diff --git a/.gitignore b/.gitignore index 4cde3de..782109a 100644 --- a/.gitignore +++ b/.gitignore @@ -1,17 +1,35 @@ -*.o -*.so -#* -*~ -*.pyc -*.pyo - +**.o +**.lo +**.la +**.so +**/#* +**~ +**.pyc +**.pyo +.libs +**/.deps +**/.dirstamp +libsharp-uninstalled.pc +libsharp-uninstalled.sh +libsharp.pc +libsharp.pc.in +perf.data* /auto /autom4te.cache -/config.log -/config.status -/config/config.auto +/m4 +config.log +config.guess +config.status +config.sub +ltmain.sh +compile +libtool +missing +/comp /configure -/sharp_oracle.inc - -/python/libsharp/libsharp.c -/python/libsharp/libsharp_mpi.c +/Makefile +/Makefile.in +/aclocal.m4 +/ar-lib +/depcomp +/install-sh diff --git a/COMPILE b/COMPILE new file mode 100644 index 0000000..8a5f3cd --- /dev/null +++ b/COMPILE @@ -0,0 +1,85 @@ +Libsharp is configured, compiled and installed using GNU autotools. + +If you have cloned the libsharp repository, you have to run +"autoreconf -i" before starting the configuration, which requires several +GNU developer tools to be available on your system. + +When using a release tarball, configuration is done via + +[CC=...] [CFLAGS=...] ./configure + +The following sections briefly describe possible choices for compilers and +flags. + + +Fast math +--------- + +Specifying "-ffast-math" is important for all compilers, since it allows the +compiler to fuse multiplications and additions into FMA instructions, which is +forbidden by the C99 standard. Since FMAs are a central aspect of the algorithm, +they are needed for optimum performance. + +If you are calling libsharp from other code which requires strict adherence +to the C99 standard, you should still be able to compile libsharp with +"-ffast-math" without any problems. + + +Runtime CPU selection with gcc +------------------------------ + +When using a recent gcc (6.0 and newer) on an x86_64 platform, the build +machinery will compile the time-critical functions for several different +architectures (SSE2, AVX, AVX2, FMA3, FMA4, AVX512F), and the appropriate +implementation will be selected at runtime. +This only happens if you do _not_ explicitly specify a target architecture via +the compiler flags. I.e., please do _not_ specify "-march=native" or +"-mtarget=avx" or similar if you want a portable binary that will run +efficiently on different x86_64 CPUs. +If you are compiling libsharp for a particular target CPU only, or if you are +using a different compiler, however, "-march-native" should be used. The +resulting binary will most likely not run on other computers, though. + + +OpenMP +------ + +OpenMP should be switched on for maximum performance, and at runtime +OMP_NUM_THREADS should be set to the number of hardware threads (not physical +cores) of the system. +(Usually this is already the default setting when OMP_NUM_THREADS is not +specified.) + + +MPI +--- + +MPI support is enabled by using the MPI compiler (typically "mpicc") _and_ +adding the flag "-DUSE_MPI". +When using MPI and OpenMP simultaneously, the product of MPI tasks per node +and OMP_NUM_THREADS should be equal to the number of hardware threads available +on the node. One MPI task per node should result in the best performance. + + +Example configure invocations +============================= + +GCC, OpenMP, portable binary: +CFLAGS="-std=c99 -O3 -ffast-math -fopenmp" ./configure + +GCC, no OpenMP, portable binary: +CFLAGS="-std=c99 -O3 -ffast-math" ./configure + +Clang, OpenMP, nonportable binary: +CC=clang CFLAGS="-std=c99 -O3 -march=native -ffast-math -fopenmp" ./configure + +Intel C compiler, OpenMP, nonportable binary: +CC=icc CFLAGS="-std=c99 -O3 -march=native -ffast-math -fopenmp" ./configure + +MPI support, nonportable binary: +CC=mpicc CFLAGS="-DUSE_MPI -std=c99 -O3 -march=native -ffast-math" ./configure + +Additional GCC flags for pedantic warning and debugging: + +-Wall -Wextra -Wshadow -Wmissing-prototypes -Wfatal-errors -pedantic -g +-fsanitize=address diff --git a/Makefile b/Makefile deleted file mode 100644 index 5a3184c..0000000 --- a/Makefile +++ /dev/null @@ -1,80 +0,0 @@ -SHARP_TARGET?=auto -ifndef SHARP_TARGET - SHARP_TARGET:=$(error SHARP_TARGET undefined. Please see README.compilation for help)UNDEFINED -endif - -default: compile_all -SRCROOT:=$(shell pwd) -include $(SRCROOT)/config/config.$(SHARP_TARGET) -include $(SRCROOT)/config/rules.common - -all_hdr:= -all_lib:= -all_cbin:= - -FULL_INCLUDE:= - -include c_utils/planck.make -include libfftpack/planck.make -include libsharp/planck.make -include docsrc/planck.make - -CYTHON_MODULES=python/libsharp/libsharp.so $(if $(MPI_CFLAGS), python/libsharp/libsharp_mpi.so) - -$(all_lib): %: | $(LIBDIR)_mkdir - @echo "# creating library $*" - $(ARCREATE) $@ $^ - -$(all_cbin): %: | $(BINDIR)_mkdir - @echo "# linking C binary $*" - $(CL) -o $@ $^ $(CLFLAGS) - -compile_all: $(all_cbin) hdrcopy - -hdrclean: - @if [ -d $(INCDIR) ]; then rm -rf $(INCDIR)/* ; fi - -hdrcopy: | $(INCDIR)_mkdir - @if [ "$(all_hdr)" ]; then cp -p $(all_hdr) $(INCDIR); fi - -$(notdir $(all_cbin)) : % : $(BINDIR)/% - -test: compile_all - $(BINDIR)/sharp_testsuite acctest && \ - $(BINDIR)/sharp_testsuite test healpix 2048 -1 1024 -1 0 1 && \ - $(BINDIR)/sharp_testsuite test fejer1 2047 -1 -1 4096 2 1 && \ - $(BINDIR)/sharp_testsuite test gauss 2047 -1 -1 4096 0 2 - -perftest: compile_all - $(BINDIR)/sharp_testsuite test healpix 2048 -1 1024 -1 0 1 && \ - $(BINDIR)/sharp_testsuite test gauss 63 -1 -1 128 0 1 && \ - $(BINDIR)/sharp_testsuite test gauss 127 -1 -1 256 0 1 && \ - $(BINDIR)/sharp_testsuite test gauss 255 -1 -1 512 0 1 && \ - $(BINDIR)/sharp_testsuite test gauss 511 -1 -1 1024 0 1 && \ - $(BINDIR)/sharp_testsuite test gauss 1023 -1 -1 2048 0 1 && \ - $(BINDIR)/sharp_testsuite test gauss 2047 -1 -1 4096 0 1 && \ - $(BINDIR)/sharp_testsuite test gauss 4095 -1 -1 8192 0 1 && \ - $(BINDIR)/sharp_testsuite test gauss 8191 -1 -1 16384 0 1 - -%.c: %.c.in -# Only do this if the md5sum changed, in order to avoid Python and Jinja -# dependency when not modifying the c.in file - grep `md5sum $< | cut -d ' ' -f 1` $@ || ./runjinja.py < $< > $@ - -genclean: - rm libsharp/sharp_legendre.c || exit 0 - -$(CYTHON_MODULES): %.so: %.pyx -ifndef PIC_CFLAGS - $(error Python extension must be built using the --enable-pic configure option.) -endif - cython $< - $(CC) $(DEBUG_CFLAGS) $(OPENMP_CFLAGS) $(PIC_CFLAGS) `python-config --cflags` -I$(INCDIR) -o $(<:.pyx=.o) -c $(<:.pyx=.c) - $(CL) -shared $(<:.pyx=.o) $(OPENMP_CFLAGS) $(CYTHON_OBJ) -L$(LIBDIR) -lsharp -lfftpack -lc_utils -L`python-config --prefix`/lib `python-config --ldflags` -o $@ - -python: $(all_lib) hdrcopy $(CYTHON_MODULES) - -# the following test files are automatic; the sht wrapper test -# must be run manually and requires MPI at the moment.. -pytest: python - cd python && nosetests --nocapture libsharp/tests/test_legendre_table.py libsharp/tests/test_legendre.py diff --git a/Makefile.am b/Makefile.am new file mode 100644 index 0000000..bcf53ff --- /dev/null +++ b/Makefile.am @@ -0,0 +1,49 @@ +ACLOCAL_AMFLAGS = -I m4 + +lib_LTLIBRARIES = libsharp.la + +src_sharp = \ + c_utils/c_utils.c \ + c_utils/c_utils.h \ + pocketfft/pocketfft.c \ + pocketfft/pocketfft.h \ + libsharp/sharp.c \ + libsharp/sharp_almhelpers.c \ + libsharp/sharp_core.c \ + libsharp/sharp_core_avx.c \ + libsharp/sharp_core_avx2.c \ + libsharp/sharp_core_fma.c \ + libsharp/sharp_core_fma4.c \ + libsharp/sharp_core_avx512f.c \ + libsharp/sharp_geomhelpers.c \ + libsharp/sharp_legendre_roots.c \ + libsharp/sharp_ylmgen_c.c \ + libsharp/sharp_internal.h \ + libsharp/sharp_legendre_roots.h \ + libsharp/sharp_vecsupport.h \ + libsharp/sharp_ylmgen_c.h + +include_HEADERS = \ + libsharp/sharp.h \ + libsharp/sharp_geomhelpers.h \ + libsharp/sharp_almhelpers.h \ + libsharp/sharp_cxx.h + +EXTRA_DIST = \ + libsharp/sharp_core_inc.c \ + runtest.sh + +libsharp_la_SOURCES = $(src_sharp) + +check_PROGRAMS = sharp_testsuite +sharp_testsuite_SOURCES = libsharp/sharp_testsuite.c c_utils/memusage.c c_utils/memusage.h c_utils/walltime_c.c c_utils/walltime_c.h +sharp_testsuite_LDADD = libsharp.la + +TESTS = runtest.sh + +AM_CFLAGS = -I$(top_srcdir)/c_utils -I$(top_srcdir)/libsharp @AM_CFLAGS@ + +pkgconfigdir = $(libdir)/pkgconfig +nodist_pkgconfig_DATA = @PACKAGE_NAME@.pc + +DISTCLEANFILES=@PACKAGE_NAME@.pc @PACKAGE_NAME@.pc.in @PACKAGE_NAME@-uninstalled.pc @PACKAGE_NAME@-uninstalled.sh diff --git a/README.md b/README.md index 24652b2..c993dd1 100644 --- a/README.md +++ b/README.md @@ -1,43 +1,35 @@ # Libsharp -*IMPORTANT NOTE*: It appears that the default branch upon cloning from -github.com/dagss/libsharp was an outdated 'dagss' branch instead of -the 'master' branch. To get the latest copy, -please do `git checkout master; git pull`. New clones are no longer affected. +Library for efficient spherical harmonic transforms at arbitrary spins, +supporting CPU vectorization, OpenMP and MPI. ## Paper https://arxiv.org/abs/1303.4945 +## News + +### January 2019 + +This update features significant speedups thanks to important algorithmic +discoveries by Keiichi Ishioka +(https://www.jstage.jst.go.jp/article/jmsj/96/2/96_2018-019/_article and +personal communication). + +These improvements reduce the fraction of CPU time spent on evaluating the +recurrences for Y_lm coefficients, which means that computing multiple +simultaneous SHTs no longer have a big performance advantage compared to SHTs +done one after the other. +As a consequence, libsharp support for simultaneous SHTs was dropped, making +its interface much simpler. + +With the proper compilers and flags (see the file COMPILE for details) libsharp +is now built with support for SSE2, AVX, AVX2, FMA3, FMA4 and AVX512f and the +appropriate implementation is selected dynamically at runtime. This should +provide a very significant performance boost for everyone using pre-compiled +portable binaries. + ## Compilation -GNU make is required for compilation. - -Libsharp compilation has been successfully tested with GNU and Intel compilers. -When using gcc, version 4.x is required [1]. -Since libsharp was written in standard C99, other compilers should work fine, -but SSE2/AVX support will most likely be deactivated. - -If you obtained libsharp directly from the git repository, you will also -need a copy of the GNU autotools. In this case, run "autoconf" in libsharp's -main directory before any other steps. -For libsharp releases distributed as a .tar.gz file, this step is not necessary. - -Afterwards, simply run "./configure"; if this fails, please refer to the output -of "./configure --help" for additional hints and, if necessary, provide -additional flags to the configure script. -Once the script finishes successfully, run "make" -(or "gmake"). This should install the compilation products in the -subdirectory "auto/". - -Documentation can be created by the command "(g)make doc". -However this requires the doxygen application to be installed -on your system. -The documentation will be created in the subdirectory doc/. - - -[1] Some versions of the gcc 4.4.x release series contain a bug which causes -the compiler to crash during libsharp compilation. This appears to be fixed -in the gcc 4.4.7 release. It is possible to work around this problem by adding -the compiler flag "-fno-tree-fre" after the other optimization flags - the -configure script should do this automatically. +The library uses the standard `autotools` mechanism for configuration, +compilation and installation. See the file `COMPILE` for configuration hints. diff --git a/c_utils/c_utils.c b/c_utils/c_utils.c index 96bd765..9344a6d 100644 --- a/c_utils/c_utils.c +++ b/c_utils/c_utils.c @@ -25,7 +25,7 @@ /* * Convenience functions * - * Copyright (C) 2008, 2009, 2010, 2011, 2012 Max-Planck-Society + * Copyright (C) 2008-2017 Max-Planck-Society * Author: Martin Reinecke */ @@ -44,7 +44,7 @@ void util_warn_ (const char *file, int line, const char *func, const char *msg) /* This function tries to avoid allocations with a total size close to a high power of two (called the "critical stride" here), by adding a few more bytes - if necssary. This lowers the probability that two arrays differ by a multiple + if necessary. This lowers the probability that two arrays differ by a multiple of the critical stride in their starting address, which in turn lowers the risk of cache line contention. */ static size_t manipsize(size_t sz) @@ -61,7 +61,7 @@ void *util_malloc_ (size_t sz) { void *res; if (sz==0) return NULL; - res = _mm_malloc(manipsize(sz),16); + res = _mm_malloc(manipsize(sz),32); UTIL_ASSERT(res,"_mm_malloc() failed"); return res; } diff --git a/c_utils/c_utils.h b/c_utils/c_utils.h index 0503449..01c64ad 100644 --- a/c_utils/c_utils.h +++ b/c_utils/c_utils.h @@ -25,7 +25,7 @@ /*! \file c_utils.h * Convenience functions * - * Copyright (C) 2008, 2009, 2010, 2011 Max-Planck-Society + * Copyright (C) 2008-2017 Max-Planck-Society * \author Martin Reinecke * \note This file should only be included from .c files, NOT from .h files. */ @@ -144,4 +144,10 @@ void util_free_ (void *ptr); } #endif +#ifdef __GNUC__ +#define NOINLINE __attribute__((noinline)) +#else +#define NOINLINE +#endif + #endif diff --git a/c_utils/planck.make b/c_utils/planck.make deleted file mode 100644 index 4f0ccb1..0000000 --- a/c_utils/planck.make +++ /dev/null @@ -1,18 +0,0 @@ -PKG:=c_utils - -SD:=$(SRCROOT)/$(PKG) -OD:=$(BLDROOT)/$(PKG) - -FULL_INCLUDE+= -I$(SD) - -HDR_$(PKG):=$(SD)/*.h -LIB_$(PKG):=$(LIBDIR)/libc_utils.a - -OBJ:=c_utils.o walltime_c.o memusage.o -OBJ:=$(OBJ:%=$(OD)/%) - -$(OBJ): $(HDR_$(PKG)) | $(OD)_mkdir -$(LIB_$(PKG)): $(OBJ) - -all_hdr+=$(HDR_$(PKG)) -all_lib+=$(LIB_$(PKG)) diff --git a/c_utils/walltime_c.c b/c_utils/walltime_c.c index c9dce3a..8f4ac0c 100644 --- a/c_utils/walltime_c.c +++ b/c_utils/walltime_c.c @@ -25,7 +25,7 @@ /* * Functionality for reading wall clock time * - * Copyright (C) 2010, 2011 Max-Planck-Society + * Copyright (C) 2010-2016 Max-Planck-Society * Author: Martin Reinecke */ @@ -33,6 +33,8 @@ #include #elif defined (USE_MPI) #include "mpi.h" +#elif defined (_WIN32) +#include #else #include #include @@ -46,6 +48,17 @@ double wallTime(void) return omp_get_wtime(); #elif defined (USE_MPI) return MPI_Wtime(); +#elif defined (_WIN32) + static double inv_freq = -1.; + if (inv_freq<0) + { + LARGE_INTEGER freq; + QueryPerformanceFrequency(&freq); + inv_freq = 1. / double(freq.QuadPart); + } + LARGE_INTEGER count; + QueryPerformanceCounter(&count); + return count.QuadPart*inv_freq; #else struct timeval t; gettimeofday(&t, NULL); diff --git a/config/config.auto.in b/config/config.auto.in deleted file mode 100644 index 841cec0..0000000 --- a/config/config.auto.in +++ /dev/null @@ -1,12 +0,0 @@ -@SILENT_RULE@ - -CC=@CC@ -CL=@CC@ -CCFLAGS_NO_C=@CCFLAGS_NO_C@ -CCFLAGS=$(CCFLAGS_NO_C) -c -CLFLAGS=-L. -L$(LIBDIR) @LDCCFLAGS@ -lm -DEBUG_CFLAGS=@DEBUG_CFLAGS@ -MPI_CFLAGS=@MPI_CFLAGS@ -OPENMP_CFLAGS=@OPENMP_CFLAGS@ -PIC_CFLAGS=@PIC_CFLAGS@ -ARCREATE=@ARCREATE@ diff --git a/config/rules.common b/config/rules.common deleted file mode 100644 index bac2a2c..0000000 --- a/config/rules.common +++ /dev/null @@ -1,33 +0,0 @@ -BLDROOT = $(SRCROOT)/build.