2003-11-14 23:09:38 +00:00
|
|
|
# $FreeBSD$
|
|
|
|
|
|
|
|
# Part of a unified Makefile for building kernels. This part includes all
|
|
|
|
# the definitions that need to be after all the % directives except %RULES
|
|
|
|
# and ones that act like they are part of %RULES.
|
2001-11-02 21:34:20 +00:00
|
|
|
#
|
2002-01-05 06:21:06 +00:00
|
|
|
# Most make variables should not be defined in this file. Instead, they
|
|
|
|
# should be defined in the kern.pre.mk so that port makefiles can
|
2002-01-06 20:33:55 +00:00
|
|
|
# override or augment them.
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2003-05-31 21:16:12 +00:00
|
|
|
# In case the config had a makeoptions DESTDIR...
|
|
|
|
.if defined(DESTDIR)
|
|
|
|
MKMODULESENV+= DESTDIR="${DESTDIR}"
|
|
|
|
.endif
|
2009-05-04 20:25:56 +00:00
|
|
|
SYSDIR?= ${S:C;^[^/];${.CURDIR}/&;}
|
|
|
|
MKMODULESENV+= KERNBUILDDIR="${.CURDIR}" SYSDIR="${SYSDIR}"
|
2003-05-31 21:16:12 +00:00
|
|
|
|
2010-03-25 13:47:21 +00:00
|
|
|
.if defined(CONF_CFLAGS)
|
|
|
|
MKMODULESENV+= CONF_CFLAGS="${CONF_CFLAGS}"
|
|
|
|
.endif
|
|
|
|
|
2010-04-02 06:55:31 +00:00
|
|
|
.if defined(WITH_CTF)
|
|
|
|
MKMODULESENV+= WITH_CTF="${WITH_CTF}"
|
|
|
|
.endif
|
|
|
|
|
2015-09-24 16:55:22 +00:00
|
|
|
# Allow overriding the kernel debug directory, so kernel and user debug may be
|
|
|
|
# installed in different directories. Setting it to "" restores the historical
|
|
|
|
# behavior of installing debug files in the kernel directory.
|
|
|
|
KERN_DEBUGDIR?= ${DEBUGDIR}
|
|
|
|
|
2003-02-21 11:02:49 +00:00
|
|
|
.MAIN: all
|
|
|
|
|
2003-03-03 09:35:44 +00:00
|
|
|
.for target in all clean cleandepend cleandir clobber depend install \
|
|
|
|
obj reinstall tags
|
2003-02-21 11:02:49 +00:00
|
|
|
${target}: kernel-${target}
|
|
|
|
.if !defined(MODULES_WITH_WORLD) && !defined(NO_MODULES) && exists($S/modules)
|
|
|
|
${target}: modules-${target}
|
|
|
|
modules-${target}:
|
2003-03-03 09:35:44 +00:00
|
|
|
cd $S/modules; ${MKMODULESENV} ${MAKE} \
|
|
|
|
${target:S/^reinstall$/install/:S/^clobber$/cleandir/}
|
2003-02-21 11:02:49 +00:00
|
|
|
.endif
|
|
|
|
.endfor
|
|
|
|
|
2012-06-14 19:30:31 +00:00
|
|
|
# Handle ports (as defined by the user) that build kernel modules
|
2005-06-09 05:38:18 +00:00
|
|
|
.if !defined(NO_MODULES) && defined(PORTS_MODULES)
|
2012-06-14 19:30:31 +00:00
|
|
|
#
|
|
|
|
# The ports tree needs some environment variables defined to match the new kernel
|
|
|
|
#
|
|
|
|
# Ports search for some dependencies in PATH, so add the location of the installed files
|
|
|
|
LOCALBASE?= /usr/local
|
|
|
|
# SRC_BASE is how the ports tree refers to the location of the base source files
|
|
|
|
.if !defined(SRC_BASE)
|
2016-02-16 02:14:12 +00:00
|
|
|
SRC_BASE= ${SYSDIR:H:tA}
|
2012-06-14 19:30:31 +00:00
|
|
|
.endif
|
|
|
|
# OSVERSION is used by some ports to determine build options
|
|
|
|
.if !defined(OSRELDATE)
|
|
|
|
# Definition copied from src/Makefile.inc1
|
|
|
|
OSRELDATE!= awk '/^\#define[[:space:]]*__FreeBSD_version/ { print $$3 }' \
|
|
|
|
${MAKEOBJDIRPREFIX}${SRC_BASE}/include/osreldate.h
|
|
|
|
.endif
|
|
|
|
# Keep the related ports builds in the obj directory so that they are only rebuilt once per kernel build
|
|
|
|
WRKDIRPREFIX?= ${MAKEOBJDIRPREFIX}${SRC_BASE}/sys/${KERNCONF}
|
|
|
|
PORTSMODULESENV=\
|
|
|
|
PATH=${PATH}:${LOCALBASE}/bin:${LOCALBASE}/sbin \
|
|
|
|
SRC_BASE=${SRC_BASE} \
|
|
|
|
OSVERSION=${OSRELDATE} \
|
|
|
|
WRKDIRPREFIX=${WRKDIRPREFIX}
|
2012-07-02 22:14:58 +00:00
|
|
|
|
|
|
|
# The WRKDIR needs to be cleaned before building, and trying to change the target
|
|
|
|
# with a :C pattern below results in install -> instclean
|
|
|
|
all:
|
|
|
|
.for __i in ${PORTS_MODULES}
|
|
|
|
@${ECHO} "===> Ports module ${__i} (all)"
|
|
|
|
cd $${PORTSDIR:-/usr/ports}/${__i}; ${PORTSMODULESENV} ${MAKE} -B clean all
|
|
|
|
.endfor
|
|
|
|
|
|
|
|
.for __target in install reinstall clean
|
2005-02-25 05:34:45 +00:00
|
|
|
${__target}: ports-${__target}
|
2005-02-25 02:25:38 +00:00
|
|
|
ports-${__target}:
|
2004-11-11 23:58:14 +00:00
|
|
|
.for __i in ${PORTS_MODULES}
|
2012-07-02 22:14:58 +00:00
|
|
|
@${ECHO} "===> Ports module ${__i} (${__target})"
|
|
|
|
cd $${PORTSDIR:-/usr/ports}/${__i}; ${PORTSMODULESENV} ${MAKE} -B ${__target:C/install/deinstall reinstall/:C/reinstall/deinstall reinstall/}
|
2004-11-11 23:58:14 +00:00
|
|
|
.endfor
|
|
|
|
.endfor
|
|
|
|
.endif
|
|
|
|
|
2003-02-21 11:02:49 +00:00
|
|
|
.ORDER: kernel-install modules-install
|
|
|
|
|
2007-03-23 21:55:59 +00:00
|
|
|
kernel-all: ${KERNEL_KO} ${KERNEL_EXTRA}
|
2003-02-21 11:02:49 +00:00
|
|
|
|
2006-02-07 13:37:26 +00:00
|
|
|
kernel-cleandir: kernel-clean kernel-cleandepend
|
2003-03-03 09:35:44 +00:00
|
|
|
|
|
|
|
kernel-clobber:
|
2003-03-02 21:25:00 +00:00
|
|
|
find . -maxdepth 1 ! -type d ! -name version -delete
|
|
|
|
|
|
|
|
kernel-obj:
|
2003-02-21 11:02:49 +00:00
|
|
|
|
|
|
|
.if !defined(MODULES_WITH_WORLD) && !defined(NO_MODULES) && exists($S/modules)
|
|
|
|
modules: modules-all
|
|
|
|
|
|
|
|
.if !defined(NO_MODULES_OBJ)
|
|
|
|
modules-all modules-depend: modules-obj
|
|
|
|
.endif
|
|
|
|
.endif
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2005-10-24 23:54:39 +00:00
|
|
|
.if !defined(DEBUG)
|
|
|
|
FULLKERNEL= ${KERNEL_KO}
|
|
|
|
.else
|
2015-09-24 16:55:22 +00:00
|
|
|
FULLKERNEL= ${KERNEL_KO}.full
|
|
|
|
${KERNEL_KO}: ${FULLKERNEL} ${KERNEL_KO}.debug
|
|
|
|
${OBJCOPY} --strip-debug --add-gnu-debuglink=${KERNEL_KO}.debug \
|
2005-10-27 14:24:45 +00:00
|
|
|
${FULLKERNEL} ${.TARGET}
|
2015-09-24 16:55:22 +00:00
|
|
|
${KERNEL_KO}.debug: ${FULLKERNEL}
|
2005-10-27 14:24:45 +00:00
|
|
|
${OBJCOPY} --only-keep-debug ${FULLKERNEL} ${.TARGET}
|
2005-10-25 09:05:07 +00:00
|
|
|
install.debug reinstall.debug: gdbinit
|
|
|
|
cd ${.CURDIR}; ${MAKE} ${.TARGET:R}
|
2003-06-12 08:09:48 +00:00
|
|
|
|
2003-12-29 11:28:21 +00:00
|
|
|
# Install gdbinit files for kernel debugging.
|
2003-06-12 08:09:48 +00:00
|
|
|
gdbinit:
|
2003-12-29 11:28:21 +00:00
|
|
|
grep -v '# XXX' ${S}/../tools/debugscripts/dot.gdbinit | \
|
|
|
|
sed "s:MODPATH:${.OBJDIR}/modules:" > .gdbinit
|
2004-09-23 08:34:50 +00:00
|
|
|
cp ${S}/../tools/debugscripts/gdbinit.kernel ${.CURDIR}
|
2010-08-23 22:24:11 +00:00
|
|
|
.if exists(${S}/../tools/debugscripts/gdbinit.${MACHINE_CPUARCH})
|
|
|
|
cp ${S}/../tools/debugscripts/gdbinit.${MACHINE_CPUARCH} \
|
2003-12-29 11:28:21 +00:00
|
|
|
${.CURDIR}/gdbinit.machine
|
2001-11-02 21:34:20 +00:00
|
|
|
.endif
|
2004-01-21 08:23:46 +00:00
|
|
|
.endif
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2015-08-13 15:16:34 +00:00
|
|
|
${FULLKERNEL}: ${SYSTEM_DEP} vers.o
|
2001-11-02 21:34:20 +00:00
|
|
|
@rm -f ${.TARGET}
|
|
|
|
@echo linking ${.TARGET}
|
|
|
|
${SYSTEM_LD}
|
Fix MFS builds when both MD_ROOT_SIZE and MFS_IMAGE are specified
MD_ROOT_SIZE and embed_mfs.sh were basically retired as part of
https://reviews.freebsd.org/D2903 .
