Switch Windows Checked build to -O2 from -O1
#59235
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
|
Tagging subscribers to this area: @hoyosjs Issue DetailsThat is, optimize for speed, not for size.
|
|
Does it make superpmi faster? |
BruceForstall
force-pushed
the
SwitchCheckedWindowsBuildToO2
branch
from
017370e to
134704d
Compare
|
For win-x64 Checked build, I measure a code size increase of: clrjit.dll: 3,620,864 => 4,058,112 For runtime, superpmi replay (which is almost 100% JIT time) is 11% faster with -O2 across our current collections. |
BruceForstall
force-pushed
the
SwitchCheckedWindowsBuildToO2
branch
from
134704d to
64dd823
Compare
|
/azp run runtime-coreclr outerloop, runtime-coreclr jitstress |
|
Azure Pipelines successfully started running 2 pipeline(s). |
BruceForstall
added a commit
to BruceForstall/runtime
that referenced
this pull request
Oct 6, 2021
The win-x64 ABI requires allocating 32 bytes of outgoing argument stack space for every call, which can be used by callees. In the case where there are no calls in a function, there is no need to allocate that space. The JIT computes the required argument space during `fgSimpleLowering` by processing the IR and looking for call instructions, but also accounting for a few special cases where calls are not visible in the IR. One case was missing, leading to a zero-sized outgoing argument space even though a call was still generated. This can cause corruption in the caller stack frame. The case is where there are no explicit calls but the JIT is going to generate a PInvoke frame initialization helper call. This shouldn't happen, however, it occurs because the JIT eliminates a PInvoke that is statically dead code, but still generates the prolog setup code. A general solution would be to eliminate the PInvoke setup code if there are no remaining PInvokes in the function. The simple change made here is to ensure there is sufficient outgoing argument space for the PInvoke helper call if that will be generated. This was found by seeing test crashes due to stack corruption when building the CLR with the VC++ compiler `-O2` optimization switch, which stored data in the incoming argument stack space of the PInvoke helper function (with PR dotnet#59235). There are 5 SPMI diffs: 2 in coreclr tests (the ones that failed at runtime), 2 in libraries tests, and 1 in System.Net.Http.WinHttpChannelBinding:.ctor(SafeWinHttpHandle):this
BruceForstall
force-pushed
the
SwitchCheckedWindowsBuildToO2
branch
from
64dd823 to
983194e
Compare
|
/azp run runtime-coreclr outerloop, runtime-coreclr jitstress |
|
Azure Pipelines successfully started running 2 pipeline(s). |
BruceForstall
added a commit
that referenced
this pull request
Oct 6, 2021
…per (#60050) The win-x64 ABI requires allocating 32 bytes of outgoing argument stack space for every call, which can be used by callees. In the case where there are no calls in a function, there is no need to allocate that space. The JIT computes the required argument space during `fgSimpleLowering` by processing the IR and looking for call instructions, but also accounting for a few special cases where calls are not visible in the IR. One case was missing, leading to a zero-sized outgoing argument space even though a call was still generated. This can cause corruption in the caller stack frame. The case is where there are no explicit calls but the JIT is going to generate a PInvoke frame initialization helper call. This shouldn't happen, however, it occurs because the JIT eliminates a PInvoke that is statically dead code, but still generates the prolog setup code. A general solution would be to eliminate the PInvoke setup code if there are no remaining PInvokes in the function. The simple change made here is to ensure there is sufficient outgoing argument space for the PInvoke helper call if that will be generated. This was found by seeing test crashes due to stack corruption when building the CLR with the VC++ compiler `-O2` optimization switch, which stored data in the incoming argument stack space of the PInvoke helper function (with PR #59235). There are 5 SPMI diffs: 2 in coreclr tests (the ones that failed at runtime), 2 in libraries tests, and 1 in System.Net.Http.WinHttpChannelBinding:.ctor(SafeWinHttpHandle):this
|
The OSX test failure is due to infra (out of disk space). |
That is, optimize for speed, not for size. Don't do this for Windows x86, due to dotnet#59845
BruceForstall
force-pushed
the
SwitchCheckedWindowsBuildToO2
branch
from
983194e to
a380ba7
Compare
|
libraries failure is #60097 |
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.
That is, optimize for speed, not for size.