Stream: git-wasmtime

Topic: wasmtime / Issue #2248 [machinst x64]: enable packed inte...

Wasmtime GitHub notifications bot (Sep 30 2020 at 18:30):

github-actions[bot] commented on Issue #2248:

Subscribe to Label Action

cc @bnjbvr

<details>
This issue or pull request has been labeled: "cranelift", "cranelift:area:x64"

Thus the following users have been cc'd because of the following labels:

• bnjbvr: cranelift

To subscribe or unsubscribe from this label, edit the <code>.github/subscribe-to-label.json</code> configuration file.

Learn more.
</details>

Wasmtime GitHub notifications bot (Oct 01 2020 at 20:01):

abrown commented on Issue #2248:

Not a huge fan of this; instead could we just xorps the destination register, in the place where it's used? (There's already special handling of xorps X X in regalloc.) The slow down caused by the additional instruction in the pipeline may be recouped by the speedup caused by the lane dependency clearing.

@bnjbvr, me neither really. I didn't like the idea of adding extra instructions either without data on the performance impact so I left it in and moved it to the bottom of the Inst enum. Maybe there is something that can be done to the context that will provide an escape hatch? E.g. some way to "unsafely" mess with the use/defs?

Wasmtime GitHub notifications bot (Oct 02 2020 at 09:34):

bnjbvr commented on Issue #2248:

without data on the performance impact

Could we make a tiny benchmark out of this to compare the two variants?

Wasmtime GitHub notifications bot (Oct 02 2020 at 15:41):

abrown commented on Issue #2248:

@bnjbvr, created https://github.com/bytecodealliance/wasmtime/issues/2256 to try to get rid of XmmUninitializedValue.

Last updated: Nov 22 2024 at 16:03 UTC