Stream: git-wasmtime

Topic: wasmtime / issue #4450 cranelift: Use requested ISA Flags...

Wasmtime GitHub notifications bot (Jul 15 2022 at 17:43):

afonso360 commented on issue #4450:

It looks like the existing behavior is to select an ISA with all native-supported features enabled.

The new behavior is to select an ISA with all features disabled unless you specifically ask for them, and then skip the test if the features you asked for aren't available.

Is that right?

Yes, except that we don't start with all features disabled, we start with the defaults enabled by cranelift. So on x86_64 for example we start with has_sse41 and a bunch of others. These can be explicitly disabled to test the lowering without those features.

Wasmtime GitHub notifications bot (Jul 15 2022 at 17:55):

afonso360 edited a comment on issue #4450:

It looks like the existing behavior is to select an ISA with all native-supported features enabled.

The new behavior is to select an ISA with all features disabled unless you specifically ask for them, and then skip the test if the features you asked for aren't available.

Is that right?

Yes, except that we don't start with all features disabled, we start with the defaults enabled by cranelift. So on x86_64 for example we start with has_sse41 and a bunch of others. These can be explicitly disabled to test the lowering without those features.

Starting from an empty set of features would be valuable, I think we can try that.

Wasmtime GitHub notifications bot (Jul 15 2022 at 18:33):

afonso360 edited a comment on issue #4450:

It looks like the existing behavior is to select an ISA with all native-supported features enabled.

The new behavior is to select an ISA with all features disabled unless you specifically ask for them, and then skip the test if the features you asked for aren't available.

Is that right?

Yes, except that we don't start with all features disabled, we start with the defaults enabled by cranelift. So on x86_64 for example we start with has_sse41 and a bunch of others. These can be explicitly disabled to test the lowering without those features.

Starting from an empty set of features would be valuable, I think we can try that. But I'd like to hear what people think is preferable.

Also CC @abrown and @bnjbvr from the original issue #1891

Last updated: Nov 22 2024 at 17:03 UTC