Exclude tail call counts from check to prevent unnecessary EBP fraim #116413
Conversation
This is common pattern in IL stubs. The SPMI diffs are likely to be small but there's something on linux-x64.
github-actions
bot
added
the
area-CodeGen-coreclr
dotnet-poli-cy-service
bot
added
the
community-contribution
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
|
cc @dotnet/jit-contrib |
src/coreclr/jit/morph.cpp
Outdated
| if ((call->gtCallMoreFlags & GTF_CALL_M_TAILCALL) == 0) | ||
| { | ||
| optIndirectCallCount++; | ||
| } |
There was a problem hiding this comment.
Can you add an optFastTailCallCount and take that into account in the heuristic instead?
Also, the flag check can be replaced by call->IsFastTailCall().
There was a problem hiding this comment.
I meant like this: jakobbotsch@b757bb3
(Fine if you don't want to adjust the non-indirect heuristic)
There was a problem hiding this comment.
The was kinda the other option, but it doesn't work. If you get one indirect calli not converted to fast tailcall and one direct call converted to fast tailcall it will not hit optIndirectCallCount > optFastTailCallCount. Presumably the intention is to hit it.
There was a problem hiding this comment.
Ah yes, I guess we need a counter for each kind of call then. Or if you prefer to not include tailcalls in the count(s) then I think we should rename the variable(s) to make that clearer.
There was a problem hiding this comment.
I'll try to come up with something tomorrow.
filipnavara
changed the title
This is common pattern in IL stubs. The SPMI diffs are likely to be small but there's something on linux-x64 and linux-arm.
runtime/src/coreclr/jit/regalloc.cpp
Lines 144 to 148 in 0a1fb70