From 81042523c6e1ae50682f46ae3da4ede22bfd596e Mon Sep 17 00:00:00 2001 From: Matthias Krüger Date: Fri, 5 Sep 2025 15:38:01 +0200 Subject: remove couple of clones --- compiler/rustc_codegen_ssa/src/lib.rs | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'compiler/rustc_codegen_ssa/src') diff --git a/compiler/rustc_codegen_ssa/src/lib.rs b/compiler/rustc_codegen_ssa/src/lib.rs index 5b90ffaa056..23146661f27 100644 --- a/compiler/rustc_codegen_ssa/src/lib.rs +++ b/compiler/rustc_codegen_ssa/src/lib.rs @@ -119,7 +119,7 @@ impl ModuleCodegen { }); CompiledModule { - name: self.name.clone(), + name: self.name, object, dwarf_object, bytecode, -- cgit 1.4.1-3-g733a5 From f279ae1b05fe0f12259b3e24d35f5598b18f5ea9 Mon Sep 17 00:00:00 2001 From: Yotam Ofek Date: Fri, 5 Sep 2025 18:43:43 +0000 Subject: Use `Itertools::all_equal_value()` where applicable --- compiler/rustc_codegen_ssa/src/mir/rvalue.rs | 27 ++++++++++------------ compiler/rustc_mir_transform/src/gvn.rs | 17 +++++++------- .../src/lint_tail_expr_drop_order.rs | 5 ++-- compiler/rustc_mir_transform/src/simplify.rs | 18 +++++---------- compiler/rustc_resolve/src/diagnostics.rs | 14 ++++------- 5 files changed, 35 insertions(+), 46 deletions(-) (limited to 'compiler/rustc_codegen_ssa/src') diff --git a/compiler/rustc_codegen_ssa/src/mir/rvalue.rs b/compiler/rustc_codegen_ssa/src/mir/rvalue.rs index 8a67b8d6e5f..f6f2e3f2a3a 100644 --- a/compiler/rustc_codegen_ssa/src/mir/rvalue.rs +++ b/compiler/rustc_codegen_ssa/src/mir/rvalue.rs @@ -1,3 +1,4 @@ +use itertools::Itertools as _; use rustc_abi::{self as abi, FIRST_VARIANT}; use rustc_middle::ty::adjustment::PointerCoercion; use rustc_middle::ty::layout::{HasTyCtxt, HasTypingEnv, LayoutOf, TyAndLayout}; @@ -111,14 +112,13 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> { let size = bx.const_usize(dest.layout.size.bytes()); // Use llvm.memset.p0i8.* to initialize all same byte arrays - if let Some(int) = bx.cx().const_to_opt_u128(v, false) { - let bytes = &int.to_le_bytes()[..cg_elem.layout.size.bytes_usize()]; - let first = bytes[0]; - if bytes[1..].iter().all(|&b| b == first) { - let fill = bx.cx().const_u8(first); - bx.memset(start, fill, size, dest.val.align, MemFlags::empty()); - return true; - } + if let Some(int) = bx.cx().const_to_opt_u128(v, false) + && let bytes = &int.to_le_bytes()[..cg_elem.layout.size.bytes_usize()] + && let Ok(&byte) = bytes.iter().all_equal_value() + { + let fill = bx.cx().const_u8(byte); + bx.memset(start, fill, size, dest.val.align, MemFlags::empty()); + return true; } // Use llvm.memset.p0i8.* to initialize byte arrays @@ -130,13 +130,10 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> { false }; - match cg_elem.val { - OperandValue::Immediate(v) => { - if try_init_all_same(bx, v) { - return; - } - } - _ => (), + if let OperandValue::Immediate(v) = cg_elem.val + && try_init_all_same(bx, v) + { + return; } let count = self diff --git a/compiler/rustc_mir_transform/src/gvn.rs b/compiler/rustc_mir_transform/src/gvn.rs index 5a13394543b..9da6e034c3b 100644 --- a/compiler/rustc_mir_transform/src/gvn.rs +++ b/compiler/rustc_mir_transform/src/gvn.rs @@ -87,6 +87,7 @@ use std::borrow::Cow; use either::Either; +use itertools::Itertools as _; use rustc_abi::{self as abi, BackendRepr, FIRST_VARIANT, FieldIdx, Primitive, Size, VariantIdx}; use rustc_const_eval::const_eval::DummyMachine; use rustc_const_eval::interpret::{ @@ -1023,15 +1024,15 @@ impl<'body, 'tcx> VnState<'body, 'tcx> { } }; - if ty.is_array() && fields.len() > 4 { - let first = fields[0]; - if fields.iter().all(|&v| v == first) { - let len = ty::Const::from_target_usize(self.tcx, fields.len().try_into().unwrap()); - if let Some(op) = self.try_as_operand(first, location) { - *rvalue = Rvalue::Repeat(op, len); - } - return Some(self.insert(ty, Value::Repeat(first, len))); + if ty.is_array() + && fields.len() > 4 + && let Ok(&first) = fields.iter().all_equal_value() + { + let len = ty::Const::from_target_usize(self.tcx, fields.len().try_into().unwrap()); + if let Some(op) = self.try_as_operand(first, location) { + *rvalue = Rvalue::Repeat(op, len); } + return Some(self.insert(ty, Value::Repeat(first, len))); } if let Some(value) = diff --git a/compiler/rustc_mir_transform/src/lint_tail_expr_drop_order.rs b/compiler/rustc_mir_transform/src/lint_tail_expr_drop_order.rs index 794984d2f3e..61c9bbe3123 100644 --- a/compiler/rustc_mir_transform/src/lint_tail_expr_drop_order.rs +++ b/compiler/rustc_mir_transform/src/lint_tail_expr_drop_order.rs @@ -2,6 +2,7 @@ use std::cell::RefCell; use std::collections::hash_map; use std::rc::Rc; +use itertools::Itertools as _; use rustc_data_structures::fx::{FxHashMap, FxHashSet, FxIndexMap}; use rustc_data_structures::unord::{UnordMap, UnordSet}; use rustc_errors::Subdiagnostic; @@ -339,9 +340,9 @@ pub(crate) fn run_lint<'tcx>(tcx: TyCtxt<'tcx>, def_id: LocalDefId, body: &Body< // Suppose that all BIDs point into the same local, // we can remove the this local from the observed drops, // so that we can focus our diagnosis more on the others. - if candidates.iter().all(|&(_, place)| candidates[0].1.local == place.local) { + if let Ok(local) = candidates.iter().map(|&(_, place)| place.local).all_equal_value() { for path_idx in all_locals_dropped.iter() { - if move_data.move_paths[path_idx].place.local == candidates[0].1.local { + if move_data.move_paths[path_idx].place.local == local { to_exclude.insert(path_idx); } } diff --git a/compiler/rustc_mir_transform/src/simplify.rs b/compiler/rustc_mir_transform/src/simplify.rs index 468ef742dfb..75917d23883 100644 --- a/compiler/rustc_mir_transform/src/simplify.rs +++ b/compiler/rustc_mir_transform/src/simplify.rs @@ -34,6 +34,7 @@ //! The normal logic that a program with UB can be changed to do anything does not apply to //! pre-"runtime" MIR! +use itertools::Itertools as _; use rustc_index::{Idx, IndexSlice, IndexVec}; use rustc_middle::mir::visit::{MutVisitor, MutatingUseContext, PlaceContext, Visitor}; use rustc_middle::mir::*; @@ -288,20 +289,13 @@ impl<'a, 'tcx> CfgSimplifier<'a, 'tcx> { return false; }; - let first_succ = { - if let Some(first_succ) = terminator.successors().next() { - if terminator.successors().all(|s| s == first_succ) { - let count = terminator.successors().count(); - self.pred_count[first_succ] -= (count - 1) as u32; - first_succ - } else { - return false; - } - } else { - return false; - } + let Ok(first_succ) = terminator.successors().all_equal_value() else { + return false; }; + let count = terminator.successors().count(); + self.pred_count[first_succ] -= (count - 1) as u32; + debug!("simplifying branch {:?}", terminator); terminator.kind = TerminatorKind::Goto { target: first_succ }; true diff --git a/compiler/rustc_resolve/src/diagnostics.rs b/compiler/rustc_resolve/src/diagnostics.rs index f458a1f17b9..689e713ef46 100644 --- a/compiler/rustc_resolve/src/diagnostics.rs +++ b/compiler/rustc_resolve/src/diagnostics.rs @@ -1,3 +1,4 @@ +use itertools::Itertools as _; use rustc_ast::visit::{self, Visitor}; use rustc_ast::{ self as ast, CRATE_NODE_ID, Crate, ItemKind, ModKind, NodeId, Path, join_path_idents, @@ -3469,16 +3470,11 @@ fn show_candidates( err.note(note.to_string()); } } else { - let (_, descr_first, _, _, _) = &inaccessible_path_strings[0]; - let descr = if inaccessible_path_strings + let descr = inaccessible_path_strings .iter() - .skip(1) - .all(|(_, descr, _, _, _)| descr == descr_first) - { - descr_first - } else { - "item" - }; + .map(|&(_, descr, _, _, _)| descr) + .all_equal_value() + .unwrap_or("item"); let plural_descr = if descr.ends_with('s') { format!("{descr}es") } else { format!("{descr}s") }; -- cgit 1.4.1-3-g733a5 From 027135976849da34250d7638502bcc2a24f741d7 Mon Sep 17 00:00:00 2001 From: bjorn3 <17426603+bjorn3@users.noreply.github.com> Date: Fri, 5 Sep 2025 09:32:56 +0000 Subject: Make the allocator shim participate in LTO again --- compiler/rustc_codegen_ssa/src/back/lto.rs | 14 ---------- compiler/rustc_codegen_ssa/src/back/write.rs | 40 +++++++++++++++++++++------- compiler/rustc_codegen_ssa/src/base.rs | 11 ++++++-- 3 files changed, 39 insertions(+), 26 deletions(-) (limited to 'compiler/rustc_codegen_ssa/src') diff --git a/compiler/rustc_codegen_ssa/src/back/lto.rs b/compiler/rustc_codegen_ssa/src/back/lto.rs index f4a9037940a..e6df6a2469f 100644 --- a/compiler/rustc_codegen_ssa/src/back/lto.rs +++ b/compiler/rustc_codegen_ssa/src/back/lto.rs @@ -1,7 +1,6 @@ use std::ffi::CString; use std::sync::Arc; -use rustc_ast::expand::allocator::AllocatorKind; use rustc_data_structures::memmap::Mmap; use rustc_hir::def_id::{CrateNum, LOCAL_CRATE}; use rustc_middle::middle::exported_symbols::{ExportedSymbol, SymbolExportInfo, SymbolExportLevel}; @@ -96,19 +95,6 @@ pub(super) fn exported_symbols_for_lto( .filter_map(|&(s, info): &(ExportedSymbol<'_>, SymbolExportInfo)| { if