Skip to content

GitLab

Commit 667cd6ab authored by illsilin's avatar illsilin
Browse files

merge from public repo

parents 7d50244e 365f39ae
Hide whitespace changes
Inline Side-by-side
Showing with 386 additions and 16 deletions
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/large_tensor/device_grouped_conv2d_fwd_xdl_large_tensor_nhwgc_gkyxc_nhwgk_int8_instance.cpp 0 → 100644
View file @ 667cd6ab
// SPDX-License-Identifier: MIT
// Copyright (c) 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_fwd/device_grouped_conv_fwd_xdl_large_tensor_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_xdl_large_tensor_nhwgc_gkyxc_nhwgk_int8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_large_tensor_int8_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_ngchw_gkyxc_ngkhw_int8_mem_inter_instance.cpp 0 → 100644
View file @ 667cd6ab
// SPDX-License-Identifier: MIT
// Copyright (c) 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_xdl_ngchw_gkyxc_ngkhw_int8_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKYXC,
Empty_Tuple,
NGKHW,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NGCHW,
GKYXC,
Empty_Tuple,
NGKHW,
ConvFwdDefault,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_ngchw_gkyxc_ngkhw_int8_mem_intra_instance.cpp 0 → 100644
View file @ 667cd6ab
// SPDX-License-Identifier: MIT
// Copyright (c) 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_xdl_ngchw_gkyxc_ngkhw_int8_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKYXC,
Empty_Tuple,
NGKHW,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NGCHW,
GKYXC,
Empty_Tuple,
NGKHW,
ConvFwdDefault,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_inter_instance.cpp 0 → 100644
View file @ 667cd6ab
// 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_inter_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Interwave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0,
Interwave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Interwave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Interwave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/mem/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_intra_instance.cpp 0 → 100644
View file @ 667cd6ab
// 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_fwd/device_grouped_conv_fwd_xdl_mem_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_int8_mem_intra_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault,
Intrawave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1P0,
Intrawave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd1x1S1P0,
Intrawave>{});
add_device_operation_instances(instances,
device_grouped_conv_fwd_xdl_int8_mem_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdOddC,
Intrawave>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/merged_groups/device_grouped_conv2d_fwd_xdl_merged_groups_ngchw_gkyxc_ngkhw_int8_instance.cpp 0 → 100644
View file @ 667cd6ab
// SPDX-License-Identifier: MIT
// Copyright (c) 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_fwd/device_grouped_conv_fwd_xdl_merged_groups_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_xdl_merged_groups_ngchw_gkyxc_ngkhw_int8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NGCHW,
GKYXC,
Empty_Tuple,
NGKHW,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_merged_groups_int8_instances<2,
NGCHW,
GKYXC,
Empty_Tuple,
NGKHW,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_merged_groups_int8_instances<2,
NGCHW,
GKYXC,
Empty_Tuple,
NGKHW,
ConvFwd3x3>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/merged_groups/device_grouped_conv2d_fwd_xdl_merged_groups_nhwgc_gkyxc_nhwgk_int8_instance.cpp 0 → 100644
View file @ 667cd6ab
// SPDX-License-Identifier: MIT
// Copyright (c) 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_fwd/device_grouped_conv_fwd_xdl_merged_groups_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
void add_device_grouped_conv2d_fwd_xdl_merged_groups_nhwgc_gkyxc_nhwgk_int8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
int8_t,
int8_t,
Empty_Tuple,
int8_t,
PassThrough,
