[Mlir-commits] [mlir] [MLIR][OpenMP] Add missing clauses to OpenMP op definitions (PR #99507)
Sergio Afonso
llvmlistbot at llvm.org
Mon Jul 29 03:31:06 PDT 2024
https://github.com/skatrak updated https://github.com/llvm/llvm-project/pull/99507
>From ae2de5c3675c035d98e6c6ea8967bd80ea01ca3e Mon Sep 17 00:00:00 2001
From: Sergio Afonso <safonsof at amd.com>
Date: Tue, 16 Jul 2024 17:18:13 +0100
Subject: [PATCH] [MLIR][OpenMP] Add missing clauses to OpenMP op definitions
This patch adds the missing `OpenMP_Clause` definitions to all existing
`OpenMP_Op`s and updates their operand structure based builders to initialize
the new arguments.
The result of this change is that operation operand structures are now based in
the same list of clauses as their tablegen counterparts. This means that all of
the information needed is now in place to automatically generate OpenMP operand
structures from tablegen defitions.
Since this change doesn't involve the introduction of actual support for these
clauses, new arguments are not initialized from values stored in the
corresponding operand structure fields but rather set to empty or null. Those
should be updated when support for these clauses on the corresponding operation
is added.
---
.../Dialect/OpenMP/OpenMPClauseOperands.h | 7 +--
mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td | 52 +++++++++--------
mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp | 49 ++++++++++------
.../OpenMP/OpenMPToLLVMIRTranslation.cpp | 57 ++++++++++++++++---
mlir/test/Dialect/OpenMP/invalid.mlir | 40 ++++++-------
mlir/test/Dialect/OpenMP/ops.mlir | 26 ++++-----
6 files changed, 144 insertions(+), 87 deletions(-)
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauseOperands.h b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauseOperands.h
index a680a85db0f25..38e4d8f245e4f 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauseOperands.h
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauseOperands.h
@@ -272,11 +272,10 @@ using ParallelOperands =
using SectionsOperands = detail::Clauses<AllocateClauseOps, NowaitClauseOps,
PrivateClauseOps, ReductionClauseOps>;
-// TODO `linear` clause.
using SimdOperands =
- detail::Clauses<AlignedClauseOps, IfClauseOps, NontemporalClauseOps,
- OrderClauseOps, PrivateClauseOps, ReductionClauseOps,
- SafelenClauseOps, SimdlenClauseOps>;
+ detail::Clauses<AlignedClauseOps, IfClauseOps, LinearClauseOps,
+ NontemporalClauseOps, OrderClauseOps, PrivateClauseOps,
+ ReductionClauseOps, SafelenClauseOps, SimdlenClauseOps>;
using SingleOperands = detail::Clauses<AllocateClauseOps, CopyprivateClauseOps,
NowaitClauseOps, PrivateClauseOps>;
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 00dadc6e6538d..16b6f624ce042 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -195,10 +195,9 @@ def TerminatorOp : OpenMP_Op<"terminator", [Terminator, Pure]> {
def TeamsOp : OpenMP_Op<"teams", traits = [
AttrSizedOperandSegments, RecursiveMemoryEffects
], clauses = [
- // TODO: Complete clause list (private).
// TODO: Sort clauses alphabetically.
OpenMP_NumTeamsClause, OpenMP_IfClause, OpenMP_ThreadLimitClause,
- OpenMP_AllocateClause, OpenMP_ReductionClause
+ OpenMP_AllocateClause, OpenMP_ReductionClause, OpenMP_PrivateClause
], singleRegion = true> {
let summary = "teams construct";
let description = [{
@@ -238,9 +237,9 @@ def SectionOp : OpenMP_Op<"section", [HasParent<"SectionsOp">],
def SectionsOp : OpenMP_Op<"sections", traits = [
AttrSizedOperandSegments
], clauses = [
- // TODO: Complete clause list (private).
// TODO: Sort clauses alphabetically.
- OpenMP_ReductionClause, OpenMP_AllocateClause, OpenMP_NowaitClause
+ OpenMP_ReductionClause, OpenMP_AllocateClause, OpenMP_NowaitClause,
+ OpenMP_PrivateClause
], singleRegion = true> {
let summary = "sections construct";
let description = [{
@@ -271,8 +270,8 @@ def SectionsOp : OpenMP_Op<"sections", traits = [
def SingleOp : OpenMP_Op<"single", traits = [
AttrSizedOperandSegments
], clauses = [
- // TODO: Complete clause list (private).
- OpenMP_AllocateClause, OpenMP_CopyprivateClause, OpenMP_NowaitClause
+ OpenMP_AllocateClause, OpenMP_CopyprivateClause, OpenMP_NowaitClause,
+ OpenMP_PrivateClause
], singleRegion = true> {
let summary = "single directive";
let description = [{
@@ -363,14 +362,15 @@ def WsloopOp : OpenMP_Op<"wsloop", traits = [
AttrSizedOperandSegments, DeclareOpInterfaceMethods<LoopWrapperInterface>,
RecursiveMemoryEffects, SingleBlock
], clauses = [
- // TODO: Complete clause list (allocate, private).
// TODO: Sort clauses alphabetically.
