1
Fork 0

Add wasm32v1-none target (compiler-team/#791)

This commit is contained in:
Graydon Hoare 2024-10-10 00:05:16 -07:00
parent e1f3068995
commit e14d6d8314
No known key found for this signature in database
GPG key ID: 69824B014EE65219
3 changed files with 92 additions and 0 deletions

View file

@ -1803,6 +1803,7 @@ supported_targets! {
("wasm32-unknown-emscripten", wasm32_unknown_emscripten),
("wasm32-unknown-unknown", wasm32_unknown_unknown),
("wasm32v1-none", wasm32v1_none),
("wasm32-wasi", wasm32_wasi),
("wasm32-wasip1", wasm32_wasip1),
("wasm32-wasip2", wasm32_wasip2),

View file

@ -0,0 +1,88 @@
//! A "bare wasm" target representing a WebAssembly output that makes zero
//! assumptions about its environment, similar to wasm32-unknown-unknown, but
//! that also specifies an _upper_ bound on the set of wasm proposals that are
//! supported.
//!
//! It is implemented as a variant on LLVM's wasm32-unknown-unknown target, with
//! the additional flags `-Ctarget-cpu=mvp` and `-Ctarget-feature=+mutable-globals`.
//!
//! This target exists to resolve a tension in Rustc's choice of WebAssembly
//! proposals to support. Since most WebAssembly users are in fact _on the web_
//! and web browsers are frequently updated with support for the latest
//! features, it is reasonable for Rustc to generate wasm code that exploits new
//! WebAssembly proposals as they gain browser support. At least by default. And
//! this is what the wasm32-unknown-unknown target does, which means that the
//! _exact_ WebAssembly features that Rustc generates will change over time.
//!
//! But a different set of users -- smaller but nonetheless worth supporting --
//! are using WebAssembly in implementations that either don't get updated very
//! often, or need to prioritize stability, implementation simplicity or
//! security over feature support. This target is for them, and it promises that
//! the wasm code it generates will not go beyond the proposals/features of the
//! W3C WebAssembly core 1.0 spec, which (as far as I can tell) is approximately
//! "the wasm MVP plus mutable globals". Mutable globals was proposed in 2018
//! and made it in.
//!
//! See https://www.w3.org/TR/wasm-core-1/
//!
//! Notably this feature-set _excludes_:
//!
//! - sign-extension operators
//! - non-trapping / saturating float-to-int conversions
//! - multi-value
//! - reference types
//! - bulk memory operations
//! - SIMD
//!
//! These are all listed as additions in the core 2.0 spec. Also they were all
//! proposed after 2020, and core 1.0 shipped in 2019. It also excludes even
//! later proposals such as:
//!
//! - exception handling
//! - tail calls
//! - extended consts
//! - function references
//! - multi-memory
//! - component model
//! - gc
//! - threads
//! - relaxed SIMD
//! - custom annotations
//! - branch hinting
//!
use crate::spec::{Cc, LinkerFlavor, Target, base};
pub(crate) fn target() -> Target {
let mut options = base::wasm::options();
options.os = "none".into();
options.cpu = "mvp".into();
options.features = "+mutable-globals".into();
options.add_pre_link_args(LinkerFlavor::WasmLld(Cc::No), &[
// For now this target just never has an entry symbol no matter the output
// type, so unconditionally pass this.
"--no-entry",
]);
options.add_pre_link_args(LinkerFlavor::WasmLld(Cc::Yes), &[
// Make sure clang uses LLD as its linker and is configured appropriately
// otherwise
"--target=wasm32-unknown-unknown",
"-Wl,--no-entry",
]);
Target {
llvm_target: "wasm32-unknown-unknown".into(),
metadata: crate::spec::TargetMetadata {
description: Some("WebAssembly".into()),
tier: Some(2),
host_tools: Some(false),
std: Some(false),
},
pointer_width: 32,
data_layout: "e-m:e-p:32:32-p10:8:8-p20:8:8-i64:64-n32:64-S128-ni:1:10:20".into(),
arch: "wasm32".into(),
options,
}
}