Range operator lowering minimizer #1004
Open
Conversation
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
Summary: During our perf tuning, we are seeing mtml_instagram_model lag behind AIT around 23% on mergenet (525662456/1670) How do we compare the perf between AIT and AOTI? Refering the idea from Oleg, I tried to implement a minimizer leveraging acc tracer + fx splitter. It is good these two backend still can use the same acc tracer so we can make sure the front end graphs are exactly the same. Then we can leverage the splitter to define a range of operator to lower. In this way, we can directly compare the perf of a small subgraph between these two backends. How to use it? 1) attach --debug_operator_range="0, 100" randomly and you will see graph nodes printed out in log ``` ``` 2) attach the precise range you want. For ex, --debug_operator_range="1613,1635" and you will see logs like this ``` INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_122 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_124 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_125 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_126 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_127 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_128 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_129 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_130 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_131 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_132 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_133 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_134 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_135 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_136 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_137 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_138 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_139 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_140 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_141 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_142 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_143 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_144 to selected nodes INFO:fx2ait.ait_splitter:Add n.op=call_function, n.target=<function linear at 0x7f01014a7b50>, n.name=linear_145 to selected nodes ``` Differential Revision: D57030900
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D57030900 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Summary:
During our perf tuning, we are seeing mtml_instagram_model lag behind AIT around 23% on mergenet (525662456/1670)
How do we compare the perf between AIT and AOTI? Refering the idea from Oleg, I tried to implement a minimizer leveraging acc tracer + fx splitter. It is good these two backend still can use the same acc tracer so we can make sure the front end graphs are exactly the same. Then we can leverage the splitter to define a range of operator to lower. In this way, we can directly compare the perf of a small subgraph between these two backends.
How to use it?
Differential Revision: D57030900