[IR] Add Freeze instruction

[aqjune]

Summary

• Define Instruction::Freeze, let it be UnaryOperator
• Add support for freeze to LLLexer/LLParser/BitcodeReader/BitcodeWriter The format is %x = freeze <ty> %v
• Add support for freeze instruction to llvm-c interface.
• Add m_Freeze in PatternMatch.
• Erase freeze when lowering IR to SelDag.

[deadalnix]

Hi,

Could you add a test case for the C API ? You can do so using the echo facility of llvm-c-test. See for instance: https://reviews.llvm.org/diffusion/L/browse/llvm/trunk/test/Bindings/llvm-c/atomics.ll

include/llvm-c/Core.h

3028

nit: please align the Name parameter properly.

[aqjune]

• Add llvm-c test (test/Bindings/llvm-c/freeze.ll)
• Update tools/llvm-c-test/echo.cpp so it can deal with Freeze instruction.

[deadalnix]

Awesome, as far as the C API goes, this is good. I'll let other the time to review everything else.

[filcab]

Just some minor nits.
You're adding some functions which will never be called. It might be easier to hold them off until the next patch (and have this one only add the instruction + necessary plumbing), or to at least post a follow-up patch on phabricator which shows usage for those functions, as that makes it easier to know it shouldn't be dead code for long.

include/llvm/IR/IRBuilder.h

1713

Maybe omit the \brief?

1715

This is not being used. Maybe submit in a later patch, with a use for it?

include/llvm/IR/Instructions.h

5064

undef?

include/llvm/IR/PatternMatch.h

931

Minor nit: Unsure if the usual pattern matching policy is to add the pattern only when you need it on when adding the instruction.

lib/Analysis/ValueTracking.cpp

3917 ↗

(On Diff #85541)

If we're just doing ConstantInt for now, why not just use the inner if (same for the FreezeInst)? Then change it after adding more, if the code ends up being too complex or has too many repeated tests.
No need to add a comment stating what an if (isa...) return is doing.
Is there a plan for adding constants, etc? Maybe add a TODO tag here, so it jumps out if someone is grepping for those.

lib/AsmParser/LLParser.cpp

5914

Please test this error.

5916

Omit the , "" since it's the default arg anyway.

lib/Bitcode/Reader/BitcodeReader.cpp

4350

Omit the , "".

[filcab]

lib/AsmParser/LLParser.cpp

5914

Even better: Should this be in the verifier only? Otherwise you're checking for the same thing in two places, which is easy to get out of sync. (Would still require a test)

[majnemer]

lib/AsmParser/LLParser.cpp

5914

its not bad for them to be in both, the error messages are much nicer in the parser.

[vsk]

Please move the bitcode round-trip in test/Bitcode/freeze.ll to test/Bitcode/compatibility.ll. This lets us test bitcode backwards-compatibility.

[aqjune]

• Modify typos, add more comments (including TODOs)
• Add tests for parsing error
• Moved test/Bitcode/freeze.ll to test/Bitcode/compatibility.ll.

[aqjune]

include/llvm/IR/IRBuilder.h

1715

Hello. Thanks for your comment.
This function is going to be used in https://reviews.llvm.org/D29016, and this patch is important because it may reduce the number of unnecessarily inserted freezes.

include/llvm/IR/PatternMatch.h

931

You're right - it is not used right now.
However every other values has its own match function currently, so I just followed convention..
If you think this is redundant now, I'll erase this.

lib/Analysis/ValueTracking.cpp

3917 ↗

(On Diff #85760)

There's no plan to add other cases now.
I added TODOs, as you recommended.

lib/AsmParser/LLParser.cpp

5914

I added 3 test cases which checks whether this error message is printed out.

5916

I removed the second argument.

[filcab]

include/llvm/IR/IRBuilder.h

1715

That patch seems to have no instance of CreateFreezeAtDef. It's better to only add this function when it's going to be used.

include/llvm/IR/PatternMatch.h

931

It's a minor nit. I'm ok with this being added now with the rest of the instruction stuff.

lib/Analysis/ValueTracking.cpp

3901 ↗

(On Diff #85760)

This function is unused.

3917 ↗

(On Diff #85760)

Maybe this?

// TODO: Deal with other Constant subclasses.
if (isa<ConstantInt>(C))
  return true;

return false;

I don't think the outer if is buying you anything for now. Anyone who adds more can deal with avoiding testing the same thing twice.
(we can even return isa<...>...;, but keeping the "itemization"-like code doesn't seem bad in my view).

lib/Bitcode/Reader/BitcodeReader.cpp

4350

If you don't test for an integer type, this will generate invalid IR. It's also not consistent with the parser.

lib/IR/Verifier.cpp

3654

Please add a test to unittests/IR/VerifierTest.cpp

test/Bitcode/freeze-vector.ll

4

No need to name the LINE variable. But you should be matching to make sure we get the correct line.
Something like (using the automatic @LINE variable):

CHECK: freeze-vector.ll:[[@LINE+2]]: ```...
The same for the other tests.

[aqjune]

• Move isGuaranteedNotToBeUndefOrPoison to https://reviews.llvm.org/D29013
• Move CreateFreezeAtDef to https://reviews.llvm.org/D29015
• Add a test to unittest/unittests/IR/VerifierTest.cpp
• Add type checking to BitcodeReader.cpp

[aqjune]

include/llvm/IR/IRBuilder.h

1715

Oh, sorry, the patch was https://reviews.llvm.org/D29015 , not https://reviews.llvm.org/D29016.
I moved the CreateFreezeAtDef to https://reviews.llvm.org/D29015.

[aqjune]

include/llvm/IR/IRBuilder.h

1713

@filcab Other functions also have \brief , so I followed the convention.

[filcab]

Looks good on my end. But please wait for someone more familiar with IR to chime in and LGTM, since I'm not sure of all places that need to be changed when adding instructions.

include/llvm/IR/IRBuilder.h

1713

Consistency is ok by me. If we ever convert these, we can do it on all at once.

[aqjune]

@filcab Thanks for your comments.

[aqjune]

Rebased to svn commit 308173

[delcypher]

lib/AsmParser/LLParser.cpp

5914

@aqjune What's the motivation to restricting the freeze instruction to scalar integers?

For vector SSA registers that contain integers that we want to "freeze" this would force doing constructing a new vector by use of extractelement, freeze on the each scalar element and then insertelement to construct the new register. This seems a bit cumbersome.

I presume the reason for not allowing floating point operands is that none of the floating point instructions are supposed to produce poison? However consider the load instruction. That could load poisoned memory and the type could be a floating point type. There's an example of a poisoned load in the IR language reference ( https://llvm.org/docs/LangRef.html#poison-values ).

[nlopes]

lib/AsmParser/LLParser.cpp

5914

You're right that in the future it may be useful to freeze vectors of integers, at least.
This is just the first version of the patch. The intent was to fix loop unswitch, which doesn't require freezing vectors yet. As uses expand we can relax the condition.
We definitely didn't want to freeze floats; I don't know yet if that's needed or what does it mean.

[delcypher]

lib/AsmParser/LLParser.cpp

5914

Thanks for clarifying. I see the instruction is documented in https://reviews.llvm.org/D29121 so I left a minor comment there about this.