$(SHARP_TARGET) -PREFIX = $(SRCROOT)/$(SHARP_TARGET) -BINDIR = $(PREFIX)/bin -INCDIR = $(PREFIX)/include -LIBDIR = $(PREFIX)/lib -DOCDIR = $(SRCROOT)/doc -PYTHONDIR = $(SRCROOT)/python/libsharp - -# do not use any suffix rules -.SUFFIXES: -# do not use any default rules -.DEFAULT: - -echo_config: - @echo using configuration \'$(SHARP_TARGET)\' - -$(BLDROOT)/%.o : $(SRCROOT)/%.c | echo_config - @echo "# compiling $*.c" - cd $(@D) && $(CC) $(FULL_INCLUDE) -I$(BLDROOT) $(CCFLAGS) $< - -$(BLDROOT)/%.o : $(SRCROOT)/%.cc | echo_config - @echo "# compiling $*.cc" - cd $(@D) && $(CXX) $(FULL_INCLUDE) -I$(BLDROOT) $(CXXCFLAGS) $< - -%_mkdir: - @if [ ! -d $* ]; then mkdir -p $* ; fi - -clean: - rm -rf $(BLDROOT) $(PREFIX) $(DOCDIR) autom4te.cache/ config.log config.status - rm -rf $(PYTHONDIR)/*.c $(PYTHONDIR)/*.o $(PYTHONDIR)/*.so - -distclean: clean - rm -f config/config.auto diff --git a/configure.ac b/configure.ac index 79b435e..1264b17 100644 --- a/configure.ac +++ b/configure.ac @@ -1,113 +1,45 @@ -AC_INIT(config/config.auto.in) +AC_INIT([libsharp], [1.0.0]) +AM_INIT_AUTOMAKE([foreign subdir-objects -Wall -Werror]) +AM_MAINTAINER_MODE([enable]) -AC_CHECK_PROG([uname_found],[uname],[1],[0]) -if test $uname_found -eq 0 ; then - echo "No uname found; setting system type to unknown." - system="unknown" -else - system=`uname -s`-`uname -r` -fi -AC_LANG([C]) -AC_TRY_COMPILE([], [@%:@ifndef __INTEL_COMPILER -choke me -@%:@endif], [ICC=[yes]], [ICC=[no]]) +dnl +dnl Needed for linking on Windows. +dnl Protect with m4_ifdef because AM_PROG_AR is required in +dnl autoconf >= 1.12 when using -Wall, but the macro is +dnl absent in old versions of autoconf. +dnl +m4_ifdef([AM_PROG_AR], [AM_PROG_AR]) -if test $ICC = yes; then GCC=no; fi -CCTYPE=unknown -if test $GCC = yes; then CCTYPE=gcc; fi -if test $ICC = yes; then CCTYPE=icc; fi -AC_OPENMP +LT_INIT +AC_CONFIG_MACRO_DIR([m4]) -SILENT_RULE=".SILENT:" -AC_ARG_ENABLE(noisy-make, - [ --enable-noisy-make enable detailed make output], - [if test "$enableval" = yes; then - SILENT_RULE="" - fi]) +dnl +dnl By default, install the headers into a subdirectory of +dnl ${prefix}/include to avoid possible header filename collisions. +dnl +includedir="${includedir}/${PACKAGE_NAME}" -ENABLE_MPI=no -AC_ARG_ENABLE(mpi, - [ --enable-mpi enable generation of MPI-parallel code], - [if test "$enableval" = yes; then - ENABLE_MPI=yes - fi]) +dnl +dnl Enable silent build rules if this version of Automake supports them +dnl +m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])]) -ENABLE_DEBUG=no -AC_ARG_ENABLE(debug, - [ --enable-debug enable generation of debugging symbols], - [if test "$enableval" = yes; then - ENABLE_DEBUG=yes - fi]) -ENABLE_PIC=no -AC_ARG_ENABLE(pic, - [ --enable-pic enable generation of position independent code], - [if test "$enableval" = yes; then - ENABLE_PIC=yes - fi]) +AC_PROG_CC_C99 -case $CCTYPE in - gcc) - CCFLAGS="-O3 -fno-tree-vectorize -ffast-math -fomit-frame-pointer -std=c99 -pedantic -Wextra -Wall -Wno-unknown-pragmas -Wshadow -Wmissing-prototypes -Wfatal-errors" - GCCVERSION="`$CC -dumpversion 2>&1`" - echo "Using gcc version $GCCVERSION" - AC_SUBST(GCCVERSION) - changequote(,) - gcc43=`echo $GCCVERSION | grep -c '^4\.[3456789]'` - gcc44=`echo $GCCVERSION | grep -c '^4\.4'` - changequote([,]) - if test $gcc43 -gt 0; then - CCFLAGS="$CCFLAGS -march=native" - fi - if test $gcc44 -gt 0; then - CCFLAGS="$CCFLAGS -fno-tree-fre" - fi - ;; - icc) - CCFLAGS="-O3 -xHOST -std=c99 -ip -Wbrief -Wall -vec-report0 -openmp-report0 -wd383,981,1419,1572" - ;; - *) - CCFLAGS="-O2" - # Don't do anything now - ;; -esac +# adding the lib to the files to link +LIBS="-lm" -case $system in - Darwin-*) - ARCREATE="libtool -static -o" - ;; - *) - ARCREATE="ar cr" - ;; -esac +AC_PROG_LIBTOOL -if test $ENABLE_DEBUG = yes; then - DEBUG_CFLAGS="-g" -fi +dnl +dnl Create pkgconfig .pc file. +dnl +AX_CREATE_PKGCONFIG_INFO(,,,,[]) +AC_SUBST([LIBS]) +AC_SUBST([AM_CFLAGS]) +AC_SUBST([AM_LDFLAGS]) -if test $ENABLE_PIC = yes; then - PIC_CFLAGS="-fPIC" -fi - -if test $ENABLE_MPI = yes; then - MPI_CFLAGS="-DUSE_MPI" -fi - -CCFLAGS="$CCFLAGS $DEBUG_CFLAGS $OPENMP_CFLAGS $PIC_CFLAGS $MPI_CFLAGS" - -CCFLAGS_NO_C="$CCFLAGS $CPPFLAGS" - -LDCCFLAGS="$LDFLAGS $CCFLAGS" - -AC_SUBST(SILENT_RULE) -AC_SUBST(CC) -AC_SUBST(CCFLAGS_NO_C) -AC_SUBST(LDCCFLAGS) -AC_SUBST(DEBUG_CFLAGS) -AC_SUBST(MPI_CFLAGS) -AC_SUBST(OPENMP_CFLAGS) -AC_SUBST(PIC_CFLAGS) -AC_SUBST(ARCREATE) - -AC_OUTPUT(config/config.auto) +AC_CONFIG_FILES([Makefile]) +AC_OUTPUT diff --git a/docsrc/c_utils.dox b/docsrc/c_utils.dox deleted file mode 100644 index daf432f..0000000 --- a/docsrc/c_utils.dox +++ /dev/null @@ -1,290 +0,0 @@ -# Doxyfile 1.8.1 - -#--------------------------------------------------------------------------- -# Project related configuration options -#--------------------------------------------------------------------------- -DOXYFILE_ENCODING = UTF-8 -PROJECT_NAME = "LevelS C support library" -PROJECT_NUMBER = 0.1 -PROJECT_BRIEF = -PROJECT_LOGO = -OUTPUT_DIRECTORY = . -CREATE_SUBDIRS = NO -OUTPUT_LANGUAGE = English -BRIEF_MEMBER_DESC = NO -REPEAT_BRIEF = YES -ABBREVIATE_BRIEF = -ALWAYS_DETAILED_SEC = NO -INLINE_INHERITED_MEMB = NO -FULL_PATH_NAMES = NO -STRIP_FROM_PATH = -STRIP_FROM_INC_PATH = -SHORT_NAMES = NO -JAVADOC_AUTOBRIEF = NO -QT_AUTOBRIEF = NO -MULTILINE_CPP_IS_BRIEF = NO -INHERIT_DOCS = YES -SEPARATE_MEMBER_PAGES = NO -TAB_SIZE = 8 -ALIASES = -TCL_SUBST = -OPTIMIZE_OUTPUT_FOR_C = YES -OPTIMIZE_OUTPUT_JAVA = NO -OPTIMIZE_FOR_FORTRAN = NO -OPTIMIZE_OUTPUT_VHDL = NO -EXTENSION_MAPPING = -MARKDOWN_SUPPORT = YES -BUILTIN_STL_SUPPORT = NO -CPP_CLI_SUPPORT = NO -SIP_SUPPORT = NO -IDL_PROPERTY_SUPPORT = YES -DISTRIBUTE_GROUP_DOC = NO -SUBGROUPING = YES -INLINE_GROUPED_CLASSES = NO -INLINE_SIMPLE_STRUCTS = NO -TYPEDEF_HIDES_STRUCT = NO -SYMBOL_CACHE_SIZE = 0 -LOOKUP_CACHE_SIZE = 0 -#--------------------------------------------------------------------------- -# Build related configuration options -#--------------------------------------------------------------------------- -EXTRACT_ALL = NO -EXTRACT_PRIVATE = NO -EXTRACT_PACKAGE = NO -EXTRACT_STATIC = NO -EXTRACT_LOCAL_CLASSES = YES -EXTRACT_LOCAL_METHODS = NO -EXTRACT_ANON_NSPACES = NO -HIDE_UNDOC_MEMBERS = YES -HIDE_UNDOC_CLASSES = YES -HIDE_FRIEND_COMPOUNDS = YES -HIDE_IN_BODY_DOCS = NO -INTERNAL_DOCS = NO -CASE_SENSE_NAMES = YES -HIDE_SCOPE_NAMES = NO -SHOW_INCLUDE_FILES = YES -FORCE_LOCAL_INCLUDES = NO -INLINE_INFO = YES -SORT_MEMBER_DOCS = NO -SORT_BRIEF_DOCS = NO -SORT_MEMBERS_CTORS_1ST = NO -SORT_GROUP_NAMES = NO -SORT_BY_SCOPE_NAME = NO -STRICT_PROTO_MATCHING = NO -GENERATE_TODOLIST = YES -GENERATE_TESTLIST = YES -GENERATE_BUGLIST = YES -GENERATE_DEPRECATEDLIST= YES -ENABLED_SECTIONS = -MAX_INITIALIZER_LINES = 30 -SHOW_USED_FILES = YES -SHOW_FILES = YES -SHOW_NAMESPACES = YES -FILE_VERSION_FILTER = -LAYOUT_FILE = -CITE_BIB_FILES = -#--------------------------------------------------------------------------- -# configuration options related to warning and progress messages -#--------------------------------------------------------------------------- -QUIET = YES -WARNINGS = YES -WARN_IF_UNDOCUMENTED = YES -WARN_IF_DOC_ERROR = YES -WARN_NO_PARAMDOC = NO -WARN_FORMAT = "$file:$line: $text" -WARN_LOGFILE = -#--------------------------------------------------------------------------- -# configuration options related to the input files -#--------------------------------------------------------------------------- -INPUT = ../c_utils -INPUT_ENCODING = UTF-8 -FILE_PATTERNS = *.h \ - *.c \ - *.dox -RECURSIVE = YES -EXCLUDE = -EXCLUDE_SYMLINKS = NO -EXCLUDE_PATTERNS = -EXCLUDE_SYMBOLS = -EXAMPLE_PATH = -EXAMPLE_PATTERNS = -EXAMPLE_RECURSIVE = NO -IMAGE_PATH = -INPUT_FILTER = -FILTER_PATTERNS = -FILTER_SOURCE_FILES = NO -FILTER_SOURCE_PATTERNS = -#--------------------------------------------------------------------------- -# configuration options related to source browsing -#--------------------------------------------------------------------------- -SOURCE_BROWSER = YES -INLINE_SOURCES = NO -STRIP_CODE_COMMENTS = NO -REFERENCED_BY_RELATION = NO -REFERENCES_RELATION = NO -REFERENCES_LINK_SOURCE = YES -USE_HTAGS = NO -VERBATIM_HEADERS = YES -#--------------------------------------------------------------------------- -# configuration options related to the alphabetical class index -#--------------------------------------------------------------------------- -ALPHABETICAL_INDEX = YES -COLS_IN_ALPHA_INDEX = 5 -IGNORE_PREFIX = -#--------------------------------------------------------------------------- -# configuration options related to the HTML output -#--------------------------------------------------------------------------- -GENERATE_HTML = YES -HTML_OUTPUT = htmldoc -HTML_FILE_EXTENSION = .html -HTML_HEADER = -HTML_FOOTER = footer.html -HTML_STYLESHEET = -HTML_EXTRA_FILES = -HTML_COLORSTYLE_HUE = 220 -HTML_COLORSTYLE_SAT = 100 -HTML_COLORSTYLE_GAMMA = 80 -HTML_TIMESTAMP = YES -HTML_DYNAMIC_SECTIONS = NO -HTML_INDEX_NUM_ENTRIES = 100 -GENERATE_DOCSET = NO -DOCSET_FEEDNAME = "Doxygen generated docs" -DOCSET_BUNDLE_ID = org.doxygen.Project -DOCSET_PUBLISHER_ID = org.doxygen.Publisher -DOCSET_PUBLISHER_NAME = Publisher -GENERATE_HTMLHELP = NO -CHM_FILE = -HHC_LOCATION = -GENERATE_CHI = NO -CHM_INDEX_ENCODING = -BINARY_TOC = NO -TOC_EXPAND = NO -GENERATE_QHP = NO -QCH_FILE = -QHP_NAMESPACE = org.doxygen.Project -QHP_VIRTUAL_FOLDER = doc -QHP_CUST_FILTER_NAME = -QHP_CUST_FILTER_ATTRS = -QHP_SECT_FILTER_ATTRS = -QHG_LOCATION = -GENERATE_ECLIPSEHELP = NO -ECLIPSE_DOC_ID = org.doxygen.Project -DISABLE_INDEX = NO -GENERATE_TREEVIEW = NO -ENUM_VALUES_PER_LINE = 4 -TREEVIEW_WIDTH = 250 -EXT_LINKS_IN_WINDOW = NO -FORMULA_FONTSIZE = 10 -FORMULA_TRANSPARENT = YES -USE_MATHJAX = NO -MATHJAX_RELPATH = http://cdn.mathjax.org/mathjax/latest -MATHJAX_EXTENSIONS = -SEARCHENGINE = NO -SERVER_BASED_SEARCH = NO -#--------------------------------------------------------------------------- -# configuration options related to the LaTeX output -#--------------------------------------------------------------------------- -GENERATE_LATEX = NO -LATEX_OUTPUT = latex -LATEX_CMD_NAME = latex -MAKEINDEX_CMD_NAME = makeindex -COMPACT_LATEX = YES -PAPER_TYPE = a4wide -EXTRA_PACKAGES = -LATEX_HEADER = -LATEX_FOOTER = -PDF_HYPERLINKS = YES -USE_PDFLATEX = YES -LATEX_BATCHMODE = NO -LATEX_HIDE_INDICES = NO -LATEX_SOURCE_CODE = NO -LATEX_BIB_STYLE = plain -#--------------------------------------------------------------------------- -# configuration options related to the RTF output -#--------------------------------------------------------------------------- -GENERATE_RTF = NO -RTF_OUTPUT = rtf -COMPACT_RTF = NO -RTF_HYPERLINKS = NO -RTF_STYLESHEET_FILE = -RTF_EXTENSIONS_FILE = -#--------------------------------------------------------------------------- -# configuration options related to the man page output -#--------------------------------------------------------------------------- -GENERATE_MAN = NO -MAN_OUTPUT = man -MAN_EXTENSION = .3 -MAN_LINKS = NO -#--------------------------------------------------------------------------- -# configuration options related to the XML output -#--------------------------------------------------------------------------- -GENERATE_XML = NO -XML_OUTPUT = xml -XML_SCHEMA = -XML_DTD = -XML_PROGRAMLISTING = YES -#--------------------------------------------------------------------------- -# configuration options for the AutoGen Definitions output -#--------------------------------------------------------------------------- -GENERATE_AUTOGEN_DEF = NO -#--------------------------------------------------------------------------- -# configuration options related to the Perl module output -#--------------------------------------------------------------------------- -GENERATE_PERLMOD = NO -PERLMOD_LATEX = NO -PERLMOD_PRETTY = YES -PERLMOD_MAKEVAR_PREFIX = -#--------------------------------------------------------------------------- -# Configuration options related to the preprocessor -#--------------------------------------------------------------------------- -ENABLE_PREPROCESSING = YES -MACRO_EXPANSION = NO -EXPAND_ONLY_PREDEF = NO -SEARCH_INCLUDES = YES -INCLUDE_PATH = -INCLUDE_FILE_PATTERNS = -PREDEFINED = -EXPAND_AS_DEFINED = -SKIP_FUNCTION_MACROS = YES -#--------------------------------------------------------------------------- -# Configuration::additions related to external references -#--------------------------------------------------------------------------- -TAGFILES = -GENERATE_TAGFILE = c_utils.tag -ALLEXTERNALS = NO -EXTERNAL_GROUPS = YES -PERL_PATH = /usr/bin/perl -#--------------------------------------------------------------------------- -# Configuration options related to the dot tool -#--------------------------------------------------------------------------- -CLASS_DIAGRAMS = YES -MSCGEN_PATH = -HIDE_UNDOC_RELATIONS = YES -HAVE_DOT = NO -DOT_NUM_THREADS = 0 -DOT_FONTNAME = FreeSans -DOT_FONTSIZE = 10 -DOT_FONTPATH = -CLASS_GRAPH = YES -COLLABORATION_GRAPH = YES -GROUP_GRAPHS = YES -UML_LOOK = NO -UML_LIMIT_NUM_FIELDS = 10 -TEMPLATE_RELATIONS = YES -INCLUDE_GRAPH = NO -INCLUDED_BY_GRAPH = NO -CALL_GRAPH = NO -CALLER_GRAPH = NO -GRAPHICAL_HIERARCHY = NO -DIRECTORY_GRAPH = YES -DOT_IMAGE_FORMAT = png -INTERACTIVE_SVG = NO -DOT_PATH = -DOTFILE_DIRS = -MSCFILE_DIRS = -DOT_GRAPH_MAX_NODES = 50 -MAX_DOT_GRAPH_DEPTH = 0 -DOT_TRANSPARENT = NO -DOT_MULTI_TARGETS = NO -GENERATE_LEGEND = YES -DOT_CLEANUP = YES diff --git a/docsrc/footer.html b/docsrc/footer.html deleted file mode 100644 index 6f5dbf0..0000000 --- a/docsrc/footer.html +++ /dev/null @@ -1,5 +0,0 @@ -
-Generated on $datetime for $projectname -
- - diff --git a/docsrc/index_code.html b/docsrc/index_code.html deleted file mode 100644 index d8a001d..0000000 --- a/docsrc/index_code.html +++ /dev/null @@ -1,15 +0,0 @@ - - -Libsharp source code documentation - -

Libsharp source code documentation

- -

C interfaces

