Improves ReduceSum performance by removing transposition. #5370
xadupre
merged
28 commits into
microsoft:master
from
redsum
Improves ReduceSum performance
8daa0b81
fix compilation issue
4a137ae9
Update reduction_ops.cc
cfcfdf53
investigate SEGFAULT
ad98b46e
still investigating
47e1559e
optimized implementation
3553ccdd
improves the PR, reduces the memory footprint
6543b4a4
simplify code
308ddd47
futher optimization, using cache
62e4946d
improves the code, handle empty reduce
f1ef6585
improve Einstein operator
db8355a1
Merge branch 'master' of https://github.com/Microsoft/onnxruntime int…
ec2199af
fix compiling issue
982645d7
Add min, max, L1, L2, logsum, sumsquare
28e84d71
fix compiling issue
222f91ec
remove unused code
933b1c62
move ReduceSumCore to the training version
7caa2a60
fix link issue for orttraining
83ea6676
Merge branch 'master' of https://github.com/Microsoft/onnxruntime int…
93150281
replace all reduce functions
8c4fdbec
remove all reduce implementation including transpose
84b6e52f
fix compilation issue
3508b566
Update reduction_ops.h
8a66d7a4
fix compiling issues
363b38c8
compilation issue, uninitialized variable
3c9c1cb3
zhanghuanrong
commented
on 2020-10-12
xadupre
changed the title
Merge branch 'master' of https://github.com/Microsoft/onnxruntime int…
0cf5ed7a
removes mutable
f6d0fc53
Merge branch 'master' of https://github.com/Microsoft/onnxruntime int…
a6aa3fc5
zhanghuanrong
approved these changes
on 2020-10-12
Login to write a write a comment.
Login via GitHub