cpu_extension.cmake 10.9 KB
Newer Older
1
2
3
include(FetchContent)

set(CMAKE_CXX_STANDARD_REQUIRED ON)
4
set(CMAKE_CXX_STANDARD 17)
5
set(CMAKE_CXX_EXTENSIONS ON)
6
7
set(CMAKE_EXPORT_COMPILE_COMMANDS ON)

8
9
10
11
12
if (${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
    set(MACOSX_FOUND TRUE)
endif()


13
14
15
#
# Define environment variables for special configurations
#
16
17
set(ENABLE_AVX512BF16 $ENV{VLLM_CPU_AVX512BF16})
set(ENABLE_AVX512VNNI $ENV{VLLM_CPU_AVX512VNNI})
18
19
20

include_directories("${CMAKE_SOURCE_DIR}/csrc")

21
22
23

set (ENABLE_NUMA TRUE)

24
25
26
#
# Check the compile flags
#
27
28

if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
29
30
    list(APPEND CXX_COMPILE_FLAGS
        "-mf16c"
31
    )
32
endif()
33

34
35
36
37
38
39
40
41
if(MACOSX_FOUND)
    list(APPEND CXX_COMPILE_FLAGS
        "-DVLLM_CPU_EXTENSION")
else()
    list(APPEND CXX_COMPILE_FLAGS
        "-fopenmp"
        "-DVLLM_CPU_EXTENSION")
endif()
42

43
44
45
46
47
48
49
if (NOT MACOSX_FOUND)
    execute_process(COMMAND cat /proc/cpuinfo
                    RESULT_VARIABLE CPUINFO_RET
                    OUTPUT_VARIABLE CPUINFO)
    if (NOT CPUINFO_RET EQUAL 0)
        message(FATAL_ERROR "Failed to check CPU features via /proc/cpuinfo")
    endif()
50
51
endif()

52

53
54
55
56
57
58
59
60
61
function (find_isa CPUINFO TARGET OUT)
    string(FIND ${CPUINFO} ${TARGET} ISA_FOUND)
    if(NOT ISA_FOUND EQUAL -1)
        set(${OUT} ON PARENT_SCOPE)
    else()
        set(${OUT} OFF PARENT_SCOPE)
    endif()
endfunction()

62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77

function(check_sysctl TARGET OUT)
    execute_process(COMMAND sysctl -n "${TARGET}"
                    RESULT_VARIABLE SYSCTL_RET
                    OUTPUT_VARIABLE SYSCTL_INFO
                    ERROR_QUIET
                    OUTPUT_STRIP_TRAILING_WHITESPACE)
    if(SYSCTL_RET EQUAL 0 AND
      (SYSCTL_INFO STREQUAL "1" OR SYSCTL_INFO GREATER 0))
        set(${OUT} ON PARENT_SCOPE)
    else()
        set(${OUT} OFF PARENT_SCOPE)
    endif()
endfunction()


78
79
80
81
82
83
84
85
86
87
88
function (is_avx512_disabled OUT)
    set(DISABLE_AVX512 $ENV{VLLM_CPU_DISABLE_AVX512})
    if(DISABLE_AVX512 AND DISABLE_AVX512 STREQUAL "true")
        set(${OUT} ON PARENT_SCOPE)
    else()
        set(${OUT} OFF PARENT_SCOPE)
    endif()
endfunction()

is_avx512_disabled(AVX512_DISABLED)

89
if (MACOSX_FOUND AND CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
90
    message(STATUS "Apple Silicon Detected")
91
    set(APPLE_SILICON_FOUND TRUE)
92
93
94
    set(ENABLE_NUMA OFF)
    check_sysctl(hw.optional.neon ASIMD_FOUND)
    check_sysctl(hw.optional.arm.FEAT_BF16 ARM_BF16_FOUND)
95
96
97
else()
    find_isa(${CPUINFO} "avx2" AVX2_FOUND)
    find_isa(${CPUINFO} "avx512f" AVX512_FOUND)
98
    find_isa(${CPUINFO} "Power11" POWER11_FOUND)
99
100
101
102
    find_isa(${CPUINFO} "POWER10" POWER10_FOUND)
    find_isa(${CPUINFO} "POWER9" POWER9_FOUND)
    find_isa(${CPUINFO} "asimd" ASIMD_FOUND) # Check for ARM NEON support
    find_isa(${CPUINFO} "bf16" ARM_BF16_FOUND) # Check for ARM BF16 support
103
    find_isa(${CPUINFO} "S390" S390_FOUND)
104
    find_isa(${CPUINFO} "v" RVV_FOUND) # Check for RISC-V RVV support
105
106
endif()

