rust/src/librustc/middle/trans/context.rs

298 lines
11 KiB
Rust
Raw Normal View History

// Copyright 2013 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.
2013-06-12 21:02:33 -05:00
use driver::session;
2013-06-16 05:52:44 -05:00
use lib::llvm::{ContextRef, ModuleRef, ValueRef};
use lib::llvm::{llvm, TargetData, TypeNames};
use lib::llvm::mk_target_data;
2013-06-12 21:02:33 -05:00
use metadata::common::LinkMeta;
use middle::astencode;
use middle::resolve;
use middle::trans::adt;
use middle::trans::base;
use middle::trans::builder::Builder;
2013-06-12 21:02:33 -05:00
use middle::trans::debuginfo;
use middle::trans::common::{C_i32, C_null};
2013-06-12 21:02:33 -05:00
use middle::ty;
2013-06-15 05:16:47 -05:00
use middle::trans::type_::Type;
use util::sha2::Sha256;
use std::cell::RefCell;
use std::c_str::ToCStr;
use std::hashmap::{HashMap, HashSet};
use std::local_data;
use std::libc::c_uint;
use syntax::ast;
2013-06-12 21:02:33 -05:00
2013-06-21 03:28:33 -05:00
use middle::trans::common::{mono_id,ExternMap,tydesc_info,BuilderRef_res,Stats};
use middle::trans::base::{decl_crate_map};
2013-06-12 21:02:33 -05:00
pub struct CrateContext {
sess: session::Session,
llmod: ModuleRef,
llcx: ContextRef,
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 19:41:01 -06:00
metadata_llmod: ModuleRef,
2013-06-12 21:02:33 -05:00
td: TargetData,
2013-06-14 21:31:52 -05:00
tn: TypeNames,
2013-06-12 21:02:33 -05:00
externs: ExternMap,
intrinsics: HashMap<&'static str, ValueRef>,
item_vals: RefCell<HashMap<ast::NodeId, ValueRef>>,
2013-06-12 21:02:33 -05:00
exp_map2: resolve::ExportMap2,
reachable: @mut HashSet<ast::NodeId>,
2013-12-18 18:53:23 -06:00
item_symbols: RefCell<HashMap<ast::NodeId, ~str>>,
2013-06-12 21:02:33 -05:00
link_meta: LinkMeta,
tydescs: HashMap<ty::t, @mut tydesc_info>,
2013-06-12 21:02:33 -05:00
// Set when running emit_tydescs to enforce that no more tydescs are
// created.
finished_tydescs: bool,
2013-06-12 21:02:33 -05:00
// Track mapping of external ids to local items imported for inlining
external: HashMap<ast::DefId, Option<ast::NodeId>>,
// Backwards version of the `external` map (inlined items to where they
// came from)
external_srcs: HashMap<ast::NodeId, ast::DefId>,
// A set of static items which cannot be inlined into other crates. This
// will pevent in ii_item() structures from being encoded into the metadata
// that is generated
non_inlineable_statics: HashSet<ast::NodeId>,
2013-06-12 21:02:33 -05:00
// Cache instances of monomorphized functions
monomorphized: RefCell<HashMap<mono_id, ValueRef>>,
monomorphizing: RefCell<HashMap<ast::DefId, uint>>,
2013-06-12 21:02:33 -05:00
// Cache generated vtables
vtables: RefCell<HashMap<(ty::t, mono_id), ValueRef>>,
2013-06-12 21:02:33 -05:00
// Cache of constant strings,
const_cstr_cache: RefCell<HashMap<@str, ValueRef>>,
2013-06-12 21:02:33 -05:00
// Reverse-direction for const ptrs cast from globals.
// Key is an int, cast from a ValueRef holding a *T,
// Val is a ValueRef holding a *[T].
