Created by: zhangting2020

PR types

Performance optimization

PR changes

OPs

Describe

Using IndexList instead of arrays of indices can speed up CPU and GPU performance. This PR use it to improve performance of dot op. For more information, please refer to #25132

performace

GPU: v100, cuda10

op	input shape	before	after	speed up
dot	[1000, 1000]	0.210704 ms	0.125058 ms	1.7x