zapcc is a caching C++ compiler based on clang, designed to perform faster compilations. zapcc uses in-memory compilation cache in client-server architecture, remembering all compilation information between runs. zapcc is the client while zapccs is the server. Each zapcc run will reuse an existing server or if none was available will start a new one.
This open source release is licensed under the LLVM Release License (University of Illinois/NCSA).
zapcc builds on
zapcc-clis not supported
zapcc was thoroughly tested on Linux x64 targetting Linux x64 and minimally on Windows. Rest are experimental, please share your experience.
The prerequisites and build process are identical to building LLVM.
sudo apt-get install ninja-build git clone https://github.com/yrnkrn/zapcc.git llvm mkdir build cd build cmake -G Ninja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_WARNINGS=OFF ../llvm ninja
You need msys2 and the mingw-builds of mingw-w64. Note there are 32- and 64- bits distribtutions of mingw-w64. To target x86_64:
Download the latest MSYS2 installerWebKit and install into the default folder
Download one of the mingw-builds personal distributions, such as x86_64-8.1.0-release-posix-seh-rt_v6-rev0.7z and open into the folder
Add the bin directories to the PATH,
Make sure you have just this gcc version available on the PATH. gcc versions outside the PATH are OK.
Either Visual C++ or the just-installed mingw-w64 may be used to build zapcc.
If building using Visual C++, target
x86_64-pc-windows-gnu must be explicitly specified,
cmake -G Ninja -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_WARNINGS=OFF -DLLVM_DEFAULT_TARGET_TRIPLE=x86_64-pc-windows-gnu ../llvm
zapcc will now target mingw-w64 and ignore Visual C++, even if installed.
To target mingw-builds 32 bits, download the appropriate 32 bits distribution of mingw-builds and replace
i686 in the configuration.
zapcc command syntax is identical to clang with the command being zapcc.
To kill the zapccs server to free memory or replace with newly-built zapcc
Full builds are 2x-5x faster, see
Typically re-compilation of one modified source file is 10x-50x faster.
Acceleration depends on the complexity of the header files vs. the complexity of the source files. It can range from no acceleration at all for plain C projects where caching is disabled to x2-x5 for build-all of heavily templated projects, up to cases of x50 speedups in developer-mode incremental change of one source file.
As a reference number, Zapcc builds the LLVM
build-all target about x2 faster compared to building LLVM using clang.
Here are ASCII movies comparing clang and zapcc fully building WebKit and incremental building Boost.
Yes, zapcc is based on heavily-modified clang code.
Yes, to the extent clang is gcc compatible.
See CATC 2017 presentation and discussion at cfe-dev.
Precompiled headers requires building your project to the exact precompiled headers rules. Most projects do not bother with using precompiled headers. Even then, precompiled headers do not cache as much as zapcc. Zapcc works within your existing build.
Precompiled headers are currently ignored by Zapcc.
As of C++17, modules are not standard, rarely used and do not support well legacy code and macros found in most existing C++ code, such as Boost. Modules require significant code refactoring in bottom-up approach everything or are slow. Even then, modules do not cache template instantiations and generated code that are specific to your code like zapcc does.
Please make sure first your project compiles successfully with Clang. If your project does not compile with Clang, Zapcc, being based on Clang, will not be able to compile any more than clang.
To avoid killing the server by using endless memory, Zapcc server has a memory limit and will automatically reset after reaching it, restarting with an empty cache and low memory usage. The memory limit is set under [MaxMemory] at bin/zapccs.config, and you can change it to optimize memory usage and the number of servers you plan to use. Usually, you should not set the -j parameter to more than the number of physical cores + 2. This is especially important for Intel CPU with hyper-threading enabled, which report twice the number of physical cores. In such cases, Zapcc may run faster with fewer servers, each using a higher memory limit.
There are patches all around LLVM & clang. Additional zapcc-only code in
tools/zapcc tools/zapccs tools/clang/test/zapcc
This open-source release was last merged with LLVM 325000 on 2018-02-13.