[aqjune]

Rebased to svn commit 331585

[xbolva00]

Ping

[aqjune]

Rebase to trunk@369887 (August 26th, 2019)

[lebedev.ri]

Thank you for working on this! This needs more cleanup.
Since the patch was initially posted, and Unary operator type was added to LLVM.
In last patch's update many parts of the patch were correctly migrated, but not all.
See D53877 for the list of potential places, i tried to point out most of them inline.

include/llvm-c/Core.h

137–140

unrelated

include/llvm/Bitcode/LLVMBitCodes.h

312–313

Did you mean to put freeze as an unary instruction instead of a function?

464–466

Wouldn't basing it on unary op work?

include/llvm/CodeGen/GlobalISel/IRTranslator.h

487 ↗

(On Diff #217187)

Is this a correctness issue?
Or does returning false here results in IRTranslator "aborting"?

include/llvm/IR/IRBuilder.h

1694

I personally have mild preference for the ::Create; not sure if this is a remark or review note.

include/llvm/IR/InstVisitor.h

209

Should you delegate to UnaryInstruction instead?

lib/AsmParser/LLParser.cpp

5913

Needs updating to match langref - vectors; floating point support?

lib/IR/Verifier.cpp

3655–3656

Needs updating to match langref - vectors; floating point support?

[kristof.beyls]

Given this seems to add a new IR instruction, shouldn't there also be good quality documentation for this new instruction in docs/LangRef.rst?

[lebedev.ri]

In D29011#1696898, @kristof.beyls wrote:

Given this seems to add a new IR instruction, shouldn't there also be good quality documentation for this new instruction in docs/LangRef.rst?

See D29121

[aqjune]

Updated the patch so UnaryOperator is used for Instruction::Freeze op.
Currently I see an error from Phabricator while uploading a new diff:

Unhandled Exception ("PhabricatorFileUploadException")	
Unable to write file: failed to write to temporary directory.

I'll retry uploading it later.

include/llvm/CodeGen/GlobalISel/IRTranslator.h

487 ↗

(On Diff #217187)

Yep, returning false makes it abort, if my understanding is correct.

[lebedev.ri]

Thanks, this looks about right.
I'm not sure about backend part of this patch, but that is likely okay.
Two main questions: should there be a constantexpr freeze, and should freeze be fully type-agnostic, like it is stated in D29121?

include/llvm/CodeGen/GlobalISel/IRTranslator.h

487 ↗

(On Diff #217187)

Okay.

lib/AsmParser/LLParser.cpp

3041–3042

Should there be constantexpr version of freeze?

5610–5611

I see no restrictions on the type to freeze in D29121,
so i think you just don't want any checking for freeze here.
And existing ParseUnaryOp is only called with /*IsFP*/true to parse fneg.
So let's change this to

bool Valid = !IsFPOnly || LHS->getType()->isFPOrFPVectorTy();

?

lib/Bitcode/Writer/BitcodeWriter.cpp

2294–2295

What about constantexpr freeze?

lib/IR/Instructions.cpp

1986–1988

I'm not seeing this restriction in D29121

lib/IR/Verifier.cpp

2842–2845

I'm not seeing this restriction in D29121

[aqjune]

should there be a constantexpr freeze

Yep, constantexpr freeze makes sense, as freeze is a scalar operation (like fneg). :)

and should freeze be fully type-agnostic, like it is stated in D29121?

I think this is a hard question, especially due to the existence of the undef pointer.
A pointer value tracks which memory block it is pointing to. If freeze i8* undef is defined to yield a random pointer to any pre-defined memory block, this will limit free moving of freeze, e.g:

p = malloc() // p is created
ptr0 = freeze i8* undef // ptr0 can point to block p
=>
ptr0 = freeze i8* undef // ptr0 can't point to block p, because p is not allocated yet
p = malloc()

[lebedev.ri]

In D29011#1697170, @aqjune wrote:

should there be a constantexpr freeze

Yep, constantexpr freeze makes sense, as freeze is a scalar operation (like fneg). :)

See inline comments then :)

and should freeze be fully type-agnostic, like it is stated in D29121?

I think this is a hard question, especially due to the existence of the undef pointer.
A pointer value tracks which memory block it is pointing to. If freeze i8* undef is defined to yield a random pointer to any pre-defined memory block, this will limit free moving of freeze, e.g:

p = malloc() // p is created
ptr0 = freeze i8* undef // ptr0 can point to block p
=>
ptr0 = freeze i8* undef // ptr0 can't point to block p, because p is not allocated yet
p = malloc()

I guess D29121 needs to explicitly single-out the pointers then.

[lebedev.ri]

(marking as reviewed as per previous comments - constantexpr, typechecking inconsistent with langref)

[aqjune]

• Support freeze constexpr
• Make freeze's type checking consistent with LangRef.

[jdoerfert]

Two minor comments from my side.

(I would like someone to include the keyword in the editor lists llvm/utils/{emacs,vim} after this landed)

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h

826

The lady of the lake says this should be:

`void visitFreeze(const User &I) { visitUnary(I, ISD::FREEZE); }`

If you have reason not to do it this way, also replace visitUnrary with visitFNeg, though I'd prefer not to.

test/Bindings/llvm-c/freeze.ll

13

Missing types, here and elsewhere I think:

• array
• struct w/ definition
• struct w/o definition (opaque)
• non-standard integer size (i666)

Missing inputs:

• undef
• null

[lebedev.ri]

In D29011#1707274, @jdoerfert wrote:

Two minor comments from my side.

Looks good to me with those two fixed.
Should you add llvm::Freze here by inheriting from UnaryOperator to make isa<Freze>(Op) possible?

(I would like someone to include the keyword in the editor lists llvm/utils/{emacs,vim} after this landed)

There is also llvm/utils/{kate}.

include/llvm/Bitcode/LLVMBitCodes.h

312–313

I think you want to rebase.

[aqjune]

In D29011#1707278, @lebedev.ri wrote:

Should you add llvm::Freeze here by inheriting from UnaryOperator to make isa<Freeze>(Op) possible?

Couldn't you kindly point which place is good to update?

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h

826

ISD::FREEZE will be added in the next patch - https://reviews.llvm.org/D29014 .
Do you want to move the definition of ISD::FREEZE to this patch? @lebedev.ri

[lebedev.ri]

In D29011#1707396, @aqjune wrote:

In D29011#1707278, @lebedev.ri wrote:

Should you add llvm::Freeze here by inheriting from UnaryOperator to make isa<Freeze>(Op) possible?

Couldn't you kindly point which place is good to update?

Right after where the llvm::UnaryOperator is defined in llvm/include/llvm/IR/InstrTypes.h i think.
See OverflowingBinaryOperator for an example.

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h

826

Oh good point, let's leave as is.

[aqjune]

• Rebase
• Add more tests to freeze.ll
• Define FreezeOperator

[aqjune]

include/llvm/IR/Operator.h

594 ↗

(On Diff #224771)

I added FreezeOperator here, as other operators including OverflowingBinaryOperator were in this file.

test/Bindings/llvm-c/freeze.ll

13

I found that freeze i8* null raised an error due to a bug in tools/llvm-c-test/echo.cpp:

// Try null
if (LLVMIsNull(Cst)) {
  check_value_kind(Cst, LLVMConstantTokenNoneValueKind);
  LLVMTypeRef Ty = TypeCloner(M).Clone(Cst);
  return LLVMConstNull(Ty);
}

Here, Cst can not only be LLVMConstantTokenNoneValueKind but also LLVMConstantPointerNullValueKind.
I'll make a separate patch that resolves this error, then I'll be able to add freeze i8* null test.

