//! Inlining pass for MIR functions use rustc::hir::CodegenFnAttrFlags; use rustc::hir::def_id::DefId; use rustc_data_structures::bit_set::BitSet; use rustc_data_structures::indexed_vec::{Idx, IndexVec}; use rustc::mir::*; use rustc::mir::visit::*; use rustc::ty::{self, Instance, InstanceDef, ParamEnv, Ty, TyCtxt}; use rustc::ty::subst::{Subst, SubstsRef}; use std::collections::VecDeque; use std::iter; use crate::transform::{MirPass, MirSource}; use super::simplify::{remove_dead_blocks, CfgSimplifier}; use syntax::attr; use rustc_target::spec::abi::Abi; const DEFAULT_THRESHOLD: usize = 50; const HINT_THRESHOLD: usize = 100; const INSTR_COST: usize = 5; const CALL_PENALTY: usize = 25; const UNKNOWN_SIZE_COST: usize = 10; pub struct Inline; #[derive(Copy, Clone, Debug)] struct CallSite<'tcx> { callee: DefId, substs: SubstsRef<'tcx>, bb: BasicBlock, location: SourceInfo, } impl MirPass for Inline { fn run_pass<'tcx>(&self, tcx: TyCtxt<'tcx, 'tcx, 'tcx>, source: MirSource<'tcx>, body: &mut Body<'tcx>) { if tcx.sess.opts.debugging_opts.mir_opt_level >= 2 { Inliner { tcx, source }.run_pass(body); } } } struct Inliner<'tcx> { tcx: TyCtxt<'tcx, 'tcx, 'tcx>, source: MirSource<'tcx>, } impl Inliner<'tcx> { fn run_pass(&self, caller_body: &mut Body<'tcx>) { // Keep a queue of callsites to try inlining on. We take // advantage of the fact that queries detect cycles here to // allow us to try and fetch the fully optimized MIR of a // call; if it succeeds, we can inline it and we know that // they do not call us. Otherwise, we just don't try to // inline. // // We use a queue so that we inline "broadly" before we inline // in depth. It is unclear if this is the best heuristic, // really, but that's true of all the heuristics in this // file. =) let mut callsites = VecDeque::new(); let param_env = self.tcx.param_env(self.source.def_id()); // Only do inlining into fn bodies. let id = self.tcx.hir().as_local_hir_id(self.source.def_id()).unwrap(); if self.tcx.hir().body_owner_kind_by_hir_id(id).is_fn_or_closure() && self.source.promoted.is_none() { for (bb, bb_data) in caller_body.basic_blocks().iter_enumerated() { if let Some(callsite) = self.get_valid_function_call(bb, bb_data, caller_body, param_env) { callsites.push_back(callsite); } } } else { return; } let mut local_change; let mut changed = false; loop { local_change = false; while let Some(callsite) = callsites.pop_front() { debug!("checking whether to inline callsite {:?}", callsite); if !self.tcx.is_mir_available(callsite.callee) { debug!("checking whether to inline callsite {:?} - MIR unavailable", callsite); continue; } let self_node_id = self.tcx.hir().as_local_node_id(self.source.def_id()).unwrap(); let callee_node_id = self.tcx.hir().as_local_node_id(callsite.callee); let callee_body = if let Some(callee_node_id) = callee_node_id { // Avoid a cycle here by only using `optimized_mir` only if we have // a lower node id than the callee. This ensures that the callee will // not inline us. This trick only works without incremental compilation. // So don't do it if that is enabled. if !self.tcx.dep_graph.is_fully_enabled() && self_node_id.as_u32() < callee_node_id.as_u32() { self.tcx.optimized_mir(callsite.callee) } else { continue; } } else { // This cannot result in a cycle since the callee MIR is from another crate // and is already optimized. self.tcx.optimized_mir(callsite.callee) }; let callee_body = if self.consider_optimizing(callsite, callee_body) { self.tcx.subst_and_normalize_erasing_regions( &callsite.substs, param_env, callee_body, ) } else { continue; }; let start = caller_body.basic_blocks().len(); debug!