[mips[microMIPS]] Adding code size reduction pass for MicroMIPS

[milena.vujosevic.janicic]

Summary

The code implements size reduction pass for MicroMIPS.

Load and store instructions are examined and transformed, if possible.
lw32 instruction is transformed into 16-bit instruction lwsp
sw32 instruction is transformed into 16-bit instruction swsp

Arithmetic instrcutions are examined and transformed, if possible.
addu32 instruction is transformed into 16-bit instruction addu16
subu32 instruction is transformed into 16-bit instruction subu16

[milena.vujosevic.janicic]

New patch version rebased to revision 259635.
Any comments to this work?

[milena.vujosevic.janicic]

New patch version rebased to revision 264141.
Any comments?

[dsanders]

Sorry, I was part way through writing them a few weeks ago but was distracted by other things.

There appears to be two optimizations in this pass with very different requirements at the moment. The first optimization is a simple substitution of an MI for an equivalent MI with a smaller encoding. This part is generally heading in the right direction. The second is a peephole optimization that reduces two or more MI's into a single MI and this is where most of my concerns are. I don't believe it's checking enough to be able to prove that this reduction is safe. For example, ReduceMIToLwpSwp checks for interfering register uses but fails to check for interfering register defs (including implicit defs and sub/super-registers), memory reads/writes (including aliases), volatile accesses, side effects, etc. I think we should remove this portion for now and proceed with the simple size reductions to begin with.

For the testing in general: We ought to make use of the MIR (http://llvm.org/docs/MIRLangRef.html) so that we're only testing this pass. However, I'm not going to make that a requirement for this patch because I haven't used it myself yet.

I haven't looked too closely at the test cases yet but they will need to check the operands since this is a key part of whether your optimization is working as intended. I'd also like the tests to be more focused than they currently are. They look like they were generated from C examples and as such have a lot of unnecessary noise.

The rest of the comments below are things I noted while reading the patch. I've included them because I've already written them but some will most likely be made moot by the above changes. If the line numbers seem odd it's because they were written for the previous diff.

lib/Target/Mips/MicroMips32SizeReduction.cpp

25

Do we really need std::vector or can we use one of the alternatives from http://llvm.org/docs/ProgrammersManual.html#sequential-containers-std-vector-std-list-etc?

For example, I see there's a std::vector<MachineOperand> below. This would probably be better as a SmallVector<MachineOperand, 4> or similar.

38

What do these enumerators mean?

Naming nit: Types and enumerators should begin with a capital. They should also have a prefix such as 'ON_'. More information can be found at http://llvm.org/docs/CodingStandards.html#name-types-functions-variables-and-enumerators-properly

602

Is the double-N meaninful?

605

Please delete the commented out code.

[dsanders]

lib/Target/Mips/MicroMips32SizeReduction.cpp

40–43

We can make this explanation appear in the doxygen documentation by writing this with '/' and '/<' comments like so:

/// Reduction type:
enum ReduceType {
  SeveralInstr, ///< Several instructions into lwm/swm.
  TwoInstr, ///< Two instructions into one.
  OneInstr ///< 32-bit instruction into 16-bit instruction.
};

Similarly for the other description comments below.

I notice that our doxygen config currently has 'EXTRACT_ANON_NSPACES=NO' but I'm going to propose that we change that.

47

Opperand -> Operand. This typo appears in a few other places too

48

Variables should begin with a capital and should be descriptive. With this style of constructor it's not ambiguous to use the same name for both the argument and the member (e.g. 'Shift(Shift)').

Similarly for the other constructors below.

59

The snr argument is never used.

64

I think I know what you're trying to say but the comment isn't very clear. I think you're referring to the way LWM16 only allows a subset of the registers that LVM32 allows.

Can we describe this in terms of register classes?

73–75

We normally use 'unsigned' for opcodes.

Also, what's the purpose of the second instruction? It's not clear from the comment

83

Why 'void *'? It seems we always pass in a 'struct ReduceEntryFA *'. We also de-reference it and take a copy immediately without nullptr checks so a reference would be better to avoid the copy and explicitly say it can't be nullptr at the same time.

124–125

Formatting.

143

Naming nit: We should probably drop the '32' so that we can re-use it for microMIPS64 in the future.

158

New code shouldn't repeat the function name in the comments.

207–208

Given that this is a static table, we should define the table as a normal array and use ArrayRef in this class

212

Is this redundant?

214–269

This table should probably be tablegen-erated but we can leave that for now and address it in later patches.

272–288

This is equivalent to GPRMM16RegClass.contains(Reg)

290–306

Similarly, this is equivalent to GPRMM16ZeroRegClass.contains(Reg)

307–312

This is only correct for the o32 and n32 ABIs. If you check for Mips::SP64 as well then it will cover the n64 ABI too.

343–344

I'd expect this to be indicative of a bug somewhere else. Should it be an assertion?

353

Operand indices are 'unsigned' rather than uint8_t

354–361

Am I right in thinking this is to check the pointer registers of each instruction are the same? If so, this should be ok but the function name should indicate that it's only suitable for pointers.

If integers are a possibility then we will also need to handle the fact that V0 != V0_64 despite being the same register.

366–381

I haven't tested this but something like:

const auto &End = GPR32RegClass.end();
const auto &I = std::find(GPR32RegClass.begin(), End, Reg1);
if (I == End || *I != Reg1)
  return false;
I++;
if (*I == Reg2)
  return true;
return false;

should be the equivalent without duplicating our register classes.

We ought to account for the '*_64' versions of these registers too which can be handled using GPR64RegClass.

385–386

Line wrapping

387–393

MathExtras.h has isShiftedInt() and isShiftedUInt() templates that are equivalent to this function

474–485

This is equivalent to this function:

MI->readsRegister(reg1) || MI->readsRegister(reg2)

If you pass the TRI argument then it will check for reads that occur because of super-register reads too. I don't think it can check for reads caused by sub-register reads though.

492

We should use C++11's range based for loop

for (const auto &I : MI->operands())

507–508

According to the tablegen definition, it's not guaranteed to be operand 2 when variable_ops for the Lwm/Swm is non-empty. It will be operand NumOps-1

516–517

Similarly, it's not guaranteed to be operand 1 when variable_ops for the Lwm/Swm is non-empty. It will be operand NumOps-2

532–533

Likewise

539

At minimum we have two sources/results ($16 and $31) along with a base address and offset so shouldn't the lower bound be 3.

Similarly: At most, we have five sources/results ($16-$19, and $31) along with the base address and offset so shouldn't the upper bound be 7?

551–560

std::find using GPR16MMRegClass and std::distance should be equivalent to this.

564

Rather than sort at startup, can we just keep the table sorted and assert std::is_sorted()? If we do want to std::sort() then the best place to put it would be in tablegen when we start tablegen-erating the array.

760

Use range-based for loop

941–942

Could you add a comment explaining why instrs[9] is special? What does '9' correspond to?

1004–1053

I think this is just mutating one MachineInst into another similar one. Do we really need to build a new instruction and transfer everything or can we just call MI->setDesc()?

lib/Target/Mips/MicroMipsInstrInfo.td

542–568

If we have explicit operands for the variable-length portion, do we still want the reglist16 operands? I believe the variable length portion covers the same operands as the reglist16's.

test/CodeGen/Mips/micromips-lwm-swm-lwp-swp-sw16.ll

1

(filename) Could you move this into a subdirectory for testing this pass? I'm thinking that the number of tests is going to grow over time and we ought to make it easy to tell which tests cover this pass.

test/CodeGen/Mips/micromips-lwsp-swsp.ll

1

(filename) Could you move this into a subdirectory for testing this pass?

[sdardis]

I believe this work should be implemented in a similar manner to ARM's codesize reduction passes, Thumb2SizeReduction.cpp and ARMLoadStoreOptimizer.cpp.

Their load store optimizer should be modifiable to work for microMIPS. Reusing their logic should avoid the tricky issue of moving loads and stores past other instructions. I'd suggest dropping all the load/store bundling from this patch and focus on the replacing a instruction with a smaller form.

Some of my comments may overlap with Daniel's as we've both looked this but I've tried to delete any ones that overlapped.

lib/Target/Mips/MicroMips32SizeReduction.cpp

1

This file should be called MicroMipsSizeReduction.cpp. This patch is for microMIPS32 but should be sufficiently general that it can be trivially extended to microMIPS64. microMIPS64 support should be a separate patch.

8

Please include a description of this pass, any relevant deficiencies and restrictions. Such as the fact is does not supprt microMIPS64. That comment should be at the bottom of the description as a TODO:.
It should look like:

<Usual LLVM boiler plate.>
//===----------------------------------------------------------------------===//
/// \file
/// This pass is used to reduce the size of instructions where applicable ...
/// ....
/// TODO: implement microMIPS64 support.
//===----------------------------------------------------------------------===//

28

"MicroMips-reduce-size" should be "micromips-reduce-size".

30

'instrs' should be 'instructions', no need to abbreviate it.

32

Here too.

155

microMIPS is the preferred spelling.

