user0/rust
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Rollup merge of #152566 - bjorn3:cg_gcc_no_thin_lto, r=antoyo

Browse source 
Remove code for ThinLTO from cg_gcc

It was just a dummy implementation to workarround the fact that thin local lto is the default in rustc. By adding a thin_lto_supported thin local lto can be automatically disabled for cg_gcc, removing the need for this dummy implementation. This makes improvements to the LTO handling on the cg_ssa side a lot easier.

cc [#rustc-codegen-gcc > thin LTO implementation](https://rust-lang.zulipchat.com/#narrow/channel/386786-rustc-codegen-gcc/topic/thin.20LTO.20implementation/with/573625132)
This should make the work on https://github.com/rust-lang/compiler-team/issues/908 easier.

r? rust-lang/wg-gcc-backend
This commit is contained in:
Jonathan Brouwer 2026-02-15 16:37:36 +01:00 • committed by GitHub
commit 4a4ea14148
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 56 additions and 409 deletions
Download patch file Download diff file Expand all files Collapse all files

395
compiler/rustc_codegen_gcc/src/back/lto.rs
View file

@ -17,15 +17,13 @@
// /usr/bin/ld: warning: type of symbol `_RNvNvNvNvNtNtNtCsAj5i4SGTR7_3std4sync4mpmc5waker17current_thread_id5DUMMY7___getit5___KEY' changed from 1 to 6 in /tmp/ccKeUSiR.ltrans0.ltrans.o // /usr/bin/ld: warning: type of symbol `_RNvNvNvNvNtNtNtCsAj5i4SGTR7_3std4sync4mpmc5waker17current_thread_id5DUMMY7___getit5___KEY' changed from 1 to 6 in /tmp/ccKeUSiR.ltrans0.ltrans.o
// /usr/bin/ld: warning: incremental linking of LTO and non-LTO objects; using -flinker-output=nolto-rel which will bypass whole program optimization // /usr/bin/ld: warning: incremental linking of LTO and non-LTO objects; using -flinker-output=nolto-rel which will bypass whole program optimization
// cSpell:enable // cSpell:enable
use std::ffi::{CStr, CString}; use std::ffi::CString;
use std::fs::{self, File}; use std::fs::{self, File};
use std::path::{Path, PathBuf}; use std::path::{Path, PathBuf};
use std::sync::Arc;
use std::sync::atomic::Ordering;
use gccjit::{Context, OutputKind}; use gccjit::OutputKind;
use object::read::archive::ArchiveFile; use object::read::archive::ArchiveFile;
use rustc_codegen_ssa::back::lto::{SerializedModule, ThinModule, ThinShared}; use rustc_codegen_ssa::back::lto::SerializedModule;
use rustc_codegen_ssa::back::write::{CodegenContext, FatLtoInput, SharedEmitter}; use rustc_codegen_ssa::back::write::{CodegenContext, FatLtoInput, SharedEmitter};
use rustc_codegen_ssa::traits::*; use rustc_codegen_ssa::traits::*;
use rustc_codegen_ssa::{ModuleCodegen, ModuleKind, looks_like_rust_object_file}; use rustc_codegen_ssa::{ModuleCodegen, ModuleKind, looks_like_rust_object_file};
@ -33,15 +31,12 @@ use rustc_data_structures::memmap::Mmap;
use rustc_data_structures::profiling::SelfProfilerRef; use rustc_data_structures::profiling::SelfProfilerRef;
use rustc_errors::{DiagCtxt, DiagCtxtHandle}; use rustc_errors::{DiagCtxt, DiagCtxtHandle};
use rustc_log::tracing::info; use rustc_log::tracing::info;
use rustc_middle::bug;
use rustc_middle::dep_graph::WorkProduct;
use rustc_session::config::Lto; use rustc_session::config::Lto;
use rustc_target::spec::RelocModel;
use tempfile::{TempDir, tempdir}; use tempfile::{TempDir, tempdir};
use crate::back::write::save_temp_bitcode; use crate::back::write::save_temp_bitcode;
use crate::errors::LtoBitcodeFromRlib; use crate::errors::LtoBitcodeFromRlib;
