Skip to content

Commit 90e05a4

Browse files
Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into add_bn_eq
2 parents 5502abb + 67bd4cd commit 90e05a4

File tree

171 files changed

+4705
-11026
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

171 files changed

+4705
-11026
lines changed

CMakeLists.txt

+17-11
Original file line numberDiff line numberDiff line change
@@ -36,8 +36,7 @@ include(simd)
3636
################################ Configurations #######################################
3737
option(WITH_GPU "Compile PaddlePaddle with NVIDIA GPU" ${CUDA_FOUND})
3838
option(WITH_AVX "Compile PaddlePaddle with AVX intrinsics" ${AVX_FOUND})
39-
option(WITH_MKLDNN "Compile PaddlePaddle with mkl-dnn support." ${AVX_FOUND})
40-
option(WITH_MKLML "Compile PaddlePaddle with mklml package." ${AVX_FOUND})
39+
option(WITH_MKL "Compile PaddlePaddle with MKL support." ${AVX_FOUND})
4140
option(WITH_DSO "Compile PaddlePaddle with dynamic linked CUDA" ON)
4241
option(WITH_TESTING "Compile PaddlePaddle with unit testing" ON)
4342
option(WITH_SWIG_PY "Compile PaddlePaddle with inference api" ON)
@@ -82,10 +81,8 @@ if(ANDROID OR IOS)
8281
"Disable PYTHON when cross-compiling for Android and iOS" FORCE)
8382
set(WITH_RDMA OFF CACHE STRING
8483
"Disable RDMA when cross-compiling for Android and iOS" FORCE)
85-
set(WITH_MKLDNN OFF CACHE STRING
86-
"Disable MKLDNN when cross-compiling for Android and iOS" FORCE)
87-
set(WITH_MKLML OFF CACHE STRING
88-
"Disable MKLML package when cross-compiling for Android and iOS" FORCE)
84+
set(WITH_MKL OFF CACHE STRING
85+
"Disable MKL when cross-compiling for Android and iOS" FORCE)
8986

9087
# Compile PaddlePaddle mobile inference library
9188
if (NOT WITH_C_API)
@@ -111,6 +108,14 @@ else()
111108
set(THIRD_PARTY_BUILD_TYPE Release)
112109
endif()
113110

111+
set(WITH_MKLML ${WITH_MKL})
112+
if (WITH_MKL AND AVX2_FOUND)
113+
set(WITH_MKLDNN ON)
114+
else()
115+
message(STATUS "Do not have AVX2 intrinsics and disabled MKL-DNN")
116+
set(WITH_MKLDNN OFF)
117+
endif()
118+
114119
########################################################################################
115120

116121
include(external/mklml) # download mklml package
@@ -158,14 +163,15 @@ set(EXTERNAL_LIBS
158163
)
159164

160165
if(WITH_GPU)
161-
list(APPEND EXTERNAL_LIBS ${CUDA_LIBRARIES} ${CUDA_rt_LIBRARY})
162-
if(NOT WITH_DSO)
163-
list(APPEND EXTERNAL_LIBS ${CUDNN_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY} ${NCCL_LIBRARY})
164-
endif(NOT WITH_DSO)
166+
include(cuda)
165167
endif(WITH_GPU)
166168

169+
if(WITH_MKLML)
170+
list(APPEND EXTERNAL_LIBS ${MKLML_IOMP_LIB})
171+
endif()
172+
167173
if(WITH_MKLDNN)
168-
list(APPEND EXTERNAL_LIBS ${MKLDNN_LIB} ${MKLDNN_IOMP_LIB})
174+
list(APPEND EXTERNAL_LIBS ${MKLDNN_LIB})
169175
endif()
170176

171177
if(USE_NNPACK)

benchmark/IntelOptimizedPaddle.md

+19-8
Original file line numberDiff line numberDiff line change
@@ -12,11 +12,11 @@ Machine:
1212

