bjoernager/rust
bjoernager/rust
1
Fork 0
Code Activity

rustc: Add a new wasm ABI

Browse source 
This commit implements the idea of a new ABI for the WebAssembly target,
one called `"wasm"`. This ABI is entirely of my own invention
and has no current precedent, but I think that the addition of this ABI
might help solve a number of issues with the WebAssembly targets.

When `wasm32-unknown-unknown` was first added to Rust I naively
"implemented an abi" for the target. I then went to write `wasm-bindgen`
which accidentally relied on details of this ABI. Turns out the ABI
definition didn't match C, which is causing issues for C/Rust interop.
Currently the compiler has a "wasm32 bindgen compat" ABI which is the
original implementation I added, and it's purely there for, well,
`wasm-bindgen`.

Another issue with the WebAssembly target is that it's not clear to me
when and if the default C ABI will change to account for WebAssembly's
multi-value feature (a feature that allows functions to return multiple
values). Even if this does happen, though, it seems like the C ABI will
be guided based on the performance of WebAssembly code and will likely
not match even what the current wasm-bindgen-compat ABI is today. This
leaves a hole in Rust's expressivity in binding WebAssembly where given
a particular import type, Rust may not be able to import that signature
with an updated C ABI for multi-value.

To fix these issues I had the idea of a new ABI for WebAssembly, one
called `wasm`. The definition of this ABI is "what you write
maps straight to wasm". The goal here is that whatever you write down in
the parameter list or in the return values goes straight into the
function's signature in the WebAssembly file. This special ABI is for
intentionally matching the ABI of an imported function from the
environment or exporting a function with the right signature.

With the addition of a new ABI, this enables rustc to:

* Eventually remove the "wasm-bindgen compat hack". Once this
  ABI is stable wasm-bindgen can switch to using it everywhere.
  Afterwards the wasm32-unknown-unknown target can have its default ABI
  updated to match C.

* Expose the ability to precisely match an ABI signature for a
  WebAssembly function, regardless of what the C ABI that clang chooses
  turns out to be.

* Continue to evolve the definition of the default C ABI to match what
  clang does on all targets, since the purpose of that ABI will be
  explicitly matching C rather than generating particular function
  imports/exports.

Naturally this is implemented as an unstable feature initially, but it
would be nice for this to get stabilized (if it works) in the near-ish
future to remove the wasm32-unknown-unknown incompatibility with the C
ABI. Doing this, however, requires the feature to be on stable because
wasm-bindgen works with stable Rust.
This commit is contained in:
Alex Crichton 2021-04-01 16:08:29 -07:00
parent 69e1d22ddb
commit 482a3d06c3
23 changed files with 405 additions and 198 deletions
Download patch file Download diff file Expand all files Collapse all files

18
compiler/rustc_target/src/abi/call/mod.rs
View file

@ -18,9 +18,7 @@ mod riscv;
mod s390x;
mod sparc;
mod sparc64;
mod wasm32;
mod wasm32_bindgen_compat;
mod wasm64;
mod wasm;
mod x86;
mod x86_64;
mod x86_win64;
@ -648,12 +646,14 @@ impl<'a, Ty> FnAbi<'a, Ty> {
"nvptx64" => nvptx64::compute_abi_info(self),
"hexagon" => hexagon::compute_abi_info(self),
"riscv32" | "riscv64" => riscv::compute_abi_info(cx, self),
"wasm32" => match cx.target_spec().os.as_str() {
"emscripten" | "wasi" => wasm32::compute_abi_info(cx, self),
_ => wasm32_bindgen_compat::compute_abi_info(self),
},
"asmjs" => wasm32::compute_abi_info(cx, self),
"wasm64" => wasm64::compute_abi_info(cx, self),
"wasm32" | "wasm64" => {
if cx.target_spec().adjust_abi(abi) == spec::abi::Abi::Wasm {
wasm::compute_wasm_abi_info(self)
} else {
wasm::compute_c_abi_info(cx, self)
}
}
"asmjs" => wasm::compute_c_abi_info(cx, self),
a => return Err(format!("unrecognized arch \"{}\" in target specification", a)),
}

27
compiler/rustc_target/src/abi/call/wasm32.rs → compiler/rustc_target/src/abi/call/wasm.rs
View file