107
if (AVX512_FOUND AND NOT AVX512_DISABLED)
108
109
110
111
112
113
114
115
    list(APPEND CXX_COMPILE_FLAGS
        "-mavx512f"
        "-mavx512vl"
        "-mavx512bw"
        "-mavx512dq")

    find_isa(${CPUINFO} "avx512_bf16" AVX512BF16_FOUND)
    if (AVX512BF16_FOUND OR ENABLE_AVX512BF16)
116
117
        if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND
            CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 12.3)
118
            list(APPEND CXX_COMPILE_FLAGS "-mavx512bf16")
119
            set(ENABLE_AVX512BF16 ON)
120
        else()
121
            set(ENABLE_AVX512BF16 OFF)
122
123
124
            message(WARNING "Disable AVX512-BF16 ISA support, requires gcc/g++ >= 12.3")
        endif()
    else()
125
        set(ENABLE_AVX512BF16 OFF)
126
127
        message(WARNING "Disable AVX512-BF16 ISA support, no avx512_bf16 found in local CPU flags." " If cross-compilation is required, please set env VLLM_CPU_AVX512BF16=1.")
    endif()
128
129

    find_isa(${CPUINFO} "avx512_vnni" AVX512VNNI_FOUND)
130
131
132
133
134
135
136
137
138
139
140
141
142
    if (AVX512VNNI_FOUND OR ENABLE_AVX512VNNI)
        if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND
            CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 12.3)
            list(APPEND CXX_COMPILE_FLAGS "-mavx512vnni")
            set(ENABLE_AVX512VNNI ON)
        else()
            set(ENABLE_AVX512VNNI OFF)
            message(WARNING "Disable AVX512-VNNI ISA support, requires gcc/g++ >= 12.3")
        endif()
    else()
        set(ENABLE_AVX512VNNI OFF)
        message(WARNING "Disable AVX512-VNNI ISA support, no avx512_vnni found in local CPU flags." " If cross-compilation is required, please set env VLLM_CPU_AVX512VNNI=1.")
    endif()
143
    
144
145
146
elseif (AVX2_FOUND)
    list(APPEND CXX_COMPILE_FLAGS "-mavx2")
    message(WARNING "vLLM CPU backend using AVX2 ISA")
147
    
148
elseif (POWER9_FOUND OR POWER10_FOUND OR POWER11_FOUND)
149
    message(STATUS "PowerPC detected")
150
151
152
153
154
155
156
157
158
159
160
    if (POWER9_FOUND)
        list(APPEND CXX_COMPILE_FLAGS
            "-mvsx"
            "-mcpu=power9"
            "-mtune=power9")
    elseif (POWER10_FOUND OR POWER11_FOUND)
        list(APPEND CXX_COMPILE_FLAGS
            "-mvsx"
            "-mcpu=power10"
            "-mtune=power10")
    endif()
161
162
163
164
165
166
167
168
169
170
171
172

elseif (ASIMD_FOUND)
    message(STATUS "ARMv8 or later architecture detected")
    if(ARM_BF16_FOUND)
        message(STATUS "BF16 extension detected")
        set(MARCH_FLAGS "-march=armv8.2-a+bf16+dotprod+fp16")
        add_compile_definitions(ARM_BF16_SUPPORT)
    else()
        message(WARNING "BF16 functionality is not available")
        set(MARCH_FLAGS "-march=armv8.2-a+dotprod+fp16")  
    endif()
    list(APPEND CXX_COMPILE_FLAGS ${MARCH_FLAGS})     
173
174
175
176
177
178
179
180
elseif (S390_FOUND)
    message(STATUS "S390 detected")
    # Check for S390 VXE support
    list(APPEND CXX_COMPILE_FLAGS
        "-mvx"
        "-mzvector"
        "-march=native"
        "-mtune=native")
181
182
183
184
185
186
elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "riscv64")
    if(RVV_FOUND)
	    message(FAIL_ERROR "Can't support rvv now.")
    else()
        list(APPEND CXX_COMPILE_FLAGS "-march=rv64gc")
    endif()
187
else()
188
    message(FATAL_ERROR "vLLM CPU backend requires AVX512, AVX2, Power9+ ISA, S390X ISA, ARMv8 or RISC-V support.")
189
190
endif()

