diff options
| author |  Smit Hinsu <hinsu@google.com> | 2018-05-21 17:42:15 -0700 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2018-05-21 17:44:41 -0700 |
| commit | b1139814f91c5216eb5ff229ee7e1982e5f4e888 (patch) | |
| tree | 7f85c8229bfd47eeba49890aa75b59c8680e619c /tensorflow/stream_executor/stream_executor_pimpl.h | |
| parent | d913a243196fa07d4728c8f7c1ce6444ecd086eb (diff) | |
Introduce an option to allocate CUDA unified memory
PiperOrigin-RevId: 197490523
Diffstat (limited to 'tensorflow/stream_executor/stream_executor_pimpl.h')
| -rw-r--r-- | tensorflow/stream_executor/stream_executor_pimpl.h | 10 |
1 files changed, 10 insertions, 0 deletions
diff --git a/tensorflow/stream_executor/stream_executor_pimpl.h b/tensorflow/stream_executor/stream_executor_pimpl.h index e426cf9931..ad80a1ba25 100644 --- a/tensorflow/stream_executor/stream_executor_pimpl.h +++ b/tensorflow/stream_executor/stream_executor_pimpl.h @@ -190,6 +190,16 @@ class StreamExecutor { // activated. void GetMemAllocs(std::map<void *, AllocRecord> *records_out); + // Allocates unified memory space of the given size, if supported. + // See + // https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#um-unified-memory-programming-hd + // for more details on unified memory. + void *UnifiedMemoryAllocate(uint64 bytes); + + // Deallocates unified memory space previously allocated with + // UnifiedMemoryAllocate. + void UnifiedMemoryDeallocate(void *location); + // Allocates a region of host memory and registers it with the platform API. // Memory allocated in this manner (or allocated and registered with // HostMemoryRegister() is required for use in asynchronous memcpy operations, |