[lebedev.ri]

Thanks, LG.
Anyone else has any comments here?

include/llvm/IR/PatternMatch.h

462–466

Let's use FreezeOperator here then?

[aqjune]

• Use FreezeOperator inside Freeze_match

[lebedev.ri]

In line with a few recent reverts, is this missing anything for ocaml binding?
(does it compile with ocaml binding enabled and do all tests still pass?)

[jdoerfert]

Fwiw, I'm OK with this (given that we have my requested changes in upcoming commits).

[jdoerfert]

In D29011#1708103, @lebedev.ri wrote:

In line with a few recent reverts, is this missing anything for ocaml binding?
(does it compile with ocaml binding enabled and do all tests still pass?)

Like D61675 ;)

[aqjune]

Sorry, I've been busy recently.

make check-llvm-bindings-ocaml wasn't raising error, but I added tests for freeze to core.ml because it is good to have the tests.

BTW, make check raises an error at test/Transforms/MergeFunc/inline-asm.ll because the output has function definitions reordered - what is the best way to resolve this? I attached the input/output .ll files.

inline-asm.tar.gz633 BDownload

[craig.topper]

Sorry to come to this late, but does it really make sense for Freeze to be inherited from UnaryOperator? UnaryOperator in my mind is like BinaryOperator which is for arithmetic operations. Freeze is different. The fact that it made sense to add a FreezeOperator hints at that. Should we instead just have a FreezeInst separate from UnaryOperator instead of FreezeOperator?

[aqjune]

In D29011#1733572, @craig.topper wrote:

Sorry to come to this late, but does it really make sense for Freeze to be inherited from UnaryOperator? UnaryOperator in my mind is like BinaryOperator which is for arithmetic operations. Freeze is different. The fact that it made sense to add a FreezeOperator hints at that. Should we instead just have a FreezeInst separate from UnaryOperator instead of FreezeOperator?

My understanding of UnaryOperator was that it is a scalar operation that has same input and output type according to the LangRef, so I thought freeze could be categorized into it.
Implementing freeze by inheriting UnaryOperator was straightforward; I did not have to specify instruction's properties, adding it to constexpr was easy, etc. I believe having it inherit UnaryOperator would be good for maintenance, but other than this I have no strong reason for this.
About the FreezeOperator, I think it is natural to have operators in Operator.h that has special purposes: it also has AddOperator / SubOperator / etc defined.

[craig.topper]

Does a ConstantExpr freeze really make sense? ConstantExprs are uniqued by the LLVMContext. So every constantexpr that has the same input ends up being the same object. Do we want that behavior do we need each freeze to be distinct?

[aqjune]

In D29011#1735007, @craig.topper wrote:

Does a ConstantExpr freeze really make sense? ConstantExprs are uniqued by the LLVMContext. So every constantexpr that has the same input ends up being the same object. Do we want that behavior do we need each freeze to be distinct?

Both semantics make sense, and I was thinking of returning the same value for same expression of freeze, because it explains copy of an instruction that has freeze constexpr as an operand.
Conceptually, we can say that freeze constexpr is evaluated at the beginning of a program, and the evaluated value is globally used inside the module.
Using this semantics, freeze constexprs should be carefully optimized, so the same freeze constexprs don't diverge into different expressions.
This is the difference from other constexprs, but I guess binary operator also has a similar diverged case, which is udiv/sdiv constexpr (that made constexpr may trap), but I have to say that having such difference isn't said to be a good idea.
What do you think?

[nlopes]

In D29011#1735007, @craig.topper wrote:

Does a ConstantExpr freeze really make sense? ConstantExprs are uniqued by the LLVMContext. So every constantexpr that has the same input ends up being the same object. Do we want that behavior do we need each freeze to be distinct?

Juneyoung and I have discussed this today.
The conclusion was that we don't see a use-case for a freeze ConstantExpr and therefore we agree that it should be removed. Also, the semantics and implementation would be tricky as you suggest. @aqjune will submit a new patch to change Freeze from being a UnaryOp to just an op on its own.

[aqjune]

The new patch is here: https://reviews.llvm.org/D69932

include/llvm-c/Core.h

typedef enum {
 
   /* Exception Handling Operators */
   LLVMResume         = 58,
   LLVMLandingPad     = 59,
   LLVMCleanupRet     = 61,
   LLVMCatchRet       = 62,
   LLVMCatchPad       = 63,
   LLVMCleanupPad     = 64,
-  LLVMCatchSwitch    = 65
+  LLVMCatchSwitch    = 65,
+
+  /* Freeze operator */
+  LLVMFreeze         = 66
 } LLVMOpcode;
 
 typedef enum {
   LLVMVoidTypeKind,        /**< type with no size */
   LLVMHalfTypeKind,        /**< 16 bit floating point type */
   LLVMFloatTypeKind,       /**< 32 bit floating point type */
   LLVMDoubleTypeKind,      /**< 64 bit floating point type */
   LLVMX86_FP80TypeKind,    /**< 80 bit floating point type (X87) */
macro(Instruction)                      \
           macro(PtrToIntInst)               \
           macro(SExtInst)                   \
           macro(SIToFPInst)                 \
           macro(TruncInst)                  \
           macro(UIToFPInst)                 \
           macro(ZExtInst)                   \
         macro(ExtractValueInst)             \
         macro(LoadInst)                     \
-        macro(VAArgInst)
+        macro(VAArgInst)                    \
+        macro(FreezeInst)
 
 /**
  * @defgroup LLVMCCoreValueGeneral General APIs
  *
  * Functions in this section work on all LLVMValueRef instances,
  * regardless of their sub-type. They correspond to functions available
  * on llvm::Value.
  *

 LLVMValueRef LLVMBuildShuffleVector(LLVMBuilderRef, LLVMValueRef V1,
                                     LLVMValueRef V2, LLVMValueRef Mask,
                                     const char *Name);
 LLVMValueRef LLVMBuildExtractValue(LLVMBuilderRef, LLVMValueRef AggVal,
                                    unsigned Index, const char *Name);
 LLVMValueRef LLVMBuildInsertValue(LLVMBuilderRef, LLVMValueRef AggVal,
                                   LLVMValueRef EltVal, unsigned Index,
                                   const char *Name);
+LLVMValueRef LLVMBuildFreeze(LLVMBuilderRef, LLVMValueRef Val,
+                             const char *Name);
 
 LLVMValueRef LLVMBuildIsNull(LLVMBuilderRef, LLVMValueRef Val,
                              const char *Name);
 LLVMValueRef LLVMBuildIsNotNull(LLVMBuilderRef, LLVMValueRef Val,
                                 const char *Name);
 LLVMValueRef LLVMBuildPtrDiff(LLVMBuilderRef, LLVMValueRef LHS,
                               LLVMValueRef RHS, const char *Name);
 LLVMValueRef LLVMBuildFence(LLVMBuilderRef B, LLVMAtomicOrdering ordering,

include/llvm/Bitcode/LLVMBitCodes.h

enum CastOpcodes {
   CAST_SITOFP = 6,
   CAST_FPTRUNC = 7,
   CAST_FPEXT = 8,
   CAST_PTRTOINT = 9,
   CAST_INTTOPTR = 10,
   CAST_BITCAST = 11,
   CAST_ADDRSPACECAST = 12
 };
 