@ -40,7 +40,8 @@ where
}
}
pub fn compute_abi_info<'a, Ty, C>(cx: &C, fn_abi: &mut FnAbi<'a, Ty>)
/// The purpose of this ABI is to match the C ABI (aka clang) exactly.
pub fn compute_c_abi_info<'a, Ty, C>(cx: &C, fn_abi: &mut FnAbi<'a, Ty>)
where
Ty: TyAndLayoutMethods<'a, C> + Copy,
C: LayoutOf<Ty = Ty, TyAndLayout = TyAndLayout<'a, Ty>> + HasDataLayout,
@ -56,3 +57,27 @@ where
classify_arg(cx, arg);
}
}
/// The purpose of this ABI is for matching the WebAssembly standard. This
/// intentionally diverges from the C ABI and is specifically crafted to take
/// advantage of LLVM's support of multiple returns in WebAssembly.
pub fn compute_wasm_abi_info<Ty>(fn_abi: &mut FnAbi<'_, Ty>) {
if !fn_abi.ret.is_ignore() {
classify_ret(&mut fn_abi.ret);
}
for arg in &mut fn_abi.args {
if arg.is_ignore() {
continue;
}
classify_arg(arg);
}
fn classify_ret<Ty>(ret: &mut ArgAbi<'_, Ty>) {
ret.extend_integer_width_to(32);
}
fn classify_arg<Ty>(arg: &mut ArgAbi<'_, Ty>) {
arg.extend_integer_width_to(32);
}
}

29
compiler/rustc_target/src/abi/call/wasm32_bindgen_compat.rs
View file

@ -1,29 +0,0 @@
// This is not and has never been a correct C ABI for WebAssembly, but
// for a long time this was the C ABI that Rust used. wasm-bindgen
// depends on ABI details for this ABI and is incompatible with the
// correct C ABI, so this ABI is being kept around until wasm-bindgen
// can be fixed to work with the correct ABI. See #63649 for further
// discussion.
use crate::abi::call::{ArgAbi, FnAbi};
fn classify_ret<Ty>(ret: &mut ArgAbi<'_, Ty>) {
ret.extend_integer_width_to(32);
}
fn classify_arg<Ty>(arg: &mut ArgAbi<'_, Ty>) {
arg.extend_integer_width_to(32);
}
pub fn compute_abi_info<Ty>(fn_abi: &mut FnAbi<'_, Ty>) {
if !fn_abi.ret.is_ignore() {
classify_ret(&mut fn_abi.ret);
}
for arg in &mut fn_abi.args {
if arg.is_ignore() {
continue;
}
classify_arg(arg);
}
}

58
compiler/rustc_target/src/abi/call/wasm64.rs
View file

@ -1,58 +0,0 @@
use crate::abi::call::{ArgAbi, FnAbi, Uniform};
use crate::abi::{HasDataLayout, LayoutOf, TyAndLayout, TyAndLayoutMethods};
fn unwrap_trivial_aggregate<'a, Ty, C>(cx: &C, val: &mut ArgAbi<'a, Ty>) -> bool
where
Ty: TyAndLayoutMethods<'a, C> + Copy,
C: LayoutOf<Ty = Ty, TyAndLayout = TyAndLayout<'a, Ty>> + HasDataLayout,
{
if val.layout.is_aggregate() {
if let Some(unit) = val.layout.homogeneous_aggregate(cx).ok().and_then(|ha| ha.unit()) {
let size = val.layout.size;
if unit.size == size {
val.cast_to(Uniform { unit, total: size });
return true;
}
}
}
false
}
fn classify_ret<'a, Ty, C>(cx: &C, ret: &mut ArgAbi<'a, Ty>)
where
Ty: TyAndLayoutMethods<'a, C> + Copy,
C: LayoutOf<Ty = Ty, TyAndLayout = TyAndLayout<'a, Ty>> + HasDataLayout,
{
ret.extend_integer_width_to(64);
if ret.layout.is_aggregate() && !unwrap_trivial_aggregate(cx, ret) {
ret.make_indirect();
}
}
fn classify_arg<'a, Ty, C>(cx: &C, arg: &mut ArgAbi<'a, Ty>)
where
Ty: TyAndLayoutMethods<'a, C> + Copy,
C: LayoutOf<Ty = Ty, TyAndLayout = TyAndLayout<'a, Ty>> + HasDataLayout,
{
arg.extend_integer_width_to(64);
if arg.layout.is_aggregate() && !unwrap_trivial_aggregate(cx, arg) {
arg.make_indirect_byval();
}
}
pub fn compute_abi_info<'a, Ty, C>(cx: &C, fn_abi: &mut FnAbi<'a, Ty>)
where
Ty: TyAndLayoutMethods<'a, C> + Copy,
C: LayoutOf<Ty = Ty, TyAndLayout = TyAndLayout<'a, Ty>> + HasDataLayout,
{
if !fn_abi.ret.is_ignore() {
classify_ret(cx, &mut fn_abi.ret);
}
for arg in &mut fn_abi.args {
if arg.is_ignore() {
continue;
}
classify_arg(cx, arg);
}
}

