[lldb] [llvm] [libcxx] [mlir] [flang] [libc] [libcxxabi] [clang-tools-extra] [clang] [compiler-rt] [mlir][flang][openacc] Support device_type on loop construct (PR #76892)
Valentin Clement バレンタイン クレメン via cfe-commits
cfe-commits at lists.llvm.org
Thu Jan 4 11:28:49 PST 2024
https://github.com/clementval updated https://github.com/llvm/llvm-project/pull/76892
>From ee784de88ed77d406eefe9f4ea65e823bde53e2e Mon Sep 17 00:00:00 2001
From: Valentin Clement <clementval at gmail.com>
Date: Tue, 2 Jan 2024 16:08:02 -0800
Subject: [PATCH 1/2] [mlir][flang][openacc] Support device_type on loop
construct
---
flang/lib/Lower/OpenACC.cpp | 183 +++++---
flang/test/Lower/OpenACC/acc-kernels-loop.f90 | 36 +-
flang/test/Lower/OpenACC/acc-loop.f90 | 41 +-
.../test/Lower/OpenACC/acc-parallel-loop.f90 | 36 +-
flang/test/Lower/OpenACC/acc-reduction.f90 | 6 +-
flang/test/Lower/OpenACC/acc-serial-loop.f90 | 36 +-
.../mlir/Dialect/OpenACC/OpenACCOps.td | 150 +++++-
mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp | 438 ++++++++++++++----
mlir/test/Dialect/OpenACC/invalid.mlir | 57 ++-
mlir/test/Dialect/OpenACC/ops.mlir | 80 ++--
10 files changed, 747 insertions(+), 316 deletions(-)
diff --git a/flang/lib/Lower/OpenACC.cpp b/flang/lib/Lower/OpenACC.cpp
index ecf70818c4ac0f..d8aedca90acf4b 100644
--- a/flang/lib/Lower/OpenACC.cpp
+++ b/flang/lib/Lower/OpenACC.cpp
@@ -1542,67 +1542,89 @@ createLoopOp(Fortran::lower::AbstractConverter &converter,
const Fortran::parser::AccClauseList &accClauseList,
bool needEarlyReturnHandling = false) {
fir::FirOpBuilder &builder = converter.getFirOpBuilder();
-
- mlir::Value workerNum;
- mlir::Value vectorNum;
- mlir::Value gangNum;
- mlir::Value gangDim;
- mlir::Value gangStatic;
llvm::SmallVector<mlir::Value> tileOperands, privateOperands,
- reductionOperands, cacheOperands;
+ reductionOperands, cacheOperands, vectorOperands, workerNumOperands,
+ gangOperands;
llvm::SmallVector<mlir::Attribute> privatizations, reductionRecipes;
- bool hasGang = false, hasVector = false, hasWorker = false;
+ llvm::SmallVector<int32_t> tileOperandsSegments, gangOperandsSegments;
+ llvm::SmallVector<int64_t> collapseValues;
+
+ llvm::SmallVector<mlir::Attribute> gangArgTypes;
+ llvm::SmallVector<mlir::Attribute> seqDeviceTypes, independentDeviceTypes,
+ autoDeviceTypes, vectorOperandsDeviceTypes, workerNumOperandsDeviceTypes,
+ vectorDeviceTypes, workerNumDeviceTypes, tileOperandsDeviceTypes,
+ collapseDeviceTypes, gangDeviceTypes, gangOperandsDeviceTypes;
+
+ // device_type attribute is set to `none` until a device_type clause is
+ // encountered.
+ auto crtDeviceTypeAttr = mlir::acc::DeviceTypeAttr::get(
+ builder.getContext(), mlir::acc::DeviceType::None);
for (const Fortran::parser::AccClause &clause : accClauseList.v) {
mlir::Location clauseLocation = converter.genLocation(clause.source);
if (const auto *gangClause =
std::get_if<Fortran::parser::AccClause::Gang>(&clause.u)) {
if (gangClause->v) {
+ auto crtGangOperands = gangOperands.size();
const Fortran::parser::AccGangArgList &x = *gangClause->v;
for (const Fortran::parser::AccGangArg &gangArg : x.v) {
if (const auto *num =
std::get_if<Fortran::parser::AccGangArg::Num>(&gangArg.u)) {
- gangNum = fir::getBase(converter.genExprValue(
- *Fortran::semantics::GetExpr(num->v), stmtCtx));
+ gangOperands.push_back(fir::getBase(converter.genExprValue(
+ *Fortran::semantics::GetExpr(num->v), stmtCtx)));
+ gangArgTypes.push_back(mlir::acc::GangArgTypeAttr::get(
+ builder.getContext(), mlir::acc::GangArgType::Num));
} else if (const auto *staticArg =
std::get_if<Fortran::parser::AccGangArg::Static>(
&gangArg.u)) {
const Fortran::parser::AccSizeExpr &sizeExpr = staticArg->v;
if (sizeExpr.v) {
- gangStatic = fir::getBase(converter.genExprValue(
- *Fortran::semantics::GetExpr(*sizeExpr.v), stmtCtx));
+ gangOperands.push_back(fir::getBase(converter.genExprValue(
+ *Fortran::semantics::GetExpr(*sizeExpr.v), stmtCtx)));
} else {
// * was passed as value and will be represented as a special
// constant.
- gangStatic = builder.createIntegerConstant(
- clauseLocation, builder.getIndexType(), starCst);
+ gangOperands.push_back(builder.createIntegerConstant(
+ clauseLocation, builder.getIndexType(), starCst));
}
+ gangArgTypes.push_back(mlir::acc::GangArgTypeAttr::get(
+ builder.getContext(), mlir::acc::GangArgType::Static));
} else if (const auto *dim =
std::get_if<Fortran::parser::AccGangArg::Dim>(
&gangArg.u)) {
- gangDim = fir::getBase(converter.genExprValue(
- *Fortran::semantics::GetExpr(dim->v), stmtCtx));
+ gangOperands.push_back(fir::getBase(converter.genExprValue(
+ *Fortran::semantics::GetExpr(dim->v), stmtCtx)));
+ gangArgTypes.push_back(mlir::acc::GangArgTypeAttr::get(
+ builder.getContext(), mlir::acc::GangArgType::Dim));
}
}
+ gangOperandsSegments.push_back(gangOperands.size() - crtGangOperands);
+ gangOperandsDeviceTypes.push_back(crtDeviceTypeAttr);
+ } else {
+ gangDeviceTypes.push_back(crtDeviceTypeAttr);
}
- hasGang = true;
} else if (const auto *workerClause =
std::get_if<Fortran::parser::AccClause::Worker>(&clause.u)) {
if (workerClause->v) {
- workerNum = fir::getBase(converter.genExprValue(
- *Fortran::semantics::GetExpr(*workerClause->v), stmtCtx));
+ workerNumOperands.push_back(fir::getBase(converter.genExprValue(
+ *Fortran::semantics::GetExpr(*workerClause->v), stmtCtx)));
+ workerNumOperandsDeviceTypes.push_back(crtDeviceTypeAttr);
+ } else {
+ workerNumDeviceTypes.push_back(crtDeviceTypeAttr);
}
- hasWorker = true;
} else if (const auto *vectorClause =
std::get_if<Fortran::parser::AccClause::Vector>(&clause.u)) {
if (vectorClause->v) {
- vectorNum = fir::getBase(converter.genExprValue(
- *Fortran::semantics::GetExpr(*vectorClause->v), stmtCtx));
+ vectorOperands.push_back(fir::getBase(converter.genExprValue(
+ *Fortran::semantics::GetExpr(*vectorClause->v), stmtCtx)));
+ vectorOperandsDeviceTypes.push_back(crtDeviceTypeAttr);
+ } else {
+ vectorDeviceTypes.push_back(crtDeviceTypeAttr);
}
- hasVector = true;
} else if (const auto *tileClause =
std::get_if<Fortran::parser::AccClause::Tile>(&clause.u)) {
const Fortran::parser::AccTileExprList &accTileExprList = tileClause->v;
+ auto crtTileOperands = tileOperands.size();
for (const auto &accTileExpr : accTileExprList.v) {
const auto &expr =
std::get<std::optional<Fortran::parser::ScalarIntConstantExpr>>(
@@ -1618,6 +1640,8 @@ createLoopOp(Fortran::lower::AbstractConverter &converter,
tileOperands.push_back(tileStar);
}
}
+ tileOperandsDeviceTypes.push_back(crtDeviceTypeAttr);
+ tileOperandsSegments.push_back(tileOperands.size() - crtTileOperands);
} else if (const auto *privateClause =
std::get_if<Fortran::parser::AccClause::Private>(
&clause.u)) {
@@ -1629,17 +1653,46 @@ createLoopOp(Fortran::lower::AbstractConverter &converter,
&clause.u)) {
genReductions(reductionClause->v, converter, semanticsContext, stmtCtx,
reductionOperands, reductionRecipes);
+ } else if (std::get_if<Fortran::parser::AccClause::Seq>(&clause.u)) {
+ seqDeviceTypes.push_back(crtDeviceTypeAttr);
+ } else if (std::get_if<Fortran::parser::AccClause::Independent>(
+ &clause.u)) {
+ independentDeviceTypes.push_back(crtDeviceTypeAttr);
+ } else if (std::get_if<Fortran::parser::AccClause::Auto>(&clause.u)) {
+ autoDeviceTypes.push_back(crtDeviceTypeAttr);
+ } else if (const auto *deviceTypeClause =
+ std::get_if<Fortran::parser::AccClause::DeviceType>(
+ &clause.u)) {
+ const Fortran::parser::AccDeviceTypeExprList &deviceTypeExprList =
+ deviceTypeClause->v;
+ assert(deviceTypeExprList.v.size() == 1 &&
+ "expect only one device_type expr");
+ crtDeviceTypeAttr = mlir::acc::DeviceTypeAttr::get(
+ builder.getContext(), getDeviceType(deviceTypeExprList.v.front().v));
+ } else if (const auto *collapseClause =
+ std::get_if<Fortran::parser::AccClause::Collapse>(
+ &clause.u)) {
+ const Fortran::parser::AccCollapseArg &arg = collapseClause->v;
+ const auto &force = std::get<bool>(arg.t);
+ if (force)
+ TODO(clauseLocation, "OpenACC collapse force modifier");
+ const auto &intExpr =
+ std::get<Fortran::parser::ScalarIntConstantExpr>(arg.t);
+ const auto *expr = Fortran::semantics::GetExpr(intExpr);
+ const std::optional<int64_t> collapseValue =
+ Fortran::evaluate::ToInt64(*expr);
+ assert(collapseValue && "expect integer value for the collapse clause");
+ collapseValues.push_back(*collapseValue);
+ collapseDeviceTypes.push_back(crtDeviceTypeAttr);
}
}
// Prepare the operand segment size attribute and the operands value range.
llvm::SmallVector<mlir::Value> operands;
llvm::SmallVector<int32_t> operandSegments;
- addOperand(operands, operandSegments, gangNum);
- addOperand(operands, operandSegments, gangDim);
- addOperand(operands, operandSegments, gangStatic);
- addOperand(operands, operandSegments, workerNum);
- addOperand(operands, operandSegments, vectorNum);
+ addOperands(operands, operandSegments, gangOperands);
+ addOperands(operands, operandSegments, workerNumOperands);
+ addOperands(operands, operandSegments, vectorOperands);
addOperands(operands, operandSegments, tileOperands);
addOperands(operands, operandSegments, cacheOperands);
addOperands(operands, operandSegments, privateOperands);
@@ -1657,12 +1710,42 @@ createLoopOp(Fortran::lower::AbstractConverter &converter,
builder, currentLocation, eval, operands, operandSegments,
/*outerCombined=*/false, retTy, yieldValue);
- if (hasGang)
- loopOp.setHasGangAttr(builder.getUnitAttr());
- if (hasWorker)
- loopOp.setHasWorkerAttr(builder.getUnitAttr());
- if (hasVector)
- loopOp.setHasVectorAttr(builder.getUnitAttr());
+ if (!gangDeviceTypes.empty())
+ loopOp.setGangAttr(builder.getArrayAttr(gangDeviceTypes));
+ if (!gangArgTypes.empty())
+ loopOp.setGangOperandsArgTypeAttr(builder.getArrayAttr(gangArgTypes));
+ if (!gangOperandsSegments.empty())
+ loopOp.setGangOperandsSegmentsAttr(
+ builder.getDenseI32ArrayAttr(gangOperandsSegments));
+ if (!gangOperandsDeviceTypes.empty())
+ loopOp.setGangOperandsDeviceTypeAttr(
+ builder.getArrayAttr(gangOperandsDeviceTypes));
+
+ if (!workerNumDeviceTypes.empty())
+ loopOp.setWorkerAttr(builder.getArrayAttr(workerNumDeviceTypes));
+ if (!workerNumOperandsDeviceTypes.empty())
+ loopOp.setWorkerNumOperandsDeviceTypeAttr(
+ builder.getArrayAttr(workerNumOperandsDeviceTypes));
+
+ if (!vectorDeviceTypes.empty())
+ loopOp.setVectorAttr(builder.getArrayAttr(vectorDeviceTypes));
+ if (!vectorOperandsDeviceTypes.empty())
+ loopOp.setVectorOperandsDeviceTypeAttr(
+ builder.getArrayAttr(vectorOperandsDeviceTypes));
+
+ if (!tileOperandsDeviceTypes.empty())
+ loopOp.setTileOperandsDeviceTypeAttr(
+ builder.getArrayAttr(tileOperandsDeviceTypes));
+ if (!tileOperandsSegments.empty())
+ loopOp.setTileOperandsSegmentsAttr(
+ builder.getDenseI32ArrayAttr(tileOperandsSegments));
+
+ if (!seqDeviceTypes.empty())
+ loopOp.setSeqAttr(builder.getArrayAttr(seqDeviceTypes));
+ if (!independentDeviceTypes.empty())
+ loopOp.setIndependentAttr(builder.getArrayAttr(independentDeviceTypes));
+ if (!autoDeviceTypes.empty())
+ loopOp.setAuto_Attr(builder.getArrayAttr(autoDeviceTypes));
if (!privatizations.empty())
loopOp.setPrivatizationsAttr(
@@ -1672,33 +1755,11 @@ createLoopOp(Fortran::lower::AbstractConverter &converter,
loopOp.setReductionRecipesAttr(
mlir::ArrayAttr::get(builder.getContext(), reductionRecipes));
- // Lower clauses mapped to attributes
- for (const Fortran::parser::AccClause &clause : accClauseList.v) {
- mlir::Location clauseLocation = converter.genLocation(clause.source);
- if (const auto *collapseClause =
- std::get_if<Fortran::parser::AccClause::Collapse>(&clause.u)) {
- const Fortran::parser::AccCollapseArg &arg = collapseClause->v;
- const auto &force = std::get<bool>(arg.t);
- if (force)
- TODO(clauseLocation, "OpenACC collapse force modifier");
- const auto &intExpr =
- std::get<Fortran::parser::ScalarIntConstantExpr>(arg.t);
- const auto *expr = Fortran::semantics::GetExpr(intExpr);
- const std::optional<int64_t> collapseValue =
- Fortran::evaluate::ToInt64(*expr);
- if (collapseValue) {
- loopOp.setCollapseAttr(builder.getI64IntegerAttr(*collapseValue));
- }
- } else if (std::get_if<Fortran::parser::AccClause::Seq>(&clause.u)) {
- loopOp.setSeqAttr(builder.getUnitAttr());
- } else if (std::get_if<Fortran::parser::AccClause::Independent>(
- &clause.u)) {
- loopOp.setIndependentAttr(builder.getUnitAttr());
- } else if (std::get_if<Fortran::parser::AccClause::Auto>(&clause.u)) {
- loopOp->setAttr(mlir::acc::LoopOp::getAutoAttrStrName(),
- builder.getUnitAttr());
- }
- }
+ if (!collapseValues.empty())
+ loopOp.setCollapseAttr(builder.getI64ArrayAttr(collapseValues));
+ if (!collapseDeviceTypes.empty())
+ loopOp.setCollapseDeviceTypeAttr(builder.getArrayAttr(collapseDeviceTypes));
+
return loopOp;
}
diff --git a/flang/test/Lower/OpenACC/acc-kernels-loop.f90 b/flang/test/Lower/OpenACC/acc-kernels-loop.f90
index 93bc699031d550..b17f2e2c80b20f 100644
--- a/flang/test/Lower/OpenACC/acc-kernels-loop.f90
+++ b/flang/test/Lower/OpenACC/acc-kernels-loop.f90
@@ -461,7 +461,7 @@ subroutine acc_kernels_loop
! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {seq}
+! CHECK-NEXT: } attributes {seq = [#acc.device_type<none>]}
! CHECK: acc.terminator
! CHECK-NEXT: }{{$}}
@@ -474,7 +474,7 @@ subroutine acc_kernels_loop
! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {auto}
+! CHECK-NEXT: } attributes {auto_ = [#acc.device_type<none>]}
! CHECK: acc.terminator
! CHECK-NEXT: }{{$}}
@@ -487,7 +487,7 @@ subroutine acc_kernels_loop
! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {independent}
+! CHECK-NEXT: } attributes {independent = [#acc.device_type<none>]}
! CHECK: acc.terminator
! CHECK-NEXT: }{{$}}
@@ -497,10 +497,10 @@ subroutine acc_kernels_loop
END DO
! CHECK: acc.kernels {
-! CHECK: acc.loop gang {
+! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: }{{$}}
+! CHECK-NEXT: } attributes {gang = [#acc.device_type<none>]}{{$}}
! CHECK: acc.terminator
! CHECK-NEXT: }{{$}}
@@ -511,7 +511,7 @@ subroutine acc_kernels_loop
! CHECK: acc.kernels {
! CHECK: [[GANGNUM1:%.*]] = arith.constant 8 : i32
-! CHECK-NEXT: acc.loop gang(num=[[GANGNUM1]] : i32) {
+! CHECK-NEXT: acc.loop gang({num=[[GANGNUM1]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -525,7 +525,7 @@ subroutine acc_kernels_loop
! CHECK: acc.kernels {
! CHECK: [[GANGNUM2:%.*]] = fir.load %{{.*}} : !fir.ref<i32>
-! CHECK-NEXT: acc.loop gang(num=[[GANGNUM2]] : i32) {
+! CHECK-NEXT: acc.loop gang({num=[[GANGNUM2]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -538,7 +538,7 @@ subroutine acc_kernels_loop
END DO
! CHECK: acc.kernels {
-! CHECK: acc.loop gang(num=%{{.*}} : i32, static=%{{.*}} : i32) {
+! CHECK: acc.loop gang({num=%{{.*}} : i32, static=%{{.*}} : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -550,10 +550,10 @@ subroutine acc_kernels_loop
a(i) = b(i)
END DO
! CHECK: acc.kernels {
-! CHECK: acc.loop vector {
+! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: }{{$}}
+! CHECK-NEXT: } attributes {vector = [#acc.device_type<none>]}{{$}}
! CHECK: acc.terminator
! CHECK-NEXT: }{{$}}
@@ -591,10 +591,10 @@ subroutine acc_kernels_loop
END DO
! CHECK: acc.kernels {
-! CHECK: acc.loop worker {
+! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: }{{$}}
+! CHECK-NEXT: } attributes {worker = [#acc.device_type<none>]}{{$}}
! CHECK: acc.terminator
! CHECK-NEXT: }{{$}}
@@ -624,7 +624,7 @@ subroutine acc_kernels_loop
! CHECK: fir.do_loop
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {collapse = 2 : i64}
+! CHECK-NEXT: } attributes {collapse = [2], collapseDeviceType = [#acc.device_type<none>]}
! CHECK: acc.terminator
! CHECK-NEXT: }{{$}}
@@ -655,7 +655,7 @@ subroutine acc_kernels_loop
! CHECK: acc.kernels {
! CHECK: [[TILESIZE:%.*]] = arith.constant 2 : i32
-! CHECK: acc.loop tile([[TILESIZE]] : i32) {
+! CHECK: acc.loop tile({[[TILESIZE]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -669,7 +669,7 @@ subroutine acc_kernels_loop
! CHECK: acc.kernels {
! CHECK: [[TILESIZEM1:%.*]] = arith.constant -1 : i32
-! CHECK: acc.loop tile([[TILESIZEM1]] : i32) {
+! CHECK: acc.loop tile({[[TILESIZEM1]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -686,7 +686,7 @@ subroutine acc_kernels_loop
! CHECK: acc.kernels {
! CHECK: [[TILESIZE1:%.*]] = arith.constant 2 : i32
! CHECK: [[TILESIZE2:%.*]] = arith.constant 2 : i32
-! CHECK: acc.loop tile([[TILESIZE1]], [[TILESIZE2]] : i32, i32) {
+! CHECK: acc.loop tile({[[TILESIZE1]] : i32, [[TILESIZE2]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -699,7 +699,7 @@ subroutine acc_kernels_loop
END DO
! CHECK: acc.kernels {
-! CHECK: acc.loop tile(%{{.*}} : i32) {
+! CHECK: acc.loop tile({%{{.*}} : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -714,7 +714,7 @@ subroutine acc_kernels_loop
END DO
! CHECK: acc.kernels {
-! CHECK: acc.loop tile(%{{.*}}, %{{.*}} : i32, i32) {
+! CHECK: acc.loop tile({%{{.*}} : i32, %{{.*}} : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
diff --git a/flang/test/Lower/OpenACC/acc-loop.f90 b/flang/test/Lower/OpenACC/acc-loop.f90
index 924574512da4c7..e7f65770498fe2 100644
--- a/flang/test/Lower/OpenACC/acc-loop.f90
+++ b/flang/test/Lower/OpenACC/acc-loop.f90
@@ -1,6 +1,5 @@
! This test checks lowering of OpenACC loop directive.