1313
System: CentOS release 6.3 (Final), Docker 1.12.1.
1414

15-
PaddlePaddle: paddlepaddle/paddle:latest (TODO: will rerun after 0.11.0)
16-
17-
- MKL-DNN tag v0.10
18-
- MKLML 2018.0.20170720
15+
PaddlePaddle: paddlepaddle/paddle:latest (for MKLML and MKL-DNN), paddlepaddle/paddle:latest-openblas (for OpenBLAS)
16+
- MKL-DNN tag v0.11
17+
- MKLML 2018.0.1.20171007
1918
- OpenBLAS v0.2.20
19+
(TODO: will rerun after 0.11.0)
2020

2121
On each machine, we will test and compare the performance of training on single node using MKL-DNN / MKLML / OpenBLAS respectively.
2222

@@ -31,15 +31,26 @@ Input image size - 3 * 224 * 224, Time: images/second
3131

3232
| BatchSize | 64 | 128 | 256 |
3333
|--------------|-------| -----| --------|
34-
| OpenBLAS | 7.82 | 8.62 | 10.34 |
35-
| MKLML | 11.02 | 12.86 | 15.33 |
36-
| MKL-DNN | 27.69 | 28.8 | 29.27 |
34+
| OpenBLAS | 7.80 | 9.00 | 10.80 |
35+
| MKLML | 12.12 | 13.70 | 16.18 |
36+
| MKL-DNN | 28.46 | 29.83 | 30.44 |
37+
38+
39+
chart on batch size 128
40+
TBD
41+
42+
- ResNet-50
43+
44+
| BatchSize | 64 | 128 | 256 |
45+
|--------------|-------| ------| -------|
46+
| OpenBLAS | 25.22 | 25.68 | 27.12 |
47+
| MKLML | 32.52 | 31.89 | 33.12 |
48+
| MKL-DNN | 81.69 | 82.35 | 84.08 |
3749

3850

3951
chart on batch size 128
4052
TBD
4153

42-
- ResNet
4354
- GoogLeNet
4455

4556
### Laptop

benchmark/paddle/image/googlenet.py

+4-1
Original file line numberDiff line numberDiff line change
@@ -5,6 +5,7 @@
55
width = 224
66
num_class = 1000
77
batch_size = get_config_arg('batch_size', int, 128)
8+
use_gpu = get_config_arg('use_gpu', bool, True)
89

910
args = {'height': height, 'width': width, 'color': True, 'num_class': num_class}
1011
define_py_data_sources2(
@@ -16,6 +17,8 @@
1617
learning_method=MomentumOptimizer(0.9),
1718
regularization=L2Regularization(0.0005 * batch_size))
1819

20+
conv_projection = conv_projection if use_gpu else img_conv_layer
21+
1922
def inception2(name, input, channels, \
2023
filter1,
2124
filter3R, filter3,
@@ -138,7 +141,7 @@ def inception(name, input, channels, \
138141
cat = concat_layer(
139142
name=name,
140143
input=[cov1, cov3, cov5, covprj],
141-
bias_attr=True,
144+
bias_attr=True if use_gpu else False,
142145
act=ReluActivation())
143146
return cat
144147

benchmark/paddle/image/run_mkldnn.sh

+2-1
Original file line numberDiff line numberDiff line change
@@ -40,6 +40,7 @@ fi
4040
for use_mkldnn in True False; do
4141
for batchsize in 64 128 256; do
4242
train vgg 19 $batchsize $use_mkldnn
43-
train resnet 50 $batchsize $use_mkldnn
43+
train resnet 50 $batchsize $use_mkldnn
44+
train googlenet v1 $batchsize $use_mkldnn
4445
done
4546
done

cmake/configure.cmake

+8-21
Original file line numberDiff line numberDiff line change
@@ -76,27 +76,14 @@ else()
7676
include_directories(${CUDA_TOOLKIT_INCLUDE})
7777
endif(NOT WITH_GPU)
7878

