MIPS: Add turbofan support for mips32.

src/compiler/mips/code-generator-mips.cc

Index: src/compiler/mips/code-generator-mips.cc
diff --git a/src/compiler/mips/code-generator-mips.cc b/src/compiler/mips/code-generator-mips.cc
new file mode 100644
index 0000000000000000000000000000000000000000..17c9b43e683563f3265b15d6d4170833d887a247
--- /dev/null
+++ b/src/compiler/mips/code-generator-mips.cc
@@ -0,0 +1,920 @@
+// Copyright 2014 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.
+
+#include "src/compiler/code-generator.h"
+#include "src/compiler/code-generator-impl.h"
+#include "src/compiler/gap-resolver.h"
+#include "src/compiler/node-matchers.h"
+#include "src/compiler/node-properties-inl.h"
+#include "src/mips/macro-assembler-mips.h"
+#include "src/scopes.h"
+
+namespace v8 {
+namespace internal {
+namespace compiler {
+
+#define __ masm()->
+
+
+// TODO(plind): Possibly avoid using these lithium names.
+#define kScratchReg kLithiumScratchReg
+#define kCompareReg kLithiumScratchReg2
+#define kScratchDoubleReg kLithiumScratchDouble
+
+
+// TODO(plind): consider renaming these macros.
+#define TRACE_MSG(msg)                                                      \
+  PrintF("code_gen: \'%s\' in function %s at line %d\n", msg, __FUNCTION__, \
+         __LINE__)
+
+#define TRACE_UNIMPL()                                                       \
+  PrintF("UNIMPLEMENTED code_generator_mips: %s at line %d\n", __FUNCTION__, \
+         __LINE__)
+
+
+// Adds Mips-specific methods to convert InstructionOperands.
+class MipsOperandConverter : public InstructionOperandConverter {
+ public:
+  MipsOperandConverter(CodeGenerator* gen, Instruction* instr)
+      : InstructionOperandConverter(gen, instr) {}
+
+  Operand InputImmediate(int index) {
+    Constant constant = ToConstant(instr_->InputAt(index));
+    switch (constant.type()) {
+      case Constant::kInt32:
+        return Operand(constant.ToInt32());
+      case Constant::kFloat64:
+        return Operand(
+            isolate()->factory()->NewNumber(constant.ToFloat64(), TENURED));
+      case Constant::kInt64:
+      case Constant::kExternalReference:
+      case Constant::kHeapObject:
+        // TODO(plind): Maybe we should handle ExtRef & HeapObj here?
+        //    maybe not done on arm due to const pool ??
+        break;
+    }
+    UNREACHABLE();
+    return Operand(zero_reg);
+  }
+
+  Operand InputOperand(int index) {
+    InstructionOperand* op = instr_->InputAt(index);
+    if (op->IsRegister()) {
+      return Operand(ToRegister(op));
+    }
+    return InputImmediate(index);
+  }
+
+  MemOperand MemoryOperand(int* first_index) {
+    const int index = *first_index;
+    switch (AddressingModeField::decode(instr_->opcode())) {
+      case kMode_None:
+        break;
+      case kMode_MRI:
+        *first_index += 2;
+        return MemOperand(InputRegister(index + 0), InputInt32(index + 1));
+      case kMode_MRR:
+        // TODO(plind): r6 address mode, to be implemented ...
+        UNREACHABLE();
+    }
+    UNREACHABLE();
+    return MemOperand(no_reg);
+  }
+
+  MemOperand MemoryOperand() {
+    int index = 0;
+    return MemoryOperand(&index);
+  }
+
+  MemOperand ToMemOperand(InstructionOperand* op) const {
+    DCHECK(op != NULL);
+    DCHECK(!op->IsRegister());
+    DCHECK(!op->IsDoubleRegister());
+    DCHECK(op->IsStackSlot() || op->IsDoubleStackSlot());
+    // The linkage computes where all spill slots are located.
+    FrameOffset offset = linkage()->GetFrameOffset(op->index(), frame(), 0);
+    return MemOperand(offset.from_stack_pointer() ? sp : fp, offset.offset());
+  }
+};
+
+
+static inline bool HasRegisterInput(Instruction* instr, int index) {
+  return instr->InputAt(index)->IsRegister();
+}
+
+
+// TODO(plind): There are only 3 shift ops, does that justify this slightly
+//    messy macro? Consider expanding this in place for sll, srl, sra ops.

[titzer, 2014/09/24 15:23:31]

I agree that you should kill this macro here.

[paul.l..., 2014/09/25 17:18:38]

Done.

