forked from InternLM/lmdeploy
-
Notifications
You must be signed in to change notification settings - Fork 0
/
CMakeLists.txt
executable file
·415 lines (356 loc) · 14.3 KB
/
CMakeLists.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
# Copyright (c) 2019-2023, NVIDIA CORPORATION. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
cmake_minimum_required(VERSION 3.11 FATAL_ERROR) # for PyTorch extensions, version should be greater than 3.13
project(TurboMind LANGUAGES CXX CUDA)
find_package(CUDA 10.2 REQUIRED)
if(${CUDA_VERSION_MAJOR} VERSION_GREATER_EQUAL "11")
add_definitions("-DENABLE_BF16")
message("CUDA_VERSION ${CUDA_VERSION_MAJOR}.${CUDA_VERSION_MINOR} is greater or equal than 11.0, enable -DENABLE_BF16 flag")
endif()
# if((${CUDA_VERSION_MAJOR} VERSION_GREATER_EQUAL "11" AND ${CUDA_VERSION_MINOR} VERSION_GREATER_EQUAL "8") OR (${CUDA_VERSION_MAJOR} VERSION_GREATER_EQUAL "12"))
# add_definitions("-DENABLE_FP8")
# option(ENABLE_FP8 "ENABLE_FP8" OFF)
# if(ENABLE_FP8)
# message("CUDA_VERSION ${CUDA_VERSION_MAJOR}.${CUDA_VERSION_MINOR} is greater or equal than 11.8, enable -DENABLE_FP8 flag")
# endif()
# endif()
set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake/Modules)
option(BUILD_PYT "Build in PyTorch TorchScript class mode" OFF)
if(NOT BUILD_MULTI_GPU)
option(BUILD_MULTI_GPU "Build project about multi-GPU" OFF)
endif()
if(NOT USE_TRITONSERVER_DATATYPE)
option(USE_TRITONSERVER_DATATYPE "Build triton backend for triton server" OFF)
endif()
option(BUILD_PY_FFI "Build python ffi" ON)
option(BUILD_TEST "Build tests" OFF)
include(FetchContent)
FetchContent_Declare(
repo-cutlass
GIT_REPOSITORY https://github.com/NVIDIA/cutlass.git
GIT_TAG 6f47420213f757831fae65c686aa471749fa8d60
GIT_SHALLOW ON
)
set(CUTLASS_ENABLE_HEADERS_ONLY ON CACHE BOOL "Enable only the header library")
FetchContent_MakeAvailable(repo-cutlass)
set(CUTLASS_HEADER_DIR ${PROJECT_SOURCE_DIR}/3rdparty/cutlass/include)
set(CUTLASS_EXTENSIONS_DIR ${PROJECT_SOURCE_DIR}/src/turbomind/cutlass_extensions/include)
option(SPARSITY_SUPPORT "Build project with Ampere sparsity feature support" OFF)
option(BUILD_FAST_MATH "Build in fast math mode" ON)
# the environment variable
# ASAN_OPTIONS=protect_shadow_gap=0,intercept_tls_get_addr=0
# must be set at runtime
# https://github.com/google/sanitizers/issues/1322
if (LMDEPLOY_ASAN_ENABLE)
add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-fsanitize=address>)
add_link_options(-fsanitize=address)
endif ()
# notice that ubsan has linker issues for ubuntu < 18.04, see
# https://stackoverflow.com/questions/50024731/ld-unrecognized-option-push-state-no-as-needed
if (LMDEPLOY_UBSAN_ENABLE)
add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-fsanitize=undefined>)
add_link_options(-fsanitize=undefined)
endif ()
if(BUILD_MULTI_GPU)
message(STATUS "Add DBUILD_MULTI_GPU, requires MPI and NCCL")
add_definitions("-DBUILD_MULTI_GPU")
set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake/Modules)
find_package(MPI REQUIRED)
find_package(NCCL REQUIRED)
set(CMAKE_MODULE_PATH "") # prevent the bugs for pytorch building
endif()
if(BUILD_PYT)
if(DEFINED ENV{NVIDIA_PYTORCH_VERSION})
if($ENV{NVIDIA_PYTORCH_VERSION} VERSION_LESS "20.03")
message(FATAL_ERROR "NVIDIA PyTorch image is too old for TorchScript mode.")