("attempting to inline callsite {:?} - body={:?}", callsite, callee_body); if !self.inline_call(callsite, caller_body, callee_body) { debug!("attempting to inline callsite {:?} - failure", callsite); continue; } debug!("attempting to inline callsite {:?} - success", callsite); // Add callsites from inlined function for (bb, bb_data) in caller_body.basic_blocks().iter_enumerated().skip(start) { if let Some(new_callsite) = self.get_valid_function_call(bb, bb_data, caller_body, param_env) { // Don't inline the same function multiple times. if callsite.callee != new_callsite.callee { callsites.push_back(new_callsite); } } } local_change = true; changed = true; } if !local_change { break; } } // Simplify if we inlined anything. if changed { debug!("Running simplify cfg on {:?}", self.source); CfgSimplifier::new(caller_body).simplify(); remove_dead_blocks(caller_body); } } fn get_valid_function_call(&self, bb: BasicBlock, bb_data: &BasicBlockData<'tcx>, caller_body: &Body<'tcx>, param_env: ParamEnv<'tcx>, ) -> Option> { // Don't inline calls that are in cleanup blocks. if bb_data.is_cleanup { return None; } // Only consider direct calls to functions let terminator = bb_data.terminator(); if let TerminatorKind::Call { func: ref op, .. } = terminator.kind { if let ty::FnDef(callee_def_id, substs) = op.ty(caller_body, self.tcx).sty { let instance = Instance::resolve(self.tcx, param_env, callee_def_id, substs)?; if let InstanceDef::Virtual(..) = instance.def { return None; } return Some(CallSite { callee: instance.def_id(), substs: instance.substs, bb, location: terminator.source_info }); } } None } fn consider_optimizing(&self, callsite: CallSite<'tcx>, callee_body: &Body<'tcx>) -> bool { debug!("consider_optimizing({:?})", callsite); self.should_inline(callsite, callee_body) && self.tcx.consider_optimizing(|| format!("Inline {:?} into {:?}", callee_body.span, callsite)) } fn should_inline(&self, callsite: CallSite<'tcx>, callee_body: &Body<'tcx>) -> bool { debug!("should_inline({:?})", callsite); let tcx = self.tcx; // Don't inline closures that have capture debuginfo // FIXME: Handle closures better if callee_body.__upvar_debuginfo_codegen_only_do_not_use.len() > 0 { debug!(" upvar debuginfo present - not inlining"); return false; } // Cannot inline generators which haven't been transformed yet if callee_body.yield_ty.is_some() { debug!(" yield ty present - not inlining"); return false; } // Do not inline {u,i}128 lang items, codegen const eval depends // on detecting calls to these lang items and intercepting them if tcx.is_binop_lang_item(callsite.callee).is_some() { debug!(" not inlining 128bit integer lang item"); return false; } let codegen_fn_attrs = tcx.codegen_fn_attrs(callsite.callee); let hinted = match codegen_fn_attrs.inline { // Just treat inline(always) as a hint for now, // there are cases that prevent inlining that we // need to check for first. attr::InlineAttr::Always => true, attr::InlineAttr::Never => { debug!("#[inline(never)] present - not inlining"); return false } attr::InlineAttr::Hint => true, attr::InlineAttr::None => false, }; // Only inline local functions if they would be eligible for cross-crate // inlining. This is to ensure that the final crate doesn't have MIR that // reference unexported symbols if callsite.callee.is_local() { if callsite.substs.non_erasable_generics().count() == 0 && !hinted { debug!