rust/src/librustc_trans/base.rs

1933 lines
72 KiB
Rust
Raw Normal View History

// Copyright 2012-2015 The Rust Project Developers. See the COPYRIGHT
// file at the top-level directory of this distribution and at
// http://rust-lang.org/COPYRIGHT.
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
// option. This file may not be copied, modified, or distributed
// except according to those terms.
//! Translate the completed AST to the LLVM IR.
//!
//! Some functions here, such as trans_block and trans_expr, return a value --
//! the result of the translation to LLVM -- while others, such as trans_fn
//! and trans_item, are called only for the side effect of adding a
//! particular definition to the LLVM IR output we're producing.
//!
//! Hopefully useful general knowledge about trans:
//!
//! * There's no way to find out the Ty type of a ValueRef. Doing so
//! would be "trying to get the eggs out of an omelette" (credit:
//! pcwalton). You can, instead, find out its TypeRef by calling val_ty,
//! but one TypeRef corresponds to many `Ty`s; for instance, tup(int, int,
//! int) and rec(x=int, y=int, z=int) will have the same TypeRef.
#![allow(non_camel_case_types)]
use super::CrateTranslation;
use super::ModuleLlvm;
use super::ModuleSource;
use super::ModuleTranslation;
use assert_module_sources;
use back::link;
use back::linker::LinkerInfo;
use llvm::{Linkage, ValueRef, Vector, get_param};
use llvm;
use rustc::hir::def::Def;
use rustc::hir::def_id::DefId;
use middle::lang_items::{LangItem, ExchangeMallocFnLangItem, StartFnLangItem};
use rustc::ty::subst::Substs;
use rustc::traits;
use rustc::ty::{self, Ty, TyCtxt, TypeFoldable};
use rustc::ty::adjustment::CustomCoerceUnsized;
use rustc::dep_graph::{DepNode, WorkProduct};
2016-03-29 08:50:44 +03:00
use rustc::hir::map as hir_map;
use rustc::util::common::time;
use session::config::{self, NoDebugInfo};
use rustc_incremental::IncrementalHashesMap;
use session::Session;
use abi::{self, Abi, FnType};
use adt;
use attributes;
use build::*;
use builder::{Builder, noname};
use callee::{Callee};
use common::{Block, C_bool, C_bytes_in_context, C_i32, C_uint};
use collector::{self, TransItemCollectionMode};
use common::{C_null, C_struct_in_context, C_u64, C_u8, C_undef};
use common::{CrateContext, FunctionContext};
use common::{Result};
use common::{fulfill_obligation};
use common::{type_is_zero_size, val_ty};
use common;
use consts;
use context::{SharedCrateContext, CrateContextList};
use debuginfo::{self, DebugLoc};
use declare;
use machine;
use machine::{llalign_of_min, llsize_of};
use meth;
use mir;
use monomorphize::{self, Instance};
use partitioning::{self, PartitioningStrategy, CodegenUnit};
use symbol_map::SymbolMap;
use symbol_names_test;
use trans_item::{TransItem, DefPathBasedNames};
use type_::Type;
use type_of;
use value::Value;
use Disr;
use util::nodemap::{NodeSet, FxHashMap, FxHashSet};
2013-06-16 22:52:44 +12:00
use arena::TypedArena;
use libc::c_uint;
std: Implement CString-related RFCs This commit is an implementation of [RFC 592][r592] and [RFC 840][r840]. These two RFCs tweak the behavior of `CString` and add a new `CStr` unsized slice type to the module. [r592]: https://github.com/rust-lang/rfcs/blob/master/text/0592-c-str-deref.md [r840]: https://github.com/rust-lang/rfcs/blob/master/text/0840-no-panic-in-c-string.md The new `CStr` type is only constructable via two methods: 1. By `deref`'ing from a `CString` 2. Unsafely via `CStr::from_ptr` The purpose of `CStr` is to be an unsized type which is a thin pointer to a `libc::c_char` (currently it is a fat pointer slice due to implementation limitations). Strings from C can be safely represented with a `CStr` and an appropriate lifetime as well. Consumers of `&CString` should now consume `&CStr` instead to allow producers to pass in C-originating strings instead of just Rust-allocated strings. A new constructor was added to `CString`, `new`, which takes `T: IntoBytes` instead of separate `from_slice` and `from_vec` methods (both have been deprecated in favor of `new`). The `new` method returns a `Result` instead of panicking. The error variant contains the relevant information about where the error happened and bytes (if present). Conversions are provided to the `io::Error` and `old_io::IoError` types via the `FromError` trait which translate to `InvalidInput`. This is a breaking change due to the modification of existing `#[unstable]` APIs and new deprecation, and more detailed information can be found in the two RFCs. Notable breakage includes: * All construction of `CString` now needs to use `new` and handle the outgoing `Result`. * Usage of `CString` as a byte slice now explicitly needs a `.as_bytes()` call. * The `as_slice*` methods have been removed in favor of just having the `as_bytes*` methods. Closes #22469 Closes #22470 [breaking-change]
2015-02-17 22:47:40 -08:00
use std::ffi::{CStr, CString};
use std::borrow::Cow;
use std::cell::{Cell, RefCell};
use std::ptr;
use std::rc::Rc;
use std::str;
use std::i32;
use syntax_pos::{Span, DUMMY_SP};
use syntax::attr;
2016-03-29 08:50:44 +03:00
use rustc::hir;
2015-07-31 00:04:06 -07:00
use syntax::ast;
thread_local! {
static TASK_LOCAL_INSN_KEY: RefCell<Option<Vec<&'static str>>> = {
RefCell::new(None)
}
}
2015-11-19 12:36:31 +01:00
pub fn with_insn_ctxt<F>(blk: F)
where F: FnOnce(&[&'static str])
2014-12-09 13:44:51 -05:00
{
TASK_LOCAL_INSN_KEY.with(move |slot| {
slot.borrow().as_ref().map(move |s| blk(s));
})
}
pub fn init_insn_ctxt() {
TASK_LOCAL_INSN_KEY.with(|slot| {
*slot.borrow_mut() = Some(Vec::new());
});
2012-09-05 15:58:43 -07:00
}
pub struct _InsnCtxt {
2015-11-19 12:36:31 +01:00
_cannot_construct_outside_of_this_module: (),
}
impl Drop for _InsnCtxt {
2013-09-16 21:18:07 -04:00
fn drop(&mut self) {
TASK_LOCAL_INSN_KEY.with(|slot| {
if let Some(ctx) = slot.borrow_mut().as_mut() {
ctx.pop();
}
})
}
}
pub fn push_ctxt(s: &'static str) -> _InsnCtxt {
debug!("new InsnCtxt: {}", s);
TASK_LOCAL_INSN_KEY.with(|slot| {
if let Some(ctx) = slot.borrow_mut().as_mut() {
ctx.push(s)
}
});
2015-11-19 12:36:31 +01:00
_InsnCtxt {
_cannot_construct_outside_of_this_module: (),
}
}
pub struct StatRecorder<'a, 'tcx: 'a> {
ccx: &'a CrateContext<'a, 'tcx>,
name: Option<String>,
istart: usize,
}
impl<'a, 'tcx> StatRecorder<'a, 'tcx> {
2015-11-19 12:36:31 +01:00
pub fn new(ccx: &'a CrateContext<'a, 'tcx>, name: String) -> StatRecorder<'a, 'tcx> {
2014-09-05 09:18:53 -07:00
let istart = ccx.stats().n_llvm_insns.get();
StatRecorder {
ccx: ccx,
name: Some(name),
istart: istart,
}
}
}
impl<'a, 'tcx> Drop for StatRecorder<'a, 'tcx> {
2013-09-16 21:18:07 -04:00
fn drop(&mut self) {
2014-03-05 16:36:01 +02:00
if self.ccx.sess().trans_stats() {
2014-09-05 09:18:53 -07:00
let iend = self.ccx.stats().n_llvm_insns.get();
2015-11-19 12:36:31 +01:00
self.ccx
.stats()
.fn_stats
.borrow_mut()
.push((self.name.take().unwrap(), iend - self.istart));
2014-09-05 09:18:53 -07:00
self.ccx.stats().n_fns.set(self.ccx.stats().n_fns.get() + 1);
// Reset LLVM insn count to avoid compound costs.
2014-09-05 09:18:53 -07:00
self.ccx.stats().n_llvm_insns.set(self.istart);
}
}
}
pub fn get_meta(bcx: Block, fat_ptr: ValueRef) -> ValueRef {
StructGEP(bcx, fat_ptr, abi::FAT_PTR_EXTRA)
}
pub fn get_dataptr(bcx: Block, fat_ptr: ValueRef) -> ValueRef {
StructGEP(bcx, fat_ptr, abi::FAT_PTR_ADDR)
}
pub fn get_meta_builder(b: &Builder, fat_ptr: ValueRef) -> ValueRef {
b.struct_gep(fat_ptr, abi::FAT_PTR_EXTRA)
}
pub fn get_dataptr_builder(b: &Builder, fat_ptr: ValueRef) -> ValueRef {
b.struct_gep(fat_ptr, abi::FAT_PTR_ADDR)
}
2015-11-19 12:36:31 +01:00
fn require_alloc_fn<'blk, 'tcx>(bcx: Block<'blk, 'tcx>, info_ty: Ty<'tcx>, it: LangItem) -> DefId {
2014-04-06 13:54:41 +03:00
match bcx.tcx().lang_items.require(it) {
Ok(id) => id,
Err(s) => {
2015-06-18 20:25:05 +03:00
bcx.sess().fatal(&format!("allocation of `{}` {}", info_ty, s));
}
}
2014-04-06 13:54:41 +03:00
}
2014-04-06 13:54:41 +03:00
// The following malloc_raw_dyn* functions allocate a box to contain
// a given type, but with a potentially dynamic size.
pub fn malloc_raw_dyn<'blk, 'tcx>(bcx: Block<'blk, 'tcx>,
llty_ptr: Type,
info_ty: Ty<'tcx>,
size: ValueRef,
align: ValueRef,
debug_loc: DebugLoc)
-> Result<'blk, 'tcx> {
2014-04-06 13:54:41 +03:00
let _icx = push_ctxt("malloc_raw_exchange");
2014-04-06 13:54:41 +03:00
// Allocate space:
let def_id = require_alloc_fn(bcx, info_ty, ExchangeMallocFnLangItem);
let r = Callee::def(bcx.ccx(), def_id, bcx.tcx().intern_substs(&[]))
.call(bcx, debug_loc, &[size, align], None);
2014-05-03 23:14:56 +12:00
Result::new(r.bcx, PointerCast(r.bcx, r.val, llty_ptr))
}
pub fn bin_op_to_icmp_predicate(op: hir::BinOp_,
2015-11-19 12:36:31 +01:00
signed: bool)
-> llvm::IntPredicate {
match op {
2015-07-31 00:04:06 -07:00
hir::BiEq => llvm::IntEQ,
hir::BiNe => llvm::IntNE,
hir::BiLt => if signed { llvm::IntSLT } else { llvm::IntULT },
hir::BiLe => if signed { llvm::IntSLE } else { llvm::IntULE },
hir::BiGt => if signed { llvm::IntSGT } else { llvm::IntUGT },
hir::BiGe => if signed { llvm::IntSGE } else { llvm::IntUGE },
op => {
bug!("comparison_op_to_icmp_predicate: expected comparison operator, \
found {:?}",
op)
}
}
}
pub fn bin_op_to_fcmp_predicate(op: hir::BinOp_) -> llvm::RealPredicate {
match op {
2015-07-31 00:04:06 -07:00
hir::BiEq => llvm::RealOEQ,
hir::BiNe => llvm::RealUNE,
hir::BiLt => llvm::RealOLT,
hir::BiLe => llvm::RealOLE,
hir::BiGt => llvm::RealOGT,
hir::BiGe => llvm::RealOGE,
op => {
bug!("comparison_op_to_fcmp_predicate: expected comparison operator, \
found {:?}",
op);
}
}
}
pub fn compare_simd_types<'blk, 'tcx>(bcx: Block<'blk, 'tcx>,
lhs: ValueRef,
rhs: ValueRef,
t: Ty<'tcx>,
ret_ty: Type,
2015-07-31 00:04:06 -07:00
op: hir::BinOp_,
debug_loc: DebugLoc)
-> ValueRef {
let signed = match t.sty {
ty::TyFloat(_) => {
let cmp = bin_op_to_fcmp_predicate(op);
return SExt(bcx, FCmp(bcx, cmp, lhs, rhs, debug_loc), ret_ty);
},
ty::TyUint(_) => false,
ty::TyInt(_) => true,
_ => bug!("compare_simd_types: invalid SIMD type"),
};
let cmp = bin_op_to_icmp_predicate(op, signed);
// LLVM outputs an `< size x i1 >`, so we need to perform a sign extension
// to get the correctly sized type. This will compile to a single instruction
// once the IR is converted to assembly if the SIMD instruction is supported
// by the target architecture.
