rust/compiler/rustc_codegen_llvm/src/lib.rs

387 lines
12 KiB
Rust
Raw Normal View History

//! The Rust compiler.
//!
//! # Note
//!
//! This API is completely unstable and subject to change.
2020-09-23 19:51:56 +00:00
#![doc(html_root_url = "https://doc.rust-lang.org/nightly/nightly-rustc/")]
2019-10-08 00:14:42 +00:00
#![feature(bool_to_option)]
#![feature(const_cstr_unchecked)]
#![feature(crate_visibility_modifier)]
#![feature(extern_types)]
#![feature(in_band_lifetimes)]
#![feature(nll)]
#![feature(or_patterns)]
#![recursion_limit = "256"]
2019-12-22 22:42:04 +00:00
use back::write::{create_informational_target_machine, create_target_machine};
2019-12-22 22:42:04 +00:00
pub use llvm_util::target_features;
use rustc_ast::expand::allocator::AllocatorKind;
2019-12-22 22:42:04 +00:00
use rustc_codegen_ssa::back::lto::{LtoModuleCodegen, SerializedModule, ThinModule};
use rustc_codegen_ssa::back::write::{
CodegenContext, FatLTOInput, ModuleConfig, TargetMachineFactoryConfig, TargetMachineFactoryFn,
};
use rustc_codegen_ssa::traits::*;
use rustc_codegen_ssa::ModuleCodegen;
use rustc_codegen_ssa::{CodegenResults, CompiledModule};
use rustc_data_structures::fx::FxHashMap;
2020-03-31 19:18:30 +00:00
use rustc_errors::{ErrorReported, FatalError, Handler};
use rustc_middle::dep_graph::{WorkProduct, WorkProductId};
2020-03-29 15:19:48 +00:00
use rustc_middle::middle::cstore::{EncodedMetadata, MetadataLoaderDyn};
use rustc_middle::ty::{self, TyCtxt};
use rustc_session::config::{OptLevel, OutputFilenames, PrintRequest};
use rustc_session::Session;
use rustc_span::symbol::Symbol;
rustc: Enable LTO and multiple codegen units This commit is a refactoring of the LTO backend in Rust to support compilations with multiple codegen units. The immediate result of this PR is to remove the artificial error emitted by rustc about `-C lto -C codegen-units-8`, but longer term this is intended to lay the groundwork for LTO with incremental compilation and ultimately be the underpinning of ThinLTO support. The problem here that needed solving is that when rustc is producing multiple codegen units in one compilation LTO needs to merge them all together. Previously only upstream dependencies were merged and it was inherently relied on that there was only one local codegen unit. Supporting this involved refactoring the optimization backend architecture for rustc, namely splitting the `optimize_and_codegen` function into `optimize` and `codegen`. After an LLVM module has been optimized it may be blocked and queued up for LTO, and only after LTO are modules code generated. Non-LTO compilations should look the same as they do today backend-wise, we'll spin up a thread for each codegen unit and optimize/codegen in that thread. LTO compilations will, however, send the LLVM module back to the coordinator thread once optimizations have finished. When all LLVM modules have finished optimizing the coordinator will invoke the LTO backend, producing a further list of LLVM modules. Currently this is always a list of one LLVM module. The coordinator then spawns further work to run LTO and code generation passes over each module. In the course of this refactoring a number of other pieces were refactored: * Management of the bytecode encoding in rlibs was centralized into one module instead of being scattered across LTO and linking. * Some internal refactorings on the link stage of the compiler was done to work directly from `CompiledModule` structures instead of lists of paths. * The trans time-graph output was tweaked a little to include a name on each bar and inflate the size of the bars a little
2017-07-23 15:14:38 +00:00
use std::any::Any;
use std::ffi::CStr;
2017-12-27 18:03:48 +00:00
mod back {
2019-03-30 14:30:07 +00:00
pub mod archive;
pub mod lto;
mod profiling;
pub mod write;
}
mod abi;
mod allocator;
mod asm;
mod attributes;
mod base;
mod builder;
mod callee;
mod common;
mod consts;
mod context;
mod coverageinfo;
mod debuginfo;
mod declare;
mod intrinsic;
// The following is a work around that replaces `pub mod llvm;` and that fixes issue 53912.
