diff --git a/INSTALL.swift b/INSTALL.swift
index 1782b75e34e2028110717d7873bc2c97365f8240..999a8d3655fa14a8ba1dcbd430b5146cc55ba791 100644
--- a/INSTALL.swift
+++ b/INSTALL.swift
@@ -96,18 +96,21 @@ SWIFT depends on a number of third party libraries that should be available
before you can build it.
- - HDF5: a HDF5 library (v. 1.8.x or higher) is required to read and
- write particle data. One of the commands "h5cc" or "h5pcc"
- should be available. If "h5pcc" is located them a parallel
- HDF5 built for the version of MPI located should be
- provided. If the command is not available then it can be
- located using the "--with-hfd5" configure option. The value
- should be the full path to the "h5cc" or "h5pcc" commands.
-
-
- - MPI: to run on more than one node an MPI library that fully
- supports MPI_THREAD_MULTIPLE. Before running configure the
- "mpirun" command should be available in the shell. If your
+ - HDF5:
+ A HDF5 library (v. 1.8.x or higher) is required to read and
+ write particle data. One of the commands "h5cc" or "h5pcc"
+ should be available. If "h5pcc" is located then a parallel
+ HDF5 built for the version of MPI located should be
+ provided. If the command is not available then it can be
+ located using the "--with-hdf5" configure option. The value
+ should be the full path to the "h5cc" or "h5pcc" commands.
+ SWIFT makes effective use of parallel HDF5 when running on more than
+ one node, so this option is highly recommended.
+
+ - MPI:
+ To run on more than one node an MPI library that fully
+ supports MPI_THREAD_MULTIPLE is required. Before running configure
+ the "mpirun" command should be available in the shell. If your
command isn't called "mpirun" then define the "MPIRUN"
environment variable, either in the shell or when running
configure.
@@ -116,57 +119,69 @@ before you can build it.
much like the CC one. Use this when your MPI compiler has a
none-standard name.
- - GSL: To use cosmological time integration, a version of the GSL
- must be available.
+ - GSL:
+ To use cosmological time integration, a version of the GSL
+ must be available.
- - libtool: The build system relies on libtool as well as the other autotools.
+ - FFTW 3.x:
+ To run with periodic gravity forces, a build of the FFTW 3
+ library must be available. Note that SWIFT does not make use
+ of the parallel capability of FFTW. Calculations are done by
+ single MPI nodes independently.
-
- Optional Dependencies
- =====================
+- libtool:
+ The build system relies on libtool as well as the other autotools.
- - METIS: a build of the METIS library can be optionally used to
- optimize the load between MPI nodes (requires an MPI
- library). This should be found in the standard installation
- directories, or pointed at using the "--with-metis"
- configuration option. In this case the top-level
- installation directory of the METIS build should be
- given. Note to use METIS you should at least supply
- "--with-metis".
+ Optional Dependencies
+ =====================
- - libNUMA: a build of the NUMA library can be used to pin the threads
- to the physical core of the machine SWIFT is running
- on. This is not always necessary as the OS scheduler may
- do a good job at distributing the threads among the
- different cores on each computing node.
+ - METIS:
+ a build of the METIS library can be optionally used to
+ optimize the load between MPI nodes (requires an MPI
+ library). This should be found in the standard installation
+ directories, or pointed at using the "--with-metis"
+ configuration option. In this case the top-level installation
+ directory of the METIS build should be given. Note to use
+ METIS you should supply at least "--with-metis".
- - TCMalloc: a build of the TCMalloc library (part of gperftools) can
- be used to obtain faster allocations than the standard C
- malloc function part of glibc. The option "-with-tcmalloc"
- should be passed to the configuration script to use it.
+- libNUMA:
+ a build of the NUMA library can be used to pin the threads to
+ the physical core of the machine SWIFT is running on. This is
+ not always necessary as the OS scheduler may do a good job at
+ distributing the threads among the different cores on each
+ computing node.
+ - tcmalloc / jemalloc / TBBmalloc:
+ a build of the tcmalloc library (part of gperftools), jemalloc
+ or TBBmalloc can be used be used to obtain faster and more
+ scalable allocations than the standard C malloc function part
+ of glibc. Using one of these is highly recommended on systems
+ with many cores per node. One of the options
+ "--with-tcmalloc", "--with-jemalloc" or "--with-tbbmalloc"
+ should be passed to the configuration script to use it.
- - gperftools: a build of gperftools can be used to obtain good
- profiling of the code. The option "-with-profiler"
- needs to be passed to the configuration script to use
- it.
