[clang] a33544b - [OpenACC][CIR] Implement 'alloca copying' for private lowering (#161382)

via cfe-commits cfe-commits at lists.llvm.org
Wed Oct 1 06:16:43 PDT 2025


Author: Erich Keane
Date: 2025-10-01T06:16:39-07:00
New Revision: a33544b83c80dcaa851fabd2979def6f68dd6e7a

URL: https://github.com/llvm/llvm-project/commit/a33544b83c80dcaa851fabd2979def6f68dd6e7a
DIFF: https://github.com/llvm/llvm-project/commit/a33544b83c80dcaa851fabd2979def6f68dd6e7a.diff

LOG: [OpenACC][CIR] Implement 'alloca copying' for private lowering (#161382)

The previous patch ensured that we correctly got the allocas put in
place. This patch takes the address of each element of each alloca, and
copies it to the previous one. This allows us to re-form the
pointer-structure for a recipe.

Added: 
    

Modified: 
    clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.cpp
    clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.h
    clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-CtorDtor.cpp
    clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-NoOps.cpp
    clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-int.cpp
    clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-CtorDtor.cpp
    clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-NoOps.cpp
    clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-int.cpp

Removed: 
    


################################################################################
diff  --git a/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.cpp b/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.cpp
index fc28ac552224c..25cacbb73eb03 100644
--- a/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.cpp
@@ -36,6 +36,75 @@ mlir::Block *OpenACCRecipeBuilderBase::createRecipeBlock(mlir::Region &region,
   llvm::SmallVector<mlir::Location> locs{types.size(), loc};
   return builder.createBlock(&region, region.end(), types, locs);
 }
+void OpenACCRecipeBuilderBase::makeAllocaCopy(mlir::Location loc,
+                                              mlir::Type copyType,
+                                              mlir::Value numEltsToCopy,
+                                              mlir::Value offsetPerSubarray,
+                                              mlir::Value destAlloca,
+                                              mlir::Value srcAlloca) {
+  mlir::OpBuilder::InsertionGuard guardCase(builder);
+
+  mlir::Type itrTy = cgf.cgm.convertType(cgf.getContext().UnsignedLongLongTy);
+  auto itrPtrTy = cir::PointerType::get(itrTy);
+  mlir::IntegerAttr itrAlign =
+      cgf.cgm.getSize(cgf.getContext().getTypeAlignInChars(
+          cgf.getContext().UnsignedLongLongTy));
+
+  auto loopBuilder = [&]() {
+    auto itr =
+        cir::AllocaOp::create(builder, loc, itrPtrTy, itrTy, "itr", itrAlign);
+    cir::ConstantOp constZero = builder.getConstInt(loc, itrTy, 0);
+    builder.CIRBaseBuilderTy::createStore(loc, constZero, itr);
+    builder.createFor(
+        loc,
+        /*condBuilder=*/
+        [&](mlir::OpBuilder &b, mlir::Location loc) {
+          // itr < numEltsToCopy
+          // Enforce a trip count of 1 if there wasn't any element count, this
+          // way we can just use this loop with a constant bounds instead of a
+          // separate code path.
+          if (!numEltsToCopy)
+            numEltsToCopy = builder.getConstInt(loc, itrTy, 1);
+
+          auto loadCur = cir::LoadOp::create(builder, loc, {itr});
+          auto cmp = builder.createCompare(loc, cir::CmpOpKind::lt, loadCur,
+                                           numEltsToCopy);
+          builder.createCondition(cmp);
+        },
+        /*bodyBuilder=*/
+        [&](mlir::OpBuilder &b, mlir::Location loc) {
+          // destAlloca[itr] = srcAlloca[offsetPerSubArray * itr];
+          auto loadCur = cir::LoadOp::create(builder, loc, {itr});
+          auto srcOffset = builder.createMul(loc, offsetPerSubarray, loadCur);
+
+          auto ptrToOffsetIntoSrc = cir::PtrStrideOp::create(
+              builder, loc, copyType, srcAlloca, srcOffset);
+
+          auto offsetIntoDecayDest = cir::PtrStrideOp::create(
+              builder, loc, builder.getPointerTo(copyType), destAlloca,
+              loadCur);
+
+          builder.CIRBaseBuilderTy::createStore(loc, ptrToOffsetIntoSrc,
+                                                offsetIntoDecayDest);
+          builder.createYield(loc);
+        },
+        /*stepBuilder=*/
+        [&](mlir::OpBuilder &b, mlir::Location loc) {
+          // Simple increment of the iterator.
+          auto load = cir::LoadOp::create(builder, loc, {itr});
+          auto inc = cir::UnaryOp::create(builder, loc, load.getType(),
+                                          cir::UnaryOpKind::Inc, load);
+          builder.CIRBaseBuilderTy::createStore(loc, inc, itr);
+          builder.createYield(loc);
+        });
+  };
+
+  cir::ScopeOp::create(builder, loc,
+                       [&](mlir::OpBuilder &b, mlir::Location loc) {
+                         loopBuilder();
+                         builder.createYield(loc);
+                       });
+}
 
