Skip to content

GitLab

Commit 0c823497 authored by muozturk's avatar muozturk
Browse files

merge

parents 334cfe1c 68f2b5e7
Hide whitespace changes
Inline Side-by-side
Showing with 365 additions and 284 deletions
docs/sphinx/requirements.in
View file @ 0c823497
rocm-docs-core>=0.20.0
sphinxcontrib-bibtex==2.5.0
sphinxcontrib-bibtex==2.6.1
docs/sphinx/requirements.txt
View file @ 0c823497
......@@ -42,7 +42,7 @@ fastjsonschema==2.18.0
# via rocm-docs-core
gitdb==4.0.10
# via gitpython
gitpython==3.1.31
gitpython==3.1.35
# via rocm-docs-core
idna==3.4
# via requests
......@@ -103,7 +103,7 @@ requests==2.28.2
# via
# pygithub
# sphinx
rocm-docs-core>=0.20.0
rocm-docs-core==0.26.0
# via -r requirements.in
six==1.16.0
# via
......@@ -139,7 +139,7 @@ sphinx-notfound-page==0.8.3
# via rocm-docs-core
sphinxcontrib-applehelp==1.0.4
# via sphinx
sphinxcontrib-bibtex==2.5.0
sphinxcontrib-bibtex==2.6.1
# via -r requirements.in
sphinxcontrib-devhelp==1.0.2
# via sphinx
......
example/01_gemm/CMakeLists.txt
View file @ 0c823497
add_custom_target(example_gemm_dl)
add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_dl example_gemm_dl_fp32)
endif()
add_example_dependencies(example_gemm_dl example_gemm_dl_fp32)
add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_dl example_gemm_dl_fp16)
endif()
add_example_dependencies(example_gemm_dl example_gemm_dl_fp16)
add_example_executable(example_gemm_dpp_fp16 gemm_dpp_fp16.cpp)
add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_dl example_gemm_dl_int8)
endif()
add_example_dependencies(example_gemm_dl example_gemm_dl_int8)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp)
add_dependencies(example_gemm_dl example_gemm_dl_int4)
add_example_dependencies(example_gemm_dl example_gemm_dl_int4)
endif(USE_BITINT_EXTENSION_INT4)
add_custom_target(example_gemm_xdl)
add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
endif()
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
endif()
add_example_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
endif()
add_example_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
add_custom_target(example_gemm_wmma)
add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
endif()
add_example_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
endif()
add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
add_example_executable(example_gemm_xdl_bf16_rtn gemm_xdl_bf16_rtn.cpp)
add_dependencies(example_gemm_xdl example_gemm_xdl_bf16_rtn)
endif()
add_example_executable(example_gemm_xdl_bf16_rtn gemm_xdl_bf16_rtn.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16_rtn)
add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_int8)
endif()
add_example_dependencies(example_gemm_xdl example_gemm_xdl_int8)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_xdl_int4 gemm_xdl_int4.cpp)
add_dependencies(example_gemm_xdl example_gemm_xdl_int4)
add_example_executable(example_gemm_xdl_int4 gemm_xdl_int4.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_int4)
endif(USE_BITINT_EXTENSION_INT4)
# FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
add_example_executable_no_testing(example_gemm_xdl_fp64 gemm_xdl_fp64.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
endif()
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
add_example_executable(example_gemm_xdl_fp8 gemm_xdl_fp8.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
add_example_executable(example_gemm_xdl_fp8 gemm_xdl_fp8.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
endif()
endif()
if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
add_example_executable(example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_fp8_bf8)
endif()
endif()
add_example_executable(example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_bf8)
add_example_executable(example_gemm_xdl_fp16_fp8 gemm_xdl_fp16_fp8.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8)
endif()
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8)
example/01_gemm/gemm_xdl_fp8.cpp
View file @ 0c823497
......@@ -7,9 +7,9 @@
using ADataType = ck::f8_t;
using BDataType = ck::f8_t;
using CDataType = ck::f8_t;
using CDataType = ck::half_t;
using AccDataType = float;
using CShuffleDataType = ck::f8_t;
using CShuffleDataType = float;
using ALayout = Row;
using BLayout = Col;
......@@ -27,7 +27,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
// ######| | | | Type| Type| Type| Type| DataType| 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|
// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 4>, 16>;
< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 4>, 8>;
// clang-format on
using ReferenceGemmInstance = ck::tensor_operation::host::
......
