From 21d096184e3c7e05646ee756a2ff6a99e90aff77 Mon Sep 17 00:00:00 2001 From: Manuel Drehwald Date: Fri, 7 Feb 2025 22:27:46 -0500 Subject: fix non-enzyme builds --- compiler/rustc_codegen_llvm/src/back/write.rs | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'compiler/rustc_codegen_llvm/src/back/write.rs') diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index 4706744f353..155c07ff635 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -551,11 +551,12 @@ pub(crate) unsafe fn llvm_optimize( let vectorize_slp; let vectorize_loop; + let run_enzyme = cfg!(llvm_enzyme); // When we build rustc with enzyme/autodiff support, we want to postpone size-increasing // optimizations until after differentiation. FIXME(ZuseZ4): Before shipping on nightly, // we should make this more granular, or at least check that the user has at least one autodiff // call in their code, to justify altering the compilation pipeline. - if skip_size_increasing_opts && cfg!(llvm_enzyme) { + if skip_size_increasing_opts && run_enzyme { unroll_loops = false; vectorize_slp = false; vectorize_loop = false; @@ -633,6 +634,7 @@ pub(crate) unsafe fn llvm_optimize( vectorize_loop, config.no_builtins, config.emit_lifetime_markers, + run_enzyme, sanitizer_options.as_ref(), pgo_gen_path.as_ref().map_or(std::ptr::null(), |s| s.as_ptr()), pgo_use_path.as_ref().map_or(std::ptr::null(), |s| s.as_ptr()), -- cgit 1.4.1-3-g733a5 From 1221cff55149d2dbbf8761345799ef06f8099b97 Mon Sep 17 00:00:00 2001 From: Manuel Drehwald Date: Mon, 10 Feb 2025 01:35:22 -0500 Subject: move second opt run to lto phase and cleanup code --- compiler/rustc_codegen_llvm/src/back/lto.rs | 25 +++++++++++-- compiler/rustc_codegen_llvm/src/back/write.rs | 41 +++++++++++++--------- .../rustc_codegen_llvm/src/builder/autodiff.rs | 35 ++++-------------- compiler/rustc_llvm/build.rs | 4 +++ compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp | 11 ++++-- src/bootstrap/src/core/build_steps/compile.rs | 9 +++-- tests/codegen/autodiff.rs | 4 +-- 7 files changed, 75 insertions(+), 54 deletions(-) (limited to 'compiler/rustc_codegen_llvm/src/back/write.rs') diff --git a/compiler/rustc_codegen_llvm/src/back/lto.rs b/compiler/rustc_codegen_llvm/src/back/lto.rs index 78c759bbe8c..8bad437eeb7 100644 --- a/compiler/rustc_codegen_llvm/src/back/lto.rs +++ b/compiler/rustc_codegen_llvm/src/back/lto.rs @@ -606,10 +606,31 @@ pub(crate) fn run_pass_manager( // If this rustc version was build with enzyme/autodiff enabled, and if users applied the // `#[autodiff]` macro at least once, then we will later call llvm_optimize a second time. - let first_run = true; debug!("running llvm pm opt pipeline"); unsafe { - write::llvm_optimize(cgcx, dcx, module, config, opt_level, opt_stage, first_run)?; + write::llvm_optimize( + cgcx, + dcx, + module, + config, + opt_level, + opt_stage, + write::AutodiffStage::DuringAD, + )?; + } + // FIXME(ZuseZ4): Make this more granular + if cfg!(llvm_enzyme) && !thin { + unsafe { + write::llvm_optimize( + cgcx, + dcx, + module, + config, + opt_level, + llvm::OptStage::FatLTO, + write::AutodiffStage::PostAD, + )?; + } } debug!("lto done"); Ok(()) diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index 155c07ff635..ae4c4d5876e 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -530,6 +530,16 @@ fn get_instr_profile_output_path(config: &ModuleConfig) -> Option { config.instrument_coverage.then(|| c"default_%m_%p.profraw".to_owned()) } +// PreAD will run llvm opts but disable size increasing opts (vectorization, loop unrolling) +// DuringAD is the same as above, but also runs the enzyme opt and autodiff passes. +// PostAD will run all opts, including size increasing opts. +#[derive(Debug, Eq, PartialEq)] +pub(crate) enum AutodiffStage { + PreAD, + DuringAD, + PostAD, +} + pub(crate) unsafe fn llvm_optimize( cgcx: &CodegenContext, dcx: DiagCtxtHandle<'_>, @@ -537,7 +547,7 @@ pub(crate) unsafe fn llvm_optimize( config: &ModuleConfig, opt_level: config::OptLevel, opt_stage: llvm::OptStage, - skip_size_increasing_opts: bool, + autodiff_stage: AutodiffStage, ) -> Result<(), FatalError> { // Enzyme: // The whole point of compiler based AD is to differentiate optimized IR instead of unoptimized @@ -550,13 +560,16 @@ pub(crate) unsafe fn llvm_optimize( let unroll_loops; let vectorize_slp; let vectorize_loop; + let run_enzyme = cfg!(llvm_enzyme) && autodiff_stage == AutodiffStage::DuringAD; - let run_enzyme = cfg!(llvm_enzyme); // When we build rustc with enzyme/autodiff support, we want to postpone size-increasing - // optimizations until after differentiation. FIXME(ZuseZ4): Before shipping on nightly, + // optimizations until after differentiation. Our pipeline is thus: (opt + enzyme), (full opt). + // We therefore have two calls to llvm_optimize, if autodiff is used. + // + // FIXME(ZuseZ4): Before shipping on nightly, // we should make this more granular, or at least check that the user has at least one autodiff // call in their code, to justify altering the compilation pipeline. - if skip_size_increasing_opts && run_enzyme { + if cfg!(llvm_enzyme) && autodiff_stage != AutodiffStage::PostAD { unroll_loops = false; vectorize_slp = false; vectorize_loop = false; @@ -566,7 +579,7 @@ pub(crate) unsafe fn llvm_optimize( vectorize_slp = config.vectorize_slp; vectorize_loop = config.vectorize_loop; } - trace!(?unroll_loops, ?vectorize_slp, ?vectorize_loop); + trace!(?unroll_loops, ?vectorize_slp, ?vectorize_loop, ?run_enzyme); let using_thin_buffers = opt_stage == llvm::OptStage::PreLinkThinLTO || config.bitcode_needed(); let pgo_gen_path = get_pgo_gen_path(config); let pgo_use_path = get_pgo_use_path(config); @@ -686,18 +699,14 @@ pub(crate) unsafe fn optimize( _ => llvm::OptStage::PreLinkNoLTO, }; - // If we know that we will later run AD, then we disable vectorization and loop unrolling - let skip_size_increasing_opts = cfg!(llvm_enzyme); + // If we know that we will later run AD, then we disable vectorization and loop unrolling. + // Otherwise we pretend AD is already done and run the normal opt pipeline (=PostAD). + // FIXME(ZuseZ4): Make this more granular, only set PreAD if we actually have autodiff + // usages, not just if we build rustc with autodiff support. + let autodiff_stage = + if cfg!