+#define ASSEMBLE_SHIFT(asm_instr)                                \
+  do {                                                           \
+    if (instr->InputAt(1)->IsRegister()) {                       \
+      __ asm_instr##v(i.OutputRegister(), i.InputRegister(0),    \
+                      i.InputRegister(1));                       \
+    } else {                                                     \
+      int32_t imm = i.InputOperand(1).immediate();               \
+      __ asm_instr(i.OutputRegister(), i.InputRegister(0), imm); \
+    }                                                            \
+  } while (0);
+
+
+// Assembles an instruction after register allocation, producing machine code.
+void CodeGenerator::AssembleArchInstruction(Instruction* instr) {
+  MipsOperandConverter i(this, instr);
+  InstructionCode opcode = instr->opcode();
+
+  switch (ArchOpcodeField::decode(opcode)) {
+    case kArchCallCodeObject: {
+      EnsureSpaceForLazyDeopt();
+      if (instr->InputAt(0)->IsImmediate()) {
+        __ Call(Handle<Code>::cast(i.InputHeapObject(0)),
+                RelocInfo::CODE_TARGET);
+      } else {
+        __ addiu(at, i.InputRegister(0), Code::kHeaderSize - kHeapObjectTag);
+        __ Call(at);
+      }
+      AddSafepointAndDeopt(instr);
+      break;
+    }
+    case kArchCallJSFunction: {
+      EnsureSpaceForLazyDeopt();
+      Register func = i.InputRegister(0);
+      if (FLAG_debug_code) {
+        // Check the function's context matches the context argument.
+        __ lw(kScratchReg, FieldMemOperand(func, JSFunction::kContextOffset));
+        __ Assert(eq, kWrongFunctionContext, cp, Operand(kScratchReg));
+      }
+
+      __ lw(at, FieldMemOperand(func, JSFunction::kCodeEntryOffset));
+      __ Call(at);
+      AddSafepointAndDeopt(instr);
+      break;
+    }
+    case kArchJmp:
+      __ Branch(code_->GetLabel(i.InputBlock(0)));
+      break;
+    case kArchNop:
+      // don't emit code for nops.
+      break;
+    case kArchRet:
+      AssembleReturn();
+      break;
+    case kArchTruncateDoubleToI:
+      __ TruncateDoubleToI(i.OutputRegister(), i.InputDoubleRegister(0));
+      break;
+    case kMipsAdd:
+      __ Addu(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsAddOvf:
+      __ AdduAndCheckForOverflow(i.OutputRegister(), i.InputRegister(0),
+                                 i.InputOperand(1), kCompareReg, kScratchReg);

[titzer, 2014/09/24 15:23:31]

The macro assembler methods here generate a lot of instructions. It may not be
an overall win to generate an add with overflow on MIPS, but instead to generate
an explicit tag check with a branch and a regular add.

[paul.l..., 2014/09/25 17:18:39]

Yes, Overflow checking is an issue for MIPS. We have two variations of add/sub,
one generates TRAP on overflow[1], the other silently overflows. We use the
silent version, but must generate extra instructions to check. I don't
understand your suggestion "generate an explicit tag check with a branch and a
regular add", can you elaborate?

The new MIPS r6 instruction set gives us a new/faster way to detect overflows,
we will implement that soon, but still need the old way for older cores.

[1] I prototyped with the TRAP version a year ago, and it made a nice speed
improvement on some benchmarks, but I thought it best to avoid installing signal
handler, it could be fragile, and maybe problematic with sandboxing. 

[titzer, 2014/09/26 10:36:21]

I meant in the front end, when we lower a v8 representation change (i.e. convert
a tagged number to a machine int32 or vice versa). If AddWithOverflow isn't very
efficient on a particular backend, we can generate a different set of machine
operators. That's something where the backend should be able to configure the
frontend to avoid expensive operations.
Ok, that is more reason to have the backend configure the frontend to avoid
certain operations that are complex/slow on specific generations of an
architecture.
Agree. And it could also be very slow for certain programs.

