[Mlir-commits] [llvm] [mlir] [mlir] `HasParallelRegion` means region may be parallel (PR #231372)
Matthias Springer
llvmlistbot at llvm.org
Tue Oct 13 20:06:54 PDT 2026
https://github.com/matthias-springer updated https://github.com/llvm/llvm-project/pull/231372
>From db36cdeb065a0cdd19b429812fb60e186de5064d Mon Sep 17 00:00:00 2001
From: Matthias Springer <me at m-sp.org>
Date: Tue, 13 Oct 2026 05:56:32 +0000
Subject: [PATCH 1/2] [mlir] `HasParallelRegion` means region may be parallel
---
mlir/include/mlir/Dialect/GPU/IR/GPUDialect.h | 1 +
mlir/include/mlir/Dialect/GPU/IR/GPUOps.td | 3 +-
.../mlir/Dialect/OpenACC/OpenACCCGOps.td | 2 +-
.../mlir/Dialect/OpenACC/OpenACCOps.td | 6 +-
.../mlir/Dialect/OpenMP/OpenMPDialect.h | 1 +
mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td | 41 ++++++------
.../mlir/Interfaces/LoopLikeInterface.h | 10 +--
.../mlir/Interfaces/LoopLikeInterface.td | 2 +-
mlir/lib/Dialect/GPU/CMakeLists.txt | 1 +
mlir/lib/Dialect/OpenMP/IR/CMakeLists.txt | 1 +
.../Transforms/buffer-loop-hoisting.mlir | 64 +++++++++++++++++++
.../llvm-project-overlay/mlir/BUILD.bazel | 4 ++
12 files changed, 107 insertions(+), 29 deletions(-)
diff --git a/mlir/include/mlir/Dialect/GPU/IR/GPUDialect.h b/mlir/include/mlir/Dialect/GPU/IR/GPUDialect.h
index 038f7e8763ee0..0d6ea9024c842 100644
--- a/mlir/include/mlir/Dialect/GPU/IR/GPUDialect.h
+++ b/mlir/include/mlir/Dialect/GPU/IR/GPUDialect.h
@@ -28,6 +28,7 @@
#include "mlir/Interfaces/FunctionInterfaces.h"
#include "mlir/Interfaces/InferIntRangeInterface.h"
#include "mlir/Interfaces/InferTypeOpInterface.h"
+#include "mlir/Interfaces/LoopLikeInterface.h"
#include "mlir/Interfaces/SideEffectInterfaces.h"
#include "llvm/ADT/STLExtras.h"
diff --git a/mlir/include/mlir/Dialect/GPU/IR/GPUOps.td b/mlir/include/mlir/Dialect/GPU/IR/GPUOps.td
index 1b6c4ec26ab68..ab5c9e0f47660 100644
--- a/mlir/include/mlir/Dialect/GPU/IR/GPUOps.td
+++ b/mlir/include/mlir/Dialect/GPU/IR/GPUOps.td
@@ -29,6 +29,7 @@ include "mlir/IR/OpAsmInterface.td"
include "mlir/Interfaces/FunctionInterfaces.td"
include "mlir/Interfaces/InferIntRangeInterface.td"
include "mlir/Interfaces/InferTypeOpInterface.td"
+include "mlir/Interfaces/LoopLikeInterface.td"
include "mlir/Interfaces/SideEffectInterfaces.td"
//===----------------------------------------------------------------------===//
@@ -811,7 +812,7 @@ def GPU_LaunchFuncOp :GPU_Op<"launch_func", [
def GPU_LaunchOp : GPU_Op<"launch", [
AffineScope, AutomaticAllocationScope, AttrSizedOperandSegments,
DeclareOpInterfaceMethods<InferIntRangeInterface, ["inferResultRanges"]>,
- GPU_AsyncOpInterface, RecursiveMemoryEffects]>,
+ GPU_AsyncOpInterface, HasParallelRegion, RecursiveMemoryEffects]>,
Arguments<(ins Variadic<GPU_AsyncToken>:$asyncDependencies,
Index:$gridSizeX, Index:$gridSizeY, Index:$gridSizeZ,
Index:$blockSizeX, Index:$blockSizeY, Index:$blockSizeZ,
diff --git a/mlir/include/mlir/Dialect/OpenACC/OpenACCCGOps.td b/mlir/include/mlir/Dialect/OpenACC/OpenACCCGOps.td
index 11efab595106e..077c1b7a0db17 100644
