cmake-native: Set --parallel for configure

The following changes since commit 14c291e1fb6324da46885b69fbd7f01b3c6b053e:

gcc-runtime: Add missing libc dependency (2018-12-18 16:19:30 +0000)

are available in the git repository at:

git://git.openembedded.org/openembedded-core-contrib rbt/cmake
http://cgit.openembedded.org/openembedded-core-contrib/log/?h=rbt/cmake

Robert Yang (1):
cmake-native: Set --parallel for configure

 meta/recipes-devtools/cmake/cmake-native_3.12.2.bb | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

2.7.4

This can save do_configure's time from 330s to 60s on my host.

---
 meta/recipes-devtools/cmake/cmake-native_3.12.2.bb | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

index 9a1390f..71b3326 100644
@@ -27,7 +27,9 @@ CMAKE_EXTRACONF = "\
 "
 
 do_configure () {
-	${S}/configure --verbose --prefix=${prefix} -- ${CMAKE_EXTRACONF}
+	${S}/configure --verbose --prefix=${prefix} \
+		--parallel=${@d.getVar('PARALLEL_MAKE').replace('-j ', '')} \
+		-- ${CMAKE_EXTRACONF}
 }
 
 do_compile() {

2.7.4

See oe.utils.parallel_make_argument()

Ross

Thanks, updated in the repo:

commit 19ac4e33e66353581191e7dfd866f6c0f73eafa6

Author: Robert Yang <[email protected]>
Date:   Tue Dec 18 17:48:22 2018 +0800

cmake-native: Set --parallel for configure

This can save do_configure's time from 330s to 60s on my host.

index 9a1390f..3086c82 100644
@@ -27,7 +27,9 @@ CMAKE_EXTRACONF = "\
 "
 
 do_configure () {
-       ${S}/configure --verbose --prefix=${prefix} -- ${CMAKE_EXTRACONF}
+       ${S}/configure --verbose --prefix=${prefix} \
+               --parallel="${@oe.utils.parallel_make_argument(d, '%d')}" \
+               -- ${CMAKE_EXTRACONF}
 }
 
 do_compile() {

// Robert

Tweaked when merging to mut to be:

+ ${S}/configure --verbose --prefix=${prefix} \
+ ${@oe.utils.parallel_make_argument(d, '--parallel=%d')} \
+ -- ${CMAKE_EXTRACONF}

I see similar wins here, walltime for cmake-native's build goes from five minutes to one minute.

(!!!)

Ross

For anyone playing along at home, this is what I did:

Cleaned tmp, applied patch to the mut branch.

$ git checkout HEAD^; bitbake cmake-native -C unpack
$ git checkout mut ; bitbake cmake-native -C unpack

Force a full build of cmake-native for both commits.

Then use buildstats-diff to compare the runs:

$ buildstats-diff /data/poky-tmp/master/buildstats/*
  PKG             TASK          ABSDIFF  RELDIFF  CPUTIME1 -> CPUTIME2
  cmake-native    do_configure    29.1s   +11.0%    265.7s -> 294.8s

Cumulative cputime:
18.0s +2.2% 13:41.4 (821.4s) -> 13:59.3 (839.3s)

Oh, that's odd: it took *more* time to run. But the output from my bitbake clearly showed bitbake taking five minutes before and one minute after.

Oh, *walltime* not *cputime*.

$ buildstats-diff /data/poky-tmp/master/buildstats/* --diff-attr walltime
  PKG             TASK          ABSDIFF  RELDIFF  WALLTIME1 -> WALLTIME2
  cmake-native    do_configure  -214.6s   -82.4%     260.4s -> 45.8s

Cumulative walltime:
-218.4s -72.3% 05:02.1 (302.1s) -> 01:23.7 (83.7s)

cputime is generally a more useful metric during a build, but for isolated tests like this walltime is also relevant. This is demonstrating that the same work is happening, but in parallel.

Ross