Skip to content

GitLab

Commit e9ecf8d1 authored by Astha Rai's avatar Astha Rai
Browse files

fixed errors in client example

parent 11001fa3
Hide whitespace changes
Inline Side-by-side
Showing with 306 additions and 65 deletions
library/include/ck/library/tensor_operation_instance/gpu/transpose/device_transpose_instance.hpp 0 → 100644
View file @ e9ecf8d1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_3d_impl.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using F16 = ck::half_t;
using F32 = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using device_transpose_f16_instances = std::tuple<
// clang-format off FOR 16, 32, 16, 32, 16
DeviceElementwise3dImpl<ck::Tuple<F16>,
ck::Tuple<F16>,
2,
2,
1,
8,
8,
8,
ck::Sequence<8>,
ck::Sequence<8>>,
DeviceElementwise3dImpl<ck::Tuple<F16>,
ck::Tuple<F16>,
2,
2,
1,
8,
8,
8,
ck::Sequence<8>,
ck::Sequence<1>>,
DeviceElementwise3dImpl<ck::Tuple<F16>,
ck::Tuple<F16>,
2,
2,
1,
8,
8,
8,
ck::Sequence<1>,
ck::Sequence<8>>,
DeviceElementwise3dImpl<ck::Tuple<F16>,
ck::Tuple<F16>,
2,
2,
1,
8,
8,
8,
ck::Sequence<1>,
ck::Sequence<1>>,
DeviceElementwise3dImpl<ck::Tuple<F16>,
ck::Tuple<F16>,
2,
2,
1,
8,
1,
1,
ck::Sequence<1>,
ck::Sequence<1>>,
DeviceElementwise3dImpl<ck::Tuple<F16>,
ck::Tuple<F16>,
2,
2,
1,
8,
1,
1,
ck::Sequence<8>,
ck::Sequence<1>>,
DeviceElementwise3dImpl<ck::Tuple<F16>,
ck::Tuple<F16>,
2,
2,
1,
8,
4,
4,
ck::Sequence<1>,
ck::Sequence<1>>,
DeviceElementwise3dImpl<ck::Tuple<F16>,
ck::Tuple<F16>,
2,
2,
1,
8,
4,
4,
ck::Sequence<8>,
ck::Sequence<8>>
// clang-format on
>;
using device_transpose_f32_instances = std::tuple<
// clang-format off // for 16, 8, 16, 32, 8 -> test with instances for fp16
DeviceElementwise3dImpl<ck::Tuple<F32>,
ck::Tuple<F32>,
2,
2,
1,
4,
4,
4,
ck::Sequence<1>,
ck::Sequence<1>>,
DeviceElementwise3dImpl<ck::Tuple<F32>,
ck::Tuple<F32>,
2,
2,
1,
4,
4,
4,
ck::Sequence<8>,
ck::Sequence<1>>,
DeviceElementwise3dImpl<ck::Tuple<F32>,
ck::Tuple<F32>,
2,
2,
1,
4,
4,
4,
ck::Sequence<8>,
ck::Sequence<8>>,
DeviceElementwise3dImpl<ck::Tuple<F32>,
ck::Tuple<F32>,
2,
2,
1,
4,
8,
4,
ck::Sequence<8>,
ck::Sequence<8>>,
DeviceElementwise3dImpl<ck::Tuple<F32>,
ck::Tuple<F32>,
2,
2,
1,
4,
8,
8,
ck::Sequence<8>,
ck::Sequence<8>>,
DeviceElementwise3dImpl<ck::Tuple<F32>,
ck::Tuple<F32>,
2,
2,
1,
4,
8,
8,
ck::Sequence<4>,
ck::Sequence<8>>,
DeviceElementwise3dImpl<ck::Tuple<F32>,
ck::Tuple<F32>,
2,
2,
1,
4,
8,
8,
ck::Sequence<4>,
ck::Sequence<4>>,
DeviceElementwise3dImpl<ck::Tuple<F32>,
ck::Tuple<F32>,
2,
2,
1,
4,
8,
8,
ck::Sequence<8>,
ck::Sequence<4>>,
DeviceElementwise3dImpl<ck::Tuple<F32>,
ck::Tuple<F32>,
2,
2,
1,
4,
4,
8,
ck::Sequence<8>,
ck::Sequence<8>>,
// clang-format on
>;
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/include/ck/library/tensor_operation_instance/gpu/transpose_3d.hpp
View file @ e9ecf8d1
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <vector>
#include <memory>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_3d_impl.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using namespace ck::transpose_op;
void add_device_transpose_f16_instances(
std::vector<std::unique_ptr<DeviceElementwise3dImpl<F16, F16, NCDHW, 3>>>& instances);
