MIPS: Initial version of optimizing compiler for MIPS64R6.

(cherry picked from commit 4dda3376b71209fae07f5c3c8ac3eb4b54207aa8)
(amended for mnc-dev)

Bug: 21555893
Change-Id: I874dc356eee6ab061a32f8f3df5f8ac3a4ab7dcf
Signed-off-by: Alexey Frunze <Alexey.Frunze@imgtec.com>
Signed-off-by: Douglas Leung <douglas.leung@imgtec.com>

3 files changed, 808 insertions, 238 deletions

diff --git a/compiler/utils/mips64/assembler_mips64.cc b/compiler/utils/mips64/assembler_mips64.cc
index 5e49b936f2..3333cd22bd 100644
--- a/compiler/utils/mips64/assembler_mips64.cc
+++ b/compiler/utils/mips64/assembler_mips64.cc
@@ -25,9 +25,9 @@
 namespace art {
 namespace mips64 {
 
-void Mips64Assembler::Emit(int32_t value) {
+void Mips64Assembler::Emit(uint32_t value) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  buffer_.Emit<int32_t>(value);
+  buffer_.Emit<uint32_t>(value);
 }
 
 void Mips64Assembler::EmitR(int opcode, GpuRegister rs, GpuRegister rt, GpuRegister rd,
@@ -35,124 +35,62 @@ void Mips64Assembler::EmitR(int opcode, GpuRegister rs, GpuRegister rt, GpuRegis
   CHECK_NE(rs, kNoGpuRegister);
   CHECK_NE(rt, kNoGpuRegister);
   CHECK_NE(rd, kNoGpuRegister);
-  int32_t encoding = opcode << kOpcodeShift |
-                     static_cast<int32_t>(rs) << kRsShift |
-                     static_cast<int32_t>(rt) << kRtShift |
-                     static_cast<int32_t>(rd) << kRdShift |
-                     shamt << kShamtShift |
-                     funct;
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      static_cast<uint32_t>(rs) << kRsShift |
+                      static_cast<uint32_t>(rt) << kRtShift |
+                      static_cast<uint32_t>(rd) << kRdShift |
+                      shamt << kShamtShift |
+                      funct;
   Emit(encoding);
 }
 
 void Mips64Assembler::EmitI(int opcode, GpuRegister rs, GpuRegister rt, uint16_t imm) {
   CHECK_NE(rs, kNoGpuRegister);
   CHECK_NE(rt, kNoGpuRegister);
-  int32_t encoding = opcode << kOpcodeShift |
-                     static_cast<int32_t>(rs) << kRsShift |
-                     static_cast<int32_t>(rt) << kRtShift |
-                     imm;
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      static_cast<uint32_t>(rs) << kRsShift |
+                      static_cast<uint32_t>(rt) << kRtShift |
+                      imm;
   Emit(encoding);
 }
 
-void Mips64Assembler::EmitJ(int opcode, int address) {
-  int32_t encoding = opcode << kOpcodeShift |
-                     address;
+void Mips64Assembler::EmitI21(int opcode, GpuRegister rs, uint32_t imm21) {
+  CHECK_NE(rs, kNoGpuRegister);
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      static_cast<uint32_t>(rs) << kRsShift |
+                      (imm21 & 0x1FFFFF);
+  Emit(encoding);
+}
+
+void Mips64Assembler::EmitJ(int opcode, uint32_t addr26) {
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      (addr26 & 0x3FFFFFF);
   Emit(encoding);
 }
 
 void Mips64Assembler::EmitFR(int opcode, int fmt, FpuRegister ft, FpuRegister fs, FpuRegister fd,
-int funct) {
+                             int funct) {
   CHECK_NE(ft, kNoFpuRegister);
   CHECK_NE(fs, kNoFpuRegister);
   CHECK_NE(fd, kNoFpuRegister);
-  int32_t encoding = opcode << kOpcodeShift |
-                     fmt << kFmtShift |
-                     static_cast<int32_t>(ft) << kFtShift |
-                     static_cast<int32_t>(fs) << kFsShift |
-                     static_cast<int32_t>(fd) << kFdShift |
-                     funct;
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      fmt << kFmtShift |
+                      static_cast<uint32_t>(ft) << kFtShift |
+                      static_cast<uint32_t>(fs) << kFsShift |
+                      static_cast<uint32_t>(fd) << kFdShift |
+                      funct;
   Emit(encoding);
 }
 
-void Mips64Assembler::EmitFI(int opcode, int fmt, FpuRegister rt, uint16_t imm) {
-  CHECK_NE(rt, kNoFpuRegister);
-  int32_t encoding = opcode << kOpcodeShift |
-                     fmt << kFmtShift |
-                     static_cast<int32_t>(rt) << kRtShift |
-                     imm;
+void Mips64Assembler::EmitFI(int opcode, int fmt, FpuRegister ft, uint16_t imm) {
+  CHECK_NE(ft, kNoFpuRegister);
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      fmt << kFmtShift |
+                      static_cast<uint32_t>(ft) << kFtShift |
+                      imm;
   Emit(encoding);
 }
 
-void Mips64Assembler::EmitBranch(GpuRegister rt, GpuRegister rs, Label* label, bool equal) {
-  int offset;
-  if (label->IsBound()) {
-    offset = label->Position() - buffer_.Size();
-  } else {
-    // Use the offset field of the branch instruction for linking the sites.
-    offset = label->position_;
-    label->LinkTo(buffer_.Size());
-  }
-  if (equal) {
-    Beq(rt, rs, (offset >> 2) & kBranchOffsetMask);
-  } else {
-    Bne(rt, rs, (offset >> 2) & kBranchOffsetMask);
-  }
-}
-
-void Mips64Assembler::EmitJump(Label* label, bool link) {
-  int offset;
-  if (label->IsBound()) {
-    offset = label->Position() - buffer_.Size();
-  } else {
-    // Use the offset field of the jump instruction for linking the sites.
-    offset = label->position_;
-    label->LinkTo(buffer_.Size());
-  }
-  if (link) {
-    Jal((offset >> 2) & kJumpOffsetMask);
-  } else {
-    J((offset >> 2) & kJumpOffsetMask);
-  }
-}
-
-int32_t Mips64Assembler::EncodeBranchOffset(int offset, int32_t inst, bool is_jump) {
-  CHECK_ALIGNED(offset, 4);
-  CHECK(IsInt<POPCOUNT(kBranchOffsetMask)>(offset)) << offset;
-
-  // Properly preserve only the bits supported in the instruction.
-  offset >>= 2;
-  if (is_jump) {
-    offset &= kJumpOffsetMask;
-    return (inst & ~kJumpOffsetMask) | offset;
-  } else {
-    offset &= kBranchOffsetMask;
-    return (inst & ~kBranchOffsetMask) | offset;
-  }
-}
-
-int Mips64Assembler::DecodeBranchOffset(int32_t inst, bool is_jump) {
-  // Sign-extend, then left-shift by 2.
-  if (is_jump) {
-    return (((inst & kJumpOffsetMask) << 6) >> 4);
-  } else {
-    return (((inst & kBranchOffsetMask) << 16) >> 14);
-  }
-}
-
-void Mips64Assembler::Bind(Label* label, bool is_jump) {
-  CHECK(!label->IsBound());
-  int bound_pc = buffer_.Size();
-  while (label->IsLinked()) {
-    int32_t position = label->Position();
-    int32_t next = buffer_.Load<int32_t>(position);
-    int32_t offset = is_jump ? bound_pc - position : bound_pc - position - 4;
-    int32_t encoded = Mips64Assembler::EncodeBranchOffset(offset, next, is_jump);
-    buffer_.Store<int32_t>(position, encoded);
-    label->position_ = Mips64Assembler::DecodeBranchOffset(next, is_jump);
-  }
-  label->BindTo(bound_pc);
-}
-
 void Mips64Assembler::Add(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, rd, 0, 0x20);
 }
@@ -169,6 +107,10 @@ void Mips64Assembler::Addiu(GpuRegister rt, GpuRegister rs, uint16_t imm16) {
   EmitI(0x9, rs, rt, imm16);
 }
 