info.level.is_below_threshold(export_threshold) || info.used { Some(symbol_name_for_instance_in_crate(tcx, s, cnum)) - } else if export_threshold == SymbolExportLevel::C - && info.rustc_std_internal_symbol - && let Some(AllocatorKind::Default) = allocator_kind_for_codegen(tcx) - { - // Export the __rdl_* exports for usage by the allocator shim when not using - // #[global_allocator]. Most of the conditions above are only used to avoid - // unnecessary expensive symbol_name_for_instance_in_crate calls. - let sym = symbol_name_for_instance_in_crate(tcx, s, cnum); - if sym.contains("__rdl_") || sym.contains("__rg_oom") { - Some(sym) - } else { - None - } } else { None } diff --git a/compiler/rustc_codegen_ssa/src/back/write.rs b/compiler/rustc_codegen_ssa/src/back/write.rs index f637e7f58db..afda7226fdd 100644 --- a/compiler/rustc_codegen_ssa/src/back/write.rs +++ b/compiler/rustc_codegen_ssa/src/back/write.rs @@ -334,6 +334,7 @@ pub struct CodegenContext { pub output_filenames: Arc, pub invocation_temp: Option, pub module_config: Arc, + pub allocator_config: Arc, pub tm_factory: TargetMachineFactoryFn, pub msvc_imps_needed: bool, pub is_pe_coff: bool, @@ -794,12 +795,19 @@ pub(crate) fn compute_per_cgu_lto_type( sess_lto: &Lto, opts: &config::Options, sess_crate_types: &[CrateType], + module_kind: ModuleKind, ) -> ComputedLtoType { // If the linker does LTO, we don't have to do it. Note that we // keep doing full LTO, if it is requested, as not to break the // assumption that the output will be a single module. let linker_does_lto = opts.cg.linker_plugin_lto.enabled(); + // When we're automatically doing ThinLTO for multi-codegen-unit + // builds we don't actually want to LTO the allocator module if + // it shows up. This is due to various linker shenanigans that + // we'll encounter later. + let is_allocator = module_kind == ModuleKind::Allocator; + // We ignore a request for full crate graph LTO if the crate type // is only an rlib, as there is no full crate graph to process, // that'll happen later. @@ -811,7 +819,7 @@ pub(crate) fn compute_per_cgu_lto_type( let is_rlib = matches!(sess_crate_types, [CrateType::Rlib]); match sess_lto { - Lto::ThinLocal if !linker_does_lto => ComputedLtoType::Thin, + Lto::ThinLocal if !linker_does_lto && !is_allocator => ComputedLtoType::Thin, Lto::Thin if !linker_does_lto && !is_rlib => ComputedLtoType::Thin, Lto::Fat if !is_rlib => ComputedLtoType::Fat, _ => ComputedLtoType::No, @@ -825,18 +833,23 @@ fn execute_optimize_work_item( let dcx = cgcx.create_dcx(); let dcx = dcx.handle(); - B::optimize(cgcx, dcx, &mut module, &cgcx.module_config); + let module_config = match module.kind { + ModuleKind::Regular => &cgcx.module_config, + ModuleKind::Allocator => &cgcx.allocator_config, + }; + + B::optimize(cgcx, dcx, &mut module, module_config); // After we've done the initial round of optimizations we need to // decide whether to synchronously codegen this module or ship it // back to the coordinator thread for further LTO processing (which // has to wait for all the initial modules to be optimized). - let lto_type = compute_per_cgu_lto_type(&cgcx.lto, &cgcx.opts, &cgcx.crate_types); + let lto_type = compute_per_cgu_lto_type(&cgcx.lto, &cgcx.opts, &cgcx.crate_types, module.kind); // If we're doing some form of incremental LTO then we need to be sure to // save our module to disk first. - let bitcode = if cgcx.module_config.emit_pre_lto_bc { + let bitcode = if module_config.emit_pre_lto_bc { let filename = pre_lto_bitcode_filename(&module.name); cgcx.incr_comp_session_dir.as_ref().map(|path| path.join(&filename)) } else { @@ -845,7 +858,7 @@ fn execute_optimize_work_item( match lto_type { ComputedLtoType::No => { - let module = B::codegen(cgcx, module, &cgcx.module_config); + let module = B::codegen(cgcx, module, module_config); WorkItemResult::Finished(module) } ComputedLtoType::Thin => { @@ -1133,6 +1146,7 @@ fn start_executing_work( diag_emitter: shared_emitter.clone(), output_filenames: Arc::clone(tcx.output_filenames(())), module_config: regular_config, + allocator_config, tm_factory: backend.target_machine_factory(tcx.sess, ol, backend_features), msvc_imps_needed: msvc_imps_needed(tcx), is_pe_coff: tcx.sess.target.is_like_windows, @@ -1147,11 +1161,6 @@ fn start_executing_work( invocation_temp: sess.invocation_temp.clone(), }; - let compiled_allocator_module = allocator_module.map(|mut