OpenMP_LinearClauseSkip<assemblyFormat = true>,
OpenMP_ReductionClauseSkip<assemblyFormat = true>,
OpenMP_ScheduleClauseSkip<assemblyFormat = true>,
OpenMP_NowaitClauseSkip<assemblyFormat = true>,
OpenMP_OrderedClauseSkip<assemblyFormat = true>,
- OpenMP_OrderClauseSkip<assemblyFormat = true>
+ OpenMP_OrderClauseSkip<assemblyFormat = true>,
+ OpenMP_AllocateClauseSkip<assemblyFormat = true>,
+ OpenMP_PrivateClauseSkip<assemblyFormat = true>
], singleRegion = true> {
let summary = "worksharing-loop construct";
let description = [{
@@ -417,6 +417,13 @@ def WsloopOp : OpenMP_Op<"wsloop", traits = [
|`nowait` $nowait
|`ordered` `(` $ordered `)`
|`order` `(` custom<OrderClause>($order, $order_mod) `)`
+ |`allocate` `(`
+ custom<AllocateAndAllocator>(
+ $allocate_vars, type($allocate_vars), $allocator_vars,
+ type($allocator_vars)) `)`
+ |`private` `(`
+ custom<PrivateList>(
+ $private_vars, type($private_vars), $private_syms) `)`
) custom<Wsloop>($region, $reduction_vars, type($reduction_vars),
$reduction_byref, $reduction_syms) attr-dict
}];
@@ -432,9 +439,9 @@ def SimdOp : OpenMP_Op<"simd", traits = [
AttrSizedOperandSegments, DeclareOpInterfaceMethods<LoopWrapperInterface>,
RecursiveMemoryEffects, SingleBlock
], clauses = [
- // TODO: Complete clause list (linear, private, reduction).
- OpenMP_AlignedClause, OpenMP_IfClause, OpenMP_NontemporalClause,
- OpenMP_OrderClause, OpenMP_SafelenClause, OpenMP_SimdlenClause
+ OpenMP_AlignedClause, OpenMP_IfClause, OpenMP_LinearClause,
+ OpenMP_NontemporalClause, OpenMP_OrderClause, OpenMP_PrivateClause,
+ OpenMP_ReductionClause, OpenMP_SafelenClause, OpenMP_SimdlenClause
], singleRegion = true> {
let summary = "simd construct";
let description = [{
@@ -499,9 +506,9 @@ def DistributeOp : OpenMP_Op<"distribute", traits = [
AttrSizedOperandSegments, DeclareOpInterfaceMethods<LoopWrapperInterface>,
RecursiveMemoryEffects, SingleBlock
], clauses = [
- // TODO: Complete clause list (private).
// TODO: Sort clauses alphabetically.
- OpenMP_DistScheduleClause, OpenMP_AllocateClause, OpenMP_OrderClause
+ OpenMP_DistScheduleClause, OpenMP_AllocateClause, OpenMP_OrderClause,
+ OpenMP_PrivateClause
], singleRegion = true> {
let summary = "distribute construct";
let description = [{
@@ -552,11 +559,12 @@ def TaskOp : OpenMP_Op<"task", traits = [
AttrSizedOperandSegments, AutomaticAllocationScope,
OutlineableOpenMPOpInterface
], clauses = [
- // TODO: Complete clause list (affinity, detach, private).
+ // TODO: Complete clause list (affinity, detach).
// TODO: Sort clauses alphabetically.
OpenMP_IfClause, OpenMP_FinalClause, OpenMP_UntiedClause,
OpenMP_MergeableClause, OpenMP_InReductionClause,
- OpenMP_PriorityClause, OpenMP_DependClause, OpenMP_AllocateClause
+ OpenMP_PriorityClause, OpenMP_DependClause, OpenMP_AllocateClause,
+ OpenMP_PrivateClause
], singleRegion = true> {
let summary = "task construct";
let description = [{
@@ -589,14 +597,13 @@ def TaskloopOp : OpenMP_Op<"taskloop", traits = [
DeclareOpInterfaceMethods<LoopWrapperInterface>, RecursiveMemoryEffects,
SingleBlock
], clauses = [
- // TODO: Complete clause list (private).
// TODO: Sort clauses alphabetically.
OpenMP_IfClause, OpenMP_FinalClause, OpenMP_UntiedClause,
OpenMP_MergeableClause,
OpenMP_InReductionClauseSkip<extraClassDeclaration = true>,
OpenMP_ReductionClauseSkip<extraClassDeclaration = true>,
OpenMP_PriorityClause, OpenMP_AllocateClause, OpenMP_GrainsizeClause,
- OpenMP_NumTasksClause, OpenMP_NogroupClause
+ OpenMP_NumTasksClause, OpenMP_NogroupClause, OpenMP_PrivateClause
], singleRegion = true> {
let summary = "taskloop construct";
let description = [{
@@ -1076,12 +1083,12 @@ def TargetUpdateOp: OpenMP_Op<"target_update", traits = [
def TargetOp : OpenMP_Op<"target", traits = [
AttrSizedOperandSegments, IsolatedFromAbove, OutlineableOpenMPOpInterface
], clauses = [
- // TODO: Complete clause list (allocate, defaultmap, in_reduction,
- // uses_allocators).
+ // TODO: Complete clause list (defaultmap, uses_allocators).
// TODO: Sort clauses alphabetically.
OpenMP_IfClause, OpenMP_DeviceClause, OpenMP_ThreadLimitClause,
OpenMP_DependClause, OpenMP_NowaitClause, OpenMP_IsDevicePtrClause,
- OpenMP_HasDeviceAddrClause, OpenMP_MapClause, OpenMP_PrivateClause
+ OpenMP_HasDeviceAddrClause, OpenMP_MapClause, OpenMP_PrivateClause,
+ OpenMP_AllocateClause, OpenMP_InReductionClause
], singleRegion = true> {
let summary = "target construct";
let description = [{
@@ -1207,7 +1214,7 @@ def OrderedRegionOp : OpenMP_Op<"ordered.region", clauses = [
//===----------------------------------------------------------------------===//
def TaskwaitOp : OpenMP_Op<"taskwait", clauses = [
- // TODO: Complete clause list (depend, nowait).