79-
if(WITH_MKLDNN)
80-
add_definitions(-DPADDLE_USE_MKLDNN)
81-
if (WITH_MKLML AND MKLDNN_IOMP_DIR)
82-
message(STATUS "Enable Intel OpenMP at ${MKLDNN_IOMP_DIR}")
83-
set(OPENMP_FLAGS "-fopenmp")
84-
set(CMAKE_C_CREATE_SHARED_LIBRARY_FORBIDDEN_FLAGS ${OPENMP_FLAGS})
85-
set(CMAKE_CXX_CREATE_SHARED_LIBRARY_FORBIDDEN_FLAGS ${OPENMP_FLAGS})
86-
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OPENMP_FLAGS}")
87-
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OPENMP_FLAGS}")
88-
else()
89-
find_package(OpenMP)
90-
if(OPENMP_FOUND)
91-
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
92-
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
93-
else()
94-
message(WARNING "Can not find OpenMP."
95-
"Some performance features in MKLDNN may not be available")
96-
endif()
97-
endif()
98-
99-
endif(WITH_MKLDNN)
79+
if (WITH_MKLML AND MKLML_IOMP_LIB)
80+
message(STATUS "Enable Intel OpenMP with ${MKLML_IOMP_LIB}")
81+
set(OPENMP_FLAGS "-fopenmp")
82+
set(CMAKE_C_CREATE_SHARED_LIBRARY_FORBIDDEN_FLAGS ${OPENMP_FLAGS})
83+
set(CMAKE_CXX_CREATE_SHARED_LIBRARY_FORBIDDEN_FLAGS ${OPENMP_FLAGS})
84+
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OPENMP_FLAGS}")
85+
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OPENMP_FLAGS}")
86+
endif()
10087

10188
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${SIMD_FLAG}")
10289
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${SIMD_FLAG}")