(llvm_enzyme) { AutodiffStage::PreAD } else { AutodiffStage::PostAD }; return unsafe { - llvm_optimize( - cgcx, - dcx, - module, - config, - opt_level, - opt_stage, - skip_size_increasing_opts, - ) + llvm_optimize(cgcx, dcx, module, config, opt_level, opt_stage, autodiff_stage) }; } Ok(()) diff --git a/compiler/rustc_codegen_llvm/src/builder/autodiff.rs b/compiler/rustc_codegen_llvm/src/builder/autodiff.rs index 474b0940203..f82449479a8 100644 --- a/compiler/rustc_codegen_llvm/src/builder/autodiff.rs +++ b/compiler/rustc_codegen_llvm/src/builder/autodiff.rs @@ -4,10 +4,9 @@ use rustc_ast::expand::autodiff_attrs::{AutoDiffAttrs, AutoDiffItem, DiffActivit use rustc_codegen_ssa::ModuleCodegen; use rustc_codegen_ssa::back::write::ModuleConfig; use rustc_errors::FatalError; -use rustc_session::config::Lto; use tracing::{debug, trace}; -use crate::back::write::{llvm_err, llvm_optimize}; +use crate::back::write::llvm_err; use crate::builder::SBuilder; use crate::context::SimpleCx; use crate::declare::declare_simple_fn; @@ -153,7 +152,7 @@ fn generate_enzyme_call<'ll>( _ => {} } - trace!("matching autodiff arguments"); + debug!("matching autodiff arguments"); // We now handle the issue that Rust level arguments not always match the llvm-ir level // arguments. A slice, `&[f32]`, for example, is represented as a pointer and a length on // llvm-ir level. The number of activities matches the number of Rust level arguments, so we @@ -222,7 +221,10 @@ fn generate_enzyme_call<'ll>( // A duplicated pointer will have the following two outer_fn arguments: // (..., ptr, ptr, ...). We add the following llvm-ir to our __enzyme call: // (..., metadata! enzyme_dup, ptr, ptr, ...). - if matches!(diff_activity, DiffActivity::Duplicated | DiffActivity::DuplicatedOnly) { + if matches!( + diff_activity, + DiffActivity::Duplicated | DiffActivity::DuplicatedOnly + ) { assert!( llvm::LLVMRustGetTypeKind(next_outer_ty) == llvm::TypeKind::Pointer ); @@ -282,7 +284,7 @@ pub(crate) fn differentiate<'ll>( module: &'ll ModuleCodegen, cgcx: &CodegenContext, diff_items: Vec, - config: &ModuleConfig, + _config: &ModuleConfig, ) -> Result<(), FatalError> { for item in &diff_items { trace!("{}", item); @@ -317,29 +319,6 @@ pub(crate) fn differentiate<'ll>( // FIXME(ZuseZ4): support SanitizeHWAddress and prevent illegal/unsupported opts - if let Some(opt_level) = config.opt_level { - let opt_stage = match cgcx.lto { - Lto::Fat => llvm::OptStage::PreLinkFatLTO, - Lto::Thin | Lto::ThinLocal => llvm::OptStage::PreLinkThinLTO, - _ if cgcx.opts.cg.linker_plugin_lto.enabled() => llvm::OptStage::PreLinkThinLTO, - _ => llvm::OptStage::PreLinkNoLTO, - }; - // This is our second opt call, so now we run all opts, - // to make sure we get the best performance. - let skip_size_increasing_opts = false; - trace!("running Module Optimization after differentiation"); - unsafe { - llvm_optimize( - cgcx, - diag_handler.handle(), - module, - config, - opt_level, - opt_stage, - skip_size_increasing_opts, - )? - }; - } trace!("done with differentiate()"); Ok(()) diff --git a/compiler/rustc_llvm/build.rs b/compiler/rustc_llvm/build.rs index d9d28299413..48806888b43 100644 --- a/compiler/rustc_llvm/build.rs +++ b/compiler/rustc_llvm/build.rs @@ -193,6 +193,10 @@ fn main() { cfg.define(&flag, None); } + if tracked_env_var_os("LLVM_ENZYME").is_some() { + cfg.define("ENZYME", None); + } + if tracked_env_var_os("LLVM_RUSTLLVM").is_some() { cfg.define("LLVM_RUSTLLVM", None); } diff --git a/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp b/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp index 0e591786d91..a6b2384f2d7 100644 --- a/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp +++ b/compiler/rustc_llvm/llvm-wrapper/PassWrapper.cpp @@ -689,7 +689,9 @@ struct LLVMRustSanitizerOptions { }; // This symbol won't be available or used when Enzyme is not enabled -extern "C" void registerEnzyme(llvm::PassBuilder &PB) __attribute__((weak)); +#ifdef ENZYME +extern "C" void registerEnzyme(llvm::PassBuilder &PB); +#endif extern "C" LLVMRustResult LLVMRustOptimize( LLVMModuleRef ModuleRef, LLVMTargetMachineRef TMRef, @@ -697,8 +699,9 @@ extern "C" LLVMRustResult LLVMRustOptimize( bool IsLinkerPluginLTO, bool NoPrepopulatePasses, bool VerifyIR, bool LintIR, bool UseThinLTOBuffers, bool MergeFunctions, bool UnrollLoops, bool SLPVectorize, bool LoopVectorize, bool DisableSimplifyLibCalls, - bool EmitLifetimeMarkers, bool RunEnzyme, LLVMRustSanitizerOptions *SanitizerOptions, - const char *PGOGenPath, const char *PGOUsePath, bool InstrumentCoverage, + bool EmitLifetimeMarkers, bool RunEnzyme, + LLVMRustSanitizerOptions *SanitizerOptions, const char *PGOGenPath, + const char *PGOUsePath, bool InstrumentCoverage, const char *InstrProfileOutput, const char *PGOSampleUsePath, bool DebugInfoForProfiling, void *LlvmSelfProfiler, LLVMRustSelfProfileBeforePassCallback BeforePassCallback, @@ -1014,6 +1017,7 @@ extern "C" LLVMRustResult LLVMRustOptimize( } // now load "-enzyme" pass: +#ifdef ENZYME if (RunEnzyme) { registerEnzyme(PB); if (auto Err = PB.parsePassPipeline(MPM, "enzyme")) { @@ -1022,6 +1026,7 @@ extern "C" LLVMRustResult LLVMRustOptimize( return LLVMRustResult::Failure; } } +#endif // Upgrade all calls to old intrinsics first. for (Module::iterator I = TheModule->begin(), E = TheModule->end(); I != E;) diff --git a/src/bootstrap/src/core/build_steps/compile.rs b/src/bootstrap/src/core/build_steps/compile.rs index 4e360e94fd6..308d4723d98 100644 --- a/src/bootstrap/src/core/build_steps/compile.rs +++ b/src/bootstrap/src/core/build_steps/compile.rs @@ -1049,9 +1049,9 @@ pub fn rustc_cargo( // . cargo.rustflag("-Zon-broken-pipe=kill"); - // We temporarily disable linking here as part of some refactoring. - // This way, people can manually use -Z llvm-plugins and -C passes=enzyme for now. - // In a follow-up PR, we will re-enable linking here and load the pass for them. + // We want to link against registerEnzyme and in the future we want to use additional + // functionality from Enzyme core. For that we need to link against Enzyme. + // FIXME(ZuseZ4): Get the LLVM version number automatically instead of hardcoding it. if builder.config.llvm_enzyme { cargo.rustflag("-l").rustflag("Enzyme-19"); } @@ -1234,6 +1234,9 @@ fn rustc_llvm_env(builder: &Builder<'_>, cargo: &mut Cargo, target: TargetSelect if builder.is_rust_llvm(target) { cargo.env("LLVM_RUSTLLVM", "1"); } + if builder.config.llvm_enzyme { + cargo.env("LLVM_ENZYME", "1"); + } let llvm::LlvmResult { llvm_config, .. } = builder.ensure(llvm::Llvm { target }); cargo.env("LLVM_CONFIG", &llvm_config); diff --git a/tests/codegen/autodiff.rs b/tests/codegen/autodiff.rs index efc5ac77f1b..abf7fcf3e4b 100644 --- a/tests/codegen/autodiff.rs +++ b/tests/codegen/autodiff.rs @@ -15,9 +15,9 @@ fn square(x: &f64) -> f64 { // CHECK-NEXT:invertstart: // CHECK-NEXT: %_0 = fmul double %x.0.val, %x.0.val // CHECK-NEXT: %0 = fadd fast double %x.0.val, %x.0.val -// CHECK-NEXT: %1 = load double, ptr %"x'", align 8, !alias.scope !17816, !noalias !17819 +// CHECK-NEXT: %1 = load double, ptr %"x'", align 8 // CHECK-NEXT: %2 = fadd fast double %1, %0 -// CHECK-NEXT: store double %2, ptr %"x'", align 8, !alias.scope !17816, !noalias !17819 +// CHECK-NEXT: store double %2, ptr %"x'", align 8 // CHECK-NEXT: ret double %_0 // CHECK-NEXT:} -- cgit 1.4.1-3-g733a5 From b2cd1b8ead95530e61ad108a3113c588e590187a Mon Sep 17 00:00:00 2001 From: Oli Scherer Date: Tue, 14 Jan 2025 12:25:16 +0000 Subject: Remove an unsafe closure invariant by inlining the closure wrapper into the called function --- compiler/rustc_codegen_llvm/src/back/write.rs | 130 ++++++++++---------------- 1 file changed, 50 insertions(+), 80 deletions(-) (limited to 'compiler/rustc_codegen_llvm/src/back/write.rs') diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index ae4c4d5876e..6a7b6996347 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -40,7 +40,7 @@ use crate::errors::{ WithLlvmError, WriteBytecode, }; use crate::llvm::diagnostic::OptimizationDiagnosticKind::*; -use crate::llvm::{self, DiagnosticInfo, PassManager}; +use crate::llvm::{self, DiagnosticInfo}; use crate::type_::Type; use crate::{LlvmCodegenBackend, ModuleLlvm, base, common, llvm_util}; @@ -54,7 +54,7 @@ pub(crate) fn llvm_err<'a>(dcx: DiagCtxtHandle<'_>, err: LlvmError<'a>) -> Fatal fn write_output_file<'ll>( dcx: DiagCtxtHandle<'_>, target: &'ll llvm::TargetMachine, - pm: &llvm::PassManager<'ll>, + no_builtins: bool, m: &'ll llvm::Module, output: &Path, dwo_output: Option<&Path>, @@ -63,16 +63,19 @@ fn write_output_file<'ll>( verify_llvm_ir: bool, ) -> Result<(), FatalError> { debug!("write_output_file output={:?} dwo_output={:?}", output, dwo_output); - unsafe { - let output_c = path_to_c_string(output); - let dwo_output_c; - let dwo_output_ptr = if let Some(dwo_output) = dwo_output { - dwo_output_c = path_to_c_string(dwo_output); - dwo_output_c.as_ptr() - } else { - std::ptr::null() - }; - let result = llvm::LLVMRustWriteOutputFile( + let output_c = path_to_c_string(output); + let dwo_output_c; + let dwo_output_ptr = if let Some(dwo_output) = dwo_output { + dwo_output_c = path_to_c_string(dwo_output); + dwo_output_c.as_ptr() + } else { + std::ptr::null() + }; + let result = unsafe { + let pm = llvm::LLVMCreatePassManager(); + llvm::LLVMAddAnalysisPasses(target, pm); + llvm::LLVMRustAddLibraryInfo(pm, m, no_builtins); + llvm::LLVMRustWriteOutputFile( target, pm, m, @@ -80,22 +83,22 @@ fn write_output_file<'ll>( dwo_output_ptr, file_type, verify_llvm_ir, - ); + ) + }; - // Record artifact sizes for self-profiling - if result == llvm::LLVMRustResult::Success { - let artifact_kind = match file_type { - llvm::FileType::ObjectFile => "object_file", - llvm::FileType::AssemblyFile => "assembly_file", - }; - record_artifact_size(self_profiler_ref, artifact_kind, output); - if let Some(dwo_file) = dwo_output { - record_artifact_size(self_profiler_ref, "dwo_file", dwo_file); - } + // Record artifact sizes for self-profiling + if result == llvm::LLVMRustResult::Success { + let artifact_kind = match file_type { + llvm::FileType::ObjectFile => "object_file", + llvm::FileType::AssemblyFile => "assembly_file", + }; + record_artifact_size(self_profiler_ref, artifact_kind, output); + if let Some(dwo_file) = dwo_output { + record_artifact_size(self_profiler_ref, "dwo_file", dwo_file); } - - result.into_result().map_err(|()| llvm_err(dcx, LlvmError::WriteOutput { path: output })) } + + result.into_result().map_err(|()| llvm_err(dcx, LlvmError::WriteOutput { path: output })) } pub(crate) fn create_informational_target_machine( @@ -755,31 +758,6 @@ pub(crate) unsafe fn codegen( create_msvc_imps(cgcx, llcx, llmod); } - // A codegen-specific pass manager is used to generate object - // files for an LLVM module. - // - // Apparently each of these pass managers is a one-shot kind of - // thing, so we create a new one for each type of output. The - // pass manager passed to the closure should be ensured to not - // escape the closure itself, and the manager should only be - // used once. - unsafe fn with_codegen<'ll, F, R>( - tm: &'ll llvm::TargetMachine, - llmod: &'ll llvm::Module, - no_builtins: bool, - f: F, - ) -> R - where - F: FnOnce(&'ll mut PassManager<'ll>) -> R, - { - unsafe { - let cpm = llvm::LLVMCreatePassManager(); - llvm::LLVMAddAnalysisPasses(tm, cpm); - llvm::LLVMRustAddLibraryInfo(cpm, llmod, no_builtins); - f(cpm) - } - } - // Note that if object files are just LLVM bitcode we write bitcode, // copy it to the .o file, and delete the bitcode if it wasn't // otherwise requested. @@ -898,21 +876,17 @@ pub(crate) unsafe fn codegen( } else { llmod }; - unsafe { - with_codegen(tm, llmod, config.no_builtins, |cpm| { - write_output_file( - dcx, - tm, - cpm, - llmod, - &path, - None, - llvm::FileType::AssemblyFile, - &cgcx.prof, - config.verify_llvm_ir, - ) - })?; - } + write_output_file( + dcx, + tm, + config.no_builtins, + llmod, + &path, + None, + llvm::FileType::AssemblyFile, + &cgcx.prof, + config.verify_llvm_ir, + )?; } match config.emit_obj { @@ -936,21 +910,17 @@ pub(crate) unsafe fn codegen( (_, SplitDwarfKind::Split) => Some(dwo_out.as_path()), }; - unsafe { - with_codegen(tm, llmod, config.no_builtins, |cpm| { - write_output_file( - dcx, - tm, - cpm, - llmod, - &obj_out, - dwo_out, - llvm::FileType::ObjectFile, - &cgcx.prof, - config.verify_llvm_ir, - ) - })?; - } + write_output_file( + dcx, + tm, + config.no_builtins, + llmod, + &obj_out, + dwo_out, + llvm::FileType::ObjectFile, + &cgcx.prof, + config.verify_llvm_ir, + )?; } EmitObj::Bitcode => { -- cgit 1.4.1-3-g733a5 From 4b83038d637c1664df94895c82cb8b26716bad1d Mon Sep 17 00:00:00 2001 From: Oli Scherer Date: Tue, 14 Jan 2025 12:25:16 +0000 Subject: Add a safe wrapper for `WriteBitcodeToFile` --- compiler/rustc_codegen_llvm/src/back/write.rs | 18 ++++++++++-------- 1 file changed, 10 insertions(+), 8 deletions(-) (limited to 'compiler/rustc_codegen_llvm/src/back/write.rs') diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index 6a7b6996347..39aa3a033dc 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -328,13 +328,17 @@ pub(crate) fn save_temp_bitcode( if !cgcx.save_temps { return; } + let ext = format!