Auto merge of #2808 - saethlin:rustup, r=oli-obk

Rustup
This commit is contained in:
bors 2023-03-11 07:21:54 +00:00
commit 0f1f7b0840
515 changed files with 16053 additions and 3769 deletions

2
.gitignore vendored
View File

@ -41,7 +41,7 @@ no_llvm_build
/inst/
/llvm/
/mingw-build/
/build/
build/
/build-rust-analyzer/
/dist/
/unicode-downloads

3
.gitmodules vendored
View File

@ -1,6 +1,3 @@
[submodule "src/rust-installer"]
path = src/tools/rust-installer
url = https://github.com/rust-lang/rust-installer.git
[submodule "src/doc/nomicon"]
path = src/doc/nomicon
url = https://github.com/rust-lang/nomicon.git

File diff suppressed because it is too large Load Diff

View File

@ -1,3 +1,99 @@
Version 1.68.0 (2023-03-09)
==========================
<a id="1.68.0-Language"></a>
Language
--------
- [Stabilize default_alloc_error_handler](https://github.com/rust-lang/rust/pull/102318/)
This allows usage of `alloc` on stable without requiring the
definition of a handler for allocation failure. Defining custom handlers is still unstable.
- [Stabilize `efiapi` calling convention.](https://github.com/rust-lang/rust/pull/105795/)
- [Remove implicit promotion for types with drop glue](https://github.com/rust-lang/rust/pull/105085/)
<a id="1.68.0-Compiler"></a>
Compiler
--------
- [Change `bindings_with_variant_name` to deny-by-default](https://github.com/rust-lang/rust/pull/104154/)
- [Allow .. to be parsed as let initializer](https://github.com/rust-lang/rust/pull/105701/)
- [Add `armv7-sony-vita-newlibeabihf` as a tier 3 target](https://github.com/rust-lang/rust/pull/105712/)
- [Always check alignment during compile-time const evaluation](https://github.com/rust-lang/rust/pull/104616/)
- [Disable "split dwarf inlining" by default.](https://github.com/rust-lang/rust/pull/106709/)
- [Add vendor to Fuchsia's target triple](https://github.com/rust-lang/rust/pull/106429/)
- [Enable sanitizers for s390x-linux](https://github.com/rust-lang/rust/pull/107127/)
<a id="1.68.0-Libraries"></a>
Libraries
---------
- [Loosen the bound on the Debug implementation of Weak.](https://github.com/rust-lang/rust/pull/90291/)
- [Make `std::task::Context` !Send and !Sync](https://github.com/rust-lang/rust/pull/95985/)
- [PhantomData layout guarantees](https://github.com/rust-lang/rust/pull/104081/)
- [Don't derive Debug for `OnceWith` & `RepeatWith`](https://github.com/rust-lang/rust/pull/104163/)
- [Implement DerefMut for PathBuf](https://github.com/rust-lang/rust/pull/105018/)
- [Add O(1) `Vec -> VecDeque` conversion guarantee](https://github.com/rust-lang/rust/pull/105128/)
- [Leak amplification for peek_mut() to ensure BinaryHeap's invariant is always met](https://github.com/rust-lang/rust/pull/105851/)
<a id="1.68.0-Stabilized-APIs"></a>
Stabilized APIs
---------------
- [`{core,std}::pin::pin!`](https://doc.rust-lang.org/stable/std/pin/macro.pin.html)
- [`impl From<bool> for {f32,f64}`](https://doc.rust-lang.org/stable/std/primitive.f32.html#impl-From%3Cbool%3E-for-f32)
- [`std::path::MAIN_SEPARATOR_STR`](https://doc.rust-lang.org/stable/std/path/constant.MAIN_SEPARATOR_STR.html)
- [`impl DerefMut for PathBuf`](https://doc.rust-lang.org/stable/std/path/struct.PathBuf.html#impl-DerefMut-for-PathBuf)
These APIs are now stable in const contexts:
- [`VecDeque::new`](https://doc.rust-lang.org/stable/std/collections/struct.VecDeque.html#method.new)
<a id="1.68.0-Cargo"></a>
Cargo
-----
- [Stabilize sparse registry support for crates.io](https://github.com/rust-lang/cargo/pull/11224/)
- [`cargo build --verbose` tells you more about why it recompiles.](https://github.com/rust-lang/cargo/pull/11407/)
- [Show progress of crates.io index update even `net.git-fetch-with-cli` option enabled](https://github.com/rust-lang/cargo/pull/11579/)
<a id="1.68.0-Misc"></a>
Misc
----
<a id="1.68.0-Compatibility-Notes"></a>
Compatibility Notes
-------------------
- [Add `SEMICOLON_IN_EXPRESSIONS_FROM_MACROS` to future-incompat report](https://github.com/rust-lang/rust/pull/103418/)
- [Only specify `--target` by default for `-Zgcc-ld=lld` on wasm](https://github.com/rust-lang/rust/pull/101792/)
- [Bump `IMPLIED_BOUNDS_ENTAILMENT` to Deny + ReportNow](https://github.com/rust-lang/rust/pull/106465/)
- [`std::task::Context` no longer implements Send and Sync](https://github.com/rust-lang/rust/pull/95985)
<a id="1.68.0-Internal-Changes"></a>
Internal Changes
----------------
These changes do not affect any public interfaces of Rust, but they represent
significant improvements to the performance or internals of rustc and related
tools.
- [Encode spans relative to the enclosing item](https://github.com/rust-lang/rust/pull/84762/)
- [Don't normalize in AstConv](https://github.com/rust-lang/rust/pull/101947/)
- [Find the right lower bound region in the scenario of partial order relations](https://github.com/rust-lang/rust/pull/104765/)
- [Fix impl block in const expr](https://github.com/rust-lang/rust/pull/104889/)
- [Check ADT fields for copy implementations considering regions](https://github.com/rust-lang/rust/pull/105102/)
- [rustdoc: simplify JS search routine by not messing with lev distance](https://github.com/rust-lang/rust/pull/105796/)
- [Enable ThinLTO for rustc on `x86_64-pc-windows-msvc`](https://github.com/rust-lang/rust/pull/103591/)
- [Enable ThinLTO for rustc on `x86_64-apple-darwin`](https://github.com/rust-lang/rust/pull/103647/)
Version 1.67.1 (2023-02-09)
===========================
@ -3765,6 +3861,8 @@ Version 1.41.1 (2020-02-27)
* [Always check types of static items][69145]
* [Always check lifetime bounds of `Copy` impls][69145]
* [Fix miscompilation in callers of `Layout::repeat`][69225]
* [Rust 1.41.0 was announced as the last Rust release with tier 1 or tier 2 support for 32-bit Apple targets][apple-32bit-drop].
That announcement did not expect a patch release. 1.41.1 also includes release binaries for these targets.
[69225]: https://github.com/rust-lang/rust/issues/69225
[69145]: https://github.com/rust-lang/rust/pull/69145
@ -3857,7 +3955,7 @@ Misc
Compatibility Notes
-------------------
- [As previously announced 1.41.0 will be the last tier 1 release for 32-bit
- [As previously announced 1.41 will be the last tier 1 release for 32-bit
Apple targets.][apple-32bit-drop] This means that the source code is still
available to build, but the targets are no longer being tested and release
binaries for those platforms will no longer be distributed by the Rust project.

View File

@ -12,7 +12,7 @@ use crate::ptr::P;
use crate::token::{self, Token};
use crate::tokenstream::*;
use rustc_data_structures::map_in_place::MapInPlace;
use rustc_data_structures::flat_map_in_place::FlatMapInPlace;
use rustc_data_structures::sync::Lrc;
use rustc_span::source_map::Spanned;
use rustc_span::symbol::Ident;

View File

@ -4,7 +4,7 @@ use crate::nll::ToRegionVid;
use crate::path_utils::allow_two_phase_borrow;
use crate::place_ext::PlaceExt;
use crate::BorrowIndex;
use rustc_data_structures::fx::{FxHashMap, FxHashSet, FxIndexMap};
use rustc_data_structures::fx::{FxIndexMap, FxIndexSet};
use rustc_index::bit_set::BitSet;
use rustc_middle::mir::traversal;
use rustc_middle::mir::visit::{MutatingUseContext, NonUseContext, PlaceContext, Visitor};
@ -26,10 +26,10 @@ pub struct BorrowSet<'tcx> {
/// NOTE: a given location may activate more than one borrow in the future
/// when more general two-phase borrow support is introduced, but for now we
/// only need to store one borrow index.
pub activation_map: FxHashMap<Location, Vec<BorrowIndex>>,
pub activation_map: FxIndexMap<Location, Vec<BorrowIndex>>,
/// Map from local to all the borrows on that local.
pub local_map: FxHashMap<mir::Local, FxHashSet<BorrowIndex>>,
pub local_map: FxIndexMap<mir::Local, FxIndexSet<BorrowIndex>>,
pub(crate) locals_state_at_exit: LocalsStateAtExit,
}
@ -175,8 +175,8 @@ struct GatherBorrows<'a, 'tcx> {
tcx: TyCtxt<'tcx>,
body: &'a Body<'tcx>,
location_map: FxIndexMap<Location, BorrowData<'tcx>>,
activation_map: FxHashMap<Location, Vec<BorrowIndex>>,
local_map: FxHashMap<mir::Local, FxHashSet<BorrowIndex>>,
activation_map: FxIndexMap<Location, Vec<BorrowIndex>>,
local_map: FxIndexMap<mir::Local, FxIndexSet<BorrowIndex>>,
/// When we encounter a 2-phase borrow statement, it will always
/// be assigning into a temporary TEMP:
@ -186,7 +186,7 @@ struct GatherBorrows<'a, 'tcx> {
/// We add TEMP into this map with `b`, where `b` is the index of
/// the borrow. When we find a later use of this activation, we
/// remove from the map (and add to the "tombstone" set below).
pending_activations: FxHashMap<mir::Local, BorrowIndex>,
pending_activations: FxIndexMap<mir::Local, BorrowIndex>,
locals_state_at_exit: LocalsStateAtExit,
}

View File

@ -1,6 +1,6 @@
#![deny(rustc::untranslatable_diagnostic)]
#![deny(rustc::diagnostic_outside_of_impl)]
use rustc_data_structures::fx::FxHashMap;
use rustc_data_structures::fx::FxIndexMap;
use rustc_index::bit_set::BitSet;
use rustc_middle::mir::{self, BasicBlock, Body, Location, Place};
use rustc_middle::ty::RegionVid;
@ -124,7 +124,7 @@ pub struct Borrows<'a, 'tcx> {
body: &'a Body<'tcx>,
borrow_set: &'a BorrowSet<'tcx>,
borrows_out_of_scope_at_location: FxHashMap<Location, Vec<BorrowIndex>>,
borrows_out_of_scope_at_location: FxIndexMap<Location, Vec<BorrowIndex>>,
}
struct StackEntry {
@ -138,7 +138,7 @@ struct OutOfScopePrecomputer<'a, 'tcx> {
visit_stack: Vec<StackEntry>,
body: &'a Body<'tcx>,
regioncx: &'a RegionInferenceContext<'tcx>,
borrows_out_of_scope_at_location: FxHashMap<Location, Vec<BorrowIndex>>,
borrows_out_of_scope_at_location: FxIndexMap<Location, Vec<BorrowIndex>>,
}
impl<'a, 'tcx> OutOfScopePrecomputer<'a, 'tcx> {
@ -148,7 +148,7 @@ impl<'a, 'tcx> OutOfScopePrecomputer<'a, 'tcx> {
visit_stack: vec![],
body,
regioncx,
borrows_out_of_scope_at_location: FxHashMap::default(),
borrows_out_of_scope_at_location: FxIndexMap::default(),
}
}
}
@ -390,6 +390,7 @@ impl<'tcx> rustc_mir_dataflow::GenKillAnalysis<'tcx> for Borrows<'_, 'tcx> {
| mir::StatementKind::Deinit(..)
| mir::StatementKind::StorageLive(..)
| mir::StatementKind::Retag { .. }
| mir::StatementKind::PlaceMention(..)
| mir::StatementKind::AscribeUserType(..)
| mir::StatementKind::Coverage(..)
| mir::StatementKind::Intrinsic(..)

View File

@ -72,6 +72,8 @@ pub fn categorize(context: PlaceContext) -> Option<DefUse> {
PlaceContext::MutatingUse(MutatingUseContext::Drop) =>
Some(DefUse::Drop),
// This statement exists to help unsafeck. It does not require the place to be live.
PlaceContext::NonUse(NonUseContext::PlaceMention) => None,
// Debug info is neither def nor use.
PlaceContext::NonUse(NonUseContext::VarDebugInfo) => None,

View File

@ -1,7 +1,7 @@
use either::Either;
use rustc_const_eval::util::CallKind;
use rustc_data_structures::captures::Captures;
use rustc_data_structures::fx::FxHashSet;
use rustc_data_structures::fx::FxIndexSet;
use rustc_errors::{
struct_span_err, Applicability, Diagnostic, DiagnosticBuilder, ErrorGuaranteed, MultiSpan,
};
@ -173,7 +173,7 @@ impl<'cx, 'tcx> MirBorrowckCtxt<'cx, 'tcx> {
let mut is_loop_move = false;
let mut in_pattern = false;
let mut seen_spans = FxHashSet::default();
let mut seen_spans = FxIndexSet::default();
for move_site in &move_site_vec {
let move_out = self.move_data.moves[(*move_site).moi];
@ -2223,8 +2223,8 @@ impl<'cx, 'tcx> MirBorrowckCtxt<'cx, 'tcx> {
}
}
let mut visited = FxHashSet::default();
let mut move_locations = FxHashSet::default();
let mut visited = FxIndexSet::default();
let mut move_locations = FxIndexSet::default();
let mut reinits = vec![];
let mut result = vec![];
@ -2351,7 +2351,7 @@ impl<'cx, 'tcx> MirBorrowckCtxt<'cx, 'tcx> {
let reinits_reachable = reinits
.into_iter()
.filter(|reinit| {
let mut visited = FxHashSet::default();
let mut visited = FxIndexSet::default();
let mut stack = vec![*reinit];
while let Some(location) = stack.pop() {
if !visited.insert(location) {

View File

@ -9,7 +9,7 @@ use crate::{
nll::ToRegionVid,
region_infer::{Cause, RegionInferenceContext},
};
use rustc_data_structures::fx::FxHashSet;
use rustc_data_structures::fx::FxIndexSet;
use rustc_middle::mir::visit::{MirVisitable, PlaceContext, Visitor};
use rustc_middle::mir::{Body, Local, Location};
use rustc_middle::ty::{RegionVid, TyCtxt};
@ -37,7 +37,7 @@ struct UseFinder<'cx, 'tcx> {
impl<'cx, 'tcx> UseFinder<'cx, 'tcx> {
fn find(&mut self) -> Option<Cause> {
let mut queue = VecDeque::new();
let mut visited = FxHashSet::default();
let mut visited = FxIndexSet::default();
queue.push_back(self.start_point);
while let Some(p) = queue.pop_front() {

View File

@ -1,7 +1,7 @@
//! Contains utilities for generating suggestions for borrowck errors related to unsatisfied
//! outlives constraints.
use rustc_data_structures::fx::FxHashSet;
use rustc_data_structures::fx::FxIndexSet;
use rustc_errors::Diagnostic;
use rustc_middle::ty::RegionVid;
use smallvec::SmallVec;
@ -87,7 +87,7 @@ impl OutlivesSuggestionBuilder {
// Keep track of variables that we have already suggested unifying so that we don't print
// out silly duplicate messages.
let mut unified_already = FxHashSet::default();
let mut unified_already = FxIndexSet::default();
for (fr, outlived) in &self.constraints_to_add {
let Some(fr_name) = self.region_vid_to_name(mbcx, *fr) else {

View File

@ -79,6 +79,8 @@ impl<'cx, 'tcx> Visitor<'tcx> for InvalidationGenerator<'cx, 'tcx> {
}
// Only relevant for mir typeck
StatementKind::AscribeUserType(..)
// Only relevant for unsafeck
| StatementKind::PlaceMention(..)
// Doesn't have any language semantics
| StatementKind::Coverage(..)
// Does not actually affect borrowck
@ -118,15 +120,6 @@ impl<'cx, 'tcx> Visitor<'tcx> for InvalidationGenerator<'cx, 'tcx> {
LocalMutationIsAllowed::Yes,
);
}
TerminatorKind::DropAndReplace {
place: drop_place,
value: new_value,
target: _,
unwind: _,
} => {
self.mutate_place(location, *drop_place, Deep);
self.consume_operand(location, new_value);
}
TerminatorKind::Call {
func,
args,

View File

@ -1,6 +1,5 @@
//! This query borrow-checks the MIR to (further) ensure it is not broken.
#![allow(rustc::potential_query_instability)]
#![feature(associated_type_bounds)]
#![feature(box_patterns)]
#![feature(let_chains)]
@ -18,7 +17,7 @@ extern crate rustc_middle;
#[macro_use]
extern crate tracing;
use rustc_data_structures::fx::{FxHashMap, FxHashSet};
use rustc_data_structures::fx::{FxIndexMap, FxIndexSet};
use rustc_data_structures::graph::dominators::Dominators;
use rustc_data_structures::vec_map::VecMap;
use rustc_errors::{Diagnostic, DiagnosticBuilder, DiagnosticMessage, SubdiagnosticMessage};
@ -404,7 +403,7 @@ fn do_mir_borrowck<'tcx>(
// Note that this set is expected to be small - only upvars from closures
// would have a chance of erroneously adding non-user-defined mutable vars
// to the set.
let temporary_used_locals: FxHashSet<Local> = mbcx
let temporary_used_locals: FxIndexSet<Local> = mbcx
.used_mut
.iter()
.filter(|&local| !mbcx.body.local_decls[*local].is_user_variable())
@ -491,7 +490,7 @@ pub struct BodyWithBorrowckFacts<'tcx> {
pub struct BorrowckInferCtxt<'cx, 'tcx> {
pub(crate) infcx: &'cx InferCtxt<'tcx>,
pub(crate) reg_var_to_origin: RefCell<FxHashMap<ty::RegionVid, RegionCtxt>>,
pub(crate) reg_var_to_origin: RefCell<FxIndexMap<ty::RegionVid, RegionCtxt>>,
}
impl<'cx, 'tcx> BorrowckInferCtxt<'cx, 'tcx> {
@ -588,7 +587,7 @@ struct MirBorrowckCtxt<'cx, 'tcx> {
/// borrow errors that is handled by the `reservation_error_reported` field as the inclusion
/// of the `Span` type (while required to mute some errors) stops the muting of the reservation
/// errors.
access_place_error_reported: FxHashSet<(Place<'tcx>, Span)>,
access_place_error_reported: FxIndexSet<(Place<'tcx>, Span)>,
/// This field keeps track of when borrow conflict errors are reported
/// for reservations, so that we don't report seemingly duplicate
/// errors for corresponding activations.
@ -596,17 +595,17 @@ struct MirBorrowckCtxt<'cx, 'tcx> {
// FIXME: ideally this would be a set of `BorrowIndex`, not `Place`s,
// but it is currently inconvenient to track down the `BorrowIndex`
// at the time we detect and report a reservation error.
reservation_error_reported: FxHashSet<Place<'tcx>>,
reservation_error_reported: FxIndexSet<Place<'tcx>>,
/// This fields keeps track of the `Span`s that we have
/// used to report extra information for `FnSelfUse`, to avoid
/// unnecessarily verbose errors.
fn_self_span_reported: FxHashSet<Span>,
fn_self_span_reported: FxIndexSet<Span>,
/// This field keeps track of errors reported in the checking of uninitialized variables,
/// so that we don't report seemingly duplicate errors.
uninitialized_error_reported: FxHashSet<PlaceRef<'tcx>>,
uninitialized_error_reported: FxIndexSet<PlaceRef<'tcx>>,
/// This field keeps track of all the local variables that are declared mut and are mutated.
/// Used for the warning issued by an unused mutable local variable.
used_mut: FxHashSet<Local>,
used_mut: FxIndexSet<Local>,
/// If the function we're checking is a closure, then we'll need to report back the list of
/// mutable upvars that have been used. This field keeps track of them.
used_mut_upvars: SmallVec<[Field; 8]>,
@ -628,7 +627,7 @@ struct MirBorrowckCtxt<'cx, 'tcx> {
/// Record the region names generated for each region in the given
/// MIR def so that we can reuse them later in help/error messages.
region_names: RefCell<FxHashMap<RegionVid, RegionName>>,
region_names: RefCell<FxIndexMap<RegionVid, RegionName>>,
/// The counter for generating new region names.
next_region_name: RefCell<usize>,
@ -691,6 +690,8 @@ impl<'cx, 'tcx> rustc_mir_dataflow::ResultsVisitor<'cx, 'tcx> for MirBorrowckCtx
}
// Only relevant for mir typeck
StatementKind::AscribeUserType(..)
// Only relevant for unsafeck
| StatementKind::PlaceMention(..)
// Doesn't have any language semantics
| StatementKind::Coverage(..)
// These do not actually affect borrowck
@ -744,15 +745,6 @@ impl<'cx, 'tcx> rustc_mir_dataflow::ResultsVisitor<'cx, 'tcx> for MirBorrowckCtx
flow_state,
);
}
TerminatorKind::DropAndReplace {
place: drop_place,
value: new_value,
target: _,
unwind: _,
} => {
self.mutate_place(loc, (*drop_place, span), Deep, flow_state);
self.consume_operand(loc, (new_value, span), flow_state);
}
TerminatorKind::Call {
func,
args,
@ -867,7 +859,6 @@ impl<'cx, 'tcx> rustc_mir_dataflow::ResultsVisitor<'cx, 'tcx> for MirBorrowckCtx
| TerminatorKind::Assert { .. }
| TerminatorKind::Call { .. }
| TerminatorKind::Drop { .. }
| TerminatorKind::DropAndReplace { .. }
| TerminatorKind::FalseEdge { real_target: _, imaginary_target: _ }
| TerminatorKind::FalseUnwind { real_target: _, unwind: _ }
| TerminatorKind::Goto { .. }
@ -2329,7 +2320,7 @@ mod error {
/// same primary span come out in a consistent order.
buffered_move_errors:
BTreeMap<Vec<MoveOutIndex>, (PlaceRef<'tcx>, DiagnosticBuilder<'tcx, ErrorGuaranteed>)>,
buffered_mut_errors: FxHashMap<Span, (DiagnosticBuilder<'tcx, ErrorGuaranteed>, usize)>,
buffered_mut_errors: FxIndexMap<Span, (DiagnosticBuilder<'tcx, ErrorGuaranteed>, usize)>,
/// Diagnostics to be reported buffer.
buffered: Vec<Diagnostic>,
/// Set to Some if we emit an error during borrowck

View File

@ -1,7 +1,7 @@
#![deny(rustc::untranslatable_diagnostic)]
#![deny(rustc::diagnostic_outside_of_impl)]
use rustc_data_structures::captures::Captures;
use rustc_data_structures::fx::FxHashMap;
use rustc_data_structures::fx::FxIndexMap;
use rustc_index::vec::IndexVec;
use rustc_middle::infer::MemberConstraint;
use rustc_middle::ty::{self, Ty};
@ -18,7 +18,7 @@ where
{
/// Stores the first "member" constraint for a given `R0`. This is an
/// index into the `constraints` vector below.
first_constraints: FxHashMap<R, NllMemberConstraintIndex>,
first_constraints: FxIndexMap<R, NllMemberConstraintIndex>,
/// Stores the data about each `R0 member of [R1..Rn]` constraint.
/// These are organized into a linked list, so each constraint
@ -132,7 +132,7 @@ where
let MemberConstraintSet { first_constraints, mut constraints, choice_regions } = self;
let mut first_constraints2 = FxHashMap::default();
let mut first_constraints2 = FxIndexMap::default();
first_constraints2.reserve(first_constraints.len());
for (r1, start1) in first_constraints {

View File

@ -10,10 +10,9 @@ use rustc_middle::mir::{
BasicBlock, Body, ClosureOutlivesSubject, ClosureRegionRequirements, LocalKind, Location,
Promoted,
};
use rustc_middle::ty::{self, OpaqueHiddenType, Region, RegionVid};
use rustc_middle::ty::{self, OpaqueHiddenType, Region, RegionVid, TyCtxt};
use rustc_span::symbol::sym;
use std::env;
use std::fmt::Debug;
use std::io;
use std::path::PathBuf;
use std::rc::Rc;
@ -325,7 +324,7 @@ pub(super) fn dump_mir_results<'tcx>(
infcx: &BorrowckInferCtxt<'_, 'tcx>,
body: &Body<'tcx>,
regioncx: &RegionInferenceContext<'tcx>,
closure_region_requirements: &Option<ClosureRegionRequirements<'_>>,
closure_region_requirements: &Option<ClosureRegionRequirements<'tcx>>,
) {
if !dump_enabled(infcx.tcx, "nll", body.source.def_id()) {
return;
@ -340,9 +339,11 @@ pub(super) fn dump_mir_results<'tcx>(
if let Some(closure_region_requirements) = closure_region_requirements {
writeln!(out, "| Free Region Constraints")?;
for_each_region_constraint(closure_region_requirements, &mut |msg| {
writeln!(out, "| {}", msg)
})?;
for_each_region_constraint(
infcx.tcx,
closure_region_requirements,
&mut |msg| writeln!(out, "| {}", msg),
)?;
writeln!(out, "|")?;
}
}
@ -375,7 +376,7 @@ pub(super) fn dump_annotation<'tcx>(
infcx: &BorrowckInferCtxt<'_, 'tcx>,
body: &Body<'tcx>,
regioncx: &RegionInferenceContext<'tcx>,
closure_region_requirements: &Option<ClosureRegionRequirements<'_>>,
closure_region_requirements: &Option<ClosureRegionRequirements<'tcx>>,
opaque_type_values: &VecMap<LocalDefId, OpaqueHiddenType<'tcx>>,
errors: &mut crate::error::BorrowckErrors<'tcx>,
) {
@ -405,7 +406,7 @@ pub(super) fn dump_annotation<'tcx>(
// Dump the region constraints we are imposing *between* those
// newly created variables.
for_each_region_constraint(closure_region_requirements, &mut |msg| {
for_each_region_constraint(tcx, closure_region_requirements, &mut |msg| {
err.note(msg);
Ok(())
})
@ -426,16 +427,19 @@ pub(super) fn dump_annotation<'tcx>(
errors.buffer_non_error_diag(err);
}
fn for_each_region_constraint(
closure_region_requirements: &ClosureRegionRequirements<'_>,
fn for_each_region_constraint<'tcx>(
tcx: TyCtxt<'tcx>,
closure_region_requirements: &ClosureRegionRequirements<'tcx>,
with_msg: &mut dyn FnMut(&str) -> io::Result<()>,
) -> io::Result<()> {
for req in &closure_region_requirements.outlives_requirements {
let subject: &dyn Debug = match &req.subject {
ClosureOutlivesSubject::Region(subject) => subject,
ClosureOutlivesSubject::Ty(ty) => ty,
let subject = match req.subject {
ClosureOutlivesSubject::Region(subject) => format!("{:?}", subject),
ClosureOutlivesSubject::Ty(ty) => {
format!("{:?}", ty.instantiate(tcx, |vid| tcx.mk_re_var(vid)))
}
};
with_msg(&format!("where {:?}: {:?}", subject, req.outlived_free_region,))?;
with_msg(&format!("where {}: {:?}", subject, req.outlived_free_region,))?;
}
Ok(())
}

View File

@ -3,7 +3,7 @@ use std::rc::Rc;
use rustc_data_structures::binary_search_util;
use rustc_data_structures::frozen::Frozen;
use rustc_data_structures::fx::{FxHashMap, FxHashSet};
use rustc_data_structures::fx::{FxIndexMap, FxIndexSet};
use rustc_data_structures::graph::scc::Sccs;
use rustc_errors::Diagnostic;
use rustc_hir::def_id::CRATE_DEF_ID;
@ -12,8 +12,9 @@ use rustc_infer::infer::outlives::test_type_match;
use rustc_infer::infer::region_constraints::{GenericKind, VarInfos, VerifyBound, VerifyIfEq};
use rustc_infer::infer::{InferCtxt, NllRegionVariableOrigin, RegionVariableOrigin};
use rustc_middle::mir::{
Body, ClosureOutlivesRequirement, ClosureOutlivesSubject, ClosureRegionRequirements,
ConstraintCategory, Local, Location, ReturnConstraint, TerminatorKind,
Body, ClosureOutlivesRequirement, ClosureOutlivesSubject, ClosureOutlivesSubjectTy,
ClosureRegionRequirements, ConstraintCategory, Local, Location, ReturnConstraint,
TerminatorKind,
};
use rustc_middle::traits::ObligationCause;
use rustc_middle::traits::ObligationCauseCode;
@ -87,7 +88,7 @@ pub struct RegionInferenceContext<'tcx> {
member_constraints_applied: Vec<AppliedMemberConstraint>,
/// Map universe indexes to information on why we created it.
universe_causes: FxHashMap<ty::UniverseIndex, UniverseInfo<'tcx>>,
universe_causes: FxIndexMap<ty::UniverseIndex, UniverseInfo<'tcx>>,
/// Contains the minimum universe of any variable within the same
/// SCC. We will ensure that no SCC contains values that are not
@ -262,7 +263,7 @@ fn sccs_info<'cx, 'tcx>(
debug!(debug_str);
let num_components = sccs.scc_data().ranges().len();
let mut components = vec![FxHashSet::default(); num_components];
let mut components = vec![FxIndexSet::default(); num_components];
for (reg_var_idx, scc_idx) in sccs.scc_indices().iter().enumerate() {
let reg_var = ty::RegionVid::from_usize(reg_var_idx);
@ -294,9 +295,9 @@ fn sccs_info<'cx, 'tcx>(
(ConstraintSccIndex::from_usize(scc_idx), repr)
})
.collect::<FxHashMap<_, _>>();
.collect::<FxIndexMap<_, _>>();
let mut scc_node_to_edges = FxHashMap::default();
let mut scc_node_to_edges = FxIndexMap::default();
for (scc_idx, repr) in components_representatives.iter() {
let edges_range = sccs.scc_data().ranges()[*scc_idx].clone();
let edges = &sccs.scc_data().all_successors()[edges_range];
@ -324,7 +325,7 @@ impl<'tcx> RegionInferenceContext<'tcx> {
universal_region_relations: Frozen<UniversalRegionRelations<'tcx>>,
outlives_constraints: OutlivesConstraintSet<'tcx>,
member_constraints_in: MemberConstraintSet<'tcx, RegionVid>,
universe_causes: FxHashMap<ty::UniverseIndex, UniverseInfo<'tcx>>,
universe_causes: FxIndexMap<ty::UniverseIndex, UniverseInfo<'tcx>>,
type_tests: Vec<TypeTest<'tcx>>,
liveness_constraints: LivenessValues<RegionVid>,
elements: &Rc<RegionValueElements>,
@ -521,6 +522,7 @@ impl<'tcx> RegionInferenceContext<'tcx> {
/// outlives `'a` and hence contains R0 and R1.
fn init_free_and_bound_regions(&mut self) {
// Update the names (if any)
// This iterator has unstable order but we collect it all into an IndexVec
for (external_name, variable) in self.universal_regions.named_universal_regions() {
debug!(
"init_universal_regions: region {:?} has external name {:?}",
@ -917,7 +919,7 @@ impl<'tcx> RegionInferenceContext<'tcx> {
// Sometimes we register equivalent type-tests that would
// result in basically the exact same error being reported to
// the user. Avoid that.
let mut deduplicate_errors = FxHashSet::default();
let mut deduplicate_errors = FxIndexSet::default();
for type_test in &self.type_tests {
debug!("check_type_test: {:?}", type_test);
@ -1084,18 +1086,10 @@ impl<'tcx> RegionInferenceContext<'tcx> {
true
}
/// When we promote a type test `T: 'r`, we have to convert the
/// type `T` into something we can store in a query result (so
/// something allocated for `'tcx`). This is problematic if `ty`
/// contains regions. During the course of NLL region checking, we
/// will have replaced all of those regions with fresh inference
/// variables. To create a test subject, we want to replace those
/// inference variables with some region from the closure
/// signature -- this is not always possible, so this is a
/// fallible process. Presuming we do find a suitable region, we
/// will use it's *external name*, which will be a `RegionKind`
/// variant that can be used in query responses such as
/// `ReEarlyBound`.
/// When we promote a type test `T: 'r`, we have to replace all region
/// variables in the type `T` with an equal universal region from the
/// closure signature.
/// This is not always possible, so this is a fallible process.
#[instrument(level = "debug", skip(self, infcx))]
fn try_promote_type_test_subject(
&self,
@ -1104,91 +1098,63 @@ impl<'tcx> RegionInferenceContext<'tcx> {
) -> Option<ClosureOutlivesSubject<'tcx>> {
let tcx = infcx.tcx;
// Opaque types' substs may include useless lifetimes.
// We will replace them with ReStatic.
struct OpaqueFolder<'tcx> {
tcx: TyCtxt<'tcx>,
}
impl<'tcx> ty::TypeFolder<TyCtxt<'tcx>> for OpaqueFolder<'tcx> {
fn interner(&self) -> TyCtxt<'tcx> {
self.tcx
}
fn fold_ty(&mut self, t: Ty<'tcx>) -> Ty<'tcx> {
use ty::TypeSuperFoldable as _;
let tcx = self.tcx;
let &ty::Alias(ty::Opaque, ty::AliasTy { substs, def_id, .. }) = t.kind() else {
return t.super_fold_with(self);
};
let substs =
std::iter::zip(substs, tcx.variances_of(def_id)).map(|(arg, v)| {
match (arg.unpack(), v) {
(ty::GenericArgKind::Lifetime(_), ty::Bivariant) => {
tcx.lifetimes.re_static.into()
}
_ => arg.fold_with(self),
}
});
tcx.mk_opaque(def_id, tcx.mk_substs_from_iter(substs))
}
}
let ty = ty.fold_with(&mut OpaqueFolder { tcx });
let ty = tcx.fold_regions(ty, |r, _depth| {
let region_vid = self.to_region_vid(r);
let r_vid = self.to_region_vid(r);
let r_scc = self.constraint_sccs.scc(r_vid);
// The challenge if this. We have some region variable `r`
// whose value is a set of CFG points and universal
// regions. We want to find if that set is *equivalent* to
// any of the named regions found in the closure.
//
// To do so, we compute the
// `non_local_universal_upper_bound`. This will be a
// non-local, universal region that is greater than `r`.
// However, it might not be *contained* within `r`, so
// then we further check whether this bound is contained
// in `r`. If so, we can say that `r` is equivalent to the
// bound.
//
// Let's work through a few examples. For these, imagine
// that we have 3 non-local regions (I'll denote them as
// `'static`, `'a`, and `'b`, though of course in the code
// they would be represented with indices) where:
//
// - `'static: 'a`
// - `'static: 'b`
//
// First, let's assume that `r` is some existential
// variable with an inferred value `{'a, 'static}` (plus
// some CFG nodes). In this case, the non-local upper
// bound is `'static`, since that outlives `'a`. `'static`
// is also a member of `r` and hence we consider `r`
// equivalent to `'static` (and replace it with
// `'static`).
//
// Now let's consider the inferred value `{'a, 'b}`. This
// means `r` is effectively `'a | 'b`. I'm not sure if
// this can come about, actually, but assuming it did, we
// would get a non-local upper bound of `'static`. Since
// `'static` is not contained in `r`, we would fail to
// find an equivalent.
let upper_bound = self.non_local_universal_upper_bound(region_vid);
if self.region_contains(region_vid, upper_bound) {
self.definitions[upper_bound].external_name.unwrap_or(r)
} else {
// In the case of a failure, use a `ReVar` result. This will
// cause the `needs_infer` later on to return `None`.
r
}
// To do so, we simply check every candidate `u_r` for equality.
self.scc_values
.universal_regions_outlived_by(r_scc)
.filter(|&u_r| !self.universal_regions.is_local_free_region(u_r))
.find(|&u_r| self.eval_equal(u_r, r_vid))
.map(|u_r| tcx.mk_re_var(u_r))
// In the case of a failure, use `ReErased`. We will eventually
// return `None` in this case.
.unwrap_or(tcx.lifetimes.re_erased)
});
debug!("try_promote_type_test_subject: folded ty = {:?}", ty);
// `needs_infer` will only be true if we failed to promote some region.
if ty.needs_infer() {
// This will be true if we failed to promote some region.
if ty.has_erased_regions() {
return None;
}
Some(ClosureOutlivesSubject::Ty(ty))
}
/// Given some universal or existential region `r`, finds a
/// non-local, universal region `r+` that outlives `r` at entry to (and
/// exit from) the closure. In the worst case, this will be
/// `'static`.
///
/// This is used for two purposes. First, if we are propagated
/// some requirement `T: r`, we can use this method to enlarge `r`
/// to something we can encode for our creator (which only knows
/// about non-local, universal regions). It is also used when
/// encoding `T` as part of `try_promote_type_test_subject` (see
/// that fn for details).
///
/// This is based on the result `'y` of `universal_upper_bound`,
/// except that it converts further takes the non-local upper
/// bound of `'y`, so that the final result is non-local.
fn non_local_universal_upper_bound(&self, r: RegionVid) -> RegionVid {
debug!("non_local_universal_upper_bound(r={:?}={})", r, self.region_value_str(r));
let lub = self.universal_upper_bound(r);
// Grow further to get smallest universal region known to
// creator.
let non_local_lub = self.universal_region_relations.non_local_upper_bound(lub);
debug!("non_local_universal_upper_bound: non_local_lub={:?}", non_local_lub);
non_local_lub
Some(ClosureOutlivesSubject::Ty(ClosureOutlivesSubjectTy::bind(tcx, ty)))
}
/// Returns a universally quantified region that outlives the
@ -1539,6 +1505,8 @@ impl<'tcx> RegionInferenceContext<'tcx> {
// the outlives suggestions or the debug output from `#[rustc_regions]` would be
// duplicated. The polonius subset errors are deduplicated here, while keeping the
// CFG-location ordering.
// We can iterate the HashMap here because the result is sorted afterwards.
#[allow(rustc::potential_query_instability)]
let mut subset_errors: Vec<_> = polonius_output
.subset_errors
.iter()

View File

@ -1,4 +1,4 @@
use rustc_data_structures::fx::{FxHashMap, FxIndexSet};
use rustc_data_structures::fx::{FxIndexMap, FxIndexSet};
use rustc_data_structures::vec_map::VecMap;
use rustc_errors::ErrorGuaranteed;
use rustc_hir::def_id::LocalDefId;
@ -65,7 +65,7 @@ impl<'tcx> RegionInferenceContext<'tcx> {
) -> VecMap<LocalDefId, OpaqueHiddenType<'tcx>> {
let mut result: VecMap<LocalDefId, OpaqueHiddenType<'tcx>> = VecMap::new();
let member_constraints: FxHashMap<_, _> = self
let member_constraints: FxIndexMap<_, _> = self
.member_constraints
.all_indices()
.map(|ci| (self.member_constraints[ci].key, ci))
@ -364,7 +364,7 @@ fn check_opaque_type_parameter_valid(
OpaqueTyOrigin::TyAlias => {}
}
let opaque_generics = tcx.generics_of(opaque_type_key.def_id);
let mut seen_params: FxHashMap<_, Vec<_>> = FxHashMap::default();
let mut seen_params: FxIndexMap<_, Vec<_>> = FxIndexMap::default();
for (i, arg) in opaque_type_key.substs.iter().enumerate() {
let arg_is_param = match arg.unpack() {
GenericArgKind::Type(ty) => matches!(ty.kind(), ty::Param(_)),

View File

@ -3,7 +3,7 @@
use crate::constraints::ConstraintSccIndex;
use crate::RegionInferenceContext;
use itertools::Itertools;
use rustc_data_structures::fx::{FxHashMap, FxHashSet};
use rustc_data_structures::fx::{FxIndexMap, FxIndexSet};
use rustc_data_structures::graph::vec_graph::VecGraph;
use rustc_data_structures::graph::WithSuccessors;
use rustc_middle::ty::RegionVid;
@ -14,7 +14,7 @@ pub(crate) struct ReverseSccGraph {
graph: VecGraph<ConstraintSccIndex>,
/// For each SCC, the range of `universal_regions` that use that SCC as
/// their value.
scc_regions: FxHashMap<ConstraintSccIndex, Range<usize>>,
scc_regions: FxIndexMap<ConstraintSccIndex, Range<usize>>,
/// All of the universal regions, in grouped so that `scc_regions` can
/// index into here.
universal_regions: Vec<RegionVid>,
@ -26,7 +26,7 @@ impl ReverseSccGraph {
&'a self,
scc0: ConstraintSccIndex,
) -> impl Iterator<Item = RegionVid> + 'a {
let mut duplicates = FxHashSet::default();
let mut duplicates = FxIndexSet::default();
self.graph
.depth_first_search(scc0)
.flat_map(move |scc1| {
@ -55,7 +55,7 @@ impl RegionInferenceContext<'_> {
paired_scc_regions.sort();
let universal_regions = paired_scc_regions.iter().map(|&(_, region)| region).collect();
let mut scc_regions = FxHashMap::default();
let mut scc_regions = FxIndexMap::default();
let mut start = 0;
for (scc, group) in &paired_scc_regions.into_iter().group_by(|(scc, _)| *scc) {
let group_size = group.count();

View File

@ -116,7 +116,9 @@ impl<'a, 'tcx> ConstraintConversion<'a, 'tcx> {
let outlived_region = closure_mapping[outlives_requirement.outlived_free_region];
let subject = match outlives_requirement.subject {
ClosureOutlivesSubject::Region(re) => closure_mapping[re].into(),
ClosureOutlivesSubject::Ty(ty) => ty.into(),
ClosureOutlivesSubject::Ty(subject_ty) => {
subject_ty.instantiate(self.tcx, |vid| closure_mapping[vid]).into()
}
};
self.category = outlives_requirement.category;

View File

@ -93,31 +93,6 @@ impl UniversalRegionRelations<'_> {
res
}
/// Returns the "postdominating" bound of the set of
/// `non_local_upper_bounds` for the given region.
pub(crate) fn non_local_upper_bound(&self, fr: RegionVid) -> RegionVid {
let upper_bounds = self.non_local_upper_bounds(fr);
// In case we find more than one, reduce to one for
// convenience. This is to prevent us from generating more
// complex constraints, but it will cause spurious errors.
let post_dom = self.inverse_outlives.mutual_immediate_postdominator(upper_bounds);
debug!("non_local_bound: post_dom={:?}", post_dom);
post_dom
.and_then(|post_dom| {
// If the mutual immediate postdom is not local, then
// there is no non-local result we can return.
if !self.universal_regions.is_local_free_region(post_dom) {
Some(post_dom)
} else {
None
}
})
.unwrap_or(self.universal_regions.fr_static)
}
/// Finds a "lower bound" for `fr` that is not local. In other
/// words, returns the largest (*) known region `fr1` that (a) is
/// outlived by `fr` and (b) is not local.