+void Mips64Assembler::Daddu(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 0, 0x2d);
+}
+
 void Mips64Assembler::Daddiu(GpuRegister rt, GpuRegister rs, uint16_t imm16) {
   EmitI(0x19, rs, rt, imm16);
 }
@@ -181,22 +123,90 @@ void Mips64Assembler::Subu(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, rd, 0, 0x23);
 }
 
-void Mips64Assembler::Mult(GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::Dsubu(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 0, 0x2f);
+}
+
+void Mips64Assembler::MultR2(GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, static_cast<GpuRegister>(0), 0, 0x18);
 }
 
-void Mips64Assembler::Multu(GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::MultuR2(GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, static_cast<GpuRegister>(0), 0, 0x19);
 }
 
-void Mips64Assembler::Div(GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::DivR2(GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, static_cast<GpuRegister>(0), 0, 0x1a);
 }
 
-void Mips64Assembler::Divu(GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::DivuR2(GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, static_cast<GpuRegister>(0), 0, 0x1b);
 }
 
+void Mips64Assembler::MulR2(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0x1c, rs, rt, rd, 0, 2);
+}
+
+void Mips64Assembler::DivR2(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  DivR2(rs, rt);
+  Mflo(rd);
+}
+
+void Mips64Assembler::ModR2(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  DivR2(rs, rt);
+  Mfhi(rd);
+}
+
+void Mips64Assembler::DivuR2(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  DivuR2(rs, rt);
+  Mflo(rd);
+}
+
+void Mips64Assembler::ModuR2(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  DivuR2(rs, rt);
+  Mfhi(rd);
+}
+
+void Mips64Assembler::MulR6(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x18);
+}
+
+void Mips64Assembler::DivR6(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x1a);
+}
+
+void Mips64Assembler::ModR6(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 3, 0x1a);
+}
+
+void Mips64Assembler::DivuR6(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x1b);
+}
+
+void Mips64Assembler::ModuR6(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 3, 0x1b);
+}
+
+void Mips64Assembler::Dmul(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x1c);
+}
+
+void Mips64Assembler::Ddiv(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x1e);
+}
+
+void Mips64Assembler::Dmod(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 3, 0x1e);
+}
+
+void Mips64Assembler::Ddivu(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x1f);
+}
+
+void Mips64Assembler::Dmodu(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 3, 0x1f);
+}
+
 void Mips64Assembler::And(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, rd, 0, 0x24);
 }
@@ -225,30 +235,80 @@ void Mips64Assembler::Nor(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, rd, 0, 0x27);
 }
 
-void Mips64Assembler::Sll(GpuRegister rd, GpuRegister rs, int shamt) {
-  EmitR(0, rs, static_cast<GpuRegister>(0), rd, shamt, 0x00);
+void Mips64Assembler::Seb(GpuRegister rd, GpuRegister rt) {
+  EmitR(0x1f, static_cast<GpuRegister>(0), rt, rd, 0x10, 0x20);
 }
 
-void Mips64Assembler::Srl(GpuRegister rd, GpuRegister rs, int shamt) {
-  EmitR(0, rs, static_cast<GpuRegister>(0), rd, shamt, 0x02);
+void Mips64Assembler::Seh(GpuRegister rd, GpuRegister rt) {
+  EmitR(0x1f, static_cast<GpuRegister>(0), rt, rd, 0x18, 0x20);
 }
 
-void Mips64Assembler::Sra(GpuRegister rd, GpuRegister rs, int shamt) {
-  EmitR(0, rs, static_cast<GpuRegister>(0), rd, shamt, 0x03);
+void Mips64Assembler::Dext(GpuRegister rt, GpuRegister rs, int pos, int size_less_one) {
+  DCHECK(0 <= pos && pos < 32) << pos;
+  DCHECK(0 <= size_less_one && size_less_one < 32) << size_less_one;
+  EmitR(0x1f, rs, rt, static_cast<GpuRegister>(size_less_one), pos, 3);
 }
 
-void Mips64Assembler::Sllv(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::Sll(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x00);
+}
+
+void Mips64Assembler::Srl(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x02);
+}
+
+void Mips64Assembler::Sra(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x03);
+}
+
+void Mips64Assembler::Sllv(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
   EmitR(0, rs, rt, rd, 0, 0x04);
 }
 
-void Mips64Assembler::Srlv(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::Srlv(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
   EmitR(0, rs, rt, rd, 0, 0x06);
 }
 
-void Mips64Assembler::Srav(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::Srav(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
   EmitR(0, rs, rt, rd, 0, 0x07);
 }
 
+void Mips64Assembler::Dsll(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x38);
+}
+
+void Mips64Assembler::Dsrl(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x3a);
+}
+
+void Mips64Assembler::Dsra(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x3b);
+}
+
+void Mips64Assembler::Dsll32(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x3c);
+}
+
+void Mips64Assembler::Dsrl32(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x3e);
+}
+
+void Mips64Assembler::Dsra32(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x3f);
+}
+
+void Mips64Assembler::Dsllv(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
+  EmitR(0, rs, rt, rd, 0, 0x14);
+}
+
+void Mips64Assembler::Dsrlv(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
+  EmitR(0, rs, rt, rd, 0, 0x16);
+}
+
+void Mips64Assembler::Dsrav(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
+  EmitR(0, rs, rt, rd, 0, 0x17);
+}
+
 void Mips64Assembler::Lb(GpuRegister rt, GpuRegister rs, uint16_t imm16) {
   EmitI(0x20, rs, rt, imm16);
 }
@@ -281,6 +341,19 @@ void Mips64Assembler::Lui(GpuRegister rt, uint16_t imm16) {
   EmitI(0xf, static_cast<GpuRegister>(0), rt, imm16);
 }
 
+void Mips64Assembler::Dahi(GpuRegister rs, uint16_t imm16) {
+  EmitI(1, rs, static_cast<GpuRegister>(6), imm16);
+}
+
+void Mips64Assembler::Dati(GpuRegister rs, uint16_t imm16) {
+  EmitI(1, rs, static_cast<GpuRegister>(0x1e), imm16);
+}
+
+void Mips64Assembler::Sync(uint32_t stype) {
+  EmitR(0, static_cast<GpuRegister>(0), static_cast<GpuRegister>(0),
+           static_cast<GpuRegister>(0), stype & 0x1f, 0xf);
+}
+
 void Mips64Assembler::Mfhi(GpuRegister rd) {
   EmitR(0, static_cast<GpuRegister>(0), static_cast<GpuRegister>(0), rd, 0, 0x10);
 }
@@ -321,34 +394,121 @@ void Mips64Assembler::Sltiu(GpuRegister rt, GpuRegister rs, uint16_t imm16) {
   EmitI(0xb, rs, rt, imm16);
 }
 
-void Mips64Assembler::Beq(GpuRegister rt, GpuRegister rs, uint16_t imm16) {
+void Mips64Assembler::Beq(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
   EmitI(0x4, rs, rt, imm16);
   Nop();
 }
 
-void Mips64Assembler::Bne(GpuRegister rt, GpuRegister rs, uint16_t imm16) {
+void Mips64Assembler::Bne(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
   EmitI(0x5, rs, rt, imm16);
   Nop();
 }
 
-void Mips64Assembler::J(uint32_t address) {
-  EmitJ(0x2, address);
+void Mips64Assembler::J(uint32_t addr26) {
+  EmitJ(0x2, addr26);
   Nop();
 }
 
-void Mips64Assembler::Jal(uint32_t address) {
-  EmitJ(0x2, address);
+void Mips64Assembler::Jal(uint32_t addr26) {
+  EmitJ(0x3, addr26);
   Nop();
 }
 
