Skip to content

359 fix fuse case readiopthenbatchreadincompletereadbackiop #360

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
morousg merged 3 commits into from
Aug 26, 2026
Merged

359 fix fuse case readiopthenbatchreadincompletereadbackiop #360

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
34d2441
Fixed OperationData
oamoros Aug 26, 2026
3106a39
Starting to work on BatchRead, to make it simpler with C++20 syntax, …
oamoros Aug 26, 2026
5afc35a
Made BatchRead specializations containing IncompleteReadBackOperation…
morousg Aug 26, 2026
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
142 changes: 48 additions & 94 deletions include/fused_kernel/core/execution_model/operation_model/batch_operations.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -116,7 +116,7 @@ namespace fk {
using OutputArrayType = std::array<Instantiable<Operation>, IOpType::Operation::BATCH>;
OutputArrayType resultingArray{};
for (int i = 0; i < IOpType::Operation::BATCH; i++) {
if constexpr (IOpType::template is<ReadType>) {
if constexpr (IOpType::template is<ReadType> || IOpType::template is<IncompleteReadBackType>) {
resultingArray[i] = Operation::build(std::forward<InstantiableType>(batchIOp).params.opData[i]);
} else {
static_assert(IOpType::template is<WriteType>, "IOpType must be ReadType or WriteType");
Expand Down Expand Up @@ -162,25 +162,20 @@ namespace fk {
Implements build(const ParamsType& params)
*/
template <size_t BATCH_, typename Op>
requires(isCompleteOperation<Op>)
struct BatchRead<PlanePolicy::PROCESS_ALL, BATCH_, Op> {
static_assert(isCompleteOperation<Op>,
"The IOp passed as template parameter is not a complete operation");
using Operation = Op;
static constexpr size_t BATCH = BATCH_;
static constexpr PlanePolicy PP = PlanePolicy::PROCESS_ALL;
private:
using SelfType = BatchRead<PlanePolicy::PROCESS_ALL, BATCH_, Op>;
using Parent = ReadOperation<typename Op::ReadDataType,
BatchReadParams<BATCH, PP, Operation>, typename Op::OutputType,
static_cast<TF>(Operation::THREAD_FUSION), SelfType,
Operation::IS_FUSED_OP>;
public:
FK_STATIC_STRUCT(BatchRead, SelfType)
using Operation = Op;
static constexpr size_t BATCH = BATCH_;
static constexpr PlanePolicy PP = PlanePolicy::PROCESS_ALL;

using ParamsType = BatchReadParams<BATCH, PP, Operation>;
using ReadDataType = typename Operation::ReadDataType;
using InstanceType = ReadType;
using OutputType = typename Operation::OutputType;
using OperationDataType = OperationData<SelfType>;
using InstantiableType = Read<SelfType>;
static constexpr bool IS_FUSED_OP = Operation::IS_FUSED_OP;
static constexpr bool THREAD_FUSION = Operation::THREAD_FUSION;
DECLARE_READ_PARENT_BASIC

FK_HOST_DEVICE_FUSE uint num_elems_x(const Point thread, const OperationDataType& opData) {
return Operation::num_elems_x(thread, opData.params.opData[thread.z]);
Expand All @@ -198,10 +193,6 @@ namespace fk {
return opData.params.activeThreads;
}

template <uint ELEMS_PER_THREAD = 1>
FK_HOST_DEVICE_FUSE auto exec(const Point thread, const OperationDataType& opData) {
return exec<ELEMS_PER_THREAD>(thread, opData.params);
}
template <uint ELEMS_PER_THREAD = 1>
FK_HOST_DEVICE_FUSE auto exec(const Point thread, const ParamsType& params) {
if constexpr (THREAD_FUSION) {
Expand All @@ -210,34 +201,23 @@ namespace fk {
return Operation::exec(thread, params.opData[thread.z]);
}
}
FK_HOST_FUSE InstantiableType build(const OperationDataType& opData) {
return InstantiableType{ opData };
}
FK_HOST_FUSE InstantiableType build(const ParamsType& params) {
return InstantiableType{ {params} };
}
};