-! RUN: bbc -fopenacc -emit-fir -hlfir=false %s -o - | FileCheck %s
! RUN: bbc -fopenacc -emit-hlfir %s -o - | FileCheck %s
! CHECK-LABEL: acc.private.recipe @privatization_ref_10x10xf32 : !fir.ref<!fir.array<10x10xf32>> init {
@@ -41,7 +40,7 @@ program acc_loop
!CHECK: acc.loop {
!CHECK: fir.do_loop
!CHECK: acc.yield
-!CHECK-NEXT: } attributes {seq}
+!CHECK-NEXT: } attributes {seq = [#acc.device_type<none>]}
!$acc loop auto
DO i = 1, n
@@ -51,7 +50,7 @@ program acc_loop
!CHECK: acc.loop {
!CHECK: fir.do_loop
!CHECK: acc.yield
-!CHECK-NEXT: } attributes {auto}
+!CHECK-NEXT: } attributes {auto_ = [#acc.device_type<none>]}
!$acc loop independent
DO i = 1, n
@@ -61,17 +60,17 @@ program acc_loop
!CHECK: acc.loop {
!CHECK: fir.do_loop
!CHECK: acc.yield
-!CHECK-NEXT: } attributes {independent}
+!CHECK-NEXT: } attributes {independent = [#acc.device_type<none>]}
!$acc loop gang
DO i = 1, n
a(i) = b(i)
END DO
-!CHECK: acc.loop gang {
+!CHECK: acc.loop {
!CHECK: fir.do_loop
!CHECK: acc.yield
-!CHECK-NEXT: }{{$}}
+!CHECK-NEXT: } attributes {gang = [#acc.device_type<none>]}{{$}}
!$acc loop gang(num: 8)
DO i = 1, n
@@ -79,7 +78,7 @@ program acc_loop
END DO
!CHECK: [[GANGNUM1:%.*]] = arith.constant 8 : i32
-!CHECK-NEXT: acc.loop gang(num=[[GANGNUM1]] : i32) {
+!CHECK-NEXT: acc.loop gang({num=[[GANGNUM1]] : i32}) {
!CHECK: fir.do_loop
!CHECK: acc.yield
!CHECK-NEXT: }{{$}}
@@ -90,7 +89,7 @@ program acc_loop
END DO
!CHECK: [[GANGNUM2:%.*]] = fir.load %{{.*}} : !fir.ref<i32>
-!CHECK-NEXT: acc.loop gang(num=[[GANGNUM2]] : i32) {
+!CHECK-NEXT: acc.loop gang({num=[[GANGNUM2]] : i32}) {
!CHECK: fir.do_loop
!CHECK: acc.yield
!CHECK-NEXT: }{{$}}
@@ -100,7 +99,7 @@ program acc_loop
a(i) = b(i)
END DO
-!CHECK: acc.loop gang(num=%{{.*}} : i32, static=%{{.*}} : i32) {
+!CHECK: acc.loop gang({num=%{{.*}} : i32, static=%{{.*}} : i32}) {
!CHECK: fir.do_loop
!CHECK: acc.yield
!CHECK-NEXT: }{{$}}
@@ -110,10 +109,10 @@ program acc_loop
a(i) = b(i)
END DO
-!CHECK: acc.loop vector {
+!CHECK: acc.loop {
!CHECK: fir.do_loop
!CHECK: acc.yield
-!CHECK-NEXT: }{{$}}
+!CHECK-NEXT: } attributes {vector = [#acc.device_type<none>]}{{$}}
!$acc loop vector(128)
DO i = 1, n
@@ -142,10 +141,10 @@ program acc_loop
a(i) = b(i)
END DO
-!CHECK: acc.loop worker {
+!CHECK: acc.loop {
!CHECK: fir.do_loop
!CHECK: acc.yield
-!CHECK-NEXT: }{{$}}
+!CHECK-NEXT: } attributes {worker = [#acc.device_type<none>]}{{$}}
!$acc loop worker(128)
DO i = 1, n
@@ -193,7 +192,7 @@ program acc_loop
a(i) = b(i)
END DO
!CHECK: [[TILESIZE:%.*]] = arith.constant 2 : i32
-!CHECK: acc.loop tile([[TILESIZE]] : i32) {
+!CHECK: acc.loop tile({[[TILESIZE]] : i32}) {
!CHECK: fir.do_loop
!CHECK: acc.yield
!CHECK-NEXT: }{{$}}
@@ -203,7 +202,7 @@ program acc_loop
a(i) = b(i)
END DO
!CHECK: [[TILESIZEM1:%.*]] = arith.constant -1 : i32
-!CHECK: acc.loop tile([[TILESIZEM1]] : i32) {
+!CHECK: acc.loop tile({[[TILESIZEM1]] : i32}) {
!CHECK: fir.do_loop
!CHECK: acc.yield
!CHECK-NEXT: }{{$}}
@@ -217,7 +216,7 @@ program acc_loop
!CHECK: [[TILESIZE1:%.*]] = arith.constant 2 : i32
!CHECK: [[TILESIZE2:%.*]] = arith.constant 2 : i32
-!CHECK: acc.loop tile([[TILESIZE1]], [[TILESIZE2]] : i32, i32) {
+!CHECK: acc.loop tile({[[TILESIZE1]] : i32, [[TILESIZE2]] : i32}) {
!CHECK: fir.do_loop
!CHECK: acc.yield
!CHECK-NEXT: }{{$}}
@@ -227,7 +226,7 @@ program acc_loop
a(i) = b(i)
END DO
-!CHECK: acc.loop tile(%{{.*}} : i32) {
+!CHECK: acc.loop tile({%{{.*}} : i32}) {
!CHECK: fir.do_loop
!CHECK: acc.yield
!CHECK-NEXT: }{{$}}
@@ -239,7 +238,7 @@ program acc_loop
END DO
END DO
-!CHECK: acc.loop tile(%{{.*}}, %{{.*}} : i32, i32) {
+!CHECK: acc.loop tile({%{{.*}} : i32, %{{.*}} : i32}) {
!CHECK: fir.do_loop
!CHECK: acc.yield
!CHECK-NEXT: }{{$}}
@@ -255,7 +254,7 @@ program acc_loop
!CHECK: fir.do_loop
!CHECK: fir.do_loop
!CHECK: acc.yield
-!CHECK-NEXT: } attributes {collapse = 2 : i64}
+!CHECK-NEXT: } attributes {collapse = [2], collapseDeviceType = [#acc.device_type<none>]}
!$acc loop
DO i = 1, n
@@ -290,7 +289,7 @@ program acc_loop
a(i) = b(i)
END DO
-!CHECK: acc.loop gang(dim=%{{.*}}, static=%{{.*}} : i32) {
+!CHECK: acc.loop gang({dim=%{{.*}}, static=%{{.*}} : i32}) {
!CHECK: fir.do_loop
!CHECK: acc.yield
!CHECK-NEXT: }{{$}}
@@ -301,7 +300,7 @@ program acc_loop
END DO
!CHECK: [[GANGDIM1:%.*]] = arith.constant 1 : i32
-!CHECK-NEXT: acc.loop gang(dim=[[GANGDIM1]] : i32) {
+!CHECK-NEXT: acc.loop gang({dim=[[GANGDIM1]] : i32}) {
!CHECK: fir.do_loop
!CHECK: acc.yield
!CHECK-NEXT: }{{$}}
diff --git a/flang/test/Lower/OpenACC/acc-parallel-loop.f90 b/flang/test/Lower/OpenACC/acc-parallel-loop.f90
index deee7089033ead..e9150a71f3826b 100644
--- a/flang/test/Lower/OpenACC/acc-parallel-loop.f90
+++ b/flang/test/Lower/OpenACC/acc-parallel-loop.f90
@@ -476,7 +476,7 @@ subroutine acc_parallel_loop
! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {seq}
+! CHECK-NEXT: } attributes {seq = [#acc.device_type<none>]}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -489,7 +489,7 @@ subroutine acc_parallel_loop
! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {auto}
+! CHECK-NEXT: } attributes {auto_ = [#acc.device_type<none>]}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -502,7 +502,7 @@ subroutine acc_parallel_loop
! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {independent}
+! CHECK-NEXT: } attributes {independent = [#acc.device_type<none>]}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -512,10 +512,10 @@ subroutine acc_parallel_loop
END DO
! CHECK: acc.parallel {
-! CHECK: acc.loop gang {
+! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: }{{$}}
+! CHECK-NEXT: } attributes {gang = [#acc.device_type<none>]}{{$}}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -526,7 +526,7 @@ subroutine acc_parallel_loop
! CHECK: acc.parallel {
! CHECK: [[GANGNUM1:%.*]] = arith.constant 8 : i32
-! CHECK-NEXT: acc.loop gang(num=[[GANGNUM1]] : i32) {
+! CHECK-NEXT: acc.loop gang({num=[[GANGNUM1]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -540,7 +540,7 @@ subroutine acc_parallel_loop
! CHECK: acc.parallel {
! CHECK: [[GANGNUM2:%.*]] = fir.load %{{.*}} : !fir.ref<i32>
-! CHECK-NEXT: acc.loop gang(num=[[GANGNUM2]] : i32) {
+! CHECK-NEXT: acc.loop gang({num=[[GANGNUM2]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -553,7 +553,7 @@ subroutine acc_parallel_loop
END DO
! CHECK: acc.parallel {
-! CHECK: acc.loop gang(num=%{{.*}} : i32, static=%{{.*}} : i32) {
+! CHECK: acc.loop gang({num=%{{.*}} : i32, static=%{{.*}} : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -565,10 +565,10 @@ subroutine acc_parallel_loop
a(i) = b(i)
END DO
! CHECK: acc.parallel {
-! CHECK: acc.loop vector {
+! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: }{{$}}
+! CHECK-NEXT: } attributes {vector = [#acc.device_type<none>]}{{$}}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -606,10 +606,10 @@ subroutine acc_parallel_loop
END DO
! CHECK: acc.parallel {
-! CHECK: acc.loop worker {
+! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: }{{$}}
+! CHECK-NEXT: } attributes {worker = [#acc.device_type<none>]}{{$}}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -639,7 +639,7 @@ subroutine acc_parallel_loop
! CHECK: fir.do_loop
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {collapse = 2 : i64}
+! CHECK-NEXT: } attributes {collapse = [2], collapseDeviceType = [#acc.device_type<none>]}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -670,7 +670,7 @@ subroutine acc_parallel_loop
! CHECK: acc.parallel {
! CHECK: [[TILESIZE:%.*]] = arith.constant 2 : i32
-! CHECK: acc.loop tile([[TILESIZE]] : i32) {
+! CHECK: acc.loop tile({[[TILESIZE]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -684,7 +684,7 @@ subroutine acc_parallel_loop
! CHECK: acc.parallel {
! CHECK: [[TILESIZEM1:%.*]] = arith.constant -1 : i32
-! CHECK: acc.loop tile([[TILESIZEM1]] : i32) {
+! CHECK: acc.loop tile({[[TILESIZEM1]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -701,7 +701,7 @@ subroutine acc_parallel_loop
! CHECK: acc.parallel {
! CHECK: [[TILESIZE1:%.*]] = arith.constant 2 : i32
! CHECK: [[TILESIZE2:%.*]] = arith.constant 2 : i32
-! CHECK: acc.loop tile([[TILESIZE1]], [[TILESIZE2]] : i32, i32) {
+! CHECK: acc.loop tile({[[TILESIZE1]] : i32, [[TILESIZE2]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -714,7 +714,7 @@ subroutine acc_parallel_loop
END DO
! CHECK: acc.parallel {
-! CHECK: acc.loop tile(%{{.*}} : i32) {
+! CHECK: acc.loop tile({%{{.*}} : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -729,7 +729,7 @@ subroutine acc_parallel_loop
END DO
! CHECK: acc.parallel {
-! CHECK: acc.loop tile(%{{.*}}, %{{.*}} : i32, i32) {
+! CHECK: acc.loop tile({%{{.*}} : i32, %{{.*}} : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
diff --git a/flang/test/Lower/OpenACC/acc-reduction.f90 b/flang/test/Lower/OpenACC/acc-reduction.f90
index a8f7e1fa81ef7e..dcfa77c9f97dbe 100644
--- a/flang/test/Lower/OpenACC/acc-reduction.f90
+++ b/flang/test/Lower/OpenACC/acc-reduction.f90
@@ -743,7 +743,7 @@ subroutine acc_reduction_add_int_array_2d(a, b)
! FIR: %[[RED_ARG1:.*]] = acc.reduction varPtr(%[[ARG1]] : !fir.ref<!fir.array<100x10xi32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<100x10xi32>> {name = "b"}
! HLFIR: %[[RED_ARG1:.*]] = acc.reduction varPtr(%[[DECLARG1]]#1 : !fir.ref<!fir.array<100x10xi32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<100x10xi32>> {name = "b"}
! CHECK: acc.loop reduction(@reduction_add_section_ext100xext10_ref_100x10xi32 -> %[[RED_ARG1]] : !fir.ref<!fir.array<100x10xi32>>) {
-! CHECK: } attributes {collapse = 2 : i64}
+! CHECK: } attributes {collapse = [2], collapseDeviceType = [#acc.device_type<none>]}
subroutine acc_reduction_add_int_array_3d(a, b)
integer :: a(100, 10, 2), b(100, 10, 2)
@@ -765,7 +765,7 @@ subroutine acc_reduction_add_int_array_3d(a, b)
! FIR: %[[RED_ARG1:.*]] = acc.reduction varPtr(%[[ARG1]] : !fir.ref<!fir.array<100x10x2xi32>>) bounds(%{{.*}}, %{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<100x10x2xi32>> {name = "b"}
! HLFIR: %[[RED_ARG1:.*]] = acc.reduction varPtr(%[[DECLARG1]]#1 : !fir.ref<!fir.array<100x10x2xi32>>) bounds(%{{.*}}, %{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<100x10x2xi32>> {name = "b"}