use crate::{GccCodegenBackend, GccContext, LTO_SUPPORTED, LtoMode, SyncContext, to_gcc_opt_level}; use crate::{GccCodegenBackend, GccContext, LtoMode, to_gcc_opt_level};
struct LtoData { struct LtoData {
// TODO(antoyo): use symbols_below_threshold. // TODO(antoyo): use symbols_below_threshold.
@ -281,385 +276,3 @@ impl ModuleBufferMethods for ModuleBuffer {
&[] &[]
} }
} }
/// Performs thin LTO by performing necessary global analysis and returning two
/// lists, one of the modules that need optimization and another for modules that
/// can simply be copied over from the incr. comp. cache.
pub(crate) fn run_thin(
cgcx: &CodegenContext,
prof: &SelfProfilerRef,
dcx: DiagCtxtHandle<'_>,
each_linked_rlib_for_lto: &[PathBuf],
modules: Vec<(String, ThinBuffer)>,
cached_modules: Vec<(SerializedModule<ModuleBuffer>, WorkProduct)>,
) -> (Vec<ThinModule<GccCodegenBackend>>, Vec<WorkProduct>) {
let lto_data = prepare_lto(cgcx, each_linked_rlib_for_lto, dcx);
if cgcx.use_linker_plugin_lto {
unreachable!(
"We should never reach this case if the LTO step \
is deferred to the linker"
);
}
thin_lto(
cgcx,
prof,
dcx,
modules,
lto_data.upstream_modules,
lto_data.tmp_path,
cached_modules,
//&lto_data.symbols_below_threshold,
)
}
pub(crate) fn prepare_thin(module: ModuleCodegen<GccContext>) -> (String, ThinBuffer) {
let name = module.name;
//let buffer = ThinBuffer::new(module.module_llvm.context, true);
let buffer = ThinBuffer::new(&module.module_llvm.context);
(name, buffer)
}
/// Prepare "thin" LTO to get run on these modules.
///
/// The general structure of ThinLTO is quite different from the structure of
/// "fat" LTO above. With "fat" LTO all LLVM modules in question are merged into
/// one giant LLVM module, and then we run more optimization passes over this
/// big module after internalizing most symbols. Thin LTO, on the other hand,
/// avoid this large bottleneck through more targeted optimization.
///
/// At a high level Thin LTO looks like:
///
/// 1. Prepare a "summary" of each LLVM module in question which describes
/// the values inside, cost of the values, etc.
/// 2. Merge the summaries of all modules in question into one "index"
/// 3. Perform some global analysis on this index
/// 4. For each module, use the index and analysis calculated previously to
/// perform local transformations on the module, for example inlining
/// small functions from other modules.
/// 5. Run thin-specific optimization passes over each module, and then code
/// generate everything at the end.
///
/// The summary for each module is intended to be quite cheap, and the global
/// index is relatively quite cheap to create as well. As a result, the goal of
/// ThinLTO is to reduce the bottleneck on LTO and enable LTO to be used in more
/// situations. For example one cheap optimization is that we can parallelize
/// all codegen modules, easily making use of all the cores on a machine.
///
/// With all that in mind, the function here is designed at specifically just
/// calculating the *index* for ThinLTO. This index will then be shared amongst
/// all of the `LtoModuleCodegen` units returned below and destroyed once
/// they all go out of scope.
fn thin_lto(
_cgcx: &CodegenContext,
prof: &SelfProfilerRef,
_dcx: DiagCtxtHandle<'_>,
modules: Vec<(String, ThinBuffer)>,
serialized_modules: Vec<(SerializedModule<ModuleBuffer>, CString)>,
tmp_path: TempDir,
cached_modules: Vec<(SerializedModule<ModuleBuffer>, WorkProduct)>,
//_symbols_below_threshold: &[String],
) -> (Vec<ThinModule<GccCodegenBackend>>, Vec<WorkProduct>) {
let _timer = prof.generic_activity("LLVM_thin_lto_global_analysis");
