[CodeGen] Generate follow-up metadata for loops with more than one transformation.

[Meinersbur]

Summary

Before this patch, CGLoop would dump all transformations for a loop into a single LoopID without encoding any order in which to apply them. rL348944 added the possibility to encode a transformation order using followup-attributes.

When a loop has more than one transformation, use the follow-up attribute define the order in which they are applied. The emitted order is the defacto order as defined by the current LLVM pass pipeline, which is:

1. LoopFullUnrollPass
2. LoopDistributePass
3. LoopVectorizePass
4. LoopUnrollAndJamPass
5. LoopUnrollPass
6. MachinePipeliner

This patch should therefore not change the assembly output, assuming that all explicit transformations can be applied, and no implicit transformations in-between. In the former case, WarnMissedTransformationsPass should emit a warning (except for MachinePipeliner which is not implemented yet). The latter could be avoided by adding 'llvm.loop.disable_nonforced' attributes.

Because LoopUnrollAndJamPass processes a loop nest, generation of the MDNode is delayed to after the inner loop metadata have been processed. A temporary LoopID is therefore used to annotate instructions and RAUW'ed by the actual LoopID later.

[rjmccall]

Sorry, this is deep in the loop analysis code, and I don't remember anything about this stuff, so I'm not sure I'm going to be a useful reviewer.

[Anastasia]

This seems like an interesting change. So far in OpenCL we have actively used only unroll. I will let others to assess in more details.

lib/CodeGen/CGLoopInfo.cpp

107 ↗

(On Diff #186043)

Will this end up emitted in the final module?

[Meinersbur]

lib/CodeGen/CGLoopInfo.cpp

107 ↗

(On Diff #186043)

This MDNode is emitted in the module generated by clang -- as a follow-up attribute -- in case there is another transformation after partial unrolling (currently just pipelining).

[Meinersbur]

ping

The Polly-powered additional transformations now also generate this kind of metadata.

[Meinersbur]

ping

[hsaito]

In D57978#1435473, @Meinersbur wrote:

ping

This is a good follow up for rL348944, but I have no familiarity to the code under Clang. So, let me thank you for doing this work and move myself to the subscriber.

[dmgreen]

Hello. I also don't feel very familiar with clang, but had a poke around and I think it looks pretty good. I see unroll and jam is being awkward again.

This could maybe do with a few extra tests. Am I correct in saying something like this:

#pragma unroll_and_jam(4)
for(int j = 0; j < n; j++) {
  #pragma unroll(4)
  for(int k = 0; k < n; k++) {
    x[j*n+k] = 10;
  }
}

would end up with a llvm.loop.unroll_and_jam.followup_inner with a llvm.loop.unroll_count?

lib/CodeGen/CGLoopInfo.cpp

500 ↗

(On Diff #186043)

I was having trouble parsing this sentance. Does it mean both the inner loop and the outer loop both have unroll-and-jam? UnrollAndJam processes loops from inner to outer, so if this is working as I think, maybe it should be put into BeforeJam.

502 ↗

(On Diff #186043)

= Attrs.UnrollAndJamEnable ?

lib/CodeGen/CGLoopInfo.h

135 ↗

(On Diff #186043)

*encodes

170 ↗

(On Diff #186043)

*encodes

[Meinersbur]

• Rebase
• Add two test cases for all attributes combined (as inner and outer loop for of an unroll-and-jam)
• Of two nested unroll-and-jams, apply the inner first
• Typos and comment clarification
• Add missing llvm.loop.isvectorized when attributes are split between BeforeJam and AfterJam

[Meinersbur]

In D57978#1440894, @dmgreen wrote:

This could maybe do with a few extra tests. Am I correct in saying something like this:

#pragma unroll_and_jam(4)
for(int j = 0; j < n; j++) {
  #pragma unroll(4)
  for(int k = 0; k < n; k++) {
    x[j*n+k] = 10;
  }
}

would end up with a llvm.loop.unroll_and_jam.followup_inner with a llvm.loop.unroll_count?

correct. I added a new test pragma-followup_inner.cpp where this can be seen.

lib/CodeGen/CGLoopInfo.cpp

500 ↗

(On Diff #186043)

Does it mean both the inner loop and the outer loop both have unroll-and-jam?

Correct. I was thinking of such a case:

#pragma unroll_and_jam(2)
for (int i = 0; i < m; i+=1)
  #pragma unroll_and_jam(3)
  for (int j = 0; j < n; j+=1)

Assuming the inner unroll-and-jam would do something, would it happen before or after the outer transformation. As you mentioned, there is an argument to put it into BeforeJam. Changed.

502 ↗

(On Diff #186043)

thanks

[dmgreen]

Yeah, OK. This looks like a good patch to me. As I said, I'm not a clang expert, but the code looks sensible enough. (Perhaps wait a couple of days in case others have objections.)

[Meinersbur]

@dmgreen Thank you for the review!

cfe/trunk/lib/CodeGen/CGLoopInfo.h

struct LoopAttributes {
   unsigned PipelineInitiationInterval;
 };
 
 /// Information used when generating a structured loop.
 class LoopInfo {
 public:
   /// Construct a new LoopInfo for the loop with entry Header.
   LoopInfo(llvm::BasicBlock *Header, const LoopAttributes &Attrs,
-           const llvm::DebugLoc &StartLoc, const llvm::DebugLoc &EndLoc);
+           const llvm::DebugLoc &StartLoc, const llvm::DebugLoc &EndLoc,
+           LoopInfo *Parent);
 
   /// Get the loop id metadata for this loop.
-  llvm::MDNode *getLoopID() const { return LoopID; }
+  llvm::MDNode *getLoopID() const { return TempLoopID.get(); }
 
   /// Get the header block of this loop.
   llvm::BasicBlock *getHeader() const { return Header; }
 
   /// Get the set of attributes active for this loop.
   const LoopAttributes &getAttributes() const { return Attrs; }
 
   /// Return this loop's access group or nullptr if it does not have one.
   llvm::MDNode *getAccessGroup() const { return AccGroup; }
 
+  /// Create the loop's metadata. Must be called after its nested loops have
+  /// been processed.
+  void finish();
+
 private:
   /// Loop ID metadata.
-  llvm::MDNode *LoopID;
+  llvm::TempMDTuple TempLoopID;
   /// Header block of this loop.
   llvm::BasicBlock *Header;
   /// The attributes for this loop.
   LoopAttributes Attrs;
   /// The access group for memory accesses parallel to this loop.
   llvm::MDNode *AccGroup = nullptr;
+  /// Start location of this loop.
+  llvm::DebugLoc StartLoc;
+  /// End location of this loop.
+  llvm::DebugLoc EndLoc;
+  /// The next outer loop, or nullptr if this is the outermost loop.
+  LoopInfo *Parent;
+  /// If this loop has unroll-and-jam metadata, this can be set by the inner
+  /// loop's LoopInfo to set the llvm.loop.unroll_and_jam.followup_inner
+  /// metadata.
+  llvm::MDNode *UnrollAndJamInnerFollowup = nullptr;
+
+  /// Create a LoopID without any transformations.
+  llvm::MDNode *
+  createLoopPropertiesMetadata(llvm::ArrayRef<llvm::Metadata *> LoopProperties);
+
+  /// Create a LoopID for transformations.
+  ///
+  /// The methods call each other in case multiple transformations are applied
+  /// to a loop. The transformation first to be applied will use LoopID of the
+  /// next transformation in its followup attribute.
+  ///
+  /// @param Attrs             The loop's transformations.
+  /// @param LoopProperties    Non-transformation properties such as debug
+  ///                          location, parallel accesses and disabled
+  ///                          transformations. These are added to the returned
+  ///                          LoopID.
+  /// @param HasUserTransforms [out] Set to true if the returned MDNode encodes
+  ///                          at least one transformation.
+  ///
+  /// @return A LoopID (metadata node) that can be used for the llvm.loop
+  ///         annotation or followup-attribute.
+  /// @{
+  llvm::MDNode *
+  createPipeliningMetadata(const LoopAttributes &Attrs,
+                           llvm::ArrayRef<llvm::Metadata *> LoopProperties,
+                           bool &HasUserTransforms);
+  llvm::MDNode *
+  createPartialUnrollMetadata(const LoopAttributes &Attrs,
+                              llvm::ArrayRef<llvm::Metadata *> LoopProperties,
+                              bool &HasUserTransforms);
+  llvm::MDNode *
+  createUnrollAndJamMetadata(const LoopAttributes &Attrs,
+                             llvm::ArrayRef<llvm::Metadata *> LoopProperties,
+                             bool &HasUserTransforms);
+  llvm::MDNode *
+  createLoopVectorizeMetadata(const LoopAttributes &Attrs,
+                              llvm::ArrayRef<llvm::Metadata *> LoopProperties,
+                              bool &HasUserTransforms);
+  llvm::MDNode *
+  createLoopDistributeMetadata(const LoopAttributes &Attrs,
+                               llvm::ArrayRef<llvm::Metadata *> LoopProperties,
+                               bool &HasUserTransforms);
+  llvm::MDNode *
+  createFullUnrollMetadata(const LoopAttributes &Attrs,
+                           llvm::ArrayRef<llvm::Metadata *> LoopProperties,
+                           bool &HasUserTransforms);
+  /// @}
+
+  /// Create a LoopID for this loop, including transformation-unspecific
+  /// metadata such as debug location.
+  ///
+  /// @param Attrs             This loop's attributes and transformations.
+  /// @param LoopProperties    Additional non-transformation properties to add
+  ///                          to the LoopID, such as transformation-specific
+  ///                          metadata that are not covered by @p Attrs.
+  /// @param HasUserTransforms [out] Set to true if the returned MDNode encodes
+  ///                          at least one transformation.
+  ///
+  /// @return A LoopID (metadata node) that can be used for the llvm.loop
+  ///         annotation.
+  llvm::MDNode *createMetadata(const LoopAttributes &Attrs,
+                               llvm::ArrayRef<llvm::Metadata *> LoopProperties,
+                               bool &HasUserTransforms);
 };
 
