onnxruntime
6dc25a60 - Make the reduction ops more consistent in checking if no transpose is required and skipping the copy of the input data if that is the case. Significantly better performance when this is done (2x faster for model calling ReduceSumSquare with input of {2048,10}). (#3265)

Commit

6 years ago

Make the reduction ops more consistent in checking if no transpose is required and skipping the copy of the input data if that is the case. Significantly better performance when this is done (2x faster for model calling ReduceSumSquare with input of {2048,10}). (#3265)

References

#3265 - Make reductions ops more consistent to improve performance

Author

skottmckay

Parents

8f00147c

onnxruntime 6dc25a60 - Make the reduction ops more consistent in checking if no transpose is required and skipping the copy of the input data if that is the case. Significantly better performance when this is done (2x faster for model calling ReduceSumSquare with input of {2048,10}). (#3265)

onnxruntime
6dc25a60 - Make the reduction ops more consistent in checking if no transpose is required and skipping the copy of the input data if that is the case. Significantly better performance when this is done (2x faster for model calling ReduceSumSquare with input of {2048,10}). (#3265)