400

This predicate is too lax. It has to check at least the same candidates as Filler:terminateSearch in MipsDelaySlotFiller.cpp, and also has to check it is not crossing control flow instructions such as wait, pause and branches or instructions such as sync which act as ordering barriers.

595

Don't use void * and casts. Instead take a pointer/reference to the relevant type.

668–681

All this post loop code should be integrated into the loop body. Rather than 'break'ing out of the loop, in case when you've identified a candidate instruction, I believe you should check the rest of your conditions and if you cannot continue, and immediately return false. If the instruction was an invalid candidate but you can continue the search, update the use set and continue, otherwise you can return ReplaceInstruction(...). Outside the loop body, you should have 'return false'.

946

This can be reduced to a unsigned Opcode = <nested ternary operator>; <newline>MIB = BuildMI(...MipsII->get(Opcode));

958

Rather than packing the operands into a vector before picking the opcode, pick the opcode then iterate over instrs structure and add the operands from that directly.

971

Rename flag to something like 'CopyOperandsForward'.

991

Check for illegal cases first before building an instruction.

test/CodeGen/Mips/micromips-lwm-swm-lwp-swp-sw16.ll

3

Can you add CHECK-LABEL: <function name> here to match the function and in all the others?

[milena.vujosevic.janicic]

The code is simplified: everything about transforming several instructions into one instruction is removed (i.e. lwm/swm and lwp/swp). Therefore, most of the comments are not applicable for this code, but will be taken into account later. Test-cases are also simplified.

[milena.vujosevic.janicic]

lib/Target/Mips/MicroMips32SizeReduction.cpp

389–395

These functions are similar but are not equivalent and cannot be used in this case. isShiftedInt is a template which should be instantiated with a constant shift value, while here the value of shift is a parameter to the function. Also, in this case, low bound and high bound does not necessary correspond to bit width.

[milena.vujosevic.janicic]

Any comments?

[sdardis]

Comments inlined. Most of them are small issues, and an omission from the reduction table for LW16, which I think should go into this patch.

There is a second short form load instruction, lwgp. That should be done as a separate patch rather than including in this revision. It will be a small patch anyway.

Thanks.

lib/Target/Mips/MicroMipsSizeReduction.cpp

13 ↗

(On Diff #52209)

Doesn't this patch do this? :)

14 ↗

(On Diff #52209)

I think we should borrow ARM's load store optimise pass rather than implementing it here.

26–27 ↗

(On Diff #52209)

Please avoid unnecessary includes.

30 ↗

(On Diff #52209)

Unnecessary include.

41 ↗

(On Diff #52209)

By convention, there should be a colon after 'TODO'. Also, spelling of extended.

151–153 ↗

(On Diff #52209)

I'm not seeing this function used anywhere. Since there are predicates for stack relative accesses and short form memory accesses, is it required?

172–183 ↗

(On Diff #52209)

LW16 is missing from this table.

212–213 ↗

(On Diff #52209)

This should be an assert as calling this function with an out of range Op is an error.

MI->getOperand(Op)

in

if (!MI->getOperand(Op).isImm())

will assert that Op < MI->getNumOperands() anyway. Returning false covers a potential bug.

220–221 ↗

(On Diff #52209)

Capitalise Value and Shift as they refer to arguments of this function.

330–334 ↗

(On Diff #52209)

These two cases can be joined together for clarity.

The second case should use isTransient(). This catches cases where MI is a debug value and other pseudo operations like EHLABEL which do not correspond to physical instruction(s).

Put a comment this check saying something like 'Don't reduce bundled instructions or pseudo operations.' so the intention is obvious.

[milena.vujosevic.janicic]

All the comments from the previous revision are taken into account.
lw16/sw16 support was excluded because it is not necessary in this moment.

The code implements size reduction pass for MicroMIPS.
Load and store instructions are examined and transformed, if possible.

lw32 instruction is transformed into 16-bit instruction lwsp
sw32 instruction is transformed into 16-bit instruction swsp

Arithmetic instrcutions are examined and transformed, if possible.

addu32 instruction is transformed into 16-bit instruction addu16
subu32 instruction is transformed into 16-bit instruction subu16

[milena.vujosevic.janicic]

Any comments?

[sdardis]

lib/Target/Mips/MicroMipsSizeReduction.cpp

158 ↗

(On Diff #79229)

Can this be reduced in size to 16 or 8 entries?

263–285 ↗

(On Diff #79229)

These two functions can be folded together as ReduceXWtoXWSP with a comment stating it covers lwsp, swsp.

310–312 ↗

(On Diff #79229)

Style: drop the '{' '} when it is a single line.

315–317 ↗

(On Diff #79229)

Modifed |= Reduced(MI); is clearer.

327 ↗

(On Diff #79229)

This should be dbgs() << ...

test/CodeGen/Mips/micromips-sizereduction/micromips-lwsp-swsp.ll

6 ↗

(On Diff #79229)

Colon after the function name so that it matches properly.

[milena.vujosevic.janicic]

All the comments from the previous revision are taken into account.

[milena.vujosevic.janicic]

Any comments?

[sdardis]

LGTM. Some small nits inlined.

lib/Target/Mips/MicroMipsSizeReduction.cpp

18 ↗

(On Diff #90327)

Spurious empty line.

185 ↗

(On Diff #90327)

This shouldn't have SP_64 in the conditional as this pass doesn't support microMIPS in 64 bit mode yet..

test/CodeGen/Mips/llvm-ir/add.ll

27 ↗

(On Diff #90327)

Add -verify-machineinstrs to the parameters here. We want to know early if the generated machine code is malformed.

test/CodeGen/Mips/llvm-ir/sub.ll

13 ↗

(On Diff #90327)

Add -verify-machineinstrs to the parameters here. We want to know early if the generated machine code is malformed.

test/CodeGen/Mips/micromips-sizereduction/micromips-lwsp-swsp.ll

1 ↗

(On Diff #90327)

Add -verify-machineinstrs to the parameters here. We want to know early if the generated machine code is malformed.

lib/Target/Mips/CMakeLists.txt

add_llvm_target(MipsCodeGen
   MipsSEFrameLowering.cpp
   MipsSEInstrInfo.cpp
   MipsSEISelDAGToDAG.cpp
   MipsSEISelLowering.cpp
   MipsSERegisterInfo.cpp
   MipsSubtarget.cpp
   MipsTargetMachine.cpp
   MipsTargetObjectFile.cpp
+  MicroMips32SizeReduction.cpp
   )
 
 add_subdirectory(InstPrinter)
 add_subdirectory(Disassembler)
 add_subdirectory(TargetInfo)
 add_subdirectory(MCTargetDesc)
 add_subdirectory(AsmParser)
 

