mirror of
https://github.com/carbon-language/carbon-lang.git
synced 2026-09-24 22:02:23 +01:00
This switches `DCHECK` and `FATAL` as well. The goal is to reduce the code size impact of these assertions so that we can keep more of them enabled. Currently, the largest cost I see from `CHECK` is not the actual check or the cold code itself, but actually the failure to inline trivial functions due to the presence of the cold code. This means that our goal isn't to reduce apparent code size in the final binary but the LLVM IR cost assessed for these routines in the inliner, which closely correlates with code size but is a bit different. As discussed in #4283, experimentation shows that a single function call with a minimal number of arguments is the lowest cost model for these. This is easily achieved with a format-string API that internally uses `llvm::formatv`. This PR is essentially the `CHECK` version of #4283. However, the check macros are substantially harder to make work with both format strings and streaming because they also take a condition. Also, unexpectedly, I was very successful at devising a regular expression based automated rewrite from the streaming to the format string form with only low 10s of manual fixes. This includes compacting strings broken up across lines, etc. Given how well that went, I've prepared this PR which just directly switches to the format string API and migrate everything to use it. One nice side-effect is that the format string approach ends up greatly simplifying the implementation here as well. This is ... *shockingly* effective. Parsing speeds up by more than 3% with just this change. And checking speeds up by **8%** with this change alone: ``` BM_CompileAPIFileDenseDecls<Phase::Parse>/256 86.3µs ± 1% 82.9µs ± 1% -3.94% (p=0.000 n=17+19) BM_CompileAPIFileDenseDecls<Phase::Parse>/1024 431µs ± 1% 415µs ± 1% -3.76% (p=0.000 n=18+19) BM_CompileAPIFileDenseDecls<Phase::Parse>/4096 1.77ms ± 1% 1.71ms ± 1% -3.18% (p=0.000 n=18+19) BM_CompileAPIFileDenseDecls<Phase::Parse>/16384 7.44ms ± 1% 7.17ms ± 2% -3.56% (p=0.000 n=18+20) BM_CompileAPIFileDenseDecls<Phase::Parse>/65536 30.7ms ± 1% 29.7ms ± 1% -3.15% (p=0.000 n=18+20) BM_CompileAPIFileDenseDecls<Phase::Parse>/262144 131ms ± 1% 127ms ± 1% -2.81% (p=0.000 n=18+18) BM_CompileAPIFileDenseDecls<Phase::Check>/256 878µs ± 2% 800µs ± 1% -8.91% (p=0.000 n=19+20) BM_CompileAPIFileDenseDecls<Phase::Check>/1024 1.88ms ± 2% 1.72ms ± 1% -8.56% (p=0.000 n=19+20) BM_CompileAPIFileDenseDecls<Phase::Check>/4096 5.78ms ± 2% 5.28ms ± 1% -8.70% (p=0.000 n=20+18) BM_CompileAPIFileDenseDecls<Phase::Check>/16384 21.9ms ± 1% 20.1ms ± 1% -8.02% (p=0.000 n=18+20) BM_CompileAPIFileDenseDecls<Phase::Check>/65536 90.4ms ± 2% 83.1ms ± 1% -8.04% (p=0.000 n=19+20) BM_CompileAPIFileDenseDecls<Phase::Check>/262144 381ms ± 2% 352ms ± 1% -7.79% (p=0.000 n=19+19) ``` --------- Co-authored-by: Richard Smith <richard@metafoo.co.uk> Co-authored-by: josh11b <15258583+josh11b@users.noreply.github.com>
219 lines
8.4 KiB
C++
219 lines
8.4 KiB
C++
// Part of the Carbon Language project, under the Apache License v2.0 with LLVM
|
|
// Exceptions. See /LICENSE for license information.