("{name}.bc"); + let cgu = Some(&module.name[..]); + let path = cgcx.output_filenames.temp_path_ext(&ext, cgu); + write_bitcode_to_file(module, &path) +} + +fn write_bitcode_to_file(module: &ModuleCodegen, path: &Path) { unsafe { - let ext = format!("{name}.bc"); - let cgu = Some(&module.name[..]); - let path = cgcx.output_filenames.temp_path_ext(&ext, cgu); - let cstr = path_to_c_string(&path); + let path = path_to_c_string(&path); let llmod = module.module_llvm.llmod(); - llvm::LLVMWriteBitcodeToFile(llmod, cstr.as_ptr()); + llvm::LLVMWriteBitcodeToFile(llmod, path.as_ptr()); } } @@ -679,7 +683,6 @@ pub(crate) unsafe fn optimize( ) -> Result<(), FatalError> { let _timer = cgcx.prof.generic_activity_with_arg("LLVM_module_optimize", &*module.name); - let llmod = module.module_llvm.llmod(); let llcx = &*module.module_llvm.llcx; let _handlers = DiagnosticHandlers::new(cgcx, dcx, llcx, module, CodegenDiagnosticsStage::Opt); @@ -688,8 +691,7 @@ pub(crate) unsafe fn optimize( if config.emit_no_opt_bc { let out = cgcx.output_filenames.temp_path_ext("no-opt.bc", module_name); - let out = path_to_c_string(&out); - unsafe { llvm::LLVMWriteBitcodeToFile(llmod, out.as_ptr()) }; + write_bitcode_to_file(module, &out) } // FIXME(ZuseZ4): support SanitizeHWAddress and prevent illegal/unsupported opts -- cgit 1.4.1-3-g733a5 From dcf1e4d72b4dbda592050dcde7d7972414b05d3c Mon Sep 17 00:00:00 2001 From: Oli Scherer Date: Tue, 14 Jan 2025 12:25:16 +0000 Subject: Document some safety constraints and use more safe wrappers --- compiler/rustc_codegen_llvm/src/allocator.rs | 4 +- compiler/rustc_codegen_llvm/src/back/archive.rs | 11 ++-- compiler/rustc_codegen_llvm/src/back/write.rs | 64 +++++++++++------------- compiler/rustc_codegen_llvm/src/common.rs | 2 +- compiler/rustc_codegen_llvm/src/consts.rs | 6 +-- compiler/rustc_codegen_llvm/src/context.rs | 10 ++-- compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs | 2 +- compiler/rustc_codegen_llvm/src/declare.rs | 2 +- compiler/rustc_codegen_llvm/src/intrinsic.rs | 2 +- compiler/rustc_codegen_llvm/src/llvm/ffi.rs | 2 +- compiler/rustc_codegen_llvm/src/llvm/mod.rs | 4 ++ 11 files changed, 50 insertions(+), 59 deletions(-) (limited to 'compiler/rustc_codegen_llvm/src/back/write.rs') diff --git a/compiler/rustc_codegen_llvm/src/allocator.rs b/compiler/rustc_codegen_llvm/src/allocator.rs index 149ded28356..66723cbf882 100644 --- a/compiler/rustc_codegen_llvm/src/allocator.rs +++ b/compiler/rustc_codegen_llvm/src/allocator.rs @@ -81,13 +81,13 @@ pub(crate) unsafe fn codegen( llvm::set_visibility(ll_g, llvm::Visibility::from_generic(tcx.sess.default_visibility())); let val = tcx.sess.opts.unstable_opts.oom.should_panic(); let llval = llvm::LLVMConstInt(i8, val as u64, False); - llvm::LLVMSetInitializer(ll_g, llval); + llvm::set_initializer(ll_g, llval); let name = NO_ALLOC_SHIM_IS_UNSTABLE; let ll_g = llvm::LLVMRustGetOrInsertGlobal(llmod, name.as_c_char_ptr(), name.len(), i8); llvm::set_visibility(ll_g, llvm::Visibility::from_generic(tcx.sess.default_visibility())); let llval = llvm::LLVMConstInt(i8, 0, False); - llvm::LLVMSetInitializer(ll_g, llval); + llvm::set_initializer(ll_g, llval); } if tcx.sess.opts.debuginfo != DebugInfo::None { diff --git a/compiler/rustc_codegen_llvm/src/back/archive.rs b/compiler/rustc_codegen_llvm/src/back/archive.rs index 33a956e552f..93553f3f364 100644 --- a/compiler/rustc_codegen_llvm/src/back/archive.rs +++ b/compiler/rustc_codegen_llvm/src/back/archive.rs @@ -11,7 +11,7 @@ use rustc_codegen_ssa::back::archive::{ use rustc_session::Session; use crate::llvm::archive_ro::{ArchiveRO, Child}; -use crate::llvm::{self, ArchiveKind}; +use crate::llvm::{self, ArchiveKind, last_error}; /// Helper for adding many files to an archive. #[must_use = "must call build() to finish building the archive"] @@ -169,6 +169,8 @@ impl<'a> LlvmArchiveBuilder<'a> { .unwrap_or_else(|kind| self.sess.dcx().emit_fatal(UnknownArchiveKind { kind })); let mut additions = mem::take(&mut self.additions); + // Values in the `members` list below will contain pointers to the strings allocated here. + // So they need to get dropped after all elements of `members` get freed. let mut strings = Vec::new(); let mut members = Vec::new(); @@ -229,12 +231,7 @@ impl<'a> LlvmArchiveBuilder<'a> { self.sess.target.arch == "arm64ec", ); let ret = if r.into_result().is_err() { - let err = llvm::LLVMRustGetLastError(); - let msg = if err.is_null() { - "failed to write archive".into() - } else { - String::from_utf8_lossy(CStr::from_ptr(err).to_bytes()) - }; + let msg = last_error().unwrap_or_else(|| "failed to write archive".into()); Err(io::Error::new(io::ErrorKind::Other, msg)) } else { Ok(!members.is_empty()) diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index 39aa3a033dc..58933a77e53 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -1049,24 +1049,18 @@ unsafe fn embed_bitcode( { // We don't need custom section flags, create LLVM globals. let llconst = common::bytes_in_context(llcx, bitcode); - let llglobal = llvm::LLVMAddGlobal( - llmod, - common::val_ty(llconst), - c"rustc.embedded.module".as_ptr(), - ); - llvm::LLVMSetInitializer(llglobal, llconst); + let llglobal = + llvm::add_global(llmod, common::val_ty(llconst), c"rustc.embedded.module"); + llvm::set_initializer(llglobal, llconst); llvm::set_section(llglobal, bitcode_section_name(cgcx)); llvm::set_linkage(llglobal, llvm::Linkage::PrivateLinkage); llvm::LLVMSetGlobalConstant(llglobal, llvm::True); let llconst = common::bytes_in_context(llcx, cmdline.as_bytes()); - let llglobal = llvm::LLVMAddGlobal( - llmod, - common::val_ty(llconst), - c"rustc.embedded.cmdline".as_ptr(), - ); - llvm::LLVMSetInitializer(llglobal, llconst); + let llglobal = + llvm::add_global(llmod, common::val_ty(llconst), c"rustc.embedded.cmdline"); + llvm::set_initializer(llglobal, llconst); let section = if cgcx.target_is_like_osx { c"__LLVM,__cmdline" } else if cgcx.target_is_like_aix { @@ -1106,31 +1100,29 @@ fn create_msvc_imps( // underscores added in front). let prefix = if cgcx.target_arch == "x86" { "\x01__imp__" } else { "\x01__imp_" }; - unsafe { - let ptr_ty = Type::ptr_llcx(llcx); - let globals = base::iter_globals(llmod) - .filter(|&val| { - llvm::get_linkage(val) == llvm::Linkage::ExternalLinkage - && llvm::LLVMIsDeclaration(val) == 0 - }) - .filter_map(|val| { - // Exclude some symbols that we know are not Rust symbols. - let name = llvm::get_value_name(val); - if ignored(name) { None } else { Some((val, name)) } - }) - .map(move |(val, name)| { - let mut imp_name = prefix.as_bytes().to_vec(); - imp_name.extend(name); - let imp_name = CString::new(imp_name).unwrap(); - (imp_name, val) - }) - .collect::>(); + let ptr_ty = Type::ptr_llcx(llcx); + let globals = base::iter_globals(llmod) + .filter(|&val| { + llvm::get_linkage(val) == llvm::Linkage::ExternalLinkage && !llvm::is_declaration(val) + }) + .filter_map(|val| { + // Exclude some symbols that we know are not Rust symbols. + let name = llvm::get_value_name(val); + if ignored(name) { None } else { Some((val, name)) } + }) + .map(move |(val, name)| { + let mut imp_name = prefix.as_bytes().to_vec(); + imp_name.extend(name); + let imp_name = CString::new(imp_name).unwrap(); + (imp_name, val) + }) + .collect::>(); - for (imp_name, val) in globals { - let imp = llvm::LLVMAddGlobal(llmod, ptr_ty, imp_name.as_ptr()); - llvm::LLVMSetInitializer(imp, val); - llvm::set_linkage(imp, llvm::Linkage::ExternalLinkage); - } + for (imp_name, val) in globals { + let imp = llvm::add_global(llmod, ptr_ty, &imp_name); + + llvm::set_initializer(imp, val); + llvm::set_linkage(imp, llvm::Linkage::ExternalLinkage); } // Use this function to exclude certain symbols from `__imp` generation. diff --git a/compiler/rustc_codegen_llvm/src/common.rs b/compiler/rustc_codegen_llvm/src/common.rs index 8c94a46ebf3..78b3a7f8541 100644 --- a/compiler/rustc_codegen_llvm/src/common.rs +++ b/compiler/rustc_codegen_llvm/src/common.rs @@ -219,8 +219,8 @@ impl<'ll, 'tcx> ConstCodegenMethods<'tcx> for CodegenCx<'ll, 'tcx> { let g = self.define_global(&sym, self.val_ty(sc)).unwrap_or_else(|| { bug!("symbol `{}` is already defined", sym); }); + llvm::set_initializer(g, sc); unsafe { - llvm::LLVMSetInitializer(g, sc); llvm::LLVMSetGlobalConstant(g, True); llvm::LLVMSetUnnamedAddress(g, llvm::UnnamedAddr::Global); } diff --git a/compiler/rustc_codegen_llvm/src/consts.rs b/compiler/rustc_codegen_llvm/src/consts.rs index c6855dd42e5..4a5491ec7a1 100644 --- a/compiler/rustc_codegen_llvm/src/consts.rs +++ b/compiler/rustc_codegen_llvm/src/consts.rs @@ -191,7 +191,7 @@ fn check_and_apply_linkage<'ll, 'tcx>( }) }); llvm::set_linkage(g2, llvm::Linkage::InternalLinkage); - unsafe { llvm::LLVMSetInitializer(g2, g1) }; + llvm::set_initializer(g2, g1); g2 } else if cx.tcx.sess.target.arch == "x86" && common::is_mingw_gnu_toolchain(&cx.tcx.sess.target) @@ -235,7 +235,7 @@ impl<'ll> CodegenCx<'ll, '_> { } _ => self.define_private_global(self.val_ty(cv)), }; - unsafe { llvm::LLVMSetInitializer(gv, cv) }; + llvm::set_initializer(gv, cv); set_global_alignment(self, gv, align); llvm::SetUnnamedAddress(gv, llvm::UnnamedAddr::Global); gv @@ -458,7 +458,7 @@ impl<'ll> CodegenCx<'ll, '_> { new_g }; set_global_alignment(self, g, alloc.align); - llvm::LLVMSetInitializer(g, v); + llvm::set_initializer(g, v); if self.should_assume_dso_local(g, true) { llvm::LLVMRustSetDSOLocal(g, true); diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs index ba4fd75fb94..7fe527a4c07 100644 --- a/compiler/rustc_codegen_llvm/src/context.rs +++ b/compiler/rustc_codegen_llvm/src/context.rs @@ -616,12 +616,10 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> { pub(crate) fn create_used_variable_impl(&self, name: &'static CStr, values: &[&'ll Value]) { let array = self.const_array(self.type_ptr(), values); - unsafe { - let g = llvm::LLVMAddGlobal(self.llmod, self.val_ty(array), name.as_ptr()); - llvm::LLVMSetInitializer(g, array); - llvm::set_linkage(g, llvm::Linkage::AppendingLinkage); - llvm::set_section(g, c"llvm.metadata"); - } + let g = llvm::add_global(self.llmod, self.val_ty(array), name); + llvm::set_initializer(g, array); + llvm::set_linkage(g, llvm::Linkage::AppendingLinkage); + llvm::set_section(g, c"llvm.metadata"); } } impl<'ll> SimpleCx<'ll> { diff --git a/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs b/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs index 2c9f1cda13a..54c5d445f66 100644 --- a/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs +++ b/compiler/rustc_codegen_llvm/src/debuginfo/gdb.rs @@ -73,7 +73,7 @@ pub(crate) fn get_or_insert_gdb_debug_scripts_section_global<'ll>( .define_global(section_var_name, llvm_type) .unwrap_or_else(|| bug!("symbol `{}` is already defined", section_var_name)); llvm::set_section(section_var, c".debug_gdb_scripts"); - llvm::LLVMSetInitializer(section_var, cx.const_bytes(section_contents)); + llvm::set_initializer(section_var, cx.const_bytes(section_contents)); llvm::LLVMSetGlobalConstant(section_var, llvm::True); llvm::LLVMSetUnnamedAddress(section_var, llvm::UnnamedAddr::Global); llvm::set_linkage(section_var, llvm::Linkage::LinkOnceODRLinkage); diff --git a/compiler/rustc_codegen_llvm/src/declare.rs b/compiler/rustc_codegen_llvm/src/declare.rs index bdc83267cca..cebceef1b93 100644 --- a/compiler/rustc_codegen_llvm/src/declare.rs +++ b/compiler/rustc_codegen_llvm/src/declare.rs @@ -235,7 +235,7 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> { /// name. pub(crate) fn get_defined_value(&self, name: &str) -> Option<&'ll Value> { self.get_declared_value(name).and_then(|val| { - let declaration = unsafe { llvm::LLVMIsDeclaration(val) != 0 }; + let declaration = llvm::is_declaration(val); if !declaration { Some(val) } else { None } }) } diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs index 3200c94d977..8b976885904 100644 --- a/compiler/rustc_codegen_llvm/src/intrinsic.rs +++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs @@ -824,7 +824,7 @@ fn codegen_msvc_try<'ll>( if bx.cx.tcx.sess.target.supports_comdat() { llvm::SetUniqueComdat(bx.llmod, tydesc); } - unsafe { llvm::LLVMSetInitializer(tydesc, type_info) }; + llvm::set_initializer(tydesc, type_info); // The flag