Implement CFI for Optimizing.

CFI is necessary for stack unwinding in gdb, lldb, and libunwind.

Change-Id: I1a3480e3a4a99f48bf7e6e63c4e83a80cfee40a2

19 files changed, 624 insertions, 284 deletions

diff --git a/build/Android.gtest.mk b/build/Android.gtest.mk
index 62a11c24e9..d9d09bcc63 100644
--- a/build/Android.gtest.mk
+++ b/build/Android.gtest.mk
@@ -214,6 +214,7 @@ COMPILER_GTEST_COMMON_SRC_FILES := \
   compiler/optimizing/live_interval_test.cc \
   compiler/optimizing/live_ranges_test.cc \
   compiler/optimizing/nodes_test.cc \
+  compiler/optimizing/optimizing_cfi_test.cc \
   compiler/optimizing/parallel_move_test.cc \
   compiler/optimizing/pretty_printer_test.cc \
   compiler/optimizing/register_allocator_test.cc \
diff --git a/compiler/compiled_method.cc b/compiler/compiled_method.cc
index eeed877444..4f7a970fdd 100644
--- a/compiler/compiled_method.cc
+++ b/compiler/compiled_method.cc
@@ -188,39 +188,6 @@ CompiledMethod* CompiledMethod::SwapAllocCompiledMethod(
   return ret;
 }
 
-CompiledMethod* CompiledMethod::SwapAllocCompiledMethodStackMap(
-    CompilerDriver* driver,
-    InstructionSet instruction_set,
-    const ArrayRef<const uint8_t>& quick_code,
-    const size_t frame_size_in_bytes,
-    const uint32_t core_spill_mask,
-    const uint32_t fp_spill_mask,
-    const ArrayRef<const uint8_t>& stack_map) {
-  SwapAllocator<CompiledMethod> alloc(driver->GetSwapSpaceAllocator());
-  CompiledMethod* ret = alloc.allocate(1);
-  alloc.construct(ret, driver, instruction_set, quick_code, frame_size_in_bytes, core_spill_mask,
-                  fp_spill_mask, nullptr, ArrayRef<const uint8_t>(), stack_map,
-                  ArrayRef<const uint8_t>(), ArrayRef<const uint8_t>(),
-                  ArrayRef<const LinkerPatch>());
-  return ret;
-}
-
-CompiledMethod* CompiledMethod::SwapAllocCompiledMethodCFI(
-    CompilerDriver* driver,
-    InstructionSet instruction_set,
-    const ArrayRef<const uint8_t>& quick_code,
-    const size_t frame_size_in_bytes,
-    const uint32_t core_spill_mask,
-    const uint32_t fp_spill_mask,
-    const ArrayRef<const uint8_t>& cfi_info) {
-  SwapAllocator<CompiledMethod> alloc(driver->GetSwapSpaceAllocator());
-  CompiledMethod* ret = alloc.allocate(1);
-  alloc.construct(ret, driver, instruction_set, quick_code, frame_size_in_bytes, core_spill_mask,
-                  fp_spill_mask, nullptr, ArrayRef<const uint8_t>(),
-                  ArrayRef<const uint8_t>(), ArrayRef<const uint8_t>(),
-                  cfi_info, ArrayRef<const LinkerPatch>());
-  return ret;
-}
 
 
 void CompiledMethod::ReleaseSwapAllocatedCompiledMethod(CompilerDriver* driver, CompiledMethod* m) {
diff --git a/compiler/compiled_method.h b/compiler/compiled_method.h
index 506b47b68a..480d021db0 100644
--- a/compiler/compiled_method.h
+++ b/compiler/compiled_method.h
@@ -320,7 +320,7 @@ class CompiledMethod FINAL : public CompiledCode {
                  const ArrayRef<const uint8_t>& vmap_table,
                  const ArrayRef<const uint8_t>& native_gc_map,
                  const ArrayRef<const uint8_t>& cfi_info,
-                 const ArrayRef<const LinkerPatch>& patches = ArrayRef<const LinkerPatch>());
+                 const ArrayRef<const LinkerPatch>& patches);
 
   virtual ~CompiledMethod();
 
@@ -336,24 +336,7 @@ class CompiledMethod FINAL : public CompiledCode {
       const ArrayRef<const uint8_t>& vmap_table,
       const ArrayRef<const uint8_t>& native_gc_map,
       const ArrayRef<const uint8_t>& cfi_info,
-      const ArrayRef<const LinkerPatch>& patches = ArrayRef<const LinkerPatch>());
-
-  static CompiledMethod* SwapAllocCompiledMethodStackMap(
-      CompilerDriver* driver,
-      InstructionSet instruction_set,
-      const ArrayRef<const uint8_t>& quick_code,
-      const size_t frame_size_in_bytes,
-      const uint32_t core_spill_mask,
-      const uint32_t fp_spill_mask,
-      const ArrayRef<const uint8_t>& stack_map);
-
-  static CompiledMethod* SwapAllocCompiledMethodCFI(CompilerDriver* driver,
-                                                    InstructionSet instruction_set,
-                                                    const ArrayRef<const uint8_t>& quick_code,
-                                                    const size_t frame_size_in_bytes,
-                                                    const uint32_t core_spill_mask,
-                                                    const uint32_t fp_spill_mask,
-                                                    const ArrayRef<const uint8_t>& cfi_info);
+      const ArrayRef<const LinkerPatch>& patches);
 
   static void ReleaseSwapAllocatedCompiledMethod(CompilerDriver* driver, CompiledMethod* m);
 
diff --git a/compiler/dex/quick/codegen_util.cc b/compiler/dex/quick/codegen_util.cc
index ff5f735255..9c9205fbb9 100644
--- a/compiler/dex/quick/codegen_util.cc
+++ b/compiler/dex/quick/codegen_util.cc
@@ -1073,7 +1073,7 @@ Mir2Lir::Mir2Lir(CompilationUnit* cu, MIRGraph* mir_graph, ArenaAllocator* arena
       pc_rel_temp_(nullptr),
       dex_cache_arrays_min_offset_(std::numeric_limits<uint32_t>::max()),
       cfi_(&last_lir_insn_,
-           cu->compiler_driver->GetCompilerOptions().GetGenerateGDBInformation(),
+           cu->compiler_driver->GetCompilerOptions().GetIncludeDebugSymbols(),
            arena),
       in_to_reg_storage_mapping_(arena) {
   switch_tables_.reserve(4);
diff --git a/compiler/dex/quick/lazy_debug_frame_opcode_writer.cc b/compiler/dex/quick/lazy_debug_frame_opcode_writer.cc
index 03cf4bef8b..5cfb0ff557 100644
--- a/compiler/dex/quick/lazy_debug_frame_opcode_writer.cc
+++ b/compiler/dex/quick/lazy_debug_frame_opcode_writer.cc
@@ -21,7 +21,7 @@ namespace art {
 namespace dwarf {
 
 const ArenaVector<uint8_t>* LazyDebugFrameOpCodeWriter::Patch(size_t code_size) {
-  if (!enable_writes_) {
+  if (!this->enabled_) {
     DCHECK(this->data()->empty());
     return this->data();
   }
diff --git a/compiler/dex/quick/lazy_debug_frame_opcode_writer.h b/compiler/dex/quick/lazy_debug_frame_opcode_writer.h
index d71a87d567..94ffd7f957 100644
--- a/compiler/dex/quick/lazy_debug_frame_opcode_writer.h
+++ b/compiler/dex/quick/lazy_debug_frame_opcode_writer.h
@@ -28,7 +28,7 @@ namespace dwarf {
 // When we are generating the CFI code, we do not know the instuction offsets,
 // this class stores the LIR references and patches the instruction stream later.
 class LazyDebugFrameOpCodeWriter FINAL
-    : private DebugFrameOpCodeWriter<ArenaAllocatorAdapter<uint8_t>> {
+    : public DebugFrameOpCodeWriter<ArenaAllocatorAdapter<uint8_t>> {
   typedef DebugFrameOpCodeWriter<ArenaAllocatorAdapter<uint8_t>> Base;
  public:
   // This method is implicitely called the by opcode writers.
@@ -38,72 +38,12 @@ class LazyDebugFrameOpCodeWriter FINAL
     advances_.push_back({this->data()->size(), *last_lir_insn_});
   }
 