View File

@ -19,7 +19,7 @@ use super::{Locations, TypeChecker};
impl<'a, 'tcx> TypeChecker<'a, 'tcx> {
/// Check explicit closure signature annotation,
/// e.g., `|x: FxHashMap<_, &'static u32>| ...`.
/// e.g., `|x: FxIndexMap<_, &'static u32>| ...`.
#[instrument(skip(self, body), level = "debug")]
pub(super) fn check_signature_annotation(&mut self, body: &Body<'tcx>) {
let mir_def_id = body.source.def_id().expect_local();

View File

@ -1,4 +1,4 @@
use rustc_data_structures::fx::{FxHashMap, FxHashSet};
use rustc_data_structures::fx::{FxIndexMap, FxIndexSet};
use rustc_index::bit_set::HybridBitSet;
use rustc_index::interval::IntervalSet;
use rustc_infer::infer::canonical::QueryRegionConstraints;
@ -56,7 +56,7 @@ pub(super) fn trace<'mir, 'tcx>(
elements,
local_use_map,
move_data,
drop_data: FxHashMap::default(),
drop_data: FxIndexMap::default(),
};
let mut results = LivenessResults::new(cx);
@ -85,7 +85,7 @@ struct LivenessContext<'me, 'typeck, 'flow, 'tcx> {
move_data: &'me MoveData<'tcx>,
/// Cache for the results of `dropck_outlives` query.
drop_data: FxHashMap<Ty<'tcx>, DropData<'tcx>>,
drop_data: FxIndexMap<Ty<'tcx>, DropData<'tcx>>,
/// Results of dataflow tracking which variables (and paths) have been
/// initialized.
@ -185,7 +185,7 @@ impl<'me, 'typeck, 'flow, 'tcx> LivenessResults<'me, 'typeck, 'flow, 'tcx> {
fn add_extra_drop_facts(
&mut self,
drop_used: Vec<(Local, Location)>,
relevant_live_locals: FxHashSet<Local>,
relevant_live_locals: FxIndexSet<Local>,
) {
let locations = IntervalSet::new(self.cx.elements.num_points());
@ -435,8 +435,7 @@ impl<'me, 'typeck, 'flow, 'tcx> LivenessResults<'me, 'typeck, 'flow, 'tcx> {
//
// What we *actually* generate is a store to a temporary
// for the call (`TMP = call()...`) and then a
// `DropAndReplace` to swap that with `X`
// (`DropAndReplace` has very particular semantics).
// `Drop(X)` followed by `X = TMP` to swap that with `X`.
}
}

View File

@ -9,7 +9,7 @@ use either::Either;
use hir::OpaqueTyOrigin;
use rustc_data_structures::frozen::Frozen;
use rustc_data_structures::fx::{FxHashMap, FxHashSet};
use rustc_data_structures::fx::{FxIndexMap, FxIndexSet};
use rustc_data_structures::vec_map::VecMap;
use rustc_hir as hir;
use rustc_hir::def::DefKind;
@ -145,7 +145,7 @@ pub(crate) fn type_check<'mir, 'tcx>(
outlives_constraints: OutlivesConstraintSet::default(),
member_constraints: MemberConstraintSet::default(),
type_tests: Vec::default(),
universe_causes: FxHashMap::default(),
universe_causes: FxIndexMap::default(),
};
let CreateResult {
@ -772,7 +772,9 @@ impl<'a, 'b, 'tcx> TypeVerifier<'a, 'b, 'tcx> {
match context {
PlaceContext::MutatingUse(_) => ty::Invariant,
PlaceContext::NonUse(StorageDead | StorageLive | VarDebugInfo) => ty::Invariant,
PlaceContext::NonUse(StorageDead | StorageLive | PlaceMention | VarDebugInfo) => {
ty::Invariant
}
PlaceContext::NonMutatingUse(
Inspect | Copy | Move | SharedBorrow | ShallowBorrow | UniqueBorrow | AddressOf
| Projection,
@ -875,7 +877,7 @@ struct TypeChecker<'a, 'tcx> {
user_type_annotations: &'a CanonicalUserTypeAnnotations<'tcx>,
region_bound_pairs: &'a RegionBoundPairs<'tcx>,
implicit_region_bound: ty::Region<'tcx>,
reported_errors: FxHashSet<(Ty<'tcx>, Span)>,
reported_errors: FxIndexSet<(Ty<'tcx>, Span)>,
borrowck_context: &'a mut BorrowCheckContext<'a, 'tcx>,
}
@ -925,7 +927,7 @@ pub(crate) struct MirTypeckRegionConstraints<'tcx> {
pub(crate) member_constraints: MemberConstraintSet<'tcx, RegionVid>,
pub(crate) universe_causes: FxHashMap<ty::UniverseIndex, UniverseInfo<'tcx>>,
pub(crate) universe_causes: FxIndexMap<ty::UniverseIndex, UniverseInfo<'tcx>>,
pub(crate) type_tests: Vec<TypeTest<'tcx>>,
}
@ -1282,6 +1284,7 @@ impl<'a, 'tcx> TypeChecker<'a, 'tcx> {
| StatementKind::Retag { .. }
| StatementKind::Coverage(..)
| StatementKind::ConstEvalCounter
| StatementKind::PlaceMention(..)
| StatementKind::Nop => {}
StatementKind::Deinit(..) | StatementKind::SetDiscriminant { .. } => {
bug!("Statement not allowed in this MIR phase")
@ -1312,24 +1315,6 @@ impl<'a, 'tcx> TypeChecker<'a, 'tcx> {
// no checks needed for these
}
TerminatorKind::DropAndReplace { place, value, target: _, unwind: _ } => {
let place_ty = place.ty(body, tcx).ty;
let rv_ty = value.ty(body, tcx);
let locations = term_location.to_locations();
if let Err(terr) =
self.sub_types(rv_ty, place_ty, locations, ConstraintCategory::Assignment)
{
span_mirbug!(
self,
term,
"bad DropAndReplace ({:?} = {:?}): {:?}",
place_ty,
rv_ty,
terr
);
}
}
TerminatorKind::SwitchInt { discr, .. } => {
self.check_operand(discr, term_location);
@ -1629,7 +1614,6 @@ impl<'a, 'tcx> TypeChecker<'a, 'tcx> {
}
TerminatorKind::Unreachable => {}
TerminatorKind::Drop { target, unwind, .. }
| TerminatorKind::DropAndReplace { target, unwind, .. }
| TerminatorKind::Assert { target, cleanup: unwind, .. } => {
self.assert_iscleanup(body, block_data, target, is_cleanup);
if let Some(unwind) = unwind {

View File

@ -312,6 +312,9 @@ impl<'tcx> UniversalRegions<'tcx> {
}
/// Gets an iterator over all the early-bound regions that have names.
/// Iteration order may be unstable, so this should only be used when
/// iteration order doesn't affect anything
#[allow(rustc::potential_query_instability)]
pub fn named_universal_regions<'s>(
&'s self,
) -> impl Iterator<Item = (ty::Region<'tcx>, ty::RegionVid)> + 's {

View File

@ -1,6 +1,6 @@
#![deny(rustc::untranslatable_diagnostic)]
#![deny(rustc::diagnostic_outside_of_impl)]
use rustc_data_structures::fx::FxHashSet;
use rustc_data_structures::fx::FxIndexSet;
use rustc_middle::mir::visit::{PlaceContext, Visitor};
use rustc_middle::mir::{
Local, Location, Place, Statement, StatementKind, Terminator, TerminatorKind,
@ -26,8 +26,8 @@ impl<'cx, 'tcx> MirBorrowckCtxt<'cx, 'tcx> {
/// See #55344 for context.
pub(crate) fn gather_used_muts(
&mut self,
temporary_used_locals: FxHashSet<Local>,
mut never_initialized_mut_locals: FxHashSet<Local>,
temporary_used_locals: FxIndexSet<Local>,
mut never_initialized_mut_locals: FxIndexSet<Local>,
) {
{
let mut visitor = GatherUsedMutsVisitor {
@ -48,8 +48,8 @@ impl<'cx, 'tcx> MirBorrowckCtxt<'cx, 'tcx> {
/// MIR visitor for collecting used mutable variables.
/// The 'visit lifetime represents the duration of the MIR walk.
struct GatherUsedMutsVisitor<'visit, 'cx, 'tcx> {
temporary_used_locals: FxHashSet<Local>,
never_initialized_mut_locals: &'visit mut FxHashSet<Local>,
temporary_used_locals: FxIndexSet<Local>,
never_initialized_mut_locals: &'visit mut FxIndexSet<Local>,
mbcx: &'visit mut MirBorrowckCtxt<'cx, 'tcx>,
}
@ -71,9 +71,6 @@ impl<'visit, 'cx, 'tcx> Visitor<'tcx> for GatherUsedMutsVisitor<'visit, 'cx, 'tc
TerminatorKind::Call { destination, .. } => {
self.remove_never_initialized_mut_locals(*destination);
}
TerminatorKind::DropAndReplace { place, .. } => {
self.remove_never_initialized_mut_locals(*place);
}
_ => {}
}

View File

@ -499,7 +499,6 @@ fn codegen_fn_body(fx: &mut FunctionCx<'_, '_, '_>, start_block: Block) {
TerminatorKind::Yield { .. }
| TerminatorKind::FalseEdge { .. }
| TerminatorKind::FalseUnwind { .. }
| TerminatorKind::DropAndReplace { .. }
| TerminatorKind::GeneratorDrop => {
bug!("shouldn't exist at codegen {:?}", bb_data.terminator());
}
@ -820,6 +819,7 @@ fn codegen_stmt<'tcx>(
| StatementKind::Nop
| StatementKind::FakeRead(..)
| StatementKind::Retag { .. }
| StatementKind::PlaceMention(..)
| StatementKind::AscribeUserType(..) => {}
StatementKind::Coverage { .. } => fx.tcx.sess.fatal("-Zcoverage is unimplemented"),

View File

@ -529,6 +529,7 @@ pub(crate) fn mir_operand_get_const_val<'tcx>(
| StatementKind::StorageDead(_)
| StatementKind::Retag(_, _)
| StatementKind::AscribeUserType(_, _)
| StatementKind::PlaceMention(..)
| StatementKind::Coverage(_)
| StatementKind::ConstEvalCounter
| StatementKind::Nop => {}
@ -543,8 +544,7 @@ pub(crate) fn mir_operand_get_const_val<'tcx>(
| TerminatorKind::Unreachable
| TerminatorKind::Drop { .. }
| TerminatorKind::Assert { .. } => {}
TerminatorKind::DropAndReplace { .. }
| TerminatorKind::Yield { .. }
TerminatorKind::Yield { .. }
| TerminatorKind::GeneratorDrop
| TerminatorKind::FalseEdge { .. }
| TerminatorKind::FalseUnwind { .. } => unreachable!(),

View File

@ -281,8 +281,12 @@ fn codegen_float_intrinsic_call<'tcx>(
sym::ceilf64 => ("ceil", 1, fx.tcx.types.f64),
sym::truncf32 => ("truncf", 1, fx.tcx.types.f32),
sym::truncf64 => ("trunc", 1, fx.tcx.types.f64),
sym::rintf32 => ("rintf", 1, fx.tcx.types.f32),
sym::rintf64 => ("rint", 1, fx.tcx.types.f64),
sym::roundf32 => ("roundf", 1, fx.tcx.types.f32),
sym::roundf64 => ("round", 1, fx.tcx.types.f64),
sym::roundevenf32 => ("roundevenf", 1, fx.tcx.types.f32),
sym::roundevenf64 => ("roundeven", 1, fx.tcx.types.f64),
sym::sinf32 => ("sinf", 1, fx.tcx.types.f32),
sym::sinf64 => ("sin", 1, fx.tcx.types.f64),
sym::cosf32 => ("cosf", 1, fx.tcx.types.f32),

View File

@ -4,36 +4,72 @@ on:
- push
- pull_request
permissions:
contents: read
env:
# Enable backtraces for easier debugging
RUST_BACKTRACE: 1
jobs:
build:
runs-on: ubuntu-latest
runs-on: ubuntu-22.04
strategy:
fail-fast: false
matrix:
libgccjit_version: ["libgccjit.so", "libgccjit_without_int128.so", "libgccjit12.so"]
libgccjit_version:
- { gcc: "libgccjit.so", extra: "", env_extra: "", artifacts_branch: "master" }
- { gcc: "libgccjit_without_int128.so", extra: "", env_extra: "", artifacts_branch: "master-without-128bit-integers" }
- { gcc: "libgccjit12.so", extra: "--no-default-features", env_extra: "TEST_FLAGS='-Cpanic=abort -Zpanic-abort-tests'", artifacts_branch: "gcc12" }
commands: [
"--mini-tests",
"--std-tests",
# FIXME: re-enable asm tests when GCC can emit in the right syntax.
# "--asm-tests",
"--test-libcore",
"--extended-rand-tests",
"--extended-regex-example-tests",
"--extended-regex-tests",
"--test-successful-rustc --nb-parts 2 --current-part 0",
"--test-successful-rustc --nb-parts 2 --current-part 1",
"--test-failing-rustc",
]
steps:
- uses: actions/checkout@v2
- uses: actions/checkout@v3
- uses: actions/checkout@v2
- uses: actions/checkout@v3
with:
repository: llvm/llvm-project
path: llvm
- name: Install packages
run: sudo apt-get install ninja-build ripgrep
# `llvm-14-tools` is needed to install the `FileCheck` binary which is used for asm tests.
run: sudo apt-get install ninja-build ripgrep llvm-14-tools
- name: Install libgccjit12
if: matrix.libgccjit_version.gcc == 'libgccjit12.so'
run: sudo apt-get install libgccjit-12-dev
- name: Download artifact
if: matrix.libgccjit_version.gcc != 'libgccjit12.so'
uses: dawidd6/action-download-artifact@v2
with:
workflow: main.yml
name: ${{ matrix.libgccjit_version }}
name: ${{ matrix.libgccjit_version.gcc }}
path: gcc-build
repo: antoyo/gcc
branch: ${{ matrix.libgccjit_version.artifacts_branch }}
event: push
search_artifacts: true # Because, instead, the action only check the last job ran and that won't work since we want multiple artifacts.
- name: Setup path to libgccjit
if: matrix.libgccjit_version.gcc == 'libgccjit12.so'
run: echo /usr/lib/gcc/x86_64-linux-gnu/12 > gcc_path
- name: Setup path to libgccjit
if: matrix.libgccjit_version.gcc != 'libgccjit12.so'
run: |
echo $(readlink -f gcc-build) > gcc_path
# NOTE: the filename is still libgccjit.so even when the artifact name is different.
@ -48,49 +84,44 @@ jobs:
- name: Set RUST_COMPILER_RT_ROOT
run: echo "RUST_COMPILER_RT_ROOT="${{ env.workspace }}/llvm/compiler-rt >> $GITHUB_ENV
# https://github.com/actions/cache/issues/133
- name: Fixup owner of ~/.cargo/
# Don't remove the trailing /. It is necessary to follow the symlink.
run: sudo chown -R $(whoami):$(id -ng) ~/.cargo/
- name: Cache cargo installed crates
uses: actions/cache@v1.1.2
uses: actions/cache@v3
with:
path: ~/.cargo/bin
key: cargo-installed-crates2-ubuntu-latest
- name: Cache cargo registry
uses: actions/cache@v1
uses: actions/cache@v3
with:
path: ~/.cargo/registry
key: ${{ runner.os }}-cargo-registry2-${{ hashFiles('**/Cargo.lock') }}
- name: Cache cargo index
uses: actions/cache@v1
uses: actions/cache@v3
with:
path: ~/.cargo/git
key: ${{ runner.os }}-cargo-index-${{ hashFiles('**/Cargo.lock') }}
- name: Cache cargo target dir
uses: actions/cache@v1.1.2
uses: actions/cache@v3
with:
path: target
key: ${{ runner.os }}-cargo-build-target-${{ hashFiles('rust-toolchain') }}
- name: Build
if: matrix.libgccjit_version != 'libgccjit12.so'
run: |
./prepare_build.sh
./build.sh
cargo test
./clean_all.sh
#- name: Cache rust repository
## We only clone the rust repository for rustc tests
#if: ${{ contains(matrix.commands, 'rustc') }}
#uses: actions/cache@v3
#id: cache-rust-repository
#with:
#path: rust
#key: ${{ runner.os }}-packages-${{ hashFiles('rust/.git/HEAD') }}
- name: Build
if: matrix.libgccjit_version == 'libgccjit12.so'
run: |
./prepare_build.sh
./build.sh --no-default-features
cargo test --no-default-features
${{ matrix.libgccjit_version.env_extra }} ./build.sh ${{ matrix.libgccjit_version.extra }}
${{ matrix.libgccjit_version.env_extra }} cargo test ${{ matrix.libgccjit_version.extra }}
./clean_all.sh
- name: Prepare dependencies
@ -106,26 +137,16 @@ jobs:
command: build
args: --release
- name: Test
if: matrix.libgccjit_version != 'libgccjit12.so'
- name: Add more failing tests for GCC 12
if: ${{ matrix.libgccjit_version.gcc == 'libgccjit12.so' }}
run: cat failing-ui-tests12.txt >> failing-ui-tests.txt
- name: Run tests
run: |
# Enable backtraces for easier debugging
export RUST_BACKTRACE=1
${{ matrix.libgccjit_version.env_extra }} ./test.sh --release --clean --build-sysroot ${{ matrix.commands }} ${{ matrix.libgccjit_version.extra }}
# Reduce amount of benchmark runs as they are slow
export COMPILE_RUNS=2
export RUN_RUNS=2
./test.sh --release
- name: Test
if: matrix.libgccjit_version == 'libgccjit12.so'
run: |
# Enable backtraces for easier debugging
export RUST_BACKTRACE=1
# Reduce amount of benchmark runs as they are slow
export COMPILE_RUNS=2
export RUN_RUNS=2
./test.sh --release --no-default-features
duplicates:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
- run: python tools/check_intrinsics_duplicates.py

View File

@ -0,0 +1,111 @@
name: CI with sysroot compiled in release mode
on:
- push
- pull_request
permissions:
contents: read
env:
# Enable backtraces for easier debugging
RUST_BACKTRACE: 1
jobs:
build:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
libgccjit_version:
- { gcc: "libgccjit.so", artifacts_branch: "master" }
commands: [
"--test-successful-rustc --nb-parts 2 --current-part 0",
"--test-successful-rustc --nb-parts 2 --current-part 1",
]
steps:
- uses: actions/checkout@v3
- uses: actions/checkout@v3
with:
repository: llvm/llvm-project
path: llvm
- name: Install packages
run: sudo apt-get install ninja-build ripgrep
- name: Download artifact
uses: dawidd6/action-download-artifact@v2
with:
workflow: main.yml
name: ${{ matrix.libgccjit_version.gcc }}
path: gcc-build
repo: antoyo/gcc
branch: ${{ matrix.libgccjit_version.artifacts_branch }}
event: push
search_artifacts: true # Because, instead, the action only check the last job ran and that won't work since we want multiple artifacts.
- name: Setup path to libgccjit
run: |
echo $(readlink -f gcc-build) > gcc_path
# NOTE: the filename is still libgccjit.so even when the artifact name is different.
ln gcc-build/libgccjit.so gcc-build/libgccjit.so.0
- name: Set env
run: |
echo "LIBRARY_PATH=$(cat gcc_path)" >> $GITHUB_ENV
echo "LD_LIBRARY_PATH=$(cat gcc_path)" >> $GITHUB_ENV
echo "workspace="$GITHUB_WORKSPACE >> $GITHUB_ENV
- name: Set RUST_COMPILER_RT_ROOT
run: echo "RUST_COMPILER_RT_ROOT="${{ env.workspace }}/llvm/compiler-rt >> $GITHUB_ENV
- name: Cache cargo installed crates
uses: actions/cache@v3
with:
path: ~/.cargo/bin
key: cargo-installed-crates2-ubuntu-latest
- name: Cache cargo registry
uses: actions/cache@v3
with:
path: ~/.cargo/registry
key: ${{ runner.os }}-cargo-registry2-${{ hashFiles('**/Cargo.lock') }}
- name: Cache cargo index
uses: actions/cache@v3
with:
path: ~/.cargo/git
key: ${{ runner.os }}-cargo-index-${{ hashFiles('**/Cargo.lock') }}
- name: Cache cargo target dir
uses: actions/cache@v3
with:
path: target
key: ${{ runner.os }}-cargo-build-target-${{ hashFiles('rust-toolchain') }}
- name: Build
run: |
./prepare_build.sh
./build.sh --release --release-sysroot
cargo test
./clean_all.sh
- name: Prepare dependencies
run: |
git config --global user.email "user@example.com"
git config --global user.name "User"
./prepare.sh
# Compile is a separate step, as the actions-rs/cargo action supports error annotations
- name: Compile
uses: actions-rs/cargo@v1.0.3
with:
command: build
args: --release
- name: Run tests
run: |
./test.sh --release --clean --release-sysroot --build-sysroot ${{ matrix.commands }}

View File

@ -0,0 +1,116 @@
name: stdarch tests with sysroot compiled in release mode
on:
- push
- pull_request
permissions:
contents: read
env:
# Enable backtraces for easier debugging
RUST_BACKTRACE: 1
jobs:
build:
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
libgccjit_version:
- { gcc: "libgccjit.so", artifacts_branch: "master" }
commands: [
"--test-successful-rustc --nb-parts 2 --current-part 0",
"--test-successful-rustc --nb-parts 2 --current-part 1",
]
steps:
- uses: actions/checkout@v3
- uses: actions/checkout@v3
with:
repository: llvm/llvm-project
path: llvm
- name: Install packages
run: sudo apt-get install ninja-build ripgrep
- name: Download artifact
uses: dawidd6/action-download-artifact@v2
with:
workflow: main.yml
name: ${{ matrix.libgccjit_version.gcc }}
path: gcc-build
repo: antoyo/gcc
branch: ${{ matrix.libgccjit_version.artifacts_branch }}
event: push
search_artifacts: true # Because, instead, the action only check the last job ran and that won't work since we want multiple artifacts.
- name: Setup path to libgccjit
run: |
echo $(readlink -f gcc-build) > gcc_path
# NOTE: the filename is still libgccjit.so even when the artifact name is different.
ln gcc-build/libgccjit.so gcc-build/libgccjit.so.0
- name: Set env
run: |
echo "LIBRARY_PATH=$(cat gcc_path)" >> $GITHUB_ENV
echo "LD_LIBRARY_PATH=$(cat gcc_path)" >> $GITHUB_ENV
echo "workspace="$GITHUB_WORKSPACE >> $GITHUB_ENV
- name: Set RUST_COMPILER_RT_ROOT
run: echo "RUST_COMPILER_RT_ROOT="${{ env.workspace }}/llvm/compiler-rt >> $GITHUB_ENV
- name: Cache cargo installed crates
uses: actions/cache@v3
with:
path: ~/.cargo/bin
key: cargo-installed-crates2-ubuntu-latest
- name: Cache cargo registry
uses: actions/cache@v3
with:
path: ~/.cargo/registry
key: ${{ runner.os }}-cargo-registry2-${{ hashFiles('**/Cargo.lock') }}
- name: Cache cargo index
uses: actions/cache@v3
with:
path: ~/.cargo/git
key: ${{ runner.os }}-cargo-index-${{ hashFiles('**/Cargo.lock') }}
- name: Cache cargo target dir
uses: actions/cache@v3
with:
path: target
key: ${{ runner.os }}-cargo-build-target-${{ hashFiles('rust-toolchain') }}
- name: Build
run: |
./prepare_build.sh
./build.sh --release --release-sysroot
cargo test
./clean_all.sh
- name: Prepare dependencies
run: |
git config --global user.email "user@example.com"
git config --global user.name "User"
./prepare.sh
# Compile is a separate step, as the actions-rs/cargo action supports error annotations
- name: Compile
uses: actions-rs/cargo@v1.0.3
with:
command: build
args: --release
- name: Run tests
run: |
./test.sh --release --clean --release-sysroot --build-sysroot --mini-tests --std-tests --test-libcore
- name: Run stdarch tests
run: |
cd build_sysroot/sysroot_src/library/stdarch/
CHANNEL=release TARGET=x86_64-unknown-linux-gnu ../../../../cargo.sh test

View File

@ -35,7 +35,7 @@ dependencies = [
[[package]]
name = "gccjit"
version = "1.0.0"
source = "git+https://github.com/antoyo/gccjit.rs#bdb86fb5092895ff5589726b33250010c64d93f6"
source = "git+https://github.com/antoyo/gccjit.rs#eefb8c662d61477f34b7c32d26bcda5f1ef08432"
dependencies = [
"gccjit_sys",
]
@ -43,7 +43,7 @@ dependencies = [
[[package]]
name = "gccjit_sys"
version = "0.0.1"
source = "git+https://github.com/antoyo/gccjit.rs#bdb86fb5092895ff5589726b33250010c64d93f6"
source = "git+https://github.com/antoyo/gccjit.rs#eefb8c662d61477f34b7c32d26bcda5f1ef08432"
dependencies = [
"libc 0.1.12",
]
@ -208,6 +208,7 @@ version = "0.1.0"
dependencies = [
"gccjit",
"lang_tester",
"smallvec",
"tempfile",
]
@ -220,6 +221,12 @@ dependencies = [
"winapi-util",
]
[[package]]
name = "smallvec"
version = "1.10.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "a507befe795404456341dfab10cef66ead4c041f62b8b11bbb92bffe5d0953e0"
[[package]]
name = "tempfile"
version = "3.2.0"

View File

@ -27,6 +27,8 @@ gccjit = { git = "https://github.com/antoyo/gccjit.rs" }
# Local copy.
#gccjit = { path = "../gccjit.rs" }
smallvec = { version = "1.6.1", features = ["union", "may_dangle"] }
[dev-dependencies]
lang_tester = "0.3.9"
tempfile = "3.1.0"

