Skip to content

GitLab

Commit 5a9c4962 authored by Adam Osewski's avatar Adam Osewski
Browse files

Merge remote-tracking branch 'origin/develop' into aosewski/ggemm_multi_d2

parents 3970cf73 43879b89
Hide whitespace changes
Inline Side-by-side
Showing with 197 additions and 231 deletions
example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp8.cpp → example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp16_fp8.cpp
View file @ 5a9c4962
...@@ -35,7 +35,7 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough; ...@@ -35,7 +35,7 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using ADataType = F16; using ADataType = F16;
using BDataType = F8; using BDataType = F8;
using AccDataType = F32; using AccDataType = F32;
using CShuffleDataType = F32; using CShuffleDataType = F16;
using DsDataType = ck::Tuple<>; using DsDataType = ck::Tuple<>;
using EDataType = F16; using EDataType = F16;
...@@ -56,7 +56,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemm_Xdl_F ...@@ -56,7 +56,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemm_Xdl_F
//######| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //######| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//######| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //######| | | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>; < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>;
// clang-format on // clang-format on
struct ProblemSize final struct ProblemSize final
......
example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
View file @ 5a9c4962
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) add_custom_target(example_gemm_reduce_xdl)
set(target 0) add_custom_target(example_gemm_reduce_xdl_max)
foreach(gpu IN LISTS GPU_TARGETS) add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
if(gpu IN_LIST gpu_list AND target EQUAL 0) add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
add_custom_target(example_gemm_reduce_xdl)
add_custom_target(example_gemm_reduce_xdl_max) add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
add_custom_target(example_gemm_reduce_xdl_mean_meansquare) add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp16)
add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp) add_example_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp16)
add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp) add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp16)
add_example_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp) add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int8)
add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp16)
add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp) add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_add_addsquare_xdl_int8)
add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int8)
add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp) add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp32)
add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_add_addsquare_xdl_int8)
add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp) add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp32)
add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp32)
add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp) add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_bf16)
add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp32)
add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp) add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_bf16)
add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_bf16)
add_example_dependencies(example_gemm_reduce_xdl
add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp) example_gemm_reduce_xdl_mean_meansquare
add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_bf16) example_gemm_reduce_xdl_max
example_gemm_add_add_mean_meansquare_xdl)
add_example_dependencies(example_gemm_reduce_xdl
example_gemm_reduce_xdl_mean_meansquare if(USE_BITINT_EXTENSION_INT4)
example_gemm_reduce_xdl_max add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
example_gemm_add_add_mean_meansquare_xdl) add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
endif()
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
endif()
set(target 1)
endif()
endforeach()
example/17_convnd_bwd_data/CMakeLists.txt
View file @ 5a9c4962
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
set(target 0) if(result EQUAL 0)
foreach(gpu IN LISTS GPU_TARGETS) target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
if(gpu IN_LIST gpu_list AND target EQUAL 0) endif()
add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
if(result EQUAL 0)
target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
endif()
set(target 1)
endif()
endforeach()
add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp) add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp)
if(result EQUAL 0) if(result EQUAL 0)
......
example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp
View file @ 5a9c4962
// SPDX-License-Identifier: MIT // 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 <iostream> #include <iostream>
#include <cstdlib> #include <cstdlib>
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_elementwise_dynamic_vector_dims_impl.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
...@@ -27,7 +27,12 @@ using DeviceElementwiseAddInstance = ...@@ -27,7 +27,12 @@ using DeviceElementwiseAddInstance =
ck::Tuple<CDataType>, ck::Tuple<CDataType>,
Add, Add,
2, 2,
64,
64,
64,
8, 8,
8,
ck::Sequence<1, 0>,
ck::Sequence<8, 8>, ck::Sequence<8, 8>,
ck::Sequence<8>>; ck::Sequence<8>>;
......
