Add runtime reasons for deopt.

Currently to help investigate. Also:
1) Log when deoptimization happens (which method and what reason)
2) Trace when deoptimization happens (to make it visible in systrace)

bug:37655083
Test: test-art-host test-art-target
Change-Id: I0c2d87b40db09e8e475cf97a7c784a034c585e97

31 files changed, 228 insertions, 93 deletions

diff --git a/compiler/optimizing/bounds_check_elimination.cc b/compiler/optimizing/bounds_check_elimination.cc
index d38d5f8684..ed630cda91 100644
--- a/compiler/optimizing/bounds_check_elimination.cc
+++ b/compiler/optimizing/bounds_check_elimination.cc
@@ -1550,7 +1550,7 @@ class BCEVisitor : public HGraphVisitor {
         HBasicBlock* block = GetPreHeader(loop, check);
         HInstruction* cond =
             new (GetGraph()->GetArena()) HEqual(array, GetGraph()->GetNullConstant());
-        InsertDeoptInLoop(loop, block, cond);
+        InsertDeoptInLoop(loop, block, cond, /* is_null_check */ true);
         ReplaceInstruction(check, array);
         return true;
       }
@@ -1616,11 +1616,16 @@ class BCEVisitor : public HGraphVisitor {
   }
 
   /** Inserts a deoptimization test in a loop preheader. */
-  void InsertDeoptInLoop(HLoopInformation* loop, HBasicBlock* block, HInstruction* condition) {
+  void InsertDeoptInLoop(HLoopInformation* loop,
+                         HBasicBlock* block,
+                         HInstruction* condition,
+                         bool is_null_check = false) {
     HInstruction* suspend = loop->GetSuspendCheck();
     block->InsertInstructionBefore(condition, block->GetLastInstruction());
+    DeoptimizationKind kind =
+        is_null_check ? DeoptimizationKind::kLoopNullBCE : DeoptimizationKind::kLoopBoundsBCE;
     HDeoptimize* deoptimize = new (GetGraph()->GetArena()) HDeoptimize(
-        GetGraph()->GetArena(), condition, HDeoptimize::Kind::kBCE, suspend->GetDexPc());
+        GetGraph()->GetArena(), condition, kind, suspend->GetDexPc());
     block->InsertInstructionBefore(deoptimize, block->GetLastInstruction());
     if (suspend->HasEnvironment()) {
       deoptimize->CopyEnvironmentFromWithLoopPhiAdjustment(
@@ -1633,7 +1638,7 @@ class BCEVisitor : public HGraphVisitor {
     HBasicBlock* block = bounds_check->GetBlock();
     block->InsertInstructionBefore(condition, bounds_check);
     HDeoptimize* deoptimize = new (GetGraph()->GetArena()) HDeoptimize(
-        GetGraph()->GetArena(), condition, HDeoptimize::Kind::kBCE, bounds_check->GetDexPc());
+        GetGraph()->GetArena(), condition, DeoptimizationKind::kBlockBCE, bounds_check->GetDexPc());
     block->InsertInstructionBefore(deoptimize, bounds_check);
     deoptimize->CopyEnvironmentFrom(bounds_check->GetEnvironment());
   }
diff --git a/compiler/optimizing/cha_guard_optimization.cc b/compiler/optimizing/cha_guard_optimization.cc
index 048073e37a..c806dbfef6 100644
--- a/compiler/optimizing/cha_guard_optimization.cc
+++ b/compiler/optimizing/cha_guard_optimization.cc
@@ -203,7 +203,7 @@ bool CHAGuardVisitor::HoistGuard(HShouldDeoptimizeFlag* flag,
     // Need a new deoptimize instruction that copies the environment
     // of the suspend instruction for the loop.
     HDeoptimize* deoptimize = new (GetGraph()->GetArena()) HDeoptimize(
-        GetGraph()->GetArena(), compare, HDeoptimize::Kind::kInline, suspend->GetDexPc());
+        GetGraph()->GetArena(), compare, DeoptimizationKind::kCHA, suspend->GetDexPc());
     pre_header->InsertInstructionBefore(deoptimize, pre_header->GetLastInstruction());
     deoptimize->CopyEnvironmentFromWithLoopPhiAdjustment(
         suspend->GetEnvironment(), loop_info->GetHeader());
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index ebd578c5cd..d037b94b4a 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -585,8 +585,13 @@ class DeoptimizationSlowPathARM : public SlowPathCodeARM {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorARM* arm_codegen = down_cast<CodeGeneratorARM*>(codegen);
     __ Bind(GetEntryLabel());
+    LocationSummary* locations = instruction_->GetLocations();
+    SaveLiveRegisters(codegen, locations);
+    InvokeRuntimeCallingConvention calling_convention;
+    __ LoadImmediate(calling_convention.GetRegisterAt(0),
+                     static_cast<uint32_t>(instruction_->AsDeoptimize()->GetDeoptimizationKind()));
     arm_codegen->InvokeRuntime(kQuickDeoptimize, instruction_, instruction_->GetDexPc(), this);
-    CheckEntrypointTypes<kQuickDeoptimize, void, void>();
+    CheckEntrypointTypes<kQuickDeoptimize, void, DeoptimizationKind>();
   }
 
   const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathARM"; }
@@ -2672,7 +2677,10 @@ void InstructionCodeGeneratorARM::VisitIf(HIf* if_instr) {
 void LocationsBuilderARM::VisitDeoptimize(HDeoptimize* deoptimize) {
   LocationSummary* locations = new (GetGraph()->GetArena())
       LocationSummary(deoptimize, LocationSummary::kCallOnSlowPath);
-  locations->SetCustomSlowPathCallerSaves(RegisterSet::Empty());  // No caller-save registers.
+  InvokeRuntimeCallingConvention calling_convention;
+  RegisterSet caller_saves = RegisterSet::Empty();
+  caller_saves.Add(Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+  locations->SetCustomSlowPathCallerSaves(caller_saves);
   if (IsBooleanValueOrMaterializedCondition(deoptimize->InputAt(0))) {
     locations->SetInAt(0, Location::RequiresRegister());
   }
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 78b627ad90..0c198b123d 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -587,8 +587,13 @@ class DeoptimizationSlowPathARM64 : public SlowPathCodeARM64 {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorARM64* arm64_codegen = down_cast<CodeGeneratorARM64*>(codegen);
     __ Bind(GetEntryLabel());
+    LocationSummary* locations = instruction_->GetLocations();
+    SaveLiveRegisters(codegen, locations);
+    InvokeRuntimeCallingConvention calling_convention;
+    __ Mov(calling_convention.GetRegisterAt(0),
+           static_cast<uint32_t>(instruction_->AsDeoptimize()->GetDeoptimizationKind()));
     arm64_codegen->InvokeRuntime(kQuickDeoptimize, instruction_, instruction_->GetDexPc(), this);
-    CheckEntrypointTypes<kQuickDeoptimize, void, void>();
+    CheckEntrypointTypes<kQuickDeoptimize, void, DeoptimizationKind>();
   }
 
