[llvm-branch-commits] [flang] [mlir] [OpenMP][MLIR] Add thread_limit with dims modifier support (PR #171825)
via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Tue Dec 16 19:41:35 PST 2025
https://github.com/skc7 updated https://github.com/llvm/llvm-project/pull/171825
>From 1bb03fd2b7580f3a8c35bb37857a982947eb3f0a Mon Sep 17 00:00:00 2001
From: skc7 <Krishna.Sankisa at amd.com>
Date: Tue, 16 Dec 2025 13:15:43 +0530
Subject: [PATCH 1/4] allow one dimsValues to be present for verifier
---
mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
index 5d2d76f2ddb34..f9dcb48c7f08b 100644
--- a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
+++ b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
@@ -2223,7 +2223,7 @@ static LogicalResult verifyDimsModifier(Operation *op,
}
return success();
} else {
- if (!dimsValues.empty()) {
+ if (dimsValues.size() > 1) {
return op->emitError(
"dims values can only be specified with dims modifier");
}
>From fa606008ce25bda40c0000ee10aa1047098bee3b Mon Sep 17 00:00:00 2001
From: skc7 <Krishna.Sankisa at amd.com>
Date: Thu, 11 Dec 2025 13:35:05 +0530
Subject: [PATCH 2/4] [OpenMP][MLIR] Add thread_limit with dims modifier
support
---
.../Optimizer/OpenMP/LowerWorkdistribute.cpp | 16 +-
.../mlir/Dialect/OpenMP/OpenMPClauses.td | 29 +++-
mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp | 72 ++++++++-
.../OpenMP/OpenMPToLLVMIRTranslation.cpp | 8 +
mlir/test/Dialect/OpenMP/invalid.mlir | 149 +++++++++++++++++-
mlir/test/Dialect/OpenMP/ops.mlir | 8 +-
6 files changed, 264 insertions(+), 18 deletions(-)
diff --git a/flang/lib/Optimizer/OpenMP/LowerWorkdistribute.cpp b/flang/lib/Optimizer/OpenMP/LowerWorkdistribute.cpp
index 7b61539984232..a3b9e5c76bdd2 100644
--- a/flang/lib/Optimizer/OpenMP/LowerWorkdistribute.cpp
+++ b/flang/lib/Optimizer/OpenMP/LowerWorkdistribute.cpp
@@ -766,6 +766,7 @@ FailureOr<omp::TargetOp> splitTargetData(omp::TargetOp targetOp,
targetOp.getInReductionSymsAttr(), targetOp.getIsDevicePtrVars(),
innerMapInfos, targetOp.getNowaitAttr(), targetOp.getPrivateVars(),
targetOp.getPrivateSymsAttr(), targetOp.getPrivateNeedsBarrierAttr(),
+ targetOp.getThreadLimitNumDimsAttr(), targetOp.getThreadLimitDimsValues(),
targetOp.getThreadLimit(), targetOp.getPrivateMapsAttr());
rewriter.inlineRegionBefore(targetOp.getRegion(), newTargetOp.getRegion(),
newTargetOp.getRegion().begin());
@@ -1485,8 +1486,9 @@ genPreTargetOp(omp::TargetOp targetOp, SmallVector<Value> &preMapOperands,
targetOp.getInReductionByrefAttr(), targetOp.getInReductionSymsAttr(),
targetOp.getIsDevicePtrVars(), preMapOperands, targetOp.getNowaitAttr(),
targetOp.getPrivateVars(), targetOp.getPrivateSymsAttr(),
- targetOp.getPrivateNeedsBarrierAttr(), targetOp.getThreadLimit(),
- targetOp.getPrivateMapsAttr());
+ targetOp.getPrivateNeedsBarrierAttr(),
+ targetOp.getThreadLimitNumDimsAttr(), targetOp.getThreadLimitDimsValues(),
+ targetOp.getThreadLimit(), targetOp.getPrivateMapsAttr());
auto *preTargetBlock = rewriter.createBlock(
&preTargetOp.getRegion(), preTargetOp.getRegion().begin(), {}, {});
IRMapping preMapping;
@@ -1575,8 +1577,9 @@ genIsolatedTargetOp(omp::TargetOp targetOp, SmallVector<Value> &postMapOperands,
targetOp.getInReductionByrefAttr(), targetOp.getInReductionSymsAttr(),
targetOp.getIsDevicePtrVars(), postMapOperands, targetOp.getNowaitAttr(),
targetOp.getPrivateVars(), targetOp.getPrivateSymsAttr(),
- targetOp.getPrivateNeedsBarrierAttr(), targetOp.getThreadLimit(),
- targetOp.getPrivateMapsAttr());
+ targetOp.getPrivateNeedsBarrierAttr(),
+ targetOp.getThreadLimitNumDimsAttr(), targetOp.getThreadLimitDimsValues(),
+ targetOp.getThreadLimit(), targetOp.getPrivateMapsAttr());
auto *isolatedTargetBlock =
rewriter.createBlock(&isolatedTargetOp.getRegion(),
isolatedTargetOp.getRegion().begin(), {}, {});
@@ -1655,8 +1658,9 @@ static omp::TargetOp genPostTargetOp(omp::TargetOp targetOp,
targetOp.getInReductionByrefAttr(), targetOp.getInReductionSymsAttr(),
targetOp.getIsDevicePtrVars(), postMapOperands, targetOp.getNowaitAttr(),
targetOp.getPrivateVars(), targetOp.getPrivateSymsAttr(),
- targetOp.getPrivateNeedsBarrierAttr(), targetOp.getThreadLimit(),
- targetOp.getPrivateMapsAttr());
+ targetOp.getPrivateNeedsBarrierAttr(),
+ targetOp.getThreadLimitNumDimsAttr(), targetOp.getThreadLimitDimsValues(),
+ targetOp.getThreadLimit(), targetOp.getPrivateMapsAttr());
// Create the block for postTargetOp
auto *postTargetBlock = rewriter.createBlock(
&postTargetOp.getRegion(), postTargetOp.getRegion().begin(), {}, {});
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
index 2beb6690f1736..ae0e2479d1d00 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
@@ -1452,16 +1452,43 @@ class OpenMP_ThreadLimitClauseSkip<
> : OpenMP_Clause<traits, arguments, assemblyFormat, description,
extraClassDeclaration> {
let arguments = (ins
+ ConfinedAttr<OptionalAttr<I64Attr>, [IntPositive]>:$thread_limit_num_dims,
+ Variadic<AnyInteger>:$thread_limit_dims_values,
Optional<AnyInteger>:$thread_limit
);
let optAssemblyFormat = [{
- `thread_limit` `(` $thread_limit `:` type($thread_limit) `)`
+ `thread_limit` `(` custom<ThreadLimitClause>(
+ $thread_limit_num_dims, $thread_limit_dims_values, type($thread_limit_dims_values),
+ $thread_limit, type($thread_limit)
+ ) `)`
}];
let description = [{
The optional `thread_limit` specifies the limit on the number of threads.