 mlir::Value OpenACCRecipeBuilderBase::makeBoundsAlloca(
     mlir::Block *block, SourceRange exprRange, mlir::Location loc,
@@ -78,6 +147,10 @@ mlir::Value OpenACCRecipeBuilderBase::makeBoundsAlloca(
 
   bool lastBoundWasArray = isArrayTy(boundTypes.back());
 
+  // Make sure we track a moving version of this so we can get our
+  // 'copying' back to correct.
+  mlir::Value lastAlloca = initialAlloca;
+
   // Since we're iterating the types in reverse, this sets up for each index
   // corresponding to the boundsRange to be the 'after application of the
   // bounds.
@@ -125,14 +198,21 @@ mlir::Value OpenACCRecipeBuilderBase::makeBoundsAlloca(
 
       mlir::Type eltTy = cgf.convertType(resultType);
       cir::PointerType ptrTy = builder.getPointerTo(eltTy);
-      builder.createAlloca(loc, ptrTy, eltTy, "openacc.init.bounds",
-                           cgf.getContext().getTypeAlignInChars(resultType),
-                           curSize);
-
-      // TODO: OpenACC : At this point we should be copying the addresses of
-      // each element of this to the last allocation.  At the moment, that is
-      // not yet implemented.
-      cgf.cgm.errorNYI(exprRange, "OpenACC recipe alloca copying");
+      mlir::Value curAlloca = builder.createAlloca(
+          loc, ptrTy, eltTy, "openacc.init.bounds",
+          cgf.getContext().getTypeAlignInChars(resultType), curSize);
+
+      makeAllocaCopy(loc, ptrTy, cumulativeElts, eltsPerSubArray, lastAlloca,
+                     curAlloca);
+      lastAlloca = curAlloca;
+    } else {
+      // In the case of an array, we just need to decay the pointer, so just do
+      // a zero-offset stride on the last alloca to decay it down an array
+      // level.
+      cir::ConstantOp constZero = builder.getConstInt(loc, itrTy, 0);
+      lastAlloca = builder.getArrayElement(loc, loc, lastAlloca,
+                                           cgf.convertType(resultType),
+                                           constZero, /*shouldDecay=*/true);
     }
 
     cumulativeElts = eltsToAlloca;

diff  --git a/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.h b/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.h
index acd187bbaee0d..d802ccb151b64 100644
--- a/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.h
+++ b/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.h
@@ -24,6 +24,10 @@
 
 namespace clang::CIRGen {
 class OpenACCRecipeBuilderBase {
+  // makes the copy of the addresses of an alloca to the previous allocation.
+  void makeAllocaCopy(mlir::Location loc, mlir::Type copyType,
+                      mlir::Value numEltsToCopy, mlir::Value offsetPerSubarray,
+                      mlir::Value destAlloca, mlir::Value srcAlloca);
   // This function generates the required alloca, similar to
   // 'emitAutoVarAlloca', except for the OpenACC array/pointer types.
   mlir::Value makeBoundsAlloca(mlir::Block *block, SourceRange exprRange,

