Skip to content

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

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Sep 9, 2025
Merged

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

merged 1 commit into from
Sep 9, 2025
+192 −17

Conversation

razvanlupusoru
Copy link
Contributor

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.

[flang][acc] Fix incorrect loop body nesting and IV value use
f61f3c3 
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.
@llvmbot llvmbot added flang Flang issues not falling into any other category flang:fir-hlfir openacc labels Sep 9, 2025
@llvmbot
Copy link
Member

llvmbot commented Sep 9, 2025 •
edited
Loading

@llvm/pr-subscribers-flang-fir-hlfir

@llvm/pr-subscribers-openacc

Author: Razvan Lupusoru (razvanlupusoru)

Changes

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.

Patch is 21.56 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/157708.diff

3 Files Affected:

  • (modified) flang/lib/Lower/Bridge.cpp (+8)
  • (modified) flang/lib/Lower/OpenACC.cpp (+8-4)
  • (modified) flang/test/Lower/OpenACC/do-loops-to-acc-loops.f90 (+176-13)
diff --git a/flang/lib/Lower/Bridge.cpp b/flang/lib/Lower/Bridge.cpp
index e91fa2db15fa2..6125ea9153662 100644
--- a/flang/lib/Lower/Bridge.cpp
+++ b/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.
diff --git a/flang/lib/Lower/OpenACC.cpp b/flang/lib/Lower/OpenACC.cpp
index bbe749f8c8805..d8a0e4d8a8fa0 100644
--- a/flang/lib/Lower/OpenACC.cpp
+++ b/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,
diff --git a/flang/test/Lower/OpenACC/do-loops-to-acc-loops.f90 b/flang/test/Lower/OpenACC/do-loops-to-acc-loops.f90
index 5f8ea03d43a5d..a75a022690fcc 100644
--- a/flang/test/Lower/OpenACC/do-loops-to-acc-loops.f90
+++ b/flang/test/Lower/OpenACC/do-loops-to-acc-loops.f90
@@ -2,6 +2,7 @@
 ! Tests the new functionality that converts Fortran iteration constructs to acc.loop with proper IV handling.
 
 ! RUN: bbc -fopenacc -emit-hlfir %s -o - | FileCheck %s
+! RUN: bbc -fopenacc -emit-hlfir --openacc-do-loop-to-acc-loop=false %s -o - | FileCheck %s --check-prefix=CHECK-NOACCLOOP
 
 ! CHECK-LABEL: func.func @_QPbasic_do_loop
 subroutine basic_do_loop()
@@ -17,10 +18,19 @@ subroutine basic_do_loop()
   !$acc end kernels
 
 ! CHECK: acc.kernels {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: %[[PRIVATE_IV:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i"}
+! CHECK: %[[PRIVATE_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_IV]] {uniq_name = "_QFbasic_do_loopEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_IV]] : !fir.ref<i32>) control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
 ! CHECK: acc.yield
 ! CHECK: attributes {auto_ = [#acc.device_type<none>], inclusiveUpperbound = array<i1: true>}
 
+! CHECK-NOACCLOOP-LABEL: func.func @_QPbasic_do_loop
+! CHECK-NOACCLOOP: acc.kernels {
+! CHECK-NOACCLOOP-NOT: acc.loop
+
 end subroutine
 
 ! CHECK-LABEL: func.func @_QPbasic_do_concurrent
@@ -37,10 +47,19 @@ subroutine basic_do_concurrent()
   !$acc end kernels
 
 ! CHECK: acc.kernels {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: %[[PRIVATE_IV:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i"}
+! CHECK: %[[PRIVATE_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_IV]] {uniq_name = "_QFbasic_do_concurrentEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_IV]] : !fir.ref<i32>) control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
 ! CHECK: acc.yield
 ! CHECK: attributes {auto_ = [#acc.device_type<none>], inclusiveUpperbound = array<i1: true>}
 
+! CHECK-NOACCLOOP-LABEL: func.func @_QPbasic_do_concurrent
+! CHECK-NOACCLOOP: acc.kernels {
+! CHECK-NOACCLOOP-NOT: acc.loop
+
 end subroutine
 
 ! CHECK-LABEL: func.func @_QPbasic_do_loop_parallel
@@ -57,10 +76,19 @@ subroutine basic_do_loop_parallel()
   !$acc end parallel
 