PassThrough,
PassThrough>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_merged_groups_int8_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwdDefault>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_merged_groups_int8_instances<2,
NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
ConvFwd3x3>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_dynamic_op/xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
View file @ 667cd6ab
...@@ -31,6 +31,8 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_bf16_instanc ...@@ -31,6 +31,8 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_bf16_instanc
Tuple<>, Tuple<>,
NHWGK, NHWGK,
ConvFwdDefault>{}); ConvFwdDefault>{});
#if 0 // Enable with dynamic op optimizations (at now generating a lot of virtual functions cause
// long compilation time)
add_device_operation_instances( add_device_operation_instances(
instances, instances,
device_grouped_conv_fwd_xdl_dynamic_op_bf16_instances<2, device_grouped_conv_fwd_xdl_dynamic_op_bf16_instances<2,
...@@ -47,6 +49,7 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_bf16_instanc ...@@ -47,6 +49,7 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_bf16_instanc
Tuple<>, Tuple<>,
NHWGK, NHWGK,
ConvFwd1x1S1P0>{}); ConvFwd1x1S1P0>{});
#endif
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_dynamic_op/xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f16_instance.cpp
View file @ 667cd6ab
...@@ -31,6 +31,8 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f16_instance ...@@ -31,6 +31,8 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f16_instance
Tuple<>, Tuple<>,
NHWGK, NHWGK,
ConvFwdDefault>{}); ConvFwdDefault>{});
#if 0 // Enable with dynamic op optimizations (at now generating a lot of virtual functions cause
// long compilation time)
add_device_operation_instances( add_device_operation_instances(
instances, instances,
device_grouped_conv_fwd_xdl_dynamic_op_f16_instances<2, device_grouped_conv_fwd_xdl_dynamic_op_f16_instances<2,
...@@ -47,6 +49,7 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f16_instance ...@@ -47,6 +49,7 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f16_instance
Tuple<>, Tuple<>,
NHWGK, NHWGK,
ConvFwd1x1S1P0>{}); ConvFwd1x1S1P0>{});
#endif
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_dynamic_op/xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f32_instance.cpp
View file @ 667cd6ab
...@@ -31,6 +31,8 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f32_instance ...@@ -31,6 +31,8 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f32_instance
Tuple<>, Tuple<>,
NHWGK, NHWGK,
ConvFwdDefault>{}); ConvFwdDefault>{});
#if 0 // Enable with dynamic op optimizations (at now generating a lot of virtual functions cause
// long compilation time)
add_device_operation_instances( add_device_operation_instances(
instances, instances,
device_grouped_conv_fwd_xdl_dynamic_op_f32_instances<2, device_grouped_conv_fwd_xdl_dynamic_op_f32_instances<2,
...@@ -47,6 +49,7 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f32_instance ...@@ -47,6 +49,7 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_f32_instance
Tuple<>, Tuple<>,
NHWGK, NHWGK,
ConvFwd1x1S1P0>{}); ConvFwd1x1S1P0>{});
#endif
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd_dynamic_op/xdl/device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_int8_instance.cpp
View file @ 667cd6ab
...@@ -30,6 +30,8 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_int8_instanc ...@@ -30,6 +30,8 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_int8_instanc
Tuple<>, Tuple<>,
NHWGK, NHWGK,
ConvFwdDefault>{}); ConvFwdDefault>{});
#if 0 // Enable with dynamic op optimizations (at now generating a lot of virtual functions cause
// long compilation time)
add_device_operation_instances( add_device_operation_instances(
instances, instances,
device_grouped_conv_fwd_xdl_dynamic_op_int8_instances<2, device_grouped_conv_fwd_xdl_dynamic_op_int8_instances<2,
...@@ -46,6 +48,7 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_int8_instanc ...@@ -46,6 +48,7 @@ void add_device_grouped_conv2d_fwd_xdl_dynamic_op_nhwgc_gkyxc_nhwgk_int8_instanc