View File

@ -1,5 +1,7 @@
# WIP libgccjit codegen backend for rust
[![Chat on IRC](https://img.shields.io/badge/irc.libera.chat-%23rustc__codegen__gcc-blue.svg)](https://web.libera.chat/#rustc_codegen_gcc)
This is a GCC codegen for rustc, which means it can be loaded by the existing rustc frontend, but benefits from GCC: more architectures are supported and GCC's optimizations are used.
**Despite its name, libgccjit can be used for ahead-of-time compilation, as is used here.**
@ -16,21 +18,61 @@ The patches in [this repository](https://github.com/antoyo/libgccjit-patches) ne
(Those patches should work when applied on master, but in case it doesn't work, they are known to work when applied on 079c23cfe079f203d5df83fea8e92a60c7d7e878.)
You can also use my [fork of gcc](https://github.com/antoyo/gcc) which already includes these patches.**
To build it (most of these instructions come from [here](https://gcc.gnu.org/onlinedocs/jit/internals/index.html), so don't hesitate to take a look there if you encounter an issue):
```bash
$ git clone https://github.com/antoyo/gcc
$ sudo apt install flex libmpfr-dev libgmp-dev libmpc3 libmpc-dev
$ mkdir gcc-build gcc-install
$ cd gcc-build
$ ../gcc/configure \
--enable-host-shared \
--enable-languages=jit \
--enable-checking=release \ # it enables extra checks which allow to find bugs
--disable-bootstrap \
--disable-multilib \
--prefix=$(pwd)/../gcc-install
$ make -j4 # You can replace `4` with another number depending on how many cores you have.
```
If you want to run libgccjit tests, you will need to also enable the C++ language in the `configure`:
```bash
--enable-languages=jit,c++
```
Then to run libgccjit tests:
```bash
$ cd gcc # from the `gcc-build` folder
$ make check-jit
# To run one specific test:
$ make check-jit RUNTESTFLAGS="-v -v -v jit.exp=jit.dg/test-asm.cc"
```
**Put the path to your custom build of libgccjit in the file `gcc_path`.**
```bash
$ git clone https://github.com/rust-lang/rustc_codegen_gcc.git
$ cd rustc_codegen_gcc
$ dirname $(readlink -f `find . -name libgccjit.so`) > gcc_path
```
You also need to set RUST_COMPILER_RT_ROOT:
```bash
$ git clone https://github.com/llvm/llvm-project llvm --depth 1 --single-branch
$ export RUST_COMPILER_RT_ROOT="$PWD/llvm/compiler-rt"
$ ./prepare_build.sh # download and patch sysroot src
$ ./build.sh --release
```
Then you can run commands like this:
```bash
$ ./prepare.sh # download and patch sysroot src and install hyperfine for benchmarking
$ LIBRARY_PATH=$(cat gcc_path) LD_LIBRARY_PATH=$(cat gcc_path) ./build.sh --release
```
To run the tests:
```bash
$ ./prepare.sh # download and patch sysroot src and install hyperfine for benchmarking
$ ./test.sh --release
```
@ -120,13 +162,52 @@ To print a debug representation of a tree:
debug_tree(expr);
```
(defined in print-tree.h)
To print a debug reprensentation of a gimple struct:
```c
debug_gimple_stmt(gimple_struct)
```
To get the `rustc` command to run in `gdb`, add the `--verbose` flag to `cargo build`.
To have the correct file paths in `gdb` instead of `/usr/src/debug/gcc/libstdc++-v3/libsupc++/eh_personality.cc`:
Maybe by calling the following at the beginning of gdb:
```
set substitute-path /usr/src/debug/gcc /path/to/gcc-repo/gcc
```
TODO(antoyo): but that's not what I remember I was doing.
### How to use a custom-build rustc
* Build the stage2 compiler (`rustup toolchain link debug-current build/x86_64-unknown-linux-gnu/stage2`).
* Clean and rebuild the codegen with `debug-current` in the file `rust-toolchain`.
### How to install a forked git-subtree
Using git-subtree with `rustc` requires a patched git to make it work.
The PR that is needed is [here](https://github.com/gitgitgadget/git/pull/493).
Use the following instructions to install it:
```
git clone git@github.com:tqc/git.git
cd git
git checkout tqc/subtree
make
make install
cd contrib/subtree
make
cp git-subtree ~/bin
```
### How to use [mem-trace](https://github.com/antoyo/mem-trace)
`rustc` needs to be built without `jemalloc` so that `mem-trace` can overload `malloc` since `jemalloc` is linked statically, so a `LD_PRELOAD`-ed library won't a chance to intercept the calls to `malloc`.
### How to build a cross-compiling libgccjit
#### Building libgccjit
@ -142,6 +223,5 @@ To get the `rustc` command to run in `gdb`, add the `--verbose` flag to `cargo b
* Since rustc doesn't support this architecture yet, set it back to `TARGET_TRIPLE="mips-unknown-linux-gnu"` (or another target having the same attributes). Alternatively, create a [target specification file](https://book.avr-rust.com/005.1-the-target-specification-json-file.html) (note that the `arch` specified in this file must be supported by the rust compiler).
* Set `linker='-Clinker=m68k-linux-gcc'`.
* Set the path to the cross-compiling libgccjit in `gcc_path`.
* Disable the 128-bit integer types if the target doesn't support them by using `let i128_type = context.new_type::<i64>();` in `context.rs` (same for u128_type).
* Comment the line: `context.add_command_line_option("-masm=intel");` in src/base.rs.
* (might not be necessary) Disable the compilation of libstd.so (and possibly libcore.so?).

View File

@ -16,7 +16,7 @@ rm Cargo.lock test_target/Cargo.lock 2>/dev/null || true
rm -r sysroot/ 2>/dev/null || true
# Build libs
export RUSTFLAGS="$RUSTFLAGS -Z force-unstable-if-unmarked -Cpanic=abort"
export RUSTFLAGS="$RUSTFLAGS -Z force-unstable-if-unmarked"
if [[ "$1" == "--release" ]]; then
sysroot_channel='release'
RUSTFLAGS="$RUSTFLAGS -Zmir-opt-level=3" cargo build --target $TARGET_TRIPLE --release

View File

@ -38,7 +38,7 @@ if [[ "$HOST_TRIPLE" != "$TARGET_TRIPLE" ]]; then
fi
fi
export RUSTFLAGS="$CG_RUSTFLAGS $linker -Cpanic=abort -Csymbol-mangling-version=v0 -Cdebuginfo=2 -Clto=off -Zpanic-abort-tests -Zcodegen-backend=$(pwd)/target/${CHANNEL:-debug}/librustc_codegen_gcc.$dylib_ext --sysroot $(pwd)/build_sysroot/sysroot"
export RUSTFLAGS="$CG_RUSTFLAGS $linker -Csymbol-mangling-version=v0 -Cdebuginfo=2 -Clto=off -Zcodegen-backend=$(pwd)/target/${CHANNEL:-debug}/librustc_codegen_gcc.$dylib_ext --sysroot $(pwd)/build_sysroot/sysroot $TEST_FLAGS"
# FIXME(antoyo): remove once the atomic shim is gone
if [[ `uname` == 'Darwin' ]]; then

View File

@ -1,4 +1,4 @@
#![feature(start, box_syntax, core_intrinsics, alloc_error_handler)]
#![feature(start, box_syntax, core_intrinsics, alloc_error_handler, lang_items)]
#![no_std]
extern crate alloc;
@ -18,16 +18,22 @@ extern "C" {
#[panic_handler]
fn panic_handler(_: &core::panic::PanicInfo) -> ! {
unsafe {
core::intrinsics::abort();
}
core::intrinsics::abort();
}
#[alloc_error_handler]
fn alloc_error_handler(_: alloc::alloc::Layout) -> ! {
unsafe {
core::intrinsics::abort();
}
core::intrinsics::abort();
}
#[lang = "eh_personality"]
fn eh_personality() -> ! {
loop {}
}
#[no_mangle]
unsafe extern "C" fn _Unwind_Resume() {
core::intrinsics::unreachable();
}
#[start]

View File

@ -1,6 +1,6 @@
#![feature(
no_core, lang_items, intrinsics, unboxed_closures, type_ascription, extern_types,
untagged_unions, decl_macro, rustc_attrs, transparent_unions, auto_traits,
decl_macro, rustc_attrs, transparent_unions, auto_traits,
thread_local
)]
#![no_core]
@ -17,6 +17,9 @@ pub trait Sized {}
#[lang = "destruct"]
pub trait Destruct {}
#[lang = "tuple_trait"]
pub trait Tuple {}
#[lang = "unsize"]
pub trait Unsize<T: ?Sized> {}
@ -39,14 +42,14 @@ impl<'a, T: ?Sized+Unsize<U>, U: ?Sized> DispatchFromDyn<&'a mut U> for &'a mut
impl<T: ?Sized+Unsize<U>, U: ?Sized> DispatchFromDyn<*const U> for *const T {}
// *mut T -> *mut U
impl<T: ?Sized+Unsize<U>, U: ?Sized> DispatchFromDyn<*mut U> for *mut T {}
impl<T: ?Sized + Unsize<U>, U: ?Sized> DispatchFromDyn<Box<U>> for Box<T> {}
impl<T: ?Sized + Unsize<U>, U: ?Sized> DispatchFromDyn<Box<U, ()>> for Box<T, ()> {}
#[lang = "receiver"]
pub trait Receiver {}
impl<T: ?Sized> Receiver for &T {}
impl<T: ?Sized> Receiver for &mut T {}
impl<T: ?Sized> Receiver for Box<T> {}
impl<T: ?Sized, A: Allocator> Receiver for Box<T, A> {}
#[lang = "copy"]
pub unsafe trait Copy {}
@ -396,7 +399,7 @@ pub struct PhantomData<T: ?Sized>;
#[lang = "fn_once"]
#[rustc_paren_sugar]
pub trait FnOnce<Args> {
pub trait FnOnce<Args: Tuple> {
#[lang = "fn_once_output"]
type Output;
@ -405,13 +408,21 @@ pub trait FnOnce<Args> {
#[lang = "fn_mut"]
#[rustc_paren_sugar]
pub trait FnMut<Args>: FnOnce<Args> {
pub trait FnMut<Args: Tuple>: FnOnce<Args> {
extern "rust-call" fn call_mut(&mut self, args: Args) -> Self::Output;
}
#[lang = "panic"]
#[track_caller]
pub fn panic(_msg: &str) -> ! {
pub fn panic(_msg: &'static str) -> ! {
unsafe {
libc::puts("Panicking\n\0" as *const str as *const u8);
intrinsics::abort();
}
}
#[lang = "panic_cannot_unwind"]
fn panic_cannot_unwind() -> ! {
unsafe {
libc::puts("Panicking\n\0" as *const str as *const u8);
intrinsics::abort();
@ -450,17 +461,32 @@ pub trait Deref {
pub trait Allocator {
}
impl Allocator for () {}
pub struct Global;
impl Allocator for Global {}
#[lang = "owned_box"]
pub struct Box<
T: ?Sized,
A: Allocator = Global,
>(*mut T, A);
#[repr(transparent)]
#[rustc_layout_scalar_valid_range_start(1)]
#[rustc_nonnull_optimization_guaranteed]
pub struct NonNull<T: ?Sized>(pub *const T);
impl<T: ?Sized + Unsize<U>, U: ?Sized> CoerceUnsized<Box<U>> for Box<T> {}
impl<T: ?Sized, U: ?Sized> CoerceUnsized<NonNull<U>> for NonNull<T> where T: Unsize<U> {}
impl<T: ?Sized, U: ?Sized> DispatchFromDyn<NonNull<U>> for NonNull<T> where T: Unsize<U> {}
pub struct Unique<T: ?Sized> {
pub pointer: NonNull<T>,
pub _marker: PhantomData<T>,
}
impl<T: ?Sized, U: ?Sized> CoerceUnsized<Unique<U>> for Unique<T> where T: Unsize<U> {}
impl<T: ?Sized, U: ?Sized> DispatchFromDyn<Unique<U>> for Unique<T> where T: Unsize<U> {}
#[lang = "owned_box"]
pub struct Box<T: ?Sized, A: Allocator = Global>(Unique<T>, A);
impl<T: ?Sized + Unsize<U>, U: ?Sized, A: Allocator> CoerceUnsized<Box<U, A>> for Box<T, A> {}
impl<T: ?Sized, A: Allocator> Drop for Box<T, A> {
fn drop(&mut self) {
@ -468,7 +494,7 @@ impl<T: ?Sized, A: Allocator> Drop for Box<T, A> {
}
}
impl<T> Deref for Box<T> {
impl<T: ?Sized, A: Allocator> Deref for Box<T, A> {
type Target = T;
fn deref(&self) -> &Self::Target {
@ -482,8 +508,8 @@ unsafe fn allocate(size: usize, _align: usize) -> *mut u8 {
}
#[lang = "box_free"]
unsafe fn box_free<T: ?Sized, A: Allocator>(ptr: *mut T, alloc: A) {
libc::free(ptr as *mut u8);
unsafe fn box_free<T: ?Sized>(ptr: Unique<T>, _alloc: ()) {
libc::free(ptr.pointer.0 as *mut u8);
}
#[lang = "drop"]
@ -505,17 +531,25 @@ pub union MaybeUninit<T> {
}
pub mod intrinsics {
use crate::Sized;
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
#[rustc_safe_intrinsic]
pub fn size_of<T>() -> usize;
pub fn size_of_val<T: ?::Sized>(val: *const T) -> usize;
pub fn size_of_val<T: ?Sized>(val: *const T) -> usize;
#[rustc_safe_intrinsic]
pub fn min_align_of<T>() -> usize;
pub fn min_align_of_val<T: ?::Sized>(val: *const T) -> usize;
pub fn min_align_of_val<T: ?Sized>(val: *const T) -> usize;
pub fn copy<T>(src: *const T, dst: *mut T, count: usize);
pub fn transmute<T, U>(e: T) -> U;
pub fn ctlz_nonzero<T>(x: T) -> T;
pub fn needs_drop<T: ?::Sized>() -> bool;
#[rustc_safe_intrinsic]
pub fn needs_drop<T: ?Sized>() -> bool;
#[rustc_safe_intrinsic]
pub fn bitreverse<T>(x: T) -> T;
#[rustc_safe_intrinsic]
pub fn bswap<T>(x: T) -> T;
pub fn write_bytes<T>(dst: *mut T, val: u8, count: usize);
pub fn unreachable() -> !;

View File

@ -85,6 +85,7 @@ fn start<T: Termination + 'static>(
main: fn() -> T,
argc: isize,
argv: *const *const u8,
_sigpipe: u8,
) -> isize {
if argc == 3 {
unsafe { puts(*argv); }
@ -228,6 +229,7 @@ fn main() {
} as Box<dyn SomeTrait>;
const FUNC_REF: Option<fn()> = Some(main);
#[allow(unreachable_code)]
match FUNC_REF {
Some(_) => {},
None => assert!(false),

View File

@ -6,9 +6,7 @@ extern {}
#[panic_handler]
fn panic_handler(_: &core::panic::PanicInfo) -> ! {
unsafe {
core::intrinsics::abort();
}
core::intrinsics::abort();
}
#[lang="eh_personality"]
@ -32,6 +30,6 @@ fn main(_argc: isize, _argv: *const *const u8) -> isize {
#[inline(never)]
fn black_box(i: u32) {
if i != 1 {
unsafe { core::intrinsics::abort(); }
core::intrinsics::abort();
}
}

View File

@ -1,5 +1,6 @@
#![feature(core_intrinsics, generators, generator_trait, is_sorted)]
#[cfg(feature="master")]
use std::arch::x86_64::*;
use std::io::Write;
use std::ops::Generator;

View File

@ -0,0 +1,68 @@
tests/ui/allocator/custom-in-block.rs
tests/ui/allocator/custom-in-submodule.rs
tests/ui/allocator/custom.rs
tests/ui/allocator/hygiene.rs
tests/ui/allocator/no_std-alloc-error-handler-custom.rs
tests/ui/allocator/no_std-alloc-error-handler-default.rs
tests/ui/allocator/xcrate-use.rs
tests/ui/allocator/xcrate-use2.rs
tests/ui/asm/may_unwind.rs
tests/ui/asm/x86_64/multiple-clobber-abi.rs
tests/ui/debuginfo/debuginfo-emit-llvm-ir-and-split-debuginfo.rs
tests/ui/functions-closures/parallel-codegen-closures.rs
tests/ui/linkage-attr/linkage1.rs
tests/ui/lto/dylib-works.rs
tests/ui/numbers-arithmetic/saturating-float-casts.rs
tests/ui/polymorphization/promoted-function.rs
tests/ui/process/nofile-limit.rs
tests/ui/sepcomp/sepcomp-cci.rs
tests/ui/sepcomp/sepcomp-extern.rs
tests/ui/sepcomp/sepcomp-fns-backwards.rs
tests/ui/sepcomp/sepcomp-fns.rs
tests/ui/sepcomp/sepcomp-statics.rs
tests/ui/simd/intrinsic/generic-arithmetic-pass.rs
tests/ui/sse2.rs
tests/ui/target-feature/missing-plusminus.rs
tests/ui/asm/x86_64/may_unwind.rs
tests/ui/backtrace.rs
tests/ui/catch-unwind-bang.rs
tests/ui/cfg/cfg-panic-abort.rs
tests/ui/drop/dynamic-drop-async.rs
tests/ui/drop/repeat-drop.rs
tests/ui/fmt/format-args-capture.rs
tests/ui/generator/panic-drops-resume.rs
tests/ui/generator/panic-drops.rs
tests/ui/intrinsics/panic-uninitialized-zeroed.rs
tests/ui/iterators/iter-sum-overflow-debug.rs
tests/ui/iterators/iter-sum-overflow-overflow-checks.rs
tests/ui/mir/mir_calls_to_shims.rs
tests/ui/mir/mir_drop_order.rs
tests/ui/mir/mir_let_chains_drop_order.rs
tests/ui/oom_unwind.rs
tests/ui/panic-runtime/abort-link-to-unwinding-crates.rs
tests/ui/panic-runtime/abort.rs
tests/ui/panic-runtime/link-to-abort.rs
tests/ui/unwind-no-uwtable.rs
tests/ui/parser/unclosed-delimiter-in-dep.rs
tests/ui/runtime/rt-explody-panic-payloads.rs
tests/ui/simd/intrinsic/ptr-cast.rs
tests/ui/binding/fn-arg-incomplete-pattern-drop-order.rs
tests/ui/consts/missing_span_in_backtrace.rs
tests/ui/drop/dynamic-drop.rs
tests/ui/dyn-star/box.rs
tests/ui/issues/issue-40883.rs
tests/ui/issues/issue-43853.rs
tests/ui/issues/issue-47364.rs
tests/ui/macros/rfc-2011-nicer-assert-messages/assert-without-captures-does-not-create-unnecessary-code.rs
tests/ui/rfc-2091-track-caller/std-panic-locations.rs
tests/ui/rfcs/rfc1857-drop-order.rs
tests/ui/simd/issue-17170.rs
tests/ui/simd/issue-39720.rs
tests/ui/simd/issue-89193.rs
tests/ui/statics/issue-91050-1.rs
tests/ui/statics/issue-91050-2.rs
tests/ui/alloc-error/default-alloc-error-hook.rs
tests/ui/generator/panic-safe.rs
tests/ui/issues/issue-14875.rs
tests/ui/issues/issue-29948.rs
tests/ui/panic-while-printing.rs

View File

@ -0,0 +1,39 @@
tests/ui/asm/x86_64/issue-96797.rs
tests/ui/intrinsics/const-eval-select-x86_64.rs
tests/ui/packed/packed-struct-drop-aligned.rs
tests/ui/packed/packed-struct-generic-layout.rs
tests/ui/packed/packed-struct-layout.rs
tests/ui/packed/packed-struct-optimized-enum.rs
tests/ui/packed/packed-struct-size.rs
tests/ui/packed/packed-struct-vec.rs
tests/ui/packed/packed-tuple-struct-layout.rs
tests/ui/simd/array-type.rs
tests/ui/simd/intrinsic/float-minmax-pass.rs
tests/ui/simd/intrinsic/generic-arithmetic-saturating-pass.rs
tests/ui/simd/intrinsic/generic-as.rs
tests/ui/simd/intrinsic/generic-cast-pass.rs
tests/ui/simd/intrinsic/generic-cast-pointer-width.rs
tests/ui/simd/intrinsic/generic-comparison-pass.rs
tests/ui/simd/intrinsic/generic-elements-pass.rs
tests/ui/simd/intrinsic/generic-reduction-pass.rs
tests/ui/simd/intrinsic/generic-select-pass.rs
tests/ui/simd/intrinsic/inlining-issue67557-ice.rs
tests/ui/simd/intrinsic/inlining-issue67557.rs
tests/ui/simd/monomorphize-shuffle-index.rs
tests/ui/simd/shuffle.rs
tests/ui/simd/simd-bitmask.rs
tests/ui/generator/resume-after-return.rs
tests/ui/iterators/iter-step-overflow-debug.rs
tests/ui/macros/rfc-2011-nicer-assert-messages/all-expr-kinds.rs
tests/ui/numbers-arithmetic/next-power-of-two-overflow-debug.rs
tests/ui/privacy/reachable-unnameable-items.rs
tests/ui/rfc-1937-termination-trait/termination-trait-in-test.rs
tests/ui/async-await/async-fn-size-moved-locals.rs
tests/ui/async-await/async-fn-size-uninit-locals.rs
tests/ui/cfg/cfg-panic.rs
tests/ui/generator/size-moved-locals.rs
tests/ui/macros/rfc-2011-nicer-assert-messages/all-not-available-cases.rs
tests/ui/simd/intrinsic/generic-gather-pass.rs
tests/ui/simd/issue-85915-simd-ptrs.rs
tests/ui/issues/issue-68010-large-zst-consts.rs
tests/ui/rust-2018/proc-macro-crate-in-paths.rs

View File

@ -60,3 +60,9 @@ codegen_gcc_invalid_monomorphization_unsupported_cast =
codegen_gcc_invalid_monomorphization_unsupported_operation =
invalid monomorphization of `{$name}` intrinsic: unsupported operation on `{$in_ty}` with element `{$in_elem}`
codegen_gcc_invalid_minimum_alignment =
invalid minimum global alignment: {$err}
codegen_gcc_tied_target_features = the target features {$features} must all be either enabled or disabled together
.help = add the missing features in a `target_feature` attribute

View File

@ -0,0 +1,39 @@
From c3821e02fbd6cb5ad6e06d759fccdc9073712375 Mon Sep 17 00:00:00 2001
From: Antoni Boucher <bouanto@zoho.com>
Date: Tue, 7 Jun 2022 21:40:13 -0400
Subject: [PATCH] Add stdarch Cargo.toml for testing
---
library/stdarch/Cargo.toml | 20 ++++++++++++++++++++
1 file changed, 20 insertions(+)
create mode 100644 library/stdarch/Cargo.toml
diff --git a/library/stdarch/Cargo.toml b/library/stdarch/Cargo.toml
new file mode 100644
index 0000000..fbe0a95
--- /dev/null
+++ b/library/stdarch/Cargo.toml
@@ -0,0 +1,20 @@
+[workspace]
+members = [
+ "crates/core_arch",
+ "crates/std_detect",
+ "crates/stdarch-gen",
+ "examples/"
+]
+exclude = [
+ "crates/wasm-assert-instr-tests"
+]
+
+[profile.release]
+debug = true
+opt-level = 3
+incremental = true
+
+[profile.bench]
+debug = 1
+opt-level = 3
+incremental = true
--
2.26.2.7.g19db9cfb68.dirty

View File

@ -0,0 +1,25 @@
From a2d53a324a02c04b76c0e9d39dc15cd443a3b8b2 Mon Sep 17 00:00:00 2001
From: Antoni Boucher <bouanto@zoho.com>
Date: Fri, 25 Nov 2022 11:18:11 -0500
Subject: [PATCH] Disable examples
---
library/stdarch/Cargo.toml | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/library/stdarch/Cargo.toml b/library/stdarch/Cargo.toml
index fbe0a95..748d72d 100644
--- a/library/stdarch/Cargo.toml
+++ b/library/stdarch/Cargo.toml
@@ -3,7 +3,7 @@ members = [
"crates/core_arch",
"crates/std_detect",
"crates/stdarch-gen",
- "examples/"
+ #"examples/"
]
exclude = [
"crates/wasm-assert-instr-tests"
--
2.26.2.7.g19db9cfb68.dirty

View File

@ -18,7 +18,7 @@ new file mode 100644
index 0000000..46fd999
--- /dev/null
+++ b/library/core/tests/Cargo.toml
@@ -0,0 +1,8 @@
@@ -0,0 +1,12 @@
+[package]
+name = "core"
+version = "0.0.0"
@ -27,37 +27,18 @@ index 0000000..46fd999
+[lib]
+name = "coretests"
+path = "lib.rs"
diff --git a/library/core/tests/num/flt2dec/mod.rs b/library/core/tests/num/flt2dec/mod.rs
index a35897e..f0bf645 100644
--- a/library/core/tests/num/flt2dec/mod.rs
+++ b/library/core/tests/num/flt2dec/mod.rs
@@ -13,7 +13,6 @@ mod strategy {
mod dragon;
mod grisu;
}
-mod random;
pub fn decode_finite<T: DecodableFloat>(v: T) -> Decoded {
match decode(v).1 {
diff --git a/library/core/tests/slice.rs b/library/core/tests/slice.rs
index 6609bc3..241b497 100644
--- a/library/core/tests/slice.rs
+++ b/library/core/tests/slice.rs
@@ -1209,6 +1209,7 @@ fn brute_force_rotate_test_1() {
}
}
+/*
#[test]
#[cfg(not(target_arch = "wasm32"))]
fn sort_unstable() {
@@ -1394,6 +1395,7 @@ fn partition_at_index() {
v.select_nth_unstable(0);
assert!(v == [0xDEADBEEF]);
}
+*/
#[test]
#[should_panic(expected = "index 0 greater than length of slice")]
+
+[dependencies]
+rand = { version = "0.8.5", default-features = false }
+rand_xorshift = { version = "0.3.0", default-features = false }
diff --git a/library/core/tests/lib.rs b/library/core/tests/lib.rs
index 42a26ae..5ac1042 100644
--- a/library/core/tests/lib.rs
+++ b/library/core/tests/lib.rs
@@ -1,3 +1,4 @@
+#![cfg(test)]
#![feature(alloc_layout_extra)]
#![feature(array_chunks)]
#![feature(array_methods)]
--
2.21.0 (Apple Git-122)

View File

@ -1,28 +0,0 @@
From b1ae000f6da1abd3b8e9b80c40bc11c89b8ae93c Mon Sep 17 00:00:00 2001
From: bjorn3 <bjorn3@users.noreply.github.com>
Date: Thu, 30 Dec 2021 16:54:40 +0100
Subject: [PATCH] [core] Disable portable-simd test
---
library/core/tests/lib.rs | 1 -
1 file changed, 1 deletion(-)
diff --git a/library/core/tests/lib.rs b/library/core/tests/lib.rs
index 06c7be0..359e2e7 100644
--- a/library/core/tests/lib.rs
+++ b/library/core/tests/lib.rs
@@ -75,7 +75,6 @@
#![feature(never_type)]
#![feature(unwrap_infallible)]
-#![feature(portable_simd)]
#![feature(ptr_metadata)]
#![feature(once_cell)]
#![feature(option_result_contains)]
@@ -127,7 +126,6 @@ mod pin;
mod pin_macro;
mod ptr;
mod result;
-mod simd;
mod slice;
mod str;
mod str_lossy;

View File

@ -1,3 +1,3 @@
[toolchain]
channel = "nightly-2022-06-06"
channel = "nightly-2023-03-02"
components = ["rust-src", "rustc-dev", "llvm-tools-preview"]

View File

@ -1,14 +0,0 @@
diff --git a/src/tools/compiletest/src/header.rs b/src/tools/compiletest/src/header.rs
index 887d27fd6dca4..2c2239f2b83d1 100644
--- a/src/tools/compiletest/src/header.rs
+++ b/src/tools/compiletest/src/header.rs
@@ -806,8 +806,8 @@ pub fn make_test_description<R: Read>(
cfg: Option<&str>,
) -> test::TestDesc {
let mut ignore = false;
#[cfg(not(bootstrap))]
- let ignore_message: Option<String> = None;
+ let ignore_message: Option<&str> = None;
let mut should_fail = false;
let rustc_has_profiler_support = env::var_os("RUSTC_PROFILER_SUPPORT").is_some();

View File

@ -1,3 +1,5 @@
#[cfg(feature="master")]
use gccjit::FnAttribute;
use gccjit::{FunctionType, GlobalKind, ToRValue};
use rustc_ast::expand::allocator::{AllocatorKind, AllocatorTy, ALLOCATOR_METHODS};
use rustc_middle::bug;
@ -50,7 +52,8 @@ pub(crate) unsafe fn codegen(tcx: TyCtxt<'_>, mods: &mut GccContext, _module_nam
let func = context.new_function(None, FunctionType::Exported, output.unwrap_or(void), &args, name, false);
if tcx.sess.target.options.default_hidden_visibility {
// TODO(antoyo): set visibility.
#[cfg(feature="master")]
func.add_attribute(FnAttribute::Visibility(gccjit::Visibility::Hidden));
}
if tcx.sess.must_emit_unwind_tables() {
// TODO(antoyo): emit unwind tables.
@ -61,7 +64,8 @@ pub(crate) unsafe fn codegen(tcx: TyCtxt<'_>, mods: &mut GccContext, _module_nam
.map(|(index, typ)| context.new_parameter(None, *typ, &format!("param{}", index)))
.collect();
let callee = context.new_function(None, FunctionType::Extern, output.unwrap_or(void), &args, callee, false);
// TODO(antoyo): set visibility.
#[cfg(feature="master")]
callee.add_attribute(FnAttribute::Visibility(gccjit::Visibility::Hidden));
let block = func.new_block("entry");
@ -90,12 +94,18 @@ pub(crate) unsafe fn codegen(tcx: TyCtxt<'_>, mods: &mut GccContext, _module_nam
.collect();
let func = context.new_function(None, FunctionType::Exported, void, &args, name, false);
if tcx.sess.target.default_hidden_visibility {
#[cfg(feature="master")]
func.add_attribute(FnAttribute::Visibility(gccjit::Visibility::Hidden));
}
let callee = alloc_error_handler_kind.fn_name(sym::oom);
let args: Vec<_> = types.iter().enumerate()
.map(|(index, typ)| context.new_parameter(None, *typ, &format!("param{}", index)))
.collect();
let callee = context.new_function(None, FunctionType::Extern, void, &args, callee, false);
//llvm::LLVMRustSetVisibility(callee, llvm::Visibility::Hidden);
#[cfg(feature="master")]
callee.add_attribute(FnAttribute::Visibility(gccjit::Visibility::Hidden));
let block = func.new_block("entry");

View File

@ -157,7 +157,7 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
use ConstraintOrRegister::*;
let (constraint, ty) = match (reg_to_gcc(reg), place) {
(Constraint(constraint), Some(place)) => (constraint, place.layout.gcc_type(self.cx, false)),
(Constraint(constraint), Some(place)) => (constraint, place.layout.gcc_type(self.cx)),
// When `reg` is a class and not an explicit register but the out place is not specified,
// we need to create an unused output variable to assign the output to. This var
// needs to be of a type that's "compatible" with the register class, but specific type
@ -226,7 +226,7 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
// This decision is also backed by the fact that LLVM needs in and out
// values to be of *exactly the same type*, not just "compatible".
// I'm not sure if GCC is so picky too, but better safe than sorry.
let ty = in_value.layout.gcc_type(self.cx, false);
let ty = in_value.layout.gcc_type(self.cx);
let tmp_var = self.current_func().new_local(None, ty, "output_register");
// If the out_place is None (i.e `inout(reg) _` syntax was used), we translate
@ -286,7 +286,7 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
continue
};
let ty = out_place.layout.gcc_type(self.cx, false);
let ty = out_place.layout.gcc_type(self.cx);
let tmp_var = self.current_func().new_local(None, ty, "output_register");
tmp_var.set_register_name(reg_name);
@ -306,7 +306,7 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
// `in("explicit register") var`
InlineAsmOperandRef::In { reg, value } => {
if let ConstraintOrRegister::Register(reg_name) = reg_to_gcc(reg) {
let ty = value.layout.gcc_type(self.cx, false);
let ty = value.layout.gcc_type(self.cx);
let reg_var = self.current_func().new_local(None, ty, "input_register");
reg_var.set_register_name(reg_name);
self.llbb().add_assignment(None, reg_var, value.immediate());
@ -325,7 +325,7 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
InlineAsmOperandRef::InOut { reg, late, in_value, out_place } => {
if let ConstraintOrRegister::Register(reg_name) = reg_to_gcc(reg) {
// See explanation in the first pass.
let ty = in_value.layout.gcc_type(self.cx, false);
let ty = in_value.layout.gcc_type(self.cx);
let tmp_var = self.current_func().new_local(None, ty, "output_register");
tmp_var.set_register_name(reg_name);
@ -353,8 +353,7 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
inputs.push(AsmInOperand {
constraint: "X".into(),
rust_idx,
val: self.cx.rvalue_as_function(get_fn(self.cx, instance))
.get_address(None),
val: get_fn(self.cx, instance).get_address(None),
});
}
@ -382,15 +381,19 @@ impl<'a, 'gcc, 'tcx> AsmBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
for piece in template {
match *piece {
InlineAsmTemplatePiece::String(ref string) => {
// TODO(@Commeownist): switch to `Iterator::intersperse` once it's stable
let mut iter = string.split('%');
if let Some(s) = iter.next() {
template_str.push_str(s);
}
for s in iter {
template_str.push_str("%%");
template_str.push_str(s);
for char in string.chars() {
// TODO(antoyo): might also need to escape | if rustc doesn't do it.
let escaped_char =
match char {
'%' => "%%",
'{' => "%{",
'}' => "%}",
_ => {
template_str.push(char);
continue;
},
};
template_str.push_str(escaped_char);
}
}
InlineAsmTemplatePiece::Placeholder { operand_idx, modifier, span: _ } => {
@ -565,39 +568,52 @@ fn reg_to_gcc(reg: InlineAsmRegOrRegClass) -> ConstraintOrRegister {
_ => unimplemented!(),
}
},
// They can be retrieved from https://gcc.gnu.org/onlinedocs/gcc/Machine-Constraints.html
InlineAsmRegOrRegClass::RegClass(reg) => match reg {
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::preg) => unimplemented!(),
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::reg) => unimplemented!(),
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::vreg) => unimplemented!(),
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::vreg_low16) => unimplemented!(),
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::reg) => unimplemented!(),
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::reg) => "r",
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::vreg) => "w",
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::vreg_low16) => "x",
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::preg) => {
unreachable!("clobber-only")
}
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::reg) => "r",
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::sreg)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::dreg_low16)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::qreg_low8) => unimplemented!(),
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::sreg_low16)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::qreg_low8)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::sreg_low16)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::dreg_low8)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::qreg_low4) => unimplemented!(),
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::dreg)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::qreg) => unimplemented!(),
InlineAsmRegClass::Avr(_) => unimplemented!(),
InlineAsmRegClass::Bpf(_) => unimplemented!(),
InlineAsmRegClass::Hexagon(HexagonInlineAsmRegClass::reg) => unimplemented!(),
InlineAsmRegClass::Mips(MipsInlineAsmRegClass::reg) => unimplemented!(),
InlineAsmRegClass::Mips(MipsInlineAsmRegClass::freg) => unimplemented!(),
InlineAsmRegClass::Msp430(_) => unimplemented!(),
InlineAsmRegClass::Nvptx(NvptxInlineAsmRegClass::reg16) => unimplemented!(),
InlineAsmRegClass::Nvptx(NvptxInlineAsmRegClass::reg32) => unimplemented!(),
InlineAsmRegClass::Nvptx(NvptxInlineAsmRegClass::reg64) => unimplemented!(),
InlineAsmRegClass::PowerPC(PowerPCInlineAsmRegClass::reg) => unimplemented!(),
InlineAsmRegClass::PowerPC(PowerPCInlineAsmRegClass::reg_nonzero) => unimplemented!(),
InlineAsmRegClass::PowerPC(PowerPCInlineAsmRegClass::freg) => unimplemented!(),
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::qreg_low4)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::dreg)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::qreg) => "t",
InlineAsmRegClass::Avr(AvrInlineAsmRegClass::reg) => "r",
InlineAsmRegClass::Avr(AvrInlineAsmRegClass::reg_upper) => "d",
InlineAsmRegClass::Avr(AvrInlineAsmRegClass::reg_pair) => "r",
InlineAsmRegClass::Avr(AvrInlineAsmRegClass::reg_iw) => "w",
InlineAsmRegClass::Avr(AvrInlineAsmRegClass::reg_ptr) => "e",
InlineAsmRegClass::Bpf(BpfInlineAsmRegClass::reg) => "r",
InlineAsmRegClass::Bpf(BpfInlineAsmRegClass::wreg) => "w",
InlineAsmRegClass::Hexagon(HexagonInlineAsmRegClass::reg) => "r",
InlineAsmRegClass::Mips(MipsInlineAsmRegClass::reg) => "d", // more specific than "r"
InlineAsmRegClass::Mips(MipsInlineAsmRegClass::freg) => "f",
InlineAsmRegClass::Msp430(Msp430InlineAsmRegClass::reg) => "r",
// https://github.com/gcc-mirror/gcc/blob/master/gcc/config/nvptx/nvptx.md -> look for
// "define_constraint".
InlineAsmRegClass::Nvptx(NvptxInlineAsmRegClass::reg16) => "h",
InlineAsmRegClass::Nvptx(NvptxInlineAsmRegClass::reg32) => "r",
InlineAsmRegClass::Nvptx(NvptxInlineAsmRegClass::reg64) => "l",
InlineAsmRegClass::PowerPC(PowerPCInlineAsmRegClass::reg) => "r",
InlineAsmRegClass::PowerPC(PowerPCInlineAsmRegClass::reg_nonzero) => "b",
InlineAsmRegClass::PowerPC(PowerPCInlineAsmRegClass::freg) => "f",
InlineAsmRegClass::PowerPC(PowerPCInlineAsmRegClass::cr)
| InlineAsmRegClass::PowerPC(PowerPCInlineAsmRegClass::xer) => {
unreachable!("clobber-only")
},
InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::reg) => unimplemented!(),
InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::freg) => unimplemented!(),
InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::vreg) => unimplemented!(),
InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::reg) => "r",
InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::freg) => "f",
InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::vreg) => {
unreachable!("clobber-only")
}
InlineAsmRegClass::X86(X86InlineAsmRegClass::reg) => "r",
InlineAsmRegClass::X86(X86InlineAsmRegClass::reg_abcd) => "Q",
InlineAsmRegClass::X86(X86InlineAsmRegClass::reg_byte) => "q",
@ -605,16 +621,18 @@ fn reg_to_gcc(reg: InlineAsmRegOrRegClass) -> ConstraintOrRegister {
| InlineAsmRegClass::X86(X86InlineAsmRegClass::ymm_reg) => "x",
InlineAsmRegClass::X86(X86InlineAsmRegClass::zmm_reg) => "v",
InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg) => "Yk",
InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg0) => unimplemented!(),
InlineAsmRegClass::Wasm(WasmInlineAsmRegClass::local) => unimplemented!(),
InlineAsmRegClass::X86(
X86InlineAsmRegClass::x87_reg | X86InlineAsmRegClass::mmx_reg | X86InlineAsmRegClass::tmm_reg,
X86InlineAsmRegClass::kreg0
| X86InlineAsmRegClass::x87_reg
| X86InlineAsmRegClass::mmx_reg
| X86InlineAsmRegClass::tmm_reg,
) => unreachable!("clobber-only"),
InlineAsmRegClass::SpirV(SpirVInlineAsmRegClass::reg) => {
bug!("GCC backend does not support SPIR-V")
}
InlineAsmRegClass::S390x(S390xInlineAsmRegClass::reg) => unimplemented!(),
InlineAsmRegClass::S390x(S390xInlineAsmRegClass::freg) => unimplemented!(),
InlineAsmRegClass::Wasm(WasmInlineAsmRegClass::local) => "r",
InlineAsmRegClass::S390x(S390xInlineAsmRegClass::reg) => "r",
InlineAsmRegClass::S390x(S390xInlineAsmRegClass::freg) => "f",
InlineAsmRegClass::Err => unreachable!(),
}
};
@ -692,21 +710,23 @@ impl<'gcc, 'tcx> AsmMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
&& options.contains(InlineAsmOptions::ATT_SYNTAX);
// Build the template string
let mut template_str = String::new();
let mut template_str = ".pushsection .text\n".to_owned();
if att_dialect {
template_str.push_str(".att_syntax\n");
}
for piece in template {
match *piece {
InlineAsmTemplatePiece::String(ref string) => {
for line in string.lines() {
let mut index = 0;
while index < string.len() {
// NOTE: gcc does not allow inline comment, so remove them.
let line =
if let Some(index) = line.rfind("//") {
&line[..index]
}
else {
line
};
template_str.push_str(line);
template_str.push('\n');
let comment_index = string[index..].find("//")
.map(|comment_index| comment_index + index)
.unwrap_or(string.len());
template_str.push_str(&string[index..comment_index]);
index = string[comment_index..].find('\n')
.map(|index| index + comment_index)
.unwrap_or(string.len());
}
},
InlineAsmTemplatePiece::Placeholder { operand_idx, modifier: _, span: _ } => {
@ -719,6 +739,8 @@ impl<'gcc, 'tcx> AsmMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
}
GlobalAsmOperandRef::SymFn { instance } => {
let function = get_fn(self, instance);
self.add_used_function(function);
// TODO(@Amanieu): Additional mangling is needed on
// some targets to add a leading underscore (Mach-O)
// or byte count suffixes (x86 Windows).
@ -727,6 +749,7 @@ impl<'gcc, 'tcx> AsmMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
}
GlobalAsmOperandRef::SymStatic { def_id } => {
// TODO(antoyo): set the global variable as used.
// TODO(@Amanieu): Additional mangling is needed on
// some targets to add a leading underscore (Mach-O).
let instance = Instance::mono(self.tcx, def_id);
@ -738,48 +761,51 @@ impl<'gcc, 'tcx> AsmMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
}
}
let template_str =
if att_dialect {
format!(".att_syntax\n\t{}\n\t.intel_syntax noprefix", template_str)
}
else {
template_str
};
if att_dialect {
template_str.push_str("\n\t.intel_syntax noprefix");
}
// NOTE: seems like gcc will put the asm in the wrong section, so set it to .text manually.
let template_str = format!(".pushsection .text\n{}\n.popsection", template_str);
template_str.push_str("\n.popsection");
self.context.add_top_level_asm(None, &template_str);
}
}
fn modifier_to_gcc(arch: InlineAsmArch, reg: InlineAsmRegClass, modifier: Option<char>) -> Option<char> {
// The modifiers can be retrieved from
// https://gcc.gnu.org/onlinedocs/gcc/Modifiers.html#Modifiers
match reg {
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::reg) => modifier,
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::preg) => modifier,
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::vreg)
| InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::vreg_low16) => {
unimplemented!()
if modifier == Some('v') { None } else { modifier }
}
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::reg) => unimplemented!(),
InlineAsmRegClass::AArch64(AArch64InlineAsmRegClass::preg) => {
unreachable!("clobber-only")
}
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::reg) => None,
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::sreg)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::sreg_low16) => unimplemented!(),
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::sreg_low16) => None,
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::dreg)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::dreg_low16)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::dreg_low8) => unimplemented!(),
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::dreg_low8) => Some('P'),
InlineAsmRegClass::Arm(ArmInlineAsmRegClass::qreg)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::qreg_low8)
| InlineAsmRegClass::Arm(ArmInlineAsmRegClass::qreg_low4) => {
unimplemented!()
if modifier.is_none() {
Some('q')
} else {
modifier
}
}
InlineAsmRegClass::Avr(_) => unimplemented!(),
InlineAsmRegClass::Bpf(_) => unimplemented!(),
InlineAsmRegClass::Hexagon(_) => unimplemented!(),
InlineAsmRegClass::Mips(_) => unimplemented!(),
InlineAsmRegClass::Msp430(_) => unimplemented!(),
InlineAsmRegClass::Nvptx(_) => unimplemented!(),
InlineAsmRegClass::PowerPC(_) => unimplemented!(),
InlineAsmRegClass::Hexagon(_) => None,
InlineAsmRegClass::Mips(_) => None,
InlineAsmRegClass::Nvptx(_) => None,
InlineAsmRegClass::PowerPC(_) => None,
InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::reg)
| InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::freg) => unimplemented!(),
InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::vreg) => unimplemented!(),
| InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::freg) => None,
InlineAsmRegClass::RiscV(RiscVInlineAsmRegClass::vreg) => {
unreachable!("clobber-only")
}
InlineAsmRegClass::X86(X86InlineAsmRegClass::reg)
| InlineAsmRegClass::X86(X86InlineAsmRegClass::reg_abcd) => match modifier {
None => if arch == InlineAsmArch::X86_64 { Some('q') } else { Some('k') },
@ -803,16 +829,29 @@ fn modifier_to_gcc(arch: InlineAsmArch, reg: InlineAsmRegClass, modifier: Option
_ => unreachable!(),
},
InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg) => None,
InlineAsmRegClass::X86(X86InlineAsmRegClass::kreg0) => None,
InlineAsmRegClass::X86(X86InlineAsmRegClass::x87_reg | X86InlineAsmRegClass::mmx_reg | X86InlineAsmRegClass::tmm_reg) => {
InlineAsmRegClass::X86(
X86InlineAsmRegClass::x87_reg
| X86InlineAsmRegClass::mmx_reg
| X86InlineAsmRegClass::kreg0
| X86InlineAsmRegClass::tmm_reg,
) => {
unreachable!("clobber-only")
}
InlineAsmRegClass::Wasm(WasmInlineAsmRegClass::local) => unimplemented!(),
InlineAsmRegClass::Wasm(WasmInlineAsmRegClass::local) => None,
InlineAsmRegClass::Bpf(_) => None,
InlineAsmRegClass::Avr(AvrInlineAsmRegClass::reg_pair)