}];
+
+ let extraClassDeclaration = [{
+ /// Returns true if the dims modifier is explicitly present
+ bool hasThreadLimitDimsModifier() {
+ return getThreadLimitNumDims().has_value() && getThreadLimitNumDims().value();
+ }
+
+ /// Returns the number of dimensions specified by dims modifier
+ unsigned getThreadLimitDimsCount() {
+ if (!hasThreadLimitDimsModifier())
+ return 1;
+ return static_cast<unsigned>(*getThreadLimitNumDims());
+ }
+
+ /// Returns the value for a specific dimension index
+ /// Index must be less than getThreadLimitDimsCount()
+ ::mlir::Value getThreadLimitDimensionValue(unsigned index) {
+ assert(index < getThreadLimitDimsCount() &&
+ "Thread limit dims index out of bounds");
+ return getThreadLimitDimsValues()[index];
+ }
+ }];
}
def OpenMP_ThreadLimitClause : OpenMP_ThreadLimitClauseSkip<>;
diff --git a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
index f9dcb48c7f08b..26cddace26c7c 100644
--- a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
+++ b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
@@ -2244,10 +2244,30 @@ void TargetOp::build(OpBuilder &builder, OperationState &state,
/*in_reduction_syms=*/nullptr, clauses.isDevicePtrVars,
clauses.mapVars, clauses.nowait, clauses.privateVars,
makeArrayAttr(ctx, clauses.privateSyms),
- clauses.privateNeedsBarrier, clauses.threadLimit,
+ clauses.privateNeedsBarrier, clauses.threadLimitNumDims,
+ clauses.threadLimitDimsValues, clauses.threadLimit,
/*private_maps=*/nullptr);
}
+// helper for thread_limit clause restrictions
+static LogicalResult
+verifyThreadLimitClause(Operation *op,
+ std::optional<IntegerAttr> threadLimitNumDims,
+ OperandRange threadLimitDimsValues, Value threadLimit) {
+ bool hasDimsModifier =
+ threadLimitNumDims.has_value() && threadLimitNumDims.value();
+
+ if (hasDimsModifier && threadLimit) {
+ return op->emitError("thread_limit with dims modifier cannot be used "
+ "together with number of threads");
+ }
+
+ if (failed(verifyDimsModifier(op, threadLimitNumDims, threadLimitDimsValues)))
+ return failure();
+
+ return success();
+}
+
LogicalResult TargetOp::verify() {
if (failed(verifyDependVarList(*this, getDependKinds(), getDependVars())))
return failure();
@@ -2259,6 +2279,11 @@ LogicalResult TargetOp::verify() {
if (failed(verifyMapClause(*this, getMapVars())))
return failure();
+ if (failed(verifyThreadLimitClause(*this, getThreadLimitNumDimsAttr(),
+ getThreadLimitDimsValues(),
+ getThreadLimit())))
+ return failure();
+
return verifyPrivateVarsMapping(*this);
}
@@ -2666,7 +2691,8 @@ void TeamsOp::build(OpBuilder &builder, OperationState &state,
/*private_needs_barrier=*/nullptr, clauses.reductionMod,
clauses.reductionVars,
makeDenseBoolArrayAttr(ctx, clauses.reductionByref),
- makeArrayAttr(ctx, clauses.reductionSyms), clauses.threadLimit);
+ makeArrayAttr(ctx, clauses.reductionSyms), clauses.threadLimitNumDims,
+ clauses.threadLimitDimsValues, clauses.threadLimit);
}
// Helper: Verify num_teams clause
@@ -2726,6 +2752,12 @@ LogicalResult TeamsOp::verify() {
return emitError(
"expected equal sizes for allocate and allocator variables");
+ // Check for thread_limit clause restrictions
+ if (failed(verifyThreadLimitClause(*this, getThreadLimitNumDimsAttr(),
+ getThreadLimitDimsValues(),
+ getThreadLimit())))
+ return failure();
+
return verifyReductionVarList(*this, getReductionSyms(), getReductionVars(),
getReductionByref());
}
@@ -4652,6 +4684,42 @@ static void printNumTeamsClause(OpAsmPrinter &p, Operation *op,
}
}
+//===----------------------------------------------------------------------===//
+// Parser and printer for thread_limit clause
+//===----------------------------------------------------------------------===//
+static ParseResult
+parseThreadLimitClause(OpAsmParser &parser, IntegerAttr &dimsAttr,
+ SmallVectorImpl<OpAsmParser::UnresolvedOperand> &values,
+ SmallVectorImpl<Type> &types,
+ std::optional<OpAsmParser::UnresolvedOperand> &bounds,
+ Type &boundsType) {
+ if (succeeded(parseDimsModifierWithValues(parser, dimsAttr, values, types))) {
+ return success();
+ }
+
+ OpAsmParser::UnresolvedOperand boundsOperand;
+ if (parser.parseOperand(boundsOperand) || parser.parseColon() ||
+ parser.parseType(boundsType)) {
+ return failure();
+ }
+ bounds = boundsOperand;
+ return success();
+}
+
+static void printThreadLimitClause(OpAsmPrinter &p, Operation *op,
+ IntegerAttr dimsAttr, OperandRange values,
+ TypeRange types, Value bounds,
+ Type boundsType) {
+ if (!values.empty()) {
+ // Multidimensional: dims(N): values : type
+ printDimsModifierWithValues(p, dimsAttr, values, types);
+ } else if (bounds) {
+ // Both bounds: bounds : type
+ p.printOperand(bounds);
+ p << " : " << boundsType;
+ }
+}
+
#define GET_ATTRDEF_CLASSES
#include "mlir/Dialect/OpenMP/OpenMPOpsAttributes.cpp.inc"
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index d910bc806288c..6c4ff65dff8fc 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -1974,6 +1974,10 @@ convertOmpTeams(omp::TeamsOp op, llvm::IRBuilderBase &builder,
return op.emitError("Lowering of num_teams with dims modifier is NYI.");
}
+ if (op.hasThreadLimitDimsModifier()) {
+ return op.emitError("Lowering of thread_limit with dims modifier is NYI.");
+ }
+
DenseMap<Value, llvm::Value *> reductionVariableMap;
unsigned numReductionVars = op.getNumReductionVars();
SmallVector<omp::DeclareReductionOp> reductionDecls;