   const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathARM64"; }
@@ -3693,7 +3698,10 @@ void InstructionCodeGeneratorARM64::VisitIf(HIf* if_instr) {
 void LocationsBuilderARM64::VisitDeoptimize(HDeoptimize* deoptimize) {
   LocationSummary* locations = new (GetGraph()->GetArena())
       LocationSummary(deoptimize, LocationSummary::kCallOnSlowPath);
-  locations->SetCustomSlowPathCallerSaves(RegisterSet::Empty());  // No caller-save registers.
+  InvokeRuntimeCallingConvention calling_convention;
+  RegisterSet caller_saves = RegisterSet::Empty();
+  caller_saves.Add(Location::RegisterLocation(calling_convention.GetRegisterAt(0).GetCode()));
+  locations->SetCustomSlowPathCallerSaves(caller_saves);
   if (IsBooleanValueOrMaterializedCondition(deoptimize->InputAt(0))) {
     locations->SetInAt(0, Location::RequiresRegister());
   }
diff --git a/compiler/optimizing/code_generator_arm_vixl.cc b/compiler/optimizing/code_generator_arm_vixl.cc
index d65b3276c8..b29f5a9d46 100644
--- a/compiler/optimizing/code_generator_arm_vixl.cc
+++ b/compiler/optimizing/code_generator_arm_vixl.cc
@@ -608,8 +608,14 @@ class DeoptimizationSlowPathARMVIXL : public SlowPathCodeARMVIXL {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorARMVIXL* arm_codegen = down_cast<CodeGeneratorARMVIXL*>(codegen);
     __ Bind(GetEntryLabel());
+        LocationSummary* locations = instruction_->GetLocations();
+    SaveLiveRegisters(codegen, locations);
+    InvokeRuntimeCallingConventionARMVIXL calling_convention;
+    __ Mov(calling_convention.GetRegisterAt(0),
+           static_cast<uint32_t>(instruction_->AsDeoptimize()->GetDeoptimizationKind()));
+
     arm_codegen->InvokeRuntime(kQuickDeoptimize, instruction_, instruction_->GetDexPc(), this);
-    CheckEntrypointTypes<kQuickDeoptimize, void, void>();
+    CheckEntrypointTypes<kQuickDeoptimize, void, DeoptimizationKind>();
   }
 
   const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathARMVIXL"; }
@@ -2704,7 +2710,10 @@ void InstructionCodeGeneratorARMVIXL::VisitIf(HIf* if_instr) {
 void LocationsBuilderARMVIXL::VisitDeoptimize(HDeoptimize* deoptimize) {
   LocationSummary* locations = new (GetGraph()->GetArena())
       LocationSummary(deoptimize, LocationSummary::kCallOnSlowPath);
-  locations->SetCustomSlowPathCallerSaves(RegisterSet::Empty());  // No caller-save registers.
+  InvokeRuntimeCallingConventionARMVIXL calling_convention;
+  RegisterSet caller_saves = RegisterSet::Empty();
+  caller_saves.Add(LocationFrom(calling_convention.GetRegisterAt(0)));
+  locations->SetCustomSlowPathCallerSaves(caller_saves);
   if (IsBooleanValueOrMaterializedCondition(deoptimize->InputAt(0))) {
     locations->SetInAt(0, Location::RequiresRegister());
   }
diff --git a/compiler/optimizing/code_generator_mips.cc b/compiler/optimizing/code_generator_mips.cc
index 357df9799c..3a7108b52e 100644
--- a/compiler/optimizing/code_generator_mips.cc
+++ b/compiler/optimizing/code_generator_mips.cc
@@ -493,8 +493,13 @@ class DeoptimizationSlowPathMIPS : public SlowPathCodeMIPS {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorMIPS* mips_codegen = down_cast<CodeGeneratorMIPS*>(codegen);
     __ Bind(GetEntryLabel());
+    LocationSummary* locations = instruction_->GetLocations();
+    SaveLiveRegisters(codegen, locations);
+    InvokeRuntimeCallingConvention calling_convention;
+    __ LoadConst32(calling_convention.GetRegisterAt(0),
+                   static_cast<uint32_t>(instruction_->AsDeoptimize()->GetDeoptimizationKind()));
     mips_codegen->InvokeRuntime(kQuickDeoptimize, instruction_, instruction_->GetDexPc(), this);
-    CheckEntrypointTypes<kQuickDeoptimize, void, void>();
+    CheckEntrypointTypes<kQuickDeoptimize, void, DeoptimizationKind>();
   }
 
   const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathMIPS"; }
@@ -5199,7 +5204,10 @@ void InstructionCodeGeneratorMIPS::VisitIf(HIf* if_instr) {
 void LocationsBuilderMIPS::VisitDeoptimize(HDeoptimize* deoptimize) {
   LocationSummary* locations = new (GetGraph()->GetArena())
       LocationSummary(deoptimize, LocationSummary::kCallOnSlowPath);
-  locations->SetCustomSlowPathCallerSaves(RegisterSet::Empty());  // No caller-save registers.
+  InvokeRuntimeCallingConvention calling_convention;
+  RegisterSet caller_saves = RegisterSet::Empty();
+  caller_saves.Add(Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+  locations->SetCustomSlowPathCallerSaves(caller_saves);
   if (IsBooleanValueOrMaterializedCondition(deoptimize->InputAt(0))) {
     locations->SetInAt(0, Location::RequiresRegister());
   }
diff --git a/compiler/optimizing/code_generator_mips64.cc b/compiler/optimizing/code_generator_mips64.cc
index a9c4964a89..f04e3841f5 100644
--- a/compiler/optimizing/code_generator_mips64.cc
+++ b/compiler/optimizing/code_generator_mips64.cc
@@ -393,8 +393,13 @@ class DeoptimizationSlowPathMIPS64 : public SlowPathCodeMIPS64 {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorMIPS64* mips64_codegen = down_cast<CodeGeneratorMIPS64*>(codegen);
     __ Bind(GetEntryLabel());
+      LocationSummary* locations = instruction_->GetLocations();
+    SaveLiveRegisters(codegen, locations);
+    InvokeRuntimeCallingConvention calling_convention;
+    __ LoadConst32(calling_convention.GetRegisterAt(0),
+                   static_cast<uint32_t>(instruction_->AsDeoptimize()->GetDeoptimizationKind()));
     mips64_codegen->InvokeRuntime(kQuickDeoptimize, instruction_, instruction_->GetDexPc(), this);
-    CheckEntrypointTypes<kQuickDeoptimize, void, void>();
+    CheckEntrypointTypes<kQuickDeoptimize, void, DeoptimizationKind>();
   }
 