cmake/cuda.cmake

+188
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,188 @@
1+
if(NOT WITH_GPU)
2+
return()
3+
endif()
4+
5+
set(paddle_known_gpu_archs "30 35 50 52 60 61 70")
6+
set(paddle_known_gpu_archs7 "30 35 50 52")
7+
set(paddle_known_gpu_archs8 "30 35 50 52 60 61")
8+
9+
######################################################################################
10+
# A function for automatic detection of GPUs installed (if autodetection is enabled)
11+
# Usage:
12+
# detect_installed_gpus(out_variable)
13+
function(detect_installed_gpus out_variable)
14+
if(NOT CUDA_gpu_detect_output)
15+
set(cufile ${PROJECT_BINARY_DIR}/detect_cuda_archs.cu)
16+
17+
file(WRITE ${cufile} ""
18+
"#include <cstdio>\n"
19+
"int main() {\n"
20+
" int count = 0;\n"
21+
" if (cudaSuccess != cudaGetDeviceCount(&count)) return -1;\n"
22+
" if (count == 0) return -1;\n"
23+
" for (int device = 0; device < count; ++device) {\n"
24+
" cudaDeviceProp prop;\n"
25+
" if (cudaSuccess == cudaGetDeviceProperties(&prop, device))\n"
26+
" std::printf(\"%d.%d \", prop.major, prop.minor);\n"
27+
" }\n"
28+
" return 0;\n"
29+
"}\n")
30+
31+
execute_process(COMMAND "${CUDA_NVCC_EXECUTABLE}" "-ccbin=${CUDA_HOST_COMPILER}"
32+
"--run" "${cufile}"
33+
WORKING_DIRECTORY "${PROJECT_BINARY_DIR}/CMakeFiles/"
34+
RESULT_VARIABLE nvcc_res OUTPUT_VARIABLE nvcc_out
35+
ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
36+
37+
if(nvcc_res EQUAL 0)
38+
# only keep the last line of nvcc_out
39+
STRING(REGEX REPLACE ";" "\\\\;" nvcc_out "${nvcc_out}")
40+
STRING(REGEX REPLACE "\n" ";" nvcc_out "${nvcc_out}")
41+
list(GET nvcc_out -1 nvcc_out)
42+
string(REPLACE "2.1" "2.1(2.0)" nvcc_out "${nvcc_out}")
43+
set(CUDA_gpu_detect_output ${nvcc_out} CACHE INTERNAL "Returned GPU architetures from detect_installed_gpus tool" FORCE)
44+
endif()
45+
endif()
46+
47+
if(NOT CUDA_gpu_detect_output)
48+
message(STATUS "Automatic GPU detection failed. Building for all known architectures.")
49+
set(${out_variable} ${paddle_known_gpu_archs} PARENT_SCOPE)
50+
else()
51+
set(${out_variable} ${CUDA_gpu_detect_output} PARENT_SCOPE)
52+
endif()
53+
endfunction()
54+
55+
56+
########################################################################
57+
# Function for selecting GPU arch flags for nvcc based on CUDA_ARCH_NAME
58+
# Usage:
59+
# select_nvcc_arch_flags(out_variable)
60+
function(select_nvcc_arch_flags out_variable)
61+
# List of arch names
62+
set(archs_names "Kepler" "Maxwell" "Pascal" "All" "Manual")
63+
set(archs_name_default "All")
64+
if(NOT CMAKE_CROSSCOMPILING)
65+
list(APPEND archs_names "Auto")
66+
endif()
67+
68+
# set CUDA_ARCH_NAME strings (so it will be seen as dropbox in CMake-Gui)
69+
set(CUDA_ARCH_NAME ${archs_name_default} CACHE STRING "Select target NVIDIA GPU achitecture.")
70+
set_property( CACHE CUDA_ARCH_NAME PROPERTY STRINGS "" ${archs_names} )
71+
mark_as_advanced(CUDA_ARCH_NAME)
72+
73+
# verify CUDA_ARCH_NAME value
74+
if(NOT ";${archs_names};" MATCHES ";${CUDA_ARCH_NAME};")
75+
string(REPLACE ";" ", " archs_names "${archs_names}")
76+
message(FATAL_ERROR "Only ${archs_names} architeture names are supported.")
77+
endif()
78+
79+
if(${CUDA_ARCH_NAME} STREQUAL "Manual")
80+
set(CUDA_ARCH_BIN ${paddle_known_gpu_archs} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
81+
set(CUDA_ARCH_PTX "50" CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")
82+
mark_as_advanced(CUDA_ARCH_BIN CUDA_ARCH_PTX)
83+
else()
84+
unset(CUDA_ARCH_BIN CACHE)
85+
unset(CUDA_ARCH_PTX CACHE)
86+
endif()
87+
88+
if(${CUDA_ARCH_NAME} STREQUAL "Kepler")
89+
set(cuda_arch_bin "30 35")
90+
elseif(${CUDA_ARCH_NAME} STREQUAL "Maxwell")
91+
set(cuda_arch_bin "50")
92+
elseif(${CUDA_ARCH_NAME} STREQUAL "Pascal")
93+
set(cuda_arch_bin "60 61")
94+
elseif(${CUDA_ARCH_NAME} STREQUAL "Volta")
95+
set(cuda_arch_bin "70")
96+
elseif(${CUDA_ARCH_NAME} STREQUAL "All")
97+
set(cuda_arch_bin ${paddle_known_gpu_archs})
98+
elseif(${CUDA_ARCH_NAME} STREQUAL "Auto")
99+
detect_installed_gpus(cuda_arch_bin)
100+
