Skip to content

GitLab

"megatron/vscode:/vscode.git/clone" did not exist on "c52c9dbf3e928ce7fbeda2b0681ac15792fc0ff5"
Commit 2edac9f1 authored by Bartlomiej Kocot's avatar Bartlomiej Kocot
Browse files

Integrate universal gemm with conv bwd data

parent 34f3dfdd
Hide whitespace changes
Inline Side-by-side
Showing with 780 additions and 97 deletions
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/mem/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_mem_inter_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
GNHWK,
GKYXC,
Empty_Tuple,
GNHWC,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_mem_instances<2,
GNHWK,
GKYXC,
Empty_Tuple,
GNHWC,
ConvBwdDataDefault,
Interwave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_mem_instances<2,
GNHWK,
GKYXC,
Empty_Tuple,
GNHWC,
ConvBwdDataFilter1x1Stride1Pad0,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/mem/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_mem_intra_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
GNHWK,
GKYXC,
Empty_Tuple,
GNHWC,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_mem_instances<2,
GNHWK,
GKYXC,
Empty_Tuple,
GNHWC,
ConvBwdDataDefault,
Intrawave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_mem_instances<2,
GNHWK,
GKYXC,
Empty_Tuple,
GNHWC,
ConvBwdDataFilter1x1Stride1Pad0,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/mem/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_mem_inter_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataDefault,
Interwave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataFilter1x1Stride1Pad0,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/mem/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_mem_intra_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataDefault,
Intrawave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataFilter1x1Stride1Pad0,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/mem/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_mem_inter_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataDefault,
Interwave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataFilter1x1Stride1Pad0,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/mem/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_mem_intra_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataDefault,
Intrawave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataFilter1x1Stride1Pad0,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/mem/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_mem_inter_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataDefault,
Interwave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataFilter1x1Stride1Pad0,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/mem/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_mem_intra_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataDefault,
Intrawave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_mem_instances<2,
NHWGK,
GKYXC,
Empty_Tuple,
NHWGC,
ConvBwdDataFilter1x1Stride1Pad0,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
View file @ 2edac9f1
# ONLY XDL_AND_WMMA_KERNELS
set(GROUPED_CONV3D_BWD_DATA
xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
xdl/comp/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_comp_instance.cpp
xdl/comp/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_comp_instance.cpp
xdl/comp/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_comp_instance.cpp
xdl/comp/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instance.cpp
xdl/comp/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instance.cpp
xdl/comp/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_comp_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_mem_inter_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_mem_intra_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_mem_intra_instance.cpp
xdl/mem/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_mem_intra_instance.cpp
wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
......@@ -17,7 +32,7 @@ set(GROUPED_CONV3D_BWD_DATA
if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "bf8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
list(APPEND GROUPED_CONV3D_BWD_DATA
xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_input_f16_comp_bf8_f8_instance.cpp)
xdl/mem/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_input_f16_comp_bf8_f8_mem_instance.cpp)
endif()
add_instance_library(device_grouped_conv3d_bwd_data_instance ${GROUPED_CONV3D_BWD_DATA})
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/comp/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_comp_instance.cpp
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_comp_instance.hpp"
namespace ck {
namespace tensor_operation {
......@@ -10,7 +10,7 @@ namespace device {
namespace instance {
// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
// wo, k]
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
GNDHWK,
GKZYXC,
......@@ -27,21 +27,21 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault>{});
device_grouped_conv_bwd_data_xdl_bf16_comp_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0>{});
device_grouped_conv_bwd_data_xdl_bf16_comp_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0>{});
}
} // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/comp/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_comp_instance.cpp
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_comp_instance.hpp"
namespace ck {
namespace tensor_operation {
......@@ -10,7 +10,7 @@ namespace device {
namespace instance {
// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
// wo, k]
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
GNDHWK,
GKZYXC,
......@@ -27,21 +27,21 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault>{});
device_grouped_conv_bwd_data_xdl_f16_comp_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0>{});
device_grouped_conv_bwd_data_xdl_f16_comp_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0>{});
}
} // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/comp/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_comp_instance.cpp
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_comp_instance.hpp"
namespace ck {
namespace tensor_operation {
......@@ -10,7 +10,7 @@ namespace device {
namespace instance {
// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
// wo, k]
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
GNDHWK,
GKZYXC,
......@@ -27,21 +27,21 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault>{});
device_grouped_conv_bwd_data_xdl_f32_comp_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0>{});
device_grouped_conv_bwd_data_xdl_f32_comp_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0>{});
}
} // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/comp/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_comp_instance.cpp
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_comp_instance.hpp"
namespace ck {
namespace tensor_operation {
......@@ -10,7 +10,7 @@ namespace device {
namespace instance {
// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
// g, k]
void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NDHWGK,
GKZYXC,
......@@ -27,21 +27,21 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataDefault>{});
device_grouped_conv_bwd_data_xdl_bf16_comp_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataDefault>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataFilter1x1Stride1Pad0>{});
device_grouped_conv_bwd_data_xdl_bf16_comp_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataFilter1x1Stride1Pad0>{});
}
} // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/comp/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_instance.cpp
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_comp_instance.hpp"
namespace ck {
namespace tensor_operation {
......@@ -10,7 +10,7 @@ namespace device {
namespace instance {
// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
// g, k]
void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NDHWGK,
GKZYXC,
......@@ -27,21 +27,21 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataDefault>{});
device_grouped_conv_bwd_data_xdl_f16_comp_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataDefault>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataFilter1x1Stride1Pad0>{});
device_grouped_conv_bwd_data_xdl_f16_comp_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataFilter1x1Stride1Pad0>{});
}
} // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp → library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/comp/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_comp_instance.cpp
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_comp_instance.hpp"
namespace ck {
namespace tensor_operation {
......@@ -10,7 +10,7 @@ namespace device {
namespace instance {
// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
// g, k]
void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_comp_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
NDHWGK,
GKZYXC,
......@@ -27,21 +27,21 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataDefault>{});
device_grouped_conv_bwd_data_xdl_f32_comp_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataDefault>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataFilter1x1Stride1Pad0>{});
device_grouped_conv_bwd_data_xdl_f32_comp_instances<3,
NDHWGK,
GKZYXC,
Empty_Tuple,
NDHWGC,
ConvBwdDataFilter1x1Stride1Pad0>{});
}
} // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_mem_inter_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
// wo, k]
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_mem_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault,
Interwave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_mem_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_mem_intra_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
// wo, k]
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
BF16,
BF16,
Empty_Tuple,
BF16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_mem_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault,
Intrawave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_bf16_mem_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_mem_inter_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
// wo, k]
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_mem_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault,
Interwave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_mem_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_mem_intra_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
// wo, k]
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
F16,
F16,
Empty_Tuple,
F16,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_mem_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault,
Intrawave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f16_mem_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/mem/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_mem_inter_instance.cpp 0 → 100644
View file @ 2edac9f1
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
// wo, k]
void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
F32,
F32,
Empty_Tuple,
F32,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
// 1. Default
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_mem_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataDefault,
Interwave>{});
// 2. Filter1x1Stride1Pad0
add_device_operation_instances(
instances,
device_grouped_conv_bwd_data_xdl_f32_mem_instances<3,
GNDHWK,
GKZYXC,
Empty_Tuple,
GNDHWC,
ConvBwdDataFilter1x1Stride1Pad0,
Interwave>{});
}
} // 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