(" callee is an exported function - not inlining"); return false; } } let mut threshold = if hinted { HINT_THRESHOLD } else { DEFAULT_THRESHOLD }; // Significantly lower the threshold for inlining cold functions if codegen_fn_attrs.flags.contains(CodegenFnAttrFlags::COLD) { threshold /= 5; } // Give a bonus functions with a small number of blocks, // We normally have two or three blocks for even // very small functions. if callee_body.basic_blocks().len() <= 3 { threshold += threshold / 4; } debug!(" final inline threshold = {}", threshold); // FIXME: Give a bonus to functions with only a single caller let param_env = tcx.param_env(self.source.def_id()); let mut first_block = true; let mut cost = 0; // Traverse the MIR manually so we can account for the effects of // inlining on the CFG. let mut work_list = vec![START_BLOCK]; let mut visited = BitSet::new_empty(callee_body.basic_blocks().len()); while let Some(bb) = work_list.pop() { if !visited.insert(bb.index()) { continue; } let blk = &callee_body.basic_blocks()[bb]; for stmt in &blk.statements { // Don't count StorageLive/StorageDead in the inlining cost. match stmt.kind { StatementKind::StorageLive(_) | StatementKind::StorageDead(_) | StatementKind::Nop => {} _ => cost += INSTR_COST } } let term = blk.terminator(); let mut is_drop = false; match term.kind { TerminatorKind::Drop { ref location, target, unwind } | TerminatorKind::DropAndReplace { ref location, target, unwind, .. } => { is_drop = true; work_list.push(target); // If the location doesn't actually need dropping, treat it like // a regular goto. let ty = location.ty(callee_body, tcx).subst(tcx, callsite.substs).ty; if ty.needs_drop(tcx, param_env) { cost += CALL_PENALTY; if let Some(unwind) = unwind { work_list.push(unwind); } } else { cost += INSTR_COST; } } TerminatorKind::Unreachable | TerminatorKind::Call { destination: None, .. } if first_block => { // If the function always diverges, don't inline // unless the cost is zero threshold = 0; } TerminatorKind::Call {func: Operand::Constant(ref f), .. } => { if let ty::FnDef(def_id, _) = f.ty.sty { // Don't give intrinsics the extra penalty for calls let f = tcx.fn_sig(def_id); if f.abi() == Abi::RustIntrinsic || f.abi() == Abi::PlatformIntrinsic { cost += INSTR_COST; } else { cost += CALL_PENALTY; } } } TerminatorKind::Assert { .. } => cost += CALL_PENALTY, _ => cost += INSTR_COST } if !is_drop { for &succ in term.successors() { work_list.push(succ); } } first_block = false; } // Count up the cost of local variables and temps, if we know the size // use that, otherwise we use a moderately-large dummy cost. let ptr_size = tcx.data_layout.pointer_size.bytes(); for v in callee_body.vars_and_temps_iter() { let v = &callee_body.local_decls[v]; let ty = v.ty.subst(tcx, callsite.substs); // Cost of the var is the size in machine-words, if we know // it. if let Some(size) = type_size_of(tcx, param_env.clone(), ty) { cost += (size / ptr_size) as usize; } else { cost += UNKNOWN_SIZE_COST; } } if let attr::InlineAttr::Always = codegen_fn_attrs.inline { debug!("INLINING {:?} because inline(always) [cost={}]", callsite, cost); true } else { if cost <= threshold { debug!("INLINING {:?} [cost={} <= threshold={}]", callsite, cost, threshold); true } else { debug!("NOT inlining {:?} [cost={} > threshold={}]", callsite, cost, threshold); false } } } fn inline_call(&self, callsite: CallSite<'tcx>, caller_body: &mut Body<'tcx>, mut callee_body: Body<'tcx>) -> bool { let terminator = caller_body[callsite.bb].terminator.take().unwrap(); match terminator.kind { // FIXME: Handle inlining of diverging calls TerminatorKind::Call { args, destination: Some(destination), cleanup, .. } => { debug!("Inlined {:?} into {:?