--- a/mlir/include/mlir/Dialect/OpenACC/OpenACCCGOps.td
+++ b/mlir/include/mlir/Dialect/OpenACC/OpenACCCGOps.td
@@ -559,7 +559,7 @@ def OpenACC_GPUAsyncTokenType : Type<
def OpenACC_ComputeRegionOp
: OpenACC_Op<"compute_region",
[OffloadRegionOpInterface, AffineScope,
- RecursiveMemoryEffects,
+ HasParallelRegion, RecursiveMemoryEffects,
SingleBlockImplicitTerminator<"YieldOp">,
IsolatedFromAbove, AttrSizedOperandSegments]> {
let summary = "Compute region for GPU execution";
diff --git a/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td b/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
index 70f8ee5c38172..b90abbe61dc09 100644
--- a/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
+++ b/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
@@ -1783,7 +1783,7 @@ def OpenACC_ReductionRecipeOp
def OpenACC_ParallelOp
: OpenACC_Op<"parallel",
[AttrSizedOperandSegments, AutomaticAllocationScope,
- RecursiveMemoryEffects,
+ HasParallelRegion, RecursiveMemoryEffects,
DeclareOpInterfaceMethods<ComputeRegionOpInterface>,
DeclareOpInterfaceMethods<RegionBranchOpInterface,
["getSuccessorInputs"]>,
@@ -2144,7 +2144,7 @@ def OpenACC_SerialOp
def OpenACC_KernelsOp
: OpenACC_Op<"kernels",
[AttrSizedOperandSegments, AutomaticAllocationScope,
- RecursiveMemoryEffects,
+ HasParallelRegion, RecursiveMemoryEffects,
DeclareOpInterfaceMethods<ComputeRegionOpInterface>,
DeclareOpInterfaceMethods<RegionBranchOpInterface,
["getSuccessorInputs"]>,
@@ -2665,7 +2665,7 @@ def OpenACC_HostDataOp
def OpenACC_LoopOp
: OpenACC_Op<
"loop", [AttrSizedOperandSegments, AutomaticAllocationScope,
- RecursiveMemoryEffects,
+ HasParallelRegion, RecursiveMemoryEffects,
DeclareOpInterfaceMethods<ComputeRegionOpInterface>,
DeclareOpInterfaceMethods<LoopLikeOpInterface>,
DeclareOpInterfaceMethods<RegionBranchOpInterface,
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPDialect.h b/mlir/include/mlir/Dialect/OpenMP/OpenMPDialect.h
index a6854752f75b6..6aed702a2da34 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPDialect.h
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPDialect.h
@@ -23,6 +23,7 @@
#include "mlir/IR/PatternMatch.h"
#include "mlir/IR/SymbolTable.h"
#include "mlir/Interfaces/ControlFlowInterfaces.h"
+#include "mlir/Interfaces/LoopLikeInterface.h"
#include "mlir/Interfaces/SideEffectInterfaces.h"
#include "llvm/Frontend/OpenMP/OMPConstants.h"
#include "llvm/Frontend/OpenMP/OMPDeviceConstants.h"
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 6bc99cf765699..576f67765073a 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -20,6 +20,7 @@ include "mlir/Dialect/OpenACCMPCommon/Interfaces/OpenACCMPOpsInterfaces.td"
include "mlir/Dialect/OpenMP/OpenMPClauses.td"
include "mlir/Dialect/OpenMP/OpenMPOpBase.td"
include "mlir/Interfaces/ControlFlowInterfaces.td"
+include "mlir/Interfaces/LoopLikeInterface.td"
include "mlir/Interfaces/SideEffectInterfaces.td"
include "mlir/IR/EnumAttr.td"
include "mlir/IR/OpAsmInterface.td"
@@ -192,7 +193,7 @@ def PrivateClauseOp
//===----------------------------------------------------------------------===//