-void Mips64Assembler::Jr(GpuRegister rs) {
-  EmitR(0, rs, static_cast<GpuRegister>(0), static_cast<GpuRegister>(0), 0, 0x09);  // Jalr zero, rs
+void Mips64Assembler::Jalr(GpuRegister rd, GpuRegister rs) {
+  EmitR(0, rs, static_cast<GpuRegister>(0), rd, 0, 0x09);
   Nop();
 }
 
 void Mips64Assembler::Jalr(GpuRegister rs) {
-  EmitR(0, rs, static_cast<GpuRegister>(0), RA, 0, 0x09);
-  Nop();
+  Jalr(RA, rs);
+}
+
+void Mips64Assembler::Jr(GpuRegister rs) {
+  Jalr(ZERO, rs);
+}
+
+void Mips64Assembler::Auipc(GpuRegister rs, uint16_t imm16) {
+  EmitI(0x3B, rs, static_cast<GpuRegister>(0x1E), imm16);
+}
+
+void Mips64Assembler::Jic(GpuRegister rt, uint16_t imm16) {
+  EmitI(0x36, static_cast<GpuRegister>(0), rt, imm16);
+}
+
+void Mips64Assembler::Jialc(GpuRegister rt, uint16_t imm16) {
+  EmitI(0x3E, static_cast<GpuRegister>(0), rt, imm16);
+}
+
+void Mips64Assembler::Bltc(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x17, rs, rt, imm16);
+}
+
+void Mips64Assembler::Bltzc(GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rt, ZERO);
+  EmitI(0x17, rt, rt, imm16);
+}
+
+void Mips64Assembler::Bgtzc(GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rt, ZERO);
+  EmitI(0x17, static_cast<GpuRegister>(0), rt, imm16);
+}
+
+void Mips64Assembler::Bgec(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x16, rs, rt, imm16);
+}
+
+void Mips64Assembler::Bgezc(GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rt, ZERO);
+  EmitI(0x16, rt, rt, imm16);
+}
+
+void Mips64Assembler::Blezc(GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rt, ZERO);
+  EmitI(0x16, static_cast<GpuRegister>(0), rt, imm16);
+}
+
+void Mips64Assembler::Bltuc(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x7, rs, rt, imm16);
+}
+
+void Mips64Assembler::Bgeuc(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x6, rs, rt, imm16);
+}
+
+void Mips64Assembler::Beqc(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x8, (rs < rt) ? rs : rt, (rs < rt) ? rt : rs, imm16);
+}
+
+void Mips64Assembler::Bnec(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x18, (rs < rt) ? rs : rt, (rs < rt) ? rt : rs, imm16);
+}
+
+void Mips64Assembler::Beqzc(GpuRegister rs, uint32_t imm21) {
+  CHECK_NE(rs, ZERO);
+  EmitI21(0x36, rs, imm21);
+}
+
+void Mips64Assembler::Bnezc(GpuRegister rs, uint32_t imm21) {
+  CHECK_NE(rs, ZERO);
+  EmitI21(0x3E, rs, imm21);
 }
 
 void Mips64Assembler::AddS(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
@@ -368,23 +528,19 @@ void Mips64Assembler::DivS(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
 }
 
 void Mips64Assembler::AddD(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
-  EmitFR(0x11, 0x11, static_cast<FpuRegister>(ft), static_cast<FpuRegister>(fs),
-         static_cast<FpuRegister>(fd), 0x0);
+  EmitFR(0x11, 0x11, ft, fs, fd, 0x0);
 }
 
 void Mips64Assembler::SubD(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
-  EmitFR(0x11, 0x11, static_cast<FpuRegister>(ft), static_cast<FpuRegister>(fs),
-         static_cast<FpuRegister>(fd), 0x1);
+  EmitFR(0x11, 0x11, ft, fs, fd, 0x1);
 }
 
 void Mips64Assembler::MulD(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
-  EmitFR(0x11, 0x11, static_cast<FpuRegister>(ft), static_cast<FpuRegister>(fs),
-         static_cast<FpuRegister>(fd), 0x2);
+  EmitFR(0x11, 0x11, ft, fs, fd, 0x2);
 }
 
 void Mips64Assembler::DivD(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
-  EmitFR(0x11, 0x11, static_cast<FpuRegister>(ft), static_cast<FpuRegister>(fs),
-         static_cast<FpuRegister>(fd), 0x3);
+  EmitFR(0x11, 0x11, ft, fs, fd, 0x3);
 }
 
 void Mips64Assembler::MovS(FpuRegister fd, FpuRegister fs) {
@@ -392,16 +548,47 @@ void Mips64Assembler::MovS(FpuRegister fd, FpuRegister fs) {
 }
 
 void Mips64Assembler::MovD(FpuRegister fd, FpuRegister fs) {
-  EmitFR(0x11, 0x11, static_cast<FpuRegister>(0), static_cast<FpuRegister>(fs),
-         static_cast<FpuRegister>(fd), 0x6);
+  EmitFR(0x11, 0x11, static_cast<FpuRegister>(0), fs, fd, 0x6);
+}
+
+void Mips64Assembler::NegS(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x10, static_cast<FpuRegister>(0), fs, fd, 0x7);
+}
+
+void Mips64Assembler::NegD(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x11, static_cast<FpuRegister>(0), fs, fd, 0x7);
+}
+
+void Mips64Assembler::Cvtsw(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x14, static_cast<FpuRegister>(0), fs, fd, 0x20);
+}
+
+void Mips64Assembler::Cvtdw(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x14, static_cast<FpuRegister>(0), fs, fd, 0x21);
+}
+
+void Mips64Assembler::Cvtsd(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x11, static_cast<FpuRegister>(0), fs, fd, 0x20);
+}
+
+void Mips64Assembler::Cvtds(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x10, static_cast<FpuRegister>(0), fs, fd, 0x21);
 }
 
 void Mips64Assembler::Mfc1(GpuRegister rt, FpuRegister fs) {
   EmitFR(0x11, 0x00, static_cast<FpuRegister>(rt), fs, static_cast<FpuRegister>(0), 0x0);
 }
 
-void Mips64Assembler::Mtc1(FpuRegister ft, GpuRegister rs) {
-  EmitFR(0x11, 0x04, ft, static_cast<FpuRegister>(rs), static_cast<FpuRegister>(0), 0x0);
+void Mips64Assembler::Mtc1(GpuRegister rt, FpuRegister fs) {
+  EmitFR(0x11, 0x04, static_cast<FpuRegister>(rt), fs, static_cast<FpuRegister>(0), 0x0);
+}
+
+void Mips64Assembler::Dmfc1(GpuRegister rt, FpuRegister fs) {
+  EmitFR(0x11, 0x01, static_cast<FpuRegister>(rt), fs, static_cast<FpuRegister>(0), 0x0);
+}
+
+void Mips64Assembler::Dmtc1(GpuRegister rt, FpuRegister fs) {
+  EmitFR(0x11, 0x05, static_cast<FpuRegister>(rt), fs, static_cast<FpuRegister>(0), 0x0);
 }
 
 void Mips64Assembler::Lwc1(FpuRegister ft, GpuRegister rs, uint16_t imm16) {
@@ -430,45 +617,330 @@ void Mips64Assembler::Nop() {
         static_cast<GpuRegister>(0), 0, 0x0);
 }
 
-void Mips64Assembler::Move(GpuRegister rt, GpuRegister rs) {
-  EmitI(0x19, rs, rt, 0);   // Daddiu
+void Mips64Assembler::Move(GpuRegister rd, GpuRegister rs) {
+  Or(rd, rs, ZERO);
 }
 
-void Mips64Assembler::Clear(GpuRegister rt) {
-  EmitR(0, static_cast<GpuRegister>(0), static_cast<GpuRegister>(0), rt, 0, 0x20);
+void Mips64Assembler::Clear(GpuRegister rd) {
+  Move(rd, ZERO);
 }
 
-void Mips64Assembler::Not(GpuRegister rt, GpuRegister rs) {
-  EmitR(0, static_cast<GpuRegister>(0), rs, rt, 0, 0x27);
+void Mips64Assembler::Not(GpuRegister rd, GpuRegister rs) {
+  Nor(rd, rs, ZERO);
 }
 