2019-12-22 22:42:04 +00:00
#[path = "llvm/mod.rs"]
mod llvm_;
pub mod llvm {
pub use super::llvm_::*;
}
mod llvm_util;
mod metadata;
2018-05-08 13:10:16 +00:00
mod mono_item;
mod type_;
mod type_of;
mod va_arg;
2019-12-22 22:42:04 +00:00
mod value;
#[derive(Clone)]
2018-05-08 13:10:16 +00:00
pub struct LlvmCodegenBackend(());
2017-09-16 15:27:29 +00:00
impl ExtraBackendMethods for LlvmCodegenBackend {
2019-06-13 21:48:52 +00:00
fn new_metadata(&self, tcx: TyCtxt<'_>, mod_name: &str) -> ModuleLlvm {
ModuleLlvm::new_metadata(tcx, mod_name)
2018-09-25 15:52:03 +00:00
}
2019-06-13 21:48:52 +00:00
fn write_compressed_metadata<'tcx>(
2018-09-25 15:52:03 +00:00
&self,
2019-06-13 21:48:52 +00:00
tcx: TyCtxt<'tcx>,
metadata: &EncodedMetadata,
llvm_module: &mut ModuleLlvm,
) {
base::write_compressed_metadata(tcx, metadata, llvm_module)
2018-09-25 15:52:03 +00:00
}
2019-06-13 21:48:52 +00:00
fn codegen_allocator<'tcx>(
&self,
2019-06-13 21:48:52 +00:00
tcx: TyCtxt<'tcx>,
mods: &mut ModuleLlvm,
kind: AllocatorKind,
has_alloc_error_handler: bool,
) {
unsafe { allocator::codegen(tcx, mods, kind, has_alloc_error_handler) }
}
fn compile_codegen_unit(
2019-12-22 22:42:04 +00:00
&self,
tcx: TyCtxt<'_>,
cgu_name: Symbol,
) -> (ModuleCodegen<ModuleLlvm>, u64) {
base::compile_codegen_unit(tcx, cgu_name)
2018-09-25 15:52:03 +00:00
}
fn target_machine_factory(
2018-09-25 15:52:03 +00:00
&self,
sess: &Session,
optlvl: OptLevel,
) -> TargetMachineFactoryFn<Self> {
back::write::target_machine_factory(sess, optlvl)
2018-09-25 15:52:03 +00:00
}
fn target_cpu<'b>(&self, sess: &'b Session) -> &'b str {
llvm_util::target_cpu(sess)
}
fn tune_cpu<'b>(&self, sess: &'b Session) -> Option<&'b str> {
llvm_util::tune_cpu(sess)
}
}
impl WriteBackendMethods for LlvmCodegenBackend {
type Module = ModuleLlvm;
type ModuleBuffer = back::lto::ModuleBuffer;
type Context = llvm::Context;
type TargetMachine = &'static mut llvm::TargetMachine;
type ThinData = back::lto::ThinData;
type ThinBuffer = back::lto::ThinBuffer;
fn print_pass_timings(&self) {
2019-12-22 22:42:04 +00:00
unsafe {
llvm::LLVMRustPrintPassTimings();
}
2018-09-25 15:52:03 +00:00
}
fn run_link(
cgcx: &CodegenContext<Self>,
diag_handler: &Handler,
modules: Vec<ModuleCodegen<Self::Module>>,
) -> Result<ModuleCodegen<Self::Module>, FatalError> {
back::write::link(cgcx, diag_handler, modules)
}
fn run_fat_lto(
cgcx: &CodegenContext<Self>,
modules: Vec<FatLTOInput<Self>>,
cached_modules: Vec<(SerializedModule<Self::ModuleBuffer>, WorkProduct)>,
) -> Result<LtoModuleCodegen<Self>, FatalError> {
2019-02-13 13:13:30 +00:00
back::lto::run_fat(cgcx, modules, cached_modules)
}
fn run_thin_lto(
cgcx: &CodegenContext<Self>,
modules: Vec<(String, Self::ThinBuffer)>,
cached_modules: Vec<(SerializedModule<Self::ModuleBuffer>, WorkProduct)>,
) -> Result<(Vec<LtoModuleCodegen<Self>>, Vec<WorkProduct>), FatalError> {
2019-02-13 13:13:30 +00:00
back::lto::run_thin(cgcx, modules, cached_modules)
2018-09-25 15:52:03 +00:00
}
unsafe fn optimize(
cgcx: &CodegenContext<Self>,
diag_handler: &Handler,
module: &ModuleCodegen<Self::Module>,
config: &ModuleConfig,
) -> Result<(), FatalError> {
2019-02-13 13:13:30 +00:00
back::write::optimize(cgcx, diag_handler, module, config)
2018-09-25 15:52:03 +00:00
}
unsafe fn optimize_thin(
cgcx: &CodegenContext<Self>,
thin: &mut ThinModule<Self>,
) -> Result<ModuleCodegen<Self::Module>, FatalError> {
2019-02-13 13:13:30 +00:00
back::lto::optimize_thin_module(thin, cgcx)
2018-09-25 15:52:03 +00:00
}
unsafe fn codegen(
cgcx: &CodegenContext<Self>,
diag_handler: &Handler,
module: ModuleCodegen<Self::Module>,
config: &ModuleConfig,
) -> Result<CompiledModule, FatalError> {