allocator_module| { - B::optimize(&cgcx, tcx.sess.dcx(), &mut allocator_module, &allocator_config); - B::codegen(&cgcx, allocator_module, &allocator_config) - }); - // This is the "main loop" of parallel work happening for parallel codegen. // It's here that we manage parallelism, schedule work, and work with // messages coming from clients. @@ -1331,6 +1340,17 @@ fn start_executing_work( let mut llvm_start_time: Option> = None; + let compiled_allocator_module = allocator_module.and_then(|allocator_module| { + match execute_optimize_work_item(&cgcx, allocator_module) { + WorkItemResult::Finished(compiled_module) => return Some(compiled_module), + WorkItemResult::NeedsFatLto(fat_lto_input) => needs_fat_lto.push(fat_lto_input), + WorkItemResult::NeedsThinLto(name, thin_buffer) => { + needs_thin_lto.push((name, thin_buffer)) + } + } + None + }); + // Run the message loop while there's still anything that needs message // processing. Note that as soon as codegen is aborted we simply want to // wait for all existing work to finish, so many of the conditions here diff --git a/compiler/rustc_codegen_ssa/src/base.rs b/compiler/rustc_codegen_ssa/src/base.rs index a9a2ae1b3db..45b028aa8ef 100644 --- a/compiler/rustc_codegen_ssa/src/base.rs +++ b/compiler/rustc_codegen_ssa/src/base.rs @@ -46,7 +46,9 @@ use crate::meth::load_vtable; use crate::mir::operand::OperandValue; use crate::mir::place::PlaceRef; use crate::traits::*; -use crate::{CachedModuleCodegen, CodegenLintLevels, CrateInfo, ModuleCodegen, errors, meth, mir}; +use crate::{ + CachedModuleCodegen, CodegenLintLevels, CrateInfo, ModuleCodegen, ModuleKind, errors, meth, mir, +}; pub(crate) fn bin_op_to_icmp_predicate(op: BinOp, signed: bool) -> IntPredicate { match (op, signed) { @@ -1124,7 +1126,12 @@ pub fn determine_cgu_reuse<'tcx>(tcx: TyCtxt<'tcx>, cgu: &CodegenUnit<'tcx>) -> // We can re-use either the pre- or the post-thinlto state. If no LTO is // being performed then we can use post-LTO artifacts, otherwise we must // reuse pre-LTO artifacts - match compute_per_cgu_lto_type(&tcx.sess.lto(), &tcx.sess.opts, tcx.crate_types()) { + match compute_per_cgu_lto_type( + &tcx.sess.lto(), + &tcx.sess.opts, + tcx.crate_types(), + ModuleKind::Regular, + ) { ComputedLtoType::No => CguReuse::PostLto, _ => CguReuse::PreLto, } -- cgit 1.4.1-3-g733a5 From 2cf94b92ca852924ad90943a0c469f01742216a6 Mon Sep 17 00:00:00 2001 From: bjorn3 <17426603+bjorn3@users.noreply.github.com> Date: Fri, 5 Sep 2025 19:09:39 +0000 Subject: Ensure fat LTO doesn't merge everything into the allocator module --- compiler/rustc_codegen_cranelift/src/driver/aot.rs | 11 ++++++++++- compiler/rustc_codegen_llvm/src/back/lto.rs | 8 +++++++- compiler/rustc_codegen_ssa/src/back/write.rs | 3 ++- compiler/rustc_codegen_ssa/src/lib.rs | 2 ++ 4 files changed, 21 insertions(+), 3 deletions(-) (limited to 'compiler/rustc_codegen_ssa/src') diff --git a/compiler/rustc_codegen_cranelift/src/driver/aot.rs b/compiler/rustc_codegen_cranelift/src/driver/aot.rs index c3adb5e767e..7e77781dc2f 100644 --- a/compiler/rustc_codegen_cranelift/src/driver/aot.rs +++ b/compiler/rustc_codegen_cranelift/src/driver/aot.rs @@ -12,7 +12,9 @@ use cranelift_object::{ObjectBuilder, ObjectModule}; use rustc_codegen_ssa::assert_module_sources::CguReuse; use rustc_codegen_ssa::back::link::ensure_removed; use rustc_codegen_ssa::base::determine_cgu_reuse; -use rustc_codegen_ssa::{CodegenResults, CompiledModule, CrateInfo, errors as ssa_errors}; +use rustc_codegen_ssa::{ + CodegenResults, CompiledModule, CrateInfo, ModuleKind, errors as ssa_errors, +}; use rustc_data_structures::profiling::SelfProfilerRef; use rustc_data_structures::stable_hasher::{HashStable, StableHasher}; use rustc_data_structures::sync::{IntoDynSyncSend, par_map}; @@ -361,6 +363,7 @@ fn emit_cgu( invocation_temp, prof, product.object, + ModuleKind::Regular, name.clone(), producer, )?; @@ -369,6 +372,7 @@ fn emit_cgu( module_regular, module_global_asm: global_asm_object_file.map(|global_asm_object_file| CompiledModule { name: format!