lib/Target/Mips/MicroMips32SizeReduction.cpp

+//=== MicroMips32SizeReduction.cpp - MicroMips size reduction pass --------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "Mips.h"
+#include "MipsInstrInfo.h"
+#include "MipsSubtarget.h"
+
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstr.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetMachine.h"
+
+#include <algorithm>
+#include <vector>
+
+using namespace llvm;
+
+#define DEBUG_TYPE "MicroMips-reduce-size"
+
+STATISTIC(NumReduced, "Number of 32-bit instrs reduced to 16-bit ones");
+STATISTIC(NumTwoOne, "Two instructions reduced to one instruction");
+STATISTIC(NumLwmSwm, "Several lw/sw instr. reduced to one lwm/swm instr.");
+
+namespace {
+
+// Order of operands
+enum opNum { NA, opAll, op01, op02, op12, op2, opLwpSwp };
+
+// Reduction type:
+// SeveralInstr - several instructions into lwm/swm
+// TwoInstr - two instructions into one
+// OneInstr - 32-bit instruction into 16-bit instruction
+enum ReduceType { SeveralInstr, TwoInstr, OneInstr };
+
+// Information about immediate field restrictions
+struct ImmField {
+  ImmField() : ImmFieldOpperand(-1), Shift(0), LBound(0), HBound(0) {}
+  ImmField(uint8_t sh, int16_t lb, int16_t hb, int8_t immf)
+      : ImmFieldOpperand(immf), Shift(sh), LBound(lb), HBound(hb) {}
+
+  int8_t ImmFieldOpperand; // Immediate operand, -1 if it does not exist
+  uint8_t Shift;           // Shift value
+  int16_t LBound;          // Low bound of the immediate operand
+  int16_t HBound;          // High bound of the immediate operand
+};
+
+// Information about opperands
+struct OpInfo {
+  OpInfo(enum opNum to, bool snr = false)
+      : TransferOperands(to), SmallerNumRegs(snr) {}
+  OpInfo() : TransferOperands(NA), SmallerNumRegs(false) {}
+
+  enum opNum TransferOperands; // Operands to transfer to the new instruction
+  bool SmallerNumRegs; // In 16 bit instr a smaller num of registers is used
+};
+
+// Information about opcodes
+struct OpCodes {
+  OpCodes(uint16_t wop, uint16_t op2, uint16_t nop)
+      : WideOpc(wop), Opc2(op2), NarrowOpc(nop) {}
+  OpCodes(uint16_t wop, uint16_t nop) : WideOpc(wop), Opc2(0), NarrowOpc(nop) {}
+
+  uint16_t WideOpc;   // Wide opcode
+  uint16_t Opc2;      // Opcode of a second instruction
+  uint16_t NarrowOpc; // Narrow opcode
+};
+
+/// ReduceTable - A static table with information on mapping from wide
+/// opcodes to narrow
+struct ReduceEntry {
+
+  enum ReduceType eRType; // Several instr. to one, Two instr. to one, 32 to 16
+  bool (*ReduceFunction)(void *v); // Pointer to reduce function
+  struct OpCodes Ops;              // All relevant OpCodes
+  struct OpInfo OpInf;             // Characteristics of operands
+  struct ImmField Imm;             // Characteristics of immediate field
+
+  ReduceEntry(enum ReduceType rtype, struct OpCodes op, bool (*f)(void *v),
+              struct OpInfo opinfo, struct ImmField imm)
+      : eRType(rtype), ReduceFunction(f), Ops(op), OpInf(opinfo), Imm(imm) {}
+
+  uint16_t NarrowOpc() const { return Ops.NarrowOpc; }
+  uint16_t WideOpc() const { return Ops.WideOpc; }
+  int16_t LBound() const { return Imm.LBound; }
+  int16_t HBound() const { return Imm.HBound; }
+  uint8_t Shift() const { return Imm.Shift; }
+  int8_t ImmField() const { return Imm.ImmFieldOpperand; }
+  enum opNum TransferOperands() const { return OpInf.TransferOperands; }
+  bool SmallerNumRegs() const { return OpInf.SmallerNumRegs; }
+  enum ReduceType RType() const { return eRType; }
+  uint16_t Opc2() const { return Ops.Opc2; }
+
+  // operator used by std::equal_range
+  bool operator<(const unsigned int r) const { return (WideOpc() < r); }
+
+  // operator used by std::equal_range
+  friend bool operator<(const unsigned int r, const struct ReduceEntry &re) {
+    return (r < re.WideOpc());
+  }
+
+  // operator used by std::sort
+  bool operator<(const struct ReduceEntry &r) const {
+    if (WideOpc() == r.WideOpc())
+      return (RType() < r.RType());
+    return (WideOpc() < r.WideOpc());
+  }
+};
+
+// Function arguments for ReduceFunction
+struct ReduceEntryFA {
+  MachineBasicBlock &MBB;                       // Basic block
+  const MachineBasicBlock::instr_iterator &MII; // Starting iterator
+  const MachineBasicBlock::instr_iterator &E;   // End iterator
+  MachineBasicBlock::instr_iterator
+      &NNextMII;            // Iterator to next instruction, if
+  const ReduceEntry &Entry; // Entry field
+
+  ReduceEntryFA(MachineBasicBlock &argMBB,
+                const MachineBasicBlock::instr_iterator &argMII,
+                const MachineBasicBlock::instr_iterator &argE,
+                MachineBasicBlock::instr_iterator &argNNextMII,
+                const ReduceEntry &argEntry)
+      : MBB(argMBB), MII(argMII), E(argE), NNextMII(argNNextMII),
+        Entry(argEntry) {}
+};
+
+struct lwmswm {
+  lwmswm() : found(false), MI(nullptr) {}
+  bool found;
+  MachineInstr *MI;
+};
+
+class MicroMips32SizeReduce : public MachineFunctionPass {
+public:
+  static char ID;
+  MicroMips32SizeReduce();
+
+  static const MipsInstrInfo *MipsII;
+  const MipsSubtarget *Subtarget;
+
+  bool runOnMachineFunction(MachineFunction &MF) override;
+
+  const char *getPassName() const override {
+    return "MicroMips32 instruction size reduction pass";
+  }
+
+private:
+  /// ReduceMBB - Reduces width of instructions in the specified basic block.
+  bool ReduceMBB(MachineBasicBlock &MBB);
+
+  /// ReduceMI - Attempts to reduce MI, returns true on success.
+  bool ReduceMI(MachineBasicBlock &MBB,
+                const MachineBasicBlock::instr_iterator &MII,
+                const MachineBasicBlock::instr_iterator &E,
+                MachineBasicBlock::instr_iterator &NextMII);
+
+  // Attempts to reduce several instruction into LWM/SWM instruction,
+  // returns true on success
+  static bool ReduceMIToLWMSWM(void *v);
+
+  // Attempts to reduce into LWP/SWP instruction, returns true on success
+  static bool ReduceMIToLwpSwp(void *v);
+
+  // Attempts to reduce SW instruction, returns true on success
+  static bool ReduceSWtoSWSP(void *v);
+  static bool ReduceSWtoSW16(void *v);
+
+  // Attempts to reduce LW instruction, returns true on success
+  static bool ReduceLWtoLWSP(void *v);
+
+  // Attempts to reduce all other Load/Store instructions,
+  // returns true on success
+  static bool ReduceLoadStore(void *v);
+
+  // Adds an instruction into machine block, instead of MI
+  // deletes MI
+  static bool ReplaceInstruction(MachineBasicBlock &MBB, MachineInstr *MI,
+                                 const ReduceEntry &Entry);
+
+  // Adds an instruction into machine block, instead of fMI and sMI,
+  // after sMI, deletes fMI and sMI
+  static bool ReplaceInstruction(MachineBasicBlock &MBB, MachineInstr *fMI,
+                                 MachineInstr *sMI, bool flag,
+                                 const ReduceEntry &Entry);
+
+  // Adds LWM/SWM instruction into machine block, instead of MI and num
+  // following instructions
+  static bool AddInstructionLWMSWM(MachineBasicBlock &MBB, MachineInstr *MI,
+                                   int64_t offset,
+                                   const SmallVector<struct lwmswm, 10> &instrs,
+                                   bool lwm);
+
+  // Deletes instructions that are reduced to LWM/SWM
+  static void DeleteInstructions(MachineBasicBlock &MBB,
+                                 const SmallVector<struct lwmswm, 10> &instrs);
+
+  // Table with transformation rules for each instruction
+  static std::vector<ReduceEntry> ReduceTable;
+};
+
+char MicroMips32SizeReduce::ID = 0;
+const MipsInstrInfo *MicroMips32SizeReduce::MipsII;
+
+std::vector<ReduceEntry> MicroMips32SizeReduce::ReduceTable = {
+
+    // ReduceType, OpCodes, ReduceFunction,
+    // OpInfo(TransferOperands, SmallerNumRegs=false),
+    // ImmField(Shift, LBound, HBound, ImmFieldPosition)
+
+    {OneInstr, OpCodes(Mips::LWM32_MM, Mips::LWM16_MM), ReduceLoadStore,
+     OpInfo(opAll), ImmField(2, 0, 16, 2)},
+    {OneInstr, OpCodes(Mips::LWM_MM, Mips::LWM16_MM), ReduceLoadStore,
+     OpInfo(opAll), ImmField(2, 0, 16, 2)},
+    {OneInstr, OpCodes(Mips::LW_MM, Mips::LWSP_MM), ReduceLWtoLWSP,
+     OpInfo(opAll), ImmField(2, 0, 32, 2)},
