| Commit message (Collapse) | Author | Age |
|
|
|
| |
PiperOrigin-RevId: 172895297
|
|
|
|
|
|
| |
with tf.linalg.adjoint(foo) or tf.transpose(foo, conjugate=True), and clean up a few places that can avoid explicit adjoints as inputs to matmul.
PiperOrigin-RevId: 172027859
|
|
|
|
|
|
|
|
|
|
|
|
| |
This way users of the class don't have to remember to capture each one manually to avoid premature deallocation and memory races for asynchronous op kernels.
* Add simple tests that run multiple ops concurrently for linalg ops that use CudaSolver.
* Put a lock around the calls to cusolverDn*getrs and cusolverDn*gesvd, which appear not to be threadsafe.
* Misc. cleanup in linalg GPU kernels.
I ran all the related tests 1000 times without failure. Before this change, tests for matrix_solve and svd would fail or hang occasionally.
PiperOrigin-RevId: 170557380
|
|
|
|
| |
PiperOrigin-RevId: 169329120
|
|
|
|
|
|
| |
Remove support support for on-the-fly transpose in internal matrix_band_part functor recently added (in anticipation of using it for QR), since it turned out to not be useful.
PiperOrigin-RevId: 169249336
|
|
|
|
| |
Change: 142080137
|
|
Change: 139959769
|