However, when building a kernel with 'options MD_ROOT_SIZE' specified, this
results in a non-working MFS, as within sys/dev/md/md.c we fall within the
wrong # ifdef.
This patch implements the following:
* Allow kernels to be built without the MD_ROOT_SIZE option, which results
in a kernel built as per D2903.
* Allow kernels to be built with the MD_ROOT_SIZE option, which results
in a kernel built similarly to the pre-D2903 way, with the following
differences:
* The MFS is now put in a separate section within the kernel (oldmfs,
so it differs from the mfs section introduced by D2903).
* embed_mfs.sh is changed, so it looks up the oldmfs section within the
kernel, gets its size and offset, sees if the MFS will fit within the
allocated oldmfs section and only if all is well does a dd of the MFS
image into the kernel.
Submitted by: Stanislav Galabov <sgalabov@gmail.com>
Reviewed by: brooks, imp
Differential Revision: https://reviews.freebsd.org/D5093
2016-02-02 07:02:51 +00:00
|
|
|
.if !empty(MD_ROOT_SIZE_CONFIGURED) && defined(MFS_IMAGE)
|
|
|
|
@sh ${S}/tools/embed_mfs.sh ${.TARGET} ${MFS_IMAGE}
|
|
|
|
.endif
|
2011-11-30 18:11:49 +00:00
|
|
|
.if ${MK_CTF} != "no"
|
2014-08-12 23:48:37 +00:00
|
|
|
@echo ${CTFMERGE} ${CTFFLAGS} -o ${.TARGET} ...
|
|
|
|
@${CTFMERGE} ${CTFFLAGS} -o ${.TARGET} ${SYSTEM_OBJS} vers.o
|
2011-11-29 08:38:47 +00:00
|
|
|
.endif
|
2004-02-13 13:21:41 +00:00
|
|
|
.if !defined(DEBUG)
|
|
|
|
${OBJCOPY} --strip-debug ${.TARGET}
|
|
|
|
.endif
|
2001-11-02 21:34:20 +00:00
|
|
|
${SYSTEM_LD_TAIL}
|
|
|
|
|
2003-06-22 17:57:56 +00:00
|
|
|
.if !exists(${.OBJDIR}/.depend)
|
2004-06-26 06:02:06 +00:00
|
|
|
${SYSTEM_OBJS}: assym.s vnode_if.h ${BEFORE_DEPEND:M*.h} ${MFILES:T:S/.m$/.h/}
|
2001-11-02 21:34:20 +00:00
|
|
|
.endif
|
|
|
|
|
2003-06-14 17:28:13 +00:00
|
|
|
LNFILES= ${CFILES:T:S/.c$/.ln/}
|
|
|
|
|
2001-11-02 21:34:20 +00:00
|
|
|
.for mfile in ${MFILES}
|
2002-08-19 01:00:37 +00:00
|
|
|
# XXX the low quality .m.o rules gnerated by config are normally used
|
|
|
|
# instead of the .m.c rules here.
|
|
|
|
${mfile:T:S/.m$/.c/}: ${mfile}
|
|
|
|
${AWK} -f $S/tools/makeobjops.awk ${mfile} -c
|
2001-11-02 21:34:20 +00:00
|
|
|
${mfile:T:S/.m$/.h/}: ${mfile}
|
2002-05-01 03:28:14 +00:00
|
|
|
${AWK} -f $S/tools/makeobjops.awk ${mfile} -h
|
2001-11-02 21:34:20 +00:00
|
|
|
.endfor
|
|
|
|
|
|
|
|
kernel-clean:
|
|
|
|
rm -f *.o *.so *.So *.ko *.s eddep errs \
|
2015-09-24 16:55:22 +00:00
|
|
|
${FULLKERNEL} ${KERNEL_KO} ${KERNEL_KO}.debug \
|
2012-10-25 04:55:04 +00:00
|
|
|
linterrs tags vers.c \
|
2005-04-22 19:58:22 +00:00
|
|
|
vnode_if.c vnode_if.h vnode_if_newproto.h vnode_if_typedef.h \
|
2003-12-29 11:28:21 +00:00
|
|
|
${MFILES:T:S/.m$/.c/} ${MFILES:T:S/.m$/.h/} \
|
2006-10-13 22:28:14 +00:00
|
|
|
${CLEAN}
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2003-06-14 17:28:13 +00:00
|
|
|
lint: ${LNFILES}
|
2003-12-29 11:28:21 +00:00
|
|
|
${LINT} ${LINTKERNFLAGS} ${CFLAGS:M-[DILU]*} ${.ALLSRC} 2>&1 | \
|
|
|
|
tee -a linterrs
|
2002-03-17 10:05:57 +00:00
|
|
|
|
2001-11-02 21:34:20 +00:00
|
|
|
# This is a hack. BFD "optimizes" away dynamic mode if there are no
|
|
|
|
# dynamic references. We could probably do a '-Bforcedynamic' mode like
|
|
|
|
# in the a.out ld. For now, this works.
|
|
|
|
HACK_EXTRA_FLAGS?= -shared
|
|
|
|
hack.So: Makefile
|
2006-08-14 13:28:53 +00:00
|
|
|
:> hack.c
|
2003-12-29 11:34:33 +00:00
|
|
|
${CC} ${HACK_EXTRA_FLAGS} -nostdlib hack.c -o hack.So
|
2001-11-02 21:34:20 +00:00
|
|
|
rm -f hack.c
|
|
|
|
|
2003-12-29 11:28:21 +00:00
|
|
|
# This rule stops ./assym.s in .depend from causing problems.
|
2001-11-02 21:34:20 +00:00
|
|
|
./assym.s: assym.s
|
|
|
|
|
|
|
|
assym.s: $S/kern/genassym.sh genassym.o
|
2015-08-14 22:58:32 +00:00
|
|
|
NM='${NM}' NMFLAGS='${NMFLAGS}' sh $S/kern/genassym.sh genassym.o > ${.TARGET}
|
2001-11-02 21:34:20 +00:00
|
|
|
|
|
|
|
genassym.o: $S/$M/$M/genassym.c
|
2003-12-30 07:54:39 +00:00
|
|
|
${CC} -c ${CFLAGS:N-fno-common} $S/$M/$M/genassym.c
|
2001-11-02 21:34:20 +00:00
|
|
|
|
|
|
|
${SYSTEM_OBJS} genassym.o vers.o: opt_global.h
|
|
|
|
|
2016-01-27 14:46:30 +00:00
|
|
|
# Normal files first
|
|
|
|
CFILES_NORMAL= ${CFILES:N*/cddl/*:N*fs/nfsclient/nfs_clkdtrace*:N*/compat/linuxkpi/common/*:N*/ofed/*:N*/dev/mlx5/*}
|
|
|
|
SFILES_NORMAL= ${SFILES:N*/cddl/*}
|
|
|
|
|
2015-06-10 15:53:39 +00:00
|
|
|
# We have "special" -I include paths for zfs/dtrace files in 'depend'.
|
|
|
|
CFILES_CDDL= ${CFILES:M*/cddl/*}
|
|
|
|
SFILES_CDDL= ${SFILES:M*/cddl/*}
|
2012-03-27 21:23:56 +00:00
|
|
|
|
2016-01-27 14:46:30 +00:00
|
|
|
# We have "special" -I include paths for LinuxKPI.
|
|
|
|
CFILES_LINUXKPI=${CFILES:M*/compat/linuxkpi/common/*}
|
|
|
|
|
|
|
|
# We have "special" -I include paths for OFED.
|
|
|
|
CFILES_OFED=${CFILES:M*/ofed/*}
|
|
|
|
|
|
|
|
# We have "special" -I include paths for MLX5.
|
|
|
|
CFILES_MLX5=${CFILES:M*/dev/mlx5/*}
|
|
|
|
|
2005-04-22 17:36:25 +00:00
|
|
|
kernel-depend: .depend
|
2002-08-18 20:41:19 +00:00
|
|
|
# The argument list can be very long, so use make -V and xargs to
|
2002-02-23 01:49:13 +00:00
|
|
|
# pass it to mkdep.
|
Add built-in ccache build support via WITH_CCACHE_BUILD option.
ccache is mostly beneficial for frequent builds where -DNO_CLEAN is not
used to achieve a safe pseudo-incremental build. This is explained in
more detail upstream [1] [2]. It incurs about a 20%-28% hit to populate the
cache, but with a full cache saves 30-50% in build times. When combined with
the WITH_FAST_DEPEND feature it saves up to 65% since ccache does cache the
resulting dependency file, which it does not do when using mkdep(1)/'CC
-E'. Stats are provided at the end of this message.
This removes the need to modify /etc/make.conf with the CC:= and CXX:=
lines which conflicted with external compiler support [3] (causing the
bootstrap compiler to not be built which lead to obscure failures [4]),
incorrectly invoked ccache in various stages, required CCACHE_CPP2 to avoid
Clang errors with parenthesis, and did not work with META_MODE.
The option name was picked to match the existing option in ports. This
feature is available for both in-src and out-of-src builds that use
/usr/share/mk.
Linking, assembly compiles, and pre-processing avoid using ccache since it is
only overhead. ccache does nothing special in these modes, although there is
no harm in calling it for them.
CCACHE_COMPILERCHECK is set to 'content' when using the in-tree bootstrap
compiler to hash the content of the compiler binary to determine if it
should be a cache miss. For external compilers the 'mtime' option is used
as it is more efficient and likely to be correct. Future work may optimize the
'content' check using the same checks as whether a bootstrap compiler is needed
to be built.
The CCACHE_CPP2 pessimization is currently default in our devel/ccache
port due to Clang requiring it. Clang's -Wparentheses-equality,
-Wtautological-compare, and -Wself-assign warnings do not mix well with
compiling already-pre-processed code that may have expanded macros that
trigger the warnings. GCC has so far not had this issue so it is allowed to
disable the CCACHE_CPP2 default in our port.