+    {OneInstr, OpCodes(Mips::LW, Mips::LWSP_MM), ReduceLWtoLWSP, OpInfo(opAll),
+     ImmField(2, 0, 32, 2)},
+    {OneInstr, OpCodes(Mips::SW_MM, Mips::SW16_MM), ReduceSWtoSW16,
+     OpInfo(opAll), ImmField(2, 0, 16, 2)},
+    {OneInstr, OpCodes(Mips::SW, Mips::SW16_MM), ReduceSWtoSW16, OpInfo(opAll),
+     ImmField(2, 0, 16, 2)},
+    {OneInstr, OpCodes(Mips::SW_MM, Mips::SWSP_MM), ReduceSWtoSWSP,
+     OpInfo(opAll), ImmField(2, 0, 32, 2)},
+    {OneInstr, OpCodes(Mips::SW, Mips::SWSP_MM), ReduceSWtoSWSP, OpInfo(opAll),
+     ImmField(2, 0, 32, 2)},
+    {OneInstr, OpCodes(Mips::SWM32_MM, Mips::SWM16_MM), ReduceLoadStore,
+     OpInfo(opAll), ImmField(2, 0, 16, 2)},
+    {OneInstr, OpCodes(Mips::SWM_MM, Mips::SWM16_MM), ReduceLoadStore,
+     OpInfo(opAll), ImmField(2, 0, 16, 2)},
+
+    // Transfer two instructions into one
+    {TwoInstr, OpCodes(Mips::LW, Mips::LW, Mips::LWP_MM), ReduceMIToLwpSwp,
+     OpInfo(opLwpSwp), ImmField(0, -2048, 2048, 2)},
+    {TwoInstr, OpCodes(Mips::LW, Mips::LW_MM, Mips::LWP_MM), ReduceMIToLwpSwp,
+     OpInfo(opLwpSwp), ImmField(0, -2048, 2048, 2)},
+    {TwoInstr, OpCodes(Mips::LW_MM, Mips::LW, Mips::LWP_MM), ReduceMIToLwpSwp,
+     OpInfo(opLwpSwp), ImmField(0, -2048, 2048, 2)},
+    {TwoInstr, OpCodes(Mips::LW_MM, Mips::LW_MM, Mips::LWP_MM),
+     ReduceMIToLwpSwp, OpInfo(opLwpSwp), ImmField(0, -2048, 2048, 2)},
+
+    {TwoInstr, OpCodes(Mips::SW, Mips::SW, Mips::SWP_MM), ReduceMIToLwpSwp,
+     OpInfo(opLwpSwp), ImmField(0, -2048, 2048, 2)},
+    {TwoInstr, OpCodes(Mips::SW, Mips::SW_MM, Mips::SWP_MM), ReduceMIToLwpSwp,
+     OpInfo(opLwpSwp), ImmField(0, -2048, 2048, 2)},
+    {TwoInstr, OpCodes(Mips::SW_MM, Mips::SW_MM, Mips::SWP_MM),
+     ReduceMIToLwpSwp, OpInfo(opLwpSwp), ImmField(0, -2048, 2048, 2)},
+    {TwoInstr, OpCodes(Mips::SW_MM, Mips::SW, Mips::SWP_MM), ReduceMIToLwpSwp,
+     OpInfo(opLwpSwp), ImmField(0, -2048, 2048, 2)},
+
+    // Transfer several instructions into one
+    {SeveralInstr, OpCodes(Mips::LW, Mips::LWM_MM), ReduceMIToLWMSWM,
+     OpInfo(NA), ImmField(0, -2048, 2048, 2)},
+    {SeveralInstr, OpCodes(Mips::LW_MM, Mips::LWM_MM), ReduceMIToLWMSWM,
+     OpInfo(NA), ImmField(0, -2048, 2048, 2)},
+    {SeveralInstr, OpCodes(Mips::SW, Mips::SWM_MM), ReduceMIToLWMSWM,
+     OpInfo(NA), ImmField(0, -2048, 2048, 2)},
+    {SeveralInstr, OpCodes(Mips::SW_MM, Mips::SWM_MM), ReduceMIToLWMSWM,
+     OpInfo(NA), ImmField(0, -2048, 2048, 2)},
+};
+}
+
+// Returns true if the register Reg is $16, $17, or $2-$7.
+static bool isMMThreeBitGPRegister(unsigned Reg) {
+  using namespace Mips;
+  switch (Reg) {
+  case S0:
+  case S1:
+  case V0:
+  case V1:
+  case A0:
+  case A1:
+  case A2:
+  case A3:
+    return true;
+  default:
+    return false;
+  }
+}
+
+// Returns true if the register Reg is $0, $17, or $2-$7.
+static bool isMMSourceRegister(unsigned Reg) {
+  using namespace Mips;
+  switch (Reg) {
+  case ZERO:
+  case S1:
+  case V0:
+  case V1:
+  case A0:
+  case A1:
+  case A2:
+  case A3:
+    return true;
+  default:
+    return false;
+  }
+}
+// Returns true if the machine operand MO is register SP
+static bool IsSP(const MachineOperand &MO) {
+  if (MO.isReg() && (MO.getReg() == Mips::SP))
+    return true;
+  return false;
+}
+
+// Returns true if the machine operand MO is register $16, $17, or $2-$7.
+static bool isMMThreeBitGPRegister(const MachineOperand &MO) {
+  if (MO.isReg() && isMMThreeBitGPRegister(MO.getReg()))
+    return true;
+  return false;
+}
+
+// Returns true if the machine operand MO is register $0, $17, or $2-$7.
+static bool isMMSourceRegister(const MachineOperand &MO) {
+  if (MO.isReg() && isMMSourceRegister(MO.getReg()))
+    return true;
+  return false;
+}
+
+// Returns true if the operand op is an immediate value
+// and writes the immediate value into variable imm
+static bool GetImm(MachineInstr *MI, unsigned op, int64_t &imm) {
+
+  if (op >= MI->getNumOperands())
+    return false;
+  if (!MI->getOperand(op).isImm())
+    return false;
+  imm = MI->getOperand(op).getImm();
+  return true;
+}
+
+// Returns true if the operand op is a register
+// and writes the register into variable reg
+static bool GetReg(MachineInstr *MI, unsigned op, unsigned &reg) {
+  if (op >= MI->getNumOperands())
+    return false;
+  if (!MI->getOperand(op).isReg())
+    return false;
+  reg = MI->getOperand(op).getReg();
+  return true;
+}
+
+// Returns true if in machine instruction operands Op1 and Op2 are equal
+// registers
+static bool EqualRegsInInstr(MachineInstr *MI, uint8_t Op1, uint8_t Op2) {
+  unsigned reg1, reg2;
+  if (!GetReg(MI, Op1, reg1))
+    return false;
+  if (!GetReg(MI, Op2, reg2))
+    return false;
+  if (reg1 != reg2)
+    return false;
+  return true;
+}
+
+// Returns true if the registers Reg1 and Reg2 are consecutive
+static bool ConsecutiveRegisters(unsigned Reg1, unsigned Reg2) {
+  static SmallVector<unsigned, 31> registers = {
+      Mips::AT, Mips::V0, Mips::V1, Mips::A0, Mips::A1, Mips::A2, Mips::A3,
+      Mips::T0, Mips::T1, Mips::T2, Mips::T3, Mips::T4, Mips::T5, Mips::T6,
+      Mips::T7, Mips::S0, Mips::S1, Mips::S2, Mips::S3, Mips::S4, Mips::S5,
+      Mips::S6, Mips::S7, Mips::T8, Mips::T9, Mips::K0, Mips::K1, Mips::GP,
+      Mips::SP, Mips::FP, Mips::RA};
+
+  for (uint8_t i = 0; i < registers.size() - 1; i++) {
+    if (registers[i] == Reg1) {
+      if (registers[i + 1] == Reg2)
+        return true;
+      else
+        return false;
+    }
+  }
+  return false;
+}
+
+// Returns true if the variable value has the number of least-significant zero
+// bits equal to shift
+// and if the shifted value is between the bounds
+static bool InRange(int64_t value, unsigned short shift, int lbound,
+                    int hbound) {
+  int64_t value2 = value >> shift;
+  if ((value2 << shift) == value && (value2 >= lbound) && (value2 < hbound))
+    return true;
+  return false;
+}
+
+// Returns true if the instruction is not appropriate to be between
+// two instructions that should be reduced to one
+static bool NotAppropriateInstruction(MachineInstr *MI) {
+
+  if (MI->isCall())
+    return true;
+
+  return false;
+}
+
+// Returns true for lw instruction
+static bool isLw(MachineInstr *MI) {
+  return (MI->getOpcode() == Mips::LW || MI->getOpcode() == Mips::LW_MM);
+}
+
+// Returns true for sw instruction
+static bool isSw(MachineInstr *MI) {
+  return (MI->getOpcode() == Mips::SW || MI->getOpcode() == Mips::SW_MM);
+}
+
+// Returns true if immediate opperand is in range
+static bool ImmInRange(MachineInstr *MI, const ReduceEntry &Entry) {
+
+  int64_t offset;
+
+  if (!GetImm(MI, Entry.ImmField(), offset))
+    return false;
+
+  if (!InRange(offset, Entry.Shift(), Entry.LBound(), Entry.HBound()))
+    return false;
+
+  return true;
+}
+
+// Returns true if MI can be reduced to lwp/swp instruciton
+static bool CheckLwpSwpInstr(MachineInstr *MI, bool lwp,
+                             const ReduceEntry &Entry) {
+
+  if (!((lwp && isLw(MI)) || (!lwp && isSw(MI))))
+    return false;
+
+  unsigned reg;
+  if (!GetReg(MI, 0, reg))
+    return false;
+  if (reg == Mips::RA)
+    return false;
+
+  if (!GetReg(MI, 1, reg))
+    return false;
+
+  if (!ImmInRange(MI, Entry))
+    return false;
+
+  if (lwp && (EqualRegsInInstr(MI, 0, 1)))
+    return false;
+
+  return true;
+}
+
+// Returns true if registers and offsets are consecutive
+static bool consecutiveInstr(MachineInstr *fMI, MachineInstr *sMI) {
+
+  int64_t offset, noffset;
+  if (!GetImm(fMI, 2, offset))
+    return false;
+  if (!GetImm(sMI, 2, noffset))
+    return false;
+
+  unsigned reg1, reg2;
+  if (!GetReg(fMI, 0, reg1))
+    return false;
+  if (!GetReg(sMI, 0, reg2))
+    return false;
+
+  return ((offset == (noffset - 4)) && (ConsecutiveRegisters(reg1, reg2)));
+}
+
+// Returns true if the instruction MI uses at least one
+// of the registers reg1 and reg2
+static bool InstrUsesRegs(MachineInstr *MI, unsigned reg1, unsigned reg2) {
+  uint8_t numOp = MI->getNumOperands();
+  unsigned reg;
+  // Iterates through the registers that this instruction uses
+  for (uint8_t i = 0; i < numOp; ++i) {
+    if (GetReg(MI, i, reg)) {
+      if ((reg == reg1) || (reg == reg2))