diff  --git a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-CtorDtor.cpp b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-CtorDtor.cpp
index e17ef90d01212..da56de005d5f2 100644
--- a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-CtorDtor.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-CtorDtor.cpp
@@ -20,6 +20,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !rec_CtorDtor, !cir.ptr<!rec_CtorDtor>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!rec_CtorDtor>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_CtorDtor>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -77,12 +101,59 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !rec_CtorDtor, !cir.ptr<!rec_CtorDtor>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!rec_CtorDtor>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_CtorDtor>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -165,6 +236,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
@@ -172,12 +267,60 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB3_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS2:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS2]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA3:.*]] = cir.alloca !rec_CtorDtor, !cir.ptr<!rec_CtorDtor>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA3]] : !cir.ptr<!rec_CtorDtor>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_CtorDtor> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -279,12 +422,59 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -311,12 +501,40 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[TL_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: %[[TL_DEREF:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !rec_CtorDtor, !cir.ptr<!rec_CtorDtor>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64} 
+// 
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!rec_CtorDtor>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_CtorDtor>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_DEREF]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -399,6 +617,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<20> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.array<!rec_CtorDtor x 5>, !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64} 
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!rec_CtorDtor x 5>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -478,6 +720,10 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[UB3:.*]] = acc.get_upperbound %[[BOUND3]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB3_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB3]] : index to !u64i 
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[TL_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!cir.ptr<!rec_CtorDtor>> x 5>>), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: %[[TL_DEREF:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
@@ -485,12 +731,58 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64} 
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB3_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_DEREF]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS2:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS2]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !rec_CtorDtor, !cir.ptr<!rec_CtorDtor>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!rec_CtorDtor>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_CtorDtor>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -589,12 +881,39 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[TL_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!cir.ptr<!rec_CtorDtor>> x 5>>), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: %[[TL_DEREF:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64} 
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_DEREF]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -622,6 +941,31 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
@@ -629,6 +973,29 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.array<!rec_CtorDtor x 5>, !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB3_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!rec_CtorDtor x 5>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -729,12 +1096,59 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<20> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.array<!rec_CtorDtor x 5>, !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!rec_CtorDtor x 5>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!rec_CtorDtor x 5>>, !cir.ptr<!cir.ptr<!cir.array<!rec_CtorDtor x 5>>> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -838,9 +1252,37 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.array<!cir.ptr<!rec_CtorDtor> x 5>, !cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>, !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
+//
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: %[[STRIDE:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
 // 
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
@@ -848,6 +1290,29 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS2]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !rec_CtorDtor, !cir.ptr<!rec_CtorDtor>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!rec_CtorDtor>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_CtorDtor> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[STRIDE]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -948,6 +1413,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<40> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.array<!cir.ptr<!rec_CtorDtor> x 5>, !cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>, !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_CtorDtor> x 5>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }

diff  --git a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-NoOps.cpp b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-NoOps.cpp
index 4398216132616..f9230dcaf1691 100644
--- a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-NoOps.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-NoOps.cpp
@@ -15,6 +15,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !rec_NoOps, !cir.ptr<!rec_NoOps>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!rec_NoOps>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_NoOps>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -41,12 +65,59 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !rec_NoOps, !cir.ptr<!rec_NoOps>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!rec_NoOps>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_NoOps>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -75,6 +146,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
@@ -82,12 +177,58 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB3_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS2:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS2]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA3:.*]] = cir.alloca !rec_NoOps, !cir.ptr<!rec_NoOps>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA3]] : !cir.ptr<!rec_NoOps>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_NoOps> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -109,12 +250,59 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -141,12 +329,39 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[TL_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: %[[TL_DEREF:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !rec_NoOps, !cir.ptr<!rec_NoOps>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64} 
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!rec_NoOps>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_NoOps>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_DEREF]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -176,6 +391,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<20> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.array<!rec_NoOps x 5>, !cir.ptr<!cir.array<!rec_NoOps x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64} 
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!rec_NoOps x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!rec_NoOps x 5>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!rec_NoOps x 5>>, !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -200,6 +439,10 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[UB3:.*]] = acc.get_upperbound %[[BOUND3]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB3_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB3]] : index to !u64i 
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[TL_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!cir.ptr<!rec_NoOps>> x 5>>), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: %[[TL_DEREF:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
@@ -207,12 +450,58 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64} 
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB3_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_DEREF]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS2:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS2]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !rec_NoOps, !cir.ptr<!rec_NoOps>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!rec_NoOps>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_NoOps>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -231,12 +520,39 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[TL_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!cir.ptr<!rec_NoOps>> x 5>>), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: %[[TL_DEREF:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64} 
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_DEREF]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -265,12 +581,59 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.array<!rec_NoOps x 5>>, !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<20> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.array<!rec_NoOps x 5>, !cir.ptr<!cir.array<!rec_NoOps x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB3_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.array<!rec_NoOps x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!rec_NoOps x 5>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!rec_NoOps x 5>>, !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -292,12 +655,59 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.array<!rec_NoOps x 5>>, !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<20> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.array<!rec_NoOps x 5>, !cir.ptr<!cir.array<!rec_NoOps x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.array<!rec_NoOps x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!rec_NoOps x 5>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!rec_NoOps x 5>>, !cir.ptr<!cir.ptr<!cir.array<!rec_NoOps x 5>>> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -328,16 +738,67 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.array<!cir.ptr<!rec_NoOps> x 5>, !cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>, !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: %[[STRIDE:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS2:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS2]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !rec_NoOps, !cir.ptr<!rec_NoOps>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!rec_NoOps>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_NoOps> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[STRIDE]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -358,6 +819,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<40> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.array<!cir.ptr<!rec_NoOps> x 5>, !cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>, !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!rec_NoOps> x 5>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }

diff  --git a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-int.cpp b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-int.cpp
index 79692d3468295..45d8b78a4a128 100644
--- a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-int.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-array-recipes-int.cpp
@@ -1,4 +1,4 @@
-// RUN: not %clang_cc1 -fopenacc -triple x86_64-linux-gnu -Wno-openacc-self-if-potential-conflict -emit-cir -fclangir -triple x86_64-linux-pc %s -o - | FileCheck %s
+// RUN: %clang_cc1 -fopenacc -triple x86_64-linux-gnu -Wno-openacc-self-if-potential-conflict -emit-cir -fclangir -triple x86_64-linux-pc %s -o - | FileCheck %s
 
 template<typename T>
 void do_things(unsigned A, unsigned B) {
@@ -13,6 +13,31 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!s32i>, %[[SRC_IDX]] : !u64i), !cir.ptr<!s32i>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -39,12 +64,59 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!s32i>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!s32i>, %[[SRC_IDX]] : !u64i), !cir.ptr<!s32i>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -73,6 +145,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
@@ -80,12 +176,57 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB3_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.ptr<!s32i>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS2:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS2]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA3:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA3]] : !cir.ptr<!s32i>, %[[SRC_IDX]] : !u64i), !cir.ptr<!s32i> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!s32i>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -107,12 +248,59 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.ptr<!s32i>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -138,12 +326,39 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[TL_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: %[[TL_DEREF:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64} 
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!s32i>, %[[SRC_IDX]] : !u64i), !cir.ptr<!s32i>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_DEREF]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!s32i>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -172,6 +387,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<20> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.array<!s32i x 5>, !cir.ptr<!cir.array<!s32i x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64} 
+// 
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!s32i x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!s32i x 5>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!s32i x 5>>, !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -197,6 +436,10 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[UB3:.*]] = acc.get_upperbound %[[BOUND3]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB3_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB3]] : index to !u64i 
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[TL_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!cir.ptr<!s32i>> x 5>>), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: %[[TL_DEREF:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
@@ -204,12 +447,58 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64} 
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB3_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!s32i>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_DEREF]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS2:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS2]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!s32i>, %[[SRC_IDX]] : !u64i), !cir.ptr<!s32i>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!s32i>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -228,12 +517,39 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[TL_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!cir.ptr<!s32i>> x 5>>), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: %[[TL_DEREF:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64} 
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_DEREF]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -262,12 +578,60 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.array<!s32i x 5>>, !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB3_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<20> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.array<!s32i x 5>, !cir.ptr<!cir.array<!s32i x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB3_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.array<!s32i x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!s32i x 5>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!s32i x 5>>, !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT:} 
@@ -289,12 +653,59 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.array<!s32i x 5>>, !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[UB2_CAST]]) : !u64i
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<20> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !cir.array<!s32i x 5>, !cir.ptr<!cir.array<!s32i x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UB2_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!cir.array<!s32i x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!s32i x 5>>
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>>
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!s32i x 5>>, !cir.ptr<!cir.ptr<!cir.array<!s32i x 5>>> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -325,16 +736,67 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.array<!cir.ptr<!s32i> x 5>, !cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>, !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
+//
 // CHECK-NEXT: %[[UB2:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB2_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB2]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[UB3_CAST]]) : !u64i
 //
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: %[[DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: %[[STRIDE:.*]] = cir.ptr_stride(%[[DECAY]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ZERO]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+//
 // CHECK-NEXT: %[[UB1:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UB1_CAST:.*]] = builtin.unrealized_conversion_cast %[[UB1]] : index to !u64i 
 // CHECK-NEXT: %[[NUM_ELTS2:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[ELT_SIZE:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS2]], %[[ELT_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA2:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB1_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA2]] : !cir.ptr<!s32i>, %[[SRC_IDX]] : !u64i), !cir.ptr<!s32i> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[STRIDE]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!s32i>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -356,6 +818,30 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[ARR_SIZE:.*]] = cir.const #cir.int<40> : !u64i
 // CHECK-NEXT: %[[ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ARR_SIZE]]) : !u64i
 // CHECK-NEXT: %[[ARR_ALLOCA:.*]] = cir.alloca !cir.array<!cir.ptr<!s32i> x 5>, !cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>, %[[ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPP_BOUND:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPP_BOUND]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]]) 
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UB2_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC:.*]] = cir.ptr_stride(%[[ARR_ALLOCA]] : !cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>> 
+// CHECK-NEXT: %[[DEST:.*]] = cir.ptr_stride(%[[TL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>> 
+// CHECK-NEXT: cir.store %[[SRC]], %[[DEST]] : !cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>, !cir.ptr<!cir.ptr<!cir.array<!cir.ptr<!s32i> x 5>>> 
+// CHECK-NEXT: cir.yield 
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i 
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i> 
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: } 
+// CHECK-NEXT: } 
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }

diff  --git a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-CtorDtor.cpp b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-CtorDtor.cpp
index 77ff3571a98b4..65b0365521bf4 100644
--- a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-CtorDtor.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-CtorDtor.cpp
@@ -29,6 +29,34 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[SIZEOF_INT_PTR:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_INT_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -44,12 +72,64 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // CHECK-NEXT: %[[INT_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUNDS1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_2:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[UPPER_BOUND_CAST]]) : !u64i
 // CHECK-NEXT: %[[SIZEOF_PTR:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_BOUND_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -71,6 +151,32 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // CHECK-NEXT: %[[INT_PTR_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUNDS2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_2:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[UPPER_BOUND_CAST]]) : !u64i
@@ -78,12 +184,65 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[SIZEOF_PTR_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_PTR_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_BOUND_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // CHECK-NEXT: %[[INT_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_3:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS_2:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_3]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[SIZEOF_INT:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS_2]], %[[SIZEOF_INT]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_ALLOCA:.*]] = cir.alloca !rec_CtorDtor, !cir.ptr<!rec_CtorDtor>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_3]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_ALLOCA]] : !cir.ptr<!rec_CtorDtor>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_CtorDtor>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -193,6 +352,33 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[SIZEOF_INT_PTR:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_INT_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -208,12 +394,64 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_CtorDtor>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // CHECK-NEXT: %[[INT_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUNDS1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_2:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[UPPER_BOUND_CAST]]) : !u64i
 // CHECK-NEXT: %[[SIZEOF_INT:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[SIZEOF_INT]]) : !u64i
 // CHECK-NEXT: %[[INT_VLA_ALLOCA:.*]] = cir.alloca !rec_CtorDtor, !cir.ptr<!rec_CtorDtor>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_BOUND_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_VLA_ALLOCA]] : !cir.ptr<!rec_CtorDtor>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_CtorDtor>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {
@@ -299,6 +537,34 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[SIZEOF_CTORDTOR:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_CTORDTOR]]) : !u64i
 // CHECK-NEXT: %[[INT_VLA_ALLOCA:.*]] = cir.alloca !rec_CtorDtor, !cir.ptr<!rec_CtorDtor>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_VLA_ALLOCA]] : !cir.ptr<!rec_CtorDtor>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_CtorDtor>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_CtorDtor>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!rec_CtorDtor>, !cir.ptr<!cir.ptr<!rec_CtorDtor>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: } destroy {