template <size_t BATCH_, typename Op>
requires(isCompleteOperation<Op>)
struct BatchRead<PlanePolicy::CONDITIONAL_WITH_DEFAULT, BATCH_, Op> {
static_assert(isCompleteOperation<Op>,
"The IOp passed as template parameter is not a complete operation");
private:
using SelfType = BatchRead<PlanePolicy::CONDITIONAL_WITH_DEFAULT, BATCH_, Op>;
public:
FK_STATIC_STRUCT(BatchRead, SelfType)
using Operation = Op;
static constexpr size_t BATCH = BATCH_;
static constexpr PlanePolicy PP = PlanePolicy::CONDITIONAL_WITH_DEFAULT;

using ParamsType = BatchReadParams<BATCH, PP, Operation, typename Operation::OutputType>;
using ReadDataType = typename Operation::ReadDataType;
using InstanceType = ReadType;
using OutputType = typename Operation::OutputType;
using OperationDataType = OperationData<SelfType>;
using InstantiableType = Read<SelfType>;
static constexpr bool IS_FUSED_OP = Operation::IS_FUSED_OP;
static constexpr bool THREAD_FUSION = false;
private:
using SelfType = BatchRead<PlanePolicy::CONDITIONAL_WITH_DEFAULT, BATCH, Op>;
using Parent = ReadOperation<typename Op::ReadDataType,
BatchReadParams<BATCH, PP, Operation, typename Op::OutputType>,
typename Op::OutputType,
static_cast<TF>(Operation::THREAD_FUSION), SelfType, false>;
public:
FK_STATIC_STRUCT(BatchRead, SelfType)
DECLARE_READ_PARENT_BASIC

FK_HOST_DEVICE_FUSE uint num_elems_x(const Point thread, const OperationDataType& opData) {
return Operation::num_elems_x(thread, opData.params.opData[thread.z]);
Expand All @@ -254,11 +234,6 @@ namespace fk {
FK_HOST_DEVICE_FUSE ActiveThreads getActiveThreads(const OperationDataType& opData) {
return opData.params.activeThreads;
}

template <uint ELEMS_PER_THREAD = 1>
FK_HOST_DEVICE_FUSE auto exec(const Point thread, const OperationDataType& opData) {
return exec<ELEMS_PER_THREAD>(thread, opData.params);
}
template <uint ELEMS_PER_THREAD = 1>
FK_HOST_DEVICE_FUSE auto exec(const Point thread, const ParamsType& params) {
if (params.usedPlanes <= thread.z) {
Expand All @@ -271,13 +246,6 @@ namespace fk {
}
}
}

FK_HOST_FUSE InstantiableType build(const OperationDataType& opData) {
return InstantiableType{ opData };
}
FK_HOST_FUSE InstantiableType build(const ParamsType& params) {
return InstantiableType{ {params} };
}
};

/*
Expand All @@ -286,23 +254,19 @@ namespace fk {
ParamsType = BatchReadParams<PlanePolicy::PROCESS_ALL, BATCH, Op>
*/
template <size_t BATCH_, typename Op>
requires(isIncompleteReadBackOperation<Op>)
struct BatchRead<PlanePolicy::PROCESS_ALL, BATCH_, TypeList<Op>> {
private:
using SelfType = BatchRead<PlanePolicy::PROCESS_ALL, BATCH_, TypeList<Op>>;
public:
FK_STATIC_STRUCT(BatchRead, SelfType)
using Operation = Op;
static constexpr size_t BATCH = BATCH_;
static constexpr PlanePolicy PP = PlanePolicy::PROCESS_ALL;

using ParamsType = BatchReadParams<BATCH, PlanePolicy::PROCESS_ALL, Operation>;
using ReadDataType = typename Operation::ReadDataType;
using InstanceType = ReadType;
using OutputType = typename Operation::OutputType;
using OperationDataType = OperationData<SelfType>;
using InstantiableType = Read<SelfType>;
static constexpr bool IS_FUSED_OP = Operation::IS_FUSED_OP;
static constexpr bool THREAD_FUSION = Operation::THREAD_FUSION;
private:
using SelfType = BatchRead<PP, BATCH, TypeList<Op>>;
using Parent = IncompleteReadBackOperation<typename Operation::ReadDataType,
BatchReadParams<BATCH, PP, Operation>, NullType,
typename Operation::OutputType, SelfType>;
public:
FK_STATIC_STRUCT(BatchRead, SelfType)
DECLARE_INCOMPLETEREADBACK_PARENT_BASIC