 /// A stack of loop information corresponding to loop nesting levels.
 /// This stack can be used to prepare attributes which are applied when a loop
 /// is emitted.
 class LoopInfoStack {
   LoopInfoStack(const LoopInfoStack &) = delete;
   void operator=(const LoopInfoStack &) = delete;

cfe/trunk/lib/CodeGen/CGLoopInfo.cpp

 #include "llvm/IR/CFG.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/InstrTypes.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/Metadata.h"
 using namespace clang::CodeGen;
 using namespace llvm;
 
-static MDNode *createMetadata(LLVMContext &Ctx, const LoopAttributes &Attrs,
-                              const llvm::DebugLoc &StartLoc,
-                              const llvm::DebugLoc &EndLoc, MDNode *&AccGroup) {
+MDNode *
+LoopInfo::createLoopPropertiesMetadata(ArrayRef<Metadata *> LoopProperties) {
+  LLVMContext &Ctx = Header->getContext();
+  SmallVector<Metadata *, 4> NewLoopProperties;
+  TempMDTuple TempNode = MDNode::getTemporary(Ctx, None);
+  NewLoopProperties.push_back(TempNode.get());
+  NewLoopProperties.append(LoopProperties.begin(), LoopProperties.end());
 
-  if (!Attrs.IsParallel && Attrs.VectorizeWidth == 0 &&
-      Attrs.InterleaveCount == 0 && Attrs.UnrollCount == 0 &&
-      Attrs.UnrollAndJamCount == 0 && !Attrs.PipelineDisabled &&
-      Attrs.PipelineInitiationInterval == 0 &&
-      Attrs.VectorizeEnable == LoopAttributes::Unspecified &&
-      Attrs.UnrollEnable == LoopAttributes::Unspecified &&
-      Attrs.UnrollAndJamEnable == LoopAttributes::Unspecified &&
-      Attrs.DistributeEnable == LoopAttributes::Unspecified && !StartLoc &&
-      !EndLoc)
-    return nullptr;
+  MDNode *LoopID = MDNode::getDistinct(Ctx, NewLoopProperties);
+  LoopID->replaceOperandWith(0, LoopID);
+  return LoopID;
+}
+
+MDNode *LoopInfo::createPipeliningMetadata(const LoopAttributes &Attrs,
+                                           ArrayRef<Metadata *> LoopProperties,
+                                           bool &HasUserTransforms) {
+  LLVMContext &Ctx = Header->getContext();
+
+  Optional<bool> Enabled;
+  if (Attrs.PipelineDisabled)
+    Enabled = false;
+  else if (Attrs.PipelineInitiationInterval != 0)
+    Enabled = true;
+
+  if (Enabled != true) {
+    SmallVector<Metadata *, 4> NewLoopProperties;
+    if (Enabled == false) {
+      NewLoopProperties.append(LoopProperties.begin(), LoopProperties.end());
+      NewLoopProperties.push_back(
+          MDNode::get(Ctx, {MDString::get(Ctx, "llvm.loop.pipeline.disable"),
+                            ConstantAsMetadata::get(ConstantInt::get(
+                                llvm::Type::getInt1Ty(Ctx), 1))}));
+      LoopProperties = NewLoopProperties;
+    }
+    return createLoopPropertiesMetadata(LoopProperties);
+  }
 
   SmallVector<Metadata *, 4> Args;
-  // Reserve operand 0 for loop id self reference.
-  auto TempNode = MDNode::getTemporary(Ctx, None);
+  TempMDTuple TempNode = MDNode::getTemporary(Ctx, None);
   Args.push_back(TempNode.get());
+  Args.append(LoopProperties.begin(), LoopProperties.end());
 
-  // If we have a valid start debug location for the loop, add it.
-  if (StartLoc) {
-    Args.push_back(StartLoc.getAsMDNode());
-
-    // If we also have a valid end debug location for the loop, add it.
-    if (EndLoc)
-      Args.push_back(EndLoc.getAsMDNode());
-  }
-
-  // Setting vectorize.width
-  if (Attrs.VectorizeWidth > 0) {
-    Metadata *Vals[] = {MDString::get(Ctx, "llvm.loop.vectorize.width"),
-                        ConstantAsMetadata::get(ConstantInt::get(
-                            Type::getInt32Ty(Ctx), Attrs.VectorizeWidth))};
+  if (Attrs.PipelineInitiationInterval > 0) {
+    Metadata *Vals[] = {
+        MDString::get(Ctx, "llvm.loop.pipeline.initiationinterval"),
+            llvm::Type::getInt32Ty(Ctx), Attrs.PipelineInitiationInterval))};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
-  // Setting interleave.count
-  if (Attrs.InterleaveCount > 0) {
-    Metadata *Vals[] = {MDString::get(Ctx, "llvm.loop.interleave.count"),
-                        ConstantAsMetadata::get(ConstantInt::get(
-                            Type::getInt32Ty(Ctx), Attrs.InterleaveCount))};
-    Args.push_back(MDNode::get(Ctx, Vals));
-  }
+  // No follow-up: This is the last transformation.
+
+  MDNode *LoopID = MDNode::getDistinct(Ctx, Args);
+  LoopID->replaceOperandWith(0, LoopID);
+  HasUserTransforms = true;
+  return LoopID;
 