   const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathMIPS64"; }
@@ -3905,7 +3910,10 @@ void InstructionCodeGeneratorMIPS64::VisitIf(HIf* if_instr) {
 void LocationsBuilderMIPS64::VisitDeoptimize(HDeoptimize* deoptimize) {
   LocationSummary* locations = new (GetGraph()->GetArena())
       LocationSummary(deoptimize, LocationSummary::kCallOnSlowPath);
-  locations->SetCustomSlowPathCallerSaves(RegisterSet::Empty());  // No caller-save registers.
+  InvokeRuntimeCallingConvention calling_convention;
+  RegisterSet caller_saves = RegisterSet::Empty();
+  caller_saves.Add(Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+  locations->SetCustomSlowPathCallerSaves(caller_saves);
   if (IsBooleanValueOrMaterializedCondition(deoptimize->InputAt(0))) {
     locations->SetInAt(0, Location::RequiresRegister());
   }
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 1e867dde51..cf2d5cbee3 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -384,8 +384,14 @@ class DeoptimizationSlowPathX86 : public SlowPathCode {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorX86* x86_codegen = down_cast<CodeGeneratorX86*>(codegen);
     __ Bind(GetEntryLabel());
+    LocationSummary* locations = instruction_->GetLocations();
+    SaveLiveRegisters(codegen, locations);
+    InvokeRuntimeCallingConvention calling_convention;
+    x86_codegen->Load32BitValue(
+        calling_convention.GetRegisterAt(0),
+        static_cast<uint32_t>(instruction_->AsDeoptimize()->GetDeoptimizationKind()));
     x86_codegen->InvokeRuntime(kQuickDeoptimize, instruction_, instruction_->GetDexPc(), this);
-    CheckEntrypointTypes<kQuickDeoptimize, void, void>();
+    CheckEntrypointTypes<kQuickDeoptimize, void, DeoptimizationKind>();
   }
 
   const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathX86"; }
@@ -1688,7 +1694,10 @@ void InstructionCodeGeneratorX86::VisitIf(HIf* if_instr) {
 void LocationsBuilderX86::VisitDeoptimize(HDeoptimize* deoptimize) {
   LocationSummary* locations = new (GetGraph()->GetArena())
       LocationSummary(deoptimize, LocationSummary::kCallOnSlowPath);
-  locations->SetCustomSlowPathCallerSaves(RegisterSet::Empty());  // No caller-save registers.
+  InvokeRuntimeCallingConvention calling_convention;
+  RegisterSet caller_saves = RegisterSet::Empty();
+  caller_saves.Add(Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+  locations->SetCustomSlowPathCallerSaves(caller_saves);
   if (IsBooleanValueOrMaterializedCondition(deoptimize->InputAt(0))) {
     locations->SetInAt(0, Location::Any());
   }
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index f413739a68..f2ed52b5a5 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -397,8 +397,14 @@ class DeoptimizationSlowPathX86_64 : public SlowPathCode {
   void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
     CodeGeneratorX86_64* x86_64_codegen = down_cast<CodeGeneratorX86_64*>(codegen);
     __ Bind(GetEntryLabel());
+    LocationSummary* locations = instruction_->GetLocations();
+    SaveLiveRegisters(codegen, locations);
+    InvokeRuntimeCallingConvention calling_convention;
+    x86_64_codegen->Load32BitValue(
+        CpuRegister(calling_convention.GetRegisterAt(0)),
+        static_cast<uint32_t>(instruction_->AsDeoptimize()->GetDeoptimizationKind()));
     x86_64_codegen->InvokeRuntime(kQuickDeoptimize, instruction_, instruction_->GetDexPc(), this);
-    CheckEntrypointTypes<kQuickDeoptimize, void, void>();
+    CheckEntrypointTypes<kQuickDeoptimize, void, DeoptimizationKind>();
   }
 
   const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathX86_64"; }
@@ -1710,7 +1716,10 @@ void InstructionCodeGeneratorX86_64::VisitIf(HIf* if_instr) {
 void LocationsBuilderX86_64::VisitDeoptimize(HDeoptimize* deoptimize) {
   LocationSummary* locations = new (GetGraph()->GetArena())
       LocationSummary(deoptimize, LocationSummary::kCallOnSlowPath);
-  locations->SetCustomSlowPathCallerSaves(RegisterSet::Empty());  // No caller-save registers.
+  InvokeRuntimeCallingConvention calling_convention;
+  RegisterSet caller_saves = RegisterSet::Empty();
+  caller_saves.Add(Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+  locations->SetCustomSlowPathCallerSaves(caller_saves);
   if (IsBooleanValueOrMaterializedCondition(deoptimize->InputAt(0))) {
     locations->SetInAt(0, Location::Any());
   }
diff --git a/compiler/optimizing/inliner.cc b/compiler/optimizing/inliner.cc
index 3b681c13a5..8674e727bb 100644
--- a/compiler/optimizing/inliner.cc
+++ b/compiler/optimizing/inliner.cc
@@ -783,7 +783,7 @@ void HInliner::AddCHAGuard(HInstruction* invoke_instruction,
   HInstruction* compare = new (graph_->GetArena()) HNotEqual(
       deopt_flag, graph_->GetIntConstant(0, dex_pc));
   HInstruction* deopt = new (graph_->GetArena()) HDeoptimize(
-      graph_->GetArena(), compare, HDeoptimize::Kind::kInline, dex_pc);
+      graph_->GetArena(), compare, DeoptimizationKind::kCHA, dex_pc);
 
   if (cursor != nullptr) {
     bb_cursor->InsertInstructionAfter(deopt_flag, cursor);
@@ -860,7 +860,9 @@ HInstruction* HInliner::AddTypeGuard(HInstruction* receiver,
         graph_->GetArena(),
         compare,
         receiver,
-        HDeoptimize::Kind::kInline,
+        Runtime::Current()->IsAotCompiler()
+            ? DeoptimizationKind::kAotInlineCache
+            : DeoptimizationKind::kJitInlineCache,
         invoke_instruction->GetDexPc());
     bb_cursor->InsertInstructionAfter(deoptimize, compare);
     deoptimize->CopyEnvironmentFrom(invoke_instruction->GetEnvironment());
@@ -1147,7 +1149,7 @@ bool HInliner::TryInlinePolymorphicCallToSameTarget(
         graph_->GetArena(),
         compare,
         receiver,
-        HDeoptimize::Kind::kInline,
+        DeoptimizationKind::kJitSameTarget,
         invoke_instruction->GetDexPc());
     bb_cursor->InsertInstructionAfter(deoptimize, compare);
     deoptimize->CopyEnvironmentFrom(invoke_instruction->GetEnvironment());
diff --git a/compiler/optimizing/nodes.cc b/compiler/optimizing/nodes.cc
index f250c1a10b..a8bfe610de 100644
--- a/compiler/optimizing/nodes.cc
+++ b/compiler/optimizing/nodes.cc
@@ -1422,18 +1422,6 @@ std::ostream& operator<<(std::ostream& os, const ComparisonBias& rhs) {
   }
 }
 
