cpu_extension.cmake 7.51 KB
Newer Older
1
2
3
4
include(FetchContent)

set(CMAKE_CXX_STANDARD_REQUIRED ON)
set(CMAKE_CXX_EXTENSIONS ON)
5
6
set(CMAKE_EXPORT_COMPILE_COMMANDS ON)

7
8
9
10
11
if (${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
    set(MACOSX_FOUND TRUE)
endif()


12
13
14
15
16
17
18
19
20
#
# Define environment variables for special configurations
#
if(DEFINED ENV{VLLM_CPU_AVX512BF16})
    set(ENABLE_AVX512BF16 ON)
endif()

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
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)

73
74
75
76
77
if (MACOSX_FOUND AND CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
    set(APPLE_SILICON_FOUND TRUE)
else()
    find_isa(${CPUINFO} "avx2" AVX2_FOUND)
    find_isa(${CPUINFO} "avx512f" AVX512_FOUND)
78
    find_isa(${CPUINFO} "Power11" POWER11_FOUND)
79
80
81
82
    find_isa(${CPUINFO} "POWER10" POWER10_FOUND)
    find_isa(${CPUINFO} "POWER9" POWER9_FOUND)
    find_isa(${CPUINFO} "asimd" ASIMD_FOUND) # Check for ARM NEON support
    find_isa(${CPUINFO} "bf16" ARM_BF16_FOUND) # Check for ARM BF16 support
83
    find_isa(${CPUINFO} "S390" S390_FOUND)
84
85
endif()

86

87
if (AVX512_FOUND AND NOT AVX512_DISABLED)
88
89
90
91
92
93
94
95
    list(APPEND CXX_COMPILE_FLAGS
        "-mavx512f"
        "-mavx512vl"
        "-mavx512bw"
        "-mavx512dq")

    find_isa(${CPUINFO} "avx512_bf16" AVX512BF16_FOUND)
    if (AVX512BF16_FOUND OR ENABLE_AVX512BF16)
96
97
        if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND
            CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 12.3)
98
            list(APPEND CXX_COMPILE_FLAGS "-mavx512bf16")
99
            set(ENABLE_AVX512BF16 ON)
100
        else()
101
            set(ENABLE_AVX512BF16 OFF)
102
103
104
            message(WARNING "Disable AVX512-BF16 ISA support, requires gcc/g++ >= 12.3")
        endif()
    else()
105
        set(ENABLE_AVX512BF16 OFF)
106
107
        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()
108
109
110
111
112
113

    find_isa(${CPUINFO} "avx512_vnni" AVX512VNNI_FOUND)
    if (AVX512VNNI_FOUND)
        list(APPEND CXX_COMPILE_FLAGS "-mavx512vnni")
        set(ENABLE_AVX512VNNI ON) 
    endif() 
114
    
115
116
117
elseif (AVX2_FOUND)
    list(APPEND CXX_COMPILE_FLAGS "-mavx2")
    message(WARNING "vLLM CPU backend using AVX2 ISA")
118
    
119
elseif (POWER9_FOUND OR POWER10_FOUND OR POWER11_FOUND)
120
    message(STATUS "PowerPC detected")
121
122
123
124
125
126
127
128
129
130
131
    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()
132
133
134
135
136
137
138
139
140
141
142
143

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})     
144
145
146
elseif(APPLE_SILICON_FOUND)
    message(STATUS "Apple Silicon Detected")
    set(ENABLE_NUMA OFF)
147
148
149
150
151
152
153
154
elseif (S390_FOUND)
    message(STATUS "S390 detected")
    # Check for S390 VXE support
    list(APPEND CXX_COMPILE_FLAGS
        "-mvx"
        "-mzvector"
        "-march=native"
        "-mtune=native")
155
else()
156
    message(FATAL_ERROR "vLLM CPU backend requires AVX512, AVX2, Power9+ ISA, S390X ISA or ARMv8 support.")
157
158
endif()

159
160
161
162
163
164
165
#
# Build oneDNN for W8A8 GEMM kernels (only for x86-AVX512 platforms)
#
if (AVX512_FOUND AND NOT AVX512_DISABLED)
    FetchContent_Declare(
        oneDNN
        GIT_REPOSITORY https://github.com/oneapi-src/oneDNN.git
166
        GIT_TAG  v3.7.1
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
        GIT_PROGRESS TRUE
        GIT_SHALLOW TRUE
    )

    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")
    set(CMAKE_POLICY_DEFAULT_CMP0077 NEW)

    FetchContent_MakeAvailable(oneDNN)
    
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
    list(APPEND LIBS dnnl)
elseif(POWER10_FOUND)
    FetchContent_Declare(
        oneDNN
        GIT_REPOSITORY https://github.com/oneapi-src/oneDNN.git
        GIT_TAG v3.7.2
        GIT_PROGRESS TRUE
        GIT_SHALLOW TRUE
    )

    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")
    set(CMAKE_POLICY_DEFAULT_CMP0077 NEW)

    set(DNNL_CPU_RUNTIME "OMP")

    FetchContent_MakeAvailable(oneDNN)

213
214
215
    list(APPEND LIBS dnnl)
endif()

216
217
message(STATUS "CPU extension compile flags: ${CXX_COMPILE_FLAGS}")

218
219
220
221
222
223
if(ENABLE_NUMA)
    list(APPEND LIBS numa)
else()
    message(STATUS "NUMA is disabled")
    add_compile_definitions(-DVLLM_NUMA_DISABLED)
endif()
224

225
226
227
228
229
230
231
#
# _C extension
#
set(VLLM_EXT_SRC
    "csrc/cpu/activation.cpp"
    "csrc/cpu/attention.cpp"
    "csrc/cpu/cache.cpp"
232
    "csrc/cpu/utils.cpp"
233
    "csrc/cpu/layernorm.cpp"
Thien Tran's avatar
Thien Tran committed
234
    "csrc/cpu/mla_decode.cpp"
235
    "csrc/cpu/pos_encoding.cpp"
236
    "csrc/cpu/torch_bindings.cpp")
237

238
239
240
if (AVX512_FOUND AND NOT AVX512_DISABLED)
    set(VLLM_EXT_SRC
        "csrc/cpu/quant.cpp"
241
        "csrc/cpu/shm.cpp"
242
        ${VLLM_EXT_SRC})
243
244
245
246
247
248
249
250
251
252
253
    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()
254
255
256
257
elseif(POWER10_FOUND)
    set(VLLM_EXT_SRC
        "csrc/cpu/quant.cpp"
        ${VLLM_EXT_SRC})
258
259
260
261
262
263
endif()

#
# Define extension targets
#

264
265
266
267
268
define_gpu_extension_target(
    _C
    DESTINATION vllm
    LANGUAGE CXX
    SOURCES ${VLLM_EXT_SRC}
269
    LIBRARIES ${LIBS}
270
    COMPILE_FLAGS ${CXX_COMPILE_FLAGS}
271
272
    USE_SABI 3
    WITH_SOABI
273
274
)

275
message(STATUS "Enabling C extension.")