Andy Wingo (Dec 08 2021 at 14:01): Andy Wingo (Dec 08 2021 at 14:01):Hello people! Quick question -- does cranelift require reducible control-flow graphs?
bjorn3 (Dec 08 2021 at 14:35):Cranelift doesn't require it.
Andy Wingo (Dec 08 2021 at 14:39):cool, tx. (also how do you know? wasm graphs won't contain these initially; you can only get them after jump threading or such...)
Andy Wingo (Dec 08 2021 at 14:40):context is i was thinking of having a poke at funclets or similar idioms
bjorn3 (Dec 08 2021 at 14:41):rustc_codegen_cranelift doesn't turn potentially irreducable control flow from rustc MIR into reducable control flow when generating cranelift ir
bjorn3 (Dec 08 2021 at 14:41):I never got any issues from this.
bjorn3 (Dec 08 2021 at 14:42):Andy Wingo said:
context is i was thinking of having a poke at funclets or similar idioms
that isn't supported by cranelift ir yet
Andy Wingo (Dec 08 2021 at 14:44):that's why i was thinking of having a poke at it
bjorn3 (Dec 08 2021 at 14:46):Have you seen https://github.com/bytecodealliance/wasmtime/issues/3427, https://github.com/bytecodealliance/wasmtime/issues/2049 and https://github.com/bytecodealliance/wasmtime/issues/1677?
Andy Wingo (Dec 08 2021 at 14:48):no i had not, thank you for those pointers!
bjorn3 (Dec 08 2021 at 14:50):https://github.com/bjorn3/wasmtime/tree/eh_cleanup was an experiment of mine for implementing it. Never finished it though.
Chris Fallin (Dec 08 2021 at 16:44):@Andy Wingo to add some confirmation here, the backend parts (regalloc, cfg lowering to branches) were explicitly designed to handle irreducible control flow, and if there's something that breaks with it, I'd consider it a bug; happy to jump on any issues that arise!
Last updated: Dec 23 2024 at 12:05 UTC