+        return true;
+    }
+  }
+  return false;
+}
+
+// Adds all the registers that MI uses into the set RegistersUsed
+static void AddRegisters(MachineInstr *MI,
+                         SmallSet<unsigned, 32> &RegistersUsed) {
+  uint8_t numOp = MI->getNumOperands();
+  unsigned reg;
+  for (uint8_t i = 0; i < numOp; ++i) {
+    if (GetReg(MI, i, reg))
+      RegistersUsed.insert(reg);
+  }
+}
+
+// Returns true if the instruction MI can be part of lwm/swm instruction
+// and fills offset, base and register
+static bool getLwmSwmOperands(MachineInstr *MI, int64_t &offset, unsigned &reg0,
+                              unsigned &base, bool lwm,
+                              const ReduceEntry &Entry) {
+
+  if (!(lwm && isLw(MI)) && !(!lwm && isSw(MI)))
+    return false;
+
+  if (!GetImm(MI, 2, offset))
+    return false;
+
+  if (!ImmInRange(MI, Entry))
+    return false;
+
+  if (!GetReg(MI, 0, reg0))
+    return false;
+
+  if (!GetReg(MI, 1, base))
+    return false;
+
+  return true;
+}
+
+// Checks 16-bit lwm/swm instruction can be generated
+static bool LwmSwm16Bit(MachineInstr *MI,
+                        const std::vector<MachineOperand> &operands) {
+
+  unsigned endReg = Mips::RA;
+
+  if (!IsSP(MI->getOperand(1)))
+    return false;
+
+  int64_t offset;
+  if (!GetImm(MI, 2, offset))
+    return false;
+
+  if (!InRange(offset, 2, 0, 16))
+    return false;
+
+  unsigned num = operands.size();
+  if (num < 2 || num > 5)
+    return false;
+
+  if (operands[num - 1].getReg() != endReg)
+    return false;
+
+  return true;
+}
+
+// Finds the postion of the register reg in the array, starting from the
+// start position, or returns false
+static bool findLWMSWMPositon(unsigned reg, unsigned &Position) {
+  SmallVector<unsigned, 10> regs = {Mips::S0, Mips::S1, Mips::S2, Mips::S3,
+                                    Mips::S4, Mips::S5, Mips::S6, Mips::S7,
+                                    Mips::FP, Mips::RA};
+  for (uint8_t k = 0; k < regs.size(); k++)
+    if (reg == regs[k]) {
+      Position = k;
+      return true;
+    }
+
+  return false;
+}
+
+MicroMips32SizeReduce::MicroMips32SizeReduce() : MachineFunctionPass(ID) {
+  std::sort(ReduceTable.begin(), ReduceTable.end());
+}
+
+bool MicroMips32SizeReduce::ReduceMI(
+    MachineBasicBlock &MBB, const MachineBasicBlock::instr_iterator &MII,
+    const MachineBasicBlock::instr_iterator &E,
+    MachineBasicBlock::instr_iterator &NNextMII) {
+
+  MachineInstr *MI = &*MII;
+  unsigned Opcode = MI->getOpcode();
+
+  // Search the table.
+  std::vector<ReduceEntry>::const_iterator Start = std::begin(ReduceTable);
+  std::vector<ReduceEntry>::const_iterator End = std::end(ReduceTable);
+
+  std::pair<std::vector<ReduceEntry>::const_iterator,
+            std::vector<ReduceEntry>::const_iterator> Range =
+      std::equal_range(Start, End, Opcode);
+
+  if (Range.first == Range.second)
+    return false;
+
+  for (std::vector<ReduceEntry>::const_iterator Entry = Range.first;
+       Entry != Range.second; ++Entry) {
+    struct ReduceEntryFA s(MBB, MII, E, NNextMII, *Entry);
+    if (((*Entry).ReduceFunction)(&s))
+      return true;
+  }
+  return false;
+}
+
+bool MicroMips32SizeReduce::ReduceMIToLwpSwp(void *v) {
+
+  ReduceEntryFA fa = *(ReduceEntryFA *)v;
+  MachineBasicBlock &MBB = fa.MBB;
+  const MachineBasicBlock::instr_iterator &MII = fa.MII;
+  const MachineBasicBlock::instr_iterator &E = fa.E;
+  MachineBasicBlock::instr_iterator &NNextMII = fa.NNextMII;
+  const ReduceEntry &Entry = fa.Entry;
+
+  MachineBasicBlock::instr_iterator NextMII; //, StartMII = MII;
+
+  // First instruction
+  MachineInstr *fMI = &*MII;
+  // Second instruction
+  MachineInstr *sMI = nullptr;
+
+  bool lwp = isLw(fMI);   // lwp==true -> transform to lwp instruction
+  if (!lwp && !isSw(fMI)) // lwp==false && isSw -> transform to swp instruction
+    return false;
+
+  if (!CheckLwpSwpInstr(fMI, lwp, Entry))
+    return false;
+
+  bool found = false;
+  bool consecutiveForward = false;
+  bool consecutiveBackward = false;
+
+  unsigned reg1, reg2;
+  if (!GetReg(fMI, 0, reg1))
+    return false;
+  if (!GetReg(fMI, 1, reg2)) // equal to GetReg(sMI,1,reg2)
+    return false;
+
+  SmallSet<unsigned, 32> RegistersUsed;
+
+  // Iterate through block to find second instruction
+  MachineBasicBlock::instr_iterator iMII = std::next(MII);
+  for (; iMII != E; iMII = NextMII) {
+
+    NextMII = std::next(iMII);
+    if (NextMII == E)
+      break;
+
+    sMI = &*iMII;
+
+    if (CheckLwpSwpInstr(sMI, lwp, Entry)) {
+      unsigned reg;
+      if (GetReg(sMI, 1, reg) && (reg2 == reg)) {
+        consecutiveForward = consecutiveInstr(fMI, sMI);
+        consecutiveBackward = consecutiveInstr(sMI, fMI);
+        found = consecutiveForward || consecutiveBackward;
+      }
+    }
+
+    if (found)
+      break;
+
+    // if the instruction is not appropriate, the
+    // reduction is not possible
+    if (lwp && isSw(sMI))
+      return false;
+    if (!lwp && isLw(sMI))
+      return false;
+    if (NotAppropriateInstruction(sMI))
+      return false;
+    if (InstrUsesRegs(sMI, reg1, reg2))
+      return false;
+
+    // memorize registers used by intermediate instructions
+    AddRegisters(sMI, RegistersUsed);
+  }
+
+  if (!found)
+    return false;
+
+  unsigned reg;
+  if (!GetReg(sMI, 0, reg))
+    return false;
+
+  // If some intermediate instruction uses reg,
+  // then reduction is not possible
+  if (RegistersUsed.count(reg))
+    return false;
+
+  NNextMII = std::next(iMII);
+  return ReplaceInstruction(MBB, fMI, sMI, consecutiveForward, Entry);
+}
+
+bool MicroMips32SizeReduce::ReduceLWtoLWSP(void *v) {
+  ReduceEntryFA fa = *(ReduceEntryFA *)v;
+  MachineBasicBlock &MBB = fa.MBB;
+  MachineInstr *MI = &*(fa.MII);
+  const ReduceEntry &Entry = fa.Entry;
+
+  if (!ImmInRange(MI, Entry))
+    return false;
+
+  if (!IsSP(MI->getOperand(1)))
+    return false;
+
+  return ReplaceInstruction(MBB, MI, Entry);
+}
+
+bool MicroMips32SizeReduce::ReduceSWtoSW16(void *v) {
+  ReduceEntryFA fa = *(ReduceEntryFA *)v;
+  MachineBasicBlock &MBB = fa.MBB;
+  MachineInstr *MI = &*(fa.MII);
+  const ReduceEntry &Entry = fa.Entry;
+
+  if (!ImmInRange(MI, Entry))
+    return false;
+
+  if (!(isMMSourceRegister(MI->getOperand(0)) &&
+        isMMThreeBitGPRegister(MI->getOperand(1))))
+    return false;
+
+  return ReplaceInstruction(MBB, MI, Entry);
+}
+
+bool MicroMips32SizeReduce::ReduceSWtoSWSP(void *v) {
+  ReduceEntryFA fa = *(ReduceEntryFA *)v;
+  MachineBasicBlock &MBB = fa.MBB;
+  MachineInstr *MI = &*(fa.MII);
+  const ReduceEntry &Entry = fa.Entry;
+
+  if (!ImmInRange(MI, Entry))
+    return false;
+
+  if (!IsSP(MI->getOperand(1)))
+    return false;
+
+  return ReplaceInstruction(MBB, MI, Entry);
+}
+
+bool MicroMips32SizeReduce::ReduceLoadStore(void *v) {
+
+  ReduceEntryFA fa = *(ReduceEntryFA *)v;
+  MachineBasicBlock &MBB = fa.MBB;
+  MachineInstr *MI = &*(fa.MII);
+  const ReduceEntry &Entry = fa.Entry;
+
+  if (!ImmInRange(MI, Entry))
+    return false;
+
+  // Check LWM/SWM instruction
+  if ((Entry.WideOpc() == Mips::SWM32_MM || Entry.WideOpc() == Mips::SWM_MM) ||
+      (Entry.WideOpc() == Mips::LWM32_MM || Entry.WideOpc() == Mips::LWM_MM)) {
+
+    if (!IsSP(MI->getOperand(1)))
+      return false;
+
+    int64_t reglist;
+    if (!GetImm(MI, 0, reglist))
+      return false;
+
+    if (!InRange(reglist, 0, 17, 22))
+      return false;
+  }
+
+  return ReplaceInstruction(MBB, MI, Entry);
+}
+
+void MicroMips32SizeReduce::DeleteInstructions(
+    MachineBasicBlock &MBB, const SmallVector<struct lwmswm, 10> &instrs) {
+
+  for (unsigned i = 0; i < instrs.size(); i++)
+    if (instrs[i].found)
+      MBB.erase(instrs[i].MI);
+}
+
+bool MicroMips32SizeReduce::ReduceMIToLWMSWM(void *v) {
+  ReduceEntryFA fa = *(ReduceEntryFA *)v;
+  MachineBasicBlock &MBB = fa.MBB;
+  const MachineBasicBlock::instr_iterator &MII = fa.MII;
+  const MachineBasicBlock::instr_iterator &E = fa.E;
+  MachineBasicBlock::instr_iterator &NextMII = fa.NNextMII;
+  const ReduceEntry &Entry = fa.Entry;
+
+  MachineBasicBlock::instr_iterator nMII;
+  MachineBasicBlock::instr_iterator iMII = MII;
+  MachineBasicBlock::instr_iterator startingMII = MII;