info!("going for that thin, thin LTO");
/*let green_modules: FxHashMap<_, _> =
cached_modules.iter().map(|(_, wp)| (wp.cgu_name.clone(), wp.clone())).collect();*/
let full_scope_len = modules.len() + serialized_modules.len() + cached_modules.len();
let mut thin_buffers = Vec::with_capacity(modules.len());
let mut module_names = Vec::with_capacity(full_scope_len);
//let mut thin_modules = Vec::with_capacity(full_scope_len);
for (i, (name, buffer)) in modules.into_iter().enumerate() {
info!("local module: {} - {}", i, name);
let cname = CString::new(name.as_bytes()).unwrap();
/*thin_modules.push(llvm::ThinLTOModule {
identifier: cname.as_ptr(),
data: buffer.data().as_ptr(),
len: buffer.data().len(),
});*/
thin_buffers.push(buffer);
module_names.push(cname);
}
// FIXME: All upstream crates are deserialized internally in the
// function below to extract their summary and modules. Note that
// unlike the loop above we *must* decode and/or read something
// here as these are all just serialized files on disk. An
// improvement, however, to make here would be to store the
// module summary separately from the actual module itself. Right
// now this is store in one large bitcode file, and the entire
// file is deflate-compressed. We could try to bypass some of the
// decompression by storing the index uncompressed and only
// lazily decompressing the bytecode if necessary.
//
// Note that truly taking advantage of this optimization will
// likely be further down the road. We'd have to implement
// incremental ThinLTO first where we could actually avoid
// looking at upstream modules entirely sometimes (the contents,
// we must always unconditionally look at the index).
let mut serialized = Vec::with_capacity(serialized_modules.len() + cached_modules.len());
let cached_modules =
cached_modules.into_iter().map(|(sm, wp)| (sm, CString::new(wp.cgu_name).unwrap()));
for (module, name) in serialized_modules.into_iter().chain(cached_modules) {
info!("upstream or cached module {:?}", name);
/*thin_modules.push(llvm::ThinLTOModule {
identifier: name.as_ptr(),
data: module.data().as_ptr(),
len: module.data().len(),
});*/
match module {
SerializedModule::Local(_) => {
//let path = module_buffer.0.to_str().expect("path");
//let my_path = PathBuf::from(path);
//let exists = my_path.exists();
/*module.module_llvm.should_combine_object_files = true;
module
.module_llvm
.context
.add_driver_option(module_buffer.0.to_str().expect("path"));*/
}
SerializedModule::FromRlib(_) => unimplemented!("from rlib"),
SerializedModule::FromUncompressedFile(_) => {
unimplemented!("from uncompressed file")
}
}
serialized.push(module);
module_names.push(name);
}
// Sanity check
//assert_eq!(thin_modules.len(), module_names.len());
// Delegate to the C++ bindings to create some data here. Once this is a
// tried-and-true interface we may wish to try to upstream some of this
// to LLVM itself, right now we reimplement a lot of what they do
// upstream...
/*let data = llvm::LLVMRustCreateThinLTOData(
thin_modules.as_ptr(),
thin_modules.len() as u32,
symbols_below_threshold.as_ptr(),
symbols_below_threshold.len() as u32,
)
.ok_or_else(|| write::llvm_err(dcx, LlvmError::PrepareThinLtoContext))?;
*/
let data = ThinData; //(Arc::new(tmp_path))/*(data)*/;
info!("thin LTO data created");
/*let (key_map_path, prev_key_map, curr_key_map) =
if let Some(ref incr_comp_session_dir) = cgcx.incr_comp_session_dir {
let path = incr_comp_session_dir.join(THIN_LTO_KEYS_INCR_COMP_FILE_NAME);
// If the previous file was deleted, or we get an IO error