@@ -5644,6 +5648,8 @@ extractHostEvalClauses(omp::TargetOp targetOp, Value &numThreads,
// num_teams dims and values are not yet supported
assert(!teamsOp.hasNumTeamsDimsModifier() &&
"Lowering of num_teams with dims modifier is NYI.");
+ assert(!teamsOp.hasThreadLimitDimsModifier() &&
+ "Lowering of thread_limit with dims modifier is NYI.");
if (teamsOp.getNumTeamsLower() == blockArg)
numTeamsLower = hostEvalVar;
else if (teamsOp.getNumTeamsUpper() == blockArg)
@@ -5769,6 +5775,8 @@ initTargetDefaultAttrs(omp::TargetOp targetOp, Operation *capturedOp,
// num_teams dims and values are not yet supported
assert(!teamsOp.hasNumTeamsDimsModifier() &&
"Lowering of num_teams with dims modifier is NYI.");
+ assert(!teamsOp.hasThreadLimitDimsModifier() &&
+ "Lowering of thread_limit with dims modifier is NYI.");
numTeamsLower = teamsOp.getNumTeamsLower();
numTeamsUpper = teamsOp.getNumTeamsUpper();
threadLimit = teamsOp.getThreadLimit();
diff --git a/mlir/test/Dialect/OpenMP/invalid.mlir b/mlir/test/Dialect/OpenMP/invalid.mlir
index dd367aba8da27..7f69ca5aaa064 100644
--- a/mlir/test/Dialect/OpenMP/invalid.mlir
+++ b/mlir/test/Dialect/OpenMP/invalid.mlir
@@ -1438,7 +1438,7 @@ func.func @omp_teams_allocate(%data_var : memref<i32>) {
// expected-error @below {{expected equal sizes for allocate and allocator variables}}
"omp.teams" (%data_var) ({
omp.terminator
- }) {operandSegmentSizes = array<i32: 1,0,0,0,0,0,0,0,0>} : (memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 1,0,0,0,0,0,0,0,0,0>} : (memref<i32>) -> ()
omp.terminator
}
return
@@ -1451,7 +1451,7 @@ func.func @omp_teams_num_teams1(%lb : i32) {
// expected-error @below {{expected num_teams upper bound to be defined if the lower bound is defined}}
"omp.teams" (%lb) ({
omp.terminator
- }) {operandSegmentSizes = array<i32: 0,0,0,0,1,0,0,0,0>} : (i32) -> ()
+ }) {operandSegmentSizes = array<i32: 0,0,0,0,1,0,0,0,0,0>} : (i32) -> ()
omp.terminator
}
return
@@ -1466,7 +1466,7 @@ func.func @omp_teams_num_teams_dims_mismatch() {
// expected-error @below {{dims(3) specified but 2 values provided}}
"omp.teams" (%v0, %v1) ({
omp.terminator
- }) {num_teams_num_dims = 3 : i64, operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0>} : (i32, i32) -> ()
+ }) {num_teams_num_dims = 3 : i64, operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0,0>} : (i32, i32) -> ()
omp.terminator
}
return
@@ -1483,7 +1483,7 @@ func.func @omp_teams_num_teams_dims_with_bounds() {
// expected-error @below {{num_teams with dims modifier cannot be used together with lower/upper bounds}}
"omp.teams" (%v0, %v1, %lb, %ub) ({
omp.terminator
- }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,2,1,1,0,0,0>} : (i32, i32, i32, i32) -> ()
+ }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,2,1,1,0,0,0,0>} : (i32, i32, i32, i32) -> ()
omp.terminator
}
return
@@ -1498,7 +1498,7 @@ func.func @omp_teams_num_teams_values_without_dims() {
// expected-error @below {{dims values can only be specified with dims modifier}}
"omp.teams" (%v0, %v1) ({
omp.terminator
- }) {operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0>} : (i32, i32) -> ()
+ }) {operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0,0>} : (i32, i32) -> ()
omp.terminator
}
return
@@ -1511,7 +1511,7 @@ func.func @omp_teams_num_teams_dims_no_values() {
// expected-error @below {{dims modifier requires values to be specified}}
"omp.teams" () ({
omp.terminator
- }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0>} : () -> ()
+ }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0>} : () -> ()
omp.terminator
}
return
@@ -1526,7 +1526,7 @@ func.func @omp_teams_num_teams_dims_type_mismatch() {
// expected-error @below {{dims modifier requires all values to have the same type}}
"omp.teams" (%v0, %v1) ({
omp.terminator
- }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0>} : (i32, i64) -> ()
+ }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0,0>} : (i32, i64) -> ()
omp.terminator
}
return
@@ -1547,6 +1547,139 @@ func.func @omp_teams_num_teams2(%lb : i32, %ub : i16) {
// -----
+func.func @omp_teams_thread_limit_dims_mismatch() {
+ omp.target {
+ %v0 = arith.constant 1 : i32
+ %v1 = arith.constant 2 : i32
+ // expected-error @below {{dims(3) specified but 2 values provided}}
+ "omp.teams" (%v0, %v1) ({
+ omp.terminator
+ }) {thread_limit_num_dims = 3 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2,0>} : (i32, i32) -> ()
+ omp.terminator
+ }
+ return
+}
+
+// -----
+
+func.func @omp_teams_thread_limit_dims_with_scalar() {
+ omp.target {
+ %v0 = arith.constant 1 : i32
+ %v1 = arith.constant 2 : i32
+ %tl = arith.constant 4 : i32
+ // expected-error @below {{thread_limit with dims modifier cannot be used together with number of threads}}
+ "omp.teams" (%v0, %v1, %tl) ({
+ omp.terminator
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2,1>} : (i32, i32, i32) -> ()
+ omp.terminator
+ }
+ return
+}
+
+// -----
+
+func.func @omp_teams_thread_limit_dims_no_values() {
+ omp.target {
+ // expected-error @below {{dims modifier requires values to be specified}}
+ "omp.teams" () ({
+ omp.terminator
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0>} : () -> ()
+ omp.terminator
+ }
+ return
+}
+
+// -----
+
+func.func @omp_teams_thread_limit_values_without_dims() {
+ omp.target {
+ %v0 = arith.constant 1 : i32
+ %v1 = arith.constant 2 : i32