FK_HOST_DEVICE_FUSE uint num_elems_x(const Point thread, const OperationDataType& opData) {
return Operation::num_elems_x(thread, opData.params.opData[thread.z]);
Expand All @@ -319,13 +283,6 @@ namespace fk {
FK_HOST_DEVICE_FUSE ActiveThreads getActiveThreads(const OperationDataType& opData) {
return opData.params.activeThreads;
}

FK_HOST_FUSE InstantiableType build(const OperationDataType& opData) {
return InstantiableType{ opData };
}
FK_HOST_FUSE InstantiableType build(const ParamsType& params) {
return InstantiableType{ {params} };
}
};

/*
Expand All @@ -334,23 +291,19 @@ namespace fk {
ParamsType = BatchReadParams<BATCH, PlanePolicy::CONDITIONAL_WITH_DEFAULT, Op, DefType>
*/
template <size_t BATCH_, typename Op, typename DefType>
requires(isIncompleteReadBackOperation<Op>)
struct BatchRead<PlanePolicy::CONDITIONAL_WITH_DEFAULT, BATCH_, TypeList<Op, DefType>> {
private:
using SelfType = BatchRead<PlanePolicy::CONDITIONAL_WITH_DEFAULT, BATCH_, TypeList<Op, DefType>>;
public:
FK_STATIC_STRUCT(BatchRead, SelfType)
using Operation = Op;
static constexpr size_t BATCH = BATCH_;
static constexpr PlanePolicy PP = PlanePolicy::CONDITIONAL_WITH_DEFAULT;

using ParamsType = BatchReadParams<BATCH, PlanePolicy::CONDITIONAL_WITH_DEFAULT, Operation, DefType>;
using ReadDataType = typename Operation::ReadDataType;
using InstanceType = ReadType;
using OutputType = typename Operation::OutputType;
using OperationDataType = OperationData<SelfType>;
using InstantiableType = Read<SelfType>;
static constexpr bool IS_FUSED_OP = Operation::IS_FUSED_OP;
static constexpr bool THREAD_FUSION = false;
private:
using SelfType = BatchRead<PP, BATCH, TypeList<Op, DefType>>;
using Parent = IncompleteReadBackOperation<typename Operation::ReadDataType,
BatchReadParams<BATCH, PP, Operation, DefType>, NullType,
typename Operation::OutputType, SelfType>;
public:
FK_STATIC_STRUCT(BatchRead, SelfType)
DECLARE_INCOMPLETEREADBACK_PARENT_BASIC

FK_HOST_DEVICE_FUSE uint num_elems_x(const Point thread, const OperationDataType& opData) {
return Operation::num_elems_x(thread, opData.params.opData[thread.z]);
Expand All @@ -367,13 +320,6 @@ namespace fk {
FK_HOST_DEVICE_FUSE ActiveThreads getActiveThreads(const OperationDataType& opData) {
return opData.params.activeThreads;
}

FK_HOST_FUSE InstantiableType build(const OperationDataType& opData) {
return InstantiableType{ opData };
}
FK_HOST_FUSE InstantiableType build(const ParamsType& params) {
return InstantiableType{ {params} };
}
};

/*
Expand Down Expand Up @@ -420,7 +366,11 @@ namespace fk {
paramsStore.opData[i] = iOps[i];
}

return BatchReadType::build(paramsStore);
if constexpr (isCompleteOperation<NewOperation>) {
return BatchReadType::build(paramsStore);
} else {
return BatchReadType::build(paramsStore, typename BatchReadType::BackIOp{});
}
}
};