+      break;
+    case kMipsSub:
+      __ Subu(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsSubOvf:
+      __ SubuAndCheckForOverflow(i.OutputRegister(), i.InputRegister(0),
+                                 i.InputOperand(1), kCompareReg, kScratchReg);
+      break;
+    case kMipsMul:
+      __ Mul(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsDiv:
+      __ Div(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsDivU:
+      __ Divu(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsMod:
+      __ Mod(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsModU:
+      __ Modu(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsAnd:
+      __ And(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsOr:
+      __ Or(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsXor:
+      __ Xor(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsShl:
+      ASSEMBLE_SHIFT(sll);
+      break;
+    case kMipsShr:
+      ASSEMBLE_SHIFT(srl);
+      break;
+    case kMipsSar:
+      ASSEMBLE_SHIFT(sra);
+      break;
+    case kMipsRor:
+      __ Ror(i.OutputRegister(), i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsTst:
+      // Psuedo-instruction used for tst/branch.
+      __ And(kCompareReg, i.InputRegister(0), i.InputOperand(1));
+      break;
+    case kMipsCmp:
+      // Psuedo-instruction used for cmp/branch. No opcode emitted here.
+      break;
+    case kMipsMov:
+      // TODO(plind): Should we combine mov/li like this, or use separate instr?
+      //    - Also see x64 ASSEMBLE_BINOP & RegisterOrOperandType
+      if (HasRegisterInput(instr, 0)) {
+        __ mov(i.OutputRegister(), i.InputRegister(0));
+      } else {
+        __ li(i.OutputRegister(), i.InputOperand(0));
+      }
+      break;
+
+    case kMipsFloat64Cmp:
+      // Psuedo-instruction used for FP cmp/branch. No opcode emitted here.
+      break;
+    case kMipsFloat64Add:
+      // TODO(plind): add special case: combine mult & add.
+      __ add_d(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
+               i.InputDoubleRegister(1));
+      break;
+    case kMipsFloat64Sub:
+      __ sub_d(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
+               i.InputDoubleRegister(1));
+      break;
+    case kMipsFloat64Mul:
+      // TODO(plind): add special case: right op is -1.0, see arm port.
+      __ mul_d(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
+               i.InputDoubleRegister(1));
+      break;
+    case kMipsFloat64Div:
+      __ div_d(i.OutputDoubleRegister(), i.InputDoubleRegister(0),
+               i.InputDoubleRegister(1));
+      break;
+    case kMipsFloat64Mod: {
+      // TODO(bmeurer): We should really get rid of this special instruction,
+      // and generate a CallAddress instruction instead.
+      FrameScope scope(masm(), StackFrame::MANUAL);
+      __ PrepareCallCFunction(0, 2, kScratchReg);
+      __ MovToFloatParameters(i.InputDoubleRegister(0),
+                              i.InputDoubleRegister(1));
+      __ CallCFunction(ExternalReference::mod_two_doubles_operation(isolate()),
+                       0, 2);
+      // Move the result in the double result register.
+      __ MovFromFloatResult(i.OutputDoubleRegister());
+      break;
+    }
+    case kMipsInt32ToFloat64: {
+      FPURegister scratch = kScratchDoubleReg;
+      __ mtc1(i.InputRegister(0), scratch);
+      __ cvt_d_w(i.OutputDoubleRegister(), scratch);
+      break;
+    }
+    case kMipsUint32ToFloat64: {
+      FPURegister scratch = kScratchDoubleReg;
+      __ Cvt_d_uw(i.OutputDoubleRegister(), i.InputRegister(0), scratch);
+      break;
+    }
+    case kMipsFloat64ToInt32: {
+      FPURegister scratch = kScratchDoubleReg;
+      // Other arches use round to zero here, so we follow.
+      __ trunc_w_d(scratch, i.InputDoubleRegister(0));
+      __ mfc1(i.OutputRegister(), scratch);
+      break;
+    }
+    case kMipsFloat64ToUint32: {
+      FPURegister scratch = kScratchDoubleReg;
+      // TODO(plind): Fix wrong param order of Trunc_uw_d() macro-asm function.
+      __ Trunc_uw_d(i.InputDoubleRegister(0), i.OutputRegister(), scratch);
+      break;
+    }
+    // ... more basic instructions ...
+
+    case kMipsLbu:
+      __ lbu(i.OutputRegister(), i.MemoryOperand());
+      break;
+    case kMipsLb:
+      __ lb(i.OutputRegister(), i.MemoryOperand());
+      break;
+    case kMipsSb:
+      __ sb(i.InputRegister(2), i.MemoryOperand());
+      break;
+    case kMipsLhu:
+      __ lhu(i.OutputRegister(), i.MemoryOperand());
+      break;
+    case kMipsLh:
+      __ lh(i.OutputRegister(), i.MemoryOperand());
+      break;
+    case kMipsSh:
+      __ sh(i.InputRegister(2), i.MemoryOperand());
+      break;
+    case kMipsLw:
+      __ lw(i.OutputRegister(), i.MemoryOperand());
+      break;
+    case kMipsSw:
+      __ sw(i.InputRegister(2), i.MemoryOperand());
+      break;
+    case kMipsLwc1: {
+      FPURegister scratch = kScratchDoubleReg;
+      __ lwc1(scratch, i.MemoryOperand());
+      __ cvt_d_s(i.OutputDoubleRegister(), scratch);
+      break;
+    }
+    case kMipsSwc1: {
+      int index = 0;
+      FPURegister scratch = kScratchDoubleReg;
+      MemOperand operand = i.MemoryOperand(&index);
+      __ cvt_s_d(scratch, i.InputDoubleRegister(index));
+      __ swc1(scratch, operand);
+      break;
+    }
+    case kMipsLdc1:
+      __ ldc1(i.OutputDoubleRegister(), i.MemoryOperand());
+      break;
+    case kMipsSdc1:
+      __ sdc1(i.InputDoubleRegister(2), i.MemoryOperand());
+      break;
+    case kMipsPush:
+      __ Push(i.InputRegister(0));
+      break;
+    case kMipsStoreWriteBarrier:
+      Register object = i.InputRegister(0);
+      Register index = i.InputRegister(1);
+      Register value = i.InputRegister(2);
+      __ addu(index, object, index);
+      __ sw(value, MemOperand(index));
+      SaveFPRegsMode mode =
+          frame()->DidAllocateDoubleRegisters() ? kSaveFPRegs : kDontSaveFPRegs;
+      RAStatus ra_status = kRAHasNotBeenSaved;
+      __ RecordWrite(object, index, value, ra_status, mode);
+      break;
+  }
+}
+
+
+#define UNSUPPORTED_COND(opcode, condition)                                  \
+  OFStream out(stdout);                                                      \
+  out << "Unsupported " << #opcode << " condition: \"" << condition << "\""; \
+  UNIMPLEMENTED();
+
+// Assembles branches after an instruction.
+void CodeGenerator::AssembleArchBranch(Instruction* instr,
+                                       FlagsCondition condition) {