 /// BinaryOpcodes - These are values used in the bitcode files to encode which
 /// binop a CST_CODE_CE_BINOP or a XXX refers to.  The values of these enums
 /// have no fixed relation to the LLVM IR enum values.  Changing these will
 /// break compatibility with old files.
 enum BinaryOpcodes {
   BINOP_ADD = 0,
   BINOP_SUB = 1,
   BINOP_MUL = 2,
   BINOP_UDIV = 3,
enum FunctionCodes {
   FUNC_CODE_INST_CLEANUPRET = 48,  // CLEANUPRET: [val] or [val,bb#]
   FUNC_CODE_INST_CATCHRET = 49,    // CATCHRET: [val,bb#]
   FUNC_CODE_INST_CATCHPAD = 50,    // CATCHPAD: [bb#,bb#,num,args...]
   FUNC_CODE_INST_CLEANUPPAD = 51,  // CLEANUPPAD: [num,args...]
   FUNC_CODE_INST_CATCHSWITCH =
       52, // CATCHSWITCH: [num,args...] or [num,args...,bb]
   // 53 is unused.
   // 54 is unused.
   FUNC_CODE_OPERAND_BUNDLE = 55, // OPERAND_BUNDLE: [tag#, value...]
+  FUNC_CODE_INST_FREEZE = 56,    // FREEZE: [opty, opval]
 };
 
 enum UseListCodes {
   USELIST_CODE_DEFAULT = 1, // DEFAULT: [index..., value-id]
   USELIST_CODE_BB = 2       // BB: [index..., bb-id]
 };
 
 enum AttributeKindCodes {
   // = 0 is unused

include/llvm/IR/IRBuilder.h

Value *CreateInsertValue(Value *Agg, Value *Val,
     return Insert(InsertValueInst::Create(Agg, Val, Idxs), Name);
   }
 
   LandingPadInst *CreateLandingPad(Type *Ty, unsigned NumClauses,
                                    const Twine &Name = "") {
     return Insert(LandingPadInst::Create(Ty, NumClauses), Name);
   }
 
+  Value *CreateFreeze(Value *V, const Twine &Name = "") {
+    return Insert(new FreezeInst(V), Name);
+  }
+
   //===--------------------------------------------------------------------===//
   // Utility creation methods
   //===--------------------------------------------------------------------===//
 
   /// \brief Return an i1 value testing if \p Arg is null.
   Value *CreateIsNull(Value *Arg, const Twine &Name = "") {
     return CreateICmpEQ(Arg, Constant::getNullValue(Arg->getType()),
                         Name);
   }
 
   /// \brief Return an i1 value testing if \p Arg is not null.
   Value *CreateIsNotNull(Value *Arg, const Twine &Name = "") {
     return CreateICmpNE(Arg, Constant::getNullValue(Arg->getType()),
                         Name);
   }
 
   /// \brief Return the i64 difference between two pointer values, dividing out
   /// the size of the pointed-to objects.
   ///
   /// This is intended to implement C-style pointer subtraction. As such, the
   /// pointers must be appropriately aligned for their element types and
   /// pointing into the same object.
   Value *CreatePtrDiff(Value *LHS, Value *RHS, const Twine &Name = "") {
     assert(LHS->getType() == RHS->getType() &&
            "Pointer subtraction operand types must match!");
     PointerType *ArgType = cast<PointerType>(LHS->getType());
     Value *LHS_int = CreatePtrToInt(LHS, Type::getInt64Ty(Context));

include/llvm/IR/InstVisitor.h

#include "llvm/IR/Instruction.def"
   RetTy visitInsertElementInst(InsertElementInst &I) { DELEGATE(Instruction);}
   RetTy visitShuffleVectorInst(ShuffleVectorInst &I) { DELEGATE(Instruction);}
   RetTy visitExtractValueInst(ExtractValueInst &I){ DELEGATE(UnaryInstruction);}
   RetTy visitInsertValueInst(InsertValueInst &I)  { DELEGATE(Instruction); }
   RetTy visitLandingPadInst(LandingPadInst &I)    { DELEGATE(Instruction); }
   RetTy visitFuncletPadInst(FuncletPadInst &I) { DELEGATE(Instruction); }
   RetTy visitCleanupPadInst(CleanupPadInst &I) { DELEGATE(FuncletPadInst); }
   RetTy visitCatchPadInst(CatchPadInst &I)     { DELEGATE(FuncletPadInst); }
+  RetTy visitFreezeInst(FreezeInst &I)         { DELEGATE(Instruction); }
 
   // Handle the special instrinsic instruction classes.
   RetTy visitDbgDeclareInst(DbgDeclareInst &I)    { DELEGATE(DbgInfoIntrinsic);}
   RetTy visitDbgValueInst(DbgValueInst &I)        { DELEGATE(DbgInfoIntrinsic);}
   RetTy visitDbgInfoIntrinsic(DbgInfoIntrinsic &I) { DELEGATE(IntrinsicInst); }
   RetTy visitMemSetInst(MemSetInst &I)            { DELEGATE(MemIntrinsic); }
   RetTy visitMemCpyInst(MemCpyInst &I)            { DELEGATE(MemTransferInst); }
   RetTy visitMemMoveInst(MemMoveInst &I)          { DELEGATE(MemTransferInst); }

include/llvm/IR/Instruction.def

 HANDLE_OTHER_INST(57, UserOp2, Instruction)  // Internal to passes only
 HANDLE_OTHER_INST(58, VAArg  , VAArgInst  )  // vaarg instruction
 HANDLE_OTHER_INST(59, ExtractElement, ExtractElementInst)// extract from vector
 HANDLE_OTHER_INST(60, InsertElement, InsertElementInst)  // insert into vector
 HANDLE_OTHER_INST(61, ShuffleVector, ShuffleVectorInst)  // shuffle two vectors.
 HANDLE_OTHER_INST(62, ExtractValue, ExtractValueInst)// extract from aggregate
 HANDLE_OTHER_INST(63, InsertValue, InsertValueInst)  // insert into aggregate
 HANDLE_OTHER_INST(64, LandingPad, LandingPadInst)  // Landing pad instruction.
-  LAST_OTHER_INST(64)
+HANDLE_OTHER_INST(65, Freeze, FreezeInst) // Freeze instruction.
+  LAST_OTHER_INST(65)
 
 #undef  FIRST_TERM_INST
 #undef HANDLE_TERM_INST
 #undef   LAST_TERM_INST
 
 #undef  FIRST_BINARY_INST
 #undef HANDLE_BINARY_INST
 #undef   LAST_BINARY_INST

include/llvm/IR/Instructions.h

public:
   }
 
   /// Returns the address space of the result.
   unsigned getDestAddressSpace() const {
     return getType()->getPointerAddressSpace();
   }
 };
 