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
|
|
#include "toolchain/lower/function_context.h"
|
|
|
|
#include "common/vlog.h"
|
|
#include "toolchain/base/kind_switch.h"
|
|
#include "toolchain/sem_ir/file.h"
|
|
|
|
namespace Carbon::Lower {
|
|
|
|
FunctionContext::FunctionContext(FileContext& file_context,
|
|
llvm::Function* function,
|
|
llvm::DISubprogram* di_subprogram,
|
|
llvm::raw_ostream* vlog_stream)
|
|
: file_context_(&file_context),
|
|
function_(function),
|
|
builder_(file_context.llvm_context(), llvm::ConstantFolder(),
|
|
Inserter(file_context.inst_namer())),
|
|
di_subprogram_(di_subprogram),
|
|
vlog_stream_(vlog_stream) {
|
|
function_->setSubprogram(di_subprogram_);
|
|
}
|
|
|
|
auto FunctionContext::GetBlock(SemIR::InstBlockId block_id)
|
|
-> llvm::BasicBlock* {
|
|
auto result = blocks_.Insert(block_id, [&] {
|
|
llvm::StringRef label_name;
|
|
if (const auto* inst_namer = file_context_->inst_namer()) {
|
|
label_name = inst_namer->GetUnscopedLabelFor(block_id);
|
|
}
|
|
return llvm::BasicBlock::Create(llvm_context(), label_name, function_);
|
|
});
|
|
return result.value();
|
|
}
|
|
|
|
auto FunctionContext::TryToReuseBlock(SemIR::InstBlockId block_id,
|
|
llvm::BasicBlock* block) -> bool {
|
|
if (!blocks_.Insert(block_id, block).is_inserted()) {
|
|
return false;
|
|
}
|
|
if (block == synthetic_block_) {
|
|
synthetic_block_ = nullptr;
|
|
}
|
|
if (const auto* inst_namer = file_context_->inst_namer()) {
|
|
block->setName(inst_namer->GetUnscopedLabelFor(block_id));
|
|
}
|
|
return true;
|
|
}
|
|
|
|
auto FunctionContext::LowerBlock(SemIR::InstBlockId block_id) -> void {
|
|
for (auto inst_id : sem_ir().inst_blocks().Get(block_id)) {
|
|
LowerInst(inst_id);
|
|
}
|
|
}
|
|
|
|
// Handles typed instructions for LowerInst. Many instructions lower using
|
|
// HandleInst, but others are unsupported or have trivial lowering.
|
|
//
|
|
// This only calls HandleInst for versions that should have implementations. A
|
|
// different approach would be to have the logic below implemented as HandleInst
|
|
// overloads. However, forward declarations of HandleInst exist for all `InstT`
|
|
// types, which would make getting the right overload resolution complex.
|
|
template <typename InstT>
|
|
static auto LowerInstHelper(FunctionContext& context, SemIR::InstId inst_id,
|
|
InstT inst) {
|
|
if constexpr (!InstT::Kind.is_lowered()) {
|
|
CARBON_FATAL(
|
|
"Encountered an instruction that isn't expected to lower. It's "
|
|
"possible that logic needs to be changed in order to stop showing this "
|
|
"instruction in lowered contexts. Instruction: {0}",
|
|
inst);
|
|
} else if constexpr (InstT::Kind.constant_kind() ==
|
|
SemIR::InstConstantKind::Always) {
|
|
CARBON_FATAL("Missing constant value for constant instruction {0}", inst);
|
|
} else if constexpr (InstT::Kind.is_type() == SemIR::InstIsType::Always) {
|
|
// For instructions that are always of type `type`, produce the trivial
|
|
// runtime representation of type `type`.
|
|
context.SetLocal(inst_id, context.GetTypeAsValue());
|
|
} else {
|
|
HandleInst(context, inst_id, inst);
|
|
}
|
|
}
|
|
|
|
// TODO: Consider renaming Handle##Name, instead relying on typed_inst overload
|
|
// resolution. That would allow putting the nonexistent handler implementations
|
|
// in `requires`-style overloads.
|
|
// NOLINTNEXTLINE(readability-function-size): The define confuses lint.
|
|
auto FunctionContext::LowerInst(SemIR::InstId inst_id) -> void {
|
|
// Skip over constants. `FileContext::GetGlobal` lowers them as needed.
|
|
if (sem_ir().constant_values().Get(inst_id).is_constant()) {
|
|
return;
|
|
}
|
|
|
|
auto inst = sem_ir().insts().Get(inst_id);
|
|
CARBON_VLOG("Lowering {0}: {1}\n", inst_id, inst);
|
|
builder_.getInserter().SetCurrentInstId(inst_id);
|
|
if (di_subprogram_) {
|
|
auto loc = file_context_->GetLocForDI(inst_id);
|
|
CARBON_CHECK(loc.filename == di_subprogram_->getFile()->getFilename(),
|
|
"Instructions located in a different file from their "
|
|
"enclosing function aren't handled yet");
|
|
builder_.SetCurrentDebugLocation(
|
|
llvm::DILocation::get(builder_.getContext(), loc.line_number,
|
|
loc.column_number, di_subprogram_));
|
|
}
|
|
|
|
CARBON_KIND_SWITCH(inst) {
|
|
#define CARBON_SEM_IR_INST_KIND(Name) \
|
|
case CARBON_KIND(SemIR::Name typed_inst): { \
|
|
LowerInstHelper(*this, inst_id, typed_inst); \
|
|
break; \
|
|
}
|
|
#include "toolchain/sem_ir/inst_kind.def"
|
|
}
|
|
|
|
builder_.getInserter().SetCurrentInstId(SemIR::InstId::Invalid);
|
|
if (di_subprogram_) {
|
|
builder_.SetCurrentDebugLocation(llvm::DebugLoc());
|
|
}
|
|
}
|
|
|
|
auto FunctionContext::GetBlockArg(SemIR::InstBlockId block_id,
|
|
SemIR::TypeId type_id) -> llvm::PHINode* {
|
|
llvm::BasicBlock* block = GetBlock(block_id);
|
|
|
|
// Find the existing phi, if any.