value of 8 indicates that we are catching the exception by // reference instead of by value. We can't use catch by value because diff --git a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs index 4d6a76b23ea..441d144ce50 100644 --- a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs +++ b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs @@ -2359,7 +2359,7 @@ unsafe extern "C" { ); pub fn LLVMRustWriteOutputFile<'a>( T: &'a TargetMachine, - PM: &PassManager<'a>, + PM: *mut PassManager<'a>, M: &'a Module, Output: *const c_char, DwoOutput: *const c_char, diff --git a/compiler/rustc_codegen_llvm/src/llvm/mod.rs b/compiler/rustc_codegen_llvm/src/llvm/mod.rs index 707aeba22cc..7becba4ccd4 100644 --- a/compiler/rustc_codegen_llvm/src/llvm/mod.rs +++ b/compiler/rustc_codegen_llvm/src/llvm/mod.rs @@ -241,6 +241,10 @@ pub fn set_linkage(llglobal: &Value, linkage: Linkage) { } } +pub fn is_declaration(llglobal: &Value) -> bool { + unsafe { LLVMIsDeclaration(llglobal) == ffi::True } +} + pub fn get_visibility(llglobal: &Value) -> Visibility { unsafe { LLVMGetVisibility(llglobal) }.to_rust() } -- cgit 1.4.1-3-g733a5 From 296625613386532e8a6963f821716db119c2ab62 Mon Sep 17 00:00:00 2001 From: clubby789 Date: Mon, 13 Jan 2025 15:52:08 +0000 Subject: Make `-O` mean `-C opt-level=3` --- compiler/rustc_codegen_cranelift/src/lib.rs | 2 +- compiler/rustc_codegen_gcc/src/lib.rs | 2 +- compiler/rustc_codegen_llvm/src/back/write.rs | 4 ++-- compiler/rustc_codegen_ssa/src/back/linker.rs | 16 ++++++------- compiler/rustc_codegen_ssa/src/back/write.rs | 4 ++-- compiler/rustc_codegen_ssa/src/base.rs | 6 ++--- compiler/rustc_mir_transform/src/inline.rs | 3 +-- compiler/rustc_session/src/config.rs | 26 +++++++++++++--------- src/doc/rustc/src/codegen-options/index.md | 2 +- src/doc/rustc/src/command-line-arguments.md | 2 +- .../run-make-support/src/external_deps/rustc.rs | 2 +- tests/run-make/rustc-help/help-v.stdout | 2 +- tests/run-make/rustc-help/help.stdout | 2 +- 13 files changed, 39 insertions(+), 34 deletions(-) (limited to 'compiler/rustc_codegen_llvm/src/back/write.rs') diff --git a/compiler/rustc_codegen_cranelift/src/lib.rs b/compiler/rustc_codegen_cranelift/src/lib.rs index b0153419903..a3f43744875 100644 --- a/compiler/rustc_codegen_cranelift/src/lib.rs +++ b/compiler/rustc_codegen_cranelift/src/lib.rs @@ -289,7 +289,7 @@ fn build_isa(sess: &Session) -> Arc { flags_builder.set("opt_level", "none").unwrap(); } OptLevel::Less - | OptLevel::Default + | OptLevel::More | OptLevel::Size | OptLevel::SizeMin | OptLevel::Aggressive => { diff --git a/compiler/rustc_codegen_gcc/src/lib.rs b/compiler/rustc_codegen_gcc/src/lib.rs index ce88ac39021..6455bcec685 100644 --- a/compiler/rustc_codegen_gcc/src/lib.rs +++ b/compiler/rustc_codegen_gcc/src/lib.rs @@ -476,7 +476,7 @@ fn to_gcc_opt_level(optlevel: Option) -> OptimizationLevel { Some(level) => match level { OptLevel::No => OptimizationLevel::None, OptLevel::Less => OptimizationLevel::Limited, - OptLevel::Default => OptimizationLevel::Standard, + OptLevel::More => OptimizationLevel::Standard, OptLevel::Aggressive => OptimizationLevel::Aggressive, OptLevel::Size | OptLevel::SizeMin => OptimizationLevel::Limited, }, diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index 58933a77e53..9fa10e96068 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -141,7 +141,7 @@ fn to_llvm_opt_settings(cfg: config::OptLevel) -> (llvm::CodeGenOptLevel, llvm:: match cfg { No => (llvm::CodeGenOptLevel::None, llvm::CodeGenOptSizeNone), Less => (llvm::CodeGenOptLevel::Less, llvm::CodeGenOptSizeNone), - Default => (llvm::CodeGenOptLevel::Default, llvm::CodeGenOptSizeNone), + More => (llvm::CodeGenOptLevel::Default, llvm::CodeGenOptSizeNone), Aggressive => (llvm::CodeGenOptLevel::Aggressive, llvm::CodeGenOptSizeNone), Size => (llvm::CodeGenOptLevel::Default, llvm::CodeGenOptSizeDefault), SizeMin => (llvm::CodeGenOptLevel::Default, llvm::CodeGenOptSizeAggressive), @@ -153,7 +153,7 @@ fn to_pass_builder_opt_level(cfg: config::OptLevel) -> llvm::PassBuilderOptLevel match cfg { No => llvm::PassBuilderOptLevel::O0, Less => llvm::PassBuilderOptLevel::O1, - Default => llvm::PassBuilderOptLevel::O2, + More => llvm::PassBuilderOptLevel::O2, Aggressive => llvm::PassBuilderOptLevel::O3, Size => llvm::PassBuilderOptLevel::Os, SizeMin => llvm::PassBuilderOptLevel::Oz, diff --git a/compiler/rustc_codegen_ssa/src/back/linker.rs b/compiler/rustc_codegen_ssa/src/back/linker.rs index 8fb831471a9..05d6ff35751 100644 --- a/compiler/rustc_codegen_ssa/src/back/linker.rs +++ b/compiler/rustc_codegen_ssa/src/back/linker.rs @@ -410,7 +410,7 @@ impl<'a> GccLinker<'a> { let opt_level = match self.sess.opts.optimize { config::OptLevel::No => "O0", config::OptLevel::Less => "O1", - config::OptLevel::Default | config::OptLevel::Size | config::OptLevel::SizeMin => "O2", + config::OptLevel::More | config::OptLevel::Size | config::OptLevel::SizeMin => "O2", config::OptLevel::Aggressive => "O3", }; @@ -685,7 +685,7 @@ impl<'a> Linker for GccLinker<'a> { // GNU-style linkers support optimization with -O. GNU ld doesn't // need a numeric argument, but other linkers do. - if self.sess.opts.optimize == config::OptLevel::Default + if self.sess.opts.optimize == config::OptLevel::More || self.sess.opts.optimize == config::OptLevel::Aggressive { self.link_arg("-O1"); @@ -1213,7 +1213,7 @@ impl<'a> Linker for EmLinker<'a> { self.cc_arg(match self.sess.opts.optimize { OptLevel::No => "-O0", OptLevel::Less => "-O1", - OptLevel::Default => "-O2", + OptLevel::More => "-O2", OptLevel::Aggressive => "-O3", OptLevel::Size => "-Os", OptLevel::SizeMin => "-Oz", @@ -1384,7 +1384,7 @@ impl<'a> Linker for WasmLd<'a> { self.link_arg(match self.sess.opts.optimize { OptLevel::No => "-O0", OptLevel::Less => "-O1", - OptLevel::Default => "-O2", + OptLevel::More => "-O2", OptLevel::Aggressive => "-O3", // Currently LLD doesn't support `Os` and `Oz`, so pass through `O2` // instead. @@ -1451,7 +1451,7 @@ impl<'a> WasmLd<'a> { let opt_level = match self.sess.opts.optimize { config::OptLevel::No => "O0", config::OptLevel::Less => "O1", - config::OptLevel::Default => "O2", + config::OptLevel::More => "O2", config::OptLevel::Aggressive => "O3", // wasm-ld only handles integer LTO opt levels. Use O2 config::OptLevel::Size | config::OptLevel::SizeMin => "O2", @@ -1525,7 +1525,7 @@ impl<'a> Linker for L4Bender<'a> { fn optimize(&mut self) { // GNU-style linkers support optimization with -O. GNU ld doesn't // need a numeric argument, but other linkers do. - if self.sess.opts.optimize == config::OptLevel::Default + if self.sess.opts.optimize == config::OptLevel::More || self.sess.opts.optimize == config::OptLevel::Aggressive { self.link_arg("-O1"); @@ -1929,7 +1929,7 @@ impl<'a> Linker for LlbcLinker<'a> { match self.sess.opts.optimize { OptLevel::No => "-O0", OptLevel::Less => "-O1", - OptLevel::Default => "-O2", + OptLevel::More => "-O2", OptLevel::Aggressive => "-O3", OptLevel::Size => "-Os", OptLevel::SizeMin => "-Oz", @@ -2006,7 +2006,7 @@ impl<'a> Linker for BpfLinker<'a> { self.link_arg(match self.sess.opts.optimize { OptLevel::No => "-O0", OptLevel::Less => "-O1", - OptLevel::Default => "-O2", + OptLevel::More => "-O2", OptLevel::Aggressive => "-O3", OptLevel::Size => "-Os", OptLevel::SizeMin => "-Oz", diff --git a/compiler/rustc_codegen_ssa/src/back/write.rs b/compiler/rustc_codegen_ssa/src/back/write.rs index ce2161a07eb..f029c08a808 100644 --- a/compiler/rustc_codegen_ssa/src/back/write.rs +++ b/compiler/rustc_codegen_ssa/src/back/write.rs @@ -236,7 +236,7 @@ impl ModuleConfig { // Copy what clang does by turning on loop vectorization at O2 and // slp vectorization at O3. vectorize_loop: !sess.opts.cg.no_vectorize_loops - && (sess.opts.optimize == config::OptLevel::Default + && (sess.opts.optimize == config::OptLevel::More || sess.opts.optimize == config::OptLevel::Aggressive), vectorize_slp: !sess.opts.cg.no_vectorize_slp && sess.opts.optimize == config::OptLevel::Aggressive, @@ -260,7 +260,7 @@ impl ModuleConfig { MergeFunctions::Trampolines | MergeFunctions::Aliases => { use config::OptLevel::*; match sess.opts.optimize { - Aggressive | Default | SizeMin | Size => true, + Aggressive | More | SizeMin | Size => true, Less | No => false, } } diff --git a/compiler/rustc_codegen_ssa/src/base.rs b/compiler/rustc_codegen_ssa/src/base.rs index d9fbf539fd3..e800492dad0 100644 --- a/compiler/rustc_codegen_ssa/src/base.rs +++ b/compiler/rustc_codegen_ssa/src/base.rs @@ -1054,12 +1054,12 @@ pub(crate) fn provide(providers: &mut Providers) { config::OptLevel::No => return config::OptLevel::No, // If globally optimise-speed is already specified, just use that level. config::OptLevel::Less => return config::OptLevel::Less, - config::OptLevel::Default => return config::OptLevel::Default, + config::OptLevel::More => return config::OptLevel::More, config::OptLevel::Aggressive => return config::OptLevel::Aggressive, // If globally optimize-for-size has been requested, use -O2 instead (if optimize(size) // are present). - config::OptLevel::Size => config::OptLevel::Default, - config::OptLevel::SizeMin => config::OptLevel::Default, + config::OptLevel::Size => config::OptLevel::More, + config::OptLevel::SizeMin => config::OptLevel::More, }; let defids = tcx.collect_and_partition_mono_items(cratenum).all_mono_items; diff --git a/compiler/rustc_mir_transform/src/inline.rs b/compiler/rustc_mir_transform/src/inline.rs index ab617e2ce6f..1f8392b2118 100644 --- a/compiler/rustc_mir_transform/src/inline.rs +++ b/compiler/rustc_mir_transform/src/inline.rs @@ -49,8 +49,7 @@ impl<'tcx> crate::MirPass<'tcx> for Inline { match sess.mir_opt_level() { 0 | 1 => false, 2 => { - (sess.opts.optimize == OptLevel::Default - || sess.opts.optimize == OptLevel::Aggressive) + (sess.opts.optimize == OptLevel::More || sess.opts.optimize == OptLevel::Aggressive) && sess.opts.incremental == None } _ => true, diff --git a/compiler/rustc_session/src/config.rs b/compiler/rustc_session/src/config.rs index 2fb4b27b889..7d473b86ff5 100644 --- a/compiler/rustc_session/src/config.rs +++ b/compiler/rustc_session/src/config.rs @@ -86,12 +86,18 @@ pub enum CFProtection { #[derive(Clone, Copy, Debug, PartialEq, Hash, HashStable_Generic)] pub enum OptLevel { - No, // -O0 - Less, // -O1 - Default, // -O2 - Aggressive, // -O3 - Size, // -Os - SizeMin, // -Oz + /// `-Copt-level=0` + No, + /// `-Copt-level=1` + Less, + /// `-Copt-level=2` + More, + /// `-Copt-level=3` / `-O` + Aggressive, + /// `-Copt-level=s` + Size, + /// `-Copt-level=z` + SizeMin, } /// This is what the `LtoCli` values get mapped to after resolving defaults and @@ -1253,7 +1259,7 @@ impl Options { Some(setting) => setting, None => match self.optimize { OptLevel::No | OptLevel::Less | OptLevel::Size | OptLevel::SizeMin => true, - OptLevel::Default | OptLevel::Aggressive => false, + OptLevel::More | OptLevel::Aggressive => false, }, } } @@ -1572,7 +1578,7 @@ pub fn rustc_optgroups() -> Vec { stack-protector-strategies|link-args|deployment-target]", ), opt(Stable, FlagMulti, "g", "", "Equivalent to -C debuginfo=2", ""), - opt(Stable, FlagMulti, "O", "", "Equivalent to -C opt-level=2", ""), + opt(Stable, FlagMulti, "O", "", "Equivalent to -C opt-level=3", ""), opt(Stable, Opt, "o", "", "Write output to ", "FILENAME"), opt(Stable, Opt, "", "out-dir", "Write output to compiler-chosen filename in ", "DIR"), opt( @@ -2127,12 +2133,12 @@ fn parse_opt_level( }) .max(); if max_o > max_c { - OptLevel::Default + OptLevel::Aggressive } else { match cg.opt_level.as_ref() { "0" => OptLevel::No, "1" => OptLevel::Less, - "2" => OptLevel::Default, + "2" => OptLevel::More, "3" => OptLevel::Aggressive, "s" => OptLevel::Size, "z" => OptLevel::SizeMin, diff --git a/src/doc/rustc/src/codegen-options/index.md b/src/doc/rustc/src/codegen-options/index.md index f45217c69ff..8c1769a8c77 100644 --- a/src/doc/rustc/src/codegen-options/index.md +++ b/src/doc/rustc/src/codegen-options/index.md @@ -368,7 +368,7 @@ This flag controls the optimization level. * `s`: optimize for binary size. * `z`: optimize for binary size, but also turn off loop vectorization. -Note: The [`-O` flag][option-o-optimize] is an alias for `-C opt-level=2`. +Note: The [`-O` flag][option-o-optimize] is an alias for `-C opt-level=3`. The default is `0`. diff --git a/src/doc/rustc/src/command-line-arguments.md b/src/doc/rustc/src/command-line-arguments.md index 0942b5ebfee..9dd2e7de1b3 100644 --- a/src/doc/rustc/src/command-line-arguments.md +++ b/src/doc/rustc/src/command-line-arguments.md @@ -308,7 +308,7 @@ A synonym for [`-C debuginfo=2`](codegen-options/index.md#debuginfo). ## `-O`: optimize your code -A synonym for [`-C