diff  --git a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-NoOps.cpp b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-NoOps.cpp
index 4822dd70227f8..07e06f86924fb 100644
--- a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-NoOps.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-NoOps.cpp
@@ -23,6 +23,33 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[SIZEOF_INT_PTR:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_INT_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -38,12 +65,65 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
+//
 // CHECK-NEXT: %[[INT_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_2:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[UPPER_BOUND_CAST]]) : !u64i
 // CHECK-NEXT: %[[SIZEOF_PTR:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_BOUND_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -63,6 +143,33 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
+//
 // CHECK-NEXT: %[[INT_PTR_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_2:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[UPPER_BOUND_CAST]]) : !u64i
@@ -70,12 +177,66 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[SIZEOF_PTR_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_PTR_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_BOUND_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
+//
 // CHECK-NEXT: %[[INT_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_3:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS_2:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_3]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[SIZEOF_INT:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS_2]], %[[SIZEOF_INT]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_ALLOCA:.*]] = cir.alloca !rec_NoOps, !cir.ptr<!rec_NoOps>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_3]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_ALLOCA]] : !cir.ptr<!rec_NoOps>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_NoOps>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -106,6 +267,33 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[SIZEOF_INT_PTR:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_INT_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+// 
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -121,12 +309,64 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, !cir.ptr<!cir.ptr<!cir.ptr<!rec_NoOps>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // CHECK-NEXT: %[[INT_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_2:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[UPPER_BOUND_CAST]]) : !u64i
 // CHECK-NEXT: %[[SIZEOF_INT:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[SIZEOF_INT]]) : !u64i
 // CHECK-NEXT: %[[INT_VLA_ALLOCA:.*]] = cir.alloca !rec_NoOps, !cir.ptr<!rec_NoOps>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_BOUND_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_VLA_ALLOCA]] : !cir.ptr<!rec_NoOps>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_NoOps>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -155,6 +395,34 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[SIZEOF_NOOPS:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_NOOPS]]) : !u64i
 // CHECK-NEXT: %[[INT_VLA_ALLOCA:.*]] = cir.alloca !rec_NoOps, !cir.ptr<!rec_NoOps>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_VLA_ALLOCA]] : !cir.ptr<!rec_NoOps>, %[[SRC_IDX]] : !u64i), !cir.ptr<!rec_NoOps>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!rec_NoOps>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!rec_NoOps>, !cir.ptr<!cir.ptr<!rec_NoOps>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }

diff  --git a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-int.cpp b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-int.cpp
index ddf25de34f74e..a3b7dcac7689d 100644
--- a/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-int.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/private-clause-pointer-recipes-int.cpp
@@ -1,4 +1,4 @@
-// RUN: not %clang_cc1 -fopenacc -triple x86_64-linux-gnu -Wno-openacc-self-if-potential-conflict -emit-cir -fclangir -triple x86_64-linux-pc %s -o - | FileCheck %s
+// RUN: %clang_cc1 -fopenacc -triple x86_64-linux-gnu -Wno-openacc-self-if-potential-conflict -emit-cir -fclangir -triple x86_64-linux-pc %s -o - | FileCheck %s
 
