[turbofan] Preprocess feedback for global accesses (partially)
Main changes: - Rename ProcessedFeedback to ElementAccessFeedback and introduce a base class with the old name ProcessedFeedback. - Introduce another kind of ProcessedFeedback, namely GlobalAccessFeedback for the LoadGlobal/StoreGlobal IC. It's either a PropertyCell or a script context slot. - Produce such processed feedback in the serializer, when visiting LdaGlobal and similar bytecodes. - Consume it, and disallow heap access, in JSNativeContextSpecialization's ReduceJSLoadGlobal and ReduceJSStoreGlobal (for --concurrent-inlining). Minor changes: - Introduce a FeedbackSource class (pair of FeedbackVector and FeedbackSlot) that is used as the key of the processed feedback hash table. We already have two similar classes, FeedbackNexus and VectorSlotPair, but both are unsuitable for technical reasons (e.g. FeedbackNexus construction accesses the heap). Eventually we should remove VectorSlotPair. - Processed feedback is now returned as a pointer, which is nullptr if the original feedback wasn't interesting (e.g. megamorphic). The title says "partially" because the CL doesn't yet take into account named accesses where the receiver happens to be the global proxy. Bug: v8:7790 Change-Id: I4404d98636b91a8f2d5667115944bae4773a4770 Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1518184 Commit-Queue: Georg Neis <neis@chromium.org> Reviewed-by: Maya Lekova <mslekova@chromium.org> Cr-Commit-Position: refs/heads/master@{#60240}
Showing
This diff is collapsed.
This diff is collapsed.
Please
register
or
sign in
to comment