def ParallelOp : OpenMP_Op<"parallel", traits = [
- AttrSizedOperandSegments, AutomaticAllocationScope,
+ AttrSizedOperandSegments, AutomaticAllocationScope, HasParallelRegion,
DeclareOpInterfaceMethods<ComposableOpInterface>,
DeclareOpInterfaceMethods<OutlineableOpenMPOpInterface>,
RecursiveMemoryEffects
@@ -242,7 +243,7 @@ def TerminatorOp : OpenMP_Op<"terminator", [Terminator, Pure]> {
//===----------------------------------------------------------------------===//
def TeamsOp : OpenMP_Op<"teams", traits = [
AttrSizedOperandSegments, DeclareOpInterfaceMethods<ComposableOpInterface>,
- RecursiveMemoryEffects, OutlineableOpenMPOpInterface
+ HasParallelRegion, RecursiveMemoryEffects, OutlineableOpenMPOpInterface
], clauses = [
OpenMP_AllocateClause, OpenMP_DynGroupprivateClause, OpenMP_IfClause,
OpenMP_NumTeamsClause, OpenMP_PrivateClause, OpenMP_ReductionClause,
@@ -298,7 +299,8 @@ def SectionOp : OpenMP_Op<"section", traits = [
}
def SectionsOp : OpenMP_Op<"sections", traits = [
- AttrSizedOperandSegments, DeclareOpInterfaceMethods<ComposableOpInterface>
+ AttrSizedOperandSegments, DeclareOpInterfaceMethods<ComposableOpInterface>,
+ HasParallelRegion
], clauses = [
OpenMP_AllocateClause, OpenMP_NowaitClause, OpenMP_PrivateClause,
OpenMP_ReductionClause
@@ -682,7 +684,8 @@ def FuseOp
//===----------------------------------------------------------------------===//
def WorkshareOp : OpenMP_Op<"workshare", traits = [
- DeclareOpInterfaceMethods<ComposableOpInterface>, RecursiveMemoryEffects
+ DeclareOpInterfaceMethods<ComposableOpInterface>, HasParallelRegion,
+ RecursiveMemoryEffects
], clauses = [
OpenMP_NowaitClause,
], singleRegion = true> {
@@ -706,7 +709,7 @@ def WorkshareOp : OpenMP_Op<"workshare", traits = [
def WorkshareLoopWrapperOp : OpenMP_Op<"workshare.loop_wrapper", traits = [
DeclareOpInterfaceMethods<LoopWrapperInterface>, HasAncestor<"WorkshareOp">,
- NoTerminator, RecursiveMemoryEffects, SingleBlock
+ HasParallelRegion, NoTerminator, RecursiveMemoryEffects, SingleBlock
], singleRegion = true> {
let summary = "contains loop nests to be parallelized by workshare";
let description = [{
@@ -726,7 +729,7 @@ def WorkshareLoopWrapperOp : OpenMP_Op<"workshare.loop_wrapper", traits = [
//===----------------------------------------------------------------------===//
def LoopNestOp : OpenMP_Op<"loop_nest", traits = [
- RecursiveMemoryEffects, SameVariadicOperandSize
+ HasParallelRegion, RecursiveMemoryEffects, SameVariadicOperandSize
], clauses = [
OpenMP_CollapseClause,
OpenMP_LoopRelatedClause,
@@ -797,7 +800,7 @@ def LoopNestOp : OpenMP_Op<"loop_nest", traits = [
def LoopOp : OpenMP_Op<"loop", traits = [
AttrSizedOperandSegments, DeclareOpInterfaceMethods<LoopWrapperInterface>,
- NoTerminator, SingleBlock
+ HasParallelRegion, NoTerminator, SingleBlock
], clauses = [
OpenMP_BindClause, OpenMP_PrivateClause, OpenMP_OrderClause,
OpenMP_ReductionClause
@@ -845,8 +848,8 @@ def LoopOp : OpenMP_Op<"loop", traits = [
def WsloopOp : OpenMP_Op<"wsloop", traits = [
