Lower complex struct in alloca used in private pointer phi (#1261)

Ref #1221

lib/LowerPrivatePointerPHIPass.cpp

@@ -76,6 +76,26 @@ Value *makeNewGEP(const DataLayout &DL, IRBuilder<> &B, Instruction *Src,
       clspv::AddressSpace::Private);
   return B.CreateGEP(SrcTy, Src, Idxs, "", true);
 }
+
+Type *getSmallestTypeInStruct(const DataLayout &DL, StructType *Ty) {
+  Type *SmallestTy = Ty;
+  for (auto ElTy : Ty->elements()) {
+    Type *Prev = nullptr;
+    while (Prev != ElTy) {
+      Prev = ElTy;
+      if (auto ElSTy = dyn_cast<StructType>(ElTy)) {
+        ElTy = getSmallestTypeInStruct(DL, ElSTy);
+      } else {
+        ElTy = BitcastUtils::GetEleType(ElTy);
+      }
+    }
+    if (BitcastUtils::SizeInBits(DL, SmallestTy) >
+        BitcastUtils::SizeInBits(DL, ElTy)) {
+      SmallestTy = ElTy;
+    }
+  }
+  return SmallestTy;
+}
 } // namespace
 
 llvm::PreservedAnalyses
@@ -115,6 +135,22 @@ void clspv::LowerPrivatePointerPHIPass::runOnFunction(Function &F) {
   WeakInstructions ToBeErased;
   DenseMap<PHINode *, PHINode *> PHIMap;
   for (auto alloca : worklist) {
+    auto allocaSTy = dyn_cast<StructType>(alloca->getAllocatedType());
+    if (allocaSTy && BitcastUtils::IsComplexStruct(DL, allocaSTy)) {
+      auto SmallestTypeInStruct = getSmallestTypeInStruct(DL, allocaSTy);
+      IRBuilder<> B(alloca);
+      auto nb_elem = alloca->getAllocationSizeInBits(DL).value() /
+                     BitcastUtils::SizeInBits(DL, SmallestTypeInStruct);
+      if (nb_elem > 1) {
+        SmallestTypeInStruct = ArrayType::get(SmallestTypeInStruct, nb_elem);
+      }
+      auto new_alloca =
+          B.CreateAlloca(SmallestTypeInStruct, alloca->getAddressSpace());
+      alloca->replaceAllUsesWith(new_alloca);
+      ToBeErased.push_back(alloca);
+      alloca = new_alloca;
+    }
+
     SmallVector<std::tuple<Value *, Instruction *, uint64_t, Value *>> nodes;
     for (auto use : alloca->users()) {
       nodes.push_back(std::make_tuple(use, alloca, 0, nullptr));

test/PrivatePointerPHI/complex_struct.ll

@@ -0,0 +1,30 @@
+; RUN: clspv-opt %s -o %t.ll --passes=lower-private-pointer-phi
+; RUN: FileCheck %s < %t.ll
+
+; CHECK: [[alloca:%[^ ]+]] = alloca [68 x i32], align 4
+; CHECK: loop:
+; CHECK-NEXT: [[phi:%[^ ]+]] = phi i32 [ 0, %entry ], [ [[add:%[^ ]+]], %loop ]
+; CHECK: [[gep:%[^ ]+]] = getelementptr inbounds [68 x i32], ptr [[alloca]], i32 0, i32 [[phi]]
+; CHECK: load i32, ptr [[gep]], align 4
+; CHECK: [[add]] = add i32 1, [[phi]]
+
+target datalayout = "e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir-unknown-unknown"
+
+define void @test() {
+entry:
+  %alloca = alloca { [4 x i32], [64 x i32] }
+  br label %loop
+
+loop:
+  %phi = phi ptr [ %alloca, %entry ], [ %gep, %loop ]
+  %count = phi i32 [ 0, %entry ], [ %next, %loop ]
+  %load = load i32, ptr %phi
+  %gep = getelementptr { [64 x i32], [32 x i32] }, ptr %phi, i32 0, i32 0, i32 1
+  %next = add i32 %count, 1
+  %cmp = icmp eq i32 %next, 64
+  br i1 %cmp, label %exit, label %loop
+
+exit:
+  ret void
+}