Sharing a cache between multiple checkouts, or systems, is explained in
the ccache manual. Sharing a cache over NFS would likely not be worth
it, but syncing cache directories between systems may be useful for an
organization. There is also a memcached backend available [5]. Due to using
an object directory outside of the source directory though you will need to
ensure that both are in the same prefix and all users use the same layout. A
possible working layout is as follows:
Source: /some/prefix/src1
Source: /some/prefix/src2
Source: /some/prefix/src3
Objdir: /some/prefix/obj
Environment: CCACHE_BASEDIR='${SRCTOP:H}' MAKEOBJDIRPREFIX='${SRCTOP:H}/obj'
This will use src*/../obj as the MAKEOBJDIRPREFIX and tells ccache to replace
all absolute paths to be relative. Using something like this is required due
to -I and -o flags containing both SRC and OBJDIR absolute paths that ccache
adds into its hash for the object without CCACHE_BASEDIR.
distcc can be hooked into by setting CCACHE_PREFIX=/usr/local/bin/distcc.
I have not personally tested this and assume it will not mix well with
using the bootstrap compiler.
The cache from buildworld can be reused in a subdir by first running
'make buildenv' (from r290424).
Note that the cache is currently different depending on whether -j is
used or not due to ccache enabling -fdiagnostics-color automatically if
stderr is a TTY, which bmake only does if not using -j.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_CCACHE_BUILD=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files and ccache objects.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
The WITH_FAST_DEPEND feature was used for comparison here as well to show
the dramatic time savings with a full cache.
buildworld:
x buildworld-before
+ buildworld-ccache-empty
* buildworld-ccache-full
% buildworld-ccache-full-fastdep
# buildworld-fastdep
+-------------------------------------------------------------------------------+
|% * # +|
|% * # +|
|% * # xxx +|
| |A |
| A|
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 4519 4525.04 4520.73 4521.59 3.1104823
Difference at 95.0% confidence
758.027 +/- 43.4565
20.1412% +/- 1.15466%
(Student's t, pooled s = 19.1726)
* 3 1823.08 1827.2 1825.62 1825.3 2.0785572
Difference at 95.0% confidence
-1938.26 +/- 43.298
-51.5007% +/- 1.15045%
(Student's t, pooled s = 19.1026)
% 3 1266.96 1279.37 1270.47 1272.2667 6.3971113
Difference at 95.0% confidence
-2491.3 +/- 44.3704
-66.1952% +/- 1.17895%
(Student's t, pooled s = 19.5758)
# 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-ccache-empty
* buildkernel-ccache-empty-fastdep
% buildkernel-ccache-full
# buildkernel-ccache-full-fastdep
@ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|# @ % * |
|# @ % * x + |
|# @ % * xx ++|
| MA |
| MA|
| A |
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 727.97 731.91 728.06 729.31333 2.2492295
Difference at 95.0% confidence
156.88 +/- 4.17129
27.4058% +/- 0.728695%
(Student's t, pooled s = 1.84034)
* 3 527.1 528.29 528.08 527.82333 0.63516402
Difference at 95.0% confidence
-44.61 +/- 2.33254
-7.79305% +/- 0.407478%
(Student's t, pooled s = 1.02909)
% 3 400.4 401.05 400.62 400.69 0.3306055
Difference at 95.0% confidence
-171.743 +/- 2.16453
-30.0023% +/- 0.378128%
(Student's t, pooled s = 0.954969)
# 3 201.94 203.34 202.28 202.52 0.73020545
Difference at 95.0% confidence
-369.913 +/- 2.40293
-64.6212% +/- 0.419774%
(Student's t, pooled s = 1.06015)
@ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
[1] https://ccache.samba.org/performance.html
[2] http://www.mail-archive.com/ccache@lists.samba.org/msg00576.html
[3] https://reviews.freebsd.org/D3484
[5] https://github.com/jrosdahl/ccache/pull/30
PR: 182944 [4]
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
Relnotes: yes
2015-11-08 00:50:18 +00:00
|
|
|
_MKDEPCC:= ${CC:N${CCACHE_BIN}}
|
2005-11-30 18:15:06 +00:00
|
|
|
SRCS= assym.s vnode_if.h ${BEFORE_DEPEND} ${CFILES} \
|
|
|
|
${SYSTEM_CFILES} ${GEN_CFILES} ${SFILES} \
|
|
|
|
${MFILES:T:S/.m$/.h/}
|
Add a FAST_DEPEND option, off by default, which speeds up the build significantly.
This speeds up buildworld by 16% on my system and buildkernel by 35%.
Rather than calling mkdep(1), which is just a wrapper around 'cc -E',
use the modern -MD -MT -MF flags to gather and generate dependencies during
compilation. This flag was introduced in GCC "a long time ago", in GCC 3.0,
and is also supported by Clang. (It appears that ICC also supports this but I
do not have access to test it). This avoids running the preprocessor *twice*
for every build, in both 'make depend' and 'make all'. This is especially
noticeable when using ccache since it does not cache preprocessor results from
mkdep(1) / 'cc -E', but still speeds up compilation with the -MD flags.
For 'make depend' a tree-walk is still done to ensure that all DPSRCS
are generated when expected, and that beforedepend/afterdepend and
_EXTRADEPEND are all still respected. In time this may change but for now
I've been conservative. The time for a tree-walk with -j combined with
SUBDIR_PARALLEL is not significant. For example, it takes about 9 seconds
with -j15 to walk all of src/ for 'make depend' now on my system.
A .depend file is still generated with the various rules that apply to
the final target, or custom rules. Otherwise there are now
per-built-object-file .depend files, such as .depend.filename.o. These
are included directly by make rather than populating .depend with a loop
and .depend lines, which only added overhead to the now almost-NOP 'make
depend' phase.
Before this I experimented with having mkdep(1) called in parallel per-file.
While this improved the kernel and lib/libc 'make depend' phase, it resulted
in slower build times overall.
The -M flags are removed from CFLAGS when linking since they have no effect.
Enabling this by default, for src or out-of-src, can be done once more testing
has been done, such as a ports exp-run, and with more compilers.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_FAST_DEPEND=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
buildworld:
x buildworld-before
+ buildworld-fastdep
+-------------------------------------------------------------------------------+
|+ |
|+ |
|+ xx x|
| |_MA___||
|A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|+ x |
|++ xx|
| A||
|A| |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
Sponsored by: EMC / Isilon Storage Division
MFC after: 3 weeks
Relnotes: yes
2015-11-06 04:45:29 +00:00
|
|
|
DEPENDFILES= .depend
|
2016-01-27 01:24:11 +00:00
|
|
|
.if ${MK_FAST_DEPEND} == "yes" && \
|
|
|
|
(${.MAKE.MODE:Unormal:Mmeta} == "" || ${.MAKE.MODE:Unormal:Mnofilemon} != "")
|
Add a FAST_DEPEND option, off by default, which speeds up the build significantly.
This speeds up buildworld by 16% on my system and buildkernel by 35%.
Rather than calling mkdep(1), which is just a wrapper around 'cc -E',
use the modern -MD -MT -MF flags to gather and generate dependencies during
compilation. This flag was introduced in GCC "a long time ago", in GCC 3.0,
and is also supported by Clang. (It appears that ICC also supports this but I
do not have access to test it). This avoids running the preprocessor *twice*
for every build, in both 'make depend' and 'make all'. This is especially
noticeable when using ccache since it does not cache preprocessor results from
mkdep(1) / 'cc -E', but still speeds up compilation with the -MD flags.
For 'make depend' a tree-walk is still done to ensure that all DPSRCS
are generated when expected, and that beforedepend/afterdepend and
_EXTRADEPEND are all still respected. In time this may change but for now
I've been conservative. The time for a tree-walk with -j combined with
SUBDIR_PARALLEL is not significant. For example, it takes about 9 seconds
with -j15 to walk all of src/ for 'make depend' now on my system.
A .depend file is still generated with the various rules that apply to
the final target, or custom rules. Otherwise there are now
per-built-object-file .depend files, such as .depend.filename.o. These
are included directly by make rather than populating .depend with a loop
and .depend lines, which only added overhead to the now almost-NOP 'make
depend' phase.
Before this I experimented with having mkdep(1) called in parallel per-file.
While this improved the kernel and lib/libc 'make depend' phase, it resulted
in slower build times overall.
The -M flags are removed from CFLAGS when linking since they have no effect.
Enabling this by default, for src or out-of-src, can be done once more testing
has been done, such as a ports exp-run, and with more compilers.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_FAST_DEPEND=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
buildworld:
x buildworld-before
+ buildworld-fastdep
+-------------------------------------------------------------------------------+
|+ |
|+ |
|+ xx x|
| |_MA___||
|A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|+ x |
|++ xx|
| A||
|A| |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
Sponsored by: EMC / Isilon Storage Division
MFC after: 3 weeks
Relnotes: yes
2015-11-06 04:45:29 +00:00
|
|
|
DEPENDFILES+= .depend.*
|
|
|
|
DEPEND_CFLAGS+= -MD -MP -MF.depend.${.TARGET}
|
|
|
|
DEPEND_CFLAGS+= -MT${.TARGET}
|
|
|
|
CFLAGS+= ${DEPEND_CFLAGS}
|
2016-01-15 22:08:58 +00:00
|
|
|
DEPENDOBJS+= ${SYSTEM_OBJS} genassym.o
|
2016-01-15 22:08:51 +00:00
|
|
|
DEPENDFILES_OBJS= ${DEPENDOBJS:O:u:C/^/.depend./}
|
2015-11-08 00:50:09 +00:00
|
|
|
.if ${.MAKEFLAGS:M-V} == ""
|
2016-01-15 22:08:51 +00:00
|
|
|
.for __depend_obj in ${DEPENDFILES_OBJS}
|
|
|
|
.sinclude "${__depend_obj}"
|
Add a FAST_DEPEND option, off by default, which speeds up the build significantly.
This speeds up buildworld by 16% on my system and buildkernel by 35%.
Rather than calling mkdep(1), which is just a wrapper around 'cc -E',
use the modern -MD -MT -MF flags to gather and generate dependencies during
compilation. This flag was introduced in GCC "a long time ago", in GCC 3.0,
and is also supported by Clang. (It appears that ICC also supports this but I
do not have access to test it). This avoids running the preprocessor *twice*
for every build, in both 'make depend' and 'make all'. This is especially
noticeable when using ccache since it does not cache preprocessor results from
mkdep(1) / 'cc -E', but still speeds up compilation with the -MD flags.
For 'make depend' a tree-walk is still done to ensure that all DPSRCS
are generated when expected, and that beforedepend/afterdepend and
_EXTRADEPEND are all still respected. In time this may change but for now
I've been conservative. The time for a tree-walk with -j combined with
SUBDIR_PARALLEL is not significant. For example, it takes about 9 seconds
with -j15 to walk all of src/ for 'make depend' now on my system.
