13811884945

Committed 12 Mar 2025 12:55PM UTC coverage: 83.51% (-0.1%) from 83.612%

Build # 13811884945

Build Type

push

github

Committed by

web-flow

Commit Message

Refactor IR type extraction (#28)

Run Details

1007 of 1455 branches covered (69.21%)

Branch coverage included in aggregate %.

250 of 255 new or added lines in 6 files covered. (98.04%)

8 existing lines in 4 files now uncovered.

1905 of 2032 relevant lines covered (93.75%)

3689.69 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

84.75

/lib/type/DITypeExtractor.cpp


//  llvm-dimeta library
//  Copyright (c) 2022-2025 llvm-dimeta authors
//  Distributed under the BSD 3-Clause license.
//  (See accompanying file LICENSE)
//  SPDX-License-Identifier: BSD-3-Clause
//

#include "DIFinder.h"
#include "DIRootType.h"
#include "DataflowAnalysis.h"
#include "DefUseAnalysis.h"
#include "GEP.h"
#include "TBAA.h"
#include "Util.h"
#include "ValuePath.h"
#include "support/Logger.h"

#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/ilist_iterator.h"
#include "llvm/BinaryFormat/Dwarf.h"
#include "llvm/Config/llvm-config.h"
#include "llvm/IR/Argument.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DebugInfoMetadata.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalVariable.h"
#include "llvm/IR/InstIterator.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Operator.h"
#include "llvm/IR/Value.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"

#include <cassert>
#include <iterator>
#include <type_traits>
#include <unordered_map>

namespace dimeta::type {

namespace reset {

using GepToDIMemberMap = std::unordered_map<const llvm::GEPOperator*, llvm::DIDerivedType*>;

namespace detail {

template <typename T, typename InstTy>
std::optional<const T*> get_operand_to(const InstTy* memory_instruction) {
  static_assert(std::is_same_v<InstTy, llvm::StoreInst> || std::is_same_v<InstTy, llvm::LoadInst>,
                "Expected load or store instruction");
  const auto* memory_target = memory_instruction->getPointerOperand();
  if (llvm::isa<T>(memory_target)) {
    return llvm::dyn_cast<T>(memory_target);
  }

  if (auto bcast = llvm::dyn_cast<llvm::BitCastInst>(memory_target)) {
    if (llvm::isa<T>(bcast->getOperand(0))) {
      return llvm::dyn_cast<T>(bcast->getOperand(0));
    }
  }

  return {};
}

}  // namespace detail

template <typename T>
bool store_to(const llvm::StoreInst* store) {
  return detail::get_operand_to<T>(store).has_value();
}

template <typename T>
bool load_to(const llvm::LoadInst* store) {
  return detail::get_operand_to<T>(store).has_value();
}

std::optional<llvm::DIType*> reset_load_related_basic(const dataflow::ValuePath& path, llvm::DIType* type_to_reset,
                                                      const llvm::LoadInst* load) {
  auto type = type_to_reset;

  if (load_to<llvm::GlobalVariable>(load) || load_to<llvm::AllocaInst>(load)) {
    LOG_DEBUG("Do not reset DIType based on load to global,alloca")
    return type;
  }

  if (auto* maybe_ptr_to_type = llvm::dyn_cast<llvm::DIDerivedType>(type)) {
    if ((maybe_ptr_to_type->getTag() == llvm::dwarf::DW_TAG_pointer_type ||
         maybe_ptr_to_type->getTag() == llvm::dwarf::DW_TAG_reference_type)) {
      LOG_DEBUG("Load of pointer-like " << log::ditype_str(maybe_ptr_to_type))
    }

    auto base_type = maybe_ptr_to_type->getBaseType();

    if (auto* composite = llvm::dyn_cast<llvm::DICompositeType>(base_type)) {
      LOG_DEBUG("Have ptr to composite " << log::ditype_str(composite))
      auto type_tbaa = tbaa::resolve_tbaa(base_type, *load);
      if (type_tbaa) {
        return type_tbaa;
      }
    }
    return base_type;
  }