//
// Needed because LLVM loses pointer->pointee association
// when we ptrcast, and we have to ptrcast during translation
// of a [T] const because we form a slice, a [*T,int] pair, not
// a pointer to an LLVM array type.
const_globals: RefCell<HashMap<int, ValueRef>>,
2013-06-12 21:02:33 -05:00
// Cache of emitted const values
2013-12-18 19:08:56 -06:00
const_values: RefCell<HashMap<ast::NodeId, ValueRef>>,
2013-06-12 21:02:33 -05:00
// Cache of external const values
extern_const_values: RefCell<HashMap<ast::DefId, ValueRef>>,
2013-06-12 21:02:33 -05:00
impl_method_cache: RefCell<HashMap<(ast::DefId, ast::Name), ast::DefId>>,
2013-06-14 00:38:17 -05:00
module_data: RefCell<HashMap<~str, ValueRef>>,
2013-12-18 19:49:17 -06:00
lltypes: RefCell<HashMap<ty::t, Type>>,
2013-06-16 05:52:44 -05:00
llsizingtypes: HashMap<ty::t, Type>,
adt_reprs: HashMap<ty::t, @adt::Repr>,
symbol_hasher: Sha256,
type_hashcodes: HashMap<ty::t, @str>,
type_short_names: HashMap<ty::t, ~str>,
all_llvm_symbols: HashSet<@str>,
2013-06-12 21:02:33 -05:00
tcx: ty::ctxt,
maps: astencode::Maps,
2013-06-22 17:38:02 -05:00
stats: @mut Stats,
2013-06-15 05:16:47 -05:00
tydesc_type: Type,
int_type: Type,
opaque_vec_type: Type,
2013-06-12 21:02:33 -05:00
builder: BuilderRef_res,
crate_map: ValueRef,
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
2013-12-03 01:19:29 -06:00
crate_map_name: ~str,
2013-06-12 21:02:33 -05:00
// Set when at least one function uses GC. Needed so that
// decl_gc_metadata knows whether to link to the module metadata, which
// is not emitted by LLVM's GC pass when no functions use GC.
uses_gc: bool,
dbg_cx: Option<debuginfo::CrateDebugContext>,
do_not_commit_warning_issued: bool
2013-06-12 21:02:33 -05:00
}
impl CrateContext {
pub fn new(sess: session::Session,
name: &str,
tcx: ty::ctxt,
emap2: resolve::ExportMap2,
maps: astencode::Maps,
symbol_hasher: Sha256,
link_meta: LinkMeta,
reachable: @mut HashSet<ast::NodeId>)
-> CrateContext {
unsafe {
let llcx = llvm::LLVMContextCreate();
set_task_llcx(llcx);
let llmod = name.with_c_str(|buf| {
llvm::LLVMModuleCreateWithNameInContext(buf, llcx)
});
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 19:41:01 -06:00
let metadata_llmod = format!("{}_metadata", name).with_c_str(|buf| {
llvm::LLVMModuleCreateWithNameInContext(buf, llcx)
});
let data_layout: &str = sess.targ_cfg.target_strs.data_layout;
let targ_triple: &str = sess.targ_cfg.target_strs.target_triple;
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 19:41:01 -06:00
data_layout.with_c_str(|buf| {
llvm::LLVMSetDataLayout(llmod, buf);
llvm::LLVMSetDataLayout(metadata_llmod, buf);
});
targ_triple.with_c_str(|buf| {
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 19:41:01 -06:00
llvm::LLVMRustSetNormalizedTarget(llmod, buf);
llvm::LLVMRustSetNormalizedTarget(metadata_llmod, buf);
});
let targ_cfg = sess.targ_cfg;
2013-06-14 21:31:52 -05:00
let td = mk_target_data(sess.targ_cfg.target_strs.data_layout);
let mut tn = TypeNames::new();
2013-06-14 21:31:52 -05:00
let mut intrinsics = base::declare_intrinsics(llmod);
if sess.opts.extra_debuginfo {
base::declare_dbg_intrinsics(llmod, &mut intrinsics);
}
2013-06-15 05:16:47 -05:00
let int_type = Type::int(targ_cfg.arch);
let tydesc_type = Type::tydesc(targ_cfg.arch);
let opaque_vec_type = Type::opaque_vec(targ_cfg.arch);
2013-06-16 05:52:44 -05:00
let mut str_slice_ty = Type::named_struct("str_slice");
str_slice_ty.set_struct_body([Type::i8p(), int_type], false);
2013-06-15 22:45:48 -05:00
2013-06-15 05:16:47 -05:00
tn.associate_type("tydesc", &tydesc_type);
