[SROA] Conservatively do not split the alloca if ptr is laundered

A miscompilation issue has been addressed with improved checking.

Fixes: #105537.

Author: antoniofrighetto

llvm/lib/Transforms/Scalar/SROA.cpp

@@ -3522,6 +3522,10 @@ class AllocaSliceRewriter : public InstVisitor<AllocaSliceRewriter, bool> {
            "Unexpected intrinsic!");
     LLVM_DEBUG(dbgs() << "    original: " << II << "\n");
 
+    // Do not record invariant group intrinsics for deletion.
+    if (II.isLaunderOrStripInvariantGroup())
+      return true;
+
     // Record this instruction for deletion.
     Pass.DeadInsts.push_back(&II);
 
@@ -3532,9 +3536,6 @@ class AllocaSliceRewriter : public InstVisitor<AllocaSliceRewriter, bool> {
       return true;
     }
 
-    if (II.isLaunderOrStripInvariantGroup())
-      return true;
-
     assert(II.getArgOperand(1) == OldPtr);
     // Lifetime intrinsics are only promotable if they cover the whole alloca.
     // Therefore, we drop lifetime intrinsics which don't cover the whole

llvm/lib/Transforms/Utils/PromoteMemoryToRegister.cpp

@@ -81,7 +81,8 @@ bool llvm::isAllocaPromotable(const AllocaInst *AI) {
         return false;
     } else if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(U)) {
       if (!II->isLifetimeStartOrEnd() && !II->isDroppable() &&
-          II->getIntrinsicID() != Intrinsic::fake_use)
+          II->getIntrinsicID() != Intrinsic::fake_use &&
+          !II->isLaunderOrStripInvariantGroup())
         return false;
     } else if (const BitCastInst *BCI = dyn_cast<BitCastInst>(U)) {
       if (!onlyUsedByLifetimeMarkersOrDroppableInsts(BCI))

llvm/test/Transforms/SROA/invariant-group.ll

@@ -138,10 +138,13 @@ end:
   ret void
 }
 
+; Conservatively do not split the alloca. 
 define void @partial_promotion_of_alloca() {
 ; CHECK-LABEL: @partial_promotion_of_alloca(
 ; CHECK-NEXT:    [[STRUCT_PTR_SROA_2:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[STRUCT_PTR:%.*]] = alloca [[T:%.*]], align 4
 ; CHECK-NEXT:    store volatile i32 0, ptr [[STRUCT_PTR_SROA_2]], align 4
+; CHECK-NEXT:    [[BARR:%.*]] = call ptr @llvm.launder.invariant.group.p0(ptr [[STRUCT_PTR]])
 ; CHECK-NEXT:    [[STRUCT_PTR_SROA_2_0_STRUCT_PTR_SROA_2_4_LOAD_VAL:%.*]] = load volatile i32, ptr [[STRUCT_PTR_SROA_2]], align 4
 ; CHECK-NEXT:    ret void
 ;
@@ -155,6 +158,19 @@ define void @partial_promotion_of_alloca() {
   ret void
 }
 
+define void @memcpy_after_laundering_alloca(ptr %ptr) {
+; CHECK-LABEL: @memcpy_after_laundering_alloca(
+; CHECK-NEXT:    [[ALLOCA:%.*]] = alloca { i64, i64 }, align 8
+; CHECK-NEXT:    [[LAUNDER:%.*]] = call ptr @llvm.launder.invariant.group.p0(ptr [[ALLOCA]])
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[LAUNDER]], ptr [[PTR:%.*]], i64 16, i1 false)
+; CHECK-NEXT:    ret void
+;
+  %alloca = alloca { i64, i64 }, align 8
+  %launder = call ptr @llvm.launder.invariant.group.p0(ptr %alloca)
+  call void @llvm.memcpy.p0.p0.i64(ptr %launder, ptr %ptr, i64 16, i1 false)
+  ret void
+}
+
 declare void @use(ptr)
 
 !0 = !{}