- - - - diff --git a/docsrc/libfftpack.dox b/docsrc/libfftpack.dox deleted file mode 100644 index 7ff2c23..0000000 --- a/docsrc/libfftpack.dox +++ /dev/null @@ -1,290 +0,0 @@ -# Doxyfile 1.8.1 - -#--------------------------------------------------------------------------- -# Project related configuration options -#--------------------------------------------------------------------------- -DOXYFILE_ENCODING = UTF-8 -PROJECT_NAME = "LevelS FFT library" -PROJECT_NUMBER = 0.1 -PROJECT_BRIEF = -PROJECT_LOGO = -OUTPUT_DIRECTORY = . -CREATE_SUBDIRS = NO -OUTPUT_LANGUAGE = English -BRIEF_MEMBER_DESC = NO -REPEAT_BRIEF = YES -ABBREVIATE_BRIEF = -ALWAYS_DETAILED_SEC = NO -INLINE_INHERITED_MEMB = NO -FULL_PATH_NAMES = NO -STRIP_FROM_PATH = -STRIP_FROM_INC_PATH = -SHORT_NAMES = NO -JAVADOC_AUTOBRIEF = NO -QT_AUTOBRIEF = NO -MULTILINE_CPP_IS_BRIEF = NO -INHERIT_DOCS = YES -SEPARATE_MEMBER_PAGES = NO -TAB_SIZE = 8 -ALIASES = -TCL_SUBST = -OPTIMIZE_OUTPUT_FOR_C = YES -OPTIMIZE_OUTPUT_JAVA = NO -OPTIMIZE_FOR_FORTRAN = NO -OPTIMIZE_OUTPUT_VHDL = NO -EXTENSION_MAPPING = -MARKDOWN_SUPPORT = YES -BUILTIN_STL_SUPPORT = NO -CPP_CLI_SUPPORT = NO -SIP_SUPPORT = NO -IDL_PROPERTY_SUPPORT = YES -DISTRIBUTE_GROUP_DOC = NO -SUBGROUPING = YES -INLINE_GROUPED_CLASSES = NO -INLINE_SIMPLE_STRUCTS = NO -TYPEDEF_HIDES_STRUCT = NO -SYMBOL_CACHE_SIZE = 0 -LOOKUP_CACHE_SIZE = 0 -#--------------------------------------------------------------------------- -# Build related configuration options -#--------------------------------------------------------------------------- -EXTRACT_ALL = NO -EXTRACT_PRIVATE = NO -EXTRACT_PACKAGE = NO -EXTRACT_STATIC = NO -EXTRACT_LOCAL_CLASSES = YES -EXTRACT_LOCAL_METHODS = NO -EXTRACT_ANON_NSPACES = NO -HIDE_UNDOC_MEMBERS = YES -HIDE_UNDOC_CLASSES = YES -HIDE_FRIEND_COMPOUNDS = YES -HIDE_IN_BODY_DOCS = NO -INTERNAL_DOCS = NO -CASE_SENSE_NAMES = YES -HIDE_SCOPE_NAMES = NO -SHOW_INCLUDE_FILES = YES -FORCE_LOCAL_INCLUDES = NO -INLINE_INFO = YES -SORT_MEMBER_DOCS = NO -SORT_BRIEF_DOCS = NO -SORT_MEMBERS_CTORS_1ST = NO -SORT_GROUP_NAMES = NO -SORT_BY_SCOPE_NAME = NO -STRICT_PROTO_MATCHING = NO -GENERATE_TODOLIST = YES -GENERATE_TESTLIST = YES -GENERATE_BUGLIST = YES -GENERATE_DEPRECATEDLIST= YES -ENABLED_SECTIONS = -MAX_INITIALIZER_LINES = 30 -SHOW_USED_FILES = YES -SHOW_FILES = YES -SHOW_NAMESPACES = YES -FILE_VERSION_FILTER = -LAYOUT_FILE = -CITE_BIB_FILES = -#--------------------------------------------------------------------------- -# configuration options related to warning and progress messages -#--------------------------------------------------------------------------- -QUIET = YES -WARNINGS = YES -WARN_IF_UNDOCUMENTED = YES -WARN_IF_DOC_ERROR = YES -WARN_NO_PARAMDOC = NO -WARN_FORMAT = "$file:$line: $text" -WARN_LOGFILE = -#--------------------------------------------------------------------------- -# configuration options related to the input files -#--------------------------------------------------------------------------- -INPUT = ../libfftpack -INPUT_ENCODING = UTF-8 -FILE_PATTERNS = *.h \ - *.c \ - *.dox -RECURSIVE = YES -EXCLUDE = -EXCLUDE_SYMLINKS = NO -EXCLUDE_PATTERNS = -EXCLUDE_SYMBOLS = -EXAMPLE_PATH = -EXAMPLE_PATTERNS = -EXAMPLE_RECURSIVE = NO -IMAGE_PATH = -INPUT_FILTER = -FILTER_PATTERNS = -FILTER_SOURCE_FILES = NO -FILTER_SOURCE_PATTERNS = -#--------------------------------------------------------------------------- -# configuration options related to source browsing -#--------------------------------------------------------------------------- -SOURCE_BROWSER = YES -INLINE_SOURCES = NO -STRIP_CODE_COMMENTS = NO -REFERENCED_BY_RELATION = NO -REFERENCES_RELATION = NO -REFERENCES_LINK_SOURCE = YES -USE_HTAGS = NO -VERBATIM_HEADERS = YES -#--------------------------------------------------------------------------- -# configuration options related to the alphabetical class index -#--------------------------------------------------------------------------- -ALPHABETICAL_INDEX = YES -COLS_IN_ALPHA_INDEX = 5 -IGNORE_PREFIX = -#--------------------------------------------------------------------------- -# configuration options related to the HTML output -#--------------------------------------------------------------------------- -GENERATE_HTML = YES -HTML_OUTPUT = htmldoc -HTML_FILE_EXTENSION = .html -HTML_HEADER = -HTML_FOOTER = footer.html -HTML_STYLESHEET = -HTML_EXTRA_FILES = -HTML_COLORSTYLE_HUE = 220 -HTML_COLORSTYLE_SAT = 100 -HTML_COLORSTYLE_GAMMA = 80 -HTML_TIMESTAMP = YES -HTML_DYNAMIC_SECTIONS = NO -HTML_INDEX_NUM_ENTRIES = 100 -GENERATE_DOCSET = NO -DOCSET_FEEDNAME = "Doxygen generated docs" -DOCSET_BUNDLE_ID = org.doxygen.Project -DOCSET_PUBLISHER_ID = org.doxygen.Publisher -DOCSET_PUBLISHER_NAME = Publisher -GENERATE_HTMLHELP = NO -CHM_FILE = -HHC_LOCATION = -GENERATE_CHI = NO -CHM_INDEX_ENCODING = -BINARY_TOC = NO -TOC_EXPAND = NO -GENERATE_QHP = NO -QCH_FILE = -QHP_NAMESPACE = org.doxygen.Project -QHP_VIRTUAL_FOLDER = doc -QHP_CUST_FILTER_NAME = -QHP_CUST_FILTER_ATTRS = -QHP_SECT_FILTER_ATTRS = -QHG_LOCATION = -GENERATE_ECLIPSEHELP = NO -ECLIPSE_DOC_ID = org.doxygen.Project -DISABLE_INDEX = NO -GENERATE_TREEVIEW = NO -ENUM_VALUES_PER_LINE = 4 -TREEVIEW_WIDTH = 250 -EXT_LINKS_IN_WINDOW = NO -FORMULA_FONTSIZE = 10 -FORMULA_TRANSPARENT = YES -USE_MATHJAX = NO -MATHJAX_RELPATH = http://cdn.mathjax.org/mathjax/latest -MATHJAX_EXTENSIONS = -SEARCHENGINE = NO -SERVER_BASED_SEARCH = NO -#--------------------------------------------------------------------------- -# configuration options related to the LaTeX output -#--------------------------------------------------------------------------- -GENERATE_LATEX = NO -LATEX_OUTPUT = latex -LATEX_CMD_NAME = latex -MAKEINDEX_CMD_NAME = makeindex -COMPACT_LATEX = YES -PAPER_TYPE = a4wide -EXTRA_PACKAGES = -LATEX_HEADER = -LATEX_FOOTER = -PDF_HYPERLINKS = YES -USE_PDFLATEX = YES -LATEX_BATCHMODE = NO -LATEX_HIDE_INDICES = NO -LATEX_SOURCE_CODE = NO -LATEX_BIB_STYLE = plain -#--------------------------------------------------------------------------- -# configuration options related to the RTF output -#--------------------------------------------------------------------------- -GENERATE_RTF = NO -RTF_OUTPUT = rtf -COMPACT_RTF = NO -RTF_HYPERLINKS = NO -RTF_STYLESHEET_FILE = -RTF_EXTENSIONS_FILE = -#--------------------------------------------------------------------------- -# configuration options related to the man page output -#--------------------------------------------------------------------------- -GENERATE_MAN = NO -MAN_OUTPUT = man -MAN_EXTENSION = .3 -MAN_LINKS = NO -#--------------------------------------------------------------------------- -# configuration options related to the XML output -#--------------------------------------------------------------------------- -GENERATE_XML = NO -XML_OUTPUT = xml -XML_SCHEMA = -XML_DTD = -XML_PROGRAMLISTING = YES -#--------------------------------------------------------------------------- -# configuration options for the AutoGen Definitions output -#--------------------------------------------------------------------------- -GENERATE_AUTOGEN_DEF = NO -#--------------------------------------------------------------------------- -# configuration options related to the Perl module output -#--------------------------------------------------------------------------- -GENERATE_PERLMOD = NO -PERLMOD_LATEX = NO -PERLMOD_PRETTY = YES -PERLMOD_MAKEVAR_PREFIX = -#--------------------------------------------------------------------------- -# Configuration options related to the preprocessor -#--------------------------------------------------------------------------- -ENABLE_PREPROCESSING = YES -MACRO_EXPANSION = NO -EXPAND_ONLY_PREDEF = NO -SEARCH_INCLUDES = YES -INCLUDE_PATH = -INCLUDE_FILE_PATTERNS = -PREDEFINED = -EXPAND_AS_DEFINED = -SKIP_FUNCTION_MACROS = YES -#--------------------------------------------------------------------------- -# Configuration::additions related to external references -#--------------------------------------------------------------------------- -TAGFILES = c_utils.tag=../c_utils -GENERATE_TAGFILE = libfftpack.tag -ALLEXTERNALS = NO -EXTERNAL_GROUPS = YES -PERL_PATH = /usr/bin/perl -#--------------------------------------------------------------------------- -# Configuration options related to the dot tool -#--------------------------------------------------------------------------- -CLASS_DIAGRAMS = YES -MSCGEN_PATH = -HIDE_UNDOC_RELATIONS = YES -HAVE_DOT = NO -DOT_NUM_THREADS = 0 -DOT_FONTNAME = FreeSans -DOT_FONTSIZE = 10 -DOT_FONTPATH = -CLASS_GRAPH = YES -COLLABORATION_GRAPH = YES -GROUP_GRAPHS = YES -UML_LOOK = NO -UML_LIMIT_NUM_FIELDS = 10 -TEMPLATE_RELATIONS = YES -INCLUDE_GRAPH = NO -INCLUDED_BY_GRAPH = NO -CALL_GRAPH = NO -CALLER_GRAPH = NO -GRAPHICAL_HIERARCHY = NO -DIRECTORY_GRAPH = YES -DOT_IMAGE_FORMAT = png -INTERACTIVE_SVG = NO -DOT_PATH = -DOTFILE_DIRS = -MSCFILE_DIRS = -DOT_GRAPH_MAX_NODES = 50 -MAX_DOT_GRAPH_DEPTH = 0 -DOT_TRANSPARENT = NO -DOT_MULTI_TARGETS = NO -GENERATE_LEGEND = YES -DOT_CLEANUP = YES diff --git a/docsrc/libsharp.dox b/docsrc/libsharp.dox deleted file mode 100644 index b476ab4..0000000 --- a/docsrc/libsharp.dox +++ /dev/null @@ -1,291 +0,0 @@ -# Doxyfile 1.8.1 - -#--------------------------------------------------------------------------- -# Project related configuration options -#--------------------------------------------------------------------------- -DOXYFILE_ENCODING = UTF-8 -PROJECT_NAME = "LevelS SHT library" -PROJECT_NUMBER = 0.1 -PROJECT_BRIEF = -PROJECT_LOGO = -OUTPUT_DIRECTORY = . -CREATE_SUBDIRS = NO -OUTPUT_LANGUAGE = English -BRIEF_MEMBER_DESC = NO -REPEAT_BRIEF = YES -ABBREVIATE_BRIEF = -ALWAYS_DETAILED_SEC = NO -INLINE_INHERITED_MEMB = NO -FULL_PATH_NAMES = NO -STRIP_FROM_PATH = -STRIP_FROM_INC_PATH = -SHORT_NAMES = NO -JAVADOC_AUTOBRIEF = NO -QT_AUTOBRIEF = NO -MULTILINE_CPP_IS_BRIEF = NO -INHERIT_DOCS = YES -SEPARATE_MEMBER_PAGES = NO -TAB_SIZE = 8 -ALIASES = -TCL_SUBST = -OPTIMIZE_OUTPUT_FOR_C = YES -OPTIMIZE_OUTPUT_JAVA = NO -OPTIMIZE_FOR_FORTRAN = NO -OPTIMIZE_OUTPUT_VHDL = NO -EXTENSION_MAPPING = -MARKDOWN_SUPPORT = YES -BUILTIN_STL_SUPPORT = NO -CPP_CLI_SUPPORT = NO -SIP_SUPPORT = NO -IDL_PROPERTY_SUPPORT = YES -DISTRIBUTE_GROUP_DOC = NO -SUBGROUPING = YES -INLINE_GROUPED_CLASSES = NO -INLINE_SIMPLE_STRUCTS = NO -TYPEDEF_HIDES_STRUCT = NO -SYMBOL_CACHE_SIZE = 0 -LOOKUP_CACHE_SIZE = 0 -#--------------------------------------------------------------------------- -# Build related configuration options -#--------------------------------------------------------------------------- -EXTRACT_ALL = NO -EXTRACT_PRIVATE = NO -EXTRACT_PACKAGE = NO -EXTRACT_STATIC = NO -EXTRACT_LOCAL_CLASSES = YES -EXTRACT_LOCAL_METHODS = NO -EXTRACT_ANON_NSPACES = NO -HIDE_UNDOC_MEMBERS = YES -HIDE_UNDOC_CLASSES = YES -HIDE_FRIEND_COMPOUNDS = YES -HIDE_IN_BODY_DOCS = NO -INTERNAL_DOCS = NO -CASE_SENSE_NAMES = YES -HIDE_SCOPE_NAMES = NO -SHOW_INCLUDE_FILES = YES -FORCE_LOCAL_INCLUDES = NO -INLINE_INFO = YES -SORT_MEMBER_DOCS = NO -SORT_BRIEF_DOCS = NO -SORT_MEMBERS_CTORS_1ST = NO -SORT_GROUP_NAMES = NO -SORT_BY_SCOPE_NAME = NO -STRICT_PROTO_MATCHING = NO -GENERATE_TODOLIST = YES -GENERATE_TESTLIST = YES -GENERATE_BUGLIST = YES -GENERATE_DEPRECATEDLIST= YES -ENABLED_SECTIONS = -MAX_INITIALIZER_LINES = 30 -SHOW_USED_FILES = YES -SHOW_FILES = YES -SHOW_NAMESPACES = YES -FILE_VERSION_FILTER = -LAYOUT_FILE = -CITE_BIB_FILES = -#--------------------------------------------------------------------------- -# configuration options related to warning and progress messages -#--------------------------------------------------------------------------- -QUIET = YES -WARNINGS = YES -WARN_IF_UNDOCUMENTED = YES -WARN_IF_DOC_ERROR = YES -WARN_NO_PARAMDOC = NO -WARN_FORMAT = "$file:$line: $text" -WARN_LOGFILE = -#--------------------------------------------------------------------------- -# configuration options related to the input files -#--------------------------------------------------------------------------- -INPUT = ../libsharp -INPUT_ENCODING = UTF-8 -FILE_PATTERNS = *.h \ - *.c \ - *.dox -RECURSIVE = YES -EXCLUDE = -EXCLUDE_SYMLINKS = NO -EXCLUDE_PATTERNS = -EXCLUDE_SYMBOLS = -EXAMPLE_PATH = -EXAMPLE_PATTERNS = -EXAMPLE_RECURSIVE = NO -IMAGE_PATH = -INPUT_FILTER = -FILTER_PATTERNS = -FILTER_SOURCE_FILES = NO -FILTER_SOURCE_PATTERNS = -#--------------------------------------------------------------------------- -# configuration options related to source browsing -#--------------------------------------------------------------------------- -SOURCE_BROWSER = YES -INLINE_SOURCES = NO -STRIP_CODE_COMMENTS = NO -REFERENCED_BY_RELATION = NO -REFERENCES_RELATION = NO -REFERENCES_LINK_SOURCE = YES -USE_HTAGS = NO -VERBATIM_HEADERS = YES -#--------------------------------------------------------------------------- -# configuration options related to the alphabetical class index -#--------------------------------------------------------------------------- -ALPHABETICAL_INDEX = YES -COLS_IN_ALPHA_INDEX = 5 -IGNORE_PREFIX = -#--------------------------------------------------------------------------- -# configuration options related to the HTML output -#--------------------------------------------------------------------------- -GENERATE_HTML = YES -HTML_OUTPUT = htmldoc -HTML_FILE_EXTENSION = .html -HTML_HEADER = -HTML_FOOTER = footer.html -HTML_STYLESHEET = -HTML_EXTRA_FILES = -HTML_COLORSTYLE_HUE = 220 -HTML_COLORSTYLE_SAT = 100 -HTML_COLORSTYLE_GAMMA = 80 -HTML_TIMESTAMP = YES -HTML_DYNAMIC_SECTIONS = NO -HTML_INDEX_NUM_ENTRIES = 100 -GENERATE_DOCSET = NO -DOCSET_FEEDNAME = "Doxygen generated docs" -DOCSET_BUNDLE_ID = org.doxygen.Project -DOCSET_PUBLISHER_ID = org.doxygen.Publisher -DOCSET_PUBLISHER_NAME = Publisher -GENERATE_HTMLHELP = NO -CHM_FILE = -HHC_LOCATION = -GENERATE_CHI = NO -CHM_INDEX_ENCODING = -BINARY_TOC = NO -TOC_EXPAND = NO -GENERATE_QHP = NO -QCH_FILE = -QHP_NAMESPACE = org.doxygen.Project -QHP_VIRTUAL_FOLDER = doc -QHP_CUST_FILTER_NAME = -QHP_CUST_FILTER_ATTRS = -QHP_SECT_FILTER_ATTRS = -QHG_LOCATION = -GENERATE_ECLIPSEHELP = NO -ECLIPSE_DOC_ID = org.doxygen.Project -DISABLE_INDEX = NO -GENERATE_TREEVIEW = NO -ENUM_VALUES_PER_LINE = 4 -TREEVIEW_WIDTH = 250 -EXT_LINKS_IN_WINDOW = NO -FORMULA_FONTSIZE = 10 -FORMULA_TRANSPARENT = YES -USE_MATHJAX = NO -MATHJAX_RELPATH = http://cdn.mathjax.org/mathjax/latest -MATHJAX_EXTENSIONS = -SEARCHENGINE = NO -SERVER_BASED_SEARCH = NO -#--------------------------------------------------------------------------- -# configuration options related to the LaTeX output -#--------------------------------------------------------------------------- -GENERATE_LATEX = NO -LATEX_OUTPUT = latex -LATEX_CMD_NAME = latex -MAKEINDEX_CMD_NAME = makeindex -COMPACT_LATEX = YES -PAPER_TYPE = a4wide -EXTRA_PACKAGES = -LATEX_HEADER = -LATEX_FOOTER = -PDF_HYPERLINKS = YES -USE_PDFLATEX = YES -LATEX_BATCHMODE = NO -LATEX_HIDE_INDICES = NO -LATEX_SOURCE_CODE = NO -LATEX_BIB_STYLE = plain -#--------------------------------------------------------------------------- -# configuration options related to the RTF output -#--------------------------------------------------------------------------- -GENERATE_RTF = NO -RTF_OUTPUT = rtf -COMPACT_RTF = NO -RTF_HYPERLINKS = NO -RTF_STYLESHEET_FILE = -RTF_EXTENSIONS_FILE = -#--------------------------------------------------------------------------- -# configuration options related to the man page output -#--------------------------------------------------------------------------- -GENERATE_MAN = NO -MAN_OUTPUT = man -MAN_EXTENSION = .3 -MAN_LINKS = NO -#--------------------------------------------------------------------------- -# configuration options related to the XML output -#--------------------------------------------------------------------------- -GENERATE_XML = NO -XML_OUTPUT = xml -XML_SCHEMA = -XML_DTD = -XML_PROGRAMLISTING = YES -#--------------------------------------------------------------------------- -# configuration options for the AutoGen Definitions output -#--------------------------------------------------------------------------- -GENERATE_AUTOGEN_DEF = NO -#--------------------------------------------------------------------------- -# configuration options related to the Perl module output -#--------------------------------------------------------------------------- -GENERATE_PERLMOD = NO -PERLMOD_LATEX = NO -PERLMOD_PRETTY = YES -PERLMOD_MAKEVAR_PREFIX = -#--------------------------------------------------------------------------- -# Configuration options related to the preprocessor -#--------------------------------------------------------------------------- -ENABLE_PREPROCESSING = YES -MACRO_EXPANSION = NO -EXPAND_ONLY_PREDEF = NO -SEARCH_INCLUDES = YES -INCLUDE_PATH = -INCLUDE_FILE_PATTERNS = -PREDEFINED = -EXPAND_AS_DEFINED = -SKIP_FUNCTION_MACROS = YES -#--------------------------------------------------------------------------- -# Configuration::additions related to external references -#--------------------------------------------------------------------------- -TAGFILES = libfftpack.tag=../libfftpack \ - c_utils.tag=../c_utils -GENERATE_TAGFILE = libsharp.tag -ALLEXTERNALS = NO -EXTERNAL_GROUPS = YES -PERL_PATH = /usr/bin/perl -#--------------------------------------------------------------------------- -# Configuration options related to the dot tool -#--------------------------------------------------------------------------- -CLASS_DIAGRAMS = YES -MSCGEN_PATH = -HIDE_UNDOC_RELATIONS = YES -HAVE_DOT = NO -DOT_NUM_THREADS = 0 -DOT_FONTNAME = FreeSans -DOT_FONTSIZE = 10 -DOT_FONTPATH = -CLASS_GRAPH = YES -COLLABORATION_GRAPH = YES -GROUP_GRAPHS = YES -UML_LOOK = NO -UML_LIMIT_NUM_FIELDS = 10 -TEMPLATE_RELATIONS = YES -INCLUDE_GRAPH = NO -INCLUDED_BY_GRAPH = NO -CALL_GRAPH = NO -CALLER_GRAPH = NO -GRAPHICAL_HIERARCHY = NO -DIRECTORY_GRAPH = YES -DOT_IMAGE_FORMAT = png -INTERACTIVE_SVG = NO -DOT_PATH = -DOTFILE_DIRS = -MSCFILE_DIRS = -DOT_GRAPH_MAX_NODES = 50 -MAX_DOT_GRAPH_DEPTH = 0 -DOT_TRANSPARENT = NO -DOT_MULTI_TARGETS = NO -GENERATE_LEGEND = YES -DOT_CLEANUP = YES diff --git a/docsrc/planck.make b/docsrc/planck.make deleted file mode 100644 index 0d0a462..0000000 --- a/docsrc/planck.make +++ /dev/null @@ -1,20 +0,0 @@ -PKG:=docsrc - -docsrc_idx: $(DOCDIR)_mkdir - cp $(SRCROOT)/docsrc/index_code.html $(DOCDIR)/index.html - -docsrc_code_doc: $(DOCDIR)_mkdir docsrc_idx - cd $(SRCROOT)/docsrc; \ - for i in c_utils libfftpack libsharp; do \ - doxygen $${i}.dox; \ - rm -rf $(DOCDIR)/$${i}; mv htmldoc $(DOCDIR)/$${i}; \ - done; \ - rm *.tag; - -docsrc_clean: - cd $(SRCROOT)/docsrc; \ - rm -f *.tag - cd $(SRCROOT)/docsrc; \ - rm -rf htmldoc - -doc: docsrc_code_doc diff --git a/fortran/sharp.f90 b/fortran/sharp.f90 deleted file mode 100644 index 36a1d11..0000000 --- a/fortran/sharp.f90 +++ /dev/null @@ -1,286 +0,0 @@ -module sharp - use iso_c_binding - implicit none - ! alm_info flags - integer, parameter :: SHARP_PACKED = 1 - - ! sharp job types - enum, bind(c) - enumerator :: SHARP_YtW = 0 - enumerator :: SHARP_Y = 1 - enumerator :: SHARP_Yt = 2 - enumerator :: SHARP_WY = 3 - enumerator :: SHARP_ALM2MAP_DERIV1 = 4 - end enum - - ! sharp job flags - integer, parameter :: SHARP_DP = ISHFT(1, 4) - integer, parameter :: SHARP_ADD = ISHFT(1, 5) - integer, parameter :: SHARP_REAL_HARMONICS = ISHFT(1, 6) - integer, parameter :: SHARP_NO_FFT = ISHFT(1, 7) - - type sharp_geom_info - type(c_ptr) :: handle - integer(c_intptr_t) :: n_local - end type sharp_geom_info - - type sharp_alm_info - type(c_ptr) :: handle - integer(c_intptr_t) :: n_local - end type sharp_alm_info - - interface - - ! alm_info - subroutine sharp_make_general_alm_info( & - lmax, nm, stride, mval, mvstart, flags, alm_info) bind(c) - use iso_c_binding - integer(c_int), value, intent(in) :: lmax, nm, stride, flags - integer(c_int), intent(in) :: mval(nm) - integer(c_intptr_t), intent(in) :: mvstart(nm) - type(c_ptr), intent(out) :: alm_info - end subroutine sharp_make_general_alm_info - - subroutine c_sharp_make_mmajor_real_packed_alm_info( & - lmax, stride, nm, ms, alm_info) bind(c, name='sharp_make_mmajor_real_packed_alm_info') - use iso_c_binding - integer(c_int), value, intent(in) :: lmax, nm, stride - integer(c_int), intent(in), optional :: ms(nm) - type(c_ptr), intent(out) :: alm_info - end subroutine c_sharp_make_mmajor_real_packed_alm_info - - function c_sharp_alm_count(alm_info) bind(c, name='sharp_alm_count') - use iso_c_binding - integer(c_intptr_t) :: c_sharp_alm_count - type(c_ptr), value, intent(in) :: alm_info - end function c_sharp_alm_count - - subroutine c_sharp_destroy_alm_info(alm_info) bind(c, name='sharp_destroy_alm_info') - use iso_c_binding - type(c_ptr), value :: alm_info - end subroutine c_sharp_destroy_alm_info - - ! geom_info - subroutine sharp_make_subset_healpix_geom_info ( & - nside, stride, nrings, rings, weight, geom_info) bind(c) - use iso_c_binding - integer(c_int), value, intent(in) :: nside, stride, nrings - integer(c_int), intent(in), optional :: rings(nrings) - real(c_double), intent(in), optional :: weight(2 * nside) - type(c_ptr), intent(out) :: geom_info - end subroutine sharp_make_subset_healpix_geom_info - - subroutine c_sharp_destroy_geom_info(geom_info) bind(c, name='sharp_destroy_geom_info') - use iso_c_binding - type(c_ptr), value :: geom_info - end subroutine c_sharp_destroy_geom_info - - function c_sharp_map_size(info) bind(c, name='sharp_map_size') - use iso_c_binding - integer(c_intptr_t) :: c_sharp_map_size - type(c_ptr), value :: info - end function c_sharp_map_size - - - ! execute - subroutine c_sharp_execute(type, spin, alm, map, geom_info, alm_info, ntrans, & - flags, time, opcnt) bind(c, name='sharp_execute') - use iso_c_binding - integer(c_int), value :: type, spin, ntrans, flags - type(c_ptr), value :: alm_info, geom_info - real(c_double), intent(out), optional :: time - integer(c_long_long), intent(out), optional :: opcnt - type(c_ptr), intent(in) :: alm(*), map(*) - end subroutine c_sharp_execute - - subroutine c_sharp_execute_mpi(comm, type, spin, alm, map, geom_info, alm_info, ntrans, & - flags, time, opcnt) bind(c, name='sharp_execute_mpi_fortran') - use iso_c_binding - integer(c_int), value :: comm, type, spin, ntrans, flags - type(c_ptr), value :: alm_info, geom_info - real(c_double), intent(out), optional :: time - integer(c_long_long), intent(out), optional :: opcnt - type(c_ptr), intent(in) :: alm(*), map(*) - end subroutine c_sharp_execute_mpi - - ! Legendre transforms - subroutine c_sharp_legendre_transform(bl, recfac, lmax, x, out, nx) & - bind(c, name='sharp_legendre_transform') - use iso_c_binding - integer(c_intptr_t), value :: lmax, nx - real(c_double) :: bl(lmax + 1), x(nx), out(nx) - real(c_double), optional :: recfac(lmax + 1) - end subroutine c_sharp_legendre_transform - - subroutine c_sharp_legendre_transform_s(bl, recfac, lmax, x, out, nx) & - bind(c, name='sharp_legendre_transform_s') - use iso_c_binding - integer(c_intptr_t), value :: lmax, nx - real(c_float) :: bl(lmax + 1), x(nx), out(nx) - real(c_float), optional :: recfac(lmax + 1) - end subroutine c_sharp_legendre_transform_s - end interface - - interface sharp_execute - module procedure sharp_execute_d - end interface - - interface sharp_legendre_transform - module procedure sharp_legendre_transform_d, sharp_legendre_transform_s - end interface sharp_legendre_transform - -contains - ! alm info - - ! if ms is not passed, we default to using m=0..lmax. - subroutine sharp_make_mmajor_real_packed_alm_info(lmax, ms, alm_info) - use iso_c_binding - integer(c_int), value, intent(in) :: lmax - integer(c_int), intent(in), optional :: ms(:) - type(sharp_alm_info), intent(out) :: alm_info - !-- - integer(c_int), allocatable :: ms_copy(:) - integer(c_int) :: nm - - if (present(ms)) then - nm = size(ms) - allocate(ms_copy(nm)) - ms_copy = ms - call c_sharp_make_mmajor_real_packed_alm_info(lmax, 1, nm, ms_copy, alm_info=alm_info%handle) - deallocate(ms_copy) - else - call c_sharp_make_mmajor_real_packed_alm_info(lmax, 1, lmax + 1, alm_info=alm_info%handle) - end if - alm_info%n_local = c_sharp_alm_count(alm_info%handle) - end subroutine sharp_make_mmajor_real_packed_alm_info - - subroutine sharp_destroy_alm_info(alm_info) - use iso_c_binding - type(sharp_alm_info), intent(inout) :: alm_info - call c_sharp_destroy_alm_info(alm_info%handle) - alm_info%handle = c_null_ptr - end subroutine sharp_destroy_alm_info - - - ! geom info - subroutine sharp_make_healpix_geom_info(nside, rings, weight, geom_info) - integer(c_int), value :: nside - integer(c_int), optional :: rings(:) - real(c_double), intent(in), optional :: weight(2 * nside) - type(sharp_geom_info), intent(out) :: geom_info - !-- - integer(c_int) :: nrings - integer(c_int), allocatable :: rings_copy(:) - - if (present(rings)) then - nrings = size(rings) - allocate(rings_copy(nrings)) - rings_copy = rings - call sharp_make_subset_healpix_geom_info(nside, 1, nrings, rings_copy, & - weight, geom_info%handle) - deallocate(rings_copy) - else - call sharp_make_subset_healpix_geom_info(nside, 1, nrings=4 * nside - 1, & - weight=weight, geom_info=geom_info%handle) - end if - geom_info%n_local = c_sharp_map_size(geom_info%handle) - end subroutine sharp_make_healpix_geom_info - - subroutine sharp_destroy_geom_info(geom_info) - use iso_c_binding - type(sharp_geom_info), intent(inout) :: geom_info - call c_sharp_destroy_geom_info(geom_info%handle) - geom_info%handle = c_null_ptr - end subroutine sharp_destroy_geom_info - - - ! Currently the only mode supported is stacked (not interleaved) maps. - ! - ! Note that passing the exact dimension of alm/map is necesarry, it - ! prevents the caller from doing too crazy slicing prior to pass array - ! in... - ! - ! Usage: - ! - ! The alm array must have shape exactly alm(alm_info%n_local, nmaps) - ! The maps array must have shape exactly map(map_info%n_local, nmaps). - subroutine sharp_execute_d(type, spin, nmaps, alm, alm_info, map, geom_info, & - add, time, opcnt, comm) - use iso_c_binding - use mpi - implicit none - integer(c_int), value :: type, spin, nmaps - integer(c_int), optional :: comm - logical, value, optional :: add ! should add instead of replace out - - type(sharp_alm_info) :: alm_info - type(sharp_geom_info) :: geom_info - real(c_double), intent(out), optional :: time - integer(c_long_long), intent(out), optional :: opcnt - real(c_double), target, intent(inout) :: alm(0:alm_info%n_local - 1, 1:nmaps) - real(c_double), target, intent(inout) :: map(0:geom_info%n_local - 1, 1:nmaps) - !-- - integer(c_int) :: mod_flags, ntrans, k - type(c_ptr), target :: alm_ptr(nmaps) - type(c_ptr), target :: map_ptr(nmaps) - - mod_flags = SHARP_DP - if (present(add) .and. add) then - mod_flags = or(mod_flags, SHARP_ADD) - end if - - if (spin == 0) then - ntrans = nmaps - else - ntrans = nmaps / 2 - end if - - ! Set up pointer table to access maps - alm_ptr(:) = c_null_ptr - map_ptr(:) = c_null_ptr - do k = 1, nmaps - if (alm_info%n_local > 0) alm_ptr(k) = c_loc(alm(0, k)) - if (geom_info%n_local > 0) map_ptr(k) = c_loc(map(0, k)) - end do - - if (present(comm)) then - call c_sharp_execute_mpi(comm, type, spin, alm_ptr, map_ptr, & - geom_info=geom_info%handle, & - alm_info=alm_info%handle, & - ntrans=ntrans, & - flags=mod_flags, & - time=time, & - opcnt=opcnt) - else - call c_sharp_execute(type, spin, alm_ptr, map_ptr, & - geom_info=geom_info%handle, & - alm_info=alm_info%handle, & - ntrans=ntrans, & - flags=mod_flags, & - time=time, & - opcnt=opcnt) - end if - end subroutine sharp_execute_d - - subroutine sharp_legendre_transform_d(bl, x, out) - use iso_c_binding - real(c_double) :: bl(:) - real(c_double) :: x(:), out(size(x)) - !-- - integer(c_intptr_t) :: lmax, nx - call c_sharp_legendre_transform(bl, lmax=int(size(bl) - 1, c_intptr_t), & - x=x, out=out, nx=int(size(x), c_intptr_t)) - end subroutine sharp_legendre_transform_d - - subroutine sharp_legendre_transform_s(bl, x, out) - use iso_c_binding - real(c_float) :: bl(:) - real(c_float) :: x(:), out(size(x)) - !-- - integer(c_intptr_t) :: lmax, nx - call c_sharp_legendre_transform_s(bl, lmax=int(size(bl) - 1, c_intptr_t), & - x=x, out=out, nx=int(size(x), c_intptr_t)) - end subroutine sharp_legendre_transform_s - - -end module diff --git a/fortran/test_sharp.f90 b/fortran/test_sharp.f90 deleted file mode 100644 index 0b7cce2..0000000 --- a/fortran/test_sharp.f90 +++ /dev/null @@ -1,84 +0,0 @@ -program test_sharp - use mpi - use sharp - use iso_c_binding, only : c_ptr, c_double - implicit none - - integer, parameter :: lmax = 2, nside = 2 - type(sharp_alm_info) :: alm_info - type(sharp_geom_info) :: geom_info - - real(c_double), dimension(0:(lmax + 1)**2 - 1, 1:1) :: alm - real(c_double), dimension(0:12*nside**2 - 1, 1:1) :: map - - integer(c_int), dimension(1:lmax + 1) :: ms - integer(c_int), dimension(1:4 * nside - 1) :: rings - integer(c_int) :: nm, m, nrings, iring - integer :: nodecount, rank, ierr - - call MPI_Init(ierr) - call MPI_Comm_size(MPI_COMM_WORLD, nodecount, ierr) - call MPI_Comm_rank(MPI_COMM_WORLD, rank, ierr) - - nm = 0 - do m = rank, lmax, nodecount - nm = nm + 1 - ms(nm) = m - end do - - nrings = 0 - do iring = rank + 1, 4 * nside - 1, nodecount - nrings = nrings + 1 - rings(nrings) = iring - end do - - alm = 0 - map = 0 - if (rank == 0) then - alm(0, 1) = 1 - end if - - print *, ms(1:nm) - call sharp_make_mmajor_real_packed_alm_info(lmax, ms=ms(1:nm), alm_info=alm_info) - print *, 'alm_info%n_local', alm_info%n_local - call sharp_make_healpix_geom_info(nside, rings=rings(1:nrings), geom_info=geom_info) - print *, 'geom_info%n_local', geom_info%n_local - print *, 'execute' - call sharp_execute(SHARP_Y, 0, 1, alm, alm_info, map, geom_info, comm=MPI_COMM_WORLD) - - print *, alm - print *, map - - call sharp_destroy_alm_info(alm_info) - call sharp_destroy_geom_info(geom_info) - print *, 'DONE' - call MPI_Finalize(ierr) - - print *, 'LEGENDRE TRANSFORMS' - - call test_legendre_transforms() - -contains - subroutine test_legendre_transforms() - integer, parameter :: lmax = 20, nx=10 - real(c_double) :: bl(0:lmax) - real(c_double) :: x(nx), out(nx) - real(c_float) :: out_s(nx) - !-- - integer :: l, i - - do l = 0, lmax - bl(l) = 1.0 / real(l + 1, c_double) - end do - do i = 1, nx - x(i) = 1 / real(i, c_double) - end do - out = 0 - call sharp_legendre_transform(bl, x, out) - print *, out - call sharp_legendre_transform(real(bl, c_float), real(x, c_float), out_s) - print *, out_s - end subroutine test_legendre_transforms - - -end program test_sharp diff --git a/libfftpack/README b/libfftpack/README deleted file mode 100644 index 2c7e7cb..0000000 --- a/libfftpack/README +++ /dev/null @@ -1,34 +0,0 @@ -ls_fft description: - -This package is intended to calculate one-dimensional real or complex FFTs -with high accuracy and good efficiency even for lengths containing large -prime factors. -The code is written in C, but a Fortran wrapper exists as well. - -Before any FFT is executed, a plan must be generated for it. Plan creation -is designed to be fast, so that there is no significant overhead if the -plan is only used once or a few times. - -The main component of the code is based on Paul N. Swarztrauber's FFTPACK in the -double precision incarnation by Hugh C. Pumphrey -(http://www.netlib.org/fftpack/dp.tgz). - -I replaced the iterative sine and cosine calculations in radfg() and radbg() -by an exact calculation, which slightly improves the transform accuracy for -real FFTs with lengths containing large prime factors. - -Since FFTPACK becomes quite slow for FFT lengths with large prime factors -(in the worst case of prime lengths it reaches O(n*n) complexity), I -implemented Bluestein's algorithm, which computes a FFT of length n by -several FFTs of length n2>=2*n-1 and a convolution. Since n2 can be chosen -to be highly composite, this algorithm is more efficient if n has large -prime factors. The longer FFTs themselves are then computed using the FFTPACK -routines. -Bluestein's algorithm was implemented according to the description at -http://en.wikipedia.org/wiki/Bluestein's_FFT_algorithm. - -Thread-safety: -All routines can be called concurrently; all information needed by ls_fft -is stored in the plan variable. However, using the same plan variable on -multiple threads simultaneously is not supported and will lead to data -corruption. diff --git a/libfftpack/bluestein.c b/libfftpack/bluestein.c deleted file mode 100644 index 2e2005c..0000000 --- a/libfftpack/bluestein.c +++ /dev/null @@ -1,173 +0,0 @@ -/* - * This file is part of libfftpack. - * - * libfftpack is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libfftpack is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libfftpack; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libfftpack is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/* - * Copyright (C) 2005, 2006, 2007, 2008 Max-Planck-Society - * \author Martin Reinecke - */ - -#include -#include -#include "fftpack.h" -#include "bluestein.h" - -/* returns the sum of all prime factors of n */ -size_t prime_factor_sum (size_t n) - { - size_t result=0,x,limit,tmp; - while (((tmp=(n>>1))<<1)==n) - { result+=2; n=tmp; } - - limit=(size_t)sqrt(n+0.01); - for (x=3; x<=limit; x+=2) - while ((tmp=(n/x))*x==n) - { - result+=x; - n=tmp; - limit=(size_t)sqrt(n+0.01); - } - if (n>1) result+=n; - - return result; - } - -/* returns the smallest composite of 2, 3 and 5 which is >= n */ -static size_t good_size(size_t n) - { - size_t f2, f23, f235, bestfac=2*n; - if (n<=6) return n; - - for (f2=1; f2=n) bestfac=f235; - return bestfac; - } - -void bluestein_i (size_t n, double **tstorage, size_t *worksize) - { - static