example/01_gemm/gemm_xdl_fp8_bf8.cpp
View file @ 0c823497
......@@ -7,9 +7,9 @@
using ADataType = ck::f8_t;
using BDataType = ck::bf8_t;
using CDataType = ck::f8_t;
using CDataType = ck::half_t;
using AccDataType = float;
using CShuffleDataType = ck::f8_t;
using CShuffleDataType = float;
using ALayout = Row;
using BLayout = Col;
......@@ -31,7 +31,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
// ######| | | | Type| Type| Type| Type| DataType| 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|
// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 4>, 16, LoopSched, PipelineVer, ComputeTypeA, ComputeTypeB>;
< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 128, 64, 16, 16, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 4>, 8, LoopSched, PipelineVer, ComputeTypeA, ComputeTypeB>;
// clang-format on
using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
......
example/04_gemm_add_add_fastgelu/CMakeLists.txt
View file @ 0c823497
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_gemm_add_add_fastgelu_xdl)
add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_gemm_add_add_fastgelu_xdl)
add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
endif(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
set(target 1)
endif()
add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
endif()
add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
endif()
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
endif(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
endif()
set(target 1)
endif()
endforeach()
\ No newline at end of file
endforeach()
example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
View file @ 0c823497
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_convnd_fwd_reduce_xdl)
add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
endif()
add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
if(result EQUAL 0)
add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
endif()
add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
endif()
add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
if(result EQUAL 0)
add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
endif()
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
endif(USE_BITINT_EXTENSION_INT4)
set(target 1)
endif()
endforeach()
\ No newline at end of file
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_convnd_fwd_reduce_xdl)
add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
endif(USE_BITINT_EXTENSION_INT4)
set(target 1)
endif()
endforeach()
example/12_reduce/README.md
View file @ 0c823497
......@@ -2,7 +2,7 @@
## Run ```example_reduce_blockwise```
```bash
# -D <xxx> : input 3d/4d/5d tensor lengths
# -D <xxx> : input 3D/4D/5D tensor lengths
# -R <xxx> : reduce dimension ids
# -v <x> : verification (0=no, 1=yes)
#arg1: data type (0: fp16, 1: fp32, 3: int8, 5: bp16, 6: fp64, 7: int4)
......@@ -22,7 +22,7 @@ Perf: 0.238063 ms, 264.285 GB/s, DeviceReduceBlockWise<256,M_C4_S1,K_C64_S1,InSr
## Run ```example_reduce_multiblock_atomic_add```
```bash
# -D <xxx> : input 3d/4d/5d tensor lengths
# -D <xxx> : input 3D/4D/5D tensor lengths
# -R <xxx> : reduce dimension ids
# -v <x> : verification (0=no, 1=yes)
#arg1: data type (0: fp32, 1: fp64)
......