endif()
if($ENV{NVIDIA_PYTORCH_VERSION} VERSION_EQUAL "20.03")
add_definitions(-DLEGACY_THS=1)
endif()
endif()
endif()
if(USE_TRITONSERVER_DATATYPE)
message("-- USE_TRITONSERVER_DATATYPE")
add_definitions("-DUSE_TRITONSERVER_DATATYPE")
endif()
set(CXX_STD "17" CACHE STRING "C++ standard")
# enable gold linker for binary and .so
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -fuse-ld=gold")
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -fuse-ld=gold")
set(CUDA_PATH ${CUDA_TOOLKIT_ROOT_DIR})
set(TF_PATH "" CACHE STRING "TensorFlow path")
set(CUSPARSELT_PATH "" CACHE STRING "cuSPARSELt path")
if((BUILD_TF OR BUILD_TF2) AND NOT TF_PATH)
message(FATAL_ERROR "TF_PATH must be set if BUILD_TF or BUILD_TF2 (=TensorFlow mode) is on.")
endif()
list(APPEND CMAKE_MODULE_PATH ${CUDA_PATH}/lib64)
# profiling
option(USE_NVTX "Whether or not to use nvtx" ON)
if(USE_NVTX)
message(STATUS "NVTX is enabled.")
add_definitions("-DUSE_NVTX")
endif()
# setting compiler flags
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}")
set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler -Wall -ldl") # -Xptxas -v
# TODO: build for sm_72 & sm_87 on aarch64 platform (Jetson devices)
if (NOT CMAKE_CUDA_ARCHITECTURES)
set(CMAKE_CUDA_ARCHITECTURES 70-real 75-real)
if (${CUDA_VERSION} VERSION_GREATER_EQUAL "11")
list(APPEND CMAKE_CUDA_ARCHITECTURES 80-real)
endif ()
if (${CUDA_VERSION} VERSION_GREATER_EQUAL "11.1")
list(APPEND CMAKE_CUDA_ARCHITECTURES 86-real)
endif ()
if (${CUDA_VERSION} VERSION_GREATER_EQUAL "11.8")
list(APPEND CMAKE_CUDA_ARCHITECTURES 89-real 90-real)
endif ()
endif ()
message(STATUS "Building with CUDA archs: ${CMAKE_CUDA_ARCHITECTURES}")
set(CMAKE_CUDA_RUNTIME_LIBRARY Shared)
set(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG} -Wall -O0")
set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -Wall -O0")
# set(CMAKE_CUDA_FLAGS_DEBUG "${CMAKE_CUDA_FLAGS_DEBUG} -O0 -G -Xcompiler -Wall --ptxas-options=-v --resource-usage")
set(CMAKE_CUDA_FLAGS_DEBUG "${CMAKE_CUDA_FLAGS_DEBUG} -O0 -G -Xcompiler -Wall -DCUDA_PTX_FP8_F2FP_ENABLED")
set(CMAKE_CXX_STANDARD "${CXX_STD}")
set(CMAKE_CXX_STANDARD_REQUIRED ON)
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-extended-lambda")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --std=c++${CXX_STD} -DCUDA_PTX_FP8_F2FP_ENABLED")
set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -O3")
set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} -O3")
# set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} -Xcompiler -O3 --ptxas-options=--verbose")
set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} -Xcompiler -O3 -DCUDA_PTX_FP8_F2FP_ENABLED")
set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "${CMAKE_CUDA_FLAGS_RELWITHDEBINFO} -Xcompiler -O3 -DCUDA_PTX_FP8_F2FP_ENABLED")
if(BUILD_FAST_MATH)
set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} --use_fast_math")
set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "${CMAKE_CUDA_FLAGS_RELWITHDEBINFO} --use_fast_math")
message("Release build CUDA flags: ${CMAKE_CUDA_FLAGS_RELEASE}")
endif()
set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
set(COMMON_HEADER_DIRS
${PROJECT_SOURCE_DIR}
${CUDA_PATH}/include
${CUTLASS_HEADER_DIR}
)
message("-- COMMON_HEADER_DIRS: ${COMMON_HEADER_DIRS}")
set(COMMON_LIB_DIRS
${CUDA_PATH}/lib64
)
if (SPARSITY_SUPPORT)
list(APPEND COMMON_HEADER_DIRS ${CUSPARSELT_PATH}/include)
list(APPEND COMMON_LIB_DIRS ${CUSPARSELT_PATH}/lib64)
add_definitions(-DSPARSITY_ENABLED=1)
endif()
if(BUILD_TF)
list(APPEND COMMON_HEADER_DIRS ${TF_PATH}/include)
list(APPEND COMMON_LIB_DIRS ${TF_PATH})
add_definitions(-D_GLIBCXX_USE_CXX11_ABI=0)
endif()
if(BUILD_TF2)
list(APPEND COMMON_HEADER_DIRS ${TF_PATH}/include)
list(APPEND COMMON_LIB_DIRS ${TF_PATH})
add_definitions(-D_GLIBCXX_USE_CXX11_ABI=1)
endif()
set(PYTHON_PATH "python" CACHE STRING "Python path")
if(BUILD_PYT)
execute_process(COMMAND ${PYTHON_PATH} "-c" "from __future__ import print_function; import torch; print(torch.__version__,end='');"
RESULT_VARIABLE _PYTHON_SUCCESS
OUTPUT_VARIABLE TORCH_VERSION)
if (TORCH_VERSION VERSION_LESS "1.5.0")
message(FATAL_ERROR "PyTorch >= 1.5.0 is needed for TorchScript mode.")