// reading the file, then we'll just use `None` as the
// prev_key_map, which will force the code to be recompiled.
let prev =
if path.exists() { ThinLTOKeysMap::load_from_file(&path).ok() } else { None };
let curr = ThinLTOKeysMap::from_thin_lto_modules(&data, &thin_modules, &module_names);
(Some(path), prev, curr)
}
else {
// If we don't compile incrementally, we don't need to load the
// import data from LLVM.
assert!(green_modules.is_empty());
let curr = ThinLTOKeysMap::default();
(None, None, curr)
};
info!("thin LTO cache key map loaded");
info!("prev_key_map: {:#?}", prev_key_map);
info!("curr_key_map: {:#?}", curr_key_map);*/
// Throw our data in an `Arc` as we'll be sharing it across threads. We
// also put all memory referenced by the C++ data (buffers, ids, etc)
// into the arc as well. After this we'll create a thin module
// codegen per module in this data.
let shared =
Arc::new(ThinShared { data, thin_buffers, serialized_modules: serialized, module_names });
let copy_jobs = vec![];
let mut opt_jobs = vec![];
info!("checking which modules can be-reused and which have to be re-optimized.");
for (module_index, module_name) in shared.module_names.iter().enumerate() {
let module_name = module_name_to_str(module_name);
/*if let (Some(prev_key_map), true) =
(prev_key_map.as_ref(), green_modules.contains_key(module_name))
{
assert!(cgcx.incr_comp_session_dir.is_some());
// If a module exists in both the current and the previous session,
// and has the same LTO cache key in both sessions, then we can re-use it
if prev_key_map.keys.get(module_name) == curr_key_map.keys.get(module_name) {
let work_product = green_modules[module_name].clone();
copy_jobs.push(work_product);
info!(" - {}: re-used", module_name);
assert!(cgcx.incr_comp_session_dir.is_some());
continue;
}
}*/
info!(" - {}: re-compiled", module_name);
opt_jobs.push(ThinModule { shared: shared.clone(), idx: module_index });
}
// Save the current ThinLTO import information for the next compilation
// session, overwriting the previous serialized data (if any).
/*if let Some(path) = key_map_path {
if let Err(err) = curr_key_map.save_to_file(&path) {
return Err(write::llvm_err(dcx, LlvmError::WriteThinLtoKey { err }));
}
}*/
// NOTE: save the temporary directory used by LTO so that it gets deleted after linking instead
// of now.
//module.module_llvm.temp_dir = Some(tmp_path);
// TODO: save the directory so that it gets deleted later.
std::mem::forget(tmp_path);
(opt_jobs, copy_jobs)
}
pub fn optimize_thin_module(
thin_module: ThinModule<GccCodegenBackend>,
_cgcx: &CodegenContext,
) -> ModuleCodegen<GccContext> {
//let module_name = &thin_module.shared.module_names[thin_module.idx];
// Right now the implementation we've got only works over serialized
// modules, so we create a fresh new LLVM context and parse the module
// into that context. One day, however, we may do this for upstream
// crates but for locally codegened modules we may be able to reuse
// that LLVM Context and Module.
//let llcx = llvm::LLVMRustContextCreate(cgcx.fewer_names);
//let llmod_raw = parse_module(llcx, module_name, thin_module.data(), &dcx)? as *const _;
let mut lto_mode = LtoMode::None;
let context = match thin_module.shared.thin_buffers.get(thin_module.idx) {
Some(thin_buffer) => Arc::clone(&thin_buffer.context),
None => {
let context = Context::default();
let len = thin_module.shared.thin_buffers.len();
let module = &thin_module.shared.serialized_modules[thin_module.idx - len];
match *module {
SerializedModule::Local(ref module_buffer) => {
