Small polishing changes in stream executor, no functional changes.

PiperOrigin-RevId: 196665609

3 files changed, 7 insertions, 16 deletions

diff --git a/tensorflow/stream_executor/cuda/cuda_dnn.cc b/tensorflow/stream_executor/cuda/cuda_dnn.cc
index 78dbd43c2d..7ace7fd303 100644
--- a/tensorflow/stream_executor/cuda/cuda_dnn.cc
+++ b/tensorflow/stream_executor/cuda/cuda_dnn.cc
@@ -4488,9 +4488,8 @@ void initialize_cudnn() {
             cuda::CUDAExecutor* cuda_executor =
                 dynamic_cast<cuda::CUDAExecutor*>(parent);
             if (cuda_executor == nullptr) {
-              LOG(ERROR)
-                  << "Attempting to initialize an instance of the cuBLAS "
-                  << "support library with a non-CUDA StreamExecutor";
+              LOG(ERROR) << "Attempting to initialize an instance of the cuDNN "
+                         << "support library with a non-CUDA StreamExecutor";
               return nullptr;
             }
 
diff --git a/tensorflow/stream_executor/dnn.h b/tensorflow/stream_executor/dnn.h
index 5b533dedcb..38abc66079 100644
--- a/tensorflow/stream_executor/dnn.h
+++ b/tensorflow/stream_executor/dnn.h
@@ -1051,10 +1051,8 @@ class DnnSupport {
   //    convolution result.
   //  scratch_allocator: un-owned, may-be-null object that may allocate scratch
   //    space in order to speed up the convolution operation.
-  //  algorithm: specifies which algorithm should be used for the
-  //    operation. If algorithm.is_default(), the system will pick an algorithm
-  //    by default. The coding of the algorithm is be interpretted by the
-  //    underlying implementation.
+  //  algorithm_config: specifies which algorithm should be used for the
+  //    operation.
   //  output_profile_result: the output profile result for this call. The
   //    profiling is only enabled when this is not nullptr.
   //
@@ -1153,17 +1151,13 @@ class DnnSupport {
   //    convolution input.
   //  filter_descriptor: dimensions of the convolution filter.
   //  convolution_descriptor: stride of the convolution filter.
-  //  input. This can be DeviceMemory pointing to NULL only when activation_mode
-  //  is kNone.
   //  output_descriptor: dimensions of the output layer.
   //  output_data: un-owned device memory region in which to place the
   //    convolution result.
   //  scratch_allocator: un-owned, may-be-null object that may allocate scratch
   //    space in order to speed up the convolution operation.
-  //  algorithm: an integer to specify which algorithm should be used for the
-  //    operation. kDefaultAlgorithm means the system will pick an algorithm
-  //    by default. The coding of the algorithm is be interpreted by the
-  //    underlying implementation.
+  //  algorithm_config: specifies which algorithm should be used for the
+  //    operation.
   //  output_profile_result: the output profile result for this call. The
   //    profiling is only enabled when this is not nullptr.
   //
@@ -1220,6 +1214,7 @@ class DnnSupport {
       ProfileResult* output_profile_result) = 0;
 
   // Return a list of algorithms supported by the forward convolution pass.
+  // cc_major and cc_minor are the compute capabilities of the device.
   virtual bool GetConvolveAlgorithms(
       bool with_winograd_nonfused, int cc_major, int cc_minor,
       std::vector<AlgorithmDesc>* out_algorithms);
diff --git a/tensorflow/stream_executor/stream.h b/tensorflow/stream_executor/stream.h
index 2c2879b586..c6e37da6d1 100644
--- a/tensorflow/stream_executor/stream.h
+++ b/tensorflow/stream_executor/stream.h
@@ -66,9 +66,6 @@ namespace dnn {
 class BatchDescriptor;
 class FilterDescriptor;
 class ConvolutionDescriptor;
-class BatchDescriptor;
-class FilterDescriptor;
-class ConvolutionDescriptor;
 class ProfileResult;
 class AlgorithmDesc;
 }  // namespace dnn