! CHECK: acc.loop reduction(@reduction_add_section_ext100xext10xext2_ref_100x10x2xi32 -> %[[RED_ARG1]] : !fir.ref<!fir.array<100x10x2xi32>>)
-! CHECK: } attributes {collapse = 3 : i64}
+! CHECK: } attributes {collapse = [3], collapseDeviceType = [#acc.device_type<none>]}
subroutine acc_reduction_add_float(a, b)
real :: a(100), b
@@ -938,7 +938,7 @@ subroutine acc_reduction_min_float_array2d(a, b)
! FIR: %[[RED_ARG1:.*]] = acc.reduction varPtr(%[[ARG1]] : !fir.ref<!fir.array<100x10xf32>>) bounds(%3, %5) -> !fir.ref<!fir.array<100x10xf32>> {name = "b"}
! HLFIR: %[[RED_ARG1:.*]] = acc.reduction varPtr(%[[DECLARG1]]#1 : !fir.ref<!fir.array<100x10xf32>>) bounds(%{{.*}}, %{{.*}}) -> !fir.ref<!fir.array<100x10xf32>> {name = "b"}
! CHECK: acc.loop reduction(@reduction_min_section_ext100xext10_ref_100x10xf32 -> %[[RED_ARG1]] : !fir.ref<!fir.array<100x10xf32>>)
-! CHECK: attributes {collapse = 2 : i64}
+! CHECK: attributes {collapse = [2], collapseDeviceType = [#acc.device_type<none>]}
subroutine acc_reduction_max_int(a, b)
integer :: a(100)
diff --git a/flang/test/Lower/OpenACC/acc-serial-loop.f90 b/flang/test/Lower/OpenACC/acc-serial-loop.f90
index 712bfc80ce387c..6041e7fb1b4906 100644
--- a/flang/test/Lower/OpenACC/acc-serial-loop.f90
+++ b/flang/test/Lower/OpenACC/acc-serial-loop.f90
@@ -411,7 +411,7 @@ subroutine acc_serial_loop
! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {seq}
+! CHECK-NEXT: } attributes {seq = [#acc.device_type<none>]}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -424,7 +424,7 @@ subroutine acc_serial_loop
! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {auto}
+! CHECK-NEXT: } attributes {auto_ = [#acc.device_type<none>]}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -437,7 +437,7 @@ subroutine acc_serial_loop
! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {independent}
+! CHECK-NEXT: } attributes {independent = [#acc.device_type<none>]}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -447,10 +447,10 @@ subroutine acc_serial_loop
END DO
! CHECK: acc.serial {
-! CHECK: acc.loop gang {
+! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: }{{$}}
+! CHECK-NEXT: } attributes {gang = [#acc.device_type<none>]}{{$}}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -461,7 +461,7 @@ subroutine acc_serial_loop
! CHECK: acc.serial {
! CHECK: [[GANGNUM1:%.*]] = arith.constant 8 : i32
-! CHECK-NEXT: acc.loop gang(num=[[GANGNUM1]] : i32) {
+! CHECK-NEXT: acc.loop gang({num=[[GANGNUM1]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -475,7 +475,7 @@ subroutine acc_serial_loop
! CHECK: acc.serial {
! CHECK: [[GANGNUM2:%.*]] = fir.load %{{.*}} : !fir.ref<i32>
-! CHECK-NEXT: acc.loop gang(num=[[GANGNUM2]] : i32) {
+! CHECK-NEXT: acc.loop gang({num=[[GANGNUM2]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -488,7 +488,7 @@ subroutine acc_serial_loop
END DO
! CHECK: acc.serial {
-! CHECK: acc.loop gang(num=%{{.*}} : i32, static=%{{.*}} : i32) {
+! CHECK: acc.loop gang({num=%{{.*}} : i32, static=%{{.*}} : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -500,10 +500,10 @@ subroutine acc_serial_loop
a(i) = b(i)
END DO
! CHECK: acc.serial {
-! CHECK: acc.loop vector {
+! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: }{{$}}
+! CHECK-NEXT: } attributes {vector = [#acc.device_type<none>]}{{$}}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -541,10 +541,10 @@ subroutine acc_serial_loop
END DO
! CHECK: acc.serial {
-! CHECK: acc.loop worker {
+! CHECK: acc.loop {
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: }{{$}}
+! CHECK-NEXT: } attributes {worker = [#acc.device_type<none>]}{{$}}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -574,7 +574,7 @@ subroutine acc_serial_loop
! CHECK: fir.do_loop
! CHECK: fir.do_loop
! CHECK: acc.yield
-! CHECK-NEXT: } attributes {collapse = 2 : i64}
+! CHECK-NEXT: } attributes {collapse = [2], collapseDeviceType = [#acc.device_type<none>]}
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -605,7 +605,7 @@ subroutine acc_serial_loop
! CHECK: acc.serial {
! CHECK: [[TILESIZE:%.*]] = arith.constant 2 : i32
-! CHECK: acc.loop tile([[TILESIZE]] : i32) {
+! CHECK: acc.loop tile({[[TILESIZE]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -619,7 +619,7 @@ subroutine acc_serial_loop
! CHECK: acc.serial {
! CHECK: [[TILESIZEM1:%.*]] = arith.constant -1 : i32
-! CHECK: acc.loop tile([[TILESIZEM1]] : i32) {
+! CHECK: acc.loop tile({[[TILESIZEM1]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -636,7 +636,7 @@ subroutine acc_serial_loop
! CHECK: acc.serial {
! CHECK: [[TILESIZE1:%.*]] = arith.constant 2 : i32
! CHECK: [[TILESIZE2:%.*]] = arith.constant 2 : i32
-! CHECK: acc.loop tile([[TILESIZE1]], [[TILESIZE2]] : i32, i32) {
+! CHECK: acc.loop tile({[[TILESIZE1]] : i32, [[TILESIZE2]] : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -649,7 +649,7 @@ subroutine acc_serial_loop
END DO
! CHECK: acc.serial {
-! CHECK: acc.loop tile(%{{.*}} : i32) {
+! CHECK: acc.loop tile({%{{.*}} : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
@@ -664,7 +664,7 @@ subroutine acc_serial_loop
END DO
! CHECK: acc.serial {
-! CHECK: acc.loop tile(%{{.*}}, %{{.*}} : i32, i32) {
+! CHECK: acc.loop tile({%{{.*}} : i32, %{{.*}} : i32}) {
! CHECK: fir.do_loop
! CHECK: acc.yield
! CHECK-NEXT: }{{$}}
diff --git a/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td b/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
index 234c1076e14e3b..f10202846c68ea 100644
--- a/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
+++ b/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
@@ -196,6 +196,27 @@ def DeviceTypeArrayAttr :
let constBuilderCall = ?;
}
+// Gang arg type enumeration
+def OpenACC_GangArgNum : I32EnumAttrCase<"Num", 0, "Num">;
+def OpenACC_GangArgDim : I32EnumAttrCase<"Dim", 1, "Dim">;
+def OpenACC_GangArgStatic : I32EnumAttrCase<"Static", 2, "Static">;
+
+def OpenACC_GangArgType : I32EnumAttr<"GangArgType",
+ "Differentiate the different gang arg values",
+ [OpenACC_GangArgNum, OpenACC_GangArgDim, OpenACC_GangArgStatic]> {
+ let genSpecializedAttr = 0;
+ let cppNamespace = "::mlir::acc";
+}
+def OpenACC_GangArgTypeAttr : EnumAttr<OpenACC_Dialect,
+ OpenACC_GangArgType,
+ "gang_arg_type"> {
+ let assemblyFormat = [{ ```<` $value `>` }];
+}
+def GangArgTypeArrayAttr :
+ TypedArrayAttrBase<OpenACC_GangArgTypeAttr, "gang arg type array attribute"> {
+ let constBuilderCall = ?;
+}
+
// Define a resource for the OpenACC runtime counters.
def OpenACC_RuntimeCounters : Resource<"::mlir::acc::RuntimeCounters">;
@@ -976,7 +997,7 @@ def OpenACC_ParallelOp : OpenACC_Op<"parallel",
`)`
| `vector_length` `(` custom<DeviceTypeOperands>($vectorLength,
type($vectorLength), $vectorLengthDeviceType) `)`
- | `wait` `(` custom<WaitOperands>($waitOperands,
+ | `wait` `(` custom<DeviceTypeOperandsWithSegment>($waitOperands,
type($waitOperands), $waitOperandsDeviceType, $waitOperandsSegments) `)`
| `self` `(` $selfCond `)`
| `if` `(` $ifCond `)`
@@ -1075,7 +1096,7 @@ def OpenACC_SerialOp : OpenACC_Op<"serial",
| `private` `(` custom<SymOperandList>(
$gangPrivateOperands, type($gangPrivateOperands), $privatizations)
`)`
- | `wait` `(` custom<WaitOperands>($waitOperands,
+ | `wait` `(` custom<DeviceTypeOperandsWithSegment>($waitOperands,
type($waitOperands), $waitOperandsDeviceType, $waitOperandsSegments) `)`
| `self` `(` $selfCond `)`
| `if` `(` $ifCond `)`
@@ -1196,8 +1217,9 @@ def OpenACC_KernelsOp : OpenACC_Op<"kernels",
type($numWorkers), $numWorkersDeviceType) `)`
| `vector_length` `(` custom<DeviceTypeOperands>($vectorLength,
type($vectorLength), $vectorLengthDeviceType) `)`
- | `wait` `(` custom<WaitOperands>($waitOperands,
- type($waitOperands), $waitOperandsDeviceType, $waitOperandsSegments) `)`
+ | `wait` `(` custom<DeviceTypeOperandsWithSegment>($waitOperands,
+ type($waitOperands), $waitOperandsDeviceType, $waitOperandsSegments)
+ `)`
| `self` `(` $selfCond `)`
| `if` `(` $ifCond `)`
)
@@ -1432,7 +1454,7 @@ def OpenACC_LoopOp : OpenACC_Op<"loop",
Example:
```mlir
- acc.loop gang vector {
+ acc.loop {
scf.for %arg3 = %c0 to %c10 step %c1 {
scf.for %arg4 = %c0 to %c10 step %c1 {
scf.for %arg5 = %c0 to %c10 step %c1 {
@@ -1441,23 +1463,33 @@ def OpenACC_LoopOp : OpenACC_Op<"loop",
}
}
acc.yield
- } attributes { collapse = 3 }
+ } attributes {
+ collapse = [3], gang = [#acc.device_type<none>],
+ vector = [#acc.device_type<none>]
+ }
```
}];
- let arguments = (ins OptionalAttr<I64Attr>:$collapse,
- Optional<IntOrIndex>:$gangNum,
- Optional<IntOrIndex>:$gangDim,
- Optional<IntOrIndex>:$gangStatic,
- Optional<IntOrIndex>:$workerNum,
- Optional<IntOrIndex>:$vectorLength,
- UnitAttr:$seq,
- UnitAttr:$independent,
- UnitAttr:$auto_,
- UnitAttr:$hasGang,
- UnitAttr:$hasWorker,
- UnitAttr:$hasVector,
+ let arguments = (ins
+ OptionalAttr<I64ArrayAttr>:$collapse,
+ OptionalAttr<DeviceTypeArrayAttr>:$collapseDeviceType,
+ Variadic<IntOrIndex>:$gangOperands,
+ OptionalAttr<GangArgTypeArrayAttr>:$gangOperandsArgType,
+ OptionalAttr<DenseI32ArrayAttr>:$gangOperandsSegments,
+ OptionalAttr<DeviceTypeArrayAttr>:$gangOperandsDeviceType,
+ Variadic<IntOrIndex>:$workerNumOperands,
+ OptionalAttr<DeviceTypeArrayAttr>:$workerNumOperandsDeviceType,
+ Variadic<IntOrIndex>:$vectorOperands,
+ OptionalAttr<DeviceTypeArrayAttr>:$vectorOperandsDeviceType,
+ OptionalAttr<DeviceTypeArrayAttr>:$seq,
+ OptionalAttr<DeviceTypeArrayAttr>:$independent,
+ OptionalAttr<DeviceTypeArrayAttr>:$auto_,
+ OptionalAttr<DeviceTypeArrayAttr>:$gang,
+ OptionalAttr<DeviceTypeArrayAttr>:$worker,
+ OptionalAttr<DeviceTypeArrayAttr>:$vector,
Variadic<IntOrIndex>:$tileOperands,
+ OptionalAttr<DenseI32ArrayAttr>:$tileOperandsSegments,
+ OptionalAttr<DeviceTypeArrayAttr>:$tileOperandsDeviceType,
Variadic<OpenACC_PointerLikeTypeInterface>:$cacheOperands,
Variadic<OpenACC_PointerLikeTypeInterface>:$privateOperands,
OptionalAttr<SymbolRefArrayAttr>:$privatizations,
@@ -1480,18 +1512,90 @@ def OpenACC_LoopOp : OpenACC_Op<"loop",
/// The i-th data operand passed.