2019-02-13 13:13:30 +00:00
back::write::codegen(cgcx, diag_handler, module, config)
2018-09-25 15:52:03 +00:00
}
2019-12-22 22:42:04 +00:00
fn prepare_thin(module: ModuleCodegen<Self::Module>) -> (String, Self::ThinBuffer) {
back::lto::prepare_thin(module)
}
2019-12-22 22:42:04 +00:00
fn serialize_module(module: ModuleCodegen<Self::Module>) -> (String, Self::ModuleBuffer) {
(module.name, back::lto::ModuleBuffer::new(module.module_llvm.llmod()))
}
fn run_lto_pass_manager(
cgcx: &CodegenContext<Self>,
module: &ModuleCodegen<Self::Module>,
config: &ModuleConfig,
2019-12-22 22:42:04 +00:00
thin: bool,
) {
back::lto::run_pass_manager(cgcx, module, config, thin)
}
}
unsafe impl Send for LlvmCodegenBackend {} // Llvm is on a per-thread basis
unsafe impl Sync for LlvmCodegenBackend {}
2018-05-08 13:10:16 +00:00
impl LlvmCodegenBackend {
pub fn new() -> Box<dyn CodegenBackend> {
Box::new(LlvmCodegenBackend(()))
2017-09-16 15:27:29 +00:00
}
}
impl CodegenBackend for LlvmCodegenBackend {
fn init(&self, sess: &Session) {
llvm_util::init(sess); // Make sure llvm is inited
}
fn print(&self, req: PrintRequest, sess: &Session) {
match req {
PrintRequest::RelocationModels => {
println!("Available relocation models:");
for name in
&["static", "pic", "dynamic-no-pic", "ropi", "rwpi", "ropi-rwpi", "default"]
{
println!(" {}", name);
}
2019-09-13 09:36:35 +00:00
println!();
}
PrintRequest::CodeModels => {
println!("Available code models:");
2020-05-20 20:09:19 +00:00
for name in &["tiny", "small", "kernel", "medium", "large"] {
println!(" {}", name);
}
2019-09-13 09:36:35 +00:00
println!();
}
PrintRequest::TlsModels => {
println!("Available TLS models:");
for name in &["global-dynamic", "local-dynamic", "initial-exec", "local-exec"] {
println!(" {}", name);
}
2019-09-13 09:36:35 +00:00
println!();
}
req => llvm_util::print(req, sess),
}
}
2017-09-16 15:27:29 +00:00
fn print_passes(&self) {
llvm_util::print_passes();
}
fn print_version(&self) {
llvm_util::print_version();
}
fn target_features(&self, sess: &Session) -> Vec<Symbol> {
target_features(sess)
}
fn metadata_loader(&self) -> Box<MetadataLoaderDyn> {
Box::new(metadata::LlvmMetadataLoader)
2017-09-16 15:27:29 +00:00
}
fn provide(&self, providers: &mut ty::query::Providers) {
attributes::provide_both(providers);
2017-09-23 12:46:15 +00:00
}
fn provide_extern(&self, providers: &mut ty::query::Providers) {
attributes::provide_both(providers);
2017-09-16 15:27:29 +00:00
}
fn codegen_crate<'tcx>(
&self,
2019-06-13 21:48:52 +00:00
tcx: TyCtxt<'tcx>,
metadata: EncodedMetadata,
need_metadata_module: bool,
) -> Box<dyn Any> {
Box::new(rustc_codegen_ssa::base::codegen_crate(
2019-12-22 22:42:04 +00:00
LlvmCodegenBackend(()),
tcx,
metadata,
need_metadata_module,
))
2017-09-16 15:27:29 +00:00
}
fn join_codegen(
&self,
ongoing_codegen: Box<dyn Any>,
2017-09-16 15:27:29 +00:00
sess: &Session,
) -> Result<(CodegenResults, FxHashMap<WorkProductId, WorkProduct>), ErrorReported> {
2019-12-22 22:42:04 +00:00
let (codegen_results, work_products) = ongoing_codegen
.downcast::<rustc_codegen_ssa::back::write::OngoingCodegen<LlvmCodegenBackend>>()
.expect("Expected LlvmCodegenBackend's OngoingCodegen, found Box<Any>")
.join(sess);
2017-09-16 15:27:29 +00:00
sess.time("llvm_dump_timing_file", || {
if sess.opts.debugging_opts.llvm_time_trace {
llvm_util::time_trace_profiler_finish("llvm_timings.json");
}
2019-12-22 22:42:04 +00:00
});
Ok((codegen_results, work_products))
}
fn link(
&self,
sess: &Session,
codegen_results: CodegenResults,
outputs: &OutputFilenames,
) -> Result<(), ErrorReported> {
// Run the linker on any artifacts that resulted from the LLVM run.
