Fix LoopLoadElimination to keep original alignment on the inital hoisted store

This is fixing a bug where Loop Vectorization is widening a load but
with a lower alignment. Hoisting the load without propagating the alignment
will allow inst-combine to later deduce a higher alignment that what the pointer
actually is.

Differential Revision: https://reviews.llvm.org/D28408

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@291281 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/LoopLoadElimination.cpp b/lib/Transforms/Scalar/LoopLoadElimination.cpp
index 08e7acdaaf72cb43dc7134d690b294acdf03af0c..8fb580183e307a0e0117af70f466aec1199b2af1 100644 (file)
--- a/lib/Transforms/Scalar/LoopLoadElimination.cpp
+++ b/lib/Transforms/Scalar/LoopLoadElimination.cpp
@@ -415,7 +415,9 @@ public:
     Value *InitialPtr = SEE.expandCodeFor(PtrSCEV->getStart(), Ptr->getType(),
                                           PH->getTerminator());
     Value *Initial =
-        new LoadInst(InitialPtr, "load_initial", PH->getTerminator());
+        new LoadInst(InitialPtr, "load_initial", /* isVolatile */ false,
+                     Cand.Load->getAlignment(), PH->getTerminator());
+
     PHINode *PHI = PHINode::Create(Initial->getType(), 2, "store_forwarded",
                                    &L->getHeader()->front());
     PHI->addIncoming(Initial, PH);

diff --git a/test/Transforms/LoopLoadElim/forward.ll b/test/Transforms/LoopLoadElim/forward.ll
index ed0d162ab7e3c249b94df802da5d917d1fa98031..9a0e03a317c8cb9bf7b39e6a6a72376fb8302a64 100644 (file)
--- a/test/Transforms/LoopLoadElim/forward.ll
+++ b/test/Transforms/LoopLoadElim/forward.ll
@@ -16,8 +16,8 @@ define void @f(i32* %A, i32* %B, i32* %C, i64 %N) {
 ; CHECK-NOT: %found.conflict{{.*}} =
 
 entry:
-; for.body.ph:
-; CHECK: %load_initial = load i32, i32* %A
+; Make sure the hoisted load keeps the alignment
+; CHECK: %load_initial = load i32, i32* %A, align 1
   br label %for.body
 
 for.body:                                         ; preds = %for.body, %entry
@@ -34,7 +34,7 @@ for.body:                                         ; preds = %for.body, %entry
   %a_p1 = add i32 %b, 2
   store i32 %a_p1, i32* %Aidx_next, align 4
 
-  %a = load i32, i32* %Aidx, align 4
+  %a = load i32, i32* %Aidx, align 1
 ; CHECK: %c = mul i32 %store_forwarded, 2
   %c = mul i32 %a, 2
   store i32 %c, i32* %Cidx, align 4