diff options
| author | Brett Cannon <brett@python.org> | 2015-09-18 15:09:42 -0700 |
|---|---|---|
| committer | Brett Cannon <brett@python.org> | 2015-09-18 15:09:42 -0700 |
| commit | 4ff151a72a3331c514924a5e0d33747a91413fbd (patch) | |
| tree | 6a911a48917de5a71557cb5a187281f6e232a1b5 /README | |
| parent | 62f0ffa86136f06c84cee139f29aa3b9ad0385a3 (diff) | |
| download | cpython-git-4ff151a72a3331c514924a5e0d33747a91413fbd.tar.gz | |
Issue #24915: Make PGO builds support Clang and use the test suite for
profile data.
Thanks to Alecsandru Patrascu of Intel for the initial patch.
Diffstat (limited to 'README')
| -rw-r--r-- | README | 29 |
1 files changed, 29 insertions, 0 deletions
@@ -173,6 +173,11 @@ rebuilt. In this case, you may have to run make again to correctly build your desired target. The interpreter executable is built in the top level directory. +If you need an optimized version of Python, you type "make profile-opt" +in the top level directory. This will rebuild the interpreter executable +using Profile Guided Optimization (PGO). For more details, see the +section below. + Once you have built a Python interpreter, see the subsections below on testing and installation. If you run into trouble, see the next section. @@ -185,6 +190,30 @@ guidance of the setup.py script, which is run by Make after the interpreter has been built. +Profile Guided Optimization +--------------------------- + +PGO takes advantage of recent versions of the GCC or Clang compilers. +If ran, the "profile-opt" rule will do several steps. + +First, the entire Python directory is cleaned of temporary files that +may resulted in a previous compilation. + +Then, an instrumented version of the interpreter is built, using suitable +compiler flags for each flavour. Note that this is just an intermediary +step and the binary resulted after this step is not good for real life +workloads, as it has profiling instructions embedded inside. + +After this instrumented version of the interpreter is built, the Makefile +will automatically run a training workload. This is necessary in order to +profile the interpreter execution. Note also that any output, both stdout +and stderr, that may appear at this step is supressed. + +Finally, the last step is to rebuild the interpreter, using the information +collected in the previous one. The end result will be a the Python binary +that is optimized and suitable for distribution or production installation. + + Troubleshooting --------------- |