[flang][acc] Fix incorrect loop body nesting and IV value use (#157708)

Two issues are being resolved:
- Incorrect loop body nesting caused by insertion point not being
updated after the loop. The scenario is now being tested through
`nested_do_loops` function in the test.
- Incorrect IV ssa values due to incorrect handling of scoping.

Additionally, this also adds `--openacc-do-loop-to-acc-loop` flag so
that the implicit conversion can be disabled for testing.

Author: razvanlupusoru

flang/lib/Lower/Bridge.cpp

@@ -72,6 +72,7 @@
 #include "mlir/Parser/Parser.h"
 #include "mlir/Support/StateStack.h"
 #include "mlir/Transforms/RegionUtils.h"
+#include "llvm/ADT/ScopeExit.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringSet.h"
 #include "llvm/Support/CommandLine.h"
@@ -2198,6 +2199,11 @@ class FirConverter : public Fortran::lower::AbstractConverter {
     // Loops with induction variables inside OpenACC compute constructs
     // need special handling to ensure that the IVs are privatized.
     if (Fortran::lower::isInsideOpenACCComputeConstruct(*builder)) {
+      // Open up a new scope for the loop variables.
+      localSymbols.pushScope();
+      auto scopeGuard =
+          llvm::make_scope_exit([&]() { localSymbols.popScope(); });
+
       mlir::Operation *loopOp = Fortran::lower::genOpenACCLoopFromDoConstruct(
           *this, bridge.getSemanticsContext(), localSymbols, doConstruct, eval);
       bool success = loopOp != nullptr;
@@ -2214,6 +2220,8 @@ class FirConverter : public Fortran::lower::AbstractConverter {
         for (auto end = --eval.getNestedEvaluations().end(); iter != end;
              ++iter)
           genFIR(*iter, unstructuredContext);
+
+        builder->setInsertionPointAfter(loopOp);
         return;
       }
       // Fall back to normal loop handling.

flang/lib/Lower/OpenACC.cpp

@@ -61,6 +61,11 @@ static llvm::cl::opt<bool> strideIncludeLowerExtent(
         "Whether to include the lower dimensions extents in the stride."),
     llvm::cl::init(true));
 
+static llvm::cl::opt<bool> lowerDoLoopToAccLoop(
+    "openacc-do-loop-to-acc-loop",
+    llvm::cl::desc("Whether to lower do loops as `acc.loop` operations."),
+    llvm::cl::init(true));
+
 // Special value for * passed in device_type or gang clauses.
 static constexpr std::int64_t starCst = -1;
 
@@ -5005,6 +5010,9 @@ mlir::Operation *Fortran::lower::genOpenACCLoopFromDoConstruct(
     Fortran::semantics::SemanticsContext &semanticsContext,
     Fortran::lower::SymMap &localSymbols,
     const Fortran::parser::DoConstruct &doConstruct, pft::Evaluation &eval) {
+  if (!lowerDoLoopToAccLoop)
+    return nullptr;
+
   // Only convert loops which have induction variables that need privatized.
   if (!doConstruct.IsDoNormal() && !doConstruct.IsDoConcurrent())
     return nullptr;
@@ -5027,10 +5035,6 @@ mlir::Operation *Fortran::lower::genOpenACCLoopFromDoConstruct(
     return nullptr;
   }
 
-  // Open up a new scope for the loop variables.
-  localSymbols.pushScope();
-  auto scopeGuard = llvm::make_scope_exit([&]() { localSymbols.popScope(); });
-
   // Prepare empty operand vectors since there are no associated `acc loop`
   // clauses with the Fortran do loops being handled here.
   llvm::SmallVector<mlir::Value> privateOperands, gangOperands,