MAGMA Forum

Have you ever tried libsci_acc bundled with Cray OpenACC compiler?

It appears transpose routine looks OK. I am also looking into magmablas_zpermute_long2 to find any out-of-bound memory access. Just for clarification. The error occurs when: M=3584 LDA=3616 Input matrix starts at A(32,32). I am still seeing wrong answer if A and X are allocated in contiguous memory ...

I am still seeing wrong answer if A and X are allocated in contiguous memory as described below. I suspect that zinplace_tranpose is making out-of-bound memory access when M!=LDA. In my application, I finally get the correct answer after changing the source to apply zinplace_transpose for LDA=M only...

It appears zhetrd_mgpu.cpp and zlatrd_mgpu.cpp are missing. You could recover them from chetrd_mgpu.cpp and clatrd_mgpu.cpp, but I'd like to Magma developers team to fix this problem. Also, there are several undefined reference for some data transfer and BLAS kernel calls such as magma_dgetmatrix an...

Hi, I am testing dgetrf_ooc routine using testing_dgetrf.cpp. Then, I got inaccurate answer. See the output below. In the tester source, I call dgetrf_ooc and dgetrf_piv in sequence. Is there a bug in dgetrf_ooc? #if 0 magma_dgetrf( M, N, h_R, lda, ipiv, &info); #else magma_dgetrf_ooc( M, N, h_R, ld...

CUBLAS's old API does not guarantee thread safety, and the CUBLAS4.0 or higher document recommends using Version 2 API with stream created by each individual CPU thread. Current version of Magma relies on CUBLAS's old API and thay may break the execution order of kernels in a few calls using multipl...

Stan,

I can see the same problem with sgetrf_gpu, dgetrf_gpu and cgetrf_gpu.

Thanks,

Keita

Hi, I have seen strange error with zgetrf_gpu (wrong answer or segfault with memory calls after the routine) on Cray XK (Fermi+). The error occurs when (1) M=N, (2) M is multiple of 32, and (3) LDA is mulitple of 32, but not equal to M. I thought this could be a bug in zinplace_tranpsose. So I chang...