 ! CHECK: acc.parallel {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: %[[PRIVATE_IV:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i"}
+! CHECK: %[[PRIVATE_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_IV]] {uniq_name = "_QFbasic_do_loop_parallelEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_IV]] : !fir.ref<i32>) control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
 ! CHECK: acc.yield
 ! CHECK: attributes {auto_ = [#acc.device_type<none>], inclusiveUpperbound = array<i1: true>}
 
+! CHECK-NOACCLOOP-LABEL: func.func @_QPbasic_do_loop_parallel
+! CHECK-NOACCLOOP: acc.parallel {
+! CHECK-NOACCLOOP-NOT: acc.loop
+
 end subroutine
 
 ! CHECK-LABEL: func.func @_QPbasic_do_loop_serial
@@ -77,10 +105,19 @@ subroutine basic_do_loop_serial()
   !$acc end serial
 
 ! CHECK: acc.serial {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: %[[PRIVATE_IV:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i"}
+! CHECK: %[[PRIVATE_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_IV]] {uniq_name = "_QFbasic_do_loop_serialEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_IV]] : !fir.ref<i32>) control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
 ! CHECK: acc.yield
 ! CHECK: attributes {inclusiveUpperbound = array<i1: true>, seq = [#acc.device_type<none>]}
 
+! CHECK-NOACCLOOP-LABEL: func.func @_QPbasic_do_loop_serial
+! CHECK-NOACCLOOP: acc.serial {
+! CHECK-NOACCLOOP-NOT: acc.loop
+
 end subroutine
 
 ! CHECK-LABEL: func.func @_QPbasic_do_concurrent_parallel
@@ -97,10 +134,19 @@ subroutine basic_do_concurrent_parallel()
   !$acc end parallel
 
 ! CHECK: acc.parallel {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: %[[PRIVATE_IV:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i"}
+! CHECK: %[[PRIVATE_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_IV]] {uniq_name = "_QFbasic_do_concurrent_parallelEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_IV]] : !fir.ref<i32>) control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
 ! CHECK: acc.yield
 ! CHECK: attributes {inclusiveUpperbound = array<i1: true>, independent = [#acc.device_type<none>]}
 
+! CHECK-NOACCLOOP-LABEL: func.func @_QPbasic_do_concurrent_parallel
+! CHECK-NOACCLOOP: acc.parallel {
+! CHECK-NOACCLOOP-NOT: acc.loop
+
 end subroutine
 
 ! CHECK-LABEL: func.func @_QPbasic_do_concurrent_serial
@@ -117,10 +163,19 @@ subroutine basic_do_concurrent_serial()
   !$acc end serial
 
 ! CHECK: acc.serial {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: %[[PRIVATE_IV:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i"}
+! CHECK: %[[PRIVATE_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_IV]] {uniq_name = "_QFbasic_do_concurrent_serialEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_IV]] : !fir.ref<i32>) control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
 ! CHECK: acc.yield
 ! CHECK: attributes {inclusiveUpperbound = array<i1: true>, seq = [#acc.device_type<none>]}
 
+! CHECK-NOACCLOOP-LABEL: func.func @_QPbasic_do_concurrent_serial
+! CHECK-NOACCLOOP: acc.serial {
+! CHECK-NOACCLOOP-NOT: acc.loop
+
 end subroutine
 
 ! CHECK-LABEL: func.func @_QPmulti_dimension_do_concurrent
@@ -137,9 +192,29 @@ subroutine multi_dimension_do_concurrent()
   !$acc end kernels
 