+MDNode *
+LoopInfo::createPartialUnrollMetadata(const LoopAttributes &Attrs,
+                                      ArrayRef<Metadata *> LoopProperties,
+                                      bool &HasUserTransforms) {
+  LLVMContext &Ctx = Header->getContext();
+
+  Optional<bool> Enabled;
+  if (Attrs.UnrollEnable == LoopAttributes::Disable)
+    Enabled = false;
+  else if (Attrs.UnrollEnable == LoopAttributes::Full)
+    Enabled = None;
+  else if (Attrs.UnrollEnable != LoopAttributes::Unspecified ||
+           Attrs.UnrollCount != 0)
+    Enabled = true;
+
+  if (Enabled != true) {
+    // createFullUnrollMetadata will already have added llvm.loop.unroll.disable
+    // if unrolling is disabled.
+    return createPipeliningMetadata(Attrs, LoopProperties, HasUserTransforms);
+  }
+
+  SmallVector<Metadata *, 4> FollowupLoopProperties;
+
+  // Apply all loop properties to the unrolled loop.
+  FollowupLoopProperties.append(LoopProperties.begin(), LoopProperties.end());
+
+  // Don't unroll an already unrolled loop.
+  FollowupLoopProperties.push_back(
+      MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.unroll.disable")));
+
+  bool FollowupHasTransforms = false;
+  MDNode *Followup = createPipeliningMetadata(Attrs, FollowupLoopProperties,
+                                              FollowupHasTransforms);
+
+  SmallVector<Metadata *, 4> Args;
+  TempMDTuple TempNode = MDNode::getTemporary(Ctx, None);
+  Args.push_back(TempNode.get());
+  Args.append(LoopProperties.begin(), LoopProperties.end());
+
   // Setting unroll.count
   if (Attrs.UnrollCount > 0) {
     Metadata *Vals[] = {MDString::get(Ctx, "llvm.loop.unroll.count"),
                         ConstantAsMetadata::get(ConstantInt::get(
-                            Type::getInt32Ty(Ctx), Attrs.UnrollCount))};
+                            llvm::Type::getInt32Ty(Ctx), Attrs.UnrollCount))};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
+  // Setting unroll.full or unroll.disable
+  if (Attrs.UnrollEnable == LoopAttributes::Enable) {
+    Metadata *Vals[] = {MDString::get(Ctx, "llvm.loop.unroll.enable")};
+    Args.push_back(MDNode::get(Ctx, Vals));
+  }
+
+  if (FollowupHasTransforms)
+    Args.push_back(MDNode::get(
+        Ctx, {MDString::get(Ctx, "llvm.loop.unroll.followup_all"), Followup}));
+
+  MDNode *LoopID = MDNode::getDistinct(Ctx, Args);
+  LoopID->replaceOperandWith(0, LoopID);
+  HasUserTransforms = true;
+  return LoopID;
+}
+
+MDNode *
+LoopInfo::createUnrollAndJamMetadata(const LoopAttributes &Attrs,
+                                     ArrayRef<Metadata *> LoopProperties,
+                                     bool &HasUserTransforms) {
+  LLVMContext &Ctx = Header->getContext();
+
+  Optional<bool> Enabled;
+  if (Attrs.UnrollAndJamEnable == LoopAttributes::Disable)
+    Enabled = false;
+  else if (Attrs.UnrollAndJamEnable == LoopAttributes::Enable ||
+           Attrs.UnrollAndJamCount != 0)
+    Enabled = true;
+
+  if (Enabled != true) {
+    SmallVector<Metadata *, 4> NewLoopProperties;
+    if (Enabled == false) {
+      NewLoopProperties.append(LoopProperties.begin(), LoopProperties.end());
+      NewLoopProperties.push_back(MDNode::get(
+          Ctx, MDString::get(Ctx, "llvm.loop.unroll_and_jam.disable")));
+      LoopProperties = NewLoopProperties;
+    }
+    return createPartialUnrollMetadata(Attrs, LoopProperties,
+                                       HasUserTransforms);
+  }
+
+  SmallVector<Metadata *, 4> FollowupLoopProperties;
+  FollowupLoopProperties.append(LoopProperties.begin(), LoopProperties.end());
+  FollowupLoopProperties.push_back(
+      MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.unroll_and_jam.disable")));
+
+  bool FollowupHasTransforms = false;
+  MDNode *Followup = createPartialUnrollMetadata(Attrs, FollowupLoopProperties,
+                                                 FollowupHasTransforms);
+
+  SmallVector<Metadata *, 4> Args;
+  TempMDTuple TempNode = MDNode::getTemporary(Ctx, None);
+  Args.push_back(TempNode.get());
+  Args.append(LoopProperties.begin(), LoopProperties.end());
+
   // Setting unroll_and_jam.count
   if (Attrs.UnrollAndJamCount > 0) {
-    Metadata *Vals[] = {MDString::get(Ctx, "llvm.loop.unroll_and_jam.count"),
-                        ConstantAsMetadata::get(ConstantInt::get(
-                            Type::getInt32Ty(Ctx), Attrs.UnrollAndJamCount))};
+    Metadata *Vals[] = {
+        MDString::get(Ctx, "llvm.loop.unroll_and_jam.count"),
+        ConstantAsMetadata::get(ConstantInt::get(llvm::Type::getInt32Ty(Ctx),
+                                                 Attrs.UnrollAndJamCount))};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
-  // Setting vectorize.enable
-  if (Attrs.VectorizeEnable != LoopAttributes::Unspecified) {
-    Metadata *Vals[] = {MDString::get(Ctx, "llvm.loop.vectorize.enable"),
-                        ConstantAsMetadata::get(ConstantInt::get(
-                            Type::getInt1Ty(Ctx), (Attrs.VectorizeEnable ==
-                                                   LoopAttributes::Enable)))};
+  if (Attrs.UnrollAndJamEnable == LoopAttributes::Enable) {
+    Metadata *Vals[] = {MDString::get(Ctx, "llvm.loop.unroll_and_jam.enable")};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
-  // Setting unroll.full or unroll.disable
-  if (Attrs.UnrollEnable != LoopAttributes::Unspecified) {
-    std::string Name;
-    if (Attrs.UnrollEnable == LoopAttributes::Enable)
-      Name = "llvm.loop.unroll.enable";
-    else if (Attrs.UnrollEnable == LoopAttributes::Full)
-      Name = "llvm.loop.unroll.full";
-    else
-      Name = "llvm.loop.unroll.disable";
-    Metadata *Vals[] = {MDString::get(Ctx, Name)};
+  if (FollowupHasTransforms)
+    Args.push_back(MDNode::get(
+        Ctx, {MDString::get(Ctx, "llvm.loop.unroll_and_jam.followup_outer"),
+              Followup}));
+
+  if (UnrollAndJamInnerFollowup)
+    Args.push_back(MDNode::get(
+        Ctx, {MDString::get(Ctx, "llvm.loop.unroll_and_jam.followup_inner"),
+              UnrollAndJamInnerFollowup}));
+
+  MDNode *LoopID = MDNode::getDistinct(Ctx, Args);
+  LoopID->replaceOperandWith(0, LoopID);
+  HasUserTransforms = true;
+  return LoopID;
+}
+
+MDNode *
+LoopInfo::createLoopVectorizeMetadata(const LoopAttributes &Attrs,
+                                      ArrayRef<Metadata *> LoopProperties,
+                                      bool &HasUserTransforms) {
+  LLVMContext &Ctx = Header->getContext();
+
+  Optional<bool> Enabled;
+  if (Attrs.VectorizeEnable == LoopAttributes::Disable)
+    Enabled = false;
+  else if (Attrs.VectorizeEnable != LoopAttributes::Unspecified ||
+           Attrs.InterleaveCount != 0 || Attrs.VectorizeWidth != 0)
+    Enabled = true;
+
+  if (Enabled != true) {
+    SmallVector<Metadata *, 4> NewLoopProperties;
+    if (Enabled == false) {
+      NewLoopProperties.append(LoopProperties.begin(), LoopProperties.end());
+      NewLoopProperties.push_back(
+          MDNode::get(Ctx, {MDString::get(Ctx, "llvm.loop.vectorize.enable"),
+                            ConstantAsMetadata::get(ConstantInt::get(
+                                llvm::Type::getInt1Ty(Ctx), 0))}));
+      LoopProperties = NewLoopProperties;
+    }
+    return createUnrollAndJamMetadata(Attrs, LoopProperties, HasUserTransforms);
+  }
+
+  // Apply all loop properties to the vectorized loop.
+  SmallVector<Metadata *, 4> FollowupLoopProperties;
+  FollowupLoopProperties.append(LoopProperties.begin(), LoopProperties.end());
+
+  // Don't vectorize an already vectorized loop.
+  FollowupLoopProperties.push_back(
+      MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.isvectorized")));
+
+  bool FollowupHasTransforms = false;
+  MDNode *Followup = createUnrollAndJamMetadata(Attrs, FollowupLoopProperties,
+                                                FollowupHasTransforms);
+
+  SmallVector<Metadata *, 4> Args;
+  TempMDTuple TempNode = MDNode::getTemporary(Ctx, None);
+  Args.push_back(TempNode.get());
+  Args.append(LoopProperties.begin(), LoopProperties.end());
+
+  // Setting vectorize.width
+  if (Attrs.VectorizeWidth > 0) {
+    Metadata *Vals[] = {
+        MDString::get(Ctx, "llvm.loop.vectorize.width"),
+        ConstantAsMetadata::get(ConstantInt::get(llvm::Type::getInt32Ty(Ctx),
+                                                 Attrs.VectorizeWidth))};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
-  // Setting unroll_and_jam.full or unroll_and_jam.disable
-  if (Attrs.UnrollAndJamEnable != LoopAttributes::Unspecified) {
-    std::string Name;
-    if (Attrs.UnrollAndJamEnable == LoopAttributes::Enable)
-      Name = "llvm.loop.unroll_and_jam.enable";
-    else if (Attrs.UnrollAndJamEnable == LoopAttributes::Full)
-      Name = "llvm.loop.unroll_and_jam.full";
-    else
-      Name = "llvm.loop.unroll_and_jam.disable";
-    Metadata *Vals[] = {MDString::get(Ctx, Name)};
+  // Setting interleave.count
+  if (Attrs.InterleaveCount > 0) {
+    Metadata *Vals[] = {
+        MDString::get(Ctx, "llvm.loop.interleave.count"),
+        ConstantAsMetadata::get(ConstantInt::get(llvm::Type::getInt32Ty(Ctx),
+                                                 Attrs.InterleaveCount))};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
-  if (Attrs.DistributeEnable != LoopAttributes::Unspecified) {
-    Metadata *Vals[] = {MDString::get(Ctx, "llvm.loop.distribute.enable"),
-                        ConstantAsMetadata::get(ConstantInt::get(
-                            Type::getInt1Ty(Ctx), (Attrs.DistributeEnable ==
-                                                   LoopAttributes::Enable)))};
+  // Setting vectorize.enable
+  if (Attrs.VectorizeEnable != LoopAttributes::Unspecified) {
+    Metadata *Vals[] = {
+        MDString::get(Ctx, "llvm.loop.vectorize.enable"),
+            llvm::Type::getInt1Ty(Ctx),
+            (Attrs.VectorizeEnable == LoopAttributes::Enable)))};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
-  if (Attrs.IsParallel) {
-    AccGroup = MDNode::getDistinct(Ctx, {});
+  if (FollowupHasTransforms)
     Args.push_back(MDNode::get(
-        Ctx, {MDString::get(Ctx, "llvm.loop.parallel_accesses"), AccGroup}));
-  }
+        Ctx,
+        {MDString::get(Ctx, "llvm.loop.vectorize.followup_all"), Followup}));
+
+  MDNode *LoopID = MDNode::get(Ctx, Args);
+  LoopID->replaceOperandWith(0, LoopID);
+  HasUserTransforms = true;
+  return LoopID;
 