A .depend file is still generated with the various rules that apply to
the final target, or custom rules. Otherwise there are now
per-built-object-file .depend files, such as .depend.filename.o. These
are included directly by make rather than populating .depend with a loop
and .depend lines, which only added overhead to the now almost-NOP 'make
depend' phase.
Before this I experimented with having mkdep(1) called in parallel per-file.
While this improved the kernel and lib/libc 'make depend' phase, it resulted
in slower build times overall.
The -M flags are removed from CFLAGS when linking since they have no effect.
Enabling this by default, for src or out-of-src, can be done once more testing
has been done, such as a ports exp-run, and with more compilers.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_FAST_DEPEND=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
buildworld:
x buildworld-before
+ buildworld-fastdep
+-------------------------------------------------------------------------------+
|+ |
|+ |
|+ xx x|
| |_MA___||
|A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|+ x |
|++ xx|
| A||
|A| |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
Sponsored by: EMC / Isilon Storage Division
MFC after: 3 weeks
Relnotes: yes
2015-11-06 04:45:29 +00:00
|
|
|
.endfor
|
2016-01-15 22:08:51 +00:00
|
|
|
.endif
|
Add a FAST_DEPEND option, off by default, which speeds up the build significantly.
This speeds up buildworld by 16% on my system and buildkernel by 35%.
Rather than calling mkdep(1), which is just a wrapper around 'cc -E',
use the modern -MD -MT -MF flags to gather and generate dependencies during
compilation. This flag was introduced in GCC "a long time ago", in GCC 3.0,
and is also supported by Clang. (It appears that ICC also supports this but I
do not have access to test it). This avoids running the preprocessor *twice*
for every build, in both 'make depend' and 'make all'. This is especially
noticeable when using ccache since it does not cache preprocessor results from
mkdep(1) / 'cc -E', but still speeds up compilation with the -MD flags.
For 'make depend' a tree-walk is still done to ensure that all DPSRCS
are generated when expected, and that beforedepend/afterdepend and
_EXTRADEPEND are all still respected. In time this may change but for now
I've been conservative. The time for a tree-walk with -j combined with
SUBDIR_PARALLEL is not significant. For example, it takes about 9 seconds
with -j15 to walk all of src/ for 'make depend' now on my system.
A .depend file is still generated with the various rules that apply to
the final target, or custom rules. Otherwise there are now
per-built-object-file .depend files, such as .depend.filename.o. These
are included directly by make rather than populating .depend with a loop
and .depend lines, which only added overhead to the now almost-NOP 'make
depend' phase.
Before this I experimented with having mkdep(1) called in parallel per-file.
While this improved the kernel and lib/libc 'make depend' phase, it resulted
in slower build times overall.
The -M flags are removed from CFLAGS when linking since they have no effect.
Enabling this by default, for src or out-of-src, can be done once more testing
has been done, such as a ports exp-run, and with more compilers.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_FAST_DEPEND=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
buildworld:
x buildworld-before
+ buildworld-fastdep
+-------------------------------------------------------------------------------+
|+ |
|+ |
|+ xx x|
| |_MA___||
|A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|+ x |
|++ xx|
| A||
|A| |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
Sponsored by: EMC / Isilon Storage Division
MFC after: 3 weeks
Relnotes: yes
2015-11-06 04:45:29 +00:00
|
|
|
.endif # ${MK_FAST_DEPEND} == "yes"
|
|
|
|
|
|
|
|
.NOPATH: .depend ${DEPENDFILES_OBJS}
|
|
|
|
|
2006-10-13 22:28:14 +00:00
|
|
|
.depend: .PRECIOUS ${SRCS}
|
Add a FAST_DEPEND option, off by default, which speeds up the build significantly.
This speeds up buildworld by 16% on my system and buildkernel by 35%.
Rather than calling mkdep(1), which is just a wrapper around 'cc -E',
use the modern -MD -MT -MF flags to gather and generate dependencies during
compilation. This flag was introduced in GCC "a long time ago", in GCC 3.0,
and is also supported by Clang. (It appears that ICC also supports this but I
do not have access to test it). This avoids running the preprocessor *twice*
for every build, in both 'make depend' and 'make all'. This is especially
noticeable when using ccache since it does not cache preprocessor results from
mkdep(1) / 'cc -E', but still speeds up compilation with the -MD flags.
For 'make depend' a tree-walk is still done to ensure that all DPSRCS
are generated when expected, and that beforedepend/afterdepend and
_EXTRADEPEND are all still respected. In time this may change but for now
I've been conservative. The time for a tree-walk with -j combined with
SUBDIR_PARALLEL is not significant. For example, it takes about 9 seconds
with -j15 to walk all of src/ for 'make depend' now on my system.
A .depend file is still generated with the various rules that apply to
the final target, or custom rules. Otherwise there are now
per-built-object-file .depend files, such as .depend.filename.o. These
are included directly by make rather than populating .depend with a loop
and .depend lines, which only added overhead to the now almost-NOP 'make
depend' phase.
Before this I experimented with having mkdep(1) called in parallel per-file.
While this improved the kernel and lib/libc 'make depend' phase, it resulted
in slower build times overall.
The -M flags are removed from CFLAGS when linking since they have no effect.
Enabling this by default, for src or out-of-src, can be done once more testing
has been done, such as a ports exp-run, and with more compilers.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_FAST_DEPEND=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
buildworld:
x buildworld-before
+ buildworld-fastdep
+-------------------------------------------------------------------------------+
|+ |
|+ |
|+ xx x|
| |_MA___||
|A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|+ x |
|++ xx|
| A||
|A| |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
Sponsored by: EMC / Isilon Storage Division
MFC after: 3 weeks
Relnotes: yes
2015-11-06 04:45:29 +00:00
|
|
|
.if ${MK_FAST_DEPEND} == "no"
|
|
|
|
rm -f ${.TARGET}.tmp
|
2016-01-27 14:46:30 +00:00
|
|
|
# C files
|
|
|
|
${MAKE} -V CFILES_NORMAL -V SYSTEM_CFILES -V GEN_CFILES | \
|
Add built-in ccache build support via WITH_CCACHE_BUILD option.
ccache is mostly beneficial for frequent builds where -DNO_CLEAN is not
used to achieve a safe pseudo-incremental build. This is explained in
more detail upstream [1] [2]. It incurs about a 20%-28% hit to populate the
cache, but with a full cache saves 30-50% in build times. When combined with
the WITH_FAST_DEPEND feature it saves up to 65% since ccache does cache the
resulting dependency file, which it does not do when using mkdep(1)/'CC
-E'. Stats are provided at the end of this message.
This removes the need to modify /etc/make.conf with the CC:= and CXX:=
lines which conflicted with external compiler support [3] (causing the
bootstrap compiler to not be built which lead to obscure failures [4]),
incorrectly invoked ccache in various stages, required CCACHE_CPP2 to avoid
Clang errors with parenthesis, and did not work with META_MODE.
The option name was picked to match the existing option in ports. This
feature is available for both in-src and out-of-src builds that use
/usr/share/mk.
Linking, assembly compiles, and pre-processing avoid using ccache since it is
only overhead. ccache does nothing special in these modes, although there is
no harm in calling it for them.
CCACHE_COMPILERCHECK is set to 'content' when using the in-tree bootstrap
compiler to hash the content of the compiler binary to determine if it
should be a cache miss. For external compilers the 'mtime' option is used
as it is more efficient and likely to be correct. Future work may optimize the
'content' check using the same checks as whether a bootstrap compiler is needed
to be built.
The CCACHE_CPP2 pessimization is currently default in our devel/ccache
port due to Clang requiring it. Clang's -Wparentheses-equality,
-Wtautological-compare, and -Wself-assign warnings do not mix well with
compiling already-pre-processed code that may have expanded macros that
trigger the warnings. GCC has so far not had this issue so it is allowed to
disable the CCACHE_CPP2 default in our port.
Sharing a cache between multiple checkouts, or systems, is explained in
the ccache manual. Sharing a cache over NFS would likely not be worth
it, but syncing cache directories between systems may be useful for an
organization. There is also a memcached backend available [5]. Due to using
an object directory outside of the source directory though you will need to
ensure that both are in the same prefix and all users use the same layout. A
possible working layout is as follows:
Source: /some/prefix/src1
Source: /some/prefix/src2
Source: /some/prefix/src3
Objdir: /some/prefix/obj
Environment: CCACHE_BASEDIR='${SRCTOP:H}' MAKEOBJDIRPREFIX='${SRCTOP:H}/obj'
This will use src*/../obj as the MAKEOBJDIRPREFIX and tells ccache to replace
all absolute paths to be relative. Using something like this is required due
to -I and -o flags containing both SRC and OBJDIR absolute paths that ccache
adds into its hash for the object without CCACHE_BASEDIR.
distcc can be hooked into by setting CCACHE_PREFIX=/usr/local/bin/distcc.
I have not personally tested this and assume it will not mix well with
using the bootstrap compiler.
The cache from buildworld can be reused in a subdir by first running
'make buildenv' (from r290424).
