wasm2c: ensure force read constraints compile for clang on mips

[shravanrn]

This is a clean slate attempt to fix #2266. To fix this i looked into how clang and gcc handle floating point constraints across a host of platforms.

Godbolt link: https://godbolt.org/z/ddeWc6ees
Results are in table below.

Given this, it seems the simplest course of action that maintains functionality as is today, but fix the mips platform is to special case this. We can separately file a bug upstream (but best case scenario is that will likely get fixed only for future clang versions, so I'm not sure we can rely on just that).

Fyi, @glandium

gcc

x86 - both
x86-64 - both
aarch32 - only "r"
aarch64 - only "r"
mips - only "r"
mips64 - both
riscv32 - both
riscv64 - both
sparc - both
sparc64 - both
powerpc - both
powerpc64 - both

clang

x86 - both
x64 - both
aarch32 - only "r"
aarch64 - only "r"
mips - only "f"
mips64 - only "f"
riscv32 - both
riscv64 - both
powerpc64 - both

[keithw]

lgtm % comment. Thank you for doing this whole investigation!

[shravanrn]

lgtm % comment. Thank you for doing this whole investigation!

Fixed 👍

[sbc100]

I wonder if this could be accompanied by some kind CI test? Can this be verified by simply compiling some wasm2c output with clang --target=mips maybe?

[shravanrn]

I wonder if this could be accompanied by some kind CI test? Can this be verified by simply compiling some wasm2c output with clang --target=mips maybe?

Open to it, but my take is that this feels like overkill at the moment. I tested this approach on Godbolt to make sure it works. I think we can push this until/if we see other mips incompatibilities