-void Mips64Assembler::Mul(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
-  Mult(rs, rt);
-  Mflo(rd);
+void Mips64Assembler::LoadConst32(GpuRegister rd, int32_t value) {
+  if (IsUint<16>(value)) {
+    // Use OR with (unsigned) immediate to encode 16b unsigned int.
+    Ori(rd, ZERO, value);
+  } else if (IsInt<16>(value)) {
+    // Use ADD with (signed) immediate to encode 16b signed int.
+    Addiu(rd, ZERO, value);
+  } else {
+    Lui(rd, value >> 16);
+    if (value & 0xFFFF)
+      Ori(rd, rd, value);
+  }
 }
 
-void Mips64Assembler::Div(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
-  Div(rs, rt);
-  Mflo(rd);
+void Mips64Assembler::LoadConst64(GpuRegister rd, int64_t value) {
+  int bit31 = (value & UINT64_C(0x80000000)) != 0;
+
+  // Loads with 1 instruction.
+  if (IsUint<16>(value)) {
+    Ori(rd, ZERO, value);
+  } else if (IsInt<16>(value)) {
+    Daddiu(rd, ZERO, value);
+  } else if ((value & 0xFFFF) == 0 && IsInt<16>(value >> 16)) {
+    Lui(rd, value >> 16);
+  } else if (IsInt<32>(value)) {
+    // Loads with 2 instructions.
+    Lui(rd, value >> 16);
+    Ori(rd, rd, value);
+  } else if ((value & 0xFFFF0000) == 0 && IsInt<16>(value >> 32)) {
+    Ori(rd, ZERO, value);
+    Dahi(rd, value >> 32);
+  } else if ((value & UINT64_C(0xFFFFFFFF0000)) == 0) {
+    Ori(rd, ZERO, value);
+    Dati(rd, value >> 48);
+  } else if ((value & 0xFFFF) == 0 &&
+             (-32768 - bit31) <= (value >> 32) && (value >> 32) <= (32767 - bit31)) {
+    Lui(rd, value >> 16);
+    Dahi(rd, (value >> 32) + bit31);
+  } else if ((value & 0xFFFF) == 0 && ((value >> 31) & 0x1FFFF) == ((0x20000 - bit31) & 0x1FFFF)) {
+    Lui(rd, value >> 16);
+    Dati(rd, (value >> 48) + bit31);
+  } else {
+    int shift_cnt = CTZ(value);
+    int64_t tmp = value >> shift_cnt;
+    if (IsUint<16>(tmp)) {
+      Ori(rd, ZERO, tmp);
+      if (shift_cnt < 32)
+        Dsll(rd, rd, shift_cnt);
+      else
+        Dsll32(rd, rd, shift_cnt & 31);
+    } else if (IsInt<16>(tmp)) {
+      Daddiu(rd, ZERO, tmp);
+      if (shift_cnt < 32)
+        Dsll(rd, rd, shift_cnt);
+      else
+        Dsll32(rd, rd, shift_cnt & 31);
+    } else if (IsInt<32>(tmp)) {
+      // Loads with 3 instructions.
+      Lui(rd, tmp >> 16);
+      Ori(rd, rd, tmp);
+      if (shift_cnt < 32)
+        Dsll(rd, rd, shift_cnt);
+      else
+        Dsll32(rd, rd, shift_cnt & 31);
+    } else {
+      shift_cnt = 16 + CTZ(value >> 16);
+      tmp = value >> shift_cnt;
+      if (IsUint<16>(tmp)) {
+        Ori(rd, ZERO, tmp);
+        if (shift_cnt < 32)
+          Dsll(rd, rd, shift_cnt);
+        else
+          Dsll32(rd, rd, shift_cnt & 31);
+        Ori(rd, rd, value);
+      } else if (IsInt<16>(tmp)) {
+        Daddiu(rd, ZERO, tmp);
+        if (shift_cnt < 32)
+          Dsll(rd, rd, shift_cnt);
+        else
+          Dsll32(rd, rd, shift_cnt & 31);
+        Ori(rd, rd, value);
+      } else {
+        // Loads with 3-4 instructions.
+        uint64_t tmp2 = value;
+        bool used_lui = false;
+        if (((tmp2 >> 16) & 0xFFFF) != 0 || (tmp2 & 0xFFFFFFFF) == 0) {
+          Lui(rd, tmp2 >> 16);
+          used_lui = true;
+        }
+        if ((tmp2 & 0xFFFF) != 0) {
+          if (used_lui)
+            Ori(rd, rd, tmp2);
+          else
+            Ori(rd, ZERO, tmp2);
+        }
+        if (bit31) {
+          tmp2 += UINT64_C(0x100000000);
+        }
+        if (((tmp2 >> 32) & 0xFFFF) != 0) {
+          Dahi(rd, tmp2 >> 32);
+        }
+        if (tmp2 & UINT64_C(0x800000000000)) {
+          tmp2 += UINT64_C(0x1000000000000);
+        }
+        if ((tmp2 >> 48) != 0) {
+          Dati(rd, tmp2 >> 48);
+        }
+      }
+    }
+  }
 }
 
-void Mips64Assembler::Rem(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
-  Div(rs, rt);
-  Mfhi(rd);
+void Mips64Assembler::Addiu32(GpuRegister rt, GpuRegister rs, int32_t value, GpuRegister rtmp) {
+  if (IsInt<16>(value)) {
+    Addiu(rt, rs, value);
+  } else {
+    LoadConst32(rtmp, value);
+    Addu(rt, rs, rtmp);
+  }
 }
 
-void Mips64Assembler::AddConstant64(GpuRegister rt, GpuRegister rs, int32_t value) {
-  CHECK((value >= -32768) && (value <= 32766));
-  Daddiu(rt, rs, value);
+void Mips64Assembler::Daddiu64(GpuRegister rt, GpuRegister rs, int64_t value, GpuRegister rtmp) {
+  if (IsInt<16>(value)) {
+    Daddiu(rt, rs, value);
+  } else {
+    LoadConst64(rtmp, value);
+    Daddu(rt, rs, rtmp);
+  }
 }
 