else() # (${CUDA_ARCH_NAME} STREQUAL "Manual")
101+
set(cuda_arch_bin ${CUDA_ARCH_BIN})
102+
endif()
103+
104+
# remove dots and convert to lists
105+
string(REGEX REPLACE "\\." "" cuda_arch_bin "${cuda_arch_bin}")
106+
string(REGEX REPLACE "\\." "" cuda_arch_ptx "${CUDA_ARCH_PTX}")
107+
string(REGEX MATCHALL "[0-9()]+" cuda_arch_bin "${cuda_arch_bin}")
108+
string(REGEX MATCHALL "[0-9]+" cuda_arch_ptx "${cuda_arch_ptx}")
109+
list(REMOVE_DUPLICATES cuda_arch_bin)
110+
list(REMOVE_DUPLICATES cuda_arch_ptx)
111+
112+
set(nvcc_flags "")
113+
set(nvcc_archs_readable "")
114+
115+
# Tell NVCC to add binaries for the specified GPUs
116+
foreach(arch ${cuda_arch_bin})
117+
if(arch MATCHES "([0-9]+)\\(([0-9]+)\\)")
118+
# User explicitly specified PTX for the concrete BIN
119+
list(APPEND nvcc_flags -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
120+
list(APPEND nvcc_archs_readable sm_${CMAKE_MATCH_1})
121+
else()
122+
# User didn't explicitly specify PTX for the concrete BIN, we assume PTX=BIN
123+
list(APPEND nvcc_flags -gencode arch=compute_${arch},code=sm_${arch})
124+
list(APPEND nvcc_archs_readable sm_${arch})
125+
endif()
126+
endforeach()
127+
128+
# Tell NVCC to add PTX intermediate code for the specified architectures
129+
foreach(arch ${cuda_arch_ptx})
130+
list(APPEND nvcc_flags -gencode arch=compute_${arch},code=compute_${arch})
131+
list(APPEND nvcc_archs_readable compute_${arch})
132+
endforeach()
133+
134+
string(REPLACE ";" " " nvcc_archs_readable "${nvcc_archs_readable}")
135+
set(${out_variable} ${nvcc_flags} PARENT_SCOPE)
136+
set(${out_variable}_readable ${nvcc_archs_readable} PARENT_SCOPE)
137+
endfunction()
138+
139+
message(STATUS "CUDA detected: " ${CUDA_VERSION})
140+
if (${CUDA_VERSION} LESS 7.0)
141+
set(paddle_known_gpu_archs ${paddle_known_gpu_archs})
142+
elseif (${CUDA_VERSION} LESS 8.0) # CUDA 7.x
143+
set(paddle_known_gpu_archs ${paddle_known_gpu_archs7})
144+
list(APPEND CUDA_NVCC_FLAGS "-D_MWAITXINTRIN_H_INCLUDED")
145+
list(APPEND CUDA_NVCC_FLAGS "-D__STRICT_ANSI__")
146+
elseif (${CUDA_VERSION} LESS 9.0) # CUDA 8.x
147+
set(paddle_known_gpu_archs ${paddle_known_gpu_archs8})
148+
list(APPEND CUDA_NVCC_FLAGS "-D_MWAITXINTRIN_H_INCLUDED")
149+
list(APPEND CUDA_NVCC_FLAGS "-D__STRICT_ANSI__")
150+
# CUDA 8 may complain that sm_20 is no longer supported. Suppress the
151+
# warning for now.
152+
list(APPEND CUDA_NVCC_FLAGS "-Wno-deprecated-gpu-targets")
153+
endif()
154+
155+
include_directories(${CUDA_INCLUDE_DIRS})
156+
list(APPEND EXTERNAL_LIBS ${CUDA_LIBRARIES} ${CUDA_rt_LIBRARY})
157+
if(NOT WITH_DSO)
158+
list(APPEND EXTERNAL_LIBS ${CUDNN_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY} ${NCCL_LIBRARY})
159+
endif(NOT WITH_DSO)
160+
161+
# setting nvcc arch flags
162+
select_nvcc_arch_flags(NVCC_FLAGS_EXTRA)
163+
list(APPEND CUDA_NVCC_FLAGS ${NVCC_FLAGS_EXTRA})
164+
message(STATUS "Added CUDA NVCC flags for: ${NVCC_FLAGS_EXTRA_readable}")
165+
166+
# Set C++11 support
167+
set(CUDA_PROPAGATE_HOST_FLAGS OFF)
168+
169+
# Release/Debug flags set by cmake. Such as -O3 -g -DNDEBUG etc.
170+
# So, don't set these flags here.
171+
list(APPEND CUDA_NVCC_FLAGS "-std=c++11")
172+
list(APPEND CUDA_NVCC_FLAGS "--use_fast_math")
173+
list(APPEND CUDA_NVCC_FLAGS "-Xcompiler -fPIC")
174+
# Set :expt-relaxed-constexpr to suppress Eigen warnings
175+
list(APPEND CUDA_NVCC_FLAGS "--expt-relaxed-constexpr")
176+
177+
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
178+
list(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_DEBUG})
179+
elseif(CMAKE_BUILD_TYPE STREQUAL "Release")
180+
list(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_RELEASE})
181+
elseif(CMAKE_BUILD_TYPE STREQUAL "RelWithDebInfo")
182+
list(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_RELWITHDEBINFO})
183+
elseif(CMAKE_BUILD_TYPE STREQUAL "MinSizeRel")
184+
list(APPEND CUDA_NVCC_FLAGS ${CMAKE_CXX_FLAGS_MINSIZEREL})
185+
endif()
186+
187+
mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD)
188+
mark_as_advanced(CUDA_SDK_ROOT_DIR CUDA_SEPARABLE_COMPILATION)

