topic: asynchronous magmablas_sgemm calls
Replies: 9
Views: 5103

Re: asynchronous magmablas_sgemm calls

I have managed to make them run on different streams setting magmablasSetKernelStream but I have not managed to make them run simultaneously i.e. they don't overlap. Ideally I would like them to overlap in time as they don't depend on each other. How could I do it? Do you have any ideas of how? cuda...
by luiceur
Thu Nov 08, 2012 5:27 am
Forum: User discussion
Topic: asynchronous magmablas_sgemm calls
Replies: 9
Views: 5103

Re: asynchronous magmablas_sgemm calls

Hi Mark, Thanks for the info, very useful. If it is possible to execute gemm with streams then, how can I indicate with which stream sgemm should be executed? The definition of magmablas_sgemm does not have anything for streams. I have to say that in our case, with matrices of 9000x24000 float eleme...
by luiceur
Wed Nov 07, 2012 7:53 am
Forum: User discussion
Topic: asynchronous magmablas_sgemm calls
Replies: 9
Views: 5103

asynchronous magmablas_sgemm calls

I am trying to execute 2 totally independent matrix calculations at the same time using magmablas_sgemm. I am thinking of using CUDA streams, however in order to be able to do it I believe that magmablas_sgem calls should be aynchronous otherwise the focus won't be returned to the CPU therefore it w...