-void Mips64Assembler::LoadImmediate64(GpuRegister rt, int32_t value) {
-  CHECK((value >= -32768) && (value <= 32766));
-  Daddiu(rt, ZERO, value);
+//
+// MIPS64R6 branches
+//
+//
+// Unconditional (pc + 32-bit signed offset):
+//
+//   auipc    at, ofs_high
+//   jic      at, ofs_low
+//   // no delay/forbidden slot
+//
+//
+// Conditional (pc + 32-bit signed offset):
+//
+//   b<cond>c   reg, +2      // skip next 2 instructions
+//   auipc      at, ofs_high
+//   jic        at, ofs_low
+//   // no delay/forbidden slot
+//
+//
+// Unconditional (pc + 32-bit signed offset) and link:
+//
+//   auipc    reg, ofs_high
+//   daddiu   reg, ofs_low
+//   jialc    reg, 0
+//   // no delay/forbidden slot
+//
+//
+// TODO: use shorter instruction sequences whenever possible.
+//
+
+void Mips64Assembler::Bind(Label* label) {
+  CHECK(!label->IsBound());
+  int32_t bound_pc = buffer_.Size();
+
+  // Walk the list of the branches (auipc + jic pairs) referring to and preceding this label.
+  // Embed the previously unknown pc-relative addresses in them.
+  while (label->IsLinked()) {
+    int32_t position = label->Position();
+    // Extract the branch (instruction pair)
+    uint32_t auipc = buffer_.Load<uint32_t>(position);
+    uint32_t jic = buffer_.Load<uint32_t>(position + 4);  // actually, jic or daddiu
+
+    // Extract the location of the previous pair in the list (walking the list backwards;
+    // the previous pair location was stored in the immediate operands of the instructions)
+    int32_t prev = (auipc << 16) | (jic & 0xFFFF);
+
+    // Get the pc-relative address
+    uint32_t offset = bound_pc - position;
+    offset += (offset & 0x8000) << 1;  // account for sign extension in jic/daddiu
+
+    // Embed it in the two instructions
+    auipc = (auipc & 0xFFFF0000) | (offset >> 16);
+    jic = (jic & 0xFFFF0000) | (offset & 0xFFFF);
+
+    // Save the adjusted instructions
+    buffer_.Store<uint32_t>(position, auipc);
+    buffer_.Store<uint32_t>(position + 4, jic);
+
+    // On to the previous branch in the list...
+    label->position_ = prev;
+  }
+
+  // Now make the label object contain its own location
+  // (it will be used by the branches referring to and following this label)
+  label->BindTo(bound_pc);
+}
+
+void Mips64Assembler::B(Label* label) {
+  if (label->IsBound()) {
+    // Branch backwards (to a preceding label), distance is known
+    uint32_t offset = label->Position() - buffer_.Size();
+    CHECK_LE(static_cast<int32_t>(offset), 0);
+    offset += (offset & 0x8000) << 1;  // account for sign extension in jic
+    Auipc(AT, offset >> 16);
+    Jic(AT, offset);
+  } else {
+    // Branch forward (to a following label), distance is unknown
+    int32_t position = buffer_.Size();
+    // The first branch forward will have 0 in its pc-relative address (copied from label's
+    // position). It will be the terminator of the list of forward-reaching branches.
+    uint32_t prev = label->position_;
+    Auipc(AT, prev >> 16);
+    Jic(AT, prev);
+    // Now make the link object point to the location of this branch
+    // (this forms a linked list of branches preceding this label)
+    label->LinkTo(position);
+  }
+}
+
+void Mips64Assembler::Jalr(Label* label, GpuRegister indirect_reg) {
+  if (label->IsBound()) {
+    // Branch backwards (to a preceding label), distance is known
+    uint32_t offset = label->Position() - buffer_.Size();
+    CHECK_LE(static_cast<int32_t>(offset), 0);
+    offset += (offset & 0x8000) << 1;  // account for sign extension in daddiu
+    Auipc(indirect_reg, offset >> 16);
+    Daddiu(indirect_reg, indirect_reg, offset);
+    Jialc(indirect_reg, 0);
+  } else {
+    // Branch forward (to a following label), distance is unknown
+    int32_t position = buffer_.Size();
+    // The first branch forward will have 0 in its pc-relative address (copied from label's
+    // position). It will be the terminator of the list of forward-reaching branches.
+    uint32_t prev = label->position_;
+    Auipc(indirect_reg, prev >> 16);
+    Daddiu(indirect_reg, indirect_reg, prev);
+    Jialc(indirect_reg, 0);
+    // Now make the link object point to the location of this branch
+    // (this forms a linked list of branches preceding this label)
+    label->LinkTo(position);
+  }
+}
+
+void Mips64Assembler::Bltc(GpuRegister rs, GpuRegister rt, Label* label) {
+  Bgec(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bltzc(GpuRegister rt, Label* label) {
+  Bgezc(rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bgtzc(GpuRegister rt, Label* label) {
+  Blezc(rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bgec(GpuRegister rs, GpuRegister rt, Label* label) {
+  Bltc(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bgezc(GpuRegister rt, Label* label) {
+  Bltzc(rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Blezc(GpuRegister rt, Label* label) {
+  Bgtzc(rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bltuc(GpuRegister rs, GpuRegister rt, Label* label) {
+  Bgeuc(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bgeuc(GpuRegister rs, GpuRegister rt, Label* label) {
+  Bltuc(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Beqc(GpuRegister rs, GpuRegister rt, Label* label) {
+  Bnec(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bnec(GpuRegister rs, GpuRegister rt, Label* label) {
+  Beqc(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Beqzc(GpuRegister rs, Label* label) {
+  Bnezc(rs, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bnezc(GpuRegister rs, Label* label) {
+  Beqzc(rs, 2);
+  B(label);
 }
 
 void Mips64Assembler::LoadFromOffset(LoadOperandType type, GpuRegister reg, GpuRegister base,
                                      int32_t offset) {
+  if (!IsInt<16>(offset)) {
+    LoadConst32(AT, offset);
+    Daddu(AT, AT, base);
+    base = AT;
+    offset = 0;
+  }
+
   switch (type) {
     case kLoadSignedByte:
       Lb(reg, base, offset);
@@ -489,23 +961,25 @@ void Mips64Assembler::LoadFromOffset(LoadOperandType type, GpuRegister reg, GpuR
       Lwu(reg, base, offset);
       break;
     case kLoadDoubleword:
-      // TODO: alignment issues ???
       Ld(reg, base, offset);
       break;
-    default:
-      LOG(FATAL) << "UNREACHABLE";
   }
 }
 
 void Mips64Assembler::LoadFpuFromOffset(LoadOperandType type, FpuRegister reg, GpuRegister base,
                                         int32_t offset) {
-  CHECK((offset >= -32768) && (offset <= 32766));
+  if (!IsInt<16>(offset)) {
+    LoadConst32(AT, offset);
+    Daddu(AT, AT, base);
+    base = AT;
+    offset = 0;
+  }
+
   switch (type) {
     case kLoadWord:
       Lwc1(reg, base, offset);
       break;
     case kLoadDoubleword:
-      // TODO: alignment issues ???
       Ldc1(reg, base, offset);
       break;
     default:
@@ -542,6 +1016,13 @@ void Mips64Assembler::EmitLoad(ManagedRegister m_dst, GpuRegister src_register,
 
 void Mips64Assembler::StoreToOffset(StoreOperandType type, GpuRegister reg, GpuRegister base,
                                     int32_t offset) {
+  if (!IsInt<16>(offset)) {
+    LoadConst32(AT, offset);
+    Daddu(AT, AT, base);
+    base = AT;
+    offset = 0;
+  }
+
   switch (type) {
     case kStoreByte:
       Sb(reg, base, offset);
@@ -553,7 +1034,6 @@ void Mips64Assembler::StoreToOffset(StoreOperandType type, GpuRegister reg, GpuR
       Sw(reg, base, offset);
       break;
     case kStoreDoubleword:
-      // TODO: alignment issues ???
       Sd(reg, base, offset);
       break;
     default:
@@ -563,6 +1043,13 @@ void Mips64Assembler::StoreToOffset(StoreOperandType type, GpuRegister reg, GpuR
 
 void Mips64Assembler::StoreFpuToOffset(StoreOperandType type, FpuRegister reg, GpuRegister base,
                                        int32_t offset) {
+  if (!IsInt<16>(offset)) {
+    LoadConst32(AT, offset);
+    Daddu(AT, AT, base);
+    base = AT;
+    offset = 0;
+  }
+
   switch (type) {
     case kStoreWord:
       Swc1(reg, base, offset);
@@ -613,10 +1100,12 @@ void Mips64Assembler::BuildFrame(size_t frame_size, ManagedRegister method_reg,
       // only increment stack offset.
       offset += size;
     } else if (reg.IsFpuRegister()) {
-      StoreFpuToOffset((size == 4) ? kStoreWord : kStoreDoubleword, reg.AsFpuRegister(), SP, offset);
+      StoreFpuToOffset((size == 4) ? kStoreWord : kStoreDoubleword,
+          reg.AsFpuRegister(), SP, offset);
       offset += size;
     } else if (reg.IsGpuRegister()) {
-      StoreToOffset((size == 4) ? kStoreWord : kStoreDoubleword, reg.AsGpuRegister(), SP, offset);
+      StoreToOffset((size == 4) ? kStoreWord : kStoreDoubleword,
+          reg.AsGpuRegister(), SP, offset);
       offset += size;
     }
   }
@@ -650,14 +1139,14 @@ void Mips64Assembler::RemoveFrame(size_t frame_size,
 }
 
 void Mips64Assembler::IncreaseFrameSize(size_t adjust) {
-  CHECK_ALIGNED(adjust, kStackAlignment);
-  AddConstant64(SP, SP, -adjust);
+  CHECK_ALIGNED(adjust, kFramePointerSize);
+  Daddiu64(SP, SP, static_cast<int32_t>(-adjust));
   cfi_.AdjustCFAOffset(adjust);
 }
 
 void Mips64Assembler::DecreaseFrameSize(size_t adjust) {
-  CHECK_ALIGNED(adjust, kStackAlignment);
-  AddConstant64(SP, SP, adjust);
+  CHECK_ALIGNED(adjust, kFramePointerSize);
+  Daddiu64(SP, SP, static_cast<int32_t>(adjust));
   cfi_.AdjustCFAOffset(-adjust);
 }
 