| InlineAsmRegClass::Avr(AvrInlineAsmRegClass::reg_iw)
| InlineAsmRegClass::Avr(AvrInlineAsmRegClass::reg_ptr) => match modifier {
Some('h') => Some('B'),
Some('l') => Some('A'),
_ => None,
},
InlineAsmRegClass::Avr(_) => None,
InlineAsmRegClass::S390x(_) => None,
InlineAsmRegClass::Msp430(_) => None,
InlineAsmRegClass::SpirV(SpirVInlineAsmRegClass::reg) => {
bug!("LLVM backend does not support SPIR-V")
},
InlineAsmRegClass::S390x(S390xInlineAsmRegClass::reg) => unimplemented!(),
InlineAsmRegClass::S390x(S390xInlineAsmRegClass::freg) => unimplemented!(),
}
InlineAsmRegClass::Err => unreachable!(),
}
}

View File

@ -0,0 +1,113 @@
#[cfg(feature="master")]
use gccjit::FnAttribute;
use gccjit::Function;
use rustc_attr::InstructionSetAttr;
use rustc_codegen_ssa::target_features::tied_target_features;
use rustc_data_structures::fx::FxHashMap;
use rustc_middle::ty;
use rustc_session::Session;
use rustc_span::symbol::sym;
use smallvec::{smallvec, SmallVec};
use crate::{context::CodegenCx, errors::TiedTargetFeatures};
// Given a map from target_features to whether they are enabled or disabled,
// ensure only valid combinations are allowed.
pub fn check_tied_features(sess: &Session, features: &FxHashMap<&str, bool>) -> Option<&'static [&'static str]> {
for tied in tied_target_features(sess) {
// Tied features must be set to the same value, or not set at all
let mut tied_iter = tied.iter();
let enabled = features.get(tied_iter.next().unwrap());
if tied_iter.any(|feature| enabled != features.get(feature)) {
return Some(tied);
}
}
None
}
// TODO(antoyo): maybe move to a new module gcc_util.
// To find a list of GCC's names, check https://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html
fn to_gcc_features<'a>(sess: &Session, s: &'a str) -> SmallVec<[&'a str; 2]> {
let arch = if sess.target.arch == "x86_64" { "x86" } else { &*sess.target.arch };
match (arch, s) {
("x86", "sse4.2") => smallvec!["sse4.2", "crc32"],
("x86", "pclmulqdq") => smallvec!["pclmul"],
("x86", "rdrand") => smallvec!["rdrnd"],
("x86", "bmi1") => smallvec!["bmi"],
("x86", "cmpxchg16b") => smallvec!["cx16"],
("x86", "avx512vaes") => smallvec!["vaes"],
("x86", "avx512gfni") => smallvec!["gfni"],
("x86", "avx512vpclmulqdq") => smallvec!["vpclmulqdq"],
// NOTE: seems like GCC requires 'avx512bw' for 'avx512vbmi2'.
("x86", "avx512vbmi2") => smallvec!["avx512vbmi2", "avx512bw"],
// NOTE: seems like GCC requires 'avx512bw' for 'avx512bitalg'.
("x86", "avx512bitalg") => smallvec!["avx512bitalg", "avx512bw"],
("aarch64", "rcpc2") => smallvec!["rcpc-immo"],
("aarch64", "dpb") => smallvec!["ccpp"],
("aarch64", "dpb2") => smallvec!["ccdp"],
("aarch64", "frintts") => smallvec!["fptoint"],
("aarch64", "fcma") => smallvec!["complxnum"],
("aarch64", "pmuv3") => smallvec!["perfmon"],
("aarch64", "paca") => smallvec!["pauth"],
("aarch64", "pacg") => smallvec!["pauth"],
// Rust ties fp and neon together. In LLVM neon implicitly enables fp,
// but we manually enable neon when a feature only implicitly enables fp
("aarch64", "f32mm") => smallvec!["f32mm", "neon"],
("aarch64", "f64mm") => smallvec!["f64mm", "neon"],
("aarch64", "fhm") => smallvec!["fp16fml", "neon"],
("aarch64", "fp16") => smallvec!["fullfp16", "neon"],
("aarch64", "jsconv") => smallvec!["jsconv", "neon"],
("aarch64", "sve") => smallvec!["sve", "neon"],
("aarch64", "sve2") => smallvec!["sve2", "neon"],
("aarch64", "sve2-aes") => smallvec!["sve2-aes", "neon"],
("aarch64", "sve2-sm4") => smallvec!["sve2-sm4", "neon"],
("aarch64", "sve2-sha3") => smallvec!["sve2-sha3", "neon"],
("aarch64", "sve2-bitperm") => smallvec!["sve2-bitperm", "neon"],
(_, s) => smallvec![s],
}
}
/// Composite function which sets GCC attributes for function depending on its AST (`#[attribute]`)
/// attributes.
pub fn from_fn_attrs<'gcc, 'tcx>(
cx: &CodegenCx<'gcc, 'tcx>,
#[cfg_attr(not(feature="master"), allow(unused_variables))]
func: Function<'gcc>,
instance: ty::Instance<'tcx>,
) {
let codegen_fn_attrs = cx.tcx.codegen_fn_attrs(instance.def_id());
let function_features =
codegen_fn_attrs.target_features.iter().map(|features| features.as_str()).collect::<Vec<&str>>();
if let Some(features) = check_tied_features(cx.tcx.sess, &function_features.iter().map(|features| (*features, true)).collect()) {
let span = cx.tcx
.get_attr(instance.def_id(), sym::target_feature)
.map_or_else(|| cx.tcx.def_span(instance.def_id()), |a| a.span);
cx.tcx.sess.create_err(TiedTargetFeatures {
features: features.join(", "),
span,
})
.emit();
return;
}
let mut function_features = function_features
.iter()
.flat_map(|feat| to_gcc_features(cx.tcx.sess, feat).into_iter())
.chain(codegen_fn_attrs.instruction_set.iter().map(|x| match x {
InstructionSetAttr::ArmA32 => "-thumb-mode", // TODO(antoyo): support removing feature.
InstructionSetAttr::ArmT32 => "thumb-mode",
}))
.collect::<Vec<_>>();
// TODO(antoyo): check if we really need global backend features. (Maybe they could be applied
// globally?)
let mut global_features = cx.tcx.global_backend_features(()).iter().map(|s| s.as_str());
function_features.extend(&mut global_features);
let target_features = function_features.join(",");
if !target_features.is_empty() {
#[cfg(feature="master")]
func.add_attribute(FnAttribute::Target(&target_features));
}
}

View File

@ -57,6 +57,7 @@ pub(crate) unsafe fn codegen(cgcx: &CodegenContext<GccCodegenBackend>, _diag_han
if env::var("CG_GCCJIT_DUMP_TO_FILE").as_deref() == Ok("1") {
let _ = fs::create_dir("/tmp/gccjit_dumps");
let path = &format!("/tmp/gccjit_dumps/{}.c", module.name);
context.set_debug_info(true);
context.dump_to_file(path, true);
}
context.compile_to_file(OutputKind::ObjectFile, obj_out.to_str().expect("path to str"));

View File

@ -8,6 +8,8 @@ use gccjit::{
};
use rustc_middle::dep_graph;
use rustc_middle::ty::TyCtxt;
#[cfg(feature="master")]
use rustc_middle::mir::mono::Visibility;
use rustc_middle::mir::mono::Linkage;
use rustc_codegen_ssa::{ModuleCodegen, ModuleKind};
use rustc_codegen_ssa::base::maybe_create_entry_wrapper;
@ -20,6 +22,15 @@ use crate::GccContext;
use crate::builder::Builder;
use crate::context::CodegenCx;
#[cfg(feature="master")]
pub fn visibility_to_gcc(linkage: Visibility) -> gccjit::Visibility {
match linkage {
Visibility::Default => gccjit::Visibility::Default,
Visibility::Hidden => gccjit::Visibility::Hidden,
Visibility::Protected => gccjit::Visibility::Protected,
}
}
pub fn global_linkage_to_gcc(linkage: Linkage) -> GlobalKind {
match linkage {
Linkage::External => GlobalKind::Imported,
@ -76,16 +87,34 @@ pub fn compile_codegen_unit(tcx: TyCtxt<'_>, cgu_name: Symbol, supports_128bit_i
// Instantiate monomorphizations without filling out definitions yet...
//let llvm_module = ModuleLlvm::new(tcx, &cgu_name.as_str());
let context = Context::default();
context.add_command_line_option("-fexceptions");
context.add_driver_option("-fexceptions");
// TODO(antoyo): only set on x86 platforms.
context.add_command_line_option("-masm=intel");
// TODO(antoyo): only add the following cli argument if the feature is supported.
context.add_command_line_option("-msse2");
context.add_command_line_option("-mavx2");
context.add_command_line_option("-msha");
context.add_command_line_option("-mpclmul");
// FIXME(antoyo): the following causes an illegal instruction on vmovdqu64 in std_example on my CPU.
// Only add if the CPU supports it.
//context.add_command_line_option("-mavx512f");
context.add_command_line_option("-msha");
context.add_command_line_option("-mpclmul");
context.add_command_line_option("-mfma");
context.add_command_line_option("-mfma4");
context.add_command_line_option("-m64");
context.add_command_line_option("-mbmi");
context.add_command_line_option("-mgfni");
//context.add_command_line_option("-mavxvnni"); // The CI doesn't support this option.
context.add_command_line_option("-mf16c");
context.add_command_line_option("-maes");
context.add_command_line_option("-mxsavec");
context.add_command_line_option("-mbmi2");
context.add_command_line_option("-mrtm");
context.add_command_line_option("-mvaes");
context.add_command_line_option("-mvpclmulqdq");
context.add_command_line_option("-mavx");
for arg in &tcx.sess.opts.cg.llvm_args {
context.add_command_line_option(arg);
}
@ -95,12 +124,20 @@ pub fn compile_codegen_unit(tcx: TyCtxt<'_>, cgu_name: Symbol, supports_128bit_i
context.add_command_line_option("-fno-semantic-interposition");
// NOTE: Rust relies on LLVM not doing TBAA (https://github.com/rust-lang/unsafe-code-guidelines/issues/292).
context.add_command_line_option("-fno-strict-aliasing");
// NOTE: Rust relies on LLVM doing wrapping on overflow.
context.add_command_line_option("-fwrapv");
if tcx.sess.opts.unstable_opts.function_sections.unwrap_or(tcx.sess.target.function_sections) {
context.add_command_line_option("-ffunction-sections");
context.add_command_line_option("-fdata-sections");
}
if env::var("CG_GCCJIT_DUMP_RTL").as_deref() == Ok("1") {
context.add_command_line_option("-fdump-rtl-vregs");
}
if env::var("CG_GCCJIT_DUMP_TREE_ALL").as_deref() == Ok("1") {
context.add_command_line_option("-fdump-tree-all");
}
if env::var("CG_GCCJIT_DUMP_CODE").as_deref() == Ok("1") {
context.set_dump_code_on_compile(true);
}
@ -115,7 +152,7 @@ pub fn compile_codegen_unit(tcx: TyCtxt<'_>, cgu_name: Symbol, supports_128bit_i
context.set_keep_intermediates(true);
}
// TODO(bjorn3): Remove once unwinding is properly implemented
// NOTE: The codegen generates unrechable blocks.
context.set_allow_unreachable_blocks(true);
{

View File

@ -217,7 +217,7 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
let actual_ty = actual_val.get_type();
if expected_ty != actual_ty {
if !actual_ty.is_vector() && !expected_ty.is_vector() && actual_ty.is_integral() && expected_ty.is_integral() && actual_ty.get_size() != expected_ty.get_size() {
if !actual_ty.is_vector() && !expected_ty.is_vector() && (actual_ty.is_integral() && expected_ty.is_integral()) || (actual_ty.get_pointee().is_some() && expected_ty.get_pointee().is_some()) {
self.context.new_cast(None, actual_val, expected_ty)
}
else if on_stack_param_indices.contains(&index) {
@ -226,6 +226,7 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
else {
assert!(!((actual_ty.is_vector() && !expected_ty.is_vector()) || (!actual_ty.is_vector() && expected_ty.is_vector())), "{:?} ({}) -> {:?} ({}), index: {:?}[{}]", actual_ty, actual_ty.is_vector(), expected_ty, expected_ty.is_vector(), func_ptr, index);
// TODO(antoyo): perhaps use __builtin_convertvector for vector casting.
// TODO: remove bitcast now that vector types can be compared?
self.bitcast(actual_val, expected_ty)
}
}
@ -279,21 +280,30 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
}
fn function_ptr_call(&mut self, func_ptr: RValue<'gcc>, args: &[RValue<'gcc>], _funclet: Option<&Funclet>) -> RValue<'gcc> {
let args = self.check_ptr_call("call", func_ptr, args);
let gcc_func = func_ptr.get_type().dyncast_function_ptr_type().expect("function ptr");
let func_name = format!("{:?}", func_ptr);
let previous_arg_count = args.len();
let orig_args = args;
let args = {
let function_address_names = self.function_address_names.borrow();
let original_function_name = function_address_names.get(&func_ptr);
llvm::adjust_intrinsic_arguments(&self, gcc_func, args.into(), &func_name, original_function_name)
};
let args_adjusted = args.len() != previous_arg_count;
let args = self.check_ptr_call("call", func_ptr, &*args);
// gccjit requires to use the result of functions, even when it's not used.
// That's why we assign the result to a local or call add_eval().
let gcc_func = func_ptr.get_type().dyncast_function_ptr_type().expect("function ptr");
let return_type = gcc_func.get_return_type();
let void_type = self.context.new_type::<()>();
let current_func = self.block.get_function();
if return_type != void_type {
unsafe { RETURN_VALUE_COUNT += 1 };
let result = current_func.new_local(None, return_type, &format!("ptrReturnValue{}", unsafe { RETURN_VALUE_COUNT }));
let func_name = format!("{:?}", func_ptr);
let args = llvm::adjust_intrinsic_arguments(&self, gcc_func, args, &func_name);
self.block.add_assignment(None, result, self.cx.context.new_call_through_ptr(None, func_ptr, &args));
let return_value = self.cx.context.new_call_through_ptr(None, func_ptr, &args);
let return_value = llvm::adjust_intrinsic_return_value(&self, return_value, &func_name, &args, args_adjusted, orig_args);
let result = current_func.new_local(None, return_value.get_type(), &format!("ptrReturnValue{}", unsafe { RETURN_VALUE_COUNT }));
self.block.add_assignment(None, result, return_value);
result.to_rvalue()
}
else {
@ -366,10 +376,10 @@ impl<'tcx> FnAbiOfHelpers<'tcx> for Builder<'_, '_, 'tcx> {
}
}
impl<'gcc, 'tcx> Deref for Builder<'_, 'gcc, 'tcx> {
impl<'a, 'gcc, 'tcx> Deref for Builder<'a, 'gcc, 'tcx> {
type Target = CodegenCx<'gcc, 'tcx>;
fn deref(&self) -> &Self::Target {
fn deref<'b>(&'b self) -> &'a Self::Target {
self.cx
}
}
@ -387,7 +397,7 @@ impl<'gcc, 'tcx> BackendTypes for Builder<'_, 'gcc, 'tcx> {
}
impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
fn build(cx: &'a CodegenCx<'gcc, 'tcx>, block: Block<'gcc>) -> Self {
fn build(cx: &'a CodegenCx<'gcc, 'tcx>, block: Block<'gcc>) -> Builder<'a, 'gcc, 'tcx> {
Builder::with_cx(cx, block)
}
@ -444,17 +454,36 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
self.block.end_with_switch(None, value, default_block, &gcc_cases);
}
fn invoke(
&mut self,
typ: Type<'gcc>,
fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>,
func: RValue<'gcc>,
args: &[RValue<'gcc>],
then: Block<'gcc>,
catch: Block<'gcc>,
_funclet: Option<&Funclet>,
) -> RValue<'gcc> {
// TODO(bjorn3): Properly implement unwinding.
#[cfg(feature="master")]
fn invoke(&mut self, typ: Type<'gcc>, _fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>, func: RValue<'gcc>, args: &[RValue<'gcc>], then: Block<'gcc>, catch: Block<'gcc>, _funclet: Option<&Funclet>) -> RValue<'gcc> {
let try_block = self.current_func().new_block("try");
let current_block = self.block.clone();
self.block = try_block;
let call = self.call(typ, None, func, args, None); // TODO(antoyo): use funclet here?
self.block = current_block;
let return_value = self.current_func()
.new_local(None, call.get_type(), "invokeResult");
try_block.add_assignment(None, return_value, call);
try_block.end_with_jump(None, then);
if self.cleanup_blocks.borrow().contains(&catch) {
self.block.add_try_finally(None, try_block, catch);
}
else {
self.block.add_try_catch(None, try_block, catch);
}
self.block.end_with_jump(None, then);
return_value.to_rvalue()
}
#[cfg(not(feature="master"))]
fn invoke(&mut self, typ: Type<'gcc>, fn_abi: Option<&FnAbi<'tcx, Ty<'tcx>>>, func: RValue<'gcc>, args: &[RValue<'gcc>], then: Block<'gcc>, catch: Block<'gcc>, _funclet: Option<&Funclet>) -> RValue<'gcc> {
let call_site = self.call(typ, None, func, args, None);
let condition = self.context.new_rvalue_from_int(self.bool_type, 1);
self.llbb().end_with_conditional(None, condition, then, catch);
@ -542,6 +571,31 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
}
fn frem(&mut self, a: RValue<'gcc>, b: RValue<'gcc>) -> RValue<'gcc> {
// TODO(antoyo): add check in libgccjit since using the binary operator % causes the following error:
// during RTL pass: expand
// libgccjit.so: error: in expmed_mode_index, at expmed.h:240
// 0x7f0101d58dc6 expmed_mode_index
// ../../../gcc/gcc/expmed.h:240
// 0x7f0101d58e35 expmed_op_cost_ptr
// ../../../gcc/gcc/expmed.h:262
// 0x7f0101d594a1 sdiv_cost_ptr
// ../../../gcc/gcc/expmed.h:531
// 0x7f0101d594f3 sdiv_cost
// ../../../gcc/gcc/expmed.h:549
// 0x7f0101d6af7e expand_divmod(int, tree_code, machine_mode, rtx_def*, rtx_def*, rtx_def*, int, optab_methods)
// ../../../gcc/gcc/expmed.cc:4356
// 0x7f0101d94f9e expand_expr_divmod
// ../../../gcc/gcc/expr.cc:8929
// 0x7f0101d97a26 expand_expr_real_2(separate_ops*, rtx_def*, machine_mode, expand_modifier)
// ../../../gcc/gcc/expr.cc:9566
// 0x7f0101bef6ef expand_gimple_stmt_1
// ../../../gcc/gcc/cfgexpand.cc:3967
// 0x7f0101bef910 expand_gimple_stmt
// ../../../gcc/gcc/cfgexpand.cc:4028
// 0x7f0101bf6ee7 expand_gimple_basic_block
// ../../../gcc/gcc/cfgexpand.cc:6069
// 0x7f0101bf9194 execute
// ../../../gcc/gcc/cfgexpand.cc:6795
if a.get_type().is_compatible_with(self.cx.float_type) {
let fmodf = self.context.get_builtin_function("fmodf");
// FIXME(antoyo): this seems to produce the wrong result.
@ -616,24 +670,29 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
a * b
}
fn fadd_fast(&mut self, _lhs: RValue<'gcc>, _rhs: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!();
fn fadd_fast(&mut self, lhs: RValue<'gcc>, rhs: RValue<'gcc>) -> RValue<'gcc> {
// NOTE: it seems like we cannot enable fast-mode for a single operation in GCC.
lhs + rhs
}
fn fsub_fast(&mut self, _lhs: RValue<'gcc>, _rhs: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!();
fn fsub_fast(&mut self, lhs: RValue<'gcc>, rhs: RValue<'gcc>) -> RValue<'gcc> {
// NOTE: it seems like we cannot enable fast-mode for a single operation in GCC.
lhs - rhs
}
fn fmul_fast(&mut self, _lhs: RValue<'gcc>, _rhs: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!();
fn fmul_fast(&mut self, lhs: RValue<'gcc>, rhs: RValue<'gcc>) -> RValue<'gcc> {
// NOTE: it seems like we cannot enable fast-mode for a single operation in GCC.
lhs * rhs
}
fn fdiv_fast(&mut self, _lhs: RValue<'gcc>, _rhs: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!();
fn fdiv_fast(&mut self, lhs: RValue<'gcc>, rhs: RValue<'gcc>) -> RValue<'gcc> {
// NOTE: it seems like we cannot enable fast-mode for a single operation in GCC.
lhs / rhs
}
fn frem_fast(&mut self, _lhs: RValue<'gcc>, _rhs: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!();
fn frem_fast(&mut self, lhs: RValue<'gcc>, rhs: RValue<'gcc>) -> RValue<'gcc> {
// NOTE: it seems like we cannot enable fast-mode for a single operation in GCC.
self.frem(lhs, rhs)
}
fn checked_binop(&mut self, oop: OverflowOp, typ: Ty<'_>, lhs: Self::Value, rhs: Self::Value) -> (Self::Value, Self::Value) {
@ -722,7 +781,7 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
}
else if place.layout.is_gcc_immediate() {
let load = self.load(
place.layout.gcc_type(self, false),
place.layout.gcc_type(self),
place.llval,
place.align,
);
@ -733,7 +792,7 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
}
else if let abi::Abi::ScalarPair(ref a, ref b) = place.layout.abi {
let b_offset = a.size(self).align_to(b.align(self).abi);
let pair_type = place.layout.gcc_type(self, false);
let pair_type = place.layout.gcc_type(self);
let mut load = |i, scalar: &abi::Scalar, align| {
let llptr = self.struct_gep(pair_type, place.llval, i as u64);
@ -833,26 +892,31 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
}
fn gep(&mut self, _typ: Type<'gcc>, ptr: RValue<'gcc>, indices: &[RValue<'gcc>]) -> RValue<'gcc> {
let mut result = ptr;
let ptr_type = ptr.get_type();
let mut pointee_type = ptr.get_type();
// NOTE: we cannot use array indexing here like in inbounds_gep because array indexing is
// always considered in bounds in GCC (TODO(antoyo): to be verified).
// So, we have to cast to a number.
let mut result = self.context.new_bitcast(None, ptr, self.sizet_type);
// FIXME(antoyo): if there were more than 1 index, this code is probably wrong and would
// require dereferencing the pointer.
for index in indices {
result = self.context.new_array_access(None, result, *index).get_address(None).to_rvalue();
pointee_type = pointee_type.get_pointee().expect("pointee type");
let pointee_size = self.context.new_rvalue_from_int(index.get_type(), pointee_type.get_size() as i32);
result = result + self.gcc_int_cast(*index * pointee_size, self.sizet_type);
}
result
self.context.new_bitcast(None, result, ptr_type)
}
fn inbounds_gep(&mut self, _typ: Type<'gcc>, ptr: RValue<'gcc>, indices: &[RValue<'gcc>]) -> RValue<'gcc> {
// FIXME(antoyo): would be safer if doing the same thing (loop) as gep.
// TODO(antoyo): specify inbounds somehow.
match indices.len() {
1 => {
self.context.new_array_access(None, ptr, indices[0]).get_address(None)
},
2 => {
let array = ptr.dereference(None); // TODO(antoyo): assert that first index is 0?
self.context.new_array_access(None, array, indices[1]).get_address(None)
},
_ => unimplemented!(),
// NOTE: array indexing is always considered in bounds in GCC (TODO(antoyo): to be verified).
let mut indices = indices.into_iter();
let index = indices.next().expect("first index in inbounds_gep");
let mut result = self.context.new_array_access(None, ptr, *index);
for index in indices {
result = self.context.new_array_access(None, result, *index);
}
result.get_address(None)
}
fn struct_gep(&mut self, value_type: Type<'gcc>, ptr: RValue<'gcc>, idx: u64) -> RValue<'gcc> {
@ -1034,8 +1098,19 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
unimplemented!();
}
fn extract_element(&mut self, _vec: RValue<'gcc>, _idx: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!();
#[cfg(feature="master")]
fn extract_element(&mut self, vec: RValue<'gcc>, idx: RValue<'gcc>) -> RValue<'gcc> {
self.context.new_vector_access(None, vec, idx).to_rvalue()
}
#[cfg(not(feature="master"))]
fn extract_element(&mut self, vec: RValue<'gcc>, idx: RValue<'gcc>) -> RValue<'gcc> {
let vector_type = vec.get_type().unqualified().dyncast_vector().expect("Called extract_element on a non-vector type");
let element_type = vector_type.get_element_type();
let vec_num_units = vector_type.get_num_units();
let array_type = self.context.new_array_type(None, element_type, vec_num_units as u64);
let array = self.context.new_bitcast(None, vec, array_type).to_rvalue();
self.context.new_array_access(None, array, idx).to_rvalue()
}
fn vector_splat(&mut self, _num_elts: usize, _elt: RValue<'gcc>) -> RValue<'gcc> {
@ -1116,22 +1191,52 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
}
fn set_personality_fn(&mut self, _personality: RValue<'gcc>) {
// TODO(antoyo)
#[cfg(feature="master")]
{
let personality = self.rvalue_as_function(_personality);
self.current_func().set_personality_function(personality);
}
}
#[cfg(feature="master")]
fn cleanup_landing_pad(&mut self, pers_fn: RValue<'gcc>) -> (RValue<'gcc>, RValue<'gcc>) {
self.set_personality_fn(pers_fn);
// NOTE: insert the current block in a variable so that a later call to invoke knows to
// generate a try/finally instead of a try/catch for this block.
self.cleanup_blocks.borrow_mut().insert(self.block);
let eh_pointer_builtin = self.cx.context.get_target_builtin_function("__builtin_eh_pointer");
let zero = self.cx.context.new_rvalue_zero(self.int_type);
let ptr = self.cx.context.new_call(None, eh_pointer_builtin, &[zero]);
let value1_type = self.u8_type.make_pointer();
let ptr = self.cx.context.new_cast(None, ptr, value1_type);
let value1 = ptr;
let value2 = zero; // TODO(antoyo): set the proper value here (the type of exception?).
(value1, value2)
}
#[cfg(not(feature="master"))]
fn cleanup_landing_pad(&mut self, _pers_fn: RValue<'gcc>) -> (RValue<'gcc>, RValue<'gcc>) {
(
self.current_func().new_local(None, self.u8_type.make_pointer(), "landing_pad0")
.to_rvalue(),
self.current_func().new_local(None, self.i32_type, "landing_pad1").to_rvalue(),
)
// TODO(antoyo): Properly implement unwinding.
// the above is just to make the compilation work as it seems
// rustc_codegen_ssa now calls the unwinding builder methods even on panic=abort.
let value1 = self.current_func().new_local(None, self.u8_type.make_pointer(), "landing_pad0")
.to_rvalue();
let value2 = self.current_func().new_local(None, self.i32_type, "landing_pad1").to_rvalue();
(value1, value2)
}
#[cfg(feature="master")]
fn resume(&mut self, exn0: RValue<'gcc>, _exn1: RValue<'gcc>) {
let exn_type = exn0.get_type();
let exn = self.context.new_cast(None, exn0, exn_type);
let unwind_resume = self.context.get_target_builtin_function("__builtin_unwind_resume");
self.llbb().add_eval(None, self.context.new_call(None, unwind_resume, &[exn]));
self.unreachable();
}
#[cfg(not(feature="master"))]
fn resume(&mut self, _exn0: RValue<'gcc>, _exn1: RValue<'gcc>) {
// TODO(bjorn3): Properly implement unwinding.
self.unreachable();
}
@ -1160,6 +1265,15 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
fn atomic_cmpxchg(&mut self, dst: RValue<'gcc>, cmp: RValue<'gcc>, src: RValue<'gcc>, order: AtomicOrdering, failure_order: AtomicOrdering, weak: bool) -> RValue<'gcc> {
let expected = self.current_func().new_local(None, cmp.get_type(), "expected");
self.llbb().add_assignment(None, expected, cmp);
// NOTE: gcc doesn't support a failure memory model that is stronger than the success
// memory model.
let order =
if failure_order as i32 > order as i32 {
failure_order
}
else {
order
};
let success = self.compare_exchange(dst, expected, src, order, failure_order, weak);
let pair_type = self.cx.type_struct(&[src.get_type(), self.bool_type], false);
@ -1469,7 +1583,7 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
#[cfg(feature="master")]
pub fn shuffle_vector(&mut self, v1: RValue<'gcc>, v2: RValue<'gcc>, mask: RValue<'gcc>) -> RValue<'gcc> {
let struct_type = mask.get_type().is_struct().expect("mask of struct type");
let struct_type = mask.get_type().is_struct().expect("mask should be of struct type");
// TODO(antoyo): use a recursive unqualified() here.
let vector_type = v1.get_type().unqualified().dyncast_vector().expect("vector type");
@ -1501,22 +1615,17 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
vector_elements.push(self.context.new_rvalue_zero(mask_element_type));
}
let array_type = self.context.new_array_type(None, element_type, vec_num_units as i32);
let result_type = self.context.new_vector_type(element_type, mask_num_units as u64);
let (v1, v2) =
if vec_num_units < mask_num_units {
// NOTE: the mask needs to be the same length as the input vectors, so join the 2
// vectors and create a dummy second vector.
// TODO(antoyo): switch to using new_vector_access.
let array = self.context.new_bitcast(None, v1, array_type);
let mut elements = vec![];
for i in 0..vec_num_units {
elements.push(self.context.new_array_access(None, array, self.context.new_rvalue_from_int(self.int_type, i as i32)).to_rvalue());
elements.push(self.context.new_vector_access(None, v1, self.context.new_rvalue_from_int(self.int_type, i as i32)).to_rvalue());
}
// TODO(antoyo): switch to using new_vector_access.
let array = self.context.new_bitcast(None, v2, array_type);
for i in 0..(mask_num_units - vec_num_units) {
elements.push(self.context.new_array_access(None, array, self.context.new_rvalue_from_int(self.int_type, i as i32)).to_rvalue());
elements.push(self.context.new_vector_access(None, v2, self.context.new_rvalue_from_int(self.int_type, i as i32)).to_rvalue());
}
let v1 = self.context.new_rvalue_from_vector(None, result_type, &elements);
let zero = self.context.new_rvalue_zero(element_type);
@ -1536,10 +1645,8 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
// NOTE: if padding was added, only select the number of elements of the masks to
// remove that padding in the result.
let mut elements = vec![];
// TODO(antoyo): switch to using new_vector_access.
let array = self.context.new_bitcast(None, result, array_type);
for i in 0..mask_num_units {
elements.push(self.context.new_array_access(None, array, self.context.new_rvalue_from_int(self.int_type, i as i32)).to_rvalue());
elements.push(self.context.new_vector_access(None, result, self.context.new_rvalue_from_int(self.int_type, i as i32)).to_rvalue());
}
self.context.new_rvalue_from_vector(None, result_type, &elements)
}
@ -1558,18 +1665,20 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
where F: Fn(RValue<'gcc>, RValue<'gcc>, &'gcc Context<'gcc>) -> RValue<'gcc>
{
let vector_type = src.get_type().unqualified().dyncast_vector().expect("vector type");
let element_type = vector_type.get_element_type();
let mask_element_type = self.type_ix(element_type.get_size() as u64 * 8);
let element_count = vector_type.get_num_units();
let mut vector_elements = vec![];
for i in 0..element_count {
vector_elements.push(i);
}
let mask_type = self.context.new_vector_type(self.int_type, element_count as u64);
let mask_type = self.context.new_vector_type(mask_element_type, element_count as u64);
let mut shift = 1;
let mut res = src;
while shift < element_count {
let vector_elements: Vec<_> =
vector_elements.iter()
.map(|i| self.context.new_rvalue_from_int(self.int_type, ((i + shift) % element_count) as i32))
.map(|i| self.context.new_rvalue_from_int(mask_element_type, ((i + shift) % element_count) as i32))
.collect();
let mask = self.context.new_rvalue_from_vector(None, mask_type, &vector_elements);
let shifted = self.context.new_rvalue_vector_perm(None, res, res, mask);
@ -1581,7 +1690,7 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
}
#[cfg(not(feature="master"))]
pub fn vector_reduce<F>(&mut self, src: RValue<'gcc>, op: F) -> RValue<'gcc>
pub fn vector_reduce<F>(&mut self, _src: RValue<'gcc>, _op: F) -> RValue<'gcc>
where F: Fn(RValue<'gcc>, RValue<'gcc>, &'gcc Context<'gcc>) -> RValue<'gcc>
{
unimplemented!();
@ -1595,15 +1704,47 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
unimplemented!();
}
#[cfg(feature="master")]
pub fn vector_reduce_fadd(&mut self, acc: RValue<'gcc>, src: RValue<'gcc>) -> RValue<'gcc> {
let vector_type = src.get_type().unqualified().dyncast_vector().expect("vector type");
let element_count = vector_type.get_num_units();
(0..element_count).into_iter()
.map(|i| self.context
.new_vector_access(None, src, self.context.new_rvalue_from_int(self.int_type, i as _))
.to_rvalue())
.fold(acc, |x, i| x + i)
}
#[cfg(not(feature="master"))]
pub fn vector_reduce_fadd(&mut self, _acc: RValue<'gcc>, _src: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!();
}
pub fn vector_reduce_fmul_fast(&mut self, _acc: RValue<'gcc>, _src: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!();
}
#[cfg(feature="master")]
pub fn vector_reduce_fmul(&mut self, acc: RValue<'gcc>, src: RValue<'gcc>) -> RValue<'gcc> {
let vector_type = src.get_type().unqualified().dyncast_vector().expect("vector type");
let element_count = vector_type.get_num_units();
(0..element_count).into_iter()
.map(|i| self.context
.new_vector_access(None, src, self.context.new_rvalue_from_int(self.int_type, i as _))
.to_rvalue())
.fold(acc, |x, i| x * i)
}
#[cfg(not(feature="master"))]
pub fn vector_reduce_fmul(&mut self, _acc: RValue<'gcc>, _src: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!()
}
// Inspired by Hacker's Delight min implementation.
pub fn vector_reduce_min(&mut self, src: RValue<'gcc>) -> RValue<'gcc> {
self.vector_reduce(src, |a, b, context| {
let differences_or_zeros = difference_or_zero(a, b, context);
context.new_binary_op(None, BinaryOp::Minus, a.get_type(), a, differences_or_zeros)
context.new_binary_op(None, BinaryOp::Plus, b.get_type(), b, differences_or_zeros)
})
}
@ -1611,38 +1752,148 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
pub fn vector_reduce_max(&mut self, src: RValue<'gcc>) -> RValue<'gcc> {
self.vector_reduce(src, |a, b, context| {
let differences_or_zeros = difference_or_zero(a, b, context);
context.new_binary_op(None, BinaryOp::Plus, b.get_type(), b, differences_or_zeros)
context.new_binary_op(None, BinaryOp::Minus, a.get_type(), a, differences_or_zeros)
})
}
fn vector_extremum(&mut self, a: RValue<'gcc>, b: RValue<'gcc>, direction: ExtremumOperation) -> RValue<'gcc> {
let vector_type = a.get_type();
// mask out the NaNs in b and replace them with the corresponding lane in a, so when a and
// b get compared & spliced together, we get the numeric values instead of NaNs.
let b_nan_mask = self.context.new_comparison(None, ComparisonOp::NotEquals, b, b);
let mask_type = b_nan_mask.get_type();
let b_nan_mask_inverted = self.context.new_unary_op(None, UnaryOp::BitwiseNegate, mask_type, b_nan_mask);
let a_cast = self.context.new_bitcast(None, a, mask_type);
let b_cast = self.context.new_bitcast(None, b, mask_type);
let res = (b_nan_mask & a_cast) | (b_nan_mask_inverted & b_cast);
let b = self.context.new_bitcast(None, res, vector_type);
// now do the actual comparison
let comparison_op = match direction {
ExtremumOperation::Min => ComparisonOp::LessThan,
ExtremumOperation::Max => ComparisonOp::GreaterThan,
};
let cmp = self.context.new_comparison(None, comparison_op, a, b);
let cmp_inverted = self.context.new_unary_op(None, UnaryOp::BitwiseNegate, cmp.get_type(), cmp);
let res = (cmp & a_cast) | (cmp_inverted & res);
self.context.new_bitcast(None, res, vector_type)
}
pub fn vector_fmin(&mut self, a: RValue<'gcc>, b: RValue<'gcc>) -> RValue<'gcc> {
self.vector_extremum(a, b, ExtremumOperation::Min)
}
#[cfg(feature="master")]
pub fn vector_reduce_fmin(&mut self, src: RValue<'gcc>) -> RValue<'gcc> {
let vector_type = src.get_type().unqualified().dyncast_vector().expect("vector type");
let element_count = vector_type.get_num_units();
let mut acc = self.context.new_vector_access(None, src, self.context.new_rvalue_zero(self.int_type)).to_rvalue();
for i in 1..element_count {
let elem = self.context
.new_vector_access(None, src, self.context.new_rvalue_from_int(self.int_type, i as _))
.to_rvalue();
let cmp = self.context.new_comparison(None, ComparisonOp::LessThan, acc, elem);
acc = self.select(cmp, acc, elem);
}
acc
}
#[cfg(not(feature="master"))]
pub fn vector_reduce_fmin(&mut self, _src: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!();
}
pub fn vector_fmax(&mut self, a: RValue<'gcc>, b: RValue<'gcc>) -> RValue<'gcc> {
self.vector_extremum(a, b, ExtremumOperation::Max)
}
#[cfg(feature="master")]
pub fn vector_reduce_fmax(&mut self, src: RValue<'gcc>) -> RValue<'gcc> {
let vector_type = src.get_type().unqualified().dyncast_vector().expect("vector type");
let element_count = vector_type.get_num_units();
let mut acc = self.context.new_vector_access(None, src, self.context.new_rvalue_zero(self.int_type)).to_rvalue();
for i in 1..element_count {
let elem = self.context
.new_vector_access(None, src, self.context.new_rvalue_from_int(self.int_type, i as _))
.to_rvalue();
let cmp = self.context.new_comparison(None, ComparisonOp::GreaterThan, acc, elem);
acc = self.select(cmp, acc, elem);
}
acc
}
#[cfg(not(feature="master"))]
pub fn vector_reduce_fmax(&mut self, _src: RValue<'gcc>) -> RValue<'gcc> {
unimplemented!();
}
pub fn vector_select(&mut self, cond: RValue<'gcc>, then_val: RValue<'gcc>, else_val: RValue<'gcc>) -> RValue<'gcc> {
// cond is a vector of integers, not of bools.
let cond_type = cond.get_type();
let vector_type = cond_type.unqualified().dyncast_vector().expect("vector type");
let vector_type = cond.get_type().unqualified().dyncast_vector().expect("vector type");
let num_units = vector_type.get_num_units();
let element_type = vector_type.get_element_type();
#[cfg(feature="master")]
let (cond, element_type) = {
let then_val_vector_type = then_val.get_type().dyncast_vector().expect("vector type");
let then_val_element_type = then_val_vector_type.get_element_type();
let then_val_element_size = then_val_element_type.get_size();
// NOTE: the mask needs to be of the same size as the other arguments in order for the &
// operation to work.
if then_val_element_size != element_type.get_size() {
let new_element_type = self.type_ix(then_val_element_size as u64 * 8);
let new_vector_type = self.context.new_vector_type(new_element_type, num_units as u64);
let cond = self.context.convert_vector(None, cond, new_vector_type);
(cond, new_element_type)
}
else {
(cond, element_type)
}
};
let cond_type = cond.get_type();
let zeros = vec![self.context.new_rvalue_zero(element_type); num_units];
let zeros = self.context.new_rvalue_from_vector(None, cond_type, &zeros);
let result_type = then_val.get_type();
let masks = self.context.new_comparison(None, ComparisonOp::NotEquals, cond, zeros);
// NOTE: masks is a vector of integers, but the values can be vectors of floats, so use bitcast to make
// the & operation work.
let then_val = self.bitcast_if_needed(then_val, masks.get_type());
let then_vals = masks & then_val;
let ones = vec![self.context.new_rvalue_one(element_type); num_units];
let ones = self.context.new_rvalue_from_vector(None, cond_type, &ones);
let inverted_masks = masks + ones;
let minus_ones = vec![self.context.new_rvalue_from_int(element_type, -1); num_units];
let minus_ones = self.context.new_rvalue_from_vector(None, cond_type, &minus_ones);
let inverted_masks = masks ^ minus_ones;
// NOTE: sometimes, the type of else_val can be different than the type of then_val in
// libgccjit (vector of int vs vector of int32_t), but they should be the same for the AND
// operation to work.
// TODO: remove bitcast now that vector types can be compared?
let else_val = self.context.new_bitcast(None, else_val, then_val.get_type());
let else_vals = inverted_masks & else_val;
then_vals | else_vals
let res = then_vals | else_vals;
self.bitcast_if_needed(res, result_type)
}
}
fn difference_or_zero<'gcc>(a: RValue<'gcc>, b: RValue<'gcc>, context: &'gcc Context<'gcc>) -> RValue<'gcc> {
let difference = a - b;
let masks = context.new_comparison(None, ComparisonOp::GreaterThanEquals, b, a);
// NOTE: masks is a vector of integers, but the values can be vectors of floats, so use bitcast to make
// the & operation work.
let a_type = a.get_type();
let masks =
if masks.get_type() != a_type {
context.new_bitcast(None, masks, a_type)
}
else {
masks
};
difference & masks
}

View File

@ -1,9 +1,10 @@
use gccjit::{FunctionType, RValue};
use rustc_codegen_ssa::traits::BaseTypeMethods;
#[cfg(feature="master")]
use gccjit::{FnAttribute, Visibility};
use gccjit::{FunctionType, Function};
use rustc_middle::ty::{self, Instance, TypeVisitableExt};
use rustc_middle::ty::layout::{FnAbiOf, HasTyCtxt};
use crate::abi::FnAbiGccExt;
use crate::attributes;
use crate::context::CodegenCx;
/// Codegens a reference to a fn/method item, monomorphizing and
@ -13,22 +14,26 @@ use crate::context::CodegenCx;
///
/// - `cx`: the crate context
/// - `instance`: the instance to be instantiated
pub fn get_fn<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, instance: Instance<'tcx>) -> RValue<'gcc> {
pub fn get_fn<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, instance: Instance<'tcx>) -> Function<'gcc> {
let tcx = cx.tcx();
assert!(!instance.substs.needs_infer());
assert!(!instance.substs.has_escaping_bound_vars());
let sym = tcx.symbol_name(instance).name;
if let Some(&func) = cx.function_instances.borrow().get(&instance) {
return func;
}
let sym = tcx.symbol_name(instance).name;
let fn_abi = cx.fn_abi_of_instance(instance, ty::List::empty());
let func =
if let Some(func) = cx.get_declared_value(&sym) {
if let Some(_func) = cx.get_declared_value(&sym) {
// FIXME(antoyo): we never reach this because get_declared_value only returns global variables
// and here we try to get a function.
unreachable!();
/*
// Create a fn pointer with the new signature.
let ptrty = fn_abi.ptr_to_gcc_type(cx);
@ -61,13 +66,105 @@ pub fn get_fn<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, instance: Instance<'tcx>)
}
else {
func
}
}*/
}
else {
cx.linkage.set(FunctionType::Extern);
let func = cx.declare_fn(&sym, &fn_abi);
attributes::from_fn_attrs(cx, func, instance);
let instance_def_id = instance.def_id();
// TODO(antoyo): set linkage and attributes.
