kou commented on code in PR #38116:
URL: https://github.com/apache/arrow/pull/38116#discussion_r1379611177
##########
cpp/src/gandiva/configuration.cc:
##########
@@ -29,11 +29,15 @@ std::size_t Configuration::Hash() const {
size_t result = kHashSeed;
arrow::internal::hash_combine(result, static_cast<size_t>(optimize_));
arrow::internal::hash_combine(result, static_cast<size_t>(target_host_cpu_));
+ arrow::internal::hash_combine(
+ result, reinterpret_cast<std::uintptr_t>(function_registry_.get()));
+
Review Comment:
```suggestion
```
##########
cpp/src/gandiva/function_registry.cc:
##########
@@ -41,42 +62,80 @@ FunctionRegistry::iterator FunctionRegistry::back() const {
return &(pc_registry_.back());
}
-std::vector<NativeFunction> FunctionRegistry::pc_registry_;
+const NativeFunction* FunctionRegistry::LookupSignature(
+ const FunctionSignature& signature) const {
+ auto got = pc_registry_map_.find(&signature);
+ return got == pc_registry_map_.end() ? nullptr : got->second;
+}
+
+Status FunctionRegistry::Add(NativeFunction func) {
+ if (pc_registry_.size() == kMaxFunctionSignatures) {
+ return Status::CapacityError("Exceeded max function signatures limit of ",
+ kMaxFunctionSignatures);
+ }
+ pc_registry_.emplace_back(std::move(func));
+ auto const& last_func = pc_registry_.back();
+ for (auto const& func_signature : last_func.signatures()) {
+ pc_registry_map_.emplace(&func_signature, &last_func);
+ }
+ return arrow::Status::OK();
+}
-SignatureMap FunctionRegistry::pc_registry_map_ = InitPCMap();
+arrow::Result<std::unique_ptr<llvm::MemoryBuffer>> GetBufferFromFile(
+ const std::string& bitcode_file_path) {
+ auto buffer_or_error = llvm::MemoryBuffer::getFile(bitcode_file_path);
-SignatureMap FunctionRegistry::InitPCMap() {
- SignatureMap map;
+ ARROW_RETURN_IF(!buffer_or_error,
+ Status::IOError("Could not load module from bitcode file: ",
+ bitcode_file_path +
+ " Error: " +
buffer_or_error.getError().message()));
- auto v1 = GetArithmeticFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v1.begin(), v1.end());
- auto v2 = GetDateTimeFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v2.begin(), v2.end());
+ auto buffer = std::move(buffer_or_error.get());
+ return std::move(buffer);
+}
- auto v3 = GetHashFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v3.begin(), v3.end());
+Status FunctionRegistry::Register(const std::vector<NativeFunction>& funcs,
+ const std::string& bitcode_path) {
+ ARROW_ASSIGN_OR_RAISE(auto llvm_buffer, GetBufferFromFile(bitcode_path));
+ auto buffer =
std::make_unique<LLVMMemoryArrowBuffer>(std::move(llvm_buffer));
+ return Register(funcs, std::move(buffer));
+}
- auto v4 = GetMathOpsFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v4.begin(), v4.end());
+arrow::Status FunctionRegistry::Register(const std::vector<NativeFunction>&
funcs,
+ std::shared_ptr<arrow::Buffer>
bitcode_buffer) {
+ bitcode_memory_buffers_.emplace_back(std::move(bitcode_buffer));
+ for (const auto& func : funcs) {
+ ARROW_RETURN_NOT_OK(FunctionRegistry::Add(func));
+ }
+ return Status::OK();
+}
- auto v5 = GetStringFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v5.begin(), v5.end());
+const std::vector<std::shared_ptr<arrow::Buffer>>&
FunctionRegistry::GetBitcodeBuffers()
+ const {
+ return bitcode_memory_buffers_;
+}
- auto v6 = GetDateTimeArithmeticFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v6.begin(), v6.end());
- for (auto& elem : pc_registry_) {
- for (auto& func_signature : elem.signatures()) {
- map.insert(std::make_pair(&(func_signature), &elem));
+arrow::Result<std::shared_ptr<FunctionRegistry>> MakeDefaultFunctionRegistry()
{
+ auto registry = std::make_shared<FunctionRegistry>();
+ for (auto const& funcs :
+ {GetArithmeticFunctionRegistry(), GetDateTimeFunctionRegistry(),
+ GetHashFunctionRegistry(), GetMathOpsFunctionRegistry(),
+ GetStringFunctionRegistry(), GetDateTimeArithmeticFunctionRegistry()})
{
+ for (auto const& func_signature : funcs) {
+ ARROW_RETURN_NOT_OK(registry->Add(func_signature));
}
}
-
- return map;
+ return std::move(registry);
}
-const NativeFunction* FunctionRegistry::LookupSignature(
- const FunctionSignature& signature) const {
- auto got = pc_registry_map_.find(&signature);
- return got == pc_registry_map_.end() ? nullptr : got->second;
+std::shared_ptr<FunctionRegistry> default_function_registry() {
+ static auto maybe_default_registry = MakeDefaultFunctionRegistry();
+ if (!maybe_default_registry.ok()) {
+ ARROW_LOG(FATAL) << "Failed to initialize default function registry: "
+ << maybe_default_registry.status().message();
+ return nullptr;
+ }
Review Comment:
Can we avoid this check for all `default_function_registry()` calls?
##########
cpp/src/gandiva/function_registry.cc:
##########
@@ -41,42 +62,80 @@ FunctionRegistry::iterator FunctionRegistry::back() const {
return &(pc_registry_.back());
}
-std::vector<NativeFunction> FunctionRegistry::pc_registry_;
+const NativeFunction* FunctionRegistry::LookupSignature(
+ const FunctionSignature& signature) const {
+ auto got = pc_registry_map_.find(&signature);
+ return got == pc_registry_map_.end() ? nullptr : got->second;
+}
+
+Status FunctionRegistry::Add(NativeFunction func) {
+ if (pc_registry_.size() == kMaxFunctionSignatures) {
+ return Status::CapacityError("Exceeded max function signatures limit of ",
+ kMaxFunctionSignatures);
+ }
+ pc_registry_.emplace_back(std::move(func));
+ auto const& last_func = pc_registry_.back();
+ for (auto const& func_signature : last_func.signatures()) {
+ pc_registry_map_.emplace(&func_signature, &last_func);
+ }
+ return arrow::Status::OK();
+}
-SignatureMap FunctionRegistry::pc_registry_map_ = InitPCMap();
+arrow::Result<std::unique_ptr<llvm::MemoryBuffer>> GetBufferFromFile(
+ const std::string& bitcode_file_path) {
+ auto buffer_or_error = llvm::MemoryBuffer::getFile(bitcode_file_path);
-SignatureMap FunctionRegistry::InitPCMap() {
- SignatureMap map;
+ ARROW_RETURN_IF(!buffer_or_error,
+ Status::IOError("Could not load module from bitcode file: ",
+ bitcode_file_path +
+ " Error: " +
buffer_or_error.getError().message()));
- auto v1 = GetArithmeticFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v1.begin(), v1.end());
- auto v2 = GetDateTimeFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v2.begin(), v2.end());
+ auto buffer = std::move(buffer_or_error.get());
+ return std::move(buffer);
+}
- auto v3 = GetHashFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v3.begin(), v3.end());
+Status FunctionRegistry::Register(const std::vector<NativeFunction>& funcs,
+ const std::string& bitcode_path) {
+ ARROW_ASSIGN_OR_RAISE(auto llvm_buffer, GetBufferFromFile(bitcode_path));