-  // The register was unspilled.
-  void Restore(Reg reg) {
-    if (enable_writes_) {
-      Base::Restore(reg);
-    }
-  }
-
-  // Custom alias - unspill many registers based on bitmask.
-  void RestoreMany(Reg reg_base, uint32_t reg_mask) {
-    if (enable_writes_) {
-      Base::RestoreMany(reg_base, reg_mask);
-    }
-  }
-
-  // Remember the state of register spills.
-  void RememberState() {
-    if (enable_writes_) {
-      Base::RememberState();
-    }
-  }
-
-  // Restore the state of register spills.
-  void RestoreState() {
-    if (enable_writes_) {
-      Base::RestoreState();
-    }
-  }
-
-  // Set the frame pointer (CFA) to (stack_pointer + offset).
-  void DefCFAOffset(int offset) {
-    if (enable_writes_) {
-      Base::DefCFAOffset(offset);
-    }
-    this->current_cfa_offset_ = offset;
-  }
-
-  // The stack size was increased by given delta.
-  void AdjustCFAOffset(int delta) {
-    DefCFAOffset(this->current_cfa_offset_ + delta);
-  }
-
-  // The register was spilled to (stack_pointer + offset).
-  void RelOffset(Reg reg, int offset) {
-    if (enable_writes_) {
-      Base::RelOffset(reg, offset);
-    }
-  }
-
-  // Custom alias - spill many registers based on bitmask.
-  void RelOffsetForMany(Reg reg_base, int offset, uint32_t reg_mask, int reg_size) {
-    if (enable_writes_) {
-      Base::RelOffsetForMany(reg_base, offset, reg_mask, reg_size);
-    }
-  }
-
-  using Base::GetCurrentCFAOffset;
-  using Base::SetCurrentCFAOffset;
-  using Base::GetCurrentPC;
-
   const ArenaVector<uint8_t>* Patch(size_t code_size);
 
   explicit LazyDebugFrameOpCodeWriter(LIR** last_lir_insn, bool enable_writes,
                                       ArenaAllocator* allocator)
-    : Base(allocator->Adapter()),
+    : Base(enable_writes, allocator->Adapter()),
       last_lir_insn_(last_lir_insn),
-      enable_writes_(enable_writes),
       advances_(allocator->Adapter()),
       patched_(false) {
   }
@@ -114,8 +54,9 @@ class LazyDebugFrameOpCodeWriter FINAL
     LIR* last_lir_insn;
   } Advance;
 
+  using Base::data;  // Hidden. Use Patch method instead.
+
   LIR** last_lir_insn_;
-  bool enable_writes_;
   ArenaVector<Advance> advances_;
   bool patched_;
 
diff --git a/compiler/dwarf/debug_frame_opcode_writer.h b/compiler/dwarf/debug_frame_opcode_writer.h
index 85186bbc22..d0d182106f 100644
--- a/compiler/dwarf/debug_frame_opcode_writer.h
+++ b/compiler/dwarf/debug_frame_opcode_writer.h
@@ -20,6 +20,7 @@
 #include "dwarf.h"
 #include "register.h"
 #include "writer.h"
+#include "utils.h"
 
 namespace art {
 namespace dwarf {
@@ -41,45 +42,51 @@ class DebugFrameOpCodeWriter : private Writer<Allocator> {
   static constexpr int kCodeAlignmentFactor = 1;
 
   // Explicitely advance the program counter to given location.
-  void AdvancePC(int absolute_pc) {
+  void ALWAYS_INLINE AdvancePC(int absolute_pc) {
     DCHECK_GE(absolute_pc, current_pc_);
-    int delta = FactorCodeOffset(absolute_pc - current_pc_);
-    if (delta != 0) {
-      if (delta <= 0x3F) {
-        this->PushUint8(DW_CFA_advance_loc | delta);
-      } else if (delta <= UINT8_MAX) {
-        this->PushUint8(DW_CFA_advance_loc1);
-        this->PushUint8(delta);
-      } else if (delta <= UINT16_MAX) {
-        this->PushUint8(DW_CFA_advance_loc2);
-        this->PushUint16(delta);
-      } else {
-        this->PushUint8(DW_CFA_advance_loc4);
-        this->PushUint32(delta);
+    if (UNLIKELY(enabled_)) {
+      int delta = FactorCodeOffset(absolute_pc - current_pc_);
+      if (delta != 0) {
+        if (delta <= 0x3F) {
+          this->PushUint8(DW_CFA_advance_loc | delta);
+        } else if (delta <= UINT8_MAX) {
+          this->PushUint8(DW_CFA_advance_loc1);
+          this->PushUint8(delta);
+        } else if (delta <= UINT16_MAX) {
+          this->PushUint8(DW_CFA_advance_loc2);
+          this->PushUint16(delta);
+        } else {
+          this->PushUint8(DW_CFA_advance_loc4);
+          this->PushUint32(delta);
+        }
       }
+      current_pc_ = absolute_pc;
     }
-    current_pc_ = absolute_pc;
   }
 
   // Override this method to automatically advance the PC before each opcode.
   virtual void ImplicitlyAdvancePC() { }
 
   // Common alias in assemblers - spill relative to current stack pointer.
-  void RelOffset(Reg reg, int offset) {
+  void ALWAYS_INLINE RelOffset(Reg reg, int offset) {
     Offset(reg, offset - current_cfa_offset_);
   }
 
   // Common alias in assemblers - increase stack frame size.
-  void AdjustCFAOffset(int delta) {
+  void ALWAYS_INLINE AdjustCFAOffset(int delta) {
     DefCFAOffset(current_cfa_offset_ + delta);
   }
 
   // Custom alias - spill many registers based on bitmask.
-  void RelOffsetForMany(Reg reg_base, int offset, uint32_t reg_mask,
-                        int reg_size) {
+  void ALWAYS_INLINE RelOffsetForMany(Reg reg_base, int offset,
+                                      uint32_t reg_mask, int reg_size) {
     DCHECK(reg_size == 4 || reg_size == 8);
-    for (int i = 0; reg_mask != 0u; reg_mask >>= 1, i++) {
-      if ((reg_mask & 1) != 0u) {
+    if (UNLIKELY(enabled_)) {
+      for (int i = 0; reg_mask != 0u; reg_mask >>= 1, i++) {
+        // Skip zero bits and go to the set bit.
+        int num_zeros = CTZ(reg_mask);
+        i += num_zeros;
+        reg_mask >>= num_zeros;
         RelOffset(Reg(reg_base.num() + i), offset);
         offset += reg_size;
       }
@@ -87,175 +94,214 @@ class DebugFrameOpCodeWriter : private Writer<Allocator> {
   }
 
   // Custom alias - unspill many registers based on bitmask.
-  void RestoreMany(Reg reg_base, uint32_t reg_mask) {
-    for (int i = 0; reg_mask != 0u; reg_mask >>= 1, i++) {
-      if ((reg_mask & 1) != 0u) {
+  void ALWAYS_INLINE RestoreMany(Reg reg_base, uint32_t reg_mask) {
+    if (UNLIKELY(enabled_)) {
+      for (int i = 0; reg_mask != 0u; reg_mask >>= 1, i++) {
+        // Skip zero bits and go to the set bit.
+        int num_zeros = CTZ(reg_mask);
+        i += num_zeros;
+        reg_mask >>= num_zeros;
         Restore(Reg(reg_base.num() + i));
       }
     }
   }
 
-  void Nop() {
-    this->PushUint8(DW_CFA_nop);
+  void ALWAYS_INLINE Nop() {
+    if (UNLIKELY(enabled_)) {
+      this->PushUint8(DW_CFA_nop);
+    }
   }
 
-  void Offset(Reg reg, int offset) {
-    ImplicitlyAdvancePC();
-    int factored_offset = FactorDataOffset(offset);  // May change sign.
-    if (factored_offset >= 0) {
-      if (0 <= reg.num() && reg.num() <= 0x3F) {
-        this->PushUint8(DW_CFA_offset | reg.num());
-        this->PushUleb128(factored_offset);
+  void ALWAYS_INLINE Offset(Reg reg, int offset) {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      int factored_offset = FactorDataOffset(offset);  // May change sign.
+      if (factored_offset >= 0) {
+        if (0 <= reg.num() && reg.num() <= 0x3F) {
+          this->PushUint8(DW_CFA_offset | reg.num());
+          this->PushUleb128(factored_offset);
+        } else {
+          this->PushUint8(DW_CFA_offset_extended);
+          this->PushUleb128(reg.num());
+          this->PushUleb128(factored_offset);
+        }
       } else {
-        this->PushUint8(DW_CFA_offset_extended);
+        uses_dwarf3_features_ = true;
+        this->PushUint8(DW_CFA_offset_extended_sf);
         this->PushUleb128(reg.num());
-        this->PushUleb128(factored_offset);
+        this->PushSleb128(factored_offset);
       }
-    } else {
-      uses_dwarf3_features_ = true;
-      this->PushUint8(DW_CFA_offset_extended_sf);
-      this->PushUleb128(reg.num());
-      this->PushSleb128(factored_offset);
     }
   }
 
-  void Restore(Reg reg) {
-    ImplicitlyAdvancePC();
-    if (0 <= reg.num() && reg.num() <= 0x3F) {
-      this->PushUint8(DW_CFA_restore | reg.num());
-    } else {
-      this->PushUint8(DW_CFA_restore_extended);
-      this->PushUleb128(reg.num());
+  void ALWAYS_INLINE Restore(Reg reg) {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      if (0 <= reg.num() && reg.num() <= 0x3F) {
+        this->PushUint8(DW_CFA_restore | reg.num());
+      } else {
+        this->PushUint8(DW_CFA_restore_extended);
+        this->PushUleb128(reg.num());
+      }
     }
   }
 