+ OpenMP_DependClause, OpenMP_NowaitClause
]> {
let summary = "taskwait construct";
let description = [{
@@ -1218,9 +1225,6 @@ def TaskwaitOp : OpenMP_Op<"taskwait", clauses = [
let builders = [
OpBuilder<(ins CArg<"const TaskwaitOperands &">:$clauses)>
];
-
- // TODO: Remove overriden `assemblyFormat` once a clause is added.
- let assemblyFormat = "attr-dict";
}
//===----------------------------------------------------------------------===//
diff --git a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
index 84de63b7806dc..51e9dbe8c801d 100644
--- a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
+++ b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
@@ -1451,13 +1451,15 @@ void TargetOp::build(OpBuilder &builder, OperationState &state,
const TargetOperands &clauses) {
MLIRContext *ctx = builder.getContext();
// TODO Store clauses in op: allocateVars, allocatorVars, inReductionVars,
- // inReduceVarByRef, inReductionDeclSymbols, reductionVars, reduceVarByRef,
- // reductionDeclSymbols.
+ // inReductionByref, inReductionSyms.
TargetOp::build(builder, state, clauses.ifVar, clauses.device,
clauses.threadLimit, makeArrayAttr(ctx, clauses.dependKinds),
clauses.dependVars, clauses.nowait, clauses.isDevicePtrVars,
clauses.hasDeviceAddrVars, clauses.mapVars,
- clauses.privateVars, makeArrayAttr(ctx, clauses.privateSyms));
+ clauses.privateVars, makeArrayAttr(ctx, clauses.privateSyms),
+ /*allocate_vars=*/{}, /*allocator_vars=*/{},
+ /*in_reduction_vars=*/{}, /*in_reduction_byref=*/nullptr,
+ /*in_reduction_syms=*/nullptr);
}
LogicalResult TargetOp::verify() {
@@ -1584,7 +1586,8 @@ void TeamsOp::build(OpBuilder &builder, OperationState &state,
clauses.ifVar, clauses.threadLimit, clauses.allocateVars,
clauses.allocatorVars, clauses.reductionVars,
makeDenseBoolArrayAttr(ctx, clauses.reductionByref),
- makeArrayAttr(ctx, clauses.reductionSyms));
+ makeArrayAttr(ctx, clauses.reductionSyms), /*private_vars=*/{},
+ /*private_syms=*/nullptr);
}
LogicalResult TeamsOp::verify() {
@@ -1630,8 +1633,8 @@ void SectionsOp::build(OpBuilder &builder, OperationState &state,
SectionsOp::build(builder, state, clauses.reductionVars,
makeDenseBoolArrayAttr(ctx, clauses.reductionByref),
makeArrayAttr(ctx, clauses.reductionSyms),
- clauses.allocateVars, clauses.allocatorVars,
- clauses.nowait);
+ clauses.allocateVars, clauses.allocatorVars, clauses.nowait,
+ /*private_vars=*/{}, /*private_syms=*/nullptr);
}
LogicalResult SectionsOp::verify() {
@@ -1664,7 +1667,8 @@ void SingleOp::build(OpBuilder &builder, OperationState &state,
// TODO Store clauses in op: privateVars, privateSyms.
SingleOp::build(builder, state, clauses.allocateVars, clauses.allocatorVars,
clauses.copyprivateVars,
- makeArrayAttr(ctx, clauses.copyprivateSyms), clauses.nowait);
+ makeArrayAttr(ctx, clauses.copyprivateSyms), clauses.nowait,
+ /*private_vars=*/{}, /*private_syms=*/nullptr);
}
LogicalResult SingleOp::verify() {
@@ -1716,7 +1720,9 @@ void WsloopOp::build(OpBuilder &builder, OperationState &state,
/*reduction_byref=*/nullptr, /*reduction_syms=*/nullptr,
/*schedule_kind=*/nullptr, /*schedule_chunk=*/nullptr,
/*schedule_mod=*/nullptr, /*schedule_simd=*/false, /*nowait=*/false,
- /*ordered=*/nullptr, /*order=*/nullptr, /*order_mod=*/nullptr);
+ /*ordered=*/nullptr, /*order=*/nullptr, /*order_mod=*/nullptr,
+ /*allocate_vars=*/{}, /*allocator_vars=*/{}, /*private_vars=*/{},
+ /*private_syms=*/nullptr);
state.addAttributes(attributes);
}
@@ -1731,7 +1737,9 @@ void WsloopOp::build(OpBuilder &builder, OperationState &state,
makeArrayAttr(ctx, clauses.reductionSyms),
clauses.scheduleKind, clauses.scheduleChunk,
clauses.scheduleMod, clauses.scheduleSimd, clauses.nowait,
- clauses.ordered, clauses.order, clauses.orderMod);
+ clauses.ordered, clauses.order, clauses.orderMod,
+ /*allocate_vars=*/{}, /*allocator_vars=*/{},
+ /*private_vars=*/{}, /*private_syms=*/nullptr);
}
LogicalResult WsloopOp::verify() {
@@ -1756,12 +1764,15 @@ LogicalResult WsloopOp::verify() {
void SimdOp::build(OpBuilder &builder, OperationState &state,
const SimdOperands &clauses) {
MLIRContext *ctx = builder.getContext();
- // TODO Store clauses in op: privateVars, privateSyms, reductionVars,
- // reduceVarByRef, reductionDeclSymbols.