AttrSizedOperandSegments, DeclareOpInterfaceMethods<ComposableOpInterface>,
- DeclareOpInterfaceMethods<LoopWrapperInterface>, NoTerminator,
- RecursiveMemoryEffects, SingleBlock
+ DeclareOpInterfaceMethods<LoopWrapperInterface>, HasParallelRegion,
+ NoTerminator, RecursiveMemoryEffects, SingleBlock
], clauses = [
OpenMP_AllocateClause, OpenMP_LinearClause, OpenMP_NowaitClause,
OpenMP_OrderClause, OpenMP_OrderedClause, OpenMP_PrivateClause,
@@ -897,8 +900,8 @@ def WsloopOp : OpenMP_Op<"wsloop", traits = [
def SimdOp : OpenMP_Op<"simd", traits = [
AttrSizedOperandSegments, DeclareOpInterfaceMethods<ComposableOpInterface>,
- DeclareOpInterfaceMethods<LoopWrapperInterface>, NoTerminator,
- RecursiveMemoryEffects, SingleBlock
+ DeclareOpInterfaceMethods<LoopWrapperInterface>, HasParallelRegion,
+ NoTerminator, RecursiveMemoryEffects, SingleBlock
], clauses = [
OpenMP_AlignedClause, OpenMP_IfClause, OpenMP_LinearClause,
OpenMP_NontemporalClause, OpenMP_OrderClause, OpenMP_PrivateClause,
@@ -972,8 +975,8 @@ def YieldOp
//===----------------------------------------------------------------------===//
def DistributeOp : OpenMP_Op<"distribute", traits = [
AttrSizedOperandSegments, DeclareOpInterfaceMethods<ComposableOpInterface>,
- DeclareOpInterfaceMethods<LoopWrapperInterface>, NoTerminator,
- RecursiveMemoryEffects, SingleBlock
+ DeclareOpInterfaceMethods<LoopWrapperInterface>, HasParallelRegion,
+ NoTerminator, RecursiveMemoryEffects, SingleBlock
], clauses = [
OpenMP_AllocateClause, OpenMP_DistScheduleClause, OpenMP_OrderClause,
OpenMP_PrivateClause
@@ -1075,7 +1078,7 @@ def TaskOp : OpenMP_Op<"task", traits = [
// SingleBlock restriction is just to ease implementation. It would be okay to
// relax this.
def TaskloopContextOp : OpenMP_Op<"taskloop.context", traits = [
- AttrSizedOperandSegments, AutomaticAllocationScope,
+ AttrSizedOperandSegments, AutomaticAllocationScope, HasParallelRegion,
RecursiveMemoryEffects, SingleBlock,
DeclareOpInterfaceMethods<ComposableOpInterface>,
DeclareOpInterfaceMethods<OutlineableOpenMPOpInterface>
@@ -1174,8 +1177,8 @@ def TaskloopContextOp : OpenMP_Op<"taskloop.context", traits = [
def TaskloopWrapperOp : OpenMP_Op<"taskloop.wrapper", traits = [
DeclareOpInterfaceMethods<ComposableOpInterface>,
- DeclareOpInterfaceMethods<LoopWrapperInterface>, NoTerminator,
- RecursiveMemoryEffects, SingleBlock
+ DeclareOpInterfaceMethods<LoopWrapperInterface>, HasParallelRegion,
+ NoTerminator, RecursiveMemoryEffects, SingleBlock
], clauses = [
// See taskloop.context, which generates the runtime calls and outlined
// task function.
@@ -1655,8 +1658,8 @@ def TargetUpdateOp: OpenMP_Op<"target_update", traits = [
def TargetOp : OpenMP_Op<"target", traits = [
AttrSizedOperandSegments, BlockArgOpenMPOpInterface,
- DeclareOpInterfaceMethods<ComposableOpInterface>, IsolatedFromAbove,
- OutlineableOpenMPOpInterface
+ DeclareOpInterfaceMethods<ComposableOpInterface>, HasParallelRegion,
+ IsolatedFromAbove, OutlineableOpenMPOpInterface
], clauses = [
// TODO: Complete clause list (defaultmap, uses_allocators).