Note that the cache is currently different depending on whether -j is
used or not due to ccache enabling -fdiagnostics-color automatically if
stderr is a TTY, which bmake only does if not using -j.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_CCACHE_BUILD=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files and ccache objects.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
The WITH_FAST_DEPEND feature was used for comparison here as well to show
the dramatic time savings with a full cache.
buildworld:
x buildworld-before
+ buildworld-ccache-empty
* buildworld-ccache-full
% buildworld-ccache-full-fastdep
# buildworld-fastdep
+-------------------------------------------------------------------------------+
|% * # +|
|% * # +|
|% * # xxx +|
| |A |
| A|
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 4519 4525.04 4520.73 4521.59 3.1104823
Difference at 95.0% confidence
758.027 +/- 43.4565
20.1412% +/- 1.15466%
(Student's t, pooled s = 19.1726)
* 3 1823.08 1827.2 1825.62 1825.3 2.0785572
Difference at 95.0% confidence
-1938.26 +/- 43.298
-51.5007% +/- 1.15045%
(Student's t, pooled s = 19.1026)
% 3 1266.96 1279.37 1270.47 1272.2667 6.3971113
Difference at 95.0% confidence
-2491.3 +/- 44.3704
-66.1952% +/- 1.17895%
(Student's t, pooled s = 19.5758)
# 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-ccache-empty
* buildkernel-ccache-empty-fastdep
% buildkernel-ccache-full
# buildkernel-ccache-full-fastdep
@ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|# @ % * |
|# @ % * x + |
|# @ % * xx ++|
| MA |
| MA|
| A |
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 727.97 731.91 728.06 729.31333 2.2492295
Difference at 95.0% confidence
156.88 +/- 4.17129
27.4058% +/- 0.728695%
(Student's t, pooled s = 1.84034)
* 3 527.1 528.29 528.08 527.82333 0.63516402
Difference at 95.0% confidence
-44.61 +/- 2.33254
-7.79305% +/- 0.407478%
(Student's t, pooled s = 1.02909)
% 3 400.4 401.05 400.62 400.69 0.3306055
Difference at 95.0% confidence
-171.743 +/- 2.16453
-30.0023% +/- 0.378128%
(Student's t, pooled s = 0.954969)
# 3 201.94 203.34 202.28 202.52 0.73020545
Difference at 95.0% confidence
-369.913 +/- 2.40293
-64.6212% +/- 0.419774%
(Student's t, pooled s = 1.06015)
@ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
[1] https://ccache.samba.org/performance.html
[2] http://www.mail-archive.com/ccache@lists.samba.org/msg00576.html
[3] https://reviews.freebsd.org/D3484
[5] https://github.com/jrosdahl/ccache/pull/30
PR: 182944 [4]
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
Relnotes: yes
2015-11-08 00:50:18 +00:00
|
|
|
CC="${_MKDEPCC}" xargs mkdep -a -f ${.TARGET}.tmp ${CFLAGS}
|
2015-06-10 15:53:39 +00:00
|
|
|
${MAKE} -V CFILES_CDDL | \
|
Add built-in ccache build support via WITH_CCACHE_BUILD option.
ccache is mostly beneficial for frequent builds where -DNO_CLEAN is not
used to achieve a safe pseudo-incremental build. This is explained in
more detail upstream [1] [2]. It incurs about a 20%-28% hit to populate the
cache, but with a full cache saves 30-50% in build times. When combined with
the WITH_FAST_DEPEND feature it saves up to 65% since ccache does cache the
resulting dependency file, which it does not do when using mkdep(1)/'CC
-E'. Stats are provided at the end of this message.
This removes the need to modify /etc/make.conf with the CC:= and CXX:=
lines which conflicted with external compiler support [3] (causing the
bootstrap compiler to not be built which lead to obscure failures [4]),
incorrectly invoked ccache in various stages, required CCACHE_CPP2 to avoid
Clang errors with parenthesis, and did not work with META_MODE.
The option name was picked to match the existing option in ports. This
feature is available for both in-src and out-of-src builds that use
/usr/share/mk.
Linking, assembly compiles, and pre-processing avoid using ccache since it is
only overhead. ccache does nothing special in these modes, although there is
no harm in calling it for them.
CCACHE_COMPILERCHECK is set to 'content' when using the in-tree bootstrap
compiler to hash the content of the compiler binary to determine if it
should be a cache miss. For external compilers the 'mtime' option is used
as it is more efficient and likely to be correct. Future work may optimize the
'content' check using the same checks as whether a bootstrap compiler is needed
to be built.
The CCACHE_CPP2 pessimization is currently default in our devel/ccache
port due to Clang requiring it. Clang's -Wparentheses-equality,
-Wtautological-compare, and -Wself-assign warnings do not mix well with
compiling already-pre-processed code that may have expanded macros that
trigger the warnings. GCC has so far not had this issue so it is allowed to
disable the CCACHE_CPP2 default in our port.
Sharing a cache between multiple checkouts, or systems, is explained in
the ccache manual. Sharing a cache over NFS would likely not be worth
it, but syncing cache directories between systems may be useful for an
organization. There is also a memcached backend available [5]. Due to using
an object directory outside of the source directory though you will need to
ensure that both are in the same prefix and all users use the same layout. A
possible working layout is as follows:
Source: /some/prefix/src1
Source: /some/prefix/src2
Source: /some/prefix/src3
Objdir: /some/prefix/obj
Environment: CCACHE_BASEDIR='${SRCTOP:H}' MAKEOBJDIRPREFIX='${SRCTOP:H}/obj'
This will use src*/../obj as the MAKEOBJDIRPREFIX and tells ccache to replace
all absolute paths to be relative. Using something like this is required due
to -I and -o flags containing both SRC and OBJDIR absolute paths that ccache
adds into its hash for the object without CCACHE_BASEDIR.
distcc can be hooked into by setting CCACHE_PREFIX=/usr/local/bin/distcc.
I have not personally tested this and assume it will not mix well with
using the bootstrap compiler.
The cache from buildworld can be reused in a subdir by first running
'make buildenv' (from r290424).
Note that the cache is currently different depending on whether -j is
used or not due to ccache enabling -fdiagnostics-color automatically if
stderr is a TTY, which bmake only does if not using -j.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_CCACHE_BUILD=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files and ccache objects.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
The WITH_FAST_DEPEND feature was used for comparison here as well to show
the dramatic time savings with a full cache.
buildworld:
x buildworld-before
+ buildworld-ccache-empty
* buildworld-ccache-full
% buildworld-ccache-full-fastdep
# buildworld-fastdep
+-------------------------------------------------------------------------------+
|% * # +|
|% * # +|
|% * # xxx +|
| |A |
| A|
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 4519 4525.04 4520.73 4521.59 3.1104823
Difference at 95.0% confidence
758.027 +/- 43.4565
20.1412% +/- 1.15466%
(Student's t, pooled s = 19.1726)
* 3 1823.08 1827.2 1825.62 1825.3 2.0785572
Difference at 95.0% confidence
-1938.26 +/- 43.298
-51.5007% +/- 1.15045%
(Student's t, pooled s = 19.1026)
% 3 1266.96 1279.37 1270.47 1272.2667 6.3971113
Difference at 95.0% confidence
-2491.3 +/- 44.3704
-66.1952% +/- 1.17895%
(Student's t, pooled s = 19.5758)
# 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-ccache-empty
* buildkernel-ccache-empty-fastdep
% buildkernel-ccache-full
# buildkernel-ccache-full-fastdep
@ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|# @ % * |
|# @ % * x + |
|# @ % * xx ++|
| MA |
| MA|
| A |
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 727.97 731.91 728.06 729.31333 2.2492295
Difference at 95.0% confidence
156.88 +/- 4.17129
27.4058% +/- 0.728695%
(Student's t, pooled s = 1.84034)
* 3 527.1 528.29 528.08 527.82333 0.63516402
Difference at 95.0% confidence
-44.61 +/- 2.33254
-7.79305% +/- 0.407478%
(Student's t, pooled s = 1.02909)
% 3 400.4 401.05 400.62 400.69 0.3306055
Difference at 95.0% confidence
-171.743 +/- 2.16453
-30.0023% +/- 0.378128%
(Student's t, pooled s = 0.954969)
# 3 201.94 203.34 202.28 202.52 0.73020545
Difference at 95.0% confidence
-369.913 +/- 2.40293
-64.6212% +/- 0.419774%
(Student's t, pooled s = 1.06015)
@ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
[1] https://ccache.samba.org/performance.html
[2] http://www.mail-archive.com/ccache@lists.samba.org/msg00576.html
[3] https://reviews.freebsd.org/D3484
[5] https://github.com/jrosdahl/ccache/pull/30
PR: 182944 [4]
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
Relnotes: yes
2015-11-08 00:50:18 +00:00
|
|
|
CC="${_MKDEPCC}" xargs mkdep -a -f ${.TARGET}.tmp ${ZFS_CFLAGS} \
|
Add a FAST_DEPEND option, off by default, which speeds up the build significantly.
This speeds up buildworld by 16% on my system and buildkernel by 35%.
Rather than calling mkdep(1), which is just a wrapper around 'cc -E',
use the modern -MD -MT -MF flags to gather and generate dependencies during
compilation. This flag was introduced in GCC "a long time ago", in GCC 3.0,
and is also supported by Clang. (It appears that ICC also supports this but I
do not have access to test it). This avoids running the preprocessor *twice*
for every build, in both 'make depend' and 'make all'. This is especially
noticeable when using ccache since it does not cache preprocessor results from
mkdep(1) / 'cc -E', but still speeds up compilation with the -MD flags.
For 'make depend' a tree-walk is still done to ensure that all DPSRCS
are generated when expected, and that beforedepend/afterdepend and
_EXTRADEPEND are all still respected. In time this may change but for now
I've been conservative. The time for a tree-walk with -j combined with
SUBDIR_PARALLEL is not significant. For example, it takes about 9 seconds
with -j15 to walk all of src/ for 'make depend' now on my system.
A .depend file is still generated with the various rules that apply to
the final target, or custom rules. Otherwise there are now
per-built-object-file .depend files, such as .depend.filename.o. These
are included directly by make rather than populating .depend with a loop
and .depend lines, which only added overhead to the now almost-NOP 'make
depend' phase.
Before this I experimented with having mkdep(1) called in parallel per-file.
While this improved the kernel and lib/libc 'make depend' phase, it resulted
in slower build times overall.
The -M flags are removed from CFLAGS when linking since they have no effect.
Enabling this by default, for src or out-of-src, can be done once more testing
has been done, such as a ports exp-run, and with more compilers.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_FAST_DEPEND=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
buildworld:
x buildworld-before
+ buildworld-fastdep
+-------------------------------------------------------------------------------+
|+ |
|+ |
|+ xx x|
| |_MA___||
|A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|+ x |
|++ xx|
| A||
|A| |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
Sponsored by: EMC / Isilon Storage Division
MFC after: 3 weeks
Relnotes: yes
2015-11-06 04:45:29 +00:00
|
|
|
${FBT_CFLAGS} ${DTRACE_CFLAGS}
|
2016-01-27 14:46:30 +00:00
|
|
|
${MAKE} -V CFILES_LINUXKPI | \
|
|
|
|
CC="${_MKDEPCC}" xargs mkdep -a -f ${.TARGET}.tmp \
|
|
|
|
${CFLAGS} ${LINUXKPI_INCLUDES}
|
|
|
|
${MAKE} -V CFILES_OFED -V CFILES_MLX5 | \
|
|
|
|
CC="${_MKDEPCC}" xargs mkdep -a -f ${.TARGET}.tmp \
|
|
|
|
${CFLAGS} ${OFEDINCLUDES}
|
|
|
|
# Assembly files
|
|
|
|
${MAKE} -V SFILES_NORMAL | \
|
Add built-in ccache build support via WITH_CCACHE_BUILD option.
ccache is mostly beneficial for frequent builds where -DNO_CLEAN is not
used to achieve a safe pseudo-incremental build. This is explained in
more detail upstream [1] [2]. It incurs about a 20%-28% hit to populate the
cache, but with a full cache saves 30-50% in build times. When combined with
the WITH_FAST_DEPEND feature it saves up to 65% since ccache does cache the
resulting dependency file, which it does not do when using mkdep(1)/'CC
-E'. Stats are provided at the end of this message.