-  if (Attrs.PipelineDisabled) {
-    Metadata *Vals[] = {
-        MDString::get(Ctx, "llvm.loop.pipeline.disable"),
+MDNode *
+LoopInfo::createLoopDistributeMetadata(const LoopAttributes &Attrs,
+                                       ArrayRef<Metadata *> LoopProperties,
+                                       bool &HasUserTransforms) {
+  LLVMContext &Ctx = Header->getContext();
+
+  Optional<bool> Enabled;
+  if (Attrs.DistributeEnable == LoopAttributes::Disable)
+    Enabled = false;
+  if (Attrs.DistributeEnable == LoopAttributes::Enable)
+    Enabled = true;
+
+  if (Enabled != true) {
+    SmallVector<Metadata *, 4> NewLoopProperties;
+    if (Enabled == false) {
+      NewLoopProperties.append(LoopProperties.begin(), LoopProperties.end());
+      NewLoopProperties.push_back(
+          MDNode::get(Ctx, {MDString::get(Ctx, "llvm.loop.distribute.enable"),
                             ConstantAsMetadata::get(ConstantInt::get(
-            Type::getInt1Ty(Ctx), (Attrs.PipelineDisabled == true)))};
-    Args.push_back(MDNode::get(Ctx, Vals));
+                                llvm::Type::getInt1Ty(Ctx), 0))}));
+      LoopProperties = NewLoopProperties;
+    }
+    return createLoopVectorizeMetadata(Attrs, LoopProperties,
+                                       HasUserTransforms);
   }
 
-  if (Attrs.PipelineInitiationInterval > 0) {
-    Metadata *Vals[] = {
-        MDString::get(Ctx, "llvm.loop.pipeline.initiationinterval"),
+  bool FollowupHasTransforms = false;
+  MDNode *Followup =
+      createLoopVectorizeMetadata(Attrs, LoopProperties, FollowupHasTransforms);
+
+  SmallVector<Metadata *, 4> Args;
+  TempMDTuple TempNode = MDNode::getTemporary(Ctx, None);
+  Args.push_back(TempNode.get());
+  Args.append(LoopProperties.begin(), LoopProperties.end());
+
+  Metadata *Vals[] = {MDString::get(Ctx, "llvm.loop.distribute.enable"),
                       ConstantAsMetadata::get(ConstantInt::get(
-            Type::getInt32Ty(Ctx), Attrs.PipelineInitiationInterval))};
-    Args.push_back(MDNode::get(Ctx, Vals));
-  }
+                          llvm::Type::getInt1Ty(Ctx),
+                          (Attrs.DistributeEnable == LoopAttributes::Enable)))};
 
-  // Set the first operand to itself.
+  if (FollowupHasTransforms)
+    Args.push_back(MDNode::get(
+        Ctx,
+        {MDString::get(Ctx, "llvm.loop.distribute.followup_all"), Followup}));
+
   MDNode *LoopID = MDNode::get(Ctx, Args);
   LoopID->replaceOperandWith(0, LoopID);
+  HasUserTransforms = true;
+  return LoopID;
+}
+
+MDNode *LoopInfo::createFullUnrollMetadata(const LoopAttributes &Attrs,
+                                           ArrayRef<Metadata *> LoopProperties,
+                                           bool &HasUserTransforms) {
+  LLVMContext &Ctx = Header->getContext();
+
+  Optional<bool> Enabled;
+  if (Attrs.UnrollEnable == LoopAttributes::Disable)
+    Enabled = false;
+  else if (Attrs.UnrollEnable == LoopAttributes::Full)
+    Enabled = true;
+
+  if (Enabled != true) {
+    SmallVector<Metadata *, 4> NewLoopProperties;
+    if (Enabled == false) {
+      NewLoopProperties.append(LoopProperties.begin(), LoopProperties.end());
+      NewLoopProperties.push_back(
+          MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.unroll.disable")));
+      LoopProperties = NewLoopProperties;
+    }
+    return createLoopDistributeMetadata(Attrs, LoopProperties,
+                                        HasUserTransforms);
+  }
+
+  SmallVector<Metadata *, 4> Args;
+  TempMDTuple TempNode = MDNode::getTemporary(Ctx, None);
+  Args.push_back(TempNode.get());
+  Args.append(LoopProperties.begin(), LoopProperties.end());
+  Args.push_back(MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.unroll.full")));
+
+  // No follow-up: there is no loop after full unrolling.
+  // TODO: Warn if there are transformations after full unrolling.
+
+  MDNode *LoopID = MDNode::getDistinct(Ctx, Args);
+  LoopID->replaceOperandWith(0, LoopID);
+  HasUserTransforms = true;
   return LoopID;
 }
 
+MDNode *LoopInfo::createMetadata(
+    const LoopAttributes &Attrs,
+    llvm::ArrayRef<llvm::Metadata *> AdditionalLoopProperties,
+    bool &HasUserTransforms) {
+  SmallVector<Metadata *, 3> LoopProperties;
+
+  // If we have a valid start debug location for the loop, add it.
+  if (StartLoc) {
+    LoopProperties.push_back(StartLoc.getAsMDNode());
+
+    // If we also have a valid end debug location for the loop, add it.
+    if (EndLoc)
+      LoopProperties.push_back(EndLoc.getAsMDNode());
+  }
+
+  assert(!!AccGroup == Attrs.IsParallel &&
+         "There must be an access group iff the loop is parallel");
+  if (Attrs.IsParallel) {
+    LLVMContext &Ctx = Header->getContext();
+    LoopProperties.push_back(MDNode::get(
+        Ctx, {MDString::get(Ctx, "llvm.loop.parallel_accesses"), AccGroup}));
+  }
+
+  LoopProperties.insert(LoopProperties.end(), AdditionalLoopProperties.begin(),
+                        AdditionalLoopProperties.end());
+  return createFullUnrollMetadata(Attrs, LoopProperties, HasUserTransforms);
+}
+
 LoopAttributes::LoopAttributes(bool IsParallel)
     : IsParallel(IsParallel), VectorizeEnable(LoopAttributes::Unspecified),
       UnrollEnable(LoopAttributes::Unspecified),
       UnrollAndJamEnable(LoopAttributes::Unspecified), VectorizeWidth(0),
       InterleaveCount(0), UnrollCount(0), UnrollAndJamCount(0),
       DistributeEnable(LoopAttributes::Unspecified), PipelineDisabled(false),
       PipelineInitiationInterval(0) {}
 
 void LoopAttributes::clear() {
   IsParallel = false;
   VectorizeWidth = 0;
   InterleaveCount = 0;
   UnrollCount = 0;
   UnrollAndJamCount = 0;
   VectorizeEnable = LoopAttributes::Unspecified;
   UnrollEnable = LoopAttributes::Unspecified;
   UnrollAndJamEnable = LoopAttributes::Unspecified;
   DistributeEnable = LoopAttributes::Unspecified;
   PipelineDisabled = false;
   PipelineInitiationInterval = 0;
 }
 