const double pi=3.14159265358979323846; - size_t n2=good_size(n*2-1); - size_t m, coeff; - double angle, xn2; - double *bk, *bkf, *work; - double pibyn=pi/n; - *worksize=2+2*n+8*n2+16; - *tstorage = RALLOC(double,2+2*n+8*n2+16); - ((size_t *)(*tstorage))[0]=n2; - bk = *tstorage+2; - bkf = *tstorage+2+2*n; - work= *tstorage+2+2*(n+n2); - -/* initialize b_k */ - bk[0] = 1; - bk[1] = 0; - - coeff=0; - for (m=1; m=2*n) coeff-=2*n; - angle = pibyn*coeff; - bk[2*m] = cos(angle); - bk[2*m+1] = sin(angle); - } - -/* initialize the zero-padded, Fourier transformed b_k. Add normalisation. */ - xn2 = 1./n2; - bkf[0] = bk[0]*xn2; - bkf[1] = bk[1]*xn2; - for (m=2; m<2*n; m+=2) - { - bkf[m] = bkf[2*n2-m] = bk[m] *xn2; - bkf[m+1] = bkf[2*n2-m+1] = bk[m+1] *xn2; - } - for (m=2*n;m<=(2*n2-2*n+1);++m) - bkf[m]=0.; - cffti (n2,work); - cfftf (n2,bkf,work); - } - -void bluestein (size_t n, double *data, double *tstorage, int isign) - { - size_t n2=*((size_t *)tstorage); - size_t m; - double *bk, *bkf, *akf, *work; - bk = tstorage+2; - bkf = tstorage+2+2*n; - work= tstorage+2+2*(n+n2); - akf = tstorage+2+2*n+6*n2+16; - -/* initialize a_k and FFT it */ - if (isign>0) - for (m=0; m<2*n; m+=2) - { - akf[m] = data[m]*bk[m] - data[m+1]*bk[m+1]; - akf[m+1] = data[m]*bk[m+1] + data[m+1]*bk[m]; - } - else - for (m=0; m<2*n; m+=2) - { - akf[m] = data[m]*bk[m] + data[m+1]*bk[m+1]; - akf[m+1] =-data[m]*bk[m+1] + data[m+1]*bk[m]; - } - for (m=2*n; m<2*n2; ++m) - akf[m]=0; - - cfftf (n2,akf,work); - -/* do the convolution */ - if (isign>0) - for (m=0; m<2*n2; m+=2) - { - double im = -akf[m]*bkf[m+1] + akf[m+1]*bkf[m]; - akf[m ] = akf[m]*bkf[m] + akf[m+1]*bkf[m+1]; - akf[m+1] = im; - } - else - for (m=0; m<2*n2; m+=2) - { - double im = akf[m]*bkf[m+1] + akf[m+1]*bkf[m]; - akf[m ] = akf[m]*bkf[m] - akf[m+1]*bkf[m+1]; - akf[m+1] = im; - } - - -/* inverse FFT */ - cfftb (n2,akf,work); - -/* multiply by b_k* */ - if (isign>0) - for (m=0; m<2*n; m+=2) - { - data[m] = bk[m] *akf[m] - bk[m+1]*akf[m+1]; - data[m+1] = bk[m+1]*akf[m] + bk[m] *akf[m+1]; - } - else - for (m=0; m<2*n; m+=2) - { - data[m] = bk[m] *akf[m] + bk[m+1]*akf[m+1]; - data[m+1] =-bk[m+1]*akf[m] + bk[m] *akf[m+1]; - } - } diff --git a/libfftpack/bluestein.h b/libfftpack/bluestein.h deleted file mode 100644 index 91e5b28..0000000 --- a/libfftpack/bluestein.h +++ /dev/null @@ -1,48 +0,0 @@ -/* - * This file is part of libfftpack. - * - * libfftpack is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libfftpack is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libfftpack; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libfftpack is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/* - * Copyright (C) 2005 Max-Planck-Society - * \author Martin Reinecke - */ - -#ifndef PLANCK_BLUESTEIN_H -#define PLANCK_BLUESTEIN_H - -#include "c_utils.h" - -#ifdef __cplusplus -extern "C" { -#endif - -size_t prime_factor_sum (size_t n); - -void bluestein_i (size_t n, double **tstorage, size_t *worksize); -void bluestein (size_t n, double *data, double *tstorage, int isign); - -#ifdef __cplusplus -} -#endif - -#endif diff --git a/libfftpack/fftpack.c b/libfftpack/fftpack.c deleted file mode 100644 index 6d09d06..0000000 --- a/libfftpack/fftpack.c +++ /dev/null @@ -1,833 +0,0 @@ -/* - * This file is part of libfftpack. - * - * libfftpack is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libfftpack is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libfftpack; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libfftpack is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/* - fftpack.c : A set of FFT routines in C. - Algorithmically based on Fortran-77 FFTPACK by Paul N. Swarztrauber - (Version 4, 1985). - - C port by Martin Reinecke (2010) - */ - -#include -#include -#include -#include "fftpack.h" - -#define WA(x,i) wa[(i)+(x)*ido] -#define CH(a,b,c) ch[(a)+ido*((b)+l1*(c))] -#define CC(a,b,c) cc[(a)+ido*((b)+cdim*(c))] -#define PM(a,b,c,d) { a=c+d; b=c-d; } -#define PMC(a,b,c,d) { a.r=c.r+d.r; a.i=c.i+d.i; b.r=c.r-d.r; b.i=c.i-d.i; } -#define ADDC(a,b,c) { a.r=b.r+c.r; a.i=b.i+c.i; } -#define SCALEC(a,b) { a.r*=b; a.i*=b; } -#define CONJFLIPC(a) { double tmp_=a.r; a.r=-a.i; a.i=tmp_; } -/* (a+ib) = conj(c+id) * (e+if) */ -#define MULPM(a,b,c,d,e,f) { a=c*e+d*f; b=c*f-d*e; } - -typedef struct { - double r,i; -} cmplx; - -#define CONCAT(a,b) a ## b - -#define X(arg) CONCAT(passb,arg) -#define BACKWARD -#include "fftpack_inc.c" -#undef BACKWARD -#undef X - -#define X(arg) CONCAT(passf,arg) -#include "fftpack_inc.c" -#undef X - -#undef CC -#undef CH -#define CC(a,b,c) cc[(a)+ido*((b)+l1*(c))] -#define CH(a,b,c) ch[(a)+ido*((b)+cdim*(c))] - -static void radf2 (size_t ido, size_t l1, const double *cc, double *ch, - const double *wa) - { - const size_t cdim=2; - size_t i, k, ic; - double ti2, tr2; - - for (k=0; k=2*ip) aidx-=2*ip; - ar2=csarr[aidx]; - ai2=csarr[aidx+1]; - for(ik=0; ik=2*ip) aidx-=2*ip; - ar2=csarr[aidx]; - ai2=csarr[aidx+1]; - for(ik=0; ik0) ? passb4(ido, l1, p1, p2, wa+iw) - : passf4(ido, l1, p1, p2, wa+iw); - else if(ip==2) - (isign>0) ? passb2(ido, l1, p1, p2, wa+iw) - : passf2(ido, l1, p1, p2, wa+iw); - else if(ip==3) - (isign>0) ? passb3(ido, l1, p1, p2, wa+iw) - : passf3(ido, l1, p1, p2, wa+iw); - else if(ip==5) - (isign>0) ? passb5(ido, l1, p1, p2, wa+iw) - : passf5(ido, l1, p1, p2, wa+iw); - else if(ip==6) - (isign>0) ? passb6(ido, l1, p1, p2, wa+iw) - : passf6(ido, l1, p1, p2, wa+iw); - else - (isign>0) ? passbg(ido, ip, l1, p1, p2, wa+iw) - : passfg(ido, ip, l1, p1, p2, wa+iw); - SWAP(p1,p2,cmplx *); - l1=l2; - iw+=(ip-1)*ido; - } - if (p1!=c) - memcpy (c,p1,n*sizeof(cmplx)); - } - -void cfftf(size_t n, double c[], double wsave[]) - { - if (n!=1) - cfft1(n, (cmplx*)c, (cmplx*)wsave, (cmplx*)(wsave+2*n), - (size_t*)(wsave+4*n),-1); - } - -void cfftb(size_t n, double c[], double wsave[]) - { - if (n!=1) - cfft1(n, (cmplx*)c, (cmplx*)wsave, (cmplx*)(wsave+2*n), - (size_t*)(wsave+4*n),+1); - } - -static void factorize (size_t n, const size_t *pf, size_t npf, size_t *ifac) - { - size_t nl=n, nf=0, ntry=0, j=0, i; - -startloop: - j++; - ntry = (j<=npf) ? pf[j-1] : ntry+2; - do - { - size_t nq=nl / ntry; - size_t nr=nl-ntry*nq; - if (nr!=0) - goto startloop; - nf++; - ifac[nf+1]=ntry; - nl=nq; - if ((ntry==2) && (nf!=1)) - { - for (i=nf+1; i>2; --i) - ifac[i]=ifac[i-1]; - ifac[2]=2; - } - } - while(nl!=1); - ifac[0]=n; - ifac[1]=nf; - } - -static void cffti1(size_t n, double wa[], size_t ifac[]) - { - static const size_t ntryh[5]={4,6,3,2,5}; - static const double twopi=6.28318530717958647692; - size_t j, k, fi; - - double argh=twopi/n; - size_t i=0, l1=1; - factorize (n,ntryh,5,ifac); - for(k=1; k<=ifac[1]; k++) - { - size_t ip=ifac[k+1]; - size_t ido=n/(l1*ip); - for(j=1; j6) - { - wa[is ]=wa[i ]; - wa[is+1]=wa[i+1]; - } - } - l1*=ip; - } - } - -void cffti(size_t n, double wsave[]) - { if (n!=1) cffti1(n, wsave+2*n,(size_t*)(wsave+4*n)); } - - -/*---------------------------------------------------------------------- - rfftf1, rfftb1, rfftf, rfftb, rffti1, rffti. Real FFTs. - ----------------------------------------------------------------------*/ - -static void rfftf1(size_t n, double c[], double ch[], const double wa[], - const size_t ifac[]) - { - size_t k1, l1=n, nf=ifac[1], iw=n-1; - double *p1=ch, *p2=c; - - for(k1=1; k1<=nf;++k1) - { - size_t ip=ifac[nf-k1+2]; - size_t ido=n / l1; - l1 /= ip; - iw-=(ip-1)*ido; - SWAP (p1,p2,double *); - if(ip==4) - radf4(ido, l1, p1, p2, wa+iw); - else if(ip==2) - radf2(ido, l1, p1, p2, wa+iw); - else if(ip==3) - radf3(ido, l1, p1, p2, wa+iw); - else if(ip==5) - radf5(ido, l1, p1, p2, wa+iw); - else - { - if (ido==1) - SWAP (p1,p2,double *); - radfg(ido, ip, l1, ido*l1, p1, p2, wa+iw); - SWAP (p1,p2,double *); - } - } - if (p1==c) - memcpy (c,ch,n*sizeof(double)); - } - -static void rfftb1(size_t n, double c[], double ch[], const double wa[], - const size_t ifac[]) - { - size_t k1, l1=1, nf=ifac[1], iw=0; - double *p1=c, *p2=ch; - - for(k1=1; k1<=nf; k1++) - { - size_t ip = ifac[k1+1], - ido= n/(ip*l1); - if(ip==4) - radb4(ido, l1, p1, p2, wa+iw); - else if(ip==2) - radb2(ido, l1, p1, p2, wa+iw); - else if(ip==3) - radb3(ido, l1, p1, p2, wa+iw); - else if(ip==5) - radb5(ido, l1, p1, p2, wa+iw); - else - { - radbg(ido, ip, l1, ido*l1, p1, p2, wa+iw); - if (ido!=1) - SWAP (p1,p2,double *); - } - SWAP (p1,p2,double *); - l1*=ip; - iw+=(ip-1)*ido; - } - if (p1!=c) - memcpy (c,ch,n*sizeof(double)); - } - -void rfftf(size_t n, double r[], double wsave[]) - { if(n!=1) rfftf1(n, r, wsave, wsave+n,(size_t*)(wsave+2*n)); } - -void rfftb(size_t n, double r[], double wsave[]) - { if(n!=1) rfftb1(n, r, wsave, wsave+n,(size_t*)(wsave+2*n)); } - -static void rffti1(size_t n, double wa[], size_t ifac[]) - { - static const size_t ntryh[4]={4,2,3,5}; - static const double twopi=6.28318530717958647692; - size_t i, j, k, fi; - - double argh=twopi/n; - size_t is=0, l1=1; - factorize (n,ntryh,4,ifac); - for (k=1; kip) iang-=ip; - abr.r += ccl[l ].r*wal[iang].r; - abr.i += ccl[l ].i*wal[iang].r; - abi.r += ccl[lc].r*wal[iang].i; - abi.i += ccl[lc].i*wal[iang].i; - } -#ifndef BACKWARD - { abi.i=-abi.i; abi.r=-abi.r; } -#endif - CONJFLIPC(abi) - PMC(CH(i,k,j),CH(i,k,jc),abr,abi) - } - } - - DEALLOC(tarr); - - if (ido==1) return; - - for (j=1; j -
  • \ref fftgroup "Programming interface" - - */ diff --git a/libfftpack/ls_fft.c b/libfftpack/ls_fft.c deleted file mode 100644 index b1c0c96..0000000 --- a/libfftpack/ls_fft.c +++ /dev/null @@ -1,291 +0,0 @@ -/* - * This file is part of libfftpack. - * - * libfftpack is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libfftpack is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libfftpack; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libfftpack is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/* - * Copyright (C) 2005 Max-Planck-Society - * \author Martin Reinecke - */ - -#include -#include -#include -#include "bluestein.h" -#include "fftpack.h" -#include "ls_fft.h" - -complex_plan make_complex_plan (size_t length) - { - complex_plan plan = RALLOC(complex_plan_i,1); - size_t pfsum = prime_factor_sum(length); - double comp1 = (double)(length*pfsum); - double comp2 = 2*3*length*log(3.*length); - comp2*=3.; /* fudge factor that appears to give good overall performance */ - plan->length=length; - plan->bluestein = (comp2bluestein) - bluestein_i (length,&(plan->work),&(plan->worksize)); - else - { - plan->worksize=4*length+15; - plan->work=RALLOC(double,4*length+15); - cffti(length, plan->work); - } - return plan; - } - -complex_plan copy_complex_plan (complex_plan plan) - { - if (!plan) return NULL; - { - complex_plan newplan = RALLOC(complex_plan_i,1); - *newplan = *plan; - newplan->work=RALLOC(double,newplan->worksize); - memcpy(newplan->work,plan->work,sizeof(double)*newplan->worksize); - return newplan; - } - } - -void kill_complex_plan (complex_plan plan) - { - DEALLOC(plan->work); - DEALLOC(plan); - } - -void complex_plan_forward (complex_plan plan, double *data) - { - if (plan->bluestein) - bluestein (plan->length, data, plan->work, -1); - else - cfftf (plan->length, data, plan->work); - } - -void complex_plan_backward (complex_plan plan, double *data) - { - if (plan->bluestein) - bluestein (plan->length, data, plan->work, 1); - else - cfftb (plan->length, data, plan->work); - } - - -real_plan make_real_plan (size_t length) - { - real_plan plan = RALLOC(real_plan_i,1); - size_t pfsum = prime_factor_sum(length); - double comp1 = .5*length*pfsum; - double comp2 = 2*3*length*log(3.*length); - comp2*=3; /* fudge factor that appears to give good overall performance */ - plan->length=length; - plan->bluestein = (comp2bluestein) - bluestein_i (length,&(plan->work),&(plan->worksize)); - else - { - plan->worksize=2*length+15; - plan->work=RALLOC(double,2*length+15); - rffti(length, plan->work); - } - return plan; - } - -real_plan copy_real_plan (real_plan plan) - { - if (!plan) return NULL; - { - real_plan newplan = RALLOC(real_plan_i,1); - *newplan = *plan; - newplan->work=RALLOC(double,newplan->worksize); - memcpy(newplan->work,plan->work,sizeof(double)*newplan->worksize); - return newplan; - } - } - -void kill_real_plan (real_plan plan) - { - DEALLOC(plan->work); - DEALLOC(plan); - } - -void real_plan_forward_fftpack (real_plan plan, double *data) - { - if (plan->bluestein) - { - size_t m; - size_t n=plan->length; - double *tmp = RALLOC(double,2*n); - for (m=0; mwork,-1); - data[0] = tmp[0]; - memcpy (data+1, tmp+2, (n-1)*sizeof(double)); - DEALLOC(tmp); - } - else - rfftf (plan->length, data, plan->work); - } - -static void fftpack2halfcomplex (double *data, size_t n) - { - size_t m; - double *tmp = RALLOC(double,n); - tmp[0]=data[0]; - for (m=1; m<(n+1)/2; ++m) - { - tmp[m]=data[2*m-1]; - tmp[n-m]=data[2*m]; - } - if (!(n&1)) - tmp[n/2]=data[n-1]; - memcpy (data,tmp,n*sizeof(double)); - DEALLOC(tmp); - } - -static void halfcomplex2fftpack (double *data, size_t n) - { - size_t m; - double *tmp = RALLOC(double,n); - tmp[0]=data[0]; - for (m=1; m<(n+1)/2; ++m) - { - tmp[2*m-1]=data[m]; - tmp[2*m]=data[n-m]; - } - if (!(n&1)) - tmp[n-1]=data[n/2]; - memcpy (data,tmp,n*sizeof(double)); - DEALLOC(tmp); - } - -void real_plan_forward_fftw (real_plan plan, double *data) - { - real_plan_forward_fftpack (plan, data); - fftpack2halfcomplex (data,plan->length); - } - -void real_plan_backward_fftpack (real_plan plan, double *data) - { - if (plan->bluestein) - { - size_t m; - size_t n=plan->length; - double *tmp = RALLOC(double,2*n); - tmp[0]=data[0]; - tmp[1]=0.; - memcpy (tmp+2,data+1, (n-1)*sizeof(double)); - if ((n&1)==0) tmp[n+1]=0.; - for (m=2; mwork, 1); - for (m=0; mlength, data, plan->work); - } - -void real_plan_backward_fftw (real_plan plan, double *data) - { - halfcomplex2fftpack (data,plan->length); - real_plan_backward_fftpack (plan, data); - } - -void real_plan_forward_c (real_plan plan, double *data) - { - size_t m; - size_t n=plan->length; - - if (plan->bluestein) - { - for (m=1; m<2*n; m+=2) - data[m]=0; - bluestein (plan->length, data, plan->work, -1); - data[1]=0; - for (m=2; mwork); - data[0] = data[1]; - data[1] = 0; - for (m=2; mlength; - - if (plan->bluestein) - { - size_t m; - data[1]=0; - for (m=2; mlength, data, plan->work, 1); - for (m=1; m<2*n; m+=2) - data[m]=0; - } - else - { - ptrdiff_t m; - data[1] = data[0]; - rfftb (n, data+1, plan->work); - for (m=n-1; m>=0; --m) - { - data[2*m] = data[m+1]; - data[2*m+1] = 0.; - } - } - } diff --git a/libfftpack/ls_fft.h b/libfftpack/ls_fft.h deleted file mode 100644 index 2d555eb..0000000 --- a/libfftpack/ls_fft.h +++ /dev/null @@ -1,161 +0,0 @@ -/* - * This file is part of libfftpack. - * - * libfftpack is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libfftpack is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libfftpack; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libfftpack is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/*! \file ls_fft.h - * Interface for the LevelS FFT package. - * - * Copyright (C) 2004 Max-Planck-Society - * \author Martin Reinecke - */ - -#ifndef PLANCK_LS_FFT_H -#define PLANCK_LS_FFT_H - -#include "c_utils.h" - -#ifdef __cplusplus -extern "C" { -#endif - -/*!