 template<typename T>
 void do_things(unsigned A, unsigned B) {
@@ -21,6 +21,33 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[SIZEOF_INT_PTR:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_INT_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -36,12 +63,65 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+
 // CHECK-NEXT: %[[INT_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_2:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[UPPER_BOUND_CAST]]) : !u64i
 // CHECK-NEXT: %[[SIZEOF_PTR:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_BOUND_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -61,6 +141,33 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, !cir.ptr<!cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
+//
 // CHECK-NEXT: %[[INT_PTR_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUND2]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_2:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[UPPER_BOUND_CAST]]) : !u64i
@@ -68,12 +175,64 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[SIZEOF_PTR_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_PTR_ALLOCA:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_BOUND_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // CHECK-NEXT: %[[INT_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_3:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS_2:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_3]], %[[NUM_ELTS]]) : !u64i
 // CHECK-NEXT: %[[SIZEOF_INT:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS_2]], %[[SIZEOF_INT]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_ALLOCA:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[NUM_ELTS]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_3]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_ALLOCA]] : !cir.ptr<!s32i>, %[[SRC_IDX]] : !u64i), !cir.ptr<!s32i>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_PTR_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -104,6 +263,34 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[SIZEOF_INT_PTR:.*]] = cir.const #cir.int<8> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_INT_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -119,6 +306,32 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_PTR]]) : !u64i
 // CHECK-NEXT: %[[INT_PTR_VLA_ALLOCA:.*]] = cir.alloca !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 8 : i64}
 //
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[SRC_IDX]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!cir.ptr<!s32i>>, !cir.ptr<!cir.ptr<!cir.ptr<!s32i>>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // CHECK-NEXT: %[[INT_PTR_UPPER_BOUND:.*]] = acc.get_upperbound %[[BOUND1]] : (!acc.data_bounds_ty) -> index
 // CHECK-NEXT: %[[UPPER_BOUND_CAST_2:.*]] = builtin.unrealized_conversion_cast %[[INT_PTR_UPPER_BOUND]] : index to !u64i
 // CHECK-NEXT: %[[NUM_ELTS:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[UPPER_BOUND_CAST]]) : !u64i
@@ -126,6 +339,32 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[NUM_ELTS]], %[[SIZEOF_INT]]) : !u64i
 // CHECK-NEXT: %[[INT_VLA_ALLOCA:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
 //
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_BOUND_CAST]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST_2]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_VLA_ALLOCA]] : !cir.ptr<!s32i>, %[[SRC_IDX]] : !u64i), !cir.ptr<!s32i>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[INT_PTR_VLA_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
+//
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }
@@ -154,6 +393,33 @@ void do_things(unsigned A, unsigned B) {
 // CHECK-NEXT: %[[SIZEOF_INT:.*]] = cir.const #cir.int<4> : !u64i
 // CHECK-NEXT: %[[CALC_ALLOCA_SIZE:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[SIZEOF_INT]]) : !u64i
 // CHECK-NEXT: %[[INT_VLA_ALLOCA:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, %[[CALC_ALLOCA_SIZE]] : !u64i, ["openacc.init.bounds"] {alignment = 4 : i64}
+//
+// Copy array pointer to the original alloca.
+// CHECK-NEXT: cir.scope {
+// CHECK-NEXT: %[[ITR:.*]] = cir.alloca !u64i, !cir.ptr<!u64i>, ["itr"] {alignment = 8 : i64}
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0> : !u64i
+// CHECK-NEXT: cir.store %[[ZERO]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.for : cond {
+// CHECK-NEXT: %[[UPPER_LIMIT:.*]] = cir.const #cir.int<1> : !u64i
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[CMP:.*]] = cir.cmp(lt, %[[ITR_LOAD]], %[[UPPER_LIMIT]]) : !u64i, !cir.bool
+// CHECK-NEXT: cir.condition(%[[CMP]])
+//
+// CHECK-NEXT: } body {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[SRC_IDX:.*]] = cir.binop(mul, %[[UPPER_BOUND_CAST]], %[[ITR_LOAD]]) : !u64i
+// CHECK-NEXT: %[[SRC_STRIDE:.*]] = cir.ptr_stride(%[[INT_VLA_ALLOCA]] : !cir.ptr<!s32i>, %[[SRC_IDX]] : !u64i), !cir.ptr<!s32i>
+// CHECK-NEXT: %[[DEST_STRIDE:.*]] = cir.ptr_stride(%[[TOP_LEVEL_ALLOCA]] : !cir.ptr<!cir.ptr<!s32i>>, %[[ITR_LOAD]] : !u64i), !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.store %[[SRC_STRIDE]], %[[DEST_STRIDE]] : !cir.ptr<!s32i>, !cir.ptr<!cir.ptr<!s32i>>
+// CHECK-NEXT: cir.yield
+//
+// CHECK-NEXT: } step {
+// CHECK-NEXT: %[[ITR_LOAD:.*]] = cir.load %[[ITR]] : !cir.ptr<!u64i>, !u64i
+// CHECK-NEXT: %[[INC:.*]] = cir.unary(inc, %[[ITR_LOAD]]) : !u64i, !u64i
+// CHECK-NEXT: cir.store %[[INC]], %[[ITR]] : !u64i, !cir.ptr<!u64i>
+// CHECK-NEXT: cir.yield
+// CHECK-NEXT: }
+// CHECK-NEXT: }
 // TODO: Add Init here.
 // CHECK-NEXT: acc.yield
 // CHECK-NEXT: }


        


More information about the cfe-commits mailing list