 ! CHECK: acc.kernels {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32, %{{.*}} : i32, %{{.*}} : i32) = (%c1{{.*}}, %c1{{.*}}, %c1{{.*}} : i32, i32, i32) to (%{{.*}}, %{{.*}}, %{{.*}} : i32, i32, i32) step (%c1{{.*}}, %c1{{.*}}, %c1{{.*}} : i32, i32, i32)
+! CHECK-DAG: %[[PRIVATE_I:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i"}
+! CHECK-DAG: %[[PRIVATE_J:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "j"}
+! CHECK-DAG: %[[PRIVATE_K:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "k"}
+! CHECK-DAG: %[[PRIVATE_I_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_I]] {uniq_name = "_QFmulti_dimension_do_concurrentEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK-DAG: %[[PRIVATE_J_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_J]] {uniq_name = "_QFmulti_dimension_do_concurrentEj"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK-DAG: %[[PRIVATE_K_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_K]] {uniq_name = "_QFmulti_dimension_do_concurrentEk"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_I]] : !fir.ref<i32>, @privatization_ref_i32 -> %[[PRIVATE_J]] : !fir.ref<i32>, @privatization_ref_i32 -> %[[PRIVATE_K]] : !fir.ref<i32>) control(%{{.*}} : i32, %{{.*}} : i32, %{{.*}} : i32) = (%c1{{.*}}, %c1{{.*}}, %c1{{.*}} : i32, i32, i32) to (%{{.*}}, %{{.*}}, %{{.*}} : i32, i32, i32) step (%c1{{.*}}, %c1{{.*}}, %c1{{.*}} : i32, i32, i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_I_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_J_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_K_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_I_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_J_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_K_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_I_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_J_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_K_DECLARE]]#0 : !fir.ref<i32>
 ! CHECK: acc.yield
 ! CHECK: attributes {auto_ = [#acc.device_type<none>], inclusiveUpperbound = array<i1: true, true, true>}
+
+! CHECK-NOACCLOOP-LABEL: func.func @_QPmulti_dimension_do_concurrent
+! CHECK-NOACCLOOP: acc.kernels {
+! CHECK-NOACCLOOP-NOT: acc.loop
+
 end subroutine
 
 
@@ -159,13 +234,27 @@ subroutine nested_do_loops()
   !$acc end kernels
 
 ! CHECK: acc.kernels {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK-DAG: %[[PRIVATE_I:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i"}
+! CHECK-DAG: %[[PRIVATE_I_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_I]] {uniq_name = "_QFnested_do_loopsEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_I]] : !fir.ref<i32>) control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_I_DECLARE]]#0 : !fir.ref<i32>
+! CHECK-DAG: %[[PRIVATE_J:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "j"}
+! CHECK-DAG: %[[PRIVATE_J_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_J]] {uniq_name = "_QFnested_do_loopsEj"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_J]] : !fir.ref<i32>) control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_J_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_I_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_J_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_I_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_J_DECLARE]]#0 : !fir.ref<i32>
 ! CHECK: acc.yield
 ! CHECK: attributes {auto_ = [#acc.device_type<none>], inclusiveUpperbound = array<i1: true>}
 ! CHECK: acc.yield
 ! CHECK: attributes {auto_ = [#acc.device_type<none>], inclusiveUpperbound = array<i1: true>}
 
+! CHECK-NOACCLOOP-LABEL: func.func @_QPnested_do_loops
+! CHECK-NOACCLOOP: acc.kernels {
+! CHECK-NOACCLOOP-NOT: acc.loop
+
 end subroutine
 
 ! CHECK-LABEL: func.func @_QPvariable_bounds_and_step
@@ -182,10 +271,19 @@ subroutine variable_bounds_and_step(n, start_val, step_val)
   !$acc end kernels
 
 ! CHECK: acc.kernels {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: %[[PRIVATE_IV:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i"}
+! CHECK: %[[PRIVATE_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_IV]] {uniq_name = "_QFvariable_bounds_and_stepEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_IV]] : !fir.ref<i32>) control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_DECLARE]]#0 : !fir.ref<i32>
 ! CHECK: acc.yield
 ! CHECK: attributes {auto_ = [#acc.device_type<none>], inclusiveUpperbound = array<i1: true>}
 
+! CHECK-NOACCLOOP-LABEL: func.func @_QPvariable_bounds_and_step
+! CHECK-NOACCLOOP: acc.kernels {
+! CHECK-NOACCLOOP-NOT: acc.loop
+
 end subroutine
 
 ! CHECK-LABEL: func.func @_QPdifferent_iv_types
@@ -216,11 +314,76 @@ subroutine different_iv_types()
   !$acc end kernels
 