-std::ostream& operator<<(std::ostream& os, const HDeoptimize::Kind& rhs) {
-  switch (rhs) {
-    case HDeoptimize::Kind::kBCE:
-      return os << "bce";
-    case HDeoptimize::Kind::kInline:
-      return os << "inline";
-    default:
-      LOG(FATAL) << "Unknown Deoptimization kind: " << static_cast<int>(rhs);
-      UNREACHABLE();
-  }
-}
-
 bool HCondition::IsBeforeWhenDisregardMoves(HInstruction* instruction) const {
   return this == instruction->GetPreviousDisregardingMoves();
 }
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index e40361eb5d..b4da20b558 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -30,6 +30,7 @@
 #include "base/transform_array_ref.h"
 #include "dex_file.h"
 #include "dex_file_types.h"
+#include "deoptimization_kind.h"
 #include "entrypoints/quick/quick_entrypoints_enum.h"
 #include "handle.h"
 #include "handle_scope.h"
@@ -2991,15 +2992,9 @@ class HTryBoundary FINAL : public HTemplateInstruction<0> {
 // Deoptimize to interpreter, upon checking a condition.
 class HDeoptimize FINAL : public HVariableInputSizeInstruction {
  public:
-  enum class Kind {
-    kBCE,
-    kInline,
-    kLast = kInline
-  };
-
   // Use this constructor when the `HDeoptimize` acts as a barrier, where no code can move
   // across.
-  HDeoptimize(ArenaAllocator* arena, HInstruction* cond, Kind kind, uint32_t dex_pc)
+  HDeoptimize(ArenaAllocator* arena, HInstruction* cond, DeoptimizationKind kind, uint32_t dex_pc)
       : HVariableInputSizeInstruction(
             SideEffects::All(),
             dex_pc,
@@ -3019,7 +3014,7 @@ class HDeoptimize FINAL : public HVariableInputSizeInstruction {
   HDeoptimize(ArenaAllocator* arena,
               HInstruction* cond,
               HInstruction* guard,
-              Kind kind,
+              DeoptimizationKind kind,
               uint32_t dex_pc)
       : HVariableInputSizeInstruction(
             SideEffects::CanTriggerGC(),
@@ -3043,7 +3038,7 @@ class HDeoptimize FINAL : public HVariableInputSizeInstruction {
 
   bool CanThrow() const OVERRIDE { return true; }
 
-  Kind GetKind() const { return GetPackedField<DeoptimizeKindField>(); }
+  DeoptimizationKind GetDeoptimizationKind() const { return GetPackedField<DeoptimizeKindField>(); }
 
   Primitive::Type GetType() const OVERRIDE {
     return GuardsAnInput() ? GuardedInput()->GetType() : Primitive::kPrimVoid;
@@ -3068,18 +3063,17 @@ class HDeoptimize FINAL : public HVariableInputSizeInstruction {
   static constexpr size_t kFieldCanBeMoved = kNumberOfGenericPackedBits;
   static constexpr size_t kFieldDeoptimizeKind = kNumberOfGenericPackedBits + 1;
   static constexpr size_t kFieldDeoptimizeKindSize =
-      MinimumBitsToStore(static_cast<size_t>(Kind::kLast));
+      MinimumBitsToStore(static_cast<size_t>(DeoptimizationKind::kLast));
   static constexpr size_t kNumberOfDeoptimizePackedBits =
       kFieldDeoptimizeKind + kFieldDeoptimizeKindSize;
   static_assert(kNumberOfDeoptimizePackedBits <= kMaxNumberOfPackedBits,
                 "Too many packed fields.");
-  using DeoptimizeKindField = BitField<Kind, kFieldDeoptimizeKind, kFieldDeoptimizeKindSize>;
+  using DeoptimizeKindField =
+      BitField<DeoptimizationKind, kFieldDeoptimizeKind, kFieldDeoptimizeKindSize>;
 
   DISALLOW_COPY_AND_ASSIGN(HDeoptimize);
 };
 
-std::ostream& operator<<(std::ostream& os, const HDeoptimize::Kind& rhs);
-
 // Represents a should_deoptimize flag. Currently used for CHA-based devirtualization.
 // The compiled code checks this flag value in a guard before devirtualized call and
 // if it's true, starts to do deoptimization.
diff --git a/compiler/optimizing/ssa_liveness_analysis_test.cc b/compiler/optimizing/ssa_liveness_analysis_test.cc
index a1016d1d47..029eb4ba61 100644
--- a/compiler/optimizing/ssa_liveness_analysis_test.cc
+++ b/compiler/optimizing/ssa_liveness_analysis_test.cc
@@ -190,7 +190,7 @@ TEST_F(SsaLivenessAnalysisTest, TestDeoptimize) {
   HInstruction* ae = new (&allocator_) HAboveOrEqual(index, length);
   block->AddInstruction(ae);
   HInstruction* deoptimize =
-      new(&allocator_) HDeoptimize(&allocator_, ae, HDeoptimize::Kind::kBCE, /* dex_pc */ 0u);
+      new(&allocator_) HDeoptimize(&allocator_, ae, DeoptimizationKind::kBlockBCE, /* dex_pc */ 0u);
   block->AddInstruction(deoptimize);
   HEnvironment* deoptimize_env = new (&allocator_) HEnvironment(&allocator_,
                                                                 /* number_of_vregs */ 5,
diff --git a/runtime/arch/arm/quick_entrypoints_arm.S b/runtime/arch/arm/quick_entrypoints_arm.S
index a277edfa29..8b978ccb6c 100644
--- a/runtime/arch/arm/quick_entrypoints_arm.S
+++ b/runtime/arch/arm/quick_entrypoints_arm.S
@@ -1681,8 +1681,8 @@ END art_quick_instrumentation_entry
     .extern artDeoptimize
 ENTRY art_quick_deoptimize
     SETUP_SAVE_ALL_CALLEE_SAVES_FRAME r0
-    mov    r0, r9         @ Set up args.
-    blx    artDeoptimize  @ artDeoptimize(Thread*)
+    mov    r0, r9         @ pass Thread::Current
+    blx    artDeoptimize  @ (Thread*)
 END art_quick_deoptimize
 
