abrown opened Issue #2409:
As discussed in https://github.com/bytecodealliance/wasmtime/pull/2397#discussion_r522410217, the Cranelift semantics for shift amounts must be masked to the number of bits that are present in the value to be shifted; e.g. if we try to shift a 32-bit integer by
33 we would actually shift by
33 mod 32 = 1. Currently, these semantics (which may the Wasm semantics) are enforced in the translation from Wasm to CLIF (i.e.
code_translator.rs`) and not by the compiler backends. This means that users of Cranelift are able to use shifts with the incorrect semantics.My proposed fix is to remove the masking from
code_translator.rs
and add it directly to the x64 backend--might need some help for the other backends (cc: @akirilov-arm, @cfallin). Another fix would be to change the semantics, but this seems less desirable so I don't plan to pursue it (let me know in the comments if you disagree!).
abrown edited Issue #2409:
As discussed in https://github.com/bytecodealliance/wasmtime/pull/2397#discussion_r522410217, the Cranelift semantics for shift amounts must be masked to the number of bits that are present in the value to be shifted; e.g. if we try to shift a 32-bit integer by
33
we would actually shift by33 mod 32 = 1
. Currently, these semantics (which match the Wasm semantics) are enforced in the translation from Wasm to CLIF (i.e.code_translator.rs
) and not by the compiler backends. This means that users of Cranelift are able to use shifts with the incorrect semantics.My proposed fix is to remove the masking from
code_translator.rs
and add it directly to the x64 backend--might need some help for the other backends (cc: @akirilov-arm, @cfallin). Another fix would be to change the semantics, but this seems less desirable so I don't plan to pursue it (let me know in the comments if you disagree!).
Last updated: Nov 22 2024 at 16:03 UTC