Build mobile inference library for minimum size #4509
Conversation
…ines when compile mobile inference library.
hedaoyuan
changed the title
CMakeLists.txt
Outdated
| set(MOBILE_INFERENCE ON) | ||
| add_definitions(-DPADDLE_MOBILE_INFERENCE) | ||
| endif() | ||
| set(WITH_TESTING OFF CACHE STRING "Disable TESTING when cross-compiling |
There was a problem hiding this comment.
I'd like to not set WITH_TESTING to OFF, because it is useful to check the linking error.
There was a problem hiding this comment.
I agree that the WITH_TESTING switch needs to be opened, but this PR will lead to some test program compiler failure, so I turn off the WITH_TESTING for temporarily.
CMakeLists.txt
Outdated
| @@ -86,6 +86,14 @@ if(ANDROID OR IOS) | |||
| "Disable MKLDNN when cross-compiling for Android and iOS" FORCE) | |||
| set(WITH_MKLML OFF CACHE STRING | |||
| "Disable MKLML package when cross-compiling for Android and iOS" FORCE) | |||
|
|
|||
| if(WITH_C_API) | |||
There was a problem hiding this comment.
Do we need to always set WITH_C_API to ON?
| ${CMAKE_DL_LIBS} | ||
| ${RDMA_LD_FLAGS} | ||
| ${RDMA_LIBS}) | ||
| endif() |
There was a problem hiding this comment.
Maybe we can use a new variable to specify which modules need to link, like:
if(MOBILE_INFERENCE)
set(PADDLE_LIBS
paddle_math
paddle_utils
paddle_parameter
paddle_proto
paddle_cuda
)
else()
set(PADDLE_LIBS
paddle_pserver
paddle_trainer_lib
paddle_network
paddle_math
paddle_utils
paddle_parameter
paddle_proto
paddle_cuda
paddle_optimizer
)
endif()
target_circle_link_libraries(${TARGET_NAME}
ARCHIVE_START
paddle_gserver
paddle_function
ARCHIVE_END
${PADDLE_LIBS}
${EXTERNAL_LIBS}
${CMAKE_THREAD_LIBS_INIT}
${CMAKE_DL_LIBS}
${RDMA_LD_FLAGS}
${RDMA_LIBS})
endif()
There was a problem hiding this comment.
I will fix this commit in the next pr.
paddle/capi/CMakeLists.txt
Outdated
| @@ -50,7 +48,9 @@ if(NOT IOS) | |||
| add_library(paddle_capi_shared SHARED ${CAPI_SOURCES}) | |||
| set_target_properties(paddle_capi_shared PROPERTIES LINK_FLAGS "${LINK_FLAGS}") | |||
| target_include_directories(paddle_capi_shared PUBLIC ${CMAKE_CURRENT_BINARY_DIR}) | |||
| link_paddle_exe(paddle_capi_shared) | |||
|
|
|||
| link_paddle_exe(paddle_capi_shared) | |||
|
|
||
| # Remove useless layers | ||
| list(REMOVE_ITEM GSERVER_SOURCES | ||
| layers/RecurrentLayerGroup.cpp) |
There was a problem hiding this comment.
I am not sure whether layers/RecurrentLayerGroup.cpp can be removed.
And cost layers can be removed too?
There was a problem hiding this comment.
For RecurrentLayerGroup.cpp, first remove Evaluator, then remove the RecurrentGradientMachine which depends on Evaluator, and last remove 'RecurrentLayerGroup' which depends on 'RecurrentGradientMachine'.
For cost layers, can be considered in the next PR.
There was a problem hiding this comment.
对这里https://github.com/hedaoyuan/Paddle/blob/6627801a586bf93f1d872c643c121e19d5c2f1bf/paddle/gserver/layers/Layer.cpp#L18
也应该加一个这个
#ifndef PADDLE_MOBILE_INFERENCE
#endif
|
Have tested, LGTM |
There was a problem hiding this comment.
LGTM. Just sadly found that if I want to test this PR on Linux, I need to make following modifications:
+set(MOBILE_INFERENCE ON)
+add_definitions(-DPADDLE_MOBILE_INFERENCE)
if(ANDROID OR IOS)
if(ANDROID)
if(${CMAKE_SYSTEM_VERSION} VERSION_LESS "16")
@@ -92,8 +94,6 @@ if(ANDROID OR IOS)
set(WITH_C_API ON CACHE STRING
"Always compile the C_API when cross-compiling for Android and iOS" FORCE)
endif()
- set(MOBILE_INFERENCE ON)
- add_definitions(-DPADDLE_MOBILE_INFERENCE)
endif()
fix #1845