diff options
| author |  Sanjoy Das <sanjoy@google.com> | 2018-08-03 15:21:37 -0700 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2018-08-03 15:26:07 -0700 |
| commit | 4d58bfb2324dbc647b294824e29ba9e75eece3ba (patch) | |
| tree | 9e7ca892ae12004930a9a90ba2dc9fc750684ed3 /tensorflow/docs_src | |
| parent | 8743b3f21f8877d3d61f0c7bcfec021fbdb4907c (diff) | |
[XLA:CPU] Migrate aot/runtine.{h,cc} to xla_compiled_cpu_function.{h,cc}
As a follow-on cleanup for cl/206980796 ("Overhaul XLA:CPU's calling
convention.") I want to introduce a BufferInfo class that encapsulates whether a
buffer is a constant, an entry parameter or a temp without using the fragile
"size < 0" scheme I have today. To do this efficiently I need a place to put
the BufferInfo class that will be visible to MallocContiguousBuffers. Instead
of creating (what seemed to me) an odd layering with BufferInfo in aot/runtime.h
I decided to pull in the runtime into xla_compiled_cpu_function since that's the
only user.
PiperOrigin-RevId: 207333245
Diffstat (limited to 'tensorflow/docs_src')
| -rw-r--r-- | tensorflow/docs_src/performance/xla/tfcompile.md | 5 |
1 files changed, 1 insertions, 4 deletions
diff --git a/tensorflow/docs_src/performance/xla/tfcompile.md b/tensorflow/docs_src/performance/xla/tfcompile.md index 8521d7eacb..e4b803164f 100644 --- a/tensorflow/docs_src/performance/xla/tfcompile.md +++ b/tensorflow/docs_src/performance/xla/tfcompile.md @@ -205,10 +205,7 @@ representing the inputs, `results` representing the outputs, and `temps` representing temporary buffers used internally to perform the computation. By default, each instance of the generated class allocates and manages all of these buffers for you. The `AllocMode` constructor argument may be used to change this -behavior. A convenience library is provided in -[`tensorflow/compiler/aot/runtime.h`](https://www.tensorflow.org/code/tensorflow/compiler/aot/runtime.h) -to help with manual buffer allocation; usage of this library is optional. All -buffers should be aligned to 32-byte boundaries. +behavior. All buffers are aligned to 64-byte boundaries. The generated C++ class is just a wrapper around the low-level code generated by XLA. |