Tuple<>, Tuple<>,
NHWGK, NHWGK,
ConvFwd1x1S1P0>{}); ConvFwd1x1S1P0>{});
#endif
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_dynamic_op/xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
View file @ 667cd6ab
...@@ -31,6 +31,8 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_bf16_inst ...@@ -31,6 +31,8 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_bf16_inst
Tuple<>, Tuple<>,
NDHWGK, NDHWGK,
ConvFwdDefault>{}); ConvFwdDefault>{});
#if 0 // Enable with dynamic op optimizations (at now generating a lot of virtual functions cause
// long compilation time)
add_device_operation_instances( add_device_operation_instances(
instances, instances,
device_grouped_conv_fwd_xdl_dynamic_op_bf16_instances<3, device_grouped_conv_fwd_xdl_dynamic_op_bf16_instances<3,
...@@ -47,6 +49,7 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_bf16_inst ...@@ -47,6 +49,7 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_bf16_inst
Tuple<>, Tuple<>,
NDHWGK, NDHWGK,
ConvFwd1x1S1P0>{}); ConvFwd1x1S1P0>{});
#endif
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_dynamic_op/xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
View file @ 667cd6ab
...@@ -31,6 +31,8 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f16_insta ...@@ -31,6 +31,8 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f16_insta
Tuple<>, Tuple<>,
NDHWGK, NDHWGK,
ConvFwdDefault>{}); ConvFwdDefault>{});
#if 0 // Enable with dynamic op optimizations (at now generating a lot of virtual functions cause
// long compilation time)
add_device_operation_instances( add_device_operation_instances(
instances, instances,
device_grouped_conv_fwd_xdl_dynamic_op_f16_instances<3, device_grouped_conv_fwd_xdl_dynamic_op_f16_instances<3,
...@@ -47,6 +49,7 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f16_insta ...@@ -47,6 +49,7 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f16_insta
Tuple<>, Tuple<>,
NDHWGK, NDHWGK,
ConvFwd1x1S1P0>{}); ConvFwd1x1S1P0>{});
#endif
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_dynamic_op/xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
View file @ 667cd6ab
...@@ -31,6 +31,8 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f32_insta ...@@ -31,6 +31,8 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f32_insta
Tuple<>, Tuple<>,
NDHWGK, NDHWGK,
ConvFwdDefault>{}); ConvFwdDefault>{});
#if 0 // Enable with dynamic op optimizations (at now generating a lot of virtual functions cause
// long compilation time)
add_device_operation_instances( add_device_operation_instances(
instances, instances,
device_grouped_conv_fwd_xdl_dynamic_op_f32_instances<3, device_grouped_conv_fwd_xdl_dynamic_op_f32_instances<3,
...@@ -47,6 +49,7 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f32_insta ...@@ -47,6 +49,7 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_f32_insta
Tuple<>, Tuple<>,
NDHWGK, NDHWGK,
ConvFwd1x1S1P0>{}); ConvFwd1x1S1P0>{});
#endif
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_dynamic_op/xdl/device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
View file @ 667cd6ab
...@@ -30,6 +30,8 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_int8_inst ...@@ -30,6 +30,8 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_int8_inst
Tuple<>, Tuple<>,
NDHWGK, NDHWGK,
ConvFwdDefault>{}); ConvFwdDefault>{});
#if 0 // Enable with dynamic op optimizations (at now generating a lot of virtual functions cause
// long compilation time)
add_device_operation_instances( add_device_operation_instances(
instances, instances,
device_grouped_conv_fwd_xdl_dynamic_op_int8_instances<3, device_grouped_conv_fwd_xdl_dynamic_op_int8_instances<3,
...@@ -46,6 +48,7 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_int8_inst ...@@ -46,6 +48,7 @@ void add_device_grouped_conv3d_fwd_xdl_dynamic_op_ndhwgc_gkzyxc_ndhwgk_int8_inst
Tuple<>, Tuple<>,
NDHWGK, NDHWGK,
ConvFwd1x1S1P0>{}); ConvFwd1x1S1P0>{});
#endif
} }
} // namespace instance } // namespace instance
......
library/src/tensor_operation_instance/gpu/mha/CMakeLists.txt
View file @ 667cd6ab
...@@ -27,11 +27,6 @@ rocm_install(FILES ${MHA_HEADERS} DESTINATION include/ck_tile/ops) ...@@ -27,11 +27,6 @@ rocm_install(FILES ${MHA_HEADERS} DESTINATION include/ck_tile/ops)
# headers for building lib # headers for building lib
file(COPY ${MHA_HEADERS} DESTINATION ${FMHA_CPP_FOLDER}) file(COPY ${MHA_HEADERS} DESTINATION ${FMHA_CPP_FOLDER})
# Delete the blob file if it exists to avoid append of old content.
if(EXISTS ${FMHA_CPP_FOLDER}/blob_list.txt)
file(REMOVE ${FMHA_CPP_FOLDER}/blob_list.txt)
endif()
set(FMHA_KNOWN_APIS "fwd,fwd_splitkv,fwd_appendkv,bwd") set(FMHA_KNOWN_APIS "fwd,fwd_splitkv,fwd_appendkv,bwd")
# generate a list of kernels, but not actually emit files at config stage # generate a list of kernels, but not actually emit files at config stage
......
profiler/src/profile_gemm_universal.cpp
View file @ 667cd6ab
...@@ -101,7 +101,9 @@ int profile_gemm_universal(int argc, char* argv[]) ...@@ -101,7 +101,9 @@ int profile_gemm_universal(int argc, char* argv[])
using F32 = float; using F32 = float;
using F16 = ck::half_t; using F16 = ck::half_t;
using BF16 = ck::bhalf_t; using BF16 = ck::bhalf_t;
using F8 = ck::f8_t; #if defined(CK_USE_FP8_ON_UNSUPPORTED_ARCH)
using F8 = ck::f8_t;
#endif
using Row = ck::tensor_layout::gemm::RowMajor; using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor; using Col = ck::tensor_layout::gemm::ColumnMajor;
...@@ -162,6 +164,7 @@ int profile_gemm_universal(int argc, char* argv[]) ...@@ -162,6 +164,7 @@ int profile_gemm_universal(int argc, char* argv[])
{ {
return profile(F16{}, F16{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{}); return profile(F16{}, F16{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{});
} }
#if defined(CK_USE_FP8_ON_UNSUPPORTED_ARCH)
else if(data_type == GemmDataType::F16_F8_F16 && layout == GemmMatrixLayout::MK_KN_MN) else if(data_type == GemmDataType::F16_F8_F16 && layout == GemmMatrixLayout::MK_KN_MN)
{ {
return profile(F16{}, F8{}, F16{}, F32{}, F16{}, Row{}, Row{}, Row{}); return profile(F16{}, F8{}, F16{}, F32{}, F16{}, Row{}, Row{}, Row{});
...@@ -178,6 +181,7 @@ int profile_gemm_universal(int argc, char* argv[]) ...@@ -178,6 +181,7 @@ int profile_gemm_universal(int argc, char* argv[])
{ {
return profile(F8{}, F16{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{}); return profile(F8{}, F16{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{});
} }
#endif
else if(data_type == GemmDataType::BF16_BF16_BF16 && layout == GemmMatrixLayout::MK_KN_MN) else if(data_type == GemmDataType::BF16_BF16_BF16 && layout == GemmMatrixLayout::MK_KN_MN)
{ {
return profile(BF16{}, BF16{}, BF16{}, F32{}, BF16{}, Row{}, Row{}, Row{}); return profile(BF16{}, BF16{}, BF16{}, F32{}, BF16{}, Row{}, Row{}, Row{});
...@@ -194,6 +198,7 @@ int profile_gemm_universal(int argc, char* argv[]) ...@@ -194,6 +198,7 @@ int profile_gemm_universal(int argc, char* argv[])
{ {
return profile(BF16{}, BF16{}, BF16{}, F32{}, BF16{}, Col{}, Row{}, Row{}); return profile(BF16{}, BF16{}, BF16{}, F32{}, BF16{}, Col{}, Row{}, Row{});
} }
#if defined(CK_USE_FP8_ON_UNSUPPORTED_ARCH)
else if(data_type == GemmDataType::F8_F8_BF16 && layout == GemmMatrixLayout::MK_KN_MN) else if(data_type == GemmDataType::F8_F8_BF16 && layout == GemmMatrixLayout::MK_KN_MN)
{ {
return profile(F8{}, F8{}, F8{}, F32{}, BF16{}, Row{}, Row{}, Row{}); return profile(F8{}, F8{}, F8{}, F32{}, BF16{}, Row{}, Row{}, Row{});
...@@ -202,6 +207,7 @@ int profile_gemm_universal(int argc, char* argv[]) ...@@ -202,6 +207,7 @@ int profile_gemm_universal(int argc, char* argv[])
{ {
return profile(F8{}, F8{}, F8{}, F32{}, BF16{}, Row{}, Col{}, Row{}); return profile(F8{}, F8{}, F8{}, F32{}, BF16{}, Row{}, Col{}, Row{});
} }
#endif
else else
{ {
std::cout << "this data_type & layout is not implemented" << std::endl; std::cout << "this data_type & layout is not implemented" << std::endl;
......