|
|
auto phis = block->phis();
|
|
if (!phis.empty()) {
|
|
CARBON_CHECK(std::next(phis.begin()) == phis.end(),
|
|
"Expected at most one phi, found {0}",
|
|
std::distance(phis.begin(), phis.end()));
|
|
return &*phis.begin();
|
|
}
|
|
|
|
// The number of predecessor slots to reserve.
|
|
static constexpr unsigned NumReservedPredecessors = 2;
|
|
auto* phi = llvm::PHINode::Create(GetType(type_id), NumReservedPredecessors);
|
|
phi->insertInto(block, block->begin());
|
|
return phi;
|
|
}
|
|
|
|
auto FunctionContext::MakeSyntheticBlock() -> llvm::BasicBlock* {
|
|
synthetic_block_ = llvm::BasicBlock::Create(llvm_context(), "", function_);
|
|
return synthetic_block_;
|
|
}
|
|
|
|
auto FunctionContext::FinishInit(SemIR::TypeId type_id, SemIR::InstId dest_id,
|
|
SemIR::InstId source_id) -> void {
|
|
switch (SemIR::InitRepr::ForType(sem_ir(), type_id).kind) {
|
|
case SemIR::InitRepr::None:
|
|
break;
|
|
case SemIR::InitRepr::InPlace:
|
|
if (sem_ir().constant_values().Get(source_id).is_constant()) {
|
|
// When initializing from a constant, emission of the source doesn't
|
|
// initialize the destination. Copy the constant value instead.
|
|
CopyValue(type_id, source_id, dest_id);
|
|
}
|
|
break;
|
|
case SemIR::InitRepr::ByCopy:
|
|
CopyValue(type_id, source_id, dest_id);
|
|
break;
|
|
case SemIR::InitRepr::Incomplete:
|
|
CARBON_FATAL("Lowering aggregate initialization of incomplete type {0}",
|
|
sem_ir().types().GetAsInst(type_id));
|
|
}
|
|
}
|
|
|
|
auto FunctionContext::CopyValue(SemIR::TypeId type_id, SemIR::InstId source_id,
|
|
SemIR::InstId dest_id) -> void {
|
|
switch (auto rep = SemIR::ValueRepr::ForType(sem_ir(), type_id); rep.kind) {
|
|
case SemIR::ValueRepr::Unknown:
|
|
CARBON_FATAL("Attempt to copy incomplete type");
|
|
case SemIR::ValueRepr::None:
|
|
break;
|
|
case SemIR::ValueRepr::Copy:
|
|
builder().CreateStore(GetValue(source_id), GetValue(dest_id));
|
|
break;
|
|
case SemIR::ValueRepr::Pointer:
|
|
CopyObject(type_id, source_id, dest_id);
|
|
break;
|
|
case SemIR::ValueRepr::Custom:
|
|
CARBON_FATAL("TODO: Add support for CopyValue with custom value rep");
|
|
}
|
|
}
|
|
|
|
auto FunctionContext::CopyObject(SemIR::TypeId type_id, SemIR::InstId source_id,
|
|
SemIR::InstId dest_id) -> void {
|
|
const auto& layout = llvm_module().getDataLayout();
|
|
auto* type = GetType(type_id);
|
|
// TODO: Compute known alignment of the source and destination, which may
|
|
// be greater than the alignment computed by LLVM.
|
|
auto align = layout.getABITypeAlign(type);
|
|
|
|
// TODO: Attach !tbaa.struct metadata indicating which portions of the
|
|
// type we actually need to copy and which are padding.
|
|
builder().CreateMemCpy(GetValue(dest_id), align, GetValue(source_id), align,
|
|
layout.getTypeAllocSize(type));
|
|
}
|
|
|
|
auto FunctionContext::Inserter::InsertHelper(
|
|
llvm::Instruction* inst, const llvm::Twine& name,
|
|
llvm::BasicBlock::iterator insert_pt) const -> void {
|
|
llvm::StringRef base_name;
|
|
llvm::StringRef separator;
|
|
if (inst_namer_ && !inst->getType()->isVoidTy()) {
|
|
base_name = inst_namer_->GetUnscopedNameFor(inst_id_);
|
|
}
|
|
if (!base_name.empty() && !name.isTriviallyEmpty()) {
|
|
separator = ".";
|
|
}
|
|
|
|
IRBuilderDefaultInserter::InsertHelper(inst, base_name + separator + name,
|
|
insert_pt);
|
|
}
|
|
|
|
} // namespace Carbon::Lower
|