|
@@ -15,7 +15,6 @@
|
|
|
#include "llvm/ADT/STLExtras.h"
|
|
#include "llvm/ADT/STLExtras.h"
|
|
|
#include "llvm/ADT/Sequence.h"
|
|
#include "llvm/ADT/Sequence.h"
|
|
|
#include "llvm/Linker/Linker.h"
|
|
#include "llvm/Linker/Linker.h"
|
|
|
-#include "llvm/Support/BLAKE3.h"
|
|
|
|
|
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
|
|
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
|
|
|
#include "llvm/Transforms/Utils/ModuleUtils.h"
|
|
#include "llvm/Transforms/Utils/ModuleUtils.h"
|
|
|
#include "toolchain/base/kind_switch.h"
|
|
#include "toolchain/base/kind_switch.h"
|
|
@@ -23,6 +22,7 @@
|
|
|
#include "toolchain/lower/constant.h"
|
|
#include "toolchain/lower/constant.h"
|
|
|
#include "toolchain/lower/function_context.h"
|
|
#include "toolchain/lower/function_context.h"
|
|
|
#include "toolchain/lower/mangler.h"
|
|
#include "toolchain/lower/mangler.h"
|
|
|
|
|
+#include "toolchain/lower/specific_coalescer.h"
|
|
|
#include "toolchain/sem_ir/absolute_node_id.h"
|
|
#include "toolchain/sem_ir/absolute_node_id.h"
|
|
|
#include "toolchain/sem_ir/diagnostic_loc_converter.h"
|
|
#include "toolchain/sem_ir/diagnostic_loc_converter.h"
|
|
|
#include "toolchain/sem_ir/entry_point.h"
|
|
#include "toolchain/sem_ir/entry_point.h"
|
|
@@ -54,9 +54,7 @@ FileContext::FileContext(Context& context, const SemIR::File& sem_ir,
|
|
|
constants_(LoweredConstantStore::MakeWithExplicitSize(
|
|
constants_(LoweredConstantStore::MakeWithExplicitSize(
|
|
|
sem_ir.insts().size(), nullptr)),
|
|
sem_ir.insts().size(), nullptr)),
|
|
|
lowered_specifics_(sem_ir.generics(), {}),
|
|
lowered_specifics_(sem_ir.generics(), {}),
|
|
|
- lowered_specifics_type_fingerprint_(sem_ir.specifics(), {}),
|
|
|
|
|
- lowered_specific_fingerprint_(sem_ir.specifics(), {}),
|
|
|
|
|
- equivalent_specifics_(sem_ir.specifics(), SemIR::SpecificId::None) {
|
|
|
|
|
|
|
+ coalescer_(vlog_stream_, sem_ir.specifics()) {
|
|
|
// Initialization that relies on invariants of the class.
|
|
// Initialization that relies on invariants of the class.
|
|
|
cpp_code_generator_ = CreateCppCodeGenerator();
|
|
cpp_code_generator_ = CreateCppCodeGenerator();
|
|
|
CARBON_CHECK(!sem_ir.has_errors(),
|
|
CARBON_CHECK(!sem_ir.has_errors(),
|
|
@@ -161,243 +159,8 @@ auto FileContext::Finalize() -> void {
|
|
|
|
|
|
|
|
// Find equivalent specifics (from the same generic), replace all uses and
|
|
// Find equivalent specifics (from the same generic), replace all uses and
|
|
|
// remove duplicately lowered function definitions.
|
|
// remove duplicately lowered function definitions.