("{name}.asm"), + kind: ModuleKind::Regular, object: Some(global_asm_object_file), dwarf_object: None, bytecode: None, @@ -385,6 +389,7 @@ fn emit_module( invocation_temp: Option<&str>, prof: &SelfProfilerRef, mut object: cranelift_object::object::write::Object<'_>, + kind: ModuleKind, name: String, producer_str: &str, ) -> Result { @@ -425,6 +430,7 @@ fn emit_module( Ok(CompiledModule { name, + kind, object: Some(tmp_file), dwarf_object: None, bytecode: None, @@ -479,6 +485,7 @@ fn reuse_workproduct_for_cgu( Ok(ModuleCodegenResult { module_regular: CompiledModule { name: cgu.name().to_string(), + kind: ModuleKind::Regular, object: Some(obj_out_regular), dwarf_object: None, bytecode: None, @@ -488,6 +495,7 @@ fn reuse_workproduct_for_cgu( }, module_global_asm: source_file_global_asm.map(|source_file| CompiledModule { name: cgu.name().to_string(), + kind: ModuleKind::Regular, object: Some(obj_out_global_asm), dwarf_object: None, bytecode: None, @@ -643,6 +651,7 @@ fn emit_allocator_module(tcx: TyCtxt<'_>) -> Option { tcx.sess.invocation_temp.as_deref(), &tcx.sess.prof, product.object, + ModuleKind::Allocator, "allocator_shim".to_owned(), &crate::debuginfo::producer(tcx.sess), ) { diff --git a/compiler/rustc_codegen_llvm/src/back/lto.rs b/compiler/rustc_codegen_llvm/src/back/lto.rs index 326b876e7e6..ad2e722cfef 100644 --- a/compiler/rustc_codegen_llvm/src/back/lto.rs +++ b/compiler/rustc_codegen_llvm/src/back/lto.rs @@ -11,7 +11,7 @@ use object::{Object, ObjectSection}; use rustc_codegen_ssa::back::lto::{SerializedModule, ThinModule, ThinShared}; use rustc_codegen_ssa::back::write::{CodegenContext, FatLtoInput}; use rustc_codegen_ssa::traits::*; -use rustc_codegen_ssa::{ModuleCodegen, looks_like_rust_object_file}; +use rustc_codegen_ssa::{ModuleCodegen, ModuleKind, looks_like_rust_object_file}; use rustc_data_structures::fx::FxHashMap; use rustc_data_structures::memmap::Mmap; use rustc_errors::DiagCtxtHandle; @@ -225,9 +225,15 @@ fn fat_lto( // All the other modules will be serialized and reparsed into the new // context, so this hopefully avoids serializing and parsing the largest // codegen unit. + // + // Additionally use a regular module as the base here to ensure that various + // file copy operations in the backend work correctly. The only other kind + // of module here should be an allocator one, and if your crate is smaller + // than the allocator module then the size doesn't really matter anyway. let costliest_module = in_memory .iter() .enumerate() + .filter(|&(_, module)| module.kind == ModuleKind::Regular) .map(|(i, module)| { let cost = unsafe { llvm::LLVMRustModuleCost(module.module_llvm.llmod()) }; (cost, i) diff --git a/compiler/rustc_codegen_ssa/src/back/write.rs b/compiler/rustc_codegen_ssa/src/back/write.rs index afda7226fdd..95e02a7c6db 100644 --- a/compiler/rustc_codegen_ssa/src/back/write.rs +++ b/compiler/rustc_codegen_ssa/src/back/write.rs @@ -490,7 +490,7 @@ fn copy_all_cgu_workproducts_to_incr_comp_cache_dir( let _timer = sess.timer("copy_all_cgu_workproducts_to_incr_comp_cache_dir"); - for module in &compiled_modules.modules { + for module in compiled_modules.modules.iter().filter(|m| m.kind == ModuleKind::Regular) { let mut files = Vec::new(); if let Some(object_file_path) = &module.object { files.push((OutputType::Object.extension(), object_file_path.as_path())); @@ -960,6 +960,7 @@ fn execute_copy_from_cache_work_item( WorkItemResult::Finished(CompiledModule { links_from_incr_cache, + kind: ModuleKind::Regular, name: module.name, object, dwarf_object, diff --git a/compiler/rustc_codegen_ssa/src/lib.rs b/compiler/rustc_codegen_ssa/src/lib.rs index 23146661f27..baba8f9ca3e 100644 --- a/compiler/rustc_codegen_ssa/src/lib.rs +++ b/compiler/rustc_codegen_ssa/src/lib.rs @@ -120,6 +120,7 @@ impl ModuleCodegen { CompiledModule { name: self.name, + kind: self.kind, object, dwarf_object, bytecode, @@ -133,6 +134,7 @@ impl ModuleCodegen { #[derive(Debug, Encodable, Decodable)] pub struct CompiledModule { pub name: String, + pub kind: ModuleKind, pub object: Option, pub dwarf_object: Option, pub bytecode: Option, -- cgit 1.4.1-3-g733a5 From e3d0b7d6480d7e4dbbbea24635f56e28dfe735b3 Mon Sep 17 00:00:00 2001 From: bjorn3 <17426603+bjorn3@users.noreply.github.com> Date: Fri, 15 Aug 2025 13:12:19 +0000 Subject: Remove thin_link_data method from ThinBufferMethods It is only used within cg_llvm. --- compiler/rustc_codegen_gcc/src/back/lto.rs | 4 ---- compiler/rustc_codegen_llvm/src/back/lto.rs | 16 ++++++++-------- compiler/rustc_codegen_ssa/src/traits/write.rs | 1 - 3 files changed, 8 insertions(+), 13 deletions(-) (limited to 'compiler/rustc_codegen_ssa/src') diff --git a/compiler/rustc_codegen_gcc/src/back/lto.rs b/compiler/rustc_codegen_gcc/src/back/lto.rs index 9d8ce2383f2..d475d2c7da9 100644 --- a/compiler/rustc_codegen_gcc/src/back/lto.rs +++ b/compiler/rustc_codegen_gcc/src/back/lto.rs @@ -650,10 +650,6 @@ impl ThinBufferMethods for ThinBuffer { fn data(&self) -> &[u8] { &[] } - - fn thin_link_data(&self) -> &[u8] { - unimplemented!