OpenMP_AllocateClause, OpenMP_DependClause, OpenMP_DeviceClause,
@@ -2654,7 +2657,7 @@ def FreeSharedMemOp : OpenMP_Op<"free_shared_mem", traits = [
//===----------------------------------------------------------------------===//
def WorkdistributeOp : OpenMP_Op<"workdistribute", traits = [
- DeclareOpInterfaceMethods<ComposableOpInterface>
+ DeclareOpInterfaceMethods<ComposableOpInterface>, HasParallelRegion
]> {
let summary = "workdistribute directive";
let description = [{
diff --git a/mlir/include/mlir/Interfaces/LoopLikeInterface.h b/mlir/include/mlir/Interfaces/LoopLikeInterface.h
index 9925fc6ce6ca9..1ccfcef7e21e5 100644
--- a/mlir/include/mlir/Interfaces/LoopLikeInterface.h
+++ b/mlir/include/mlir/Interfaces/LoopLikeInterface.h
@@ -35,10 +35,12 @@ LogicalResult verifyLoopLikeOpInterface(Operation *op);
//===----------------------------------------------------------------------===//
namespace OpTrait {
-// A trait indicating that the single region contained in the operation has
-// parallel execution semantics. This may have implications in a certain pass.
-// For example, buffer hoisting is illegal in parallel loops, and local buffers
-// may be accessed by parallel threads simultaneously.
+// A trait indicating that the single region contained in the operation may have
+// parallel execution semantics. Whether it executes in parallel may depend on
+// the operation's attributes or enclosing context. Transformations must account
+// for possible parallel execution unless they can prove sequential execution.
+// For example, hoisting a buffer allocation out of the region may cause parallel
+// executions to share a buffer that was previously private.
template <typename ConcreteType>
class HasParallelRegion : public TraitBase<ConcreteType, HasParallelRegion> {
public:
diff --git a/mlir/include/mlir/Interfaces/LoopLikeInterface.td b/mlir/include/mlir/Interfaces/LoopLikeInterface.td
index 0bb82c1e4638e..15ec7ab87438e 100644
--- a/mlir/include/mlir/Interfaces/LoopLikeInterface.td
+++ b/mlir/include/mlir/Interfaces/LoopLikeInterface.td
@@ -487,7 +487,7 @@ def LoopLikeOpInterface : OpInterface<"LoopLikeOpInterface"> {
// Traits
//===----------------------------------------------------------------------===//
-// Op contains a region with parallel execution semantics
+// Op contains a region that may have parallel execution semantics.
def HasParallelRegion : NativeOpTrait<"HasParallelRegion">;
#endif // MLIR_INTERFACES_LOOPLIKEINTERFACE
diff --git a/mlir/lib/Dialect/GPU/CMakeLists.txt b/mlir/lib/Dialect/GPU/CMakeLists.txt
index 6726bf0cab1d2..ae7ed6f4257b5 100644
--- a/mlir/lib/Dialect/GPU/CMakeLists.txt
+++ b/mlir/lib/Dialect/GPU/CMakeLists.txt
@@ -24,6 +24,7 @@ add_mlir_dialect_library(MLIRGPUDialect
MLIRFunctionInterfaces
MLIRInferIntRangeInterface
MLIRIR
+ MLIRLoopLikeInterface
MLIRMathDialect
MLIRMemRefDialect
MLIRSideEffectInterfaces
diff --git a/mlir/lib/Dialect/OpenMP/IR/CMakeLists.txt b/mlir/lib/Dialect/OpenMP/IR/CMakeLists.txt
index 1beea2098d3bb..5edb7581feb76 100644
--- a/mlir/lib/Dialect/OpenMP/IR/CMakeLists.txt
+++ b/mlir/lib/Dialect/OpenMP/IR/CMakeLists.txt
@@ -17,5 +17,6 @@ add_mlir_dialect_library(MLIROpenMPDialect
MLIRIR
MLIRLLVMDialect
MLIRFuncDialect
+ MLIRLoopLikeInterface
MLIROpenACCMPCommon
)
diff --git a/mlir/test/Dialect/Bufferization/Transforms/buffer-loop-hoisting.mlir b/mlir/test/Dialect/Bufferization/Transforms/buffer-loop-hoisting.mlir
index 7498f7af8c5d1..7e4e5540bc483 100644
--- a/mlir/test/Dialect/Bufferization/Transforms/buffer-loop-hoisting.mlir
+++ b/mlir/test/Dialect/Bufferization/Transforms/buffer-loop-hoisting.mlir
@@ -546,6 +546,70 @@ func.func @hoist_alloca(
// -----
+// Each parallel OpenACC iteration must retain its own allocation.
