diff options
Diffstat (limited to 'tensorflow/stream_executor/cuda/cuda_gpu_executor.h')
-rw-r--r-- | tensorflow/stream_executor/cuda/cuda_gpu_executor.h | 12 |
1 files changed, 6 insertions, 6 deletions
diff --git a/tensorflow/stream_executor/cuda/cuda_gpu_executor.h b/tensorflow/stream_executor/cuda/cuda_gpu_executor.h index e8ebbc3220..1481dcc19a 100644 --- a/tensorflow/stream_executor/cuda/cuda_gpu_executor.h +++ b/tensorflow/stream_executor/cuda/cuda_gpu_executor.h @@ -71,16 +71,16 @@ class CUDAExecutor : public internal::StreamExecutorInterface { const KernelArgsArrayBase &args) override; int CalculateOccupancy(const DeviceDescription& device_description, + uint64 registers_per_thread, + uint64 shared_memory_per_block, + const ThreadDim& thread_dims, CUfunction func); + + int CompareOccupancy(int* initial_blocks, + const DeviceDescription& device_description, uint64 registers_per_thread, uint64 shared_memory_per_block, const ThreadDim& thread_dims, CUfunction func); - int CompareOccupancy(int* initial_blocks, - const DeviceDescription& device_description, - uint64 registers_per_thread, - uint64 shared_memory_per_block, - const ThreadDim& thread_dims, CUfunction func); - void *Allocate(uint64 size) override; void *AllocateSubBuffer(DeviceMemoryBase *mem, uint64 offset_bytes, |