example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp
View file @ 5a9c4962
// SPDX-License-Identifier: MIT // 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 <iostream> #include <iostream>
#include <cstdlib> #include <cstdlib>
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_elementwise_dynamic_vector_dims_impl.hpp"
#include "ck/library/utility/algorithm.hpp" #include "ck/library/utility/algorithm.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
...@@ -27,9 +27,14 @@ using DeviceElementwiseAddInstance = ...@@ -27,9 +27,14 @@ using DeviceElementwiseAddInstance =
ck::Tuple<CDataType>, ck::Tuple<CDataType>,
Add, Add,
3, 3,
8, 64,
ck::Sequence<1, 8>, 16,
ck::Sequence<8>>; 16,
2,
2,
ck::Sequence<1, 0>,
ck::Sequence<1, 2>,
ck::Sequence<2>>;
template <typename HostTensorA, typename HostTensorB, typename HostTensorC, typename Functor> template <typename HostTensorA, typename HostTensorB, typename HostTensorC, typename Functor>
void host_broadcast3D_am_bmnk(HostTensorC& C, void host_broadcast3D_am_bmnk(HostTensorC& C,
......
example/19_binary_elementwise/elementwise_add_1d.cpp
View file @ 5a9c4962
// SPDX-License-Identifier: MIT // 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 <iostream> #include <iostream>
#include <cstdlib> #include <cstdlib>
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_elementwise_dynamic_vector_dims_impl.hpp"
#include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
...@@ -25,9 +25,14 @@ using DeviceElementwiseAddInstance = ...@@ -25,9 +25,14 @@ using DeviceElementwiseAddInstance =
ck::Tuple<CDataType>, ck::Tuple<CDataType>,
Add, Add,
1, 1,
8, 64,
ck::Sequence<8, 8>, 16,
ck::Sequence<8>>; 16,
2,
2,
ck::Sequence<1, 0>,
ck::Sequence<2, 2>,
ck::Sequence<2>>;
template <typename HostTensorA, typename HostTensorB, typename HostTensorC, typename Functor> template <typename HostTensorA, typename HostTensorB, typename HostTensorC, typename Functor>
void host_elementwise1D( void host_elementwise1D(
......
example/19_binary_elementwise/elementwise_add_4d.cpp
View file @ 5a9c4962
// SPDX-License-Identifier: MIT // 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 <iostream> #include <iostream>
#include <cstdlib> #include <cstdlib>
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_elementwise_dynamic_vector_dims_impl.hpp"
#include "ck/library/utility/algorithm.hpp" #include "ck/library/utility/algorithm.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
...@@ -27,9 +27,14 @@ using DeviceElementwiseAddInstance = ...@@ -27,9 +27,14 @@ using DeviceElementwiseAddInstance =
ck::Tuple<CDataType>, ck::Tuple<CDataType>,
Add, Add,
4, 4,
8, 64,
ck::Sequence<8, 8>, 2,
ck::Sequence<8>>; 128,
2,
2,
ck::Sequence<1, 0>,
ck::Sequence<2, 2>,
ck::Sequence<2>>;
template <typename HostTensorA, typename HostTensorB, typename HostTensorC, typename Functor> template <typename HostTensorA, typename HostTensorB, typename HostTensorC, typename Functor>
void host_elementwise4D(HostTensorC& C, void host_elementwise4D(HostTensorC& C,
......
example/20_grouped_conv_bwd_weight/CMakeLists.txt
View file @ 5a9c4962
list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942) add_custom_target(example_grouped_conv_bwd_weight)
list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102) add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
set(target 0) add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0)
add_custom_target(example_grouped_conv_bwd_weight)
add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16)
add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp) add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16) add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16)
add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8 grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp) add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8 grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp)
add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8) add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8)
set(target 1)
endif()
if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0) add_example_executable(example_grouped_conv_bwd_weight_wmma_fp16 grouped_conv_bwd_weight_wmma_fp16.cpp)
add_custom_target(example_grouped_conv_bwd_weight) add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_wmma_fp16)
add_example_executable(example_grouped_conv_bwd_weight_wmma_fp16 grouped_conv_bwd_weight_wmma_fp16.cpp)
add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_wmma_fp16)
set(target 1)
endif()
endforeach()
add_custom_target(example_grouped_conv_bwd_weight_dl)
add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp) add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp)
add_example_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16) add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_dl_fp16)
example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp
View file @ 5a9c4962
// SPDX-License-Identifier: MIT // 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 "common.hpp" #include "common.hpp"
...@@ -78,6 +78,9 @@ using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWe ...@@ -78,6 +78,9 @@ using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWe
InElementOp, InElementOp,
WeiElementOp, WeiElementOp,
OutElementOp, OutElementOp,
0,
0,
0,
ComputeTypeA, ComputeTypeA,
ComputeTypeB>; ComputeTypeB>;
......