-  void Undefined(Reg reg) {
-    ImplicitlyAdvancePC();
-    this->PushUint8(DW_CFA_undefined);
-    this->PushUleb128(reg.num());
+  void ALWAYS_INLINE Undefined(Reg reg) {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      this->PushUint8(DW_CFA_undefined);
+      this->PushUleb128(reg.num());
+    }
   }
 
-  void SameValue(Reg reg) {
-    ImplicitlyAdvancePC();
-    this->PushUint8(DW_CFA_same_value);
-    this->PushUleb128(reg.num());
+  void ALWAYS_INLINE SameValue(Reg reg) {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      this->PushUint8(DW_CFA_same_value);
+      this->PushUleb128(reg.num());
+    }
   }
 
   // The previous value of "reg" is stored in register "new_reg".
-  void Register(Reg reg, Reg new_reg) {
-    ImplicitlyAdvancePC();
-    this->PushUint8(DW_CFA_register);
-    this->PushUleb128(reg.num());
-    this->PushUleb128(new_reg.num());
-  }
-
-  void RememberState() {
-    ImplicitlyAdvancePC();
-    this->PushUint8(DW_CFA_remember_state);
-  }
-
-  void RestoreState() {
-    ImplicitlyAdvancePC();
-    this->PushUint8(DW_CFA_restore_state);
+  void ALWAYS_INLINE Register(Reg reg, Reg new_reg) {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      this->PushUint8(DW_CFA_register);
+      this->PushUleb128(reg.num());
+      this->PushUleb128(new_reg.num());
+    }
   }
 
-  void DefCFA(Reg reg, int offset) {
-    ImplicitlyAdvancePC();
-    if (offset >= 0) {
-      this->PushUint8(DW_CFA_def_cfa);
-      this->PushUleb128(reg.num());
-      this->PushUleb128(offset);  // Non-factored.
-    } else {
-      uses_dwarf3_features_ = true;
-      this->PushUint8(DW_CFA_def_cfa_sf);
-      this->PushUleb128(reg.num());
-      this->PushSleb128(FactorDataOffset(offset));
+  void ALWAYS_INLINE RememberState() {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      this->PushUint8(DW_CFA_remember_state);
     }
-    current_cfa_offset_ = offset;
   }
 
-  void DefCFARegister(Reg reg) {
-    ImplicitlyAdvancePC();
-    this->PushUint8(DW_CFA_def_cfa_register);
-    this->PushUleb128(reg.num());
+  void ALWAYS_INLINE RestoreState() {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      this->PushUint8(DW_CFA_restore_state);
+    }
   }
 
-  void DefCFAOffset(int offset) {
-    if (current_cfa_offset_ != offset) {
+  void ALWAYS_INLINE DefCFA(Reg reg, int offset) {
+    if (UNLIKELY(enabled_)) {
       ImplicitlyAdvancePC();
       if (offset >= 0) {
-        this->PushUint8(DW_CFA_def_cfa_offset);
+        this->PushUint8(DW_CFA_def_cfa);
+        this->PushUleb128(reg.num());
         this->PushUleb128(offset);  // Non-factored.
       } else {
         uses_dwarf3_features_ = true;
-        this->PushUint8(DW_CFA_def_cfa_offset_sf);
+        this->PushUint8(DW_CFA_def_cfa_sf);
+        this->PushUleb128(reg.num());
         this->PushSleb128(FactorDataOffset(offset));
       }
-      current_cfa_offset_ = offset;
     }
+    current_cfa_offset_ = offset;
   }
 
-  void ValOffset(Reg reg, int offset) {
-    ImplicitlyAdvancePC();
-    uses_dwarf3_features_ = true;
-    int factored_offset = FactorDataOffset(offset);  // May change sign.
-    if (factored_offset >= 0) {
-      this->PushUint8(DW_CFA_val_offset);
-      this->PushUleb128(reg.num());
-      this->PushUleb128(factored_offset);
-    } else {
-      this->PushUint8(DW_CFA_val_offset_sf);
+  void ALWAYS_INLINE DefCFARegister(Reg reg) {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      this->PushUint8(DW_CFA_def_cfa_register);
       this->PushUleb128(reg.num());
-      this->PushSleb128(factored_offset);
     }
   }
 
-  void DefCFAExpression(void* expr, int expr_size) {
-    ImplicitlyAdvancePC();
-    uses_dwarf3_features_ = true;
-    this->PushUint8(DW_CFA_def_cfa_expression);
-    this->PushUleb128(expr_size);
-    this->PushData(expr, expr_size);
+  void ALWAYS_INLINE DefCFAOffset(int offset) {
+    if (UNLIKELY(enabled_)) {
+      if (current_cfa_offset_ != offset) {
+        ImplicitlyAdvancePC();
+        if (offset >= 0) {
+          this->PushUint8(DW_CFA_def_cfa_offset);
+          this->PushUleb128(offset);  // Non-factored.
+        } else {
+          uses_dwarf3_features_ = true;
+          this->PushUint8(DW_CFA_def_cfa_offset_sf);
+          this->PushSleb128(FactorDataOffset(offset));
+        }
+      }
+    }
+    // Uncoditional so that the user can still get and check the value.
+    current_cfa_offset_ = offset;
   }
 
-  void Expression(Reg reg, void* expr, int expr_size) {
-    ImplicitlyAdvancePC();
-    uses_dwarf3_features_ = true;
-    this->PushUint8(DW_CFA_expression);
-    this->PushUleb128(reg.num());
-    this->PushUleb128(expr_size);
-    this->PushData(expr, expr_size);
+  void ALWAYS_INLINE ValOffset(Reg reg, int offset) {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      uses_dwarf3_features_ = true;
+      int factored_offset = FactorDataOffset(offset);  // May change sign.
+      if (factored_offset >= 0) {
+        this->PushUint8(DW_CFA_val_offset);
+        this->PushUleb128(reg.num());
+        this->PushUleb128(factored_offset);
+      } else {
+        this->PushUint8(DW_CFA_val_offset_sf);
+        this->PushUleb128(reg.num());
+        this->PushSleb128(factored_offset);
+      }
+    }
   }
 
-  void ValExpression(Reg reg, void* expr, int expr_size) {
-    ImplicitlyAdvancePC();
-    uses_dwarf3_features_ = true;
-    this->PushUint8(DW_CFA_val_expression);
-    this->PushUleb128(reg.num());
-    this->PushUleb128(expr_size);
-    this->PushData(expr, expr_size);
+  void ALWAYS_INLINE DefCFAExpression(void * expr, int expr_size) {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      uses_dwarf3_features_ = true;
+      this->PushUint8(DW_CFA_def_cfa_expression);
+      this->PushUleb128(expr_size);
+      this->PushData(expr, expr_size);
+    }
   }
 
-  int GetCurrentPC() const {
-    return current_pc_;
+  void ALWAYS_INLINE Expression(Reg reg, void * expr, int expr_size) {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      uses_dwarf3_features_ = true;
+      this->PushUint8(DW_CFA_expression);
+      this->PushUleb128(reg.num());
+      this->PushUleb128(expr_size);
+      this->PushData(expr, expr_size);
+    }
   }
 
-  int GetCurrentCFAOffset() const {
-    return current_cfa_offset_;
+  void ALWAYS_INLINE ValExpression(Reg reg, void * expr, int expr_size) {
+    if (UNLIKELY(enabled_)) {
+      ImplicitlyAdvancePC();
+      uses_dwarf3_features_ = true;
+      this->PushUint8(DW_CFA_val_expression);
+      this->PushUleb128(reg.num());
+      this->PushUleb128(expr_size);
+      this->PushData(expr, expr_size);
+    }
   }
 
-  void SetCurrentCFAOffset(int offset) {
-    current_cfa_offset_ = offset;
-  }
+  bool IsEnabled() const { return enabled_; }
+
+  void SetEnabled(bool value) { enabled_ = value; }
+
+  int GetCurrentPC() const { return current_pc_; }
+
+  int GetCurrentCFAOffset() const { return current_cfa_offset_; }
+
+  void SetCurrentCFAOffset(int offset) { current_cfa_offset_ = offset; }
 
   using Writer<Allocator>::data;
 