let path = module_buffer.0.to_str().expect("path");
context.add_driver_option(path);
lto_mode = LtoMode::Thin;
/*module.module_llvm.should_combine_object_files = true;
module
.module_llvm
.context
.add_driver_option(module_buffer.0.to_str().expect("path"));*/
}
SerializedModule::FromRlib(_) => unimplemented!("from rlib"),
SerializedModule::FromUncompressedFile(_) => {
unimplemented!("from uncompressed file")
}
}
Arc::new(SyncContext::new(context))
}
};
let lto_supported = LTO_SUPPORTED.load(Ordering::SeqCst);
let module = ModuleCodegen::new_regular(
thin_module.name().to_string(),
GccContext {
context,
lto_mode,
lto_supported,
// TODO(antoyo): use the correct relocation model here.
relocation_model: RelocModel::Pic,
temp_dir: None,
},
);
/*{
let target = &*module.module_llvm.tm;
let llmod = module.module_llvm.llmod();
save_temp_bitcode(cgcx, &module, "thin-lto-input");
// Up next comes the per-module local analyses that we do for Thin LTO.
// Each of these functions is basically copied from the LLVM
// implementation and then tailored to suit this implementation. Ideally
// each of these would be supported by upstream LLVM but that's perhaps
// a patch for another day!
//
// You can find some more comments about these functions in the LLVM
// bindings we've got (currently `PassWrapper.cpp`)
{
let _timer =
cgcx.prof.generic_activity_with_arg("LLVM_thin_lto_rename", thin_module.name());
unsafe { llvm::LLVMRustPrepareThinLTORename(thin_module.shared.data.0, llmod, target) };
save_temp_bitcode(cgcx, &module, "thin-lto-after-rename");
}
{
let _timer = cgcx
.prof
.generic_activity_with_arg("LLVM_thin_lto_resolve_weak", thin_module.name());
if !llvm::LLVMRustPrepareThinLTOResolveWeak(thin_module.shared.data.0, llmod) {
return Err(write::llvm_err(&dcx, LlvmError::PrepareThinLtoModule));
}
save_temp_bitcode(cgcx, &module, "thin-lto-after-resolve");
}
{
let _timer = cgcx
.prof
.generic_activity_with_arg("LLVM_thin_lto_internalize", thin_module.name());
if !llvm::LLVMRustPrepareThinLTOInternalize(thin_module.shared.data.0, llmod) {
return Err(write::llvm_err(&dcx, LlvmError::PrepareThinLtoModule));
}
save_temp_bitcode(cgcx, &module, "thin-lto-after-internalize");
}
{
let _timer =
cgcx.prof.generic_activity_with_arg("LLVM_thin_lto_import", thin_module.name());
if !llvm::LLVMRustPrepareThinLTOImport(thin_module.shared.data.0, llmod, target) {
return Err(write::llvm_err(&dcx, LlvmError::PrepareThinLtoModule));
}
save_temp_bitcode(cgcx, &module, "thin-lto-after-import");
}
// Alright now that we've done everything related to the ThinLTO
// analysis it's time to run some optimizations! Here we use the same
// `run_pass_manager` as the "fat" LTO above except that we tell it to
// populate a thin-specific pass manager, which presumably LLVM treats a
// little differently.
{
info!("running thin lto passes over {}", module.name);
run_pass_manager(cgcx, &dcx, &mut module, true)?;
save_temp_bitcode(cgcx, &module, "thin-lto-after-pm");
}
}*/
// FIXME: switch to #[expect] when the clippy bug is fixed.
#[allow(clippy::let_and_return)]
module
}
pub struct ThinBuffer {
context: Arc<SyncContext>,
}
impl ThinBuffer {
pub(crate) fn new(context: &Arc<SyncContext>) -> Self {
Self { context: Arc::clone(context) }
}
}
impl ThinBufferMethods for ThinBuffer {
fn data(&self) -> &[u8] {
&[]
}
}
pub struct ThinData; //(Arc<TempDir>);
fn module_name_to_str(c_str: &CStr) -> &str {
c_str.to_str().unwrap_or_else(|e| {
bug!("Encountered non-utf8 GCC module name `{}`: {}", c_str.to_string_lossy(), e)
})
}