     /*
@@ -1691,9 +1691,9 @@ END art_quick_deoptimize
      */
     .extern artDeoptimizeFromCompiledCode
 ENTRY art_quick_deoptimize_from_compiled_code
-    SETUP_SAVE_EVERYTHING_FRAME r0
-    mov    r0, r9                         @ Set up args.
-    blx    artDeoptimizeFromCompiledCode  @ artDeoptimizeFromCompiledCode(Thread*)
+    SETUP_SAVE_EVERYTHING_FRAME r1
+    mov    r1, r9                         @ pass Thread::Current
+    blx    artDeoptimizeFromCompiledCode  @ (DeoptimizationKind, Thread*)
 END art_quick_deoptimize_from_compiled_code
 
     /*
diff --git a/runtime/arch/arm64/quick_entrypoints_arm64.S b/runtime/arch/arm64/quick_entrypoints_arm64.S
index c555126668..18015b572e 100644
--- a/runtime/arch/arm64/quick_entrypoints_arm64.S
+++ b/runtime/arch/arm64/quick_entrypoints_arm64.S
@@ -2219,7 +2219,7 @@ END art_quick_instrumentation_exit
 ENTRY art_quick_deoptimize
     SETUP_SAVE_ALL_CALLEE_SAVES_FRAME
     mov    x0, xSELF          // Pass thread.
-    bl     artDeoptimize      // artDeoptimize(Thread*)
+    bl     artDeoptimize      // (Thread*)
     brk 0
 END art_quick_deoptimize
 
@@ -2230,8 +2230,8 @@ END art_quick_deoptimize
     .extern artDeoptimizeFromCompiledCode
 ENTRY art_quick_deoptimize_from_compiled_code
     SETUP_SAVE_EVERYTHING_FRAME
-    mov    x0, xSELF                      // Pass thread.
-    bl     artDeoptimizeFromCompiledCode  // artDeoptimizeFromCompiledCode(Thread*)
+    mov    x1, xSELF                      // Pass thread.
+    bl     artDeoptimizeFromCompiledCode  // (DeoptimizationKind, Thread*)
     brk 0
 END art_quick_deoptimize_from_compiled_code
 
diff --git a/runtime/arch/mips/quick_entrypoints_mips.S b/runtime/arch/mips/quick_entrypoints_mips.S
index 7bbcbf096f..e628a9f40d 100644
--- a/runtime/arch/mips/quick_entrypoints_mips.S
+++ b/runtime/arch/mips/quick_entrypoints_mips.S
@@ -1978,8 +1978,7 @@ END art_quick_instrumentation_exit
 ENTRY art_quick_deoptimize
     SETUP_SAVE_ALL_CALLEE_SAVES_FRAME
     la       $t9, artDeoptimize
-    jalr     $t9            # artDeoptimize(Thread*)
-                            # Returns caller method's frame size.
+    jalr     $t9            # (Thread*)
     move     $a0, rSELF     # pass Thread::current
 END art_quick_deoptimize
 
@@ -1991,9 +1990,8 @@ END art_quick_deoptimize
 ENTRY art_quick_deoptimize_from_compiled_code
     SETUP_SAVE_EVERYTHING_FRAME
     la       $t9, artDeoptimizeFromCompiledCode
-    jalr     $t9                            # artDeoptimizeFromCompiledCode(Thread*)
-                                            # Returns caller method's frame size.
-    move     $a0, rSELF                     # pass Thread::current
+    jalr     $t9                            # (DeoptimizationKind, Thread*)
+    move     $a1, rSELF                     # pass Thread::current
 END art_quick_deoptimize_from_compiled_code
 
     /*
diff --git a/runtime/arch/mips64/quick_entrypoints_mips64.S b/runtime/arch/mips64/quick_entrypoints_mips64.S
index 8f713a1928..40bad16291 100644
--- a/runtime/arch/mips64/quick_entrypoints_mips64.S
+++ b/runtime/arch/mips64/quick_entrypoints_mips64.S
@@ -1913,8 +1913,7 @@ END art_quick_instrumentation_exit
     .extern artEnterInterpreterFromDeoptimize
 ENTRY art_quick_deoptimize
     SETUP_SAVE_ALL_CALLEE_SAVES_FRAME
-    jal      artDeoptimize     # artDeoptimize(Thread*, SP)
-                               # Returns caller method's frame size.
+    jal      artDeoptimize     # artDeoptimize(Thread*)
     move     $a0, rSELF        # pass Thread::current
 END art_quick_deoptimize
 
@@ -1925,9 +1924,8 @@ END art_quick_deoptimize
     .extern artDeoptimizeFromCompiledCode
 ENTRY_NO_GP art_quick_deoptimize_from_compiled_code
     SETUP_SAVE_EVERYTHING_FRAME
-    jal      artDeoptimizeFromCompiledCode    # artDeoptimizeFromCompiledCode(Thread*, SP)
-                                              # Returns caller method's frame size.
-    move     $a0, rSELF                       # pass Thread::current
+    jal      artDeoptimizeFromCompiledCode    # (DeoptimizationKind, Thread*)
+    move     $a1, rSELF                       # pass Thread::current
 END art_quick_deoptimize_from_compiled_code
 
   .set push
diff --git a/runtime/arch/x86/quick_entrypoints_x86.S b/runtime/arch/x86/quick_entrypoints_x86.S
index 3694c3e7a6..2222f5cc3b 100644
--- a/runtime/arch/x86/quick_entrypoints_x86.S
+++ b/runtime/arch/x86/quick_entrypoints_x86.S
@@ -2028,7 +2028,7 @@ DEFINE_FUNCTION art_quick_deoptimize
     CFI_ADJUST_CFA_OFFSET(12)
     pushl %fs:THREAD_SELF_OFFSET  // Pass Thread::Current().
     CFI_ADJUST_CFA_OFFSET(4)
-    call SYMBOL(artDeoptimize)    // artDeoptimize(Thread*)
+    call SYMBOL(artDeoptimize)    // (Thread*)
     UNREACHABLE
 END_FUNCTION art_quick_deoptimize
 
@@ -2038,11 +2038,12 @@ END_FUNCTION art_quick_deoptimize
      */
 DEFINE_FUNCTION art_quick_deoptimize_from_compiled_code
     SETUP_SAVE_EVERYTHING_FRAME ebx, ebx
-    subl LITERAL(12), %esp                      // Align stack.
-    CFI_ADJUST_CFA_OFFSET(12)
+    subl LITERAL(8), %esp                      // Align stack.
+    CFI_ADJUST_CFA_OFFSET(8)
     pushl %fs:THREAD_SELF_OFFSET                // Pass Thread::Current().
     CFI_ADJUST_CFA_OFFSET(4)
-    call SYMBOL(artDeoptimizeFromCompiledCode)  // artDeoptimizeFromCompiledCode(Thread*)
+    PUSH eax
+    call SYMBOL(artDeoptimizeFromCompiledCode)  // (DeoptimizationKind, Thread*)
     UNREACHABLE
 END_FUNCTION art_quick_deoptimize_from_compiled_code
 
diff --git a/runtime/arch/x86_64/quick_entrypoints_x86_64.S b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
index ad7c2b3765..41651d8f1a 100644
--- a/runtime/arch/x86_64/quick_entrypoints_x86_64.S
+++ b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
@@ -1983,7 +1983,7 @@ DEFINE_FUNCTION art_quick_deoptimize
     SETUP_SAVE_ALL_CALLEE_SAVES_FRAME
                                        // Stack should be aligned now.
     movq %gs:THREAD_SELF_OFFSET, %rdi  // Pass Thread.
-    call SYMBOL(artDeoptimize)         // artDeoptimize(Thread*)
+    call SYMBOL(artDeoptimize)         // (Thread*)
     UNREACHABLE
 END_FUNCTION art_quick_deoptimize
 
@@ -1994,8 +1994,8 @@ END_FUNCTION art_quick_deoptimize
 DEFINE_FUNCTION art_quick_deoptimize_from_compiled_code
     SETUP_SAVE_EVERYTHING_FRAME
                                                 // Stack should be aligned now.
-    movq %gs:THREAD_SELF_OFFSET, %rdi           // Pass Thread.
-    call SYMBOL(artDeoptimizeFromCompiledCode)  // artDeoptimizeFromCompiledCode(Thread*)
+    movq %gs:THREAD_SELF_OFFSET, %rsi           // Pass Thread.
+    call SYMBOL(artDeoptimizeFromCompiledCode)  // (DeoptimizationKind, Thread*)
     UNREACHABLE
 END_FUNCTION art_quick_deoptimize_from_compiled_code
 
diff --git a/runtime/deoptimization_kind.h b/runtime/deoptimization_kind.h
new file mode 100644
index 0000000000..14e189c5d1
--- /dev/null
+++ b/runtime/deoptimization_kind.h
@@ -0,0 +1,53 @@
+/*
+ * Copyright (C) 2017 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.
+ */
+
+#ifndef ART_RUNTIME_DEOPTIMIZATION_KIND_H_
+#define ART_RUNTIME_DEOPTIMIZATION_KIND_H_
+
+namespace art {
+
+enum class DeoptimizationKind {
+  kAotInlineCache = 0,
+  kJitInlineCache,
+  kJitSameTarget,
+  kLoopBoundsBCE,
+  kLoopNullBCE,
+  kBlockBCE,
+  kCHA,
+  kFullFrame,
+  kLast = kFullFrame
+};
+
+inline const char* GetDeoptimizationKindName(DeoptimizationKind kind) {
+  switch (kind) {
+    case DeoptimizationKind::kAotInlineCache: return "AOT inline cache";
+    case DeoptimizationKind::kJitInlineCache: return "JIT inline cache";
+    case DeoptimizationKind::kJitSameTarget: return "JIT same target";
+    case DeoptimizationKind::kLoopBoundsBCE: return "loop bounds check elimination";
+    case DeoptimizationKind::kLoopNullBCE: return "loop bounds check elimination on null";
+    case DeoptimizationKind::kBlockBCE: return "block bounds check elimination";
+    case DeoptimizationKind::kCHA: return "class hierarchy analysis";
+    case DeoptimizationKind::kFullFrame: return "full frame";
+  }
+  LOG(FATAL) << "Unexpected kind " << static_cast<size_t>(kind);
+  UNREACHABLE();
+}
+
+std::ostream& operator<<(std::ostream& os, const DeoptimizationKind& kind);
+
+}  // namespace art
+
+#endif  // ART_RUNTIME_DEOPTIMIZATION_KIND_H_
diff --git a/runtime/entrypoints/quick/quick_deoptimization_entrypoints.cc b/runtime/entrypoints/quick/quick_deoptimization_entrypoints.cc
index 3820d854f9..5762e4f00a 100644
--- a/runtime/entrypoints/quick/quick_deoptimization_entrypoints.cc
+++ b/runtime/entrypoints/quick/quick_deoptimization_entrypoints.cc
@@ -16,6 +16,7 @@
 
 #include "base/logging.h"
 #include "base/mutex.h"
+#include "base/systrace.h"
 #include "callee_save_frame.h"
 #include "interpreter/interpreter.h"
 #include "obj_ptr-inl.h"  // TODO: Find the other include that isn't complete, and clean this up.
@@ -24,8 +25,9 @@
 
 namespace art {
 
-NO_RETURN static void artDeoptimizeImpl(Thread* self, bool single_frame)
+NO_RETURN static void artDeoptimizeImpl(Thread* self, DeoptimizationKind kind, bool single_frame)
     REQUIRES_SHARED(Locks::mutator_lock_) {
+  Runtime::Current()->IncrementDeoptimizationCount(kind);
   if (VLOG_IS_ON(deopt)) {
     if (single_frame) {
       // Deopt logging will be in DeoptimizeSingleFrame. It is there to take advantage of the
@@ -38,10 +40,13 @@ NO_RETURN static void artDeoptimizeImpl(Thread* self, bool single_frame)
 
   self->AssertHasDeoptimizationContext();
   QuickExceptionHandler exception_handler(self, true);
-  if (single_frame) {
-    exception_handler.DeoptimizeSingleFrame();
-  } else {
-    exception_handler.DeoptimizeStack();
+  {
+    ScopedTrace trace(std::string("Deoptimization ") + GetDeoptimizationKindName(kind));
+    if (single_frame) {
+      exception_handler.DeoptimizeSingleFrame(kind);
+    } else {
+      exception_handler.DeoptimizeStack();
+    }
   }
   uintptr_t return_pc = exception_handler.UpdateInstrumentationStack();
   if (exception_handler.IsFullFragmentDone()) {
@@ -57,18 +62,18 @@ NO_RETURN static void artDeoptimizeImpl(Thread* self, bool single_frame)
 
 extern "C" NO_RETURN void artDeoptimize(Thread* self) REQUIRES_SHARED(Locks::mutator_lock_) {
   ScopedQuickEntrypointChecks sqec(self);
-  artDeoptimizeImpl(self, false);
+  artDeoptimizeImpl(self, DeoptimizationKind::kFullFrame, false);
 }
 