-  DebugFrameOpCodeWriter(const Allocator& alloc = Allocator())
+  DebugFrameOpCodeWriter(bool enabled = true,
+                         const Allocator& alloc = Allocator())
       : Writer<Allocator>(&opcodes_),
+        enabled_(enabled),
         opcodes_(alloc),
         current_cfa_offset_(0),
         current_pc_(0),
         uses_dwarf3_features_(false) {
+    if (enabled) {
+      // Best guess based on couple of observed outputs.
+      opcodes_.reserve(16);
+    }
   }
 
   virtual ~DebugFrameOpCodeWriter() { }
@@ -271,6 +317,7 @@ class DebugFrameOpCodeWriter : private Writer<Allocator> {
     return offset / kCodeAlignmentFactor;
   }
 
+  bool enabled_;  // If disabled all writes are no-ops.
   std::vector<uint8_t, Allocator> opcodes_;
   int current_cfa_offset_;
   int current_pc_;
diff --git a/compiler/elf_writer_quick.cc b/compiler/elf_writer_quick.cc
index 354c71ec12..193cbe281e 100644
--- a/compiler/elf_writer_quick.cc
+++ b/compiler/elf_writer_quick.cc
@@ -679,7 +679,7 @@ static void WriteDebugSymbols(const CompilerDriver* compiler_driver,
   }
 
   if (!method_info.empty() &&
-      compiler_driver->GetCompilerOptions().GetGenerateGDBInformation()) {
+      compiler_driver->GetCompilerOptions().GetIncludeDebugSymbols()) {
     ElfRawSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> debug_info(".debug_info",
                                                                    SHT_PROGBITS,
                                                                    0, nullptr, 0, 1, 0);
diff --git a/compiler/jni/quick/jni_compiler.cc b/compiler/jni/quick/jni_compiler.cc
index 49b7cd1aa5..8a14038074 100644
--- a/compiler/jni/quick/jni_compiler.cc
+++ b/compiler/jni/quick/jni_compiler.cc
@@ -28,6 +28,7 @@
 #include "compiled_method.h"
 #include "dex_file-inl.h"
 #include "driver/compiler_driver.h"
+#include "driver/compiler_options.h"
 #include "entrypoints/quick/quick_entrypoints.h"
 #include "jni_env_ext.h"
 #include "mirror/art_method.h"
@@ -93,6 +94,7 @@ CompiledMethod* ArtJniCompileMethodInternal(CompilerDriver* driver,
 
   // Assembler that holds generated instructions
   std::unique_ptr<Assembler> jni_asm(Assembler::Create(instruction_set));
+  jni_asm->cfi().SetEnabled(driver->GetCompilerOptions().GetIncludeDebugSymbols());
 
   // Offsets into data structures
   // TODO: if cross compiling these offsets are for the host not the target
@@ -435,13 +437,18 @@ CompiledMethod* ArtJniCompileMethodInternal(CompilerDriver* driver,
   MemoryRegion code(&managed_code[0], managed_code.size());
   __ FinalizeInstructions(code);
 
-  return CompiledMethod::SwapAllocCompiledMethodCFI(driver,
-                                                    instruction_set,
-                                                    ArrayRef<const uint8_t>(managed_code),
-                                                    frame_size,
-                                                    main_jni_conv->CoreSpillMask(),
-                                                    main_jni_conv->FpSpillMask(),
-                                                    ArrayRef<const uint8_t>(*jni_asm->cfi().data()));
+  return CompiledMethod::SwapAllocCompiledMethod(driver,
+                                                 instruction_set,
+                                                 ArrayRef<const uint8_t>(managed_code),
+                                                 frame_size,
+                                                 main_jni_conv->CoreSpillMask(),
+                                                 main_jni_conv->FpSpillMask(),
+                                                 nullptr,  // src_mapping_table.
+                                                 ArrayRef<const uint8_t>(),  // mapping_table.
+                                                 ArrayRef<const uint8_t>(),  // vmap_table.
+                                                 ArrayRef<const uint8_t>(),  // native_gc_map.
+                                                 ArrayRef<const uint8_t>(*jni_asm->cfi().data()),
+                                                 ArrayRef<const LinkerPatch>());
 }
 
 // Copy a single parameter from the managed to the JNI calling convention
diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index da28dc7ecb..8736374306 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -82,6 +82,7 @@ void CodeGenerator::CompileInternal(CodeAllocator* allocator, bool is_baseline)
   HGraphVisitor* instruction_visitor = GetInstructionVisitor();
   DCHECK_EQ(current_block_index_, 0u);
   GenerateFrameEntry();
+  DCHECK_EQ(GetAssembler()->cfi().GetCurrentCFAOffset(), static_cast<int>(frame_size_));
   for (size_t e = block_order_->Size(); current_block_index_ < e; ++current_block_index_) {
     HBasicBlock* block = block_order_->Get(current_block_index_);
     // Don't generate code for an empty block. Its predecessors will branch to its successor
@@ -415,7 +416,16 @@ void CodeGenerator::BuildNativeGCMap(
   }
 }
 
-void CodeGenerator::BuildMappingTable(std::vector<uint8_t>* data, DefaultSrcMap* src_map) const {
+void CodeGenerator::BuildSourceMap(DefaultSrcMap* src_map) const {
+  for (size_t i = 0; i < pc_infos_.Size(); i++) {
+    struct PcInfo pc_info = pc_infos_.Get(i);
+    uint32_t pc2dex_offset = pc_info.native_pc;
+    int32_t pc2dex_dalvik_offset = pc_info.dex_pc;
+    src_map->push_back(SrcMapElem({pc2dex_offset, pc2dex_dalvik_offset}));
+  }
+}
+
+void CodeGenerator::BuildMappingTable(std::vector<uint8_t>* data) const {
   uint32_t pc2dex_data_size = 0u;
   uint32_t pc2dex_entries = pc_infos_.Size();
   uint32_t pc2dex_offset = 0u;
@@ -425,19 +435,12 @@ void CodeGenerator::BuildMappingTable(std::vector<uint8_t>* data, DefaultSrcMap*
   uint32_t dex2pc_offset = 0u;
   int32_t dex2pc_dalvik_offset = 0;
 
-  if (src_map != nullptr) {
-    src_map->reserve(pc2dex_entries);
-  }
-
   for (size_t i = 0; i < pc2dex_entries; i++) {
     struct PcInfo pc_info = pc_infos_.Get(i);
     pc2dex_data_size += UnsignedLeb128Size(pc_info.native_pc - pc2dex_offset);
     pc2dex_data_size += SignedLeb128Size(pc_info.dex_pc - pc2dex_dalvik_offset);
     pc2dex_offset = pc_info.native_pc;
     pc2dex_dalvik_offset = pc_info.dex_pc;
-    if (src_map != nullptr) {
-      src_map->push_back(SrcMapElem({pc2dex_offset, pc2dex_dalvik_offset}));
-    }
   }
 
   // Walk over the blocks and find which ones correspond to catch block entries.
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index 07ca6b1ccf..b888aca264 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -205,7 +205,8 @@ class CodeGenerator {
     slow_paths_.Add(slow_path);
   }
 
-  void BuildMappingTable(std::vector<uint8_t>* vector, DefaultSrcMap* src_map) const;
+  void BuildSourceMap(DefaultSrcMap* src_map) const;
+  void BuildMappingTable(std::vector<uint8_t>* vector) const;
   void BuildVMapTable(std::vector<uint8_t>* vector) const;
   void BuildNativeGCMap(
       std::vector<uint8_t>* vector, const DexCompilationUnit& dex_compilation_unit) const;
@@ -425,6 +426,8 @@ class CodeGenerator {
 
   StackMapStream stack_map_stream_;
 
+  friend class OptimizingCFITest;
+
   DISALLOW_COPY_AND_ASSIGN(CodeGenerator);
 };
 
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index cfc798a34e..a799a519c0 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -513,6 +513,14 @@ void CodeGeneratorARM::ComputeSpillMask() {
   }
 }
 
+static dwarf::Reg DWARFReg(Register reg) {
+    return dwarf::Reg::ArmCore(static_cast<int>(reg));
+}
+
+static dwarf::Reg DWARFReg(SRegister reg) {
+    return dwarf::Reg::ArmFp(static_cast<int>(reg));
+}
+
 void CodeGeneratorARM::GenerateFrameEntry() {
   bool skip_overflow_check =
       IsLeafMethod() && !FrameNeedsStackCheck(GetFrameSize(), InstructionSet::kArm);
@@ -531,12 +539,19 @@ void CodeGeneratorARM::GenerateFrameEntry() {
 