endif()
execute_process(COMMAND ${PYTHON_PATH} "-c" "from __future__ import print_function; import os; import torch;
print(os.path.dirname(torch.__file__),end='');"
RESULT_VARIABLE _PYTHON_SUCCESS
OUTPUT_VARIABLE TORCH_DIR)
if (NOT _PYTHON_SUCCESS MATCHES 0)
message(FATAL_ERROR "Torch config Error.")
endif()
list(APPEND CMAKE_PREFIX_PATH ${TORCH_DIR})
find_package(Torch REQUIRED)
execute_process(COMMAND ${PYTHON_PATH} "-c" "from __future__ import print_function; from distutils import sysconfig;
print(sysconfig.get_python_inc());"
RESULT_VARIABLE _PYTHON_SUCCESS
OUTPUT_VARIABLE PY_INCLUDE_DIR)
if (NOT _PYTHON_SUCCESS MATCHES 0)
message(FATAL_ERROR "Python config Error.")
endif()
list(APPEND COMMON_HEADER_DIRS ${PY_INCLUDE_DIR})
execute_process(COMMAND ${PYTHON_PATH} "-c" "from __future__ import print_function; import torch;
print(torch._C._GLIBCXX_USE_CXX11_ABI,end='');"
RESULT_VARIABLE _PYTHON_SUCCESS
OUTPUT_VARIABLE USE_CXX11_ABI)
message("-- USE_CXX11_ABI=${USE_CXX11_ABI}")
if (USE_CXX11_ABI)
set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "${CMAKE_CUDA_FLAGS_RELWITHDEBINFO} -D_GLIBCXX_USE_CXX11_ABI=1")
set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} -D_GLIBCXX_USE_CXX11_ABI=1")
set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} -D_GLIBCXX_USE_CXX11_ABI=1")
set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -D_GLIBCXX_USE_CXX11_ABI=1")
set(CMAKE_CUDA_FLAGS_DEBUG "${CMAKE_CUDA_FLAGS_DEBUG} -D_GLIBCXX_USE_CXX11_ABI=1")
set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -D_GLIBCXX_USE_CXX11_ABI=1")
else()
set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "${CMAKE_CUDA_FLAGS_RELWITHDEBINFO} -D_GLIBCXX_USE_CXX11_ABI=0")
set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} -D_GLIBCXX_USE_CXX11_ABI=0")
set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} -D_GLIBCXX_USE_CXX11_ABI=0")
set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -D_GLIBCXX_USE_CXX11_ABI=0")
set(CMAKE_CUDA_FLAGS_DEBUG "${CMAKE_CUDA_FLAGS_DEBUG} -D_GLIBCXX_USE_CXX11_ABI=0")
set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -D_GLIBCXX_USE_CXX11_ABI=0")
endif()
endif()
# turn off warnings on windows
if (MSVC)
foreach(
flag_var
CMAKE_CXX_FLAGS
CMAKE_CXX_FLAGS_DEBUG
CMAKE_CXX_FLAGS_RELEASE
CMAKE_CXX_FLAGS_MINSIZEREL
CMAKE_CXX_FLAGS_RELWITHDEBINFO
CMAKE_C_FLAGS
CMAKE_C_FLAGS_DEBUG
CMAKE_C_FLAGS_RELEASE
CMAKE_C_FLAGS_MINSIZEREL
CMAKE_C_FLAGS_RELWITHDEBINFO
CMAKE_CUDA_FLAGS
CMAKE_CUDA_FLAGS_DEBUG
CMAKE_CUDA_FLAGS_RELEASE
CMAKE_CUDA_FLAGS_MINSIZEREL
CMAKE_CUDA_FLAGS_RELWITHDEBINFO)
string(REGEX REPLACE "-Wall" " /W0 " ${flag_var} "${${flag_var}}")
endforeach()
endif()
if (BUILD_MULTI_GPU)
list(APPEND COMMON_HEADER_DIRS ${MPI_INCLUDE_PATH})
endif()
if(USE_TRITONSERVER_DATATYPE)
list(APPEND COMMON_HEADER_DIRS ${PROJECT_SOURCE_DIR}/../repo-core-src/include)
endif()
include_directories(
${COMMON_HEADER_DIRS}
)
link_directories(
${COMMON_LIB_DIRS}
)
# add_subdirectory(3rdparty)
add_subdirectory(src)
add_subdirectory(examples)
if(BUILD_TEST)
add_subdirectory(tests/csrc)
endif()
