ITK/Release 4/Wrapping/ccache: Difference between revisions
mNo edit summary |
Agouaillard (talk | contribs) No edit summary |
||
Line 1: | Line 1: | ||
= Using CCache for WrapITK = | = Using CCache for WrapITK = | ||
== Installation of CCACHE == | |||
Using CCache[http://ccache.samba.org] to build WrapITK can improve the build times considerably. For all unix systems (including MacOS X), ccache is enabled by creating symbolic links to the g++/gcc compiler: | Using CCache[http://ccache.samba.org] to build WrapITK can improve the build times considerably. For all unix systems (including MacOS X), ccache is enabled by creating symbolic links to the g++/gcc compiler: | ||
ln -s /usr/bin/ccache /usr/local/bin/g++ | ln -s /usr/bin/ccache /usr/local/bin/g++ | ||
Line 6: | Line 8: | ||
$ sudo port install ccache | $ sudo port install ccache | ||
$ export PATH= /opt/local/libexec/ccache:$PATH | $ export PATH= /opt/local/libexec/ccache:$PATH | ||
== ITK Specific configuration Flags == | |||
Note that ccache is installed outside of ITK, and will benefit any software project that uses gcc. | |||
The other parts of WrapITK which can benefit from ccache are gccxml and swig. To enable ccache usage with them, just turn ON the option WRAP_ITK_USE_CCACHE and everything is gonna be set up for you automatically. | |||
== Example of speed-ups == | |||
=== 8-cores (8 physical threads) === | |||
All the subsequent builds of WrapITK would use ccache and we can see a speedup of as much as 40 minutes for repeated invocations. For example, following lists the results for building WrapITK with & without ccache (for 8-core Quad-Core Intel Xeon with Mac OSX Snow Leopard): | All the subsequent builds of WrapITK would use ccache and we can see a speedup of as much as 40 minutes for repeated invocations. For example, following lists the results for building WrapITK with & without ccache (for 8-core Quad-Core Intel Xeon with Mac OSX Snow Leopard): | ||
Line 58: | Line 71: | ||
max cache size 1.0 Gbytes | max cache size 1.0 Gbytes | ||
=== 12-cores (24 physical threads) === | |||
to be updated by AG or ashish. |
Revision as of 14:04, 1 May 2011
Using CCache for WrapITK
Installation of CCACHE
Using CCache[1] to build WrapITK can improve the build times considerably. For all unix systems (including MacOS X), ccache is enabled by creating symbolic links to the g++/gcc compiler:
ln -s /usr/bin/ccache /usr/local/bin/g++ ln -s /usr/bin/ccache /usr/local/bin/gcc
Alternatively, on MacOS X, Install ccache with Macports. The Macports installation would automatically create the symbolic links to g++/gcc compilers and put them in the directory /opt/local/libexec/ccache. The only remaining thing then is to add this directory to the PATH variable (alternatively you can add it to the .bash_profile or .profile in home directory)
$ sudo port install ccache $ export PATH= /opt/local/libexec/ccache:$PATH
ITK Specific configuration Flags
Note that ccache is installed outside of ITK, and will benefit any software project that uses gcc.
The other parts of WrapITK which can benefit from ccache are gccxml and swig. To enable ccache usage with them, just turn ON the option WRAP_ITK_USE_CCACHE and everything is gonna be set up for you automatically.
Example of speed-ups
8-cores (8 physical threads)
All the subsequent builds of WrapITK would use ccache and we can see a speedup of as much as 40 minutes for repeated invocations. For example, following lists the results for building WrapITK with & without ccache (for 8-core Quad-Core Intel Xeon with Mac OSX Snow Leopard):
Without cache:
time make -j 12 real 46m58.307s user 41m49.829s sys 4m22.580s
Repeated invocation:
make clean time make -j 12 real 44m3.667s user 39m29.606s sys 3m49.870s
On enabling ccache, the build times reduce considerably as noted below:
With cache:
time make -j 12 real 45m44.488s user 40m3.881s sys 4m25.381s ccache -s cache hit (direct) 19 cache hit (preprocessed) 1 cache miss 2658 called for link 396 compile failed 22 preprocessor error 12 no input file 21 files in cache 5307 cache size 593.9 Mbytes max cache size 1.0 Gbytes
Repeated invocation:
make clean time make -j 12
real 2m48.801s user 1m28.743s sys 0m40.986s
ccache -s cache hit (direct) 2435 cache hit (preprocessed) 0 cache miss 0 called for link 152 files in cache 5307 cache size 593.9 Mbytes max cache size 1.0 Gbytes
12-cores (24 physical threads)
to be updated by AG or ashish.