cmake/external/mkldnn.cmake

+7-7
Original file line numberDiff line numberDiff line change
@@ -40,10 +40,9 @@ INCLUDE_DIRECTORIES(${MKLDNN_INC_DIR})
4040

4141
IF(${CBLAS_PROVIDER} STREQUAL "MKLML")
4242
SET(MKLDNN_DEPENDS ${MKLML_PROJECT})
43-
SET(MKLDNN_MKLROOT ${MKLML_ROOT})
44-
SET(MKLDNN_IOMP_LIB ${MKLML_IOMP_LIB})
45-
SET(MKLDNN_IOMP_DIR ${MKLML_LIB_DIR})
46-
MESSAGE(STATUS "Build MKLDNN with ${MKLDNN_MKLROOT}")
43+
MESSAGE(STATUS "Build MKLDNN with MKLML ${MKLML_ROOT}")
44+
ELSE()
45+
MESSAGE(FATAL_ERROR "Should enable MKLML when build MKLDNN")
4746
ENDIF()
4847

4948
SET(MKLDNN_CFLAG "${CMAKE_C_FLAGS} -Wno-error=strict-overflow")
@@ -57,15 +56,16 @@ ExternalProject_Add(
5756
PREFIX ${MKLDNN_SOURCES_DIR}
5857
UPDATE_COMMAND ""
5958
CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${MKLDNN_INSTALL_DIR}
60-
CMAKE_ARGS -DMKLROOT=${MKLDNN_MKLROOT}
59+
CMAKE_ARGS -DMKLROOT=${MKLML_ROOT}
6160
CMAKE_ARGS -DCMAKE_C_FLAGS=${MKLDNN_CFLAG}
6261
CMAKE_ARGS -DCMAKE_CXX_FLAGS=${MKLDNN_CXXFLAG}
6362
CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${MKLDNN_INSTALL_DIR}
64-
-DMKLROOT:PATH=${MKLDNN_MKLROOT}
63+
-DMKLROOT:PATH=${MKLML_ROOT}
6564
)
6665

6766
ADD_LIBRARY(mkldnn SHARED IMPORTED GLOBAL)
6867
SET_PROPERTY(TARGET mkldnn PROPERTY IMPORTED_LOCATION ${MKLDNN_LIB})
6968
ADD_DEPENDENCIES(mkldnn ${MKLDNN_PROJECT})
70-
MESSAGE(STATUS "Mkldnn library: ${MKLDNN_LIB}")
69+
MESSAGE(STATUS "MKLDNN library: ${MKLDNN_LIB}")
70+
add_definitions(-DPADDLE_USE_MKLDNN)
7171
LIST(APPEND external_project_dependencies mkldnn)

0 commit comments

Comments
 (0)