   // PC is in the list of callee-save to mimic Quick, but we need to push
   // LR at entry instead.
-  __ PushList((core_spill_mask_ & (~(1 << PC))) | 1 << LR);
+  uint32_t push_mask = (core_spill_mask_ & (~(1 << PC))) | 1 << LR;
+  __ PushList(push_mask);
+  __ cfi().AdjustCFAOffset(kArmWordSize * POPCOUNT(push_mask));
+  __ cfi().RelOffsetForMany(DWARFReg(Register(0)), 0, push_mask, kArmWordSize);
   if (fpu_spill_mask_ != 0) {
     SRegister start_register = SRegister(LeastSignificantBit(fpu_spill_mask_));
     __ vpushs(start_register, POPCOUNT(fpu_spill_mask_));
+    __ cfi().AdjustCFAOffset(kArmWordSize * POPCOUNT(fpu_spill_mask_));
+    __ cfi().RelOffsetForMany(DWARFReg(SRegister(0)), 0, fpu_spill_mask_, kArmWordSize);
   }
-  __ AddConstant(SP, -(GetFrameSize() - FrameEntrySpillSize()));
+  int adjust = GetFrameSize() - FrameEntrySpillSize();
+  __ AddConstant(SP, -adjust);
+  __ cfi().AdjustCFAOffset(adjust);
   __ StoreToOffset(kStoreWord, R0, SP, 0);
 }
 
@@ -545,10 +560,14 @@ void CodeGeneratorARM::GenerateFrameExit() {
     __ bx(LR);
     return;
   }
-  __ AddConstant(SP, GetFrameSize() - FrameEntrySpillSize());
+  int adjust = GetFrameSize() - FrameEntrySpillSize();
+  __ AddConstant(SP, adjust);
+  __ cfi().AdjustCFAOffset(-adjust);
   if (fpu_spill_mask_ != 0) {
     SRegister start_register = SRegister(LeastSignificantBit(fpu_spill_mask_));
     __ vpops(start_register, POPCOUNT(fpu_spill_mask_));
+    __ cfi().AdjustCFAOffset(-kArmPointerSize * POPCOUNT(fpu_spill_mask_));
+    __ cfi().RestoreMany(DWARFReg(SRegister(0)), fpu_spill_mask_);
   }
   __ PopList(core_spill_mask_);
 }
@@ -1190,7 +1209,10 @@ void LocationsBuilderARM::VisitReturnVoid(HReturnVoid* ret) {
 
 void InstructionCodeGeneratorARM::VisitReturnVoid(HReturnVoid* ret) {
   UNUSED(ret);
+  __ cfi().RememberState();
   codegen_->GenerateFrameExit();
+  __ cfi().RestoreState();
+  __ cfi().DefCFAOffset(codegen_->GetFrameSize());
 }
 
 void LocationsBuilderARM::VisitReturn(HReturn* ret) {
@@ -1201,7 +1223,10 @@ void LocationsBuilderARM::VisitReturn(HReturn* ret) {
 
 void InstructionCodeGeneratorARM::VisitReturn(HReturn* ret) {
   UNUSED(ret);
+  __ cfi().RememberState();
   codegen_->GenerateFrameExit();
+  __ cfi().RestoreState();
+  __ cfi().DefCFAOffset(codegen_->GetFrameSize());
 }
 
 void LocationsBuilderARM::VisitInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) {
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 439e85ca6c..12eb0c78ea 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -465,20 +465,67 @@ void CodeGeneratorARM64::GenerateFrameEntry() {
     //      ...                       : reserved frame space.
     //      sp[0]                     : current method.
     __ Str(kArtMethodRegister, MemOperand(sp, -frame_size, PreIndex));
-    __ PokeCPURegList(GetFramePreservedCoreRegisters(), frame_size - GetCoreSpillSize());
-    __ PokeCPURegList(GetFramePreservedFPRegisters(), frame_size - FrameEntrySpillSize());
+    GetAssembler()->cfi().AdjustCFAOffset(frame_size);
+    SpillRegisters(GetFramePreservedCoreRegisters(), frame_size - GetCoreSpillSize());
+    SpillRegisters(GetFramePreservedFPRegisters(), frame_size - FrameEntrySpillSize());
   }
 }
 
 void CodeGeneratorARM64::GenerateFrameExit() {
   if (!HasEmptyFrame()) {
     int frame_size = GetFrameSize();
-    __ PeekCPURegList(GetFramePreservedFPRegisters(), frame_size - FrameEntrySpillSize());
-    __ PeekCPURegList(GetFramePreservedCoreRegisters(), frame_size - GetCoreSpillSize());
+    UnspillRegisters(GetFramePreservedFPRegisters(), frame_size - FrameEntrySpillSize());
+    UnspillRegisters(GetFramePreservedCoreRegisters(), frame_size - GetCoreSpillSize());
     __ Drop(frame_size);
+    GetAssembler()->cfi().AdjustCFAOffset(-frame_size);
   }
 }
 
+static inline dwarf::Reg DWARFReg(CPURegister reg) {
+  if (reg.IsFPRegister()) {
+    return dwarf::Reg::Arm64Fp(reg.code());
+  } else {
+    DCHECK_LT(reg.code(), 31u);  // X0 - X30.
+    return dwarf::Reg::Arm64Core(reg.code());
+  }
+}
+
+void CodeGeneratorARM64::SpillRegisters(vixl::CPURegList registers, int offset) {
+  int size = registers.RegisterSizeInBytes();
+  while (registers.Count() >= 2) {
+    const CPURegister& dst0 = registers.PopLowestIndex();
+    const CPURegister& dst1 = registers.PopLowestIndex();
+    __ Stp(dst0, dst1, MemOperand(__ StackPointer(), offset));
+    GetAssembler()->cfi().RelOffset(DWARFReg(dst0), offset);
+    GetAssembler()->cfi().RelOffset(DWARFReg(dst1), offset + size);
+    offset += 2 * size;
+  }
+  if (!registers.IsEmpty()) {
+    const CPURegister& dst0 = registers.PopLowestIndex();
+    __ Str(dst0, MemOperand(__ StackPointer(), offset));
+    GetAssembler()->cfi().RelOffset(DWARFReg(dst0), offset);
+  }
+  DCHECK(registers.IsEmpty());
+}
+
+void CodeGeneratorARM64::UnspillRegisters(vixl::CPURegList registers, int offset) {
+  int size = registers.RegisterSizeInBytes();
+  while (registers.Count() >= 2) {
+    const CPURegister& dst0 = registers.PopLowestIndex();
+    const CPURegister& dst1 = registers.PopLowestIndex();
+    __ Ldp(dst0, dst1, MemOperand(__ StackPointer(), offset));
+    GetAssembler()->cfi().Restore(DWARFReg(dst0));
+    GetAssembler()->cfi().Restore(DWARFReg(dst1));
+    offset += 2 * size;
+  }
+  if (!registers.IsEmpty()) {
+    const CPURegister& dst0 = registers.PopLowestIndex();
+    __ Ldr(dst0, MemOperand(__ StackPointer(), offset));
+    GetAssembler()->cfi().Restore(DWARFReg(dst0));
+  }
+  DCHECK(registers.IsEmpty());
+}
+
 void CodeGeneratorARM64::Bind(HBasicBlock* block) {
   __ Bind(GetLabelOf(block));
 }
@@ -2403,8 +2450,11 @@ void LocationsBuilderARM64::VisitReturn(HReturn* instruction) {
 
 void InstructionCodeGeneratorARM64::VisitReturn(HReturn* instruction) {
   UNUSED(instruction);
+  GetAssembler()->cfi().RememberState();
   codegen_->GenerateFrameExit();
   __ Ret();
+  GetAssembler()->cfi().RestoreState();
+  GetAssembler()->cfi().DefCFAOffset(codegen_->GetFrameSize());
 }
 
 void LocationsBuilderARM64::VisitReturnVoid(HReturnVoid* instruction) {
@@ -2413,8 +2463,11 @@ void LocationsBuilderARM64::VisitReturnVoid(HReturnVoid* instruction) {
 
 void InstructionCodeGeneratorARM64::VisitReturnVoid(HReturnVoid* instruction) {
   UNUSED(instruction);
+  GetAssembler()->cfi().RememberState();
   codegen_->GenerateFrameExit();
   __ Ret();
+  GetAssembler()->cfi().RestoreState();
+  GetAssembler()->cfi().DefCFAOffset(codegen_->GetFrameSize());
 }
 
 void LocationsBuilderARM64::VisitShl(HShl* shl) {
diff --git a/compiler/optimizing/code_generator_arm64.h b/compiler/optimizing/code_generator_arm64.h
index 7edb129880..9430e31037 100644
--- a/compiler/optimizing/code_generator_arm64.h
+++ b/compiler/optimizing/code_generator_arm64.h
@@ -227,6 +227,8 @@ class CodeGeneratorARM64 : public CodeGenerator {
 
   void GenerateFrameEntry() OVERRIDE;
   void GenerateFrameExit() OVERRIDE;
+  void SpillRegisters(vixl::CPURegList registers, int offset);
+  void UnspillRegisters(vixl::CPURegList registers, int offset);
 
   vixl::CPURegList GetFramePreservedCoreRegisters() const {
     return vixl::CPURegList(vixl::CPURegister::kRegister, vixl::kXRegSize,
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 92b62e2c84..8fbc64ea5e 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -459,7 +459,12 @@ InstructionCodeGeneratorX86::InstructionCodeGeneratorX86(HGraph* graph, CodeGene
         assembler_(codegen->GetAssembler()),
         codegen_(codegen) {}
 