+  MachineInstr *MI = &*MII;
+  MachineInstr *startingMI = &*MII;
+
+  bool lwm = false;
+  if (isLw(MI))
+    lwm = true;
+  else if (!isSw(MI))
+    return false;
+
+  unsigned startingBase;
+  int64_t offset;
+  unsigned reg0, base;
+  if (!getLwmSwmOperands(MI, offset, reg0, startingBase, lwm, Entry))
+    return false;
+
+  unsigned endReg = Mips::RA;
+
+  SmallVector<struct lwmswm, 10> instrs(10);
+
+  unsigned position;
+  if (findLWMSWMPositon(reg0, position)) {
+    instrs[position].found = true;
+    instrs[position].MI = MI;
+  } else
+    return false;
+
+  int64_t startingOffset = 0;
+  bool b = false;
+  if (reg0 != endReg) {
+    startingOffset = offset - position * 4;
+    b = true;
+  }
+
+  if (lwm) {
+    if (startingBase == reg0)
+      return false;
+  }
+
+  for (iMII = std::next(startingMII); iMII != E; iMII = nMII) {
+    MachineInstr *MI = &*iMII;
+    nMII = std::next(iMII);
+
+    if (!getLwmSwmOperands(MI, offset, reg0, base, lwm, Entry))
+      break;
+
+    if (startingBase != base)
+      break;
+
+    if (!findLWMSWMPositon(reg0, position))
+      break;
+
+    if (lwm) {
+      if (base == reg0)
+        return false;
+    }
+
+    if (instrs[position].found)
+      break;
+
+    if (!b) {
+      if (reg0 != endReg) {
+        startingOffset = offset - position * 4;
+        b = true;
+      }
+    }
+
+    if ((reg0 != endReg) && (startingOffset != offset - position * 4))
+      break;
+
+    instrs[position].found = true;
+    instrs[position].MI = MI;
+  }
+
+  unsigned i;
+  if (!(instrs[0].found))
+    return false;
+
+  for (i = 1; i < instrs.size(); i++)
+    if (!(instrs[i].found))
+      break;
+
+  unsigned num = i;
+
+  for (; i < (instrs.size() - 1); i++)
+    if (instrs[i].found)
+      instrs[i].found = false;
+
+  if (instrs[9].found) {
+    getLwmSwmOperands(instrs[9].MI, offset, reg0, base, lwm, Entry);
+
+    if (num != instrs.size()) {
+      if (startingOffset != offset - num * 4)
+        instrs[9].found = false;
+      else
+        num++;
+    } else if (startingOffset != (offset - (num - 1) * 4))
+      instrs[9].found = false;
+  }
+
+  if (num == 1)
+    return false;
+
+  AddInstructionLWMSWM(MBB, startingMI, startingOffset, instrs, lwm);
+  NextMII = nMII;
+  DeleteInstructions(MBB, instrs);
+
+  return true;
+}
+
+bool MicroMips32SizeReduce::ReduceMBB(MachineBasicBlock &MBB) {
+  bool Modified = false;
+  MachineBasicBlock::instr_iterator MII = MBB.instr_begin(),
+                                    E = MBB.instr_end();
+  MachineBasicBlock::instr_iterator NextMII;
+
+  bool ModifiedInstructions = false;
+
+  do {
+
+    ModifiedInstructions = false;
+    MII = MBB.instr_begin();
+    E = MBB.instr_end();
+
+    // Iterate through the instructions in the basic block
+    for (; MII != E; MII = NextMII) {
+      NextMII = std::next(MII);
+      MachineInstr *MI = &*MII;
+
+      if (MI->isBundle()) {
+        continue;
+      }
+      if (MI->isDebugValue())
+        continue;
+
+      // Try to reduce several instructions into one instruction
+      // Try to reduce two instructions into one instruction
+      // Try to reduce 32-bit instruction into 16-bit instruction
+      if (ReduceMI(MBB, MII, E, NextMII)) {
+        Modified = true;
+        ModifiedInstructions = true;
+      }
+    }
+  } while (ModifiedInstructions);
+
+  return Modified;
+}
+
+bool MicroMips32SizeReduce::AddInstructionLWMSWM(
+    MachineBasicBlock &MBB, MachineInstr *MI, int64_t offset,
+    const SmallVector<struct lwmswm, 10> &instrs, bool lwm) {
+
+  MachineInstr *iMI = MI;
+  DebugLoc dl = iMI->getDebugLoc();
+  MachineInstrBuilder MIB;
+
+  std::vector<MachineOperand> operands;
+
+  unsigned i;
+  for (i = 0; i < instrs.size(); i++) {
+    if (!(instrs[i].found))
+      break;
+    iMI = instrs[i].MI;
+    operands.push_back(iMI->getOperand(0));
+  }
+  if (i < instrs.size() && instrs[9].found)
+    operands.push_back(instrs[9].MI->getOperand(0));
+
+  // Check if 16-bit lwm/swm instruction can be generated
+  bool bit16 = LwmSwm16Bit(instrs[0].MI, operands);
+
+  if (bit16) {
+    if (lwm)
+      MIB = BuildMI(MBB, MI, dl, MipsII->get(Mips::LWM16_MM));
+    else
+      MIB = BuildMI(MBB, MI, dl, MipsII->get(Mips::SWM16_MM));
+  } else { // 32 bit lwm/swm
+    if (lwm)
+      MIB = BuildMI(MBB, MI, dl, MipsII->get(Mips::LWM32_MM));
+    else
+      MIB = BuildMI(MBB, MI, dl, MipsII->get(Mips::SWM32_MM));
+  }
+
+  for (unsigned int i = 0; i < operands.size(); i++)
+    MIB.addOperand(operands[i]);
+
+  MIB.addOperand(MI->getOperand(1));
+  MIB.addImm(offset);
+
+  DEBUG(errs() << "Converted to: " << *MIB);
+  ++NumLwmSwm;
+  return true;
+}
+
+bool MicroMips32SizeReduce::ReplaceInstruction(MachineBasicBlock &MBB,
+                                               MachineInstr *fMI,
+                                               MachineInstr *sMI, bool flag,
+                                               const ReduceEntry &Entry) {
+
+  const MCInstrDesc &NewMCID = MipsII->get(Entry.NarrowOpc());
+  DebugLoc dl = sMI->getDebugLoc();
+  MachineInstrBuilder MIB = BuildMI(MBB, sMI, dl, NewMCID);
+
+  if (Entry.TransferOperands() == opLwpSwp) {
+    if (flag) {
+      MIB.addOperand(fMI->getOperand(0));
+      MIB.addOperand(sMI->getOperand(0));
+      MIB.addOperand(fMI->getOperand(1));
+      MIB.addOperand(fMI->getOperand(2));
+    } else { // consecutive backward
+      MIB.addOperand(sMI->getOperand(0));
+      MIB.addOperand(fMI->getOperand(0));
+      MIB.addOperand(sMI->getOperand(1));
+      MIB.addOperand(sMI->getOperand(2));
+    }
+  } else
+    return false;
+
+  DEBUG(errs() << "\nConverted " << *fMI << "and " << *sMI
+               << "       to 16-bit: " << *MIB);
+
+  MBB.erase_instr(fMI);
+  MBB.erase_instr(sMI);
+  ++NumTwoOne;
+
+  return true;
+}
+
+bool MicroMips32SizeReduce::ReplaceInstruction(MachineBasicBlock &MBB,
+                                               MachineInstr *MI,
+                                               const ReduceEntry &Entry) {
+
+  // Add the 16-bit instruction.
+  const MCInstrDesc &NewMCID = MipsII->get(Entry.NarrowOpc());
+  DebugLoc dl = MI->getDebugLoc();
+  MachineInstrBuilder MIB = BuildMI(MBB, MI, dl, NewMCID);
+
+  enum opNum opNums = Entry.TransferOperands();
+  if (opNums == op02) {
+    MIB.addOperand(MI->getOperand(0));
+    MIB.addOperand(MI->getOperand(2));
+  } else if (opNums == op01) {
+    MIB.addOperand(MI->getOperand(0));
+    MIB.addOperand(MI->getOperand(1));
+  } else if (opNums == op12) {
+    MIB.addOperand(MI->getOperand(1));
+    MIB.addOperand(MI->getOperand(2));
+  } else if (opNums == op2) {
+    MIB.addOperand(MI->getOperand(2));
+  } else if ((opNums == opAll) && Entry.SmallerNumRegs()) {
+    if (EqualRegsInInstr(MI, 0, 1)) {
+      MIB.addOperand(MI->getOperand(0));
+      MIB.addOperand(MI->getOperand(2));
+      MIB.addOperand(MI->getOperand(1));
+    } else {
+      MIB.addOperand(MI->getOperand(0));
+      MIB.addOperand(MI->getOperand(1));
+      MIB.addOperand(MI->getOperand(2));
+    }
+  } else if (opNums == opAll)
+    for (uint8_t i = 0, e = MI->getNumOperands(); i != e; ++i) {
+      const MachineOperand &MO = MI->getOperand(i);
+      MIB.addOperand(MO);
+    }
+  else
+    return false;
+
+  // Transfer memoperands.
+  MIB->setMemRefs(MI->memoperands_begin(), MI->memoperands_end());
+
+  // Transfer MI flags.
+  MIB.setMIFlags(MI->getFlags());
+
+  DEBUG(errs() << "Converted 32-bit: " << *MI << "       to 16-bit: " << *MIB);
+  MBB.erase_instr(MI);
+  ++NumReduced;
+  return true;
+}
+
+bool MicroMips32SizeReduce::runOnMachineFunction(MachineFunction &MF) {
+
+  Subtarget = &static_cast<const MipsSubtarget &>(MF.getSubtarget());
+
+  if (!Subtarget->inMicroMipsMode())
+    return false;
+
+  MipsII = static_cast<const MipsInstrInfo *>(Subtarget->getInstrInfo());
+
+  bool Modified = false;
+  MachineFunction::iterator I = MF.begin(), E = MF.end();
+
+  for (; I != E; ++I)
+    Modified |= ReduceMBB(*I);
+  return Modified;
+}
+
+/// createMicroMipsSizeReductionPass - Returns an instance of the MicroMips size
+/// reduction pass.
+FunctionPass *llvm::createMicroMips32SizeReductionPass() {
+  return new MicroMips32SizeReduce();
+}