This removes the need to modify /etc/make.conf with the CC:= and CXX:=
lines which conflicted with external compiler support [3] (causing the
bootstrap compiler to not be built which lead to obscure failures [4]),
incorrectly invoked ccache in various stages, required CCACHE_CPP2 to avoid
Clang errors with parenthesis, and did not work with META_MODE.
The option name was picked to match the existing option in ports. This
feature is available for both in-src and out-of-src builds that use
/usr/share/mk.
Linking, assembly compiles, and pre-processing avoid using ccache since it is
only overhead. ccache does nothing special in these modes, although there is
no harm in calling it for them.
CCACHE_COMPILERCHECK is set to 'content' when using the in-tree bootstrap
compiler to hash the content of the compiler binary to determine if it
should be a cache miss. For external compilers the 'mtime' option is used
as it is more efficient and likely to be correct. Future work may optimize the
'content' check using the same checks as whether a bootstrap compiler is needed
to be built.
The CCACHE_CPP2 pessimization is currently default in our devel/ccache
port due to Clang requiring it. Clang's -Wparentheses-equality,
-Wtautological-compare, and -Wself-assign warnings do not mix well with
compiling already-pre-processed code that may have expanded macros that
trigger the warnings. GCC has so far not had this issue so it is allowed to
disable the CCACHE_CPP2 default in our port.
Sharing a cache between multiple checkouts, or systems, is explained in
the ccache manual. Sharing a cache over NFS would likely not be worth
it, but syncing cache directories between systems may be useful for an
organization. There is also a memcached backend available [5]. Due to using
an object directory outside of the source directory though you will need to
ensure that both are in the same prefix and all users use the same layout. A
possible working layout is as follows:
Source: /some/prefix/src1
Source: /some/prefix/src2
Source: /some/prefix/src3
Objdir: /some/prefix/obj
Environment: CCACHE_BASEDIR='${SRCTOP:H}' MAKEOBJDIRPREFIX='${SRCTOP:H}/obj'
This will use src*/../obj as the MAKEOBJDIRPREFIX and tells ccache to replace
all absolute paths to be relative. Using something like this is required due
to -I and -o flags containing both SRC and OBJDIR absolute paths that ccache
adds into its hash for the object without CCACHE_BASEDIR.
distcc can be hooked into by setting CCACHE_PREFIX=/usr/local/bin/distcc.
I have not personally tested this and assume it will not mix well with
using the bootstrap compiler.
The cache from buildworld can be reused in a subdir by first running
'make buildenv' (from r290424).
Note that the cache is currently different depending on whether -j is
used or not due to ccache enabling -fdiagnostics-color automatically if
stderr is a TTY, which bmake only does if not using -j.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_CCACHE_BUILD=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files and ccache objects.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
The WITH_FAST_DEPEND feature was used for comparison here as well to show
the dramatic time savings with a full cache.
buildworld:
x buildworld-before
+ buildworld-ccache-empty
* buildworld-ccache-full
% buildworld-ccache-full-fastdep
# buildworld-fastdep
+-------------------------------------------------------------------------------+
|% * # +|
|% * # +|
|% * # xxx +|
| |A |
| A|
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 4519 4525.04 4520.73 4521.59 3.1104823
Difference at 95.0% confidence
758.027 +/- 43.4565
20.1412% +/- 1.15466%
(Student's t, pooled s = 19.1726)
* 3 1823.08 1827.2 1825.62 1825.3 2.0785572
Difference at 95.0% confidence
-1938.26 +/- 43.298
-51.5007% +/- 1.15045%
(Student's t, pooled s = 19.1026)
% 3 1266.96 1279.37 1270.47 1272.2667 6.3971113
Difference at 95.0% confidence
-2491.3 +/- 44.3704
-66.1952% +/- 1.17895%
(Student's t, pooled s = 19.5758)
# 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-ccache-empty
* buildkernel-ccache-empty-fastdep
% buildkernel-ccache-full
# buildkernel-ccache-full-fastdep
@ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|# @ % * |
|# @ % * x + |
|# @ % * xx ++|
| MA |
| MA|
| A |
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 727.97 731.91 728.06 729.31333 2.2492295
Difference at 95.0% confidence
156.88 +/- 4.17129
27.4058% +/- 0.728695%
(Student's t, pooled s = 1.84034)
* 3 527.1 528.29 528.08 527.82333 0.63516402
Difference at 95.0% confidence
-44.61 +/- 2.33254
-7.79305% +/- 0.407478%
(Student's t, pooled s = 1.02909)
% 3 400.4 401.05 400.62 400.69 0.3306055
Difference at 95.0% confidence
-171.743 +/- 2.16453
-30.0023% +/- 0.378128%
(Student's t, pooled s = 0.954969)
# 3 201.94 203.34 202.28 202.52 0.73020545
Difference at 95.0% confidence
-369.913 +/- 2.40293
-64.6212% +/- 0.419774%
(Student's t, pooled s = 1.06015)
@ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
[1] https://ccache.samba.org/performance.html
[2] http://www.mail-archive.com/ccache@lists.samba.org/msg00576.html
[3] https://reviews.freebsd.org/D3484
[5] https://github.com/jrosdahl/ccache/pull/30
PR: 182944 [4]
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
Relnotes: yes
2015-11-08 00:50:18 +00:00
|
|
|
CC="${_MKDEPCC}" xargs mkdep -a -f ${.TARGET}.tmp ${ASM_CFLAGS}
|
2015-06-10 15:53:39 +00:00
|
|
|
${MAKE} -V SFILES_CDDL | \
|
Add built-in ccache build support via WITH_CCACHE_BUILD option.
ccache is mostly beneficial for frequent builds where -DNO_CLEAN is not
used to achieve a safe pseudo-incremental build. This is explained in
more detail upstream [1] [2]. It incurs about a 20%-28% hit to populate the
cache, but with a full cache saves 30-50% in build times. When combined with
the WITH_FAST_DEPEND feature it saves up to 65% since ccache does cache the
resulting dependency file, which it does not do when using mkdep(1)/'CC
-E'. Stats are provided at the end of this message.
This removes the need to modify /etc/make.conf with the CC:= and CXX:=
lines which conflicted with external compiler support [3] (causing the
bootstrap compiler to not be built which lead to obscure failures [4]),
incorrectly invoked ccache in various stages, required CCACHE_CPP2 to avoid
Clang errors with parenthesis, and did not work with META_MODE.
The option name was picked to match the existing option in ports. This
feature is available for both in-src and out-of-src builds that use
/usr/share/mk.
Linking, assembly compiles, and pre-processing avoid using ccache since it is
only overhead. ccache does nothing special in these modes, although there is
no harm in calling it for them.
CCACHE_COMPILERCHECK is set to 'content' when using the in-tree bootstrap
compiler to hash the content of the compiler binary to determine if it
should be a cache miss. For external compilers the 'mtime' option is used
as it is more efficient and likely to be correct. Future work may optimize the
'content' check using the same checks as whether a bootstrap compiler is needed
to be built.
The CCACHE_CPP2 pessimization is currently default in our devel/ccache
port due to Clang requiring it. Clang's -Wparentheses-equality,
-Wtautological-compare, and -Wself-assign warnings do not mix well with
compiling already-pre-processed code that may have expanded macros that
trigger the warnings. GCC has so far not had this issue so it is allowed to
disable the CCACHE_CPP2 default in our port.
Sharing a cache between multiple checkouts, or systems, is explained in
the ccache manual. Sharing a cache over NFS would likely not be worth
it, but syncing cache directories between systems may be useful for an
organization. There is also a memcached backend available [5]. Due to using
an object directory outside of the source directory though you will need to
ensure that both are in the same prefix and all users use the same layout. A
possible working layout is as follows:
Source: /some/prefix/src1
Source: /some/prefix/src2
Source: /some/prefix/src3
Objdir: /some/prefix/obj
Environment: CCACHE_BASEDIR='${SRCTOP:H}' MAKEOBJDIRPREFIX='${SRCTOP:H}/obj'
This will use src*/../obj as the MAKEOBJDIRPREFIX and tells ccache to replace
all absolute paths to be relative. Using something like this is required due
to -I and -o flags containing both SRC and OBJDIR absolute paths that ccache
adds into its hash for the object without CCACHE_BASEDIR.
distcc can be hooked into by setting CCACHE_PREFIX=/usr/local/bin/distcc.
I have not personally tested this and assume it will not mix well with
using the bootstrap compiler.
The cache from buildworld can be reused in a subdir by first running
'make buildenv' (from r290424).