test/CMakeLists.txt
View file @ 667cd6ab
...@@ -64,11 +64,11 @@ function(add_test_executable TEST_NAME) ...@@ -64,11 +64,11 @@ function(add_test_executable TEST_NAME)
#only continue if there are some source files left on the list #only continue if there are some source files left on the list
if(ARGN) if(ARGN)
if(ARGN MATCHES "_xdl") if(ARGN MATCHES "_xdl")
list(REMOVE_ITEM TEST_TARGETS gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1200 gfx1201) list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1200 gfx1201)
elseif(ARGN MATCHES "_wmma") elseif(ARGN MATCHES "_wmma")
list(REMOVE_ITEM TEST_TARGETS gfx908 gfx90a gfx940 gfx941 gfx942 gfx1030) list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx940 gfx941 gfx942 gfx1030)
elseif(ARGN MATCHES "_smfmac") elseif(ARGN MATCHES "_smfmac")
list(REMOVE_ITEM TEST_TARGETS gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx908 gfx90a gfx1200 gfx1201) list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx908 gfx90a gfx1200 gfx1201)
endif() endif()
set_source_files_properties(${ARGN} PROPERTIES LANGUAGE HIP) set_source_files_properties(${ARGN} PROPERTIES LANGUAGE HIP)
add_executable(${TEST_NAME} ${ARGN}) add_executable(${TEST_NAME} ${ARGN})
...@@ -141,11 +141,11 @@ function(add_gtest_executable TEST_NAME) ...@@ -141,11 +141,11 @@ function(add_gtest_executable TEST_NAME)
#only continue if there are some source files left on the list #only continue if there are some source files left on the list
if(ARGN) if(ARGN)
if(ARGN MATCHES "_xdl") if(ARGN MATCHES "_xdl")
list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1200 gfx1201) list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx1200 gfx1201)
elseif(ARGN MATCHES "_wmma") elseif(ARGN MATCHES "_wmma")
list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx908 gfx90a gfx940 gfx941 gfx942 gfx1030) list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx908:xnack+ gfx908:xnack- gfx90a:xnack+ gfx90a:xnack- gfx908 gfx90a gfx940 gfx941 gfx942 gfx1030)
elseif(ARGN MATCHES "_smfmac") elseif(ARGN MATCHES "_smfmac")
list(REMOVE_ITEM TEST_TARGETS gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx908 gfx90a gfx1200 gfx1201) list(REMOVE_ITEM TEST_TARGETS gfx900 gfx906 gfx906:xnack- gfx1030 gfx1100 gfx1101 gfx1102 gfx1103 gfx908 gfx90a gfx1200 gfx1201)
endif() endif()
set_source_files_properties(${ARGN} PROPERTIES LANGUAGE HIP) set_source_files_properties(${ARGN} PROPERTIES LANGUAGE HIP)
add_executable(${TEST_NAME} ${ARGN}) add_executable(${TEST_NAME} ${ARGN})
......