  return type;
}

std::optional<llvm::DIType*> reset_store_related_basic(const dataflow::ValuePath&, llvm::DIType* type_to_reset,
                                                       const llvm::StoreInst* store_inst) {
  auto type = type_to_reset;

  if (store_to<llvm::GlobalVariable>(store_inst) || store_to<llvm::AllocaInst>(store_inst)) {
    // Relevant in "heap_lulesh_mock_char.cpp"
    LOG_DEBUG("Store to alloca/global, return " << log::ditype_str(type))
    return type;
  }

  if (!llvm::isa<llvm::DIDerivedType>(type)) {
    LOG_DEBUG("Store resolved, return " << log::ditype_str(type))
    return type;
  }

  auto* derived_type = llvm::cast<llvm::DIDerivedType>(type);

  if (derived_type->getTag() == llvm::dwarf::DW_TAG_member) {
    auto* member_base               = derived_type->getBaseType();
    const bool is_array_type_member = member_base->getTag() == llvm::dwarf::DW_TAG_array_type;
    // Need to look at base type for array-type member of struct. Tests w.r.t. gep:
    // 1. array_composite.c
    // 2. array_composite_offset_zero.c
    // 3. array_composite_sub_offset_zero.c
    // 4. global_nested.c
    if (is_array_type_member) {
      return llvm::cast<llvm::DICompositeType>(member_base)->getBaseType();
    }
    return member_base;
  }

  const bool is_pointer = derived_type->getTag() == llvm::dwarf::DW_TAG_pointer_type ||
                          derived_type->getTag() == llvm::dwarf::DW_TAG_reference_type;
  if (is_pointer) {
    if (auto* may_be_ptr_to_ptr = llvm::dyn_cast<llvm::DIDerivedType>(derived_type->getBaseType())) {
      // Pointer to pointer by default remove one level for RHS assignment type w.r.t. store:
      const auto is_ptr_to_ptr = may_be_ptr_to_ptr->getTag() == llvm::dwarf::DW_TAG_pointer_type ||
                                 may_be_ptr_to_ptr->getTag() == llvm::dwarf::DW_TAG_reference_type;
      if (is_ptr_to_ptr) {
        LOG_DEBUG("Store to ptr-ptr, return " << log::ditype_str(may_be_ptr_to_ptr))
        return may_be_ptr_to_ptr;
      }
    }

    if (auto* ptr_to_composite = llvm::dyn_cast<llvm::DICompositeType>(derived_type->getBaseType())) {
      if (store_to<llvm::LoadInst>(store_inst)) {
        // Triggers for "heap_lhs_obj_opt.c" (llvm 14/15)
        auto composite_members = ptr_to_composite->getElements();
        assert(!composite_members.empty() && "Store to composite assumed to be store to first member!");
        auto store_di_target = llvm::dyn_cast<llvm::DIDerivedType>(composite_members[0])->getBaseType();
        LOG_DEBUG("Store to a 'load of a composite type', assume first member as target "
                  << log::ditype_str(store_di_target))
        return store_di_target;
      }
    }
  }

  LOG_DEBUG("Store resolved, return " << log::ditype_str(type))
  return type;
}

template <typename Iter, typename Iter2>
std::optional<llvm::DIType*> reset_ditype(llvm::DIType* type_to_reset, const dataflow::ValuePath& path,
                                          const Iter& path_iter, const Iter2&) {
  std::optional<llvm::DIType*> type = type_to_reset;
  if (!type) {
    LOG_DEBUG("No type to reset!")
    return {};
  }

  auto next_value = path_iter;
  LOG_DEBUG("Type to reset: " << log::ditype_str(*type));
  LOG_DEBUG(">> based on IR: " << **next_value);