Note that the cache is currently different depending on whether -j is
used or not due to ccache enabling -fdiagnostics-color automatically if
stderr is a TTY, which bmake only does if not using -j.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_CCACHE_BUILD=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files and ccache objects.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
The WITH_FAST_DEPEND feature was used for comparison here as well to show
the dramatic time savings with a full cache.
buildworld:
x buildworld-before
+ buildworld-ccache-empty
* buildworld-ccache-full
% buildworld-ccache-full-fastdep
# buildworld-fastdep
+-------------------------------------------------------------------------------+
|% * # +|
|% * # +|
|% * # xxx +|
| |A |
| A|
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 4519 4525.04 4520.73 4521.59 3.1104823
Difference at 95.0% confidence
758.027 +/- 43.4565
20.1412% +/- 1.15466%
(Student's t, pooled s = 19.1726)
* 3 1823.08 1827.2 1825.62 1825.3 2.0785572
Difference at 95.0% confidence
-1938.26 +/- 43.298
-51.5007% +/- 1.15045%
(Student's t, pooled s = 19.1026)
% 3 1266.96 1279.37 1270.47 1272.2667 6.3971113
Difference at 95.0% confidence
-2491.3 +/- 44.3704
-66.1952% +/- 1.17895%
(Student's t, pooled s = 19.5758)
# 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-ccache-empty
* buildkernel-ccache-empty-fastdep
% buildkernel-ccache-full
# buildkernel-ccache-full-fastdep
@ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|# @ % * |
|# @ % * x + |
|# @ % * xx ++|
| MA |
| MA|
| A |
| A |
|A |
| A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 727.97 731.91 728.06 729.31333 2.2492295
Difference at 95.0% confidence
156.88 +/- 4.17129
27.4058% +/- 0.728695%
(Student's t, pooled s = 1.84034)
* 3 527.1 528.29 528.08 527.82333 0.63516402
Difference at 95.0% confidence
-44.61 +/- 2.33254
-7.79305% +/- 0.407478%
(Student's t, pooled s = 1.02909)
% 3 400.4 401.05 400.62 400.69 0.3306055
Difference at 95.0% confidence
-171.743 +/- 2.16453
-30.0023% +/- 0.378128%
(Student's t, pooled s = 0.954969)
# 3 201.94 203.34 202.28 202.52 0.73020545
Difference at 95.0% confidence
-369.913 +/- 2.40293
-64.6212% +/- 0.419774%
(Student's t, pooled s = 1.06015)
@ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
[1] https://ccache.samba.org/performance.html
[2] http://www.mail-archive.com/ccache@lists.samba.org/msg00576.html
[3] https://reviews.freebsd.org/D3484
[5] https://github.com/jrosdahl/ccache/pull/30
PR: 182944 [4]
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
Relnotes: yes
2015-11-08 00:50:18 +00:00
|
|
|
CC="${_MKDEPCC}" xargs mkdep -a -f ${.TARGET}.tmp ${ZFS_ASM_CFLAGS}
|
Add a FAST_DEPEND option, off by default, which speeds up the build significantly.
This speeds up buildworld by 16% on my system and buildkernel by 35%.
Rather than calling mkdep(1), which is just a wrapper around 'cc -E',
use the modern -MD -MT -MF flags to gather and generate dependencies during
compilation. This flag was introduced in GCC "a long time ago", in GCC 3.0,
and is also supported by Clang. (It appears that ICC also supports this but I
do not have access to test it). This avoids running the preprocessor *twice*
for every build, in both 'make depend' and 'make all'. This is especially
noticeable when using ccache since it does not cache preprocessor results from
mkdep(1) / 'cc -E', but still speeds up compilation with the -MD flags.
For 'make depend' a tree-walk is still done to ensure that all DPSRCS
are generated when expected, and that beforedepend/afterdepend and
_EXTRADEPEND are all still respected. In time this may change but for now
I've been conservative. The time for a tree-walk with -j combined with
SUBDIR_PARALLEL is not significant. For example, it takes about 9 seconds
with -j15 to walk all of src/ for 'make depend' now on my system.
A .depend file is still generated with the various rules that apply to
the final target, or custom rules. Otherwise there are now
per-built-object-file .depend files, such as .depend.filename.o. These
are included directly by make rather than populating .depend with a loop
and .depend lines, which only added overhead to the now almost-NOP 'make
depend' phase.
Before this I experimented with having mkdep(1) called in parallel per-file.
While this improved the kernel and lib/libc 'make depend' phase, it resulted
in slower build times overall.
The -M flags are removed from CFLAGS when linking since they have no effect.
Enabling this by default, for src or out-of-src, can be done once more testing
has been done, such as a ports exp-run, and with more compilers.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_FAST_DEPEND=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
buildworld:
x buildworld-before
+ buildworld-fastdep
+-------------------------------------------------------------------------------+
|+ |
|+ |
|+ xx x|
| |_MA___||
|A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|+ x |
|++ xx|
| A||
|A| |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
Sponsored by: EMC / Isilon Storage Division
MFC after: 3 weeks
Relnotes: yes
2015-11-06 04:45:29 +00:00
|
|
|
mv ${.TARGET}.tmp ${.TARGET}
|
|
|
|
.else
|
|
|
|
: > ${.TARGET}
|
|
|
|
.endif
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2005-11-30 18:15:06 +00:00
|
|
|
_ILINKS= machine
|
2015-10-08 17:42:08 +00:00
|
|
|
.if ${MACHINE} != ${MACHINE_CPUARCH} && ${MACHINE} != "arm64"
|
2010-07-13 13:11:18 +00:00
|
|
|
_ILINKS+= ${MACHINE_CPUARCH}
|
2005-11-30 18:15:06 +00:00
|
|
|
.endif
|
2010-11-01 17:34:04 +00:00
|
|
|
.if ${MACHINE_CPUARCH} == "i386" || ${MACHINE_CPUARCH} == "amd64"
|
|
|
|
_ILINKS+= x86
|
|
|
|
.endif
|
2005-11-30 18:15:06 +00:00
|
|
|
|
|
|
|
# Ensure that the link exists without depending on it when it exists.
|
|
|
|
.for _link in ${_ILINKS}
|
|
|
|
.if !exists(${.OBJDIR}/${_link})
|
2013-08-14 22:19:29 +00:00
|
|
|
${SRCS} ${CLEAN:M*.o}: ${_link}
|
2005-11-30 18:15:06 +00:00
|
|
|
.endif
|
|
|
|
.endfor
|
|
|
|
|
|
|
|
${_ILINKS}:
|
|
|
|
@case ${.TARGET} in \
|
|
|
|
machine) \
|
|
|
|
path=${S}/${MACHINE}/include ;; \
|
2010-11-01 17:34:04 +00:00
|
|
|
*) \
|
|
|
|
path=${S}/${.TARGET}/include ;; \
|
2005-11-30 18:15:06 +00:00
|
|
|
esac ; \
|
|
|
|
${ECHO} ${.TARGET} "->" $$path ; \
|
|
|
|
ln -s $$path ${.TARGET}
|
|
|
|
|
2006-10-13 22:28:14 +00:00
|
|
|
# .depend needs include links so we remove them only together.
|
Add a FAST_DEPEND option, off by default, which speeds up the build significantly.
This speeds up buildworld by 16% on my system and buildkernel by 35%.
Rather than calling mkdep(1), which is just a wrapper around 'cc -E',
use the modern -MD -MT -MF flags to gather and generate dependencies during
compilation. This flag was introduced in GCC "a long time ago", in GCC 3.0,
and is also supported by Clang. (It appears that ICC also supports this but I
do not have access to test it). This avoids running the preprocessor *twice*
for every build, in both 'make depend' and 'make all'. This is especially
noticeable when using ccache since it does not cache preprocessor results from
mkdep(1) / 'cc -E', but still speeds up compilation with the -MD flags.
For 'make depend' a tree-walk is still done to ensure that all DPSRCS
are generated when expected, and that beforedepend/afterdepend and
_EXTRADEPEND are all still respected. In time this may change but for now
I've been conservative. The time for a tree-walk with -j combined with
SUBDIR_PARALLEL is not significant. For example, it takes about 9 seconds
with -j15 to walk all of src/ for 'make depend' now on my system.
A .depend file is still generated with the various rules that apply to
the final target, or custom rules. Otherwise there are now
per-built-object-file .depend files, such as .depend.filename.o. These
are included directly by make rather than populating .depend with a loop
and .depend lines, which only added overhead to the now almost-NOP 'make
depend' phase.
Before this I experimented with having mkdep(1) called in parallel per-file.
While this improved the kernel and lib/libc 'make depend' phase, it resulted
in slower build times overall.
The -M flags are removed from CFLAGS when linking since they have no effect.
Enabling this by default, for src or out-of-src, can be done once more testing
has been done, such as a ports exp-run, and with more compilers.
The system I used for testing was:
WITNESS
Build options: -j20 WITH_LLDB=yes WITH_DEBUG_FILES=yes WITH_FAST_DEPEND=yes
DISK: ZFS 3-way mirror with very slow disks using SSD l2arc/log.
The arc was fully populated with src tree files.