 LoopInfo::LoopInfo(BasicBlock *Header, const LoopAttributes &Attrs,
-                   const llvm::DebugLoc &StartLoc, const llvm::DebugLoc &EndLoc)
-    : LoopID(nullptr), Header(Header), Attrs(Attrs) {
-  LoopID =
-      createMetadata(Header->getContext(), Attrs, StartLoc, EndLoc, AccGroup);
+                   const llvm::DebugLoc &StartLoc, const llvm::DebugLoc &EndLoc,
+                   LoopInfo *Parent)
+    : Header(Header), Attrs(Attrs), StartLoc(StartLoc), EndLoc(EndLoc),
+      Parent(Parent) {
+
+  if (Attrs.IsParallel) {
+    // Create an access group for this loop.
+    LLVMContext &Ctx = Header->getContext();
+    AccGroup = MDNode::getDistinct(Ctx, {});
+  }
+
+  if (!Attrs.IsParallel && Attrs.VectorizeWidth == 0 &&
+      Attrs.InterleaveCount == 0 && Attrs.UnrollCount == 0 &&
+      Attrs.UnrollAndJamCount == 0 && !Attrs.PipelineDisabled &&
+      Attrs.PipelineInitiationInterval == 0 &&
+      Attrs.VectorizeEnable == LoopAttributes::Unspecified &&
+      Attrs.UnrollEnable == LoopAttributes::Unspecified &&
+      Attrs.UnrollAndJamEnable == LoopAttributes::Unspecified &&
+      Attrs.DistributeEnable == LoopAttributes::Unspecified && !StartLoc &&
+      !EndLoc)
+    return;
+
+  TempLoopID = MDNode::getTemporary(Header->getContext(), None);
+}
+
+void LoopInfo::finish() {
+  // We did not annotate the loop body instructions because there are no
+  // attributes for this loop.
+  if (!TempLoopID)
+    return;
+
+  MDNode *LoopID;
+  LoopAttributes CurLoopAttr = Attrs;
+  LLVMContext &Ctx = Header->getContext();
+
+  if (Parent && (Parent->Attrs.UnrollAndJamEnable ||
+                 Parent->Attrs.UnrollAndJamCount != 0)) {
+    // Parent unroll-and-jams this loop.
+    // Split the transformations in those that happens before the unroll-and-jam
+    // and those after.
+
+    LoopAttributes BeforeJam, AfterJam;
+
+    BeforeJam.IsParallel = AfterJam.IsParallel = Attrs.IsParallel;
+
+    BeforeJam.VectorizeWidth = Attrs.VectorizeWidth;
+    BeforeJam.InterleaveCount = Attrs.InterleaveCount;
+    BeforeJam.VectorizeEnable = Attrs.VectorizeEnable;
+    BeforeJam.DistributeEnable = Attrs.DistributeEnable;
+
+    switch (Attrs.UnrollEnable) {
+    case LoopAttributes::Unspecified:
+    case LoopAttributes::Disable:
+      BeforeJam.UnrollEnable = Attrs.UnrollEnable;
+      AfterJam.UnrollEnable = Attrs.UnrollEnable;
+      break;
+    case LoopAttributes::Full:
+      BeforeJam.UnrollEnable = LoopAttributes::Full;
+      break;
+    case LoopAttributes::Enable:
+      AfterJam.UnrollEnable = LoopAttributes::Enable;
+      break;
+    }
+
+    AfterJam.UnrollCount = Attrs.UnrollCount;
+    AfterJam.PipelineDisabled = Attrs.PipelineDisabled;
+    AfterJam.PipelineInitiationInterval = Attrs.PipelineInitiationInterval;
+
+    // If this loop is subject of an unroll-and-jam by the parent loop, and has
+    // an unroll-and-jam annotation itself, we have to decide whether to first
+    // apply the parent's unroll-and-jam or this loop's unroll-and-jam. The
+    // UnrollAndJam pass processes loops from inner to outer, so we apply the
+    // inner first.
+    BeforeJam.UnrollAndJamCount = Attrs.UnrollAndJamCount;
+    BeforeJam.UnrollAndJamEnable = Attrs.UnrollAndJamEnable;
+
+    // Set the inner followup metadata to process by the outer loop. Only
+    // consider the first inner loop.
+    if (!Parent->UnrollAndJamInnerFollowup) {
+      // Splitting the attributes into a BeforeJam and an AfterJam part will
+      // stop 'llvm.loop.isvectorized' (generated by vectorization in BeforeJam)
+      // to be forwarded to the AfterJam part. We detect the situation here and
+      // add it manually.
+      SmallVector<Metadata *, 1> BeforeLoopProperties;
+      if (BeforeJam.VectorizeEnable != LoopAttributes::Unspecified ||
+          BeforeJam.InterleaveCount != 0 || BeforeJam.VectorizeWidth != 0)
+        BeforeLoopProperties.push_back(
+            MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.isvectorized")));
+
+      bool InnerFollowupHasTransform = false;
+      MDNode *InnerFollowup = createMetadata(AfterJam, BeforeLoopProperties,
+                                             InnerFollowupHasTransform);
+      if (InnerFollowupHasTransform)
+        Parent->UnrollAndJamInnerFollowup = InnerFollowup;
+    }
+
+    CurLoopAttr = BeforeJam;
+  }
+
+  bool HasUserTransforms = false;
+  LoopID = createMetadata(CurLoopAttr, {}, HasUserTransforms);
+  TempLoopID->replaceAllUsesWith(LoopID);
 }
 
 void LoopInfoStack::push(BasicBlock *Header, const llvm::DebugLoc &StartLoc,
                          const llvm::DebugLoc &EndLoc) {
-  Active.push_back(LoopInfo(Header, StagedAttrs, StartLoc, EndLoc));
+  Active.push_back(LoopInfo(Header, StagedAttrs, StartLoc, EndLoc,
+                            Active.empty() ? nullptr : &Active.back()));
   // Clear the attributes so nested loops do not inherit them.
   StagedAttrs.clear();
 }
 
 void LoopInfoStack::push(BasicBlock *Header, clang::ASTContext &Ctx,
                          ArrayRef<const clang::Attr *> Attrs,
                          const llvm::DebugLoc &StartLoc,
                          const llvm::DebugLoc &EndLoc) {
void LoopInfoStack::push(BasicBlock *Header, clang::ASTContext &Ctx,
   }
 
   /// Stage the attributes.
   push(Header, StartLoc, EndLoc);
 }
 
 void LoopInfoStack::pop() {
   assert(!Active.empty() && "No active loops to pop");
+  Active.back().finish();
   Active.pop_back();
 }
 
 void LoopInfoStack::InsertHelper(Instruction *I) const {
   if (I->mayReadOrWriteMemory()) {
     SmallVector<Metadata *, 4> AccessGroups;
     for (const LoopInfo &AL : Active) {
       // Here we assume that every loop that has an access group is parallel.

cfe/trunk/test/CodeGenCXX/pragma-followup_inner.cpp

+// RUN: %clang_cc1 -triple x86_64-apple-darwin -std=c++11 -emit-llvm -o - %s | FileCheck %s
+
+extern "C" void followup_inner(int n, int *x) {
+#pragma unroll_and_jam(4)
+  for(int j = 0; j < n; j++) {
+#pragma clang loop pipeline_initiation_interval(10)
+#pragma clang loop unroll_count(4)
+#pragma clang loop vectorize(assume_safety)
+#pragma clang loop distribute(enable)
+    for(int i = 0; i < n; i++)
+      x[j+i*n] = 10;
+
+  }
+}
+
+
+// CHECK-LABEL: define void @followup_inner
+// CHECK: br label %for.cond1, !llvm.loop ![[INNERLOOP_3:[0-9]+]]
+// CHECK: br label %for.cond, !llvm.loop ![[OUTERLOOP_9:[0-9]+]]
+
+// CHECK-DAG: ![[ACCESSGROUP_2:[0-9]+]] = distinct !{}
+
+// CHECK-DAG: ![[INNERLOOP_3:[0-9]+]] = distinct !{![[INNERLOOP_3:[0-9]+]], ![[PARALLEL_ACCESSES_4:[0-9]+]], ![[DISTRIBUTE_5:[0-9]+]], ![[DISTRIBUTE_FOLLOWUP_6:[0-9]+]]}
+// CHECK-DAG: ![[PARALLEL_ACCESSES_4:[0-9]+]] = !{!"llvm.loop.parallel_accesses", !2}
+// CHECK-DAG: ![[DISTRIBUTE_5:[0-9]+]] = !{!"llvm.loop.distribute.enable", i1 true}
+// CHECK-DAG: ![[DISTRIBUTE_FOLLOWUP_6:[0-9]+]] = !{!"llvm.loop.distribute.followup_all", ![[LOOP_7:[0-9]+]]}
+
+// CHECK-DAG: ![[LOOP_7:[0-9]+]] = distinct !{![[LOOP_7:[0-9]+]], ![[PARALLEL_ACCESSES_4:[0-9]+]], ![[VECTORIZE_8:[0-9]+]]}
+// CHECK-DAG: ![[VECTORIZE_8:[0-9]+]] = !{!"llvm.loop.vectorize.enable", i1 true}
+
+// CHECK-DAG: ![[OUTERLOOP_9:[0-9]+]] = distinct !{![[OUTERLOOP_9:[0-9]+]], ![[UNROLLANDJAM_COUNT_10:[0-9]+]], ![[UNROLLANDJAM_FOLLOWUPINNER_11:[0-9]+]]}
+// CHECK-DAG: ![[UNROLLANDJAM_COUNT_10:[0-9]+]] = !{!"llvm.loop.unroll_and_jam.count", i32 4}
+// CHECK-DAG: ![[UNROLLANDJAM_FOLLOWUPINNER_11:[0-9]+]] = !{!"llvm.loop.unroll_and_jam.followup_inner", !12}
+
+// CHECK-DAG: ![[LOOP_12:[0-9]+]] = distinct !{![[LOOP_12:[0-9]+]], ![[PARALLEL_ACCESSES_4:[0-9]+]], ![[ISVECTORIZED_13:[0-9]+]], ![[UNROLL_COUNT_13:[0-9]+]], ![[UNROLL_FOLLOWUP_14:[0-9]+]]}
+// CHECK-DAG: ![[ISVECTORIZED_13:[0-9]+]] = !{!"llvm.loop.isvectorized"}
+// CHECK-DAG: ![[UNROLL_COUNT_13:[0-9]+]] = !{!"llvm.loop.unroll.count", i32 4}
+// CHECK-DAG: ![[UNROLL_FOLLOWUP_14:[0-9]+]] = !{!"llvm.loop.unroll.followup_all", ![[LOOP_15:[0-9]+]]}
+
+// CHECK-DAG: ![[LOOP_15:[0-9]+]] = distinct !{![[LOOP_15:[0-9]+]], ![[PARALLEL_ACCESSES_4:[0-9]+]], ![[ISVECTORIZED_13:[0-9]+]], ![[UNROLL_DISABLE_16:[0-9]+]], ![[PIPELINE_17:[0-9]+]]}
+// CHECK-DAG: ![[UNROLL_DISABLE_16:[0-9]+]] = !{!"llvm.loop.unroll.disable"}
+// CHECK-DAG: ![[PIPELINE_17:[0-9]+]] = !{!"llvm.loop.pipeline.initiationinterval", i32 10}