example/15_grouped_gemm/CMakeLists.txt
View file @ 0c823497
add_custom_target(example_grouped_gemm_xdl)
add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32)
endif()
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32)
add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp16)
endif()
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp16)
add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_multiple_d_dl_fp16)
endif()
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_multiple_d_dl_fp16)
add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_splitk_fp16)
endif()
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_splitk_fp16)
add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp16 grouped_gemm_xdl_fixed_nk_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp16)
endif()
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp16)
add_example_executable(example_grouped_gemm_xdl_fixed_nk_bias_fp16 grouped_gemm_xdl_fixed_nk_bias_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_bias_fp16)
endif()
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_bias_fp16)
add_example_executable(example_grouped_gemm_xdl_bf16 grouped_gemm_xdl_bf16.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bf16)
endif()
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bf16)
add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int8)
endif()
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int8)
add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp8 grouped_gemm_xdl_fixed_nk_fp8.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp8)
endif()
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp8)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
endif()
add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
endif()
example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
View file @ 0c823497
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_gemm_reduce_xdl)
add_custom_target(example_gemm_reduce_xdl_max)
add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp16)
endif()
add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
endif()
add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp16)
endif()
add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int8)
endif()
add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_add_addsquare_xdl_int8)
endif()
add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp32)
endif()
add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp32)
endif()
add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_bf16)
endif()
add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_bf16)
endif()
add_dependencies(example_gemm_reduce_xdl
example_gemm_reduce_xdl_mean_meansquare
example_gemm_reduce_xdl_max
example_gemm_add_add_mean_meansquare_xdl)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
if(result EQUAL 0)
add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
endif()
endif()
set(target 1)
endif()
if(gpu IN_LIST gpu_list AND target EQUAL 0)
add_custom_target(example_gemm_reduce_xdl)
add_custom_target(example_gemm_reduce_xdl_max)
add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp16)
add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
add_example_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp16)
add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int8)
add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_add_addsquare_xdl_int8)
add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp32)
add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp32)
add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_bf16)
add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_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
example_gemm_reduce_xdl_mean_meansquare
example_gemm_reduce_xdl_max
example_gemm_add_add_mean_meansquare_xdl)
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/20_grouped_conv_bwd_weight/CMakeLists.txt
View file @ 0c823497
list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942)
list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102)
set(target 0)
foreach(gpu IN LISTS GPU_TARGETS)
if(gpu IN_LIST gpu_list 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)
if(result EQUAL 0)
add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16)
endif()
add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16)
endif()
if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8 grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp)
if(result EQUAL 0)
add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8)
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_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_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_custom_target(example_grouped_conv_bwd_weight)
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()
endif()
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)
if(result EQUAL 0)
add_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16)
endif()
add_example_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16)
example/20_grouped_conv_bwd_weight/common.hpp
View file @ 0c823497
......@@ -46,25 +46,21 @@ struct CommonLayoutSetting
using OutputLayout = OutputLay;
};
template <ck::index_t NDimSpatial>
struct CommonLayoutSettingSelector;
namespace ctl = ck::tensor_layout::convolution;
template <>
struct CommonLayoutSettingSelector<1> final : CommonLayoutSetting<ctl::GNWC, ctl::GKXC, ctl::GNWK>
{
};
template <>
struct CommonLayoutSettingSelector<2> final
: CommonLayoutSetting<ctl::GNHWC, ctl::GKYXC, ctl::GNHWK>
{
};
template <>
struct CommonLayoutSettingSelector<3> final
: CommonLayoutSetting<ctl::GNDHWC, ctl::GKZYXC, ctl::GNDHWK>
template <ck::index_t NDimSpatial>
struct CommonLayoutSettingSelector
: CommonLayoutSetting<ck::tuple_element_t<NDimSpatial - 1,
ck::Tuple<ck::tensor_layout::convolution::GNWC,
ck::tensor_layout::convolution::GNHWC,
ck::tensor_layout::convolution::GNDHWC>>,
ck::tuple_element_t<NDimSpatial - 1,
ck::Tuple<ck::tensor_layout::convolution::GKXC,
ck::tensor_layout::convolution::GKYXC,
ck::tensor_layout::convolution::GKZYXC>>,
ck::tuple_element_t<NDimSpatial - 1,
ck::Tuple<ck::tensor_layout::convolution::GNWK,
ck::tensor_layout::convolution::GNHWK,
ck::tensor_layout::convolution::GNDHWK>>>
{
};
......@@ -84,10 +80,10 @@ struct ExecutionConfig final
bool time_kernel = false;
};
#define DefaultConvParam \
ck::utils::conv::ConvParam \
{ \
2, 4, 1, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, { 1, 1 } \
#define DefaultConvParam \
ck::utils::conv::ConvParam \
{ \
3, 4, 1, 128, 256, {3, 3, 3}, {14, 14, 14}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, { 1, 1, 1 } \
}
inline void print_help_msg()
......
