[LoopVectorize] Propagate known metadata to vectorized instructions

There are some kinds of metadata that are safe to propagate from the scalar
instructions to the vector instructions (fpmath and tbaa currently).

Regarding TBAA, one might worry about propagating it on if-converted loads and
stores, because the metadata might have had a control dependency on the
condition, and thus actually aliased with some other non-speculated memory
access when the condition was false. However, this would be caught by the
runtime overlap checks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213452 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index 6e82590..9a0f1df 100644 (file)
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -518,6 +518,34 @@ static std::string getDebugLocString(const Loop *L) {
 }
 #endif
 
+/// \brief Propagate known metadata from one instruction to another.
+static void propagateMetadata(Instruction *To, const Instruction *From) {
+  SmallVector<std::pair<unsigned, MDNode *>, 4> Metadata;
+  From->getAllMetadataOtherThanDebugLoc(Metadata);
+
+  for (auto M : Metadata) {
+    unsigned Kind = M.first;
+
+    // These are safe to transfer (this is safe for TBAA, even when we
+    // if-convert, because should that metadata have had a control dependency
+    // on the condition, and thus actually aliased with some other
+    // non-speculated memory access when the condition was false, this would be
+    // caught by the runtime overlap checks).
+    if (Kind != LLVMContext::MD_tbaa &&
+        Kind != LLVMContext::MD_fpmath)
+      continue;
+
+    To->setMetadata(Kind, M.second);
+  }
+}
+
+/// \brief Propagate known metadata from one instruction to a vector of others.
+static void propagateMetadata(SmallVectorImpl<Value *> &To, const Instruction *From) {
+  for (Value *V : To)
+    if (Instruction *I = dyn_cast<Instruction>(V))
+      propagateMetadata(I, From);
+}
+
 /// LoopVectorizationLegality checks if it is legal to vectorize a loop, and
 /// to what vectorization factor.
 /// This class does not look at the profitability of vectorization, only the
@@ -1734,7 +1762,9 @@ void InnerLoopVectorizer::vectorizeMemoryInstruction(Instruction *Instr) {
 
       Value *VecPtr = Builder.CreateBitCast(PartPtr,
                                             DataTy->getPointerTo(AddressSpace));
-      Builder.CreateStore(StoredVal[Part], VecPtr)->setAlignment(Alignment);
+      StoreInst *NewSI = Builder.CreateStore(StoredVal[Part], VecPtr);
+      NewSI->setAlignment(Alignment);
+      propagateMetadata(NewSI, SI);
     }
     return;
   }
@@ -1755,9 +1785,10 @@ void InnerLoopVectorizer::vectorizeMemoryInstruction(Instruction *Instr) {
 
     Value *VecPtr = Builder.CreateBitCast(PartPtr,
                                           DataTy->getPointerTo(AddressSpace));
-    Value *LI = Builder.CreateLoad(VecPtr, "wide.load");
-    cast<LoadInst>(LI)->setAlignment(Alignment);
-    Entry[Part] = Reverse ? reverseVector(LI) :  LI;
+    LoadInst *NewLI = Builder.CreateLoad(VecPtr, "wide.load");
+    NewLI->setAlignment(Alignment);
+    propagateMetadata(NewLI, LI);
+    Entry[Part] = Reverse ? reverseVector(NewLI) :  NewLI;
   }
 }
 
@@ -3135,6 +3166,8 @@ void InnerLoopVectorizer::vectorizeBlockInLoop(BasicBlock *BB, PhiVector *PV) {
 
         Entry[Part] = V;
       }
+
+      propagateMetadata(Entry, it);
       break;
     }
     case Instruction::Select: {
@@ -3162,6 +3195,8 @@ void InnerLoopVectorizer::vectorizeBlockInLoop(BasicBlock *BB, PhiVector *PV) {
           Op0[Part],
           Op1[Part]);
       }
+
+      propagateMetadata(Entry, it);
       break;
     }
 
