Inline unimplemented intrinsics

There are intrinsics that are unimplemented i.e. we didn't
hand-craft code for them. Allow the inliner to inline those.

Since our optimizations expect InvokeVirtual, I stopped the
de-virtualization of intrinsics. That could be re-added
if we modify optimizations like TryReplaceStringBuilderAppend.

Test: art/test/testrunner/testrunner.py --host --64 --optimizing -b
Test: art/test/testrunner/testrunner.py --target --64 --optimizing
Test: compiling the APK in the bug and seeing the inline
Bug: 262585898
Fixes: 262585898
Change-Id: I501b69c4ffd9082ca8ffacb1cd1cd5d1ab3668a8

15 files changed, 348 insertions, 217 deletions

diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index c1bc32aa1e..c9f42b52f5 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -1055,7 +1055,8 @@ CodeGenerator::CodeGenerator(HGraph* graph,
                              uint32_t core_callee_save_mask,
                              uint32_t fpu_callee_save_mask,
                              const CompilerOptions& compiler_options,
-                             OptimizingCompilerStats* stats)
+                             OptimizingCompilerStats* stats,
+                             const art::ArrayRef<const bool>& unimplemented_intrinsics)
     : frame_size_(0),
       core_spill_mask_(0),
       fpu_spill_mask_(0),
@@ -1080,7 +1081,8 @@ CodeGenerator::CodeGenerator(HGraph* graph,
       is_leaf_(true),
       needs_suspend_check_entry_(false),
       requires_current_method_(false),
-      code_generation_data_() {
+      code_generation_data_(),
+      unimplemented_intrinsics_(unimplemented_intrinsics) {
   if (GetGraph()->IsCompilingOsr()) {
     // Make OSR methods have all registers spilled, this simplifies the logic of
     // jumping to the compiled code directly.
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index ee80357791..9872efaa4a 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -304,6 +304,12 @@ class CodeGenerator : public DeletableArenaObject<kArenaAllocCodeGenerator> {
   // Returns whether we should split long moves in parallel moves.
   virtual bool ShouldSplitLongMoves() const { return false; }
 
+  // Returns true if `invoke` is an implemented intrinsic in this codegen's arch.
+  bool IsImplementedIntrinsic(HInvoke* invoke) const {
+    return invoke->IsIntrinsic() &&
+           !unimplemented_intrinsics_[static_cast<size_t>(invoke->GetIntrinsic())];
+  }
+
   size_t GetNumberOfCoreCalleeSaveRegisters() const {
     return POPCOUNT(core_callee_save_mask_);
   }
@@ -749,7 +755,8 @@ class CodeGenerator : public DeletableArenaObject<kArenaAllocCodeGenerator> {
                 uint32_t core_callee_save_mask,
                 uint32_t fpu_callee_save_mask,
                 const CompilerOptions& compiler_options,
-                OptimizingCompilerStats* stats);
+                OptimizingCompilerStats* stats,
+                const art::ArrayRef<const bool>& unimplemented_intrinsics);
 
   virtual HGraphVisitor* GetLocationBuilder() = 0;
   virtual HGraphVisitor* GetInstructionVisitor() = 0;
@@ -893,6 +900,9 @@ class CodeGenerator : public DeletableArenaObject<kArenaAllocCodeGenerator> {
   // CodeGenerator::Compile() and remains alive until the CodeGenerator is destroyed.
   std::unique_ptr<CodeGenerationData> code_generation_data_;
 
+  // Which intrinsics we don't have handcrafted code for.
+  art::ArrayRef<const bool> unimplemented_intrinsics_;
+
   friend class OptimizingCFITest;
   ART_FRIEND_TEST(CodegenTest, ARM64FrameSizeSIMD);
   ART_FRIEND_TEST(CodegenTest, ARM64FrameSizeNoSIMD);
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 079c440033..03b2f9eff2 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -936,6 +936,33 @@ Location CriticalNativeCallingConventionVisitorARM64::GetMethodLocation() const
   return Location::RegisterLocation(x15.GetCode());
 }
 
+namespace detail {
+// Mark which intrinsics we don't have handcrafted code for.
+template <Intrinsics T>
+struct IsUnimplemented {
+  bool is_unimplemented = false;
+};
+
+#define TRUE_OVERRIDE(Name)                     \
+  template <>                                   \
+  struct IsUnimplemented<Intrinsics::k##Name> { \
+    bool is_unimplemented = true;               \
+  };
+UNIMPLEMENTED_INTRINSIC_LIST_ARM64(TRUE_OVERRIDE)
+#undef TRUE_OVERRIDE
+
+#include "intrinsics_list.h"
+static constexpr bool kIsIntrinsicUnimplemented[] = {
+  false,  // kNone
+#define IS_UNIMPLEMENTED(Intrinsic, ...) \
+  IsUnimplemented<Intrinsics::k##Intrinsic>().is_unimplemented,
+  INTRINSICS_LIST(IS_UNIMPLEMENTED)
+#undef IS_UNIMPLEMENTED
+};
+#undef INTRINSICS_LIST
+
+}  // namespace detail
+
 CodeGeneratorARM64::CodeGeneratorARM64(HGraph* graph,
                                        const CompilerOptions& compiler_options,
                                        OptimizingCompilerStats* stats)
@@ -946,7 +973,8 @@ CodeGeneratorARM64::CodeGeneratorARM64(HGraph* graph,
                     callee_saved_core_registers.GetList(),
                     callee_saved_fp_registers.GetList(),
                     compiler_options,
-                    stats),
+                    stats,
+                    ArrayRef<const bool>(detail::kIsIntrinsicUnimplemented)),
       block_labels_(graph->GetAllocator()->Adapter(kArenaAllocCodeGenerator)),
       jump_tables_(graph->GetAllocator()->Adapter(kArenaAllocCodeGenerator)),
       location_builder_neon_(graph, this),
diff --git a/compiler/optimizing/code_generator_arm64.h b/compiler/optimizing/code_generator_arm64.h
index deba88b860..6190364d1d 100644
--- a/compiler/optimizing/code_generator_arm64.h
+++ b/compiler/optimizing/code_generator_arm64.h
@@ -123,6 +123,41 @@ const vixl::aarch64::CPURegList callee_saved_fp_registers(vixl::aarch64::CPURegi
                                                           vixl::aarch64::d15.GetCode());
 Location ARM64ReturnLocation(DataType::Type return_type);
 
+#define UNIMPLEMENTED_INTRINSIC_LIST_ARM64(V) \
+  V(StringStringIndexOf)                      \
+  V(StringStringIndexOfAfter)                 \
+  V(StringBufferAppend)                       \
+  V(StringBufferLength)                       \
+  V(StringBufferToString)                     \
+  V(StringBuilderAppendObject)                \
+  V(StringBuilderAppendString)                \
+  V(StringBuilderAppendCharSequence)          \
+  V(StringBuilderAppendCharArray)             \
+  V(StringBuilderAppendBoolean)               \
+  V(StringBuilderAppendChar)                  \
+  V(StringBuilderAppendInt)                   \
+  V(StringBuilderAppendLong)                  \
+  V(StringBuilderAppendFloat)                 \
+  V(StringBuilderAppendDouble)                \
+  V(StringBuilderLength)                      \
+  V(StringBuilderToString)                    \
+  V(SystemArrayCopyByte)                      \
+  V(SystemArrayCopyInt)                       \
+  /* 1.8 */                                   \
+  V(UnsafeGetAndAddInt)                       \
+  V(UnsafeGetAndAddLong)                      \
+  V(UnsafeGetAndSetInt)                       \
+  V(UnsafeGetAndSetLong)                      \
+  V(UnsafeGetAndSetObject)                    \
+  V(MethodHandleInvokeExact)                  \
+  V(MethodHandleInvoke)                       \
+  /* OpenJDK 11 */                            \
+  V(JdkUnsafeGetAndAddInt)                    \
+  V(JdkUnsafeGetAndAddLong)                   \
+  V(JdkUnsafeGetAndSetInt)                    \
+  V(JdkUnsafeGetAndSetLong)                   \
+  V(JdkUnsafeGetAndSetObject)
+
 class SlowPathCodeARM64 : public SlowPathCode {
  public:
   explicit SlowPathCodeARM64(HInstruction* instruction)
diff --git a/compiler/optimizing/code_generator_arm_vixl.cc b/compiler/optimizing/code_generator_arm_vixl.cc
index 51d6a46ddb..cc34e76861 100644
--- a/compiler/optimizing/code_generator_arm_vixl.cc
+++ b/compiler/optimizing/code_generator_arm_vixl.cc
@@ -1907,6 +1907,33 @@ vixl32::Label* CodeGeneratorARMVIXL::GetFinalLabel(HInstruction* instruction,
   return final_label;
 }
 
+namespace detail {
+// Mark which intrinsics we don't have handcrafted code for.
+template <Intrinsics T>
+struct IsUnimplemented {
+  bool is_unimplemented = false;
+};
+
+#define TRUE_OVERRIDE(Name)                     \
+  template <>                                   \
+  struct IsUnimplemented<Intrinsics::k##Name> { \
+    bool is_unimplemented = true;               \
+  };
+UNIMPLEMENTED_INTRINSIC_LIST_ARM(TRUE_OVERRIDE)
+#undef TRUE_OVERRIDE
+
+#include "intrinsics_list.h"
+static constexpr bool kIsIntrinsicUnimplemented[] = {
+  false,  // kNone
+#define IS_UNIMPLEMENTED(Intrinsic, ...) \
+  IsUnimplemented<Intrinsics::k##Intrinsic>().is_unimplemented,
+  INTRINSICS_LIST(IS_UNIMPLEMENTED)
+#undef IS_UNIMPLEMENTED
+};
+#undef INTRINSICS_LIST
+
+}  // namespace detail
+
 CodeGeneratorARMVIXL::CodeGeneratorARMVIXL(HGraph* graph,
                                            const CompilerOptions& compiler_options,
                                            OptimizingCompilerStats* stats)
@@ -1917,7 +1944,8 @@ CodeGeneratorARMVIXL::CodeGeneratorARMVIXL(HGraph* graph,
                     kCoreCalleeSaves.GetList(),
                     ComputeSRegisterListMask(kFpuCalleeSaves),
                     compiler_options,
-                    stats),
+                    stats,
+                    ArrayRef<const bool>(detail::kIsIntrinsicUnimplemented)),
       block_labels_(graph->GetAllocator()->Adapter(kArenaAllocCodeGenerator)),
       jump_tables_(graph->GetAllocator()->Adapter(kArenaAllocCodeGenerator)),
       location_builder_(graph, this),
diff --git a/compiler/optimizing/code_generator_arm_vixl.h b/compiler/optimizing/code_generator_arm_vixl.h
index 872a17b285..f5abe6951a 100644
--- a/compiler/optimizing/code_generator_arm_vixl.h
+++ b/compiler/optimizing/code_generator_arm_vixl.h
@@ -119,6 +119,65 @@ class CodeGeneratorARMVIXL;
 using VIXLInt32Literal = vixl::aarch32::Literal<int32_t>;
 using VIXLUInt32Literal = vixl::aarch32::Literal<uint32_t>;
 
+#define UNIMPLEMENTED_INTRINSIC_LIST_ARM(V)                                \
+  V(MathRoundDouble) /* Could be done by changing rounding mode, maybe? */ \
+  V(UnsafeCASLong)   /* High register pressure */                          \
+  V(SystemArrayCopyChar)                                                   \
+  V(LongDivideUnsigned)                                                    \
+  V(CRC32Update)                                                           \
+  V(CRC32UpdateBytes)                                                      \
+  V(CRC32UpdateByteBuffer)                                                 \
+  V(FP16ToFloat)                                                           \
+  V(FP16ToHalf)                                                            \
+  V(FP16Floor)                                                             \
+  V(FP16Ceil)                                                              \
+  V(FP16Rint)                                                              \
+  V(FP16Greater)                                                           \
+  V(FP16GreaterEquals)                                                     \
+  V(FP16Less)                                                              \
+  V(FP16LessEquals)                                                        \
+  V(FP16Compare)                                                           \
+  V(FP16Min)                                                               \
+  V(FP16Max)                                                               \
+  V(MathMultiplyHigh)                                                      \
+  V(StringStringIndexOf)                                                   \
+  V(StringStringIndexOfAfter)                                              \
+  V(StringBufferAppend)                                                    \
+  V(StringBufferLength)                                                    \
+  V(StringBufferToString)                                                  \
+  V(StringBuilderAppendObject)                                             \
+  V(StringBuilderAppendString)                                             \
+  V(StringBuilderAppendCharSequence)                                       \
+  V(StringBuilderAppendCharArray)                                          \
+  V(StringBuilderAppendBoolean)                                            \
+  V(StringBuilderAppendChar)                                               \
+  V(StringBuilderAppendInt)                                                \
+  V(StringBuilderAppendLong)                                               \
+  V(StringBuilderAppendFloat)                                              \
+  V(StringBuilderAppendDouble)                                             \
+  V(StringBuilderLength)                                                   \
+  V(StringBuilderToString)                                                 \
+  V(SystemArrayCopyByte)                                                   \
+  V(SystemArrayCopyInt)                                                    \
+  /* 1.8 */                                                                \
+  V(MathFmaDouble)                                                         \
+  V(MathFmaFloat)                                                          \
+  V(UnsafeGetAndAddInt)                                                    \
+  V(UnsafeGetAndAddLong)                                                   \
+  V(UnsafeGetAndSetInt)                                                    \
+  V(UnsafeGetAndSetLong)                                                   \
+  V(UnsafeGetAndSetObject)                                                 \
+  V(MethodHandleInvokeExact)                                               \
+  V(MethodHandleInvoke)                                                    \
+  /* OpenJDK 11 */                                                         \
+  V(JdkUnsafeCASLong) /* High register pressure */                         \
+  V(JdkUnsafeGetAndAddInt)                                                 \
+  V(JdkUnsafeGetAndAddLong)                                                \
+  V(JdkUnsafeGetAndSetInt)                                                 \
+  V(JdkUnsafeGetAndSetLong)                                                \
+  V(JdkUnsafeGetAndSetObject)                                              \
+  V(JdkUnsafeCompareAndSetLong)
+
 class JumpTableARMVIXL : public DeletableArenaObject<kArenaAllocSwitchTable> {
  public:
   explicit JumpTableARMVIXL(HPackedSwitch* switch_instr)
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 58cb56d243..0cbdbe391d 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -1106,6 +1106,33 @@ void CodeGeneratorX86::GenerateInvokeRuntime(int32_t entry_point_offset) {
   __ fs()->call(Address::Absolute(entry_point_offset));
 }
 