+static dwarf::Reg DWARFReg(Register reg) {
+    return dwarf::Reg::X86Core(static_cast<int>(reg));
+}
+
 void CodeGeneratorX86::GenerateFrameEntry() {
+  __ cfi().SetCurrentCFAOffset(kX86WordSize);  // return address
   __ Bind(&frame_entry_label_);
   bool skip_overflow_check =
       IsLeafMethod() && !FrameNeedsStackCheck(GetFrameSize(), InstructionSet::kX86);
@@ -478,10 +483,14 @@ void CodeGeneratorX86::GenerateFrameEntry() {
     Register reg = kCoreCalleeSaves[i];
     if (allocated_registers_.ContainsCoreRegister(reg)) {
       __ pushl(reg);
+      __ cfi().AdjustCFAOffset(kX86WordSize);
+      __ cfi().RelOffset(DWARFReg(reg), 0);
     }
   }
 
-  __ subl(ESP, Immediate(GetFrameSize() - FrameEntrySpillSize()));
+  int adjust = GetFrameSize() - FrameEntrySpillSize();
+  __ subl(ESP, Immediate(adjust));
+  __ cfi().AdjustCFAOffset(adjust);
   __ movl(Address(ESP, kCurrentMethodStackOffset), EAX);
 }
 
@@ -490,12 +499,16 @@ void CodeGeneratorX86::GenerateFrameExit() {
     return;
   }
 
-  __ addl(ESP, Immediate(GetFrameSize() - FrameEntrySpillSize()));
+  int adjust = GetFrameSize() - FrameEntrySpillSize();
+  __ addl(ESP, Immediate(adjust));
+  __ cfi().AdjustCFAOffset(-adjust);
 
   for (size_t i = 0; i < arraysize(kCoreCalleeSaves); ++i) {
     Register reg = kCoreCalleeSaves[i];
     if (allocated_registers_.ContainsCoreRegister(reg)) {
       __ popl(reg);
+      __ cfi().AdjustCFAOffset(-static_cast<int>(kX86WordSize));
+      __ cfi().Restore(DWARFReg(reg));
     }
   }
 }
@@ -1102,8 +1115,11 @@ void LocationsBuilderX86::VisitReturnVoid(HReturnVoid* ret) {
 
 void InstructionCodeGeneratorX86::VisitReturnVoid(HReturnVoid* ret) {
   UNUSED(ret);
+  __ cfi().RememberState();
   codegen_->GenerateFrameExit();
   __ ret();
+  __ cfi().RestoreState();
+  __ cfi().DefCFAOffset(codegen_->GetFrameSize());
 }
 
 void LocationsBuilderX86::VisitReturn(HReturn* ret) {
@@ -1161,8 +1177,11 @@ void InstructionCodeGeneratorX86::VisitReturn(HReturn* ret) {
         LOG(FATAL) << "Unknown return type " << ret->InputAt(0)->GetType();
     }
   }
+  __ cfi().RememberState();
   codegen_->GenerateFrameExit();
   __ ret();
+  __ cfi().RestoreState();
+  __ cfi().DefCFAOffset(codegen_->GetFrameSize());
 }
 
 void LocationsBuilderX86::VisitInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) {
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index cdbc7780a8..b5e038fb05 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -481,7 +481,15 @@ void CodeGeneratorX86_64::SetupBlockedRegisters(bool is_baseline) const {
   }
 }
 
+static dwarf::Reg DWARFReg(Register reg) {
+    return dwarf::Reg::X86_64Core(static_cast<int>(reg));
+}
+static dwarf::Reg DWARFReg(FloatRegister reg) {
+    return dwarf::Reg::X86_64Fp(static_cast<int>(reg));
+}
+
 void CodeGeneratorX86_64::GenerateFrameEntry() {
+  __ cfi().SetCurrentCFAOffset(kX86_64WordSize);  // return address
   __ Bind(&frame_entry_label_);
   bool skip_overflow_check = IsLeafMethod()
       && !FrameNeedsStackCheck(GetFrameSize(), InstructionSet::kX86_64);
@@ -501,17 +509,22 @@ void CodeGeneratorX86_64::GenerateFrameEntry() {
     Register reg = kCoreCalleeSaves[i];
     if (allocated_registers_.ContainsCoreRegister(reg)) {
       __ pushq(CpuRegister(reg));
+      __ cfi().AdjustCFAOffset(kX86_64WordSize);
+      __ cfi().RelOffset(DWARFReg(reg), 0);
     }
   }
 
-  __ subq(CpuRegister(RSP), Immediate(GetFrameSize() - GetCoreSpillSize()));
+  int adjust = GetFrameSize() - GetCoreSpillSize();
+  __ subq(CpuRegister(RSP), Immediate(adjust));
+  __ cfi().AdjustCFAOffset(adjust);
   uint32_t xmm_spill_location = GetFpuSpillStart();
   size_t xmm_spill_slot_size = GetFloatingPointSpillSlotSize();
 
   for (int i = arraysize(kFpuCalleeSaves) - 1; i >= 0; --i) {
     if (allocated_registers_.ContainsFloatingPointRegister(kFpuCalleeSaves[i])) {
-      __ movsd(Address(CpuRegister(RSP), xmm_spill_location + (xmm_spill_slot_size * i)),
-               XmmRegister(kFpuCalleeSaves[i]));
+      int offset = xmm_spill_location + (xmm_spill_slot_size * i);
+      __ movsd(Address(CpuRegister(RSP), offset), XmmRegister(kFpuCalleeSaves[i]));
+      __ cfi().RelOffset(DWARFReg(kFpuCalleeSaves[i]), offset);
     }
   }
 
@@ -526,17 +539,22 @@ void CodeGeneratorX86_64::GenerateFrameExit() {
   size_t xmm_spill_slot_size = GetFloatingPointSpillSlotSize();
   for (size_t i = 0; i < arraysize(kFpuCalleeSaves); ++i) {
     if (allocated_registers_.ContainsFloatingPointRegister(kFpuCalleeSaves[i])) {
-      __ movsd(XmmRegister(kFpuCalleeSaves[i]),
-               Address(CpuRegister(RSP), xmm_spill_location + (xmm_spill_slot_size * i)));
+      int offset = xmm_spill_location + (xmm_spill_slot_size * i);
+      __ movsd(XmmRegister(kFpuCalleeSaves[i]), Address(CpuRegister(RSP), offset));
+      __ cfi().Restore(DWARFReg(kFpuCalleeSaves[i]));
     }
   }
 
-  __ addq(CpuRegister(RSP), Immediate(GetFrameSize() - GetCoreSpillSize()));
+  int adjust = GetFrameSize() - GetCoreSpillSize();
+  __ addq(CpuRegister(RSP), Immediate(adjust));
+  __ cfi().AdjustCFAOffset(-adjust);
 
   for (size_t i = 0; i < arraysize(kCoreCalleeSaves); ++i) {
     Register reg = kCoreCalleeSaves[i];
     if (allocated_registers_.ContainsCoreRegister(reg)) {
       __ popq(CpuRegister(reg));
+      __ cfi().AdjustCFAOffset(-static_cast<int>(kX86_64WordSize));
+      __ cfi().Restore(DWARFReg(reg));
     }
   }
 }
@@ -1123,8 +1141,11 @@ void LocationsBuilderX86_64::VisitReturnVoid(HReturnVoid* ret) {
 
 void InstructionCodeGeneratorX86_64::VisitReturnVoid(HReturnVoid* ret) {
   UNUSED(ret);
+  __ cfi().RememberState();
   codegen_->GenerateFrameExit();
   __ ret();
+  __ cfi().RestoreState();
+  __ cfi().DefCFAOffset(codegen_->GetFrameSize());
 }
 
 void LocationsBuilderX86_64::VisitReturn(HReturn* ret) {
@@ -1175,8 +1196,11 @@ void InstructionCodeGeneratorX86_64::VisitReturn(HReturn* ret) {
         LOG(FATAL) << "Unexpected return type " << ret->InputAt(0)->GetType();
     }
   }
+  __ cfi().RememberState();
   codegen_->GenerateFrameExit();
   __ ret();
+  __ cfi().RestoreState();
+  __ cfi().DefCFAOffset(codegen_->GetFrameSize());
 }
 