SExt(bcx, ICmp(bcx, cmp, lhs, rhs, debug_loc), ret_ty)
}
2015-11-11 22:02:51 +02:00
/// Retrieve the information we are losing (making dynamic) in an unsizing
/// adjustment.
///
/// The `old_info` argument is a bit funny. It is intended for use
/// in an upcast, where the new vtable for an object will be drived
/// from the old one.
pub fn unsized_info<'ccx, 'tcx>(ccx: &CrateContext<'ccx, 'tcx>,
source: Ty<'tcx>,
target: Ty<'tcx>,
old_info: Option<ValueRef>)
2015-11-11 22:02:51 +02:00
-> ValueRef {
let (source, target) = ccx.tcx().struct_lockstep_tails(source, target);
match (&source.sty, &target.sty) {
(&ty::TyArray(_, len), &ty::TySlice(_)) => C_uint(ccx, len),
(&ty::TyTrait(_), &ty::TyTrait(_)) => {
// For now, upcasts are limited to changes in marker
// traits, and hence never actually require an actual
// change to the vtable.
old_info.expect("unsized_info: missing old info for trait upcast")
}
(_, &ty::TyTrait(ref data)) => {
let trait_ref = data.principal.with_self_ty(ccx.tcx(), source);
let trait_ref = ccx.tcx().erase_regions(&trait_ref);
consts::ptrcast(meth::get_vtable(ccx, trait_ref),
2015-11-11 22:02:51 +02:00
Type::vtable_ptr(ccx))
}
_ => bug!("unsized_info: invalid unsizing {:?} -> {:?}",
2015-11-11 22:02:51 +02:00
source,
target),
2015-11-11 22:02:51 +02:00
}
}
/// Coerce `src` to `dst_ty`. `src_ty` must be a thin pointer.
pub fn unsize_thin_ptr<'blk, 'tcx>(bcx: Block<'blk, 'tcx>,
src: ValueRef,
src_ty: Ty<'tcx>,
dst_ty: Ty<'tcx>)
-> (ValueRef, ValueRef) {
debug!("unsize_thin_ptr: {:?} => {:?}", src_ty, dst_ty);
match (&src_ty.sty, &dst_ty.sty) {
(&ty::TyBox(a), &ty::TyBox(b)) |
(&ty::TyRef(_, ty::TypeAndMut { ty: a, .. }),
&ty::TyRef(_, ty::TypeAndMut { ty: b, .. })) |
(&ty::TyRef(_, ty::TypeAndMut { ty: a, .. }),
&ty::TyRawPtr(ty::TypeAndMut { ty: b, .. })) |
(&ty::TyRawPtr(ty::TypeAndMut { ty: a, .. }),
&ty::TyRawPtr(ty::TypeAndMut { ty: b, .. })) => {
assert!(common::type_is_sized(bcx.tcx(), a));
let ptr_ty = type_of::in_memory_type_of(bcx.ccx(), b).ptr_to();
(PointerCast(bcx, src, ptr_ty),
unsized_info(bcx.ccx(), a, b, None))
2015-11-11 22:02:51 +02:00
}
_ => bug!("unsize_thin_ptr: called on bad types"),
2015-11-11 22:02:51 +02:00
}
}
/// Coerce `src`, which is a reference to a value of type `src_ty`,
/// to a value of type `dst_ty` and store the result in `dst`
pub fn coerce_unsized_into<'blk, 'tcx>(bcx: Block<'blk, 'tcx>,
src: ValueRef,
src_ty: Ty<'tcx>,
dst: ValueRef,
dst_ty: Ty<'tcx>) {
match (&src_ty.sty, &dst_ty.sty) {
(&ty::TyBox(..), &ty::TyBox(..)) |
(&ty::TyRef(..), &ty::TyRef(..)) |
(&ty::TyRef(..), &ty::TyRawPtr(..)) |
(&ty::TyRawPtr(..), &ty::TyRawPtr(..)) => {
let (base, info) = if common::type_is_fat_ptr(bcx.tcx(), src_ty) {
// fat-ptr to fat-ptr unsize preserves the vtable
// i.e. &'a fmt::Debug+Send => &'a fmt::Debug
// So we need to pointercast the base to ensure
// the types match up.
let (base, info) = load_fat_ptr(bcx, src, src_ty);
let llcast_ty = type_of::fat_ptr_base_ty(bcx.ccx(), dst_ty);
let base = PointerCast(bcx, base, llcast_ty);
(base, info)
2015-11-11 22:02:51 +02:00
} else {
let base = load_ty(bcx, src, src_ty);
unsize_thin_ptr(bcx, base, src_ty, dst_ty)
};
store_fat_ptr(bcx, base, info, dst, dst_ty);
}
(&ty::TyAdt(def_a, substs_a), &ty::TyAdt(def_b, substs_b)) => {
2015-11-11 22:02:51 +02:00
assert_eq!(def_a, def_b);
let src_fields = def_a.variants[0].fields.iter().map(|f| {
monomorphize::field_ty(bcx.tcx(), substs_a, f)
});
let dst_fields = def_b.variants[0].fields.iter().map(|f| {
monomorphize::field_ty(bcx.tcx(), substs_b, f)
});
2015-11-11 22:02:51 +02:00
let src = adt::MaybeSizedValue::sized(src);
let dst = adt::MaybeSizedValue::sized(dst);
let iter = src_fields.zip(dst_fields).enumerate();
2015-11-11 22:02:51 +02:00
for (i, (src_fty, dst_fty)) in iter {
2015-11-19 12:36:31 +01:00
if type_is_zero_size(bcx.ccx(), dst_fty) {
continue;
}
2015-11-11 22:02:51 +02:00
let src_f = adt::trans_field_ptr(bcx, src_ty, src, Disr(0), i);
let dst_f = adt::trans_field_ptr(bcx, dst_ty, dst, Disr(0), i);
2015-11-11 22:02:51 +02:00
if src_fty == dst_fty {
memcpy_ty(bcx, dst_f, src_f, src_fty);
} else {
2015-11-19 12:36:31 +01:00
coerce_unsized_into(bcx, src_f, src_fty, dst_f, dst_fty);
2015-11-11 22:02:51 +02:00
}
}
}
_ => bug!("coerce_unsized_into: invalid coercion {:?} -> {:?}",
src_ty,
dst_ty),
2015-11-11 22:02:51 +02:00
}
}
pub fn custom_coerce_unsize_info<'scx, 'tcx>(scx: &SharedCrateContext<'scx, 'tcx>,
source_ty: Ty<'tcx>,
target_ty: Ty<'tcx>)
-> CustomCoerceUnsized {
let trait_ref = ty::Binder(ty::TraitRef {
def_id: scx.tcx().lang_items.coerce_unsized_trait().unwrap(),
substs: scx.tcx().mk_substs_trait(source_ty, &[target_ty])
});
match fulfill_obligation(scx, DUMMY_SP, trait_ref) {
traits::VtableImpl(traits::VtableImplData { impl_def_id, .. }) => {
scx.tcx().custom_coerce_unsized_kind(impl_def_id)
}
vtable => {
bug!("invalid CoerceUnsized vtable: {:?}", vtable);
}
}
}
2015-11-19 12:36:31 +01:00
pub fn cast_shift_expr_rhs(cx: Block, op: hir::BinOp_, lhs: ValueRef, rhs: ValueRef) -> ValueRef {
cast_shift_rhs(op, lhs, rhs, |a, b| Trunc(cx, a, b), |a, b| ZExt(cx, a, b))
}
2015-11-19 12:36:31 +01:00
pub fn cast_shift_const_rhs(op: hir::BinOp_, lhs: ValueRef, rhs: ValueRef) -> ValueRef {
cast_shift_rhs(op,
lhs,
rhs,
2013-06-16 22:52:44 +12:00
|a, b| unsafe { llvm::LLVMConstTrunc(a, b.to_ref()) },
|a, b| unsafe { llvm::LLVMConstZExt(a, b.to_ref()) })
}
2015-07-31 00:04:06 -07:00
fn cast_shift_rhs<F, G>(op: hir::BinOp_,
lhs: ValueRef,
rhs: ValueRef,
trunc: F,
zext: G)
2015-11-19 12:36:31 +01:00
-> ValueRef
where F: FnOnce(ValueRef, Type) -> ValueRef,
G: FnOnce(ValueRef, Type) -> ValueRef
2014-12-09 13:44:51 -05:00
{
// Shifts may have any size int on the rhs
if op.is_shift() {
let mut rhs_llty = val_ty(rhs);
let mut lhs_llty = val_ty(lhs);
2015-11-19 12:36:31 +01:00
if rhs_llty.kind() == Vector {
rhs_llty = rhs_llty.element_type()
}
if lhs_llty.kind() == Vector {
lhs_llty = lhs_llty.element_type()
}
let rhs_sz = rhs_llty.int_width();
let lhs_sz = lhs_llty.int_width();
if lhs_sz < rhs_sz {
trunc(rhs, lhs_llty)
} else if lhs_sz > rhs_sz {
// FIXME (#1877: If shifting by negative
// values becomes not undefined then this is wrong.
zext(rhs, lhs_llty)
} else {
rhs
}
} else {
rhs
}
}
pub fn invoke<'blk, 'tcx>(bcx: Block<'blk, 'tcx>,
llfn: ValueRef,
llargs: &[ValueRef],
debug_loc: DebugLoc)
-> (ValueRef, Block<'blk, 'tcx>) {
let _icx = push_ctxt("invoke_");
if bcx.unreachable.get() {
return (C_null(Type::i8(bcx.ccx())), bcx);
}
if need_invoke(bcx) {
debug!("invoking {:?} at {:?}", Value(llfn), bcx.llbb);
2015-01-31 12:20:46 -05:00
for &llarg in llargs {
debug!("arg: {:?}", Value(llarg));
}
let normal_bcx = bcx.fcx.new_block("normal-return");
let landing_pad = bcx.fcx.get_landing_pad();
let llresult = Invoke(bcx,
llfn,
&llargs[..],
normal_bcx.llbb,
landing_pad,
debug_loc);
return (llresult, normal_bcx);
} else {
debug!("calling {:?} at {:?}", Value(llfn), bcx.llbb);
2015-01-31 12:20:46 -05:00
for &llarg in llargs {
debug!("arg: {:?}", Value(llarg));
}
let llresult = Call(bcx, llfn, &llargs[..], debug_loc);
return (llresult, bcx);
}
}
/// Returns whether this session's target will use SEH-based unwinding.
///
/// This is only true for MSVC targets, and even then the 64-bit MSVC target
/// currently uses SEH-ish unwinding with DWARF info tables to the side (same as
/// 64-bit MinGW) instead of "full SEH".
pub fn wants_msvc_seh(sess: &Session) -> bool {
trans: Reimplement unwinding on MSVC This commit transitions the compiler to using the new exception handling instructions in LLVM for implementing unwinding for MSVC. This affects both 32 and 64-bit MSVC as they're both now using SEH-based strategies. In terms of standard library support, lots more details about how SEH unwinding is implemented can be found in the commits. In terms of trans, this change necessitated a few modifications: * Branches were added to detect when the old landingpad instruction is used or the new cleanuppad instruction is used to `trans::cleanup`. * The return value from `cleanuppad` is not stored in an `alloca` (because it cannot be). * Each block in trans now has an `Option<LandingPad>` instead of `is_lpad: bool` for indicating whether it's in a landing pad or not. The new exception handling intrinsics require that on MSVC each `call` inside of a landing pad is annotated with which landing pad that it's in. This change to the basic block means that whenever a `call` or `invoke` instruction is generated we know whether to annotate it as part of a cleanuppad or not. * Lots of modifications were made to the instruction builders to construct the new instructions as well as pass the tagging information for the call/invoke instructions. * The translation of the `try` intrinsics for MSVC has been overhauled to use the new `catchpad` instruction. The filter function is now also a rustc-generated function instead of a purely libstd-defined function. The libstd definition still exists, it just has a stable ABI across architectures and leaves some of the really weird implementation details to the compiler (e.g. the `localescape` and `localrecover` intrinsics).