15
compiler/rustc_target/src/spec/abi.rs
View file

@ -34,6 +34,7 @@ pub enum Abi {
AvrInterrupt,
AvrNonBlockingInterrupt,
CCmseNonSecureCall,
Wasm,
// Multiplatform / generic ABIs
System { unwind: bool },
@ -83,6 +84,7 @@ const AbiDatas: &[AbiData] = &[
generic: false,
},
AbiData { abi: Abi::CCmseNonSecureCall, name: "C-cmse-nonsecure-call", generic: false },
AbiData { abi: Abi::Wasm, name: "wasm", generic: false },
// Cross-platform ABIs
AbiData { abi: Abi::System { unwind: false }, name: "system", generic: true },
AbiData { abi: Abi::System { unwind: true }, name: "system-unwind", generic: true },
@ -131,13 +133,14 @@ impl Abi {
AvrInterrupt => 18,
AvrNonBlockingInterrupt => 19,
CCmseNonSecureCall => 20,
Wasm => 21,
// Cross-platform ABIs
System { unwind: false } => 21,
System { unwind: true } => 22,
RustIntrinsic => 23,
RustCall => 24,
PlatformIntrinsic => 25,
Unadjusted => 26,
System { unwind: false } => 22,
System { unwind: true } => 23,
RustIntrinsic => 24,
RustCall => 25,
PlatformIntrinsic => 26,
Unadjusted => 27,
};
debug_assert!(
AbiDatas

22
compiler/rustc_target/src/spec/mod.rs
View file

@ -1254,6 +1254,9 @@ pub struct TargetOptions {
/// enabled can generated on this target, but the necessary supporting libraries are not
/// distributed with the target, the sanitizer should still appear in this list for the target.
pub supported_sanitizers: SanitizerSet,
/// If present it's a default value to use for adjusting the C ABI.
pub default_adjusted_cabi: Option<Abi>,
}
impl Default for TargetOptions {
@ -1357,6 +1360,7 @@ impl Default for TargetOptions {
has_thumb_interworking: false,
split_debuginfo: SplitDebuginfo::Off,
supported_sanitizers: SanitizerSet::empty(),
default_adjusted_cabi: None,
}
}
}
@ -1408,6 +1412,9 @@ impl Target {
Abi::C { unwind: false }
}
}
Abi::C { unwind } => self.default_adjusted_cabi.unwrap_or(Abi::C { unwind }),
abi => abi,
}
}
@ -1742,6 +1749,16 @@ impl Target {
}
}
} );
($key_name:ident, Option<Abi>) => ( {
let name = (stringify!($key_name)).replace("_", "-");
obj.find(&name[..]).and_then(|o| o.as_string().and_then(|s| {
match lookup_abi(s) {
Some(abi) => base.$key_name = Some(abi),
_ => return Some(Err(format!("'{}' is not a valid value for abi", s))),
}
Some(Ok(()))
})).unwrap_or(Ok(()))
} );
}
if let Some(s) = obj.find("target-endian").and_then(Json::as_string) {
@ -1841,6 +1858,7 @@ impl Target {
key!(has_thumb_interworking, bool);
key!(split_debuginfo, SplitDebuginfo)?;
key!(supported_sanitizers, SanitizerSet)?;
key!(default_adjusted_cabi, Option<Abi>)?;
// NB: The old name is deprecated, but support for it is retained for
// compatibility.
@ -2081,6 +2099,10 @@ impl ToJson for Target {
target_option_val!(split_debuginfo);
target_option_val!(supported_sanitizers);
if let Some(abi) = self.default_adjusted_cabi {
d.insert("default-adjusted-cabi".to_string(), Abi::name(abi).to_json());
}
if default.unsupported_abis != self.unsupported_abis {
d.insert(
"unsupported-abis".to_string(),

12
compiler/rustc_target/src/spec/wasm32_unknown_unknown.rs
View file

@ -12,11 +12,23 @@
use super::wasm_base;
use super::{LinkerFlavor, LldFlavor, Target};
use crate::spec::abi::Abi;
pub fn target() -> Target {
let mut options = wasm_base::options();
options.os = "unknown".to_string();
options.linker_flavor = LinkerFlavor::Lld(LldFlavor::Wasm);
// This is a default for backwards-compatibility with the original
// definition of this target oh-so-long-ago. Once the "wasm" ABI is
// stable and the wasm-bindgen project has switched to using it then there's
// no need for this and it can be removed.
//
// Currently this is the reason that this target's ABI is mismatched with
// clang's ABI. This means that, in the limit, you can't merge C and Rust
// code on this target due to this ABI mismatch.
options.default_adjusted_cabi = Some(Abi::Wasm);
let clang_args = options.pre_link_args.entry(LinkerFlavor::Gcc).or_default();
// Make sure clang uses LLD as its linker and is configured appropriately