+ // expected-error @below {{dims values can only be specified with dims modifier}}
+ "omp.teams" (%v0, %v1) ({
+ omp.terminator
+ }) {operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2,0>} : (i32, i32) -> ()
+ omp.terminator
+ }
+ return
+}
+
+// -----
+
+func.func @omp_teams_thread_limit_dims_type_mismatch() {
+ omp.target {
+ %v0 = arith.constant 1 : i32
+ %v1 = arith.constant 2 : i64
+ // expected-error @below {{dims modifier requires all values to have the same type}}
+ "omp.teams" (%v0, %v1) ({
+ omp.terminator
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2,0>} : (i32, i64) -> ()
+ omp.terminator
+ }
+ return
+}
+
+// -----
+
+func.func @omp_target_thread_limit_dims_mismatch() {
+ %v0 = arith.constant 1 : i32
+ %v1 = arith.constant 2 : i32
+ // expected-error @below {{dims(3) specified but 2 values provided}}
+ "omp.target" (%v0, %v1) ({
+ omp.terminator
+ }) {thread_limit_num_dims = 3 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2,0>} : (i32, i32) -> ()
+ return
+}
+
+// -----
+
+func.func @omp_target_thread_limit_dims_with_scalar() {
+ %v0 = arith.constant 1 : i32
+ %v1 = arith.constant 2 : i32
+ %tl = arith.constant 4 : i32
+ // expected-error @below {{thread_limit with dims modifier cannot be used together with number of threads}}
+ "omp.target" (%v0, %v1, %tl) ({
+ omp.terminator
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2,1>} : (i32, i32, i32) -> ()
+ return
+}
+
+// -----
+
+func.func @omp_target_thread_limit_dims_no_values() {
+ // expected-error @below {{dims modifier requires values to be specified}}
+ "omp.target" () ({
+ omp.terminator
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,0,0>} : () -> ()
+ return
+}
+
+// -----
+
+func.func @omp_target_thread_limit_values_without_dims() {
+ %v0 = arith.constant 1 : i32
+ %v1 = arith.constant 2 : i32
+ // expected-error @below {{dims values can only be specified with dims modifier}}
+ "omp.target" (%v0, %v1) ({
+ omp.terminator
+ }) {operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2,0>} : (i32, i32) -> ()
+ return
+}
+
+// -----
+
+func.func @omp_target_thread_limit_dims_type_mismatch() {
+ %v0 = arith.constant 1 : i32
+ %v1 = arith.constant 2 : i64
+ // expected-error @below {{dims modifier requires all values to have the same type}}
+ "omp.target" (%v0, %v1) ({
+ omp.terminator
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2,0>} : (i32, i64) -> ()
+ return
+}
+
+// -----
+
func.func @omp_sections(%data_var : memref<i32>) -> () {
// expected-error @below {{expected equal sizes for allocate and allocator variables}}
"omp.sections" (%data_var) ({
@@ -2533,7 +2666,7 @@ func.func @omp_target_depend(%data_var: memref<i32>) {
// expected-error @below {{op expected as many depend values as depend variables}}
"omp.target"(%data_var) ({
"omp.terminator"() : () -> ()
- }) {depend_kinds = [], operandSegmentSizes = array<i32: 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0>} : (memref<i32>) -> ()
+ }) {depend_kinds = [], operandSegmentSizes = array<i32: 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0>} : (memref<i32>) -> ()
"func.return"() : () -> ()
}
diff --git a/mlir/test/Dialect/OpenMP/ops.mlir b/mlir/test/Dialect/OpenMP/ops.mlir
index 3633a4be1eb62..a1146fbb263e6 100644
--- a/mlir/test/Dialect/OpenMP/ops.mlir
+++ b/mlir/test/Dialect/OpenMP/ops.mlir
@@ -823,7 +823,7 @@ func.func @omp_target(%if_cond : i1, %device : si32, %num_threads : i32, %devic
"omp.target"(%device, %if_cond, %num_threads) ({
// CHECK: omp.terminator
omp.terminator
- }) {nowait, operandSegmentSizes = array<i32: 0,0,0,1,0,0,1,0,0,0,0,1>} : ( si32, i1, i32 ) -> ()
+ }) {nowait, operandSegmentSizes = array<i32: 0,0,0,1,0,0,1,0,0,0,0,0,1>} : ( si32, i1, i32 ) -> ()
// Test with optional map clause.
// CHECK: %[[MAP_A:.*]] = omp.map.info var_ptr(%[[VAL_1:.*]] : memref<?xi32>, tensor<?xi32>) map_clauses(always, to) capture(ByRef) -> memref<?xi32> {name = ""}
@@ -1128,6 +1128,12 @@ func.func @omp_teams(%lb : i32, %ub : i32, %if_cond : i1, %num_threads : i32,
omp.terminator
}
+ // CHECK: omp.teams thread_limit(dims(2): %{{.*}}, %{{.*}} : i32)
+ omp.teams thread_limit(dims(2): %lb, %ub : i32) {
+ // CHECK: omp.terminator
+ omp.terminator
+ }
+
// Test reduction.
%c1 = arith.constant 1 : i32
%0 = llvm.alloca %c1 x i32 : (i32) -> !llvm.ptr
>From 6acbc14e66de2c21b30c89667ab3ca92dbc15b85 Mon Sep 17 00:00:00 2001
From: skc7 <Krishna.Sankisa at amd.com>
Date: Thu, 11 Dec 2025 18:35:55 +0530
Subject: [PATCH 3/4] update thread_limit description
---
mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td | 13 ++++++++++++-
1 file changed, 12 insertions(+), 1 deletion(-)
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
index ae0e2479d1d00..32e6c92d3da18 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
@@ -1465,7 +1465,18 @@ class OpenMP_ThreadLimitClauseSkip<
}];
let description = [{
- The optional `thread_limit` specifies the limit on the number of threads.
+ The `thread_limit` clause specifies the limit on the number of threads.
+
+ With dims modifier:
+ - The number of dimensions is specified by the `thread_limit_num_dims` attribute.
+ - The values for each dimension are specified by the `thread_limit_dims_values` attribute.
+ - Format: `thread_limit(dims(N): values : type)`
+ - Example: `thread_limit(dims(2): %n, %m : i64)`
+
+ Without dims modifier:
+ - The number of threads is specified by the `thread_limit`.