2015-10-23 18:18:44 -07:00
sess.target.target.options.is_like_msvc
}
pub fn avoid_invoke(bcx: Block) -> bool {
2016-02-08 11:53:06 +01:00
bcx.sess().no_landing_pads() || bcx.lpad().is_some()
}
pub fn need_invoke(bcx: Block) -> bool {
if avoid_invoke(bcx) {
false
} else {
bcx.fcx.needs_invoke()
}
}
pub fn call_assume<'a, 'tcx>(b: &Builder<'a, 'tcx>, val: ValueRef) {
let assume_intrinsic = b.ccx.get_intrinsic("llvm.assume");
b.call(assume_intrinsic, &[val], None);
}
/// Helper for loading values from memory. Does the necessary conversion if the in-memory type
/// differs from the type used for SSA values. Also handles various special cases where the type
/// gives us better information about what we are loading.
2015-11-19 12:36:31 +01:00
pub fn load_ty<'blk, 'tcx>(cx: Block<'blk, 'tcx>, ptr: ValueRef, t: Ty<'tcx>) -> ValueRef {
if cx.unreachable.get() {
return C_undef(type_of::type_of(cx.ccx(), t));
}
load_ty_builder(&B(cx), ptr, t)
}
pub fn load_ty_builder<'a, 'tcx>(b: &Builder<'a, 'tcx>, ptr: ValueRef, t: Ty<'tcx>) -> ValueRef {
let ccx = b.ccx;
if type_is_zero_size(ccx, t) {
return C_undef(type_of::type_of(ccx, t));
}
unsafe {
let global = llvm::LLVMIsAGlobalVariable(ptr);
if !global.is_null() && llvm::LLVMIsGlobalConstant(global) == llvm::True {
let val = llvm::LLVMGetInitializer(global);
if !val.is_null() {
if t.is_bool() {
return llvm::LLVMConstTrunc(val, Type::i1(ccx).to_ref());
}
return val;
}
}
}
if t.is_bool() {
b.trunc(b.load_range_assert(ptr, 0, 2, llvm::False), Type::i1(ccx))
} else if t.is_char() {
// a char is a Unicode codepoint, and so takes values from 0
// to 0x10FFFF inclusive only.
b.load_range_assert(ptr, 0, 0x10FFFF + 1, llvm::False)
} else if (t.is_region_ptr() || t.is_unique()) &&
!common::type_is_fat_ptr(ccx.tcx(), t) {
b.load_nonnull(ptr)
} else {
b.load(ptr)
}
}
/// Helper for storing values in memory. Does the necessary conversion if the in-memory type
/// differs from the type used for SSA values.
pub fn store_ty<'blk, 'tcx>(cx: Block<'blk, 'tcx>, v: ValueRef, dst: ValueRef, t: Ty<'tcx>) {
if cx.unreachable.get() {
return;
}
debug!("store_ty: {:?} : {:?} <- {:?}", Value(dst), t, Value(v));
if common::type_is_fat_ptr(cx.tcx(), t) {
let lladdr = ExtractValue(cx, v, abi::FAT_PTR_ADDR);
let llextra = ExtractValue(cx, v, abi::FAT_PTR_EXTRA);
store_fat_ptr(cx, lladdr, llextra, dst, t);
} else {
Store(cx, from_immediate(cx, v), dst);
}
}
pub fn store_fat_ptr<'blk, 'tcx>(cx: Block<'blk, 'tcx>,
data: ValueRef,
extra: ValueRef,
dst: ValueRef,
_ty: Ty<'tcx>) {
// FIXME: emit metadata
Store(cx, data, get_dataptr(cx, dst));
Store(cx, extra, get_meta(cx, dst));
}
pub fn load_fat_ptr<'blk, 'tcx>(cx: Block<'blk, 'tcx>,
src: ValueRef,
ty: Ty<'tcx>)
-> (ValueRef, ValueRef)
{
if cx.unreachable.get() {
// FIXME: remove me
return (Load(cx, get_dataptr(cx, src)),
Load(cx, get_meta(cx, src)));
}
load_fat_ptr_builder(&B(cx), src, ty)
}
pub fn load_fat_ptr_builder<'a, 'tcx>(
b: &Builder<'a, 'tcx>,
src: ValueRef,
t: Ty<'tcx>)
-> (ValueRef, ValueRef)
{
let ptr = get_dataptr_builder(b, src);
let ptr = if t.is_region_ptr() || t.is_unique() {
b.load_nonnull(ptr)
} else {
b.load(ptr)
};
// FIXME: emit metadata on `meta`.
let meta = b.load(get_meta_builder(b, src));
(ptr, meta)
}
pub fn from_immediate(bcx: Block, val: ValueRef) -> ValueRef {
if val_ty(val) == Type::i1(bcx.ccx()) {
ZExt(bcx, val, Type::i8(bcx.ccx()))
} else {
val
}
}
pub fn to_immediate(bcx: Block, val: ValueRef, ty: Ty) -> ValueRef {
if ty.is_bool() {
Trunc(bcx, val, Type::i1(bcx.ccx()))
} else {
val
}
}
2015-11-19 12:36:31 +01:00
pub fn with_cond<'blk, 'tcx, F>(bcx: Block<'blk, 'tcx>, val: ValueRef, f: F) -> Block<'blk, 'tcx>
where F: FnOnce(Block<'blk, 'tcx>) -> Block<'blk, 'tcx>
2014-12-09 13:44:51 -05:00
{
let _icx = push_ctxt("with_cond");
if bcx.unreachable.get() || common::const_to_opt_uint(val) == Some(0) {
return bcx;
}
let fcx = bcx.fcx;
let next_cx = fcx.new_block("next");
let cond_cx = fcx.new_block("cond");
CondBr(bcx, val, cond_cx.llbb, next_cx.llbb, DebugLoc::None);
let after_cx = f(cond_cx);
if !after_cx.terminated.get() {
Br(after_cx, next_cx.llbb, DebugLoc::None);
}
next_cx
}
pub enum Lifetime { Start, End }
// If LLVM lifetime intrinsic support is enabled (i.e. optimizations
// on), and `ptr` is nonzero-sized, then extracts the size of `ptr`
// and the intrinsic for `lt` and passes them to `emit`, which is in
// charge of generating code to call the passed intrinsic on whatever
// block of generated code is targetted for the intrinsic.
//
// If LLVM lifetime intrinsic support is disabled (i.e. optimizations
// off) or `ptr` is zero-sized, then no-op (does not call `emit`).
fn core_lifetime_emit<'blk, 'tcx, F>(ccx: &'blk CrateContext<'blk, 'tcx>,
ptr: ValueRef,
lt: Lifetime,
emit: F)
where F: FnOnce(&'blk CrateContext<'blk, 'tcx>, machine::llsize, ValueRef)
{
if ccx.sess().opts.optimize == config::OptLevel::No {
Emit LLVM lifetime intrinsics to improve stack usage and codegen in general Lifetime intrinsics help to reduce stack usage, because LLVM can apply stack coloring to reuse the stack slots of dead allocas for new ones. For example these functions now both use the same amount of stack, while previous `bar()` used five times as much as `foo()`: ````rust fn foo() { println("{}", 5); } fn bar() { println("{}", 5); println("{}", 5); println("{}", 5); println("{}", 5); println("{}", 5); } ```` On top of that, LLVM can also optimize out certain operations when it knows that memory is dead after a certain point. For example, it can sometimes remove the zeroing used to cancel the drop glue. This is possible when the glue drop itself was already removed because the zeroing dominated the drop glue call. For example in: ````rust pub fn bar(x: (Box<int>, int)) -> (Box<int>, int) { x } ```` With optimizations, this currently results in: ````llvm define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 { "_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit": %2 = bitcast { i64*, i64 }* %1 to i8* %3 = bitcast { i64*, i64 }* %0 to i8* tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false) tail call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 16, i32 8, i1 false) ret void } ```` But with lifetime intrinsics we get: ````llvm define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 { "_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit": %2 = bitcast { i64*, i64 }* %1 to i8* %3 = bitcast { i64*, i64 }* %0 to i8* tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false) tail call void @llvm.lifetime.end(i64 16, i8* %2) ret void } ```` Fixes #15665
2014-05-01 19:32:07 +02:00
return;
}
let _icx = push_ctxt(match lt {
Lifetime::Start => "lifetime_start",
Lifetime::End => "lifetime_end"
});
Emit LLVM lifetime intrinsics to improve stack usage and codegen in general Lifetime intrinsics help to reduce stack usage, because LLVM can apply stack coloring to reuse the stack slots of dead allocas for new ones. For example these functions now both use the same amount of stack, while previous `bar()` used five times as much as `foo()`: ````rust fn foo() { println("{}", 5); } fn bar() { println("{}", 5); println("{}", 5); println("{}", 5); println("{}", 5); println("{}", 5); } ```` On top of that, LLVM can also optimize out certain operations when it knows that memory is dead after a certain point. For example, it can sometimes remove the zeroing used to cancel the drop glue. This is possible when the glue drop itself was already removed because the zeroing dominated the drop glue call. For example in: ````rust pub fn bar(x: (Box<int>, int)) -> (Box<int>, int) { x } ```` With optimizations, this currently results in: ````llvm define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 { "_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit": %2 = bitcast { i64*, i64 }* %1 to i8* %3 = bitcast { i64*, i64 }* %0 to i8* tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false) tail call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 16, i32 8, i1 false) ret void } ```` But with lifetime intrinsics we get: ````llvm define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 { "_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit": %2 = bitcast { i64*, i64 }* %1 to i8* %3 = bitcast { i64*, i64 }* %0 to i8* tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false) tail call void @llvm.lifetime.end(i64 16, i8* %2) ret void } ```` Fixes #15665
2014-05-01 19:32:07 +02:00
let size = machine::llsize_of_alloc(ccx, val_ty(ptr).element_type());
if size == 0 {
return;
}
let lifetime_intrinsic = ccx.get_intrinsic(match lt {
Lifetime::Start => "llvm.lifetime.start",
Lifetime::End => "llvm.lifetime.end"
});
emit(ccx, size, lifetime_intrinsic)
Emit LLVM lifetime intrinsics to improve stack usage and codegen in general Lifetime intrinsics help to reduce stack usage, because LLVM can apply stack coloring to reuse the stack slots of dead allocas for new ones. For example these functions now both use the same amount of stack, while previous `bar()` used five times as much as `foo()`: ````rust fn foo() { println("{}", 5); } fn bar() { println("{}", 5); println("{}", 5); println("{}", 5); println("{}", 5); println("{}", 5); } ```` On top of that, LLVM can also optimize out certain operations when it knows that memory is dead after a certain point. For example, it can sometimes remove the zeroing used to cancel the drop glue. This is possible when the glue drop itself was already removed because the zeroing dominated the drop glue call. For example in: ````rust pub fn bar(x: (Box<int>, int)) -> (Box<int>, int) { x } ```` With optimizations, this currently results in: ````llvm define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 { "_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit": %2 = bitcast { i64*, i64 }* %1 to i8* %3 = bitcast { i64*, i64 }* %0 to i8* tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false) tail call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 16, i32 8, i1 false) ret void } ```` But with lifetime intrinsics we get: ````llvm define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 { "_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit": %2 = bitcast { i64*, i64 }* %1 to i8* %3 = bitcast { i64*, i64 }* %0 to i8* tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false) tail call void @llvm.lifetime.end(i64 16, i8* %2) ret void } ```` Fixes #15665
2014-05-01 19:32:07 +02:00
}
impl Lifetime {
pub fn call(self, b: &Builder, ptr: ValueRef) {
core_lifetime_emit(b.ccx, ptr, self, |ccx, size, lifetime_intrinsic| {
let ptr = b.pointercast(ptr, Type::i8p(ccx));
b.call(lifetime_intrinsic, &[C_u64(ccx, size), ptr], None);
});
}
}
pub fn call_lifetime_start(bcx: Block, ptr: ValueRef) {
if !bcx.unreachable.get() {
Lifetime::Start.call(&bcx.build(), ptr);
}
}
pub fn call_lifetime_end(bcx: Block, ptr: ValueRef) {
if !bcx.unreachable.get() {
Lifetime::End.call(&bcx.build(), ptr);
}
Emit LLVM lifetime intrinsics to improve stack usage and codegen in general Lifetime intrinsics help to reduce stack usage, because LLVM can apply stack coloring to reuse the stack slots of dead allocas for new ones. For example these functions now both use the same amount of stack, while previous `bar()` used five times as much as `foo()`: ````rust fn foo() { println("{}", 5); } fn bar() { println("{}", 5); println("{}", 5); println("{}", 5); println("{}", 5); println("{}", 5); } ```` On top of that, LLVM can also optimize out certain operations when it knows that memory is dead after a certain point. For example, it can sometimes remove the zeroing used to cancel the drop glue. This is possible when the glue drop itself was already removed because the zeroing dominated the drop glue call. For example in: ````rust pub fn bar(x: (Box<int>, int)) -> (Box<int>, int) { x } ```` With optimizations, this currently results in: ````llvm define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 { "_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit": %2 = bitcast { i64*, i64 }* %1 to i8* %3 = bitcast { i64*, i64 }* %0 to i8* tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false) tail call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 16, i32 8, i1 false) ret void } ```` But with lifetime intrinsics we get: ````llvm define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 { "_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit": %2 = bitcast { i64*, i64 }* %1 to i8* %3 = bitcast { i64*, i64 }* %0 to i8* tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false) tail call void @llvm.lifetime.end(i64 16, i8* %2) ret void } ```` Fixes #15665
2014-05-01 19:32:07 +02:00
}
// Generates code for resumption of unwind at the end of a landing pad.
