Stream: git-wasmtime

Topic: wasmtime / PR #8945 Resolve aliases before inserting valu...


view this post on Zulip Wasmtime GitHub notifications bot (Jul 12 2024 at 16:58):

fitzgen opened PR #8945 from fitzgen:fix-fuzz-bug-found-in-8941 to bytecodealliance:main:

This fixes a fuzz bug found in the development of https://github.com/bytecodealliance/wasmtime/pull/8941

view this post on Zulip Wasmtime GitHub notifications bot (Jul 12 2024 at 16:58):

fitzgen requested alexcrichton for a review on PR #8945.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 12 2024 at 16:58):

fitzgen requested wasmtime-fuzz-reviewers for a review on PR #8945.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 12 2024 at 16:58):

fitzgen requested abrown for a review on PR #8945.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 12 2024 at 16:58):

fitzgen requested wasmtime-compiler-reviewers for a review on PR #8945.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 12 2024 at 16:58):

fitzgen requested wasmtime-default-reviewers for a review on PR #8945.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 12 2024 at 17:05):

alexcrichton submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 12 2024 at 17:06):

bjorn3 commented on PR #8945:

Why does the alias resolving need to be done? cranelift-frontend can and does change aliases on the fly. It feels iffy to immediately resolve them inside cranelift-frontend as there is nothing preventing the alias that was resolved to be changed later.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 12 2024 at 17:10):

fitzgen commented on PR #8945:

Why does the alias resolving need to be done?

Because otherwise we would need to have a reverse-aliases map from a value to all of the values that alias it. This is necessary because when we find a definition, we kill it from the live set, so at that point we only have the non-alias value.

It feels iffy to immediately resolve them inside cranelift-frontend as there is nothing preventing the alias that was resolved to be changed later.

Note that the safepoint spilling and its liveness analysis happens as part of finishing a FunctionBuilder, so we don't need to worry about anything changing under our feet here.

view this post on Zulip Wasmtime GitHub notifications bot (Jul 12 2024 at 17:24):

fitzgen merged PR #8945.


Last updated: Nov 22 2024 at 16:03 UTC