// This should produce either a finished executable or library.
sess.time("link_crate", || {
2019-03-30 14:30:07 +00:00
use crate::back::archive::LlvmArchiveBuilder;
2019-12-22 22:42:04 +00:00
use rustc_codegen_ssa::back::link::link_binary;
2019-03-30 14:30:07 +00:00
let target_cpu = crate::llvm_util::target_cpu(sess);
link_binary::<LlvmArchiveBuilder<'_>>(
sess,
&codegen_results,
outputs,
&codegen_results.crate_name.as_str(),
target_cpu,
);
});
Ok(())
2017-09-16 15:27:29 +00:00
}
}
2018-09-25 15:52:03 +00:00
pub struct ModuleLlvm {
llcx: &'static mut llvm::Context,
llmod_raw: *const llvm::Module,
tm: &'static mut llvm::TargetMachine,
}
2019-12-22 22:42:04 +00:00
unsafe impl Send for ModuleLlvm {}
unsafe impl Sync for ModuleLlvm {}
rustc: Enable LTO and multiple codegen units This commit is a refactoring of the LTO backend in Rust to support compilations with multiple codegen units. The immediate result of this PR is to remove the artificial error emitted by rustc about `-C lto -C codegen-units-8`, but longer term this is intended to lay the groundwork for LTO with incremental compilation and ultimately be the underpinning of ThinLTO support. The problem here that needed solving is that when rustc is producing multiple codegen units in one compilation LTO needs to merge them all together. Previously only upstream dependencies were merged and it was inherently relied on that there was only one local codegen unit. Supporting this involved refactoring the optimization backend architecture for rustc, namely splitting the `optimize_and_codegen` function into `optimize` and `codegen`. After an LLVM module has been optimized it may be blocked and queued up for LTO, and only after LTO are modules code generated. Non-LTO compilations should look the same as they do today backend-wise, we'll spin up a thread for each codegen unit and optimize/codegen in that thread. LTO compilations will, however, send the LLVM module back to the coordinator thread once optimizations have finished. When all LLVM modules have finished optimizing the coordinator will invoke the LTO backend, producing a further list of LLVM modules. Currently this is always a list of one LLVM module. The coordinator then spawns further work to run LTO and code generation passes over each module. In the course of this refactoring a number of other pieces were refactored: * Management of the bytecode encoding in rlibs was centralized into one module instead of being scattered across LTO and linking. * Some internal refactorings on the link stage of the compiler was done to work directly from `CompiledModule` structures instead of lists of paths. * The trans time-graph output was tweaked a little to include a name on each bar and inflate the size of the bars a little
2017-07-23 15:14:38 +00:00
impl ModuleLlvm {
2019-06-13 21:48:52 +00:00
fn new(tcx: TyCtxt<'_>, mod_name: &str) -> Self {
unsafe {
let llcx = llvm::LLVMRustContextCreate(tcx.sess.fewer_names());
let llmod_raw = context::create_module(tcx, llcx, mod_name) as *const _;
ModuleLlvm { llmod_raw, llcx, tm: create_target_machine(tcx, mod_name) }
}
}
2019-06-13 21:48:52 +00:00
fn new_metadata(tcx: TyCtxt<'_>, mod_name: &str) -> Self {
unsafe {
let llcx = llvm::LLVMRustContextCreate(tcx.sess.fewer_names());
let llmod_raw = context::create_module(tcx, llcx, mod_name) as *const _;
ModuleLlvm { llmod_raw, llcx, tm: create_informational_target_machine(tcx.sess) }
}
}
fn parse(
cgcx: &CodegenContext<LlvmCodegenBackend>,
name: &CStr,
buffer: &[u8],
handler: &Handler,
) -> Result<Self, FatalError> {
unsafe {
let llcx = llvm::LLVMRustContextCreate(cgcx.fewer_names);
let llmod_raw = back::lto::parse_module(llcx, name, buffer, handler)?;
let split_dwarf_file = cgcx
.output_filenames
.split_dwarf_filename(cgcx.split_dwarf_kind, Some(name.to_str().unwrap()));
let tm_factory_config = TargetMachineFactoryConfig { split_dwarf_file };
let tm = match (cgcx.tm_factory)(tm_factory_config) {
Ok(m) => m,
Err(e) => {
handler.struct_err(&e).emit();
2019-12-22 22:42:04 +00:00
return Err(FatalError);
}
};
2019-12-22 22:42:04 +00:00
Ok(ModuleLlvm { llmod_raw, llcx, tm })
}
}
fn llmod(&self) -> &llvm::Module {
2019-12-22 22:42:04 +00:00
unsafe { &*self.llmod_raw }
}
}
impl Drop for ModuleLlvm {
