Skip to content

[flang][cuda] Inline this_warp() calls #146134

New issue

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

Sign up for GitHub

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

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jun 27, 2025
Merged

[flang][cuda] Inline this_warp() calls #146134

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
997b083
[flang][cuda] Inline this_warp() calls
clementval Jun 27, 2025
acf6cd6
Add comments
clementval Jun 27, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions flang/include/flang/Optimizer/Builder/IntrinsicCall.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -443,6 +443,7 @@ struct IntrinsicLibrary {
fir::ExtendedValue genTranspose(mlir::Type,
llvm::ArrayRef<fir::ExtendedValue>);
mlir::Value genThisGrid(mlir::Type, llvm::ArrayRef<mlir::Value>);
mlir::Value genThisWarp(mlir::Type, llvm::ArrayRef<mlir::Value>);
void genThreadFence(llvm::ArrayRef<fir::ExtendedValue>);
void genThreadFenceBlock(llvm::ArrayRef<fir::ExtendedValue>);
void genThreadFenceSystem(llvm::ArrayRef<fir::ExtendedValue>);
Expand Down
40 changes: 40 additions & 0 deletions flang/lib/Optimizer/Builder/IntrinsicCall.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -933,6 +933,7 @@ static constexpr IntrinsicHandler handlers[]{
/*isElemental=*/false},
{"tand", &I::genTand},
{"this_grid", &I::genThisGrid, {}, /*isElemental=*/false},
{"this_warp", &I::genThisWarp, {}, /*isElemental=*/false},
{"threadfence", &I::genThreadFence, {}, /*isElemental=*/false},
{"threadfence_block", &I::genThreadFenceBlock, {}, /*isElemental=*/false},
{"threadfence_system", &I::genThreadFenceSystem, {}, /*isElemental=*/false},
Expand Down Expand Up @@ -8194,6 +8195,45 @@ mlir::Value IntrinsicLibrary::genThisGrid(mlir::Type resultType,
return res;
}

// THIS_WARP
mlir::Value IntrinsicLibrary::genThisWarp(mlir::Type resultType,
llvm::ArrayRef<mlir::Value> args) {
assert(args.size() == 0);
auto recTy = mlir::cast<fir::RecordType>(resultType);
assert(recTy && "RecordType expepected");
mlir::Value res = builder.create<fir::AllocaOp>(loc, resultType);
mlir::Type i32Ty = builder.getI32Type();

// coalesced_group%size = 32
mlir::Value size = builder.createIntegerConstant(loc, i32Ty, 32);
auto sizeFieldName = recTy.getTypeList()[1].first;
mlir::Type sizeFieldTy = recTy.getTypeList()[1].second;
mlir::Type fieldIndexType = fir::FieldType::get(resultType.getContext());
mlir::Value sizeFieldIndex = builder.create<fir::FieldIndexOp>(
loc, fieldIndexType, sizeFieldName, recTy,
/*typeParams=*/mlir::ValueRange{});
mlir::Value sizeCoord = builder.create<fir::CoordinateOp>(
loc, builder.getRefType(sizeFieldTy), res, sizeFieldIndex);
builder.create<fir::StoreOp>(loc, size, sizeCoord);

// coalesced_group%rank = threadIdx.x & 31 + 1
mlir::Value threadIdX = builder.create<mlir::NVVM::ThreadIdXOp>(loc, i32Ty);
mlir::Value mask = builder.createIntegerConstant(loc, i32Ty, 31);
mlir::Value one = builder.createIntegerConstant(loc, i32Ty, 1);
mlir::Value masked =
builder.create<mlir::arith::AndIOp>(loc, threadIdX, mask);
mlir::Value rank = builder.create<mlir::arith::AddIOp>(loc, masked, one);
auto rankFieldName = recTy.getTypeList()[2].first;
mlir::Type rankFieldTy = recTy.getTypeList()[2].second;
mlir::Value rankFieldIndex = builder.create<fir::FieldIndexOp>(
loc, fieldIndexType, rankFieldName, recTy,
/*typeParams=*/mlir::ValueRange{});
mlir::Value rankCoord = builder.create<fir::CoordinateOp>(
loc, builder.getRefType(rankFieldTy), res, rankFieldIndex);
builder.create<fir::StoreOp>(loc, rank, rankCoord);
return res;
}