@@ -702,7 +1191,7 @@ void Mips64Assembler::StoreImmediateToFrame(FrameOffset dest, uint32_t imm,
                                             ManagedRegister mscratch) {
   Mips64ManagedRegister scratch = mscratch.AsMips64();
   CHECK(scratch.IsGpuRegister()) << scratch;
-  LoadImmediate64(scratch.AsGpuRegister(), imm);
+  LoadConst32(scratch.AsGpuRegister(), imm);
   StoreToOffset(kStoreWord, scratch.AsGpuRegister(), SP, dest.Int32Value());
 }
 
@@ -710,7 +1199,9 @@ void Mips64Assembler::StoreImmediateToThread64(ThreadOffset<8> dest, uint32_t im
                                                ManagedRegister mscratch) {
   Mips64ManagedRegister scratch = mscratch.AsMips64();
   CHECK(scratch.IsGpuRegister()) << scratch;
-  LoadImmediate64(scratch.AsGpuRegister(), imm);
+  // TODO: it's unclear wether 32 or 64 bits need to be stored (Arm64 and x86/x64 disagree?).
+  // Is this function even referenced anywhere else in the code?
+  LoadConst32(scratch.AsGpuRegister(), imm);
   StoreToOffset(kStoreDoubleword, scratch.AsGpuRegister(), S1, dest.Int32Value());
 }
 
@@ -719,7 +1210,7 @@ void Mips64Assembler::StoreStackOffsetToThread64(ThreadOffset<8> thr_offs,
                                                  ManagedRegister mscratch) {
   Mips64ManagedRegister scratch = mscratch.AsMips64();
   CHECK(scratch.IsGpuRegister()) << scratch;
-  AddConstant64(scratch.AsGpuRegister(), SP, fr_offs.Int32Value());
+  Daddiu64(scratch.AsGpuRegister(), SP, fr_offs.Int32Value());
   StoreToOffset(kStoreDoubleword, scratch.AsGpuRegister(), S1, thr_offs.Int32Value());
 }
 
@@ -757,20 +1248,24 @@ void Mips64Assembler::LoadRef(ManagedRegister mdest, ManagedRegister base, Membe
   LoadFromOffset(kLoadUnsignedWord, dest.AsGpuRegister(),
                  base.AsMips64().AsGpuRegister(), offs.Int32Value());
   if (kPoisonHeapReferences && poison_reference) {
-    Subu(dest.AsGpuRegister(), ZERO, dest.AsGpuRegister());
+    // TODO: review
+    // Negate the 32-bit ref
+    Dsubu(dest.AsGpuRegister(), ZERO, dest.AsGpuRegister());
+    // And constrain it to 32 bits (zero-extend into bits 32 through 63) as on Arm64 and x86/64
+    Dext(dest.AsGpuRegister(), dest.AsGpuRegister(), 0, 31);
   }
 }
 
 void Mips64Assembler::LoadRawPtr(ManagedRegister mdest, ManagedRegister base,
-                               Offset offs) {
+                                 Offset offs) {
   Mips64ManagedRegister dest = mdest.AsMips64();
-  CHECK(dest.IsGpuRegister() && dest.IsGpuRegister()) << dest;
+  CHECK(dest.IsGpuRegister() && base.AsMips64().IsGpuRegister());
   LoadFromOffset(kLoadDoubleword, dest.AsGpuRegister(),
                  base.AsMips64().AsGpuRegister(), offs.Int32Value());
 }
 
 void Mips64Assembler::LoadRawPtrFromThread64(ManagedRegister mdest,
-                                         ThreadOffset<8> offs) {
+                                             ThreadOffset<8> offs) {
   Mips64ManagedRegister dest = mdest.AsMips64();
   CHECK(dest.IsGpuRegister());
   LoadFromOffset(kLoadDoubleword, dest.AsGpuRegister(), S1, offs.Int32Value());
@@ -849,7 +1344,7 @@ void Mips64Assembler::Copy(FrameOffset dest, FrameOffset src,
 }
 
 void Mips64Assembler::Copy(FrameOffset dest, ManagedRegister src_base, Offset src_offset,
-                         ManagedRegister mscratch, size_t size) {
+                           ManagedRegister mscratch, size_t size) {
   GpuRegister scratch = mscratch.AsMips64().AsGpuRegister();
   CHECK(size == 4 || size == 8) << size;
   if (size == 4) {
@@ -866,7 +1361,7 @@ void Mips64Assembler::Copy(FrameOffset dest, ManagedRegister src_base, Offset sr
 }
 
 void Mips64Assembler::Copy(ManagedRegister dest_base, Offset dest_offset, FrameOffset src,
-                         ManagedRegister mscratch, size_t size) {
+                           ManagedRegister mscratch, size_t size) {
   GpuRegister scratch = mscratch.AsMips64().AsGpuRegister();
   CHECK(size == 4 || size == 8) << size;
   if (size == 4) {
@@ -888,8 +1383,8 @@ void Mips64Assembler::Copy(FrameOffset /*dest*/, FrameOffset /*src_base*/, Offse
 }
 
 void Mips64Assembler::Copy(ManagedRegister dest, Offset dest_offset,
-                         ManagedRegister src, Offset src_offset,
-                         ManagedRegister mscratch, size_t size) {
+                           ManagedRegister src, Offset src_offset,
+                           ManagedRegister mscratch, size_t size) {
   GpuRegister scratch = mscratch.AsMips64().AsGpuRegister();
   CHECK(size == 4 || size == 8) << size;
   if (size == 4) {
@@ -912,12 +1407,14 @@ void Mips64Assembler::Copy(FrameOffset /*dest*/, Offset /*dest_offset*/, FrameOf
 }
 
 void Mips64Assembler::MemoryBarrier(ManagedRegister) {
+  // TODO: sync?
   UNIMPLEMENTED(FATAL) << "no mips64 implementation";
 }
 
 void Mips64Assembler::CreateHandleScopeEntry(ManagedRegister mout_reg,
-                                    FrameOffset handle_scope_offset,
-                                    ManagedRegister min_reg, bool null_allowed) {
+                                             FrameOffset handle_scope_offset,
+                                             ManagedRegister min_reg,
+                                             bool null_allowed) {
   Mips64ManagedRegister out_reg = mout_reg.AsMips64();
   Mips64ManagedRegister in_reg = min_reg.AsMips64();
   CHECK(in_reg.IsNoRegister() || in_reg.IsGpuRegister()) << in_reg;
@@ -933,20 +1430,20 @@ void Mips64Assembler::CreateHandleScopeEntry(ManagedRegister mout_reg,
       in_reg = out_reg;
     }
     if (!out_reg.Equals(in_reg)) {
-      LoadImmediate64(out_reg.AsGpuRegister(), 0);
+      LoadConst32(out_reg.AsGpuRegister(), 0);
     }
-    EmitBranch(in_reg.AsGpuRegister(), ZERO, &null_arg, true);
-    AddConstant64(out_reg.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
-    Bind(&null_arg, false);
+    Beqzc(in_reg.AsGpuRegister(), &null_arg);
+    Daddiu64(out_reg.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
+    Bind(&null_arg);
   } else {
-    AddConstant64(out_reg.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
+    Daddiu64(out_reg.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
   }
 }
 
 void Mips64Assembler::CreateHandleScopeEntry(FrameOffset out_off,
-                                    FrameOffset handle_scope_offset,
-                                    ManagedRegister mscratch,
-                                    bool null_allowed) {
+                                             FrameOffset handle_scope_offset,
+                                             ManagedRegister mscratch,
+                                             bool null_allowed) {
   Mips64ManagedRegister scratch = mscratch.AsMips64();
   CHECK(scratch.IsGpuRegister()) << scratch;
   if (null_allowed) {
@@ -956,30 +1453,30 @@ void Mips64Assembler::CreateHandleScopeEntry(FrameOffset out_off,
     // Null values get a handle scope entry value of 0.  Otherwise, the handle scope entry is
     // the address in the handle scope holding the reference.
     // e.g. scratch = (scratch == 0) ? 0 : (SP+handle_scope_offset)
-    EmitBranch(scratch.AsGpuRegister(), ZERO, &null_arg, true);
-    AddConstant64(scratch.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
-    Bind(&null_arg, false);
+    Beqzc(scratch.AsGpuRegister(), &null_arg);
+    Daddiu64(scratch.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
+    Bind(&null_arg);
   } else {
-    AddConstant64(scratch.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
+    Daddiu64(scratch.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
   }
   StoreToOffset(kStoreDoubleword, scratch.AsGpuRegister(), SP, out_off.Int32Value());
 }
 