pub fn trans_unwind_resume(bcx: Block, lpval: ValueRef) {
if !bcx.sess().target.target.options.custom_unwind_resume {
Resume(bcx, lpval);
} else {
let exc_ptr = ExtractValue(bcx, lpval, 0);
bcx.fcx.eh_unwind_resume()
.call(bcx, DebugLoc::None, &[exc_ptr], None);
}
}
pub fn call_memcpy<'bcx, 'tcx>(b: &Builder<'bcx, 'tcx>,
dst: ValueRef,
src: ValueRef,
n_bytes: ValueRef,
align: u32) {
let _icx = push_ctxt("call_memcpy");
let ccx = b.ccx;
let ptr_width = &ccx.sess().target.target.target_pointer_width[..];
let key = format!("llvm.memcpy.p0i8.p0i8.i{}", ptr_width);
2014-04-09 19:56:31 -04:00
let memcpy = ccx.get_intrinsic(&key);
let src_ptr = b.pointercast(src, Type::i8p(ccx));
let dst_ptr = b.pointercast(dst, Type::i8p(ccx));
let size = b.intcast(n_bytes, ccx.int_type());
let align = C_i32(ccx, align as i32);
let volatile = C_bool(ccx, false);
b.call(memcpy, &[dst_ptr, src_ptr, size, align, volatile], None);
}
2015-11-19 12:36:31 +01:00
pub fn memcpy_ty<'blk, 'tcx>(bcx: Block<'blk, 'tcx>, dst: ValueRef, src: ValueRef, t: Ty<'tcx>) {
let _icx = push_ctxt("memcpy_ty");
let ccx = bcx.ccx();
if type_is_zero_size(ccx, t) || bcx.unreachable.get() {
return;
}
if t.is_structural() {
2013-05-25 09:52:25 -04:00
let llty = type_of::type_of(ccx, t);
let llsz = llsize_of(ccx, llty);
2014-08-06 11:59:40 +02:00
let llalign = type_of::align_of(ccx, t);
call_memcpy(&B(bcx), dst, src, llsz, llalign as u32);
} else if common::type_is_fat_ptr(bcx.tcx(), t) {
let (data, extra) = load_fat_ptr(bcx, src, t);
store_fat_ptr(bcx, data, extra, dst, t);
} else {
store_ty(bcx, load_ty(bcx, src, t), dst, t);
2011-07-27 14:19:39 +02:00
}
}
pub fn init_zero_mem<'blk, 'tcx>(cx: Block<'blk, 'tcx>, llptr: ValueRef, t: Ty<'tcx>) {
2015-11-19 12:36:31 +01:00
if cx.unreachable.get() {
return;
}
let _icx = push_ctxt("init_zero_mem");
let bcx = cx;
memfill(&B(bcx), llptr, t, 0);
}
// Always use this function instead of storing a constant byte to the memory
// in question. e.g. if you store a zero constant, LLVM will drown in vreg
// allocation for large data structures, and the generated code will be
// awful. (A telltale sign of this is large quantities of
// `mov [byte ptr foo],0` in the generated code.)
fn memfill<'a, 'tcx>(b: &Builder<'a, 'tcx>, llptr: ValueRef, ty: Ty<'tcx>, byte: u8) {
let _icx = push_ctxt("memfill");
let ccx = b.ccx;
2014-08-06 11:59:40 +02:00
let llty = type_of::type_of(ccx, ty);
let llptr = b.pointercast(llptr, Type::i8(ccx).ptr_to());
let llzeroval = C_u8(ccx, byte);
2014-08-06 11:59:40 +02:00
let size = machine::llsize_of(ccx, llty);
let align = C_i32(ccx, type_of::align_of(ccx, ty) as i32);
call_memset(b, llptr, llzeroval, size, align, false);
}
pub fn call_memset<'bcx, 'tcx>(b: &Builder<'bcx, 'tcx>,
ptr: ValueRef,
fill_byte: ValueRef,
size: ValueRef,
align: ValueRef,
volatile: bool) {
let ccx = b.ccx;
let ptr_width = &ccx.sess().target.target.target_pointer_width[..];
let intrinsic_key = format!("llvm.memset.p0i8.i{}", ptr_width);
let llintrinsicfn = ccx.get_intrinsic(&intrinsic_key);
let volatile = C_bool(ccx, volatile);
b.call(llintrinsicfn, &[ptr, fill_byte, size, align, volatile], None);
}
pub fn alloc_ty<'blk, 'tcx>(bcx: Block<'blk, 'tcx>,
ty: Ty<'tcx>,
name: &str) -> ValueRef {
assert!(!ty.has_param_types());
alloca(bcx, type_of::type_of(bcx.ccx(), ty), name)
}
pub fn alloca(cx: Block, ty: Type, name: &str) -> ValueRef {
let _icx = push_ctxt("alloca");
if cx.unreachable.get() {
unsafe {
return llvm::LLVMGetUndef(ty.ptr_to().to_ref());
}
}
DebugLoc::None.apply(cx.fcx);
Alloca(cx, ty, name)
}
impl<'blk, 'tcx> FunctionContext<'blk, 'tcx> {
/// Create a function context for the given function.
/// Beware that you must call `fcx.init` or `fcx.bind_args`
/// before doing anything with the returned function context.
pub fn new(ccx: &'blk CrateContext<'blk, 'tcx>,
llfndecl: ValueRef,
fn_ty: FnType,
definition: Option<(Instance<'tcx>, &ty::FnSig<'tcx>, Abi)>,
block_arena: &'blk TypedArena<common::BlockS<'blk, 'tcx>>)
-> FunctionContext<'blk, 'tcx> {
let (param_substs, def_id) = match definition {
2016-08-26 19:23:42 +03:00
Some((instance, ..)) => {
common::validate_substs(instance.substs);
(instance.substs, Some(instance.def))
}
None => (ccx.tcx().intern_substs(&[]), None)
};
let local_id = def_id.and_then(|id| ccx.tcx().map.as_local_node_id(id));
debug!("FunctionContext::new({})",
definition.map_or(String::new(), |d| d.0.to_string()));
let no_debug = if let Some(id) = local_id {
ccx.tcx().map.attrs(id)
.iter().any(|item| item.check_name("no_debug"))
} else if let Some(def_id) = def_id {
ccx.sess().cstore.item_attrs(def_id)
.iter().any(|item| item.check_name("no_debug"))
} else {
false
};
2016-10-28 13:55:49 +03:00
let mir = def_id.map(|id| ccx.tcx().item_mir(id));
let debug_context = if let (false, Some((instance, sig, abi)), &Some(ref mir)) =
(no_debug, definition, &mir) {
debuginfo::create_function_debug_context(ccx, instance, sig, abi, llfndecl, mir)
} else {
debuginfo::empty_function_debug_context(ccx)
};
FunctionContext {
mir: mir,
llfn: llfndecl,
llretslotptr: Cell::new(None),
param_env: ccx.tcx().empty_parameter_environment(),
alloca_insert_pt: Cell::new(None),
landingpad_alloca: Cell::new(None),
fn_ty: fn_ty,
param_substs: param_substs,
span: None,
block_arena: block_arena,
lpad_arena: TypedArena::new(),
ccx: ccx,
debug_context: debug_context,
scopes: RefCell::new(Vec::new()),
}
}
/// Performs setup on a newly created function, creating the entry
/// scope block and allocating space for the return pointer.
pub fn init(&'blk self, skip_retptr: bool) -> Block<'blk, 'tcx> {
let entry_bcx = self.new_block("entry-block");
// Use a dummy instruction as the insertion point for all allocas.
// This is later removed in FunctionContext::cleanup.
self.alloca_insert_pt.set(Some(unsafe {
Load(entry_bcx, C_null(Type::i8p(self.ccx)));
llvm::LLVMGetFirstInstruction(entry_bcx.llbb)
}));
if !self.fn_ty.ret.is_ignore() && !skip_retptr {
// We normally allocate the llretslotptr, unless we
// have been instructed to skip it for immediate return
// values, or there is nothing to return at all.
// We create an alloca to hold a pointer of type `ret.original_ty`
// which will hold the pointer to the right alloca which has the
// final ret value
let llty = self.fn_ty.ret.memory_ty(self.ccx);
// But if there are no nested returns, we skip the indirection
// and have a single retslot
let slot = if self.fn_ty.ret.is_indirect() {
get_param(self.llfn, 0)
} else {
AllocaFcx(self, llty, "sret_slot")
};
self.llretslotptr.set(Some(slot));
}
entry_bcx
}
/// Ties up the llstaticallocas -> llloadenv -> lltop edges,
/// and builds the return block.
pub fn finish(&'blk self, ret_cx: Block<'blk, 'tcx>,
ret_debug_loc: DebugLoc) {
let _icx = push_ctxt("FunctionContext::finish");
self.build_return_block(ret_cx, ret_debug_loc);
DebugLoc::None.apply(self);
self.cleanup();
}
// Builds the return block for a function.
pub fn build_return_block(&self, ret_cx: Block<'blk, 'tcx>,
ret_debug_location: DebugLoc) {
if self.llretslotptr.get().is_none() ||
ret_cx.unreachable.get() ||
self.fn_ty.ret.is_indirect() {
return RetVoid(ret_cx, ret_debug_location);
}
let retslot = self.llretslotptr.get().unwrap();
let retptr = Value(retslot);
let llty = self.fn_ty.ret.original_ty;
match (retptr.get_dominating_store(ret_cx), self.fn_ty.ret.cast) {
// If there's only a single store to the ret slot, we can directly return
// the value that was stored and omit the store and the alloca.
// However, we only want to do this when there is no cast needed.