+ // TODO Store clauses in op: linearVars, linearStepVars, privateVars,
+ // privateSyms, reductionVars, reductionByref, reductionSyms.
SimdOp::build(builder, state, clauses.alignedVars,
makeArrayAttr(ctx, clauses.alignments), clauses.ifVar,
+ /*linear_vars=*/{}, /*linear_step_vars=*/{},
clauses.nontemporalVars, clauses.order, clauses.orderMod,
- clauses.safelen, clauses.simdlen);
+ /*private_vars=*/{}, /*private_syms=*/nullptr,
+ /*reduction_vars=*/{}, /*reduction_byref=*/nullptr,
+ /*reduction_syms=*/nullptr, clauses.safelen, clauses.simdlen);
}
LogicalResult SimdOp::verify() {
@@ -1795,7 +1806,8 @@ void DistributeOp::build(OpBuilder &builder, OperationState &state,
// TODO Store clauses in op: privateVars, privateSyms.
DistributeOp::build(builder, state, clauses.distScheduleStatic,
clauses.distScheduleChunkSize, clauses.allocateVars,
- clauses.allocatorVars, clauses.order, clauses.orderMod);
+ clauses.allocatorVars, clauses.order, clauses.orderMod,
+ /*private_vars=*/{}, /*private_syms=*/nullptr);
}
LogicalResult DistributeOp::verify() {
@@ -1927,7 +1939,8 @@ void TaskOp::build(OpBuilder &builder, OperationState &state,
makeDenseBoolArrayAttr(ctx, clauses.inReductionByref),
makeArrayAttr(ctx, clauses.inReductionSyms), clauses.priority,
makeArrayAttr(ctx, clauses.dependKinds), clauses.dependVars,
- clauses.allocateVars, clauses.allocatorVars);
+ clauses.allocateVars, clauses.allocatorVars,
+ /*private_vars=*/{}, /*private_syms=*/nullptr);
}
LogicalResult TaskOp::verify() {
@@ -1975,7 +1988,8 @@ void TaskloopOp::build(OpBuilder &builder, OperationState &state,
makeDenseBoolArrayAttr(ctx, clauses.reductionByref),
makeArrayAttr(ctx, clauses.reductionSyms), clauses.priority,
clauses.allocateVars, clauses.allocatorVars,
- clauses.grainsize, clauses.numTasks, clauses.nogroup);
+ clauses.grainsize, clauses.numTasks, clauses.nogroup,
+ /*private_vars=*/{}, /*private_syms=*/nullptr);
}
SmallVector<Value> TaskloopOp::getAllReductionVars() {
@@ -2225,8 +2239,9 @@ LogicalResult OrderedRegionOp::verify() {
void TaskwaitOp::build(OpBuilder &builder, OperationState &state,
const TaskwaitOperands &clauses) {
- // TODO Store clauses in op: dependTypeAttrs, dependVars, nowaitAttr.
- TaskwaitOp::build(builder, state);
+ // TODO Store clauses in op: dependKinds, dependVars, nowait.
+ TaskwaitOp::build(builder, state, /*depend_kinds=*/nullptr,
+ /*depend_vars=*/{}, /*nowait=*/nullptr);
}
//===----------------------------------------------------------------------===//
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index fe3f258156690..ddee117838697 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -837,9 +837,9 @@ convertOmpSections(Operation &opInst, llvm::IRBuilderBase &builder,
// TODO: Support the following clauses: private, firstprivate, lastprivate,
// allocate
if (!sectionsOp.getAllocateVars().empty() ||
- !sectionsOp.getAllocatorVars().empty())
- return emitError(sectionsOp.getLoc())
- << "allocate clause is not supported for sections construct";
+ !sectionsOp.getAllocatorVars().empty() ||
+ !sectionsOp.getPrivateVars().empty() || sectionsOp.getPrivateSyms())
+ return opInst.emitError("unhandled clauses for translation to LLVM IR");
llvm::ArrayRef<bool> isByRef = getIsByRef(sectionsOp.getReductionByref());
assert(isByRef.size() == sectionsOp.getNumReductionVars());
@@ -945,6 +945,9 @@ convertOmpSingle(omp::SingleOp &singleOp, llvm::IRBuilderBase &builder,
using InsertPointTy = llvm::OpenMPIRBuilder::InsertPointTy;
llvm::OpenMPIRBuilder::LocationDescription ompLoc(builder);
LogicalResult bodyGenStatus = success();
+ if (!singleOp.getPrivateVars().empty() || singleOp.getPrivateSyms())
+ return singleOp.emitError("unhandled clauses for translation to LLVM IR");
+
auto bodyCB = [&](InsertPointTy allocaIP, InsertPointTy codegenIP) {
builder.restoreIP(codegenIP);
convertOmpOpRegions(singleOp.getRegion(), "omp.single.region", builder,
@@ -976,7 +979,8 @@ convertOmpTeams(omp::TeamsOp op, llvm::IRBuilderBase &builder,
LLVM::ModuleTranslation &moduleTranslation) {
using InsertPointTy = llvm::OpenMPIRBuilder::InsertPointTy;
LogicalResult bodyGenStatus = success();
- if (!op.getAllocatorVars().empty() || op.getReductionSyms())
+ if (!op.getAllocatorVars().empty() || op.getReductionSyms() ||
+ !op.getPrivateVars().empty() || op.getPrivateSyms())
return op.emitError("unhandled clauses for translation to LLVM IR");
auto bodyCB = [&](InsertPointTy allocaIP, InsertPointTy codegenIP) {
@@ -1043,7 +1047,8 @@ convertOmpTaskOp(omp::TaskOp taskOp, llvm::IRBuilderBase &builder,
LogicalResult bodyGenStatus = success();
if (taskOp.getUntiedAttr() || taskOp.getMergeableAttr() ||
taskOp.getInReductionSyms() || taskOp.getPriority() ||
- !taskOp.getAllocateVars().empty()) {
+ !taskOp.getAllocateVars().empty() || !taskOp.getPrivateVars().empty() ||
+ taskOp.getPrivateSyms()) {
return taskOp.emitError("unhandled clauses for translation to LLVM IR");
}
auto bodyCB = [&](InsertPointTy allocaIP, InsertPointTy codegenIP) {
@@ -1091,11 +1096,28 @@ convertOmpTaskgroupOp(omp::TaskgroupOp tgOp, llvm::IRBuilderBase &builder,
ompLoc, allocaIP, bodyCB));
return bodyGenStatus;
}
+
+static LogicalResult
+convertOmpTaskwaitOp(omp::TaskwaitOp twOp, llvm::IRBuilderBase &builder,
+ LLVM::ModuleTranslation &moduleTranslation) {
+ if (!twOp.getDependVars().empty() || twOp.getDependKinds() ||
+ twOp.getNowait())
+ return twOp.emitError("unhandled clauses for translation to LLVM IR");
+
+ moduleTranslation.getOpenMPBuilder()->createTaskwait(builder.saveIP());
+ return success();
+}
+
/// Converts an OpenMP workshare loop into LLVM IR using OpenMPIRBuilder.