 ! CHECK: acc.kernels {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i64) = (%{{.*}} : i64) to (%{{.*}} : i64) step (%{{.*}} : i64)
+! CHECK: %[[PRIVATE_I8:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i64>) -> !fir.ref<i64> {implicit = true, name = "i8"}
+! CHECK: %[[PRIVATE_I8_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_I8]] {uniq_name = "_QFdifferent_iv_typesEi8"} : (!fir.ref<i64>) -> (!fir.ref<i64>, !fir.ref<i64>)
+! CHECK: acc.loop private(@privatization_ref_i64 -> %[[PRIVATE_I8]] : !fir.ref<i64>) control(%{{.*}} : i64) = (%{{.*}} : i64) to (%{{.*}} : i64) step (%{{.*}} : i64)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_I8_DECLARE]]#0 : !fir.ref<i64>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_I8_DECLARE]]#0 : !fir.ref<i64>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_I8_DECLARE]]#0 : !fir.ref<i64>
 ! CHECK: acc.kernels {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: %[[PRIVATE_I4:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i4"}
+! CHECK: %[[PRIVATE_I4_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_I4]] {uniq_name = "_QFdifferent_iv_typesEi4"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_I4]] : !fir.ref<i32>) control(%{{.*}} : i32) = (%{{.*}} : i32) to (%{{.*}} : i32) step (%{{.*}} : i32)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_I4_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_I4_DECLARE]]#0 : !fir.ref<i32>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_I4_DECLARE]]#0 : !fir.ref<i32>
 ! CHECK: acc.kernels {
-! CHECK: acc.loop {{.*}} control(%{{.*}} : i16) = (%{{.*}} : i16) to (%{{.*}} : i16) step (%{{.*}} : i16)
+! CHECK: %[[PRIVATE_I2:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i16>) -> !fir.ref<i16> {implicit = true, name = "i2"}
+! CHECK: %[[PRIVATE_I2_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_I2]] {uniq_name = "_QFdifferent_iv_typesEi2"} : (!fir.ref<i16>) -> (!fir.ref<i16>, !fir.ref<i16>)
+! CHECK: acc.loop private(@privatization_ref_i16 -> %[[PRIVATE_I2]] : !fir.ref<i16>) control(%{{.*}} : i16) = (%{{.*}} : i16) to (%{{.*}} : i16) step (%{{.*}} : i16)
+! CHECK: fir.store %{{.*}} to %[[PRIVATE_I2_DECLARE]]#0 : !fir.ref<i16>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_I2_DECLARE]]#0 : !fir.ref<i16>
+! CHECK: %{{.*}} = fir.load %[[PRIVATE_I2_DECLARE]]#0 : !fir.ref<i16>
+
+! CHECK-NOACCLOOP-LABEL: func.func @_QPdifferent_iv_types
+! CHECK-NOACCLOOP: acc.kernels {
+! CHECK-NOACCLOOP-NOT: acc.loop
+
+end subroutine
+
+! CHECK-LABEL: func.func @_QPnested_loop_with_reduction
+subroutine nested_loop_with_reduction(x, y)
+  integer :: x, y
+  integer :: i, j
+
+  ! Nested loop with reduction variables - check that reduction operations
+  ! are correctly scoped (outer loop reduction should not be inside inner loop)
+  !$acc parallel
+  !$acc loop reduction(+:x,y)
+  do i = 1, 10
+    do j = 1, 20
+      y = y + 1
+    end do
+    x = x + 1
+  end do
+  !$acc end parallel
+
+! CHECK: acc.parallel {
+! CHECK: %[[REDUCTION_X:.*]] = acc.reduction varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {name = "x"}
+! CHECK: %[[REDUCTION_Y:.*]] = acc.reduction varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {name = "y"}
+! CHECK: %[[PRIVATE_I:.*]] = acc.private varPtr(%{{.*}} : !fir.ref<i32>) -> !fir.ref<i32> {implicit = true, name = "i"}
+! CHECK: %[[PRIVATE_I_DECLARE:.*]]:2 = hlfir.declare %[[PRIVATE_I]] {uniq_name = "_QFnested_loop_with_reductionEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: acc.loop private(@privatization_ref_i32 -> %[[PRIVATE_I]] : !fir.ref<i32>) re...
[truncated]

clementval
clementval approved these changes Sep 9, 2025
View reviewed changes
Copy link
Contributor

@clementval clementval left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@razvanlupusoru razvanlupusoru merged commit 2a2296b into llvm:main Sep 9, 2025
13 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@clementval clementval clementval approved these changes

@rscottmanley rscottmanley Awaiting requested review from rscottmanley

Assignees
No one assigned
Labels
flang:fir-hlfir flang Flang issues not falling into any other category openacc
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@razvanlupusoru @llvmbot @clementval