// TRAILZ
mlir::Value IntrinsicLibrary::genTrailz(mlir::Type resultType,
llvm::ArrayRef<mlir::Value> args) {
Expand Down
13 changes: 13 additions & 0 deletions flang/module/cooperative_groups.f90
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,11 +20,24 @@ module cooperative_groups
integer(4) :: rank
end type grid_group

type :: coalesced_group
type(c_devptr), private :: handle
integer(4) :: size
integer(4) :: rank
end type coalesced_group

interface
attributes(device) function this_grid()
import
type(grid_group) :: this_grid
end function
end interface

interface this_warp
attributes(device) function this_warp()
import
type(coalesced_group) :: this_warp
end function
end interface

end module
21 changes: 21 additions & 0 deletions flang/test/Lower/CUDA/cuda-cooperative.cuf
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,3 +50,24 @@ end subroutine
! CHECK: fir.store %[[SIZE]] to %[[COORD_SIZE]] : !fir.ref<i32>
! CHECK: %[[COORD_RANK:.*]] = fir.coordinate_of %[[RES]], rank : (!fir.ref<!fir.type<_QMcooperative_groupsTgrid_group{_QMcooperative_groupsTgrid_group.handle:!fir.type<_QM__fortran_builtinsT__builtin_c_devptr{cptr:!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>}>,size:i32,rank:i32}>>) -> !fir.ref<i32>
! CHECK: fir.store %[[RANK]] to %[[COORD_RANK]] : !fir.ref<i32>

attributes(grid_global) subroutine w1()
use cooperative_groups
type(coalesced_group) :: gg
gg = this_warp()
end subroutine

! CHECK: %[[WARPSIZE:.*]] = fir.alloca i32 {bindc_name = "__builtin_warpsize", uniq_name = "_QM__fortran_builtinsEC__builtin_warpsize"}
! CHECK: %[[WARPSIZE_DECL:.*]]:2 = hlfir.declare %[[WARPSIZE]] {uniq_name = "_QM__fortran_builtinsEC__builtin_warpsize"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
! CHECK: %[[COALESCED_GROUP:.*]] = fir.alloca !fir.type<_QMcooperative_groupsTcoalesced_group{_QMcooperative_groupsTcoalesced_group.handle:!fir.type<_QM__fortran_builtinsT__builtin_c_devptr{cptr:!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>}>,size:i32,rank:i32}>
! CHECK: %[[C32:.*]] = arith.constant 32 : i32
! CHECK: %[[SIZE_COORD:.*]] = fir.coordinate_of %[[COALESCED_GROUP]], size : (!fir.ref<!fir.type<_QMcooperative_groupsTcoalesced_group{_QMcooperative_groupsTcoalesced_group.handle:!fir.type<_QM__fortran_builtinsT__builtin_c_devptr{cptr:!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>}>,size:i32,rank:i32}>>) -> !fir.ref<i32>
! CHECK: fir.store %[[C32]] to %[[SIZE_COORD]] : !fir.ref<i32>

! CHECK: %[[THREAD_ID:.*]] = nvvm.read.ptx.sreg.tid.x : i32
! CHECK: %[[C31:.*]] = arith.constant 31 : i32
! CHECK: %[[C1:.*]] = arith.constant 1 : i32
! CHECK: %[[AND:.*]] = arith.andi %[[THREAD_ID]], %[[C31]] : i32
! CHECK: %[[RANK:.*]] = arith.addi %[[AND]], %[[C1]] : i32
! CHECK: %[[RANK_COORD:.*]] = fir.coordinate_of %{{.*}}, rank : (!fir.ref<!fir.type<_QMcooperative_groupsTcoalesced_group{_QMcooperative_groupsTcoalesced_group.handle:!fir.type<_QM__fortran_builtinsT__builtin_c_devptr{cptr:!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>}>,size:i32,rank:i32}>>) -> !fir.ref<i32>
! CHECK: fir.store %[[RANK]] to %[[RANK_COORD]] : !fir.ref<i32>
Loading