+namespace detail {
+// Mark which intrinsics we don't have handcrafted code for.
+template <Intrinsics T>
+struct IsUnimplemented {
+  bool is_unimplemented = false;
+};
+
+#define TRUE_OVERRIDE(Name)                     \
+  template <>                                   \
+  struct IsUnimplemented<Intrinsics::k##Name> { \
+    bool is_unimplemented = true;               \
+  };
+UNIMPLEMENTED_INTRINSIC_LIST_X86(TRUE_OVERRIDE)
+#undef TRUE_OVERRIDE
+
+#include "intrinsics_list.h"
+static constexpr bool kIsIntrinsicUnimplemented[] = {
+  false,  // kNone
+#define IS_UNIMPLEMENTED(Intrinsic, ...) \
+  IsUnimplemented<Intrinsics::k##Intrinsic>().is_unimplemented,
+  INTRINSICS_LIST(IS_UNIMPLEMENTED)
+#undef IS_UNIMPLEMENTED
+};
+#undef INTRINSICS_LIST
+
+}  // namespace detail
+
 CodeGeneratorX86::CodeGeneratorX86(HGraph* graph,
                                    const CompilerOptions& compiler_options,
                                    OptimizingCompilerStats* stats)
@@ -1118,7 +1145,8 @@ CodeGeneratorX86::CodeGeneratorX86(HGraph* graph,
                         | (1 << kFakeReturnRegister),
                     0,
                     compiler_options,
-                    stats),
+                    stats,
+                    ArrayRef<const bool>(detail::kIsIntrinsicUnimplemented)),
       block_labels_(nullptr),
       location_builder_(graph, this),
       instruction_visitor_(graph, this),
diff --git a/compiler/optimizing/code_generator_x86.h b/compiler/optimizing/code_generator_x86.h
index 9f09e171fe..d27155f31d 100644
--- a/compiler/optimizing/code_generator_x86.h
+++ b/compiler/optimizing/code_generator_x86.h
@@ -48,6 +48,61 @@ static constexpr XmmRegister kRuntimeParameterFpuRegisters[] = { XMM0, XMM1, XMM
 static constexpr size_t kRuntimeParameterFpuRegistersLength =
     arraysize(kRuntimeParameterFpuRegisters);
 
+#define UNIMPLEMENTED_INTRINSIC_LIST_X86(V) \
+  V(MathRoundDouble)                        \
+  V(FloatIsInfinite)                        \
+  V(DoubleIsInfinite)                       \
+  V(IntegerHighestOneBit)                   \
+  V(LongHighestOneBit)                      \
+  V(LongDivideUnsigned)                     \
+  V(CRC32Update)                            \
+  V(CRC32UpdateBytes)                       \
+  V(CRC32UpdateByteBuffer)                  \
+  V(FP16ToFloat)                            \
+  V(FP16ToHalf)                             \
+  V(FP16Floor)                              \
+  V(FP16Ceil)                               \
+  V(FP16Rint)                               \
+  V(FP16Greater)                            \
+  V(FP16GreaterEquals)                      \
+  V(FP16Less)                               \
+  V(FP16LessEquals)                         \
+  V(FP16Compare)                            \
+  V(FP16Min)                                \
+  V(FP16Max)                                \
+  V(MathMultiplyHigh)                       \
+  V(StringStringIndexOf)                    \
+  V(StringStringIndexOfAfter)               \
+  V(StringBufferAppend)                     \
+  V(StringBufferLength)                     \
+  V(StringBufferToString)                   \
+  V(StringBuilderAppendObject)              \
+  V(StringBuilderAppendString)              \
+  V(StringBuilderAppendCharSequence)        \
+  V(StringBuilderAppendCharArray)           \
+  V(StringBuilderAppendBoolean)             \
+  V(StringBuilderAppendChar)                \
+  V(StringBuilderAppendInt)                 \
+  V(StringBuilderAppendLong)                \
+  V(StringBuilderAppendFloat)               \
+  V(StringBuilderAppendDouble)              \
+  V(StringBuilderLength)                    \
+  V(StringBuilderToString)                  \
+  /* 1.8 */                                 \
+  V(UnsafeGetAndAddInt)                     \
+  V(UnsafeGetAndAddLong)                    \
+  V(UnsafeGetAndSetInt)                     \
+  V(UnsafeGetAndSetLong)                    \
+  V(UnsafeGetAndSetObject)                  \
+  V(MethodHandleInvokeExact)                \
+  V(MethodHandleInvoke)                     \
+  /* OpenJDK 11 */                          \
+  V(JdkUnsafeGetAndAddInt)                  \
+  V(JdkUnsafeGetAndAddLong)                 \
+  V(JdkUnsafeGetAndSetInt)                  \
+  V(JdkUnsafeGetAndSetLong)                 \
+  V(JdkUnsafeGetAndSetObject)
+
 class InvokeRuntimeCallingConvention : public CallingConvention<Register, XmmRegister> {
  public:
   InvokeRuntimeCallingConvention()
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index 2d7dc441ea..47de888f32 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -1494,6 +1494,33 @@ void CodeGeneratorX86_64::GenerateInvokeRuntime(int32_t entry_point_offset) {
   __ gs()->call(Address::Absolute(entry_point_offset, /* no_rip= */ true));
 }
 