\defgroup fftgroup FFT interface -This package is intended to calculate one-dimensional real or complex FFTs -with high accuracy and good efficiency even for lengths containing large -prime factors. -The code is written in C, but a Fortran wrapper exists as well. - -Before any FFT is executed, a plan must be generated for it. Plan creation -is designed to be fast, so that there is no significant overhead if the -plan is only used once or a few times. - -The main component of the code is based on Paul N. Swarztrauber's FFTPACK in the -double precision incarnation by Hugh C. Pumphrey -(http://www.netlib.org/fftpack/dp.tgz). - -I replaced the iterative sine and cosine calculations in radfg() and radbg() -by an exact calculation, which slightly improves the transform accuracy for -real FFTs with lengths containing large prime factors. - -Since FFTPACK becomes quite slow for FFT lengths with large prime factors -(in the worst case of prime lengths it reaches \f$\mathcal{O}(n^2)\f$ -complexity), I implemented Bluestein's algorithm, which computes a FFT of length -\f$n\f$ by several FFTs of length \f$n_2\ge 2n-1\f$ and a convolution. Since -\f$n_2\f$ can be chosen to be highly composite, this algorithm is more efficient -if \f$n\f$ has large prime factors. The longer FFTs themselves are then computed -using the FFTPACK routines. -Bluestein's algorithm was implemented according to the description on Wikipedia -( -http://en.wikipedia.org/wiki/Bluestein%27s_FFT_algorithm). - -\b Thread-safety: -All routines can be called concurrently; all information needed by -ls_fft is stored in the plan variable. However, using the same plan -variable on multiple threads simultaneously is not supported and will lead to -data corruption. -*/ -/*! \{ */ - -typedef struct - { - double *work; - size_t length, worksize; - int bluestein; - } complex_plan_i; - -/*! The opaque handle type for complex-FFT plans. */ -typedef complex_plan_i * complex_plan; - -/*! Returns a plan for a complex FFT with \a length elements. */ -complex_plan make_complex_plan (size_t length); -/*! Constructs a copy of \a plan. */ -complex_plan copy_complex_plan (complex_plan plan); -/*! Destroys a plan for a complex FFT. */ -void kill_complex_plan (complex_plan plan); -/*! Computes a complex forward FFT on \a data, using \a plan. - \a Data has the form r0, i0, r1, i1, ..., - r[length-1], i[length-1]. */ -void complex_plan_forward (complex_plan plan, double *data); -/*! Computes a complex backward FFT on \a data, using \a plan. - \a Data has the form r0, i0, r1, i1, ..., - r[length-1], i[length-1]. */ -void complex_plan_backward (complex_plan plan, double *data); - -typedef struct - { - double *work; - size_t length, worksize; - int bluestein; - } real_plan_i; - -/*! The opaque handle type for real-FFT plans. */ -typedef real_plan_i * real_plan; - -/*! Returns a plan for a real FFT with \a length elements. */ -real_plan make_real_plan (size_t length); -/*! Constructs a copy of \a plan. */ -real_plan copy_real_plan (real_plan plan); -/*! Destroys a plan for a real FFT. */ -void kill_real_plan (real_plan plan); -/*! Computes a real forward FFT on \a data, using \a plan - and assuming the FFTPACK storage scheme: - - on entry, \a data has the form r0, r1, ..., r[length-1]; - - on exit, it has the form r0, r1, i1, r2, i2, ... - (a total of \a length values). */ -void real_plan_forward_fftpack (real_plan plan, double *data); -/*! Computes a real backward FFT on \a data, using \a plan - and assuming the FFTPACK storage scheme: - - on entry, \a data has the form r0, r1, i1, r2, i2, ... - (a total of \a length values); - - on exit, it has the form r0, r1, ..., r[length-1]. */ -void real_plan_backward_fftpack (real_plan plan, double *data); -/*! Computes a real forward FFT on \a data, using \a plan - and assuming the FFTW halfcomplex storage scheme: - - on entry, \a data has the form r0, r1, ..., r[length-1]; - - on exit, it has the form r0, r1, r2, ..., i2, i1. */ -void real_plan_forward_fftw (real_plan plan, double *data); -/*! Computes a real backward FFT on \a data, using \a plan - and assuming the FFTW halfcomplex storage scheme: - - on entry, \a data has the form r0, r1, r2, ..., i2, i1. - - on exit, it has the form r0, r1, ..., r[length-1]. */ -void real_plan_backward_fftw (real_plan plan, double *data); -/*! Computes a real forward FFT on \a data, using \a plan - and assuming a full-complex storage scheme: - - on entry, \a data has the form r0, [ignored], r1, [ignored], ..., - r[length-1], [ignored]; - - on exit, it has the form r0, i0, r1, i1, ..., - r[length-1], i[length-1]. */ -void real_plan_forward_c (real_plan plan, double *data); -/*! Computes a real backward FFT on \a data, using \a plan - and assuming a full-complex storage scheme: - - on entry, \a data has the form r0, i0, r1, i1, ..., - r[length-1], i[length-1]; - - on exit, it has the form r0, 0, r1, 0, ..., r[length-1], 0. */ -void real_plan_backward_c (real_plan plan, double *data); - -/*! \} */ - -#ifdef __cplusplus -} -#endif - -#endif diff --git a/libfftpack/planck.make b/libfftpack/planck.make deleted file mode 100644 index c171367..0000000 --- a/libfftpack/planck.make +++ /dev/null @@ -1,21 +0,0 @@ -PKG:=libfftpack - -SD:=$(SRCROOT)/$(PKG) -OD:=$(BLDROOT)/$(PKG) - -FULL_INCLUDE+= -I$(SD) - -HDR_$(PKG):=$(SD)/*.h -LIB_$(PKG):=$(LIBDIR)/libfftpack.a -OBJ:=fftpack.o bluestein.o ls_fft.o -OBJ:=$(OBJ:%=$(OD)/%) - -ODEP:=$(HDR_$(PKG)) $(HDR_c_utils) - -$(OD)/fftpack.o: $(SD)/fftpack_inc.c - -$(OBJ): $(ODEP) | $(OD)_mkdir -$(LIB_$(PKG)): $(OBJ) - -all_hdr+=$(HDR_$(PKG)) -all_lib+=$(LIB_$(PKG)) diff --git a/libsharp/libsharp.dox b/libsharp/libsharp.dox index 4c441f1..06361ee 100644 --- a/libsharp/libsharp.dox +++ b/libsharp/libsharp.dox @@ -70,8 +70,8 @@ libsharp supports shared-memory parallelisation via OpenMP; this feature will be automatically enabled if the compiler supports it. - Libsharp will also make use of SSE2 and AVX instructions when compiled for a - platform known to support them. + Libsharp will also make use of SSE2/AVX/AVX512 instructions when compiled + for a platform known to support them. Support for MPI-parallel transforms is also available; in this mode, every MPI task must provide a unique subset of the map and a_lm coefficients. diff --git a/libsharp/planck.make b/libsharp/planck.make deleted file mode 100644 index 76d534f..0000000 --- a/libsharp/planck.make +++ /dev/null @@ -1,29 +0,0 @@ -PKG:=libsharp - -SD:=$(SRCROOT)/$(PKG) -OD:=$(BLDROOT)/$(PKG) - -FULL_INCLUDE+= -I$(SD) - -HDR_$(PKG):=$(SD)/*.h -LIB_$(PKG):=$(LIBDIR)/libsharp.a -BIN:=sharp_testsuite -LIBOBJ:=sharp_ylmgen_c.o sharp.o sharp_announce.o sharp_geomhelpers.o sharp_almhelpers.o sharp_core.o sharp_legendre.o sharp_legendre_roots.o sharp_legendre_table.o -ALLOBJ:=$(LIBOBJ) sharp_testsuite.o -LIBOBJ:=$(LIBOBJ:%=$(OD)/%) -ALLOBJ:=$(ALLOBJ:%=$(OD)/%) - -ODEP:=$(HDR_$(PKG)) $(HDR_libfftpack) $(HDR_c_utils) -$(OD)/sharp_core.o: $(SD)/sharp_core_inchelper.c $(SD)/sharp_core_inc.c $(SD)/sharp_core_inc2.c -$(OD)/sharp.o: $(SD)/sharp_mpi.c -BDEP:=$(LIB_$(PKG)) $(LIB_libfftpack) $(LIB_c_utils) - -$(LIB_$(PKG)): $(LIBOBJ) - -$(ALLOBJ): $(ODEP) | $(OD)_mkdir -BIN:=$(BIN:%=$(BINDIR)/%) -$(BIN): $(BINDIR)/% : $(OD)/%.o $(BDEP) - -all_hdr+=$(HDR_$(PKG)) -all_lib+=$(LIB_$(PKG)) -all_cbin+=$(BIN) diff --git a/libsharp/sharp.c b/libsharp/sharp.c index 1eb8857..456b69c 100644 --- a/libsharp/sharp.c +++ b/libsharp/sharp.c @@ -25,17 +25,16 @@ /*! \file sharp.c * Spherical transform library * - * Copyright (C) 2006-2013 Max-Planck-Society + * Copyright (C) 2006-2016 Max-Planck-Society * \author Martin Reinecke \author Dag Sverre Seljebotn */ #include -#include "ls_fft.h" +#include +#include "pocketfft/pocketfft.h" #include "sharp_ylmgen_c.h" #include "sharp_internal.h" #include "c_utils.h" -#include "sharp_core.h" -#include "sharp_vecutil.h" #include "walltime_c.h" #include "sharp_almhelpers.h" #include "sharp_geomhelpers.h" @@ -63,7 +62,7 @@ static void get_chunk_info (int ndata, int nmult, int *nchunks, int *chunksize) *nchunks = (ndata+(*chunksize)-1)/(*chunksize); } -int sharp_get_mlim (int lmax, int spin, double sth, double cth) +NOINLINE int sharp_get_mlim (int lmax, int spin, double sth, double cth) { double ofs=lmax*0.01; if (ofs<100.) ofs=100.; @@ -82,24 +81,25 @@ typedef struct double phi0_; dcmplx *shiftarr; int s_shift; - real_plan plan; + rfft_plan plan; + int length; int norot; } ringhelper; static void ringhelper_init (ringhelper *self) { - static ringhelper rh_null = { 0, NULL, 0, NULL, 0 }; + static ringhelper rh_null = { 0, NULL, 0, NULL, 0, 0 }; *self = rh_null; } static void ringhelper_destroy (ringhelper *self) { - if (self->plan) kill_real_plan(self->plan); + if (self->plan) destroy_rfft_plan(self->plan); DEALLOC(self->shiftarr); ringhelper_init(self); } -static void ringhelper_update (ringhelper *self, int nph, int mmax, double phi0) +NOINLINE static void ringhelper_update (ringhelper *self, int nph, int mmax, double phi0) { self->norot = (fabs(phi0)<1e-14); if (!(self->norot)) @@ -108,14 +108,18 @@ static void ringhelper_update (ringhelper *self, int nph, int mmax, double phi0) RESIZE (self->shiftarr,dcmplx,mmax+1); self->s_shift = mmax+1; self->phi0_ = phi0; +// FIXME: improve this by using sincos2pibyn(nph) etc. for (int m=0; m<=mmax; ++m) self->shiftarr[m] = cos(m*phi0) + _Complex_I*sin(m*phi0); +// double *tmp=(double *) self->shiftarr; +// sincos_multi (mmax+1, phi0, &tmp[1], &tmp[0], 2); } - if (!self->plan) self->plan=make_real_plan(nph); - if (nph!=(int)self->plan->length) +// if (!self->plan) self->plan=make_rfft_plan(nph); + if (nph!=(int)self->length) { - kill_real_plan(self->plan); - self->plan=make_real_plan(nph); + if (self->plan) destroy_rfft_plan(self->plan); + self->plan=make_rfft_plan(nph); + self->length=nph; } } @@ -127,6 +131,7 @@ static int ringinfo_compare (const void *xa, const void *xb) static int ringpair_compare (const void *xa, const void *xb) { const sharp_ringpair *a=xa, *b=xb; +// return (a->r1.sth < b->r1.sth) ? -1 : (a->r1.sth > b->r1.sth) ? 1 : 0; if (a->r1.nph==b->r1.nph) return (a->r1.phi0 < b->r1.phi0) ? -1 : ((a->r1.phi0 > b->r1.phi0) ? 1 : @@ -261,6 +266,7 @@ void sharp_destroy_geom_info (sharp_geom_info *geom_info) distribution are permissible. */ static int sharp_get_mmax (int *mval, int nm) { + //FIXME: if gaps are allowed, we have to search the maximum m in the array int *mcheck=RALLOC(int,nm); SET_ARRAY(mcheck,0,nm,0); for (int i=0; i=2*mmax+1) { - for (int m=0; m<=mmax; ++m) - { - dcmplx tmp = phase[m*pstride]*wgt; - if(!self->norot) tmp*=self->shiftarr[m]; - data[2*m]=creal(tmp); - data[2*m+1]=cimag(tmp); - } + if (self->norot) + for (int m=0; m<=mmax; ++m) + { + data[2*m]=creal(phase[m*pstride])*wgt; + data[2*m+1]=cimag(phase[m*pstride])*wgt; + } + else + for (int m=0; m<=mmax; ++m) + { + dcmplx tmp = phase[m*pstride]*self->shiftarr[m]; + data[2*m]=creal(tmp)*wgt; + data[2*m+1]=cimag(tmp)*wgt; + } for (int m=2*(mmax+1); mplan, &(data[1])); + rfft_backward (self->plan, &(data[1]), 1.); } -static void ringhelper_ring2phase (ringhelper *self, +NOINLINE static void ringhelper_ring2phase (ringhelper *self, const sharp_ringinfo *info, double *data, int mmax, dcmplx *phase, int pstride, int flags) { @@ -342,7 +354,7 @@ static void ringhelper_ring2phase (ringhelper *self, if (flags&SHARP_REAL_HARMONICS) wgt *= sqrt_two; - real_plan_forward_fftpack (self->plan, &(data[1])); + rfft_forward (self->plan, &(data[1]), 1.); data[0]=data[1]; data[1]=data[nph+1]=0.; @@ -376,7 +388,7 @@ static void ringhelper_ring2phase (ringhelper *self, phase[m*pstride]=0.; } -static void fill_map (const sharp_geom_info *ginfo, void *map, double value, +NOINLINE static void clear_map (const sharp_geom_info *ginfo, void *map, int flags) { if (flags & SHARP_NO_FFT) @@ -386,50 +398,55 @@ static void fill_map (const sharp_geom_info *ginfo, void *map, double value, if (flags&SHARP_DP) { for (ptrdiff_t i=0;ipair[j].r1.nph;++i) - ((dcmplx *)map)[ginfo->pair[j].r1.ofs+i*ginfo->pair[j].r1.stride] - =value; + ((dcmplx *)map)[ginfo->pair[j].r1.ofs+i*ginfo->pair[j].r1.stride]=0; for (ptrdiff_t i=0;ipair[j].r2.nph;++i) - ((dcmplx *)map)[ginfo->pair[j].r2.ofs+i*ginfo->pair[j].r2.stride] - =value; + ((dcmplx *)map)[ginfo->pair[j].r2.ofs+i*ginfo->pair[j].r2.stride]=0; } else { for (ptrdiff_t i=0;ipair[j].r1.nph;++i) - ((fcmplx *)map)[ginfo->pair[j].r1.ofs+i*ginfo->pair[j].r1.stride] - =(float)value; + ((fcmplx *)map)[ginfo->pair[j].r1.ofs+i*ginfo->pair[j].r1.stride]=0; for (ptrdiff_t i=0;ipair[j].r2.nph;++i) - ((fcmplx *)map)[ginfo->pair[j].r2.ofs+i*ginfo->pair[j].r2.stride] - =(float)value; + ((fcmplx *)map)[ginfo->pair[j].r2.ofs+i*ginfo->pair[j].r2.stride]=0; } } } else { - for (int j=0;jnpairs;++j) + if (flags&SHARP_DP) { - if (flags&SHARP_DP) + for (int j=0;jnpairs;++j) { - for (ptrdiff_t i=0;ipair[j].r1.nph;++i) - ((double *)map)[ginfo->pair[j].r1.ofs+i*ginfo->pair[j].r1.stride] - =value; - for (ptrdiff_t i=0;ipair[j].r2.nph;++i) - ((double *)map)[ginfo->pair[j].r2.ofs+i*ginfo->pair[j].r2.stride] - =value; + double *dmap=(double *)map; + if (ginfo->pair[j].r1.stride==1) + memset(&dmap[ginfo->pair[j].r1.ofs],0, + ginfo->pair[j].r1.nph*sizeof(double)); + else + for (ptrdiff_t i=0;ipair[j].r1.nph;++i) + dmap[ginfo->pair[j].r1.ofs+i*ginfo->pair[j].r1.stride]=0; + if ((ginfo->pair[j].r2.nph>0)&&(ginfo->pair[j].r2.stride==1)) + memset(&dmap[ginfo->pair[j].r2.ofs],0, + ginfo->pair[j].r2.nph*sizeof(double)); + else + for (ptrdiff_t i=0;ipair[j].r2.nph;++i) + dmap[ginfo->pair[j].r2.ofs+i*ginfo->pair[j].r2.stride]=0; } - else + } + else + { + for (int j=0;jnpairs;++j) { for (ptrdiff_t i=0;ipair[j].r1.nph;++i) - ((float *)map)[ginfo->pair[j].r1.ofs+i*ginfo->pair[j].r1.stride] - =(float)value; + ((float *)map)[ginfo->pair[j].r1.ofs+i*ginfo->pair[j].r1.stride]=0; for (ptrdiff_t i=0;ipair[j].r2.nph;++i) - ((float *)map)[ginfo->pair[j].r2.ofs+i*ginfo->pair[j].r2.stride] - =(float)value; + ((float *)map)[ginfo->pair[j].r2.ofs+i*ginfo->pair[j].r2.stride]=0; } } } } -static void clear_alm (const sharp_alm_info *ainfo, void *alm, int flags) +NOINLINE static void clear_alm (const sharp_alm_info *ainfo, void *alm, + int flags) { #define CLEARLOOP(real_t,body) \ { \ @@ -465,59 +482,67 @@ static void clear_alm (const sharp_alm_info *ainfo, void *alm, int flags) } } -static void init_output (sharp_job *job) +NOINLINE static void init_output (sharp_job *job) { if (job->flags&SHARP_ADD) return; if (job->type == SHARP_MAP2ALM) - for (int i=0; intrans*job->nalm; ++i) + for (int i=0; inalm; ++i) clear_alm (job->ainfo,job->alm[i],job->flags); else - for (int i=0; intrans*job->nmaps; ++i) - fill_map (job->ginfo,job->map[i],0.,job->flags); + for (int i=0; inmaps; ++i) + clear_map (job->ginfo,job->map[i],job->flags); } -static void alloc_phase (sharp_job *job, int nm, int ntheta) +NOINLINE static void alloc_phase (sharp_job *job, int nm, int ntheta) { if (job->type==SHARP_MAP2ALM) { - if ((nm&1023)==0) nm+=3; // hack to avoid critical strides - job->s_m=2*job->ntrans*job->nmaps; + job->s_m=2*job->nmaps; + if (((job->s_m*16*nm)&1023)==0) nm+=3; // hack to avoid critical strides job->s_th=job->s_m*nm; } else { - if ((ntheta&1023)==0) ntheta+=3; // hack to avoid critical strides - job->s_th=2*job->ntrans*job->nmaps; + job->s_th=2*job->nmaps; + if (((job->s_th*16*ntheta)&1023)==0) ntheta+=3; // hack to avoid critical strides job->s_m=job->s_th*ntheta; } - job->phase=RALLOC(dcmplx,2*job->ntrans*job->nmaps*nm*ntheta); + job->phase=RALLOC(dcmplx,2*job->nmaps*nm*ntheta); } static void dealloc_phase (sharp_job *job) { DEALLOC(job->phase); } static void alloc_almtmp (sharp_job *job, int lmax) - { job->almtmp=RALLOC(dcmplx,job->ntrans*job->nalm*(lmax+1)); } + { job->almtmp=RALLOC(dcmplx,job->nalm*(lmax+2)); } static void dealloc_almtmp (sharp_job *job) { DEALLOC(job->almtmp); } -static void alm2almtmp (sharp_job *job, int lmax, int mi) +NOINLINE static void alm2almtmp (sharp_job *job, int lmax, int mi) { -#define COPY_LOOP(real_t, source_t, expr_of_x) \ - for (int l=job->ainfo->mval[mi]; l<=lmax; ++l) \ - for (int i=0; intrans*job->nalm; ++i) \ +#define COPY_LOOP(real_t, source_t, expr_of_x) \ + { \ + for (int l=m; lnalm; ++i) \ + job->almtmp[job->nalm*l+i] = 0; \ + for (int l=lmin; l<=lmax; ++l) \ + for (int i=0; inalm; ++i) \ { \ - source_t x = *(source_t *)(((real_t *)job->alm[i])+ofs+l*stride); \ - job->almtmp[job->ntrans*job->nalm*l+i] = expr_of_x; \ - } + source_t x = *(source_t *)(((real_t *)job->alm[i])+ofs+l*stride); \ + job->almtmp[job->nalm*l+i] = expr_of_x; \ + } \ + for (int i=0; inalm; ++i) \ + job->almtmp[job->nalm*(lmax+1)+i] = 0; \ + } if (job->type!