46
compiler/rustc_codegen_gcc/src/lib.rs
View file

@ -76,7 +76,6 @@ use std::path::{Path, PathBuf};
use std::sync::atomic::{AtomicBool, Ordering}; use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::{Arc, Mutex}; use std::sync::{Arc, Mutex};
use back::lto::{ThinBuffer, ThinData};
use gccjit::{CType, Context, OptimizationLevel}; use gccjit::{CType, Context, OptimizationLevel};
#[cfg(feature = "master")] #[cfg(feature = "master")]
use gccjit::{TargetInfo, Version}; use gccjit::{TargetInfo, Version};
@ -87,7 +86,9 @@ use rustc_codegen_ssa::back::write::{
}; };
use rustc_codegen_ssa::base::codegen_crate; use rustc_codegen_ssa::base::codegen_crate;
use rustc_codegen_ssa::target_features::cfg_target_feature; use rustc_codegen_ssa::target_features::cfg_target_feature;
use rustc_codegen_ssa::traits::{CodegenBackend, ExtraBackendMethods, WriteBackendMethods}; use rustc_codegen_ssa::traits::{
CodegenBackend, ExtraBackendMethods, ThinBufferMethods, WriteBackendMethods,
};
use rustc_codegen_ssa::{CodegenResults, CompiledModule, ModuleCodegen, TargetConfig}; use rustc_codegen_ssa::{CodegenResults, CompiledModule, ModuleCodegen, TargetConfig};
use rustc_data_structures::fx::FxIndexMap; use rustc_data_structures::fx::FxIndexMap;
use rustc_data_structures::profiling::SelfProfilerRef; use rustc_data_structures::profiling::SelfProfilerRef;
@ -177,8 +178,6 @@ pub struct GccCodegenBackend {
lto_supported: Arc<AtomicBool>, lto_supported: Arc<AtomicBool>,
} }
static LTO_SUPPORTED: AtomicBool = AtomicBool::new(false);
fn load_libgccjit_if_needed(libgccjit_target_lib_file: &Path) { fn load_libgccjit_if_needed(libgccjit_target_lib_file: &Path) {
if gccjit::is_loaded() { if gccjit::is_loaded() {
// Do not load a libgccjit second time. // Do not load a libgccjit second time.
@ -251,7 +250,6 @@ impl CodegenBackend for GccCodegenBackend {
#[cfg(feature = "master")] #[cfg(feature = "master")]
{ {
let lto_supported = gccjit::is_lto_supported(); let lto_supported = gccjit::is_lto_supported();
LTO_SUPPORTED.store(lto_supported, Ordering::SeqCst);
self.lto_supported.store(lto_supported, Ordering::SeqCst); self.lto_supported.store(lto_supported, Ordering::SeqCst);
gccjit::set_global_personality_function_name(b"rust_eh_personality\0"); gccjit::set_global_personality_function_name(b"rust_eh_personality\0");
@ -281,6 +279,10 @@ impl CodegenBackend for GccCodegenBackend {
} }
} }
fn thin_lto_supported(&self) -> bool {
false
}
fn provide(&self, providers: &mut Providers) { fn provide(&self, providers: &mut Providers) {
providers.queries.global_backend_features = providers.queries.global_backend_features =
|tcx, ()| gcc_util::global_gcc_features(tcx.sess) |tcx, ()| gcc_util::global_gcc_features(tcx.sess)
@ -421,11 +423,19 @@ unsafe impl Send for SyncContext {}
// FIXME(antoyo): that shouldn't be Sync. Parallel compilation is currently disabled with "CodegenBackend::supports_parallel()". // FIXME(antoyo): that shouldn't be Sync. Parallel compilation is currently disabled with "CodegenBackend::supports_parallel()".
