[Interpreter] TurboFan implementation of intrinsics.

src/interpreter/bytecodes.h

 // Copyright 2015 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef V8_INTERPRETER_BYTECODES_H_
 #define V8_INTERPRETER_BYTECODES_H_
 
 #include <iosfwd>
 
 // Clients of this interface shouldn't depend on lots of interpreter internals.
@@ skipping 243 matching lines @@
   V(ForInPrepareWide, OperandType::kRegOutTriple16)                            \
   V(ForInDone, OperandType::kReg8, OperandType::kReg8)                         \
   V(ForInNext, OperandType::kReg8, OperandType::kReg8, OperandType::kRegPair8) \
   V(ForInNextWide, OperandType::kReg16, OperandType::kReg16,                   \
     OperandType::kRegPair16)                                                   \
   V(ForInStep, OperandType::kReg8)                                             \
                                                                                \
   /* Non-local flow control */                                                 \
   V(Throw, OperandType::kNone)                                                 \
   V(ReThrow, OperandType::kNone)                                               \
-  V(Return, OperandType::kNone)
+  V(Return, OperandType::kNone)                                                \
+                                                                               \
+  /* Intrinsics */                                                             \
+  V(IntrinsicOneArg, OperandType::kIdx16, OperandType::kMaybeReg16)

[oth, 2016/02/01 09:52:33]

Can this be called this CallIntrinsic or InvokeIntrinsic?

Is the final register operand optional?

If not it should be a kReg8 operand here. Register translation will ensure a
single register operand is always reachable. So far kMaybeRegXX operands are
used for variable register arguments where it's valid to have a zero count of
operands.

If we foresee supporting other intrinsics in future with zero or >= two
arguments, then we could consider rationalizing into two bytecodes for all
cases:
CallIntrinsic OperandType::kIdx16, OperandType::kMaybeReg8,
OperandType::kRegCount8
CallIntrinsicWide OperandType::kIdx16, OperandType::kMaybeReg16,
OperandType::kRegCount16

[rmcilroy, 2016/02/01 11:04:44]

+1 for InvokeIntrinsic and the comment on using kReg8 instead of kMaybeReg8.

[epertoso, 2016/03/03 11:20:39]

Added InvokeIntrinsic and InvokeIntrinsicWide. As we might be adding support for
other intrinsics, I used kMaybeRegXX and kRegCountXX.

 
 // Enumeration of the size classes of operand types used by bytecodes.
 enum class OperandSize : uint8_t {
   kNone = 0,
   kByte = 1,
   kShort = 2,
 };
 
 
 // Enumeration of operand types used by bytecodes.
@@ skipping 195 matching lines @@
 
 std::ostream& operator<<(std::ostream& os, const Bytecode& bytecode);
 std::ostream& operator<<(std::ostream& os, const OperandType& operand_type);
 std::ostream& operator<<(std::ostream& os, const OperandSize& operand_type);
 
 }  // namespace interpreter
 }  // namespace internal
 }  // namespace v8
 
 #endif  // V8_INTERPRETER_BYTECODES_H_