+ - Format: `thread_limit(number_of_threads : type)`
+ - Example: `thread_limit(%n : i64)`
}];
let extraClassDeclaration = [{
>From 7613ab21cf20b6a50503d9d800f25bf56dde0096 Mon Sep 17 00:00:00 2001
From: skc7 <Krishna.Sankisa at amd.com>
Date: Wed, 17 Dec 2025 09:06:36 +0530
Subject: [PATCH 4/4] Remove separate thread_limit argument from clause
---
flang/lib/Lower/OpenMP/ClauseProcessor.cpp | 3 +-
flang/lib/Lower/OpenMP/OpenMP.cpp | 16 ++--
.../Optimizer/OpenMP/LowerWorkdistribute.cpp | 8 +-
.../mlir/Dialect/OpenMP/OpenMPClauses.td | 16 ++--
mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp | 74 ++++++++-----------
.../OpenMP/OpenMPToLLVMIRTranslation.cpp | 11 +--
mlir/test/Dialect/OpenMP/invalid.mlir | 38 +++++-----
mlir/test/Dialect/OpenMP/ops.mlir | 2 +-
8 files changed, 78 insertions(+), 90 deletions(-)
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index 3c31b3a07f57f..b1c0053f149ec 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -666,8 +666,9 @@ bool ClauseProcessor::processThreadLimit(
lower::StatementContext &stmtCtx,
mlir::omp::ThreadLimitClauseOps &result) const {
if (auto *clause = findUniqueClause<omp::clause::ThreadLimit>()) {
- result.threadLimit =
+ mlir::Value threadLimitVal =
fir::getBase(converter.genExprValue(clause->v, stmtCtx));
+ result.threadLimitDimsValues.push_back(threadLimitVal);
return true;
}
return false;
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 38ab42076f559..54acb252cdbba 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -102,8 +102,9 @@ class HostEvalInfo {
if (ops.numThreads)
vars.push_back(ops.numThreads);
- if (ops.threadLimit)
- vars.push_back(ops.threadLimit);
+ // Old spec: single value in threadLimitDimsValues
+ for (mlir::Value val : ops.threadLimitDimsValues)
+ vars.push_back(val);
}
/// Update \c ops, replacing all values with the corresponding block argument
@@ -116,7 +117,7 @@ class HostEvalInfo {
ops.loopLowerBounds.size() + ops.loopUpperBounds.size() +
ops.loopSteps.size() + (ops.numTeamsLower ? 1 : 0) +
(ops.numTeamsUpper ? 1 : 0) + (ops.numThreads ? 1 : 0) +
- (ops.threadLimit ? 1 : 0) &&
+ ops.threadLimitDimsValues.size() &&
"invalid block argument list");
int argIndex = 0;
for (size_t i = 0; i < ops.loopLowerBounds.size(); ++i)
@@ -137,8 +138,8 @@ class HostEvalInfo {
if (ops.numThreads)
ops.numThreads = args[argIndex++];
- if (ops.threadLimit)
- ops.threadLimit = args[argIndex++];
+ for (size_t i = 0; i < ops.threadLimitDimsValues.size(); ++i)
+ ops.threadLimitDimsValues[i] = args[argIndex++];
}
/// Update \p clauseOps and \p ivOut with the corresponding host-evaluated
@@ -185,12 +186,13 @@ class HostEvalInfo {
/// \returns whether an update was performed. If not, these clauses were not
/// evaluated in the host device.
bool apply(mlir::omp::TeamsOperands &clauseOps) {
- if (!ops.numTeamsLower && !ops.numTeamsUpper && !ops.threadLimit)
+ if (!ops.numTeamsLower && !ops.numTeamsUpper &&
+ ops.threadLimitDimsValues.empty())
return false;
clauseOps.numTeamsLower = ops.numTeamsLower;
clauseOps.numTeamsUpper = ops.numTeamsUpper;
- clauseOps.threadLimit = ops.threadLimit;
+ clauseOps.threadLimitDimsValues = ops.threadLimitDimsValues;
return true;
}
diff --git a/flang/lib/Optimizer/OpenMP/LowerWorkdistribute.cpp b/flang/lib/Optimizer/OpenMP/LowerWorkdistribute.cpp
index a3b9e5c76bdd2..4d3fec3b0710f 100644
--- a/flang/lib/Optimizer/OpenMP/LowerWorkdistribute.cpp
+++ b/flang/lib/Optimizer/OpenMP/LowerWorkdistribute.cpp
@@ -767,7 +767,7 @@ FailureOr<omp::TargetOp> splitTargetData(omp::TargetOp targetOp,
innerMapInfos, targetOp.getNowaitAttr(), targetOp.getPrivateVars(),
targetOp.getPrivateSymsAttr(), targetOp.getPrivateNeedsBarrierAttr(),
targetOp.getThreadLimitNumDimsAttr(), targetOp.getThreadLimitDimsValues(),
- targetOp.getThreadLimit(), targetOp.getPrivateMapsAttr());
+ targetOp.getPrivateMapsAttr());
rewriter.inlineRegionBefore(targetOp.getRegion(), newTargetOp.getRegion(),
newTargetOp.getRegion().begin());
rewriter.replaceOp(targetOp, targetDataOp);
@@ -1488,7 +1488,7 @@ genPreTargetOp(omp::TargetOp targetOp, SmallVector<Value> &preMapOperands,
targetOp.getPrivateVars(), targetOp.getPrivateSymsAttr(),
targetOp.getPrivateNeedsBarrierAttr(),
targetOp.getThreadLimitNumDimsAttr(), targetOp.getThreadLimitDimsValues(),
- targetOp.getThreadLimit(), targetOp.getPrivateMapsAttr());
+ targetOp.getPrivateMapsAttr());
auto *preTargetBlock = rewriter.createBlock(
&preTargetOp.getRegion(), preTargetOp.getRegion().begin(), {}, {});
IRMapping preMapping;
@@ -1579,7 +1579,7 @@ genIsolatedTargetOp(omp::TargetOp targetOp, SmallVector<Value> &postMapOperands,
targetOp.getPrivateVars(), targetOp.getPrivateSymsAttr(),
targetOp.getPrivateNeedsBarrierAttr(),
targetOp.getThreadLimitNumDimsAttr(), targetOp.getThreadLimitDimsValues(),
- targetOp.getThreadLimit(), targetOp.getPrivateMapsAttr());
+ targetOp.getPrivateMapsAttr());
auto *isolatedTargetBlock =
rewriter.createBlock(&isolatedTargetOp.getRegion(),
isolatedTargetOp.getRegion().begin(), {}, {});
@@ -1660,7 +1660,7 @@ static omp::TargetOp genPostTargetOp(omp::TargetOp targetOp,
targetOp.getPrivateVars(), targetOp.getPrivateSymsAttr(),
targetOp.getPrivateNeedsBarrierAttr(),
targetOp.getThreadLimitNumDimsAttr(), targetOp.getThreadLimitDimsValues(),
- targetOp.getThreadLimit(), targetOp.getPrivateMapsAttr());
+ targetOp.getPrivateMapsAttr());
// Create the block for postTargetOp
auto *postTargetBlock = rewriter.createBlock(
&postTargetOp.getRegion(), postTargetOp.getRegion().begin(), {}, {});
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
index 32e6c92d3da18..f39236aee0dca 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
@@ -1453,14 +1453,12 @@ class OpenMP_ThreadLimitClauseSkip<
extraClassDeclaration> {
let arguments = (ins
ConfinedAttr<OptionalAttr<I64Attr>, [IntPositive]>:$thread_limit_num_dims,
- Variadic<AnyInteger>:$thread_limit_dims_values,
- Optional<AnyInteger>:$thread_limit
+ Variadic<AnyInteger>:$thread_limit_dims_values
);
let optAssemblyFormat = [{
`thread_limit` `(` custom<ThreadLimitClause>(
- $thread_limit_num_dims, $thread_limit_dims_values, type($thread_limit_dims_values),
- $thread_limit, type($thread_limit)
+ $thread_limit_num_dims, $thread_limit_dims_values, type($thread_limit_dims_values)
) `)`
}];
@@ -1468,14 +1466,14 @@ class OpenMP_ThreadLimitClauseSkip<
The `thread_limit` clause specifies the limit on the number of threads.