|
|
|
- CoalesceEquivalentSpecifics();
|
|
|
|
|
-}
|
|
|
|
|
-
|
|
|
|
|
-auto FileContext::InsertPair(
|
|
|
|
|
- SemIR::SpecificId specific_id1, SemIR::SpecificId specific_id2,
|
|
|
|
|
- Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>& set_of_pairs)
|
|
|
|
|
- -> bool {
|
|
|
|
|
- if (specific_id1.index > specific_id2.index) {
|
|
|
|
|
- std::swap(specific_id1.index, specific_id2.index);
|
|
|
|
|
- }
|
|
|
|
|
- auto insert_result =
|
|
|
|
|
- set_of_pairs.Insert(std::make_pair(specific_id1, specific_id2));
|
|
|
|
|
- return insert_result.is_inserted();
|
|
|
|
|
-}
|
|
|
|
|
-
|
|
|
|
|
-auto FileContext::ContainsPair(
|
|
|
|
|
- SemIR::SpecificId specific_id1, SemIR::SpecificId specific_id2,
|
|
|
|
|
- const Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>& set_of_pairs)
|
|
|
|
|
- -> bool {
|
|
|
|
|
- if (specific_id1.index > specific_id2.index) {
|
|
|
|
|
- std::swap(specific_id1.index, specific_id2.index);
|
|
|
|
|
- }
|
|
|
|
|
- return set_of_pairs.Contains(std::make_pair(specific_id1, specific_id2));
|
|
|
|
|
-}
|
|
|
|
|
-
|
|
|
|
|
-auto FileContext::CoalesceEquivalentSpecifics() -> void {
|
|
|
|
|
- for (auto& specifics : lowered_specifics_.values()) {
|
|
|
|
|
- // Collect specifics to delete for each generic. Replace and remove each
|
|
|
|
|
- // after processing all specifics for a generic. Note, we could also
|
|
|
|
|
- // replace and remove all specifics after processing all generics.
|
|
|
|
|
- llvm::SmallVector<SemIR::SpecificId> specifics_to_delete;
|
|
|
|
|
- // i cannot be unsigned due to the comparison with a negative number when
|
|
|
|
|
- // the specifics vector is empty.
|
|
|
|
|
- for (int i = 0; i < static_cast<int>(specifics.size()) - 1; ++i) {
|
|
|
|
|
- // This specific was already replaced, skip it.
|
|
|
|
|
- if (equivalent_specifics_.Get(specifics[i]).has_value() &&
|
|
|
|
|
- equivalent_specifics_.Get(specifics[i]) != specifics[i]) {
|
|
|
|
|
- specifics_to_delete.push_back(specifics[i]);
|
|
|
|
|
- specifics[i] = specifics[specifics.size() - 1];
|
|
|
|
|
- specifics.pop_back();
|
|
|
|
|
- --i;
|
|
|
|
|
- continue;
|
|
|
|
|
- }
|
|
|
|
|
- // TODO: Improve quadratic behavior by using a single hash based on
|
|
|
|
|
- // `lowered_specifics_type_fingerprint_` and `common_fingerprint`.
|
|
|
|
|
- for (int j = i + 1; j < static_cast<int>(specifics.size()); ++j) {
|
|
|
|
|
- // When the specific was already replaced, skip it.
|
|
|
|
|
- if (equivalent_specifics_.Get(specifics[j]).has_value() &&
|
|
|
|
|
- equivalent_specifics_.Get(specifics[j]) != specifics[j]) {
|
|
|
|
|
- specifics_to_delete.push_back(specifics[j]);
|
|
|
|
|
- specifics[j] = specifics[specifics.size() - 1];
|
|
|
|
|
- specifics.pop_back();
|
|
|
|
|
- --j;
|
|
|
|
|
- continue;
|
|
|
|
|
- }
|
|
|
|
|
-
|
|
|
|
|
- // When the two specifics are not equivalent due to the function type
|
|
|
|
|
- // info stored in lowered_specifics_types, mark non-equivalance. This
|
|
|
|
|
- // can be reused to short-cut another path and continue the search for
|
|
|
|
|
- // other equivalences.
|
|
|
|
|
- if (!AreFunctionTypesEquivalent(specifics[i], specifics[j])) {
|
|
|
|
|
- InsertPair(specifics[i], specifics[j], non_equivalent_specifics_);
|
|
|
|
|
- continue;
|
|
|
|
|
- }
|
|
|
|
|
-
|
|
|
|
|
- Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>
|
|
|
|
|
- visited_equivalent_specifics;
|
|
|
|
|
- InsertPair(specifics[i], specifics[j], visited_equivalent_specifics);
|
|
|
|
|
- // Function type information matches; check usages inside the function
|
|
|
|
|
- // body that are dependent on the specific. This information has been
|
|
|
|
|
- // stored in lowered_states while lowering each function body.
|
|
|
|
|
- if (AreFunctionBodiesEquivalent(specifics[i], specifics[j],
|
|
|
|
|
- visited_equivalent_specifics)) {
|
|
|
|
|
- // When processing equivalences, we may change the canonical specific
|
|
|
|
|
- // multiple times, so we don't delete replaced specifics until the
|
|
|
|
|
- // end.