=SHARP_MAP2ALM) { ptrdiff_t ofs=job->ainfo->mvstart[mi]; int stride=job->ainfo->stride; int m=job->ainfo->mval[mi]; + int lmin=(mspin) ? job->spin : m; /* in the case of SHARP_REAL_HARMONICS, phase2ring scales all the coefficients by sqrt_one_half; here we must compensate to avoid scaling m=0 */ @@ -562,20 +587,20 @@ static void alm2almtmp (sharp_job *job, int lmax, int mi) } } else - SET_ARRAY(job->almtmp,job->ntrans*job->nalm*job->ainfo->mval[mi], - job->ntrans*job->nalm*(lmax+1),0.); + memset (job->almtmp+job->nalm*job->ainfo->mval[mi], 0, + job->nalm*(lmax+2-job->ainfo->mval[mi])*sizeof(dcmplx)); #undef COPY_LOOP } -static void almtmp2alm (sharp_job *job, int lmax, int mi) +NOINLINE static void almtmp2alm (sharp_job *job, int lmax, int mi) { #define COPY_LOOP(real_t, target_t, expr_of_x) \ - for (int l=job->ainfo->mval[mi]; l<=lmax; ++l) \ - for (int i=0; intrans*job->nalm; ++i) \ + for (int l=lmin; l<=lmax; ++l) \ + for (int i=0; inalm; ++i) \ { \ - dcmplx x = job->almtmp[job->ntrans*job->nalm*l+i]; \ + dcmplx x = job->almtmp[job->nalm*l+i]; \ *(target_t *)(((real_t *)job->alm[i])+ofs+l*stride) += expr_of_x; \ } @@ -583,6 +608,7 @@ static void almtmp2alm (sharp_job *job, int lmax, int mi) ptrdiff_t ofs=job->ainfo->mvstart[mi]; int stride=job->ainfo->stride; int m=job->ainfo->mval[mi]; + int lmin=(mspin) ? job->spin : m; /* in the case of SHARP_REAL_HARMONICS, ring2phase scales all the coefficients by sqrt_two; here we must compensate to avoid scaling m=0 */ @@ -629,27 +655,56 @@ static void almtmp2alm (sharp_job *job, int lmax, int mi) #undef COPY_LOOP } -static void ringtmp2ring (sharp_job *job, sharp_ringinfo *ri, double *ringtmp, - int rstride) +NOINLINE static void ringtmp2ring (sharp_job *job, sharp_ringinfo *ri, + const double *ringtmp, int rstride) { - double **dmap = (double **)job->map; - float **fmap = (float **)job->map; - for (int i=0; intrans*job->nmaps; ++i) - for (int m=0; mnph; ++m) - if (job->flags & SHARP_DP) - dmap[i][ri->ofs+m*ri->stride] += ringtmp[i*rstride+m+1]; + if (job->flags & SHARP_DP) + { + double **dmap = (double **)job->map; + for (int i=0; inmaps; ++i) + { + double *restrict p1=&dmap[i][ri->ofs]; + const double *restrict p2=&ringtmp[i*rstride+1]; + if (ri->stride==1) + { + if (job->flags&SHARP_ADD) + for (int m=0; mnph; ++m) + p1[m] += p2[m]; + else + memcpy(p1,p2,ri->nph*sizeof(double)); + } else + for (int m=0; mnph; ++m) + p1[m*ri->stride] += p2[m]; + } + } + else + { + float **fmap = (float **)job->map; + for (int i=0; inmaps; ++i) + for (int m=0; mnph; ++m) fmap[i][ri->ofs+m*ri->stride] += (float)ringtmp[i*rstride+m+1]; + } } -static void ring2ringtmp (sharp_job *job, sharp_ringinfo *ri, double *ringtmp, - int rstride) +NOINLINE static void ring2ringtmp (sharp_job *job, sharp_ringinfo *ri, + double *ringtmp, int rstride) { - for (int i=0; intrans*job->nmaps; ++i) - for (int m=0; mnph; ++m) - ringtmp[i*rstride+m+1] = (job->flags & SHARP_DP) ? - ((double *)(job->map[i]))[ri->ofs+m*ri->stride] : - ((float *)(job->map[i]))[ri->ofs+m*ri->stride]; + if (job->flags & SHARP_DP) + for (int i=0; inmaps; ++i) + { + double *restrict p1=&ringtmp[i*rstride+1], + *restrict p2=&(((double *)(job->map[i]))[ri->ofs]); + if (ri->stride==1) + memcpy(p1,p2,ri->nph*sizeof(double)); + else + for (int m=0; mnph; ++m) + p1[m] = p2[m*ri->stride]; + } + else + for (int i=0; inmaps; ++i) + for (int m=0; mnph; ++m) + ringtmp[i*rstride+m+1] = ((float *)(job->map[i]))[ri->ofs+m*ri->stride]; } static void ring2phase_direct (sharp_job *job, sharp_ringinfo *ri, int mmax, @@ -657,7 +712,7 @@ static void ring2phase_direct (sharp_job *job, sharp_ringinfo *ri, int mmax, { if (ri->nph<0) { - for (int i=0; intrans*job->nmaps; ++i) + for (int i=0; inmaps; ++i) for (int m=0; m<=mmax; ++m) phase[2*i+job->s_m*m]=0.; } @@ -667,7 +722,7 @@ static void ring2phase_direct (sharp_job *job, sharp_ringinfo *ri, int mmax, double wgt = (job->flags&SHARP_USE_WEIGHTS) ? (ri->nph*ri->weight) : 1.; if (job->flags&SHARP_REAL_HARMONICS) wgt *= sqrt_two; - for (int i=0; intrans*job->nmaps; ++i) + for (int i=0; inmaps; ++i) for (int m=0; m<=mmax; ++m) phase[2*i+job->s_m*m]= (job->flags & SHARP_DP) ? ((dcmplx *)(job->map[i]))[ri->ofs+m*ri->stride]*wgt : @@ -684,7 +739,7 @@ static void phase2ring_direct (sharp_job *job, sharp_ringinfo *ri, int mmax, double wgt = (job->flags&SHARP_USE_WEIGHTS) ? (ri->nph*ri->weight) : 1.; if (job->flags&SHARP_REAL_HARMONICS) wgt *= sqrt_one_half; - for (int i=0; intrans*job->nmaps; ++i) + for (int i=0; inmaps; ++i) for (int m=0; m<=mmax; ++m) if (job->flags & SHARP_DP) dmap[i][ri->ofs+m*ri->stride] += wgt*phase[2*i+job->s_m*m]; @@ -693,7 +748,7 @@ static void phase2ring_direct (sharp_job *job, sharp_ringinfo *ri, int mmax, } //FIXME: set phase to zero if not SHARP_MAP2ALM? -static void map2phase (sharp_job *job, int mmax, int llim, int ulim) +NOINLINE static void map2phase (sharp_job *job, int mmax, int llim, int ulim) { if (job->type != SHARP_MAP2ALM) return; int pstride = job->s_m; @@ -715,19 +770,19 @@ static void map2phase (sharp_job *job, int mmax, int llim, int ulim) ringhelper helper; ringhelper_init(&helper); int rstride=job->ginfo->nphmax+2; - double *ringtmp=RALLOC(double,job->ntrans*job->nmaps*rstride); + double *ringtmp=RALLOC(double,job->nmaps*rstride); #pragma omp for schedule(dynamic,1) for (int ith=llim; iths_th*(ith-llim); ring2ringtmp(job,&(job->ginfo->pair[ith].r1),ringtmp,rstride); - for (int i=0; intrans*job->nmaps; ++i) + for (int i=0; inmaps; ++i) ringhelper_ring2phase (&helper,&(job->ginfo->pair[ith].r1), &ringtmp[i*rstride],mmax,&job->phase[dim2+2*i],pstride,job->flags); if (job->ginfo->pair[ith].r2.nph>0) { ring2ringtmp(job,&(job->ginfo->pair[ith].r2),ringtmp,rstride); - for (int i=0; intrans*job->nmaps; ++i) + for (int i=0; inmaps; ++i) ringhelper_ring2phase (&helper,&(job->ginfo->pair[ith].r2), &ringtmp[i*rstride],mmax,&job->phase[dim2+2*i+1],pstride,job->flags); } @@ -738,7 +793,7 @@ static void map2phase (sharp_job *job, int mmax, int llim, int ulim) } } -static void phase2map (sharp_job *job, int mmax, int llim, int ulim) +NOINLINE static void phase2map (sharp_job *job, int mmax, int llim, int ulim) { if (job->type == SHARP_MAP2ALM) return; int pstride = job->s_m; @@ -760,18 +815,18 @@ static void phase2map (sharp_job *job, int mmax, int llim, int ulim) ringhelper helper; ringhelper_init(&helper); int rstride=job->ginfo->nphmax+2; - double *ringtmp=RALLOC(double,job->ntrans*job->nmaps*rstride); + double *ringtmp=RALLOC(double,job->nmaps*rstride); #pragma omp for schedule(dynamic,1) for (int ith=llim; iths_th*(ith-llim); - for (int i=0; intrans*job->nmaps; ++i) + for (int i=0; inmaps; ++i) ringhelper_phase2ring (&helper,&(job->ginfo->pair[ith].r1), &ringtmp[i*rstride],mmax,&job->phase[dim2+2*i],pstride,job->flags); ringtmp2ring(job,&(job->ginfo->pair[ith].r1),ringtmp,rstride); if (job->ginfo->pair[ith].r2.nph>0) { - for (int i=0; intrans*job->nmaps; ++i) + for (int i=0; inmaps; ++i) ringhelper_phase2ring (&helper,&(job->ginfo->pair[ith].r2), &ringtmp[i*rstride],mmax,&job->phase[dim2+2*i+1],pstride,job->flags); ringtmp2ring(job,&(job->ginfo->pair[ith].r2),ringtmp,rstride); @@ -783,7 +838,7 @@ static void phase2map (sharp_job *job, int mmax, int llim, int ulim) } } -static void sharp_execute_job (sharp_job *job) +NOINLINE static void sharp_execute_job (sharp_job *job) { double timer=wallTime(); job->opcnt=0; @@ -798,8 +853,9 @@ static void sharp_execute_job (sharp_job *job) init_output (job); int nchunks, chunksize; - get_chunk_info(job->ginfo->npairs,(job->flags&SHARP_NVMAX)*VLEN,&nchunks, - &chunksize); + get_chunk_info(job->ginfo->npairs,sharp_veclen()*sharp_max_nvec(job->spin), + &nchunks,&chunksize); +//FIXME: needs to be changed to "nm" alloc_phase (job,mmax+1,chunksize); /* chunk loop */ @@ -863,10 +919,8 @@ static void sharp_execute_job (sharp_job *job) static void sharp_build_job_common (sharp_job *job, sharp_jobtype type, int spin, void *alm, void *map, const sharp_geom_info *geom_info, - const sharp_alm_info *alm_info, int ntrans, int flags) + const sharp_alm_info *alm_info, int flags) { - UTIL_ASSERT((ntrans>0)&&(ntrans<=SHARP_MAXTRANS), - "bad number of simultaneous transforms"); if (type==SHARP_ALM2MAP_DERIV1) spin=1; if (type==SHARP_MAP2ALM) flags|=SHARP_USE_WEIGHTS; if (type==SHARP_Yt) type=SHARP_MAP2ALM; @@ -881,24 +935,21 @@ static void sharp_build_job_common (sharp_job *job, sharp_jobtype type, job->ginfo = geom_info; job->ainfo = alm_info; job->flags = flags; - if ((job->flags&SHARP_NVMAX)==0) - job->flags|=sharp_nv_oracle (type, spin, ntrans); if (alm_info->flags&SHARP_REAL_HARMONICS) job->flags|=SHARP_REAL_HARMONICS; job->time = 0.; job->opcnt = 0; - job->ntrans = ntrans; job->alm=alm; job->map=map; } void sharp_execute (sharp_jobtype type, int spin, void *alm, void *map, - const sharp_geom_info *geom_info, const sharp_alm_info *alm_info, int ntrans, + const sharp_geom_info *geom_info, const sharp_alm_info *alm_info, int flags, double *time, unsigned long long *opcnt) { sharp_job job; sharp_build_job_common (&job, type, spin, alm, map, geom_info, alm_info, - ntrans, flags); + flags); sharp_execute_job (&job); if (time!=NULL) *time = job.time; @@ -910,96 +961,16 @@ void sharp_set_chunksize_min(int new_chunksize_min) void sharp_set_nchunks_max(int new_nchunks_max) { nchunks_max=new_nchunks_max; } -int sharp_get_nv_max (void) -{ return 6; } - -static int sharp_oracle (sharp_jobtype type, int spin, int ntrans) - { - int lmax=511; - int mmax=(lmax+1)/2; - int nrings=(lmax+1)/4; - int ppring=1; - - spin = (spin!=0) ? 2 : 0; - - ptrdiff_t npix=(ptrdiff_t)nrings*ppring; - sharp_geom_info *tinfo; - sharp_make_gauss_geom_info (nrings, ppring, 0., 1, ppring, &tinfo); - - ptrdiff_t nalms = ((mmax+1)*(mmax+2))/2 + (mmax+1)*(lmax-mmax); - int ncomp = ntrans*((spin==0) ? 1 : 2); - - double **map; - ALLOC2D(map,double,ncomp,npix); - SET_ARRAY(map[0],0,npix*ncomp,0.); - - sharp_alm_info *alms; - sharp_make_triangular_alm_info(lmax,mmax,1,&alms); - - dcmplx **alm; - ALLOC2D(alm,dcmplx,ncomp,nalms); - SET_ARRAY(alm[0],0,nalms*ncomp,0.); - - double time=1e30; - int nvbest=-1; - - for (int nv=1; nv<=sharp_get_nv_max(); ++nv) - { - double time_acc=0.; - double jtime; - int ntries=0; - do - { - sharp_execute(type,spin,&alm[0],&map[0],tinfo,alms,ntrans, - nv|SHARP_DP|SHARP_NO_OPENMP,&jtime,NULL); - - if (jtime0),"bad number of simultaneous transforms"); - UTIL_ASSERT(spin>=0, "bad spin"); - ntrans=IMIN(ntrans,maxtr); - - if (nv_opt[ntrans-1][spin!=0][type]==0) - nv_opt[ntrans-1][spin!=0][type]=sharp_oracle(type,spin,ntrans); - return nv_opt[ntrans-1][spin!=0][type]; - } - #ifdef USE_MPI #include "sharp_mpi.c" int sharp_execute_mpi_maybe (void *pcomm, sharp_jobtype type, int spin, void *alm, void *map, const sharp_geom_info *geom_info, - const sharp_alm_info *alm_info, int ntrans, int flags, double *time, + const sharp_alm_info *alm_info, int flags, double *time, unsigned long long *opcnt) { MPI_Comm comm = *(MPI_Comm*)pcomm; - sharp_execute_mpi((MPI_Comm)comm, type, spin, alm, map, geom_info, alm_info, ntrans, + sharp_execute_mpi((MPI_Comm)comm, type, spin, alm, map, geom_info, alm_info, flags, time, opcnt); return 0; } @@ -1008,12 +979,12 @@ int sharp_execute_mpi_maybe (void *pcomm, sharp_jobtype type, int spin, int sharp_execute_mpi_maybe (void *pcomm, sharp_jobtype type, int spin, void *alm, void *map, const sharp_geom_info *geom_info, - const sharp_alm_info *alm_info, int ntrans, int flags, double *time, + const sharp_alm_info *alm_info, int flags, double *time, unsigned long long *opcnt) { /* Suppress unused warning: */ (void)pcomm; (void)type; (void)spin; (void)alm; (void)map; (void)geom_info; - (void)alm_info; (void)ntrans; (void)flags; (void)time; (void)opcnt; + (void)alm_info; (void)flags; (void)time; (void)opcnt; return SHARP_ERROR_NO_MPI; } diff --git a/libsharp/sharp.h b/libsharp/sharp.h index 6722aee..0e43ba9 100644 --- a/libsharp/sharp.h +++ b/libsharp/sharp.h @@ -23,24 +23,243 @@ */ /*! \file sharp.h - * Interface for the spherical transform library. + * Portable interface for the spherical transform library. * - * Copyright (C) 2006-2012 Max-Planck-Society - * \author Martin Reinecke + * Copyright (C) 2012-2019 Max-Planck-Society + * \author Martin Reinecke \author Dag Sverre Seljebotn */ #ifndef PLANCK_SHARP_H #define PLANCK_SHARP_H +#include + #ifdef __cplusplus -#error This header file cannot be included from C++, only from C +extern "C" { #endif -#include +/*! \internal + Helper type containing information about a single ring. */ +typedef struct + { + double theta, phi0, weight, cth, sth; + ptrdiff_t ofs; + int nph, stride; + } sharp_ringinfo; -#include "sharp_lowlevel.h" -#include "sharp_legendre.h" -#include "sharp_legendre_roots.h" -#include "sharp_legendre_table.h" +/*! \internal + Helper type containing information about a pair of rings with colatitudes + symmetric around the equator. */ +typedef struct + { + sharp_ringinfo r1,r2; + } sharp_ringpair; + +/*! \internal + Type holding all required information about a map geometry. */ +typedef struct + { + sharp_ringpair *pair; + int npairs, nphmax; + } sharp_geom_info; + +/*! \defgroup almgroup Helpers for dealing with a_lm */ +/*! \{ */ + +/*! \internal + Helper type for index calculation in a_lm arrays. */ +typedef struct + { + /*! Maximum \a l index of the array */ + int lmax; + /*! Number of different \a m values in this object */ + int nm; + /*! Array with \a nm entries containing the individual m values */ + int *mval; + /*! Combination of flags from sharp_almflags */ + int flags; + /*! Array with \a nm entries containing the (hypothetical) indices of + the coefficients with quantum numbers 0,\a mval[i] */ + ptrdiff_t *mvstart; + /*! Stride between a_lm and a_(l+1),m */ + ptrdiff_t stride; + } sharp_alm_info; + +/*! alm_info flags */ +typedef enum { SHARP_PACKED = 1, + /*!< m=0-coefficients are packed so that the (zero) imaginary part is + not present. mvstart is in units of *real* float/double for all + m; stride is in units of reals for m=0 and complex for m!=0 */ + SHARP_REAL_HARMONICS = 1<<6 + /*!< Use the real spherical harmonic convention. For + m==0, the alm are treated exactly the same as in + the complex case. For m!