+namespace detail {
+// Mark which intrinsics we don't have handcrafted code for.
+template <Intrinsics T>
+struct IsUnimplemented {
+  bool is_unimplemented = false;
+};
+
+#define TRUE_OVERRIDE(Name)                     \
+  template <>                                   \
+  struct IsUnimplemented<Intrinsics::k##Name> { \
+    bool is_unimplemented = true;               \
+  };
+UNIMPLEMENTED_INTRINSIC_LIST_X86_64(TRUE_OVERRIDE)
+#undef TRUE_OVERRIDE
+
+#include "intrinsics_list.h"
+static constexpr bool kIsIntrinsicUnimplemented[] = {
+  false,  // kNone
+#define IS_UNIMPLEMENTED(Intrinsic, ...) \
+  IsUnimplemented<Intrinsics::k##Intrinsic>().is_unimplemented,
+  INTRINSICS_LIST(IS_UNIMPLEMENTED)
+#undef IS_UNIMPLEMENTED
+};
+#undef INTRINSICS_LIST
+
+}  // namespace detail
+
 static constexpr int kNumberOfCpuRegisterPairs = 0;
 // Use a fake return address register to mimic Quick.
 static constexpr Register kFakeReturnRegister = Register(kLastCpuRegister + 1);
@@ -1510,7 +1537,8 @@ CodeGeneratorX86_64::CodeGeneratorX86_64(HGraph* graph,
                     ComputeRegisterMask(reinterpret_cast<const int*>(kFpuCalleeSaves),
                                         arraysize(kFpuCalleeSaves)),
                     compiler_options,
-                    stats),
+                    stats,
+                    ArrayRef<const bool>(detail::kIsIntrinsicUnimplemented)),
       block_labels_(nullptr),
       location_builder_(graph, this),
       instruction_visitor_(graph, this),
diff --git a/compiler/optimizing/code_generator_x86_64.h b/compiler/optimizing/code_generator_x86_64.h
index 1fac62fa34..dff2e799e0 100644
--- a/compiler/optimizing/code_generator_x86_64.h
+++ b/compiler/optimizing/code_generator_x86_64.h
@@ -53,6 +53,53 @@ static constexpr size_t kRuntimeParameterFpuRegistersLength =
 // these are not clobbered by any direct call to native code (such as math intrinsics).
 static constexpr FloatRegister non_volatile_xmm_regs[] = { XMM12, XMM13, XMM14, XMM15 };
 
+#define UNIMPLEMENTED_INTRINSIC_LIST_X86_64(V) \
+  V(CRC32Update)                               \
+  V(CRC32UpdateBytes)                          \
+  V(CRC32UpdateByteBuffer)                     \
+  V(FP16ToFloat)                               \
+  V(FP16ToHalf)                                \
+  V(FP16Floor)                                 \
+  V(FP16Ceil)                                  \
+  V(FP16Rint)                                  \
+  V(FP16Greater)                               \
+  V(FP16GreaterEquals)                         \
+  V(FP16Less)                                  \
+  V(FP16LessEquals)                            \
+  V(FP16Compare)                               \
+  V(FP16Min)                                   \
+  V(FP16Max)                                   \
+  V(StringStringIndexOf)                       \
+  V(StringStringIndexOfAfter)                  \
+  V(StringBufferAppend)                        \
+  V(StringBufferLength)                        \
+  V(StringBufferToString)                      \
+  V(StringBuilderAppendObject)                 \
+  V(StringBuilderAppendString)                 \
+  V(StringBuilderAppendCharSequence)           \
+  V(StringBuilderAppendCharArray)              \
+  V(StringBuilderAppendBoolean)                \
+  V(StringBuilderAppendChar)                   \
+  V(StringBuilderAppendInt)                    \
+  V(StringBuilderAppendLong)                   \
+  V(StringBuilderAppendFloat)                  \
+  V(StringBuilderAppendDouble)                 \
+  V(StringBuilderLength)                       \
+  V(StringBuilderToString)                     \
+  /* 1.8 */                                    \
+  V(UnsafeGetAndAddInt)                        \
+  V(UnsafeGetAndAddLong)                       \
+  V(UnsafeGetAndSetInt)                        \
+  V(UnsafeGetAndSetLong)                       \
+  V(UnsafeGetAndSetObject)                     \
+  V(MethodHandleInvokeExact)                   \
+  V(MethodHandleInvoke)                        \
+  /* OpenJDK 11 */                             \
+  V(JdkUnsafeGetAndAddInt)                     \
+  V(JdkUnsafeGetAndAddLong)                    \
+  V(JdkUnsafeGetAndSetInt)                     \
+  V(JdkUnsafeGetAndSetLong)                    \
+  V(JdkUnsafeGetAndSetObject)
 