(Some(s), None) => {
let mut retval = s.get_operand(0).unwrap().get();
s.erase_from_parent();
if retptr.has_no_uses() {
retptr.erase_from_parent();
}
if self.fn_ty.ret.is_indirect() {
Store(ret_cx, retval, get_param(self.llfn, 0));
RetVoid(ret_cx, ret_debug_location)
} else {
if llty == Type::i1(self.ccx) {
retval = Trunc(ret_cx, retval, llty);
}
Ret(ret_cx, retval, ret_debug_location)
}
}
(_, cast_ty) if self.fn_ty.ret.is_indirect() => {
// Otherwise, copy the return value to the ret slot.
assert_eq!(cast_ty, None);
let llsz = llsize_of(self.ccx, self.fn_ty.ret.ty);
let llalign = llalign_of_min(self.ccx, self.fn_ty.ret.ty);
call_memcpy(&B(ret_cx), get_param(self.llfn, 0),
retslot, llsz, llalign as u32);
RetVoid(ret_cx, ret_debug_location)
}
(_, Some(cast_ty)) => {
let load = Load(ret_cx, PointerCast(ret_cx, retslot, cast_ty.ptr_to()));
let llalign = llalign_of_min(self.ccx, self.fn_ty.ret.ty);
unsafe {
llvm::LLVMSetAlignment(load, llalign);
}
Ret(ret_cx, load, ret_debug_location)
}
(_, None) => {
let retval = if llty == Type::i1(self.ccx) {
let val = LoadRangeAssert(ret_cx, retslot, 0, 2, llvm::False);
Trunc(ret_cx, val, llty)
} else {
Load(ret_cx, retslot)
};
Ret(ret_cx, retval, ret_debug_location)
}
}
}
2012-02-13 16:06:56 -08:00
}
pub fn trans_instance<'a, 'tcx>(ccx: &CrateContext<'a, 'tcx>, instance: Instance<'tcx>) {
let _s = if ccx.sess().trans_stats() {
let mut instance_name = String::new();
DefPathBasedNames::new(ccx.tcx(), true, true)
.push_def_path(instance.def, &mut instance_name);
Some(StatRecorder::new(ccx, instance_name))
} else {
None
};
// this is an info! to allow collecting monomorphization statistics
// and to allow finding the last function before LLVM aborts from
// release builds.
info!("trans_instance({})", instance);
let _icx = push_ctxt("trans_instance");
let fn_ty = ccx.tcx().item_type(instance.def);
let fn_ty = ccx.tcx().erase_regions(&fn_ty);
let fn_ty = monomorphize::apply_param_substs(ccx.shared(), instance.substs, &fn_ty);
let ty::BareFnTy { abi, ref sig, .. } = *common::ty_fn_ty(ccx, fn_ty);
let sig = ccx.tcx().erase_late_bound_regions_and_normalize(sig);
let lldecl = match ccx.instances().borrow().get(&instance) {
Some(&val) => val,
None => bug!("Instance `{:?}` not already declared", instance)
};
2014-09-05 09:18:53 -07:00
ccx.stats().n_closures.set(ccx.stats().n_closures.get() + 1);
rustc: Implement custom panic runtimes This commit is an implementation of [RFC 1513] which allows applications to alter the behavior of panics at compile time. A new compiler flag, `-C panic`, is added and accepts the values `unwind` or `panic`, with the default being `unwind`. This model affects how code is generated for the local crate, skipping generation of landing pads with `-C panic=abort`. [RFC 1513]: https://github.com/rust-lang/rfcs/blob/master/text/1513-less-unwinding.md Panic implementations are then provided by crates tagged with `#![panic_runtime]` and lazily required by crates with `#![needs_panic_runtime]`. The panic strategy (`-C panic` value) of the panic runtime must match the final product, and if the panic strategy is not `abort` then the entire DAG must have the same panic strategy. With the `-C panic=abort` strategy, users can expect a stable method to disable generation of landing pads, improving optimization in niche scenarios, decreasing compile time, and decreasing output binary size. With the `-C panic=unwind` strategy users can expect the existing ability to isolate failure in Rust code from the outside world. Organizationally, this commit dismantles the `sys_common::unwind` module in favor of some bits moving part of it to `libpanic_unwind` and the rest into the `panicking` module in libstd. The custom panic runtime support is pretty similar to the custom allocator support with the only major difference being how the panic runtime is injected (takes the `-C panic` flag into account).
2016-04-08 16:18:40 -07:00
if !ccx.sess().no_landing_pads() {
attributes::emit_uwtable(lldecl, true);
rustc: Implement custom panic runtimes This commit is an implementation of [RFC 1513] which allows applications to alter the behavior of panics at compile time. A new compiler flag, `-C panic`, is added and accepts the values `unwind` or `panic`, with the default being `unwind`. This model affects how code is generated for the local crate, skipping generation of landing pads with `-C panic=abort`. [RFC 1513]: https://github.com/rust-lang/rfcs/blob/master/text/1513-less-unwinding.md Panic implementations are then provided by crates tagged with `#![panic_runtime]` and lazily required by crates with `#![needs_panic_runtime]`. The panic strategy (`-C panic` value) of the panic runtime must match the final product, and if the panic strategy is not `abort` then the entire DAG must have the same panic strategy. With the `-C panic=abort` strategy, users can expect a stable method to disable generation of landing pads, improving optimization in niche scenarios, decreasing compile time, and decreasing output binary size. With the `-C panic=unwind` strategy users can expect the existing ability to isolate failure in Rust code from the outside world. Organizationally, this commit dismantles the `sys_common::unwind` module in favor of some bits moving part of it to `libpanic_unwind` and the rest into the `panicking` module in libstd. The custom panic runtime support is pretty similar to the custom allocator support with the only major difference being how the panic runtime is injected (takes the `-C panic` flag into account).
2016-04-08 16:18:40 -07:00
}
let fn_ty = FnType::new(ccx, abi, &sig, &[]);
let (arena, fcx): (TypedArena<_>, FunctionContext);
arena = TypedArena::new();
fcx = FunctionContext::new(ccx,
lldecl,
fn_ty,
Some((instance, &sig, abi)),
&arena);
if fcx.mir.is_none() {
bug!("attempted translation of `{}` w/o MIR", instance);
}
mir::trans_mir(&fcx);
2011-06-29 19:50:50 -07:00
}
pub fn trans_ctor_shim<'a, 'tcx>(ccx: &CrateContext<'a, 'tcx>,
def_id: DefId,
substs: &'tcx Substs<'tcx>,
disr: Disr,
llfndecl: ValueRef) {
attributes::inline(llfndecl, attributes::InlineAttr::Hint);
attributes::set_frame_pointer_elimination(ccx, llfndecl);
let ctor_ty = ccx.tcx().item_type(def_id);
let ctor_ty = monomorphize::apply_param_substs(ccx.shared(), substs, &ctor_ty);
let sig = ccx.tcx().erase_late_bound_regions_and_normalize(&ctor_ty.fn_sig());
let fn_ty = FnType::new(ccx, Abi::Rust, &sig, &[]);
let (arena, fcx): (TypedArena<_>, FunctionContext);
arena = TypedArena::new();
fcx = FunctionContext::new(ccx, llfndecl, fn_ty, None, &arena);
let bcx = fcx.init(false);
if !fcx.fn_ty.ret.is_ignore() {
let dest = fcx.llretslotptr.get().unwrap();
let dest_val = adt::MaybeSizedValue::sized(dest); // Can return unsized value
let mut llarg_idx = fcx.fn_ty.ret.is_indirect() as usize;
let mut arg_idx = 0;
for (i, arg_ty) in sig.inputs.into_iter().enumerate() {
let lldestptr = adt::trans_field_ptr(bcx, sig.output, dest_val, Disr::from(disr), i);
let arg = &fcx.fn_ty.args[arg_idx];
arg_idx += 1;
let b = &bcx.build();
if common::type_is_fat_ptr(bcx.tcx(), arg_ty) {
let meta = &fcx.fn_ty.args[arg_idx];
arg_idx += 1;
arg.store_fn_arg(b, &mut llarg_idx, get_dataptr(bcx, lldestptr));
meta.store_fn_arg(b, &mut llarg_idx, get_meta(bcx, lldestptr));
} else {
arg.store_fn_arg(b, &mut llarg_idx, lldestptr);
}
}
adt::trans_set_discr(bcx, sig.output, dest, disr);
}
fcx.finish(bcx, DebugLoc::None);
}
pub fn llvm_linkage_by_name(name: &str) -> Option<Linkage> {
// Use the names from src/llvm/docs/LangRef.rst here. Most types are only
// applicable to variable declarations and may not really make sense for
// Rust code in the first place but whitelist them anyway and trust that
// the user knows what s/he's doing. Who knows, unanticipated use cases
// may pop up in the future.
//
// ghost, dllimport, dllexport and linkonce_odr_autohide are not supported
// and don't have to be, LLVM treats them as no-ops.
match name {
"appending" => Some(llvm::Linkage::AppendingLinkage),
"available_externally" => Some(llvm::Linkage::AvailableExternallyLinkage),
"common" => Some(llvm::Linkage::CommonLinkage),
"extern_weak" => Some(llvm::Linkage::ExternalWeakLinkage),
"external" => Some(llvm::Linkage::ExternalLinkage),
"internal" => Some(llvm::Linkage::InternalLinkage),
"linkonce" => Some(llvm::Linkage::LinkOnceAnyLinkage),
"linkonce_odr" => Some(llvm::Linkage::LinkOnceODRLinkage),
"private" => Some(llvm::Linkage::PrivateLinkage),
"weak" => Some(llvm::Linkage::WeakAnyLinkage),
"weak_odr" => Some(llvm::Linkage::WeakODRLinkage),
_ => None,
}
}
pub fn set_link_section(ccx: &CrateContext,
llval: ValueRef,
attrs: &[ast::Attribute]) {
if let Some(sect) = attr::first_attr_value_str_by_name(attrs, "link_section") {
if contains_null(&sect) {
ccx.sess().fatal(&format!("Illegal null byte in link_section value: `{}`", &sect));
}
unsafe {
let buf = CString::new(sect.as_bytes()).unwrap();
llvm::LLVMSetSection(llval, buf.as_ptr());
}
}
}
/// Create the `main` function which will initialise the rust runtime and call
/// users main function.
pub fn maybe_create_entry_wrapper(ccx: &CrateContext) {
let (main_def_id, span) = match *ccx.sess().entry_fn.borrow() {
Some((id, span)) => {
(ccx.tcx().map.local_def_id(id), span)
}
None => return,
};
// check for the #[rustc_error] annotation, which forces an
// error in trans. This is used to write compile-fail tests
// that actually test that compilation succeeds without
// reporting an error.
if ccx.tcx().has_attr(main_def_id, "rustc_error") {
ccx.tcx().sess.span_fatal(span, "compilation successful");
}
let instance = Instance::mono(ccx.shared(), main_def_id);
if !ccx.codegen_unit().contains_item(&TransItem::Fn(instance)) {
// We want to create the wrapper in the same codegen unit as Rust's main
// function.
return;
}
let main_llfn = Callee::def(ccx, main_def_id, instance.substs).reify(ccx);
2014-03-05 16:36:01 +02:00
let et = ccx.sess().entry_type.get().unwrap();
match et {
config::EntryMain => {
create_entry_fn(ccx, span, main_llfn, true);
}
config::EntryStart => create_entry_fn(ccx, span, main_llfn, false),
config::EntryNone => {} // Do nothing.
}
2014-03-06 18:47:24 +02:00
fn create_entry_fn(ccx: &CrateContext,
2015-03-04 11:46:55 +02:00
sp: Span,
rust_main: ValueRef,
use_start_lang_item: bool) {
2015-11-19 12:36:31 +01:00
let llfty = Type::func(&[ccx.int_type(), Type::i8p(ccx).ptr_to()], &ccx.int_type());
if declare::get_defined_value(ccx, "main").is_some() {
// FIXME: We should be smart and show a better diagnostic here.
2015-12-21 10:00:43 +13:00
ccx.sess().struct_span_err(sp, "entry symbol `main` defined multiple times")
.help("did you use #[no_mangle] on `fn main`? Use #[start] instead")
.emit();
ccx.sess().abort_if_errors();
bug!();
}
let llfn = declare::declare_cfn(ccx, "main", llfty);
// `main` should respect same config for frame pointer elimination as rest of code
attributes::set_frame_pointer_elimination(ccx, llfn);
let llbb = unsafe {
2015-11-19 12:36:31 +01:00
llvm::LLVMAppendBasicBlockInContext(ccx.llcx(), llfn, "top\0".as_ptr() as *const _)
};
2014-09-05 09:18:53 -07:00
let bld = ccx.raw_builder();
unsafe {
llvm::LLVMPositionBuilderAtEnd(bld, llbb);
2015-04-24 15:25:42 +12:00
debuginfo::gdb::insert_reference_to_gdb_debug_scripts_section_global(ccx);
let (start_fn, args) = if use_start_lang_item {
2014-09-05 09:18:53 -07:00
let start_def_id = match ccx.tcx().lang_items.require(StartFnLangItem) {
Ok(id) => id,
Err(s) => ccx.sess().fatal(&s)
};
let empty_substs = ccx.tcx().intern_substs(&[]);
let start_fn = Callee::def(ccx, start_def_id, empty_substs).reify(ccx);
let args = {
2015-11-19 12:36:31 +01:00
let opaque_rust_main =
llvm::LLVMBuildPointerCast(bld,
rust_main,
Type::i8p(ccx).to_ref(),
"rust_main\0".as_ptr() as *const _);
vec![opaque_rust_main, get_param(llfn, 0), get_param(llfn, 1)]
};
(start_fn, args)
} else {
debug!("using user-defined start fn");
2015-11-19 12:36:31 +01:00
let args = vec![get_param(llfn, 0 as c_uint), get_param(llfn, 1 as c_uint)];
(rust_main, args)
};
trans: Reimplement unwinding on MSVC This commit transitions the compiler to using the new exception handling instructions in LLVM for implementing unwinding for MSVC. This affects both 32 and 64-bit MSVC as they're both now using SEH-based strategies. In terms of standard library support, lots more details about how SEH unwinding is implemented can be found in the commits. In terms of trans, this change necessitated a few modifications: * Branches were added to detect when the old landingpad instruction is used or the new cleanuppad instruction is used to `trans::cleanup`. * The return value from `cleanuppad` is not stored in an `alloca` (because it cannot be). * Each block in trans now has an `Option<LandingPad>` instead of `is_lpad: bool` for indicating whether it's in a landing pad or not. The new exception handling intrinsics require that on MSVC each `call` inside of a landing pad is annotated with which landing pad that it's in. This change to the basic block means that whenever a `call` or `invoke` instruction is generated we know whether to annotate it as part of a cleanuppad or not. * Lots of modifications were made to the instruction builders to construct the new instructions as well as pass the tagging information for the call/invoke instructions. * The translation of the `try` intrinsics for MSVC has been overhauled to use the new `catchpad` instruction. The filter function is now also a rustc-generated function instead of a purely libstd-defined function. The libstd definition still exists, it just has a stable ABI across architectures and leaves some of the really weird implementation details to the compiler (e.g. the `localescape` and `localrecover` intrinsics).