  if (const auto* load = llvm::dyn_cast<llvm::LoadInst>(*next_value)) {
    // Re-set the DIType from the gep, if presence of:
    // - a load after a gep is likely the first element of the composite type
    // - a load also resolves to the basetype w.r.t. an array composite
    LOG_DEBUG("Reset based on load")
    return reset::reset_load_related_basic(path, type.value(), load);
  }

  if (const auto* store_inst = llvm::dyn_cast<llvm::StoreInst>(*next_value)) {
    // - a store with a ditype(array) is likely the first element of the array
    LOG_DEBUG("Reset based on store")
    return reset::reset_store_related_basic(path, type.value(), store_inst);
  }

  LOG_DEBUG(">> skipping");

  return type;
}

}  // namespace reset

std::optional<llvm::DIType*> find_type(const dataflow::CallValuePath& call_path) {
  auto type = root::find_type_root(call_path);

  if (!type) {
    LOG_DEBUG("find_type_root failed to find a type for path " << call_path.path)
    return {};
  }

  const auto path_end = call_path.path.path_to_value.rend();
  for (auto path_iter = call_path.path.path_to_value.rbegin(); path_iter != path_end; ++path_iter) {
    if (!type) {
      break;
    }
    if (llvm::isa<llvm::GEPOperator>(*path_iter)) {
      auto* gep = llvm::cast<llvm::GEPOperator>(*path_iter);
      LOG_DEBUG("Path iter gep for extraction is currently " << *gep);
      // TODO: Maybe we could somehow get more info on the underlying type from the dataflow path
      //       if this returns an empty result due to forward decls?
      const auto gep_result = gep::extract_gep_dereferenced_type(type.value(), *gep);
      type                  = gep_result.type;
      if (gep_result.member) {
        LOG_DEBUG("Using gep member type result")
        type = gep_result.member;
      }
      LOG_DEBUG("Gep reset type is " << log::ditype_str(type.value_or(nullptr)) << "\n")
      continue;
    }
    LOG_DEBUG("Extracted type w.r.t. gep: " << log::ditype_str(*type));
    type = reset::reset_ditype(type.value(), call_path.path, path_iter, path_end).value_or(type.value());
    LOG_DEBUG("reset_ditype result " << log::ditype_str(type.value_or(nullptr)) << "\n")
  }

  if (type) {
    // If last node is a store inst, try to extract type via TBAA
    const auto* const start_node = llvm::dyn_cast_or_null<llvm::StoreInst>(*call_path.path.start_value());
    if (start_node) {
      auto type_tbaa = tbaa::resolve_tbaa(type.value(), *llvm::dyn_cast<llvm::Instruction>(start_node));
      if (type_tbaa) {
        type = type_tbaa.value();
      }
    }
  }