2013-06-15 22:45:48 -05:00
tn.associate_type("str_slice", &str_slice_ty);
2013-06-15 05:16:47 -05:00
let (crate_map_name, crate_map) = decl_crate_map(sess, link_meta.clone(), llmod);
let dbg_cx = if sess.opts.debuginfo {
Some(debuginfo::CrateDebugContext::new(llmod, name.to_owned()))
} else {
None
};
if sess.count_llvm_insns() {
base::init_insn_ctxt()
}
CrateContext {
sess: sess,
llmod: llmod,
llcx: llcx,
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 19:41:01 -06:00
metadata_llmod: metadata_llmod,
td: td,
tn: tn,
externs: HashMap::new(),
intrinsics: intrinsics,
item_vals: RefCell::new(HashMap::new()),
exp_map2: emap2,
reachable: reachable,
2013-12-18 18:53:23 -06:00
item_symbols: RefCell::new(HashMap::new()),
link_meta: link_meta,
tydescs: HashMap::new(),
finished_tydescs: false,
external: HashMap::new(),
external_srcs: HashMap::new(),
non_inlineable_statics: HashSet::new(),
monomorphized: RefCell::new(HashMap::new()),
monomorphizing: RefCell::new(HashMap::new()),
vtables: RefCell::new(HashMap::new()),
const_cstr_cache: RefCell::new(HashMap::new()),
const_globals: RefCell::new(HashMap::new()),
2013-12-18 19:08:56 -06:00
const_values: RefCell::new(HashMap::new()),
extern_const_values: RefCell::new(HashMap::new()),
impl_method_cache: RefCell::new(HashMap::new()),
module_data: RefCell::new(HashMap::new()),
2013-12-18 19:49:17 -06:00
lltypes: RefCell::new(HashMap::new()),
llsizingtypes: HashMap::new(),
adt_reprs: HashMap::new(),
symbol_hasher: symbol_hasher,
type_hashcodes: HashMap::new(),
type_short_names: HashMap::new(),
all_llvm_symbols: HashSet::new(),
tcx: tcx,
maps: maps,
2013-06-22 17:38:02 -05:00
stats: @mut Stats {
n_static_tydescs: 0u,
n_glues_created: 0u,
n_null_glues: 0u,
n_real_glues: 0u,
n_fns: 0u,
n_monos: 0u,
n_inlines: 0u,
n_closures: 0u,
n_llvm_insns: 0u,
llvm_insn_ctxt: ~[],
llvm_insns: HashMap::new(),
fn_stats: ~[]
},
tydesc_type: tydesc_type,
int_type: int_type,
2013-06-15 05:16:47 -05:00
opaque_vec_type: opaque_vec_type,
2013-06-20 00:52:02 -05:00
builder: BuilderRef_res(llvm::LLVMCreateBuilderInContext(llcx)),
crate_map: crate_map,
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
2013-12-03 01:19:29 -06:00
crate_map_name: crate_map_name,
uses_gc: false,
dbg_cx: dbg_cx,
do_not_commit_warning_issued: false
}
}
}
pub fn builder(@mut self) -> Builder {
Builder::new(self)
}
pub fn const_inbounds_gepi(&self,
pointer: ValueRef,
indices: &[uint]) -> ValueRef {
debug!("const_inbounds_gepi: pointer={} indices={:?}",
self.tn.val_to_str(pointer), indices);
let v: ~[ValueRef] =
2013-08-11 13:56:43 -05:00
indices.iter().map(|i| C_i32(*i as i32)).collect();
unsafe {
llvm::LLVMConstInBoundsGEP(pointer,
v.as_ptr(),
indices.len() as c_uint)
}
}
pub fn offsetof_gep(&self,
llptr_ty: Type,
indices: &[uint]) -> ValueRef {
/*!
* Returns the offset of applying the given GEP indices
* to an instance of `llptr_ty`. Similar to `offsetof` in C,
* except that `llptr_ty` must be a pointer type.
*/
unsafe {
let null = C_null(llptr_ty);
llvm::LLVMConstPtrToInt(self.const_inbounds_gepi(null, indices),
self.int_type.to_ref())
}
}
}
#[unsafe_destructor]
impl Drop for CrateContext {
2013-09-16 20:18:07 -05:00
fn drop(&mut self) {
unset_task_llcx();
}
}
2013-09-17 01:34:40 -05:00
local_data_key!(task_local_llcx_key: @ContextRef)
pub fn task_llcx() -> ContextRef {
let opt = local_data::get(task_local_llcx_key, |k| k.map(|k| *k));
*opt.expect("task-local LLVMContextRef wasn't ever set!")
}
fn set_task_llcx(c: ContextRef) {
local_data::set(task_local_llcx_key, @c);
}
fn unset_task_llcx() {
local_data::pop(task_local_llcx_key);
}