 Location InvokeDexCallingConventionVisitor::GetNextLocation(Primitive::Type type) {
diff --git a/compiler/optimizing/optimizing_cfi_test.cc b/compiler/optimizing/optimizing_cfi_test.cc
new file mode 100644
index 0000000000..8602255cc5
--- /dev/null
+++ b/compiler/optimizing/optimizing_cfi_test.cc
@@ -0,0 +1,108 @@
+/*
+ * Copyright (C) 2015 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <memory>
+#include <vector>
+
+#include "arch/instruction_set.h"
+#include "cfi_test.h"
+#include "gtest/gtest.h"
+#include "optimizing/code_generator.h"
+#include "utils/assembler.h"
+
+#include "optimizing/optimizing_cfi_test_expected.inc"
+
+namespace art {
+
+// Run the tests only on host.
+#ifndef HAVE_ANDROID_OS
+
+class OptimizingCFITest  : public CFITest {
+ public:
+  // Enable this flag to generate the expected outputs.
+  static constexpr bool kGenerateExpected = false;
+
+  void TestImpl(InstructionSet isa, const char* isa_str,
+                const std::vector<uint8_t>& expected_asm,
+                const std::vector<uint8_t>& expected_cfi) {
+    // Setup simple context.
+    ArenaPool pool;
+    ArenaAllocator allocator(&pool);
+    CompilerOptions opts;
+    std::unique_ptr<const InstructionSetFeatures> isa_features;
+    std::string error;
+    isa_features.reset(InstructionSetFeatures::FromVariant(isa, "default", &error));
+    HGraph graph(&allocator);
+    // Generate simple frame with some spills.
+    auto code_gen = CodeGenerator::Create(&graph, isa, *isa_features.get(), opts);
+    const int frame_size = 64;
+    int core_reg = 0;
+    int fp_reg = 0;
+    for (int i = 0; i < 2; i++) {  // Two registers of each kind.
+      for (; core_reg < 32; core_reg++) {
+        if (code_gen->IsCoreCalleeSaveRegister(core_reg)) {
+          auto location = Location::RegisterLocation(core_reg);
+          code_gen->AddAllocatedRegister(location);
+          core_reg++;
+          break;
+        }
+      }
+      for (; fp_reg < 32; fp_reg++) {
+        if (code_gen->IsFloatingPointCalleeSaveRegister(fp_reg)) {
+          auto location = Location::FpuRegisterLocation(fp_reg);
+          code_gen->AddAllocatedRegister(location);
+          fp_reg++;
+          break;
+        }
+      }
+    }
+    code_gen->ComputeSpillMask();
+    code_gen->SetFrameSize(frame_size);
+    code_gen->GenerateFrameEntry();
+    code_gen->GetInstructionVisitor()->VisitReturnVoid(new (&allocator) HReturnVoid());
+    // Get the outputs.
+    Assembler* opt_asm = code_gen->GetAssembler();
+    std::vector<uint8_t> actual_asm(opt_asm->CodeSize());
+    MemoryRegion code(&actual_asm[0], actual_asm.size());
+    opt_asm->FinalizeInstructions(code);
+    const std::vector<uint8_t>& actual_cfi = *(opt_asm->cfi().data());
+
+    if (kGenerateExpected) {
+      GenerateExpected(stdout, isa, isa_str, actual_asm, actual_cfi);
+    } else {
+      EXPECT_EQ(expected_asm, actual_asm);
+      EXPECT_EQ(expected_cfi, actual_cfi);
+    }
+  }
+};
+
+#define TEST_ISA(isa) \
+  TEST_F(OptimizingCFITest, isa) { \
+    std::vector<uint8_t> expected_asm(expected_asm_##isa, \
+        expected_asm_##isa + arraysize(expected_asm_##isa)); \
+    std::vector<uint8_t> expected_cfi(expected_cfi_##isa, \
+        expected_cfi_##isa + arraysize(expected_cfi_##isa)); \
+    TestImpl(isa, #isa, expected_asm, expected_cfi); \
+  }
+
+TEST_ISA(kThumb2)
+TEST_ISA(kArm64)
+TEST_ISA(kX86)
+TEST_ISA(kX86_64)
+
+#endif  // HAVE_ANDROID_OS
+
+}  // namespace art
diff --git a/compiler/optimizing/optimizing_cfi_test_expected.inc b/compiler/optimizing/optimizing_cfi_test_expected.inc
new file mode 100644
index 0000000000..2125f6eb01
--- /dev/null
+++ b/compiler/optimizing/optimizing_cfi_test_expected.inc
@@ -0,0 +1,141 @@
+static constexpr uint8_t expected_asm_kThumb2[] = {
+    0x60, 0xB5, 0x2D, 0xED, 0x02, 0x8A, 0x8B, 0xB0, 0x00, 0x90, 0x0B, 0xB0,
+    0xBD, 0xEC, 0x02, 0x8A, 0x60, 0xBD,
+};
+static constexpr uint8_t expected_cfi_kThumb2[] = {
+    0x42, 0x0E, 0x0C, 0x85, 0x03, 0x86, 0x02, 0x8E, 0x01, 0x44, 0x0E, 0x14,
+    0x05, 0x50, 0x05, 0x05, 0x51, 0x04, 0x42, 0x0E, 0x40, 0x42, 0x0A, 0x42,
+    0x0E, 0x14, 0x44, 0x0E, 0x0C, 0x06, 0x50, 0x06, 0x51, 0x42, 0x0B, 0x0E,
+    0x40,
+};
+// 0x00000000: push {r5, r6, lr}
+// 0x00000002: .cfi_def_cfa_offset: 12
+// 0x00000002: .cfi_offset: r5 at cfa-12
+// 0x00000002: .cfi_offset: r6 at cfa-8
+// 0x00000002: .cfi_offset: r14 at cfa-4
+// 0x00000002: vpush.f32 {s16-s17}
+// 0x00000006: .cfi_def_cfa_offset: 20
+// 0x00000006: .cfi_offset_extended: r80 at cfa-20
+// 0x00000006: .cfi_offset_extended: r81 at cfa-16
+// 0x00000006: sub sp, sp, #44
+// 0x00000008: .cfi_def_cfa_offset: 64
+// 0x00000008: str r0, [sp, #0]
+// 0x0000000a: .cfi_remember_state
+// 0x0000000a: add sp, sp, #44
+// 0x0000000c: .cfi_def_cfa_offset: 20
+// 0x0000000c: vpop.f32 {s16-s17}
+// 0x00000010: .cfi_def_cfa_offset: 12
+// 0x00000010: .cfi_restore_extended: r80
+// 0x00000010: .cfi_restore_extended: r81
+// 0x00000010: pop {r5, r6, pc}
+// 0x00000012: .cfi_restore_state
+// 0x00000012: .cfi_def_cfa_offset: 64
+
+static constexpr uint8_t expected_asm_kArm64[] = {
+    0xE0, 0x0F, 0x1C, 0xB8, 0xF3, 0xD3, 0x02, 0xA9, 0xFE, 0x1F, 0x00, 0xF9,
+    0xE8, 0xA7, 0x01, 0x6D, 0xE8, 0xA7, 0x41, 0x6D, 0xF3, 0xD3, 0x42, 0xA9,
+    0xFE, 0x1F, 0x40, 0xF9, 0xFF, 0x03, 0x01, 0x91, 0xC0, 0x03, 0x5F, 0xD6,
+};
+static constexpr uint8_t expected_cfi_kArm64[] = {
+    0x44, 0x0E, 0x40, 0x44, 0x93, 0x06, 0x94, 0x04, 0x44, 0x9E, 0x02, 0x44,
+    0x05, 0x48, 0x0A, 0x05, 0x49, 0x08, 0x0A, 0x44, 0x06, 0x48, 0x06, 0x49,
+    0x44, 0xD3, 0xD4, 0x44, 0xDE, 0x44, 0x0E, 0x00, 0x44, 0x0B, 0x0E, 0x40,
+};
+// 0x00000000: str w0, [sp, #-64]!
+// 0x00000004: .cfi_def_cfa_offset: 64
+// 0x00000004: stp x19, x20, [sp, #40]
+// 0x00000008: .cfi_offset: r19 at cfa-24
+// 0x00000008: .cfi_offset: r20 at cfa-16
+// 0x00000008: str lr, [sp, #56]
+// 0x0000000c: .cfi_offset: r30 at cfa-8
+// 0x0000000c: stp d8, d9, [sp, #24]
+// 0x00000010: .cfi_offset_extended: r72 at cfa-40
+// 0x00000010: .cfi_offset_extended: r73 at cfa-32
+// 0x00000010: .cfi_remember_state
+// 0x00000010: ldp d8, d9, [sp, #24]
+// 0x00000014: .cfi_restore_extended: r72
+// 0x00000014: .cfi_restore_extended: r73
+// 0x00000014: ldp x19, x20, [sp, #40]
+// 0x00000018: .cfi_restore: r19
+// 0x00000018: .cfi_restore: r20
+// 0x00000018: ldr lr, [sp, #56]
+// 0x0000001c: .cfi_restore: r30
+// 0x0000001c: add sp, sp, #0x40 (64)
+// 0x00000020: .cfi_def_cfa_offset: 0
+// 0x00000020: ret
+// 0x00000024: .cfi_restore_state
+// 0x00000024: .cfi_def_cfa_offset: 64
+
+static constexpr uint8_t expected_asm_kX86[] = {
+    0x56, 0x55, 0x83, 0xEC, 0x34, 0x89, 0x04, 0x24, 0x83, 0xC4, 0x34, 0x5D,
+    0x5E, 0xC3,
+};
+static constexpr uint8_t expected_cfi_kX86[] = {
+    0x41, 0x0E, 0x08, 0x86, 0x02, 0x41, 0x0E, 0x0C, 0x85, 0x03, 0x43, 0x0E,
+    0x40, 0x43, 0x0A, 0x43, 0x0E, 0x0C, 0x41, 0x0E, 0x08, 0xC5, 0x41, 0x0E,
+    0x04, 0xC6, 0x41, 0x0B, 0x0E, 0x40,
+};
+// 0x00000000: push esi
+// 0x00000001: .cfi_def_cfa_offset: 8
+// 0x00000001: .cfi_offset: r6 at cfa-8
+// 0x00000001: push ebp
+// 0x00000002: .cfi_def_cfa_offset: 12
+// 0x00000002: .cfi_offset: r5 at cfa-12
+// 0x00000002: sub esp, 52
+// 0x00000005: .cfi_def_cfa_offset: 64
+// 0x00000005: mov [esp], eax
+// 0x00000008: .cfi_remember_state
+// 0x00000008: add esp, 52
+// 0x0000000b: .cfi_def_cfa_offset: 12
+// 0x0000000b: pop ebp
+// 0x0000000c: .cfi_def_cfa_offset: 8
+// 0x0000000c: .cfi_restore: r5
+// 0x0000000c: pop esi
+// 0x0000000d: .cfi_def_cfa_offset: 4
+// 0x0000000d: .cfi_restore: r6
+// 0x0000000d: ret
+// 0x0000000e: .cfi_restore_state
+// 0x0000000e: .cfi_def_cfa_offset: 64
+
+static constexpr uint8_t expected_asm_kX86_64[] = {
+    0x55, 0x53, 0x48, 0x83, 0xEC, 0x28, 0xF2, 0x44, 0x0F, 0x11, 0x6C, 0x24,
+    0x20, 0xF2, 0x44, 0x0F, 0x11, 0x64, 0x24, 0x18, 0x89, 0x3C, 0x24, 0xF2,
+    0x44, 0x0F, 0x10, 0x64, 0x24, 0x18, 0xF2, 0x44, 0x0F, 0x10, 0x6C, 0x24,
+    0x20, 0x48, 0x83, 0xC4, 0x28, 0x5B, 0x5D, 0xC3,
+};
+static constexpr uint8_t expected_cfi_kX86_64[] = {
+    0x41, 0x0E, 0x10, 0x86, 0x04, 0x41, 0x0E, 0x18, 0x83, 0x06, 0x44, 0x0E,
+    0x40, 0x47, 0x9E, 0x08, 0x47, 0x9D, 0x0A, 0x43, 0x0A, 0x47, 0xDD, 0x47,
+    0xDE, 0x44, 0x0E, 0x18, 0x41, 0x0E, 0x10, 0xC3, 0x41, 0x0E, 0x08, 0xC6,
+    0x41, 0x0B, 0x0E, 0x40,
+};
+// 0x00000000: push rbp
+// 0x00000001: .cfi_def_cfa_offset: 16
+// 0x00000001: .cfi_offset: r6 at cfa-16
+// 0x00000001: push rbx
+// 0x00000002: .cfi_def_cfa_offset: 24
+// 0x00000002: .cfi_offset: r3 at cfa-24
+// 0x00000002: subq rsp, 40
+// 0x00000006: .cfi_def_cfa_offset: 64
+// 0x00000006: movsd [rsp + 32], xmm13
+// 0x0000000d: .cfi_offset: r30 at cfa-32
+// 0x0000000d: movsd [rsp + 24], xmm12
+// 0x00000014: .cfi_offset: r29 at cfa-40
+// 0x00000014: mov [rsp], edi
+// 0x00000017: .cfi_remember_state
+// 0x00000017: movsd xmm12, [rsp + 24]
+// 0x0000001e: .cfi_restore: r29
+// 0x0000001e: movsd xmm13, [rsp + 32]
+// 0x00000025: .cfi_restore: r30
+// 0x00000025: addq rsp, 40
+// 0x00000029: .cfi_def_cfa_offset: 24
+// 0x00000029: pop rbx
+// 0x0000002a: .cfi_def_cfa_offset: 16
+// 0x0000002a: .cfi_restore: r3
+// 0x0000002a: pop rbp
+// 0x0000002b: .cfi_def_cfa_offset: 8
+// 0x0000002b: .cfi_restore: r6
+// 0x0000002b: ret
+// 0x0000002c: .cfi_restore_state
+// 0x0000002c: .cfi_def_cfa_offset: 64
+
diff --git a/compiler/optimizing/optimizing_compiler.cc b/compiler/optimizing/optimizing_compiler.cc
index 12798edac5..9cb0004aa5 100644
--- a/compiler/optimizing/optimizing_compiler.cc
+++ b/compiler/optimizing/optimizing_compiler.cc
@@ -50,6 +50,7 @@
 #include "ssa_builder.h"
 #include "ssa_phi_elimination.h"
 #include "ssa_liveness_analysis.h"
+#include "utils/assembler.h"
 #include "reference_type_propagation.h"
 