2015-10-23 18:18:44 -07:00
let result = llvm::LLVMRustBuildCall(bld,
start_fn,
args.as_ptr(),
args.len() as c_uint,
ptr::null_mut(),
trans: Reimplement unwinding on MSVC This commit transitions the compiler to using the new exception handling instructions in LLVM for implementing unwinding for MSVC. This affects both 32 and 64-bit MSVC as they're both now using SEH-based strategies. In terms of standard library support, lots more details about how SEH unwinding is implemented can be found in the commits. In terms of trans, this change necessitated a few modifications: * Branches were added to detect when the old landingpad instruction is used or the new cleanuppad instruction is used to `trans::cleanup`. * The return value from `cleanuppad` is not stored in an `alloca` (because it cannot be). * Each block in trans now has an `Option<LandingPad>` instead of `is_lpad: bool` for indicating whether it's in a landing pad or not. The new exception handling intrinsics require that on MSVC each `call` inside of a landing pad is annotated with which landing pad that it's in. This change to the basic block means that whenever a `call` or `invoke` instruction is generated we know whether to annotate it as part of a cleanuppad or not. * Lots of modifications were made to the instruction builders to construct the new instructions as well as pass the tagging information for the call/invoke instructions. * The translation of the `try` intrinsics for MSVC has been overhauled to use the new `catchpad` instruction. The filter function is now also a rustc-generated function instead of a purely libstd-defined function. The libstd definition still exists, it just has a stable ABI across architectures and leaves some of the really weird implementation details to the compiler (e.g. the `localescape` and `localrecover` intrinsics).
2015-10-23 18:18:44 -07:00
noname());
llvm::LLVMBuildRet(bld, result);
}
}
}
fn contains_null(s: &str) -> bool {
2014-09-23 12:54:16 -07:00
s.bytes().any(|b| b == 0)
}
fn write_metadata(cx: &SharedCrateContext,
reachable_ids: &NodeSet) -> Vec<u8> {
use flate;
Store metadata separately in rlib files Right now whenever an rlib file is linked against, all of the metadata from the rlib is pulled in to the final staticlib or binary. The reason for this is that the metadata is currently stored in a section of the object file. Note that this is intentional for dynamic libraries in order to distribute metadata bundled with static libraries. This commit alters the situation for rlib libraries to instead store the metadata in a separate file in the archive. In doing so, when the archive is passed to the linker, none of the metadata will get pulled into the result executable. Furthermore, the metadata file is skipped when assembling rlibs into an archive. The snag in this implementation comes with multiple output formats. When generating a dylib, the metadata needs to be in the object file, but when generating an rlib this needs to be separate. In order to accomplish this, the metadata variable is inserted into an entirely separate LLVM Module which is then codegen'd into a different location (foo.metadata.o). This is then linked into dynamic libraries and silently ignored for rlib files. While changing how metadata is inserted into archives, I have also stopped compressing metadata when inserted into rlib files. We have wanted to stop compressing metadata, but the sections it creates in object file sections are apparently too large. Thankfully if it's just an arbitrary file it doesn't matter how large it is. I have seen massive reductions in executable sizes, as well as staticlib output sizes (to confirm that this is all working).
2013-12-03 17:41:01 -08:00
#[derive(PartialEq, Eq, PartialOrd, Ord)]
enum MetadataKind {
None,
Uncompressed,
Compressed
}
let kind = cx.sess().crate_types.borrow().iter().map(|ty| {
match *ty {
config::CrateTypeExecutable |
config::CrateTypeStaticlib |
config::CrateTypeCdylib => MetadataKind::None,
config::CrateTypeRlib => MetadataKind::Uncompressed,
config::CrateTypeDylib |
config::CrateTypeProcMacro => MetadataKind::Compressed,
}
}).max().unwrap();
if kind == MetadataKind::None {
2015-11-19 12:36:31 +01:00
return Vec::new();
}
2015-11-21 01:08:09 +02:00
let cstore = &cx.tcx().sess.cstore;
let metadata = cstore.encode_metadata(cx.tcx(),
cx.export_map(),
cx.link_meta(),
2016-10-28 13:55:49 +03:00
reachable_ids);
if kind == MetadataKind::Uncompressed {
return metadata;
}
assert!(kind == MetadataKind::Compressed);
2015-11-21 01:08:09 +02:00
let mut compressed = cstore.metadata_encoding_version().to_vec();
std: Stabilize APIs for the 1.6 release This commit is the standard API stabilization commit for the 1.6 release cycle. The list of issues and APIs below have all been through their cycle-long FCP and the libs team decisions are listed below Stabilized APIs * `Read::read_exact` * `ErrorKind::UnexpectedEof` (renamed from `UnexpectedEOF`) * libcore -- this was a bit of a nuanced stabilization, the crate itself is now marked as `#[stable]` and the methods appearing via traits for primitives like `char` and `str` are now also marked as stable. Note that the extension traits themeselves are marked as unstable as they're imported via the prelude. The `try!` macro was also moved from the standard library into libcore to have the same interface. Otherwise the functions all have copied stability from the standard library now. * The `#![no_std]` attribute * `fs::DirBuilder` * `fs::DirBuilder::new` * `fs::DirBuilder::recursive` * `fs::DirBuilder::create` * `os::unix::fs::DirBuilderExt` * `os::unix::fs::DirBuilderExt::mode` * `vec::Drain` * `vec::Vec::drain` * `string::Drain` * `string::String::drain` * `vec_deque::Drain` * `vec_deque::VecDeque::drain` * `collections::hash_map::Drain` * `collections::hash_map::HashMap::drain` * `collections::hash_set::Drain` * `collections::hash_set::HashSet::drain` * `collections::binary_heap::Drain` * `collections::binary_heap::BinaryHeap::drain` * `Vec::extend_from_slice` (renamed from `push_all`) * `Mutex::get_mut` * `Mutex::into_inner` * `RwLock::get_mut` * `RwLock::into_inner` * `Iterator::min_by_key` (renamed from `min_by`) * `Iterator::max_by_key` (renamed from `max_by`) Deprecated APIs * `ErrorKind::UnexpectedEOF` (renamed to `UnexpectedEof`) * `OsString::from_bytes` * `OsStr::to_cstring` * `OsStr::to_bytes` * `fs::walk_dir` and `fs::WalkDir` * `path::Components::peek` * `slice::bytes::MutableByteVector` * `slice::bytes::copy_memory` * `Vec::push_all` (renamed to `extend_from_slice`) * `Duration::span` * `IpAddr` * `SocketAddr::ip` * `Read::tee` * `io::Tee` * `Write::broadcast` * `io::Broadcast` * `Iterator::min_by` (renamed to `min_by_key`) * `Iterator::max_by` (renamed to `max_by_key`) * `net::lookup_addr` New APIs (still unstable) * `<[T]>::sort_by_key` (added to mirror `min_by_key`) Closes #27585 Closes #27704 Closes #27707 Closes #27710 Closes #27711 Closes #27727 Closes #27740 Closes #27744 Closes #27799 Closes #27801 cc #27801 (doesn't close as `Chars` is still unstable) Closes #28968
2015-12-02 17:31:49 -08:00
compressed.extend_from_slice(&flate::deflate_bytes(&metadata));
2015-11-21 01:08:09 +02:00
let llmeta = C_bytes_in_context(cx.metadata_llcx(), &compressed[..]);
2014-11-17 21:39:01 +13:00
let llconst = C_struct_in_context(cx.metadata_llcx(), &[llmeta], false);
let name = cx.metadata_symbol_name();
std: Implement CString-related RFCs This commit is an implementation of [RFC 592][r592] and [RFC 840][r840]. These two RFCs tweak the behavior of `CString` and add a new `CStr` unsized slice type to the module. [r592]: https://github.com/rust-lang/rfcs/blob/master/text/0592-c-str-deref.md [r840]: https://github.com/rust-lang/rfcs/blob/master/text/0840-no-panic-in-c-string.md The new `CStr` type is only constructable via two methods: 1. By `deref`'ing from a `CString` 2. Unsafely via `CStr::from_ptr` The purpose of `CStr` is to be an unsized type which is a thin pointer to a `libc::c_char` (currently it is a fat pointer slice due to implementation limitations). Strings from C can be safely represented with a `CStr` and an appropriate lifetime as well. Consumers of `&CString` should now consume `&CStr` instead to allow producers to pass in C-originating strings instead of just Rust-allocated strings. A new constructor was added to `CString`, `new`, which takes `T: IntoBytes` instead of separate `from_slice` and `from_vec` methods (both have been deprecated in favor of `new`). The `new` method returns a `Result` instead of panicking. The error variant contains the relevant information about where the error happened and bytes (if present). Conversions are provided to the `io::Error` and `old_io::IoError` types via the `FromError` trait which translate to `InvalidInput`. This is a breaking change due to the modification of existing `#[unstable]` APIs and new deprecation, and more detailed information can be found in the two RFCs. Notable breakage includes: * All construction of `CString` now needs to use `new` and handle the outgoing `Result`. * Usage of `CString` as a byte slice now explicitly needs a `.as_bytes()` call. * The `as_slice*` methods have been removed in favor of just having the `as_bytes*` methods. Closes #22469 Closes #22470 [breaking-change]
2015-02-17 22:47:40 -08:00
let buf = CString::new(name).unwrap();
let llglobal = unsafe {
2015-11-19 12:36:31 +01:00
llvm::LLVMAddGlobal(cx.metadata_llmod(), val_ty(llconst).to_ref(), buf.as_ptr())
};
unsafe {
llvm::LLVMSetInitializer(llglobal, llconst);
let section_name =
2015-11-21 01:08:09 +02:00
cx.tcx().sess.cstore.metadata_section_name(&cx.sess().target.target);
let name = CString::new(section_name).unwrap();
llvm::LLVMSetSection(llglobal, name.as_ptr());
// Also generate a .section directive to force no
// flags, at least for ELF outputs, so that the
// metadata doesn't get loaded into memory.
let directive = format!(".section {}", section_name);
let directive = CString::new(directive).unwrap();
llvm::LLVMSetModuleInlineAsm(cx.metadata_llmod(), directive.as_ptr())
}
Store metadata separately in rlib files Right now whenever an rlib file is linked against, all of the metadata from the rlib is pulled in to the final staticlib or binary. The reason for this is that the metadata is currently stored in a section of the object file. Note that this is intentional for dynamic libraries in order to distribute metadata bundled with static libraries. This commit alters the situation for rlib libraries to instead store the metadata in a separate file in the archive. In doing so, when the archive is passed to the linker, none of the metadata will get pulled into the result executable. Furthermore, the metadata file is skipped when assembling rlibs into an archive. The snag in this implementation comes with multiple output formats. When generating a dylib, the metadata needs to be in the object file, but when generating an rlib this needs to be separate. In order to accomplish this, the metadata variable is inserted into an entirely separate LLVM Module which is then codegen'd into a different location (foo.metadata.o). This is then linked into dynamic libraries and silently ignored for rlib files. While changing how metadata is inserted into archives, I have also stopped compressing metadata when inserted into rlib files. We have wanted to stop compressing metadata, but the sections it creates in object file sections are apparently too large. Thankfully if it's just an arbitrary file it doesn't matter how large it is. I have seen massive reductions in executable sizes, as well as staticlib output sizes (to confirm that this is all working).