unsafe impl Sync for SyncContext {} unsafe impl Sync for SyncContext {}
pub struct ThinBuffer;
impl ThinBufferMethods for ThinBuffer {
fn data(&self) -> &[u8] {
&[]
}
}
impl WriteBackendMethods for GccCodegenBackend { impl WriteBackendMethods for GccCodegenBackend {
type Module = GccContext; type Module = GccContext;
type TargetMachine = (); type TargetMachine = ();
type ModuleBuffer = ModuleBuffer; type ModuleBuffer = ModuleBuffer;
type ThinData = ThinData; type ThinData = ();
type ThinBuffer = ThinBuffer; type ThinBuffer = ThinBuffer;
fn run_and_optimize_fat_lto( fn run_and_optimize_fat_lto(
@ -442,16 +452,16 @@ impl WriteBackendMethods for GccCodegenBackend {
} }
fn run_thin_lto( fn run_thin_lto(
cgcx: &CodegenContext, _cgcx: &CodegenContext,
prof: &SelfProfilerRef, _prof: &SelfProfilerRef,
dcx: DiagCtxtHandle<'_>, _dcx: DiagCtxtHandle<'_>,
// FIXME(bjorn3): Limit LTO exports to these symbols // FIXME(bjorn3): Limit LTO exports to these symbols
_exported_symbols_for_lto: &[String], _exported_symbols_for_lto: &[String],
each_linked_rlib_for_lto: &[PathBuf], _each_linked_rlib_for_lto: &[PathBuf],
modules: Vec<(String, Self::ThinBuffer)>, _modules: Vec<(String, Self::ThinBuffer)>,
cached_modules: Vec<(SerializedModule<Self::ModuleBuffer>, WorkProduct)>, _cached_modules: Vec<(SerializedModule<Self::ModuleBuffer>, WorkProduct)>,
) -> (Vec<ThinModule<Self>>, Vec<WorkProduct>) { ) -> (Vec<ThinModule<Self>>, Vec<WorkProduct>) {
back::lto::run_thin(cgcx, prof, dcx, each_linked_rlib_for_lto, modules, cached_modules) unreachable!()
} }
fn print_pass_timings(&self) { fn print_pass_timings(&self) {
@ -473,13 +483,13 @@ impl WriteBackendMethods for GccCodegenBackend {
} }
fn optimize_thin( fn optimize_thin(
cgcx: &CodegenContext, _cgcx: &CodegenContext,
_prof: &SelfProfilerRef, _prof: &SelfProfilerRef,
_shared_emitter: &SharedEmitter, _shared_emitter: &SharedEmitter,
_tm_factory: TargetMachineFactoryFn<Self>, _tm_factory: TargetMachineFactoryFn<Self>,
thin: ThinModule<Self>, _thin: ThinModule<Self>,
) -> ModuleCodegen<Self::Module> { ) -> ModuleCodegen<Self::Module> {
back::lto::optimize_thin_module(thin, cgcx) unreachable!()
} }
fn codegen( fn codegen(
@ -492,8 +502,8 @@ impl WriteBackendMethods for GccCodegenBackend {
back::write::codegen(cgcx, prof, shared_emitter, module, config) back::write::codegen(cgcx, prof, shared_emitter, module, config)
} }
fn prepare_thin(module: ModuleCodegen<Self::Module>) -> (String, Self::ThinBuffer) { fn prepare_thin(_module: ModuleCodegen<Self::Module>) -> (String, Self::ThinBuffer) {
back::lto::prepare_thin(module) unreachable!()
} }
fn serialize_module(_module: ModuleCodegen<Self::Module>) -> (String, Self::ModuleBuffer) { fn serialize_module(_module: ModuleCodegen<Self::Module>) -> (String, Self::ModuleBuffer) {

