MAGMA Forum

The weird part is, MAGMA is correct relative to cuBLAS, but wrong if compared to BLAS. There are also numerous error messages like BLAS : Bad memory unallocation! : 1024 0x2000c0000000 with the pointer address changing every time. Is this an issue with the testing routine, or with OpenBLAS? This co...

The y-dimension has the same limit as the z-dimensions. Some kernels do use the X dimension for batching, but most of our code use the x-y dimensions for thread configurations. Changing that would be cumbersome because of the many kernels MAGMA has. BTW, if you problem size is fixed at 3x3, you can ...

Most of the batch kernels in MAGMA use the z-dimension of the kernel grid for batching across different problems. The z-dimension has a hardware limit of 65535. That is why the routine fails. I think that for your particular case, the error comes from dgetrs. The batch dgetrf routine does not have t...

There is a batch routine for solving many small linear systems (magma_Xgesv_batched), where X specifies the precision (s, d, c, z). The routine applies LU factorization with partial pivoting to the input matrices, followed by a row interchanges step and triangular solves. For very small matrices, th...

Most of the MAGMA batched kernels use the z-dimension of the kernel grid for batching across different problems. The maximum value of this dimension is 65535 (a hardware limitation of the GPU itself). Depending on the kernel configuration, a large batch size may lead to exceeding this limit. If you ...

You should add your new CUDA file to the Makefile.src file under the magmablas subdirectory.

Ahmad

Hi Vishwak, Thank you for reporting this. I have committed a fix for reporting the correct info. Please check it out and let me know. One of the reasons was NaN checking, which was missing from the panel factorization kernel. The way the kernel was written made it encounter NaNs when factorizing a c...

Can you please give some details about your specific test case. I have tried some simple tests on my side, which seem to be always detected. Sometimes the value of info is not returned correctly (a non-zero value, but pointing to the wrong column/diagonal) but the routine always catches the error. T...

Great, thanks for letting us know.

--Ahmad

This should be fixed now if you update your local copy of the repository. Please test on your side and let us know if the issues are gone.

--Ahmad