+//===----------------------------------------------------------------------===//
+//                              FreezeInst Class
+//===----------------------------------------------------------------------===//
+
+/// This class represents a freeze function that returns random concrete
+/// value if an operand is either a poison value or an undef value
+class FreezeInst : public UnaryInstruction {
+protected:
+  // Note: Instruction needs to be a friend here to call cloneImpl.
+  friend class Instruction;
+
+  /// Clone an identical FreezeInst
+  FreezeInst *cloneImpl() const;
+
+public:
+  /// Constructor with insert-before-instruction semantics
+  FreezeInst(
+    Value *S,                           ///< The value to freeze
+    const Twine &NameStr = "",          ///< A name for the new instruction
+    Instruction *InsertBefore = nullptr ///< Where to insert the new instruction
+  );
+
+  /// Constructor with insert-at-end-of-block semantics
+  FreezeInst(
+    Value *S,                     ///< The value to freeze
+    const Twine &NameStr,         ///< A name for the new instruction
+    BasicBlock *InsertAtEnd       ///< The block to insert the instruction into
+  );
+
+  // Methods for support type inquiry through isa, cast, and dyn_cast:
+  static inline bool classof(const Instruction *I) {
+    return I->getOpcode() == Freeze;
+  }
+  static inline bool classof(const Value *V) {
+    return isa<Instruction>(V) && classof(cast<Instruction>(V));
+  }
+};
+
 } // end namespace llvm
 
 #endif // LLVM_IR_INSTRUCTIONS_H

include/llvm/IR/PatternMatch.h

inline BinaryOp_match<LHS, RHS, Instruction::FSub> m_FSub(const LHS &L,
   return BinaryOp_match<LHS, RHS, Instruction::FSub>(L, R);
 }
 
 template <typename LHS, typename RHS>
 inline BinaryOp_match<LHS, RHS, Instruction::Mul> m_Mul(const LHS &L,
                                                         const RHS &R) {
   return BinaryOp_match<LHS, RHS, Instruction::Mul>(L, R);
 }
 
 template <typename LHS, typename RHS>
 inline BinaryOp_match<LHS, RHS, Instruction::FMul> m_FMul(const LHS &L,
                                                           const RHS &R) {
   return BinaryOp_match<LHS, RHS, Instruction::FMul>(L, R);
 }
 
 template <typename LHS, typename RHS>
 inline BinaryOp_match<LHS, RHS, Instruction::UDiv> m_UDiv(const LHS &L,
                                                           const RHS &R) {
   return BinaryOp_match<LHS, RHS, Instruction::UDiv>(L, R);
 }
 
private:
   }
 };
 
 /// \brief Match a floating point negate.
 template <typename LHS> inline fneg_match<LHS> m_FNeg(const LHS &L) {
   return L;
 }
 
+template <typename Op_t> struct FreezeClass_match {
+  Op_t Op;
+
+  FreezeClass_match(const Op_t &OpMatch) : Op(OpMatch) {}
+
+  template <typename OpTy> bool match(OpTy *V) {
+    if (auto *O = dyn_cast<Operator>(V))
+      return O->getOpcode() == Instruction::Freeze &&
+             Op.match(O->getOperand(0));
+    return false;
+  }
+};
+
+/// \brief Matches Freeze.
+template <typename OpTy>
+inline FreezeClass_match<OpTy> m_Freeze(const OpTy &Op) {
+  return FreezeClass_match<OpTy>(Op);
+}
+
 //===----------------------------------------------------------------------===//
 // Matchers for control flow.
 //
 
 struct br_match {
   BasicBlock *&Succ;
   br_match(BasicBlock *&Succ) : Succ(Succ) {}
 

lib/AsmParser/LLLexer.cpp

#define INSTKEYWORD(STR, Enum)                                                 \
   INSTKEYWORD(insertvalue,    InsertValue);
   INSTKEYWORD(landingpad,     LandingPad);
   INSTKEYWORD(cleanupret,     CleanupRet);
   INSTKEYWORD(catchret,       CatchRet);
   INSTKEYWORD(catchswitch,  CatchSwitch);
   INSTKEYWORD(catchpad,     CatchPad);
   INSTKEYWORD(cleanuppad,   CleanupPad);
 
+  INSTKEYWORD(freeze,       Freeze);
+
 #undef INSTKEYWORD
 
 #define DWKEYWORD(TYPE, TOKEN)                                                 \
   do {                                                                         \
     if (Keyword.startswith("DW_" #TYPE "_")) {                                 \
       StrVal.assign(Keyword.begin(), Keyword.end());                           \
       return lltok::TOKEN;                                                     \
     }                                                                          \

lib/AsmParser/LLParser.h

#include "llvm/IR/Metadata.def"
     int ParseLoad(Instruction *&I, PerFunctionState &PFS);
     int ParseStore(Instruction *&I, PerFunctionState &PFS);
     int ParseCmpXchg(Instruction *&I, PerFunctionState &PFS);
     int ParseAtomicRMW(Instruction *&I, PerFunctionState &PFS);
     int ParseFence(Instruction *&I, PerFunctionState &PFS);
     int ParseGetElementPtr(Instruction *&I, PerFunctionState &PFS);
     int ParseExtractValue(Instruction *&I, PerFunctionState &PFS);
     int ParseInsertValue(Instruction *&I, PerFunctionState &PFS);
+    bool ParseFreeze(Instruction *&I, PerFunctionState &PFS);
 
     // Use-list order directives.
     bool ParseUseListOrder(PerFunctionState *PFS = nullptr);
     bool ParseUseListOrderBB();
     bool ParseUseListOrderIndexes(SmallVectorImpl<unsigned> &Indexes);
     bool sortUseListOrder(Value *V, ArrayRef<unsigned> Indexes, SMLoc Loc);
   };
 } // End llvm namespace
 
 #endif

lib/AsmParser/LLParser.cpp

if (Opc == Instruction::FCmp) {
       if (!Val0->getType()->isIntOrIntVectorTy() &&
           !Val0->getType()->getScalarType()->isPointerTy())
         return Error(ID.Loc, "icmp requires pointer or integer operands");
       ID.ConstantVal = ConstantExpr::getICmp(Pred, Val0, Val1);
     }
     ID.Kind = ValID::t_Constant;
     return false;
   }
 
   // Binary Operators.
   case lltok::kw_add:
   case lltok::kw_fadd:
   case lltok::kw_sub:
   case lltok::kw_fsub:
   case lltok::kw_mul:
   case lltok::kw_fmul:
   case lltok::kw_udiv:
   case lltok::kw_sdiv:
int LLParser::ParseInstruction(Instruction *&Inst, BasicBlock *BB,
   // Other.
   case lltok::kw_select:         return ParseSelect(Inst, PFS);
   case lltok::kw_va_arg:         return ParseVA_Arg(Inst, PFS);
   case lltok::kw_extractelement: return ParseExtractElement(Inst, PFS);
   case lltok::kw_insertelement:  return ParseInsertElement(Inst, PFS);
   case lltok::kw_shufflevector:  return ParseShuffleVector(Inst, PFS);
   case lltok::kw_phi:            return ParsePHI(Inst, PFS);
   case lltok::kw_landingpad:     return ParseLandingPad(Inst, PFS);
+  case lltok::kw_freeze:         return ParseFreeze(Inst, PFS);
   // Call.
   case lltok::kw_call:     return ParseCall(Inst, PFS, CallInst::TCK_None);
   case lltok::kw_tail:     return ParseCall(Inst, PFS, CallInst::TCK_Tail);
   case lltok::kw_musttail: return ParseCall(Inst, PFS, CallInst::TCK_MustTail);
   case lltok::kw_notail:   return ParseCall(Inst, PFS, CallInst::TCK_NoTail);
   // Memory.
   case lltok::kw_alloca:         return ParseAlloc(Inst, PFS);
   case lltok::kw_load:           return ParseLoad(Inst, PFS);
bool LLParser::ParseCleanupPad(Instruction *&Inst, PerFunctionState &PFS) {
   SmallVector<Value *, 8> Args;
   if (ParseExceptionArgs(Args, PFS))
     return true;
 