With dims modifier:
- - The number of dimensions is specified by the `thread_limit_num_dims` attribute.
- - The values for each dimension are specified by the `thread_limit_dims_values` attribute.
+ - The number of dimensions is specified by the `thread_limit_num_dims`.
+ - The values for each dimension are specified by the `thread_limit_dims_values`.
- Format: `thread_limit(dims(N): values : type)`
- Example: `thread_limit(dims(2): %n, %m : i64)`
Without dims modifier:
- - The number of threads is specified by the `thread_limit`.
- - Format: `thread_limit(number_of_threads : type)`
+ - The number of threads is specified by the single value in `thread_limit_dims_values`.
+ - Format: `thread_limit(value : type)`
- Example: `thread_limit(%n : i64)`
}];
@@ -1497,6 +1495,8 @@ class OpenMP_ThreadLimitClauseSkip<
::mlir::Value getThreadLimitDimensionValue(unsigned index) {
assert(index < getThreadLimitDimsCount() &&
"Thread limit dims index out of bounds");
+ if (getThreadLimitDimsValues().empty())
+ return nullptr;
return getThreadLimitDimsValues()[index];
}
}];
diff --git a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
index 26cddace26c7c..6009b46f69d40 100644
--- a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
+++ b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
@@ -2245,7 +2245,7 @@ void TargetOp::build(OpBuilder &builder, OperationState &state,
clauses.mapVars, clauses.nowait, clauses.privateVars,
makeArrayAttr(ctx, clauses.privateSyms),
clauses.privateNeedsBarrier, clauses.threadLimitNumDims,
- clauses.threadLimitDimsValues, clauses.threadLimit,
+ clauses.threadLimitDimsValues,
/*private_maps=*/nullptr);
}
@@ -2253,15 +2253,7 @@ void TargetOp::build(OpBuilder &builder, OperationState &state,
static LogicalResult
verifyThreadLimitClause(Operation *op,
std::optional<IntegerAttr> threadLimitNumDims,
- OperandRange threadLimitDimsValues, Value threadLimit) {
- bool hasDimsModifier =
- threadLimitNumDims.has_value() && threadLimitNumDims.value();
-
- if (hasDimsModifier && threadLimit) {
- return op->emitError("thread_limit with dims modifier cannot be used "
- "together with number of threads");
- }
-
+ OperandRange threadLimitDimsValues) {
if (failed(verifyDimsModifier(op, threadLimitNumDims, threadLimitDimsValues)))
return failure();
@@ -2280,8 +2272,7 @@ LogicalResult TargetOp::verify() {
return failure();
if (failed(verifyThreadLimitClause(*this, getThreadLimitNumDimsAttr(),
- getThreadLimitDimsValues(),
- getThreadLimit())))
+ getThreadLimitDimsValues())))
return failure();
return verifyPrivateVarsMapping(*this);
@@ -2299,10 +2290,9 @@ LogicalResult TargetOp::verifyRegions() {
cast<BlockArgOpenMPOpInterface>(getOperation()).getHostEvalBlockArgs()) {
for (Operation *user : hostEvalArg.getUsers()) {
if (auto teamsOp = dyn_cast<TeamsOp>(user)) {
- if (llvm::is_contained({teamsOp.getNumTeamsLower(),
- teamsOp.getNumTeamsUpper(),
- teamsOp.getThreadLimit()},
- hostEvalArg))
+ if (teamsOp.getNumTeamsLower() == hostEvalArg ||
+ teamsOp.getNumTeamsUpper() == hostEvalArg ||
+ llvm::is_contained(teamsOp.getThreadLimitDimsValues(), hostEvalArg))
continue;
return emitOpError() << "host_eval argument only legal as 'num_teams' "
@@ -2683,16 +2673,16 @@ void TeamsOp::build(OpBuilder &builder, OperationState &state,
const TeamsOperands &clauses) {
MLIRContext *ctx = builder.getContext();
// TODO Store clauses in op: privateVars, privateSyms, privateNeedsBarrier
- TeamsOp::build(
- builder, state, clauses.allocateVars, clauses.allocatorVars,
- clauses.ifExpr, clauses.numTeamsNumDims, clauses.numTeamsDimsValues,
- clauses.numTeamsLower, clauses.numTeamsUpper,
- /*private_vars=*/{}, /*private_syms=*/nullptr,
- /*private_needs_barrier=*/nullptr, clauses.reductionMod,
- clauses.reductionVars,
- makeDenseBoolArrayAttr(ctx, clauses.reductionByref),
- makeArrayAttr(ctx, clauses.reductionSyms), clauses.threadLimitNumDims,
- clauses.threadLimitDimsValues, clauses.threadLimit);
+ TeamsOp::build(builder, state, clauses.allocateVars, clauses.allocatorVars,
+ clauses.ifExpr, clauses.numTeamsNumDims,
+ clauses.numTeamsDimsValues, clauses.numTeamsLower,
+ clauses.numTeamsUpper,
+ /*private_vars=*/{}, /*private_syms=*/nullptr,
+ /*private_needs_barrier=*/nullptr, clauses.reductionMod,
+ clauses.reductionVars,
+ makeDenseBoolArrayAttr(ctx, clauses.reductionByref),
+ makeArrayAttr(ctx, clauses.reductionSyms),
+ clauses.threadLimitNumDims, clauses.threadLimitDimsValues);
}
// Helper: Verify num_teams clause
@@ -2754,8 +2744,7 @@ LogicalResult TeamsOp::verify() {
// Check for thread_limit clause restrictions
if (failed(verifyThreadLimitClause(*this, getThreadLimitNumDimsAttr(),
- getThreadLimitDimsValues(),
- getThreadLimit())))
+ getThreadLimitDimsValues())))
return failure();
return verifyReductionVarList(*this, getReductionSyms(), getReductionVars(),
@@ -4690,34 +4679,29 @@ static void printNumTeamsClause(OpAsmPrinter &p, Operation *op,
static ParseResult
parseThreadLimitClause(OpAsmParser &parser, IntegerAttr &dimsAttr,
SmallVectorImpl<OpAsmParser::UnresolvedOperand> &values,
- SmallVectorImpl<Type> &types,
- std::optional<OpAsmParser::UnresolvedOperand> &bounds,
- Type &boundsType) {
+ SmallVectorImpl<Type> &types) {
+ // Try parsing with dims modifier: dims(N): values : type
if (succeeded(parseDimsModifierWithValues(parser, dimsAttr, values, types))) {
return success();
}
- OpAsmParser::UnresolvedOperand boundsOperand;
- if (parser.parseOperand(boundsOperand) || parser.parseColon() ||
- parser.parseType(boundsType)) {
+ // Without dims modifier: value : type
+ OpAsmParser::UnresolvedOperand singleValue;
+ Type singleType;
+ if (parser.parseOperand(singleValue) || parser.parseColon() ||