example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
View file @ 5a9c4962
...@@ -119,7 +119,10 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config, ...@@ -119,7 +119,10 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
conv_param.input_right_pads_, conv_param.input_right_pads_,
InElementOp{}, InElementOp{},
WeiElementOp{}, WeiElementOp{},
OutElementOp{}); OutElementOp{},
{},
{},
{});
ref_invoker.Run(ref_argument); ref_invoker.Run(ref_argument);
......
example/21_gemm_layernorm/CMakeLists.txt
View file @ 5a9c4962
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_welford_fp16 gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp)
set(target 0) add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_naive_fp16 gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp)
foreach(gpu IN LISTS GPU_TARGETS) add_example_executable(example_gemm_layernorm_xdl_naive_fp16 gemm_layernorm_xdl_naive_fp16.cpp)
if(gpu IN_LIST gpu_list AND target EQUAL 0) add_example_executable(example_gemm_xdl_layernorm_naive_single_kernel_fp16 gemm_xdl_layernorm_naive_single_kernel_fp16.cpp)
add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_welford_fp16 gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp)
add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_naive_fp16 gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp)
add_example_executable(example_gemm_layernorm_xdl_naive_fp16 gemm_layernorm_xdl_naive_fp16.cpp)
add_example_executable(example_gemm_xdl_layernorm_naive_single_kernel_fp16 gemm_xdl_layernorm_naive_single_kernel_fp16.cpp)
set(target 1)
endif()
endforeach()
example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp
View file @ 5a9c4962
// SPDX-License-Identifier: MIT // 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 <iostream> #include <iostream>
#include <numeric> #include <numeric>
...@@ -9,7 +9,7 @@ ...@@ -9,7 +9,7 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_elementwise_dynamic_vector_dims_impl.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
...@@ -103,9 +103,14 @@ using DeviceNormalizeInstance = ck::tensor_operation::device::DeviceElementwiseI ...@@ -103,9 +103,14 @@ using DeviceNormalizeInstance = ck::tensor_operation::device::DeviceElementwiseI
ck::Tuple<LayerNormOutDataType>, // y ck::Tuple<LayerNormOutDataType>, // y
NormalizeFunctor, NormalizeFunctor,
2, 2,
8, // MPerthread 64, // BlockSize
ck::Sequence<8, 1, 1, 8, 8>, // scalarPerVector: x(gemm_out), mean, meansquare, gamma, beta 16, // MPerBlock
ck::Sequence<8>>; // scalarPerVector: y(layerNorm_out) 16, // NPerBlock
2, // MPerthread
2, // NPerthread
ck::Sequence<1, 0>, // ThreadClusterArrangeOrder
ck::Sequence<2, 1, 1, 2, 2>, // scalarPerVector: x(gemm_out), mean, meansquare, gamma, beta
ck::Sequence<2>>; // scalarPerVector: y(layerNorm_out)
auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) { auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) {
return HostTensorDescriptor({len}, {stride}); return HostTensorDescriptor({len}, {stride});
......