lib/Target/Mips/MicroMipsInstrInfo.td

def reglist16 : Operand<i32> {
   let EncoderMethod = "getRegisterListOpValue16";
   let DecoderMethod = "DecodeRegListOperand16";
   let PrintMethod = "printRegisterList";
   let ParserMatchClass = RegList16AsmOperand;
 }
 
 class StoreMultMM<string opstr,
             InstrItinClass Itin = NoItinerary, ComplexPattern Addr = addr> :
-  InstSE<(outs), (ins reglist:$rt, mem_mm_12:$addr),
+  InstSE<(outs), (ins reglist:$rt, variable_ops, mem_mm_12:$addr),
          !strconcat(opstr, "\t$rt, $addr"), [], Itin, FrmI, opstr> {
   let DecoderMethod = "DecodeMemMMImm12";
   let mayStore = 1;
 }
 
 class LoadMultMM<string opstr,
             InstrItinClass Itin = NoItinerary, ComplexPattern Addr = addr> :
-  InstSE<(outs reglist:$rt), (ins mem_mm_12:$addr),
+  InstSE<(outs reglist:$rt, variable_ops), (ins mem_mm_12:$addr),
           !strconcat(opstr, "\t$rt, $addr"), [], Itin, FrmI, opstr> {
   let DecoderMethod = "DecodeMemMMImm12";
   let mayLoad = 1;
 }
 
 class StoreMultMM16<string opstr,
                     InstrItinClass Itin = NoItinerary,
                     ComplexPattern Addr = addr> :
-  MicroMipsInst16<(outs), (ins reglist16:$rt, mem_mm_4sp:$addr),
+  MicroMipsInst16<(outs), (ins reglist16:$rt, variable_ops, mem_mm_4sp:$addr),
                   !strconcat(opstr, "\t$rt, $addr"), [], Itin, FrmI> {
   let DecoderMethod = "DecodeMemMMReglistImm4Lsl2";
   let mayStore = 1;
 }
 
 class LoadMultMM16<string opstr,
                    InstrItinClass Itin = NoItinerary,
                    ComplexPattern Addr = addr> :
-  MicroMipsInst16<(outs reglist16:$rt), (ins mem_mm_4sp:$addr),
+  MicroMipsInst16<(outs reglist16:$rt, variable_ops), (ins mem_mm_4sp:$addr),
                   !strconcat(opstr, "\t$rt, $addr"), [], Itin, FrmI> {
   let DecoderMethod = "DecodeMemMMReglistImm4Lsl2";
   let mayLoad = 1;
 }
 
 class UncondBranchMM16<string opstr> :
   MicroMipsInst16<(outs), (ins brtarget10_mm:$offset),
                   !strconcat(opstr, "\t$offset"),

lib/Target/Mips/Mips.h

namespace llvm {
   ModulePass *createMipsOs16Pass(MipsTargetMachine &TM);
   ModulePass *createMips16HardFloatPass(MipsTargetMachine &TM);
 
   FunctionPass *createMipsModuleISelDagPass(MipsTargetMachine &TM);
   FunctionPass *createMipsOptimizePICCallPass(MipsTargetMachine &TM);
   FunctionPass *createMipsDelaySlotFillerPass(MipsTargetMachine &TM);
   FunctionPass *createMipsLongBranchPass(MipsTargetMachine &TM);
   FunctionPass *createMipsConstantIslandPass(MipsTargetMachine &tm);
+  FunctionPass *createMicroMips32SizeReductionPass();
 } // end namespace llvm;
 
 #endif

lib/Target/Mips/MipsTargetMachine.cpp

TargetIRAnalysis MipsTargetMachine::getTargetIRAnalysis() {
   });
 }
 
 // Implemented by targets that want to run passes immediately before
 // machine code is emitted. return true if -print-machineinstrs should
 // print out the code after the passes.
 void MipsPassConfig::addPreEmitPass() {
   MipsTargetMachine &TM = getMipsTargetMachine();
+  addPass(createMicroMips32SizeReductionPass());
   addPass(createMipsDelaySlotFillerPass(TM));
   addPass(createMipsLongBranchPass(TM));
   addPass(createMipsConstantIslandPass(TM));
 }

