diff options
author | Peter Hawkins <phawkins@google.com> | 2016-11-29 18:55:46 -0800 |
---|---|---|
committer | TensorFlower Gardener <gardener@tensorflow.org> | 2016-11-29 19:05:55 -0800 |
commit | bada4a5339d4567419e993b7736eb23a6f3535c4 (patch) | |
tree | d793b3a70518a2ab63fbc24c0e41d72eb7d7d97f /tensorflow/stream_executor/stream_executor_internal.h | |
parent | 347d3ef2a871d8212f39c7ea2b7defe63468dfbc (diff) |
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
Diffstat (limited to 'tensorflow/stream_executor/stream_executor_internal.h')
-rw-r--r-- | tensorflow/stream_executor/stream_executor_internal.h | 5 |
1 files changed, 1 insertions, 4 deletions
diff --git a/tensorflow/stream_executor/stream_executor_internal.h b/tensorflow/stream_executor/stream_executor_internal.h index acdbb07cb7..57db7775a6 100644 --- a/tensorflow/stream_executor/stream_executor_internal.h +++ b/tensorflow/stream_executor/stream_executor_internal.h @@ -184,7 +184,7 @@ class StreamExecutorInterface { } virtual bool Launch(Stream *stream, const ThreadDim &thread_dims, const BlockDim &block_dims, const KernelBase &k, - const std::vector<KernelArg> &args) { + const KernelArgsArrayBase &args) { return false; } virtual void *Allocate(uint64 size) = 0; @@ -258,9 +258,6 @@ class StreamExecutorInterface { // caller. virtual DeviceDescription *PopulateDeviceDescription() const = 0; - virtual KernelArg DeviceMemoryToKernelArg( - const DeviceMemoryBase &gpu_mem) const = 0; - // Attempts to register the provided TraceListener with the device-specific // Executor implementation. When this is called, the PIMPL interface has // already taken ownership of the object and is managing the generic tracing |