  return type;
}
}  // namespace dimeta::type

1
2	// llvm-dimeta library
3	// Copyright (c) 2022-2025 llvm-dimeta authors
4	// Distributed under the BSD 3-Clause license.
5	// (See accompanying file LICENSE)
6	// SPDX-License-Identifier: BSD-3-Clause
7	//
8
9	#include "DIFinder.h"
10	#include "DIRootType.h"
11	#include "DataflowAnalysis.h"
12	#include "DefUseAnalysis.h"
13	#include "GEP.h"
14	#include "TBAA.h"
15	#include "Util.h"
16	#include "ValuePath.h"
17	#include "support/Logger.h"
18
19	#include "llvm/ADT/ArrayRef.h"
20	#include "llvm/ADT/STLExtras.h"
21	#include "llvm/ADT/SmallVector.h"
22	#include "llvm/ADT/ilist_iterator.h"
23	#include "llvm/BinaryFormat/Dwarf.h"
24	#include "llvm/Config/llvm-config.h"
25	#include "llvm/IR/Argument.h"
26	#include "llvm/IR/Constants.h"
27	#include "llvm/IR/DebugInfoMetadata.h"
28	#include "llvm/IR/Function.h"
29	#include "llvm/IR/GlobalVariable.h"
30	#include "llvm/IR/InstIterator.h"
31	#include "llvm/IR/InstrTypes.h"
32	#include "llvm/IR/Instructions.h"
33	#include "llvm/IR/IntrinsicInst.h"
34	#include "llvm/IR/Metadata.h"
35	#include "llvm/IR/Operator.h"
36	#include "llvm/IR/Value.h"
37	#include "llvm/Support/Casting.h"
38	#include "llvm/Support/Debug.h"
39	#include "llvm/Support/ErrorHandling.h"
40	#include "llvm/Support/raw_ostream.h"
41
42	#include <cassert>
43	#include <iterator>
44	#include <type_traits>
45	#include <unordered_map>
46
47	namespace dimeta::type {
48
49	namespace reset {
50
51	using GepToDIMemberMap = std::unordered_map<const llvm::GEPOperator, llvm::DIDerivedType>;
52
53	namespace detail {
54
55	template <typename T, typename InstTy>
56	std::optional<const T> get_operand_to(const InstTy memory_instruction) {	2,900✔
57	static_assert(std::is_same_v<InstTy, llvm::StoreInst> \|\| std::is_same_v<InstTy, llvm::LoadInst>,
58	"Expected load or store instruction");
59	const auto* memory_target = memory_instruction->getPointerOperand();	2,900✔
60	if (llvm::isa<T>(memory_target)) {	2,900!
61	return llvm::dyn_cast<T>(memory_target);	494✔
62	}
63
64	if (auto bcast = llvm::dyn_cast<llvm::BitCastInst>(memory_target)) {	2,618!
65	if (llvm::isa<T>(bcast->getOperand(0))) {	212!
UNCOV 66	return llvm::dyn_cast<T>(bcast->getOperand(0));	×
67	}
68	}	104✔
69
70	return {};	2,406✔
71	}	2,900✔
72
73	} // namespace detail
74
75	template <typename T>
76	bool store_to(const llvm::StoreInst* store) {	1,790✔
77	return detail::get_operand_to<T>(store).has_value();	1,790✔
78	}
79
80	template <typename T>
81	bool load_to(const llvm::LoadInst* store) {	1,110✔
82	return detail::get_operand_to<T>(store).has_value();	1,110✔
83	}
84
85	std::optional<llvm::DIType> reset_load_related_basic(const dataflow::ValuePath& path, llvm::DIType type_to_reset,	600✔
86	const llvm::LoadInst* load) {
87	auto type = type_to_reset;	600✔
88
89	if (load_to<llvm::GlobalVariable>(load) \|\| load_to<llvm::AllocaInst>(load)) {	600✔
90	LOG_DEBUG("Do not reset DIType based on load to global,alloca")
91	return type;	362✔
92	}
93
94	if (auto* maybe_ptr_to_type = llvm::dyn_cast<llvm::DIDerivedType>(type)) {	434!
95	if ((maybe_ptr_to_type->getTag() == llvm::dwarf::DW_TAG_pointer_type \|\|	238✔
96	maybe_ptr_to_type->getTag() == llvm::dwarf::DW_TAG_reference_type)) {	62✔
97	LOG_DEBUG("Load of pointer-like " << log::ditype_str(maybe_ptr_to_type))
98	}	154✔
99