example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
View file @ 0c823497
......@@ -76,4 +76,23 @@ using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWe
#include "run_grouped_conv_bwd_weight_example.inc"
int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
int main(int argc, char* argv[])
{
ExecutionConfig config;
ck::utils::conv::ConvParam conv_param = DefaultConvParam;
if(!parse_cmd_args(argc, argv, config, conv_param))
{
return 1;
}
switch(conv_param.num_dim_spatial_)
{
case 1: return !run_grouped_conv_bwd_weight<1>(config, conv_param);
case 2: return !run_grouped_conv_bwd_weight<2>(config, conv_param);
case 3: return !run_grouped_conv_bwd_weight<3>(config, conv_param);
default: break;
}
return 1;
}
example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_wmma_fp16.cpp 0 → 100644
View file @ 0c823497
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_wmma_cshuffle.hpp"
using InDataType = F16;
using WeiDataType = F16;
using OutDataType = F16;
using AccDataType = F32;
using InElementOp = PassThrough;
using WeiElementOp = PassThrough;
using OutElementOp = PassThrough;
template <ck::index_t NDimSpatial>
using DeviceConvBwdWeightInstance =
ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Wmma_CShuffle<
NDimSpatial,
ck::tensor_layout::convolution::GNDHWC,
ck::tensor_layout::convolution::GKZYXC,
ck::tensor_layout::convolution::GNDHWK,
InDataType, // InDataType
WeiDataType, // WeiDataType
OutDataType, // OutDataType
AccDataType, // AccDataType
InElementOp, // InElementwiseOperation
WeiElementOp, // WeiElementwiseOperation
OutElementOp, // OutElementwiseOperation
ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization
256, // BlockSize
128, // MPerBlock
128, // NPerBlock
4, // K0PerBlock
8, // K1
16, // MPerWMMA
16, // NPerWMMA
4, // MRepeat
2, // NRepeat
S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
S<0, 2, 1>, // ABlockTransferThreadClusterArrangeOrder
S<0, 2, 1>, // ABlockTransferSrcAccessOrder
1, // ABlockTransferSrcVectorDim
1, // ABlockTransferSrcScalarPerVector
8, // ABlockTransferDstScalarPerVector_AK1
true, // ABlockLdsExtraM
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
S<0, 2, 1>, // BBlockTransferThreadClusterArrangeOrder
S<0, 2, 1>, // BBlockTransferSrcAccessOrder
1, // BBlockTransferSrcVectorDim
1, // BBlockTransferSrcScalarPerVector
8, // BBlockTransferDstScalarPerVector_BK1
true, // BBlockLdsExtraN
4,
2,
S<1, 32, 1, 8>,
1>;
template <ck::index_t NDimSpatial>
using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp>;
#include "run_grouped_conv_bwd_weight_example.inc"
int main(int argc, char* argv[])
{
ExecutionConfig config;
ck::utils::conv::ConvParam conv_param = DefaultConvParam;
if(!parse_cmd_args(argc, argv, config, conv_param))
{
return 1;
}
switch(conv_param.num_dim_spatial_)
{
case 3: return !run_grouped_conv_bwd_weight<3>(config, conv_param);
default: break;
}
return 1;
}
example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
View file @ 0c823497
......@@ -78,4 +78,23 @@ using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWe
#include "run_grouped_conv_bwd_weight_example.inc"
int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
int main(int argc, char* argv[])
{
ExecutionConfig config;
ck::utils::conv::ConvParam conv_param = DefaultConvParam;
if(!parse_cmd_args(argc, argv, config, conv_param))
{
return 1;
}
switch(conv_param.num_dim_spatial_)
{
case 1: return !run_grouped_conv_bwd_weight<1>(config, conv_param);
case 2: return !run_grouped_conv_bwd_weight<2>(config, conv_param);
case 3: return !run_grouped_conv_bwd_weight<3>(config, conv_param);
default: break;
}
return 1;
}
example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
View file @ 0c823497
......@@ -77,4 +77,23 @@ using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWe
#include "run_grouped_conv_bwd_weight_example.inc"
int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
int main(int argc, char* argv[])
{
ExecutionConfig config;
ck::utils::conv::ConvParam conv_param = DefaultConvParam;
if(!parse_cmd_args(argc, argv, config, conv_param))
{
return 1;
}
switch(conv_param.num_dim_spatial_)
{