[titzer, 2014/09/24 15:23:32]

The handling of branches here is tough to follow. It may be that the flags
continuation which works OK on other architectures doesn't work so great here.
In particular it seems that materializing the extra bit for AddWithOverflow,
only to feed it into a branch at the end is a bit indirect.

[Benedikt Meurer, 2014/09/25 05:37:10]

Hm, we may need to refactor the AssembleArchBranch and AssembleArchBoolean logic
a bit to make it easier for MIPS. I'll think about that.

[paul.l..., 2014/09/25 17:18:39]

Complications arise from MIPS not having condition codes like other arch's. The
compare and branch are merged into a single instruction, which means that the
branch instruction needs the registers provided to the compare to still be
valid. The current FlagsContinuation mechanism does guarantee this. As you can
see we mostly don't emit instructions for the compare itself, but we do preserve
the type of comparison as a psuedo-op to guide the branch generation.
Yes, the code is a bit ugly :(  It "looked better" here with single-line cases: 
https://github.com/paul99/v8m-rb/blob/b8e8df9/src/compiler/mips/code-generato...
but clang-format won't take that form. I'm thinking of splitting into 4 smaller
routines, each specialized for one arch_opcode type. I'd be glad to hear your
ideas. 

Another issue is the exposed branch delay slots here, which make things faster
at the expense of readability.

[titzer, 2014/09/26 10:36:21]

Ok, in that case maybe using a FlagsContinuation on MIPS is more trouble than
its worth, and the instruction selector could just emit specific kinds of
compare instructions and not try to combine the flags usage with arithmetic
(which is our goal on x86).

But that's something we can iterate on; I'm ok with landing this as is.

