diff options
author | Yuefeng Zhou <yuefengz@google.com> | 2018-01-17 15:58:10 -0800 |
---|---|---|
committer | TensorFlower Gardener <gardener@tensorflow.org> | 2018-01-17 16:05:19 -0800 |
commit | e86fc00d097cb8cc118435ebd464497825bcad1e (patch) | |
tree | 2ae34164ad6b3cbd7624efa35423ff66cfe17f2c /tensorflow/core/profiler | |
parent | 04121a03acb16645dc0eac327c35a501453d8e8a (diff) |
Combine host and device memory proto fields.
PiperOrigin-RevId: 182284426
Diffstat (limited to 'tensorflow/core/profiler')
-rw-r--r-- | tensorflow/core/profiler/internal/testdata/run_meta | bin | 5539 -> 5631 bytes | |||
-rw-r--r-- | tensorflow/core/profiler/internal/tfprof_node.cc | 27 |
2 files changed, 15 insertions, 12 deletions
diff --git a/tensorflow/core/profiler/internal/testdata/run_meta b/tensorflow/core/profiler/internal/testdata/run_meta Binary files differindex ae76acb743..eaea62b06c 100644 --- a/tensorflow/core/profiler/internal/testdata/run_meta +++ b/tensorflow/core/profiler/internal/testdata/run_meta diff --git a/tensorflow/core/profiler/internal/tfprof_node.cc b/tensorflow/core/profiler/internal/tfprof_node.cc index 2945c9510f..86cb20de7b 100644 --- a/tensorflow/core/profiler/internal/tfprof_node.cc +++ b/tensorflow/core/profiler/internal/tfprof_node.cc @@ -133,18 +133,21 @@ void ExecStep::AddMemoryStats(const string& dev, exec_mem.set_output_bytes(total_output_bytes); if (step_stat.has_memory_stats()) { - exec_mem.set_host_temp_bytes( - exec_mem.host_temp_bytes() + - step_stat.memory_stats().host_temp_memory_size()); - exec_mem.set_host_persistent_bytes( - exec_mem.host_persistent_bytes() + - step_stat.memory_stats().host_persistent_memory_size()); - exec_mem.set_accelerator_temp_bytes( - exec_mem.accelerator_temp_bytes() + - step_stat.memory_stats().device_temp_memory_size()); - exec_mem.set_accelerator_persistent_bytes( - exec_mem.accelerator_persistent_bytes() + - step_stat.memory_stats().device_persistent_memory_size()); + if (IsPlacedOnCPU(dev)) { + // Currently we assume ops placed on gpu only allocate memory on gpu. + exec_mem.set_host_temp_bytes(exec_mem.host_temp_bytes() + + step_stat.memory_stats().temp_memory_size()); + exec_mem.set_host_persistent_bytes( + exec_mem.host_persistent_bytes() + + step_stat.memory_stats().persistent_memory_size()); + } else { + exec_mem.set_accelerator_temp_bytes( + exec_mem.accelerator_temp_bytes() + + step_stat.memory_stats().temp_memory_size()); + exec_mem.set_accelerator_persistent_bytes( + exec_mem.accelerator_persistent_bytes() + + step_stat.memory_stats().persistent_memory_size()); + } } // TODO(xpan): Make this more accurate: |