Speed up SuperPMI mcs remove dup process #33946
Merged
Conversation
This file contains 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
Dotnet-GitSync-Bot
added
the
area-CodeGen-coreclr
|
@dotnet/runtime-infrastructure It looks like almost all the CI jobs are "cancelled" with "#[error]The remote provider was unable to process the request." |
Create a "Hash" class that encapsulates the MD5 hashing that is used to determine if two MCs are equivalent. Primarily, this allows caching the Windows Crypto provider, which it is very slow to acquire. In addition, make some changes to avoid unnecessary memory allocations and other unnecessary work. The result is that `mcs -removeDup` is about 4x faster. Much of the remaining cost is that we read, deserialize the MC, then reserialize the MC (if unique), and finally destroy the in-memory MC. There is a lot of memory allocation/deallocation in this process that could possibly be avoided or improved for this scenario.
Also add `-thin`. With this, ``` mcs.exe -merge base.mch *.mc -recursive mcs.exe -removeDup -thin base.mch nodup.mch ``` can be replaced with: ``` mcs.exe -merge -recursive -dedup -thin nodup.mch *.mc ``` The main benefit is avoiding creating a potentially very large base.mch file. Related, the data being processed only needs to be loaded once.
Adjust superpmi.py script and superpmicollect.cs unit test.
Add description of `mcs -merge -dedup -thin` arguments and usage.
BruceForstall
force-pushed
the
AddRemoveDupToMerge
branch
from
9070216
to
2e67374
Compare
|
@dotnet/jit-contrib |
lpereira
reviewed
Mar 23, 2020
Comment on lines
+102
to
+105
| if (m_inFileLegacy->GetIndex(newInfo.ILCodeSize) == -1) | ||
| m_inFileLegacy->Add(newInfo.ILCodeSize, new DenseLightWeightMap<MethodContext*>()); | ||
|
|
||
| DenseLightWeightMap<MethodContext*>* ourRank = m_inFileLegacy->Get(newInfo.ILCodeSize); |
There was a problem hiding this comment.
Maybe avoid the lookup here when GetIndex() returns -1 and set ourRank to the newly-constructed DenseLightWeightMap?
|
libraries test failure is infrastructure / System.Net flakiness (of course) |
|
ping |
AndyAyersMS
approved these changes
Mar 27, 2020
I don't think I've ever used it. I should probably ask Scott sometime. We could probably get rid of all that code. |
ghost
locked as resolved and limited conversation to collaborators
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
To speed up the "remove dup" code, create a "Hash" class that encapsulates the MD5 hashing
that is used to determine if two MCs are equivalent. Primarily, this allows caching the
Windows Crypto provider, which it is very slow to acquire.
In addition, make some changes to avoid unnecessary memory allocations
and other unnecessary work.
The result is that mcs -removeDup is about 4x faster.
Much of the remaining cost is that we read, deserialize the MC,
then reserialize the MC (if unique), and finally destroy the in-memory MC.
There is a lot of memory allocation/deallocation in this process that
could possibly be avoided or improved for this scenario.
In addition, add support to
mcs -mergeto also remove duplicates at the same time.This removes the need to create merged file with duplicates, and thus significantly reduces the
disk space required to do a collection, as well as speeds up the collection processing.
Update the superpmi.py drive script, superpmicollect unit test, and documentation to match.