+  MipsOperandConverter i(this, instr);
+  Label done;
+
+  // Emit a branch. The true and false targets are always the last two inputs
+  // to the instruction.
+  BasicBlock* tblock = i.InputBlock(instr->InputCount() - 2);
+  BasicBlock* fblock = i.InputBlock(instr->InputCount() - 1);
+  bool fallthru = IsNextInAssemblyOrder(fblock);
+  Label* tlabel = code()->GetLabel(tblock);
+  Label* flabel = fallthru ? &done : code()->GetLabel(fblock);
+  Condition cc = kNoCondition;
+
+  // MIPS does not have condition code flags, so compare and branch are
+  // implemented differently than on the other arch's. The compare operations
+  // emit mips psuedo-instructions, which are handled here by branch
+  // instructions that do the actual comparison. Essential that the input
+  // registers to compare psuedo-op are not modified before this branch op, as
+  // they are tested here.
+  // TODO(plind): Add CHECK() to ensure that test/cmp and this branch were
+  //    not separated by other instructions.
+
+  if (instr->arch_opcode() == kMipsTst) {
+    // The kMipsTst psuedo-instruction emits And to 'kCompareReg' register.
+    switch (condition) {
+      case kNotEqual:
+        cc = ne;
+        break;
+      case kEqual:
+        cc = eq;
+        break;
+      default:
+        UNSUPPORTED_COND(kMipsTst, condition);
+        break;
+    }
+    __ Branch(tlabel, cc, kCompareReg, Operand(zero_reg));
+
+  } else if (instr->arch_opcode() == kMipsAddOvf ||
+             instr->arch_opcode() == kMipsSubOvf) {
+    // kMipsAddOvf, SubOvf emit negative result to 'kCompareReg' on overflow.
+    switch (condition) {
+      case kOverflow:
+        cc = lt;
+        break;
+      case kNotOverflow:
+        cc = ge;
+        break;
+      default:
+        UNSUPPORTED_COND(kMipsAddOvf, condition);
+        break;
+    }
+    __ Branch(tlabel, cc, kCompareReg, Operand(zero_reg));
+
+  } else if (instr->arch_opcode() == kMipsCmp) {
+    switch (condition) {
+      case kEqual:
+        cc = eq;
+        break;
+      case kNotEqual:
+        cc = ne;
+        break;
+      case kSignedLessThan:
+        cc = lt;
+        break;
+      case kSignedGreaterThanOrEqual:
+        cc = ge;
+        break;
+      case kSignedLessThanOrEqual:
+        cc = le;
+        break;
+      case kSignedGreaterThan:
+        cc = gt;
+        break;
+      case kUnsignedLessThan:
+        cc = lo;
+        break;
+      case kUnsignedGreaterThanOrEqual:
+        cc = hs;
+        break;
+      case kUnsignedLessThanOrEqual:
+        cc = ls;
+        break;
+      case kUnsignedGreaterThan:
+        cc = hi;
+        break;
+      default:
+        UNSUPPORTED_COND(kMipsCmp, condition);
+        break;
+    }
+    __ Branch(tlabel, cc, i.InputRegister(0), i.InputOperand(1));
+
+    if (!fallthru) __ Branch(flabel);  // no fallthru to flabel.
+    __ bind(&done);
+
+  } else if (instr->arch_opcode() == kMipsFloat64Cmp) {
+    // TODO(dusmil) optimize unordered checks to use less instructions
+    // even if we have to unfold BranchF macro.
+    Label* nan = flabel;
+    switch (condition) {
+      case kUnorderedEqual:
+        cc = eq;
+        break;
+      case kUnorderedNotEqual:
+        cc = ne;
+        nan = tlabel;
+        break;
+      case kUnorderedLessThan:
+        cc = lt;
+        break;
+      case kUnorderedGreaterThanOrEqual:
+        cc = ge;
+        nan = tlabel;
+        break;
+      case kUnorderedLessThanOrEqual:
+        cc = le;
+        break;
+      case kUnorderedGreaterThan:
+        cc = gt;
+        nan = tlabel;
+        break;
+      default:
+        UNSUPPORTED_COND(kMipsFloat64Cmp, condition);
+        break;
+    }
+    __ BranchF(tlabel, nan, cc, i.InputDoubleRegister(0),
+               i.InputDoubleRegister(1));
+
+    if (!fallthru) __ Branch(flabel);  // no fallthru to flabel.
+    __ bind(&done);
+
+  } else {
+    PrintF("AssembleArchBranch Unimplemented arch_opcode: %d\n",
+           instr->arch_opcode());
+    UNIMPLEMENTED();
+  }
+}
+
+
+// Assembles boolean materializations after an instruction.
+void CodeGenerator::AssembleArchBoolean(Instruction* instr,
+                                        FlagsCondition condition) {
+  MipsOperandConverter i(this, instr);
+  Label done;
+
+  // Materialize a full 32-bit 1 or 0 value. The result register is always the
+  // last output of the instruction.
+  Label false_value;
+  DCHECK_NE(0, instr->OutputCount());
+  Register result = i.OutputRegister(instr->OutputCount() - 1);
+  Condition cc = kNoCondition;
+
+  // MIPS does not have condition code flags, so compare and branch are
+  // implemented differently than on the other arch's. The compare operations