100	auto base_type = maybe_ptr_to_type->getBaseType();	238✔
101
102	if (auto* composite = llvm::dyn_cast<llvm::DICompositeType>(base_type)) {	296✔
103	LOG_DEBUG("Have ptr to composite " << log::ditype_str(composite))
104	auto type_tbaa = tbaa::resolve_tbaa(base_type, *load);	70✔
105	if (type_tbaa) {	70!
106	return type_tbaa;	70✔
107	}
108	}
109	return base_type;	168✔
110	}	238✔
111
UNCOV 112	return type;	×
113	}	600✔
114
115	std::optional<llvm::DIType> reset_store_related_basic(const dataflow::ValuePath&, llvm::DIType type_to_reset,	876✔
116	const llvm::StoreInst* store_inst) {
117	auto type = type_to_reset;	876✔
118
119	if (store_to<llvm::GlobalVariable>(store_inst) \|\| store_to<llvm::AllocaInst>(store_inst)) {	876!
120	// Relevant in "heap_lulesh_mock_char.cpp"
121	LOG_DEBUG("Store to alloca/global, return " << log::ditype_str(type))
122	return type;	124✔
123	}
124
125	if (!llvm::isa<llvm::DIDerivedType>(type)) {	752✔
126	LOG_DEBUG("Store resolved, return " << log::ditype_str(type))
127	return type;	56✔
128	}
129
130	auto* derived_type = llvm::cast<llvm::DIDerivedType>(type);	696✔
131
132	if (derived_type->getTag() == llvm::dwarf::DW_TAG_member) {	696✔
133	auto* member_base = derived_type->getBaseType();	304✔
134	const bool is_array_type_member = member_base->getTag() == llvm::dwarf::DW_TAG_array_type;	304✔
135	// Need to look at base type for array-type member of struct. Tests w.r.t. gep:
136	// 1. array_composite.c
137	// 2. array_composite_offset_zero.c
138	// 3. array_composite_sub_offset_zero.c
139	// 4. global_nested.c
140	if (is_array_type_member) {	304✔
141	return llvm::cast<llvm::DICompositeType>(member_base)->getBaseType();	98✔
142	}
143	return member_base;	206✔
144	}	304✔
145
146	const bool is_pointer = derived_type->getTag() == llvm::dwarf::DW_TAG_pointer_type \|\|	392✔
147	derived_type->getTag() == llvm::dwarf::DW_TAG_reference_type;	32✔
148	if (is_pointer) {	392✔
149	if (auto* may_be_ptr_to_ptr = llvm::dyn_cast<llvm::DIDerivedType>(derived_type->getBaseType())) {	680✔
150	// Pointer to pointer by default remove one level for RHS assignment type w.r.t. store:
151	const auto is_ptr_to_ptr = may_be_ptr_to_ptr->getTag() == llvm::dwarf::DW_TAG_pointer_type \|\|	308✔
152	may_be_ptr_to_ptr->getTag() == llvm::dwarf::DW_TAG_reference_type;	20✔
153	if (is_ptr_to_ptr) {	308✔
154	LOG_DEBUG("Store to ptr-ptr, return " << log::ditype_str(may_be_ptr_to_ptr))
155	return may_be_ptr_to_ptr;	288✔
156	}
157	}	246✔
158
159	if (auto* ptr_to_composite = llvm::dyn_cast<llvm::DICompositeType>(derived_type->getBaseType())) {	122✔
160	if (store_to<llvm::LoadInst>(store_inst)) {	38✔
161	// Triggers for "heap_lhs_obj_opt.c" (llvm 14/15)
162	auto composite_members = ptr_to_composite->getElements();	8✔
163	assert(!composite_members.empty() && "Store to composite assumed to be store to first member!");	16!
164	auto store_di_target = llvm::dyn_cast<llvm::DIDerivedType>(composite_members[0])->getBaseType();	8✔
165	LOG_DEBUG("Store to a 'load of a composite type', assume first member as target "
166	<< log::ditype_str(store_di_target))
167	return store_di_target;	8✔
168	}	8✔
169	}	24✔
170	}	62✔
171
172	LOG_DEBUG("Store resolved, return " << log::ditype_str(type))
173	return type;	96✔
174	}	876✔
175
176	template <typename Iter, typename Iter2>