191

192
# Build oneDNN for GEMM kernels (only for x86-AVX512 /ARM platforms)
193
if ((AVX512_FOUND AND NOT AVX512_DISABLED) OR (ASIMD_FOUND AND NOT APPLE_SILICON_FOUND) OR POWER9_FOUND OR POWER10_FOUND OR POWER11_FOUND)
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
    # Fetch and build Arm Compute Library (ACL) as oneDNN's backend for AArch64
    # TODO [fadara01]: remove this once ACL can be fetched and built automatically as a dependency of oneDNN
    if(ASIMD_FOUND)
        if(DEFINED ENV{ACL_ROOT_DIR} AND IS_DIRECTORY "$ENV{ACL_ROOT_DIR}")
            message(STATUS "Using ACL from specified source directory: $ENV{ACL_ROOT_DIR}")
        else()
            message(STATUS "Downloading Arm Compute Library (ACL) from GitHub")
            FetchContent_Populate(arm_compute
                SUBBUILD_DIR "${FETCHCONTENT_BASE_DIR}/arm_compute-subbuild"
                SOURCE_DIR   "${FETCHCONTENT_BASE_DIR}/arm_compute-src"
                GIT_REPOSITORY https://github.com/ARM-software/ComputeLibrary.git
                GIT_TAG        v52.2.0
                GIT_SHALLOW    TRUE
                GIT_PROGRESS   TRUE
            )
            set(ENV{ACL_ROOT_DIR} "${arm_compute_SOURCE_DIR}")
        endif()

        # Build ACL with scons
        include(ProcessorCount)
        ProcessorCount(_NPROC)
        execute_process(
            COMMAND scons -j${_NPROC}
                    Werror=0 debug=0 neon=1 examples=0 embed_kernels=0 os=linux
                    arch=armv8.2-a build=native benchmark_examples=0 fixed_format_kernels=1
                    multi_isa=1 openmp=1 cppthreads=0
            WORKING_DIRECTORY "$ENV{ACL_ROOT_DIR}"
            RESULT_VARIABLE _acl_rc
        )
        if(NOT _acl_rc EQUAL 0)
            message(FATAL_ERROR "ACL SCons build failed (exit ${_acl_rc}).")
        endif()

        set(ONEDNN_AARCH64_USE_ACL "ON")
        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wl,-rpath,$ENV{ACL_ROOT_DIR}/build/")
        add_compile_definitions(VLLM_USE_ACL)
    endif()

232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
    set(FETCHCONTENT_SOURCE_DIR_ONEDNN "$ENV{FETCHCONTENT_SOURCE_DIR_ONEDNN}" CACHE PATH "Path to a local oneDNN source directory.")

    if(FETCHCONTENT_SOURCE_DIR_ONEDNN)
        message(STATUS "Using oneDNN from specified source directory: ${FETCHCONTENT_SOURCE_DIR_ONEDNN}")
        FetchContent_Declare(
            oneDNN
            SOURCE_DIR ${FETCHCONTENT_SOURCE_DIR_ONEDNN}
        )
    else()
        message(STATUS "Downloading oneDNN from GitHub")
        FetchContent_Declare(
            oneDNN
            GIT_REPOSITORY https://github.com/oneapi-src/oneDNN.git
            GIT_TAG v3.9
            GIT_PROGRESS TRUE
            GIT_SHALLOW TRUE
        )
    endif()
250
251
252
253
254
255
256
257
258
259
260
261

    set(ONEDNN_LIBRARY_TYPE "STATIC")
    set(ONEDNN_BUILD_DOC "OFF")
    set(ONEDNN_BUILD_EXAMPLES "OFF")
    set(ONEDNN_BUILD_TESTS "OFF")
    set(ONEDNN_ENABLE_WORKLOAD "INFERENCE")
    set(ONEDNN_ENABLE_PRIMITIVE "MATMUL;REORDER")
    set(ONEDNN_BUILD_GRAPH "OFF")
    set(ONEDNN_ENABLE_JIT_PROFILING "OFF")
    set(ONEDNN_ENABLE_ITT_TASKS "OFF")
    set(ONEDNN_ENABLE_MAX_CPU_ISA "OFF")
    set(ONEDNN_ENABLE_CPU_ISA_HINTS "OFF")
262
    set(ONEDNN_VERBOSE "OFF")
263
264
265
    set(CMAKE_POLICY_DEFAULT_CMP0077 NEW)

    FetchContent_MakeAvailable(oneDNN)
266
267
268
269
270
271
    add_library(dnnl_ext OBJECT "csrc/cpu/dnnl_helper.cpp")
    target_include_directories(
        dnnl_ext
        PUBLIC ${oneDNN_SOURCE_DIR}/include
        PUBLIC ${oneDNN_BINARY_DIR}/include
        PRIVATE ${oneDNN_SOURCE_DIR}/src
272
    )
273
274
275
276
277
278
    target_link_libraries(dnnl_ext dnnl)
    target_compile_options(dnnl_ext PRIVATE ${CXX_COMPILE_FLAGS} -fPIC)
    list(APPEND LIBS dnnl_ext)
    set(USE_ONEDNN ON)
else()
    set(USE_ONEDNN OFF)
279
280
endif()

281
282
message(STATUS "CPU extension compile flags: ${CXX_COMPILE_FLAGS}")

283
284
285
286
287
288
if(ENABLE_NUMA)
    list(APPEND LIBS numa)
else()
    message(STATUS "NUMA is disabled")
    add_compile_definitions(-DVLLM_NUMA_DISABLED)
endif()
289

290
291
292
293
294
295
296
#
# _C extension
#
set(VLLM_EXT_SRC
    "csrc/cpu/activation.cpp"
    "csrc/cpu/attention.cpp"
    "csrc/cpu/cache.cpp"
297
    "csrc/cpu/utils.cpp"
298
    "csrc/cpu/layernorm.cpp"
Thien Tran's avatar
Thien Tran committed
299
    "csrc/cpu/mla_decode.cpp"
300
    "csrc/cpu/pos_encoding.cpp"
301
302
    "csrc/cpu/torch_bindings.cpp"
    "csrc/moe/dynamic_4bit_int_moe_cpu.cpp")
303

304
305
if (AVX512_FOUND AND NOT AVX512_DISABLED)
    set(VLLM_EXT_SRC
306
        "csrc/cpu/shm.cpp"
307
        ${VLLM_EXT_SRC})
308
309
310
311
312
313
314
315
316
317
318
    if (ENABLE_AVX512BF16 AND ENABLE_AVX512VNNI)
        set(VLLM_EXT_SRC
            "csrc/cpu/sgl-kernels/gemm.cpp"
            "csrc/cpu/sgl-kernels/gemm_int8.cpp"
            "csrc/cpu/sgl-kernels/gemm_fp8.cpp"
            "csrc/cpu/sgl-kernels/moe.cpp"
            "csrc/cpu/sgl-kernels/moe_int8.cpp"
            "csrc/cpu/sgl-kernels/moe_fp8.cpp"
            ${VLLM_EXT_SRC})
        add_compile_definitions(-DCPU_CAPABILITY_AVX512)
    endif()
319
endif()
320
321

if(USE_ONEDNN)
322
    set(VLLM_EXT_SRC
323
        "csrc/cpu/dnnl_kernels.cpp"
324
325
        ${VLLM_EXT_SRC})
endif()
326

327
328
message(STATUS "CPU extension source files: ${VLLM_EXT_SRC}")

329
330
331
332
#
# Define extension targets
#

333
334
335
336
337
define_gpu_extension_target(
    _C
    DESTINATION vllm
    LANGUAGE CXX
    SOURCES ${VLLM_EXT_SRC}
338
    LIBRARIES ${LIBS}
339
    COMPILE_FLAGS ${CXX_COMPILE_FLAGS}
340
341
    USE_SABI 3
    WITH_SOABI
342
343
)

344
message(STATUS "Enabling C extension.")