 class InvokeRuntimeCallingConvention : public CallingConvention<Register, FloatRegister> {
  public:
diff --git a/compiler/optimizing/inliner.cc b/compiler/optimizing/inliner.cc
index c24ac84d37..41dc5eb206 100644
--- a/compiler/optimizing/inliner.cc
+++ b/compiler/optimizing/inliner.cc
@@ -182,7 +182,7 @@ bool HInliner::Run() {
       HInstruction* next = instruction->GetNext();
       HInvoke* call = instruction->AsInvoke();
       // As long as the call is not intrinsified, it is worth trying to inline.
-      if (call != nullptr && call->GetIntrinsic() == Intrinsics::kNone) {
+      if (call != nullptr && !codegen_->IsImplementedIntrinsic(call)) {
         if (honor_noinline_directives) {
           // Debugging case: directives in method names control or assert on inlining.
           std::string callee_name =
@@ -1272,6 +1272,13 @@ bool HInliner::TryDevirtualize(HInvoke* invoke_instruction,
     return false;
   }
 
+  // Don't try to devirtualize intrinsics as it breaks pattern matching from later phases.
+  // TODO(solanes): This `if` could be removed if we update optimizations like
+  // TryReplaceStringBuilderAppend.
+  if (invoke_instruction->IsIntrinsic()) {
+    return false;
+  }
+
   // Don't bother trying to call directly a default conflict method. It
   // doesn't have a proper MethodReference, but also `GetCanonicalMethod`
   // will return an actual default implementation.
@@ -1344,7 +1351,7 @@ bool HInliner::TryInlineAndReplace(HInvoke* invoke_instruction,
                                    ReferenceTypeInfo receiver_type,
                                    bool do_rtp,
                                    bool is_speculative) {
-  DCHECK(!invoke_instruction->IsIntrinsic());
+  DCHECK(!codegen_->IsImplementedIntrinsic(invoke_instruction));
   HInstruction* return_replacement = nullptr;
 
   if (!TryBuildAndInline(
diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index 946d4348af..a1f79ed70b 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -5761,42 +5761,9 @@ void VarHandleSlowPathARM64::EmitByteArrayViewCode(CodeGenerator* codegen_in) {
   __ B(GetExitLabel());
 }
 
-UNIMPLEMENTED_INTRINSIC(ARM64, StringStringIndexOf);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringStringIndexOfAfter);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBufferAppend);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBufferLength);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBufferToString);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderAppendObject);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderAppendString);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderAppendCharSequence);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderAppendCharArray);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderAppendBoolean);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderAppendChar);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderAppendInt);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderAppendLong);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderAppendFloat);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderAppendDouble);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderLength);
-UNIMPLEMENTED_INTRINSIC(ARM64, StringBuilderToString);
-UNIMPLEMENTED_INTRINSIC(ARM64, SystemArrayCopyByte);
-UNIMPLEMENTED_INTRINSIC(ARM64, SystemArrayCopyInt);
-
-// 1.8.
-UNIMPLEMENTED_INTRINSIC(ARM64, UnsafeGetAndAddInt)
-UNIMPLEMENTED_INTRINSIC(ARM64, UnsafeGetAndAddLong)
-UNIMPLEMENTED_INTRINSIC(ARM64, UnsafeGetAndSetInt)
-UNIMPLEMENTED_INTRINSIC(ARM64, UnsafeGetAndSetLong)
-UNIMPLEMENTED_INTRINSIC(ARM64, UnsafeGetAndSetObject)
-
-UNIMPLEMENTED_INTRINSIC(ARM64, MethodHandleInvokeExact)
-UNIMPLEMENTED_INTRINSIC(ARM64, MethodHandleInvoke)
-
-// OpenJDK 11
-UNIMPLEMENTED_INTRINSIC(ARM64, JdkUnsafeGetAndAddInt)
-UNIMPLEMENTED_INTRINSIC(ARM64, JdkUnsafeGetAndAddLong)
-UNIMPLEMENTED_INTRINSIC(ARM64, JdkUnsafeGetAndSetInt)
-UNIMPLEMENTED_INTRINSIC(ARM64, JdkUnsafeGetAndSetLong)
-UNIMPLEMENTED_INTRINSIC(ARM64, JdkUnsafeGetAndSetObject)
+#define MARK_UNIMPLEMENTED(Name) UNIMPLEMENTED_INTRINSIC(ARM64, Name)
+UNIMPLEMENTED_INTRINSIC_LIST_ARM64(MARK_UNIMPLEMENTED);
+#undef MARK_UNIMPLEMENTED
 
 UNREACHABLE_INTRINSICS(ARM64)
 
diff --git a/compiler/optimizing/intrinsics_arm_vixl.cc b/compiler/optimizing/intrinsics_arm_vixl.cc
index 55d51fe6ef..4df1088e20 100644
--- a/compiler/optimizing/intrinsics_arm_vixl.cc
+++ b/compiler/optimizing/intrinsics_arm_vixl.cc
@@ -5549,69 +5549,9 @@ void VarHandleSlowPathARMVIXL::EmitByteArrayViewCode(CodeGenerator* codegen_in)
   __ B(GetExitLabel());
 }
 