cfe/trunk/test/CodeGenCXX/pragma-followup_outer.cpp

+// RUN: %clang_cc1 -triple x86_64-apple-darwin -std=c++11 -emit-llvm -o - %s | FileCheck %s
+
+extern "C" void followup_outer(int n, int *x) {
+#pragma clang loop pipeline_initiation_interval(10)
+#pragma clang loop unroll_count(4)
+#pragma unroll_and_jam
+#pragma clang loop vectorize(assume_safety)
+#pragma clang loop distribute(enable)
+  for(int j = 0; j < n; j++) {
+    x[j] = 10;
+  }
+}
+
+
+// CHECK-LABEL: define void @followup_outer
+// CHECK: br label %for.cond, !llvm.loop ![[LOOP_3:[0-9]+]]
+
+// CHECK-DAG: ![[ACCESSGROUP_2:[0-9]+]] = distinct !{}
+
+// CHECK-DAG: ![[LOOP_3:[0-9]+]] = distinct !{![[LOOP_3:[0-9]+]], ![[PARALLEL_ACCESSES_4:[0-9]+]], ![[DISTRIBUTE_5:[0-9]+]], ![[DISTRIBUTE_FOLLOWUP_6:[0-9]+]]}
+// CHECK-DAG: ![[PARALLEL_ACCESSES_4:[0-9]+]] = !{!"llvm.loop.parallel_accesses", ![[ACCESSGROUP_2]]}
+// CHECK-DAG: ![[DISTRIBUTE_5:[0-9]+]] = !{!"llvm.loop.distribute.enable", i1 true}
+// CHECK-DAG: ![[DISTRIBUTE_FOLLOWUP_6:[0-9]+]] = !{!"llvm.loop.distribute.followup_all", ![[LOOP_7:[0-9]+]]}
+
+// CHECK-DAG: ![[LOOP_7:[0-9]+]] = distinct !{![[LOOP_7:[0-9]+]], ![[PARALLEL_ACCESSES_4:[0-9]+]], ![[VECTORIZE_8:[0-9]+]], ![[VECTORIZE_FOLLOWUP_9:[0-9]+]]}
+// CHECK-DAG: ![[VECTORIZE_8:[0-9]+]] = !{!"llvm.loop.vectorize.enable", i1 true}
+// CHECK-DAG: ![[VECTORIZE_FOLLOWUP_9:[0-9]+]] = !{!"llvm.loop.vectorize.followup_all", ![[LOOP_10:[0-9]+]]}
+
+// CHECK-DAG: ![[LOOP_10:[0-9]+]] = distinct !{![[LOOP_10:[0-9]+]], ![[PARALLEL_ACCESSES_4:[0-9]+]], ![[ISVECTORIZED_11:[0-9]+]], ![[UNROLLANDJAM_12:[0-9]+]], ![[UNROLLANDJAM_FOLLOWUPOUTER_13:[0-9]+]]}
+// CHECK-DAG: ![[ISVECTORIZED_11:[0-9]+]] = !{!"llvm.loop.isvectorized"}
+// CHECK-DAG: ![[UNROLLANDJAM_12:[0-9]+]] =  !{!"llvm.loop.unroll_and_jam.enable"}
+// CHECK-DAG: ![[UNROLLANDJAM_FOLLOWUPOUTER_13:[0-9]+]] = !{!"llvm.loop.unroll_and_jam.followup_outer", ![[LOOP_14:[0-9]+]]}
+
+// CHECK-DAG: ![[LOOP_14:[0-9]+]] = distinct !{![[LOOP_14:[0-9]+]], ![[PARALLEL_ACCESSES_4:[0-9]+]], ![[ISVECTORIZED_11:[0-9]+]], ![[UNROLLANDJAM_DISABLE_15:[0-9]+]], ![[UNROLL_COUNT_16:[0-9]+]], ![[UNROLL_FOLLOWUP_17:[0-9]+]]}
+// CHECK-DAG: ![[UNROLLANDJAM_DISABLE_15:[0-9]+]]  = !{!"llvm.loop.unroll_and_jam.disable"}
+// CHECK-DAG: ![[UNROLL_COUNT_16:[0-9]+]] = !{!"llvm.loop.unroll.count", i32 4}
+// CHECK-DAG: ![[UNROLL_FOLLOWUP_17:[0-9]+]] = !{!"llvm.loop.unroll.followup_all", ![[LOOP_18:[0-9]+]]}
+
+// CHECK-DAG: ![[LOOP_18:[0-9]+]] = distinct !{![[LOOP_18:[0-9]+]], ![[PARALLEL_ACCESSES_4:[0-9]+]], ![[ISVECTORIZED_11:[0-9]+]], ![[UNROLLANDJAM_DISABLE_15:[0-9]+]], ![[UNROLL_DISABLE_19:[0-9]+]], ![[INITIATIONINTERVAL_20:[0-9]+]]}
+// CHECK-DAG: ![[UNROLL_DISABLE_19:[0-9]+]]  = !{!"llvm.loop.unroll.disable"}
+// CHECK-DAG: ![[INITIATIONINTERVAL_20:[0-9]+]] = !{!"llvm.loop.pipeline.initiationinterval", i32 10}

cfe/trunk/test/CodeGenCXX/pragma-loop-safety-imperfectly_nested.cpp

 // CHECK: store i32 %[[MUL]], i32* %{{.+}}, !llvm.access.group ![[ACCESS_GROUP_3:[0-9]+]]
 // CHECK: br label %{{.+}}, !llvm.loop ![[INNER_LOOPID:[0-9]+]]
 // CHECK: store i32 21, i32* %{{.+}}, !llvm.access.group ![[ACCESS_GROUP_2]]
 // CHECK: br label %{{.+}}, !llvm.loop ![[OUTER_LOOPID:[0-9]+]]
 
 // CHECK: ![[ACCESS_GROUP_2]] = distinct !{}
 // CHECK: ![[ACCESS_GROUP_LIST_3:[0-9]+]]  = !{![[ACCESS_GROUP_2]], ![[ACCESS_GROUP_4:[0-9]+]]}
 // CHECK: ![[ACCESS_GROUP_4]] = distinct !{}
-// CHECK: ![[INNER_LOOPID]] = distinct !{![[INNER_LOOPID]], {{.*}} ![[PARALLEL_ACCESSES_8:[0-9]+]]}
+// CHECK: ![[INNER_LOOPID]] = distinct !{![[INNER_LOOPID]], ![[PARALLEL_ACCESSES_8:[0-9]+]]
 // CHECK: ![[PARALLEL_ACCESSES_8]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_4]]}
-// CHECK: ![[OUTER_LOOPID]] = distinct !{![[OUTER_LOOPID]], {{.*}} ![[PARALLEL_ACCESSES_10:[0-9]+]]}
+// CHECK: ![[OUTER_LOOPID]] = distinct !{![[OUTER_LOOPID]], ![[PARALLEL_ACCESSES_10:[0-9]+]]
 // CHECK: ![[PARALLEL_ACCESSES_10]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_2]]}

cfe/trunk/test/CodeGenCXX/pragma-loop-safety-nested.cpp

 // CHECK: %[[MUL:.+]] = mul
 // CHECK: store i32 %[[MUL]], i32* %{{.+}}, !llvm.access.group ![[ACCESS_GROUP_LIST_3:[0-9]+]]
 // CHECK: br label %{{.+}}, !llvm.loop ![[INNER_LOOPID:[0-9]+]]
 // CHECK: br label %{{.+}}, !llvm.loop ![[OUTER_LOOPID:[0-9]+]]
 
 // CHECK: ![[ACCESS_GROUP_2]] = distinct !{}
 // CHECK: ![[ACCESS_GROUP_LIST_3]] = !{![[ACCESS_GROUP_2]], ![[ACCESS_GROUP_4:[0-9]+]]}
 // CHECK: ![[ACCESS_GROUP_4]] = distinct !{}
-// CHECK: ![[INNER_LOOPID]] = distinct !{![[INNER_LOOPID]], {{.*}} ![[PARALLEL_ACCESSES_8:[0-9]+]]}
+// CHECK: ![[INNER_LOOPID]] = distinct !{![[INNER_LOOPID]], ![[PARALLEL_ACCESSES_8:[0-9]+]]
 // CHECK: ![[PARALLEL_ACCESSES_8]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_4]]}
-// CHECK: ![[OUTER_LOOPID]] = distinct !{![[OUTER_LOOPID]], {{.*}} ![[PARALLEL_ACCESSES_10:[0-9]+]]}
+// CHECK: ![[OUTER_LOOPID]] = distinct !{![[OUTER_LOOPID]], ![[PARALLEL_ACCESSES_10:[0-9]+]]
 // CHECK: ![[PARALLEL_ACCESSES_10]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_2]]}

cfe/trunk/test/CodeGenCXX/pragma-loop-safety-outer.cpp

 
 // CHECK: %[[MUL:.+]] = mul
 // CHECK: store i32 %[[MUL]], i32* %{{.+}}, !llvm.access.group ![[ACCESS_GROUP_2:[0-9]+]]
 // CHECK: br label %{{.+}}, !llvm.loop ![[INNER_LOOPID:[0-9]+]]
 // CHECK: br label %{{.+}}, !llvm.loop ![[OUTER_LOOPID:[0-9]+]]
 