+ parser.parseType(singleType)) {
return failure();
}
- bounds = boundsOperand;
+ values.push_back(singleValue);
+ types.push_back(singleType);
return success();
}
static void printThreadLimitClause(OpAsmPrinter &p, Operation *op,
IntegerAttr dimsAttr, OperandRange values,
- TypeRange types, Value bounds,
- Type boundsType) {
- if (!values.empty()) {
- // Multidimensional: dims(N): values : type
- printDimsModifierWithValues(p, dimsAttr, values, types);
- } else if (bounds) {
- // Both bounds: bounds : type
- p.printOperand(bounds);
- p << " : " << boundsType;
- }
+ TypeRange types) {
+ // Multidimensional: dims(N): values : type
+ printDimsModifierWithValues(p, dimsAttr, values, types);
}
#define GET_ATTRDEF_CLASSES
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index 6c4ff65dff8fc..2771b0428131a 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -2024,7 +2024,7 @@ convertOmpTeams(omp::TeamsOp op, llvm::IRBuilderBase &builder,
numTeamsUpper = moduleTranslation.lookupValue(numTeamsUpperVar);
llvm::Value *threadLimit = nullptr;
- if (Value threadLimitVar = op.getThreadLimit())
+ if (Value threadLimitVar = op.getThreadLimitDimensionValue(0))
threadLimit = moduleTranslation.lookupValue(threadLimitVar);
llvm::Value *ifExpr = nullptr;
@@ -5654,7 +5654,7 @@ extractHostEvalClauses(omp::TargetOp targetOp, Value &numThreads,
numTeamsLower = hostEvalVar;
else if (teamsOp.getNumTeamsUpper() == blockArg)
numTeamsUpper = hostEvalVar;
- else if (teamsOp.getThreadLimit() == blockArg)
+ else if (teamsOp.getThreadLimitDimensionValue(0) == blockArg)
threadLimit = hostEvalVar;
else
llvm_unreachable("unsupported host_eval use");
@@ -5779,7 +5779,7 @@ initTargetDefaultAttrs(omp::TargetOp targetOp, Operation *capturedOp,
"Lowering of thread_limit with dims modifier is NYI.");
numTeamsLower = teamsOp.getNumTeamsLower();
numTeamsUpper = teamsOp.getNumTeamsUpper();
- threadLimit = teamsOp.getThreadLimit();
+ threadLimit = teamsOp.getThreadLimitDimensionValue(0);
}
if (auto parallelOp = castOrGetParentOfType<omp::ParallelOp>(capturedOp))
@@ -5824,7 +5824,8 @@ initTargetDefaultAttrs(omp::TargetOp targetOp, Operation *capturedOp,
// Extract 'thread_limit' clause from 'target' and 'teams' directives.
int32_t targetThreadLimitVal = -1, teamsThreadLimitVal = -1;
- setMaxValueFromClause(targetOp.getThreadLimit(), targetThreadLimitVal);
+ setMaxValueFromClause(targetOp.getThreadLimitDimensionValue(0),
+ targetThreadLimitVal);
setMaxValueFromClause(threadLimit, teamsThreadLimitVal);
// Extract 'max_threads' clause from 'parallel' or set to 1 if it's SIMD.
@@ -5903,7 +5904,7 @@ initTargetRuntimeAttrs(llvm::IRBuilderBase &builder,
teamsThreadLimit, &lowerBounds, &upperBounds, &steps);
// TODO: Handle constant 'if' clauses.
- if (Value targetThreadLimit = targetOp.getThreadLimit())
+ if (Value targetThreadLimit = targetOp.getThreadLimitDimensionValue(0))
attrs.TargetThreadLimit.front() =
moduleTranslation.lookupValue(targetThreadLimit);
diff --git a/mlir/test/Dialect/OpenMP/invalid.mlir b/mlir/test/Dialect/OpenMP/invalid.mlir
index 7f69ca5aaa064..7d4668d213e6f 100644
--- a/mlir/test/Dialect/OpenMP/invalid.mlir
+++ b/mlir/test/Dialect/OpenMP/invalid.mlir
@@ -1438,7 +1438,7 @@ func.func @omp_teams_allocate(%data_var : memref<i32>) {
// expected-error @below {{expected equal sizes for allocate and allocator variables}}
"omp.teams" (%data_var) ({
omp.terminator
- }) {operandSegmentSizes = array<i32: 1,0,0,0,0,0,0,0,0,0>} : (memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 1,0,0,0,0,0,0,0,0>} : (memref<i32>) -> ()
omp.terminator
}
return
@@ -1451,7 +1451,7 @@ func.func @omp_teams_num_teams1(%lb : i32) {
// expected-error @below {{expected num_teams upper bound to be defined if the lower bound is defined}}
"omp.teams" (%lb) ({
omp.terminator
- }) {operandSegmentSizes = array<i32: 0,0,0,0,1,0,0,0,0,0>} : (i32) -> ()
+ }) {operandSegmentSizes = array<i32: 0,0,0,0,1,0,0,0,0>} : (i32) -> ()
omp.terminator
}
return
@@ -1466,7 +1466,7 @@ func.func @omp_teams_num_teams_dims_mismatch() {
// expected-error @below {{dims(3) specified but 2 values provided}}
"omp.teams" (%v0, %v1) ({
omp.terminator
- }) {num_teams_num_dims = 3 : i64, operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0,0>} : (i32, i32) -> ()
+ }) {num_teams_num_dims = 3 : i64, operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0>} : (i32, i32) -> ()
omp.terminator
}
return
@@ -1483,7 +1483,7 @@ func.func @omp_teams_num_teams_dims_with_bounds() {
// expected-error @below {{num_teams with dims modifier cannot be used together with lower/upper bounds}}
"omp.teams" (%v0, %v1, %lb, %ub) ({
omp.terminator
- }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,2,1,1,0,0,0,0>} : (i32, i32, i32, i32) -> ()
+ }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,2,1,1,0,0,0>} : (i32, i32, i32, i32) -> ()
omp.terminator
}
return
@@ -1498,7 +1498,7 @@ func.func @omp_teams_num_teams_values_without_dims() {
// expected-error @below {{dims values can only be specified with dims modifier}}
"omp.teams" (%v0, %v1) ({
omp.terminator
- }) {operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0,0>} : (i32, i32) -> ()
+ }) {operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0>} : (i32, i32) -> ()
omp.terminator
}
return
@@ -1511,7 +1511,7 @@ func.func @omp_teams_num_teams_dims_no_values() {
// expected-error @below {{dims modifier requires values to be specified}}
"omp.teams" () ({
omp.terminator
- }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0>} : () -> ()
+ }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0>} : () -> ()
omp.terminator
}
return
@@ -1526,7 +1526,7 @@ func.func @omp_teams_num_teams_dims_type_mismatch() {