test/CodeGen/Mips/micromips-lwm-swm-lwp-swp-sw16.ll

+; RUN: llc -march=mipsel -mattr=+micromips < %s | FileCheck %s
+
+define void @ell_3m_mul_d(double* %m3, double* %m1, double* %m2) #0 {
+entry:
+; CHECK: swm16
+; CHECK: lwp
+; CHECK: lwm16
+  %arrayidx = getelementptr double, double* %m1, i32 5
+  %0 = load double, double* %arrayidx, align 8
+  %arrayidx1 = getelementptr double, double* %m2, i32 1
+  %1 = load double, double* %arrayidx1, align 8
+  %mul = fmul double %0, %1
+  %arrayidx2 = getelementptr double, double* %m1, i32 8
+  %2 = load double, double* %arrayidx2, align 8
+  %arrayidx3 = getelementptr double, double* %m2, i32 2
+  %3 = load double, double* %arrayidx3, align 8
+  %mul4 = fmul double %2, %3
+  %add = fadd double %mul, %mul4
+  %arrayidx5 = getelementptr double, double* %m3, i32 2
+  store double %add, double* %arrayidx5, align 8
+  ret void
+}
+
+define void @ell_4m_inv_d(double* %i, double* %m) #0 {
+entry:
+; CHECK: swm32
+; CHECK: lwp
+; CHECK: lwp
+; CHECK: lwp
+; CHECK: lwp
+; CHECK: lwm32
+  %0 = load double, double* %m, align 8
+  %arrayidx1 = getelementptr double, double* %m, i32 5
+  %1 = load double, double* %arrayidx1, align 8
+  %arrayidx2 = getelementptr double, double* %m, i32 10
+  %2 = load double, double* %arrayidx2, align 8
+  %mul = fmul double %1, %2
+  %arrayidx3 = getelementptr double, double* %m, i32 15
+  %3 = load double, double* %arrayidx3, align 8
+  %mul4 = fmul double %mul, %3
+  %mul5 = fmul double %0, %mul4
+  %arrayidx11 = getelementptr double, double* %m, i32 9
+  %4 = load double, double* %arrayidx11, align 8
+  %arrayidx12 = getelementptr double, double* %m, i32 14
+  %5 = load double, double* %arrayidx12, align 8
+  %mul13 = fmul double %4, %5
+  %add = fadd double %mul4, %mul13
+  %div = fdiv double %add, %mul5
+  store double %div, double* %i, align 8
+  ret void
+}
+
+define i32 @sw16(i32* nocapture %r) #0 {
+entry:
+; CHECK: sw16
+  store i32 0, i32* %r, align 4
+  ret i32 0
+}
+
+%union.expr_rec = type { %struct.constant_rec }
+%struct.constant_rec = type { i32, %union.Type_Rec*, i32, i32, i32, [4 x i8*], i32, i32, [4 x %union.scalar_constant_rec] }
+%union.Type_Rec = type { %struct.TypeStruct_Rec }
+%struct.TypeStruct_Rec = type { i32, i32, %union.Type_Rec*, %struct.Scope_Rec*, %struct.SourceLoc_Rec, i32, i32, i32, i32, i8*, i32, i8* }
+%struct.Scope_Rec = type { %struct.Scope_Rec*, %struct.Scope_Rec*, %struct.Scope_Rec*, %struct.Scope_Rec*, %struct.MemoryPool_rec*, %struct.Symbol_Rec*, %struct.Symbol_Rec*, %struct.Symbol_Rec*, %union.Type_Rec*, i32, i32, i32, i32, i32, i32, i32, i32, %struct.SymbolList_Rec*, %union.stmt_rec* }
+%struct.MemoryPool_rec = type opaque
+%struct.Symbol_Rec = type { %struct.Symbol_Rec*, %struct.Symbol_Rec*, %struct.Symbol_Rec*, i32, %union.Type_Rec*, %struct.SourceLoc_Rec, i32, i32, i32, i32, i8*, i8*, %union.anon }
+%union.anon = type { %struct.FunSymbol_Rec }
+%struct.FunSymbol_Rec = type { %struct.Scope_Rec*, %struct.Symbol_Rec*, %union.stmt_rec*, %struct.Symbol_Rec*, i32, i16, i16, i8 }
+%struct.SymbolList_Rec = type { %struct.SymbolList_Rec*, %struct.Symbol_Rec* }
+%union.stmt_rec = type { %struct.for_stmt_rec }
+%struct.for_stmt_rec = type { i32, %union.stmt_rec*, %struct.SourceLoc_Rec, %union.stmt_rec*, %union.expr_rec*, %union.stmt_rec*, %union.stmt_rec* }
+%struct.SourceLoc_Rec = type { i16, i16 }
+%union.scalar_constant_rec = type { float }
+%struct.binary_rec = type { i32, %union.Type_Rec*, i32, i32, i32, [4 x i8*], i32, i32, %union.expr_rec*, %union.expr_rec* }
+
+; Function Attrs: nounwind
+define %union.expr_rec* @GenVAssign(%union.expr_rec* %fVar, %union.expr_rec* %fExpr, i32 signext %base, i32 signext %len) {
+entry:
+; CHECK: swp
+; CHECK: lwp
+; CHECK: lwm16
+  %and = shl i32 %len, 4
+  %shl = and i32 %and, 240
+  %and1 = and i32 %base, 15
+  %or = or i32 %shl, %and1
+  %call = tail call %struct.binary_rec* @NewBinopSubNode(i32 signext 106, i32 signext %or, %union.expr_rec* %fVar, %union.expr_rec* %fExpr)
+  %0 = bitcast %struct.binary_rec* %call to %union.expr_rec*
+  %call2 = tail call %union.Type_Rec* @GetStandardType(i32 signext %base, i32 signext %len, i32 signext 0)
+  %type = getelementptr inbounds %struct.binary_rec, %struct.binary_rec* %call, i32 0, i32 1
+  store %union.Type_Rec* %call2, %union.Type_Rec** %type, align 4
+  ret %union.expr_rec* %0
+}
+
+declare %struct.binary_rec* @NewBinopSubNode(i32 signext, i32 signext, %union.expr_rec*, %union.expr_rec*)
+declare %union.Type_Rec* @GetStandardType(i32 signext, i32 signext, i32 signext)
+
+attributes #0 = { nounwind "use-soft-float"="true"  }

test/CodeGen/Mips/micromips-lwsp-swsp.ll

+; RUN: llc -march=mipsel -mattr=+micromips -filetype=asm -asm-show-inst < %s | FileCheck %s
+
+%struct.inflate_blocks_state = type { i32, %union.anon, i32, i32, i32, %struct.inflate_huft_s*, i8*, i8*, i8*, i8*, i32 (i32, i8*, i32)*, i32 }
+%union.anon = type { %struct.anon }
+%struct.anon = type { i32, i32, i32*, i32, %struct.inflate_huft_s* }
+%struct.inflate_huft_s = type { %union.anon.0, i32 }
+%union.anon.0 = type { i32 }
+%struct.z_stream_s = type { i8*, i32, i32, i8*, i32, i32, i8*, %struct.internal_state*, i8* (i8*, i32, i32)*, void (i8*, i8*)*, i8*, i32, i32, i32 }
+%struct.internal_state = type { i32 }
+
+@inflate_mask = global [17 x i32] [i32 0, i32 1, i32 3, i32 7, i32 15, i32 31, i32 63, i32 127, i32 255, i32 511, i32 1023, i32 2047, i32 4095, i32 8191, i32 16383, i32 32767, i32 65535], align 4
+
+; Function Attrs: nounwind
+define i32 @inflate_flush(%struct.inflate_blocks_state* nocapture %s, %struct.z_stream_s* nocapture %z, i32 signext %r) {
+entry:
+; CHECK: SWSP_MM
+; CHECK: SWSP_MM
+; CHECK: SWSP_MM
+; CHECK: LWSP_MM
+; CHECK: LWSP_MM
+; CHECK: LWSP_MM
+  %next_out = getelementptr inbounds %struct.z_stream_s, %struct.z_stream_s* %z, i32 0, i32 3
+  %0 = load i8*, i8** %next_out, align 4
+  %read = getelementptr inbounds %struct.inflate_blocks_state, %struct.inflate_blocks_state* %s, i32 0, i32 8
+  %1 = load i8*, i8** %read, align 4
+  %write = getelementptr inbounds %struct.inflate_blocks_state, %struct.inflate_blocks_state* %s, i32 0, i32 9
+  %2 = load i8*, i8** %write, align 4
+  %cmp = icmp ugt i8* %1, %2
+  br label %cond.end
+
+cond.end:                                         ; preds = %entry
+  %sub.ptr.lhs.cast = ptrtoint i8* %2 to i32
+  %sub.ptr.rhs.cast = ptrtoint i8* %1 to i32
+  %sub.ptr.sub = sub i32 %sub.ptr.lhs.cast, %sub.ptr.rhs.cast
+  %avail_out = getelementptr inbounds %struct.z_stream_s, %struct.z_stream_s* %z, i32 0, i32 4
+  %3 = load i32, i32* %avail_out, align 4
+  %cmp2 = icmp ugt i32 %sub.ptr.sub, %3
+  %.sub.ptr.sub = select i1 %cmp2, i32 %3, i32 %sub.ptr.sub
+  %sub = sub i32 %3, %.sub.ptr.sub
+  store i32 %sub, i32* %avail_out, align 4
+  %total_out = getelementptr inbounds %struct.z_stream_s, %struct.z_stream_s* %z, i32 0, i32 5
+  %4 = load i32, i32* %total_out, align 4
+  %checkfn = getelementptr inbounds %struct.inflate_blocks_state, %struct.inflate_blocks_state* %s, i32 0, i32 10
+  %5 = load i32 (i32, i8*, i32)*, i32 (i32, i8*, i32)** %checkfn, align 4
+  %cmp8 = icmp eq i32 (i32, i8*, i32)* %5, null
+  br i1 %cmp8, label %if.end.12, label %if.then.9
+
+if.then.9:                                        ; preds = %cond.end
+  %check = getelementptr inbounds %struct.inflate_blocks_state, %struct.inflate_blocks_state* %s, i32 0, i32 11
+  %6 = load i32, i32* %check, align 4
+  %call = tail call i32 %5(i32 signext %6, i8* %1, i32 signext %.sub.ptr.sub)
+  store i32 %call, i32* %check, align 4
+  br label %if.end.12
+
+if.end.12:                                        ; preds = %cond.end, %if.then.9
+  tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %0, i8* %1, i32 %.sub.ptr.sub, i32 1, i1 false)
+  ret i32 0
+}
+
+; Function Attrs: nounwind argmemonly
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture readonly, i32, i32, i1)