5
compiler/rustc_codegen_ssa/src/traits/backend.rs
View file

@ -80,6 +80,11 @@ pub trait CodegenBackend {
vec![] vec![]
} }
/// Is ThinLTO supported by this backend?
fn thin_lto_supported(&self) -> bool {
true
}
/// Value printed by `--print=backend-has-zstd`. /// Value printed by `--print=backend-has-zstd`.
/// ///
/// Used by compiletest to determine whether tests involving zstd compression /// Used by compiletest to determine whether tests involving zstd compression

1
compiler/rustc_interface/src/interface.rs
View file

@ -463,6 +463,7 @@ pub fn run_compiler<R: Send>(config: Config, f: impl FnOnce(&Compiler) -> R + Se
codegen_backend.init(&sess); codegen_backend.init(&sess);
sess.replaced_intrinsics = FxHashSet::from_iter(codegen_backend.replaced_intrinsics()); sess.replaced_intrinsics = FxHashSet::from_iter(codegen_backend.replaced_intrinsics());
sess.thin_lto_supported = codegen_backend.thin_lto_supported();
let cfg = parse_cfg(sess.dcx(), config.crate_cfg); let cfg = parse_cfg(sess.dcx(), config.crate_cfg);
let mut cfg = config::build_configuration(&sess, cfg); let mut cfg = config::build_configuration(&sess, cfg);

1
compiler/rustc_macros/src/diagnostics/message.rs
View file

@ -101,6 +101,7 @@ const ALLOWED_CAPITALIZED_WORDS: &[&str] = &[
"NaNs", "NaNs",
"OK", "OK",
"Rust", "Rust",
"ThinLTO",
"Unicode", "Unicode",
"VS", "VS",
// tidy-alphabetical-end // tidy-alphabetical-end

4
compiler/rustc_session/src/errors.rs
View file

@ -537,3 +537,7 @@ pub(crate) struct UnexpectedBuiltinCfg {
pub(crate) cfg_name: Symbol, pub(crate) cfg_name: Symbol,
pub(crate) controlled_by: &'static str, pub(crate) controlled_by: &'static str,
} }
#[derive(Diagnostic)]
#[diag("ThinLTO is not supported by the codegen backend")]
pub(crate) struct ThinLtoNotSupportedByBackend;

13
compiler/rustc_session/src/session.rs
View file

@ -158,6 +158,9 @@ pub struct Session {
/// The names of intrinsics that the current codegen backend replaces /// The names of intrinsics that the current codegen backend replaces
/// with its own implementations. /// with its own implementations.
pub replaced_intrinsics: FxHashSet<Symbol>, pub replaced_intrinsics: FxHashSet<Symbol>,
/// Does the codegen backend support ThinLTO?
pub thin_lto_supported: bool,
} }
#[derive(Clone, Copy)] #[derive(Clone, Copy)]
@ -606,10 +609,19 @@ impl Session {
} }
config::LtoCli::Thin => { config::LtoCli::Thin => {
// The user explicitly asked for ThinLTO // The user explicitly asked for ThinLTO
if !self.thin_lto_supported {
// Backend doesn't support ThinLTO, disable LTO.
self.dcx().emit_warn(errors::ThinLtoNotSupportedByBackend);
return config::Lto::No;
}
return config::Lto::Thin; return config::Lto::Thin;
} }
} }
if !self.thin_lto_supported {
return config::Lto::No;
}
// Ok at this point the target doesn't require anything and the user // Ok at this point the target doesn't require anything and the user
// hasn't asked for anything. Our next decision is whether or not // hasn't asked for anything. Our next decision is whether or not
// we enable "auto" ThinLTO where we use multiple codegen units and // we enable "auto" ThinLTO where we use multiple codegen units and
@ -1088,6 +1100,7 @@ pub fn build_session(
host_filesearch, host_filesearch,
invocation_temp, invocation_temp,
replaced_intrinsics: FxHashSet::default(), // filled by `run_compiler` replaced_intrinsics: FxHashSet::default(), // filled by `run_compiler`
thin_lto_supported: true, // filled by `run_compiler`
}; };
validate_commandline_args_with_session_available(&sess); validate_commandline_args_with_session_available(&sess);