Recommit r158407: Allow SROA to look at a vector type and see if the offset is out of range to be replaced with a scalar access. Now with additional fix and test for indexing into a vector inside a struct

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@158454 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 33 insertions, 9 deletions

diff --git a/lib/Transforms/Scalar/ScalarReplAggregates.cpp b/lib/Transforms/Scalar/ScalarReplAggregates.cpp
index 113397fc11..a6c3a5e3e9 100644
--- a/lib/Transforms/Scalar/ScalarReplAggregates.cpp
+++ b/lib/Transforms/Scalar/ScalarReplAggregates.cpp
@@ -1764,6 +1764,12 @@ bool SROA::TypeHasComponent(Type *T, uint64_t Offset, uint64_t Size) {
     if (Offset >= AT->getNumElements() * EltSize)
       return false;
     Offset %= EltSize;
+  } else if (VectorType *VT = dyn_cast<VectorType>(T)) {
+    EltTy = VT->getElementType();
+    EltSize = TD->getTypeAllocSize(EltTy);
+    if (Offset >= VT->getNumElements() * EltSize)
+      return false;
+    Offset %= EltSize;
   } else {
     return false;
   }
@@ -1931,9 +1937,16 @@ uint64_t SROA::FindElementAndOffset(Type *&T, uint64_t &Offset,
     Offset -= Layout->getElementOffset(Idx);
     IdxTy = Type::getInt32Ty(T->getContext());
     return Idx;
+  } else if (ArrayType *AT = dyn_cast<ArrayType>(T)) {
+    T = AT->getElementType();
+    uint64_t EltSize = TD->getTypeAllocSize(T);
+    Idx = Offset / EltSize;
+    Offset -= Idx * EltSize;
+    IdxTy = Type::getInt64Ty(T->getContext());
+    return Idx;
   }
-  ArrayType *AT = cast<ArrayType>(T);
-  T = AT->getElementType();
+  VectorType *VT = cast<VectorType>(T);
+  T = VT->getElementType();
   uint64_t EltSize = TD->getTypeAllocSize(T);
   Idx = Offset / EltSize;
   Offset -= Idx * EltSize;
diff --git a/test/Transforms/ScalarRepl/2011-06-08-VectorExtractValue.ll b/test/Transforms/ScalarRepl/2011-06-08-VectorExtractValue.ll
index 98fa1c684c..459688547f 100644
--- a/test/Transforms/ScalarRepl/2011-06-08-VectorExtractValue.ll
+++ b/test/Transforms/ScalarRepl/2011-06-08-VectorExtractValue.ll
@@ -10,8 +10,7 @@ target triple = "x86_64-apple-macosx10.7.0"
 
 ; CHECK: main
 ; CHECK-NOT: alloca
-; CHECK: %[[A:[a-z0-9]*]] = and i128
-; CHECK: %[[B:[a-z0-9]*]] = trunc i128 %[[A]] to i32
+; CHECK: extractelement <2 x float> zeroinitializer, i32 0
 
 define void @main() uwtable ssp {
 entry:
@@ -28,8 +27,7 @@ entry:
 
 ; CHECK: test1
 ; CHECK-NOT: alloca
-; CHECK: %[[A:[a-z0-9]*]] = and i128
-; CHECK: %[[B:[a-z0-9]*]] = trunc i128 %[[A]] to i32
+; CHECK: extractelement <2 x float> zeroinitializer, i32 0
 
 define void @test1() uwtable ssp {
 entry:
@@ -43,9 +41,8 @@ entry:
 
 ; CHECK: test2
 ; CHECK-NOT: alloca
-; CHECK: and i128
-; CHECK: or i128
-; CHECK: trunc i128
+; CHECK: %[[A:[a-z0-9]*]] = extractelement <2 x float> zeroinitializer, i32 0
+; CHECK: fadd float %[[A]], 1.000000e+00
 ; CHECK-NOT: insertelement
 ; CHECK-NOT: extractelement
 
@@ -62,3 +59,17 @@ entry:
   %r = fadd float %r1, %r2
   ret float %r
 }
+
+; CHECK: test3
+; CHECK: %[[A:[a-z0-9]*]] = extractelement <2 x float> <float 2.000000e+00, float 3.000000e+00>, i32 1
+; CHECK: ret float %[[A]]
+
+define float @test3() {
+entry:
+  %ai = alloca { <2 x float>, <2 x float> }, align 8
+  store { <2 x float>, <2 x float> } {<2 x float> <float 0.0, float 1.0>, <2 x float> <float 2.0, float 3.0>}, { <2 x float>, <2 x float> }* %ai, align 8
+  %tmpcast = bitcast { <2 x float>, <2 x float> }* %ai to [4 x float]*
+  %arrayidx = getelementptr inbounds [4 x float]* %tmpcast, i64 0, i64 3
+  %f = load float* %arrayidx, align 4
+  ret float %f
+}