|
|
|
|
|
- visited_equivalent_specifics.ForEach(
|
|
|
|
|
- [&](std::pair<SemIR::SpecificId, SemIR::SpecificId>
|
|
|
|
|
- equivalent_entry) {
|
|
|
|
|
- CARBON_VLOG("Found equivalent specifics: {0}, {1}",
|
|
|
|
|
- equivalent_entry.first, equivalent_entry.second);
|
|
|
|
|
- ProcessSpecificEquivalence(equivalent_entry);
|
|
|
|
|
- });
|
|
|
|
|
-
|
|
|
|
|
- // Removed the replaced specific from the list of emitted specifics.
|
|
|
|
|
- // Only the top level, since the others are somewhere else in the
|
|
|
|
|
- // vector, they will be found and removed during processing.
|
|
|
|
|
- specifics_to_delete.push_back(specifics[j]);
|
|
|
|
|
- specifics[j] = specifics[specifics.size() - 1];
|
|
|
|
|
- specifics.pop_back();
|
|
|
|
|
- --j;
|
|
|
|
|
- } else {
|
|
|
|
|
- // Only mark non-equivalence based on state for starting specifics.
|
|
|
|
|
- InsertPair(specifics[i], specifics[j], non_equivalent_specifics_);
|
|
|
|
|
- }
|
|
|
|
|
- }
|
|
|
|
|
- }
|
|
|
|
|
-
|
|
|
|
|
- // Once all equivalences are found for a generic, update and delete up
|
|
|
|
|
- // equivalent specifics.
|
|
|
|
|
- for (auto specific_id : specifics_to_delete) {
|
|
|
|
|
- UpdateAndDeleteLLVMFunction(specific_id);
|
|
|
|
|
- }
|
|
|
|
|
- }
|
|
|
|
|
-}
|
|
|
|
|
-
|
|
|
|
|
-auto FileContext::ProcessSpecificEquivalence(
|
|
|
|
|
- std::pair<SemIR::SpecificId, SemIR::SpecificId> pair) -> void {
|
|
|
|
|
- auto [specific_id1, specific_id2] = pair;
|
|
|
|
|
- CARBON_CHECK(specific_id1.has_value() && specific_id2.has_value(),
|
|
|
|
|
- "Expected values in equivalence check");
|
|
|
|
|
-
|
|
|
|
|
- auto get_canon = [&](SemIR::SpecificId specific_id) {
|
|
|
|
|
- auto equiv_id = equivalent_specifics_.Get(specific_id);
|
|
|
|
|
- return equiv_id.has_value() ? equiv_id : specific_id;
|
|
|
|
|
- };
|
|
|
|
|
- auto canon_id1 = get_canon(specific_id1);
|
|
|
|
|
- auto canon_id2 = get_canon(specific_id2);
|
|
|
|
|
-
|
|
|
|
|
- if (canon_id1 == canon_id2) {
|
|
|
|
|
- // Already equivalent, there was a previous replacement.
|
|
|
|
|
- return;
|
|
|
|
|
- }
|
|
|
|
|
-
|
|
|
|
|
- if (canon_id1.index >= canon_id2.index) {
|
|
|
|
|
- // Prefer the earlier index for canonical values.
|
|
|
|
|
- std::swap(canon_id1, canon_id2);
|
|
|
|
|
- }
|
|
|
|
|
-
|
|
|
|
|
- // Update equivalent_specifics_ for all. This is used as an indicator that
|
|
|
|
|
- // this specific_id may be the canonical one when reducing the equivalence
|
|
|
|
|
- // chains in `IsKnownEquivalence`.
