Stream: git-wasmtime

Topic: wasmtime / PR #14412 Leverage same-`vmctx` analysis resul...

view this post on Zulip Wasmtime GitHub notifications bot (Sep 24 2026 at 19:49):

fitzgen opened PR #14412 from fitzgen:use-same-vmctx-analysis to bytecodealliance:main:

This allows GVN to deduplicate the callee vmctx loads for multiple calls into
imported functions from the same instance.

This shaves off another ~10% of time from the increment-each-byte compile-time
builtins benchmark.

There is still some store-to-load forwarding that I would have expected this to unlock that is not happening, which if it did happen I'd expect to see that 10% number jump up to maybe 25% or 50%. Seems related to Cranelift's alias analysis tho, not related to anything here, so I wouldn't delay this PR on that ongoing investigation.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 24 2026 at 19:49):

fitzgen requested cfallin for a review on PR #14412.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 24 2026 at 19:49):

fitzgen requested wasmtime-compiler-reviewers for a review on PR #14412.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 24 2026 at 19:49):

fitzgen requested wasmtime-core-reviewers for a review on PR #14412.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 24 2026 at 19:49):

fitzgen requested alexcrichton for a review on PR #14412.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 24 2026 at 19:49):

fitzgen requested wasmtime-default-reviewers for a review on PR #14412.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 24 2026 at 19:49):

fitzgen edited PR #14412:

Depends on https://github.com/bytecodealliance/wasmtime/pull/14411

This allows GVN to deduplicate the callee vmctx loads for multiple calls into
imported functions from the same instance.

This shaves off another ~10% of time from the increment-each-byte compile-time
builtins benchmark.

There is still some store-to-load forwarding that I would have expected this to unlock that is not happening, which if it did happen I'd expect to see that 10% number jump up to maybe 25% or 50%. Seems related to Cranelift's alias analysis tho, not related to anything here, so I wouldn't delay this PR on that ongoing investigation.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 24 2026 at 21:05):

:thumbs_up: alexcrichton submitted PR review.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 25 2026 at 03:39):

fitzgen updated PR #14412.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 25 2026 at 03:39):

fitzgen has enabled auto merge for PR #14412.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 25 2026 at 03:48):

fitzgen added PR #14412 Leverage same-vmctx analysis results in Wasm-to-CLIF translation to the merge queue.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 25 2026 at 04:14):

:check: fitzgen merged PR #14412.

view this post on Zulip Wasmtime GitHub notifications bot (Sep 25 2026 at 04:14):

fitzgen removed PR #14412 Leverage same-vmctx analysis results in Wasm-to-CLIF translation from the merge queue.

Last updated: Oct 11 2026 at 04:10 UTC