diff options
Diffstat (limited to 'tensorflow/stream_executor/device_description.h')
-rw-r--r-- | tensorflow/stream_executor/device_description.h | 17 |
1 files changed, 0 insertions, 17 deletions
diff --git a/tensorflow/stream_executor/device_description.h b/tensorflow/stream_executor/device_description.h index d335b9b875..b15ce31216 100644 --- a/tensorflow/stream_executor/device_description.h +++ b/tensorflow/stream_executor/device_description.h @@ -24,7 +24,6 @@ limitations under the License. #include <memory> #include "tensorflow/stream_executor/platform/port.h" -#include "tensorflow/stream_executor/cuda/cuda_driver.h" #include "tensorflow/stream_executor/launch_dim.h" #include "tensorflow/stream_executor/platform/port.h" @@ -324,22 +323,6 @@ void CalculateDimensionality(const DeviceDescription &device_description, uint64 element_count, uint64 *threads_per_block, uint64 *block_count); -// Compute and return maximum blocks per core (occupancy) based on the -// device description, some kernel characteristics and the number of threads per -// block. If unable to compute occupancy, zero is returned. -int CalculateOccupancy(const DeviceDescription& device_description, - uint64 registers_per_thread, - uint64 shared_memory_per_block, - const ThreadDim& thread_dims, CUfunction func); - -// Compute and return the suggested thread count to acheive ideal occupancy. -// If the provided thread dimensions match this number, zero is returned. -int CompareOccupancy(int* initial_blocks, - const DeviceDescription& device_description, - uint64 registers_per_thread, - uint64 shared_memory_per_block, - const ThreadDim& thread_dims, CUfunction func); - } // namespace stream_executor #endif // TENSORFLOW_STREAM_EXECUTOR_DEVICE_DESCRIPTION_H_ |