|
|
|
|
|
- equivalent_specifics_.Set(specific_id1, canon_id1);
|
|
|
|
|
- equivalent_specifics_.Set(specific_id2, canon_id1);
|
|
|
|
|
- equivalent_specifics_.Set(canon_id1, canon_id1);
|
|
|
|
|
- equivalent_specifics_.Set(canon_id2, canon_id1);
|
|
|
|
|
-}
|
|
|
|
|
-
|
|
|
|
|
-auto FileContext::UpdateEquivalentSpecific(SemIR::SpecificId specific_id)
|
|
|
|
|
- -> void {
|
|
|
|
|
- if (!equivalent_specifics_.Get(specific_id).has_value()) {
|
|
|
|
|
- return;
|
|
|
|
|
- }
|
|
|
|
|
-
|
|
|
|
|
- llvm::SmallVector<SemIR::SpecificId> stack;
|
|
|
|
|
- SemIR::SpecificId specific_to_update = specific_id;
|
|
|
|
|
- SemIR::SpecificId equivalent = equivalent_specifics_.Get(specific_to_update);
|
|
|
|
|
- SemIR::SpecificId equivalent_next = equivalent_specifics_.Get(equivalent);
|
|
|
|
|
- while (equivalent != equivalent_next) {
|
|
|
|
|
- stack.push_back(specific_to_update);
|
|
|
|
|
- specific_to_update = equivalent;
|
|
|
|
|
- equivalent = equivalent_next;
|
|
|
|
|
- equivalent_next = equivalent_specifics_.Get(equivalent_next);
|
|
|
|
|
- }
|
|
|
|
|
-
|
|
|
|
|
- for (auto specific : stack) {
|
|
|
|
|
- equivalent_specifics_.Set(specific, equivalent);
|
|
|
|
|
- }
|
|
|
|
|
-}
|
|
|
|
|
-
|
|
|
|
|
-auto FileContext::UpdateAndDeleteLLVMFunction(SemIR::SpecificId specific_id)
|
|
|
|
|
- -> void {
|
|
|
|
|
- UpdateEquivalentSpecific(specific_id);
|
|
|
|
|
- auto* old_function = specific_functions_.Get(specific_id);
|
|
|
|
|
- auto* new_function =
|
|
|
|
|
- specific_functions_.Get(equivalent_specifics_.Get(specific_id));
|
|
|
|
|
- old_function->replaceAllUsesWith(new_function);
|
|
|
|
|
- old_function->eraseFromParent();
|
|
|
|
|
- specific_functions_.Set(specific_id, new_function);
|
|
|
|
|
-}
|
|
|
|
|
-
|
|
|
|
|
-auto FileContext::IsKnownEquivalence(SemIR::SpecificId specific_id1,
|
|
|
|
|
- SemIR::SpecificId specific_id2) -> bool {
|
|
|
|
|
- if (!equivalent_specifics_.Get(specific_id1).has_value() ||
|
|
|
|
|
- !equivalent_specifics_.Get(specific_id2).has_value()) {
|
|
|
|
|
- return false;
|
|
|
|
|
- }
|
|
|
|
|
-
|
|
|
|
|
- UpdateEquivalentSpecific(specific_id1);
|
|
|
|
|
- UpdateEquivalentSpecific(specific_id2);
|
|
|
|
|
-
|
|
|
|
|
- return equivalent_specifics_.Get(specific_id1) ==
|
|
|
|
|
- equivalent_specifics_.Get(specific_id2);
|
|
|
|
|
-}
|
|
|
|
|
-
|
|
|
|
|
-auto FileContext::AreFunctionTypesEquivalent(SemIR::SpecificId specific_id1,
|
|
|
|
|
- SemIR::SpecificId specific_id2)
|
|
|
|
|
- -> bool {
|
|
|
|
|
- CARBON_CHECK(specific_id1.has_value() && specific_id2.has_value());
|
|
|
|
|
- return lowered_specifics_type_fingerprint_.Get(specific_id1) ==
|
|
|
|
|
- lowered_specifics_type_fingerprint_.Get(specific_id2);
|
|
|
|
|
-}
|
|
|
|
|
-
|
|
|
|
|
-auto FileContext::AreFunctionBodiesEquivalent(
|
|
|
|
|
- SemIR::SpecificId specific_id1, SemIR::SpecificId specific_id2,
|
|
|
|
|
- Set<std::pair<SemIR::SpecificId, SemIR::SpecificId>>&
|
|
|
|
|
- visited_equivalent_specifics) -> bool {
|
|
|
|
|
- llvm::SmallVector<std::pair<SemIR::SpecificId, SemIR::SpecificId>> worklist;
|
|
|
|
|
- worklist.push_back({specific_id1, specific_id2});
|
|
|
|
|
-
|
|
|
|
|
- while (!worklist.empty()) {
|
|
|
|
|
- auto outer_pair = worklist.pop_back_val();
|
|
|
|
|
- auto [specific_id1, specific_id2] = outer_pair;
|
|
|
|
|
-
|
|
|
|
|
- auto state1 = lowered_specific_fingerprint_.Get(specific_id1);
|
|
|
|
|
- auto state2 = lowered_specific_fingerprint_.Get(specific_id2);
|
|
|
|
|
- if (state1.common_fingerprint != state2.common_fingerprint) {
|
|
|
|
|
- InsertPair(specific_id1, specific_id2, non_equivalent_specifics_);
|
|
|
|
|
- return false;
|
|
|
|
|
- }
|
|
|
|
|
- if (state1.specific_fingerprint == state2.specific_fingerprint) {
|
|
|
|
|
- continue;
|
|
|
|
|
- }
|
|
|
|
|
-
|
|
|
|
|
- // A size difference should have been detected by the common fingerprint.