Value getDataOperand(unsigned i);
+
+ /// Return true if the op has the auto attribute for the
+ /// mlir::acc::DeviceType::None device_type.
+ bool hasAuto();
+ /// Return true if the op has the auto attribute for the given device_type.
+ bool hasAuto(mlir::acc::DeviceType deviceType);
+ /// Return true if the op has the independent attribute for the
+ /// mlir::acc::DeviceType::None device_type.
+ bool hasIndependent();
+ /// Return true if the op has the independent attribute for the given
+ /// device_type.
+ bool hasIndependent(mlir::acc::DeviceType deviceType);
+ /// Return true if the op has the seq attribute for the
+ /// mlir::acc::DeviceType::None device_type.
+ bool hasSeq();
+ /// Return true if the op has the seq attribute for the given device_type.
+ bool hasSeq(mlir::acc::DeviceType deviceType);
+
+ /// Return the value of the vector clause if present.
+ mlir::Value getVectorValue();
+ /// Return the value of the vector clause for the given device_type
+ /// if present.
+ mlir::Value getVectorValue(mlir::acc::DeviceType deviceType);
+ /// Return true if the op has the vector attribute for the
+ /// mlir::acc::DeviceType::None device_type.
+ bool hasVector();
+ /// Return true if the op has the vector attribute for the given
+ /// device_type.
+ bool hasVector(mlir::acc::DeviceType deviceType);
+
+ /// Return the value of the worker clause if present.
+ mlir::Value getWorkerValue();
+ /// Return the value of the worker clause for the given device_type
+ /// if present.
+ mlir::Value getWorkerValue(mlir::acc::DeviceType deviceType);
+ /// Return true if the op has the worker attribute for the
+ /// mlir::acc::DeviceType::None device_type.
+ bool hasWorker();
+ /// Return true if the op has the worker attribute for the given
+ /// device_type.
+ bool hasWorker(mlir::acc::DeviceType deviceType);
+
+ /// Return the values of the tile clause if present.
+ mlir::Operation::operand_range getTileValues();
+ /// Return the values of the tile clause for the given device_type if
+ /// present.
+ mlir::Operation::operand_range
+ getTileValues(mlir::acc::DeviceType deviceType);
+
+ /// Return the value of the collapse clause if present.
+ std::optional<int64_t> getCollapseValue();
+ /// Return the value of the collapse clause for the given device_type
+ /// if present.
+ std::optional<int64_t> getCollapseValue(mlir::acc::DeviceType deviceType);
+
+ /// Return true if the op has the gang attribute for the
+ /// mlir::acc::DeviceType::None device_type.
+ bool hasGangOnly();
+ /// Return true if the op has the gang attribute for the given
+ /// device_type.
+ bool hasGangOnly(mlir::acc::DeviceType deviceType);
+
+ /// Return the value of the worker clause if present.
+ mlir::Value getGangValue(mlir::acc::GangArgType gangArgType);
+ /// Return the value of the worker clause for the given device_type
+ /// if present.
+ mlir::Value getGangValue(mlir::acc::GangArgType gangArgType, mlir::acc::DeviceType deviceType);
}];
let hasCustomAssemblyFormat = 1;
let assemblyFormat = [{
oilist(
- `gang` `` custom<GangClause>($gangNum, type($gangNum), $gangDim, type($gangDim), $gangStatic, type($gangStatic), $hasGang)
- | `worker` `` custom<WorkerClause>($workerNum, type($workerNum), $hasWorker)
- | `vector` `` custom<VectorClause>($vectorLength, type($vectorLength), $hasVector)
+ `gang` `` `(` custom<GangClause>($gangOperands, type($gangOperands),
+ $gangOperandsArgType, $gangOperandsDeviceType,
+ $gangOperandsSegments) `)`
+ | `worker` `` `(` custom<DeviceTypeOperands>($workerNumOperands,
+ type($workerNumOperands), $workerNumOperandsDeviceType) `)`
+ | `vector` `` `(` custom<DeviceTypeOperands>($vectorOperands,
+ type($vectorOperands), $vectorOperandsDeviceType) `)`
| `private` `(` custom<SymOperandList>(
- $privateOperands, type($privateOperands), $privatizations)
+ $privateOperands, type($privateOperands), $privatizations) `)`
+ | `tile` `(` custom<DeviceTypeOperandsWithSegment>($tileOperands,
+ type($tileOperands), $tileOperandsDeviceType, $tileOperandsSegments)
`)`
- | `tile` `(` $tileOperands `:` type($tileOperands) `)`
| `reduction` `(` custom<SymOperandList>(
$reductionOperands, type($reductionOperands), $reductionRecipes)
`)`
diff --git a/mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp b/mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp
index f484eda3268db0..3f0cacf9e44804 100644
--- a/mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp
+++ b/mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp
@@ -16,6 +16,7 @@
#include "mlir/IR/Matchers.h"
#include "mlir/IR/OpImplementation.h"
#include "mlir/Transforms/DialectConversion.h"
+#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/TypeSwitch.h"
using namespace mlir;
@@ -874,7 +875,7 @@ static void printNumGangs(mlir::OpAsmPrinter &p, mlir::Operation *op,
}
}
-static ParseResult parseWaitOperands(
+static ParseResult parseDeviceTypeOperandsWithSegment(
mlir::OpAsmParser &parser,
llvm::SmallVectorImpl<mlir::OpAsmParser::UnresolvedOperand> &operands,
llvm::SmallVectorImpl<Type> &types, mlir::ArrayAttr &deviceTypes,
@@ -886,6 +887,8 @@ static ParseResult parseWaitOperands(
if (failed(parser.parseLBrace()))
return failure();
+ int32_t crtOperandsSize = operands.size();
+
if (failed(parser.parseCommaSeparatedList(
mlir::AsmParser::Delimiter::None, [&]() {
if (parser.parseOperand(operands.emplace_back()) ||
@@ -895,7 +898,7 @@ static ParseResult parseWaitOperands(
})))
return failure();
- seg.push_back(operands.size());
+ seg.push_back(operands.size() - crtOperandsSize);
if (failed(parser.parseRBrace()))
return failure();
@@ -918,11 +921,10 @@ static ParseResult parseWaitOperands(
return success();
}
-static void printWaitOperands(mlir::OpAsmPrinter &p, mlir::Operation *op,
- mlir::OperandRange operands,
- mlir::TypeRange types,
- std::optional<mlir::ArrayAttr> deviceTypes,
- std::optional<mlir::DenseI32ArrayAttr> segments) {
+static void printDeviceTypeOperandsWithSegment(
+ mlir::OpAsmPrinter &p, mlir::Operation *op, mlir::OperandRange operands,
+ mlir::TypeRange types, std::optional<mlir::ArrayAttr> deviceTypes,
+ std::optional<mlir::DenseI32ArrayAttr> segments) {
unsigned opIdx = 0;
for (unsigned i = 0; i < deviceTypes->size(); ++i) {
if (i != 0)
@@ -1208,16 +1210,19 @@ void acc::HostDataOp::getCanonicalizationPatterns(RewritePatternSet &results,
// LoopOp
//===----------------------------------------------------------------------===//
-static ParseResult
-parseGangValue(OpAsmParser &parser, llvm::StringRef keyword,
- std::optional<OpAsmParser::UnresolvedOperand> &value,
- Type &valueType, bool &needComa, bool &newValue) {
+static ParseResult parseGangValue(
+ OpAsmParser &parser, llvm::StringRef keyword,
+ llvm::SmallVectorImpl<mlir::OpAsmParser::UnresolvedOperand> &operands,
+ llvm::SmallVectorImpl<Type> &types,
+ llvm::SmallVector<GangArgTypeAttr> &attributes, GangArgTypeAttr gangArgType,
+ bool &needComa, bool &newValue) {
if (succeeded(parser.parseOptionalKeyword(keyword))) {
if (parser.parseEqual())
return failure();
- value = OpAsmParser::UnresolvedOperand{};
- if (parser.parseOperand(*value) || parser.parseColonType(valueType))
+ if (parser.parseOperand(operands.emplace_back()) ||
+ parser.parseColonType(types.emplace_back()))
return failure();
+ attributes.push_back(gangArgType);
needComa = true;
newValue = true;
}
@@ -1225,19 +1230,27 @@ parseGangValue(OpAsmParser &parser, llvm::StringRef keyword,
}
static ParseResult parseGangClause(
- OpAsmParser &parser, std::optional<OpAsmParser::UnresolvedOperand> &gangNum,
- Type &gangNumType, std::optional<OpAsmParser::UnresolvedOperand> &gangDim,
- Type &gangDimType,
- std::optional<OpAsmParser::UnresolvedOperand> &gangStatic,
- Type &gangStaticType, UnitAttr &hasGang) {
- hasGang = UnitAttr::get(parser.getBuilder().getContext());
- gangNum = std::nullopt;
- gangDim = std::nullopt;
- gangStatic = std::nullopt;
+ OpAsmParser &parser,
+ llvm::SmallVectorImpl<mlir::OpAsmParser::UnresolvedOperand> &gangOperands,
+ llvm::SmallVectorImpl<Type> &gangOperandsType, mlir::ArrayAttr &gangArgType,
+ mlir::ArrayAttr &deviceType, mlir::DenseI32ArrayAttr &segments) {
+ llvm::SmallVector<GangArgTypeAttr> attributes;
+ llvm::SmallVector<DeviceTypeAttr> deviceTypeAttributes;
+ llvm::SmallVector<int32_t> seg;
bool needComa = false;
- // optional gang operands
- if (succeeded(parser.parseOptionalLParen())) {
+ auto argNum = mlir::acc::GangArgTypeAttr::get(parser.getContext(),
+ mlir::acc::GangArgType::Num);
+ auto argDim = mlir::acc::GangArgTypeAttr::get(parser.getContext(),
+ mlir::acc::GangArgType::Dim);
+ auto argStatic = mlir::acc::GangArgTypeAttr::get(
+ parser.getContext(), mlir::acc::GangArgType::Static);
+
+ do {
+ if (failed(parser.parseLBrace()))
+ return failure();
+
+ int32_t crtOperandsSize = gangOperands.size();
while (true) {
bool newValue = false;
bool needValue = false;
@@ -1248,15 +1261,17 @@ static ParseResult parseGangClause(
break;
}
- if (failed(parseGangValue(parser, LoopOp::getGangNumKeyword(), gangNum,
- gangNumType, needComa, newValue)))