RAM: 76GiB
CPU: Intel(R) Xeon(R) CPU L5520 @2.27GHz
2 package(s) x 4 core(s) x 2 SMT threads = hw.ncpu=16
buildworld:
x buildworld-before
+ buildworld-fastdep
+-------------------------------------------------------------------------------+
|+ |
|+ |
|+ xx x|
| |_MA___||
|A |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 3744.13 3794.31 3752.25 3763.5633 26.935139
+ 3 3153.34 3155.16 3154.2 3154.2333 0.91045776
Difference at 95.0% confidence
-609.33 +/- 43.1943
-16.1902% +/- 1.1477%
(Student's t, pooled s = 19.0569)
buildkernel:
x buildkernel-before
+ buildkernel-fastdep
+-------------------------------------------------------------------------------+
|+ x |
|++ xx|
| A||
|A| |
+-------------------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 3 571.57 573.94 571.79 572.43333 1.3094401
+ 3 369.12 370.57 369.3 369.66333 0.79033748
Difference at 95.0% confidence
-202.77 +/- 2.45131
-35.4225% +/- 0.428227%
(Student's t, pooled s = 1.0815)
Sponsored by: EMC / Isilon Storage Division
MFC after: 3 weeks
Relnotes: yes
2015-11-06 04:45:29 +00:00
|
|
|
kernel-cleandepend: .PHONY
|
|
|
|
rm -f ${DEPENDFILES} ${_ILINKS}
|
2001-11-02 21:34:20 +00:00
|
|
|
|
|
|
|
kernel-tags:
|
|
|
|
@[ -f .depend ] || { echo "you must make depend first"; exit 1; }
|
|
|
|
sh $S/conf/systags.sh
|
|
|
|
|
2004-03-20 22:47:30 +00:00
|
|
|
kernel-install:
|
2005-10-27 14:24:45 +00:00
|
|
|
@if [ ! -f ${KERNEL_KO} ] ; then \
|
2001-11-02 21:34:20 +00:00
|
|
|
echo "You must build a kernel first." ; \
|
|
|
|
exit 1 ; \
|
|
|
|
fi
|
|
|
|
.if exists(${DESTDIR}${KODIR})
|
|
|
|
-thiskernel=`sysctl -n kern.bootfile` ; \
|
2005-04-28 14:45:57 +00:00
|
|
|
if [ ! "`dirname "$$thiskernel"`" -ef ${DESTDIR}${KODIR} ] ; then \
|
2001-11-02 21:34:20 +00:00
|
|
|
chflags -R noschg ${DESTDIR}${KODIR} ; \
|
|
|
|
rm -rf ${DESTDIR}${KODIR} ; \
|
2015-09-24 16:55:22 +00:00
|
|
|
rm -rf ${DESTDIR}${KERN_DEBUGDIR}${KODIR} ; \
|
2001-11-02 21:34:20 +00:00
|
|
|
else \
|
|
|
|
if [ -d ${DESTDIR}${KODIR}.old ] ; then \
|
|
|
|
chflags -R noschg ${DESTDIR}${KODIR}.old ; \
|
|
|
|
rm -rf ${DESTDIR}${KODIR}.old ; \
|
|
|
|
fi ; \
|
|
|
|
mv ${DESTDIR}${KODIR} ${DESTDIR}${KODIR}.old ; \
|
2015-09-24 16:55:22 +00:00
|
|
|
if [ -n "${KERN_DEBUGDIR}" -a \
|
|
|
|
-d ${DESTDIR}${KERN_DEBUGDIR}${KODIR} ]; then \
|
|
|
|
rm -rf ${DESTDIR}${KERN_DEBUGDIR}${KODIR}.old ; \
|
|
|
|
mv ${DESTDIR}${KERN_DEBUGDIR}${KODIR} ${DESTDIR}${KERN_DEBUGDIR}${KODIR}.old ; \
|
|
|
|
fi ; \
|
2003-08-07 19:12:06 +00:00
|
|
|
sysctl kern.bootfile=${DESTDIR}${KODIR}.old/"`basename "$$thiskernel"`" ; \
|
2001-11-02 21:34:20 +00:00
|
|
|
fi
|
|
|
|
.endif
|
|
|
|
mkdir -p ${DESTDIR}${KODIR}
|
2015-10-17 05:49:07 +00:00
|
|
|
${INSTALL} -p -m 555 -o ${KMODOWN} -g ${KMODGRP} ${KERNEL_KO} ${DESTDIR}${KODIR}/
|
2012-07-03 05:01:00 +00:00
|
|
|
.if defined(DEBUG) && !defined(INSTALL_NODEBUG) && ${MK_KERNEL_SYMBOLS} != "no"
|
2015-09-24 16:55:22 +00:00
|
|
|
mkdir -p ${DESTDIR}${KERN_DEBUGDIR}${KODIR}
|
2015-10-17 05:49:07 +00:00
|
|
|
${INSTALL} -p -m 555 -o ${KMODOWN} -g ${KMODGRP} ${KERNEL_KO}.debug ${DESTDIR}${KERN_DEBUGDIR}${KODIR}/
|
2005-10-24 23:54:39 +00:00
|
|
|
.endif
|
2007-03-23 21:55:59 +00:00
|
|
|
.if defined(KERNEL_EXTRA_INSTALL)
|
2015-10-17 05:49:07 +00:00
|
|
|
${INSTALL} -p -m 555 -o ${KMODOWN} -g ${KMODGRP} ${KERNEL_EXTRA_INSTALL} ${DESTDIR}${KODIR}/
|
2007-03-23 21:55:59 +00:00
|
|
|
.endif
|
|
|
|
|
|
|
|
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2002-04-24 11:26:19 +00:00
|
|
|
kernel-reinstall:
|
2001-11-02 21:34:20 +00:00
|
|
|
@-chflags -R noschg ${DESTDIR}${KODIR}
|
2015-10-17 05:49:07 +00:00
|
|
|
${INSTALL} -p -m 555 -o ${KMODOWN} -g ${KMODGRP} ${KERNEL_KO} ${DESTDIR}${KODIR}/
|
2012-07-03 05:01:00 +00:00
|
|
|
.if defined(DEBUG) && !defined(INSTALL_NODEBUG) && ${MK_KERNEL_SYMBOLS} != "no"
|
2015-10-17 05:49:07 +00:00
|
|
|
${INSTALL} -p -m 555 -o ${KMODOWN} -g ${KMODGRP} ${KERNEL_KO}.debug ${DESTDIR}${KERN_DEBUGDIR}${KODIR}/
|
2005-10-24 23:54:39 +00:00
|
|
|
.endif
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2005-03-29 11:23:25 +00:00
|
|
|
config.o env.o hints.o vers.o vnode_if.o:
|
2001-11-02 21:34:20 +00:00
|
|
|
${NORMAL_C}
|
2011-11-30 18:11:49 +00:00
|
|
|
${NORMAL_CTFCONVERT}
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2005-03-29 11:23:25 +00:00
|
|
|
config.ln env.ln hints.ln vers.ln vnode_if.ln:
|
2003-06-14 17:28:13 +00:00
|
|
|
${NORMAL_LINT}
|
|
|
|
|
2001-11-02 21:34:20 +00:00
|
|
|
vers.c: $S/conf/newvers.sh $S/sys/param.h ${SYSTEM_DEP}
|
2004-09-17 09:17:33 +00:00
|
|
|
MAKE=${MAKE} sh $S/conf/newvers.sh ${KERN_IDENT}
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2002-03-01 01:21:29 +00:00
|
|
|
vnode_if.c: $S/tools/vnode_if.awk $S/kern/vnode_if.src
|
|
|
|
${AWK} -f $S/tools/vnode_if.awk $S/kern/vnode_if.src -c
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2006-10-26 19:04:20 +00:00
|
|
|
vnode_if.h vnode_if_newproto.h vnode_if_typedef.h: $S/tools/vnode_if.awk \
|
|
|
|
$S/kern/vnode_if.src
|
|
|
|
vnode_if.h: vnode_if_newproto.h vnode_if_typedef.h
|
2002-03-01 01:21:29 +00:00
|
|
|
${AWK} -f $S/tools/vnode_if.awk $S/kern/vnode_if.src -h
|
2006-10-26 19:04:20 +00:00
|
|
|
vnode_if_newproto.h:
|
2004-12-01 23:16:38 +00:00
|
|
|
${AWK} -f $S/tools/vnode_if.awk $S/kern/vnode_if.src -p
|
2006-10-26 19:04:20 +00:00
|
|
|
vnode_if_typedef.h:
|
2004-12-01 23:16:38 +00:00
|
|
|
${AWK} -f $S/tools/vnode_if.awk $S/kern/vnode_if.src -q
|
2001-11-02 21:34:20 +00:00
|
|
|
|
2015-08-13 15:16:34 +00:00
|
|
|
.if ${MFS_IMAGE:Uno} != "no"
|
Fix MFS builds when both MD_ROOT_SIZE and MFS_IMAGE are specified
MD_ROOT_SIZE and embed_mfs.sh were basically retired as part of
https://reviews.freebsd.org/D2903 .
However, when building a kernel with 'options MD_ROOT_SIZE' specified, this
results in a non-working MFS, as within sys/dev/md/md.c we fall within the
wrong # ifdef.
This patch implements the following:
* Allow kernels to be built without the MD_ROOT_SIZE option, which results
in a kernel built as per D2903.
* Allow kernels to be built with the MD_ROOT_SIZE option, which results
in a kernel built similarly to the pre-D2903 way, with the following
differences:
* The MFS is now put in a separate section within the kernel (oldmfs,
so it differs from the mfs section introduced by D2903).
* embed_mfs.sh is changed, so it looks up the oldmfs section within the
kernel, gets its size and offset, sees if the MFS will fit within the
allocated oldmfs section and only if all is well does a dd of the MFS
image into the kernel.
Submitted by: Stanislav Galabov <sgalabov@gmail.com>
Reviewed by: brooks, imp
Differential Revision: https://reviews.freebsd.org/D5093
2016-02-02 07:02:51 +00:00
|
|
|
.if empty(MD_ROOT_SIZE_CONFIGURED)
|
2015-08-13 15:16:34 +00:00
|
|
|
# Generate an object file from the file system image to embed in the kernel
|
|
|
|
# via linking. Make sure the contents are in the mfs section and rename the
|
|
|
|
# start/end/size variables to __start_mfs, __stop_mfs, and mfs_size,
|
|
|
|
# respectively.
|
|
|
|
embedfs_${MFS_IMAGE:T:R}.o: ${MFS_IMAGE}
|
|
|
|
${OBJCOPY} --input-target binary \
|
|
|
|
--output-target ${EMBEDFS_FORMAT.${MACHINE_ARCH}} \
|
|
|
|
--binary-architecture ${EMBEDFS_ARCH.${MACHINE_ARCH}} \
|
|
|
|
${MFS_IMAGE} ${.TARGET}
|
|
|
|
${OBJCOPY} \
|
|
|
|
--rename-section .data=mfs,contents,alloc,load,readonly,data \
|
|
|
|
--redefine-sym \
|
|
|
|
_binary_${MFS_IMAGE:C,[^[:alnum:]],_,g}_size=__mfs_root_size \
|
|
|
|
--redefine-sym \
|
|
|
|
_binary_${MFS_IMAGE:C,[^[:alnum:]],_,g}_start=mfs_root \
|
|
|
|
--redefine-sym \
|
|
|
|
_binary_${MFS_IMAGE:C,[^[:alnum:]],_,g}_end=mfs_root_end \
|
|
|
|
${.TARGET}
|
|
|
|
.endif
|
Fix MFS builds when both MD_ROOT_SIZE and MFS_IMAGE are specified
MD_ROOT_SIZE and embed_mfs.sh were basically retired as part of
https://reviews.freebsd.org/D2903 .
However, when building a kernel with 'options MD_ROOT_SIZE' specified, this
results in a non-working MFS, as within sys/dev/md/md.c we fall within the
wrong # ifdef.
This patch implements the following:
* Allow kernels to be built without the MD_ROOT_SIZE option, which results
in a kernel built as per D2903.
* Allow kernels to be built with the MD_ROOT_SIZE option, which results
in a kernel built similarly to the pre-D2903 way, with the following
differences:
* The MFS is now put in a separate section within the kernel (oldmfs,
so it differs from the mfs section introduced by D2903).
* embed_mfs.sh is changed, so it looks up the oldmfs section within the
kernel, gets its size and offset, sees if the MFS will fit within the
allocated oldmfs section and only if all is well does a dd of the MFS
image into the kernel.
Submitted by: Stanislav Galabov <sgalabov@gmail.com>
Reviewed by: brooks, imp
Differential Revision: https://reviews.freebsd.org/D5093
2016-02-02 07:02:51 +00:00
|
|
|
.endif
|
2015-08-13 15:16:34 +00:00
|
|
|
|
2003-06-22 17:57:56 +00:00
|
|
|
# XXX strictly, everything depends on Makefile because changes to ${PROF}
|
|
|
|
# only appear there, but we don't handle that.
|
2003-06-14 17:28:13 +00:00
|
|
|
|
2003-02-28 22:12:17 +00:00
|
|
|
.include "kern.mk"
|