static LogicalResult
convertOmpWsloop(Operation &opInst, llvm::IRBuilderBase &builder,
LLVM::ModuleTranslation &moduleTranslation) {
auto wsloopOp = cast<omp::WsloopOp>(opInst);
+ if (!wsloopOp.getAllocateVars().empty() ||
+ !wsloopOp.getAllocatorVars().empty() ||
+ !wsloopOp.getPrivateVars().empty() || wsloopOp.getPrivateSyms())
+ return opInst.emitError("unhandled clauses for translation to LLVM IR");
+
// FIXME: Here any other nested wrappers (e.g. omp.simd) are skipped, so
// codegen for composite constructs like 'DO/FOR SIMD' will be the same as for
// 'DO/FOR'.
@@ -1607,6 +1629,12 @@ convertOmpSimd(Operation &opInst, llvm::IRBuilderBase &builder,
auto simdOp = cast<omp::SimdOp>(opInst);
auto loopOp = cast<omp::LoopNestOp>(simdOp.getWrappedLoop());
+ if (!simdOp.getLinearVars().empty() || !simdOp.getLinearStepVars().empty() ||
+ !simdOp.getPrivateVars().empty() || simdOp.getPrivateSyms() ||
+ !simdOp.getReductionVars().empty() || simdOp.getReductionByref() ||
+ simdOp.getReductionSyms())
+ return opInst.emitError("unhandled clauses for translation to LLVM IR");
+
llvm::OpenMPIRBuilder::LocationDescription ompLoc(builder);
// Generator of the canonical loop body.
@@ -3035,6 +3063,18 @@ static bool targetOpSupported(Operation &opInst) {
return false;
}
+ if (!targetOp.getAllocateVars().empty() ||
+ !targetOp.getAllocatorVars().empty()) {
+ opInst.emitError("Allocate clause not yet supported");
+ return false;
+ }
+
+ if (!targetOp.getInReductionVars().empty() ||
+ targetOp.getInReductionByref() || targetOp.getInReductionSyms()) {
+ opInst.emitError("In reduction clause not yet supported");
+ return false;
+ }
+
return true;
}
@@ -3434,10 +3474,6 @@ convertHostOrTargetOperation(Operation *op, llvm::IRBuilderBase &builder,
ompBuilder->createBarrier(builder.saveIP(), llvm::omp::OMPD_barrier);
return success();
})
- .Case([&](omp::TaskwaitOp) {
- ompBuilder->createTaskwait(builder.saveIP());
- return success();
- })
.Case([&](omp::TaskyieldOp) {
ompBuilder->createTaskyield(builder.saveIP());
return success();
@@ -3505,6 +3541,9 @@ convertHostOrTargetOperation(Operation *op, llvm::IRBuilderBase &builder,
.Case([&](omp::TaskgroupOp op) {
return convertOmpTaskgroupOp(op, builder, moduleTranslation);
})
+ .Case([&](omp::TaskwaitOp op) {
+ return convertOmpTaskwaitOp(op, builder, moduleTranslation);
+ })
.Case<omp::YieldOp, omp::TerminatorOp, omp::DeclareReductionOp,
omp::CriticalDeclareOp>([](auto op) {
// `yield` and `terminator` can be just omitted. The block structure
diff --git a/mlir/test/Dialect/OpenMP/invalid.mlir b/mlir/test/Dialect/OpenMP/invalid.mlir
index a29e34a9d5518..28542587fb9bb 100644
--- a/mlir/test/Dialect/OpenMP/invalid.mlir
+++ b/mlir/test/Dialect/OpenMP/invalid.mlir
@@ -421,7 +421,7 @@ func.func @omp_simd_aligned_mismatch(%arg0 : index, %arg1 : index,
omp.yield
}
}) {alignments = [128],
- operandSegmentSizes = array<i32: 2, 0, 0>} : (memref<i32>, memref<i32>) -> ()
+ operandSegmentSizes = array<i32: 2, 0, 0, 0, 0, 0, 0>} : (memref<i32>, memref<i32>) -> ()
return
}
@@ -435,7 +435,7 @@ func.func @omp_simd_aligned_negative(%arg0 : index, %arg1 : index,
omp.loop_nest (%iv) : index = (%arg0) to (%arg1) step (%arg2) {
omp.yield
}
- }) {alignments = [-1, 128], operandSegmentSizes = array<i32: 2, 0, 0>} : (memref<i32>, memref<i32>) -> ()
+ }) {alignments = [-1, 128], operandSegmentSizes = array<i32: 2, 0, 0, 0, 0, 0, 0>} : (memref<i32>, memref<i32>) -> ()
return
}
@@ -463,7 +463,7 @@ func.func @omp_simd_aligned_float(%arg0 : index, %arg1 : index,
omp.loop_nest (%iv) : index = (%arg0) to (%arg1) step (%arg2) {
omp.yield
}
- }) {alignments = [1.5, 128], operandSegmentSizes = array<i32: 2, 0, 0>} : (memref<i32>, memref<i32>) -> ()
+ }) {alignments = [1.5, 128], operandSegmentSizes = array<i32: 2, 0, 0, 0, 0, 0, 0>} : (memref<i32>, memref<i32>) -> ()
return