+ - gperftools:
+ a build of gperftools can be used to obtain good profiling of
+ the code. The option "--with-profiler" needs to be passed to
+ the configuration script to use it.
+ - DOXYGEN:
+ the doxygen library is required to create the SWIFT API
+ documentation.
- - DOXYGEN: the doxygen library is required to create the SWIFT API
- documentation.
+ - python:
+ Examples and solution script use python and rely on the numpy
+ library version 1.8.2 or higher.
- - python: Examples and solution script use python and rely on the
- numpy library version 1.8.2 or higher.
SWIFT Coding style
==================
-The SWIFT source code is using a variation of the 'Google' style. The
-script 'format.sh' in the root directory applies the clang-format-3.8
-tool with our style choices to all the SWIFT C source file. Please
-apply the formatting script to the files before submitting a merge
-request.
+The SWIFT source code uses a variation of 'Google' style. The script
+'format.sh' in the root directory applies the clang-format-3.8 tool with our
+style choices to all the SWIFT C source file. Please apply the formatting
+script to the files before submitting a merge request.
diff --git a/configure.ac b/configure.ac
index 583169d146f8c7e94c45c941e006243badcefe51..316155b4962cb853d4c3bb51e5ffbf6c1c0e038e 100644
--- a/configure.ac
+++ b/configure.ac
@@ -625,15 +625,51 @@ AC_SUBST([FFTW_LIBS])
AC_SUBST([FFTW_INCS])
AM_CONDITIONAL([HAVEFFTW],[test -n "$FFTW_LIBS"])
+# Check for -lprofiler usually part of the gperftools along with tcmalloc.
+have_profiler="no"
+AC_ARG_WITH([profiler],
+ [AS_HELP_STRING([--with-profiler=PATH],
+ [use cpu profiler library or specify the directory with lib @<:@yes/no@:>@]
+ )],
+ [with_profiler="$withval"],
+ [with_profiler="no"]
+)
+if test "x$with_profiler" != "xno"; then
+ if test "x$with_profiler" != "xyes" -a "x$with_profiler" != "x"; then
+ proflibs="-L$with_profiler -lprofiler"
+ else
+ proflibs="-lprofiler"
+ fi
+ AC_CHECK_LIB([profiler],[ProfilerFlush],
+ [have_profiler="yes"
+ AC_DEFINE([WITH_PROFILER],1,[Link against the gperftools profiling library.])],
+ [have_profiler="no"], $proflibs)
+
+ if test "$have_profiler" = "yes"; then
+ PROFILER_LIBS="$proflibs"
+ else
+ PROFILER_LIBS=""
+ fi
+fi
+AC_SUBST([PROFILER_LIBS])
+AM_CONDITIONAL([HAVEPROFILER],[test -n "$PROFILER_LIBS"])
+
+# Check for special allocators
+have_special_allocator="no"
+
# Check for tcmalloc a fast malloc that is part of the gperftools.
have_tcmalloc="no"
AC_ARG_WITH([tcmalloc],
- [AS_HELP_STRING([--with-tcmalloc],
+ [AS_HELP_STRING([--with-tcmalloc=PATH],
[use tcmalloc library or specify the directory with lib @<:@yes/no@:>@]
)],
[with_tcmalloc="$withval"],
[with_tcmalloc="no"]
)
+if test "x$with_tcmalloc" != "xno" -a "x$have_special_allocator" != "xno"; then
+ AC_MSG_ERROR("Cannot activate more than one alternative malloc library")
+fi
+
if test "x$with_tcmalloc" != "xno"; then
if test "x$with_tcmalloc" != "xyes" -a "x$with_tcmalloc" != "x"; then
tclibs="-L$with_tcmalloc -ltcmalloc"
@@ -657,10 +693,17 @@ if test "x$with_tcmalloc" != "xno"; then
if test "$have_tcmalloc" = "yes"; then
TCMALLOC_LIBS="$tclibs"
- # These are recommended for GCC.
- if test "$ax_cv_c_compiler_vendor" = "gnu"; then
- CFLAGS="$CFLAGS -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-free"
- fi
+ AC_DEFINE([HAVE_TCMALLOC],1,[The tcmalloc library appears to be present.])
+
+ have_special_allocator="tcmalloc"
+
+ # Prevent compilers that replace the calls with built-ins (GNU 99) from doing so.
+ case "$ax_cv_c_compiler_vendor" in
+ intel | gnu | clang)
+ CFLAGS="$CFLAGS -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-free"
+ ;;
+ esac
+
else
TCMALLOC_LIBS=""
fi
@@ -668,44 +711,19 @@ fi
AC_SUBST([TCMALLOC_LIBS])
AM_CONDITIONAL([HAVETCMALLOC],[test -n "$TCMALLOC_LIBS"])
-# Check for -lprofiler usually part of the gperftools along with tcmalloc.