// Apply an appropriate linkage/visibility value to our item that we
// just declared.
//
// This is sort of subtle. Inside our codegen unit we started off
// compilation by predefining all our own `MonoItem` instances. That
// is, everything we're codegenning ourselves is already defined. That
// means that anything we're actually codegenning in this codegen unit
// will have hit the above branch in `get_declared_value`. As a result,
// we're guaranteed here that we're declaring a symbol that won't get
// defined, or in other words we're referencing a value from another
// codegen unit or even another crate.
//
// So because this is a foreign value we blanket apply an external
// linkage directive because it's coming from a different object file.
// The visibility here is where it gets tricky. This symbol could be
// referencing some foreign crate or foreign library (an `extern`
// block) in which case we want to leave the default visibility. We may
// also, though, have multiple codegen units. It could be a
// monomorphization, in which case its expected visibility depends on
// whether we are sharing generics or not. The important thing here is
// that the visibility we apply to the declaration is the same one that
// has been applied to the definition (wherever that definition may be).
let is_generic = instance.substs.non_erasable_generics().next().is_some();
if is_generic {
// This is a monomorphization. Its expected visibility depends
// on whether we are in share-generics mode.
if cx.tcx.sess.opts.share_generics() {
// We are in share_generics mode.
if let Some(instance_def_id) = instance_def_id.as_local() {
// This is a definition from the current crate. If the
// definition is unreachable for downstream crates or
// the current crate does not re-export generics, the
// definition of the instance will have been declared
// as `hidden`.
if cx.tcx.is_unreachable_local_definition(instance_def_id)
|| !cx.tcx.local_crate_exports_generics()
{
#[cfg(feature="master")]
func.add_attribute(FnAttribute::Visibility(Visibility::Hidden));
}
} else {
// This is a monomorphization of a generic function
// defined in an upstream crate.
if instance.upstream_monomorphization(tcx).is_some() {
// This is instantiated in another crate. It cannot
// be `hidden`.
} else {
// This is a local instantiation of an upstream definition.
// If the current crate does not re-export it
// (because it is a C library or an executable), it
// will have been declared `hidden`.
if !cx.tcx.local_crate_exports_generics() {
#[cfg(feature="master")]
func.add_attribute(FnAttribute::Visibility(Visibility::Hidden));
}
}
}
} else {
// When not sharing generics, all instances are in the same
// crate and have hidden visibility
#[cfg(feature="master")]
func.add_attribute(FnAttribute::Visibility(Visibility::Hidden));
}
} else {
// This is a non-generic function
if cx.tcx.is_codegened_item(instance_def_id) {
// This is a function that is instantiated in the local crate
if instance_def_id.is_local() {
// This is function that is defined in the local crate.
// If it is not reachable, it is hidden.
if !cx.tcx.is_reachable_non_generic(instance_def_id) {
#[cfg(feature="master")]
func.add_attribute(FnAttribute::Visibility(Visibility::Hidden));
}
} else {
// This is a function from an upstream crate that has
// been instantiated here. These are always hidden.
#[cfg(feature="master")]
func.add_attribute(FnAttribute::Visibility(Visibility::Hidden));
}
}
}
func
};

View File

@ -36,7 +36,7 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
pub fn bytes_in_context<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, bytes: &[u8]) -> RValue<'gcc> {
let context = &cx.context;
let byte_type = context.new_type::<u8>();
let typ = context.new_array_type(None, byte_type, bytes.len() as i32);
let typ = context.new_array_type(None, byte_type, bytes.len() as u64);
let elements: Vec<_> =
bytes.iter()
.map(|&byte| context.new_rvalue_from_int(byte_type, byte as i32))
@ -115,8 +115,8 @@ impl<'gcc, 'tcx> ConstMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
self.const_uint(self.usize_type, i)
}
fn const_u8(&self, _i: u8) -> RValue<'gcc> {
unimplemented!();
fn const_u8(&self, i: u8) -> RValue<'gcc> {
self.const_uint(self.type_u8(), i as u64)
}
fn const_real(&self, typ: Type<'gcc>, val: f64) -> RValue<'gcc> {
@ -133,7 +133,7 @@ impl<'gcc, 'tcx> ConstMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
.1;
let len = s.len();
let cs = self.const_ptrcast(str_global.get_address(None),
self.type_ptr_to(self.layout_of(self.tcx.types.str_).gcc_type(self, true)),
self.type_ptr_to(self.layout_of(self.tcx.types.str_).gcc_type(self)),
);
(cs, self.const_usize(len as u64))
}
@ -174,8 +174,18 @@ impl<'gcc, 'tcx> ConstMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
}
let value = self.const_uint_big(self.type_ix(bitsize), data);
// TODO(bjorn3): assert size is correct
self.const_bitcast(value, ty)
let bytesize = layout.size(self).bytes();
if bitsize > 1 && ty.is_integral() && bytesize as u32 == ty.get_size() {
// NOTE: since the intrinsic _xabort is called with a bitcast, which
// is non-const, but expects a constant, do a normal cast instead of a bitcast.
// FIXME(antoyo): fix bitcast to work in constant contexts.
// TODO(antoyo): perhaps only use bitcast for pointers?
self.context.new_cast(None, value, ty)
}
else {
// TODO(bjorn3): assert size is correct
self.const_bitcast(value, ty)
}
}
Scalar::Ptr(ptr, _size) => {
let (alloc_id, offset) = ptr.into_parts();
@ -227,11 +237,11 @@ impl<'gcc, 'tcx> ConstMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
fn from_const_alloc(&self, layout: TyAndLayout<'tcx>, alloc: ConstAllocation<'tcx>, offset: Size) -> PlaceRef<'tcx, RValue<'gcc>> {
assert_eq!(alloc.inner().align, layout.align.abi);
let ty = self.type_ptr_to(layout.gcc_type(self, true));
let ty = self.type_ptr_to(layout.gcc_type(self));
let value =
if layout.size == Size::ZERO {
let value = self.const_usize(alloc.inner().align.bytes());
self.context.new_cast(None, value, ty)
self.const_bitcast(value, ty)
}
else {
let init = const_alloc_to_gcc(self, alloc);

View File

@ -1,8 +1,8 @@
use gccjit::{GlobalKind, LValue, RValue, ToRValue, Type};
#[cfg(feature = "master")]
use gccjit::FnAttribute;
use gccjit::{Function, GlobalKind, LValue, RValue, ToRValue, Type};
use rustc_codegen_ssa::traits::{BaseTypeMethods, ConstMethods, DerivedTypeMethods, StaticMethods};
use rustc_hir as hir;
use rustc_hir::Node;
use rustc_middle::{bug, span_bug};
use rustc_middle::span_bug;
use rustc_middle::middle::codegen_fn_attrs::{CodegenFnAttrFlags, CodegenFnAttrs};
use rustc_middle::mir::mono::MonoItem;
use rustc_middle::ty::{self, Instance, Ty};
@ -13,6 +13,7 @@ use rustc_target::abi::{self, Align, HasDataLayout, Primitive, Size, WrappingRan
use crate::base;
use crate::context::CodegenCx;
use crate::errors::InvalidMinimumAlignment;
use crate::type_of::LayoutGccExt;
impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
@ -30,6 +31,21 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
}
}
fn set_global_alignment<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, gv: LValue<'gcc>, mut align: Align) {
// The target may require greater alignment for globals than the type does.
// Note: GCC and Clang also allow `__attribute__((aligned))` on variables,
// which can force it to be smaller. Rust doesn't support this yet.
if let Some(min) = cx.sess().target.min_global_align {
match Align::from_bits(min) {
Ok(min) => align = align.max(min),
Err(err) => {
cx.sess().emit_err(InvalidMinimumAlignment { err });
}
}
}
gv.set_alignment(align.bytes() as i32);
}
impl<'gcc, 'tcx> StaticMethods for CodegenCx<'gcc, 'tcx> {
fn static_addr_of(&self, cv: RValue<'gcc>, align: Align, kind: Option<&str>) -> RValue<'gcc> {
// TODO(antoyo): implement a proper rvalue comparison in libgccjit instead of doing the
@ -79,9 +95,9 @@ impl<'gcc, 'tcx> StaticMethods for CodegenCx<'gcc, 'tcx> {
let instance = Instance::mono(self.tcx, def_id);
let ty = instance.ty(self.tcx, ty::ParamEnv::reveal_all());
let gcc_type = self.layout_of(ty).gcc_type(self, true);
let gcc_type = self.layout_of(ty).gcc_type(self);
// TODO(antoyo): set alignment.
set_global_alignment(self, global, self.align_of(ty));
let value = self.bitcast_if_needed(value, gcc_type);
global.global_set_initializer_rvalue(value);
@ -158,12 +174,19 @@ impl<'gcc, 'tcx> StaticMethods for CodegenCx<'gcc, 'tcx> {
// TODO(antoyo)
}
fn add_compiler_used_global(&self, _global: RValue<'gcc>) {
// TODO(antoyo)
fn add_compiler_used_global(&self, global: RValue<'gcc>) {
// NOTE: seems like GCC does not make the distinction between compiler.used and used.
self.add_used_global(global);
}
}
impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
#[cfg_attr(not(feature="master"), allow(unused_variables))]
pub fn add_used_function(&self, function: Function<'gcc>) {
#[cfg(feature = "master")]
function.add_attribute(FnAttribute::Used);
}
pub fn static_addr_of_mut(&self, cv: RValue<'gcc>, align: Align, kind: Option<&str>) -> RValue<'gcc> {
let global =
match kind {
@ -208,82 +231,59 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
let sym = self.tcx.symbol_name(instance).name;
let global =
if let Some(def_id) = def_id.as_local() {
let id = self.tcx.hir().local_def_id_to_hir_id(def_id);
let llty = self.layout_of(ty).gcc_type(self, true);
// FIXME: refactor this to work without accessing the HIR
let global = match self.tcx.hir().get(id) {
Node::Item(&hir::Item { span, kind: hir::ItemKind::Static(..), .. }) => {
if let Some(global) = self.get_declared_value(&sym) {
if self.val_ty(global) != self.type_ptr_to(llty) {
span_bug!(span, "Conflicting types for static");
}
}
let is_tls = fn_attrs.flags.contains(CodegenFnAttrFlags::THREAD_LOCAL);
let global = self.declare_global(
&sym,
llty,
GlobalKind::Exported,
is_tls,
fn_attrs.link_section,
);
if !self.tcx.is_reachable_non_generic(def_id) {
// TODO(antoyo): set visibility.
}
global
}
Node::ForeignItem(&hir::ForeignItem {
span: _,
kind: hir::ForeignItemKind::Static(..),
..
}) => {
let fn_attrs = self.tcx.codegen_fn_attrs(def_id);
check_and_apply_linkage(&self, &fn_attrs, ty, sym)
}
item => bug!("get_static: expected static, found {:?}", item),
};
global
}
else {
// FIXME(nagisa): perhaps the map of externs could be offloaded to llvm somehow?
//debug!("get_static: sym={} item_attr={:?}", sym, self.tcx.item_attrs(def_id));
let attrs = self.tcx.codegen_fn_attrs(def_id);
let global = check_and_apply_linkage(&self, &attrs, ty, sym);
let needs_dll_storage_attr = false; // TODO(antoyo)
// If this assertion triggers, there's something wrong with commandline
// argument validation.
debug_assert!(
!(self.tcx.sess.opts.cg.linker_plugin_lto.enabled()
&& self.tcx.sess.target.options.is_like_msvc
&& self.tcx.sess.opts.cg.prefer_dynamic)
);
if needs_dll_storage_attr {
// This item is external but not foreign, i.e., it originates from an external Rust
// crate. Since we don't know whether this crate will be linked dynamically or
// statically in the final application, we always mark such symbols as 'dllimport'.
// If final linkage happens to be static, we rely on compiler-emitted __imp_ stubs
// to make things work.
//
// However, in some scenarios we defer emission of statics to downstream
// crates, so there are cases where a static with an upstream DefId
// is actually present in the current crate. We can find out via the
// is_codegened_item query.
if !self.tcx.is_codegened_item(def_id) {
unimplemented!();
}
if def_id.is_local() && !self.tcx.is_foreign_item(def_id) {
let llty = self.layout_of(ty).gcc_type(self);
if let Some(global) = self.get_declared_value(sym) {
if self.val_ty(global) != self.type_ptr_to(llty) {
span_bug!(self.tcx.def_span(def_id), "Conflicting types for static");
}
global
};
}
let is_tls = fn_attrs.flags.contains(CodegenFnAttrFlags::THREAD_LOCAL);
let global = self.declare_global(
&sym,
llty,
GlobalKind::Exported,
is_tls,
fn_attrs.link_section,
);
if !self.tcx.is_reachable_non_generic(def_id) {
// TODO(antoyo): set visibility.
}
global
} else {
check_and_apply_linkage(&self, &fn_attrs, ty, sym)
};
if !def_id.is_local() {
let needs_dll_storage_attr = false; // TODO(antoyo)
// If this assertion triggers, there's something wrong with commandline
// argument validation.
debug_assert!(
!(self.tcx.sess.opts.cg.linker_plugin_lto.enabled()
&& self.tcx.sess.target.options.is_like_msvc
&& self.tcx.sess.opts.cg.prefer_dynamic)
);
if needs_dll_storage_attr {
// This item is external but not foreign, i.e., it originates from an external Rust
// crate. Since we don't know whether this crate will be linked dynamically or
// statically in the final application, we always mark such symbols as 'dllimport'.
// If final linkage happens to be static, we rely on compiler-emitted __imp_ stubs
// to make things work.
//
// However, in some scenarios we defer emission of statics to downstream
// crates, so there are cases where a static with an upstream DefId
// is actually present in the current crate. We can find out via the
// is_codegened_item query.
if !self.tcx.is_codegened_item(def_id) {
unimplemented!();
}
}
}
// TODO(antoyo): set dll storage class.
@ -357,7 +357,7 @@ pub fn codegen_static_initializer<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, def_id
fn check_and_apply_linkage<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, attrs: &CodegenFnAttrs, ty: Ty<'tcx>, sym: &str) -> LValue<'gcc> {
let is_tls = attrs.flags.contains(CodegenFnAttrFlags::THREAD_LOCAL);
let llty = cx.layout_of(ty).gcc_type(cx, true);
let gcc_type = cx.layout_of(ty).gcc_type(cx);
if let Some(linkage) = attrs.import_linkage {
// Declare a symbol `foo` with the desired linkage.
let global1 = cx.declare_global_with_linkage(&sym, cx.type_i8(), base::global_linkage_to_gcc(linkage));
@ -370,9 +370,10 @@ fn check_and_apply_linkage<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, attrs: &Codeg
// zero.
let mut real_name = "_rust_extern_with_linkage_".to_string();
real_name.push_str(&sym);
let global2 = cx.define_global(&real_name, llty, is_tls, attrs.link_section);
let global2 = cx.define_global(&real_name, gcc_type, is_tls, attrs.link_section);
// TODO(antoyo): set linkage.
global2.global_set_initializer_rvalue(global1.get_address(None));
let value = cx.const_ptrcast(global1.get_address(None), gcc_type);
global2.global_set_initializer_rvalue(value);
// TODO(antoyo): use global_set_initializer() when it will work.
global2
}
@ -386,6 +387,6 @@ fn check_and_apply_linkage<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, attrs: &Codeg
// don't do this then linker errors can be generated where the linker
// complains that one object files has a thread local version of the
// symbol and another one doesn't.
cx.declare_global(&sym, llty, GlobalKind::Imported, is_tls, attrs.link_section)
cx.declare_global(&sym, gcc_type, GlobalKind::Imported, is_tls, attrs.link_section)
}
}

View File

@ -1,9 +1,10 @@
use std::cell::{Cell, RefCell};
use gccjit::{Block, CType, Context, Function, FunctionPtrType, FunctionType, LValue, RValue, Struct, Type};
use gccjit::{Block, CType, Context, Function, FunctionPtrType, FunctionType, LValue, RValue, Type};
use rustc_codegen_ssa::base::wants_msvc_seh;
use rustc_codegen_ssa::traits::{
BackendTypes,
BaseTypeMethods,
MiscMethods,
};
use rustc_data_structures::base_n;
@ -11,7 +12,7 @@ use rustc_data_structures::fx::{FxHashMap, FxHashSet};
use rustc_middle::span_bug;
use rustc_middle::mir::mono::CodegenUnit;
use rustc_middle::ty::{self, Instance, ParamEnv, PolyExistentialTraitRef, Ty, TyCtxt};
use rustc_middle::ty::layout::{FnAbiError, FnAbiOfHelpers, FnAbiRequest, HasParamEnv, HasTyCtxt, LayoutError, TyAndLayout, LayoutOfHelpers};
use rustc_middle::ty::layout::{FnAbiError, FnAbiOf, FnAbiOfHelpers, FnAbiRequest, HasParamEnv, HasTyCtxt, LayoutError, TyAndLayout, LayoutOfHelpers};
use rustc_session::Session;
use rustc_span::{Span, source_map::respan};
use rustc_target::abi::{call::FnAbi, HasDataLayout, PointeeInfo, Size, TargetDataLayout, VariantIdx};
@ -33,6 +34,7 @@ pub struct CodegenCx<'gcc, 'tcx> {
// TODO(bjorn3): Can this field be removed?
pub current_func: RefCell<Option<Function<'gcc>>>,
pub normal_function_addresses: RefCell<FxHashSet<RValue<'gcc>>>,
pub function_address_names: RefCell<FxHashMap<RValue<'gcc>, String>>,
pub functions: RefCell<FxHashMap<String, Function<'gcc>>>,
pub intrinsics: RefCell<FxHashMap<String, Function<'gcc>>>,
@ -78,12 +80,10 @@ pub struct CodegenCx<'gcc, 'tcx> {
pub struct_types: RefCell<FxHashMap<Vec<Type<'gcc>>, Type<'gcc>>>,
pub types_with_fields_to_set: RefCell<FxHashMap<Type<'gcc>, (Struct<'gcc>, TyAndLayout<'tcx>)>>,
/// Cache instances of monomorphic and polymorphic items
pub instances: RefCell<FxHashMap<Instance<'tcx>, LValue<'gcc>>>,
/// Cache function instances of monomorphic and polymorphic items
pub function_instances: RefCell<FxHashMap<Instance<'tcx>, RValue<'gcc>>>,
pub function_instances: RefCell<FxHashMap<Instance<'tcx>, Function<'gcc>>>,
/// Cache generated vtables
pub vtables: RefCell<FxHashMap<(Ty<'tcx>, Option<ty::PolyExistentialTraitRef<'tcx>>), RValue<'gcc>>>,
@ -110,6 +110,7 @@ pub struct CodegenCx<'gcc, 'tcx> {
local_gen_sym_counter: Cell<usize>,
eh_personality: Cell<Option<RValue<'gcc>>>,
pub rust_try_fn: Cell<Option<(Type<'gcc>, Function<'gcc>)>>,
pub pointee_infos: RefCell<FxHashMap<(Ty<'tcx>, Size), Option<PointeeInfo>>>,
@ -119,6 +120,8 @@ pub struct CodegenCx<'gcc, 'tcx> {
/// they can be dereferenced later.
/// FIXME(antoyo): fix the rustc API to avoid having this hack.
pub structs_as_pointer: RefCell<FxHashSet<RValue<'gcc>>>,
pub cleanup_blocks: RefCell<FxHashSet<Block<'gcc>>>,
}
impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
@ -194,6 +197,7 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
context,
current_func: RefCell::new(None),
normal_function_addresses: Default::default(),
function_address_names: Default::default(),
functions: RefCell::new(functions),
intrinsics: RefCell::new(FxHashMap::default()),
@ -243,11 +247,12 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
types: Default::default(),
tcx,
struct_types: Default::default(),
types_with_fields_to_set: Default::default(),
local_gen_sym_counter: Cell::new(0),
eh_personality: Cell::new(None),
rust_try_fn: Cell::new(None),
pointee_infos: Default::default(),
structs_as_pointer: Default::default(),
cleanup_blocks: Default::default(),
}
}
@ -327,8 +332,9 @@ impl<'gcc, 'tcx> MiscMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
fn get_fn(&self, instance: Instance<'tcx>) -> RValue<'gcc> {
let func = get_fn(self, instance);
*self.current_func.borrow_mut() = Some(self.rvalue_as_function(func));
func
*self.current_func.borrow_mut() = Some(func);
// FIXME(antoyo): this is a wrong cast. That requires changing the compiler API.
unsafe { std::mem::transmute(func) }
}
fn get_fn_addr(&self, instance: Instance<'tcx>) -> RValue<'gcc> {
@ -339,8 +345,7 @@ impl<'gcc, 'tcx> MiscMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
self.intrinsics.borrow()[func_name].clone()
}
else {
let func = get_fn(self, instance);
self.rvalue_as_function(func)
get_fn(self, instance)
};
let ptr = func.get_address(None);
@ -348,6 +353,7 @@ impl<'gcc, 'tcx> MiscMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
// FIXME(antoyo): the rustc API seems to call get_fn_addr() when not needed (e.g. for FFI).
self.normal_function_addresses.borrow_mut().insert(ptr);
self.function_address_names.borrow_mut().insert(ptr, func_name.to_string());
ptr
}
@ -377,31 +383,40 @@ impl<'gcc, 'tcx> MiscMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
return llpersonality;
}
let tcx = self.tcx;
let llfn = match tcx.lang_items().eh_personality() {
Some(def_id) if !wants_msvc_seh(self.sess()) => self.get_fn_addr(
ty::Instance::resolve(
tcx,
ty::ParamEnv::reveal_all(),
def_id,
ty::List::empty(),
)
.unwrap().unwrap(),
),
_ => {
let _name = if wants_msvc_seh(self.sess()) {
"__CxxFrameHandler3"
} else {
"rust_eh_personality"
};
//let func = self.declare_func(name, self.type_i32(), &[], true);
// FIXME(antoyo): this hack should not be needed. That will probably be removed when
// unwinding support is added.
self.context.new_rvalue_from_int(self.int_type, 0)
}
};
let func =
match tcx.lang_items().eh_personality() {
Some(def_id) if !wants_msvc_seh(self.sess()) => {
let instance =
ty::Instance::resolve(
tcx,
ty::ParamEnv::reveal_all(),
def_id,
ty::List::empty(),
)
.unwrap().unwrap();
let symbol_name = tcx.symbol_name(instance).name;
let fn_abi = self.fn_abi_of_instance(instance, ty::List::empty());
self.linkage.set(FunctionType::Extern);
let func = self.declare_fn(symbol_name, &fn_abi);
let func: RValue<'gcc> = unsafe { std::mem::transmute(func) };
func
},
_ => {
let name =
if wants_msvc_seh(self.sess()) {
"__CxxFrameHandler3"
}
else {
"rust_eh_personality"
};
let func = self.declare_func(name, self.type_i32(), &[], true);
unsafe { std::mem::transmute(func) }
}
};
// TODO(antoyo): apply target cpu attributes.
self.eh_personality.set(Some(llfn));
llfn
self.eh_personality.set(Some(func));
func
}
fn sess(&self) -> &Session {

View File

@ -38,12 +38,10 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
global
}
/*pub fn declare_func(&self, name: &str, return_type: Type<'gcc>, params: &[Type<'gcc>], variadic: bool) -> RValue<'gcc> {
self.linkage.set(FunctionType::Exported);
let func = declare_raw_fn(self, name, () /*llvm::CCallConv*/, return_type, params, variadic);
// FIXME(antoyo): this is a wrong cast. That requires changing the compiler API.
unsafe { std::mem::transmute(func) }
}*/
pub fn declare_func(&self, name: &str, return_type: Type<'gcc>, params: &[Type<'gcc>], variadic: bool) -> Function<'gcc> {
self.linkage.set(FunctionType::Extern);
declare_raw_fn(self, name, () /*llvm::CCallConv*/, return_type, params, variadic)
}
pub fn declare_global(&self, name: &str, ty: Type<'gcc>, global_kind: GlobalKind, is_tls: bool, link_section: Option<Symbol>) -> LValue<'gcc> {
let global = self.context.new_global(None, global_kind, ty, name);
@ -79,12 +77,11 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
unsafe { std::mem::transmute(func) }
}
pub fn declare_fn(&self, name: &str, fn_abi: &FnAbi<'tcx, Ty<'tcx>>) -> RValue<'gcc> {
pub fn declare_fn(&self, name: &str, fn_abi: &FnAbi<'tcx, Ty<'tcx>>) -> Function<'gcc> {
let (return_type, params, variadic, on_stack_param_indices) = fn_abi.gcc_type(self);
let func = declare_raw_fn(self, name, () /*fn_abi.llvm_cconv()*/, return_type, &params, variadic);
self.on_stack_function_params.borrow_mut().insert(func, on_stack_param_indices);
// FIXME(antoyo): this is a wrong cast. That requires changing the compiler API.
unsafe { std::mem::transmute(func) }
func
}
pub fn define_global(&self, name: &str, ty: Type<'gcc>, is_tls: bool, link_section: Option<Symbol>) -> LValue<'gcc> {

View File

@ -221,3 +221,18 @@ pub(crate) struct UnwindingInlineAsm {
#[primary_span]
pub span: Span,
}
#[derive(Diagnostic)]
#[diag(codegen_gcc_invalid_minimum_alignment)]
pub(crate) struct InvalidMinimumAlignment {
pub err: String,
}
#[derive(Diagnostic)]
#[diag(codegen_gcc_tied_target_features)]
#[help]
pub(crate) struct TiedTargetFeatures {
#[primary_span]
pub span: Span,
pub features: String,
}

View File

@ -389,18 +389,22 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
};
self.context.new_comparison(None, op, cmp, self.context.new_rvalue_from_int(self.int_type, limit))
}
else if a_type.get_pointee().is_some() && b_type.get_pointee().is_some() {
// NOTE: gcc cannot compare pointers to different objects, but rustc does that, so cast them to usize.
lhs = self.context.new_bitcast(None, lhs, self.usize_type);
rhs = self.context.new_bitcast(None, rhs, self.usize_type);
self.context.new_comparison(None, op.to_gcc_comparison(), lhs, rhs)
}
else {
let left_type = lhs.get_type();
let right_type = rhs.get_type();
if left_type != right_type {
if a_type != b_type {
// NOTE: because libgccjit cannot compare function pointers.
if left_type.dyncast_function_ptr_type().is_some() && right_type.dyncast_function_ptr_type().is_some() {
if a_type.dyncast_function_ptr_type().is_some() && b_type.dyncast_function_ptr_type().is_some() {
lhs = self.context.new_cast(None, lhs, self.usize_type.make_pointer());
rhs = self.context.new_cast(None, rhs, self.usize_type.make_pointer());
}
// NOTE: hack because we try to cast a vector type to the same vector type.
else if format!("{:?}", left_type) != format!("{:?}", right_type) {
rhs = self.context.new_cast(None, rhs, left_type);
else if format!("{:?}", a_type) != format!("{:?}", b_type) {
rhs = self.context.new_cast(None, rhs, a_type);
}
}
self.context.new_comparison(None, op.to_gcc_comparison(), lhs, rhs)

File diff suppressed because it is too large Load Diff

View File

@ -1,159 +1,387 @@
use std::borrow::Cow;
use gccjit::{Function, FunctionPtrType, RValue, ToRValue};
use gccjit::{Function, FunctionPtrType, RValue, ToRValue, UnaryOp};
use rustc_codegen_ssa::traits::BuilderMethods;
use crate::{context::CodegenCx, builder::Builder};
pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>(builder: &Builder<'a, 'gcc, 'tcx>, gcc_func: FunctionPtrType<'gcc>, mut args: Cow<'b, [RValue<'gcc>]>, func_name: &str) -> Cow<'b, [RValue<'gcc>]> {
pub fn adjust_intrinsic_arguments<'a, 'b, 'gcc, 'tcx>(builder: &Builder<'a, 'gcc, 'tcx>, gcc_func: FunctionPtrType<'gcc>, mut args: Cow<'b, [RValue<'gcc>]>, func_name: &str, original_function_name: Option<&String>) -> Cow<'b, [RValue<'gcc>]> {
// Some LLVM intrinsics do not map 1-to-1 to GCC intrinsics, so we add the missing
// arguments here.
if gcc_func.get_param_count() != args.len() {
match &*func_name {
"__builtin_ia32_pmuldq512_mask" | "__builtin_ia32_pmuludq512_mask"
// FIXME(antoyo): the following intrinsics has 4 (or 5) arguments according to the doc, but is defined with 2 (or 3) arguments in library/stdarch/crates/core_arch/src/x86/avx512f.rs.
// NOTE: the following intrinsics have a different number of parameters in LLVM and GCC.
"__builtin_ia32_prold512_mask" | "__builtin_ia32_pmuldq512_mask" | "__builtin_ia32_pmuludq512_mask"
| "__builtin_ia32_pmaxsd512_mask" | "__builtin_ia32_pmaxsq512_mask" | "__builtin_ia32_pmaxsq256_mask"
| "__builtin_ia32_pmaxsq128_mask" | "__builtin_ia32_maxps512_mask" | "__builtin_ia32_maxpd512_mask"
| "__builtin_ia32_pmaxud512_mask" | "__builtin_ia32_pmaxuq512_mask" | "__builtin_ia32_pmaxuq256_mask"
| "__builtin_ia32_pmaxuq128_mask"
| "__builtin_ia32_pmaxsq128_mask" | "__builtin_ia32_pmaxud512_mask" | "__builtin_ia32_pmaxuq512_mask"
| "__builtin_ia32_pminsd512_mask" | "__builtin_ia32_pminsq512_mask" | "__builtin_ia32_pminsq256_mask"
| "__builtin_ia32_pminsq128_mask" | "__builtin_ia32_minps512_mask" | "__builtin_ia32_minpd512_mask"
| "__builtin_ia32_pminud512_mask" | "__builtin_ia32_pminuq512_mask" | "__builtin_ia32_pminuq256_mask"
| "__builtin_ia32_pminuq128_mask" | "__builtin_ia32_sqrtps512_mask" | "__builtin_ia32_sqrtpd512_mask"
| "__builtin_ia32_pminsq128_mask" | "__builtin_ia32_pminud512_mask" | "__builtin_ia32_pminuq512_mask"
| "__builtin_ia32_prolq512_mask" | "__builtin_ia32_prorq512_mask" | "__builtin_ia32_pslldi512_mask"
| "__builtin_ia32_psrldi512_mask" | "__builtin_ia32_psllqi512_mask" | "__builtin_ia32_psrlqi512_mask"
| "__builtin_ia32_pslld512_mask" | "__builtin_ia32_psrld512_mask" | "__builtin_ia32_psllq512_mask"
| "__builtin_ia32_psrlq512_mask" | "__builtin_ia32_psrad512_mask" | "__builtin_ia32_psraq512_mask"
| "__builtin_ia32_psradi512_mask" | "__builtin_ia32_psraqi512_mask" | "__builtin_ia32_psrav16si_mask"
| "__builtin_ia32_psrav8di_mask" | "__builtin_ia32_prolvd512_mask" | "__builtin_ia32_prorvd512_mask"
| "__builtin_ia32_prolvq512_mask" | "__builtin_ia32_prorvq512_mask" | "__builtin_ia32_psllv16si_mask"
| "__builtin_ia32_psrlv16si_mask" | "__builtin_ia32_psllv8di_mask" | "__builtin_ia32_psrlv8di_mask"
| "__builtin_ia32_permvarsi512_mask" | "__builtin_ia32_vpermilvarps512_mask"
| "__builtin_ia32_vpermilvarpd512_mask" | "__builtin_ia32_permvardi512_mask"
| "__builtin_ia32_permvarsf512_mask" | "__builtin_ia32_permvarqi512_mask"
| "__builtin_ia32_permvarqi256_mask" | "__builtin_ia32_permvarqi128_mask"
| "__builtin_ia32_vpmultishiftqb512_mask" | "__builtin_ia32_vpmultishiftqb256_mask"
| "__builtin_ia32_vpmultishiftqb128_mask"
=> {
// TODO: refactor by separating those intrinsics outside of this branch.
let add_before_last_arg =
match &*func_name {
"__builtin_ia32_maxps512_mask" | "__builtin_ia32_maxpd512_mask"
| "__builtin_ia32_minps512_mask" | "__builtin_ia32_minpd512_mask"
| "__builtin_ia32_sqrtps512_mask" | "__builtin_ia32_sqrtpd512_mask" => true,
_ => false,
};
let new_first_arg_is_zero =
match &*func_name {
"__builtin_ia32_pmaxuq256_mask" | "__builtin_ia32_pmaxuq128_mask"
| "__builtin_ia32_pminuq256_mask" | "__builtin_ia32_pminuq128_mask" => true,
_ => false
};
let arg3_index =
match &*func_name {
"__builtin_ia32_sqrtps512_mask" | "__builtin_ia32_sqrtpd512_mask" => 1,
_ => 2,
};
let mut new_args = args.to_vec();
let arg3_type = gcc_func.get_param_type(arg3_index);
let first_arg =
if new_first_arg_is_zero {
let vector_type = arg3_type.dyncast_vector().expect("vector type");
let zero = builder.context.new_rvalue_zero(vector_type.get_element_type());
let num_units = vector_type.get_num_units();
builder.context.new_rvalue_from_vector(None, arg3_type, &vec![zero; num_units])
}
else {
builder.current_func().new_local(None, arg3_type, "undefined_for_intrinsic").to_rvalue()
};
if add_before_last_arg {
new_args.insert(new_args.len() - 1, first_arg);
let mut new_args = args.to_vec();
let arg3_type = gcc_func.get_param_type(2);
let first_arg = builder.current_func().new_local(None, arg3_type, "undefined_for_intrinsic").to_rvalue();
new_args.push(first_arg);
let arg4_type = gcc_func.get_param_type(3);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
new_args.push(minus_one);
args = new_args.into();
},
"__builtin_ia32_pmaxuq256_mask" | "__builtin_ia32_pmaxuq128_mask" | "__builtin_ia32_pminuq256_mask"
| "__builtin_ia32_pminuq128_mask" | "__builtin_ia32_prold256_mask" | "__builtin_ia32_prold128_mask"
| "__builtin_ia32_prord512_mask" | "__builtin_ia32_prord256_mask" | "__builtin_ia32_prord128_mask"
| "__builtin_ia32_prolq256_mask" | "__builtin_ia32_prolq128_mask" | "__builtin_ia32_prorq256_mask"
| "__builtin_ia32_prorq128_mask" | "__builtin_ia32_psraq256_mask" | "__builtin_ia32_psraq128_mask"
| "__builtin_ia32_psraqi256_mask" | "__builtin_ia32_psraqi128_mask" | "__builtin_ia32_psravq256_mask"
| "__builtin_ia32_psravq128_mask" | "__builtin_ia32_prolvd256_mask" | "__builtin_ia32_prolvd128_mask"
| "__builtin_ia32_prorvd256_mask" | "__builtin_ia32_prorvd128_mask" | "__builtin_ia32_prolvq256_mask"
| "__builtin_ia32_prolvq128_mask" | "__builtin_ia32_prorvq256_mask" | "__builtin_ia32_prorvq128_mask"
| "__builtin_ia32_permvardi256_mask" | "__builtin_ia32_permvardf512_mask" | "__builtin_ia32_permvardf256_mask"
| "__builtin_ia32_pmulhuw512_mask" | "__builtin_ia32_pmulhw512_mask" | "__builtin_ia32_pmulhrsw512_mask"
| "__builtin_ia32_pmaxuw512_mask" | "__builtin_ia32_pmaxub512_mask" | "__builtin_ia32_pmaxsw512_mask"
| "__builtin_ia32_pmaxsb512_mask" | "__builtin_ia32_pminuw512_mask" | "__builtin_ia32_pminub512_mask"
| "__builtin_ia32_pminsw512_mask" | "__builtin_ia32_pminsb512_mask"
| "__builtin_ia32_pmaddwd512_mask" | "__builtin_ia32_pmaddubsw512_mask" | "__builtin_ia32_packssdw512_mask"
| "__builtin_ia32_packsswb512_mask" | "__builtin_ia32_packusdw512_mask" | "__builtin_ia32_packuswb512_mask"
| "__builtin_ia32_pavgw512_mask" | "__builtin_ia32_pavgb512_mask" | "__builtin_ia32_psllw512_mask"
| "__builtin_ia32_psllwi512_mask" | "__builtin_ia32_psllv32hi_mask" | "__builtin_ia32_psrlw512_mask"
| "__builtin_ia32_psrlwi512_mask" | "__builtin_ia32_psllv16hi_mask" | "__builtin_ia32_psllv8hi_mask"
| "__builtin_ia32_psrlv32hi_mask" | "__builtin_ia32_psraw512_mask" | "__builtin_ia32_psrawi512_mask"
| "__builtin_ia32_psrlv16hi_mask" | "__builtin_ia32_psrlv8hi_mask" | "__builtin_ia32_psrav32hi_mask"
| "__builtin_ia32_permvarhi512_mask" | "__builtin_ia32_pshufb512_mask" | "__builtin_ia32_psrav16hi_mask"
| "__builtin_ia32_psrav8hi_mask" | "__builtin_ia32_permvarhi256_mask" | "__builtin_ia32_permvarhi128_mask"
=> {
let mut new_args = args.to_vec();
let arg3_type = gcc_func.get_param_type(2);
let vector_type = arg3_type.dyncast_vector().expect("vector type");
let zero = builder.context.new_rvalue_zero(vector_type.get_element_type());
let num_units = vector_type.get_num_units();
let first_arg = builder.context.new_rvalue_from_vector(None, arg3_type, &vec![zero; num_units]);
new_args.push(first_arg);
let arg4_type = gcc_func.get_param_type(3);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
new_args.push(minus_one);
args = new_args.into();
},
"__builtin_ia32_dbpsadbw512_mask" | "__builtin_ia32_dbpsadbw256_mask" | "__builtin_ia32_dbpsadbw128_mask" => {
let mut new_args = args.to_vec();
let arg4_type = gcc_func.get_param_type(3);
let vector_type = arg4_type.dyncast_vector().expect("vector type");
let zero = builder.context.new_rvalue_zero(vector_type.get_element_type());
let num_units = vector_type.get_num_units();
let first_arg = builder.context.new_rvalue_from_vector(None, arg4_type, &vec![zero; num_units]);
new_args.push(first_arg);
let arg5_type = gcc_func.get_param_type(4);
let minus_one = builder.context.new_rvalue_from_int(arg5_type, -1);
new_args.push(minus_one);
args = new_args.into();
},
"__builtin_ia32_vplzcntd_512_mask" | "__builtin_ia32_vplzcntd_256_mask" | "__builtin_ia32_vplzcntd_128_mask"
| "__builtin_ia32_vplzcntq_512_mask" | "__builtin_ia32_vplzcntq_256_mask" | "__builtin_ia32_vplzcntq_128_mask" => {
let mut new_args = args.to_vec();
// Remove last arg as it doesn't seem to be used in GCC and is always false.
new_args.pop();
let arg2_type = gcc_func.get_param_type(1);
let vector_type = arg2_type.dyncast_vector().expect("vector type");
let zero = builder.context.new_rvalue_zero(vector_type.get_element_type());
let num_units = vector_type.get_num_units();
let first_arg = builder.context.new_rvalue_from_vector(None, arg2_type, &vec![zero; num_units]);
new_args.push(first_arg);
let arg3_type = gcc_func.get_param_type(2);
let minus_one = builder.context.new_rvalue_from_int(arg3_type, -1);
new_args.push(minus_one);
args = new_args.into();
},
"__builtin_ia32_vpconflictsi_512_mask" | "__builtin_ia32_vpconflictsi_256_mask"
| "__builtin_ia32_vpconflictsi_128_mask" | "__builtin_ia32_vpconflictdi_512_mask"
| "__builtin_ia32_vpconflictdi_256_mask" | "__builtin_ia32_vpconflictdi_128_mask" => {
let mut new_args = args.to_vec();
let arg2_type = gcc_func.get_param_type(1);
let vector_type = arg2_type.dyncast_vector().expect("vector type");
let zero = builder.context.new_rvalue_zero(vector_type.get_element_type());
let num_units = vector_type.get_num_units();
let first_arg = builder.context.new_rvalue_from_vector(None, arg2_type, &vec![zero; num_units]);
new_args.push(first_arg);
let arg3_type = gcc_func.get_param_type(2);
let minus_one = builder.context.new_rvalue_from_int(arg3_type, -1);
new_args.push(minus_one);
args = new_args.into();
},
"__builtin_ia32_pternlogd512_mask" | "__builtin_ia32_pternlogd256_mask"
| "__builtin_ia32_pternlogd128_mask" | "__builtin_ia32_pternlogq512_mask"
| "__builtin_ia32_pternlogq256_mask" | "__builtin_ia32_pternlogq128_mask" => {
let mut new_args = args.to_vec();
let arg5_type = gcc_func.get_param_type(4);
let minus_one = builder.context.new_rvalue_from_int(arg5_type, -1);
new_args.push(minus_one);
args = new_args.into();
},
"__builtin_ia32_vfmaddps512_mask" | "__builtin_ia32_vfmaddpd512_mask" => {
let mut new_args = args.to_vec();
let mut last_arg = None;
if args.len() == 4 {
last_arg = new_args.pop();
}
let arg4_type = gcc_func.get_param_type(3);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
new_args.push(minus_one);
if args.len() == 3 {
// Both llvm.fma.v16f32 and llvm.x86.avx512.vfmadd.ps.512 maps to
// the same GCC intrinsic, but the former has 3 parameters and the
// latter has 4 so it doesn't require this additional argument.
let arg5_type = gcc_func.get_param_type(4);
new_args.push(builder.context.new_rvalue_from_int(arg5_type, 4));
}
if let Some(last_arg) = last_arg {
new_args.push(last_arg);
}
args = new_args.into();
},
"__builtin_ia32_addps512_mask" | "__builtin_ia32_addpd512_mask"
| "__builtin_ia32_subps512_mask" | "__builtin_ia32_subpd512_mask"
| "__builtin_ia32_mulps512_mask" | "__builtin_ia32_mulpd512_mask"
| "__builtin_ia32_divps512_mask" | "__builtin_ia32_divpd512_mask"
| "__builtin_ia32_maxps512_mask" | "__builtin_ia32_maxpd512_mask"
| "__builtin_ia32_minps512_mask" | "__builtin_ia32_minpd512_mask" => {
let mut new_args = args.to_vec();
let last_arg = new_args.pop().expect("last arg");
let arg3_type = gcc_func.get_param_type(2);
let undefined = builder.current_func().new_local(None, arg3_type, "undefined_for_intrinsic").to_rvalue();
new_args.push(undefined);
let arg4_type = gcc_func.get_param_type(3);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
new_args.push(minus_one);
new_args.push(last_arg);
args = new_args.into();
},
"__builtin_ia32_vfmaddsubps512_mask" | "__builtin_ia32_vfmaddsubpd512_mask" => {
let mut new_args = args.to_vec();
let last_arg = new_args.pop().expect("last arg");
let arg4_type = gcc_func.get_param_type(3);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
new_args.push(minus_one);
new_args.push(last_arg);
args = new_args.into();
},
"__builtin_ia32_vpermi2vard512_mask" | "__builtin_ia32_vpermi2vard256_mask"
| "__builtin_ia32_vpermi2vard128_mask" | "__builtin_ia32_vpermi2varq512_mask"
| "__builtin_ia32_vpermi2varq256_mask" | "__builtin_ia32_vpermi2varq128_mask"
| "__builtin_ia32_vpermi2varps512_mask" | "__builtin_ia32_vpermi2varps256_mask"
| "__builtin_ia32_vpermi2varps128_mask" | "__builtin_ia32_vpermi2varpd512_mask"
| "__builtin_ia32_vpermi2varpd256_mask" | "__builtin_ia32_vpermi2varpd128_mask" | "__builtin_ia32_vpmadd52huq512_mask"
| "__builtin_ia32_vpmadd52luq512_mask" | "__builtin_ia32_vpmadd52huq256_mask" | "__builtin_ia32_vpmadd52luq256_mask"
| "__builtin_ia32_vpmadd52huq128_mask"
=> {
let mut new_args = args.to_vec();
let arg4_type = gcc_func.get_param_type(3);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
new_args.push(minus_one);
args = new_args.into();
},
"__builtin_ia32_cvtdq2ps512_mask" | "__builtin_ia32_cvtudq2ps512_mask"
| "__builtin_ia32_sqrtps512_mask" | "__builtin_ia32_sqrtpd512_mask" => {
let mut new_args = args.to_vec();
let last_arg = new_args.pop().expect("last arg");
let arg2_type = gcc_func.get_param_type(1);
let undefined = builder.current_func().new_local(None, arg2_type, "undefined_for_intrinsic").to_rvalue();
new_args.push(undefined);
let arg3_type = gcc_func.get_param_type(2);
let minus_one = builder.context.new_rvalue_from_int(arg3_type, -1);
new_args.push(minus_one);
new_args.push(last_arg);
args = new_args.into();
},
"__builtin_ia32_stmxcsr" => {
args = vec![].into();
},
"__builtin_ia32_addcarryx_u64" | "__builtin_ia32_sbb_u64" | "__builtin_ia32_addcarryx_u32" | "__builtin_ia32_sbb_u32" => {
let mut new_args = args.to_vec();
let arg2_type = gcc_func.get_param_type(1);
let variable = builder.current_func().new_local(None, arg2_type, "addcarryResult");
new_args.push(variable.get_address(None));