void add_device_transpose_f32_instances(
std::vector<std::unique_ptr<DeviceElementwise3dImpl<F32, F32, NCDHW, 3>>>& instances);
template <typename InDataTypeTuple,
typename OutDataTypeTuple,
typename ElementwiseOperation,
index_t NumDim>
struct DeviceOperationInstanceFactory<
ck::tensor_operation::device::
DeviceElementwise3dImpl<InDataTypeTuple, OutDataTypeTuple, ElementwiseOperation, NumDim>>
{
using DeviceOp = DeviceElementwise3dImpl<InDataTypeTuple,
OutDataTypeTuple,
ElementwiseOperation,
NumDim_m, // choose how to set dims
NumDim_n,
NumDim_k,
MPerThread,
NPerThread,
KPerThread,
InScalarPerVectorSeq,
OutScalarPerVectorSeq>;
static auto GetInstances()
{
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
{
add_device_transpose_f32_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
{
add_device_transpose_f16_instances(op_ptrs);
}
}
return op_ptrs;
}
};
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <vector>
#include <memory>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_3d_impl.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_transpose_f16_instances(
std::vector<std::unique_ptr<DeviceElementwise3dImpl<F16, F16, NCDHW, 3>>>& instances);
void add_device_transpose_f32_instances(
std::vector<std::unique_ptr<DeviceElementwise3dImpl<F32, F32, NCDHW, 3>>>& instances);
template <typename InDataTypeTuple,
typename OutDataTypeTuple,
typename ElementwiseOperation,
index_t NumDim>
struct DeviceOperationInstanceFactory<
ck::tensor_operation::device::
DeviceElementwise3dImpl<InDataTypeTuple, OutDataTypeTuple, ElementwiseOperation, NumDim>>
{
using DeviceOp = DeviceElementwise3dImpl<InDataTypeTuple,
OutDataTypeTuple,
ElementwiseOperation,
NumDim_m, // choose how to set dims
NumDim_n,
NumDim_k,
MPerThread,
NPerThread,
KPerThread,
InScalarPerVectorSeq,
OutScalarPerVectorSeq>;
static auto GetInstances()
{
std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
{
add_device_transpose_f32_instances(op_ptrs);
}
else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
{
add_device_transpose_f16_instances(op_ptrs);
}
}
return op_ptrs;
}
};
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/transpose/CMakeLists.txt 0 → 100644
View file @ e9ecf8d1
add_instance_library(device_transpose_instance
device_transpose_instances_3d.cpp
)
library/src/tensor_operation_instance/gpu/transpose/device_transpose_instances_3d.cpp 0 → 100644
View file @ e9ecf8d1
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/transpose/device_transpose_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
void add_device_transpose_f16_instances(
std::vector<std::unique_ptr<DeviceElementwise3dImpl<F16, F16, NCDHW, 3>>>& instances)
{
#ifdef CK_ENABLE_FP16
add_device_operation_instances(instances, device_transpose_f16_instances<F16, F16, NCDHW, 3>{});
#else
ignore = instances;
#endif
}
void add_device_transpose_f32_instances(
std::vector<std::unique_ptr<DeviceElementwise3dImpl<F32, F32, NCDHW, 3>>>& instances)
{
#ifdef CK_ENABLE_FP32
add_device_operation_instances(instances, device_transpose_f32_instances<F32, F32, NCDHW, 3>{});
#else
ignore = instances;
#endif
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment