Stream: git-wasmtime

Topic: wasmtime / PR #8457 cranelift: Don't recompute allocatabl...

view this post on Zulip Wasmtime GitHub notifications bot (Apr 24 2024 at 06:30):

jameysharp requested elliottt for a review on PR #8457.

view this post on Zulip Wasmtime GitHub notifications bot (Apr 24 2024 at 06:30):

jameysharp requested wasmtime-compiler-reviewers for a review on PR #8457.

view this post on Zulip Wasmtime GitHub notifications bot (Apr 24 2024 at 06:30):

jameysharp opened PR #8457 from jameysharp:cache-allocatable to bytecodealliance:main:

The implementation of PRegSet::from(self.vcode.machine_env()) does a surprising amount of work: it lazily initializes a OnceLock in the backend, then loops over six vectors of registers and adds them all to the PRegSet.

We could likely implement that better, but in the meantime at least we can avoid repeating that work for every single machine instruction. The PRegSet itself only takes a few words to store so it's cheap to just keep it around.

I discovered this because when the call to self.vcode.machine_env() is in the middle of the loop, that prevents holding a mutable borrow on parts of self.vcode, which I want to be able to do in another PR.

view this post on Zulip Wasmtime GitHub notifications bot (Apr 24 2024 at 15:00):

cfallin commented on PR #8457:

This one really surprised me so I went looking at history to figure out if we ever measured it: it looks like first appeared in #6957; previously the PRegSet was computed once when the lowering backend was created; and we didn't benchmark that. The PR did note caching the MachineEnv but we missed caching the derived information.

Out of curiosity have you seen a perf improvement with this PR? (We should do it even if not, but it seems to me it should improve compile time)

view this post on Zulip Wasmtime GitHub notifications bot (Apr 24 2024 at 15:01):

cfallin submitted PR review:

also I'll go ahead and r+ since I dug into this (sorry to preempt reviewers!).

view this post on Zulip Wasmtime GitHub notifications bot (Apr 26 2024 at 03:56):

jameysharp commented on PR #8457:

According to hyperfine, this commit is 1.02 ± 0.01 times faster when running wasmtime compile -C cache=n ../sightglass/benchmarks/spidermonkey/benchmark.wasm. That's actually a bigger effect than I expected since I thought operand collection is a pretty small part of the overall time.

view this post on Zulip Wasmtime GitHub notifications bot (Apr 26 2024 at 04:15):

jameysharp merged PR #8457.

Last updated: Dec 23 2024 at 12:05 UTC