+  // emit mips psuedo-instructions, which are checked and handled here.
+
+  // For materializations, we use delay slot to set the result true, and
+  // in the false case, where we fall thru the branch, we reset the result
+  // false.
+
+  // TODO(plind): Add CHECK() to ensure that test/cmp and this branch were
+  //    not separated by other instructions.
+  if (instr->arch_opcode() == kMipsTst) {
+    // The kMipsTst psuedo-instruction emits And to 'kCompareReg' register.
+    switch (condition) {
+      case kNotEqual:
+        cc = ne;
+        break;
+      case kEqual:
+        cc = eq;
+        break;
+      default:
+        UNSUPPORTED_COND(kMipsTst, condition);
+        break;
+    }
+    __ Branch(USE_DELAY_SLOT, &done, cc, kCompareReg, Operand(zero_reg));
+    __ li(result, Operand(1));  // In delay slot.
+
+  } else if (instr->arch_opcode() == kMipsAddOvf ||
+             instr->arch_opcode() == kMipsSubOvf) {
+    // kMipsAddOvf, SubOvf emits negative result to 'kCompareReg' on overflow.
+    switch (condition) {
+      case kOverflow:
+        cc = lt;
+        break;
+      case kNotOverflow:
+        cc = ge;
+        break;
+      default:
+        UNSUPPORTED_COND(kMipsAddOvf, condition);
+        break;
+    }
+    __ Branch(USE_DELAY_SLOT, &done, cc, kCompareReg, Operand(zero_reg));
+    __ li(result, Operand(1));  // In delay slot.
+
+
+  } else if (instr->arch_opcode() == kMipsCmp) {
+    Register left = i.InputRegister(0);
+    Operand right = i.InputOperand(1);
+    switch (condition) {
+      case kEqual:
+        cc = eq;
+        break;
+      case kNotEqual:
+        cc = ne;
+        break;
+      case kSignedLessThan:
+        cc = lt;
+        break;
+      case kSignedGreaterThanOrEqual:
+        cc = ge;
+        break;
+      case kSignedLessThanOrEqual:
+        cc = le;
+        break;
+      case kSignedGreaterThan:
+        cc = gt;
+        break;
+      case kUnsignedLessThan:
+        cc = lo;
+        break;
+      case kUnsignedGreaterThanOrEqual:
+        cc = hs;
+        break;
+      case kUnsignedLessThanOrEqual:
+        cc = ls;
+        break;
+      case kUnsignedGreaterThan:
+        cc = hi;
+        break;
+      default:
+        UNSUPPORTED_COND(kMipsCmp, condition);
+        break;
+    }
+    __ Branch(USE_DELAY_SLOT, &done, cc, left, right);
+    __ li(result, Operand(1));  // In delay slot.
+
+  } else if (instr->arch_opcode() == kMipsFloat64Cmp) {
+    FPURegister left = i.InputDoubleRegister(0);
+    FPURegister right = i.InputDoubleRegister(1);
+    // TODO(plind): Provide NaN-testing macro-asm function without need for
+    // BranchF.
+    FPURegister dummy1 = f0;
+    FPURegister dummy2 = f2;
+    switch (condition) {
+      case kUnorderedEqual:
+        // TODO(plind):  improve the NaN testing throughout this function.
+        __ BranchF(NULL, &false_value, kNoCondition, dummy1, dummy2);
+        cc = eq;
+        break;
+      case kUnorderedNotEqual:
+        __ BranchF(USE_DELAY_SLOT, NULL, &done, kNoCondition, dummy1, dummy2);
+        __ li(result, Operand(1));  // In delay slot - returns 1 on NaN.
+        cc = ne;
+        break;
+      case kUnorderedLessThan:
+        __ BranchF(NULL, &false_value, kNoCondition, dummy1, dummy2);
+        cc = lt;
+        break;
+      case kUnorderedGreaterThanOrEqual:
+        __ BranchF(USE_DELAY_SLOT, NULL, &done, kNoCondition, dummy1, dummy2);
+        __ li(result, Operand(1));  // In delay slot - returns 1 on NaN.
+        cc = ge;
+        break;
+      case kUnorderedLessThanOrEqual:
+        __ BranchF(NULL, &false_value, kNoCondition, dummy1, dummy2);
+        cc = le;
+        break;
+      case kUnorderedGreaterThan:
+        __ BranchF(USE_DELAY_SLOT, NULL, &done, kNoCondition, dummy1, dummy2);
+        __ li(result, Operand(1));  // In delay slot - returns 1 on NaN.
+        cc = gt;
+        break;
+      default:
+        UNSUPPORTED_COND(kMipsCmp, condition);
+        break;
+    }
+    __ BranchF(USE_DELAY_SLOT, &done, NULL, cc, left, right);
+    __ li(result, Operand(1));  // In delay slot - branch taken returns 1.
+                                // Fall-thru (branch not taken) returns 0.
+
+  } else {
+    PrintF("AssembleArchBranch Unimplemented arch_opcode is : %d\n",
+           instr->arch_opcode());
+    TRACE_UNIMPL();
+    UNIMPLEMENTED();
+  }
+  // Fallthru case is the false materialization.
+  __ bind(&false_value);
+  __ li(result, Operand(0));
+  __ bind(&done);
+}
+
+
+void CodeGenerator::AssembleDeoptimizerCall(int deoptimization_id) {
+  Address deopt_entry = Deoptimizer::GetDeoptimizationEntry(
+      isolate(), deoptimization_id, Deoptimizer::LAZY);
+  __ Call(deopt_entry, RelocInfo::RUNTIME_ENTRY);
+}
+
+
+void CodeGenerator::AssemblePrologue() {