example/21_gemm_layernorm/gemm_layernorm_xdl_naive_fp16.cpp
View file @ 5a9c4962
// SPDX-License-Identifier: MIT // 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 <iostream> #include <iostream>
#include <numeric> #include <numeric>
...@@ -9,7 +9,7 @@ ...@@ -9,7 +9,7 @@
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp" #include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_elementwise_dynamic_vector_dims_impl.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
...@@ -102,9 +102,14 @@ using DeviceNormalizeInstance = ck::tensor_operation::device::DeviceElementwiseI ...@@ -102,9 +102,14 @@ using DeviceNormalizeInstance = ck::tensor_operation::device::DeviceElementwiseI
ck::Tuple<LayerNormOutDataType>, // y ck::Tuple<LayerNormOutDataType>, // y
NormalizeFunctor, NormalizeFunctor,
2, 2,
8, // MPerthread 64, // BlockSize
ck::Sequence<8, 1, 1, 8, 8>, // scalarPerVector: x(gemm_out), mean, meansquare, gamma, beta 16, // MPerBlock
ck::Sequence<8>>; // scalarPerVector: y(layerNorm_out) 16, // NPerBlock
2, // MPerthread
2, // NPerthread
ck::Sequence<1, 0>, // ThreadClusterArrangeOrder
ck::Sequence<2, 1, 1, 2, 2>, // scalarPerVector: x(gemm_out), mean, meansquare, gamma, beta
ck::Sequence<2>>; // scalarPerVector: y(layerNorm_out)
auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) { auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) {
return HostTensorDescriptor({len}, {stride}); return HostTensorDescriptor({len}, {stride});
......
example/26_contraction/CMakeLists.txt
View file @ 5a9c4962
...@@ -4,49 +4,49 @@ add_custom_target(example_contraction_bilinear) ...@@ -4,49 +4,49 @@ add_custom_target(example_contraction_bilinear)
# FP32 # FP32
add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp) add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp)
add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32) add_example_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32)
add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp) add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp)
add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32) add_example_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32)
add_example_executable(example_contraction_bilinear_xdl_fp32_compute_bf16 contraction_bilinear_xdl_fp32_compute_bf16.cpp) add_example_executable(example_contraction_bilinear_xdl_fp32_compute_bf16 contraction_bilinear_xdl_fp32_compute_bf16.cpp)
add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32_compute_bf16) add_example_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32_compute_bf16)
add_example_executable(example_contraction_scale_xdl_fp32_compute_bf16 contraction_scale_xdl_fp32_compute_bf16.cpp) add_example_executable(example_contraction_scale_xdl_fp32_compute_bf16 contraction_scale_xdl_fp32_compute_bf16.cpp)
add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32_compute_bf16) add_example_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32_compute_bf16)
add_example_executable(example_contraction_bilinear_xdl_fp32_compute_fp16 contraction_bilinear_xdl_fp32_compute_fp16.cpp) add_example_executable(example_contraction_bilinear_xdl_fp32_compute_fp16 contraction_bilinear_xdl_fp32_compute_fp16.cpp)
add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32_compute_fp16) add_example_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32_compute_fp16)
add_example_executable(example_contraction_scale_xdl_fp32_compute_fp16 contraction_scale_xdl_fp32_compute_fp16.cpp) add_example_executable(example_contraction_scale_xdl_fp32_compute_fp16 contraction_scale_xdl_fp32_compute_fp16.cpp)
add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32_compute_fp16) add_example_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32_compute_fp16)
# FP64 # FP64
add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp) add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp)
add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp64) add_example_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp64)
add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp) add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp)
add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp64) add_example_dependencies(example_contraction_scale example_contraction_scale_xdl_fp64)
add_example_executable(example_contraction_bilinear_xdl_fp64_compute_fp32 contraction_bilinear_xdl_fp64_compute_fp32.cpp) add_example_executable(example_contraction_bilinear_xdl_fp64_compute_fp32 contraction_bilinear_xdl_fp64_compute_fp32.cpp)
add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp64_compute_fp32) add_example_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp64_compute_fp32)