+ auto buffer =
std::make_unique<LLVMMemoryArrowBuffer>(std::move(llvm_buffer));
Review Comment:
```suggestion
auto buffer =
std::make_shared<LLVMMemoryArrowBuffer>(std::move(llvm_buffer));
```
##########
cpp/src/gandiva/function_registry.cc:
##########
@@ -41,42 +62,80 @@ FunctionRegistry::iterator FunctionRegistry::back() const {
return &(pc_registry_.back());
}
-std::vector<NativeFunction> FunctionRegistry::pc_registry_;
+const NativeFunction* FunctionRegistry::LookupSignature(
+ const FunctionSignature& signature) const {
+ auto got = pc_registry_map_.find(&signature);
+ return got == pc_registry_map_.end() ? nullptr : got->second;
+}
+
+Status FunctionRegistry::Add(NativeFunction func) {
+ if (pc_registry_.size() == kMaxFunctionSignatures) {
+ return Status::CapacityError("Exceeded max function signatures limit of ",
+ kMaxFunctionSignatures);
+ }
+ pc_registry_.emplace_back(std::move(func));
+ auto const& last_func = pc_registry_.back();
+ for (auto const& func_signature : last_func.signatures()) {
+ pc_registry_map_.emplace(&func_signature, &last_func);
+ }
+ return arrow::Status::OK();
+}
-SignatureMap FunctionRegistry::pc_registry_map_ = InitPCMap();
+arrow::Result<std::unique_ptr<llvm::MemoryBuffer>> GetBufferFromFile(
+ const std::string& bitcode_file_path) {
+ auto buffer_or_error = llvm::MemoryBuffer::getFile(bitcode_file_path);
-SignatureMap FunctionRegistry::InitPCMap() {
- SignatureMap map;
+ ARROW_RETURN_IF(!buffer_or_error,
+ Status::IOError("Could not load module from bitcode file: ",
+ bitcode_file_path +
+ " Error: " +
buffer_or_error.getError().message()));
- auto v1 = GetArithmeticFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v1.begin(), v1.end());
- auto v2 = GetDateTimeFunctionRegistry();
- pc_registry_.insert(std::end(pc_registry_), v2.begin(), v2.end());
+ auto buffer = std::move(buffer_or_error.get());
+ return std::move(buffer);
Review Comment:
```suggestion
return buffer_or_error.get();
```
##########
cpp/src/gandiva/tree_expr_test.cc:
##########
@@ -45,7 +45,7 @@ class TestExprTree : public ::testing::Test {
FieldPtr i1_; // int32
FieldPtr b0_; // bool
- FunctionRegistry registry_;
+ FunctionRegistry& registry_ = *gandiva::default_function_registry();
Review Comment:
Can we use `std::shared_ptr<FunctionRegistry>` here too?
##########
cpp/src/gandiva/llvm_generator.cc:
##########
@@ -36,11 +36,16 @@ namespace gandiva {
AddTrace(__VA_ARGS__); \
}
-LLVMGenerator::LLVMGenerator(bool cached) : cached_(cached),
enable_ir_traces_(false) {}
+LLVMGenerator::LLVMGenerator(bool cached,
+ std::shared_ptr<FunctionRegistry>
function_registry)
+ : cached_(cached),
+ function_registry_(std::move(function_registry)),
+ enable_ir_traces_(false) {}
-Status LLVMGenerator::Make(std::shared_ptr<Configuration> config, bool cached,
+Status LLVMGenerator::Make(const std::shared_ptr<Configuration>& config, bool
cached,
std::unique_ptr<LLVMGenerator>* llvm_generator) {
- std::unique_ptr<LLVMGenerator> llvmgen_obj(new LLVMGenerator(cached));
+ std::unique_ptr<LLVMGenerator> llvmgen_obj(
+ new LLVMGenerator(cached, config->function_registry()));
Review Comment:
```suggestion
auto llvmgen_obj = std::make_unique<LLVMGenerator>(cached,
config->function_registry());
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]