opt-level=2`](codegen-options/index.md#opt-level). +A synonym for [`-C opt-level=3`](codegen-options/index.md#opt-level). ## `-o`: filename of the output diff --git a/src/tools/run-make-support/src/external_deps/rustc.rs b/src/tools/run-make-support/src/external_deps/rustc.rs index 33dc57cbc07..fd4a20278ad 100644 --- a/src/tools/run-make-support/src/external_deps/rustc.rs +++ b/src/tools/run-make-support/src/external_deps/rustc.rs @@ -84,7 +84,7 @@ impl Rustc { self } - /// Specify default optimization level `-O` (alias for `-C opt-level=2`). + /// Specify default optimization level `-O` (alias for `-C opt-level=3`). pub fn opt(&mut self) -> &mut Self { self.cmd.arg("-O"); self diff --git a/tests/run-make/rustc-help/help-v.stdout b/tests/run-make/rustc-help/help-v.stdout index 8f6fde69c29..382b1c96682 100644 --- a/tests/run-make/rustc-help/help-v.stdout +++ b/tests/run-make/rustc-help/help-v.stdout @@ -32,7 +32,7 @@ Options: --print [crate-name|file-names|sysroot|target-libdir|cfg|check-cfg|calling-conventions|target-list|target-cpus|target-features|relocation-models|code-models|tls-models|target-spec-json|all-target-specs-json|native-static-libs|stack-protector-strategies|link-args|deployment-target] Compiler information to print on stdout -g Equivalent to -C debuginfo=2 - -O Equivalent to -C opt-level=2 + -O Equivalent to -C opt-level=3 -o FILENAME Write output to --out-dir DIR Write output to compiler-chosen filename in --explain OPT Provide a detailed explanation of an error message diff --git a/tests/run-make/rustc-help/help.stdout b/tests/run-make/rustc-help/help.stdout index 131efa93282..08bb7f85098 100644 --- a/tests/run-make/rustc-help/help.stdout +++ b/tests/run-make/rustc-help/help.stdout @@ -32,7 +32,7 @@ Options: --print [crate-name|file-names|sysroot|target-libdir|cfg|check-cfg|calling-conventions|target-list|target-cpus|target-features|relocation-models|code-models|tls-models|target-spec-json|all-target-specs-json|native-static-libs|stack-protector-strategies|link-args|deployment-target] Compiler information to print on stdout -g Equivalent to -C debuginfo=2 - -O Equivalent to -C opt-level=2 + -O Equivalent to -C opt-level=3 -o FILENAME Write output to --out-dir DIR Write output to compiler-chosen filename in --explain OPT Provide a detailed explanation of an error message -- cgit 1.4.1-3-g733a5 From f4e2218b131b82097c720d28c1c2ea10922b0d47 Mon Sep 17 00:00:00 2001 From: Manuel Drehwald Date: Fri, 21 Feb 2025 21:47:48 -0500 Subject: clean up autodiff code/comments --- compiler/rustc_ast/src/expand/autodiff_attrs.rs | 1 - compiler/rustc_codegen_llvm/src/back/write.rs | 15 +++++---------- compiler/rustc_monomorphize/src/partitioning/autodiff.rs | 2 +- 3 files changed, 6 insertions(+), 12 deletions(-) (limited to 'compiler/rustc_codegen_llvm/src/back/write.rs') diff --git a/compiler/rustc_ast/src/expand/autodiff_attrs.rs b/compiler/rustc_ast/src/expand/autodiff_attrs.rs index 70222f4acab..c8ec185ee5e 100644 --- a/compiler/rustc_ast/src/expand/autodiff_attrs.rs +++ b/compiler/rustc_ast/src/expand/autodiff_attrs.rs @@ -17,7 +17,6 @@ use crate::{Ty, TyKind}; /// functions. The proper solution is to recognize and resolve this DAG of autodiff invocations, /// as it's already done in the C++ and Julia frontend of Enzyme. /// -/// (FIXME) remove *First variants. /// Documentation for using [reverse](https://enzyme.mit.edu/rust/rev.html) and /// [forward](https://enzyme.mit.edu/rust/fwd.html) mode is available online. #[derive(Clone, Copy, Eq, PartialEq, Encodable, Decodable, Debug, HashStable_Generic)] diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index 9fa10e96068..b67890c0465 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -564,19 +564,16 @@ pub(crate) unsafe fn llvm_optimize( // FIXME(ZuseZ4): In a future update we could figure out how to only optimize individual functions getting // differentiated. + let consider_ad = cfg!(llvm_enzyme) && config.autodiff.contains(&config::AutoDiff::Enable); + let run_enzyme = autodiff_stage == AutodiffStage::DuringAD; let unroll_loops; let vectorize_slp; let vectorize_loop; - let run_enzyme = cfg!(llvm_enzyme) && autodiff_stage == AutodiffStage::DuringAD; // When we build rustc with enzyme/autodiff support, we want to postpone size-increasing // optimizations until after differentiation. Our pipeline is thus: (opt + enzyme), (full opt). // We therefore have two calls to llvm_optimize, if autodiff is used. - // - // FIXME(ZuseZ4): Before shipping on nightly, - // we should make this more granular, or at least check that the user has at least one autodiff - // call in their code, to justify altering the compilation pipeline. - if cfg!(llvm_enzyme) && autodiff_stage != AutodiffStage::PostAD { + if consider_ad && autodiff_stage != AutodiffStage::PostAD { unroll_loops = false; vectorize_slp = false; vectorize_loop = false; @@ -706,10 +703,8 @@ pub(crate) unsafe fn optimize( // If we know that we will later run AD, then we disable vectorization and loop unrolling. // Otherwise we pretend AD is already done and run the normal opt pipeline (=PostAD). - // FIXME(ZuseZ4): Make this more granular, only set PreAD if we actually have autodiff - // usages, not just if we build rustc with autodiff support. - let autodiff_stage = - if cfg!(llvm_enzyme) { AutodiffStage::PreAD } else { AutodiffStage::PostAD }; + let consider_ad = cfg!(llvm_enzyme) && config.autodiff.contains(&config::AutoDiff::Enable); + let autodiff_stage = if consider_ad { AutodiffStage::PreAD } else { AutodiffStage::PostAD }; return unsafe { llvm_optimize(cgcx, dcx, module, config, opt_level, opt_stage, autodiff_stage) }; diff --git a/compiler/rustc_monomorphize/src/partitioning/autodiff.rs b/compiler/rustc_monomorphize/src/partitioning/autodiff.rs index bce31bf0748..0c855508434 100644 --- a/compiler/rustc_monomorphize/src/partitioning/autodiff.rs +++ b/compiler/rustc_monomorphize/src/partitioning/autodiff.rs @@ -66,7 +66,7 @@ pub(crate) fn find_autodiff_source_functions<'tcx>( let mut autodiff_items: Vec = vec![]; for (item, instance) in autodiff_mono_items { let target_id = instance.def_id(); - let cg_fn_attr = tcx.codegen_fn_attrs(target_id).autodiff_item.clone(); + let cg_fn_attr = &tcx.codegen_fn_attrs(target_id).autodiff_item; let Some(target_attrs) = cg_fn_attr else { continue; }; -- cgit 1.4.1-3-g733a5