2013-12-03 17:41:01 -08:00
return metadata;
}
/// Find any symbols that are defined in one compilation unit, but not declared
/// in any other compilation unit. Give these symbols internal linkage.
2016-07-22 10:39:30 -04:00
fn internalize_symbols<'a, 'tcx>(sess: &Session,
ccxs: &CrateContextList<'a, 'tcx>,
symbol_map: &SymbolMap<'tcx>,
reachable: &FxHashSet<&str>) {
let scx = ccxs.shared();
let tcx = scx.tcx();
2016-07-22 10:39:30 -04:00
// In incr. comp. mode, we can't necessarily see all refs since we
// don't generate LLVM IR for reused modules, so skip this
// step. Later we should get smarter.
if sess.opts.debugging_opts.incremental.is_some() {
return;
}
// 'unsafe' because we are holding on to CStr's from the LLVM module within
// this block.
unsafe {
let mut referenced_somewhere = FxHashSet();
// Collect all symbols that need to stay externally visible because they
// are referenced via a declaration in some other codegen unit.
for ccx in ccxs.iter_need_trans() {
for val in iter_globals(ccx.llmod()).chain(iter_functions(ccx.llmod())) {
let linkage = llvm::LLVMRustGetLinkage(val);
// We only care about external declarations (not definitions)
// and available_externally definitions.
let is_available_externally = linkage == llvm::Linkage::AvailableExternallyLinkage;
let is_decl = llvm::LLVMIsDeclaration(val) != 0;
if is_decl || is_available_externally {
let symbol_name = CStr::from_ptr(llvm::LLVMGetValueName(val));
referenced_somewhere.insert(symbol_name);
}
}
}
// Also collect all symbols for which we cannot adjust linkage, because
// it is fixed by some directive in the source code (e.g. #[no_mangle]).
let linkage_fixed_explicitly: FxHashSet<_> = scx
.translation_items()
.borrow()
.iter()
.cloned()
.filter(|trans_item|{
trans_item.explicit_linkage(tcx).is_some()
})
.map(|trans_item| symbol_map.get_or_compute(scx, trans_item))
.collect();
// Examine each external definition. If the definition is not used in
// any other compilation unit, and is not reachable from other crates,
// then give it internal linkage.
for ccx in ccxs.iter_need_trans() {
for val in iter_globals(ccx.llmod()).chain(iter_functions(ccx.llmod())) {
let linkage = llvm::LLVMRustGetLinkage(val);
let is_externally_visible = (linkage == llvm::Linkage::ExternalLinkage) ||
(linkage == llvm::Linkage::LinkOnceODRLinkage) ||
(linkage == llvm::Linkage::WeakODRLinkage);
let is_definition = llvm::LLVMIsDeclaration(val) == 0;
// If this is a definition (as opposed to just a declaration)
// and externally visible, check if we can internalize it
if is_definition && is_externally_visible {
let name_cstr = CStr::from_ptr(llvm::LLVMGetValueName(val));
let name_str = name_cstr.to_str().unwrap();
let name_cow = Cow::Borrowed(name_str);
let is_referenced_somewhere = referenced_somewhere.contains(&name_cstr);
let is_reachable = reachable.contains(&name_str);
let has_fixed_linkage = linkage_fixed_explicitly.contains(&name_cow);
if !is_referenced_somewhere && !is_reachable && !has_fixed_linkage {
llvm::LLVMRustSetLinkage(val, llvm::Linkage::InternalLinkage);
2016-08-03 00:25:19 +03:00
llvm::LLVMSetDLLStorageClass(val,
llvm::DLLStorageClass::Default);
llvm::UnsetComdat(val);
}
}
}
}
}
}
// Create a `__imp_<symbol> = &symbol` global for every public static `symbol`.
// This is required to satisfy `dllimport` references to static data in .rlibs
// when using MSVC linker. We do this only for data, as linker can fix up
// code references on its own.
// See #26591, #27438
fn create_imps(cx: &CrateContextList) {
// The x86 ABI seems to require that leading underscores are added to symbol
// names, so we need an extra underscore on 32-bit. There's also a leading
// '\x01' here which disables LLVM's symbol mangling (e.g. no extra
// underscores added in front).
let prefix = if cx.shared().sess().target.target.target_pointer_width == "32" {
"\x01__imp__"
} else {
"\x01__imp_"
};
unsafe {
for ccx in cx.iter_need_trans() {
let exported: Vec<_> = iter_globals(ccx.llmod())
2015-11-19 12:36:31 +01:00
.filter(|&val| {
llvm::LLVMRustGetLinkage(val) ==
llvm::Linkage::ExternalLinkage &&
2015-11-19 12:36:31 +01:00
llvm::LLVMIsDeclaration(val) == 0
})
.collect();
let i8p_ty = Type::i8p(&ccx);
for val in exported {
let name = CStr::from_ptr(llvm::LLVMGetValueName(val));
let mut imp_name = prefix.as_bytes().to_vec();
imp_name.extend(name.to_bytes());
let imp_name = CString::new(imp_name).unwrap();
2015-11-19 12:36:31 +01:00
let imp = llvm::LLVMAddGlobal(ccx.llmod(),
i8p_ty.to_ref(),
imp_name.as_ptr() as *const _);
let init = llvm::LLVMConstBitCast(val, i8p_ty.to_ref());
llvm::LLVMSetInitializer(imp, init);
llvm::LLVMRustSetLinkage(imp, llvm::Linkage::ExternalLinkage);
}
}
}
}
struct ValueIter {
cur: ValueRef,
step: unsafe extern "C" fn(ValueRef) -> ValueRef,
}
impl Iterator for ValueIter {
type Item = ValueRef;
fn next(&mut self) -> Option<ValueRef> {
let old = self.cur;
if !old.is_null() {
self.cur = unsafe { (self.step)(old) };
Some(old)
} else {
None
}
}
}
fn iter_globals(llmod: llvm::ModuleRef) -> ValueIter {
unsafe {
ValueIter {
cur: llvm::LLVMGetFirstGlobal(llmod),
step: llvm::LLVMGetNextGlobal,
}
}
}
2015-01-01 23:54:03 -05:00
fn iter_functions(llmod: llvm::ModuleRef) -> ValueIter {
unsafe {
ValueIter {
cur: llvm::LLVMGetFirstFunction(llmod),
step: llvm::LLVMGetNextFunction,
}
}
}
/// The context provided lists a set of reachable ids as calculated by
/// middle::reachable, but this contains far more ids and symbols than we're
/// actually exposing from the object file. This function will filter the set in
/// the context to the set of ids which correspond to symbols that are exposed
/// from the object file being generated.
///
/// This list is later used by linkers to determine the set of symbols needed to
/// be exposed from a dynamic library and it's also encoded into the metadata.
pub fn filter_reachable_ids(tcx: TyCtxt, reachable: NodeSet) -> NodeSet {
reachable.into_iter().filter(|&id| {
// Next, we want to ignore some FFI functions that are not exposed from
// this crate. Reachable FFI functions can be lumped into two
// categories:
//
// 1. Those that are included statically via a static library
// 2. Those included otherwise (e.g. dynamically or via a framework)
//
// Although our LLVM module is not literally emitting code for the
// statically included symbols, it's an export of our library which
// needs to be passed on to the linker and encoded in the metadata.
//
// As a result, if this id is an FFI item (foreign item) then we only
// let it through if it's included statically.
match tcx.map.get(id) {
2015-07-31 00:04:06 -07:00
hir_map::NodeForeignItem(..) => {
tcx.sess.cstore.is_statically_included_foreign_item(id)
}
// Only consider nodes that actually have exported symbols.
hir_map::NodeItem(&hir::Item {
node: hir::ItemStatic(..), .. }) |
hir_map::NodeItem(&hir::Item {
node: hir::ItemFn(..), .. }) |
hir_map::NodeImplItem(&hir::ImplItem {
node: hir::ImplItemKind::Method(..), .. }) => {
let def_id = tcx.map.local_def_id(id);
let generics = tcx.item_generics(def_id);
let attributes = tcx.get_attrs(def_id);
(generics.parent_types == 0 && generics.types.is_empty()) &&
// Functions marked with #[inline] are only ever translated
// with "internal" linkage and are never exported.
!attr::requests_inline(&attributes[..])
}
_ => false
}
}).collect()
}
pub fn trans_crate<'a, 'tcx>(tcx: TyCtxt<'a, 'tcx, 'tcx>,
analysis: ty::CrateAnalysis,
incremental_hashes_map: &IncrementalHashesMap)
-> CrateTranslation {
let _task = tcx.dep_graph.in_task(DepNode::TransCrate);
// Be careful with this krate: obviously it gives access to the
// entire contents of the krate. So if you push any subtasks of
// `TransCrate`, you need to be careful to register "reads" of the
// particular items that will be processed.
2014-09-07 20:09:06 +03:00
let krate = tcx.map.krate();
let ty::CrateAnalysis { export_map, reachable, name, .. } = analysis;
let reachable = filter_reachable_ids(tcx, reachable);
let check_overflow = if let Some(v) = tcx.sess.opts.debugging_opts.force_overflow_checks {
v
} else {
tcx.sess.opts.debug_assertions
};
let link_meta = link::build_link_meta(incremental_hashes_map, name);
let shared_ccx = SharedCrateContext::new(tcx,
export_map,
link_meta.clone(),
reachable,
check_overflow);
// Translate the metadata.
let metadata = time(tcx.sess.time_passes(), "write metadata", || {
write_metadata(&shared_ccx, shared_ccx.reachable())
});
let metadata_module = ModuleTranslation {
name: "metadata".to_string(),
symbol_name_hash: 0, // we always rebuild metadata, at least for now
source: ModuleSource::Translated(ModuleLlvm {
llcx: shared_ccx.metadata_llcx(),
llmod: shared_ccx.metadata_llmod(),
}),
};
let no_builtins = attr::contains_name(&krate.attrs, "no_builtins");
// Run the translation item collector and partition the collected items into
// codegen units.
let (codegen_units, symbol_map) = collect_and_partition_translation_items(&shared_ccx);
let symbol_map = Rc::new(symbol_map);
let previous_work_products = trans_reuse_previous_work_products(tcx,
&codegen_units,
&symbol_map);
let crate_context_list = CrateContextList::new(&shared_ccx,
codegen_units,
previous_work_products,
symbol_map.clone());
let modules: Vec<_> = crate_context_list.iter_all()
.map(|ccx| {
let source = match ccx.previous_work_product() {
Some(buf) => ModuleSource::Preexisting(buf.clone()),
None => ModuleSource::Translated(ModuleLlvm {
llcx: ccx.llcx(),
llmod: ccx.llmod(),
}),
};
ModuleTranslation {
name: String::from(ccx.codegen_unit().name()),
symbol_name_hash: ccx.codegen_unit().compute_symbol_name_hash(tcx, &symbol_map),
source: source,
}
})
.collect();
assert_module_sources::assert_module_sources(tcx, &modules);
// Skip crate items and just output metadata in -Z no-trans mode.
if tcx.sess.opts.debugging_opts.no_trans {
let linker_info = LinkerInfo::new(&shared_ccx, &[]);
return CrateTranslation {
modules: modules,
metadata_module: metadata_module,
link: link_meta,
metadata: metadata,
reachable: vec![],
no_builtins: no_builtins,
linker_info: linker_info,
windows_subsystem: None,
};
}
// Instantiate translation items without filling out definitions yet...
for ccx in crate_context_list.iter_need_trans() {
let cgu = ccx.codegen_unit();
let trans_items = cgu.items_in_deterministic_order(tcx, &symbol_map);
tcx.dep_graph.with_task(cgu.work_product_dep_node(), || {
for (trans_item, linkage) in trans_items {
trans_item.predefine(&ccx, linkage);
}
});
}
// ... and now that we have everything pre-defined, fill out those definitions.