 // Given a handle scope entry, load the associated reference.
 void Mips64Assembler::LoadReferenceFromHandleScope(ManagedRegister mout_reg,
-                                          ManagedRegister min_reg) {
+                                                   ManagedRegister min_reg) {
   Mips64ManagedRegister out_reg = mout_reg.AsMips64();
   Mips64ManagedRegister in_reg = min_reg.AsMips64();
   CHECK(out_reg.IsGpuRegister()) << out_reg;
   CHECK(in_reg.IsGpuRegister()) << in_reg;
   Label null_arg;
   if (!out_reg.Equals(in_reg)) {
-    LoadImmediate64(out_reg.AsGpuRegister(), 0);
+    LoadConst32(out_reg.AsGpuRegister(), 0);
   }
-  EmitBranch(in_reg.AsGpuRegister(), ZERO, &null_arg, true);
+  Beqzc(in_reg.AsGpuRegister(), &null_arg);
   LoadFromOffset(kLoadDoubleword, out_reg.AsGpuRegister(),
                  in_reg.AsGpuRegister(), 0);
-  Bind(&null_arg, false);
+  Bind(&null_arg);
 }
 
 void Mips64Assembler::VerifyObject(ManagedRegister /*src*/, bool /*could_be_null*/) {
@@ -1022,7 +1519,7 @@ void Mips64Assembler::GetCurrentThread(ManagedRegister tr) {
 }
 
 void Mips64Assembler::GetCurrentThread(FrameOffset offset,
-                                     ManagedRegister /*mscratch*/) {
+                                       ManagedRegister /*mscratch*/) {
   StoreToOffset(kStoreDoubleword, S1, SP, offset.Int32Value());
 }
 
@@ -1032,13 +1529,13 @@ void Mips64Assembler::ExceptionPoll(ManagedRegister mscratch, size_t stack_adjus
   buffer_.EnqueueSlowPath(slow);
   LoadFromOffset(kLoadDoubleword, scratch.AsGpuRegister(),
                  S1, Thread::ExceptionOffset<8>().Int32Value());
-  EmitBranch(scratch.AsGpuRegister(), ZERO, slow->Entry(), false);
+  Bnezc(scratch.AsGpuRegister(), slow->Entry());
 }
 
 void Mips64ExceptionSlowPath::Emit(Assembler* sasm) {
   Mips64Assembler* sp_asm = down_cast<Mips64Assembler*>(sasm);
 #define __ sp_asm->
-  __ Bind(&entry_, false);
+  __ Bind(&entry_);
   if (stack_adjust_ != 0) {  // Fix up the frame.
     __ DecreaseFrameSize(stack_adjust_);
   }
@@ -1048,6 +1545,7 @@ void Mips64ExceptionSlowPath::Emit(Assembler* sasm) {
   // Set up call to Thread::Current()->pDeliverException
   __ LoadFromOffset(kLoadDoubleword, T9, S1,
                     QUICK_ENTRYPOINT_OFFSET(8, pDeliverException).Int32Value());
+  // TODO: check T9 usage
   __ Jr(T9);
   // Call never returns
   __ Break();
diff --git a/compiler/utils/mips64/assembler_mips64.h b/compiler/utils/mips64/assembler_mips64.h
index 38419abbac..88cc4bcd00 100644
--- a/compiler/utils/mips64/assembler_mips64.h
+++ b/compiler/utils/mips64/assembler_mips64.h
@@ -56,13 +56,31 @@ class Mips64Assembler FINAL : public Assembler {
   void Addi(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Addu(GpuRegister rd, GpuRegister rs, GpuRegister rt);
   void Addiu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void Daddiu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
+  void Daddu(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64
+  void Daddiu(GpuRegister rt, GpuRegister rs, uint16_t imm16);  // MIPS64
   void Sub(GpuRegister rd, GpuRegister rs, GpuRegister rt);
   void Subu(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-  void Mult(GpuRegister rs, GpuRegister rt);
-  void Multu(GpuRegister rs, GpuRegister rt);
-  void Div(GpuRegister rs, GpuRegister rt);
-  void Divu(GpuRegister rs, GpuRegister rt);
+  void Dsubu(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64
+
+  void MultR2(GpuRegister rs, GpuRegister rt);  // R2
+  void MultuR2(GpuRegister rs, GpuRegister rt);  // R2
+  void DivR2(GpuRegister rs, GpuRegister rt);  // R2
+  void DivuR2(GpuRegister rs, GpuRegister rt);  // R2
+  void MulR2(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R2
+  void DivR2(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R2
+  void ModR2(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R2
+  void DivuR2(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R2
+  void ModuR2(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R2
+  void MulR6(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R6
+  void DivR6(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R6
+  void ModR6(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R6
+  void DivuR6(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R6
+  void ModuR6(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R6
+  void Dmul(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64 R6
+  void Ddiv(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64 R6
+  void Dmod(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64 R6
+  void Ddivu(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64 R6
+  void Dmodu(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64 R6
 
   void And(GpuRegister rd, GpuRegister rs, GpuRegister rt);
   void Andi(GpuRegister rt, GpuRegister rs, uint16_t imm16);
@@ -72,40 +90,72 @@ class Mips64Assembler FINAL : public Assembler {
   void Xori(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Nor(GpuRegister rd, GpuRegister rs, GpuRegister rt);
 
-  void Sll(GpuRegister rd, GpuRegister rs, int shamt);
-  void Srl(GpuRegister rd, GpuRegister rs, int shamt);
-  void Sra(GpuRegister rd, GpuRegister rs, int shamt);
-  void Sllv(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-  void Srlv(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-  void Srav(GpuRegister rd, GpuRegister rs, GpuRegister rt);
+  void Seb(GpuRegister rd, GpuRegister rt);  // R2+
+  void Seh(GpuRegister rd, GpuRegister rt);  // R2+
+  void Dext(GpuRegister rs, GpuRegister rt, int pos, int size_less_one);  // MIPS64
+
+  void Sll(GpuRegister rd, GpuRegister rt, int shamt);
+  void Srl(GpuRegister rd, GpuRegister rt, int shamt);
+  void Sra(GpuRegister rd, GpuRegister rt, int shamt);
+  void Sllv(GpuRegister rd, GpuRegister rt, GpuRegister rs);
+  void Srlv(GpuRegister rd, GpuRegister rt, GpuRegister rs);
+  void Srav(GpuRegister rd, GpuRegister rt, GpuRegister rs);
+  void Dsll(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsrl(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsra(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsll32(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsrl32(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsra32(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsllv(GpuRegister rd, GpuRegister rt, GpuRegister rs);  // MIPS64
+  void Dsrlv(GpuRegister rd, GpuRegister rt, GpuRegister rs);  // MIPS64
+  void Dsrav(GpuRegister rd, GpuRegister rt, GpuRegister rs);  // MIPS64
 
   void Lb(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Lh(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Lw(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void Ld(GpuRegister rt, GpuRegister rs, uint16_t imm16);
+  void Ld(GpuRegister rt, GpuRegister rs, uint16_t imm16);  // MIPS64
   void Lbu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Lhu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void Lwu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
+  void Lwu(GpuRegister rt, GpuRegister rs, uint16_t imm16);  // MIPS64
   void Lui(GpuRegister rt, uint16_t imm16);
-  void Mfhi(GpuRegister rd);
-  void Mflo(GpuRegister rd);
+  void Dahi(GpuRegister rs, uint16_t imm16);  // MIPS64 R6
+  void Dati(GpuRegister rs, uint16_t imm16);  // MIPS64 R6
+  void Sync(uint32_t stype);
+  void Mfhi(GpuRegister rd);  // R2
+  void Mflo(GpuRegister rd);  // R2
 
   void Sb(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Sh(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Sw(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void Sd(GpuRegister rt, GpuRegister rs, uint16_t imm16);
+  void Sd(GpuRegister rt, GpuRegister rs, uint16_t imm16);  // MIPS64
 
   void Slt(GpuRegister rd, GpuRegister rs, GpuRegister rt);
   void Sltu(GpuRegister rd, GpuRegister rs, GpuRegister rt);
   void Slti(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Sltiu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
 