# install python api
if (BUILD_PY_FFI)
install(TARGETS _turbomind DESTINATION ${CMAKE_SOURCE_DIR}/lmdeploy/lib)
endif ()
if (MSVC)
return()
endif ()
# # Mesaure the compile time
option(MEASURE_BUILD_TIME "Measure the build time of each module" OFF)
if (MEASURE_BUILD_TIME)
set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE "${CMAKE_COMMAND} -E time")
set_property(GLOBAL PROPERTY RULE_LAUNCH_CUSTOM "${CMAKE_COMMAND} -E time")
set_property(GLOBAL PROPERTY RULE_LAUNCH_LINK "${CMAKE_COMMAND} -E time")
endif()
########################################
add_library(transformer-shared SHARED
$<TARGET_OBJECTS:BaseSamplingLayer>
$<TARGET_OBJECTS:DynamicDecodeLayer>
$<TARGET_OBJECTS:Llama>
$<TARGET_OBJECTS:LlamaTritonBackend>
$<TARGET_OBJECTS:gemm_s4_f16>
$<TARGET_OBJECTS:TopKSamplingLayer>
$<TARGET_OBJECTS:TopPSamplingLayer>
$<TARGET_OBJECTS:TransformerTritonBackend>
$<TARGET_OBJECTS:activation_kernels>
$<TARGET_OBJECTS:ban_bad_words>
$<TARGET_OBJECTS:cublasAlgoMap>
$<TARGET_OBJECTS:cublasMMWrapper>
$<TARGET_OBJECTS:cuda_utils>
$<TARGET_OBJECTS:custom_ar_comm>
$<TARGET_OBJECTS:custom_ar_kernels>
$<TARGET_OBJECTS:attention>
$<TARGET_OBJECTS:decoding_kernels>
$<TARGET_OBJECTS:gpt_kernels>
$<TARGET_OBJECTS:logprob_kernels>
$<TARGET_OBJECTS:logger>
$<TARGET_OBJECTS:memory_utils>
$<TARGET_OBJECTS:mpi_utils>
$<TARGET_OBJECTS:nccl_utils>
$<TARGET_OBJECTS:nvtx_utils>
$<TARGET_OBJECTS:anomaly_handler>
$<TARGET_OBJECTS:sampling_penalty_kernels>
$<TARGET_OBJECTS:sampling_topk_kernels>
$<TARGET_OBJECTS:sampling_topp_kernels>
$<TARGET_OBJECTS:stop_criteria>
$<TARGET_OBJECTS:tensor>
$<TARGET_OBJECTS:unfused_attention_kernels>
)
if (BUILD_MULTI_GPU)
target_link_libraries(transformer-shared PUBLIC
${MPI_CXX_LIBRARIES}
${NCCL_LIBRARIES}
)
endif()
if(USE_NVTX)
target_link_libraries(transformer-shared PUBLIC
-lnvToolsExt
)
endif()
set_target_properties(transformer-shared PROPERTIES POSITION_INDEPENDENT_CODE ON)
set_target_properties(transformer-shared PROPERTIES CUDA_RESOLVE_DEVICE_SYMBOLS ON)
set_target_properties(transformer-shared PROPERTIES LINKER_LANGUAGE CXX)
target_link_libraries(transformer-shared PUBLIC -lcudart -lcublas -lcublasLt -lcurand)
include(GNUInstallDirs)
set(INSTALL_CONFIGDIR ${CMAKE_INSTALL_LIBDIR}/cmake/TurboMind)
include(CMakePackageConfigHelpers)
configure_package_config_file(
${CMAKE_CURRENT_LIST_DIR}/cmake/TurboMindConfig.cmake.in
${CMAKE_CURRENT_BINARY_DIR}/TurboMindConfig.cmake
INSTALL_DESTINATION ${INSTALL_CONFIGDIR}
)
install(
FILES
${CMAKE_CURRENT_BINARY_DIR}/TurboMindConfig.cmake
DESTINATION ${INSTALL_CONFIGDIR}
)
install(
TARGETS
transformer-shared
EXPORT
transformer-shared-targets
LIBRARY DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/turbomind
ARCHIVE DESTINATION ${CMAKE_INSTALL_PREFIX}/backends/turbomind
RUNTIME DESTINATION ${CMAKE_INSTALL_PREFIX}/bin
)
install(
EXPORT
transformer-shared-targets
FILE
TurboMindTargets.cmake
DESTINATION
${INSTALL_CONFIGDIR}
)
export(
EXPORT
transformer-shared-targets
FILE
${CMAKE_CURRENT_BINARY_DIR}/TurboMindTargets.cmake
NAMESPACE
TritonCore::
)
export(PACKAGE TurboMind)