-have_profiler="no"
-AC_ARG_WITH([profiler],
- [AS_HELP_STRING([--with-profiler],
- [use cpu profiler library or specify the directory with lib @<:@yes/no@:>@]
- )],
- [with_profiler="$withval"],
- [with_profiler="no"]
-)
-if test "x$with_profiler" != "xno"; then
- if test "x$with_profiler" != "xyes" -a "x$with_profiler" != "x"; then
- proflibs="-L$with_profiler -lprofiler"
- else
- proflibs="-lprofiler"
- fi
- AC_CHECK_LIB([profiler],[ProfilerFlush],
- [have_profiler="yes"
- AC_DEFINE([WITH_PROFILER],1,[Link against the gperftools profiling library.])],
- [have_profiler="no"], $proflibs)
-
- if test "$have_profiler" = "yes"; then
- PROFILER_LIBS="$proflibs"
- else
- PROFILER_LIBS=""
- fi
-fi
-AC_SUBST([PROFILER_LIBS])
-AM_CONDITIONAL([HAVEPROFILER],[test -n "$PROFILER_LIBS"])
-
# Check for jemalloc another fast malloc that is good with contention.
have_jemalloc="no"
AC_ARG_WITH([jemalloc],
- [AS_HELP_STRING([--with-jemalloc],
+ [AS_HELP_STRING([--with-jemalloc=PATH],
[use jemalloc library or specify the directory with lib @<:@yes/no@:>@]
)],
[with_jemalloc="$withval"],
[with_jemalloc="no"]
)
+if test "x$with_jemalloc" != "xno" -a "x$have_special_allocator" != "xno"; then
+ AC_MSG_ERROR("Cannot activate more than one alternative malloc library")
+fi
+
if test "x$with_jemalloc" != "xno"; then
if test "x$with_jemalloc" != "xyes" -a "x$with_jemalloc" != "x"; then
jelibs="-L$with_jemalloc -ljemalloc"
@@ -717,6 +735,18 @@ if test "x$with_jemalloc" != "xno"; then
if test "$have_jemalloc" = "yes"; then
JEMALLOC_LIBS="$jelibs"
+
+ AC_DEFINE([HAVE_JEMALLOC],1,[The jemalloc library appears to be present.])
+
+ have_special_allocator="jemalloc"
+
+ # Prevent compilers that replace the regular calls with built-ins (GNU 99) from doing so.
+ case "$ax_cv_c_compiler_vendor" in
+ intel | gnu | clang)
+ CFLAGS="$CFLAGS -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-free"
+ ;;
+ esac
+
else
JEMALLOC_LIBS=""
fi
@@ -724,11 +754,49 @@ fi
AC_SUBST([JEMALLOC_LIBS])
AM_CONDITIONAL([HAVEJEMALLOC],[test -n "$JEMALLOC_LIBS"])
-# Don't allow both tcmalloc and jemalloc.
-if test "x$have_tcmalloc" != "xno" -a "x$have_jemalloc" != "xno"; then
- AC_MSG_ERROR([Cannot use tcmalloc at same time as jemalloc])
+# Check for tbbmalloc, Intel's fast and parallel allocator
+have_tbbmalloc="no"
+AC_ARG_WITH([tbbmalloc],
+ [AS_HELP_STRING([--with-tbbmalloc=PATH],
+ [use tbbmalloc library or specify the directory with lib @<:@yes/no@:>@]
+ )],
+ [with_tbbmalloc="$withval"],
+ [with_tbbmalloc="no"]
+)
+if test "x$with_tbbmalloc" != "xno" -a "x$have_special_allocator" != "xno"; then
+ AC_MSG_ERROR("Cannot activate more than one alternative malloc library")
fi
+if test "x$with_tbbmalloc" != "xno"; then
+ if test "x$with_tbbmalloc" != "xyes" -a "x$with_tbbmalloc" != "x"; then
+ tbblibs="-L$with_tbbmalloc -ltbbmalloc_proxy -ltbbmalloc"
+ else
+ tbblibs="-ltbbmalloc_proxy -ltbbmalloc"
+ fi
+ AC_CHECK_LIB([tbbmalloc],[scalable_malloc],[have_tbbmalloc="yes"],[have_tbbmalloc="no"],
+ $tbblibs)
+
+ if test "$have_tbbmalloc" = "yes"; then
+ TBBMALLOC_LIBS="$tbblibs"
+
+ AC_DEFINE([HAVE_TBBMALLOC],1,[The TBBmalloc library appears to be present.])