case 1: return !run_grouped_conv_bwd_weight<1>(config, conv_param);
case 2: return !run_grouped_conv_bwd_weight<2>(config, conv_param);
case 3: return !run_grouped_conv_bwd_weight<3>(config, conv_param);
default: break;
}
return 1;
}
example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp
View file @ 0c823497
......@@ -83,4 +83,23 @@ using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWe
#include "run_grouped_conv_bwd_weight_example.inc"
int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
int main(int argc, char* argv[])
{
ExecutionConfig config;
ck::utils::conv::ConvParam conv_param = DefaultConvParam;
if(!parse_cmd_args(argc, argv, config, conv_param))
{
return 1;
}
switch(conv_param.num_dim_spatial_)
{
case 1: return !run_grouped_conv_bwd_weight<1>(config, conv_param);
case 2: return !run_grouped_conv_bwd_weight<2>(config, conv_param);
case 3: return !run_grouped_conv_bwd_weight<3>(config, conv_param);
default: break;
}
return 1;
}
example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
View file @ 0c823497
......@@ -5,7 +5,7 @@ template <ck::index_t NDimSpatial>
bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
const ck::utils::conv::ConvParam& conv_param)
{
// Dl op doesn't support split_k > 1
// Dl and WMMA ops don't support split_k > 1
constexpr ck::index_t split_k = 1;
const auto in_g_n_c_wis_desc =
......@@ -143,23 +143,3 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
return true;
}
bool run_grouped_conv_bwd_weight_example(int argc, char* argv[])
{
ExecutionConfig config;
ck::utils::conv::ConvParam conv_param = DefaultConvParam;
if(!parse_cmd_args(argc, argv, config, conv_param))
{
return false;
}
switch(conv_param.num_dim_spatial_)
{
case 1: return run_grouped_conv_bwd_weight<1>(config, conv_param);
case 2: return run_grouped_conv_bwd_weight<2>(config, conv_param);
case 3: return run_grouped_conv_bwd_weight<3>(config, conv_param);
}
return false;
}
example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp
View file @ 0c823497
......@@ -114,12 +114,15 @@ void host_gemm_layernorm(Tensor<HDataType>& h_m_n,
BetaDataType,
HDataType,
AccDataType,
AccDataType,
HElementOp,
2,
1>;
Tensor<EMeanVarDataType> e_m_n(HostTensorDescriptor{M, N});
Tensor<AccDataType> c_m_n(HostTensorDescriptor{M, N});
Tensor<AccDataType> save_mean({M});
Tensor<AccDataType> save_inv_std({M});
auto ref_gemm = ReferenceGemm{};
auto ref_gemm_invoker = ref_gemm.MakeInvoker();
......@@ -145,7 +148,7 @@ void host_gemm_layernorm(Tensor<HDataType>& h_m_n,
auto ref_layernorm_invoker = ref_layernorm.MakeInvoker();
auto ref_layernorm_argument = ref_layernorm.MakeArgument(
e_m_n, gamma_n, beta_n, h_m_n, h_element_op, {M, N}, {1}, epsilon);
e_m_n, gamma_n, beta_n, h_m_n, save_mean, save_inv_std, h_element_op, {M, N}, {1}, epsilon);
ref_layernorm_invoker.Run(ref_layernorm_argument);
}
......
example/22_cgemm/CMakeLists.txt
View file @ 0c823497
add_custom_target(example_cgemm_xdl)
add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp)
if(result EQUAL 0)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_bf16)
endif()
add_example_dependencies(example_cgemm_xdl example_cgemm_xdl_bf16)
add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp)
if(result EQUAL 0)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp16)
endif()
add_example_dependencies(example_cgemm_xdl example_cgemm_xdl_fp16)
add_example_executable(example_cgemm_xdl_fp32 cgemm_xdl_fp32.cpp)
if(result EQUAL 0)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp32)
endif()
add_example_dependencies(example_cgemm_xdl example_cgemm_xdl_fp32)
add_example_executable(example_cgemm_xdl_int8 cgemm_xdl_int8.cpp)
if(result EQUAL 0)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_int8)
endif()
add_example_dependencies(example_cgemm_xdl example_cgemm_xdl_int8)
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_cgemm_xdl_int4 cgemm_xdl_int4.cpp)
add_dependencies(example_cgemm_xdl example_cgemm_xdl_int4)
add_example_executable(example_cgemm_xdl_int4 cgemm_xdl_int4.cpp)
add_example_dependencies(example_cgemm_xdl example_cgemm_xdl_int4)
endif()
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