}", callsite.callee, self.source); let mut local_map = IndexVec::with_capacity(callee_body.local_decls.len()); let mut scope_map = IndexVec::with_capacity(callee_body.source_scopes.len()); let mut promoted_map = IndexVec::with_capacity(callee_body.promoted.len()); for mut scope in callee_body.source_scopes.iter().cloned() { if scope.parent_scope.is_none() { scope.parent_scope = Some(callsite.location.scope); scope.span = callee_body.span; } scope.span = callsite.location.span; let idx = caller_body.source_scopes.push(scope); scope_map.push(idx); } for loc in callee_body.vars_and_temps_iter() { let mut local = callee_body.local_decls[loc].clone(); local.source_info.scope = scope_map[local.source_info.scope]; local.source_info.span = callsite.location.span; local.visibility_scope = scope_map[local.visibility_scope]; let idx = caller_body.local_decls.push(local); local_map.push(idx); } promoted_map.extend( callee_body.promoted.iter().cloned().map(|p| caller_body.promoted.push(p)) ); // If the call is something like `a[*i] = f(i)`, where // `i : &mut usize`, then just duplicating the `a[*i]` // Place could result in two different locations if `f` // writes to `i`. To prevent this we need to create a temporary // borrow of the place and pass the destination as `*temp` instead. fn dest_needs_borrow(place: &Place<'_>) -> bool { place.iterate(|place_base, place_projection| { for proj in place_projection { match proj.elem { ProjectionElem::Deref | ProjectionElem::Index(_) => return true, _ => {} } } match place_base { // Static variables need a borrow because the callee // might modify the same static. PlaceBase::Static(_) => true, _ => false } }) } let dest = if dest_needs_borrow(&destination.0) { debug!("Creating temp for return destination"); let dest = Rvalue::Ref( self.tcx.lifetimes.re_erased, BorrowKind::Mut { allow_two_phase_borrow: false }, destination.0); let ty = dest.ty(caller_body, self.tcx); let temp = LocalDecl::new_temp(ty, callsite.location.span); let tmp = caller_body.local_decls.push(temp); let tmp = Place::Base(PlaceBase::Local(tmp)); let stmt = Statement { source_info: callsite.location, kind: StatementKind::Assign(tmp.clone(), box dest) }; caller_body[callsite.bb] .statements.push(stmt); tmp.deref() } else { destination.0 }; let return_block = destination.1; // Copy the arguments if needed. let args: Vec<_> = self.make_call_args(args, &callsite, caller_body); let bb_len = caller_body.basic_blocks().len(); let mut integrator = Integrator { block_idx: bb_len, args: &args, local_map, scope_map, promoted_map, _callsite: callsite, destination: dest, return_block, cleanup_block: cleanup, in_cleanup_block: false }; for (bb, mut block) in callee_body.basic_blocks_mut().drain_enumerated(..) { integrator.visit_basic_block_data(bb, &mut block); caller_body.basic_blocks_mut().push(block); } let terminator = Terminator { source_info: callsite.location, kind: TerminatorKind::Goto { target: BasicBlock::new(bb_len) } }; caller_body[callsite.bb].terminator = Some(terminator); true } kind => { caller_body[callsite.bb].terminator = Some(Terminator { source_info: terminator.source_info, kind, }); false } } } fn make_call_args( &self, args: Vec>, callsite: &CallSite<'tcx>, caller_body: &mut Body<'tcx>, ) -> Vec { let tcx = self.tcx; // There is a bit of a mismatch between the *caller* of a closure and the *callee*. // The caller provides the arguments wrapped up in a tuple: // // tuple_tmp = (a, b, c) // Fn::call(closure_ref, tuple_tmp) // // meanwhile the closure body expects the arguments (here, `a`, `b`, and `c`) // as distinct arguments. (This is the "rust-call" ABI hack.) Normally, codegen has // the job of unpacking this tuple. But here, we are codegen. =) So we want to create // a vector like // // [closure_ref, tuple_tmp.0, tuple_tmp.1, tuple_tmp.2] // // Except for one tiny wrinkle: we don't actually want `tuple_tmp.0`. It's more convenient // if we "spill" that into *another* temporary, so that we can map the argument // variable in the callee MIR directly to an argument variable on our side. // So we introduce temporaries like: // // tmp0 = tuple_tmp.0 // tmp1 = tuple_tmp.1 // tmp2 = tuple_tmp.2 // // and the vector is `[closure_ref, tmp0, tmp1, tmp2]`. if tcx.is_closure(callsite.callee) { let mut args = args.into_iter(); let self_ = self.create_temp_if_necessary(args.next().unwrap(), callsite, caller_body); let tuple = self.create_temp_if_necessary(args.next().unwrap(), callsite, caller_body); assert!(args.next().is_none()); let tuple = Place::Base(PlaceBase::Local(tuple)); let tuple_tys = if let ty::Tuple(s) = tuple.ty(caller_body, tcx).ty.sty { s } else { bug!("Closure arguments are not passed as a tuple"); }; // The `closure_ref` in our example above. let closure_ref_arg = iter::once(self_); // The `tmp0`, `tmp1`, and `tmp2` in our example abonve. let tuple_tmp_args = tuple_tys.iter().enumerate().map(|(i, ty)| { // This is e.g., `tuple_tmp.0` in our example above. let tuple_field = Operand::Move(tuple.clone().field( Field::new(i), ty.expect_ty(), )); // Spill to a local to make e.g., `tmp0`. self.create_temp_if_necessary(tuple_field, callsite, caller_body) }); closure_ref_arg.chain(tuple_tmp_args).collect() } else { args.into_iter() .map(|a| self.create_temp_if_necessary(a, callsite, caller_body)) .collect() } } /// If `arg` is already a temporary, returns it. Otherwise, introduces a fresh /// temporary `T` and an instruction `T = arg`, and returns `T`. fn create_temp_if_necessary( &self, arg: Operand<'tcx>, callsite: &CallSite<'tcx>, caller_body: &mut Body<'tcx>, ) -> Local { // FIXME: Analysis of the usage of the arguments to avoid // unnecessary temporaries. if let Operand::Move(Place::Base(PlaceBase::Local(local))) = arg { if caller_body.local_kind(local) == LocalKind::Temp { // Reuse the operand if it's a temporary already return local; } } debug!("Creating temp for argument {:?}", arg); // Otherwise, create a temporary for the arg let arg = Rvalue::Use(arg); let ty = arg.ty(caller_body, self.tcx); let arg_tmp = LocalDecl::new_temp(ty, callsite.location.span); let arg_tmp = caller_body.local_decls.push(arg_tmp); let stmt = Statement { source_info: callsite.location, kind: StatementKind::Assign(Place::Base(PlaceBase::Local(arg_tmp)), box arg), }; caller_body[callsite.bb].statements.push(stmt); arg_tmp } } fn type_size_of<'tcx>(tcx: TyCtxt<'tcx, 'tcx, 'tcx>, param_env: ty::ParamEnv<'tcx>, ty: Ty<'tcx>) -> Option { tcx.layout_of(param_env.and(ty)).ok().map(|layout| layout.size.bytes()) } /** * Integrator. * * Integrates blocks from the callee function into the calling function. * Updates block indices, references to locals and other control flow * stuff. */ struct Integrator<'a, 'tcx: 'a> { block_idx: usize, args: &'a [Local], local_map: IndexVec, scope_map: IndexVec, promoted_map: IndexVec, _callsite: CallSite<'tcx>, destination: Place<'tcx>, return_block: BasicBlock, cleanup_block: Option, in_cleanup_block: bool, } impl<'a, 'tcx> Integrator<'a, 'tcx> { fn update_target(&self, tgt: BasicBlock) -> BasicBlock { let new = BasicBlock::new(tgt.index() + self.block_idx); debug!("Updating target `{:?}`, new: `{:?}`", tgt, new); new } } impl<'a, 'tcx> MutVisitor<'tcx> for Integrator<'a, 'tcx> { fn visit_local(&mut self, local: &mut Local, _ctxt: PlaceContext, _location: Location) { if *local == RETURN_PLACE { match self.destination { Place::Base(PlaceBase::Local(l)) => { *local = l; return; }, ref place => bug!