args = new_args.into();
},
"__builtin_ia32_vpermt2varqi512_mask" | "__builtin_ia32_vpermt2varqi256_mask"
| "__builtin_ia32_vpermt2varqi128_mask" | "__builtin_ia32_vpermt2varhi512_mask"
| "__builtin_ia32_vpermt2varhi256_mask" | "__builtin_ia32_vpermt2varhi128_mask"
=> {
let new_args = args.to_vec();
let arg4_type = gcc_func.get_param_type(3);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
args = vec![new_args[1], new_args[0], new_args[2], minus_one].into();
},
"__builtin_ia32_xrstor" | "__builtin_ia32_xsavec" => {
let new_args = args.to_vec();
let thirty_two = builder.context.new_rvalue_from_int(new_args[1].get_type(), 32);
let arg2 = new_args[1] << thirty_two | new_args[2];
let arg2_type = gcc_func.get_param_type(1);
let arg2 = builder.context.new_cast(None, arg2, arg2_type);
args = vec![new_args[0], arg2].into();
},
"__builtin_prefetch" => {
let mut new_args = args.to_vec();
new_args.pop();
args = new_args.into();
},
_ => (),
}
}
else {
match &*func_name {
"__builtin_ia32_rndscaless_mask_round" | "__builtin_ia32_rndscalesd_mask_round" => {
let new_args = args.to_vec();
let arg3_type = gcc_func.get_param_type(2);
let arg3 = builder.context.new_cast(None, new_args[4], arg3_type);
let arg4_type = gcc_func.get_param_type(3);
let arg4 = builder.context.new_bitcast(None, new_args[2], arg4_type);
args = vec![new_args[0], new_args[1], arg3, arg4, new_args[3], new_args[5]].into();
},
// NOTE: the LLVM intrinsic receives 3 floats, but the GCC builtin requires 3 vectors.
// FIXME: the intrinsics like _mm_mask_fmadd_sd should probably directly call the GCC
// instrinsic to avoid this.
"__builtin_ia32_vfmaddss3_round" => {
let new_args = args.to_vec();
let arg1_type = gcc_func.get_param_type(0);
let arg2_type = gcc_func.get_param_type(1);
let arg3_type = gcc_func.get_param_type(2);
let a = builder.context.new_rvalue_from_vector(None, arg1_type, &[new_args[0]; 4]);
let b = builder.context.new_rvalue_from_vector(None, arg2_type, &[new_args[1]; 4]);
let c = builder.context.new_rvalue_from_vector(None, arg3_type, &[new_args[2]; 4]);
args = vec![a, b, c, new_args[3]].into();
},
"__builtin_ia32_vfmaddsd3_round" => {
let new_args = args.to_vec();
let arg1_type = gcc_func.get_param_type(0);
let arg2_type = gcc_func.get_param_type(1);
let arg3_type = gcc_func.get_param_type(2);
let a = builder.context.new_rvalue_from_vector(None, arg1_type, &[new_args[0]; 2]);
let b = builder.context.new_rvalue_from_vector(None, arg2_type, &[new_args[1]; 2]);
let c = builder.context.new_rvalue_from_vector(None, arg3_type, &[new_args[2]; 2]);
args = vec![a, b, c, new_args[3]].into();
},
"__builtin_ia32_vfmaddsubpd256" | "__builtin_ia32_vfmaddsubps" | "__builtin_ia32_vfmaddsubps256"
| "__builtin_ia32_vfmaddsubpd" => {
if let Some(original_function_name) = original_function_name {
match &**original_function_name {
"llvm.x86.fma.vfmsubadd.pd.256" | "llvm.x86.fma.vfmsubadd.ps" | "llvm.x86.fma.vfmsubadd.ps.256"
| "llvm.x86.fma.vfmsubadd.pd" => {
// NOTE: since both llvm.x86.fma.vfmsubadd.ps and llvm.x86.fma.vfmaddsub.ps maps to
// __builtin_ia32_vfmaddsubps, only add minus if this comes from a
// subadd LLVM intrinsic, e.g. _mm256_fmsubadd_pd.
let mut new_args = args.to_vec();
let arg3 = &mut new_args[2];
*arg3 = builder.context.new_unary_op(None, UnaryOp::Minus, arg3.get_type(), *arg3);
args = new_args.into();
},
_ => (),
}
else {
new_args.push(first_arg);
}
let arg4_index =
match &*func_name {
"__builtin_ia32_sqrtps512_mask" | "__builtin_ia32_sqrtpd512_mask" => 2,
_ => 3,
};
let arg4_type = gcc_func.get_param_type(arg4_index);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
if add_before_last_arg {
new_args.insert(new_args.len() - 1, minus_one);
}
else {
new_args.push(minus_one);
}
args = new_args.into();
},
"__builtin_ia32_pternlogd512_mask" | "__builtin_ia32_pternlogd256_mask"
| "__builtin_ia32_pternlogd128_mask" | "__builtin_ia32_pternlogq512_mask"
| "__builtin_ia32_pternlogq256_mask" | "__builtin_ia32_pternlogq128_mask" => {
let mut new_args = args.to_vec();
let arg5_type = gcc_func.get_param_type(4);
let minus_one = builder.context.new_rvalue_from_int(arg5_type, -1);
new_args.push(minus_one);
args = new_args.into();
},
"__builtin_ia32_vfmaddps512_mask" | "__builtin_ia32_vfmaddpd512_mask" => {
let mut new_args = args.to_vec();
let mut last_arg = None;
if args.len() == 4 {
last_arg = new_args.pop();
}
let arg4_type = gcc_func.get_param_type(3);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
new_args.push(minus_one);
if args.len() == 3 {
// Both llvm.fma.v16f32 and llvm.x86.avx512.vfmadd.ps.512 maps to
// the same GCC intrinsic, but the former has 3 parameters and the
// latter has 4 so it doesn't require this additional argument.
let arg5_type = gcc_func.get_param_type(4);
new_args.push(builder.context.new_rvalue_from_int(arg5_type, 4));
}
if let Some(last_arg) = last_arg {
new_args.push(last_arg);
}
args = new_args.into();
},
"__builtin_ia32_addps512_mask" | "__builtin_ia32_addpd512_mask"
| "__builtin_ia32_subps512_mask" | "__builtin_ia32_subpd512_mask"
| "__builtin_ia32_mulps512_mask" | "__builtin_ia32_mulpd512_mask"
| "__builtin_ia32_divps512_mask" | "__builtin_ia32_divpd512_mask" => {
let mut new_args = args.to_vec();
let last_arg = new_args.pop().expect("last arg");
let arg3_type = gcc_func.get_param_type(2);
let undefined = builder.current_func().new_local(None, arg3_type, "undefined_for_intrinsic").to_rvalue();
new_args.push(undefined);
let arg4_type = gcc_func.get_param_type(3);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
new_args.push(minus_one);
new_args.push(last_arg);
args = new_args.into();
},
"__builtin_ia32_vfmaddsubps512_mask" | "__builtin_ia32_vfmaddsubpd512_mask" => {
let mut new_args = args.to_vec();
let last_arg = new_args.pop().expect("last arg");
let arg4_type = gcc_func.get_param_type(3);
let minus_one = builder.context.new_rvalue_from_int(arg4_type, -1);
new_args.push(minus_one);
new_args.push(last_arg);
args = new_args.into();
},
_ => (),
}
},
"__builtin_ia32_ldmxcsr" => {
// The builtin __builtin_ia32_ldmxcsr takes an integer value while llvm.x86.sse.ldmxcsr takes a pointer,
// so dereference the pointer.
let mut new_args = args.to_vec();
let uint_ptr_type = builder.uint_type.make_pointer();
let arg1 = builder.context.new_cast(None, args[0], uint_ptr_type);
new_args[0] = arg1.dereference(None).to_rvalue();
args = new_args.into();
},
"__builtin_ia32_rcp14sd_mask" | "__builtin_ia32_rcp14ss_mask" | "__builtin_ia32_rsqrt14sd_mask"
| "__builtin_ia32_rsqrt14ss_mask" => {
let new_args = args.to_vec();
args = vec![new_args[1], new_args[0], new_args[2], new_args[3]].into();
},
"__builtin_ia32_sqrtsd_mask_round" | "__builtin_ia32_sqrtss_mask_round" => {
let new_args = args.to_vec();
args = vec![new_args[1], new_args[0], new_args[2], new_args[3], new_args[4]].into();
},
_ => (),
}
}
args
}
pub fn adjust_intrinsic_return_value<'a, 'gcc, 'tcx>(builder: &Builder<'a, 'gcc, 'tcx>, mut return_value: RValue<'gcc>, func_name: &str, args: &[RValue<'gcc>], args_adjusted: bool, orig_args: &[RValue<'gcc>]) -> RValue<'gcc> {
match func_name {
"__builtin_ia32_vfmaddss3_round" | "__builtin_ia32_vfmaddsd3_round" => {
#[cfg(feature="master")]
{
let zero = builder.context.new_rvalue_zero(builder.int_type);
return_value = builder.context.new_vector_access(None, return_value, zero).to_rvalue();
}
},
"__builtin_ia32_addcarryx_u64" | "__builtin_ia32_sbb_u64" | "__builtin_ia32_addcarryx_u32" | "__builtin_ia32_sbb_u32" => {
// Both llvm.x86.addcarry.32 and llvm.x86.addcarryx.u32 points to the same GCC builtin,
// but only the former requires adjusting the return value.
// Those 2 LLVM intrinsics differ by their argument count, that's why we check if the
// arguments were adjusted.
if args_adjusted {
let last_arg = args.last().expect("last arg");
let field1 = builder.context.new_field(None, builder.u8_type, "carryFlag");
let field2 = builder.context.new_field(None, args[1].get_type(), "carryResult");
let struct_type = builder.context.new_struct_type(None, "addcarryResult", &[field1, field2]);
return_value = builder.context.new_struct_constructor(None, struct_type.as_type(), None, &[return_value, last_arg.dereference(None).to_rvalue()]);
}
},
"__builtin_ia32_stmxcsr" => {
// The builtin __builtin_ia32_stmxcsr returns a value while llvm.x86.sse.stmxcsr writes
// the result in its pointer argument.
// We removed the argument since __builtin_ia32_stmxcsr takes no arguments, so we need
// to get back the original argument to get the pointer we need to write the result to.
let uint_ptr_type = builder.uint_type.make_pointer();
let ptr = builder.context.new_cast(None, orig_args[0], uint_ptr_type);
builder.llbb().add_assignment(None, ptr.dereference(None), return_value);
// The return value was assigned to the result pointer above. In order to not call the
// builtin twice, we overwrite the return value with a dummy value.
return_value = builder.context.new_rvalue_zero(builder.int_type);
},
_ => (),
}
return_value
}
pub fn ignore_arg_cast(func_name: &str, index: usize, args_len: usize) -> bool {
// NOTE: these intrinsics have missing parameters before the last one, so ignore the
// last argument type check.
// FIXME(antoyo): find a way to refactor in order to avoid this hack.
match func_name {
// NOTE: these intrinsics have missing parameters before the last one, so ignore the
// last argument type check.
"__builtin_ia32_maxps512_mask" | "__builtin_ia32_maxpd512_mask"
| "__builtin_ia32_minps512_mask" | "__builtin_ia32_minpd512_mask" | "__builtin_ia32_sqrtps512_mask"
| "__builtin_ia32_sqrtpd512_mask" | "__builtin_ia32_addps512_mask" | "__builtin_ia32_addpd512_mask"
| "__builtin_ia32_subps512_mask" | "__builtin_ia32_subpd512_mask"
| "__builtin_ia32_mulps512_mask" | "__builtin_ia32_mulpd512_mask"
| "__builtin_ia32_divps512_mask" | "__builtin_ia32_divpd512_mask"
| "__builtin_ia32_vfmaddsubps512_mask" | "__builtin_ia32_vfmaddsubpd512_mask" => {
| "__builtin_ia32_vfmaddsubps512_mask" | "__builtin_ia32_vfmaddsubpd512_mask"
| "__builtin_ia32_cvtdq2ps512_mask" | "__builtin_ia32_cvtudq2ps512_mask" => {
if index == args_len - 1 {
return true;
}
},
"__builtin_ia32_rndscaless_mask_round" | "__builtin_ia32_rndscalesd_mask_round" => {
if index == 2 || index == 3 {
return true;
}
},
"__builtin_ia32_vfmaddps512_mask" | "__builtin_ia32_vfmaddpd512_mask" => {
// Since there are two LLVM intrinsics that map to each of these GCC builtins and only
// one of them has a missing parameter before the last one, we check the number of
@ -162,6 +390,14 @@ pub fn ignore_arg_cast(func_name: &str, index: usize, args_len: usize) -> bool {
return true;
}
},
// NOTE: the LLVM intrinsic receives 3 floats, but the GCC builtin requires 3 vectors.
"__builtin_ia32_vfmaddss3_round" | "__builtin_ia32_vfmaddsd3_round" => return true,
"__builtin_ia32_vplzcntd_512_mask" | "__builtin_ia32_vplzcntd_256_mask" | "__builtin_ia32_vplzcntd_128_mask"
| "__builtin_ia32_vplzcntq_512_mask" | "__builtin_ia32_vplzcntq_256_mask" | "__builtin_ia32_vplzcntq_128_mask" => {
if index == args_len - 1 {
return true;
}
},
_ => (),
}
@ -171,7 +407,7 @@ pub fn ignore_arg_cast(func_name: &str, index: usize, args_len: usize) -> bool {
#[cfg(not(feature="master"))]
pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function<'gcc> {
match name {
"llvm.x86.xgetbv" => {
"llvm.x86.xgetbv" | "llvm.x86.sse2.pause" => {
let gcc_name = "__builtin_trap";
let func = cx.context.get_builtin_function(gcc_name);
cx.functions.borrow_mut().insert(gcc_name.to_string(), func);
@ -183,24 +419,26 @@ pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function
#[cfg(feature="master")]
pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function<'gcc> {
match name {
"llvm.prefetch" => {
let gcc_name = "__builtin_prefetch";
let func = cx.context.get_builtin_function(gcc_name);
cx.functions.borrow_mut().insert(gcc_name.to_string(), func);
return func
},
_ => (),
}
let gcc_name = match name {
"llvm.x86.xgetbv" => "__builtin_ia32_xgetbv",
// NOTE: this doc specifies the equivalent GCC builtins: http://huonw.github.io/llvmint/llvmint/x86/index.html
"llvm.sqrt.v2f64" => "__builtin_ia32_sqrtpd",
"llvm.x86.avx512.pmul.dq.512" => "__builtin_ia32_pmuldq512_mask",
"llvm.x86.avx512.pmulu.dq.512" => "__builtin_ia32_pmuludq512_mask",
"llvm.x86.avx512.mask.pmaxs.q.256" => "__builtin_ia32_pmaxsq256_mask",
"llvm.x86.avx512.mask.pmaxs.q.128" => "__builtin_ia32_pmaxsq128_mask",
"llvm.x86.avx512.max.ps.512" => "__builtin_ia32_maxps512_mask",
"llvm.x86.avx512.max.pd.512" => "__builtin_ia32_maxpd512_mask",
"llvm.x86.avx512.mask.pmaxu.q.256" => "__builtin_ia32_pmaxuq256_mask",
"llvm.x86.avx512.mask.pmaxu.q.128" => "__builtin_ia32_pmaxuq128_mask",
"llvm.x86.avx512.mask.pmins.q.256" => "__builtin_ia32_pminsq256_mask",
"llvm.x86.avx512.mask.pmins.q.128" => "__builtin_ia32_pminsq128_mask",
"llvm.x86.avx512.min.ps.512" => "__builtin_ia32_minps512_mask",
"llvm.x86.avx512.min.pd.512" => "__builtin_ia32_minpd512_mask",
"llvm.x86.avx512.mask.pminu.q.256" => "__builtin_ia32_pminuq256_mask",
"llvm.x86.avx512.mask.pminu.q.128" => "__builtin_ia32_pminuq128_mask",
"llvm.fma.v16f32" => "__builtin_ia32_vfmaddps512_mask",
"llvm.fma.v8f64" => "__builtin_ia32_vfmaddpd512_mask",
"llvm.x86.avx512.vfmaddsub.ps.512" => "__builtin_ia32_vfmaddsubps512_mask",
@ -221,6 +459,153 @@ pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function
"llvm.x86.avx512.div.pd.512" => "__builtin_ia32_divpd512_mask",
"llvm.x86.avx512.vfmadd.ps.512" => "__builtin_ia32_vfmaddps512_mask",
"llvm.x86.avx512.vfmadd.pd.512" => "__builtin_ia32_vfmaddpd512_mask",
"llvm.x86.avx512.sitofp.round.v16f32.v16i32" => "__builtin_ia32_cvtdq2ps512_mask",
"llvm.x86.avx512.uitofp.round.v16f32.v16i32" => "__builtin_ia32_cvtudq2ps512_mask",
"llvm.x86.avx512.mask.ucmp.d.512" => "__builtin_ia32_ucmpd512_mask",
"llvm.x86.avx512.mask.ucmp.d.256" => "__builtin_ia32_ucmpd256_mask",
"llvm.x86.avx512.mask.ucmp.d.128" => "__builtin_ia32_ucmpd128_mask",
"llvm.x86.avx512.mask.cmp.d.512" => "__builtin_ia32_cmpd512_mask",
"llvm.x86.avx512.mask.cmp.d.256" => "__builtin_ia32_cmpd256_mask",
"llvm.x86.avx512.mask.cmp.d.128" => "__builtin_ia32_cmpd128_mask",
"llvm.x86.avx512.mask.ucmp.q.512" => "__builtin_ia32_ucmpq512_mask",
"llvm.x86.avx512.mask.ucmp.q.256" => "__builtin_ia32_ucmpq256_mask",
"llvm.x86.avx512.mask.ucmp.q.128" => "__builtin_ia32_ucmpq128_mask",
"llvm.x86.avx512.mask.cmp.q.512" => "__builtin_ia32_cmpq512_mask",
"llvm.x86.avx512.mask.cmp.q.256" => "__builtin_ia32_cmpq256_mask",
"llvm.x86.avx512.mask.cmp.q.128" => "__builtin_ia32_cmpq128_mask",
"llvm.x86.avx512.mask.max.ss.round" => "__builtin_ia32_maxss_mask_round",
"llvm.x86.avx512.mask.max.sd.round" => "__builtin_ia32_maxsd_mask_round",
"llvm.x86.avx512.mask.min.ss.round" => "__builtin_ia32_minss_mask_round",
"llvm.x86.avx512.mask.min.sd.round" => "__builtin_ia32_minsd_mask_round",
"llvm.x86.avx512.mask.sqrt.ss" => "__builtin_ia32_sqrtss_mask_round",
"llvm.x86.avx512.mask.sqrt.sd" => "__builtin_ia32_sqrtsd_mask_round",
"llvm.x86.avx512.mask.getexp.ss" => "__builtin_ia32_getexpss_mask_round",
"llvm.x86.avx512.mask.getexp.sd" => "__builtin_ia32_getexpsd_mask_round",
"llvm.x86.avx512.mask.getmant.ss" => "__builtin_ia32_getmantss_mask_round",
"llvm.x86.avx512.mask.getmant.sd" => "__builtin_ia32_getmantsd_mask_round",
"llvm.x86.avx512.mask.rndscale.ss" => "__builtin_ia32_rndscaless_mask_round",
"llvm.x86.avx512.mask.rndscale.sd" => "__builtin_ia32_rndscalesd_mask_round",
"llvm.x86.avx512.mask.scalef.ss" => "__builtin_ia32_scalefss_mask_round",
"llvm.x86.avx512.mask.scalef.sd" => "__builtin_ia32_scalefsd_mask_round",
"llvm.x86.avx512.vfmadd.f32" => "__builtin_ia32_vfmaddss3_round",
"llvm.x86.avx512.vfmadd.f64" => "__builtin_ia32_vfmaddsd3_round",
"llvm.ceil.v4f64" => "__builtin_ia32_ceilpd256",
"llvm.ceil.v8f32" => "__builtin_ia32_ceilps256",
"llvm.floor.v4f64" => "__builtin_ia32_floorpd256",
"llvm.floor.v8f32" => "__builtin_ia32_floorps256",
"llvm.sqrt.v4f64" => "__builtin_ia32_sqrtpd256",
"llvm.x86.sse.stmxcsr" => "__builtin_ia32_stmxcsr",
"llvm.x86.sse.ldmxcsr" => "__builtin_ia32_ldmxcsr",
"llvm.ctpop.v16i32" => "__builtin_ia32_vpopcountd_v16si",
"llvm.ctpop.v8i32" => "__builtin_ia32_vpopcountd_v8si",
"llvm.ctpop.v4i32" => "__builtin_ia32_vpopcountd_v4si",
"llvm.ctpop.v8i64" => "__builtin_ia32_vpopcountq_v8di",
"llvm.ctpop.v4i64" => "__builtin_ia32_vpopcountq_v4di",
"llvm.ctpop.v2i64" => "__builtin_ia32_vpopcountq_v2di",
"llvm.x86.addcarry.64" => "__builtin_ia32_addcarryx_u64",
"llvm.x86.subborrow.64" => "__builtin_ia32_sbb_u64",
"llvm.floor.v2f64" => "__builtin_ia32_floorpd",
"llvm.floor.v4f32" => "__builtin_ia32_floorps",
"llvm.ceil.v2f64" => "__builtin_ia32_ceilpd",
"llvm.ceil.v4f32" => "__builtin_ia32_ceilps",
"llvm.fma.v2f64" => "__builtin_ia32_vfmaddpd",
"llvm.fma.v4f64" => "__builtin_ia32_vfmaddpd256",
"llvm.fma.v4f32" => "__builtin_ia32_vfmaddps",
"llvm.fma.v8f32" => "__builtin_ia32_vfmaddps256",
"llvm.ctlz.v16i32" => "__builtin_ia32_vplzcntd_512_mask",
"llvm.ctlz.v8i32" => "__builtin_ia32_vplzcntd_256_mask",
"llvm.ctlz.v4i32" => "__builtin_ia32_vplzcntd_128_mask",
"llvm.ctlz.v8i64" => "__builtin_ia32_vplzcntq_512_mask",
"llvm.ctlz.v4i64" => "__builtin_ia32_vplzcntq_256_mask",
"llvm.ctlz.v2i64" => "__builtin_ia32_vplzcntq_128_mask",
"llvm.ctpop.v32i16" => "__builtin_ia32_vpopcountw_v32hi",
"llvm.x86.fma.vfmsub.sd" => "__builtin_ia32_vfmsubsd3",
"llvm.x86.fma.vfmsub.ss" => "__builtin_ia32_vfmsubss3",
"llvm.x86.fma.vfmsubadd.pd" => "__builtin_ia32_vfmaddsubpd",
"llvm.x86.fma.vfmsubadd.pd.256" => "__builtin_ia32_vfmaddsubpd256",
"llvm.x86.fma.vfmsubadd.ps" => "__builtin_ia32_vfmaddsubps",
"llvm.x86.fma.vfmsubadd.ps.256" => "__builtin_ia32_vfmaddsubps256",
"llvm.x86.fma.vfnmadd.sd" => "__builtin_ia32_vfnmaddsd3",
"llvm.x86.fma.vfnmadd.ss" => "__builtin_ia32_vfnmaddss3",
"llvm.x86.fma.vfnmsub.sd" => "__builtin_ia32_vfnmsubsd3",
"llvm.x86.fma.vfnmsub.ss" => "__builtin_ia32_vfnmsubss3",
"llvm.x86.avx512.conflict.d.512" => "__builtin_ia32_vpconflictsi_512_mask",
"llvm.x86.avx512.conflict.d.256" => "__builtin_ia32_vpconflictsi_256_mask",
"llvm.x86.avx512.conflict.d.128" => "__builtin_ia32_vpconflictsi_128_mask",
"llvm.x86.avx512.conflict.q.512" => "__builtin_ia32_vpconflictdi_512_mask",
"llvm.x86.avx512.conflict.q.256" => "__builtin_ia32_vpconflictdi_256_mask",
"llvm.x86.avx512.conflict.q.128" => "__builtin_ia32_vpconflictdi_128_mask",
"llvm.x86.avx512.vpermi2var.qi.512" => "__builtin_ia32_vpermt2varqi512_mask",
"llvm.x86.avx512.vpermi2var.qi.256" => "__builtin_ia32_vpermt2varqi256_mask",
"llvm.x86.avx512.vpermi2var.qi.128" => "__builtin_ia32_vpermt2varqi128_mask",
"llvm.x86.avx512.permvar.qi.512" => "__builtin_ia32_permvarqi512_mask",
"llvm.x86.avx512.permvar.qi.256" => "__builtin_ia32_permvarqi256_mask",
"llvm.x86.avx512.permvar.qi.128" => "__builtin_ia32_permvarqi128_mask",
"llvm.x86.avx512.pmultishift.qb.512" => "__builtin_ia32_vpmultishiftqb512_mask",
"llvm.x86.avx512.pmultishift.qb.256" => "__builtin_ia32_vpmultishiftqb256_mask",
"llvm.x86.avx512.pmultishift.qb.128" => "__builtin_ia32_vpmultishiftqb128_mask",
"llvm.ctpop.v16i16" => "__builtin_ia32_vpopcountw_v16hi",
"llvm.ctpop.v8i16" => "__builtin_ia32_vpopcountw_v8hi",
"llvm.ctpop.v64i8" => "__builtin_ia32_vpopcountb_v64qi",
"llvm.ctpop.v32i8" => "__builtin_ia32_vpopcountb_v32qi",
"llvm.ctpop.v16i8" => "__builtin_ia32_vpopcountb_v16qi",
"llvm.x86.avx512.mask.vpshufbitqmb.512" => "__builtin_ia32_vpshufbitqmb512_mask",
"llvm.x86.avx512.mask.vpshufbitqmb.256" => "__builtin_ia32_vpshufbitqmb256_mask",
"llvm.x86.avx512.mask.vpshufbitqmb.128" => "__builtin_ia32_vpshufbitqmb128_mask",
"llvm.x86.avx512.mask.ucmp.w.512" => "__builtin_ia32_ucmpw512_mask",
"llvm.x86.avx512.mask.ucmp.w.256" => "__builtin_ia32_ucmpw256_mask",
"llvm.x86.avx512.mask.ucmp.w.128" => "__builtin_ia32_ucmpw128_mask",
"llvm.x86.avx512.mask.ucmp.b.512" => "__builtin_ia32_ucmpb512_mask",
"llvm.x86.avx512.mask.ucmp.b.256" => "__builtin_ia32_ucmpb256_mask",
"llvm.x86.avx512.mask.ucmp.b.128" => "__builtin_ia32_ucmpb128_mask",
"llvm.x86.avx512.mask.cmp.w.512" => "__builtin_ia32_cmpw512_mask",
"llvm.x86.avx512.mask.cmp.w.256" => "__builtin_ia32_cmpw256_mask",
"llvm.x86.avx512.mask.cmp.w.128" => "__builtin_ia32_cmpw128_mask",
"llvm.x86.avx512.mask.cmp.b.512" => "__builtin_ia32_cmpb512_mask",
"llvm.x86.avx512.mask.cmp.b.256" => "__builtin_ia32_cmpb256_mask",
"llvm.x86.avx512.mask.cmp.b.128" => "__builtin_ia32_cmpb128_mask",
"llvm.x86.xrstor" => "__builtin_ia32_xrstor",
"llvm.x86.xsavec" => "__builtin_ia32_xsavec",
"llvm.x86.addcarry.32" => "__builtin_ia32_addcarryx_u32",
"llvm.x86.subborrow.32" => "__builtin_ia32_sbb_u32",
"llvm.x86.avx512.mask.compress.store.w.512" => "__builtin_ia32_compressstoreuhi512_mask",
"llvm.x86.avx512.mask.compress.store.w.256" => "__builtin_ia32_compressstoreuhi256_mask",
"llvm.x86.avx512.mask.compress.store.w.128" => "__builtin_ia32_compressstoreuhi128_mask",
"llvm.x86.avx512.mask.compress.store.b.512" => "__builtin_ia32_compressstoreuqi512_mask",
"llvm.x86.avx512.mask.compress.store.b.256" => "__builtin_ia32_compressstoreuqi256_mask",
"llvm.x86.avx512.mask.compress.store.b.128" => "__builtin_ia32_compressstoreuqi128_mask",
"llvm.x86.avx512.mask.compress.w.512" => "__builtin_ia32_compresshi512_mask",
"llvm.x86.avx512.mask.compress.w.256" => "__builtin_ia32_compresshi256_mask",
"llvm.x86.avx512.mask.compress.w.128" => "__builtin_ia32_compresshi128_mask",
"llvm.x86.avx512.mask.compress.b.512" => "__builtin_ia32_compressqi512_mask",
"llvm.x86.avx512.mask.compress.b.256" => "__builtin_ia32_compressqi256_mask",
"llvm.x86.avx512.mask.compress.b.128" => "__builtin_ia32_compressqi128_mask",
"llvm.x86.avx512.mask.expand.w.512" => "__builtin_ia32_expandhi512_mask",
"llvm.x86.avx512.mask.expand.w.256" => "__builtin_ia32_expandhi256_mask",
"llvm.x86.avx512.mask.expand.w.128" => "__builtin_ia32_expandhi128_mask",
"llvm.x86.avx512.mask.expand.b.512" => "__builtin_ia32_expandqi512_mask",
"llvm.x86.avx512.mask.expand.b.256" => "__builtin_ia32_expandqi256_mask",
"llvm.x86.avx512.mask.expand.b.128" => "__builtin_ia32_expandqi128_mask",
"llvm.fshl.v8i64" => "__builtin_ia32_vpshldv_v8di",
"llvm.fshl.v4i64" => "__builtin_ia32_vpshldv_v4di",
"llvm.fshl.v2i64" => "__builtin_ia32_vpshldv_v2di",
"llvm.fshl.v16i32" => "__builtin_ia32_vpshldv_v16si",
"llvm.fshl.v8i32" => "__builtin_ia32_vpshldv_v8si",
"llvm.fshl.v4i32" => "__builtin_ia32_vpshldv_v4si",
"llvm.fshl.v32i16" => "__builtin_ia32_vpshldv_v32hi",
"llvm.fshl.v16i16" => "__builtin_ia32_vpshldv_v16hi",
"llvm.fshl.v8i16" => "__builtin_ia32_vpshldv_v8hi",
"llvm.fshr.v8i64" => "__builtin_ia32_vpshrdv_v8di",
"llvm.fshr.v4i64" => "__builtin_ia32_vpshrdv_v4di",
"llvm.fshr.v2i64" => "__builtin_ia32_vpshrdv_v2di",
"llvm.fshr.v16i32" => "__builtin_ia32_vpshrdv_v16si",
"llvm.fshr.v8i32" => "__builtin_ia32_vpshrdv_v8si",
"llvm.fshr.v4i32" => "__builtin_ia32_vpshrdv_v4si",
"llvm.fshr.v32i16" => "__builtin_ia32_vpshrdv_v32hi",
"llvm.fshr.v16i16" => "__builtin_ia32_vpshrdv_v16hi",
"llvm.fshr.v8i16" => "__builtin_ia32_vpshrdv_v8hi",
"llvm.x86.fma.vfmadd.sd" => "__builtin_ia32_vfmaddsd3",
"llvm.x86.fma.vfmadd.ss" => "__builtin_ia32_vfmaddss3",
// The above doc points to unknown builtins for the following, so override them:
"llvm.x86.avx2.gather.d.d" => "__builtin_ia32_gathersiv4si",
@ -239,7 +624,151 @@ pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function
"llvm.x86.avx2.gather.q.q.256" => "__builtin_ia32_gatherdiv4di",
"llvm.x86.avx2.gather.q.pd" => "__builtin_ia32_gatherdiv2df",
"llvm.x86.avx2.gather.q.pd.256" => "__builtin_ia32_gatherdiv4df",
"" => "",
"llvm.x86.avx512.pslli.d.512" => "__builtin_ia32_pslldi512_mask",
"llvm.x86.avx512.psrli.d.512" => "__builtin_ia32_psrldi512_mask",
"llvm.x86.avx512.pslli.q.512" => "__builtin_ia32_psllqi512_mask",
"llvm.x86.avx512.psrli.q.512" => "__builtin_ia32_psrlqi512_mask",
"llvm.x86.avx512.psll.d.512" => "__builtin_ia32_pslld512_mask",
"llvm.x86.avx512.psrl.d.512" => "__builtin_ia32_psrld512_mask",
"llvm.x86.avx512.psll.q.512" => "__builtin_ia32_psllq512_mask",
"llvm.x86.avx512.psrl.q.512" => "__builtin_ia32_psrlq512_mask",
"llvm.x86.avx512.psra.d.512" => "__builtin_ia32_psrad512_mask",
"llvm.x86.avx512.psra.q.512" => "__builtin_ia32_psraq512_mask",
"llvm.x86.avx512.psra.q.256" => "__builtin_ia32_psraq256_mask",
"llvm.x86.avx512.psra.q.128" => "__builtin_ia32_psraq128_mask",
"llvm.x86.avx512.psrai.d.512" => "__builtin_ia32_psradi512_mask",
"llvm.x86.avx512.psrai.q.512" => "__builtin_ia32_psraqi512_mask",
"llvm.x86.avx512.psrai.q.256" => "__builtin_ia32_psraqi256_mask",
"llvm.x86.avx512.psrai.q.128" => "__builtin_ia32_psraqi128_mask",
"llvm.x86.avx512.psrav.d.512" => "__builtin_ia32_psrav16si_mask",
"llvm.x86.avx512.psrav.q.512" => "__builtin_ia32_psrav8di_mask",
"llvm.x86.avx512.psrav.q.256" => "__builtin_ia32_psravq256_mask",
"llvm.x86.avx512.psrav.q.128" => "__builtin_ia32_psravq128_mask",
"llvm.x86.avx512.psllv.d.512" => "__builtin_ia32_psllv16si_mask",
"llvm.x86.avx512.psrlv.d.512" => "__builtin_ia32_psrlv16si_mask",
"llvm.x86.avx512.psllv.q.512" => "__builtin_ia32_psllv8di_mask",
"llvm.x86.avx512.psrlv.q.512" => "__builtin_ia32_psrlv8di_mask",
"llvm.x86.avx512.permvar.si.512" => "__builtin_ia32_permvarsi512_mask",
"llvm.x86.avx512.vpermilvar.ps.512" => "__builtin_ia32_vpermilvarps512_mask",
"llvm.x86.avx512.vpermilvar.pd.512" => "__builtin_ia32_vpermilvarpd512_mask",
"llvm.x86.avx512.permvar.di.512" => "__builtin_ia32_permvardi512_mask",
"llvm.x86.avx512.permvar.di.256" => "__builtin_ia32_permvardi256_mask",
"llvm.x86.avx512.permvar.sf.512" => "__builtin_ia32_permvarsf512_mask",
"llvm.x86.avx512.permvar.df.512" => "__builtin_ia32_permvardf512_mask",
"llvm.x86.avx512.permvar.df.256" => "__builtin_ia32_permvardf256_mask",
"llvm.x86.avx512.vpermi2var.d.512" => "__builtin_ia32_vpermi2vard512_mask",
"llvm.x86.avx512.vpermi2var.d.256" => "__builtin_ia32_vpermi2vard256_mask",
"llvm.x86.avx512.vpermi2var.d.128" => "__builtin_ia32_vpermi2vard128_mask",
"llvm.x86.avx512.vpermi2var.q.512" => "__builtin_ia32_vpermi2varq512_mask",
"llvm.x86.avx512.vpermi2var.q.256" => "__builtin_ia32_vpermi2varq256_mask",
"llvm.x86.avx512.vpermi2var.q.128" => "__builtin_ia32_vpermi2varq128_mask",
"llvm.x86.avx512.vpermi2var.ps.512" => "__builtin_ia32_vpermi2varps512_mask",
"llvm.x86.avx512.vpermi2var.ps.256" => "__builtin_ia32_vpermi2varps256_mask",
"llvm.x86.avx512.vpermi2var.ps.128" => "__builtin_ia32_vpermi2varps128_mask",
"llvm.x86.avx512.vpermi2var.pd.512" => "__builtin_ia32_vpermi2varpd512_mask",
"llvm.x86.avx512.vpermi2var.pd.256" => "__builtin_ia32_vpermi2varpd256_mask",
"llvm.x86.avx512.vpermi2var.pd.128" => "__builtin_ia32_vpermi2varpd128_mask",
"llvm.x86.avx512.mask.add.ss.round" => "__builtin_ia32_addss_mask_round",
"llvm.x86.avx512.mask.add.sd.round" => "__builtin_ia32_addsd_mask_round",
"llvm.x86.avx512.mask.sub.ss.round" => "__builtin_ia32_subss_mask_round",
"llvm.x86.avx512.mask.sub.sd.round" => "__builtin_ia32_subsd_mask_round",
"llvm.x86.avx512.mask.mul.ss.round" => "__builtin_ia32_mulss_mask_round",
"llvm.x86.avx512.mask.mul.sd.round" => "__builtin_ia32_mulsd_mask_round",
"llvm.x86.avx512.mask.div.ss.round" => "__builtin_ia32_divss_mask_round",
"llvm.x86.avx512.mask.div.sd.round" => "__builtin_ia32_divsd_mask_round",
"llvm.x86.avx512.mask.cvtss2sd.round" => "__builtin_ia32_cvtss2sd_mask_round",
"llvm.x86.avx512.mask.cvtsd2ss.round" => "__builtin_ia32_cvtsd2ss_mask_round",
"llvm.x86.avx512.mask.range.ss" => "__builtin_ia32_rangess128_mask_round",
"llvm.x86.avx512.mask.range.sd" => "__builtin_ia32_rangesd128_mask_round",
"llvm.x86.avx512.rcp28.ss" => "__builtin_ia32_rcp28ss_mask_round",
"llvm.x86.avx512.rcp28.sd" => "__builtin_ia32_rcp28sd_mask_round",
"llvm.x86.avx512.rsqrt28.ss" => "__builtin_ia32_rsqrt28ss_mask_round",
"llvm.x86.avx512.rsqrt28.sd" => "__builtin_ia32_rsqrt28sd_mask_round",
"llvm.x86.avx512fp16.mask.add.sh.round" => "__builtin_ia32_addsh_mask_round",
"llvm.x86.avx512fp16.mask.div.sh.round" => "__builtin_ia32_divsh_mask_round",
"llvm.x86.avx512fp16.mask.getmant.sh" => "__builtin_ia32_getmantsh_mask_round",
"llvm.x86.avx512fp16.mask.max.sh.round" => "__builtin_ia32_maxsh_mask_round",
"llvm.x86.avx512fp16.mask.min.sh.round" => "__builtin_ia32_minsh_mask_round",
"llvm.x86.avx512fp16.mask.mul.sh.round" => "__builtin_ia32_mulsh_mask_round",
"llvm.x86.avx512fp16.mask.rndscale.sh" => "__builtin_ia32_rndscalesh_mask_round",
"llvm.x86.avx512fp16.mask.scalef.sh" => "__builtin_ia32_scalefsh_mask_round",
"llvm.x86.avx512fp16.mask.sub.sh.round" => "__builtin_ia32_subsh_mask_round",
"llvm.x86.avx512fp16.mask.vcvtsd2sh.round" => "__builtin_ia32_vcvtsd2sh_mask_round",
"llvm.x86.avx512fp16.mask.vcvtsh2sd.round" => "__builtin_ia32_vcvtsh2sd_mask_round",
"llvm.x86.avx512fp16.mask.vcvtsh2ss.round" => "__builtin_ia32_vcvtsh2ss_mask_round",
"llvm.x86.avx512fp16.mask.vcvtss2sh.round" => "__builtin_ia32_vcvtss2sh_mask_round",
"llvm.x86.aesni.aesenc.256" => "__builtin_ia32_vaesenc_v32qi",
"llvm.x86.aesni.aesenclast.256" => "__builtin_ia32_vaesenclast_v32qi",
"llvm.x86.aesni.aesdec.256" => "__builtin_ia32_vaesdec_v32qi",
"llvm.x86.aesni.aesdeclast.256" => "__builtin_ia32_vaesdeclast_v32qi",
"llvm.x86.aesni.aesenc.512" => "__builtin_ia32_vaesenc_v64qi",
"llvm.x86.aesni.aesenclast.512" => "__builtin_ia32_vaesenclast_v64qi",
"llvm.x86.aesni.aesdec.512" => "__builtin_ia32_vaesdec_v64qi",
"llvm.x86.aesni.aesdeclast.512" => "__builtin_ia32_vaesdeclast_v64qi",
"llvm.x86.avx512bf16.cvtne2ps2bf16.128" => "__builtin_ia32_cvtne2ps2bf16_v8bf",
"llvm.x86.avx512bf16.cvtne2ps2bf16.256" => "__builtin_ia32_cvtne2ps2bf16_v16bf",
"llvm.x86.avx512bf16.cvtne2ps2bf16.512" => "__builtin_ia32_cvtne2ps2bf16_v32bf",
"llvm.x86.avx512bf16.cvtneps2bf16.256" => "__builtin_ia32_cvtneps2bf16_v8sf",
"llvm.x86.avx512bf16.cvtneps2bf16.512" => "__builtin_ia32_cvtneps2bf16_v16sf",
"llvm.x86.avx512bf16.dpbf16ps.128" => "__builtin_ia32_dpbf16ps_v4sf",
"llvm.x86.avx512bf16.dpbf16ps.256" => "__builtin_ia32_dpbf16ps_v8sf",
"llvm.x86.avx512bf16.dpbf16ps.512" => "__builtin_ia32_dpbf16ps_v16sf",
"llvm.x86.pclmulqdq.512" => "__builtin_ia32_vpclmulqdq_v8di",
"llvm.x86.pclmulqdq.256" => "__builtin_ia32_vpclmulqdq_v4di",
"llvm.x86.avx512.pmulhu.w.512" => "__builtin_ia32_pmulhuw512_mask",
"llvm.x86.avx512.pmulh.w.512" => "__builtin_ia32_pmulhw512_mask",
"llvm.x86.avx512.pmul.hr.sw.512" => "__builtin_ia32_pmulhrsw512_mask",
"llvm.x86.avx512.pmaddw.d.512" => "__builtin_ia32_pmaddwd512_mask",
"llvm.x86.avx512.pmaddubs.w.512" => "__builtin_ia32_pmaddubsw512_mask",
"llvm.x86.avx512.packssdw.512" => "__builtin_ia32_packssdw512_mask",
"llvm.x86.avx512.packsswb.512" => "__builtin_ia32_packsswb512_mask",
"llvm.x86.avx512.packusdw.512" => "__builtin_ia32_packusdw512_mask",
"llvm.x86.avx512.packuswb.512" => "__builtin_ia32_packuswb512_mask",
"llvm.x86.avx512.pavg.w.512" => "__builtin_ia32_pavgw512_mask",
"llvm.x86.avx512.pavg.b.512" => "__builtin_ia32_pavgb512_mask",
"llvm.x86.avx512.psll.w.512" => "__builtin_ia32_psllw512_mask",
"llvm.x86.avx512.pslli.w.512" => "__builtin_ia32_psllwi512_mask",
"llvm.x86.avx512.psllv.w.512" => "__builtin_ia32_psllv32hi_mask",
"llvm.x86.avx512.psllv.w.256" => "__builtin_ia32_psllv16hi_mask",
"llvm.x86.avx512.psllv.w.128" => "__builtin_ia32_psllv8hi_mask",
"llvm.x86.avx512.psrl.w.512" => "__builtin_ia32_psrlw512_mask",
"llvm.x86.avx512.psrli.w.512" => "__builtin_ia32_psrlwi512_mask",
"llvm.x86.avx512.psrlv.w.512" => "__builtin_ia32_psrlv32hi_mask",
"llvm.x86.avx512.psrlv.w.256" => "__builtin_ia32_psrlv16hi_mask",
"llvm.x86.avx512.psrlv.w.128" => "__builtin_ia32_psrlv8hi_mask",
"llvm.x86.avx512.psra.w.512" => "__builtin_ia32_psraw512_mask",
"llvm.x86.avx512.psrai.w.512" => "__builtin_ia32_psrawi512_mask",
"llvm.x86.avx512.psrav.w.512" => "__builtin_ia32_psrav32hi_mask",
"llvm.x86.avx512.psrav.w.256" => "__builtin_ia32_psrav16hi_mask",
"llvm.x86.avx512.psrav.w.128" => "__builtin_ia32_psrav8hi_mask",
"llvm.x86.avx512.vpermi2var.hi.512" => "__builtin_ia32_vpermt2varhi512_mask",
"llvm.x86.avx512.vpermi2var.hi.256" => "__builtin_ia32_vpermt2varhi256_mask",
"llvm.x86.avx512.vpermi2var.hi.128" => "__builtin_ia32_vpermt2varhi128_mask",
"llvm.x86.avx512.permvar.hi.512" => "__builtin_ia32_permvarhi512_mask",
"llvm.x86.avx512.permvar.hi.256" => "__builtin_ia32_permvarhi256_mask",
"llvm.x86.avx512.permvar.hi.128" => "__builtin_ia32_permvarhi128_mask",
"llvm.x86.avx512.pshuf.b.512" => "__builtin_ia32_pshufb512_mask",
"llvm.x86.avx512.dbpsadbw.512" => "__builtin_ia32_dbpsadbw512_mask",
"llvm.x86.avx512.dbpsadbw.256" => "__builtin_ia32_dbpsadbw256_mask",
"llvm.x86.avx512.dbpsadbw.128" => "__builtin_ia32_dbpsadbw128_mask",
"llvm.x86.avx512.vpmadd52h.uq.512" => "__builtin_ia32_vpmadd52huq512_mask",
"llvm.x86.avx512.vpmadd52l.uq.512" => "__builtin_ia32_vpmadd52luq512_mask",
"llvm.x86.avx512.vpmadd52h.uq.256" => "__builtin_ia32_vpmadd52huq256_mask",
"llvm.x86.avx512.vpmadd52l.uq.256" => "__builtin_ia32_vpmadd52luq256_mask",
"llvm.x86.avx512.vpmadd52h.uq.128" => "__builtin_ia32_vpmadd52huq128_mask",
"llvm.x86.avx512.vpdpwssd.512" => "__builtin_ia32_vpdpwssd_v16si",
"llvm.x86.avx512.vpdpwssd.256" => "__builtin_ia32_vpdpwssd_v8si",
"llvm.x86.avx512.vpdpwssd.128" => "__builtin_ia32_vpdpwssd_v4si",
"llvm.x86.avx512.vpdpwssds.512" => "__builtin_ia32_vpdpwssds_v16si",
"llvm.x86.avx512.vpdpwssds.256" => "__builtin_ia32_vpdpwssds_v8si",
"llvm.x86.avx512.vpdpwssds.128" => "__builtin_ia32_vpdpwssds_v4si",
"llvm.x86.avx512.vpdpbusd.512" => "__builtin_ia32_vpdpbusd_v16si",
"llvm.x86.avx512.vpdpbusd.256" => "__builtin_ia32_vpdpbusd_v8si",
"llvm.x86.avx512.vpdpbusd.128" => "__builtin_ia32_vpdpbusd_v4si",
"llvm.x86.avx512.vpdpbusds.512" => "__builtin_ia32_vpdpbusds_v16si",
"llvm.x86.avx512.vpdpbusds.256" => "__builtin_ia32_vpdpbusds_v8si",
"llvm.x86.avx512.vpdpbusds.128" => "__builtin_ia32_vpdpbusds_v4si",
// NOTE: this file is generated by https://github.com/GuillaumeGomez/llvmint/blob/master/generate_list.py
_ => include!("archs.rs"),
};

View File

@ -1,6 +1,9 @@
pub mod llvm;
mod simd;
#[cfg(feature="master")]
use std::iter;
use gccjit::{ComparisonOp, Function, RValue, ToRValue, Type, UnaryOp, FunctionType};
use rustc_codegen_ssa::MemFlags;
use rustc_codegen_ssa::base::wants_msvc_seh;
@ -8,15 +11,23 @@ use rustc_codegen_ssa::common::IntPredicate;
use rustc_codegen_ssa::mir::operand::{OperandRef, OperandValue};
use rustc_codegen_ssa::mir::place::PlaceRef;
use rustc_codegen_ssa::traits::{ArgAbiMethods, BaseTypeMethods, BuilderMethods, ConstMethods, IntrinsicCallMethods};
#[cfg(feature="master")]
use rustc_codegen_ssa::traits::{DerivedTypeMethods, MiscMethods};
use rustc_middle::bug;
use rustc_middle::ty::{self, Instance, Ty};
use rustc_middle::ty::layout::LayoutOf;
#[cfg(feature="master")]
use rustc_middle::ty::layout::{FnAbiOf, HasTyCtxt};
use rustc_span::{Span, Symbol, symbol::kw, sym};
use rustc_target::abi::HasDataLayout;
use rustc_target::abi::call::{ArgAbi, FnAbi, PassMode};
use rustc_target::spec::PanicStrategy;
#[cfg(feature="master")]
use rustc_target::spec::abi::Abi;
use crate::abi::GccType;
#[cfg(feature="master")]
use crate::abi::FnAbiGccExt;
use crate::builder::Builder;
use crate::common::{SignType, TypeReflection};
use crate::context::CodegenCx;
@ -68,6 +79,8 @@ fn get_simple_intrinsic<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, name: Symbol) ->
sym::nearbyintf64 => "nearbyint",
sym::roundf32 => "roundf",
sym::roundf64 => "round",
sym::roundevenf32 => "roundevenf",
sym::roundevenf64 => "roundeven",
sym::abort => "abort",
_ => return None,
};
@ -91,7 +104,7 @@ impl<'a, 'gcc, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'a, 'gcc, 'tcx> {
let name = tcx.item_name(def_id);
let name_str = name.as_str();
let llret_ty = self.layout_of(ret_ty).gcc_type(self, true);
let llret_ty = self.layout_of(ret_ty).gcc_type(self);
let result = PlaceRef::new_sized(llresult, fn_abi.ret.layout);
let simple = get_simple_intrinsic(self, name);
@ -404,7 +417,7 @@ impl<'gcc, 'tcx> ArgAbiExt<'gcc, 'tcx> for ArgAbi<'tcx, Ty<'tcx>> {
/// Gets the LLVM type for a place of the original Rust type of
/// this argument/return, i.e., the result of `type_of::type_of`.
fn memory_ty(&self, cx: &CodegenCx<'gcc, 'tcx>) -> Type<'gcc> {
self.layout.gcc_type(cx, true)
self.layout.gcc_type(cx)
}
/// Stores a direct/indirect value described by this ArgAbi into a
@ -1120,10 +1133,8 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
}
}
fn try_intrinsic<'gcc, 'tcx>(bx: &mut Builder<'_, 'gcc, 'tcx>, try_func: RValue<'gcc>, data: RValue<'gcc>, _catch_func: RValue<'gcc>, dest: RValue<'gcc>) {
// NOTE: the `|| true` here is to use the panic=abort strategy with panic=unwind too
if bx.sess().panic_strategy() == PanicStrategy::Abort || true {
// TODO(bjorn3): Properly implement unwinding and remove the `|| true` once this is done.
fn try_intrinsic<'a, 'b, 'gcc, 'tcx>(bx: &'b mut Builder<'a, 'gcc, 'tcx>, try_func: RValue<'gcc>, data: RValue<'gcc>, _catch_func: RValue<'gcc>, dest: RValue<'gcc>) {
if bx.sess().panic_strategy() == PanicStrategy::Abort {
bx.call(bx.type_void(), None, try_func, &[data], None);
// Return 0 unconditionally from the intrinsic call;
// we can never unwind.
@ -1134,6 +1145,141 @@ fn try_intrinsic<'gcc, 'tcx>(bx: &mut Builder<'_, 'gcc, 'tcx>, try_func: RValue<
unimplemented!();
}
else {
#[cfg(feature="master")]
codegen_gnu_try(bx, try_func, data, _catch_func, dest);
#[cfg(not(feature="master"))]
unimplemented!();
}
}
// Definition of the standard `try` function for Rust using the GNU-like model
// of exceptions (e.g., the normal semantics of LLVM's `landingpad` and `invoke`
// instructions).
//
// This codegen is a little surprising because we always call a shim
// function instead of inlining the call to `invoke` manually here. This is done
// because in LLVM we're only allowed to have one personality per function
// definition. The call to the `try` intrinsic is being inlined into the
// function calling it, and that function may already have other personality
// functions in play. By calling a shim we're guaranteed that our shim will have
// the right personality function.
#[cfg(feature="master")]
fn codegen_gnu_try<'gcc>(bx: &mut Builder<'_, 'gcc, '_>, try_func: RValue<'gcc>, data: RValue<'gcc>, catch_func: RValue<'gcc>, dest: RValue<'gcc>) {
let cx: &CodegenCx<'gcc, '_> = bx.cx;
let (llty, func) = get_rust_try_fn(cx, &mut |mut bx| {
// Codegens the shims described above:
//
// bx:
// invoke %try_func(%data) normal %normal unwind %catch
//
// normal:
// ret 0
//
// catch:
// (%ptr, _) = landingpad
// call %catch_func(%data, %ptr)
// ret 1
let then = bx.append_sibling_block("then");
let catch = bx.append_sibling_block("catch");
let func = bx.current_func();
let try_func = func.get_param(0).to_rvalue();
let data = func.get_param(1).to_rvalue();
let catch_func = func.get_param(2).to_rvalue();
let try_func_ty = bx.type_func(&[bx.type_i8p()], bx.type_void());
let current_block = bx.block.clone();
bx.switch_to_block(then);
bx.ret(bx.const_i32(0));
// Type indicator for the exception being thrown.
//
// The value is a pointer to the exception object
// being thrown.
bx.switch_to_block(catch);
bx.set_personality_fn(bx.eh_personality());
let eh_pointer_builtin = bx.cx.context.get_target_builtin_function("__builtin_eh_pointer");
let zero = bx.cx.context.new_rvalue_zero(bx.int_type);
let ptr = bx.cx.context.new_call(None, eh_pointer_builtin, &[zero]);
let catch_ty = bx.type_func(&[bx.type_i8p(), bx.type_i8p()], bx.type_void());
bx.call(catch_ty, None, catch_func, &[data, ptr], None);
bx.ret(bx.const_i32(1));
// NOTE: the blocks must be filled before adding the try/catch, otherwise gcc will not
// generate a try/catch.
// FIXME(antoyo): add a check in the libgccjit API to prevent this.
bx.switch_to_block(current_block);
bx.invoke(try_func_ty, None, try_func, &[data], then, catch, None);
});
let func = unsafe { std::mem::transmute(func) };
// Note that no invoke is used here because by definition this function
// can't panic (that's what it's catching).
let ret = bx.call(llty, None, func, &[try_func, data, catch_func], None);
let i32_align = bx.tcx().data_layout.i32_align.abi;
bx.store(ret, dest, i32_align);
}
// Helper function used to get a handle to the `__rust_try` function used to
// catch exceptions.
//
// This function is only generated once and is then cached.
#[cfg(feature="master")]
fn get_rust_try_fn<'a, 'gcc, 'tcx>(cx: &'a CodegenCx<'gcc, 'tcx>, codegen: &mut dyn FnMut(Builder<'a, 'gcc, 'tcx>)) -> (Type<'gcc>, Function<'gcc>) {
if let Some(llfn) = cx.rust_try_fn.get() {
return llfn;
}
// Define the type up front for the signature of the rust_try function.
let tcx = cx.tcx;
let i8p = tcx.mk_mut_ptr(tcx.types.i8);
// `unsafe fn(*mut i8) -> ()`
let try_fn_ty = tcx.mk_fn_ptr(ty::Binder::dummy(tcx.mk_fn_sig(
iter::once(i8p),
tcx.mk_unit(),
false,
rustc_hir::Unsafety::Unsafe,
Abi::Rust,
)));
// `unsafe fn(*mut i8, *mut i8) -> ()`
let catch_fn_ty = tcx.mk_fn_ptr(ty::Binder::dummy(tcx.mk_fn_sig(
[i8p, i8p].iter().cloned(),
tcx.mk_unit(),
false,
rustc_hir::Unsafety::Unsafe,
Abi::Rust,
)));
// `unsafe fn(unsafe fn(*mut i8) -> (), *mut i8, unsafe fn(*mut i8, *mut i8) -> ()) -> i32`
let rust_fn_sig = ty::Binder::dummy(cx.tcx.mk_fn_sig(
[try_fn_ty, i8p, catch_fn_ty],
tcx.types.i32,
false,
rustc_hir::Unsafety::Unsafe,
Abi::Rust,
));
let rust_try = gen_fn(cx, "__rust_try", rust_fn_sig, codegen);
cx.rust_try_fn.set(Some(rust_try));
rust_try
}
// Helper function to give a Block to a closure to codegen a shim function.
// This is currently primarily used for the `try` intrinsic functions above.
#[cfg(feature="master")]
fn gen_fn<'a, 'gcc, 'tcx>(cx: &'a CodegenCx<'gcc, 'tcx>, name: &str, rust_fn_sig: ty::PolyFnSig<'tcx>, codegen: &mut dyn FnMut(Builder<'a, 'gcc, 'tcx>)) -> (Type<'gcc>, Function<'gcc>) {
let fn_abi = cx.fn_abi_of_fn_ptr(rust_fn_sig, ty::List::empty());
let (typ, _, _, _) = fn_abi.gcc_type(cx);
// FIXME(eddyb) find a nicer way to do this.
cx.linkage.set(FunctionType::Internal);
let func = cx.declare_fn(name, fn_abi);
let func_val = unsafe { std::mem::transmute(func) };
cx.set_frame_pointer_type(func_val);
cx.apply_target_cpu_attr(func_val);
let block = Builder::append_block(cx, func_val, "entry-block");
let bx = Builder::build(cx, block);
codegen(bx);
(typ, func)
}

View File

@ -1,8 +1,13 @@
use std::cmp::Ordering;
#[cfg(feature="master")]
use gccjit::{ComparisonOp, UnaryOp};
use gccjit::ToRValue;
use gccjit::{BinaryOp, RValue, Type};
use gccjit::{BinaryOp, RValue, ToRValue, Type};
use rustc_codegen_ssa::base::compare_simd_types;
use rustc_codegen_ssa::common::TypeKind;
use rustc_codegen_ssa::common::{IntPredicate, TypeKind};
#[cfg(feature="master")]
use rustc_codegen_ssa::errors::ExpectedPointerMutability;
use rustc_codegen_ssa::errors::InvalidMonomorphization;
use rustc_codegen_ssa::mir::operand::OperandRef;
use rustc_codegen_ssa::mir::place::PlaceRef;
use rustc_codegen_ssa::traits::{BaseTypeMethods, BuilderMethods};
@ -14,18 +19,21 @@ use rustc_span::{sym, Span, Symbol};
use rustc_target::abi::Align;
use crate::builder::Builder;
#[cfg(feature="master")]
use crate::context::CodegenCx;
#[cfg(feature="master")]
use crate::errors::{InvalidMonomorphizationExpectedSignedUnsigned, InvalidMonomorphizationInsertedType};
use crate::errors::{
InvalidMonomorphizationExpectedSignedUnsigned, InvalidMonomorphizationExpectedSimd,
InvalidMonomorphizationInsertedType, InvalidMonomorphizationInvalidBitmask,
InvalidMonomorphizationExpectedSimd,
InvalidMonomorphizationInvalidBitmask,
InvalidMonomorphizationInvalidFloatVector, InvalidMonomorphizationMaskType,
InvalidMonomorphizationMismatchedLengths, InvalidMonomorphizationNotFloat,
InvalidMonomorphizationReturnElement, InvalidMonomorphizationReturnIntegerType,
InvalidMonomorphizationReturnLength, InvalidMonomorphizationReturnLengthInputType,
InvalidMonomorphizationReturnType, InvalidMonomorphizationSimdShuffle,
InvalidMonomorphizationUnrecognized, InvalidMonomorphizationUnsupportedCast,
InvalidMonomorphizationUnsupportedElement, InvalidMonomorphizationUnsupportedOperation,
InvalidMonomorphizationUnrecognized, InvalidMonomorphizationUnsupportedElement,
InvalidMonomorphizationUnsupportedOperation,
};
use crate::intrinsic;
pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
bx: &mut Builder<'a, 'gcc, 'tcx>,
@ -105,14 +113,19 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
let arg1_vector_type = arg1_type.unqualified().dyncast_vector().expect("vector type");
let arg1_element_type = arg1_vector_type.get_element_type();
// NOTE: since the arguments can be vectors of floats, make sure the mask is a vector of
// integer.
let mask_element_type = bx.type_ix(arg1_element_type.get_size() as u64 * 8);
let vector_mask_type = bx.context.new_vector_type(mask_element_type, arg1_vector_type.get_num_units() as u64);
let mut elements = vec![];
let one = bx.context.new_rvalue_one(mask.get_type());
for _ in 0..len {
let element = bx.context.new_cast(None, mask & one, arg1_element_type);
let element = bx.context.new_cast(None, mask & one, mask_element_type);
elements.push(element);
mask = mask >> one;
}
let vector_mask = bx.context.new_rvalue_from_vector(None, arg1_type, &elements);
let vector_mask = bx.context.new_rvalue_from_vector(None, vector_mask_type, &elements);
return Ok(bx.vector_select(vector_mask, arg1, args[2].immediate()));
}
@ -210,48 +223,12 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
let vector = args[0].immediate();
let index = args[1].immediate();
let value = args[2].immediate();
// TODO(antoyo): use a recursive unqualified() here.
let vector_type = vector.get_type().unqualified().dyncast_vector().expect("vector type");
let element_type = vector_type.get_element_type();
// NOTE: we cannot cast to an array and assign to its element here because the value might
// not be an l-value. So, call a builtin to set the element.
// TODO(antoyo): perhaps we could create a new vector or maybe there's a GIMPLE instruction for that?
// TODO(antoyo): don't use target specific builtins here.
let func_name = match in_len {
2 => {
if element_type == bx.i64_type {
"__builtin_ia32_vec_set_v2di"
} else {
unimplemented!();
}
}
4 => {
if element_type == bx.i32_type {
"__builtin_ia32_vec_set_v4si"
} else {
unimplemented!();
}
}
8 => {
if element_type == bx.i16_type {
"__builtin_ia32_vec_set_v8hi"
} else {
unimplemented!();
}
}
_ => unimplemented!("Len: {}", in_len),
};
let builtin = bx.context.get_target_builtin_function(func_name);
let param1_type = builtin.get_param(0).to_rvalue().get_type();
// TODO(antoyo): perhaps use __builtin_convertvector for vector casting.
let vector = bx.cx.bitcast_if_needed(vector, param1_type);
let result = bx.context.new_call(
None,
builtin,
&[vector, value, bx.context.new_cast(None, index, bx.int_type)],
);
// TODO(antoyo): perhaps use __builtin_convertvector for vector casting.
return Ok(bx.context.new_bitcast(None, result, vector.get_type()));
let variable = bx.current_func().new_local(None, vector.get_type(), "new_vector");
bx.llbb().add_assignment(None, variable, vector);
let lvalue = bx.context.new_vector_access(None, variable.to_rvalue(), index);
// TODO(antoyo): if simd_insert is constant, use BIT_REF.
bx.llbb().add_assignment(None, lvalue, value);
return Ok(variable.to_rvalue());
}
#[cfg(feature = "master")]
@ -280,7 +257,8 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
return Ok(bx.vector_select(args[0].immediate(), args[1].immediate(), args[2].immediate()));
}
if name == sym::simd_cast {
#[cfg(feature="master")]
if name == sym::simd_cast || name == sym::simd_as {
require_simd!(ret_ty, "return");
let (out_len, out_elem) = ret_ty.simd_size_and_type(bx.tcx());
require!(
@ -301,125 +279,40 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
enum Style {
Float,
Int(/* is signed? */ bool),
Int,
Unsupported,
}
let (in_style, in_width) = match in_elem.kind() {
// vectors of pointer-sized integers should've been
// disallowed before here, so this unwrap is safe.
ty::Int(i) => (
Style::Int(true),
i.normalize(bx.tcx().sess.target.pointer_width).bit_width().unwrap(),
),
ty::Uint(u) => (
Style::Int(false),
u.normalize(bx.tcx().sess.target.pointer_width).bit_width().unwrap(),
),
ty::Float(f) => (Style::Float, f.bit_width()),
_ => (Style::Unsupported, 0),
};
let (out_style, out_width) = match out_elem.kind() {
ty::Int(i) => (
Style::Int(true),
i.normalize(bx.tcx().sess.target.pointer_width).bit_width().unwrap(),
),
ty::Uint(u) => (
Style::Int(false),
u.normalize(bx.tcx().sess.target.pointer_width).bit_width().unwrap(),
),
ty::Float(f) => (Style::Float, f.bit_width()),
_ => (Style::Unsupported, 0),
};
let extend = |in_type, out_type| {
let vector_type = bx.context.new_vector_type(out_type, 8);
let vector = args[0].immediate();
let array_type = bx.context.new_array_type(None, in_type, 8);
// TODO(antoyo): switch to using new_vector_access or __builtin_convertvector for vector casting.
let array = bx.context.new_bitcast(None, vector, array_type);
let cast_vec_element = |index| {
let index = bx.context.new_rvalue_from_int(bx.int_type, index);
bx.context.new_cast(
None,
bx.context.new_array_access(None, array, index).to_rvalue(),
out_type,
)
let in_style =
match in_elem.kind() {
ty::Int(_) | ty::Uint(_) => Style::Int,
ty::Float(_) => Style::Float,
_ => Style::Unsupported,
};
bx.context.new_rvalue_from_vector(
None,
vector_type,
&[
cast_vec_element(0),
cast_vec_element(1),
cast_vec_element(2),
cast_vec_element(3),
cast_vec_element(4),
cast_vec_element(5),
cast_vec_element(6),
cast_vec_element(7),
],
)
};
let out_style =
match out_elem.kind() {
ty::Int(_) | ty::Uint(_) => Style::Int,
ty::Float(_) => Style::Float,
_ => Style::Unsupported,
};
match (in_style, out_style) {
(Style::Int(in_is_signed), Style::Int(_)) => {
return Ok(match in_width.cmp(&out_width) {
Ordering::Greater => bx.trunc(args[0].immediate(), llret_ty),
Ordering::Equal => args[0].immediate(),
Ordering::Less => {
if in_is_signed {
match (in_width, out_width) {
// FIXME(antoyo): the function _mm_cvtepi8_epi16 should directly
// call an intrinsic equivalent to __builtin_ia32_pmovsxbw128 so that
// we can generate a call to it.
(8, 16) => extend(bx.i8_type, bx.i16_type),
(8, 32) => extend(bx.i8_type, bx.i32_type),
(8, 64) => extend(bx.i8_type, bx.i64_type),
(16, 32) => extend(bx.i16_type, bx.i32_type),
(32, 64) => extend(bx.i32_type, bx.i64_type),
(16, 64) => extend(bx.i16_type, bx.i64_type),
_ => unimplemented!("in: {}, out: {}", in_width, out_width),
}
} else {
match (in_width, out_width) {
(8, 16) => extend(bx.u8_type, bx.u16_type),
(8, 32) => extend(bx.u8_type, bx.u32_type),
(8, 64) => extend(bx.u8_type, bx.u64_type),
(16, 32) => extend(bx.u16_type, bx.u32_type),
(16, 64) => extend(bx.u16_type, bx.u64_type),
(32, 64) => extend(bx.u32_type, bx.u64_type),
_ => unimplemented!("in: {}, out: {}", in_width, out_width),
}
}
(Style::Unsupported, Style::Unsupported) => {
require!(
false,
InvalidMonomorphization::UnsupportedCast {
span,
name,
in_ty,
in_elem,
ret_ty,
out_elem
}
});
}
(Style::Int(_), Style::Float) => {
// TODO: add support for internal functions in libgccjit to get access to IFN_VEC_CONVERT which is
// doing like __builtin_convertvector?
// Or maybe provide convert_vector as an API since it might not easy to get the
// types of internal functions.
unimplemented!();
}
(Style::Float, Style::Int(_)) => {
unimplemented!();
}
(Style::Float, Style::Float) => {
unimplemented!();
}
_ => { /* Unsupported. Fallthrough. */ }
);
},
_ => return Ok(bx.context.convert_vector(None, args[0].immediate(), llret_ty)),
}
return_error!(InvalidMonomorphizationUnsupportedCast {
span,
name,
in_ty,
in_elem,
ret_ty,
out_elem
});
}
macro_rules! arith_binary {
@ -436,6 +329,71 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
}
}
if name == sym::simd_bitmask {
// The `fn simd_bitmask(vector) -> unsigned integer` intrinsic takes a
// vector mask and returns the most significant bit (MSB) of each lane in the form
// of either:
// * an unsigned integer
// * an array of `u8`
// If the vector has less than 8 lanes, a u8 is returned with zeroed trailing bits.
//
// The bit order of the result depends on the byte endianness, LSB-first for little
// endian and MSB-first for big endian.
let vector = args[0].immediate();
let vector_type = vector.get_type().dyncast_vector().expect("vector type");
let elem_type = vector_type.get_element_type();
let expected_int_bits = in_len.max(8);
let expected_bytes = expected_int_bits / 8 + ((expected_int_bits % 8 > 0) as u64);
// FIXME(antoyo): that's not going to work for masks bigger than 128 bits.
let result_type = bx.type_ix(expected_int_bits);
let mut result = bx.context.new_rvalue_zero(result_type);
let elem_size = elem_type.get_size() * 8;
let sign_shift = bx.context.new_rvalue_from_int(elem_type, elem_size as i32 - 1);
let one = bx.context.new_rvalue_one(elem_type);
let mut shift = 0;
for i in 0..in_len {
let elem = bx.extract_element(vector, bx.context.new_rvalue_from_int(bx.int_type, i as i32));
let shifted = elem >> sign_shift;
let masked = shifted & one;
result = result | (bx.context.new_cast(None, masked, result_type) << bx.context.new_rvalue_from_int(result_type, shift));
shift += 1;
}
match ret_ty.kind() {
ty::Uint(i) if i.bit_width() == Some(expected_int_bits) => {
// Zero-extend iN to the bitmask type:
return Ok(result);
}
ty::Array(elem, len)
if matches!(elem.kind(), ty::Uint(ty::UintTy::U8))
&& len.try_eval_target_usize(bx.tcx, ty::ParamEnv::reveal_all())
== Some(expected_bytes) =>
{
// Zero-extend iN to the array length:
let ze = bx.zext(result, bx.type_ix(expected_bytes * 8));
// Convert the integer to a byte array
let ptr = bx.alloca(bx.type_ix(expected_bytes * 8), Align::ONE);
bx.store(ze, ptr, Align::ONE);
let array_ty = bx.type_array(bx.type_i8(), expected_bytes);
let ptr = bx.pointercast(ptr, bx.cx.type_ptr_to(array_ty));
return Ok(bx.load(array_ty, ptr, Align::ONE));
}
_ => return_error!(InvalidMonomorphization::CannotReturn {
span,
name,
ret_ty,
expected_int_bits,
expected_bytes
}),
}
}
fn simd_simple_float_intrinsic<'gcc, 'tcx>(
name: Symbol,
in_elem: Ty<'_>,
@ -451,55 +409,66 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
return Err(());
}};
}
let (elem_ty_str, elem_ty) = if let ty::Float(f) = in_elem.kind() {
let elem_ty = bx.cx.type_float_from_ty(*f);
match f.bit_width() {
32 => ("f32", elem_ty),
64 => ("f64", elem_ty),
_ => {
return_error!(InvalidMonomorphizationInvalidFloatVector {
span,
name,
elem_ty: f.name_str(),
vec_ty: in_ty
});
let (elem_ty_str, elem_ty) =
if let ty::Float(f) = in_elem.kind() {
let elem_ty = bx.cx.type_float_from_ty(*f);
match f.bit_width() {
32 => ("f", elem_ty),
64 => ("", elem_ty),
_ => {
return_error!(InvalidMonomorphizationInvalidFloatVector { span, name, elem_ty: f.name_str(), vec_ty: in_ty });
}
}
}
} else {
return_error!(InvalidMonomorphizationNotFloat { span, name, ty: in_ty });
};
else {
return_error!(InvalidMonomorphizationNotFloat { span, name, ty: in_ty });
};
let vec_ty = bx.cx.type_vector(elem_ty, in_len);
let (intr_name, fn_ty) = match name {
sym::simd_ceil => ("ceil", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_fabs => ("fabs", bx.type_func(&[vec_ty], vec_ty)), // TODO(antoyo): pand with 170141183420855150465331762880109871103
sym::simd_fcos => ("cos", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_fexp2 => ("exp2", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_fexp => ("exp", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_flog10 => ("log10", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_flog2 => ("log2", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_flog => ("log", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_floor => ("floor", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_fma => ("fma", bx.type_func(&[vec_ty, vec_ty, vec_ty], vec_ty)),
sym::simd_fpowi => ("powi", bx.type_func(&[vec_ty, bx.type_i32()], vec_ty)),
sym::simd_fpow => ("pow", bx.type_func(&[vec_ty, vec_ty], vec_ty)),
sym::simd_fsin => ("sin", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_fsqrt => ("sqrt", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_round => ("round", bx.type_func(&[vec_ty], vec_ty)),
sym::simd_trunc => ("trunc", bx.type_func(&[vec_ty], vec_ty)),
_ => return_error!(InvalidMonomorphizationUnrecognized { span, name }),
};
let llvm_name = &format!("llvm.{0}.v{1}{2}", intr_name, in_len, elem_ty_str);
let function = intrinsic::llvm::intrinsic(llvm_name, &bx.cx);
let function: RValue<'gcc> = unsafe { std::mem::transmute(function) };
let c = bx.call(
fn_ty,
None,
function,
&args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(),
None,
);
let intr_name =
match name {
sym::simd_ceil => "ceil",
sym::simd_fabs => "fabs", // TODO(antoyo): pand with 170141183420855150465331762880109871103
sym::simd_fcos => "cos",
sym::simd_fexp2 => "exp2",
sym::simd_fexp => "exp",
sym::simd_flog10 => "log10",
sym::simd_flog2 => "log2",
sym::simd_flog => "log",
sym::simd_floor => "floor",
sym::simd_fma => "fma",
sym::simd_fpowi => "__builtin_powi",
sym::simd_fpow => "pow",
sym::simd_fsin => "sin",
sym::simd_fsqrt => "sqrt",
sym::simd_round => "round",
sym::simd_trunc => "trunc",
_ => return_error!(InvalidMonomorphizationUnrecognized { span, name })
};
let builtin_name = format!("{}{}", intr_name, elem_ty_str);
let funcs = bx.cx.functions.borrow();
let function = funcs.get(&builtin_name).unwrap_or_else(|| panic!("unable to find builtin function {}", builtin_name));
// TODO(antoyo): add platform-specific behavior here for architectures that have these
// intrinsics as instructions (for instance, gpus)
let mut vector_elements = vec![];
for i in 0..in_len {
let index = bx.context.new_rvalue_from_long(bx.ulong_type, i as i64);
// we have to treat fpowi specially, since fpowi's second argument is always an i32
let arguments = if name == sym::simd_fpowi {
vec![
bx.extract_element(args[0].immediate(), index).to_rvalue(),
args[1].immediate(),
]
} else {
args.iter()
.map(|arg| bx.extract_element(arg.immediate(), index).to_rvalue())
.collect()
};
vector_elements.push(bx.context.new_call(None, *function, &arguments));
}
let c = bx.context.new_rvalue_from_vector(None, vec_ty, &vector_elements);
Ok(c)
}
@ -525,6 +494,297 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
return simd_simple_float_intrinsic(name, in_elem, in_ty, in_len, bx, span, args);
}
#[cfg(feature="master")]
fn vector_ty<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, elem_ty: Ty<'tcx>, vec_len: u64) -> Type<'gcc> {
// FIXME: use cx.layout_of(ty).llvm_type() ?
let elem_ty = match *elem_ty.kind() {
ty::Int(v) => cx.type_int_from_ty(v),
ty::Uint(v) => cx.type_uint_from_ty(v),
ty::Float(v) => cx.type_float_from_ty(v),
_ => unreachable!(),
};
cx.type_vector(elem_ty, vec_len)
}
#[cfg(feature="master")]
fn gather<'a, 'gcc, 'tcx>(default: RValue<'gcc>, pointers: RValue<'gcc>, mask: RValue<'gcc>, pointer_count: usize, bx: &mut Builder<'a, 'gcc, 'tcx>, in_len: u64, underlying_ty: Ty<'tcx>, invert: bool) -> RValue<'gcc> {
let vector_type =
if pointer_count > 1 {
bx.context.new_vector_type(bx.usize_type, in_len)
}
else {
vector_ty(bx, underlying_ty, in_len)
};
let elem_type = vector_type.dyncast_vector().expect("vector type").get_element_type();
let mut values = vec![];
for i in 0..in_len {
let index = bx.context.new_rvalue_from_long(bx.i32_type, i as i64);
let int = bx.context.new_vector_access(None, pointers, index).to_rvalue();
let ptr_type = elem_type.make_pointer();
let ptr = bx.context.new_bitcast(None, int, ptr_type);
let value = ptr.dereference(None).to_rvalue();
values.push(value);
}
let vector = bx.context.new_rvalue_from_vector(None, vector_type, &values);
let mut mask_types = vec![];
let mut mask_values = vec![];
for i in 0..in_len {
let index = bx.context.new_rvalue_from_long(bx.i32_type, i as i64);
mask_types.push(bx.context.new_field(None, bx.i32_type, "m"));
let mask_value = bx.context.new_vector_access(None, mask, index).to_rvalue();
let masked = bx.context.new_rvalue_from_int(bx.i32_type, in_len as i32) & mask_value;
let value = index + masked;
mask_values.push(value);
}
let mask_type = bx.context.new_struct_type(None, "mask_type", &mask_types);
let mask = bx.context.new_struct_constructor(None, mask_type.as_type(), None, &mask_values);
if invert {
bx.shuffle_vector(vector, default, mask)
}
else {
bx.shuffle_vector(default, vector, mask)
}
}
#[cfg(feature="master")]
if name == sym::simd_gather {
// simd_gather(values: <N x T>, pointers: <N x *_ T>,
// mask: <N x i{M}>) -> <N x T>
// * N: number of elements in the input vectors
// * T: type of the element to load
// * M: any integer width is supported, will be truncated to i1
// All types must be simd vector types
require_simd!(in_ty, "first");
require_simd!(arg_tys[1], "second");
require_simd!(arg_tys[2], "third");
require_simd!(ret_ty, "return");
// Of the same length:
let (out_len, _) = arg_tys[1].simd_size_and_type(bx.tcx());
let (out_len2, _) = arg_tys[2].simd_size_and_type(bx.tcx());
require!(
in_len == out_len,
InvalidMonomorphization::SecondArgumentLength {
span,
name,
in_len,
in_ty,
arg_ty: arg_tys[1],
out_len
}
);
require!(
in_len == out_len2,
InvalidMonomorphization::ThirdArgumentLength {
span,
name,
in_len,
in_ty,
arg_ty: arg_tys[2],
out_len: out_len2
}
);
// The return type must match the first argument type
require!(
ret_ty == in_ty,
InvalidMonomorphization::ExpectedReturnType { span, name, in_ty, ret_ty }
);
// This counts how many pointers
fn ptr_count(t: Ty<'_>) -> usize {
match t.kind() {
ty::RawPtr(p) => 1 + ptr_count(p.ty),
_ => 0,
}
}
// Non-ptr type
fn non_ptr(t: Ty<'_>) -> Ty<'_> {
match t.kind() {
ty::RawPtr(p) => non_ptr(p.ty),
_ => t,
}
}
// The second argument must be a simd vector with an element type that's a pointer
// to the element type of the first argument
let (_, element_ty0) = arg_tys[0].simd_size_and_type(bx.tcx());
let (_, element_ty1) = arg_tys[1].simd_size_and_type(bx.tcx());
let (pointer_count, underlying_ty) = match element_ty1.kind() {
ty::RawPtr(p) if p.ty == in_elem => (ptr_count(element_ty1), non_ptr(element_ty1)),
_ => {
require!(
false,
InvalidMonomorphization::ExpectedElementType {
span,
name,
expected_element: element_ty1,
second_arg: arg_tys[1],
in_elem,
in_ty,
mutability: ExpectedPointerMutability::Not,
}
);
unreachable!();
}
};
assert!(pointer_count > 0);
assert_eq!(pointer_count - 1, ptr_count(element_ty0));
assert_eq!(underlying_ty, non_ptr(element_ty0));
// The element type of the third argument must be a signed integer type of any width:
let (_, element_ty2) = arg_tys[2].simd_size_and_type(bx.tcx());
match element_ty2.kind() {
ty::Int(_) => (),
_ => {
require!(
false,
InvalidMonomorphization::ThirdArgElementType {
span,
name,
expected_element: element_ty2,
third_arg: arg_tys[2]
}
);
}
}
return Ok(gather(args[0].immediate(), args[1].immediate(), args[2].immediate(), pointer_count, bx, in_len, underlying_ty, false));
}
#[cfg(feature="master")]
if name == sym::simd_scatter {
// simd_scatter(values: <N x T>, pointers: <N x *mut T>,
// mask: <N x i{M}>) -> ()
// * N: number of elements in the input vectors
// * T: type of the element to load
// * M: any integer width is supported, will be truncated to i1
// All types must be simd vector types
require_simd!(in_ty, "first");
require_simd!(arg_tys[1], "second");
require_simd!(arg_tys[2], "third");
// Of the same length:
let (element_len1, _) = arg_tys[1].simd_size_and_type(bx.tcx());
let (element_len2, _) = arg_tys[2].simd_size_and_type(bx.tcx());
require!(
in_len == element_len1,
InvalidMonomorphization::SecondArgumentLength {
span,
name,
in_len,
in_ty,
arg_ty: arg_tys[1],
out_len: element_len1
}
);
require!(
in_len == element_len2,
InvalidMonomorphization::ThirdArgumentLength {
span,
name,
in_len,
in_ty,
arg_ty: arg_tys[2],
out_len: element_len2
}
);
// This counts how many pointers
fn ptr_count(t: Ty<'_>) -> usize {
match t.kind() {
ty::RawPtr(p) => 1 + ptr_count(p.ty),
_ => 0,
}
}
// Non-ptr type
fn non_ptr(t: Ty<'_>) -> Ty<'_> {
match t.kind() {
ty::RawPtr(p) => non_ptr(p.ty),
_ => t,
}
}
// The second argument must be a simd vector with an element type that's a pointer
// to the element type of the first argument
let (_, element_ty0) = arg_tys[0].simd_size_and_type(bx.tcx());
let (_, element_ty1) = arg_tys[1].simd_size_and_type(bx.tcx());
let (_, element_ty2) = arg_tys[2].simd_size_and_type(bx.tcx());
let (pointer_count, underlying_ty) = match element_ty1.kind() {
ty::RawPtr(p) if p.ty == in_elem && p.mutbl == hir::Mutability::Mut => {
(ptr_count(element_ty1), non_ptr(element_ty1))
}
_ => {
require!(
false,
InvalidMonomorphization::ExpectedElementType {
span,
name,
expected_element: element_ty1,
second_arg: arg_tys[1],
in_elem,
in_ty,
mutability: ExpectedPointerMutability::Mut,
}
);
unreachable!();
}
};
assert!(pointer_count > 0);
assert_eq!(pointer_count - 1, ptr_count(element_ty0));
assert_eq!(underlying_ty, non_ptr(element_ty0));
// The element type of the third argument must be a signed integer type of any width:
match element_ty2.kind() {
ty::Int(_) => (),
_ => {
require!(
false,
InvalidMonomorphization::ThirdArgElementType {
span,
name,
expected_element: element_ty2,
third_arg: arg_tys[2]
}
);
}
}
let result = gather(args[0].immediate(), args[1].immediate(), args[2].immediate(), pointer_count, bx, in_len, underlying_ty, true);
let pointers = args[1].immediate();
let vector_type =
if pointer_count > 1 {
bx.context.new_vector_type(bx.usize_type, in_len)
}
else {
vector_ty(bx, underlying_ty, in_len)
};
let elem_type = vector_type.dyncast_vector().expect("vector type").get_element_type();
for i in 0..in_len {
let index = bx.context.new_rvalue_from_int(bx.int_type, i as i32);
let value = bx.context.new_vector_access(None, result, index);
let int = bx.context.new_vector_access(None, pointers, index).to_rvalue();
let ptr_type = elem_type.make_pointer();
let ptr = bx.context.new_bitcast(None, int, ptr_type);
bx.llbb().add_assignment(None, ptr.dereference(None), value);
}
return Ok(bx.context.new_rvalue_zero(bx.i32_type));
}
arith_binary! {
simd_add: Uint, Int => add, Float => fadd;
simd_sub: Uint, Int => sub, Float => fsub;
@ -536,6 +796,8 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
simd_and: Uint, Int => and;
simd_or: Uint, Int => or; // FIXME(antoyo): calling `or` might not work on vectors.
simd_xor: Uint, Int => xor;
simd_fmin: Float => vector_fmin;
simd_fmax: Float => vector_fmax;
}
macro_rules! arith_unary {
@ -562,10 +824,11 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
let rhs = args[1].immediate();
let is_add = name == sym::simd_saturating_add;
let ptr_bits = bx.tcx().data_layout.pointer_size.bits() as _;
let (signed, elem_width, elem_ty) = match *in_elem.kind() {
ty::Int(i) => (true, i.bit_width().unwrap_or(ptr_bits), bx.cx.type_int_from_ty(i)),
ty::Uint(i) => (false, i.bit_width().unwrap_or(ptr_bits), bx.cx.type_uint_from_ty(i)),
_ => {
let (signed, elem_width, elem_ty) =
match *in_elem.kind() {
ty::Int(i) => (true, i.bit_width().unwrap_or(ptr_bits) / 8, bx.cx.type_int_from_ty(i)),
ty::Uint(i) => (false, i.bit_width().unwrap_or(ptr_bits) / 8, bx.cx.type_uint_from_ty(i)),
_ => {
return_error!(InvalidMonomorphizationExpectedSignedUnsigned {
span,
name,
@ -574,33 +837,78 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
});
}
};
let builtin_name = match (signed, is_add, in_len, elem_width) {
(true, true, 32, 8) => "__builtin_ia32_paddsb256", // TODO(antoyo): cast arguments to unsigned.
(false, true, 32, 8) => "__builtin_ia32_paddusb256",
(true, true, 16, 16) => "__builtin_ia32_paddsw256",
(false, true, 16, 16) => "__builtin_ia32_paddusw256",
(true, false, 16, 16) => "__builtin_ia32_psubsw256",
(false, false, 16, 16) => "__builtin_ia32_psubusw256",
(true, false, 32, 8) => "__builtin_ia32_psubsb256",
(false, false, 32, 8) => "__builtin_ia32_psubusb256",
_ => unimplemented!(
"signed: {}, is_add: {}, in_len: {}, elem_width: {}",
signed,
is_add,
in_len,
elem_width
),
};
let vec_ty = bx.cx.type_vector(elem_ty, in_len as u64);
let func = bx.context.get_target_builtin_function(builtin_name);
let param1_type = func.get_param(0).to_rvalue().get_type();
let param2_type = func.get_param(1).to_rvalue().get_type();
let lhs = bx.cx.bitcast_if_needed(lhs, param1_type);
let rhs = bx.cx.bitcast_if_needed(rhs, param2_type);
let result = bx.context.new_call(None, func, &[lhs, rhs]);
// TODO(antoyo): perhaps use __builtin_convertvector for vector casting.
return Ok(bx.context.new_bitcast(None, result, vec_ty));
let result =
match (signed, is_add) {
(false, true) => {
let res = lhs + rhs;
let cmp = bx.context.new_comparison(None, ComparisonOp::LessThan, res, lhs);
res | cmp
},
(true, true) => {
// Algorithm from: https://codereview.stackexchange.com/questions/115869/saturated-signed-addition
// TODO(antoyo): improve using conditional operators if possible.
let arg_type = lhs.get_type();
// TODO(antoyo): convert lhs and rhs to unsigned.
let sum = lhs + rhs;
let vector_type = arg_type.dyncast_vector().expect("vector type");
let unit = vector_type.get_num_units();
let a = bx.context.new_rvalue_from_int(elem_ty, ((elem_width as i32) << 3) - 1);
let width = bx.context.new_rvalue_from_vector(None, lhs.get_type(), &vec![a; unit]);
let xor1 = lhs ^ rhs;
let xor2 = lhs ^ sum;
let and = bx.context.new_unary_op(None, UnaryOp::BitwiseNegate, arg_type, xor1) & xor2;
let mask = and >> width;
let one = bx.context.new_rvalue_one(elem_ty);
let ones = bx.context.new_rvalue_from_vector(None, lhs.get_type(), &vec![one; unit]);
let shift1 = ones << width;
let shift2 = sum >> width;
let mask_min = shift1 ^ shift2;
let and1 = bx.context.new_unary_op(None, UnaryOp::BitwiseNegate, arg_type, mask) & sum;
let and2 = mask & mask_min;
and1 + and2
},
(false, false) => {
let res = lhs - rhs;
let cmp = bx.context.new_comparison(None, ComparisonOp::LessThanEquals, res, lhs);
res & cmp
},
(true, false) => {
let arg_type = lhs.get_type();
// TODO(antoyo): this uses the same algorithm from saturating add, but add the
// negative of the right operand. Find a proper subtraction algorithm.
let rhs = bx.context.new_unary_op(None, UnaryOp::Minus, arg_type, rhs);
// TODO(antoyo): convert lhs and rhs to unsigned.
let sum = lhs + rhs;
let vector_type = arg_type.dyncast_vector().expect("vector type");
let unit = vector_type.get_num_units();
let a = bx.context.new_rvalue_from_int(elem_ty, ((elem_width as i32) << 3) - 1);
let width = bx.context.new_rvalue_from_vector(None, lhs.get_type(), &vec![a; unit]);
let xor1 = lhs ^ rhs;
let xor2 = lhs ^ sum;
let and = bx.context.new_unary_op(None, UnaryOp::BitwiseNegate, arg_type, xor1) & xor2;
let mask = and >> width;
let one = bx.context.new_rvalue_one(elem_ty);
let ones = bx.context.new_rvalue_from_vector(None, lhs.get_type(), &vec![one; unit]);
let shift1 = ones << width;
let shift2 = sum >> width;
let mask_min = shift1 ^ shift2;
let and1 = bx.context.new_unary_op(None, UnaryOp::BitwiseNegate, arg_type, mask) & sum;
let and2 = mask & mask_min;
and1 + and2
}
};
return Ok(result);
}
macro_rules! arith_red {
@ -650,33 +958,50 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
add,
0.0 // TODO: Use this argument.
);
arith_red!(simd_reduce_mul_unordered: BinaryOp::Mult, vector_reduce_fmul_fast, false, mul, 1.0);
arith_red!(
simd_reduce_mul_unordered: BinaryOp::Mult,
vector_reduce_fmul_fast,
false,
mul,
1.0
);
arith_red!(
simd_reduce_add_ordered: BinaryOp::Plus,
vector_reduce_fadd,
true,
add,
0.0
);
arith_red!(
simd_reduce_mul_ordered: BinaryOp::Mult,
vector_reduce_fmul,
true,
mul,
1.0
);
macro_rules! minmax_red {
($name:ident: $reduction:ident) => {
($name:ident: $int_red:ident, $float_red:ident) => {
if name == sym::$name {
require!(
ret_ty == in_elem,
InvalidMonomorphizationReturnType { span, name, in_elem, in_ty, ret_ty }
);
return match in_elem.kind() {
ty::Int(_) | ty::Uint(_) | ty::Float(_) => {
Ok(bx.$reduction(args[0].immediate()))
}
_ => return_error!(InvalidMonomorphizationUnsupportedElement {
span,
name,
in_ty,
elem_ty: in_elem,
ret_ty
}),
ty::Int(_) | ty::Uint(_) => Ok(bx.$int_red(args[0].immediate())),
ty::Float(_) => Ok(bx.$float_red(args[0].immediate())),
_ => return_error!(InvalidMonomorphizationUnsupportedElement { span, name, in_ty, elem_ty: in_elem, ret_ty }),
};
}
};
}
minmax_red!(simd_reduce_min: vector_reduce_min);
minmax_red!(simd_reduce_max: vector_reduce_max);
minmax_red!(simd_reduce_min: vector_reduce_min, vector_reduce_fmin);
minmax_red!(simd_reduce_max: vector_reduce_max, vector_reduce_fmax);
// TODO(sadlerap): revisit these intrinsics to generate more optimal reductions
minmax_red!(simd_reduce_min_nanless: vector_reduce_min, vector_reduce_fmin);
minmax_red!(simd_reduce_max_nanless: vector_reduce_max, vector_reduce_fmax);
macro_rules! bitwise_red {
($name:ident : $op:expr, $boolean:expr) => {
@ -699,15 +1024,12 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
}),
}
// boolean reductions operate on vectors of i1s:
let i1 = bx.type_i1();
let i1xn = bx.type_vector(i1, in_len as u64);
bx.trunc(args[0].immediate(), i1xn)
args[0].immediate()
};
return match in_elem.kind() {
ty::Int(_) | ty::Uint(_) => {
let r = bx.vector_reduce_op(input, $op);
Ok(if !$boolean { r } else { bx.zext(r, bx.type_bool()) })
Ok(if !$boolean { r } else { bx.icmp(IntPredicate::IntNE, r, bx.context.new_rvalue_zero(r.get_type())) })
}
_ => return_error!(InvalidMonomorphizationUnsupportedElement {
span,
@ -723,6 +1045,9 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
bitwise_red!(simd_reduce_and: BinaryOp::BitwiseAnd, false);
bitwise_red!(simd_reduce_or: BinaryOp::BitwiseOr, false);
bitwise_red!(simd_reduce_xor: BinaryOp::BitwiseXor, false);
bitwise_red!(simd_reduce_all: BinaryOp::BitwiseAnd, true);
bitwise_red!(simd_reduce_any: BinaryOp::BitwiseOr, true);
unimplemented!("simd {}", name);
}

View File

@ -1,7 +1,7 @@
/*
* TODO(antoyo): implement equality in libgccjit based on https://zpz.github.io/blog/overloading-equality-operator-in-cpp-class-hierarchy/ (for type equality?)
* TODO(antoyo): support #[inline] attributes.
* TODO(antoyo): support LTO (gcc's equivalent to Thin LTO is enabled by -fwhopr: https://stackoverflow.com/questions/64954525/does-gcc-have-thin-lto).
* TODO(antoyo): support LTO (gcc's equivalent to Full LTO is -flto -flto-partition=one https://documentation.suse.com/sbp/all/html/SBP-GCC-10/index.html).
*
* TODO(antoyo): remove the patches.
*/
@ -23,6 +23,7 @@
extern crate rustc_apfloat;
extern crate rustc_ast;
extern crate rustc_attr;
extern crate rustc_codegen_ssa;
extern crate rustc_data_structures;
extern crate rustc_errors;
@ -43,6 +44,7 @@ mod abi;
mod allocator;
mod archive;
mod asm;
mod attributes;
mod back;
mod base;
mod builder;
@ -314,9 +316,12 @@ pub fn target_features(sess: &Session, allow_unstable: bool) -> Vec<Symbol> {
.filter(|_feature| {
// TODO(antoyo): implement a way to get enabled feature in libgccjit.
// Probably using the equivalent of __builtin_cpu_supports.
// TODO(antoyo): maybe use whatever outputs the following command:
// gcc -march=native -Q --help=target
#[cfg(feature="master")]
{
_feature.contains("sse") || _feature.contains("avx")
// NOTE: the CPU in the CI doesn't support sse4a, so disable it to make the stdarch tests pass in the CI.
(_feature.contains("sse") || _feature.contains("avx")) && !_feature.contains("avx512") && !_feature.contains("sse4a")
}
#[cfg(not(feature="master"))]
{

View File

@ -1,38 +1,66 @@
#[cfg(feature="master")]
use gccjit::{VarAttribute, FnAttribute};
use rustc_codegen_ssa::traits::PreDefineMethods;
use rustc_hir::def_id::{DefId, LOCAL_CRATE};
use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrFlags;
use rustc_middle::mir::mono::{Linkage, Visibility};
use rustc_middle::ty::{self, Instance, TypeVisitableExt};
use rustc_middle::ty::layout::{FnAbiOf, LayoutOf};
use rustc_span::def_id::DefId;
use crate::attributes;
use crate::base;
use crate::context::CodegenCx;
use crate::type_of::LayoutGccExt;
impl<'gcc, 'tcx> PreDefineMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
fn predefine_static(&self, def_id: DefId, _linkage: Linkage, _visibility: Visibility, symbol_name: &str) {
#[cfg_attr(not(feature="master"), allow(unused_variables))]
fn predefine_static(&self, def_id: DefId, _linkage: Linkage, visibility: Visibility, symbol_name: &str) {
let attrs = self.tcx.codegen_fn_attrs(def_id);
let instance = Instance::mono(self.tcx, def_id);
let ty = instance.ty(self.tcx, ty::ParamEnv::reveal_all());
let gcc_type = self.layout_of(ty).gcc_type(self, true);
let gcc_type = self.layout_of(ty).gcc_type(self);
let is_tls = attrs.flags.contains(CodegenFnAttrFlags::THREAD_LOCAL);
let global = self.define_global(symbol_name, gcc_type, is_tls, attrs.link_section);
#[cfg(feature="master")]
global.add_attribute(VarAttribute::Visibility(base::visibility_to_gcc(visibility)));
// TODO(antoyo): set linkage and visibility.
// TODO(antoyo): set linkage.
self.instances.borrow_mut().insert(instance, global);
}
fn predefine_fn(&self, instance: Instance<'tcx>, linkage: Linkage, _visibility: Visibility, symbol_name: &str) {
#[cfg_attr(not(feature="master"), allow(unused_variables))]
fn predefine_fn(&self, instance: Instance<'tcx>, linkage: Linkage, visibility: Visibility, symbol_name: &str) {
assert!(!instance.substs.needs_infer());
let fn_abi = self.fn_abi_of_instance(instance, ty::List::empty());
self.linkage.set(base::linkage_to_gcc(linkage));
let _decl = self.declare_fn(symbol_name, &fn_abi);
let decl = self.declare_fn(symbol_name, &fn_abi);
//let attrs = self.tcx.codegen_fn_attrs(instance.def_id());
attributes::from_fn_attrs(self, decl, instance);
// If we're compiling the compiler-builtins crate, e.g., the equivalent of
// compiler-rt, then we want to implicitly compile everything with hidden
// visibility as we're going to link this object all over the place but
// don't want the symbols to get exported.
if linkage != Linkage::Internal
&& linkage != Linkage::Private
&& self.tcx.is_compiler_builtins(LOCAL_CRATE)
{
#[cfg(feature="master")]
decl.add_attribute(FnAttribute::Visibility(gccjit::Visibility::Hidden));
}
else {
#[cfg(feature="master")]
decl.add_attribute(FnAttribute::Visibility(base::visibility_to_gcc(visibility)));
}
// TODO(antoyo): call set_link_section() to allow initializing argc/argv.
// TODO(antoyo): set unique comdat.
// TODO(antoyo): use inline attribute from there in linkage.set() above.
self.functions.borrow_mut().insert(symbol_name.to_string(), decl);
self.function_instances.borrow_mut().insert(instance, unsafe { std::mem::transmute(decl) });
}
}

View File

@ -1,5 +1,3 @@
use std::convert::TryInto;
use gccjit::{RValue, Struct, Type};
use rustc_codegen_ssa::traits::{BaseTypeMethods, DerivedTypeMethods, TypeMembershipMethods};