+
+ have_special_allocator="TBBmalloc"
+
+ # Prevent compilers that replace the calls with built-ins (GNU 99) from doing so.
+ case "$ax_cv_c_compiler_vendor" in
+ intel | gnu | clang)
+ CFLAGS="$CFLAGS -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-free"
+ ;;
+ esac
+
+ else
+ TBBMALLOC_LIBS=""
+ fi
+fi
+AC_SUBST([TBBMALLOC_LIBS])
+AM_CONDITIONAL([HAVETBBMALLOC],[test -n "$TBBMALLOC_LIBS"])
+
# Check for HDF5. This is required.
AX_LIB_HDF5
if test "$with_hdf5" != "yes"; then
@@ -750,7 +818,8 @@ if test "$with_hdf5" = "yes"; then
if test "$enable_parallel_hdf5" = "yes"; then
AC_MSG_CHECKING([for HDF5 parallel support])
-# Check if the library is capable, the header should define H5_HAVE_PARALLEL.
+
+ # Check if the library is capable, the header should define H5_HAVE_PARALLEL.
AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
#include "hdf5.h"
@@ -837,7 +906,6 @@ if test "$enable_warn" != "no"; then
;;
esac
-
# Add a "choke on warning" flag if it exists
if test "$enable_warn" = "error"; then
case "$ax_cv_c_compiler_vendor" in
@@ -1228,22 +1296,21 @@ AC_MSG_RESULT([
$PACKAGE_NAME v.$PACKAGE_VERSION
- Compiler : $CC
- - vendor : $ax_cv_c_compiler_vendor
- - version : $ax_cv_c_compiler_version
- - flags : $CFLAGS
- MPI enabled : $enable_mpi
- HDF5 enabled : $with_hdf5
- - parallel : $have_parallel_hdf5
- Metis enabled : $have_metis
- FFTW3 enabled : $have_fftw
- GSL enabled : $have_gsl
- libNUMA enabled : $have_numa
- GRACKLE enabled : $have_grackle
- Using tcmalloc : $have_tcmalloc
- Using jemalloc : $have_jemalloc
- CPU profiler : $have_profiler
- Pthread barriers : $have_pthread_barrier
+ Compiler : $CC
+ - vendor : $ax_cv_c_compiler_vendor
+ - version : $ax_cv_c_compiler_version
+ - flags : $CFLAGS
+ MPI enabled : $enable_mpi
+ HDF5 enabled : $with_hdf5
+ - parallel : $have_parallel_hdf5
+ Metis enabled : $have_metis
+ FFTW3 enabled : $have_fftw
+ GSL enabled : $have_gsl
+ libNUMA enabled : $have_numa
+ GRACKLE enabled : $have_grackle
+ Special allocators : $have_special_allocator
+ CPU profiler : $have_profiler
+ Pthread barriers : $have_pthread_barrier
Hydro scheme : $with_hydro
Dimensionality : $with_dimension
diff --git a/examples/Makefile.am b/examples/Makefile.am
index 95057c9bee7d3b4cc001d6c19ca39cab5d8544c4..9de393ba6c77f9bd50e1d8aff12c0d83ba8e7ddc 100644
--- a/examples/Makefile.am
+++ b/examples/Makefile.am
@@ -24,7 +24,7 @@ AM_CFLAGS = -I$(top_srcdir)/src $(HDF5_CPPFLAGS) $(GSL_INCS) $(FFTW_INCS)
AM_LDFLAGS = $(HDF5_LDFLAGS)
# Extra libraries.
-EXTRA_LIBS = $(HDF5_LIBS) $(FFTW_LIBS) $(PROFILER_LIBS) $(TCMALLOC_LIBS) $(JEMALLOC_LIBS) $(GRACKLE_LIBS) $(GSL_LIBS)
+EXTRA_LIBS = $(HDF5_LIBS) $(FFTW_LIBS) $(PROFILER_LIBS) $(TCMALLOC_LIBS) $(JEMALLOC_LIBS) $(TBBMALLOC_LIBS) $(GRACKLE_LIBS) $(GSL_LIBS)
# MPI libraries.
MPI_LIBS = $(METIS_LIBS) $(MPI_THREAD_LIBS)
diff --git a/src/Makefile.am b/src/Makefile.am
index 374d2c4569efe0e07ca1b2558b679b33564da432..d89133d3fa2c23b5740b35574f91f3fe6808bfae 100644