|
|
|
|
|
- CARBON_CHECK(state1.calls.size() == state2.calls.size(),
|
|
|
|
|
- "Number of specific calls expected to be the same.");
|
|
|
|
|
-
|
|
|
|
|
- for (auto [state1_call, state2_call] :
|
|
|
|
|
- llvm::zip(state1.calls, state2.calls)) {
|
|
|
|
|
- if (state1_call != state2_call) {
|
|
|
|
|
- if (ContainsPair(state1_call, state2_call, non_equivalent_specifics_)) {
|
|
|
|
|
- return false;
|
|
|
|
|
- }
|
|
|
|
|
- if (IsKnownEquivalence(state1_call, state2_call)) {
|
|
|
|
|
- continue;
|
|
|
|
|
- }
|
|
|
|
|
- if (!InsertPair(state1_call, state2_call,
|
|
|
|
|
- visited_equivalent_specifics)) {
|
|
|
|
|
- continue;
|
|
|
|
|
- }
|
|
|
|
|
- // Leave the added equivalence pair in place and continue.
|
|
|
|
|
- worklist.push_back({state1_call, state2_call});
|
|
|
|
|
- }
|
|
|
|
|
- }
|
|
|
|
|
- }
|
|
|
|
|
- return true;
|
|
|
|
|
|
|
+ coalescer_.CoalesceEquivalentSpecifics(lowered_specifics_,
|
|
|
|
|
+ specific_functions_);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
auto FileContext::CreateCppCodeGenerator()
|
|
auto FileContext::CreateCppCodeGenerator()
|
|
@@ -634,12 +397,7 @@ auto FileContext::HandleReferencedSpecificFunction(
|
|
|
// For now, we compute the function type fingerprint only for specifics,
|
|
// For now, we compute the function type fingerprint only for specifics,
|
|
|
// though we might need it for all functions in order to create a canonical
|
|
// though we might need it for all functions in order to create a canonical
|
|
|
// fingerprint across translation units.
|
|
// fingerprint across translation units.
|
|
|
- llvm::BLAKE3 function_type_fingerprint;
|
|
|
|
|
- RawStringOstream os;
|
|
|
|
|
- llvm_type->print(os);
|
|
|
|
|
- function_type_fingerprint.update(os.TakeStr());
|
|
|
|
|
- function_type_fingerprint.final(
|
|
|
|
|
- lowered_specifics_type_fingerprint_.Get(specific_id));
|
|
|
|
|
|
|
+ coalescer_.CreateTypeFingerprint(specific_id, llvm_type);
|
|
|
}
|
|
}
|
|
|
|
|
|
|
|
auto FileContext::BuildFunctionDecl(SemIR::FunctionId function_id,
|
|
auto FileContext::BuildFunctionDecl(SemIR::FunctionId function_id,
|
|
@@ -817,7 +575,7 @@ auto FileContext::BuildFunctionBody(SemIR::FunctionId function_id,
|
|
|
|
|
|
|
|
FunctionContext function_lowering(
|
|
FunctionContext function_lowering(
|
|
|
definition_context, llvm_function, *this, specific_id,
|
|
definition_context, llvm_function, *this, specific_id,
|
|
|
- InitializeFingerprintForSpecific(specific_id),
|
|
|
|
|
|
|
+ coalescer_.InitializeFingerprintForSpecific(specific_id),
|
|
|
definition_context.BuildDISubprogram(definition_function, llvm_function),
|
|
definition_context.BuildDISubprogram(definition_function, llvm_function),
|
|
|
vlog_stream_);
|
|
vlog_stream_);
|
|
|
|
|
|