// expected-error @below {{dims modifier requires all values to have the same type}}
"omp.teams" (%v0, %v1) ({
omp.terminator
- }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0,0>} : (i32, i64) -> ()
+ }) {num_teams_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,2,0,0,0,0,0>} : (i32, i64) -> ()
omp.terminator
}
return
@@ -1554,7 +1554,7 @@ func.func @omp_teams_thread_limit_dims_mismatch() {
// expected-error @below {{dims(3) specified but 2 values provided}}
"omp.teams" (%v0, %v1) ({
omp.terminator
- }) {thread_limit_num_dims = 3 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2,0>} : (i32, i32) -> ()
+ }) {thread_limit_num_dims = 3 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2>} : (i32, i32) -> ()
omp.terminator
}
return
@@ -1567,10 +1567,10 @@ func.func @omp_teams_thread_limit_dims_with_scalar() {
%v0 = arith.constant 1 : i32
%v1 = arith.constant 2 : i32
%tl = arith.constant 4 : i32
- // expected-error @below {{thread_limit with dims modifier cannot be used together with number of threads}}
+ // expected-error @below {{dims(2) specified but 3 values provided}}
"omp.teams" (%v0, %v1, %tl) ({
omp.terminator
- }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2,1>} : (i32, i32, i32) -> ()
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,3>} : (i32, i32, i32) -> ()
omp.terminator
}
return
@@ -1598,7 +1598,7 @@ func.func @omp_teams_thread_limit_values_without_dims() {
// expected-error @below {{dims values can only be specified with dims modifier}}
"omp.teams" (%v0, %v1) ({
omp.terminator
- }) {operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2,0>} : (i32, i32) -> ()
+ }) {operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2>} : (i32, i32) -> ()
omp.terminator
}
return
@@ -1613,7 +1613,7 @@ func.func @omp_teams_thread_limit_dims_type_mismatch() {
// expected-error @below {{dims modifier requires all values to have the same type}}
"omp.teams" (%v0, %v1) ({
omp.terminator
- }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2,0>} : (i32, i64) -> ()
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,2>} : (i32, i64) -> ()
omp.terminator
}
return
@@ -1627,7 +1627,7 @@ func.func @omp_target_thread_limit_dims_mismatch() {
// expected-error @below {{dims(3) specified but 2 values provided}}
"omp.target" (%v0, %v1) ({
omp.terminator
- }) {thread_limit_num_dims = 3 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2,0>} : (i32, i32) -> ()
+ }) {thread_limit_num_dims = 3 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2>} : (i32, i32) -> ()
return
}
@@ -1637,10 +1637,10 @@ func.func @omp_target_thread_limit_dims_with_scalar() {
%v0 = arith.constant 1 : i32
%v1 = arith.constant 2 : i32
%tl = arith.constant 4 : i32
- // expected-error @below {{thread_limit with dims modifier cannot be used together with number of threads}}
+ // expected-error @below {{dims(2) specified but 3 values provided}}
"omp.target" (%v0, %v1, %tl) ({
omp.terminator
- }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2,1>} : (i32, i32, i32) -> ()
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,3>} : (i32, i32, i32) -> ()
return
}
@@ -1650,7 +1650,7 @@ func.func @omp_target_thread_limit_dims_no_values() {
// expected-error @below {{dims modifier requires values to be specified}}
"omp.target" () ({
omp.terminator
- }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,0,0>} : () -> ()
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,0>} : () -> ()
return
}
@@ -1662,7 +1662,7 @@ func.func @omp_target_thread_limit_values_without_dims() {
// expected-error @below {{dims values can only be specified with dims modifier}}
"omp.target" (%v0, %v1) ({
omp.terminator
- }) {operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2,0>} : (i32, i32) -> ()
+ }) {operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2>} : (i32, i32) -> ()
return
}
@@ -1674,7 +1674,7 @@ func.func @omp_target_thread_limit_dims_type_mismatch() {
// expected-error @below {{dims modifier requires all values to have the same type}}
"omp.target" (%v0, %v1) ({
omp.terminator
- }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2,0>} : (i32, i64) -> ()
+ }) {thread_limit_num_dims = 2 : i64, operandSegmentSizes = array<i32: 0,0,0,0,0,0,0,0,0,0,0,2>} : (i32, i64) -> ()
return
}
@@ -2666,7 +2666,7 @@ func.func @omp_target_depend(%data_var: memref<i32>) {
// expected-error @below {{op expected as many depend values as depend variables}}
"omp.target"(%data_var) ({
"omp.terminator"() : () -> ()
- }) {depend_kinds = [], operandSegmentSizes = array<i32: 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0>} : (memref<i32>) -> ()
+ }) {depend_kinds = [], operandSegmentSizes = array<i32: 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0>} : (memref<i32>) -> ()
"func.return"() : () -> ()
}
diff --git a/mlir/test/Dialect/OpenMP/ops.mlir b/mlir/test/Dialect/OpenMP/ops.mlir
index a1146fbb263e6..a7d043404e37b 100644
--- a/mlir/test/Dialect/OpenMP/ops.mlir
+++ b/mlir/test/Dialect/OpenMP/ops.mlir
@@ -823,7 +823,7 @@ func.func @omp_target(%if_cond : i1, %device : si32, %num_threads : i32, %devic
"omp.target"(%device, %if_cond, %num_threads) ({
// CHECK: omp.terminator
omp.terminator
- }) {nowait, operandSegmentSizes = array<i32: 0,0,0,1,0,0,1,0,0,0,0,0,1>} : ( si32, i1, i32 ) -> ()
+ }) {nowait, operandSegmentSizes = array<i32: 0,0,0,1,0,0,1,0,0,0,0,1>} : ( si32, i1, i32 ) -> ()
// Test with optional map clause.
// CHECK: %[[MAP_A:.*]] = omp.map.info var_ptr(%[[VAL_1:.*]] : memref<?xi32>, tensor<?xi32>) map_clauses(always, to) capture(ByRef) -> memref<?xi32> {name = ""}
More information about the llvm-branch-commits
mailing list