add_example_executable(example_contraction_scale_xdl_fp64_compute_fp32 contraction_scale_xdl_fp64_compute_fp32.cpp) add_example_executable(example_contraction_scale_xdl_fp64_compute_fp32 contraction_scale_xdl_fp64_compute_fp32.cpp)
add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp64_compute_fp32) add_example_dependencies(example_contraction_scale example_contraction_scale_xdl_fp64_compute_fp32)
# FP16 # FP16
add_example_executable(example_contraction_bilinear_xdl_fp16_compute_fp32 contraction_bilinear_xdl_fp16_compute_fp32.cpp) add_example_executable(example_contraction_bilinear_xdl_fp16_compute_fp32 contraction_bilinear_xdl_fp16_compute_fp32.cpp)
add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp16_compute_fp32) add_example_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp16_compute_fp32)
add_example_executable(example_contraction_scale_xdl_fp16_compute_fp32 contraction_scale_xdl_fp16_compute_fp32.cpp) add_example_executable(example_contraction_scale_xdl_fp16_compute_fp32 contraction_scale_xdl_fp16_compute_fp32.cpp)
add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp16_compute_fp32) add_example_dependencies(example_contraction_scale example_contraction_scale_xdl_fp16_compute_fp32)
# BF16 # BF16
add_example_executable(example_contraction_bilinear_xdl_bf16_compute_fp32 contraction_bilinear_xdl_bf16_compute_fp32.cpp) add_example_executable(example_contraction_bilinear_xdl_bf16_compute_fp32 contraction_bilinear_xdl_bf16_compute_fp32.cpp)
add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_bf16_compute_fp32) add_example_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_bf16_compute_fp32)
add_example_executable(example_contraction_scale_xdl_bf16_compute_fp32 contraction_scale_xdl_bf16_compute_fp32.cpp) add_example_executable(example_contraction_scale_xdl_bf16_compute_fp32 contraction_scale_xdl_bf16_compute_fp32.cpp)
add_dependencies(example_contraction_scale example_contraction_scale_xdl_bf16_compute_fp32) add_example_dependencies(example_contraction_scale example_contraction_scale_xdl_bf16_compute_fp32)
add_dependencies(example_contraction example_contraction_scale) add_example_dependencies(example_contraction example_contraction_scale)
add_dependencies(example_contraction example_contraction_bilinear) add_example_dependencies(example_contraction example_contraction_bilinear)
example/29_batched_gemm_bias_e_permute/CMakeLists.txt
View file @ 5a9c4962
add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp) add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
if(GPU_TARGETS MATCHES "gfx11")
add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
endif()
example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
View file @ 5a9c4962
list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942) add_custom_target(example_grouped_conv_fwd_multiple_d)
list(APPEND gpu_list2 gfx1100 gfx1101 gfx1102) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
set(target 0) add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
foreach(gpu IN LISTS GPU_TARGETS) add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
add_custom_target(example_grouped_conv_fwd_multiple_d)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16) add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16) add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32) add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp) if(USE_BITINT_EXTENSION_INT4)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16) add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
endif() # USE_BITINT_EXTENSION_INT4
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp) add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8) add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
endif() # USE_BITINT_EXTENSION_INT4
set(target 1)
endif()
endforeach()
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
set(target 1)
endif()
endforeach()
example/31_batched_gemm_gemm/CMakeLists.txt
View file @ 5a9c4962
list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942) add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
set(target 0) add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
foreach(gpu IN LISTS GPU_TARGETS) if(USE_BITINT_EXTENSION_INT4)
if(gpu IN_LIST gpu_list1 AND target EQUAL 0) add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp) endif(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
endif(USE_BITINT_EXTENSION_INT4)
set(target 1)
endif()
endforeach()
if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1") if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp) add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
......