}
@@ -477,7 +477,7 @@ func.func @omp_simd_aligned_the_same_var(%arg0 : index, %arg1 : index,
omp.loop_nest (%iv) : index = (%arg0) to (%arg1) step (%arg2) {
omp.yield
}
- }) {alignments = [1, 128], operandSegmentSizes = array<i32: 2, 0, 0>} : (memref<i32>, memref<i32>) -> ()
+ }) {alignments = [1, 128], operandSegmentSizes = array<i32: 2, 0, 0, 0, 0, 0, 0>} : (memref<i32>, memref<i32>) -> ()
return
}
@@ -491,7 +491,7 @@ func.func @omp_simd_nontemporal_the_same_var(%arg0 : index, %arg1 : index,
omp.loop_nest (%iv) : index = (%arg0) to (%arg1) step (%arg2) {
omp.yield
}
- }) {operandSegmentSizes = array<i32: 0, 0, 2>} : (memref<i32>, memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 0, 0, 0, 0, 2, 0, 0>} : (memref<i32>, memref<i32>) -> ()
return
}
@@ -1394,7 +1394,7 @@ 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: 0,0,0,0,1,0,0>} : (memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 0,0,0,0,1,0,0,0>} : (memref<i32>) -> ()
omp.terminator
}
return
@@ -1407,7 +1407,7 @@ 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: 1,0,0,0,0,0,0>} : (i32) -> ()
+ }) {operandSegmentSizes = array<i32: 1,0,0,0,0,0,0,0>} : (i32) -> ()
omp.terminator
}
return
@@ -1432,7 +1432,7 @@ func.func @omp_sections(%data_var : memref<i32>) -> () {
// expected-error @below {{expected equal sizes for allocate and allocator variables}}
"omp.sections" (%data_var) ({
omp.terminator
- }) {operandSegmentSizes = array<i32: 0,1,0>} : (memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 0,1,0,0>} : (memref<i32>) -> ()
return
}
@@ -1442,7 +1442,7 @@ func.func @omp_sections(%data_var : memref<i32>) -> () {
// expected-error @below {{expected as many reduction symbol references as reduction variables}}
"omp.sections" (%data_var) ({
omp.terminator
- }) {operandSegmentSizes = array<i32: 1,0,0>} : (memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 1,0,0,0>} : (memref<i32>) -> ()
return
}
@@ -1557,7 +1557,7 @@ func.func @omp_single(%data_var : memref<i32>) -> () {
// expected-error @below {{expected equal sizes for allocate and allocator variables}}
"omp.single" (%data_var) ({
omp.barrier
- }) {operandSegmentSizes = array<i32: 1,0,0>} : (memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 1,0,0,0>} : (memref<i32>) -> ()
return
}
@@ -1567,7 +1567,7 @@ func.func @omp_single_copyprivate(%data_var : memref<i32>) -> () {
// expected-error @below {{inconsistent number of copyprivate vars (= 1) and functions (= 0), both must be equal}}
"omp.single" (%data_var) ({
omp.barrier
- }) {operandSegmentSizes = array<i32: 0,0,1>} : (memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 0,0,1,0>} : (memref<i32>) -> ()
return
}
@@ -1623,7 +1623,7 @@ func.func @omp_task_depend(%data_var: memref<i32>) {
// expected-error @below {{op expected as many depend values as depend variables}}
"omp.task"(%data_var) ({
"omp.terminator"() : () -> ()
- }) {depend_kinds = [], operandSegmentSizes = array<i32: 0, 0, 0, 0, 1, 0, 0>} : (memref<i32>) -> ()
+ }) {depend_kinds = [], operandSegmentSizes = array<i32: 0, 0, 0, 0, 1, 0, 0, 0>} : (memref<i32>) -> ()
"func.return"() : () -> ()
}
@@ -1820,7 +1820,7 @@ func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
omp.yield
}
- }) {operandSegmentSizes = array<i32: 0, 0, 0, 0, 0, 1, 0, 0, 0>} : (memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 0, 0, 0, 0, 0, 1, 0, 0, 0, 0>} : (memref<i32>) -> ()
return
}
@@ -1834,7 +1834,7 @@ func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
omp.yield
}
- }) {operandSegmentSizes = array<i32: 0, 0, 0, 2, 0, 0, 0, 0, 0>, reduction_syms = [@add_f32]} : (!llvm.ptr, !llvm.ptr) -> ()
+ }) {operandSegmentSizes = array<i32: 0, 0, 0, 2, 0, 0, 0, 0, 0, 0>, reduction_syms = [@add_f32]} : (!llvm.ptr, !llvm.ptr) -> ()
return
}
@@ -1847,7 +1847,7 @@ func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
omp.yield
}
- }) {operandSegmentSizes = array<i32: 0, 0, 0, 1, 0, 0, 0, 0, 0>, reduction_syms = [@add_f32, @add_f32]} : (!llvm.ptr) -> ()
+ }) {operandSegmentSizes = array<i32: 0, 0, 0, 1, 0, 0, 0, 0, 0, 0>, reduction_syms = [@add_f32, @add_f32]} : (!llvm.ptr) -> ()
return
}
@@ -1861,7 +1861,7 @@ func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
omp.yield
}
- }) {in_reduction_syms = [@add_f32], operandSegmentSizes = array<i32: 0, 0, 2, 0, 0, 0, 0, 0, 0>} : (!llvm.ptr, !llvm.ptr) -> ()
+ }) {in_reduction_syms = [@add_f32], operandSegmentSizes = array<i32: 0, 0, 2, 0, 0, 0, 0, 0, 0, 0>} : (!llvm.ptr, !llvm.ptr) -> ()
return
}
@@ -1874,7 +1874,7 @@ func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
omp.yield
}
- }) {in_reduction_syms = [@add_f32, @add_f32], operandSegmentSizes = array<i32: 0, 0, 1, 0, 0, 0, 0, 0, 0>} : (!llvm.ptr) -> ()
+ }) {in_reduction_syms = [@add_f32, @add_f32], operandSegmentSizes = array<i32: 0, 0, 1, 0, 0, 0, 0, 0, 0, 0>} : (!llvm.ptr) -> ()
return
}
@@ -2129,7 +2129,7 @@ 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, 0, 1, 0, 0, 0, 0>} : (memref<i32>) -> ()
+ }) {depend_kinds = [], operandSegmentSizes = array<i32: 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0>} : (memref<i32>) -> ()
"func.return"() : () -> ()
}
@@ -2137,7 +2137,7 @@ func.func @omp_target_depend(%data_var: memref<i32>) {
func.func @omp_distribute_schedule(%chunk_size : i32) -> () {
// expected-error @below {{op chunk size set without dist_schedule_static being present}}
- "omp.distribute"(%chunk_size) <{operandSegmentSizes = array<i32: 1, 0, 0>}> ({
+ "omp.distribute"(%chunk_size) <{operandSegmentSizes = array<i32: 1, 0, 0, 0>}> ({
"omp.terminator"() : () -> ()
}) : (i32) -> ()
}
@@ -2146,7 +2146,7 @@ func.func @omp_distribute_schedule(%chunk_size : i32) -> () {
func.func @omp_distribute_allocate(%data_var : memref<i32>) -> () {
// expected-error @below {{expected equal sizes for allocate and allocator variables}}
- "omp.distribute"(%data_var) <{operandSegmentSizes = array<i32: 0, 1, 0>}> ({
+ "omp.distribute"(%data_var) <{operandSegmentSizes = array<i32: 0, 1, 0, 0>}> ({
"omp.terminator"() : () -> ()
}) : (memref<i32>) -> ()
}
diff --git a/mlir/test/Dialect/OpenMP/ops.mlir b/mlir/test/Dialect/OpenMP/ops.mlir
index f639af9b86722..057d54c0c2f7c 100644
--- a/mlir/test/Dialect/OpenMP/ops.mlir
+++ b/mlir/test/Dialect/OpenMP/ops.mlir
@@ -382,7 +382,7 @@ func.func @omp_wsloop(%lb : index, %ub : index, %step : index, %data_var : memre
omp.yield
}
omp.terminator
- }) {operandSegmentSizes = array<i32: 0,0,0,0>, ordered = 1} :
+ }) {operandSegmentSizes = array<i32: 0,0,0,0,0,0,0>, ordered = 1} :
() -> ()
// CHECK: omp.wsloop linear(%{{.*}} = %{{.*}} : memref<i32>) schedule(static) {
@@ -392,7 +392,7 @@ func.func @omp_wsloop(%lb : index, %ub : index, %step : index, %data_var : memre
omp.yield
}
omp.terminator
- }) {operandSegmentSizes = array<i32: 1,1,0,0>, schedule_kind = #omp<schedulekind static>} :
+ }) {operandSegmentSizes = array<i32: 1,1,0,0,0,0,0>, schedule_kind = #omp<schedulekind static>} :
(memref<i32>, i32) -> ()
// CHECK: omp.wsloop linear(%{{.*}} = %{{.*}} : memref<i32>, %{{.*}} = %{{.*}} : memref<i32>) schedule(static) {
@@ -402,7 +402,7 @@ func.func @omp_wsloop(%lb : index, %ub : index, %step : index, %data_var : memre
omp.yield
}
omp.terminator
- }) {operandSegmentSizes = array<i32: 2,2,0,0>, schedule_kind = #omp<schedulekind static>} :
+ }) {operandSegmentSizes = array<i32: 2,2,0,0,0,0,0>, schedule_kind = #omp<schedulekind static>} :
(memref<i32>, memref<i32>, i32, i32) -> ()
// CHECK: omp.wsloop linear(%{{.*}} = %{{.*}} : memref<i32>) schedule(dynamic = %{{.*}}) ordered(2) {
@@ -412,7 +412,7 @@ func.func @omp_wsloop(%lb : index, %ub : index, %step : index, %data_var : memre
omp.yield
}
omp.terminator
- }) {operandSegmentSizes = array<i32: 1,1,0,1>, schedule_kind = #omp<schedulekind dynamic>, ordered = 2} :
+ }) {operandSegmentSizes = array<i32: 1,1,0,1,0,0,0>, schedule_kind = #omp<schedulekind dynamic>, ordered = 2} :
(memref<i32>, i32, i32) -> ()
// CHECK: omp.wsloop schedule(auto) nowait {
@@ -422,7 +422,7 @@ func.func @omp_wsloop(%lb : index, %ub : index, %step : index, %data_var : memre
omp.yield
}
omp.terminator
- }) {operandSegmentSizes = array<i32: 0,0,0,0>, nowait, schedule_kind = #omp<schedulekind auto>} :
+ }) {operandSegmentSizes = array<i32: 0,0,0,0,0,0,0>, nowait, schedule_kind = #omp<schedulekind auto>} :
() -> ()
// CHECK: omp.wsloop {
@@ -575,7 +575,7 @@ func.func @omp_simd_aligned_list(%arg0 : index, %arg1 : index, %arg2 : index,
}) : (index, index, index) -> ()
"omp.terminator"() : () -> ()
}) {alignments = [32, 128],
- operandSegmentSizes = array<i32: 2, 0, 0>} : (memref<i32>, memref<i32>) -> ()
+ operandSegmentSizes = array<i32: 2, 0, 0, 0, 0, 0, 0>} : (memref<i32>, memref<i32>) -> ()
return
}
@@ -590,7 +590,7 @@ func.func @omp_simd_aligned_single(%arg0 : index, %arg1 : index, %arg2 : index,
}) : (index, index, index) -> ()
"omp.terminator"() : () -> ()
}) {alignments = [32],
- operandSegmentSizes = array<i32: 1, 0, 0>} : (memref<i32>) -> ()
+ operandSegmentSizes = array<i32: 1, 0, 0, 0, 0, 0, 0>} : (memref<i32>) -> ()
return
}
@@ -605,7 +605,7 @@ func.func @omp_simd_nontemporal_list(%arg0 : index, %arg1 : index,
"omp.yield"() : () -> ()
}) : (index, index, index) -> ()
"omp.terminator"() : () -> ()
- }) {operandSegmentSizes = array<i32: 0, 0, 2>} : (memref<i32>, memref<i64>) -> ()
+ }) {operandSegmentSizes = array<i32: 0, 0, 0, 0, 2, 0, 0>} : (memref<i32>, memref<i64>) -> ()
return
}
@@ -620,7 +620,7 @@ func.func @omp_simd_nontemporal_single(%arg0 : index, %arg1 : index,
"omp.yield"() : () -> ()
}) : (index, index, index) -> ()
"omp.terminator"() : () -> ()
- }) {operandSegmentSizes = array<i32: 0, 0, 1>} : (memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 0, 0, 0, 0, 1, 0, 0>} : (memref<i32>) -> ()
return
}
@@ -809,7 +809,7 @@ func.func @omp_target(%if_cond : i1, %device : si32, %num_threads : i32, %devic
"omp.target"(%if_cond, %device, %num_threads) ({
// CHECK: omp.terminator
omp.terminator
- }) {nowait, operandSegmentSizes = array<i32: 1,1,1,0,0,0,0,0>} : ( i1, si32, i32 ) -> ()
+ }) {nowait, operandSegmentSizes = array<i32: 1,1,1,0,0,0,0,0,0,0,0>} : ( i1, si32, i32 ) -> ()
// Test with optional map clause.
// CHECK: %[[MAP_A:.*]] = omp.map.info var_ptr(%[[VAL_1:.*]] : memref<?xi32>, tensor<?xi32>) map_clauses(tofrom) capture(ByRef) -> memref<?xi32> {name = ""}
@@ -1874,13 +1874,13 @@ func.func @omp_sectionsop(%data_var1 : memref<i32>, %data_var2 : memref<i32>,
"omp.sections" (%data_var1, %data_var1) ({
// CHECK: omp.terminator
omp.terminator
- }) {operandSegmentSizes = array<i32: 0,1,1>} : (memref<i32>, memref<i32>) -> ()
+ }) {operandSegmentSizes = array<i32: 0,1,1,0>} : (memref<i32>, memref<i32>) -> ()
// CHECK: omp.sections reduction(@add_f32 -> %{{.*}} : !llvm.ptr)
"omp.sections" (%redn_var) ({
// CHECK: omp.terminator
omp.terminator
- }) {operandSegmentSizes = array<i32: 1,0,0>, reduction_byref = array<i1: false>, reduction_syms=[@add_f32]} : (!llvm.ptr) -> ()
+ }) {operandSegmentSizes = array<i32: 1,0,0,0>, reduction_byref = array<i1: false>, reduction_syms=[@add_f32]} : (!llvm.ptr) -> ()
// CHECK: omp.sections nowait {
omp.sections nowait {
@@ -2136,7 +2136,7 @@ func.func @omp_target_depend(%arg0: memref<i32>, %arg1: memref<i32>) {
omp.target depend(taskdependin -> %arg0 : memref<i32>, taskdependin -> %arg1 : memref<i32>, taskdependinout -> %arg0 : memref<i32>) {
// CHECK: omp.terminator
omp.terminator
- } {operandSegmentSizes = array<i32: 0,0,0,3,0>}
+ } {operandSegmentSizes = array<i32: 0,0,0,3,0,0,0,0>}
return
}
More information about the Mlir-commits
mailing list