-// This is called directly from compiled code by an HDepptimize.
-extern "C" NO_RETURN void artDeoptimizeFromCompiledCode(Thread* self)
+// This is called directly from compiled code by an HDeoptimize.
+extern "C" NO_RETURN void artDeoptimizeFromCompiledCode(DeoptimizationKind kind, Thread* self)
     REQUIRES_SHARED(Locks::mutator_lock_) {
   ScopedQuickEntrypointChecks sqec(self);
   // Before deoptimizing to interpreter, we must push the deoptimization context.
   JValue return_value;
   return_value.SetJ(0);  // we never deoptimize from compiled code with an invoke result.
   self->PushDeoptimizationContext(return_value, false, /* from_code */ true, self->GetException());
-  artDeoptimizeImpl(self, true);
+  artDeoptimizeImpl(self, kind, true);
 }
 
 }  // namespace art
diff --git a/runtime/entrypoints/quick/quick_entrypoints.h b/runtime/entrypoints/quick/quick_entrypoints.h
index 915f18ed71..6cd9dc1d71 100644
--- a/runtime/entrypoints/quick/quick_entrypoints.h
+++ b/runtime/entrypoints/quick/quick_entrypoints.h
@@ -21,6 +21,7 @@
 
 #include "base/macros.h"
 #include "base/mutex.h"
+#include "deoptimization_kind.h"
 #include "offsets.h"
 
 #define QUICK_ENTRYPOINT_OFFSET(ptr_size, x) \
diff --git a/runtime/entrypoints/quick/quick_entrypoints_list.h b/runtime/entrypoints/quick/quick_entrypoints_list.h
index e0a2e3cf8f..e2d45acb34 100644
--- a/runtime/entrypoints/quick/quick_entrypoints_list.h
+++ b/runtime/entrypoints/quick/quick_entrypoints_list.h
@@ -140,7 +140,7 @@
   V(ThrowNullPointer, void, void) \
   V(ThrowStackOverflow, void, void*) \
   V(ThrowStringBounds, void, int32_t, int32_t) \
-  V(Deoptimize, void, void) \
+  V(Deoptimize, void, DeoptimizationKind) \
 \
   V(A64Load, int64_t, volatile const int64_t *) \
   V(A64Store, void, volatile int64_t *, int64_t) \
diff --git a/runtime/entrypoints/runtime_asm_entrypoints.h b/runtime/entrypoints/runtime_asm_entrypoints.h
index 4ca52de2a2..fa287cb0ad 100644
--- a/runtime/entrypoints/runtime_asm_entrypoints.h
+++ b/runtime/entrypoints/runtime_asm_entrypoints.h
@@ -17,6 +17,8 @@
 #ifndef ART_RUNTIME_ENTRYPOINTS_RUNTIME_ASM_ENTRYPOINTS_H_
 #define ART_RUNTIME_ENTRYPOINTS_RUNTIME_ASM_ENTRYPOINTS_H_
 
+#include "deoptimization_kind.h"
+
 namespace art {
 
 #ifndef BUILDING_LIBART
@@ -77,7 +79,7 @@ static inline const void* GetQuickInstrumentationEntryPoint() {
 }
 
 // Stub to deoptimize from compiled code.
-extern "C" void art_quick_deoptimize_from_compiled_code();
+extern "C" void art_quick_deoptimize_from_compiled_code(DeoptimizationKind);
 
 // The return_pc of instrumentation exit stub.
 extern "C" void art_quick_instrumentation_exit();
diff --git a/runtime/jit/jit.cc b/runtime/jit/jit.cc
index b32b272a31..1dfb0f6e9a 100644
--- a/runtime/jit/jit.cc
+++ b/runtime/jit/jit.cc
@@ -353,6 +353,7 @@ Jit::~Jit() {
   DCHECK(!profile_saver_options_.IsEnabled() || !ProfileSaver::IsStarted());
   if (dump_info_on_shutdown_) {
     DumpInfo(LOG_STREAM(INFO));
+    Runtime::Current()->DumpDeoptimizations(LOG_STREAM(INFO));
   }
   DeleteThreadPool();
   if (jit_compiler_handle_ != nullptr) {
diff --git a/runtime/jit/jit_code_cache.cc b/runtime/jit/jit_code_cache.cc
index e9a5ae5fa9..81b87f15fd 100644
--- a/runtime/jit/jit_code_cache.cc
+++ b/runtime/jit/jit_code_cache.cc
@@ -149,7 +149,6 @@ JitCodeCache::JitCodeCache(MemMap* code_map,
       used_memory_for_code_(0),
       number_of_compilations_(0),
       number_of_osr_compilations_(0),
-      number_of_deoptimizations_(0),
       number_of_collections_(0),
       histogram_stack_map_memory_use_("Memory used for stack maps", 16),
       histogram_code_memory_use_("Memory used for compiled code", 16),
@@ -1416,8 +1415,6 @@ void JitCodeCache::InvalidateCompiledCodeFor(ArtMethod* method,
       osr_code_map_.erase(it);
     }
   }
-  MutexLock mu(Thread::Current(), lock_);
-  number_of_deoptimizations_++;
 }
 
 uint8_t* JitCodeCache::AllocateCode(size_t code_size) {
@@ -1456,7 +1453,6 @@ void JitCodeCache::Dump(std::ostream& os) {
      << "Total number of JIT compilations: " << number_of_compilations_ << "\n"
      << "Total number of JIT compilations for on stack replacement: "
         << number_of_osr_compilations_ << "\n"
-     << "Total number of deoptimizations: " << number_of_deoptimizations_ << "\n"
      << "Total number of JIT code cache collections: " << number_of_collections_ << std::endl;
   histogram_stack_map_memory_use_.PrintMemoryUse(os);
   histogram_code_memory_use_.PrintMemoryUse(os);
diff --git a/runtime/jit/jit_code_cache.h b/runtime/jit/jit_code_cache.h
index db214e7983..612d06ba1c 100644
--- a/runtime/jit/jit_code_cache.h
+++ b/runtime/jit/jit_code_cache.h
@@ -384,9 +384,6 @@ class JitCodeCache {
   // Number of compilations for on-stack-replacement done throughout the lifetime of the JIT.
   size_t number_of_osr_compilations_ GUARDED_BY(lock_);
 