+// CHECK-LABEL: func @no_hoist_acc_parallel_loop(
+// CHECK-NOT: memref.alloc
+// CHECK: acc.parallel
+// CHECK-NOT: memref.alloc
+// CHECK: acc.loop gang vector
+// CHECK-NEXT: %[[ALLOC:.*]] = memref.alloc()
+// CHECK-NEXT: memref.store {{.*}}, %[[ALLOC]]
+// CHECK-NEXT: %[[VALUE:.*]] = memref.load %[[ALLOC]]
+// CHECK-NEXT: memref.store %[[VALUE]]
+func.func @no_hoist_acc_parallel_loop(%out: memref<2xindex>) {
+ %c0 = arith.constant 0 : index
+ %c1 = arith.constant 1 : index
+ %c2 = arith.constant 2 : index
+ acc.parallel {
+ acc.loop gang vector control(%i : index) = (%c0 : index)
+ to (%c2 : index) step (%c1 : index) {
+ %buffer = memref.alloc() : memref<1xindex>
+ memref.store %i, %buffer[%c0] : memref<1xindex>
+ %value = memref.load %buffer[%c0] : memref<1xindex>
+ memref.store %value, %out[%i] : memref<2xindex>
+ acc.yield
+ } independent
+ acc.yield
+ }
+ return
+}
+
+// -----
+
+// Hoist out of an inner sequential loop, but keep the allocation private to
+// each parallel OpenACC iteration.
+// CHECK-LABEL: func @hoist_within_acc_parallel_loop(
+// CHECK-NOT: memref.alloc
+// CHECK: acc.parallel
+// CHECK-NOT: memref.alloc
+// CHECK: acc.loop gang vector
+// CHECK-NEXT: %[[ALLOC:.*]] = memref.alloc()
+// CHECK-NEXT: scf.for
+// CHECK-NEXT: memref.store {{.*}}, %[[ALLOC]]
+// CHECK-NEXT: %[[VALUE:.*]] = memref.load %[[ALLOC]]
+// CHECK-NEXT: memref.store %[[VALUE]]
+func.func @hoist_within_acc_parallel_loop(%out: memref<2x2xindex>) {
+ %c0 = arith.constant 0 : index
+ %c1 = arith.constant 1 : index
+ %c2 = arith.constant 2 : index
+ acc.parallel {
+ acc.loop gang vector control(%i : index) = (%c0 : index)
+ to (%c2 : index) step (%c1 : index) {
+ scf.for %j = %c0 to %c2 step %c1 {
+ %buffer = memref.alloc() : memref<1xindex>
+ memref.store %j, %buffer[%c0] : memref<1xindex>
+ %value = memref.load %buffer[%c0] : memref<1xindex>
+ memref.store %value, %out[%i, %j] : memref<2x2xindex>
+ }
+ acc.yield
+ } independent
+ acc.yield
+ }
+ return
+}
+
+// -----
+
// A nested entry block can be reachable even when its enclosing block is not.