-  void Beq(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void Bne(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void J(uint32_t address);
-  void Jal(uint32_t address);
-  void Jr(GpuRegister rs);
+  void Beq(GpuRegister rs, GpuRegister rt, uint16_t imm16);
+  void Bne(GpuRegister rs, GpuRegister rt, uint16_t imm16);
+  void J(uint32_t addr26);
+  void Jal(uint32_t addr26);
+  void Jalr(GpuRegister rd, GpuRegister rs);
   void Jalr(GpuRegister rs);
+  void Jr(GpuRegister rs);
+  void Auipc(GpuRegister rs, uint16_t imm16);  // R6
+  void Jic(GpuRegister rt, uint16_t imm16);  // R6
+  void Jialc(GpuRegister rt, uint16_t imm16);  // R6
+  void Bltc(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Bltzc(GpuRegister rt, uint16_t imm16);  // R6
+  void Bgtzc(GpuRegister rt, uint16_t imm16);  // R6
+  void Bgec(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Bgezc(GpuRegister rt, uint16_t imm16);  // R6
+  void Blezc(GpuRegister rt, uint16_t imm16);  // R6
+  void Bltuc(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Bgeuc(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Beqc(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Bnec(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Beqzc(GpuRegister rs, uint32_t imm21);  // R6
+  void Bnezc(GpuRegister rs, uint32_t imm21);  // R6
 
   void AddS(FpuRegister fd, FpuRegister fs, FpuRegister ft);
   void SubS(FpuRegister fd, FpuRegister fs, FpuRegister ft);
@@ -117,9 +167,18 @@ class Mips64Assembler FINAL : public Assembler {
   void DivD(FpuRegister fd, FpuRegister fs, FpuRegister ft);
   void MovS(FpuRegister fd, FpuRegister fs);
   void MovD(FpuRegister fd, FpuRegister fs);
+  void NegS(FpuRegister fd, FpuRegister fs);
+  void NegD(FpuRegister fd, FpuRegister fs);
+
+  void Cvtsw(FpuRegister fd, FpuRegister fs);
+  void Cvtdw(FpuRegister fd, FpuRegister fs);
+  void Cvtsd(FpuRegister fd, FpuRegister fs);
+  void Cvtds(FpuRegister fd, FpuRegister fs);
 
   void Mfc1(GpuRegister rt, FpuRegister fs);
-  void Mtc1(FpuRegister ft, GpuRegister rs);
+  void Mtc1(GpuRegister rt, FpuRegister fs);
+  void Dmfc1(GpuRegister rt, FpuRegister fs);  // MIPS64
+  void Dmtc1(GpuRegister rt, FpuRegister fs);  // MIPS64
   void Lwc1(FpuRegister ft, GpuRegister rs, uint16_t imm16);
   void Ldc1(FpuRegister ft, GpuRegister rs, uint16_t imm16);
   void Swc1(FpuRegister ft, GpuRegister rs, uint16_t imm16);
@@ -127,15 +186,33 @@ class Mips64Assembler FINAL : public Assembler {
 
   void Break();
   void Nop();
-  void Move(GpuRegister rt, GpuRegister rs);
-  void Clear(GpuRegister rt);
-  void Not(GpuRegister rt, GpuRegister rs);
-  void Mul(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-  void Div(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-  void Rem(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-
-  void AddConstant64(GpuRegister rt, GpuRegister rs, int32_t value);
-  void LoadImmediate64(GpuRegister rt, int32_t value);
+  void Move(GpuRegister rd, GpuRegister rs);
+  void Clear(GpuRegister rd);
+  void Not(GpuRegister rd, GpuRegister rs);
+
+  // Higher level composite instructions
+  void LoadConst32(GpuRegister rd, int32_t value);
+  void LoadConst64(GpuRegister rd, int64_t value);  // MIPS64
+
+  void Addiu32(GpuRegister rt, GpuRegister rs, int32_t value, GpuRegister rtmp = AT);
+  void Daddiu64(GpuRegister rt, GpuRegister rs, int64_t value, GpuRegister rtmp = AT);  // MIPS64
+
+  void Bind(Label* label);  // R6
+  void B(Label* label);  // R6
+  void Jalr(Label* label, GpuRegister indirect_reg = RA);  // R6
+  // TODO: implement common for R6 and non-R6 interface for conditional branches?
+  void Bltc(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Bltzc(GpuRegister rt, Label* label);  // R6
+  void Bgtzc(GpuRegister rt, Label* label);  // R6
+  void Bgec(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Bgezc(GpuRegister rt, Label* label);  // R6
+  void Blezc(GpuRegister rt, Label* label);  // R6
+  void Bltuc(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Bgeuc(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Beqc(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Bnec(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Beqzc(GpuRegister rs, Label* label);  // R6
+  void Bnezc(GpuRegister rs, Label* label);  // R6
 
   void EmitLoad(ManagedRegister m_dst, GpuRegister src_register, int32_t src_offset, size_t size);
   void LoadFromOffset(LoadOperandType type, GpuRegister reg, GpuRegister base, int32_t offset);
@@ -144,10 +221,7 @@ class Mips64Assembler FINAL : public Assembler {
   void StoreFpuToOffset(StoreOperandType type, FpuRegister reg, GpuRegister base, int32_t offset);
 
   // Emit data (e.g. encoded instruction or immediate) to the instruction stream.
-  void Emit(int32_t value);
-  void EmitBranch(GpuRegister rt, GpuRegister rs, Label* label, bool equal);
-  void EmitJump(Label* label, bool link);
-  void Bind(Label* label, bool is_jump);
+  void Emit(uint32_t value);
 
   //
   // Overridden common assembler high-level functionality
@@ -269,13 +343,11 @@ class Mips64Assembler FINAL : public Assembler {
  private:
   void EmitR(int opcode, GpuRegister rs, GpuRegister rt, GpuRegister rd, int shamt, int funct);
   void EmitI(int opcode, GpuRegister rs, GpuRegister rt, uint16_t imm);
-  void EmitJ(int opcode, int address);
+  void EmitI21(int opcode, GpuRegister rs, uint32_t imm21);
+  void EmitJ(int opcode, uint32_t addr26);
   void EmitFR(int opcode, int fmt, FpuRegister ft, FpuRegister fs, FpuRegister fd, int funct);
   void EmitFI(int opcode, int fmt, FpuRegister rt, uint16_t imm);
 
-  int32_t EncodeBranchOffset(int offset, int32_t inst, bool is_jump);
-  int DecodeBranchOffset(int32_t inst, bool is_jump);
-
   DISALLOW_COPY_AND_ASSIGN(Mips64Assembler);
 };
 
diff --git a/compiler/utils/mips64/constants_mips64.h b/compiler/utils/mips64/constants_mips64.h
index 8b7697cac3..f57498d34f 100644
--- a/compiler/utils/mips64/constants_mips64.h
+++ b/compiler/utils/mips64/constants_mips64.h
@@ -67,7 +67,7 @@ class Instr {
   static const uint32_t kBreakPointInstruction = 0x0000000D;
 
   bool IsBreakPoint() {
-    return ((*reinterpret_cast<const uint32_t*>(this)) & 0xFC0000CF) == kBreakPointInstruction;
+    return ((*reinterpret_cast<const uint32_t*>(this)) & 0xFC00003F) == kBreakPointInstruction;
   }
 
   // Instructions are read out of a code stream. The only way to get a