 // CHECK: ![[ACCESS_GROUP_2]] = distinct !{}
 // CHECK: ![[INNER_LOOPID]] = distinct !{![[INNER_LOOPID]],
-// CHECK: ![[OUTER_LOOPID]] = distinct !{![[OUTER_LOOPID]], {{.*}} ![[PARALLEL_ACCESSES_9:[0-9]+]]}
+// CHECK: ![[OUTER_LOOPID]] = distinct !{![[OUTER_LOOPID]], ![[PARALLEL_ACCESSES_9:[0-9]+]]
 // CHECK: ![[PARALLEL_ACCESSES_9]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_2]]}

cfe/trunk/test/CodeGenCXX/pragma-loop-safety.cpp

for (int i = 0; i < Length; i++) {
     // CHECK-NEXT: br label [[LOOP2_INC:[^,]+]]
     List[i] = i * 2;
 
     // CHECK: br label [[LOOP2_COND:[^,]+]], !llvm.loop ![[LOOP2_HINTS:[0-9]+]]
   }
 }
 
 // CHECK: ![[ACCESS_GROUP_2]] = distinct !{}
-// CHECK: ![[LOOP1_HINTS]] = distinct !{![[LOOP1_HINTS]], ![[INTERLEAVE_1:[0-9]+]], ![[INTENABLE_1:[0-9]+]], ![[UNROLL_DISABLE:[0-9]+]], ![[PARALLEL_ACCESSES_7:[0-9]+]]}
+// CHECK: ![[LOOP1_HINTS]] = distinct !{![[LOOP1_HINTS]], ![[PARALLEL_ACCESSES_7:[0-9]+]], ![[UNROLL_DISABLE:[0-9]+]], ![[INTERLEAVE_1:[0-9]+]], ![[INTENABLE_1:[0-9]+]]}
+// CHECK: ![[PARALLEL_ACCESSES_7]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_2]]}
+// CHECK: ![[UNROLL_DISABLE]] = !{!"llvm.loop.unroll.disable"}
 // CHECK: ![[INTERLEAVE_1]] = !{!"llvm.loop.interleave.count", i32 1}
 // CHECK: ![[INTENABLE_1]] = !{!"llvm.loop.vectorize.enable", i1 true}
-// CHECK: ![[UNROLL_DISABLE]] = !{!"llvm.loop.unroll.disable"}
-// CHECK: ![[PARALLEL_ACCESSES_7]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_2]]}
 // CHECK: ![[ACCESS_GROUP_8]] = distinct !{}
-// CHECK: ![[LOOP2_HINTS]] = distinct !{![[LOOP2_HINTS]], ![[WIDTH_1:[0-9]+]], ![[INTENABLE_1]], ![[UNROLL_DISABLE]], ![[PARALLEL_ACCESSES_11:[0-9]+]]}
-// CHECK: ![[WIDTH_1]] = !{!"llvm.loop.vectorize.width", i32 1}
+// CHECK: ![[LOOP2_HINTS]] = distinct !{![[LOOP2_HINTS]], ![[PARALLEL_ACCESSES_11:[0-9]+]], ![[UNROLL_DISABLE]], ![[WIDTH_1:[0-9]+]], ![[INTENABLE_1]]}
 // CHECK: ![[PARALLEL_ACCESSES_11]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_8]]}
+// CHECK: ![[WIDTH_1]] = !{!"llvm.loop.vectorize.width", i32 1}

cfe/trunk/test/CodeGenCXX/pragma-loop.cpp

 void template_test(double *List, int Length) {
   double Value = 10;
 
   for_template_test<double>(List, Length, Value);
   for_template_define_test<double, int>(List, Length, Value);
   for_template_constant_expression_test<double, 2, 4, 8>(List, Length);
 }
 
-// CHECK: ![[LOOP_1]] = distinct !{![[LOOP_1]], ![[WIDTH_4:.*]], ![[INTERLEAVE_4:.*]], ![[INTENABLE_1:.*]], ![[UNROLL_FULL:.*]], ![[DISTRIBUTE_ENABLE:.*]]}
-// CHECK: ![[WIDTH_4]] = !{!"llvm.loop.vectorize.width", i32 4}
-// CHECK: ![[INTERLEAVE_4]] = !{!"llvm.loop.interleave.count", i32 4}
-// CHECK: ![[INTENABLE_1]] = !{!"llvm.loop.vectorize.enable", i1 true}
+// CHECK: ![[LOOP_1]] = distinct !{![[LOOP_1]], ![[UNROLL_FULL:.*]]}
 // CHECK: ![[UNROLL_FULL]] = !{!"llvm.loop.unroll.full"}
-// CHECK: ![[DISTRIBUTE_ENABLE]] = !{!"llvm.loop.distribute.enable", i1 true}
-// CHECK: ![[LOOP_2]] = distinct !{![[LOOP_2:.*]], ![[WIDTH_8:.*]], ![[INTERLEAVE_4:.*]], ![[UNROLL_DISABLE:.*]], ![[DISTRIBUTE_DISABLE:.*]]}
-// CHECK: ![[WIDTH_8]] = !{!"llvm.loop.vectorize.width", i32 8}
+
+// CHECK: ![[LOOP_2]] = distinct !{![[LOOP_2]], ![[UNROLL_DISABLE:.*]], ![[DISTRIBUTE_DISABLE:.*]], ![[WIDTH_8:.*]], ![[INTERLEAVE_4:.*]]}
 // CHECK: ![[UNROLL_DISABLE]] = !{!"llvm.loop.unroll.disable"}
 // CHECK: ![[DISTRIBUTE_DISABLE]] = !{!"llvm.loop.distribute.enable", i1 false}
-// CHECK: ![[LOOP_3]] = distinct !{![[LOOP_3]], ![[INTERLEAVE_4:.*]], ![[UNROLL_8:.*]], ![[INTENABLE_1:.*]]}
+// CHECK: ![[WIDTH_8]] = !{!"llvm.loop.vectorize.width", i32 8}
+// CHECK: ![[INTERLEAVE_4]] = !{!"llvm.loop.interleave.count", i32 4}
+
+// CHECK: ![[LOOP_3]] = distinct !{![[LOOP_3]], ![[INTERLEAVE_4:.*]], ![[INTENABLE_1:.*]], ![[FOLLOWUP_VECTOR_3:.*]]}
+// CHECK: ![[INTENABLE_1]] = !{!"llvm.loop.vectorize.enable", i1 true}
+// CHECK: ![[FOLLOWUP_VECTOR_3]] = !{!"llvm.loop.vectorize.followup_all", ![[AFTER_VECTOR_3:.*]]}
+// CHECK: ![[AFTER_VECTOR_3]] = distinct !{![[AFTER_VECTOR_3]], ![[ISVECTORIZED:.*]], ![[UNROLL_8:.*]]}
+// CHECK: ![[ISVECTORIZED]] = !{!"llvm.loop.isvectorized"}
 // CHECK: ![[UNROLL_8]] = !{!"llvm.loop.unroll.count", i32 8}
+
 // CHECK: ![[LOOP_4]] = distinct !{![[LOOP_4]], ![[WIDTH_2:.*]], ![[INTERLEAVE_2:.*]]}
 // CHECK: ![[WIDTH_2]] = !{!"llvm.loop.vectorize.width", i32 2}
 // CHECK: ![[INTERLEAVE_2]] = !{!"llvm.loop.interleave.count", i32 2}
-// CHECK: ![[LOOP_5]] = distinct !{![[LOOP_5]], ![[WIDTH_1:.*]], ![[UNROLL_DISABLE:.*]], ![[DISTRIBUTE_DISABLE:.*]]}
+
+// CHECK: ![[LOOP_5]] = distinct !{![[LOOP_5]], ![[UNROLL_DISABLE:.*]], ![[DISTRIBUTE_DISABLE:.*]], ![[WIDTH_1:.*]]}
 // CHECK: ![[WIDTH_1]] = !{!"llvm.loop.vectorize.width", i32 1}
-// CHECK: ![[LOOP_6]] = distinct !{![[LOOP_6]], ![[WIDTH_2:.*]], ![[INTERLEAVE_2:.*]], ![[UNROLL_8:.*]]}
+
+// CHECK: ![[LOOP_6]] = distinct !{![[LOOP_6]], ![[WIDTH_2:.*]], ![[INTERLEAVE_2:.*]], ![[FOLLOWUP_VECTOR_6:.*]]}
+// CHECK: ![[FOLLOWUP_VECTOR_6]] = !{!"llvm.loop.vectorize.followup_all", ![[AFTER_VECTOR_6:.*]]}
+// CHECK: ![[AFTER_VECTOR_6]] = distinct !{![[AFTER_VECTOR_6]], ![[ISVECTORIZED:.*]], ![[UNROLL_8:.*]]}
+
 // CHECK: ![[LOOP_7]] = distinct !{![[LOOP_7]], ![[WIDTH_5:.*]]}
 // CHECK: ![[WIDTH_5]] = !{!"llvm.loop.vectorize.width", i32 5}
+
 // CHECK: ![[LOOP_8]] = distinct !{![[LOOP_8]], ![[WIDTH_5:.*]]}
-// CHECK: ![[LOOP_9]] = distinct !{![[LOOP_9]], ![[WIDTH_8:.*]], ![[INTERLEAVE_8:.*]], ![[UNROLL_8:.*]]}
-// CHECK: ![[INTERLEAVE_8]] = !{!"llvm.loop.interleave.count", i32 8}
-// CHECK: ![[LOOP_10]] = distinct !{![[LOOP_10]], ![[WIDTH_2:.*]], ![[INTERLEAVE_2:.*]], ![[UNROLL_8:.*]]}
-// CHECK: ![[LOOP_11]] = distinct !{![[LOOP_11]], ![[WIDTH_2:.*]], ![[INTERLEAVE_4:.*]], ![[UNROLL_8:.*]]}
-// CHECK: ![[LOOP_12]] = distinct !{![[LOOP_12]], ![[WIDTH_6:.*]], ![[INTERLEAVE_10:.*]], ![[UNROLL_24:.*]]}
-// CHECK: ![[WIDTH_6]] = !{!"llvm.loop.vectorize.width", i32 6}
-// CHECK: ![[INTERLEAVE_10]] = !{!"llvm.loop.interleave.count", i32 10}
+
+// CHECK: ![[LOOP_9]] = distinct !{![[LOOP_9]], ![[WIDTH_8:.*]], ![[INTERLEAVE_8:.*]], ![[FOLLOWUP_VECTOR_9:.*]]}
+// CHECK: ![[FOLLOWUP_VECTOR_9]] = !{!"llvm.loop.vectorize.followup_all", ![[AFTER_VECTOR_9:.*]]}
+// CHECK: ![[AFTER_VECTOR_9]] = distinct !{![[AFTER_VECTOR_9]], ![[ISVECTORIZED:.*]], ![[UNROLL_8:.*]]}
+
+// CHECK: ![[LOOP_10]] = distinct !{![[LOOP_10]], ![[WIDTH_2:.*]], ![[INTERLEAVE_2:.*]], ![[FOLLOWUP_VECTOR_10:.*]]}
+// CHECK: ![[FOLLOWUP_VECTOR_10]] = !{!"llvm.loop.vectorize.followup_all", ![[AFTER_VECTOR_10:.*]]}
+// CHECK: ![[AFTER_VECTOR_10]] = distinct !{![[AFTER_VECTOR_10]], ![[ISVECTORIZED:.*]], ![[UNROLL_8:.*]]}
+
+// CHECK: ![[LOOP_11]] = distinct !{![[LOOP_11]], ![[WIDTH_2:.*]], ![[INTERLEAVE_4:.*]], ![[FOLLOWUP_VECTOR_11:.*]]}
+// CHECK: ![[FOLLOWUP_VECTOR_11]] = !{!"llvm.loop.vectorize.followup_all", ![[AFTER_VECTOR_11:.*]]}
+// CHECK: ![[AFTER_VECTOR_11]] = distinct !{![[AFTER_VECTOR_11]], ![[ISVECTORIZED:.*]], ![[UNROLL_8:.*]]}
+
+// CHECK: ![[LOOP_12]] = distinct !{![[LOOP_12]], ![[WIDTH_6:.*]], ![[INTERLEAVE_10:.*]], ![[FOLLOWUP_VECTOR_12:.*]]}
+// CHECK: ![[FOLLOWUP_VECTOR_12]] = !{!"llvm.loop.vectorize.followup_all", ![[AFTER_VECTOR_12:.*]]}
+// CHECK: ![[AFTER_VECTOR_12]] = distinct !{![[AFTER_VECTOR_12]], ![[ISVECTORIZED:.*]], ![[UNROLL_24:.*]]}
 // CHECK: ![[UNROLL_24]] = !{!"llvm.loop.unroll.count", i32 24}
-// CHECK: ![[LOOP_13]] = distinct !{![[LOOP_13]], ![[WIDTH_8:.*]], ![[INTERLEAVE_16:.*]], ![[UNROLL_32:.*]]}
+
+// CHECK: ![[LOOP_13]] = distinct !{![[LOOP_13]], ![[WIDTH_8:.*]], ![[INTERLEAVE_16:.*]], ![[FOLLOWUP_VECTOR_13:.*]]}
 // CHECK: ![[INTERLEAVE_16]] = !{!"llvm.loop.interleave.count", i32 16}
+// CHECK: ![[FOLLOWUP_VECTOR_13]] = !{!"llvm.loop.vectorize.followup_all", ![[AFTER_VECTOR_13:.*]]}
+// CHECK: ![[AFTER_VECTOR_13]] = distinct !{![[AFTER_VECTOR_13]], ![[ISVECTORIZED:.*]], ![[UNROLL_32:.*]]}
 // CHECK: ![[UNROLL_32]] = !{!"llvm.loop.unroll.count", i32 32}
+
 // CHECK: ![[LOOP_14]] = distinct !{![[LOOP_14]], ![[WIDTH_10:.*]]}
 // CHECK: ![[WIDTH_10]] = !{!"llvm.loop.vectorize.width", i32 10}

cfe/trunk/test/CodeGenCXX/pragma-unroll-and-jam.cpp

 }
 
 // CHECK: ![[LOOP_1]] = distinct !{![[LOOP_1]], ![[UNJ_ENABLE:.*]]}
 // CHECK: ![[UNJ_ENABLE]] = !{!"llvm.loop.unroll_and_jam.enable"}
 // CHECK: ![[LOOP_2]] = distinct !{![[LOOP_2]], ![[UNJ_4:.*]]}
 // CHECK: ![[UNJ_4]] = !{!"llvm.loop.unroll_and_jam.count", i32 4}
 // CHECK: ![[LOOP_3]] = distinct !{![[LOOP_3]], ![[UNJ_DISABLE:.*]]}
 // CHECK: ![[UNJ_DISABLE]] = !{!"llvm.loop.unroll_and_jam.disable"}
-// CHECK: ![[LOOP_7]] = distinct !{![[LOOP_7]], ![[UNROLL_4:.*]], ![[UNJ_DISABLE:.*]]}
+// CHECK: ![[LOOP_7]] = distinct !{![[LOOP_7]], ![[UNJ_DISABLE:.*]], ![[UNROLL_4:.*]]}
 // CHECK: ![[UNROLL_4]] = !{!"llvm.loop.unroll.count", i32 4}

cfe/trunk/test/OpenMP/simd_metadata.c

 
 // Metadata for h1:
 // CHECK: [[LOOP_H1_HEADER:![0-9]+]] = distinct !{[[LOOP_H1_HEADER]], [[LOOP_WIDTH_16:![0-9]+]], [[LOOP_VEC_ENABLE:![0-9]+]]}
 // CHECK: [[LOOP_WIDTH_16]] = !{!"llvm.loop.vectorize.width", i32 16}
 // CHECK: [[LOOP_VEC_ENABLE]] = !{!"llvm.loop.vectorize.enable", i1 true}
 // CHECK: [[LOOP_H1_HEADER:![0-9]+]] = distinct !{[[LOOP_H1_HEADER]], [[LOOP_WIDTH_8:![0-9]+]], [[LOOP_VEC_ENABLE]]}
 // CHECK: [[LOOP_WIDTH_8]] = !{!"llvm.loop.vectorize.width", i32 8}
 // CHECK: ![[ACCESS_GROUP_7]] = distinct !{}
-// CHECK: [[LOOP_H1_HEADER:![0-9]+]] = distinct !{[[LOOP_H1_HEADER]], [[LOOP_WIDTH_8]], [[LOOP_VEC_ENABLE]], ![[PARALLEL_ACCESSES_9:[0-9]+]]}
+// CHECK: [[LOOP_H1_HEADER:![0-9]+]] = distinct !{[[LOOP_H1_HEADER]], ![[PARALLEL_ACCESSES_9:[0-9]+]], [[LOOP_WIDTH_8]], [[LOOP_VEC_ENABLE]]}
 // CHECK: ![[PARALLEL_ACCESSES_9]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_7]]}
 //
 // Metadata for h2:
 // CHECK: ![[ACCESS_GROUP_10]] = distinct !{}
-// CHECK: [[LOOP_H2_HEADER]] = distinct !{[[LOOP_H2_HEADER]], [[LOOP_VEC_ENABLE]], ![[PARALLEL_ACCESSES_12:[0-9]+]]}
+// CHECK: [[LOOP_H2_HEADER]] = distinct !{[[LOOP_H2_HEADER]], ![[PARALLEL_ACCESSES_12:[0-9]+]], [[LOOP_VEC_ENABLE]]}
 // CHECK: ![[PARALLEL_ACCESSES_12]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_10]]}
 //
 // Metadata for h3:
 // CHECK: ![[ACCESS_GROUP_13]] = distinct !{}
-// CHECK: [[LOOP_H3_HEADER]] = distinct !{[[LOOP_H3_HEADER]], [[LOOP_VEC_ENABLE]], ![[PARALLEL_ACCESSES_15:[0-9]+]]}
+// CHECK: [[LOOP_H3_HEADER]] = distinct !{[[LOOP_H3_HEADER]], ![[PARALLEL_ACCESSES_15:[0-9]+]], [[LOOP_VEC_ENABLE]]}
 // CHECK: ![[PARALLEL_ACCESSES_15]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP_13]]}
 //