   Inst = CleanupPadInst::Create(ParentPad, Args);
   return false;
 }
 
 //===----------------------------------------------------------------------===//
 // Binary Operators.
 //===----------------------------------------------------------------------===//
 
 /// ParseArithmetic
 ///  ::= ArithmeticOps TypeAndValue ',' Value
 ///
 /// If OperandType is 0, then any FP or integer operand is allowed.  If it is 1,
 /// then any integer operand is allowed, if it is 2, any fp operand is allowed.
 bool LLParser::ParseArithmetic(Instruction *&Inst, PerFunctionState &PFS,
if (!CV)
       return Error(VLoc, "clause argument must be a constant");
     LP->addClause(CV);
   }
 
   Inst = LP.release();
   return false;
 }
 
+/// ParseFreeze
+///   ::= 'freeze' Type Value
+bool LLParser::ParseFreeze(Instruction *&Inst, PerFunctionState &PFS) {
+  LocTy Loc;
+  Value *Op;
+  if (ParseTypeAndValue(Op, Loc, PFS))
+    return true;
+
+  if (!Op->getType()->isIntegerTy())
+    return Error(Loc,"cannot freeze non-integer type");
+
+  Inst = new FreezeInst(Op);
+  return false;
+}
+
 /// ParseCall
 ///   ::= 'call' OptionalFastMathFlags OptionalCallingConv
 ///           OptionalAttrs Type Value ParameterList OptionalAttrs
 ///   ::= 'tail' 'call' OptionalFastMathFlags OptionalCallingConv
 ///           OptionalAttrs Type Value ParameterList OptionalAttrs
 ///   ::= 'musttail' 'call' OptionalFastMathFlags OptionalCallingConv
 ///           OptionalAttrs Type Value ParameterList OptionalAttrs
 ///   ::= 'notail' 'call'  OptionalFastMathFlags OptionalCallingConv

lib/AsmParser/LLToken.h

enum Kind {
 
   kw_extractelement,
   kw_insertelement,
   kw_shufflevector,
   kw_extractvalue,
   kw_insertvalue,
   kw_blockaddress,
 
+  kw_freeze,
+
   // Metadata types.
   kw_distinct,
 
   // Use-list order directives.
   kw_uselistorder,
   kw_uselistorder_bb,
 
   // Unsigned Valued tokens (UIntVal).

lib/Bitcode/Reader/BitcodeReader.cpp

case bitc::FUNC_CODE_OPERAND_BUNDLE: {
         if (getValueTypePair(Record, OpNum, NextValueNo, Op))
           return error("Invalid record");
         Inputs.push_back(Op);
       }
 
       OperandBundles.emplace_back(BundleTags[Record[0]], std::move(Inputs));
       continue;
     }
+
+    case bitc::FUNC_CODE_INST_FREEZE: { // FREEZE: [opty,opval]
+      unsigned OpNum = 0;
+      Value *Op = nullptr;
+      if (getValueTypePair(Record, OpNum, NextValueNo, Op))
+        return error("Invalid record");
+      if (OpNum != Record.size())
+        return error("Invalid record");
+      if (!Op->getType()->isIntegerTy())
+        return error("Invalid type for value");
+
+      I = new FreezeInst(Op);
+      InstructionList.push_back(I);
+      break;
+    }
     }
 
     // Add instruction to end of current BB.  If there is no current BB, reject
     // this file.
     if (!CurBB) {
       delete I;
       return error("Invalid instruction with no BB");
     }

lib/Bitcode/Writer/BitcodeWriter.cpp

if (C->isNullValue()) {
           Code = bitc::CST_CODE_CE_BINOP;
           Record.push_back(getEncodedBinaryOpcode(CE->getOpcode()));
           Record.push_back(VE.getValueID(C->getOperand(0)));
           Record.push_back(VE.getValueID(C->getOperand(1)));
           uint64_t Flags = getOptimizationFlags(CE);
           if (Flags != 0)
             Record.push_back(Flags);
         }
         break;
       case Instruction::GetElementPtr: {
         Code = bitc::CST_CODE_CE_GEP;
         const auto *GO = cast<GEPOperator>(C);
         Record.push_back(VE.getTypeID(GO->getSourceElementType()));
         if (Optional<unsigned> Idx = GO->getInRangeIndex()) {
           Code = bitc::CST_CODE_CE_GEP_WITH_INRANGE_INDEX;
           Record.push_back((*Idx << 1) | GO->isInBounds());
         } else if (GO->isInBounds())
           Code = bitc::CST_CODE_CE_INBOUNDS_GEP;
case Instruction::Call: {
     break;
   }
   case Instruction::VAArg:
     Code = bitc::FUNC_CODE_INST_VAARG;
     Vals.push_back(VE.getTypeID(I.getOperand(0)->getType()));   // valistty
     pushValue(I.getOperand(0), InstID, Vals);                   // valist.
     Vals.push_back(VE.getTypeID(I.getType())); // restype.
     break;
+  case Instruction::Freeze:
+    Code = bitc::FUNC_CODE_INST_FREEZE;
+    pushValueAndType(I.getOperand(0), InstID, Vals);
+    break;
   }
 
   Stream.EmitRecord(Code, Vals, AbbrevToUse);
   Vals.clear();
 }
 
 /// Emit names for globals/functions etc. \p IsModuleLevel is true when
 /// we are writing the module-level VST, where we are including a function

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h

public:
   void visitJumpTableHeader(JumpTable &JT, JumpTableHeader &JTH,
                             MachineBasicBlock *SwitchBB);
 
 private:
   // These all get lowered before this pass.
   void visitInvoke(const InvokeInst &I);
   void visitResume(const ResumeInst &I);
 
   void visitBinary(const User &I, unsigned OpCode);
   void visitShift(const User &I, unsigned Opcode);
   void visitAdd(const User &I)  { visitBinary(I, ISD::ADD); }
   void visitFAdd(const User &I) { visitBinary(I, ISD::FADD); }
   void visitSub(const User &I)  { visitBinary(I, ISD::SUB); }
   void visitFSub(const User &I);
   void visitMul(const User &I)  { visitBinary(I, ISD::MUL); }
   void visitFMul(const User &I) { visitBinary(I, ISD::FMUL); }
   void visitURem(const User &I) { visitBinary(I, ISD::UREM); }
private:
   bool visitStrLenCall(const CallInst &I);
   bool visitStrNLenCall(const CallInst &I);
   bool visitUnaryFloatCall(const CallInst &I, unsigned Opcode);
   bool visitBinaryFloatCall(const CallInst &I, unsigned Opcode);
   void visitAtomicLoad(const LoadInst &I);
   void visitAtomicStore(const StoreInst &I);
   void visitLoadFromSwiftError(const LoadInst &I);
   void visitStoreToSwiftError(const StoreInst &I);
+  void visitFreeze(const FreezeInst &I);
 
   void visitInlineAsm(ImmutableCallSite CS);
   const char *visitIntrinsicCall(const CallInst &I, unsigned Intrinsic);
   void visitTargetIntrinsic(const CallInst &I, unsigned Intrinsic);
 
   void visitVAStart(const CallInst &I);
   void visitVAArg(const VAArgInst &I);
   void visitVAEnd(const CallInst &I);

