Explode store of arrays in instcombine

[deadalnix]

Summary

This is the last step toward supporting aggregate memory access in instcombine. This explodes stores of arrays into a serie of stores for each element, allowing them to be optimized.

[majnemer]

LGTM

lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

for (unsigned i = 0; i < Count; i++) {
       IC.Builder->CreateAlignedStore(Val, Ptr, EltAlign);
     }
 
     return true;
   }
 
   if (auto *AT = dyn_cast<ArrayType>(T)) {
     // If the array only have one element, we unpack.
-    if (AT->getNumElements() == 1) {
+    auto NumElements = AT->getNumElements();
+    if (NumElements == 1) {
       V = IC.Builder->CreateExtractValue(V, 0);
       combineStoreToNewValue(IC, SI, V);
       return true;
     }
+
+    const DataLayout &DL = IC.getDataLayout();
+    auto EltSize = DL.getTypeAllocSize(AT->getElementType());
+    auto Align = SI.getAlignment();
+    if (!Align)
+      Align = DL.getABITypeAlignment(T);
+
+    SmallString<16> EltName = V->getName();
+    EltName += ".elt";
+    auto *Addr = SI.getPointerOperand();
+    SmallString<16> AddrName = Addr->getName();
+    AddrName += ".repack";
+
+    auto *IdxType = Type::getInt64Ty(T->getContext());
+    auto *Zero = ConstantInt::get(IdxType, 0);
+
+    uint64_t Offset = 0;
+    for (uint64_t i = 0; i < NumElements; i++) {
+      Value *Indices[2] = {
+        Zero,
+        ConstantInt::get(IdxType, i),
+      };
+      auto *Ptr = IC.Builder->CreateInBoundsGEP(AT, Addr, makeArrayRef(Indices),
+                                                AddrName);
+      auto *Val = IC.Builder->CreateExtractValue(V, i, EltName);
+      auto EltAlign = MinAlign(Align, Offset);
+      IC.Builder->CreateAlignedStore(Val, Ptr, EltAlign);
+      Offset += EltSize;
+    }
+
+    return true;
   }
 
   return false;
 }
 
 /// equivalentAddressValues - Test if A and B will obviously have the same
 /// value. This includes recognizing that %t0 and %t1 will have the same
 /// value in code like this:

test/Transforms/InstCombine/unpack-fca.ll

 ; CHECK-LABEL: storeStructOfArrayOfA
 ; CHECK-NEXT: [[GEP:%[a-z0-9\.]+]] = getelementptr inbounds { [1 x %A] }, { [1 x %A] }* %saa.ptr, i64 0, i32 0, i64 0, i32 0
 ; CHECK-NEXT: store %A__vtbl* @A__vtblZ, %A__vtbl** [[GEP]], align 8
 ; CHECK-NEXT: ret void
   store { [1 x %A] } { [1 x %A] [%A { %A__vtbl* @A__vtblZ }] }, { [1 x %A] }* %saa.ptr, align 8
   ret void
 }
 
+define void @storeArrayOfB([2 x %B]* %ab.ptr, [2 x %B] %ab) {
+; CHECK-LABEL: storeArrayOfB
+; CHECK-NEXT: [[EVB0:%[a-z0-9\.]+]] = extractvalue [2 x %B] %ab, 0
+; CHECK-NEXT: [[GEP0:%[a-z0-9\.]+]] = getelementptr inbounds [2 x %B], [2 x %B]* %ab.ptr, i64 0, i64 0, i32 0
+; CHECK-NEXT: [[EV0:%[a-z0-9\.]+]] = extractvalue %B [[EVB0]], 0
+; CHECK-NEXT: store i8* [[EV0]], i8** [[GEP0]], align 8
+; CHECK-NEXT: [[GEP1:%[a-z0-9\.]+]] = getelementptr inbounds [2 x %B], [2 x %B]* %ab.ptr, i64 0, i64 0, i32 1
+; CHECK-NEXT: [[EV1:%[a-z0-9\.]+]] = extractvalue %B [[EVB0]], 1
+; CHECK-NEXT: store i64 [[EV1]], i64* [[GEP1]], align 8
+; CHECK-NEXT: [[EVB1:%[a-z0-9\.]+]] = extractvalue [2 x %B] %ab, 1
+; CHECK-NEXT: [[GEP2:%[a-z0-9\.]+]] = getelementptr inbounds [2 x %B], [2 x %B]* %ab.ptr, i64 0, i64 1, i32 0
+; CHECK-NEXT: [[EV2:%[a-z0-9\.]+]] = extractvalue %B [[EVB1]], 0
+; CHECK-NEXT: store i8* [[EV2]], i8** [[GEP2]], align 8
+; CHECK-NEXT: [[GEP3:%[a-z0-9\.]+]] = getelementptr inbounds [2 x %B], [2 x %B]* %ab.ptr, i64 0, i64 1, i32 1
+; CHECK-NEXT: [[EV3:%[a-z0-9\.]+]] = extractvalue %B [[EVB1]], 1
+; CHECK-NEXT: store i64 [[EV3]], i64* [[GEP3]], align 8
+; CHECK-NEXT: ret void
+  store [2 x %B] %ab, [2 x %B]* %ab.ptr, align 8
+  ret void
+}
+
 define %A @loadA(%A* %a.ptr) {
 ; CHECK-LABEL: loadA
 ; CHECK-NEXT: [[GEP:%[a-z0-9\.]+]] = getelementptr inbounds %A, %A* %a.ptr, i64 0, i32 0
 ; CHECK-NEXT: [[LOAD:%[a-z0-9\.]+]] = load %A__vtbl*, %A__vtbl** [[GEP]], align 8
 ; CHECK-NEXT: [[IV:%[a-z0-9\.]+]] = insertvalue %A undef, %A__vtbl* [[LOAD]], 0
 ; CHECK-NEXT: ret %A [[IV]]
   %1 = load %A, %A* %a.ptr, align 8
   ret %A %1