("Return place is {:?}, not local", place) } } let idx = local.index() - 1; if idx < self.args.len() { *local = self.args[idx]; return; } *local = self.local_map[Local::new(idx - self.args.len())]; } fn visit_place(&mut self, place: &mut Place<'tcx>, _ctxt: PlaceContext, _location: Location) { match place { Place::Base(PlaceBase::Local(RETURN_PLACE)) => { // Return pointer; update the place itself *place = self.destination.clone(); }, Place::Base( PlaceBase::Static(box Static { kind: StaticKind::Promoted(promoted), .. }) ) => { if let Some(p) = self.promoted_map.get(*promoted).cloned() { *promoted = p; } }, _ => self.super_place(place, _ctxt, _location) } } fn visit_basic_block_data(&mut self, block: BasicBlock, data: &mut BasicBlockData<'tcx>) { self.in_cleanup_block = data.is_cleanup; self.super_basic_block_data(block, data); self.in_cleanup_block = false; } fn visit_retag( &mut self, kind: &mut RetagKind, place: &mut Place<'tcx>, loc: Location, ) { self.super_retag(kind, place, loc); // We have to patch all inlined retags to be aware that they are no longer // happening on function entry. if *kind == RetagKind::FnEntry { *kind = RetagKind::Default; } } fn visit_terminator_kind(&mut self, kind: &mut TerminatorKind<'tcx>, loc: Location) { self.super_terminator_kind(kind, loc); match *kind { TerminatorKind::GeneratorDrop | TerminatorKind::Yield { .. } => bug!(), TerminatorKind::Goto { ref mut target} => { *target = self.update_target(*target); } TerminatorKind::SwitchInt { ref mut targets, .. } => { for tgt in targets { *tgt = self.update_target(*tgt); } } TerminatorKind::Drop { ref mut target, ref mut unwind, .. } | TerminatorKind::DropAndReplace { ref mut target, ref mut unwind, .. } => { *target = self.update_target(*target); if let Some(tgt) = *unwind { *unwind = Some(self.update_target(tgt)); } else if !self.in_cleanup_block { // Unless this drop is in a cleanup block, add an unwind edge to // the original call's cleanup block *unwind = self.cleanup_block; } } TerminatorKind::Call { ref mut destination, ref mut cleanup, .. } => { if let Some((_, ref mut tgt)) = *destination { *tgt = self.update_target(*tgt); } if let Some(tgt) = *cleanup { *cleanup = Some(self.update_target(tgt)); } else if !self.in_cleanup_block { // Unless this call is in a cleanup block, add an unwind edge to // the original call's cleanup block *cleanup = self.cleanup_block; } } TerminatorKind::Assert { ref mut target, ref mut cleanup, .. } => { *target = self.update_target(*target); if let Some(tgt) = *cleanup { *cleanup = Some(self.update_target(tgt)); } else if !self.in_cleanup_block { // Unless this assert is in a cleanup block, add an unwind edge to // the original call's cleanup block *cleanup = self.cleanup_block; } } TerminatorKind::Return => { *kind = TerminatorKind::Goto { target: self.return_block }; } TerminatorKind::Resume => { if let Some(tgt) = self.cleanup_block { *kind = TerminatorKind::Goto { target: tgt } } } TerminatorKind::Abort => { } TerminatorKind::Unreachable => { } TerminatorKind::FalseEdges { ref mut real_target, ref mut imaginary_targets } => { *real_target = self.update_target(*real_target); for target in imaginary_targets { *target = self.update_target(*target); } } TerminatorKind::FalseUnwind { real_target: _ , unwind: _ } => // see the ordering of passes in the optimized_mir query. bug!("False unwinds should have been removed before inlining") } } fn visit_source_scope(&mut self, scope: &mut SourceScope) { *scope = self.scope_map[*scope]; } }