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

if (NumClusters > 3 && TM.getOptLevel() != CodeGenOpt::None &&
       // For optimized builds, lower large range as a balanced binary tree.
       splitWorkItem(WorkList, W, SI.getCondition(), SwitchMBB);
       continue;
     }
 
     lowerWorkItem(W, SI.getCondition(), SwitchMBB, DefaultMBB);
   }
 }
+
+void SelectionDAGBuilder::visitFreeze(const FreezeInst &I) {
+  SDValue N = getValue(I.getOperand(0));
+  setValue(&I, N);
+}

lib/CodeGen/TargetLoweringBase.cpp

#include "llvm/IR/Instruction.def"
   case UserOp2:        return 0;
   case VAArg:          return 0;
   case ExtractElement: return ISD::EXTRACT_VECTOR_ELT;
   case InsertElement:  return ISD::INSERT_VECTOR_ELT;
   case ShuffleVector:  return ISD::VECTOR_SHUFFLE;
   case ExtractValue:   return ISD::MERGE_VALUES;
   case InsertValue:    return ISD::MERGE_VALUES;
   case LandingPad:     return 0;
+  case Freeze:         return 0;
   }
 
   llvm_unreachable("Unknown instruction type encountered!");
 }
 
 std::pair<int, MVT>
 TargetLoweringBase::getTypeLegalizationCost(const DataLayout &DL,
                                             Type *Ty) const {

lib/IR/Core.cpp

 
 LLVMValueRef LLVMBuildInsertValue(LLVMBuilderRef B, LLVMValueRef AggVal,
                                   LLVMValueRef EltVal, unsigned Index,
                                   const char *Name) {
   return wrap(unwrap(B)->CreateInsertValue(unwrap(AggVal), unwrap(EltVal),
                                            Index, Name));
 }
 
+LLVMValueRef LLVMBuildFreeze(LLVMBuilderRef B, LLVMValueRef Val,
+                             const char *Name) {
+  return wrap(unwrap(B)->CreateFreeze(unwrap(Val), Name));
+}
+
 LLVMValueRef LLVMBuildIsNull(LLVMBuilderRef B, LLVMValueRef Val,
                              const char *Name) {
   return wrap(unwrap(B)->CreateIsNull(unwrap(Val), Name));
 }
 
 LLVMValueRef LLVMBuildIsNotNull(LLVMBuilderRef B, LLVMValueRef Val,
                                 const char *Name) {
   return wrap(unwrap(B)->CreateIsNotNull(unwrap(Val), Name));

lib/IR/Instruction.cpp

const char *Instruction::getOpcodeName(unsigned OpCode) {
   case VAArg:          return "va_arg";
   case ExtractElement: return "extractelement";
   case InsertElement:  return "insertelement";
   case ShuffleVector:  return "shufflevector";
   case ExtractValue:   return "extractvalue";
   case InsertValue:    return "insertvalue";
   case LandingPad:     return "landingpad";
   case CleanupPad:     return "cleanuppad";
+  case Freeze:         return "freeze";
 
   default: return "<Invalid operator> ";
   }
 }
 
 /// Return true if both instructions have the same special state. This must be
 /// kept in sync with FunctionComparator::cmpOperations in
 /// lib/Transforms/IPO/MergeFunctions.cpp.

lib/IR/Instructions.cpp

Type *ExtractValueInst::getIndexedType(Type *Agg,
   }
   return const_cast<Type*>(Agg);
 }
 
 //===----------------------------------------------------------------------===//
 //                             BinaryOperator Class
 //===----------------------------------------------------------------------===//
 
 BinaryOperator::BinaryOperator(BinaryOps iType, Value *S1, Value *S2,
                                Type *Ty, const Twine &Name,
                                Instruction *InsertBefore)
   : Instruction(Ty, iType,
                 OperandTraits<BinaryOperator>::op_begin(this),
                 OperandTraits<BinaryOperator>::operands(this),
                 InsertBefore) {
   Op<0>() = S1;
   Op<1>() = S2;
   init(iType);
   setName(Name);

 unsigned IndirectBrInst::getNumSuccessorsV() const {
   return getNumSuccessors();
 }
 void IndirectBrInst::setSuccessorV(unsigned idx, BasicBlock *B) {
   setSuccessor(idx, B);
 }
 
 //===----------------------------------------------------------------------===//
+//                            FreezeInst Implementation
+//===----------------------------------------------------------------------===//
+
+FreezeInst::FreezeInst(Value *S,
+                       const Twine &Name, Instruction *InsertBefore)
+    : UnaryInstruction(S->getType(), Freeze, S, InsertBefore) {
+  setName(Name);
+}
+
+FreezeInst::FreezeInst(Value *S,
+                       const Twine &Name, BasicBlock *InsertAtEnd)
+    : UnaryInstruction(S->getType(), Freeze, S, InsertAtEnd) {
+  setName(Name);
+}
+
+//===----------------------------------------------------------------------===//
 //                           cloneImpl() implementations
 //===----------------------------------------------------------------------===//
 
 // Define these methods here so vtables don't get emitted into every translation
 // unit that uses these classes.
 
 GetElementPtrInst *GetElementPtrInst::cloneImpl() const {
   return new (getNumOperands()) GetElementPtrInst(*this);

 FuncletPadInst *FuncletPadInst::cloneImpl() const {
   return new (getNumOperands()) FuncletPadInst(*this);
 }
 
 UnreachableInst *UnreachableInst::cloneImpl() const {
   LLVMContext &Context = getContext();
   return new UnreachableInst(Context);
 }
+
+FreezeInst *FreezeInst::cloneImpl() const {
+  return new FreezeInst(getOperand(0));
+}

lib/IR/Verifier.cpp

#include "llvm/IR/Metadata.def"
   void visitLandingPadInst(LandingPadInst &LPI);
   void visitResumeInst(ResumeInst &RI);
   void visitCatchPadInst(CatchPadInst &CPI);
   void visitCatchReturnInst(CatchReturnInst &CatchReturn);
   void visitCleanupPadInst(CleanupPadInst &CPI);
   void visitFuncletPadInst(FuncletPadInst &FPI);
   void visitCatchSwitchInst(CatchSwitchInst &CatchSwitch);
   void visitCleanupReturnInst(CleanupReturnInst &CRI);
+  void visitFreezeInst(FreezeInst &FI);
 
   void verifyCallSite(CallSite CS);
   void verifySwiftErrorCallSite(CallSite CS, const Value *SwiftErrorVal);
   void verifySwiftErrorValue(const Value *SwiftErrorVal);
   void verifyMustTailCall(CallInst &CI);
   bool performTypeCheck(Intrinsic::ID ID, Function *F, Type *Ty, int VT,
                         unsigned ArgNo, std::string &Suffix);
   bool verifyAttributeCount(AttributeSet Attrs, unsigned Params);
void Verifier::visitInvokeInst(InvokeInst &II) {
 