 namespace art {
@@ -395,12 +396,17 @@ CompiledMethod* OptimizingCompiler::CompileOptimized(HGraph* graph,
   CodeVectorAllocator allocator;
   codegen->CompileOptimized(&allocator);
 
+  DefaultSrcMap src_mapping_table;
+  if (compiler_driver->GetCompilerOptions().GetIncludeDebugSymbols()) {
+    codegen->BuildSourceMap(&src_mapping_table);
+  }
+
   std::vector<uint8_t> stack_map;
   codegen->BuildStackMaps(&stack_map);
 
   compilation_stats_.RecordStat(MethodCompilationStat::kCompiledOptimized);
 
-  return CompiledMethod::SwapAllocCompiledMethodStackMap(
+  return CompiledMethod::SwapAllocCompiledMethod(
       compiler_driver,
       codegen->GetInstructionSet(),
       ArrayRef<const uint8_t>(allocator.GetMemory()),
@@ -410,7 +416,12 @@ CompiledMethod* OptimizingCompiler::CompileOptimized(HGraph* graph,
       codegen->HasEmptyFrame() ? 0 : codegen->GetFrameSize(),
       codegen->GetCoreSpillMask(),
       codegen->GetFpuSpillMask(),
-      ArrayRef<const uint8_t>(stack_map));
+      &src_mapping_table,
+      ArrayRef<const uint8_t>(),  // mapping_table.
+      ArrayRef<const uint8_t>(stack_map),
+      ArrayRef<const uint8_t>(),  // native_gc_map.
+      ArrayRef<const uint8_t>(*codegen->GetAssembler()->cfi().data()),
+      ArrayRef<const LinkerPatch>());
 }
 
 
@@ -422,9 +433,11 @@ CompiledMethod* OptimizingCompiler::CompileBaseline(
   codegen->CompileBaseline(&allocator);
 
   std::vector<uint8_t> mapping_table;
+  codegen->BuildMappingTable(&mapping_table);
   DefaultSrcMap src_mapping_table;
-  bool include_debug_symbol = compiler_driver->GetCompilerOptions().GetIncludeDebugSymbols();
-  codegen->BuildMappingTable(&mapping_table, include_debug_symbol ? &src_mapping_table : nullptr);
+  if (compiler_driver->GetCompilerOptions().GetIncludeDebugSymbols()) {
+    codegen->BuildSourceMap(&src_mapping_table);
+  }
   std::vector<uint8_t> vmap_table;
   codegen->BuildVMapTable(&vmap_table);
   std::vector<uint8_t> gc_map;
@@ -445,7 +458,8 @@ CompiledMethod* OptimizingCompiler::CompileBaseline(
       AlignVectorSize(mapping_table),
       AlignVectorSize(vmap_table),
       AlignVectorSize(gc_map),
-      ArrayRef<const uint8_t>());
+      ArrayRef<const uint8_t>(*codegen->GetAssembler()->cfi().data()),
+      ArrayRef<const LinkerPatch>());
 }
 
 CompiledMethod* OptimizingCompiler::TryCompile(const DexFile::CodeItem* code_item,
@@ -511,6 +525,8 @@ CompiledMethod* OptimizingCompiler::TryCompile(const DexFile::CodeItem* code_ite
     compilation_stats_.RecordStat(MethodCompilationStat::kNotCompiledNoCodegen);
     return nullptr;
   }
+  codegen->GetAssembler()->cfi().SetEnabled(
+      compiler_driver->GetCompilerOptions().GetIncludeDebugSymbols());
 
   PassInfoPrinter pass_info_printer(graph,
                                     method_name.c_str(),