for ccx in crate_context_list.iter_need_trans() {
let cgu = ccx.codegen_unit();
let trans_items = cgu.items_in_deterministic_order(tcx, &symbol_map);
tcx.dep_graph.with_task(cgu.work_product_dep_node(), || {
for (trans_item, _) in trans_items {
trans_item.define(&ccx);
}
// If this codegen unit contains the main function, also create the
// wrapper here
maybe_create_entry_wrapper(&ccx);
2016-03-16 15:00:20 -04:00
// Run replace-all-uses-with for statics that need it
for &(old_g, new_g) in ccx.statics_to_rauw().borrow().iter() {
unsafe {
let bitcast = llvm::LLVMConstPointerCast(new_g, llvm::LLVMTypeOf(old_g));
llvm::LLVMReplaceAllUsesWith(old_g, bitcast);
llvm::LLVMDeleteGlobal(old_g);
}
}
// Finalize debuginfo
if ccx.sess().opts.debuginfo != NoDebugInfo {
debuginfo::finalize(&ccx);
}
});
}
symbol_names_test::report_symbol_names(&shared_ccx);
if shared_ccx.sess().trans_stats() {
let stats = shared_ccx.stats();
println!("--- trans stats ---");
println!("n_glues_created: {}", stats.n_glues_created.get());
println!("n_null_glues: {}", stats.n_null_glues.get());
println!("n_real_glues: {}", stats.n_real_glues.get());
println!("n_fns: {}", stats.n_fns.get());
println!("n_inlines: {}", stats.n_inlines.get());
println!("n_closures: {}", stats.n_closures.get());
println!("fn stats:");
stats.fn_stats.borrow_mut().sort_by(|&(_, insns_a), &(_, insns_b)| {
insns_b.cmp(&insns_a)
});
for tuple in stats.fn_stats.borrow().iter() {
match *tuple {
(ref name, insns) => {
println!("{} insns, {}", insns, *name);
2013-07-08 11:05:52 -04:00
}
}
}
}
if shared_ccx.sess().count_llvm_insns() {
for (k, v) in shared_ccx.stats().llvm_insns.borrow().iter() {
println!("{:7} {}", *v, *k);
}
}
let sess = shared_ccx.sess();
let mut reachable_symbols = shared_ccx.reachable().iter().map(|&id| {
let def_id = shared_ccx.tcx().map.local_def_id(id);
symbol_for_def_id(def_id, &shared_ccx, &symbol_map)
}).collect::<Vec<_>>();
if sess.entry_fn.borrow().is_some() {
reachable_symbols.push("main".to_string());
}
Implement LTO This commit implements LTO for rust leveraging LLVM's passes. What this means is: * When compiling an rlib, in addition to insdering foo.o into the archive, also insert foo.bc (the LLVM bytecode) of the optimized module. * When the compiler detects the -Z lto option, it will attempt to perform LTO on a staticlib or binary output. The compiler will emit an error if a dylib or rlib output is being generated. * The actual act of performing LTO is as follows: 1. Force all upstream libraries to have an rlib version available. 2. Load the bytecode of each upstream library from the rlib. 3. Link all this bytecode into the current LLVM module (just using llvm apis) 4. Run an internalization pass which internalizes all symbols except those found reachable for the local crate of compilation. 5. Run the LLVM LTO pass manager over this entire module 6a. If assembling an archive, then add all upstream rlibs into the output archive. This ignores all of the object/bitcode/metadata files rust generated and placed inside the rlibs. 6b. If linking a binary, create copies of all upstream rlibs, remove the rust-generated object-file, and then link everything as usual. As I have explained in #10741, this process is excruciatingly slow, so this is *not* turned on by default, and it is also why I have decided to hide it behind a -Z flag for now. The good news is that the binary sizes are about as small as they can be as a result of LTO, so it's definitely working. Closes #10741 Closes #10740
2013-12-02 23:19:29 -08:00
if sess.crate_types.borrow().contains(&config::CrateTypeDylib) {
reachable_symbols.push(shared_ccx.metadata_symbol_name());
}
// For the purposes of LTO or when creating a cdylib, we add to the
// reachable set all of the upstream reachable extern fns. These functions
// are all part of the public ABI of the final product, so we need to
// preserve them.
//
// Note that this happens even if LTO isn't requested or we're not creating
// a cdylib. In those cases, though, we're not even reading the
// `reachable_symbols` list later on so it should be ok.
for cnum in sess.cstore.crates() {
let syms = sess.cstore.reachable_ids(cnum);
reachable_symbols.extend(syms.into_iter().filter(|&def_id| {
let applicable = match sess.cstore.describe_def(def_id) {
Some(Def::Static(..)) => true,
Some(Def::Fn(_)) => {
shared_ccx.tcx().item_generics(def_id).types.is_empty()
}
_ => false
};
if applicable {
let attrs = shared_ccx.tcx().get_attrs(def_id);
attr::contains_extern_indicator(sess.diagnostic(), &attrs)
} else {
false
}
}).map(|did| {
symbol_for_def_id(did, &shared_ccx, &symbol_map)
}));
}
time(shared_ccx.sess().time_passes(), "internalize symbols", || {
2016-07-22 10:39:30 -04:00
internalize_symbols(sess,
&crate_context_list,
&symbol_map,
&reachable_symbols.iter()
.map(|s| &s[..])
.collect())
});
if sess.target.target.options.is_like_msvc &&
2015-09-25 18:53:14 -07:00
sess.crate_types.borrow().iter().any(|ct| *ct == config::CrateTypeRlib) {
create_imps(&crate_context_list);
}
let linker_info = LinkerInfo::new(&shared_ccx, &reachable_symbols);
let subsystem = attr::first_attr_value_str_by_name(&krate.attrs,
"windows_subsystem");
let windows_subsystem = subsystem.map(|subsystem| {
if subsystem != "windows" && subsystem != "console" {
tcx.sess.fatal(&format!("invalid windows subsystem `{}`, only \
`windows` and `console` are allowed",
subsystem));
}
subsystem.to_string()
});
CrateTranslation {
modules: modules,
metadata_module: metadata_module,
link: link_meta,
Store metadata separately in rlib files Right now whenever an rlib file is linked against, all of the metadata from the rlib is pulled in to the final staticlib or binary. The reason for this is that the metadata is currently stored in a section of the object file. Note that this is intentional for dynamic libraries in order to distribute metadata bundled with static libraries. This commit alters the situation for rlib libraries to instead store the metadata in a separate file in the archive. In doing so, when the archive is passed to the linker, none of the metadata will get pulled into the result executable. Furthermore, the metadata file is skipped when assembling rlibs into an archive. The snag in this implementation comes with multiple output formats. When generating a dylib, the metadata needs to be in the object file, but when generating an rlib this needs to be separate. In order to accomplish this, the metadata variable is inserted into an entirely separate LLVM Module which is then codegen'd into a different location (foo.metadata.o). This is then linked into dynamic libraries and silently ignored for rlib files. While changing how metadata is inserted into archives, I have also stopped compressing metadata when inserted into rlib files. We have wanted to stop compressing metadata, but the sections it creates in object file sections are apparently too large. Thankfully if it's just an arbitrary file it doesn't matter how large it is. I have seen massive reductions in executable sizes, as well as staticlib output sizes (to confirm that this is all working).
2013-12-03 17:41:01 -08:00
metadata: metadata,
reachable: reachable_symbols,
no_builtins: no_builtins,
linker_info: linker_info,
windows_subsystem: windows_subsystem,
}
}
/// For each CGU, identify if we can reuse an existing object file (or
/// maybe other context).
fn trans_reuse_previous_work_products(tcx: TyCtxt,
codegen_units: &[CodegenUnit],
symbol_map: &SymbolMap)
-> Vec<Option<WorkProduct>> {
debug!("trans_reuse_previous_work_products()");
codegen_units
.iter()
.map(|cgu| {
let id = cgu.work_product_id();
let hash = cgu.compute_symbol_name_hash(tcx, symbol_map);
debug!("trans_reuse_previous_work_products: id={:?} hash={}", id, hash);
if let Some(work_product) = tcx.dep_graph.previous_work_product(&id) {
if work_product.input_hash == hash {
debug!("trans_reuse_previous_work_products: reusing {:?}", work_product);
return Some(work_product);
} else {
debug!("trans_reuse_previous_work_products: \
not reusing {:?} because hash changed to {:?}",
work_product, hash);
}
}
None
})
.collect()
}
fn collect_and_partition_translation_items<'a, 'tcx>(scx: &SharedCrateContext<'a, 'tcx>)
-> (Vec<CodegenUnit<'tcx>>, SymbolMap<'tcx>) {
let time_passes = scx.sess().time_passes();
let collection_mode = match scx.sess().opts.debugging_opts.print_trans_items {
Some(ref s) => {
let mode_string = s.to_lowercase();
let mode_string = mode_string.trim();
if mode_string == "eager" {
TransItemCollectionMode::Eager
} else {
if mode_string != "lazy" {
let message = format!("Unknown codegen-item collection mode '{}'. \
Falling back to 'lazy' mode.",
mode_string);
scx.sess().warn(&message);
}
TransItemCollectionMode::Lazy
}
}
None => TransItemCollectionMode::Lazy
};
let (items, inlining_map) =
time(time_passes, "translation item collection", || {
collector::collect_crate_translation_items(&scx, collection_mode)
});
let symbol_map = SymbolMap::build(scx, items.iter().cloned());
let strategy = if scx.sess().opts.debugging_opts.incremental.is_some() {
PartitioningStrategy::PerModule
} else {
PartitioningStrategy::FixedUnitCount(scx.sess().opts.cg.codegen_units)
};
let codegen_units = time(time_passes, "codegen unit partitioning", || {
partitioning::partition(scx,
items.iter().cloned(),
strategy,
&inlining_map)
});
assert!(scx.tcx().sess.opts.cg.codegen_units == codegen_units.len() ||
scx.tcx().sess.opts.debugging_opts.incremental.is_some());
{
let mut ccx_map = scx.translation_items().borrow_mut();
for trans_item in items.iter().cloned() {
ccx_map.insert(trans_item);
}
}
if scx.sess().opts.debugging_opts.print_trans_items.is_some() {
let mut item_to_cgus = FxHashMap();
for cgu in &codegen_units {
for (&trans_item, &linkage) in cgu.items() {
item_to_cgus.entry(trans_item)
.or_insert(Vec::new())
.push((cgu.name().clone(), linkage));
}
}
let mut item_keys: Vec<_> = items
.iter()
.map(|i| {
let mut output = i.to_string(scx.tcx());
output.push_str(" @@");
let mut empty = Vec::new();
let mut cgus = item_to_cgus.get_mut(i).unwrap_or(&mut empty);
cgus.as_mut_slice().sort_by_key(|&(ref name, _)| name.clone());
cgus.dedup();
for &(ref cgu_name, linkage) in cgus.iter() {
output.push_str(" ");
output.push_str(&cgu_name[..]);
let linkage_abbrev = match linkage {
llvm::Linkage::ExternalLinkage => "External",
llvm::Linkage::AvailableExternallyLinkage => "Available",
llvm::Linkage::LinkOnceAnyLinkage => "OnceAny",
llvm::Linkage::LinkOnceODRLinkage => "OnceODR",
llvm::Linkage::WeakAnyLinkage => "WeakAny",
llvm::Linkage::WeakODRLinkage => "WeakODR",
llvm::Linkage::AppendingLinkage => "Appending",
llvm::Linkage::InternalLinkage => "Internal",
llvm::Linkage::PrivateLinkage => "Private",
llvm::Linkage::ExternalWeakLinkage => "ExternalWeak",
llvm::Linkage::CommonLinkage => "Common",
};
output.push_str("[");
output.push_str(linkage_abbrev);
output.push_str("]");
}
output
})
.collect();
item_keys.sort();
for item in item_keys {
println!("TRANS_ITEM {}", item);
}
}
(codegen_units, symbol_map)
}
fn symbol_for_def_id<'a, 'tcx>(def_id: DefId,
scx: &SharedCrateContext<'a, 'tcx>,
symbol_map: &SymbolMap<'tcx>)
-> String {
// Just try to look things up in the symbol map. If nothing's there, we
// recompute.
if let Some(node_id) = scx.tcx().map.as_local_node_id(def_id) {
if let Some(sym) = symbol_map.get(TransItem::Static(node_id)) {
return sym.to_owned();
}
}
let instance = Instance::mono(scx, def_id);
symbol_map.get(TransItem::Fn(instance))
.map(str::to_owned)
.unwrap_or_else(|| instance.symbol_name(scx))
}