summaryrefslogtreecommitdiff
path: root/INSTALL
diff options
context:
space:
mode:
authorTheodore Ts'o <tytso@mit.edu>2012-02-06 06:00:17 (GMT)
committerJunio C Hamano <gitster@pobox.com>2012-02-06 08:15:12 (GMT)
commitf2d713fc3e8e0b7be89584f04b421808aa99c403 (patch)
treef648f664829a3cb9351a14798227907a9b83e6c8 /INSTALL
parent828ea97de486c1693d6e4f2c7347acb50235a85d (diff)
downloadgit-f2d713fc3e8e0b7be89584f04b421808aa99c403.zip
git-f2d713fc3e8e0b7be89584f04b421808aa99c403.tar.gz
git-f2d713fc3e8e0b7be89584f04b421808aa99c403.tar.bz2
Fix build problems related to profile-directed optimization
There was a number of problems I ran into when trying the profile-directed optimizations added by Andi Kleen in git commit 7ddc2710b9. (This was using gcc 4.4 found on many enterprise distros.) 1) The -fprofile-generate and -fprofile-use commands are incompatible with ccache; the code ends up looking in the wrong place for the gcda files based on the ccache object names. 2) If the makefile notices that CFLAGS are different, it will rebuild all of the binaries. Hence the recipe originally specified by the INSTALL file ("make profile-all" followed by "make install") doesn't work. It will appear to work, but the binaries will end up getting built with no optimization. This patch fixes this by using an explicit set of options passed via the PROFILE variable then using this to directly manipulate CFLAGS and EXTLIBS. The developer can run "make PROFILE=BUILD all ; sudo make PROFILE=BUILD install" automatically run a two-pass build with the test suite run in between as the sample workload for the purpose of recording profiling information to do the profile-directed optimization. Alternatively, the profiling version of binaries can be built using: make PROFILE=GEN PROFILE_DIR=/var/cache/profile all make PROFILE=GEN install and then after git has been used for a while, the optimized version of the binary can be built as follows: make PROFILE=USE PROFILE_DIR=/var/cache/profile all make PROFILE=USE install Signed-off-by: "Theodore Ts'o" <tytso@mit.edu> Cc: Andi Kleen <ak@linux.intel.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Diffstat (limited to 'INSTALL')
-rw-r--r--INSTALL17
1 files changed, 13 insertions, 4 deletions
diff --git a/INSTALL b/INSTALL
index 6fa83fe..58b2b86 100644
--- a/INSTALL
+++ b/INSTALL
@@ -28,16 +28,25 @@ set up install paths (via config.mak.autogen), so you can write instead
If you're willing to trade off (much) longer build time for a later
faster git you can also do a profile feedback build with
- $ make profile-all
- # make prefix=... install
+ $ make prefix=/usr PROFILE=BUILD all
+ # make prefix=/usr PROFILE=BUILD install
This will run the complete test suite as training workload and then
rebuild git with the generated profile feedback. This results in a git
which is a few percent faster on CPU intensive workloads. This
may be a good tradeoff for distribution packagers.
-Note that the profile feedback build stage currently generates
-a lot of additional compiler warnings.
+Or if you just want to install a profile-optimized version of git into
+your home directory, you could run:
+
+ $ make PROFILE=BUILD install
+
+As a caveat: a profile-optimized build takes a *lot* longer since the
+git tree must be built twice, and in order for the profiling
+measurements to work properly, ccache must be disabled and the test
+suite has to be run using only a single CPU. In addition, the profile
+feedback build stage currently generates a lot of additional compiler
+warnings.
Issues of note: