StreamExecutor: Optimize kernel argument packing

Create a single class to hold all kernel arguments and optimize how they are added into this class.
Change: 140556725

1 files changed, 1 insertions, 6 deletions

diff --git a/tensorflow/stream_executor/stream_executor_pimpl.cc b/tensorflow/stream_executor/stream_executor_pimpl.cc
index 2fdd1e4b49..7739d31662 100644
--- a/tensorflow/stream_executor/stream_executor_pimpl.cc
+++ b/tensorflow/stream_executor/stream_executor_pimpl.cc
@@ -394,7 +394,7 @@ rng::RngSupport *StreamExecutor::AsRng() {
 bool StreamExecutor::Launch(Stream *stream, const ThreadDim &thread_dims,
                             const BlockDim &block_dims,
                             const KernelBase &kernel,
-                            const std::vector<KernelArg> &args) {
+                            const KernelArgsArrayBase &args) {
   SubmitTrace(&TraceListener::LaunchSubmit, stream, thread_dims, block_dims,
               kernel, args);
 
@@ -659,11 +659,6 @@ bool StreamExecutor::DeviceMemoryUsage(int64 *free, int64 *total) const {
   return implementation_->DeviceMemoryUsage(free, total);
 }
 
-KernelArg StreamExecutor::DeviceMemoryToKernelArg(
-    const DeviceMemoryBase &gpu_mem) const {
-  return implementation_->DeviceMemoryToKernelArg(gpu_mem);
-}
-
 void StreamExecutor::EnqueueOnBackgroundThread(std::function<void()> task) {
   background_threads_->Schedule(task);
 }