-  // Number of deoptimizations done throughout the lifetime of the JIT.
-  size_t number_of_deoptimizations_ GUARDED_BY(lock_);
-
   // Number of code cache collections done throughout the lifetime of the JIT.
   size_t number_of_collections_ GUARDED_BY(lock_);
 
diff --git a/runtime/quick_exception_handler.cc b/runtime/quick_exception_handler.cc
index b8669412fc..db10103c4b 100644
--- a/runtime/quick_exception_handler.cc
+++ b/runtime/quick_exception_handler.cc
@@ -530,7 +530,7 @@ void QuickExceptionHandler::DeoptimizeStack() {
   PrepareForLongJumpToInvokeStubOrInterpreterBridge();
 }
 
-void QuickExceptionHandler::DeoptimizeSingleFrame() {
+void QuickExceptionHandler::DeoptimizeSingleFrame(DeoptimizationKind kind) {
   DCHECK(is_deoptimization_);
 
   if (VLOG_IS_ON(deopt) || kDebugExceptionDelivery) {
@@ -544,6 +544,10 @@ void QuickExceptionHandler::DeoptimizeSingleFrame() {
   // Compiled code made an explicit deoptimization.
   ArtMethod* deopt_method = visitor.GetSingleFrameDeoptMethod();
   DCHECK(deopt_method != nullptr);
+  LOG(INFO) << "Deoptimizing "
+            << deopt_method->PrettyMethod()
+            << " due to "
+            << GetDeoptimizationKindName(kind);
   if (Runtime::Current()->UseJitCompilation()) {
     Runtime::Current()->GetJit()->GetCodeCache()->InvalidateCompiledCodeFor(
         deopt_method, visitor.GetSingleFrameDeoptQuickMethodHeader());
diff --git a/runtime/quick_exception_handler.h b/runtime/quick_exception_handler.h
index 3ead7dbe64..8090f9b035 100644
--- a/runtime/quick_exception_handler.h
+++ b/runtime/quick_exception_handler.h
@@ -20,6 +20,7 @@
 #include "base/logging.h"
 #include "base/macros.h"
 #include "base/mutex.h"
+#include "deoptimization_kind.h"
 #include "stack_reference.h"
 
 namespace art {
@@ -62,7 +63,7 @@ class QuickExceptionHandler {
   //   the result of IsDeoptimizeable().
   // - It can be either full-fragment, or partial-fragment deoptimization, depending
   //   on whether that single frame covers full or partial fragment.
-  void DeoptimizeSingleFrame() REQUIRES_SHARED(Locks::mutator_lock_);
+  void DeoptimizeSingleFrame(DeoptimizationKind kind) REQUIRES_SHARED(Locks::mutator_lock_);
 
   void DeoptimizePartialFragmentFixup(uintptr_t return_pc)
       REQUIRES_SHARED(Locks::mutator_lock_);
diff --git a/runtime/runtime.cc b/runtime/runtime.cc
index b1acec6e82..60fa0828a0 100644
--- a/runtime/runtime.cc
+++ b/runtime/runtime.cc
@@ -262,6 +262,9 @@ Runtime::Runtime()
   std::fill(callee_save_methods_, callee_save_methods_ + arraysize(callee_save_methods_), 0u);
   interpreter::CheckInterpreterAsmConstants();
   callbacks_.reset(new RuntimeCallbacks());
+  for (size_t i = 0; i <= static_cast<size_t>(DeoptimizationKind::kLast); ++i) {
+    deoptimization_counts_[i] = 0u;
+  }
 }
 
 Runtime::~Runtime() {
@@ -1575,6 +1578,23 @@ void Runtime::RegisterRuntimeNativeMethods(JNIEnv* env) {
   register_sun_misc_Unsafe(env);
 }
 
+std::ostream& operator<<(std::ostream& os, const DeoptimizationKind& kind) {
+  os << GetDeoptimizationKindName(kind);
+  return os;
+}
+
+void Runtime::DumpDeoptimizations(std::ostream& os) {
+  for (size_t i = 0; i <= static_cast<size_t>(DeoptimizationKind::kLast); ++i) {
+    if (deoptimization_counts_[i] != 0) {
+      os << "Number of "
+         << GetDeoptimizationKindName(static_cast<DeoptimizationKind>(i))
+         << " deoptimizations: "
+         << deoptimization_counts_[i]
+         << "\n";
+    }
+  }
+}
+
 void Runtime::DumpForSigQuit(std::ostream& os) {
   GetClassLinker()->DumpForSigQuit(os);
   GetInternTable()->DumpForSigQuit(os);
@@ -1586,6 +1606,7 @@ void Runtime::DumpForSigQuit(std::ostream& os) {
   } else {
     os << "Running non JIT\n";
   }
+  DumpDeoptimizations(os);
   TrackedAllocators::Dump(os);
   os << "\n";
 
diff --git a/runtime/runtime.h b/runtime/runtime.h
index 3ba0f2cd42..a2505e2292 100644
--- a/runtime/runtime.h
+++ b/runtime/runtime.h
@@ -29,6 +29,7 @@
 #include "arch/instruction_set.h"
 #include "base/macros.h"
 #include "base/mutex.h"
+#include "deoptimization_kind.h"
 #include "dex_file_types.h"
 #include "experimental_flags.h"
 #include "gc_root.h"
@@ -235,6 +236,7 @@ class Runtime {
   // Detaches the current native thread from the runtime.
   void DetachCurrentThread() REQUIRES(!Locks::mutator_lock_);
 
+  void DumpDeoptimizations(std::ostream& os);
   void DumpForSigQuit(std::ostream& os);
   void DumpLockHolders(std::ostream& os);
 
@@ -682,6 +684,11 @@ class Runtime {
     dump_gc_performance_on_shutdown_ = value;
   }
 
+  void IncrementDeoptimizationCount(DeoptimizationKind kind) {
+    DCHECK_LE(kind, DeoptimizationKind::kLast);
+    deoptimization_counts_[static_cast<size_t>(kind)]++;
+  }
+
  private:
   static void InitPlatformSignalHandlers();
 
@@ -941,6 +948,8 @@ class Runtime {
 
   std::unique_ptr<RuntimeCallbacks> callbacks_;
 
+  std::atomic<uint32_t> deoptimization_counts_[static_cast<uint32_t>(DeoptimizationKind::kLast)];
+
   DISALLOW_COPY_AND_ASSIGN(Runtime);
 };
 std::ostream& operator<<(std::ostream& os, const Runtime::CalleeSaveType& rhs);