JIT: handle large OSR funclet frame initial SP delta for arm64 type 5… #66124
Conversation
… frame Because OSR funclet frames must pad with Tier0 frame size, initial SP delta may exceed 240 bytes. Fixes dotnet#65996.
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT Issue Details… frame Because OSR funclet frames must pad with Tier0 frame size, initial SP delta Fixes #65996.
|
|
@BruceForstall PTAL |
| $(CLRTestBatchPreCommands) | ||
| set COMPlus_TieredCompilation=1 | ||
| set COMPlus_TC_QuickJitForLoops=1 | ||
| set COMPlus_TC_OnStackReplacement=1 | ||
| set COMPlus_OSR_HitLimit=2 | ||
| set COMPlus_JitRandomOnStackReplacement=15 | ||
| set COMPlus_JitStress=2 | ||
| ]]></CLRTestBatchPreCommands> | ||
| <BashCLRTestPreCommands><![CDATA[ | ||
| $(BashCLRTestPreCommands) | ||
| export COMPlus_TieredCompilation=1 | ||
| export COMPlus_TC_QuickJitForLoops=1 | ||
| export COMPlus_TC_OnStackReplacement=1 | ||
| export COMPlus_OSR_HitLimit=2 | ||
| export COMPlus_JitRandomOnStackReplacement=15 | ||
| export COMPlus_JitStress=2 | ||
| ]]></BashCLRTestPreCommands> |
There was a problem hiding this comment.
Is this going to be the first Pri-0 OSR test case?
There was a problem hiding this comment.
All the tests in this directory are OSR tests. This will be the first that relies on "Random" OSR and aggressive OSR transitions. Normally those are only tested in the experimental CI legs.
I could probably rewrite the test to trigger with default OSR settings, if you think that's preferable.
There was a problem hiding this comment.
Hmmm, "random" doesn't seem great in terms of repeatability? But if the seed is the method hash, then it's not actually random? Seems ok.
Thanks -- I will check. |
Yeah, it seems wrong. For the main method this bit of SP adjust happens in the Tier0 frame, but for funclets it has to happen in the funclet prolog, and type 5 frames are missing this bit. |
|
Turns out (because of patchpoint info skew when cross-jitting) that running natively on arm64 we don't generate a type 5 frame. So probably need to find a better test case, but for now this is better than nothing. |
|
Trying to cleanly restart the SPMI jobs. |
|
/azp run runtime-coreclr superpmi-replay |
|
/azp run runtime-coreclr superpmi-asmdiffs |
|
Azure Pipelines successfully started running 1 pipeline(s). |
1 similar comment
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
Failures seem to be known issues. |
dotnet#66124) Because OSR funclet frames must pad with Tier0 frame size, initial SP delta may exceed 240 bytes. Fixes dotnet#65996.
ghost
locked as resolved and limited conversation to collaborators
… frame
Because OSR funclet frames must pad with Tier0 frame size, initial SP delta
may exceed 240 bytes.
Fixes #65996.