(); - } } pub struct ThinData; //(Arc); diff --git a/compiler/rustc_codegen_llvm/src/back/lto.rs b/compiler/rustc_codegen_llvm/src/back/lto.rs index ad2e722cfef..319f80beb06 100644 --- a/compiler/rustc_codegen_llvm/src/back/lto.rs +++ b/compiler/rustc_codegen_llvm/src/back/lto.rs @@ -698,21 +698,21 @@ impl ThinBuffer { let mut ptr = NonNull::new(ptr).unwrap(); ThinBuffer(unsafe { ptr.as_mut() }) } -} -impl ThinBufferMethods for ThinBuffer { - fn data(&self) -> &[u8] { + pub(crate) fn thin_link_data(&self) -> &[u8] { unsafe { - let ptr = llvm::LLVMRustThinLTOBufferPtr(self.0) as *const _; - let len = llvm::LLVMRustThinLTOBufferLen(self.0); + let ptr = llvm::LLVMRustThinLTOBufferThinLinkDataPtr(self.0) as *const _; + let len = llvm::LLVMRustThinLTOBufferThinLinkDataLen(self.0); slice::from_raw_parts(ptr, len) } } +} - fn thin_link_data(&self) -> &[u8] { +impl ThinBufferMethods for ThinBuffer { + fn data(&self) -> &[u8] { unsafe { - let ptr = llvm::LLVMRustThinLTOBufferThinLinkDataPtr(self.0) as *const _; - let len = llvm::LLVMRustThinLTOBufferThinLinkDataLen(self.0); + let ptr = llvm::LLVMRustThinLTOBufferPtr(self.0) as *const _; + let len = llvm::LLVMRustThinLTOBufferLen(self.0); slice::from_raw_parts(ptr, len) } } diff --git a/compiler/rustc_codegen_ssa/src/traits/write.rs b/compiler/rustc_codegen_ssa/src/traits/write.rs index cc7c4e46d7b..40f546ebc92 100644 --- a/compiler/rustc_codegen_ssa/src/traits/write.rs +++ b/compiler/rustc_codegen_ssa/src/traits/write.rs @@ -59,7 +59,6 @@ pub trait WriteBackendMethods: Clone + 'static { pub trait ThinBufferMethods: Send + Sync { fn data(&self) -> &[u8]; - fn thin_link_data(&self) -> &[u8]; } pub trait ModuleBufferMethods: Send + Sync { -- cgit 1.4.1-3-g733a5 From f2933b34a8331204270fa2bdbdcfd79fcffbb302 Mon Sep 17 00:00:00 2001 From: bjorn3 <17426603+bjorn3@users.noreply.github.com> Date: Thu, 4 Sep 2025 15:16:55 +0000 Subject: Remove want_summary argument from prepare_thin It is always false nowadays. ThinLTO summary writing is instead done by llvm_optimize. --- compiler/rustc_codegen_gcc/src/back/lto.rs | 7 ++----- compiler/rustc_codegen_gcc/src/lib.rs | 7 ++----- compiler/rustc_codegen_llvm/src/back/lto.rs | 11 ++++------- compiler/rustc_codegen_llvm/src/back/write.rs | 2 +- compiler/rustc_codegen_llvm/src/lib.rs | 7 ++----- compiler/rustc_codegen_llvm/src/llvm/ffi.rs | 1 - compiler/rustc_codegen_ssa/src/back/write.rs | 2 +- compiler/rustc_codegen_ssa/src/traits/write.rs | 5 +---- compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp | 11 +++-------- 9 files changed, 16 insertions(+), 37 deletions(-) (limited to 'compiler/rustc_codegen_ssa/src') diff --git a/compiler/rustc_codegen_gcc/src/back/lto.rs b/compiler/rustc_codegen_gcc/src/back/lto.rs index d475d2c7da9..d29bba2570f 100644 --- a/compiler/rustc_codegen_gcc/src/back/lto.rs +++ b/compiler/rustc_codegen_gcc/src/back/lto.rs @@ -305,12 +305,9 @@ pub(crate) fn run_thin( ) } -pub(crate) fn prepare_thin( - module: ModuleCodegen, - _emit_summary: bool, -) -> (String, ThinBuffer) { +pub(crate) fn prepare_thin(module: ModuleCodegen) -> (String, ThinBuffer) { let name = module.name; - //let buffer = ThinBuffer::new(module.module_llvm.context, true, emit_summary); + //let buffer = ThinBuffer::new(module.module_llvm.context, true); let buffer = ThinBuffer::new(&module.module_llvm.context); (name, buffer) } diff --git a/compiler/rustc_codegen_gcc/src/lib.rs b/compiler/rustc_codegen_gcc/src/lib.rs index 2d7df79ba95..f76f933cad4 100644 --- a/compiler/rustc_codegen_gcc/src/lib.rs +++ b/compiler/rustc_codegen_gcc/src/lib.rs @@ -408,11 +408,8 @@ impl WriteBackendMethods for GccCodegenBackend { back::write::codegen(cgcx, module, config) } - fn prepare_thin( - module: ModuleCodegen, - emit_summary: bool, - ) -> (String, Self::ThinBuffer) { - back::lto::prepare_thin(module, emit_summary) + fn prepare_thin(module: ModuleCodegen) -> (String, Self::ThinBuffer) { + back::lto::prepare_thin(module) } fn serialize_module(_module: ModuleCodegen) -> (String, Self::ModuleBuffer) { diff --git a/compiler/rustc_codegen_llvm/src/back/lto.rs b/compiler/rustc_codegen_llvm/src/back/lto.rs index 319f80beb06..f571716d9dd 100644 --- a/compiler/rustc_codegen_llvm/src/back/lto.rs +++ b/compiler/rustc_codegen_llvm/src/back/lto.rs @@ -185,12 +185,9 @@ pub(crate) fn run_thin( thin_lto(cgcx, dcx, modules, upstream_modules, cached_modules, &symbols_below_threshold) } -pub(crate) fn prepare_thin( - module: ModuleCodegen, - emit_summary: bool, -) -> (String, ThinBuffer) { +pub(crate) fn prepare_thin(module: ModuleCodegen) -> (String, ThinBuffer) { let name = module.name; - let buffer = ThinBuffer::new(module.module_llvm.llmod(), true, emit_summary); + let buffer = ThinBuffer::new(module.module_llvm.llmod(), true); (name, buffer) } @@ -687,9 +684,9 @@ unsafe impl Send for ThinBuffer {} unsafe impl Sync for ThinBuffer {} impl ThinBuffer { - pub(crate) fn new(m: &llvm::Module, is_thin: bool, emit_summary: bool) -> ThinBuffer { + pub(crate) fn new(m: &llvm::Module, is_thin: bool) -> ThinBuffer { unsafe { - let buffer = llvm::LLVMRustThinLTOBufferCreate(m, is_thin, emit_summary); + let buffer = llvm::LLVMRustThinLTOBufferCreate(m, is_thin); ThinBuffer(buffer) } } diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index 7ea2ae6673b..423f0da4878 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -837,7 +837,7 @@ pub(crate) fn codegen( "LLVM_module_codegen_make_bitcode", &*module.name, ); - ThinBuffer::new(llmod, config.emit_thin_lto, false) + ThinBuffer::new(llmod, config.emit_thin_lto) }; let data = thin.data(); let _timer = cgcx diff --git a/compiler/rustc_codegen_llvm/src/lib.rs b/compiler/rustc_codegen_llvm/src/lib.rs index 628cb34fd9e..6fb23d09843 100644 --- a/compiler/rustc_codegen_llvm/src/lib.rs +++ b/compiler/rustc_codegen_llvm/src/lib.rs @@ -211,11 +211,8 @@ impl WriteBackendMethods for LlvmCodegenBackend { ) -> CompiledModule { back::write::codegen(cgcx, module, config) } - fn prepare_thin( - module: ModuleCodegen, - emit_summary: bool, - ) -> (String, Self::ThinBuffer) { - back::lto::prepare_thin(module, emit_summary) + fn prepare_thin(module: ModuleCodegen) -> (String, Self::ThinBuffer) { + back::lto::prepare_thin(module) } fn serialize_module(module: ModuleCodegen) -> (String, Self::ModuleBuffer) { (module.name, back::lto::ModuleBuffer::new(module.module_llvm.llmod())) diff --git a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs index b66fc157b3c..0679f55ab7f 100644 --- a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs +++ b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs @@ -2602,7 +2602,6 @@ unsafe extern "C" { pub(crate) fn LLVMRustThinLTOBufferCreate( M: &Module, is_thin: bool, - emit_summary: bool, ) -> &'static mut ThinLTOBuffer; pub(crate) fn LLVMRustThinLTOBufferFree(M: &'static mut ThinLTOBuffer); pub(crate) fn LLVMRustThinLTOBufferPtr(M: &ThinLTOBuffer) -> *const c_char; diff --git a/compiler/rustc_codegen_ssa/src/back/write.rs b/compiler/rustc_codegen_ssa/src/back/write.rs index 95e02a7c6db..365c2a08980 100644 --- a/compiler/rustc_codegen_ssa/src/back/write.rs +++ b/compiler/rustc_codegen_ssa/src/back/write.rs @@ -862,7 +862,7 @@ fn execute_optimize_work_item( WorkItemResult::Finished(module) } ComputedLtoType::Thin => { - let (name, thin_buffer) = B::prepare_thin(module, false); + let (name, thin_buffer) = B::prepare_thin(module); if let Some(path) = bitcode { fs::write(&path, thin_buffer.data()).unwrap_or_else(|e| { panic!("Error writing pre-lto-bitcode file `{}`: {}", path.display(), e); diff --git a/compiler/rustc_codegen_ssa/src/traits/write.rs b/compiler/rustc_codegen_ssa/src/traits/write.rs index 40f546ebc92..1ac1d7ef2e2 100644 --- a/compiler/rustc_codegen_ssa/src/traits/write.rs +++ b/compiler/rustc_codegen_ssa/src/traits/write.rs @@ -50,10 +50,7 @@ pub trait WriteBackendMethods: Clone + 'static { module: ModuleCodegen, config: &ModuleConfig, ) -> CompiledModule; - fn prepare_thin( - module: ModuleCodegen, - want_summary: bool, - ) -> (String, Self::ThinBuffer); + fn prepare_thin(module: ModuleCodegen) -> (String, Self::ThinBuffer); fn serialize_module(module: ModuleCodegen) -> (String, Self::ModuleBuffer); } diff --git a/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp b/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp index dd492325814..3bb1533c2fe 100644 --- a/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp +++ b/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp @@ -1568,12 +1568,11 @@ extern "C" bool LLVMRustPrepareThinLTOImport(const LLVMRustThinLTOData *Data, return true; } -extern "C" LLVMRustThinLTOBuffer * -LLVMRustThinLTOBufferCreate(LLVMModuleRef