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathRoundDouble)   // Could be done by changing rounding mode, maybe?
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, UnsafeCASLong)     // High register pressure.
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, SystemArrayCopyChar)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, LongDivideUnsigned)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, CRC32Update)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, CRC32UpdateBytes)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, CRC32UpdateByteBuffer)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16ToFloat)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16ToHalf)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16Floor)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16Ceil)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16Rint)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16Greater)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16GreaterEquals)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16Less)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16LessEquals)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16Compare)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16Min)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16Max)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathMultiplyHigh)
-
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringStringIndexOf);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringStringIndexOfAfter);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBufferAppend);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBufferLength);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBufferToString);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderAppendObject);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderAppendString);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderAppendCharSequence);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderAppendCharArray);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderAppendBoolean);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderAppendChar);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderAppendInt);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderAppendLong);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderAppendFloat);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderAppendDouble);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderLength);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringBuilderToString);
-
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, SystemArrayCopyByte);
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, SystemArrayCopyInt);
-
-// 1.8.
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathFmaDouble)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathFmaFloat)
-
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, UnsafeGetAndAddInt)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, UnsafeGetAndAddLong)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, UnsafeGetAndSetInt)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, UnsafeGetAndSetLong)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, UnsafeGetAndSetObject)
-
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, MethodHandleInvokeExact)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, MethodHandleInvoke)
-
-// OpenJDK 11
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, JdkUnsafeCASLong)      // High register pressure.
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, JdkUnsafeGetAndAddInt)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, JdkUnsafeGetAndAddLong)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, JdkUnsafeGetAndSetInt)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, JdkUnsafeGetAndSetLong)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, JdkUnsafeGetAndSetObject)
-UNIMPLEMENTED_INTRINSIC(ARMVIXL, JdkUnsafeCompareAndSetLong)
+#define MARK_UNIMPLEMENTED(Name) UNIMPLEMENTED_INTRINSIC(ARMVIXL, Name)
+UNIMPLEMENTED_INTRINSIC_LIST_ARM(MARK_UNIMPLEMENTED);
+#undef MARK_UNIMPLEMENTED
 
 UNREACHABLE_INTRINSICS(ARMVIXL)
 
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index a76b773a3c..868fd4a120 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -4832,64 +4832,9 @@ void IntrinsicLocationsBuilderX86::VisitMathFmaFloat(HInvoke* invoke) {
   }
 }
 
-UNIMPLEMENTED_INTRINSIC(X86, MathRoundDouble)
-UNIMPLEMENTED_INTRINSIC(X86, FloatIsInfinite)
-UNIMPLEMENTED_INTRINSIC(X86, DoubleIsInfinite)
-UNIMPLEMENTED_INTRINSIC(X86, IntegerHighestOneBit)
-UNIMPLEMENTED_INTRINSIC(X86, LongHighestOneBit)
-UNIMPLEMENTED_INTRINSIC(X86, LongDivideUnsigned)
-UNIMPLEMENTED_INTRINSIC(X86, CRC32Update)
-UNIMPLEMENTED_INTRINSIC(X86, CRC32UpdateBytes)
-UNIMPLEMENTED_INTRINSIC(X86, CRC32UpdateByteBuffer)
-UNIMPLEMENTED_INTRINSIC(X86, FP16ToFloat)
-UNIMPLEMENTED_INTRINSIC(X86, FP16ToHalf)
-UNIMPLEMENTED_INTRINSIC(X86, FP16Floor)
-UNIMPLEMENTED_INTRINSIC(X86, FP16Ceil)
-UNIMPLEMENTED_INTRINSIC(X86, FP16Rint)
-UNIMPLEMENTED_INTRINSIC(X86, FP16Greater)
-UNIMPLEMENTED_INTRINSIC(X86, FP16GreaterEquals)
-UNIMPLEMENTED_INTRINSIC(X86, FP16Less)
-UNIMPLEMENTED_INTRINSIC(X86, FP16LessEquals)
-UNIMPLEMENTED_INTRINSIC(X86, FP16Compare)
-UNIMPLEMENTED_INTRINSIC(X86, FP16Min)
-UNIMPLEMENTED_INTRINSIC(X86, FP16Max)
-UNIMPLEMENTED_INTRINSIC(X86, MathMultiplyHigh)
-
-UNIMPLEMENTED_INTRINSIC(X86, StringStringIndexOf);
-UNIMPLEMENTED_INTRINSIC(X86, StringStringIndexOfAfter);
-UNIMPLEMENTED_INTRINSIC(X86, StringBufferAppend);
-UNIMPLEMENTED_INTRINSIC(X86, StringBufferLength);
-UNIMPLEMENTED_INTRINSIC(X86, StringBufferToString);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderAppendObject);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderAppendString);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderAppendCharSequence);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderAppendCharArray);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderAppendBoolean);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderAppendChar);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderAppendInt);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderAppendLong);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderAppendFloat);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderAppendDouble);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderLength);
-UNIMPLEMENTED_INTRINSIC(X86, StringBuilderToString);
-
-// 1.8.
-
-UNIMPLEMENTED_INTRINSIC(X86, UnsafeGetAndAddInt)
-UNIMPLEMENTED_INTRINSIC(X86, UnsafeGetAndAddLong)
-UNIMPLEMENTED_INTRINSIC(X86, UnsafeGetAndSetInt)
-UNIMPLEMENTED_INTRINSIC(X86, UnsafeGetAndSetLong)
-UNIMPLEMENTED_INTRINSIC(X86, UnsafeGetAndSetObject)
-
-UNIMPLEMENTED_INTRINSIC(X86, MethodHandleInvokeExact)
-UNIMPLEMENTED_INTRINSIC(X86, MethodHandleInvoke)
-
-// OpenJDK 11
-UNIMPLEMENTED_INTRINSIC(X86, JdkUnsafeGetAndAddInt)
-UNIMPLEMENTED_INTRINSIC(X86, JdkUnsafeGetAndAddLong)
-UNIMPLEMENTED_INTRINSIC(X86, JdkUnsafeGetAndSetInt)
-UNIMPLEMENTED_INTRINSIC(X86, JdkUnsafeGetAndSetLong)
-UNIMPLEMENTED_INTRINSIC(X86, JdkUnsafeGetAndSetObject)
+#define MARK_UNIMPLEMENTED(Name) UNIMPLEMENTED_INTRINSIC(X86, Name)
+UNIMPLEMENTED_INTRINSIC_LIST_X86(MARK_UNIMPLEMENTED);
+#undef MARK_UNIMPLEMENTED
 