rustc: Enable LTO and multiple codegen units This commit is a refactoring of the LTO backend in Rust to support compilations with multiple codegen units. The immediate result of this PR is to remove the artificial error emitted by rustc about `-C lto -C codegen-units-8`, but longer term this is intended to lay the groundwork for LTO with incremental compilation and ultimately be the underpinning of ThinLTO support. The problem here that needed solving is that when rustc is producing multiple codegen units in one compilation LTO needs to merge them all together. Previously only upstream dependencies were merged and it was inherently relied on that there was only one local codegen unit. Supporting this involved refactoring the optimization backend architecture for rustc, namely splitting the `optimize_and_codegen` function into `optimize` and `codegen`. After an LLVM module has been optimized it may be blocked and queued up for LTO, and only after LTO are modules code generated. Non-LTO compilations should look the same as they do today backend-wise, we'll spin up a thread for each codegen unit and optimize/codegen in that thread. LTO compilations will, however, send the LLVM module back to the coordinator thread once optimizations have finished. When all LLVM modules have finished optimizing the coordinator will invoke the LTO backend, producing a further list of LLVM modules. Currently this is always a list of one LLVM module. The coordinator then spawns further work to run LTO and code generation passes over each module. In the course of this refactoring a number of other pieces were refactored: * Management of the bytecode encoding in rlibs was centralized into one module instead of being scattered across LTO and linking. * Some internal refactorings on the link stage of the compiler was done to work directly from `CompiledModule` structures instead of lists of paths. * The trans time-graph output was tweaked a little to include a name on each bar and inflate the size of the bars a little
2017-07-23 15:14:38 +00:00
fn drop(&mut self) {
unsafe {
llvm::LLVMContextDispose(&mut *(self.llcx as *mut _));
llvm::LLVMRustDisposeTargetMachine(&mut *(self.tm as *mut _));
rustc: Enable LTO and multiple codegen units This commit is a refactoring of the LTO backend in Rust to support compilations with multiple codegen units. The immediate result of this PR is to remove the artificial error emitted by rustc about `-C lto -C codegen-units-8`, but longer term this is intended to lay the groundwork for LTO with incremental compilation and ultimately be the underpinning of ThinLTO support. The problem here that needed solving is that when rustc is producing multiple codegen units in one compilation LTO needs to merge them all together. Previously only upstream dependencies were merged and it was inherently relied on that there was only one local codegen unit. Supporting this involved refactoring the optimization backend architecture for rustc, namely splitting the `optimize_and_codegen` function into `optimize` and `codegen`. After an LLVM module has been optimized it may be blocked and queued up for LTO, and only after LTO are modules code generated. Non-LTO compilations should look the same as they do today backend-wise, we'll spin up a thread for each codegen unit and optimize/codegen in that thread. LTO compilations will, however, send the LLVM module back to the coordinator thread once optimizations have finished. When all LLVM modules have finished optimizing the coordinator will invoke the LTO backend, producing a further list of LLVM modules. Currently this is always a list of one LLVM module. The coordinator then spawns further work to run LTO and code generation passes over each module. In the course of this refactoring a number of other pieces were refactored: * Management of the bytecode encoding in rlibs was centralized into one module instead of being scattered across LTO and linking. * Some internal refactorings on the link stage of the compiler was done to work directly from `CompiledModule` structures instead of lists of paths. * The trans time-graph output was tweaked a little to include a name on each bar and inflate the size of the bars a little
2017-07-23 15:14:38 +00:00
}
}
}