@@ -3181,6 +3216,8 @@ void InnerLoopVectorizer::vectorizeBlockInLoop(BasicBlock *BB, PhiVector *PV) {
           C = Builder.CreateICmp(Cmp->getPredicate(), A[Part], B[Part]);
         Entry[Part] = C;
       }
+
+      propagateMetadata(Entry, it);
       break;
     }
 
@@ -3213,6 +3250,7 @@ void InnerLoopVectorizer::vectorizeBlockInLoop(BasicBlock *BB, PhiVector *PV) {
         Value *Broadcasted = getBroadcastInstrs(ScalarCast);
         for (unsigned Part = 0; Part < UF; ++Part)
           Entry[Part] = getConsecutiveVector(Broadcasted, VF * Part, false);
+        propagateMetadata(Entry, it);
         break;
       }
       /// Vectorize casts.
@@ -3222,6 +3260,7 @@ void InnerLoopVectorizer::vectorizeBlockInLoop(BasicBlock *BB, PhiVector *PV) {
       VectorParts &A = getVectorValue(it->getOperand(0));
       for (unsigned Part = 0; Part < UF; ++Part)
         Entry[Part] = Builder.CreateCast(CI->getOpcode(), A[Part], DestTy);
+      propagateMetadata(Entry, it);
       break;
     }
 
@@ -3259,6 +3298,8 @@ void InnerLoopVectorizer::vectorizeBlockInLoop(BasicBlock *BB, PhiVector *PV) {
           Function *F = Intrinsic::getDeclaration(M, ID, Tys);
           Entry[Part] = Builder.CreateCall(F, Args);
         }
+
+        propagateMetadata(Entry, it);
         break;
       }
       break;

diff --git a/test/Transforms/LoopVectorize/metadata.ll b/test/Transforms/LoopVectorize/metadata.ll
new file mode 100644 (file)
index 0000000..bdcf1c9
--- /dev/null
+++ b/test/Transforms/LoopVectorize/metadata.ll
@@ -0,0 +1,44 @@
+; RUN: opt < %s  -loop-vectorize -force-vector-unroll=1 -force-vector-width=4 -S | FileCheck %s
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; Function Attrs: nounwind uwtable
+define i32 @test1(i32* nocapture %a, float* nocapture readonly %b) #0 {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds float* %b, i64 %indvars.iv
+  %0 = load float* %arrayidx, align 4, !tbaa !0
+  %conv = fptosi float %0 to i32
+  %arrayidx2 = getelementptr inbounds i32* %a, i64 %indvars.iv
+  store i32 %conv, i32* %arrayidx2, align 4, !tbaa !4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1600
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body
+  ret i32 0
+}
+
+; CHECK-LABEL: @test1
+; CHECK: load <4 x float>* %{{.*}}, align 4, !tbaa ![[TFLT:[0-9]+]]
+; CHECK: store <4 x i32> %{{.*}}, <4 x i32>* %{{.*}}, align 4, !tbaa ![[TINT:[0-9]+]]
+; CHECK: ret i32 0
+
+; CHECK-DAG: ![[TFLT]] = metadata !{metadata ![[TFLT1:[0-9]+]]
+; CHECK-DAG: ![[TFLT1]] = metadata !{metadata !"float"
+
+; CHECK-DAG: ![[TINT]] = metadata !{metadata ![[TINT1:[0-9]+]]
+; CHECK-DAG: ![[TINT1]] = metadata !{metadata !"int"
+
+attributes #0 = { nounwind uwtable }
+
+!0 = metadata !{metadata !1, metadata !1, i64 0}
+!1 = metadata !{metadata !"float", metadata !2, i64 0}
+!2 = metadata !{metadata !"omnipotent char", metadata !3, i64 0}
+!3 = metadata !{metadata !"Simple C/C++ TBAA"}
+!4 = metadata !{metadata !5, metadata !5, i64 0}
+!5 = metadata !{metadata !"int", metadata !2, i64 0}
+