+ if (failed(parseGangValue(parser, LoopOp::getGangNumKeyword(),
+ gangOperands, gangOperandsType, attributes,
+ argNum, needComa, newValue)))
return failure();
- if (failed(parseGangValue(parser, LoopOp::getGangDimKeyword(), gangDim,
- gangDimType, needComa, newValue)))
+ if (failed(parseGangValue(parser, LoopOp::getGangDimKeyword(),
+ gangOperands, gangOperandsType, attributes,
+ argDim, needComa, newValue)))
return failure();
if (failed(parseGangValue(parser, LoopOp::getGangStaticKeyword(),
- gangStatic, gangStaticType, needComa,
- newValue)))
+ gangOperands, gangOperandsType, attributes,
+ argStatic, needComa, newValue)))
return failure();
if (!newValue && needValue) {
@@ -1269,86 +1284,168 @@ static ParseResult parseGangClause(
break;
}
- if (!gangNum && !gangDim && !gangStatic) {
- parser.emitError(parser.getCurrentLocation(),
- "expect at least one of num, dim or static values");
+ if (gangOperands.empty())
+ return parser.emitError(
+ parser.getCurrentLocation(),
+ "expect at least one of num, dim or static values");
+
+ if (failed(parser.parseRBrace()))
return failure();
+
+ if (succeeded(parser.parseOptionalLSquare())) {
+ if (parser.parseAttribute(deviceTypeAttributes.emplace_back()) ||
+ parser.parseRSquare())
+ return failure();
+ } else {
+ deviceTypeAttributes.push_back(mlir::acc::DeviceTypeAttr::get(
+ parser.getContext(), mlir::acc::DeviceType::None));
}
- if (failed(parser.parseRParen()))
- return failure();
- }
+ seg.push_back(gangOperands.size() - crtOperandsSize);
+
+ } while (succeeded(parser.parseOptionalComma()));
+
+ llvm::SmallVector<mlir::Attribute> arrayAttr(attributes.begin(),
+ attributes.end());
+ gangArgType = ArrayAttr::get(parser.getContext(), arrayAttr);
+
+ llvm::SmallVector<mlir::Attribute> deviceTypeAttr(
+ deviceTypeAttributes.begin(), deviceTypeAttributes.end());
+ deviceType = ArrayAttr::get(parser.getContext(), deviceTypeAttr);
+ segments = DenseI32ArrayAttr::get(parser.getContext(), seg);
return success();
}
-void printGangClause(OpAsmPrinter &p, Operation *op, Value gangNum,
- Type gangNumType, Value gangDim, Type gangDimType,
- Value gangStatic, Type gangStaticType, UnitAttr hasGang) {
- if (gangNum || gangStatic || gangDim) {
- p << "(";
- if (gangNum) {
- p << LoopOp::getGangNumKeyword() << "=" << gangNum << " : "
- << gangNumType;
- if (gangStatic || gangDim)
- p << ", ";
- }
- if (gangDim) {
- p << LoopOp::getGangDimKeyword() << "=" << gangDim << " : "
- << gangDimType;
- if (gangStatic)
+void printGangClause(OpAsmPrinter &p, Operation *op,
+ mlir::OperandRange operands, mlir::TypeRange types,
+ std::optional<mlir::ArrayAttr> gangArgTypes,
+ std::optional<mlir::ArrayAttr> deviceTypes,
+ std::optional<mlir::DenseI32ArrayAttr> segments) {
+ unsigned opIdx = 0;
+ for (unsigned i = 0; i < deviceTypes->size(); ++i) {
+ if (i != 0)
+ p << ", ";
+ p << "{";
+ for (int32_t j = 0; j < (*segments)[i]; ++j) {
+ if (j != 0)
p << ", ";
+ auto gangArgTypeAttr =
+ mlir::dyn_cast<mlir::acc::GangArgTypeAttr>((*gangArgTypes)[opIdx]);
+ if (gangArgTypeAttr.getValue() == mlir::acc::GangArgType::Num)
+ p << LoopOp::getGangNumKeyword();
+ else if (gangArgTypeAttr.getValue() == mlir::acc::GangArgType::Dim)
+ p << LoopOp::getGangDimKeyword();
+ else if (gangArgTypeAttr.getValue() == mlir::acc::GangArgType::Static)
+ p << LoopOp::getGangStaticKeyword();
+ p << "=" << operands[opIdx] << " : " << operands[opIdx].getType();
+ ++opIdx;
}
- if (gangStatic)
- p << LoopOp::getGangStaticKeyword() << "=" << gangStatic << " : "
- << gangStaticType;
- p << ")";
- }
-}
-static ParseResult
-parseWorkerClause(OpAsmParser &parser,
- std::optional<OpAsmParser::UnresolvedOperand> &workerNum,
- Type &workerNumType, UnitAttr &hasWorker) {
- hasWorker = UnitAttr::get(parser.getBuilder().getContext());
- if (succeeded(parser.parseOptionalLParen())) {
- workerNum = OpAsmParser::UnresolvedOperand{};
- if (parser.parseOperand(*workerNum) ||
- parser.parseColonType(workerNumType) || parser.parseRParen())
- return failure();
+ p << "}";
+ auto deviceTypeAttr =
+ mlir::dyn_cast<mlir::acc::DeviceTypeAttr>((*deviceTypes)[i]);
+ if (deviceTypeAttr.getValue() != mlir::acc::DeviceType::None)
+ p << " [" << (*deviceTypes)[i] << "]";
}
- return success();
}
-void printWorkerClause(OpAsmPrinter &p, Operation *op, Value workerNum,
- Type workerNumType, UnitAttr hasWorker) {
- if (workerNum)
- p << "(" << workerNum << " : " << workerNumType << ")";
+bool hasDuplicateDeviceTypes(
+ std::optional<mlir::ArrayAttr> segments,
+ llvm::SmallSet<mlir::acc::DeviceType, 3> &deviceTypes) {
+ if (!segments)
+ return false;
+ for (auto attr : *segments) {
+ auto deviceTypeAttr = mlir::dyn_cast<mlir::acc::DeviceTypeAttr>(attr);
+ if (deviceTypes.contains(deviceTypeAttr.getValue()))
+ return true;
+ deviceTypes.insert(deviceTypeAttr.getValue());
+ }
+ return false;
}
-static ParseResult
-parseVectorClause(OpAsmParser &parser,
- std::optional<OpAsmParser::UnresolvedOperand> &vectorLength,
- Type &vectorLengthType, UnitAttr &hasVector) {
- hasVector = UnitAttr::get(parser.getBuilder().getContext());
- if (succeeded(parser.parseOptionalLParen())) {
- vectorLength = OpAsmParser::UnresolvedOperand{};
- if (parser.parseOperand(*vectorLength) ||
- parser.parseColonType(vectorLengthType) || parser.parseRParen())
+/// Check for duplicates in the DeviceType array attribute.
+LogicalResult checkDeviceTypes(mlir::ArrayAttr deviceTypes) {
+ llvm::SmallSet<mlir::acc::DeviceType, 3> crtDeviceTypes;
+ if (!deviceTypes)
+ return success();
+ for (auto attr : deviceTypes) {
+ auto deviceTypeAttr =
+ mlir::dyn_cast_or_null<mlir::acc::DeviceTypeAttr>(attr);
+ if (!deviceTypeAttr)
+ return failure();
+ if (crtDeviceTypes.contains(deviceTypeAttr.getValue()))
return failure();
+ crtDeviceTypes.insert(deviceTypeAttr.getValue());
}
return success();
}
-void printVectorClause(OpAsmPrinter &p, Operation *op, Value vectorLength,
- Type vectorLengthType, UnitAttr hasVector) {
- if (vectorLength)
- p << "(" << vectorLength << " : " << vectorLengthType << ")";
-}
-
LogicalResult acc::LoopOp::verify() {
+ // Check collapse
+ if (getCollapseAttr() && !getCollapseDeviceTypeAttr())
+ return emitOpError() << "collapse device_type attr must be define when"
+ << " collapse attr is present";
+
+ if (getCollapseAttr() && getCollapseDeviceTypeAttr() &&
+ getCollapseAttr().getValue().size() !=
+ getCollapseDeviceTypeAttr().getValue().size())
+ return emitOpError() << "collapse attribute count must match collapse"
+ << " device_type count";
+ if (failed(checkDeviceTypes(getCollapseDeviceTypeAttr())))
+ return emitOpError()
+ << "duplicate device_type found in collapseDeviceType attribute";
+
+ // Check gang
+ if (!getGangOperands().empty()) {
+ if (!getGangOperandsArgType())
+ return emitOpError() << "gangOperandsArgType attribute must be defined"
+ << " when gang operands are present";
+
+ if (getGangOperands().size() !=
+ getGangOperandsArgTypeAttr().getValue().size())
+ return emitOpError() << "gangOperandsArgType attribute count must match"
+ << " gangOperands count";
+ }
+ if (getGangAttr() && failed(checkDeviceTypes(getGangAttr())))
+ return emitOpError() << "duplicate device_type found in gang attribute";
+
+ if (failed(verifyDeviceTypeAndSegmentCountMatch(
+ *this, getGangOperands(), getGangOperandsSegmentsAttr(),
+ getGangOperandsDeviceTypeAttr(), "gang")))
+ return failure();
+
+ // Check worker
+ if (failed(checkDeviceTypes(getWorkerAttr())))
+ return emitOpError() << "duplicate device_type found in worker attribute";
+ if (failed(checkDeviceTypes(getWorkerNumOperandsDeviceTypeAttr())))
+ return emitOpError() << "duplicate device_type found in "
+ "workerNumOperandsDeviceType attribute";
+ if (failed(verifyDeviceTypeCountMatch(*this, getWorkerNumOperands(),
+ getWorkerNumOperandsDeviceTypeAttr(),
+ "worker")))
+ return failure();
+
+ // Check vector
+ if (failed(checkDeviceTypes(getVectorAttr())))
+ return emitOpError() << "duplicate device_type found in vector attribute";
+ if (failed(checkDeviceTypes(getVectorOperandsDeviceTypeAttr())))
+ return emitOpError() << "duplicate device_type found in "
+ "vectorOperandsDeviceType attribute";
+ if (failed(verifyDeviceTypeCountMatch(*this, getVectorOperands(),
+ getVectorOperandsDeviceTypeAttr(),
+ "vector")))
+ return failure();
+
+ if (failed(verifyDeviceTypeAndSegmentCountMatch(
+ *this, getTileOperands(), getTileOperandsSegmentsAttr(),
+ getTileOperandsDeviceTypeAttr(), "tile")))
+ return failure();
+
// auto, independent and seq attribute are mutually exclusive.
- if ((getAuto_() && (getIndependent() || getSeq())) ||
- (getIndependent() && getSeq())) {
+ llvm::SmallSet<mlir::acc::DeviceType, 3> deviceTypes;
+ if (hasDuplicateDeviceTypes(getAuto_(), deviceTypes) ||
+ hasDuplicateDeviceTypes(getIndependent(), deviceTypes) ||
+ hasDuplicateDeviceTypes(getSeq(), deviceTypes)) {
return emitError() << "only one of \"" << acc::LoopOp::getAutoAttrStrName()
<< "\", " << getIndependentAttrName() << ", "
<< getSeqAttrName()
@@ -1356,8 +1453,24 @@ LogicalResult acc::LoopOp::verify() {
}
// Gang, worker and vector are incompatible with seq.