example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
View file @ 5a9c4962
if(GPU_TARGETS MATCHES "gfx11") add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_wmma_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_wmma_fp16.cpp)
add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_wmma_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_wmma_fp16.cpp) add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_wmma_fp16 batched_gemm_scale_softmax_gemm_permute_wmma_fp16.cpp)
add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_wmma_fp16 batched_gemm_scale_softmax_gemm_permute_wmma_fp16.cpp) add_example_executable(example_self_attention_forward_wmma_fp16 self_attention_forward_wmma_fp16.cpp)
add_example_executable(example_self_attention_forward_wmma_fp16 self_attention_forward_wmma_fp16.cpp) add_example_executable(example_cross_attention_forward_wmma_fp16 cross_attention_forward_wmma_fp16.cpp)
add_example_executable(example_cross_attention_forward_wmma_fp16 cross_attention_forward_wmma_fp16.cpp) add_example_executable(example_multi_query_attention_forward_wmma_fp16 multi_query_attention_forward_wmma_fp16.cpp)
add_example_executable(example_multi_query_attention_forward_wmma_fp16 multi_query_attention_forward_wmma_fp16.cpp) add_example_executable(example_grouped_query_attention_forward_wmma_fp16 grouped_query_attention_forward_wmma_fp16.cpp)
add_example_executable(example_grouped_query_attention_forward_wmma_fp16 grouped_query_attention_forward_wmma_fp16.cpp)
endif()
add_custom_target(example_gemm_scale_softmax_gemm) add_custom_target(example_gemm_scale_softmax_gemm)
......
example/34_batchnorm/batchnorm_infer_impl.hpp
View file @ 5a9c4962
// SPDX-License-Identifier: MIT // 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.
#pragma once #pragma once
...@@ -10,7 +10,7 @@ ...@@ -10,7 +10,7 @@
#include "ck/utility/sequence.hpp" #include "ck/utility/sequence.hpp"
#include "ck/utility/tuple.hpp" #include "ck/utility/tuple.hpp"
#include "ck/utility/reduction_operator.hpp" #include "ck/utility/reduction_operator.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp" #include "ck/tensor_operation/gpu/device/impl/device_elementwise_dynamic_vector_dims_impl.hpp"
#include "batchnorm_common.hpp" #include "batchnorm_common.hpp"
...@@ -54,7 +54,12 @@ int bnorm_infer( ...@@ -54,7 +54,12 @@ int bnorm_infer(
ck::Tuple<YDataType>, // y ck::Tuple<YDataType>, // y
NormalizeInInfer, NormalizeInInfer,
Rank, Rank,
2, // MPerthread 64, // BlockSize
32, // MPerBlock
32, // NPerBlock
4, // MPerthread
4, // NPerthread
ck::Sequence<1, 0>, // ThreadClusterArrangeOrder
ck::Sequence<1, 1, 1, 1, 1>, // x, mean, variance, scale, bias ck::Sequence<1, 1, 1, 1, 1>, // x, mean, variance, scale, bias
ck::Sequence<1>>; // scalarPerVector: y ck::Sequence<1>>; // scalarPerVector: y
......
example/35_splitK_gemm/CMakeLists.txt
View file @ 5a9c4962
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942) add_custom_target(example_splitK_gemm_xdl)
set(target 0) add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
foreach(gpu IN LISTS GPU_TARGETS) add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_splitK_gemm_xdl)
add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp) add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32) add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16)
add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp) add_example_executable(example_splitK_gemm_xdl_fp16_fp8 splitK_gemm_xdl_fp16_fp8.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16) add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16_fp8)
add_example_executable(example_splitK_gemm_xdl_fp16_fp8 splitK_gemm_xdl_fp16_fp8.cpp) add_example_executable(example_splitK_gemm_xdl_lds_direct_load_fp16 splitK_gemm_xdl_lds_direct_load_fp16.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16_fp8) add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_lds_direct_load_fp16)
add_example_executable(example_splitK_gemm_xdl_lds_direct_load_fp16 splitK_gemm_xdl_lds_direct_load_fp16.cpp) add_example_executable(example_splitK_gemm_xdl_bf16 splitK_gemm_xdl_bf16.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_lds_direct_load_fp16) add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bf16)
add_example_executable(example_splitK_gemm_xdl_bf16 splitK_gemm_xdl_bf16.cpp) add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bf16) add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8)
add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp) if(USE_BITINT_EXTENSION_INT4)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8) add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
if(USE_BITINT_EXTENSION_INT4) endif()
add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
endif()
set(target 1)
endif()
endforeach()
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