177	std::optional<llvm::DIType> reset_ditype(llvm::DIType type_to_reset, const dataflow::ValuePath& path,	2,830✔
178	const Iter& path_iter, const Iter2&) {
179	std::optional<llvm::DIType*> type = type_to_reset;	2,830✔
180	if (!type) {	2,830!
181	LOG_DEBUG("No type to reset!")
182	return {};	×
183	}
184
185	auto next_value = path_iter;	2,830✔
186	LOG_DEBUG("Type to reset: " << log::ditype_str(*type));
187	LOG_DEBUG(">> based on IR: " << **next_value);
188
189	if (const auto* load = llvm::dyn_cast<llvm::LoadInst>(*next_value)) {	3,316✔
190	// Re-set the DIType from the gep, if presence of:
191	// - a load after a gep is likely the first element of the composite type
192	// - a load also resolves to the basetype w.r.t. an array composite
193	LOG_DEBUG("Reset based on load")
194	return reset::reset_load_related_basic(path, type.value(), load);	600✔
195	}
196
197	if (const auto* store_inst = llvm::dyn_cast<llvm::StoreInst>(*next_value)) {	2,932✔
198	// - a store with a ditype(array) is likely the first element of the array
199	LOG_DEBUG("Reset based on store")
200	return reset::reset_store_related_basic(path, type.value(), store_inst);	876✔
201	}
202
203	LOG_DEBUG(">> skipping");
204
205	return type;	1,354✔
206	}	2,830✔
207
208	} // namespace reset
209
210	std::optional<llvm::DIType*> find_type(const dataflow::CallValuePath& call_path) {	1,036✔
211	auto type = root::find_type_root(call_path);	1,036✔
212
213	if (!type) {	1,036✔
214	LOG_DEBUG("find_type_root failed to find a type for path " << call_path.path)
215	return {};	10✔
216	}
217
218	const auto path_end = call_path.path.path_to_value.rend();	1,026✔
219	for (auto path_iter = call_path.path.path_to_value.rbegin(); path_iter != path_end; ++path_iter) {	4,750✔
220	if (!type) {	3,724!
221	break;	×
222	}
223	if (llvm::isa<llvm::GEPOperator>(*path_iter)) {	3,724✔
224	auto* gep = llvm::cast<llvm::GEPOperator>(*path_iter);	894✔
225	LOG_DEBUG("Path iter gep for extraction is currently " << *gep);
226	// TODO: Maybe we could somehow get more info on the underlying type from the dataflow path
227	// if this returns an empty result due to forward decls?
228	const auto gep_result = gep::extract_gep_dereferenced_type(type.value(), *gep);	894✔
229	type = gep_result.type;	894✔
230	if (gep_result.member) {	894✔
231	LOG_DEBUG("Using gep member type result")
232	type = gep_result.member;	542✔
233	}	542✔
234	LOG_DEBUG("Gep reset type is " << log::ditype_str(type.value_or(nullptr)) << "\n")
235	continue;
236	}	894✔
237	LOG_DEBUG("Extracted type w.r.t. gep: " << log::ditype_str(*type));
238	type = reset::reset_ditype(type.value(), call_path.path, path_iter, path_end).value_or(type.value());	2,830✔
239	LOG_DEBUG("reset_ditype result " << log::ditype_str(type.value_or(nullptr)) << "\n")
240	}	2,830✔
241
242	if (type) {	1,026!
243	// If last node is a store inst, try to extract type via TBAA
244	const auto* const start_node = llvm::dyn_cast_or_null<llvm::StoreInst>(*call_path.path.start_value());	1,026✔
245	if (start_node) {	1,026✔
246	auto type_tbaa = tbaa::resolve_tbaa(type.value(), *llvm::dyn_cast<llvm::Instruction>(start_node));	876✔
247	if (type_tbaa) {	876✔
248	type = type_tbaa.value();	856✔
249	}	856✔
250	}	876✔
251	}	1,026✔
252
253	return type;	1,026✔
254	}	1,036✔
255	} // namespace dimeta::type

ahueck / llvm-dimeta / 13811884945

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous