* We need to look at Maven, and Nexus as use cases and figure out what each of them needs to be
 * able to do
 *
 * - now what is really the difference between this and loading a component in plexus - custom
 * classloading capability - remote resolution of dependencies - do we want a model for sharing
 * information among plugins, is this more like an extension point - do we need a sort of bus for

# This is the same as the official TensorFlow builds.
# See https://developer.nvidia.com/cuda-gpus#compute
# `compute_XY` enables PTX embedding in addition to SASS. PTX
# is forward compatible beyond the current compute capability major
# release while SASS is only forward compatible inside the current
# major release. Example: sm_80 kernels can run on sm_89 GPUs but
# not on sm_90 GPUs. compute_80 kernels though can also run on sm_90 GPUs.

  EXPECT_EQ(4, product[1]);
  EXPECT_EQ(6, product[2]);
  EXPECT_EQ(8, product[3]);

  // When we turn on grappler, confirm that the tf.Add has been rewritten into a
  // tf.Mul.
  // This capability of checking the executed op names is currently only enabled
  // for TFRT debug build, for performance and simplicity reasons.
  if (use_tfrt) {
    TF_Buffer* buf = TF_NewBuffer();

* Fix `cstring.h` missing file issue with the Libtensorflow archive.

# Release 2.17.0

## TensorFlow

### Breaking Changes

* GPU
    * Support for NVIDIA GPUs with compute capability 5.x (Maxwell generation) has been removed from TF binary distributions (Python wheels).

### Major Features and Improvements

* **[stable]** Volume plugin for ScaleIO enabling pods to seamlessly access and use data stored on Dell EMC ScaleIO volumes.
* **[stable]** Volume plugin for Portworx added capability to use [Portworx](http://www.portworx.com) as a storage provider for Kubernetes clusters. Portworx pools server capacity and turns servers or cloud instances into converged, highly available compute and storage nodes.