bjoernager/rust
bjoernager/rust
1
Fork 0
Code Issues Pull requests Activity

Change signature of target_features_cfg.

Browse source 
Currently it is called twice, once with `allow_unstable` set to true and
once with it set to false. This results in some duplicated work. Most
notably, for the LLVM backend, `LLVMRustHasFeature` is called twice for
every feature, and it's moderately slow. For very short running
compilations on platforms with many features (e.g. a `check` build of
hello-world on x86) this is a significant fraction of runtime.

This commit changes `target_features_cfg` so it is only called once, and
it now returns a pair of feature sets. This halves the number of
`LLVMRustHasFeature` calls.
This commit is contained in:
Nicholas Nethercote 2025-02-25 15:25:54 +11:00
parent 2df8e657f2
commit 936a8232df
6 changed files with 85 additions and 71 deletions
Download patch file Download diff file Expand all files Collapse all files

54
compiler/rustc_codegen_llvm/src/llvm_util.rs
View file

@ -306,7 +306,7 @@ pub(crate) fn to_llvm_features<'a>(sess: &Session, s: &'a str) -> Option<LLVMFea
/// Must express features in the way Rust understands them.
///
/// We do not have to worry about RUSTC_SPECIFIC_FEATURES here, those are handled outside codegen.
pub(crate) fn target_features_cfg(sess: &Session, allow_unstable: bool) -> Vec<Symbol> {
pub(crate) fn target_features_cfg(sess: &Session) -> (Vec<Symbol>, Vec<Symbol>) {
let mut features: FxHashSet<Symbol> = Default::default();
// Add base features for the target.
@ -331,6 +331,9 @@ pub(crate) fn target_features_cfg(sess: &Session, allow_unstable: bool) -> Vec<S
if let Some(feat) = to_llvm_features(sess, feature) {
for llvm_feature in feat {
let cstr = SmallCStr::new(llvm_feature);
// `LLVMRustHasFeature` is moderately expensive. On targets with many
// features (e.g. x86) these calls take a non-trivial fraction of runtime
// when compiling very small programs.
if !unsafe { llvm::LLVMRustHasFeature(target_machine.raw(), cstr.as_ptr()) }
{
return false;
@ -371,11 +374,7 @@ pub(crate) fn target_features_cfg(sess: &Session, allow_unstable: bool) -> Vec<S
// `features.contains` below.
#[allow(rustc::potential_query_instability)]
features.retain(|f| {
if sess
.target
.implied_target_features(f.as_str())
.contains(&feature.as_str())
{
if sess.target.implied_target_features(f.as_str()).contains(&feature.as_str()) {
// If `f` if implies `feature`, then `!feature` implies `!f`, so we have to
// remove `f`. (This is the standard logical contraposition principle.)
false
@ -387,24 +386,31 @@ pub(crate) fn target_features_cfg(sess: &Session, allow_unstable: bool) -> Vec<S
}
}
// Filter enabled features based on feature gates
sess.target
.rust_target_features()
.iter()
.filter_map(|(feature, gate, _)| {
// The `allow_unstable` set is used by rustc internally to determined which target
// features are truly available, so we want to return even perma-unstable "forbidden"
// features.
if allow_unstable
|| (gate.in_cfg() && (sess.is_nightly_build() || gate.requires_nightly().is_none()))
{
Some(Symbol::intern(feature))
} else {
None
}
})
.filter(|feature| features.contains(&feature))
.collect()
// Filter enabled features based on feature gates.
let f = |allow_unstable| {
sess.target
.rust_target_features()
.iter()
.filter_map(|(feature, gate, _)| {
// The `allow_unstable` set is used by rustc internally to determined which target
// features are truly available, so we want to return even perma-unstable
// "forbidden" features.
if allow_unstable
|| (gate.in_cfg()
&& (sess.is_nightly_build() || gate.requires_nightly().is_none()))
{
Some(Symbol::intern(feature))
} else {
None
}
})
.filter(|feature| features.contains(&feature))
.collect()
};
let target_features = f(false);
let unstable_target_features = f(true);
(target_features, unstable_target_features)
}
pub(crate) fn print_version() {