High CPU usage in v8::internal::Runtime_LoadIC_Miss

[Wayakorn Vadhanasin]

I am working on updating my Electron app from Electron 4 (Chromium 69) to Electron 8 (Chromium 80). I am occasionally experiencing random delays in my app while calling a JS function to process HTML DOMs (no external callouts).

Looking closely at the profiling data collected from the Windows Performance Analysis tool, I am seeing the following calls during the repro:

  v8::internal::Runtime_LoadIC_Miss - 55%
  v8::internal::LoadIC::Load - 13%

More details on the LoadIC_Miss:

  v8::internal::IC::UpdatePolymorphicIC
  v8::internal::IC::SetCache
  v8::internal::LoadIC::UpdateCaches
  v8::internal::LoadIC::Load
  v8::internal::Runtime_LoadIC_Miss

Very few instances of the above occurred in a normal condition, only when there was a perf issue.

I'd like to understand more what inline cache is doing. Have there been many churns in this area since Chromium 69? Any hints or suggestions on the next step of investigation is appreciated. Thanks in advance.

Way

[Camillo Bruni]

Hi,

Inline Caches are a crucial optimization to speed up property accesses.

Usually, you get misses the first time an inline cache is used (for initialization) or when the shape of objects is different (different order, different types of properties).

Spending 55% of your time in LoadIC_Miss seems really high. Without access to the repro this is hard to assess.

cheers,

Camillo

--
--
v8-users mailing list
v8-u...@googlegroups.com
http://groups.google.com/group/v8-users
---
You received this message because you are subscribed to the Google Groups "v8-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to v8-users+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/v8-users/ad7fe487-be79-491a-96ee-37c286274c12n%40googlegroups.com.

[Seth Brenith]

A few brief observations based on stepping through the problem area in the debugger:

• The generated code for this JS function calls directly into Builtins_LoadICTrampoline_Megamorphic.
• However, the feedback vector slot for that load is in monomorphic state.
• The map we're currently loading from matches the map in the feedback vector.
• We don't find the relevant data in the global megamorphic lookup cache.
• UpdatePolymorphicIC returns true, so IC::SetCache returns without running UpdateMegamorphicCache, so this property lookup never has any chance to get faster.
  

Does anybody have advice about how a LoadIC could initially get into this state where its feedback says monomorphic but the generated code is executing a megamorphic load?

To unsubscribe from this group and stop receiving emails from it, send an email to v8-u...@googlegroups.com.

[Seth Brenith]

After trying and failing to generate a small repro (I always got soft deopts instead of the bug), I realized that it's because Leszek already fixed it! https://chromium-review.googlesource.com/c/v8/v8/+/2150586