Optimizations for precision conversion operations in nGraph reference implementations#3974
Merged
dmitry-gorokhov merged 13 commits intoopenvinotoolkit:masterfrom Feb 8, 2021
Conversation
a3f6ec7 to
0896d9e
Compare
4027bf2 to
3414af4
Compare
85c7c33 to
23e4f88
Compare
4e3db8b to
e1552e8
Compare
Contributor
GlebKazantaev
left a comment
There was a problem hiding this comment.
Transformations part looks good,
GlebKazantaev
approved these changes
Jan 26, 2021
e1552e8 to
2e1c6b1
Compare
ilyachur
reviewed
Jan 29, 2021
| for (size_t i = 0; i < size; ++i) { | ||
| dst_data[i] = convert_value<src_type, dst_type>(src_data[i]); | ||
| for (size_t i = 0; i < size; ++i) { | ||
| dst_data[i] = convert_value<src_type, dst_type>(src_data[i]); |
Contributor
There was a problem hiding this comment.
Why cannot I always use the reference implementation?
I thought that reference implementation should support any precisions
Contributor
Author
There was a problem hiding this comment.
The behavior of two convert operations is slightly different. The nGraph conversion operation does not check upper and lower bounds. In my cases, for int8->fp16 and fp16->fp32 conversions, I can use nGraph implementations as-is, because the target type values range is wider than the source.
Contributor
Author
There was a problem hiding this comment.
May be we should align conversion operations in Graph and transformations library.
ilya-lavrenov
approved these changes
Feb 2, 2021
| @@ -0,0 +1,273 @@ | |||
| //***************************************************************************** | |||
| // Copyright 2017-2020 Intel Corporation | |||
There was a problem hiding this comment.
please fix in other files as well
ilyachur
reviewed
Feb 3, 2021
2e1c6b1 to
8da0575
Compare
…onymous namespace
257c35b to
5dbf375
Compare
ilyachur
approved these changes
Feb 5, 2021
5dbf375 to
b1bdd55
Compare
dmitry-gorokhov
approved these changes
Feb 8, 2021
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Data precision conversion operations take up a significant part of the load time. On average ConvertPrecision transformation pass takes 34% of load time for FP16 models. This fix significantly reduce time of int8->fp16 and fp16->fp32 conversions.