test/gemm_universal/test_gemm_universal_xdl.cpp
View file @ 667cd6ab
...@@ -56,7 +56,7 @@ class TestGemmUniversal_KM_NK ...@@ -56,7 +56,7 @@ class TestGemmUniversal_KM_NK
using KernelTypes_MK_KN = ::testing::Types< using KernelTypes_MK_KN = ::testing::Types<
// ADataType, BDataType, ComputeDataType, CDataType // ADataType, BDataType, ComputeDataType, CDataType
std::tuple< F16, F16, F16, F16>, std::tuple< F16, F16, F16, F16>,
#if (defined CK_ENABLE_FP8) #if defined(CK_ENABLE_FP8) && defined(CK_USE_FP8_ON_UNSUPPORTED_ARCH)
std::tuple< F16, F8, F16, F16>, std::tuple< F16, F8, F16, F16>,
std::tuple< F8, F16, F16, F16>, std::tuple< F8, F16, F16, F16>,
std::tuple< F8, F8, F8, BF16>, std::tuple< F8, F8, F8, BF16>,
...@@ -66,7 +66,7 @@ using KernelTypes_MK_KN = ::testing::Types< ...@@ -66,7 +66,7 @@ using KernelTypes_MK_KN = ::testing::Types<
using KernelTypes_MK_NK = ::testing::Types< using KernelTypes_MK_NK = ::testing::Types<
// ADataType, BDataType, ComputeDataType, CDataType // ADataType, BDataType, ComputeDataType, CDataType
std::tuple< F16, F16, F16, F16>, std::tuple< F16, F16, F16, F16>,
#if (defined CK_ENABLE_FP8) #if defined(CK_ENABLE_FP8) && defined(CK_USE_FP8_ON_UNSUPPORTED_ARCH)
std::tuple< F16, F8, F16, F16>, std::tuple< F16, F8, F16, F16>,
std::tuple< F8, F16, F16, F16>, std::tuple< F8, F16, F16, F16>,
std::tuple< F8, F8, F8, BF16>, std::tuple< F8, F8, F8, BF16>,
......
test/grouped_convnd_fwd/test_grouped_convnd_fwd.cpp
View file @ 667cd6ab
...@@ -58,13 +58,13 @@ using KernelTypes1d = ::testing::Types<std::tuple<float, GNWC, GKXC, GNWK>, ...@@ -58,13 +58,13 @@ using KernelTypes1d = ::testing::Types<std::tuple<float, GNWC, GKXC, GNWK>,
using KernelTypes2d = ::testing::Types<std::tuple<float, GNHWC, GKYXC, GNHWK>, using KernelTypes2d = ::testing::Types<std::tuple<float, GNHWC, GKYXC, GNHWK>,
std::tuple<ck::half_t, GNHWC, GKYXC, GNHWK>, std::tuple<ck::half_t, GNHWC, GKYXC, GNHWK>,
std::tuple<ck::bhalf_t, GNHWC, GKYXC, GNHWK>, std::tuple<ck::bhalf_t, GNHWC, GKYXC, GNHWK>,
std::tuple<int8_t, GNHWC, GKYXC, GNHWK>,
std::tuple<float, NHWGC, GKYXC, NHWGK>, std::tuple<float, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::half_t, NHWGC, GKYXC, NHWGK>, std::tuple<ck::half_t, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::bhalf_t, NHWGC, GKYXC, NHWGK>, std::tuple<ck::bhalf_t, NHWGC, GKYXC, NHWGK>,
std::tuple<int8_t, NHWGC, GKYXC, NHWGK>, std::tuple<int8_t, NHWGC, GKYXC, NHWGK>,
std::tuple<float, NGCHW, GKYXC, NGKHW>, std::tuple<float, NGCHW, GKYXC, NGKHW>,
std::tuple<ck::half_t, NGCHW, GKYXC, NGKHW>>; std::tuple<ck::half_t, NGCHW, GKYXC, NGKHW>,
std::tuple<int8_t, NGCHW, GKYXC, NGKHW>>;
using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWC, GKZYXC, GNDHWK>, using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<ck::half_t, GNDHWC, GKZYXC, GNDHWK>, std::tuple<ck::half_t, GNDHWC, GKZYXC, GNDHWK>,
......
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