Expand Down Expand Up @@ -453,7 +403,11 @@ namespace fk {
paramsStore.opData[i] = iOps[i];
}

return BatchReadType::build(paramsStore);
if constexpr (isCompleteOperation<NewOperation>) {
return BatchReadType::build(paramsStore);
} else {
return BatchReadType::build(paramsStore, typename BatchReadType::BackIOp{});
}
}
};
// ##################### END BATCH_READ #####################
Expand Down
39 changes: 34 additions & 5 deletions include/fused_kernel/core/execution_model/operation_model/operation_data.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,24 +75,53 @@ namespace fk {
constexpr bool hasParamsArray =
hasParams_v<Operation> && std::is_array_v<typename Operation::ParamsType>;

template <typename Operation>
constexpr bool hasParamsAndBackIOpNoArray =
hasParamsAndBackIOp_v<Operation> && !std::is_array_v<typename Operation::ParamsType> &&
!std::is_array_v<typename Operation::BackIOp>;

template <typename Operation>
constexpr bool hasParamsAndBackIOpArray =
hasParamsAndBackIOp_v<Operation> && std::is_array_v<typename Operation::ParamsType> &&
std::is_array_v<typename Operation::BackIOp>;

// OperationData implementations
template <typename Operation, typename Enabler = void>
template <typename Operation>
struct OperationData;

using ParamsTypes = TypeList<BinaryType, ReadType, WriteType, MidWriteType, OpenType, ClosedType>;
using ParamsAndBackIOpTypes = TypeList<ReadBackType, IncompleteReadBackType, TernaryType>;

template <typename Operation>
struct OperationData<Operation, std::enable_if_t<one_of_v<typename Operation::InstanceType, ParamsTypes>>> {
requires (one_of_v<typename Operation::InstanceType, ParamsTypes>)
struct OperationData<Operation> {
FK_HOST_DEVICE_CNST OperationData() {};
FK_HOST_DEVICE_CNST OperationData(const typename Operation::ParamsType &params_) : params(params_) {}
FK_HOST_DEVICE_CNST OperationData(const typename Operation::ParamsType &params_)
requires(hasParamsNoArray<Operation>) : params(params_) {}
FK_HOST_DEVICE_CNST OperationData(const typename Operation::ParamsType& params_)
requires(hasParamsArray<Operation>) : params{} {
for (size_t i = 0; i < std::extent_v<typename Operation::ParamsType>; ++i) {
params[i] = params_[i];
}
}
typename Operation::ParamsType params{};
};

template <typename Operation>
struct OperationData<Operation, std::enable_if_t<one_of_v<typename Operation::InstanceType, ParamsAndBackIOpTypes>>> {
requires (one_of_v<typename Operation::InstanceType, ParamsAndBackIOpTypes>)
struct OperationData<Operation> {
FK_HOST_DEVICE_CNST OperationData() {};
FK_HOST_DEVICE_CNST OperationData(const typename Operation::ParamsType& params_, const typename Operation::BackIOp& backIOp_) : params(params_), backIOp(backIOp_) {}
FK_HOST_DEVICE_CNST OperationData(const typename Operation::ParamsType& params_, const typename Operation::BackIOp& backIOp_)
requires (hasParamsAndBackIOpNoArray<Operation>) : params(params_), backIOp(backIOp_) {}
FK_HOST_DEVICE_CNST OperationData(const typename Operation::ParamsType& params_, const typename Operation::BackIOp& backIOp_)
requires (hasParamsAndBackIOpArray<Operation>) : params{}, backIOp{} {
for (size_t i = 0; i < std::extent_v<typename Operation::ParamsType>; ++i) {
params[i] = params_[i];
}
for (size_t i = 0; i < std::extent_v<typename Operation::BackIOp>; ++i) {
backIOp[i] = backIOp_[i];
}
}
typename Operation::ParamsType params{};
typename Operation::BackIOp backIOp{};
};
Expand Down
3 changes: 3 additions & 0 deletions include/fused_kernel/core/execution_model/operation_model/operation_types.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -170,6 +170,9 @@ namespace fk {
template <typename T>
constexpr bool isCompleteOperation = IsCompleteOperation<T>::value;

template <typename T>
constexpr bool isIncompleteReadBackOperation = opIs<IncompleteReadBackType, T>;

template <typename Enabler, typename T>
struct is_fused_operation_ : std::false_type {};

Expand Down
Loading