 UNREACHABLE_INTRINSICS(X86)
 
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index cf3d63b4f4..9d0d5f155e 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -4990,57 +4990,9 @@ void VarHandleSlowPathX86_64::EmitByteArrayViewCode(CodeGeneratorX86_64* codegen
   __ jmp(GetExitLabel());
 }
 
-UNIMPLEMENTED_INTRINSIC(X86_64, CRC32Update)
-UNIMPLEMENTED_INTRINSIC(X86_64, CRC32UpdateBytes)
-UNIMPLEMENTED_INTRINSIC(X86_64, CRC32UpdateByteBuffer)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16ToFloat)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16ToHalf)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16Floor)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16Ceil)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16Rint)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16Greater)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16GreaterEquals)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16Less)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16LessEquals)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16Compare)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16Min)
-UNIMPLEMENTED_INTRINSIC(X86_64, FP16Max)
-
-UNIMPLEMENTED_INTRINSIC(X86_64, StringStringIndexOf);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringStringIndexOfAfter);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBufferAppend);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBufferLength);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBufferToString);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderAppendObject);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderAppendString);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderAppendCharSequence);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderAppendCharArray);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderAppendBoolean);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderAppendChar);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderAppendInt);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderAppendLong);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderAppendFloat);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderAppendDouble);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderLength);
-UNIMPLEMENTED_INTRINSIC(X86_64, StringBuilderToString);
-
-// 1.8.
-
-UNIMPLEMENTED_INTRINSIC(X86_64, UnsafeGetAndAddInt)
-UNIMPLEMENTED_INTRINSIC(X86_64, UnsafeGetAndAddLong)
-UNIMPLEMENTED_INTRINSIC(X86_64, UnsafeGetAndSetInt)
-UNIMPLEMENTED_INTRINSIC(X86_64, UnsafeGetAndSetLong)
-UNIMPLEMENTED_INTRINSIC(X86_64, UnsafeGetAndSetObject)
-
-UNIMPLEMENTED_INTRINSIC(X86_64, MethodHandleInvokeExact)
-UNIMPLEMENTED_INTRINSIC(X86_64, MethodHandleInvoke)
-
-// OpenJDK 11
-UNIMPLEMENTED_INTRINSIC(X86_64, JdkUnsafeGetAndAddInt)
-UNIMPLEMENTED_INTRINSIC(X86_64, JdkUnsafeGetAndAddLong)
-UNIMPLEMENTED_INTRINSIC(X86_64, JdkUnsafeGetAndSetInt)
-UNIMPLEMENTED_INTRINSIC(X86_64, JdkUnsafeGetAndSetLong)
-UNIMPLEMENTED_INTRINSIC(X86_64, JdkUnsafeGetAndSetObject)
+#define MARK_UNIMPLEMENTED(Name) UNIMPLEMENTED_INTRINSIC(X86_64, Name)
+UNIMPLEMENTED_INTRINSIC_LIST_X86_64(MARK_UNIMPLEMENTED);
+#undef MARK_UNIMPLEMENTED
 
 UNREACHABLE_INTRINSICS(X86_64)