tree 673d5a59a10dba8100e521434faf442396cbe09b
parent e7bd72d41e638107d0a4cccd62513cf456e18810
author Sameer Agarwal <sameeragarwal@google.com> 1684167214 -0700
committer Sameer Agarwal <sameeragarwal@google.com> 1684178738 -0700

Use page locked memory in BlockSparseMatrix

If using CUDA_SPARSE for an iterative solve on the GPU,
allocate the values array in BlockSparseMatrix to make copying
to the GPU faster.

Change-Id: I63c1d2512babd74fc275b277ac8c3eabf3ec1144