+  CallDescriptor* descriptor = linkage()->GetIncomingDescriptor();
+  if (descriptor->kind() == CallDescriptor::kCallAddress) {
+    __ Push(ra, fp);
+    __ mov(fp, sp);
+    const RegList saves = descriptor->CalleeSavedRegisters();
+    if (saves != 0) {  // Save callee-saved registers.
+      // TODO(plind): make callee save size const, possibly DCHECK it.
+      int register_save_area_size = 0;
+      for (int i = Register::kNumRegisters - 1; i >= 0; i--) {
+        if (!((1 << i) & saves)) continue;
+        register_save_area_size += kPointerSize;
+      }
+      frame()->SetRegisterSaveAreaSize(register_save_area_size);
+      __ MultiPush(saves);
+    }
+  } else if (descriptor->IsJSFunctionCall()) {
+    CompilationInfo* info = linkage()->info();
+    __ Prologue(info->IsCodePreAgingActive());
+    frame()->SetRegisterSaveAreaSize(
+        StandardFrameConstants::kFixedFrameSizeFromFp);
+
+    // Sloppy mode functions and builtins need to replace the receiver with the
+    // global proxy when called as functions (without an explicit receiver
+    // object).
+    // TODO(mstarzinger/verwaest): Should this be moved back into the CallIC?
+    if (info->strict_mode() == SLOPPY && !info->is_native()) {
+      Label ok;
+      // +2 for return address and saved frame pointer.
+      int receiver_slot = info->scope()->num_parameters() + 2;
+      __ lw(a2, MemOperand(fp, receiver_slot * kPointerSize));
+      __ LoadRoot(at, Heap::kUndefinedValueRootIndex);
+      __ Branch(&ok, ne, a2, Operand(at));
+
+      __ lw(a2, GlobalObjectOperand());
+      __ lw(a2, FieldMemOperand(a2, GlobalObject::kGlobalProxyOffset));
+      __ sw(a2, MemOperand(fp, receiver_slot * kPointerSize));
+      __ bind(&ok);
+    }
+  } else {
+    __ StubPrologue();
+    frame()->SetRegisterSaveAreaSize(
+        StandardFrameConstants::kFixedFrameSizeFromFp);
+  }
+  int stack_slots = frame()->GetSpillSlotCount();
+  if (stack_slots > 0) {
+    __ Subu(sp, sp, Operand(stack_slots * kPointerSize));
+  }
+}
+
+
+void CodeGenerator::AssembleReturn() {
+  CallDescriptor* descriptor = linkage()->GetIncomingDescriptor();
+  if (descriptor->kind() == CallDescriptor::kCallAddress) {
+    if (frame()->GetRegisterSaveAreaSize() > 0) {
+      // Remove this frame's spill slots first.
+      int stack_slots = frame()->GetSpillSlotCount();
+      if (stack_slots > 0) {
+        __ Addu(sp, sp, Operand(stack_slots * kPointerSize));
+      }
+      // Restore registers.
+      const RegList saves = descriptor->CalleeSavedRegisters();
+      if (saves != 0) {
+        __ MultiPop(saves);
+      }
+    }
+    __ mov(sp, fp);
+    __ Pop(ra, fp);
+    __ Ret();
+  } else {
+    __ mov(sp, fp);
+    __ Pop(ra, fp);
+    int pop_count = descriptor->IsJSFunctionCall()
+                        ? static_cast<int>(descriptor->JSParameterCount())
+                        : 0;
+    __ DropAndRet(pop_count);
+  }
+}
+
+
+void CodeGenerator::AssembleMove(InstructionOperand* source,
+                                 InstructionOperand* destination) {
+  MipsOperandConverter g(this, NULL);
+  // Dispatch on the source and destination operand kinds.  Not all
+  // combinations are possible.
+  if (source->IsRegister()) {
+    DCHECK(destination->IsRegister() || destination->IsStackSlot());
+    Register src = g.ToRegister(source);
+    if (destination->IsRegister()) {
+      __ mov(g.ToRegister(destination), src);
+    } else {
+      __ sw(src, g.ToMemOperand(destination));
+    }
+  } else if (source->IsStackSlot()) {
+    DCHECK(destination->IsRegister() || destination->IsStackSlot());
+    MemOperand src = g.ToMemOperand(source);
+    if (destination->IsRegister()) {
+      __ lw(g.ToRegister(destination), src);
+    } else {
+      Register temp = kScratchReg;
+      __ lw(temp, src);
+      __ sw(temp, g.ToMemOperand(destination));
+    }
+  } else if (source->IsConstant()) {
+    if (destination->IsRegister() || destination->IsStackSlot()) {
+      Register dst =
+          destination->IsRegister() ? g.ToRegister(destination) : kScratchReg;
+      Constant src = g.ToConstant(source);
+      switch (src.type()) {
+        case Constant::kInt32:
+          __ li(dst, Operand(src.ToInt32()));
+          break;
+        case Constant::kInt64:
+          UNREACHABLE();
+          break;
+        case Constant::kFloat64:
+          __ li(dst, isolate()->factory()->NewNumber(src.ToFloat64(), TENURED));
+          break;
+        case Constant::kExternalReference:
+          __ li(dst, Operand(src.ToExternalReference()));
+          break;
+        case Constant::kHeapObject:
+          __ li(dst, src.ToHeapObject());
+          break;
+      }
