cc @dotnet/jit-contrib

We probably also want to override the default genCreateAddrMode and just look for base+offset, or (shorter term) just refuse to form any modes.

Should I update the create as well, and then tweak this so that it asserts that there's no idx?

Should I update the create as well, and then tweak this so that it asserts that there's no idx?

@kg it seems that inside genCreateAddrMode there are a few ifdefs for RISC-V which is also reg + cns only, so you can add WASM defines there.

The following indirections are valid address modes on riscv64:

Although, I am not sure RISC-V and WASM have the same rules for the offset (it should never be negative for WASM, right?)

Although, I am not sure RISC-V and WASM have the same rules for the offset (it should never be negative for WASM, right?)

The WASM rules are that:

1. Displacement is >= 0.
2. The addition doesn't overflow.

2 is the 'hard' part. Providing for it involves far-reaching refactoring. It may be best to give up on any address modes as NYI for the time being.

It may be best to give up on any address modes as NYI for the time being.

Yeah, that was my first reaction as well, I thought we started optimizing along the way 🙂

so genCreateAddrMode should just return false on wasm

It may be best to give up on any address modes as NYI for the time being.

Yeah, that was my first reaction as well, I thought we started optimizing along the way 🙂

so genCreateAddrMode should just return false on wasm

Done, thanks all

/ba-g known ios/osx device flakiness