M, bool is_thin, bool emit_summary) { +extern "C" LLVMRustThinLTOBuffer *LLVMRustThinLTOBufferCreate(LLVMModuleRef M, + bool is_thin) { auto Ret = std::make_unique(); { auto OS = raw_string_ostream(Ret->data); - auto ThinLinkOS = raw_string_ostream(Ret->thin_link_data); { if (is_thin) { PassBuilder PB; @@ -1587,11 +1586,7 @@ LLVMRustThinLTOBufferCreate(LLVMModuleRef M, bool is_thin, bool emit_summary) { PB.registerLoopAnalyses(LAM); PB.crossRegisterProxies(LAM, FAM, CGAM, MAM); ModulePassManager MPM; - // We only pass ThinLinkOS to be filled in if we want the summary, - // because otherwise LLVM does extra work and may double-emit some - // errors or warnings. - MPM.addPass( - ThinLTOBitcodeWriterPass(OS, emit_summary ? &ThinLinkOS : nullptr)); + MPM.addPass(ThinLTOBitcodeWriterPass(OS, nullptr)); MPM.run(*unwrap(M), MAM); } else { WriteBitcodeToFile(*unwrap(M), OS); -- cgit 1.4.1-3-g733a5 From 3a1ae064a71f7e265e1e24c75b26fe3cd7edd3b1 Mon Sep 17 00:00:00 2001 From: bjorn3 <17426603+bjorn3@users.noreply.github.com> Date: Thu, 4 Sep 2025 09:27:12 +0000 Subject: Move timers into execute_*_work_item --- compiler/rustc_codegen_ssa/src/back/write.rs | 47 ++++++++++++---------------- 1 file changed, 20 insertions(+), 27 deletions(-) (limited to 'compiler/rustc_codegen_ssa/src') diff --git a/compiler/rustc_codegen_ssa/src/back/write.rs b/compiler/rustc_codegen_ssa/src/back/write.rs index 365c2a08980..cbaf67d7345 100644 --- a/compiler/rustc_codegen_ssa/src/back/write.rs +++ b/compiler/rustc_codegen_ssa/src/back/write.rs @@ -830,6 +830,8 @@ fn execute_optimize_work_item( cgcx: &CodegenContext, mut module: ModuleCodegen, ) -> WorkItemResult { + let _timer = cgcx.prof.generic_activity_with_arg("codegen_module_optimize", &*module.name); + let dcx = cgcx.create_dcx(); let dcx = dcx.handle(); @@ -890,6 +892,10 @@ fn execute_copy_from_cache_work_item( cgcx: &CodegenContext, module: CachedModuleCodegen, ) -> WorkItemResult { + let _timer = cgcx + .prof + .generic_activity_with_arg("codegen_copy_artifacts_from_incr_cache", &*module.name); + let incr_comp_session_dir = cgcx.incr_comp_session_dir.as_ref().unwrap(); let mut links_from_incr_cache = Vec::new(); @@ -977,6 +983,8 @@ fn execute_fat_lto_work_item( mut needs_fat_lto: Vec>, import_only_modules: Vec<(SerializedModule, WorkProduct)>, ) -> WorkItemResult { + let _timer = cgcx.prof.generic_activity_with_arg("codegen_module_perform_lto", "everything"); + for (module, wp) in import_only_modules { needs_fat_lto.push(FatLtoInput::Serialized { name: wp.cgu_name, buffer: module }) } @@ -995,6 +1003,8 @@ fn execute_thin_lto_work_item( cgcx: &CodegenContext, module: lto::ThinModule, ) -> WorkItemResult { + let _timer = cgcx.prof.generic_activity_with_arg("codegen_module_perform_lto", module.name()); + let module = B::optimize_thin(cgcx, module); let module = B::codegen(cgcx, module, &cgcx.module_config); WorkItemResult::Finished(module) @@ -1714,38 +1724,21 @@ fn spawn_work<'a, B: ExtraBackendMethods>( B::spawn_named_thread(cgcx.time_trace, work.short_description(), move || { let result = std::panic::catch_unwind(AssertUnwindSafe(|| match work { - WorkItem::Optimize(m) => { - let _timer = - cgcx.prof.generic_activity_with_arg("codegen_module_optimize", &*m.name); - execute_optimize_work_item(&cgcx, m) - } - WorkItem::CopyPostLtoArtifacts(m) => { - let _timer = cgcx - .prof - .generic_activity_with_arg("codegen_copy_artifacts_from_incr_cache", &*m.name); - execute_copy_from_cache_work_item(&cgcx, m) - } + WorkItem::Optimize(m) => execute_optimize_work_item(&cgcx, m), + WorkItem::CopyPostLtoArtifacts(m) => execute_copy_from_cache_work_item(&cgcx, m), WorkItem::FatLto { exported_symbols_for_lto, each_linked_rlib_for_lto, needs_fat_lto, import_only_modules, - } => { - let _timer = - cgcx.prof.generic_activity_with_arg("codegen_module_perform_lto", "everything"); - execute_fat_lto_work_item( - &cgcx, - &exported_symbols_for_lto, - &each_linked_rlib_for_lto, - needs_fat_lto, - import_only_modules, - ) - } - WorkItem::ThinLto(m) => { - let _timer = - cgcx.prof.generic_activity_with_arg("codegen_module_perform_lto", m.name()); - execute_thin_lto_work_item(&cgcx, m) - } + } => execute_fat_lto_work_item( + &cgcx, + &exported_symbols_for_lto, + &each_linked_rlib_for_lto, + needs_fat_lto, + import_only_modules, + ), + WorkItem::ThinLto(m) => execute_thin_lto_work_item(&cgcx, m), })); let msg = match result { -- cgit 1.4.1-3-g733a5