+      if (destination->IsStackSlot()) __ sw(dst, g.ToMemOperand(destination));
+    } else if (destination->IsDoubleRegister()) {
+      FPURegister result = g.ToDoubleRegister(destination);
+      __ Move(result, g.ToDouble(source));
+    } else {
+      DCHECK(destination->IsDoubleStackSlot());
+      FPURegister temp = kScratchDoubleReg;
+      __ Move(temp, g.ToDouble(source));
+      __ sdc1(temp, g.ToMemOperand(destination));
+    }
+  } else if (source->IsDoubleRegister()) {
+    FPURegister src = g.ToDoubleRegister(source);
+    if (destination->IsDoubleRegister()) {
+      FPURegister dst = g.ToDoubleRegister(destination);
+      __ Move(dst, src);
+    } else {
+      DCHECK(destination->IsDoubleStackSlot());
+      __ sdc1(src, g.ToMemOperand(destination));
+    }
+  } else if (source->IsDoubleStackSlot()) {
+    DCHECK(destination->IsDoubleRegister() || destination->IsDoubleStackSlot());
+    MemOperand src = g.ToMemOperand(source);
+    if (destination->IsDoubleRegister()) {
+      __ ldc1(g.ToDoubleRegister(destination), src);
+    } else {
+      FPURegister temp = kScratchDoubleReg;
+      __ ldc1(temp, src);
+      __ sdc1(temp, g.ToMemOperand(destination));
+    }
+  } else {
+    UNREACHABLE();
+  }
+}
+
+
+void CodeGenerator::AssembleSwap(InstructionOperand* source,
+                                 InstructionOperand* destination) {
+  MipsOperandConverter g(this, NULL);
+  // Dispatch on the source and destination operand kinds.  Not all
+  // combinations are possible.
+  if (source->IsRegister()) {
+    // Register-register.
+    Register temp = kScratchReg;
+    Register src = g.ToRegister(source);
+    if (destination->IsRegister()) {
+      Register dst = g.ToRegister(destination);
+      __ Move(temp, src);
+      __ Move(src, dst);
+      __ Move(dst, temp);
+    } else {
+      DCHECK(destination->IsStackSlot());
+      MemOperand dst = g.ToMemOperand(destination);
+      __ mov(temp, src);
+      __ lw(src, dst);
+      __ sw(temp, dst);
+    }
+  } else if (source->IsStackSlot()) {
+    DCHECK(destination->IsStackSlot());
+    Register temp_0 = kScratchReg;
+    Register temp_1 = kCompareReg;
+    MemOperand src = g.ToMemOperand(source);
+    MemOperand dst = g.ToMemOperand(destination);
+    __ lw(temp_0, src);
+    __ lw(temp_1, dst);
+    __ sw(temp_0, dst);
+    __ sw(temp_1, src);
+  } else if (source->IsDoubleRegister()) {
+    FPURegister temp = kScratchDoubleReg;
+    FPURegister src = g.ToDoubleRegister(source);
+    if (destination->IsDoubleRegister()) {
+      FPURegister dst = g.ToDoubleRegister(destination);
+      __ Move(temp, src);
+      __ Move(src, dst);
+      __ Move(dst, temp);
+    } else {
+      DCHECK(destination->IsDoubleStackSlot());
+      MemOperand dst = g.ToMemOperand(destination);
+      __ Move(temp, src);
+      __ ldc1(src, dst);
+      __ sdc1(temp, dst);
+    }
+  } else if (source->IsDoubleStackSlot()) {
+    DCHECK(destination->IsDoubleStackSlot());
+    Register temp_0 = kScratchReg;
+    FPURegister temp_1 = kScratchDoubleReg;
+    MemOperand src0 = g.ToMemOperand(source);
+    MemOperand src1(src0.rm(), src0.offset() + kPointerSize);
+    MemOperand dst0 = g.ToMemOperand(destination);
+    MemOperand dst1(dst0.rm(), dst0.offset() + kPointerSize);
+    __ ldc1(temp_1, dst0);  // Save destination in temp_1.
+    __ lw(temp_0, src0);    // Then use temp_0 to copy source to destination.
+    __ sw(temp_0, dst0);
+    __ lw(temp_0, src1);
+    __ sw(temp_0, dst1);
+    __ sdc1(temp_1, src0);
+  } else {
+    // No other combinations are possible.
+    UNREACHABLE();
+  }
+}
+
+
+void CodeGenerator::AddNopForSmiCodeInlining() {
+  // Unused on 32-bit ARM. Still exists on 64-bit arm.
+  // TODO(plind): Unclear when this is called now. Understand, fix if needed.
+  __ nop();  // Maybe PROPERTY_ACCESS_INLINED?
+}
+
+
+void CodeGenerator::EnsureSpaceForLazyDeopt() {
+  int space_needed = Deoptimizer::patch_size();
+  if (!linkage()->info()->IsStub()) {
+    // Ensure that we have enough space after the previous lazy-bailout
+    // instruction for patching the code here.
+    int current_pc = masm()->pc_offset();
+    if (current_pc < last_lazy_deopt_pc_ + space_needed) {
+      // Block tramoline pool emission for duration of padding.
+      v8::internal::Assembler::BlockTrampolinePoolScope block_trampoline_pool(
+          masm());
+      int padding_size = last_lazy_deopt_pc_ + space_needed - current_pc;
+      DCHECK_EQ(0, padding_size % v8::internal::Assembler::kInstrSize);
+      while (padding_size > 0) {
+        __ nop();
+        padding_size -= v8::internal::Assembler::kInstrSize;
+      }
+    }
+  }
+  MarkLazyDeoptSite();
+}
+
+#undef __
+
+}  // namespace compiler
+}  // namespace internal
+}  // namespace v8