--- a/src/Makefile.am
+++ b/src/Makefile.am
@@ -25,7 +25,7 @@ AM_LDFLAGS = $(HDF5_LDFLAGS) $(FFTW_LIBS) -version-info 0:0:0
GIT_CMD = @GIT_CMD@
# Additional dependencies for shared libraries.
-EXTRA_LIBS = $(HDF5_LIBS) $(PROFILER_LIBS) $(TCMALLOC_LIBS) $(JEMALLOC_LIBS) $(GRACKLE_LIB) $(GSL_LIBS)
+EXTRA_LIBS = $(HDF5_LIBS) $(FFTW_LIBS) $(PROFILER_LIBS) $(TCMALLOC_LIBS) $(JEMALLOC_LIBS) $(TBBMALLOC_LIBS) $(GRACKLE_LIB) $(GSL_LIBS)
# MPI libraries.
MPI_LIBS = $(METIS_LIBS) $(MPI_THREAD_LIBS)
diff --git a/src/common_io.c b/src/common_io.c
index 8b173adb7b5e5a014b0967b4fd04aef5ee6606e9..88374e30cccfa619352aed5c7e401bf163d6db63 100644
--- a/src/common_io.c
+++ b/src/common_io.c
@@ -340,6 +340,7 @@ void io_write_code_description(hid_t h_file) {
io_write_attribute_s(h_grpcode, "CFLAGS", compilation_cflags());
io_write_attribute_s(h_grpcode, "HDF5 library version", hdf5_version());
io_write_attribute_s(h_grpcode, "Thread barriers", thread_barrier_version());
+ io_write_attribute_s(h_grpcode, "Allocators", allocator_version());
#ifdef HAVE_FFTW
io_write_attribute_s(h_grpcode, "FFTW library version", fftw3_version());
#endif
diff --git a/src/version.c b/src/version.c
index 54749721de96bde010f56965152c536b08672230..a0a2ae42a5ce1a24ffce43f7ae3f59e1d6ed4168 100644
--- a/src/version.c
+++ b/src/version.c
@@ -368,6 +368,26 @@ const char *thread_barrier_version(void) {
return version;
}
+/**
+ * @brief return the allocator library used in SWIFT.
+ *
+ * @result description of the allocation library
+ */
+const char *allocator_version(void) {
+
+ static char version[256] = {0};
+#if defined(HAVE_TBBMALLOC)
+ sprintf(version, "TBB malloc");
+#elif defined(HAVE_TCMALLOC)
+ sprintf(version, "tc-malloc");
+#elif defined(HAVE_JEMALLOC)
+ sprintf(version, "je-malloc");
+#else
+ sprintf(version, "Compiler version (probably glibc)");
+#endif
+ return version;
+}
+
/**
* @brief Prints a greeting message to the standard output containing code
* version and revision number
diff --git a/src/version.h b/src/version.h
index 3163f242c50e56c64cc709b13dfe926f93672a00..44119b6a3bbdf57c3f0195bae5ff329d05c61fd5 100644
--- a/src/version.h
+++ b/src/version.h
@@ -36,6 +36,7 @@ const char* hdf5_version(void);
const char* fftw3_version(void);
const char* libgsl_version(void);
const char* thread_barrier_version(void);
+const char* allocator_version(void);
void greetings(void);
#endif /* SWIFT_VERSION_H */
diff --git a/tests/Makefile.am b/tests/Makefile.am
index 891eef3f518f83c17b66623e3dac1832512d31f3..564076ed501409083db017de4c79c17f3b2a7c15 100644
--- a/tests/Makefile.am
+++ b/tests/Makefile.am
@@ -17,7 +17,7 @@
# Add the source directory and the non-standard paths to the included library headers to CFLAGS
AM_CFLAGS = -I$(top_srcdir)/src $(HDF5_CPPFLAGS) $(GSL_INCS) $(FFTW_INCS)
-AM_LDFLAGS = ../src/.libs/libswiftsim.a $(HDF5_LDFLAGS) $(HDF5_LIBS) $(FFTW_LIBS) $(GRACKLE_LIBS) $(GSL_LIBS) $(PROFILER_LIBS)
+AM_LDFLAGS = ../src/.libs/libswiftsim.a $(HDF5_LDFLAGS) $(HDF5_LIBS) $(FFTW_LIBS) $(TCMALLOC_LIBS) $(JEMALLOC_LIBS) $(TBBMALLOC_LIBS) $(GRACKLE_LIBS) $(GSL_LIBS) $(PROFILER_LIBS)
# List of programs and scripts to run in the test suite
TESTS = testGreetings testMaths testReading.sh testSingle testKernel testSymmetry \