- if (getSeq() && (getHasGang() || getHasWorker() || getHasVector()))
- return emitError("gang, worker or vector cannot appear with the seq attr");
+ if (getSeqAttr()) {
+ for (auto attr : getSeqAttr()) {
+ auto deviceTypeAttr = mlir::dyn_cast<mlir::acc::DeviceTypeAttr>(attr);
+ if (hasVector(deviceTypeAttr.getValue()) ||
+ getVectorValue(deviceTypeAttr.getValue()) ||
+ hasWorker(deviceTypeAttr.getValue()) ||
+ getWorkerValue(deviceTypeAttr.getValue()) ||
+ hasGangOnly(deviceTypeAttr.getValue()) ||
+ getGangValue(mlir::acc::GangArgType::Num,
+ deviceTypeAttr.getValue()) ||
+ getGangValue(mlir::acc::GangArgType::Dim,
+ deviceTypeAttr.getValue()) ||
+ getGangValue(mlir::acc::GangArgType::Static,
+ deviceTypeAttr.getValue()))
+ return emitError()
+ << "gang, worker or vector cannot appear with the seq attr";
+ }
+ }
if (failed(checkSymOperandList<mlir::acc::PrivateRecipeOp>(
*this, getPrivatizations(), getPrivateOperands(), "private",
@@ -1381,16 +1494,149 @@ unsigned LoopOp::getNumDataOperands() {
}
Value LoopOp::getDataOperand(unsigned i) {
- unsigned numOptional = getGangNum() ? 1 : 0;
- numOptional += getGangDim() ? 1 : 0;
- numOptional += getGangStatic() ? 1 : 0;
- numOptional += getVectorLength() ? 1 : 0;
- numOptional += getWorkerNum() ? 1 : 0;
+ unsigned numOptional = getGangOperands().size();
+ numOptional += getVectorOperands().size();
+ numOptional += getWorkerNumOperands().size();
numOptional += getTileOperands().size();
numOptional += getCacheOperands().size();
return getOperand(numOptional + i);
}
+bool LoopOp::hasAuto() { return hasAuto(mlir::acc::DeviceType::None); }
+
+bool LoopOp::hasAuto(mlir::acc::DeviceType deviceType) {
+ if (auto arrayAttr = getAuto_()) {
+ if (findSegment(*arrayAttr, deviceType))
+ return true;
+ }
+ return false;
+}
+
+bool LoopOp::hasIndependent() {
+ return hasIndependent(mlir::acc::DeviceType::None);
+}
+
+bool LoopOp::hasIndependent(mlir::acc::DeviceType deviceType) {
+ if (auto arrayAttr = getIndependent()) {
+ if (findSegment(*arrayAttr, deviceType))
+ return true;
+ }
+ return false;
+}
+
+bool LoopOp::hasSeq() { return hasSeq(mlir::acc::DeviceType::None); }
+
+bool LoopOp::hasSeq(mlir::acc::DeviceType deviceType) {
+ if (auto arrayAttr = getSeq()) {
+ if (findSegment(*arrayAttr, deviceType))
+ return true;
+ }
+ return false;
+}
+
+mlir::Value LoopOp::getVectorValue() {
+ return getVectorValue(mlir::acc::DeviceType::None);
+}
+
+mlir::Value LoopOp::getVectorValue(mlir::acc::DeviceType deviceType) {
+ return getValueInDeviceTypeSegment(getVectorOperandsDeviceType(),
+ getVectorOperands(), deviceType);
+}
+
+bool LoopOp::hasVector() { return hasVector(mlir::acc::DeviceType::None); }
+
+bool LoopOp::hasVector(mlir::acc::DeviceType deviceType) {
+ if (auto arrayAttr = getVector()) {
+ if (findSegment(*arrayAttr, deviceType))
+ return true;
+ }
+ return false;
+}
+
+mlir::Value LoopOp::getWorkerValue() {
+ return getWorkerValue(mlir::acc::DeviceType::None);
+}
+
+mlir::Value LoopOp::getWorkerValue(mlir::acc::DeviceType deviceType) {
+ return getValueInDeviceTypeSegment(getWorkerNumOperandsDeviceType(),
+ getWorkerNumOperands(), deviceType);
+}
+
+bool LoopOp::hasWorker() { return hasWorker(mlir::acc::DeviceType::None); }
+
+bool LoopOp::hasWorker(mlir::acc::DeviceType deviceType) {
+ if (auto arrayAttr = getWorker()) {
+ if (findSegment(*arrayAttr, deviceType))
+ return true;
+ }
+ return false;
+}
+
+mlir::Operation::operand_range LoopOp::getTileValues() {
+ return getTileValues(mlir::acc::DeviceType::None);
+}
+
+mlir::Operation::operand_range
+LoopOp::getTileValues(mlir::acc::DeviceType deviceType) {
+ return getValuesFromSegments(getTileOperandsDeviceType(), getTileOperands(),
+ getTileOperandsSegments(), deviceType);
+}
+
+std::optional<int64_t> LoopOp::getCollapseValue() {
+ return getCollapseValue(mlir::acc::DeviceType::None);
+}
+
+std::optional<int64_t>
+LoopOp::getCollapseValue(mlir::acc::DeviceType deviceType) {
+ if (!getCollapseAttr())
+ return std::nullopt;
+ if (auto pos = findSegment(getCollapseDeviceTypeAttr(), deviceType)) {
+ auto intAttr =
+ mlir::dyn_cast<IntegerAttr>(getCollapseAttr().getValue()[*pos]);
+ return intAttr.getValue().getZExtValue();
+ }
+ return std::nullopt;
+}
+
+mlir::Value LoopOp::getGangValue(mlir::acc::GangArgType gangArgType) {
+ return getGangValue(gangArgType, mlir::acc::DeviceType::None);
+}
+
+mlir::Value LoopOp::getGangValue(mlir::acc::GangArgType gangArgType,
+ mlir::acc::DeviceType deviceType) {
+ if (getGangOperands().empty())
+ return {};
+ if (auto pos = findSegment(*getGangOperandsDeviceType(), deviceType)) {
+ int32_t nbOperandsBefore = 0;
+ for (unsigned i = 0; i < *pos; ++i)
+ nbOperandsBefore += (*getGangOperandsSegments())[i];
+ mlir::Operation::operand_range values =
+ getGangOperands()
+ .drop_front(nbOperandsBefore)
+ .take_front((*getGangOperandsSegments())[*pos]);
+
+ int32_t argTypeIdx = nbOperandsBefore;
+ for (auto value : values) {
+ auto gangArgTypeAttr = mlir::dyn_cast<mlir::acc::GangArgTypeAttr>(
+ (*getGangOperandsArgType())[argTypeIdx]);
+ if (gangArgTypeAttr.getValue() == gangArgType)
+ return value;
+ ++argTypeIdx;
+ }
+ }
+ return {};
+}
+
+bool LoopOp::hasGangOnly() { return hasGangOnly(mlir::acc::DeviceType::None); }
+
+bool LoopOp::hasGangOnly(mlir::acc::DeviceType deviceType) {
+ if (auto arrayAttr = getGang()) {
+ if (findSegment(*arrayAttr, deviceType))
+ return true;
+ }
+ return false;
+}
+
//===----------------------------------------------------------------------===//
// DataOp
//===----------------------------------------------------------------------===//
diff --git a/mlir/test/Dialect/OpenACC/invalid.mlir b/mlir/test/Dialect/OpenACC/invalid.mlir
index c18d964b370f2c..5dcdb3a37e4e3b 100644
--- a/mlir/test/Dialect/OpenACC/invalid.mlir
+++ b/mlir/test/Dialect/OpenACC/invalid.mlir
@@ -1,58 +1,58 @@
// RUN: mlir-opt -split-input-file -verify-diagnostics %s
// expected-error at +1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop gang {
+acc.loop {
"test.openacc_dummy_op"() : () -> ()
acc.yield
-} attributes {seq}
+} attributes {seq = [#acc.device_type<none>], gang = [#acc.device_type<none>]}
// -----
// expected-error at +1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop worker {
+acc.loop {
"test.openacc_dummy_op"() : () -> ()
acc.yield
-} attributes {seq}
+} attributes {seq = [#acc.device_type<none>], worker = [#acc.device_type<none>]}
// -----
// expected-error at +1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop vector {
+acc.loop {
"test.openacc_dummy_op"() : () -> ()
acc.yield
-} attributes {seq}
+} attributes {seq = [#acc.device_type<none>], vector = [#acc.device_type<none>]}
// -----
// expected-error at +1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop gang worker {
+acc.loop {
"test.openacc_dummy_op"() : () -> ()
acc.yield
-} attributes {seq}
+} attributes {seq = [#acc.device_type<none>], worker = [#acc.device_type<none>], gang = [#acc.device_type<none>]}
// -----
// expected-error at +1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop gang vector {
+acc.loop {
"test.openacc_dummy_op"() : () -> ()
acc.yield
-} attributes {seq}
+} attributes {seq = [#acc.device_type<none>], vector = [#acc.device_type<none>], gang = [#acc.device_type<none>]}
// -----
// expected-error at +1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop worker vector {
+acc.loop {
"test.openacc_dummy_op"() : () -> ()
acc.yield
-} attributes {seq}
+} attributes {seq = [#acc.device_type<none>], vector = [#acc.device_type<none>], worker = [#acc.device_type<none>]}
// -----
// expected-error at +1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop gang worker vector {
+acc.loop {
"test.openacc_dummy_op"() : () -> ()
acc.yield
-} attributes {seq}
+} attributes {seq = [#acc.device_type<none>], vector = [#acc.device_type<none>], worker = [#acc.device_type<none>], gang = [#acc.device_type<none>]}
// -----
@@ -62,10 +62,31 @@ acc.loop {
// -----
+// expected-error at +1 {{'acc.loop' op duplicate device_type found in gang attribute}}
+acc.loop {
+ acc.yield
+} attributes {gang = [#acc.device_type<none>, #acc.device_type<none>]}
+
+// -----
+
+// expected-error at +1 {{'acc.loop' op duplicate device_type found in worker attribute}}
+acc.loop {
+ acc.yield
+} attributes {worker = [#acc.device_type<none>, #acc.device_type<none>]}
+
+// -----
+
+// expected-error at +1 {{'acc.loop' op duplicate device_type found in vector attribute}}
+acc.loop {
+ acc.yield
+} attributes {vector = [#acc.device_type<none>, #acc.device_type<none>]}
+
+// -----
+
// expected-error at +1 {{only one of "auto", "independent", "seq" can be present at the same time}}
acc.loop {
acc.yield
-} attributes {auto_, seq}
+} attributes {auto_ = [#acc.device_type<none>], seq = [#acc.device_type<none>]}
// -----
@@ -368,7 +389,7 @@ acc.firstprivate.recipe @privatization_i32 : i32 init {
// -----
// expected-error at +1 {{expected ')'}}
-acc.loop gang(static=%i64Value: i64, num=%i64Value: i64 {
+acc.loop gang({static=%i64Value: i64, num=%i64Value: i64} {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
@@ -437,7 +458,7 @@ acc.reduction.recipe @reduction_i64 : i64 reduction_operator<add> init {
// -----
// expected-error at +1 {{new value expected after comma}}
-acc.loop gang(static=%i64Value: i64, ) {
+acc.loop gang({static=%i64Value: i64, ) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
@@ -454,7 +475,7 @@ func.func @fct1(%0 : !llvm.ptr) -> () {
// -----
// expected-error at +1 {{expect at least one of num, dim or static values}}
-acc.loop gang() {
+acc.loop gang({}) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
diff --git a/mlir/test/Dialect/OpenACC/ops.mlir b/mlir/test/Dialect/OpenACC/ops.mlir
index 5a95811685f845..83a133f45cb63e 100644
--- a/mlir/test/Dialect/OpenACC/ops.mlir
+++ b/mlir/test/Dialect/OpenACC/ops.mlir
@@ -11,7 +11,7 @@ func.func @compute1(%A: memref<10x10xf32>, %B: memref<10x10xf32>, %C: memref<10x
%async = arith.constant 1 : i64
acc.parallel async(%async: i64) {
- acc.loop gang vector {
+ acc.loop {
scf.for %arg3 = %c0 to %c10 step %c1 {
scf.for %arg4 = %c0 to %c10 step %c1 {
scf.for %arg5 = %c0 to %c10 step %c1 {
@@ -25,7 +25,7 @@ func.func @compute1(%A: memref<10x10xf32>, %B: memref<10x10xf32>, %C: memref<10x
}
}
acc.yield
- } attributes { collapse = 3 }
+ } attributes { collapse = [3], collapseDeviceType = [#acc.device_type<none>], vector = [#acc.device_type<none>], gang = [#acc.device_type<none>]}
acc.yield
}
@@ -38,7 +38,7 @@ func.func @compute1(%A: memref<10x10xf32>, %B: memref<10x10xf32>, %C: memref<10x
// CHECK-NEXT: %{{.*}} = arith.constant 1 : index
// CHECK-NEXT: [[ASYNC:%.*]] = arith.constant 1 : i64
// CHECK-NEXT: acc.parallel async([[ASYNC]] : i64) {
-// CHECK-NEXT: acc.loop gang vector {
+// CHECK-NEXT: acc.loop {
// CHECK-NEXT: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
// CHECK-NEXT: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
// CHECK-NEXT: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
@@ -52,7 +52,7 @@ func.func @compute1(%A: memref<10x10xf32>, %B: memref<10x10xf32>, %C: memref<10x
// CHECK-NEXT: }
// CHECK-NEXT: }
// CHECK-NEXT: acc.yield
-// CHECK-NEXT: } attributes {collapse = 3 : i64}
+// CHECK-NEXT: } attributes {collapse = [3], collapseDeviceType = [#acc.device_type<none>], gang = [#acc.device_type<none>], vector = [#acc.device_type<none>]}
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
// CHECK-NEXT: return %{{.*}} : memref<10x10xf32>
@@ -80,7 +80,7 @@ func.func @compute2(%A: memref<10x10xf32>, %B: memref<10x10xf32>, %C: memref<10x
}
}
acc.yield
- } attributes {seq}
+ } attributes {seq = [#acc.device_type<none>]}
acc.yield
}
@@ -106,7 +106,7 @@ func.func @compute2(%A: memref<10x10xf32>, %B: memref<10x10xf32>, %C: memref<10x
// CHECK-NEXT: }
// CHECK-NEXT: }
// CHECK-NEXT: acc.yield
-// CHECK-NEXT: } attributes {seq}
+// CHECK-NEXT: } attributes {seq = [#acc.device_type<none>]}
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
// CHECK-NEXT: return %{{.*}} : memref<10x10xf32>
@@ -138,9 +138,9 @@ func.func @compute3(%a: memref<10x10xf32>, %b: memref<10x10xf32>, %c: memref<10x
acc.data dataOperands(%pa, %pb, %pc, %pd: memref<10x10xf32>, memref<10x10xf32>, memref<10xf32>, memref<10xf32>) {
%private = acc.private varPtr(%c : memref<10xf32>) -> memref<10xf32>
acc.parallel num_gangs({%numGangs: i64}) num_workers(%numWorkers: i64 [#acc.device_type<nvidia>]) private(@privatization_memref_10_f32 -> %private : memref<10xf32>) {
- acc.loop gang {
+ acc.loop {
scf.for %x = %lb to %c10 step %st {
- acc.loop worker {
+ acc.loop {
scf.for %y = %lb to %c10 step %st {
%axy = memref.load %a[%x, %y] : memref<10x10xf32>
%bxy = memref.load %b[%x, %y] : memref<10x10xf32>
@@ -148,7 +148,7 @@ func.func @compute3(%a: memref<10x10xf32>, %b: memref<10x10xf32>, %c: memref<10x
memref.store %tmp, %c[%y] : memref<10xf32>
}
acc.yield
- }
+ } attributes {worker = [#acc.device_type<none>]}
acc.loop {
// for i = 0 to 10 step 1
@@ -160,10 +160,10 @@ func.func @compute3(%a: memref<10x10xf32>, %b: memref<10x10xf32>, %c: memref<10x
memref.store %z, %d[%x] : memref<10xf32>
}
acc.yield
- } attributes {seq}
+ } attributes {seq = [#acc.device_type<none>]}
}
acc.yield
- }
+ } attributes {gang = [#acc.device_type<none>]}
acc.yield
}
acc.terminator
@@ -181,9 +181,9 @@ func.func @compute3(%a: memref<10x10xf32>, %b: memref<10x10xf32>, %c: memref<10x
// CHECK: acc.data dataOperands(%{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : memref<10x10xf32>, memref<10x10xf32>, memref<10xf32>, memref<10xf32>) {
// CHECK-NEXT: %[[P_ARG2:.*]] = acc.private varPtr([[ARG2]] : memref<10xf32>) -> memref<10xf32>
// CHECK-NEXT: acc.parallel num_gangs({[[NUMGANG]] : i64}) num_workers([[NUMWORKERS]] : i64 [#acc.device_type<nvidia>]) private(@privatization_memref_10_f32 -> %[[P_ARG2]] : memref<10xf32>) {
-// CHECK-NEXT: acc.loop gang {
+// CHECK-NEXT: acc.loop {
// CHECK-NEXT: scf.for %{{.*}} = [[C0]] to [[C10]] step [[C1]] {
-// CHECK-NEXT: acc.loop worker {
+// CHECK-NEXT: acc.loop {
// CHECK-NEXT: scf.for %{{.*}} = [[C0]] to [[C10]] step [[C1]] {
// CHECK-NEXT: %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
// CHECK-NEXT: %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
@@ -191,7 +191,7 @@ func.func @compute3(%a: memref<10x10xf32>, %b: memref<10x10xf32>, %c: memref<10x
// CHECK-NEXT: memref.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32>
// CHECK-NEXT: }
// CHECK-NEXT: acc.yield
-// CHECK-NEXT: }
+// CHECK-NEXT: } attributes {worker = [#acc.device_type<none>]}
// CHECK-NEXT: acc.loop {
// CHECK-NEXT: scf.for %{{.*}} = [[C0]] to [[C10]] step [[C1]] {
// CHECK-NEXT: %{{.*}} = memref.load %{{.*}}[%{{.*}}] : memref<10xf32>
@@ -200,10 +200,10 @@ func.func @compute3(%a: memref<10x10xf32>, %b: memref<10x10xf32>, %c: memref<10x
// CHECK-NEXT: memref.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32>
// CHECK-NEXT: }
// CHECK-NEXT: acc.yield
-// CHECK-NEXT: } attributes {seq}
+// CHECK-NEXT: } attributes {seq = [#acc.device_type<none>]}
// CHECK-NEXT: }
// CHECK-NEXT: acc.yield
-// CHECK-NEXT: }
+// CHECK-NEXT: } attributes {gang = [#acc.device_type<none>]}
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
// CHECK-NEXT: acc.terminator
@@ -218,15 +218,15 @@ func.func @testloopop(%a : memref<10xf32>) -> () {
%i32Value = arith.constant 128 : i32
%idxValue = arith.constant 8 : index
- acc.loop gang worker vector {
+ acc.loop {
"test.openacc_dummy_op"() : () -> ()
acc.yield
- }
- acc.loop gang(num=%i64Value: i64) {
+ } attributes {vector = [#acc.device_type<none>], worker = [#acc.device_type<none>], gang = [#acc.device_type<none>]}
+ acc.loop gang({num=%i64Value: i64}) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
- acc.loop gang(static=%i64Value: i64) {
+ acc.loop gang({static=%i64Value: i64}) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
@@ -254,31 +254,31 @@ func.func @testloopop(%a : memref<10xf32>) -> () {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
- acc.loop gang(num=%i64Value: i64) worker vector {
+ acc.loop gang({num=%i64Value: i64}) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
- }
- acc.loop gang(num=%i64Value: i64, static=%i64Value: i64) worker(%i64Value: i64) vector(%i64Value: i64) {
+ } attributes {vector = [#acc.device_type<none>], worker = [#acc.device_type<none>]}
+ acc.loop gang({num=%i64Value: i64, static=%i64Value: i64}) worker(%i64Value: i64) vector(%i64Value: i64) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
- acc.loop gang(num=%i32Value: i32, static=%idxValue: index) {
+ acc.loop gang({num=%i32Value: i32, static=%idxValue: index}) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
- acc.loop tile(%i64Value, %i64Value : i64, i64) {
+ acc.loop tile({%i64Value : i64, %i64Value : i64}) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
- acc.loop tile(%i32Value, %i32Value : i32, i32) {
+ acc.loop tile({%i32Value : i32, %i32Value : i32}) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
- acc.loop gang(static=%i64Value: i64, num=%i64Value: i64) {
+ acc.loop gang({static=%i64Value: i64, num=%i64Value: i64}) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
- acc.loop gang(dim=%i64Value : i64, static=%i64Value: i64) {
+ acc.loop gang({dim=%i64Value : i64, static=%i64Value: i64}) {
"test.openacc_dummy_op"() : () -> ()
acc.yield
}
@@ -293,15 +293,15 @@ func.func @testloopop(%a : memref<10xf32>) -> () {
// CHECK: [[I64VALUE:%.*]] = arith.constant 1 : i64
// CHECK-NEXT: [[I32VALUE:%.*]] = arith.constant 128 : i32
// CHECK-NEXT: [[IDXVALUE:%.*]] = arith.constant 8 : index
-// CHECK: acc.loop gang worker vector {
+// CHECK: acc.loop {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
-// CHECK-NEXT: }
-// CHECK: acc.loop gang(num=[[I64VALUE]] : i64) {
+// CHECK-NEXT: } attributes {gang = [#acc.device_type<none>], vector = [#acc.device_type<none>], worker = [#acc.device_type<none>]}
+// CHECK: acc.loop gang({num=[[I64VALUE]] : i64}) {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
-// CHECK: acc.loop gang(static=[[I64VALUE]] : i64) {
+// CHECK: acc.loop gang({static=[[I64VALUE]] : i64}) {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
@@ -329,31 +329,31 @@ func.func @testloopop(%a : memref<10xf32>) -> () {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
-// CHECK: acc.loop gang(num=[[I64VALUE]] : i64) worker vector {
+// CHECK: acc.loop gang({num=[[I64VALUE]] : i64}) {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
-// CHECK-NEXT: }
-// CHECK: acc.loop gang(num=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64) worker([[I64VALUE]] : i64) vector([[I64VALUE]] : i64) {
+// CHECK-NEXT: } attributes {vector = [#acc.device_type<none>], worker = [#acc.device_type<none>]}
+// CHECK: acc.loop gang({num=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64}) worker([[I64VALUE]] : i64) vector([[I64VALUE]] : i64) {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
-// CHECK: acc.loop gang(num=[[I32VALUE]] : i32, static=[[IDXVALUE]] : index) {
+// CHECK: acc.loop gang({num=[[I32VALUE]] : i32, static=[[IDXVALUE]] : index}) {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
-// CHECK: acc.loop tile([[I64VALUE]], [[I64VALUE]] : i64, i64) {
+// CHECK: acc.loop tile({[[I64VALUE]] : i64, [[I64VALUE]] : i64}) {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
-// CHECK: acc.loop tile([[I32VALUE]], [[I32VALUE]] : i32, i32) {
+// CHECK: acc.loop tile({[[I32VALUE]] : i32, [[I32VALUE]] : i32}) {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
-// CHECK: acc.loop gang(num=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64) {
+// CHECK: acc.loop gang({static=[[I64VALUE]] : i64, num=[[I64VALUE]] : i64}) {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
-// CHECK: acc.loop gang(dim=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64) {
+// CHECK: acc.loop gang({dim=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64}) {
// CHECK-NEXT: "test.openacc_dummy_op"() : () -> ()
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
>From d1f82104c7b8b05ca4ee2fe8bf17f0292b2ce56e Mon Sep 17 00:00:00 2001
From: Valentin Clement <clementval at gmail.com>
Date: Thu, 4 Jan 2024 11:09:33 -0800
Subject: [PATCH 2/2] hasGangOnly -> hasGang
---
mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td | 4 ++--
mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp | 6 +++---
2 files changed, 5 insertions(+), 5 deletions(-)
diff --git a/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td b/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
index f10202846c68ea..f3a510bbc26be9 100644
--- a/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
+++ b/mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td
@@ -1569,10 +1569,10 @@ def OpenACC_LoopOp : OpenACC_Op<"loop",
/// Return true if the op has the gang attribute for the
/// mlir::acc::DeviceType::None device_type.
- bool hasGangOnly();
+ bool hasGang();
/// Return true if the op has the gang attribute for the given
/// device_type.
- bool hasGangOnly(mlir::acc::DeviceType deviceType);
+ bool hasGang(mlir::acc::DeviceType deviceType);
/// Return the value of the worker clause if present.
mlir::Value getGangValue(mlir::acc::GangArgType gangArgType);
diff --git a/mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp b/mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp
index 3f0cacf9e44804..f7f67bb3e3fd34 100644
--- a/mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp
+++ b/mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp
@@ -1460,7 +1460,7 @@ LogicalResult acc::LoopOp::verify() {
getVectorValue(deviceTypeAttr.getValue()) ||
hasWorker(deviceTypeAttr.getValue()) ||
getWorkerValue(deviceTypeAttr.getValue()) ||
- hasGangOnly(deviceTypeAttr.getValue()) ||
+ hasGang(deviceTypeAttr.getValue()) ||
getGangValue(mlir::acc::GangArgType::Num,
deviceTypeAttr.getValue()) ||
getGangValue(mlir::acc::GangArgType::Dim,
@@ -1627,9 +1627,9 @@ mlir::Value LoopOp::getGangValue(mlir::acc::GangArgType gangArgType,
return {};
}
-bool LoopOp::hasGangOnly() { return hasGangOnly(mlir::acc::DeviceType::None); }
+bool LoopOp::hasGang() { return hasGang(mlir::acc::DeviceType::None); }
-bool LoopOp::hasGangOnly(mlir::acc::DeviceType deviceType) {
+bool LoopOp::hasGang(mlir::acc::DeviceType deviceType) {
if (auto arrayAttr = getGang()) {
if (findSegment(*arrayAttr, deviceType))
return true;
More information about the cfe-commits
mailing list