   // Verify that the first non-PHI instruction of the unwind destination is an
   // exception handling instruction.
   Assert(
       II.getUnwindDest()->isEHPad(),
       "The unwind destination does not have an exception handling instruction!",
       &II);
 
   visitTerminatorInst(II);
 }
 
 /// visitBinaryOperator - Check that both arguments to the binary operator are
 /// of the same type!
 ///
 void Verifier::visitBinaryOperator(BinaryOperator &B) {
   Assert(B.getOperand(0)->getType() == B.getOperand(1)->getType(),
          "Both operands to a binary operator are not of the same type!", &B);
 
   switch (B.getOpcode()) {
   // Check that integer arithmetic operators are only used with
Assert(I->isEHPad() && !isa<LandingPadInst>(I),
            "CleanupReturnInst must unwind to an EH block which is not a "
            "landingpad.",
            &CRI);
   }
 
   visitTerminatorInst(CRI);
 }
 
+void Verifier::visitFreezeInst(FreezeInst &FI) {
+  Assert(FI.getOperand(0)->getType()->isIntegerTy(),
+         "Cannot freeze non-integer type!", &FI);
+
+  visitInstruction(FI);
+}
+
 void Verifier::verifyDominatesUse(Instruction &I, unsigned i) {
   Instruction *Op = cast<Instruction>(I.getOperand(i));
   // If the we have an invalid invoke, don't try to compute the dominance.
   // We already reject it in the invoke specific checks and the dominance
   // computation doesn't handle multiple edges.
   if (InvokeInst *II = dyn_cast<InvokeInst>(Op)) {
     if (II->getNormalDest() == II->getUnwindDest())
       return;

test/Bindings/llvm-c/freeze.ll

+; RUN: llvm-as < %s | llvm-dis > %t.orig
+; RUN: llvm-as < %s | llvm-c-test --echo > %t.echo
+; RUN: diff -w %t.orig %t.echo
+
+define i32 @f(i32 %arg) {
+  %1 = freeze i32 %arg
+  %2 = freeze i32 10
+  %3 = freeze i32 %1
+  ret i32 %1
+}

test/Bitcode/compatibility.ll

exit:
   call void @f.nobuiltin() builtin
   ; CHECK: call void @f.nobuiltin() #40
 
   call fastcc noalias i32* @f.noalias() noinline
   ; CHECK: call fastcc noalias i32* @f.noalias() #12
   tail call ghccc nonnull i32* @f.nonnull() minsize
   ; CHECK: tail call ghccc nonnull i32* @f.nonnull() #7
 
+  freeze i32 %op1
+  ; CHECK: freeze i32 %op1
+  freeze i32 10
+  ; CHECK: freeze i32 10
+
   ret void
 }
 
 define void @instructions.call_musttail(i8* inalloca %val) {
   musttail call void @f.param.inalloca(i8* inalloca %val)
   ; CHECK: musttail call void @f.param.inalloca(i8* inalloca %val)
 
   ret void

test/Bitcode/freeze-float.ll

+; RUN: not llvm-as %s -o /dev/null 2>&1 | FileCheck %s
+
+; CHECK: freeze-float.ll:[[@LINE+2]]:15: error: cannot freeze non-integer type
+define float @float(float %x) {
+  %y = freeze float %x
+  ret float %y
+}

test/Bitcode/freeze-pointer.ll

+; RUN: not llvm-as %s -o /dev/null 2>&1 | FileCheck %s
+
+; CHECK: freeze-pointer.ll:[[@LINE+2]]:15: error: cannot freeze non-integer type
+define i32* @ptr(i32* %x) {
+  %y = freeze i32* %x
+  ret i32* %y
+}

test/Bitcode/freeze-vector.ll

+; RUN: not llvm-as %s -o /dev/null 2>&1 | FileCheck %s
+
+; CHECK: freeze-vector.ll:[[@LINE+2]]:15: error: cannot freeze non-integer type
+define <2 x i32> @vec(<2 x i32> %x) {
+  %y = freeze <2 x i32> %x
+  ret <2 x i32> %y
+}

tools/llvm-c-test/echo.cpp

switch(Op) {
         LLVMValueRef Agg = CloneValue(LLVMGetOperand(Src, 0));
         LLVMValueRef V = CloneValue(LLVMGetOperand(Src, 1));
         if (LLVMGetNumIndices(Src) != 1)
           report_fatal_error("Expected only one indice");
         auto I = LLVMGetIndices(Src)[0];
         Dst = LLVMBuildInsertValue(Builder, Agg, V, I, Name);
         break;
       }
+      case LLVMFreeze: {
+        LLVMValueRef Arg = CloneValue(LLVMGetOperand(Src, 0));
+        Dst = LLVMBuildFreeze(Builder, Arg, Name);
+        break;
+      }
       default:
         break;
     }
 
     if (Dst == nullptr) {
       fprintf(stderr, "%d is not a supported opcode\n", Op);
       exit(-1);
     }

unittests/IR/VerifierTest.cpp

TEST(VerifierTest, Branch_i1) {
   // ... then use setOperand to redirect it to a value of different type.
 
   Constant *Zero32 = ConstantInt::get(IntegerType::get(C, 32), 0);
   BI->setOperand(0, Zero32);
 
   EXPECT_TRUE(verifyFunction(*F));
 }
 
+TEST(VerifierTest, Freeze) {
+  LLVMContext C;
+  Module M("M", C);
+  FunctionType *FTy = FunctionType::get(Type::getVoidTy(C), /*isVarArg=*/false);
+  Function *F = cast<Function>(M.getOrInsertFunction("foo", FTy));
+  BasicBlock *Entry = BasicBlock::Create(C, "entry", F);
+  ReturnInst *RI = ReturnInst::Create(C, Entry);
+
+  IntegerType *ITy = IntegerType::get(C, 32);
+  ConstantInt *CI = ConstantInt::get(ITy, 0);
+
+  // Invalid type : freeze(<2 x i32>)
+  Constant *CV = ConstantVector::getSplat(2, CI);
+  FreezeInst *FI_vec = new FreezeInst(CV);
+  FI_vec->insertBefore(RI);
+
+  EXPECT_TRUE(verifyFunction(*F));
+
+  FI_vec->eraseFromParent();
+
+  // Invalid type : freeze(float)
+  Constant *CFP = ConstantFP::get(Type::getDoubleTy(C), 0.0);
+  FreezeInst *FI_dbl = new FreezeInst(CFP);
+  FI_dbl->insertBefore(RI);
+
+  EXPECT_TRUE(verifyFunction(*F));
+
+  FI_dbl->eraseFromParent();
+
+  // Invalid type : freeze(i32*)
+  PointerType *PT = PointerType::get(ITy, 0);
+  ConstantPointerNull *CPN = ConstantPointerNull::get(PT);
+  FreezeInst *FI_ptr = new FreezeInst(CPN);
+  FI_ptr->insertBefore(RI);
+
+  EXPECT_TRUE(verifyFunction(*F));
+
+  FI_ptr->eraseFromParent();
+
+  // Valid type : freeze(int)
+  FreezeInst *FI = new FreezeInst(CI);
+  FI->insertBefore(RI);
+
+  EXPECT_FALSE(verifyFunction(*F));
+
+  FI_ptr->eraseFromParent();
+}
+
 TEST(VerifierTest, InvalidRetAttribute) {
   LLVMContext C;
   Module M("M", C);
   FunctionType *FTy = FunctionType::get(Type::getInt32Ty(C), /*isVarArg=*/false);
   Function *F = cast<Function>(M.getOrInsertFunction("foo", FTy));
   AttributeSet AS = F->getAttributes();
   F->setAttributes(AS.addAttribute(C, AttributeSet::ReturnIndex,
                                    Attribute::UWTable));