Skip to content

GitLab

Commit e305e41e authored by aska-0096's avatar aska-0096
Browse files

Merge branch 'develop' of... 

Merge branch 'develop' of https://github.com/ROCmSoftwarePlatform/composable_kernel into e2e_kernellib
parents bee4e344 a35456a3
Hide whitespace changes
Inline Side-by-side
Showing with 36 additions and 34 deletions
example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm.inc
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
int run(int argc, char* argv[]) int run(int argc, char* argv[])
{ {
......
example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
int run(int argc, char* argv[]) int run(int argc, char* argv[])
{ {
......
example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
int run(int argc, char* argv[]) int run(int argc, char* argv[])
{ {
......
example/33_multiple_reduce/dual_reduce_common.hpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream> #include <iostream>
#include <cstdlib> #include <cstdlib>
......
example/33_multiple_reduce/dual_reduce_multiblock.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream> #include <iostream>
#include <cstdlib> #include <cstdlib>
......
example/33_multiple_reduce/dual_reduce_threadwise.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream> #include <iostream>
#include <cstdlib> #include <cstdlib>
......
example/34_batchnorm/batchnorm_backward_nhwc.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <limits> #include <limits>
#include <iostream> #include <iostream>
......
example/34_batchnorm/batchnorm_common.hpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once #pragma once
......
example/34_batchnorm/batchnorm_forward_inferring_nhwc.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <limits> #include <limits>
#include <iostream> #include <iostream>
......
example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <limits> #include <limits>
#include <iostream> #include <iostream>
......
example/34_batchnorm/batchnorm_infer_impl.hpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once #pragma once
......
example/35_splitK_gemm/CMakeLists.txt
View file @ e305e41e
add_custom_target(example_splitK_gemm_xdl) if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
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_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp) add_dependencies(example_splitK_gemm_xdl
add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
add_dependencies(example_splitK_gemm_xdl
example_splitK_gemm_xdl_fp32 example_splitK_gemm_xdl_fp32
example_splitK_gemm_xdl_fp16 example_splitK_gemm_xdl_fp16
example_splitK_gemm_xdl_bfp16 example_splitK_gemm_xdl_bfp16
example_splitK_gemm_xdl_int8) example_splitK_gemm_xdl_int8)
if(USE_BITINT_EXTENSION_INT4) if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp) add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4) add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
endif()
endif() endif()
example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream> #include <iostream>
#include <numeric> #include <numeric>
......
example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream> #include <iostream>
#include <numeric> #include <numeric>
......
example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream> #include <iostream>
#include <numeric> #include <numeric>
......
example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream> #include <iostream>
#include <numeric> #include <numeric>
......
example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream> #include <iostream>
#include <numeric> #include <numeric>
......
example/36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream> #include <iostream>
#include <numeric> #include <numeric>
......
example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
View file @ e305e41e
// SPDX-License-Identifier: MIT // SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
/* /*
Computes C_m_o = Relu(A0[m, k] * B0[n, k] + D00[m, n] + D01[mn]) * B1[n, o] + D1[m, o] Computes C_m_o = Relu(A0[m, k] * B0[n, k] + D00[m, n] + D01[mn]) * B1[n, o] + D1[m, o]
......
example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
View file @ e305e41e
add_custom_target(example_grouped_conv_bwd_data) if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
add_custom_target(example_grouped_conv_bwd_data)
add_example_executable(example_grouped_conv_bwd_data_fp16 grouped_conv_bwd_data_fp16.cpp)
add_example_executable(example_grouped_conv_bwd_data_bias_relu_fp16 grouped_conv_bwd_data_bias_relu_fp16.cpp)
add_example_executable(example_grouped_conv_bwd_data_fp16 grouped_conv_bwd_data_fp16.cpp) add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_fp16)
add_example_executable(example_grouped_conv_bwd_data_bias_relu_fp16 grouped_conv_bwd_data_bias_relu_fp16.cpp) add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
endif()
add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_fp16) \ No newline at end of file
add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
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