This repository has been archived by the owner on Jan 24, 2024. It is now read-only.
How to build mobile inference library with minimum size #21
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Merge the updated develop branch |
Xreki
reviewed
Nov 2, 2017
deployment/build_for_minimum_size.md
Outdated
| @@ -0,0 +1,27 @@ | |||
| # Build mobile inference library with minimum size | |||
|
|
|||
| In many mobile applications, the size of the executable program will have some requirements. | |||
There was a problem hiding this comment.
the size of the executable program will have some requirements -> there usually are some limitations to the size of libraries.
deployment/build_for_minimum_size.md
Outdated
| # Build mobile inference library with minimum size | ||
|
|
||
| In many mobile applications, the size of the executable program will have some requirements. | ||
| Here we explore how to compile the inference library with minimum size. |
There was a problem hiding this comment.
with minimum size -> for minimum size
deployment/build_for_minimum_size.md
Outdated
|
|
||
| **Note:** | ||
| In the original PaddlePaddle, all computationally relevant code is implemented in Matrix.cpp and BaseMatrix.cu, | ||
| this causes the compiled Matrix.o and BaseMatrix.o files to be large and can not be split. |
There was a problem hiding this comment.
this causes large compiled Matrix.o and BaseMatrix.o files which cannot be split
deployment/build_for_minimum_size.md
Outdated
| In the original PaddlePaddle, all computationally relevant code is implemented in Matrix.cpp and BaseMatrix.cu, | ||
| this causes the compiled Matrix.o and BaseMatrix.o files to be large and can not be split. | ||
| The module of Layer can be split, but the Layer forward and backward computing is included in the same file. | ||
| The configuration definition in proto has some redundancy. These all will lead to the size of inference library larger. |
There was a problem hiding this comment.
has some redundancy -> is redundant
And I suggest to move These all will lead to the size of inference library larger. to line 7, as Three reasons lead to the large size of PaddlePaddle's inference library:, then list the three reasons.
There was a problem hiding this comment.
Done.
Maybe not only three. I think we can fix this later.
deployment/build_for_minimum_size.md
Outdated
| ## How to build with minimum size | ||
| Here we mainly introduce some optimization in the compilation process to reduce the size of the inference library. These methods are also used in the refactored PaddlePaddle. | ||
|
|
||
| - MinSizeRel: By specify the build type(CMAKE_BUILD_TYPE) as MinSizeRel, the -Os option is used by the compiler during the build for minimum size code. |
There was a problem hiding this comment.
By specify the build type(CMAKE_BUILD_TYPE) as MinSizeRel -> by specifing the build type (CMAKE_BUILD_TYPE) to MinSizeRel
deployment/build_for_minimum_size.md
Outdated
| - protobuf-lite: With the `--cpp_out lite` option, the generated proto configuration rely on MessageLite instead of Message, | ||
| so only libprotobuf-lite is needed for the link, and the size of the final executable file can be reduced. | ||
| - whole-archive: In `libpaddle_capi_layers.a` contains all the object file of the layers, need use the `--whole-archive` option to ensure that all layers be linked to the executable file. | ||
| But don't forget to use `-no-whole-archive after` after `libpaddle_capi_layers.a`, avoid this option affect other libraries. |
There was a problem hiding this comment.
Reduntant "after": -no-whole-archive after -> -no-whole-archive
deployment/build_for_minimum_size.md
Outdated
| ## How to build with minimum size | ||
| Here we mainly introduce some optimization in the compilation process to reduce the size of the inference library. These methods are also used in the refactored PaddlePaddle. | ||
|
|
||
| - MinSizeRel: By specify the build type(CMAKE_BUILD_TYPE) as MinSizeRel, the -Os option is used by the compiler during the build for minimum size code. |
There was a problem hiding this comment.
MinSizeRel: -> Set build type to MinSizeRel:
deployment/build_for_minimum_size.md
Outdated
| Here we mainly introduce some optimization in the compilation process to reduce the size of the inference library. These methods are also used in the refactored PaddlePaddle. | ||
|
|
||
| - MinSizeRel: By specify the build type(CMAKE_BUILD_TYPE) as MinSizeRel, the -Os option is used by the compiler during the build for minimum size code. | ||
| - protobuf-lite: With the `--cpp_out lite` option, the generated proto configuration rely on MessageLite instead of Message, |
There was a problem hiding this comment.
protobuf-lite: -> Use protobuf-lite instead of protobuf:
deployment/build_for_minimum_size.md
Outdated
| - MinSizeRel: By specify the build type(CMAKE_BUILD_TYPE) as MinSizeRel, the -Os option is used by the compiler during the build for minimum size code. | ||
| - protobuf-lite: With the `--cpp_out lite` option, the generated proto configuration rely on MessageLite instead of Message, | ||
| so only libprotobuf-lite is needed for the link, and the size of the final executable file can be reduced. | ||
| - whole-archive: In `libpaddle_capi_layers.a` contains all the object file of the layers, need use the `--whole-archive` option to ensure that all layers be linked to the executable file. |
There was a problem hiding this comment.
whole-archive: -> How to link static libraries:
deployment/build_for_minimum_size.md
Outdated
| - whole-archive: In `libpaddle_capi_layers.a` contains all the object file of the layers, need use the `--whole-archive` option to ensure that all layers be linked to the executable file. | ||
| But don't forget to use `-no-whole-archive after` after `libpaddle_capi_layers.a`, avoid this option affect other libraries. | ||
| [Here's an example](https://github.com/PaddlePaddle/Mobile/blob/develop/benchmark/tool/C/CMakeLists.txt#L41) | ||
| - Shared library: When building a shared library by `libpaddle_capi_layers.a` and `libpaddle_capi_engine.a`, |
There was a problem hiding this comment.
Shared library: -> How to generate shared library use PaddlePaddle's static libraries
Perhaps it is still not precise...
hedaoyuan
changed the title
Xreki
approved these changes
Nov 10, 2017
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.