// CHECK-LABEL: func @loop_unreachable_parent(
// CHECK: return
diff --git a/utils/bazel/llvm-project-overlay/mlir/BUILD.bazel b/utils/bazel/llvm-project-overlay/mlir/BUILD.bazel
index be52547a9da73..61b2fb788defa 100644
--- a/utils/bazel/llvm-project-overlay/mlir/BUILD.bazel
+++ b/utils/bazel/llvm-project-overlay/mlir/BUILD.bazel
@@ -5713,6 +5713,7 @@ td_library(
":FunctionInterfacesTdFiles",
":InferIntRangeInterfaceTdFiles",
":LLVMOpsTdFiles",
+ ":LoopLikeInterfaceTdFiles",
":OpBaseTdFiles",
":SCFTdFiles",
":SideEffectInterfacesTdFiles",
@@ -5806,6 +5807,7 @@ cc_library(
":InferIntRangeInterface",
":InferTypeOpInterface",
":InliningUtils",
+ ":LoopLikeInterface",
":MathDialect",
":MemRefDialect",
":SCFDialect",
@@ -10966,6 +10968,7 @@ td_library(
deps = [
":AtomicInterfacesTdFiles",
":LLVMOpsTdFiles",
+ ":LoopLikeInterfaceTdFiles",
":OpBaseTdFiles",
],
)
@@ -11060,6 +11063,7 @@ cc_library(
":FuncDialect",
":IR",
":LLVMDialect",
+ ":LoopLikeInterface",
":OpenACCMPOpsInterfacesIncGen",
":OpenMPInterfacesIncGen",
":OpenMPOpsIncGen",
>From 708d1adfc789b811d22b71bd836f63e440909c87 Mon Sep 17 00:00:00 2001
From: Matthias Springer <me at m-sp.org>
Date: Wed, 14 Oct 2026 03:06:23 +0000
Subject: [PATCH 2/2] address comments
---
mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td | 2 +-
mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td | 4 ++--
mlir/include/mlir/Interfaces/LoopLikeInterface.h | 5 +++--
3 files changed, 6 insertions(+), 5 deletions(-)
diff --git a/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td b/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
index b90abbe61dc09..53f42a3e8ab44 100644
--- a/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
+++ b/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
@@ -2144,7 +2144,7 @@ def OpenACC_SerialOp
def OpenACC_KernelsOp
: OpenACC_Op<"kernels",
[AttrSizedOperandSegments, AutomaticAllocationScope,
- HasParallelRegion, RecursiveMemoryEffects,
+ RecursiveMemoryEffects,
DeclareOpInterfaceMethods<ComputeRegionOpInterface>,
DeclareOpInterfaceMethods<RegionBranchOpInterface,
["getSuccessorInputs"]>,
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 576f67765073a..23d8198fa3a73 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -1658,8 +1658,8 @@ def TargetUpdateOp: OpenMP_Op<"target_update", traits = [
def TargetOp : OpenMP_Op<"target", traits = [
AttrSizedOperandSegments, BlockArgOpenMPOpInterface,
- DeclareOpInterfaceMethods<ComposableOpInterface>, HasParallelRegion,
- IsolatedFromAbove, OutlineableOpenMPOpInterface
+ DeclareOpInterfaceMethods<ComposableOpInterface>, IsolatedFromAbove,
+ OutlineableOpenMPOpInterface
], clauses = [
// TODO: Complete clause list (defaultmap, uses_allocators).
OpenMP_AllocateClause, OpenMP_DependClause, OpenMP_DeviceClause,
diff --git a/mlir/include/mlir/Interfaces/LoopLikeInterface.h b/mlir/include/mlir/Interfaces/LoopLikeInterface.h
index 1ccfcef7e21e5..66f17d3c1d5d8 100644
--- a/mlir/include/mlir/Interfaces/LoopLikeInterface.h
+++ b/mlir/include/mlir/Interfaces/LoopLikeInterface.h
@@ -39,8 +39,9 @@ namespace OpTrait {
// parallel execution semantics. Whether it executes in parallel may depend on
// the operation's attributes or enclosing context. Transformations must account
// for possible parallel execution unless they can prove sequential execution.
-// For example, hoisting a buffer allocation out of the region may cause parallel
-// executions to share a buffer that was previously private.
+// For example, hoisting a buffer allocation out of the region may cause
+// parallel executions to share a buffer that was previously private.
+// TODO: Turn this trait into an op interface.
template <typename ConcreteType>
class HasParallelRegion : public TraitBase<ConcreteType, HasParallelRegion> {
public:
More information about the Mlir-commits
mailing list