use rustc_codegen_ssa::common::TypeKind;
@ -202,8 +200,9 @@ impl<'gcc, 'tcx> BaseTypeMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
value.get_type()
}
fn type_array(&self, ty: Type<'gcc>, mut len: u64) -> Type<'gcc> {
if let Some(struct_type) = ty.is_struct() {
fn type_array(&self, ty: Type<'gcc>, len: u64) -> Type<'gcc> {
// TODO: remove this as well?
/*if let Some(struct_type) = ty.is_struct() {
if struct_type.get_field_count() == 0 {
// NOTE: since gccjit only supports i32 for the array size and libcore's tests uses a
// size of usize::MAX in test_binary_search, we workaround this by setting the size to
@ -211,14 +210,7 @@ impl<'gcc, 'tcx> BaseTypeMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
// FIXME(antoyo): fix gccjit API.
len = 0;
}
}
// NOTE: see note above. Some other test uses usize::MAX.
if len == u64::MAX {
len = 0;
}
let len: i32 = len.try_into().expect("array len");
}*/
self.context.new_array_type(None, ty, len)
}
@ -247,10 +239,6 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
pub fn type_named_struct(&self, name: &str) -> Struct<'gcc> {
self.context.new_opaque_struct_type(None, name)
}
pub fn type_bool(&self) -> Type<'gcc> {
self.context.new_type::<bool>()
}
}
pub fn struct_fields<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, layout: TyAndLayout<'tcx>) -> (Vec<Type<'gcc>>, bool) {
@ -273,7 +261,7 @@ pub fn struct_fields<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, layout: TyAndLayout
assert_eq!(offset.align_to(padding_align) + padding, target_offset);
result.push(cx.type_padding_filler(padding, padding_align));
result.push(field.gcc_type(cx, !field.ty.is_any_ptr())); // FIXME(antoyo): might need to check if the type is inside another, like Box<Type>.
result.push(field.gcc_type(cx));
offset = target_offset + field.size;
prev_effective_align = effective_field_align;
}

View File

@ -6,7 +6,7 @@ use rustc_middle::bug;
use rustc_middle::ty::{self, Ty, TypeVisitableExt};
use rustc_middle::ty::layout::{FnAbiOf, LayoutOf, TyAndLayout};
use rustc_middle::ty::print::with_no_trimmed_paths;
use rustc_target::abi::{self, Abi, F32, F64, FieldsShape, Int, Integer, Pointer, PointeeInfo, Size, TyAbiInterface, Variants};
use rustc_target::abi::{self, Abi, Align, F32, F64, FieldsShape, Int, Integer, Pointer, PointeeInfo, Size, TyAbiInterface, Variants};
use rustc_target::abi::call::{CastTarget, FnAbi, Reg};
use crate::abi::{FnAbiGccExt, GccType};
@ -50,11 +50,25 @@ impl<'gcc, 'tcx> CodegenCx<'gcc, 'tcx> {
}
}
pub fn uncached_gcc_type<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, layout: TyAndLayout<'tcx>, defer: &mut Option<(Struct<'gcc>, TyAndLayout<'tcx>)>) -> Type<'gcc> {
impl<'a, 'tcx> CodegenCx<'a, 'tcx> {
pub fn align_of(&self, ty: Ty<'tcx>) -> Align {
self.layout_of(ty).align.abi
}
}
fn uncached_gcc_type<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, layout: TyAndLayout<'tcx>, defer: &mut Option<(Struct<'gcc>, TyAndLayout<'tcx>)>) -> Type<'gcc> {
match layout.abi {
Abi::Scalar(_) => bug!("handled elsewhere"),
Abi::Vector { ref element, count } => {
let element = layout.scalar_gcc_type_at(cx, element, Size::ZERO);
let element =
// NOTE: gcc doesn't allow pointer types in vectors.
if element.get_pointee().is_some() {
cx.usize_type
}
else {
element
};
return cx.context.new_vector_type(element, count);
},
Abi::ScalarPair(..) => {
@ -114,7 +128,7 @@ pub fn uncached_gcc_type<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, layout: TyAndLa
},
}
}
FieldsShape::Array { count, .. } => cx.type_array(layout.field(cx, 0).gcc_type(cx, true), count),
FieldsShape::Array { count, .. } => cx.type_array(layout.field(cx, 0).gcc_type(cx), count),
FieldsShape::Arbitrary { .. } =>
match name {
None => {
@ -133,7 +147,7 @@ pub fn uncached_gcc_type<'gcc, 'tcx>(cx: &CodegenCx<'gcc, 'tcx>, layout: TyAndLa
pub trait LayoutGccExt<'tcx> {
fn is_gcc_immediate(&self) -> bool;
fn is_gcc_scalar_pair(&self) -> bool;
fn gcc_type<'gcc>(&self, cx: &CodegenCx<'gcc, 'tcx>, set_fields: bool) -> Type<'gcc>;
fn gcc_type<'gcc>(&self, cx: &CodegenCx<'gcc, 'tcx>) -> Type<'gcc>;
fn immediate_gcc_type<'gcc>(&self, cx: &CodegenCx<'gcc, 'tcx>) -> Type<'gcc>;
fn scalar_gcc_type_at<'gcc>(&self, cx: &CodegenCx<'gcc, 'tcx>, scalar: &abi::Scalar, offset: Size) -> Type<'gcc>;
fn scalar_pair_element_gcc_type<'gcc>(&self, cx: &CodegenCx<'gcc, 'tcx>, index: usize, immediate: bool) -> Type<'gcc>;
@ -168,8 +182,7 @@ impl<'tcx> LayoutGccExt<'tcx> for TyAndLayout<'tcx> {
/// with the inner-most trailing unsized field using the "minimal unit"
/// of that field's type - this is useful for taking the address of
/// that field and ensuring the struct has the right alignment.
//TODO(antoyo): do we still need the set_fields parameter?
fn gcc_type<'gcc>(&self, cx: &CodegenCx<'gcc, 'tcx>, set_fields: bool) -> Type<'gcc> {
fn gcc_type<'gcc>(&self, cx: &CodegenCx<'gcc, 'tcx>) -> Type<'gcc> {
if let Abi::Scalar(ref scalar) = self.abi {
// Use a different cache for scalars because pointers to DSTs
// can be either fat or thin (data pointers of fat pointers).
@ -179,10 +192,10 @@ impl<'tcx> LayoutGccExt<'tcx> for TyAndLayout<'tcx> {
let ty =
match *self.ty.kind() {
ty::Ref(_, ty, _) | ty::RawPtr(ty::TypeAndMut { ty, .. }) => {
cx.type_ptr_to(cx.layout_of(ty).gcc_type(cx, set_fields))
cx.type_ptr_to(cx.layout_of(ty).gcc_type(cx))
}
ty::Adt(def, _) if def.is_box() => {
cx.type_ptr_to(cx.layout_of(self.ty.boxed_ty()).gcc_type(cx, true))
cx.type_ptr_to(cx.layout_of(self.ty.boxed_ty()).gcc_type(cx))
}
ty::FnPtr(sig) => cx.fn_ptr_backend_type(&cx.fn_abi_of_fn_ptr(sig, ty::List::empty())),
_ => self.scalar_gcc_type_at(cx, scalar, Size::ZERO),
@ -199,13 +212,6 @@ impl<'tcx> LayoutGccExt<'tcx> for TyAndLayout<'tcx> {
};
let cached_type = cx.types.borrow().get(&(self.ty, variant_index)).cloned();
if let Some(ty) = cached_type {
let type_to_set_fields = cx.types_with_fields_to_set.borrow_mut().remove(&ty);
if let Some((struct_type, layout)) = type_to_set_fields {
// Since we might be trying to generate a type containing another type which is not
// completely generated yet, we deferred setting the fields until now.
let (fields, packed) = struct_fields(cx, layout);
cx.set_struct_body(struct_type, &fields, packed);
}
return ty;
}
@ -222,7 +228,7 @@ impl<'tcx> LayoutGccExt<'tcx> for TyAndLayout<'tcx> {
if let Some(v) = variant_index {
layout = layout.for_variant(cx, v);
}
layout.gcc_type(cx, true)
layout.gcc_type(cx)
}
else {
uncached_gcc_type(cx, *self, &mut defer)
@ -230,9 +236,9 @@ impl<'tcx> LayoutGccExt<'tcx> for TyAndLayout<'tcx> {
cx.types.borrow_mut().insert((self.ty, variant_index), ty);
if let Some((ty, layout)) = defer {
if let Some((deferred_ty, layout)) = defer {
let (fields, packed) = struct_fields(cx, layout);
cx.set_struct_body(ty, &fields, packed);
cx.set_struct_body(deferred_ty, &fields, packed);
}
ty
@ -244,7 +250,7 @@ impl<'tcx> LayoutGccExt<'tcx> for TyAndLayout<'tcx> {
return cx.type_i1();
}
}
self.gcc_type(cx, true)
self.gcc_type(cx)
}
fn scalar_gcc_type_at<'gcc>(&self, cx: &CodegenCx<'gcc, 'tcx>, scalar: &abi::Scalar, offset: Size) -> Type<'gcc> {
@ -273,7 +279,7 @@ impl<'tcx> LayoutGccExt<'tcx> for TyAndLayout<'tcx> {
// pointee types, to avoid bitcasting every `OperandRef::deref`.
match self.ty.kind() {
ty::Ref(..) | ty::RawPtr(_) => {
return self.field(cx, index).gcc_type(cx, true);
return self.field(cx, index).gcc_type(cx);
}
// only wide pointer boxes are handled as pointers
// thin pointer boxes with scalar allocators are handled by the general logic below
@ -343,7 +349,7 @@ impl<'tcx> LayoutGccExt<'tcx> for TyAndLayout<'tcx> {
impl<'gcc, 'tcx> LayoutTypeMethods<'tcx> for CodegenCx<'gcc, 'tcx> {
fn backend_type(&self, layout: TyAndLayout<'tcx>) -> Type<'gcc> {
layout.gcc_type(self, true)
layout.gcc_type(self)
}
fn immediate_backend_type(&self, layout: TyAndLayout<'tcx>) -> Type<'gcc> {

View File

@ -17,17 +17,20 @@ export LIBRARY_PATH="$GCC_PATH"
flags=
gcc_master_branch=1
channel="debug"
func=all
funcs=()
build_only=0
nb_parts=0
current_part=0
while [[ $# -gt 0 ]]; do
case $1 in
--release)
codegen_channel=release
channel="release"
shift
;;
--release-sysroot)
sysroot_channel=release
sysroot_channel="--release"
shift
;;
--no-default-features)
@ -40,43 +43,83 @@ while [[ $# -gt 0 ]]; do
flags="$flags --features $1"
shift
;;
--release)
channel="release"
"--test-rustc")
funcs+=(test_rustc)
shift
;;
"--test-rustc")
func=test_rustc
"--test-successful-rustc")
funcs+=(test_successful_rustc)
shift
;;
"--test-failing-rustc")
funcs+=(test_failing_rustc)
shift
;;
"--test-libcore")
func=test_libcore
funcs+=(test_libcore)
shift
;;
"--clean-ui-tests")
func=clean_ui_tests
funcs+=(clean_ui_tests)
shift
;;
"--clean")
funcs+=(clean)
shift
;;
"--std-tests")
func=std_tests
funcs+=(std_tests)
shift
;;
"--asm-tests")
funcs+=(asm_tests)
shift
;;
"--extended-tests")
func=extended_sysroot_tests
funcs+=(extended_sysroot_tests)
shift
;;
"--extended-rand-tests")
funcs+=(extended_rand_tests)
shift
;;
"--extended-regex-example-tests")
funcs+=(extended_regex_example_tests)
shift
;;
"--extended-regex-tests")
funcs+=(extended_regex_tests)
shift
;;
"--mini-tests")
funcs+=(mini_tests)
shift
;;
"--build-sysroot")
func=build_sysroot
funcs+=(build_sysroot)
shift
;;
"--build")
build_only=1
shift
;;
"--nb-parts")
shift
nb_parts=$1
shift
;;
"--current-part")
shift
current_part=$1
shift
;;
*)
echo "Unknown option $1"
exit 1
@ -87,7 +130,6 @@ done
if [[ $channel == "release" ]]; then
export CHANNEL='release'
CARGO_INCREMENTAL=1 cargo rustc --release $flags
shift
else
echo $LD_LIBRARY_PATH
export CHANNEL='debug'
@ -95,6 +137,7 @@ else
fi
if (( $build_only == 1 )); then
echo "Since it's 'build-only', exiting..."
exit
fi
@ -119,7 +162,7 @@ function mini_tests() {
function build_sysroot() {
echo "[BUILD] sysroot"
time ./build_sysroot/build_sysroot.sh
time ./build_sysroot/build_sysroot.sh $sysroot_channel
}
function std_tests() {
@ -148,17 +191,57 @@ function std_tests() {
$RUN_WRAPPER ./target/out/std_example --target $TARGET_TRIPLE
echo "[AOT] subslice-patterns-const-eval"
$RUSTC example/subslice-patterns-const-eval.rs --crate-type bin -Cpanic=abort --target $TARGET_TRIPLE
$RUSTC example/subslice-patterns-const-eval.rs --crate-type bin $TEST_FLAGS --target $TARGET_TRIPLE
$RUN_WRAPPER ./target/out/subslice-patterns-const-eval
echo "[AOT] track-caller-attribute"
$RUSTC example/track-caller-attribute.rs --crate-type bin -Cpanic=abort --target $TARGET_TRIPLE
$RUSTC example/track-caller-attribute.rs --crate-type bin $TEST_FLAGS --target $TARGET_TRIPLE
$RUN_WRAPPER ./target/out/track-caller-attribute
echo "[BUILD] mod_bench"
$RUSTC example/mod_bench.rs --crate-type bin --target $TARGET_TRIPLE
}
function setup_rustc() {
rust_toolchain=$(cat rust-toolchain | grep channel | sed 's/channel = "\(.*\)"/\1/')
git clone https://github.com/rust-lang/rust.git || true
cd rust
git fetch
git checkout $(rustc -V | cut -d' ' -f3 | tr -d '(')
export RUSTFLAGS=
rm config.toml || true
cat > config.toml <<EOF
[rust]
codegen-backends = []
deny-warnings = false
[build]
cargo = "$(which cargo)"
local-rebuild = true
rustc = "$HOME/.rustup/toolchains/$rust_toolchain-$TARGET_TRIPLE/bin/rustc"
[target.x86_64-unknown-linux-gnu]
llvm-filecheck = "`which FileCheck-10 || which FileCheck-11 || which FileCheck-12 || which FileCheck-13 || which FileCheck-14`"
[llvm]
download-ci-llvm = false
EOF
rustc -V | cut -d' ' -f3 | tr -d '('
git checkout $(rustc -V | cut -d' ' -f3 | tr -d '(') tests
}
function asm_tests() {
setup_rustc
echo "[TEST] rustc test suite"
RUSTC_ARGS="-Zpanic-abort-tests -Csymbol-mangling-version=v0 -Zcodegen-backend="$(pwd)"/../target/"$CHANNEL"/librustc_codegen_gcc."$dylib_ext" --sysroot "$(pwd)"/../build_sysroot/sysroot -Cpanic=abort"
COMPILETEST_FORCE_STAGE0=1 ./x.py test --run always --stage 0 tests/assembly/asm --rustc-args "$RUSTC_ARGS"
}
# FIXME(antoyo): linker gives multiple definitions error on Linux
#echo "[BUILD] sysroot in release mode"
#./build_sysroot/build_sysroot.sh --release
@ -187,7 +270,7 @@ function test_libcore() {
#echo "[BENCH RUN] mod_bench"
#hyperfine --runs ${RUN_RUNS:-10} ./target/out/mod_bench{,_inline} ./target/out/mod_bench_llvm_*
function extended_sysroot_tests() {
function extended_rand_tests() {
if (( $gcc_master_branch == 0 )); then
return
fi
@ -197,17 +280,12 @@ function extended_sysroot_tests() {
echo "[TEST] rust-random/rand"
../cargo.sh test --workspace
popd
}
#pushd simple-raytracer
#echo "[BENCH COMPILE] ebobby/simple-raytracer"
#hyperfine --runs "${RUN_RUNS:-10}" --warmup 1 --prepare "cargo clean" \
#"RUSTC=rustc RUSTFLAGS='' cargo build" \
#"../cargo.sh build"
#echo "[BENCH RUN] ebobby/simple-raytracer"
#cp ./target/debug/main ./raytracer_cg_gcc
#hyperfine --runs "${RUN_RUNS:-10}" ./raytracer_cg_llvm ./raytracer_cg_gcc
#popd
function extended_regex_example_tests() {
if (( $gcc_master_branch == 0 )); then
return
fi
pushd regex
echo "[TEST] rust-lang/regex example shootout-regex-dna"
@ -219,41 +297,43 @@ function extended_sysroot_tests() {
| ../cargo.sh run --example shootout-regex-dna \
| grep -v "Spawned thread" > res.txt
diff -u res.txt examples/regexdna-output.txt
popd
}
function extended_regex_tests() {
if (( $gcc_master_branch == 0 )); then
return
fi
pushd regex
echo "[TEST] rust-lang/regex tests"
export CG_RUSTFLAGS="--cap-lints warn" # newer aho_corasick versions throw a deprecation warning
../cargo.sh test --tests -- --exclude-should-panic --test-threads 1 -Zunstable-options -q
popd
}
function extended_sysroot_tests() {
#pushd simple-raytracer
#echo "[BENCH COMPILE] ebobby/simple-raytracer"
#hyperfine --runs "${RUN_RUNS:-10}" --warmup 1 --prepare "cargo clean" \
#"RUSTC=rustc RUSTFLAGS='' cargo build" \
#"../cargo.sh build"
#echo "[BENCH RUN] ebobby/simple-raytracer"
#cp ./target/debug/main ./raytracer_cg_gcc
#hyperfine --runs "${RUN_RUNS:-10}" ./raytracer_cg_llvm ./raytracer_cg_gcc
#popd
extended_rand_tests
extended_regex_example_tests
extended_regex_tests
}
function test_rustc() {
echo
echo "[TEST] rust-lang/rust"
rust_toolchain=$(cat rust-toolchain | grep channel | sed 's/channel = "\(.*\)"/\1/')
git clone https://github.com/rust-lang/rust.git || true
cd rust
git fetch
git checkout $(rustc -V | cut -d' ' -f3 | tr -d '(')
export RUSTFLAGS=
git apply ../rustc_patches/compile_test.patch || true
rm config.toml || true
cat > config.toml <<EOF
[rust]
codegen-backends = []
deny-warnings = false
[build]
cargo = "$(which cargo)"
local-rebuild = true
rustc = "$HOME/.rustup/toolchains/$rust_toolchain-$TARGET_TRIPLE/bin/rustc"
EOF
rustc -V | cut -d' ' -f3 | tr -d '('
git checkout $(rustc -V | cut -d' ' -f3 | tr -d '(') tests
setup_rustc
for test in $(rg -i --files-with-matches "//(\[\w+\])?~|// error-pattern:|// build-fail|// run-fail|-Cllvm-args" tests/ui); do
rm $test
@ -261,21 +341,61 @@ EOF
git checkout -- tests/ui/issues/auxiliary/issue-3136-a.rs # contains //~ERROR, but shouldn't be removed
rm -r tests/ui/{abi*,extern/,panic-runtime/,panics/,unsized-locals/,proc-macro/,threads-sendsync/,thinlto/,borrowck/,test*,*lto*.rs} || true
for test in $(rg --files-with-matches "catch_unwind|should_panic|thread|lto" tests/ui); do
rm -r tests/ui/{abi*,extern/,unsized-locals/,proc-macro/,threads-sendsync/,thinlto/,borrowck/,chalkify/bugs/,test*,*lto*.rs,consts/const-float-bits-reject-conv.rs,consts/issue-miri-1910.rs} || true
rm tests/ui/mir/mir_heavy_promoted.rs # this tests is oom-killed in the CI.
for test in $(rg --files-with-matches "thread|lto" tests/ui); do
rm $test
done
git checkout tests/ui/lto/auxiliary/dylib.rs
git checkout tests/ui/type-alias-impl-trait/auxiliary/cross_crate_ice.rs
git checkout tests/ui/type-alias-impl-trait/auxiliary/cross_crate_ice2.rs
git checkout tests/ui/macros/rfc-2011-nicer-assert-messages/auxiliary/common.rs
RUSTC_ARGS="-Zpanic-abort-tests -Csymbol-mangling-version=v0 -Zcodegen-backend="$(pwd)"/../target/"$CHANNEL"/librustc_codegen_gcc."$dylib_ext" --sysroot "$(pwd)"/../build_sysroot/sysroot -Cpanic=abort"
RUSTC_ARGS="$TEST_FLAGS -Csymbol-mangling-version=v0 -Zcodegen-backend="$(pwd)"/../target/"$CHANNEL"/librustc_codegen_gcc."$dylib_ext" --sysroot "$(pwd)"/../build_sysroot/sysroot"
if [ $# -eq 0 ]; then
# No argument supplied to the function. Doing nothing.
echo "No argument provided. Keeping all UI tests"
elif [ $1 = "0" ]; then
# Removing the failing tests.
xargs -a ../failing-ui-tests.txt -d'\n' rm
else
# Removing all tests.
find tests/ui -type f -name '*.rs' -not -path '*/auxiliary/*' -delete
# Putting back only the failing ones.
xargs -a ../failing-ui-tests.txt -d'\n' git checkout --
fi
if [ $nb_parts -gt 0 ]; then
echo "Splitting ui_test into $nb_parts parts (and running part $current_part)"
find tests/ui -type f -name '*.rs' -not -path "*/auxiliary/*" > ui_tests
# To ensure it'll be always the same sub files, we sort the content.
sort ui_tests -o ui_tests
count=$((`wc -l < ui_tests` / $nb_parts))
# We increment the number of tests by one because if this is an odd number, we would skip
# one test.
count=$((count + 1))
split -d -l $count -a 1 ui_tests ui_tests.split
# Removing all tests.
find tests/ui -type f -name '*.rs' -not -path "*/auxiliary/*" -delete
# Putting back only the ones we want to test.
xargs -a "ui_tests.split$current_part" -d'\n' git checkout --
fi
echo "[TEST] rustc test suite"
COMPILETEST_FORCE_STAGE0=1 ./x.py test --run always --stage 0 tests/ui/ --rustc-args "$RUSTC_ARGS"
}
function test_failing_rustc() {
test_rustc "1"
}
function test_successful_rustc() {
test_rustc "0"
}
function clean_ui_tests() {
find rust/build/x86_64-unknown-linux-gnu/test/ui/ -name stamp -exec rm -rf {} \;
find rust/build/x86_64-unknown-linux-gnu/test/ui/ -name stamp -delete
}
function all() {
@ -283,9 +403,17 @@ function all() {
mini_tests
build_sysroot
std_tests
#asm_tests
test_libcore
extended_sysroot_tests
test_rustc
}
$func
if [ ${#funcs[@]} -eq 0 ]; then
echo "No command passed, running '--all'..."
all
else
for t in ${funcs[@]}; do
$t
done
fi

View File

@ -46,11 +46,15 @@ pub fn main_inner(profile: Profile) {
&format!("-Zcodegen-backend={}/target/debug/librustc_codegen_gcc.so", current_dir),
"--sysroot", &format!("{}/build_sysroot/sysroot/", current_dir),
"-Zno-parallel-llvm",
"-C", "panic=abort",
"-C", "link-arg=-lc",
"-o", exe.to_str().expect("to_str"),
path.to_str().expect("to_str"),
]);
if let Some(flags) = option_env!("TEST_FLAGS") {
for flag in flags.split_whitespace() {
compiler.arg(&flag);
}
}
match profile {
Profile::Debug => {}
Profile::Release => {

View File

@ -33,6 +33,7 @@ mod intrinsics {
use super::Sized;
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}

View File

@ -33,6 +33,7 @@ mod intrinsics {
use super::Sized;
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}

View File

@ -79,7 +79,7 @@ pub unsafe fn drop_in_place<T: ?Sized>(to_drop: *mut T) {
#[lang = "panic"]
#[track_caller]
#[no_mangle]
pub fn panic(_msg: &str) -> ! {
pub fn panic(_msg: &'static str) -> ! {
unsafe {
libc::puts("Panicking\0" as *const str as *const u8);
intrinsics::abort();
@ -105,6 +105,7 @@ fn panic_bounds_check(index: usize, len: usize) -> ! {
mod intrinsics {
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}

View File

@ -57,6 +57,7 @@ mod libc {
mod intrinsics {
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}
@ -64,7 +65,7 @@ mod intrinsics {
#[lang = "panic"]
#[track_caller]
#[no_mangle]
pub fn panic(_msg: &str) -> ! {
pub fn panic(_msg: &'static str) -> ! {
unsafe {
libc::puts("Panicking\0" as *const str as *const u8);
libc::fflush(libc::stdout);

View File

@ -97,10 +97,14 @@ fn panic_bounds_check(index: usize, len: usize) -> ! {
mod intrinsics {
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}
#[lang = "tuple_trait"]
pub trait Tuple {}
#[lang = "unsize"]
pub trait Unsize<T: ?Sized> {}
@ -114,7 +118,7 @@ impl<T: ?Sized + Unsize<U>, U: ?Sized> CoerceUnsized<*mut U> for *mut T {}
#[lang = "fn_once"]
#[rustc_paren_sugar]
pub trait FnOnce<Args> {
pub trait FnOnce<Args: Tuple> {
#[lang = "fn_once_output"]
type Output;
@ -123,7 +127,7 @@ pub trait FnOnce<Args> {
#[lang = "fn_mut"]
#[rustc_paren_sugar]
pub trait FnMut<Args>: FnOnce<Args> {
pub trait FnMut<Args: Tuple>: FnOnce<Args> {
extern "rust-call" fn call_mut(&mut self, args: Args) -> Self::Output;
}
@ -177,7 +181,7 @@ impl Add for isize {
#[lang = "panic"]
#[track_caller]
#[no_mangle]
pub fn panic(_msg: &str) -> ! {
pub fn panic(_msg: &'static str) -> ! {
unsafe {
libc::puts("Panicking\0" as *const str as *const u8);
intrinsics::abort();

View File

@ -82,7 +82,7 @@ pub unsafe fn drop_in_place<T: ?Sized>(to_drop: *mut T) {
#[lang = "panic"]
#[track_caller]
#[no_mangle]
pub fn panic(_msg: &str) -> ! {
pub fn panic(_msg: &'static str) -> ! {
unsafe {
libc::puts("Panicking\0" as *const str as *const u8);
intrinsics::abort();
@ -108,6 +108,7 @@ fn panic_bounds_check(index: usize, len: usize) -> ! {
mod intrinsics {
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}

View File

@ -76,7 +76,7 @@ pub unsafe fn drop_in_place<T: ?Sized>(to_drop: *mut T) {
#[lang = "panic"]
#[track_caller]
#[no_mangle]
pub fn panic(_msg: &str) -> ! {
pub fn panic(_msg: &'static str) -> ! {
unsafe {
libc::puts("Panicking\0" as *const str as *const u8);
intrinsics::abort();
@ -102,6 +102,7 @@ fn panic_bounds_check(index: usize, len: usize) -> ! {
mod intrinsics {
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}

View File

@ -3,22 +3,14 @@
// Run-time:
// status: 0
#![feature(const_black_box, core_intrinsics, start)]
#![no_std]
#[panic_handler]
fn panic_handler(_: &core::panic::PanicInfo) -> ! {
core::intrinsics::abort();
}
#![feature(const_black_box)]
/*
* Code
*/
#[start]
fn main(_argc: isize, _argv: *const *const u8) -> isize {
use core::hint::black_box;
fn main() {
use std::hint::black_box;
macro_rules! check {
($ty:ty, $expr:expr) => {
@ -335,6 +327,4 @@ fn main(_argc: isize, _argv: *const *const u8) -> isize {
const VAL5: T = 73236519889708027473620326106273939584_i128;
check_ops128!();
}
0
}

View File

@ -55,6 +55,7 @@ mod libc {
mod intrinsics {
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}
@ -62,7 +63,7 @@ mod intrinsics {
#[lang = "panic"]
#[track_caller]
#[no_mangle]
pub fn panic(_msg: &str) -> ! {
pub fn panic(_msg: &'static str) -> ! {
unsafe {
// Panicking is expected iff overflow checking is enabled.
#[cfg(debug_assertions)]

View File

@ -59,6 +59,7 @@ mod libc {
mod intrinsics {
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}
@ -66,7 +67,7 @@ mod intrinsics {
#[lang = "panic"]
#[track_caller]
#[no_mangle]
pub fn panic(_msg: &str) -> ! {
pub fn panic(_msg: &'static str) -> ! {
unsafe {
libc::puts("Panicking\0" as *const str as *const u8);
libc::fflush(libc::stdout);

View File

@ -65,6 +65,7 @@ mod libc {
mod intrinsics {
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}
@ -72,7 +73,7 @@ mod intrinsics {
#[lang = "panic"]
#[track_caller]
#[no_mangle]
pub fn panic(_msg: &str) -> ! {
pub fn panic(_msg: &'static str) -> ! {
unsafe {
libc::puts("Panicking\0" as *const str as *const u8);
libc::fflush(libc::stdout);

View File

@ -76,7 +76,7 @@ pub unsafe fn drop_in_place<T: ?Sized>(to_drop: *mut T) {
#[lang = "panic"]
#[track_caller]
#[no_mangle]
pub fn panic(_msg: &str) -> ! {
pub fn panic(_msg: &'static str) -> ! {
unsafe {
libc::puts("Panicking\0" as *const str as *const u8);
intrinsics::abort();
@ -102,6 +102,7 @@ fn panic_bounds_check(index: usize, len: usize) -> ! {
mod intrinsics {
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}

View File

@ -102,6 +102,7 @@ mod intrinsics {
use super::Sized;
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}

View File

@ -45,6 +45,7 @@ mod intrinsics {
use super::Sized;
extern "rust-intrinsic" {
#[rustc_safe_intrinsic]
pub fn abort() -> !;
}
}

View File

@ -0,0 +1,67 @@
import sys
def check_duplicates():
auto_content = ""
manual_content = ""
with open("src/intrinsic/llvm.rs", "r", encoding="utf8") as f:
manual_content = f.read()
with open("src/intrinsic/archs.rs", "r", encoding="utf8") as f:
auto_content = f.read()
intrinsics_map = {}
for line in auto_content.splitlines():
line = line.strip()
if not line.startswith('"'):
continue
parts = line.split('"')
if len(parts) != 5:
continue
intrinsics_map[parts[1]] = parts[3]
if len(intrinsics_map) == 0:
print("No intrinsics found in auto code... Aborting.")
return 1
print("Found {} intrinsics in auto code".format(len(intrinsics_map)))
errors = []
lines = manual_content.splitlines()
pos = 0
found = 0
while pos < len(lines):
line = lines[pos].strip()
# This is our marker.
if line == "let gcc_name = match name {":
while pos < len(lines):
line = lines[pos].strip()
pos += 1
if line == "};":
# We're done!
if found == 0:
print("No intrinsics found in manual code even though we found the "
"marker... Aborting...")
return 1
for error in errors:
print("ERROR => {}".format(error))
return 1 if len(errors) != 0 else 0
parts = line.split('"')
if len(parts) != 5:
continue
found += 1
if parts[1] in intrinsics_map:
if parts[3] != intrinsics_map[parts[1]]:
print("Same intrinsics (`{}` at line {}) but different GCC "
"translations: `{}` != `{}`".format(
parts[1], pos, intrinsics_map[parts[1]], parts[3]))
else:
errors.append("Duplicated intrinsics: `{}` at line {}. Please remove it "
" from manual code".format(parts[1], pos))
# Weird but whatever...
return 1 if len(errors) != 0 else 0
pos += 1
print("No intrinsics found in manual code... Aborting")
return 1
if __name__ == "__main__":
sys.exit(check_duplicates())

View File

@ -13,7 +13,7 @@ def run_command(command, cwd=None):
sys.exit(1)
def clone_repository(repo_name, path, repo_url, sub_path=None):
def clone_repository(repo_name, path, repo_url, sub_paths=None):
if os.path.exists(path):
while True:
choice = input("There is already a `{}` folder, do you want to update it? [y/N]".format(path))
@ -27,12 +27,12 @@ def clone_repository(repo_name, path, repo_url, sub_path=None):
else:
print("Didn't understand answer...")
print("Cloning {} repository...".format(repo_name))
if sub_path is None:
if sub_paths is None:
run_command(["git", "clone", repo_url, "--depth", "1", path])
else:
run_command(["git", "clone", repo_url, "--filter=tree:0", "--no-checkout", path])
run_command(["git", "sparse-checkout", "init"], cwd=path)
run_command(["git", "sparse-checkout", "set", "add", sub_path], cwd=path)
run_command(["git", "sparse-checkout", "set", *sub_paths], cwd=path)
run_command(["git", "checkout"], cwd=path)
@ -40,56 +40,45 @@ def append_intrinsic(array, intrinsic_name, translation):
array.append((intrinsic_name, translation))
def extract_instrinsics(intrinsics, file):
print("Extracting intrinsics from `{}`...".format(file))
with open(file, "r", encoding="utf8") as f:
content = f.read()
lines = content.splitlines()
pos = 0
current_arch = None
while pos < len(lines):
line = lines[pos].strip()
if line.startswith("let TargetPrefix ="):
current_arch = line.split('"')[1].strip()
if len(current_arch) == 0:
current_arch = None
elif current_arch is None:
pass
elif line == "}":
current_arch = None
elif line.startswith("def "):
content = ""
while not content.endswith(";") and not content.endswith("}") and pos < len(lines):
line = lines[pos].split(" // ")[0].strip()
content += line
pos += 1
entries = re.findall('GCCBuiltin<"(\\w+)">', content)
if len(entries) > 0:
intrinsic = content.split("def ")[1].strip().split(":")[0].strip()
intrinsic = intrinsic.split("_")
if len(intrinsic) < 2 or intrinsic[0] != "int":
continue
intrinsic[0] = "llvm"
intrinsic = ".".join(intrinsic)
if current_arch not in intrinsics:
intrinsics[current_arch] = []
for entry in entries:
append_intrinsic(intrinsics[current_arch], intrinsic, entry)
continue
pos += 1
continue
print("Done!")
def convert_to_string(content):
if content.__class__.__name__ == 'bytes':
return content.decode('utf-8')
return content
def extract_instrinsics_from_llvm(llvm_path, intrinsics):
files = []
intrinsics_path = os.path.join(llvm_path, "llvm/include/llvm/IR")
for (dirpath, dirnames, filenames) in walk(intrinsics_path):
files.extend([os.path.join(intrinsics_path, f) for f in filenames if f.endswith(".td")])
for file in files:
extract_instrinsics(intrinsics, file)
p = subprocess.Popen(
["llvm-tblgen", "llvm/IR/Intrinsics.td"],
cwd=os.path.join(llvm_path, "llvm/include"),
stdout=subprocess.PIPE)
output, err = p.communicate()
lines = convert_to_string(output).splitlines()
pos = 0
while pos < len(lines):
line = lines[pos]
if not line.startswith("def "):
pos += 1
continue
intrinsic = line.split(" ")[1].strip()
content = line
while pos < len(lines):
line = lines[pos].split(" // ")[0].strip()
content += line
pos += 1
if line == "}":
break
entries = re.findall('string ClangBuiltinName = "(\\w+)";', content)
current_arch = re.findall('string TargetPrefix = "(\\w+)";', content)
if len(entries) == 1 and len(current_arch) == 1:
current_arch = current_arch[0]
intrinsic = intrinsic.split("_")
if len(intrinsic) < 2 or intrinsic[0] != "int":
continue
intrinsic[0] = "llvm"
intrinsic = ".".join(intrinsic)
if current_arch not in intrinsics:
intrinsics[current_arch] = []
append_intrinsic(intrinsics[current_arch], intrinsic, entries[0])
def append_translation(json_data, p, array):
@ -193,6 +182,8 @@ def update_intrinsics(llvm_path, llvmint, llvmint2):
for entry in intrinsics[arch]:
if entry[2] == True: # if it is a duplicate
out.write(' // [DUPLICATE]: "{}" => "{}",\n'.format(entry[0], entry[1]))
elif "_round_mask" in entry[1]:
out.write(' // [INVALID CONVERSION]: "{}" => "{}",\n'.format(entry[0], entry[1]))
else:
out.write(' "{}" => "{}",\n'.format(entry[0], entry[1]))
out.write(' _ => unimplemented!("***** unsupported LLVM intrinsic {}", name),\n')
@ -219,7 +210,7 @@ def main():
"llvm-project",
llvm_path,
"https://github.com/llvm/llvm-project",
sub_path="llvm/include/llvm/IR",
sub_paths=["llvm/include/llvm/IR", "llvm/include/llvm/CodeGen/"],
)
clone_repository(
"llvmint",

View File

@ -735,9 +735,13 @@ impl<'ll> CodegenCx<'ll, '_> {
ifn!("llvm.copysign.f32", fn(t_f32, t_f32) -> t_f32);
ifn!("llvm.copysign.f64", fn(t_f64, t_f64) -> t_f64);
ifn!("llvm.round.f32", fn(t_f32) -> t_f32);
ifn!("llvm.round.f64", fn(t_f64) -> t_f64);
ifn!("llvm.roundeven.f32", fn(t_f32) -> t_f32);
ifn!("llvm.roundeven.f64", fn(t_f64) -> t_f64);
ifn!("llvm.rint.f32", fn(t_f32) -> t_f32);
ifn!("llvm.rint.f64", fn(t_f64) -> t_f64);
ifn!("llvm.nearbyint.f32", fn(t_f32) -> t_f32);

View File

@ -71,6 +71,8 @@ fn get_simple_intrinsic<'ll>(
sym::roundf32 => "llvm.round.f32",
sym::roundf64 => "llvm.round.f64",
sym::ptr_mask => "llvm.ptrmask",
sym::roundevenf32 => "llvm.roundeven.f32",
sym::roundevenf64 => "llvm.roundeven.f64",
_ => return None,
};
Some(cx.get_intrinsic(llvm_name))

View File

@ -640,9 +640,6 @@ extern "C" {
pub struct Builder<'a>(InvariantOpaque<'a>);
#[repr(C)]
pub struct PassManager<'a>(InvariantOpaque<'a>);
extern "C" {
pub type PassManagerBuilder;
}
extern "C" {
pub type Pass;
}

View File

@ -58,7 +58,7 @@ fn reachable_non_generics_provider(tcx: TyCtxt<'_>, cnum: CrateNum) -> DefIdMap<
let mut reachable_non_generics: DefIdMap<_> = tcx
.reachable_set(())
.iter()
.items()
.filter_map(|&def_id| {
// We want to ignore some FFI functions that are not exposed from
// this crate. Reachable FFI functions can be lumped into two
@ -136,7 +136,7 @@ fn reachable_non_generics_provider(tcx: TyCtxt<'_>, cnum: CrateNum) -> DefIdMap<
};
(def_id.to_def_id(), info)
})
.collect();
.into();
if let Some(id) = tcx.proc_macro_decls_static(()) {
reachable_non_generics.insert(

View File

@ -61,25 +61,29 @@ fn codegen_fn_attrs(tcx: TyCtxt<'_>, did: DefId) -> CodegenFnAttrs {
let supported_target_features = tcx.supported_target_features(LOCAL_CRATE);
// In some cases, attribute are only valid on functions, but it's the `check_attr`
// pass that check that they aren't used anywhere else, rather this module.
// In these cases, we bail from performing further checks that are only meaningful for
// functions (such as calling `fn_sig`, which ICEs if given a non-function). We also
// report a delayed bug, just in case `check_attr` isn't doing its job.
let validate_fn_only_attr = |attr_sp| -> bool {
let def_kind = tcx.def_kind(did);
if let DefKind::Fn | DefKind::AssocFn | DefKind::Variant | DefKind::Ctor(..) = def_kind {
true
} else {
tcx.sess.delay_span_bug(attr_sp, "this attribute can only be applied to functions");
false
}
};
let mut inline_span = None;
let mut link_ordinal_span = None;
let mut no_sanitize_span = None;
for attr in attrs.iter() {
// In some cases, attribute are only valid on functions, but it's the `check_attr`
// pass that check that they aren't used anywhere else, rather this module.
// In these cases, we bail from performing further checks that are only meaningful for
// functions (such as calling `fn_sig`, which ICEs if given a non-function). We also
// report a delayed bug, just in case `check_attr` isn't doing its job.
let fn_sig = || {
use DefKind::*;
let def_kind = tcx.def_kind(did);
if let Fn | AssocFn | Variant | Ctor(..) = def_kind {
Some(tcx.fn_sig(did))
} else {
tcx.sess
.delay_span_bug(attr.span, "this attribute can only be applied to functions");
None
}
};
if attr.has_name(sym::cold) {
codegen_fn_attrs.flags |= CodegenFnAttrFlags::COLD;
} else if attr.has_name(sym::rustc_allocator) {
@ -169,8 +173,8 @@ fn codegen_fn_attrs(tcx: TyCtxt<'_>, did: DefId) -> CodegenFnAttrs {
}
}
} else if attr.has_name(sym::cmse_nonsecure_entry) {
if validate_fn_only_attr(attr.span)
&& !matches!(tcx.fn_sig(did).skip_binder().abi(), abi::Abi::C { .. })
if let Some(fn_sig) = fn_sig()
&& !matches!(fn_sig.skip_binder().abi(), abi::Abi::C { .. })
{
struct_span_err!(
tcx.sess,
@ -189,8 +193,8 @@ fn codegen_fn_attrs(tcx: TyCtxt<'_>, did: DefId) -> CodegenFnAttrs {
codegen_fn_attrs.flags |= CodegenFnAttrFlags::THREAD_LOCAL;
} else if attr.has_name(sym::track_caller) {
if !tcx.is_closure(did.to_def_id())
&& validate_fn_only_attr(attr.span)
&& tcx.fn_sig(did).skip_binder().abi() != abi::Abi::Rust
&& let Some(fn_sig) = fn_sig()
&& fn_sig.skip_binder().abi() != abi::Abi::Rust
{
struct_span_err!(tcx.sess, attr.span, E0737, "`#[track_caller]` requires Rust ABI")
.emit();
@ -222,7 +226,8 @@ fn codegen_fn_attrs(tcx: TyCtxt<'_>, did: DefId) -> CodegenFnAttrs {
}
} else if attr.has_name(sym::target_feature) {
if !tcx.is_closure(did.to_def_id())
&& tcx.fn_sig(did).skip_binder().unsafety() == hir::Unsafety::Normal
&& let Some(fn_sig) = fn_sig()
&& fn_sig.skip_binder().unsafety() == hir::Unsafety::Normal
{
if tcx.sess.target.is_like_wasm || tcx.sess.opts.actually_rustdoc {
// The `#[target_feature]` attribute is allowed on

View File

@ -295,7 +295,6 @@ pub fn cleanup_kinds(mir: &mir::Body<'_>) -> IndexVec<mir::BasicBlock, CleanupKi
TerminatorKind::Call { cleanup: unwind, .. }
| TerminatorKind::InlineAsm { cleanup: unwind, .. }
| TerminatorKind::Assert { cleanup: unwind, .. }
| TerminatorKind::DropAndReplace { unwind, .. }
| TerminatorKind::Drop { unwind, .. } => {
if let Some(unwind) = unwind {
debug!(

View File

@ -1305,10 +1305,6 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
mergeable_succ(),
),
mir::TerminatorKind::DropAndReplace { .. } => {
bug!("undesugared DropAndReplace in codegen: {:?}", terminator);
}
mir::TerminatorKind::Call {
ref func,
ref args,

View File

@ -92,6 +92,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
| mir::StatementKind::Retag { .. }
| mir::StatementKind::AscribeUserType(..)
| mir::StatementKind::ConstEvalCounter
| mir::StatementKind::PlaceMention(..)
| mir::StatementKind::Nop => {}
}
}

View File

@ -1,7 +1,7 @@
use rustc_ast::ast;
use rustc_attr::InstructionSetAttr;
use rustc_data_structures::fx::FxHashMap;
use rustc_data_structures::fx::FxHashSet;
use rustc_data_structures::fx::FxIndexSet;
use rustc_errors::Applicability;
use rustc_hir::def::DefKind;
use rustc_hir::def_id::DefId;
@ -418,7 +418,7 @@ pub fn from_target_feature(
/// Computes the set of target features used in a function for the purposes of
/// inline assembly.
fn asm_target_features(tcx: TyCtxt<'_>, did: DefId) -> &FxHashSet<Symbol> {
fn asm_target_features(tcx: TyCtxt<'_>, did: DefId) -> &FxIndexSet<Symbol> {
let mut target_features = tcx.sess.unstable_target_features.clone();
if tcx.def_kind(did).has_codegen_attrs() {
let attrs = tcx.codegen_fn_attrs(did);

View File

@ -1,3 +1,4 @@
use rustc_attr as attr;
use rustc_hir as hir;
use rustc_hir::def::DefKind;
use rustc_hir::def_id::{DefId, LocalDefId};
@ -5,11 +6,17 @@ use rustc_middle::ty::query::Providers;
use rustc_middle::ty::TyCtxt;
use rustc_span::symbol::Symbol;
/// Whether the `def_id` is an unstable const fn and what feature gate is necessary to enable it
pub fn is_unstable_const_fn(tcx: TyCtxt<'_>, def_id: DefId) -> Option<Symbol> {
/// Whether the `def_id` is an unstable const fn and what feature gate(s) are necessary to enable
/// it.
pub fn is_unstable_const_fn(tcx: TyCtxt<'_>, def_id: DefId) -> Option<(Symbol, Option<Symbol>)> {
if tcx.is_const_fn_raw(def_id) {
let const_stab = tcx.lookup_const_stability(def_id)?;
if const_stab.is_const_unstable() { Some(const_stab.feature) } else { None }
match const_stab.level {
attr::StabilityLevel::Unstable { implied_by, .. } => {
Some((const_stab.feature, implied_by))
}
attr::StabilityLevel::Stable { .. } => None,
}
} else {
None
}

View File

@ -67,12 +67,12 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
}
Pointer(PointerCast::ReifyFnPointer) => {
// All reifications must be monomorphic, bail out otherwise.
ensure_monomorphic_enough(*self.tcx, src.layout.ty)?;
// The src operand does not matter, just its type
match *src.layout.ty.kind() {
ty::FnDef(def_id, substs) => {
// All reifications must be monomorphic, bail out otherwise.
ensure_monomorphic_enough(*self.tcx, src.layout.ty)?;
let instance = ty::Instance::resolve_for_fn_ptr(
*self.tcx,
self.param_env,
@ -100,12 +100,12 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
}
Pointer(PointerCast::ClosureFnPointer(_)) => {
// All reifications must be monomorphic, bail out otherwise.
ensure_monomorphic_enough(*self.tcx, src.layout.ty)?;
// The src operand does not matter, just its type
match *src.layout.ty.kind() {
ty::Closure(def_id, substs) => {
// All reifications must be monomorphic, bail out otherwise.
ensure_monomorphic_enough(*self.tcx, src.layout.ty)?;
let instance = ty::Instance::resolve_closure(
*self.tcx,
def_id,
@ -359,8 +359,11 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
let val = Immediate::new_dyn_trait(ptr, vtable, &*self.tcx);
self.write_immediate(val, dest)
}
_ => {
// Do not ICE if we are not monomorphic enough.
ensure_monomorphic_enough(*self.tcx, src.layout.ty)?;
ensure_monomorphic_enough(*self.tcx, cast_ty)?;
span_bug!(
self.cur_span(),
"invalid pointer unsizing {:?} -> {:?}",
@ -404,12 +407,18 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
}
Ok(())
}
_ => span_bug!(
self.cur_span(),
"unsize_into: invalid conversion: {:?} -> {:?}",
src.layout,
dest.layout
),
_ => {
// Do not ICE if we are not monomorphic enough.
ensure_monomorphic_enough(*self.tcx, src.layout.ty)?;
ensure_monomorphic_enough(*self.tcx, cast_ty.ty)?;
span_bug!(
self.cur_span(),
"unsize_into: invalid conversion: {:?} -> {:?}",
src.layout,
dest.layout
)
}
}
}
}

Some files were not shown because too many files have changed in this diff Show More