=0, alm[i] represent a + pair (+abs(m), -abs(m)) instead of (real, imag), + and the coefficients are scaled by a factor of + sqrt(2) relative to the complex case. In other + words, (sqrt(.5) * alm[i]) recovers the + corresponding complex coefficient (when accessed + as complex). + */ + } sharp_almflags; + + + +/*! Creates an a_lm data structure from the following parameters: + \param lmax maximum \a l quantum number (>=0) + \param mmax maximum \a m quantum number (0<= \a mmax <= \a lmax) + \param stride the stride between entries with identical \a m, and \a l + differing by 1. + \param mstart the index of the (hypothetical) coefficient with the + quantum numbers 0,\a m. Must have \a mmax+1 entries. + \param alm_info will hold a pointer to the newly created data structure + */ +void sharp_make_alm_info (int lmax, int mmax, int stride, + const ptrdiff_t *mstart, sharp_alm_info **alm_info); +/*! Creates an a_lm data structure which from the following parameters: + \param lmax maximum \a l quantum number (\a >=0) + \param nm number of different \a m (\a 0<=nm<=lmax+1) + \param stride the stride between entries with identical \a m, and \a l + differing by 1. + \param mval array with \a nm entries containing the individual m values + \param mvstart array with \a nm entries containing the (hypothetical) + indices of the coefficients with the quantum numbers 0,\a mval[i] + \param flags a combination of sharp_almflags (pass 0 unless you know you need this) + \param alm_info will hold a pointer to the newly created data structure + */ +void sharp_make_general_alm_info (int lmax, int nm, int stride, const int *mval, + const ptrdiff_t *mvstart, int flags, sharp_alm_info **alm_info); +/*! Returns the index of the coefficient with quantum numbers \a l, + \a mval[mi]. + \note for a \a sharp_alm_info generated by sharp_make_alm_info() this is + the index for the coefficient with the quantum numbers \a l, \a mi. */ +ptrdiff_t sharp_alm_index (const sharp_alm_info *self, int l, int mi); +/*! Returns the number of alm coefficients described by \a self. If the SHARP_PACKED + flag is set, this is number of "real" coeffecients (for m < 0 and m >= 0), + otherwise it is the number of complex coefficients (with m>=0). */ +ptrdiff_t sharp_alm_count(const sharp_alm_info *self); +/*! Deallocates the a_lm info object. */ +void sharp_destroy_alm_info (sharp_alm_info *info); + +/*! \} */ + +/*! \defgroup geominfogroup Functions for dealing with geometry information */ +/*! \{ */ + +/*! Creates a geometry information from a set of ring descriptions. + All arrays passed to this function must have \a nrings elements. + \param nrings the number of rings in the map + \param nph the number of pixels in each ring + \param ofs the index of the first pixel in each ring in the map array + \param stride the stride between consecutive pixels + \param phi0 the azimuth (in radians) of the first pixel in each ring + \param theta the colatitude (in radians) of each ring + \param wgt the pixel weight to be used for the ring in map2alm + and adjoint map2alm transforms. + Pass NULL to use 1.0 as weight for all rings. + \param geom_info will hold a pointer to the newly created data structure + */ +void sharp_make_geom_info (int nrings, const int *nph, const ptrdiff_t *ofs, + const int *stride, const double *phi0, const double *theta, + const double *wgt, sharp_geom_info **geom_info); + +/*! Counts the number of grid points needed for (the local part of) a map described + by \a info. + */ +ptrdiff_t sharp_map_size(const sharp_geom_info *info); + +/*! Deallocates the geometry information in \a info. */ +void sharp_destroy_geom_info (sharp_geom_info *info); + +/*! \} */ + +/*! \defgroup lowlevelgroup Low-level libsharp SHT interface */ +/*! \{ */ + +/*! Enumeration of SHARP job types. */ +typedef enum { SHARP_YtW=0, /*!< analysis */ + SHARP_MAP2ALM=SHARP_YtW, /*!< analysis */ + SHARP_Y=1, /*!< synthesis */ + SHARP_ALM2MAP=SHARP_Y, /*!< synthesis */ + SHARP_Yt=2, /*!< adjoint synthesis */ + SHARP_WY=3, /*!< adjoint analysis */ + SHARP_ALM2MAP_DERIV1=4 /*!< synthesis of first derivatives */ + } sharp_jobtype; + +/*! Job flags */ +typedef enum { SHARP_DP = 1<<4, + /*!< map and a_lm are in double precision */ + SHARP_ADD = 1<<5, + /*!< results are added to the output arrays, instead of + overwriting them */ + + /* NOTE: SHARP_REAL_HARMONICS, 1<<6, is also available in sharp_jobflags, + but its use here is deprecated in favor of having it in the sharp_alm_info */ + + SHARP_NO_FFT = 1<<7, + + SHARP_USE_WEIGHTS = 1<<20, /* internal use only */ + SHARP_NO_OPENMP = 1<<21, /* internal use only */ + } sharp_jobflags; + +/*! Performs a libsharp SHT job. The interface deliberately does not use + the C99 "complex" data type, in order to be callable from C89 and C++. + \param type the type of SHT + \param spin the spin of the quantities to be transformed + \param alm contains pointers to the a_lm coefficients. If \a spin==0, + alm[0] points to the a_lm of the SHT. If \a spin>0, alm[0] and alm[1] + point to the two a_lm sets of the SHT. The exact data type of \a alm + depends on whether the SHARP_DP flag is set. + \param map contains pointers to the maps. If \a spin==0, + map[0] points to the map of the SHT. If \a spin>0, or \a type is + SHARP_ALM2MAP_DERIV1, map[0] and map[1] point to the two maps of the SHT. + The exact data type of \a map depends on whether the SHARP_DP flag is set. + \param geom_info A \c sharp_geom_info object compatible with the provided + \a map arrays. + \param alm_info A \c sharp_alm_info object compatible with the provided + \a alm arrays. All \c m values from 0 to some \c mmax<=lmax must be present + exactly once. + \param flags See sharp_jobflags. In particular, if SHARP_DP is set, then + \a alm is expected to have the type "complex double **" and \a map is + expected to have the type "double **"; otherwise, the expected + types are "complex float **" and "float **", respectively. + \param time If not NULL, the wall clock time required for this SHT + (in seconds) will be written here. + \param opcnt If not NULL, a conservative estimate of the total floating point + operation count for this SHT will be written here. */ +void sharp_execute (sharp_jobtype type, int spin, void *alm, void *map, + const sharp_geom_info *geom_info, const sharp_alm_info *alm_info, + int flags, double *time, unsigned long long *opcnt); + +void sharp_set_chunksize_min(int new_chunksize_min); +void sharp_set_nchunks_max(int new_nchunks_max); + + +typedef enum { SHARP_ERROR_NO_MPI = 1, + /*!< libsharp not compiled with MPI support */ + } sharp_errors; + +/*! Works like sharp_execute_mpi, but is always present whether or not libsharp + is compiled with USE_MPI. This is primarily useful for wrapper code etc. + + Note that \a pcomm has the type MPI_Comm*, except we declare void* to avoid + pulling in MPI headers. I.e., the comm argument of sharp_execute_mpi + is *(MPI_Comm*)pcomm. + + Other parameters are the same as sharp_execute_mpi. + + Returns 0 if successful, or SHARP_ERROR_NO_MPI if MPI is not available + (in which case nothing is done). + */ +int sharp_execute_mpi_maybe (void *pcomm, sharp_jobtype type, int spin, + void *alm, void *map, const sharp_geom_info *geom_info, + const sharp_alm_info *alm_info, int flags, double *time, + unsigned long long *opcnt); + +/*! \} */ + +#ifdef __cplusplus +} +#endif #endif diff --git a/libsharp/sharp_almhelpers.c b/libsharp/sharp_almhelpers.c index 6a98309..12ce600 100644 --- a/libsharp/sharp_almhelpers.c +++ b/libsharp/sharp_almhelpers.c @@ -25,7 +25,7 @@ /*! \file sharp_almhelpers.c * Spherical transform library * - * Copyright (C) 2008-2013 Max-Planck-Society + * Copyright (C) 2008-2016 Max-Planck-Society * \author Martin Reinecke */ diff --git a/libsharp/sharp_almhelpers.h b/libsharp/sharp_almhelpers.h index 3bff317..06bee8f 100644 --- a/libsharp/sharp_almhelpers.h +++ b/libsharp/sharp_almhelpers.h @@ -25,14 +25,14 @@ /*! \file sharp_almhelpers.h * SHARP helper function for the creation of a_lm data structures * - * Copyright (C) 2008-2011 Max-Planck-Society + * Copyright (C) 2008-2019 Max-Planck-Society * \author Martin Reinecke */ #ifndef PLANCK_SHARP_ALMHELPERS_H #define PLANCK_SHARP_ALMHELPERS_H -#include "sharp_lowlevel.h" +#include "sharp.h" #ifdef __cplusplus extern "C" { diff --git a/libsharp/sharp_announce.c b/libsharp/sharp_announce.c deleted file mode 100644 index bc6ee50..0000000 --- a/libsharp/sharp_announce.c +++ /dev/null @@ -1,98 +0,0 @@ -/* - * This file is part of libc_utils. - * - * libc_utils is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libc_utils is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libc_utils; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libc_utils is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/*! \file sharp_announce.c - * Banner for module startup - * - * Copyright (C) 2012 Max-Planck-Society - * \author Martin Reinecke - */ - -#include -#include -#include -#ifdef _OPENMP -#include -#endif -#ifdef USE_MPI -#include -#endif - -#include "sharp_announce.h" -#include "sharp_vecutil.h" - -static void OpenMP_status(void) - { -#ifndef _OPENMP - printf("OpenMP: not supported by this binary\n"); -#else - int threads = omp_get_max_threads(); - if (threads>1) - printf("OpenMP active: max. %d threads.\n",threads); - else - printf("OpenMP active, but running with 1 thread only.\n"); -#endif - } - -static void MPI_status(void) - { -#ifndef USE_MPI - printf("MPI: not supported by this binary\n"); -#else - int tasks; - MPI_Comm_size(MPI_COMM_WORLD,&tasks); - if (tasks>1) - printf("MPI active with %d tasks.\n",tasks); - else - printf("MPI active, but running with 1 task only.\n"); -#endif - } - -static void vecmath_status(void) - { printf("Supported vector length: %d\n",VLEN); } - -void sharp_announce (const char *name) - { - size_t m, nlen=strlen(name); - printf("\n+-"); - for (m=0; m -#include -#include "sharp_vecsupport.h" - -#define UNSAFE_CODE - -#if (VLEN==1) - -static inline complex double vhsum_cmplx(Tv a, Tv b) - { return a+_Complex_I*b; } - -static inline void vhsum_cmplx2 (Tv a, Tv b, Tv c, Tv d, - complex double * restrict c1, complex double * restrict c2) - { *c1 += a+_Complex_I*b; *c2 += c+_Complex_I*d; } - -#endif - -#if (VLEN==2) - -static inline complex double vhsum_cmplx (Tv a, Tv b) - { -#if defined(__SSE3__) - Tv tmp = _mm_hadd_pd(a,b); -#else - Tv tmp = vadd(_mm_shuffle_pd(a,b,_MM_SHUFFLE2(0,1)), - _mm_shuffle_pd(a,b,_MM_SHUFFLE2(1,0))); -#endif - union {Tv v; complex double c; } u; - u.v=tmp; return u.c; - } - -static inline void vhsum_cmplx2 (Tv a, Tv b, Tv c, - Tv d, complex double * restrict c1, complex double * restrict c2) - { -#ifdef UNSAFE_CODE -#if defined(__SSE3__) - vaddeq(*((__m128d *)c1),_mm_hadd_pd(a,b)); - vaddeq(*((__m128d *)c2),_mm_hadd_pd(c,d)); -#else - vaddeq(*((__m128d *)c1),vadd(_mm_shuffle_pd(a,b,_MM_SHUFFLE2(0,1)), - _mm_shuffle_pd(a,b,_MM_SHUFFLE2(1,0)))); - vaddeq(*((__m128d *)c2),vadd(_mm_shuffle_pd(c,d,_MM_SHUFFLE2(0,1)), - _mm_shuffle_pd(c,d,_MM_SHUFFLE2(1,0)))); -#endif -#else - union {Tv v; complex double c; } u1, u2; -#if defined(__SSE3__) - u1.v = _mm_hadd_pd(a,b); u2.v=_mm_hadd_pd(c,d); -#else - u1.v = vadd(_mm_shuffle_pd(a,b,_MM_SHUFFLE2(0,1)), - _mm_shuffle_pd(a,b,_MM_SHUFFLE2(1,0))); - u2.v = vadd(_mm_shuffle_pd(c,d,_MM_SHUFFLE2(0,1)), - _mm_shuffle_pd(c,d,_MM_SHUFFLE2(1,0))); -#endif - *c1+=u1.c; *c2+=u2.c; -#endif - } - -#endif - -#if (VLEN==4) - -static inline complex double vhsum_cmplx (Tv a, Tv b) - { - Tv tmp=_mm256_hadd_pd(a,b); - Tv tmp2=_mm256_permute2f128_pd(tmp,tmp,1); - tmp=_mm256_add_pd(tmp,tmp2); -#ifdef UNSAFE_CODE - complex double ret; - *((__m128d *)&ret)=_mm256_extractf128_pd(tmp, 0); - return ret; -#else - union {Tv v; complex double c[2]; } u; - u.v=tmp; return u.c[0]; -#endif - } - -static inline void vhsum_cmplx2 (Tv a, Tv b, Tv c, Tv d, - complex double * restrict c1, complex double * restrict c2) - { - Tv tmp1=_mm256_hadd_pd(a,b), tmp2=_mm256_hadd_pd(c,d); - Tv tmp3=_mm256_permute2f128_pd(tmp1,tmp2,49), - tmp4=_mm256_permute2f128_pd(tmp1,tmp2,32); - tmp1=vadd(tmp3,tmp4); -#ifdef UNSAFE_CODE - *((__m128d *)c1)=_mm_add_pd(*((__m128d *)c1),_mm256_extractf128_pd(tmp1, 0)); - *((__m128d *)c2)=_mm_add_pd(*((__m128d *)c2),_mm256_extractf128_pd(tmp1, 1)); -#else - union {Tv v; complex double c[2]; } u; - u.v=tmp1; - *c1+=u.c[0]; *c2+=u.c[1]; -#endif - } - -#endif - -#if (VLEN==8) - -static inline complex double vhsum_cmplx(Tv a, Tv b) - { return _mm512_reduce_add_pd(a)+_Complex_I*_mm512_reduce_add_pd(b); } - -static inline void vhsum_cmplx2 (Tv a, Tv b, Tv c, Tv d, - complex double * restrict c1, complex double * restrict c2) - { - *c1 += _mm512_reduce_add_pd(a)+_Complex_I*_mm512_reduce_add_pd(b); - *c2 += _mm512_reduce_add_pd(c)+_Complex_I*_mm512_reduce_add_pd(d); - } - -#endif - -#endif diff --git a/libsharp/sharp_core.c b/libsharp/sharp_core.c index 7cd2f17..036a8ed 100644 --- a/libsharp/sharp_core.c +++ b/libsharp/sharp_core.c @@ -1,240 +1,116 @@ -/* - * This file is part of libsharp. - * - * libsharp is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libsharp is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libsharp; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ +#define XCONCATX(a,b) a##_##b +#define XCONCATX2(a,b) XCONCATX(a,b) +#define XARCH(a) XCONCATX2(a,ARCH) -/* - * libsharp is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ +#define ARCH default +#include "sharp_core_inc.c" +#undef ARCH -/*! \file sharp_core.c - * Computational core - * - * Copyright (C) 2012-2013 Max-Planck-Society - * \author Martin Reinecke - */ +typedef void (*t_inner_loop) (sharp_job *job, const int *ispair, + const double *cth_, const double *sth_, int llim, int ulim, + sharp_Ylmgen_C *gen, int mi, const int *mlim); +typedef int (*t_veclen) (void); +typedef int (*t_max_nvec) (int spin); +typedef const char *(*t_architecture) (void); -#include -#include -#include -#include "sharp_vecsupport.h" -#include "sharp_complex_hacks.h" -#include "sharp_ylmgen_c.h" -#include "sharp.h" -#include "sharp_core.h" -#include "c_utils.h" +static t_inner_loop inner_loop_ = NULL; +static t_veclen veclen_ = NULL; +static t_max_nvec max_nvec_ = NULL; +static t_architecture architecture_ = NULL; -typedef complex double dcmplx; +#if defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=6) -// must be in the range [0;6] -#define MAXJOB_SPECIAL 2 +#define DECL(arch) \ +static int XCONCATX2(have,arch)(void) \ + { \ + static int res=-1; \ + if (res<0) \ + { \ + __builtin_cpu_init(); \ + res = __builtin_cpu_supports(#arch); \ + } \ + return res; \ + } \ +\ +void XCONCATX2(inner_loop,arch) (sharp_job *job, const int *ispair, \ + const double *cth_, const double *sth_, int llim, int ulim, \ + sharp_Ylmgen_C *gen, int mi, const int *mlim); \ +int XCONCATX2(sharp_veclen,arch) (void); \ +int XCONCATX2(sharp_max_nvec,arch) (int spin); \ +const char *XCONCATX2(sharp_architecture,arch) (void); -#define XCONCAT2(a,b) a##_##b -#define CONCAT2(a,b) XCONCAT2(a,b) -#define XCONCAT3(a,b,c) a##_##b##_##c -#define CONCAT3(a,b,c) XCONCAT3(a,b,c) +#if (!defined(__AVX512F__)) +DECL(avx512f) +#endif +#if (!defined(__FMA4__)) +DECL(fma4) +#endif +#if (!defined(__FMA__)) +DECL(fma) +#endif +#if (!defined(__AVX2__)) +DECL(avx2) +#endif +#if (!defined(__AVX__)) +DECL(avx) +#endif -#define nvec 1 -#include "sharp_core_inchelper.c" -#undef nvec +#endif -#define nvec 2 -#include "sharp_core_inchelper.c" -#undef nvec +static void assign_funcs(void) + { +#if defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=6) +#define DECL2(arch) \ + if (XCONCATX2(have,arch)()) \ + { \ + inner_loop_ = XCONCATX2(inner_loop,arch); \ + veclen_ = XCONCATX2(sharp_veclen,arch); \ + max_nvec_ = XCONCATX2(sharp_max_nvec,arch); \ + architecture_ = XCONCATX2(sharp_architecture,arch); \ + return; \ + } +#if (!defined(__AVX512F__)) +DECL2(avx512f) +#endif +#if (!defined(__FMA4__)) +DECL2(fma4) +#endif +#if (!defined(__FMA__)) +DECL2(fma) +#endif +#if (!defined(__AVX2__)) +DECL2(avx2) +#endif +#if (!defined(__AVX__)) +DECL2(avx) +#endif +#endif + inner_loop_ = inner_loop_default; + veclen_ = sharp_veclen_default; + max_nvec_ = sharp_max_nvec_default; + architecture_ = sharp_architecture_default; + } -#define nvec 3 -#include "sharp_core_inchelper.c" -#undef nvec - -#define nvec 4 -#include "sharp_core_inchelper.c" -#undef nvec - -#define nvec 5 -#include "sharp_core_inchelper.c" -#undef nvec - -#define nvec 6 -#include "sharp_core_inchelper.c" -#undef nvec void inner_loop (sharp_job *job, const int *ispair,const double *cth, const double *sth, int llim, int ulim, sharp_Ylmgen_C *gen, int mi, const int *mlim) { - int njobs=job->ntrans, nv=job->flags&SHARP_NVMAX; - if (njobs<=MAXJOB_SPECIAL) - { - switch (njobs*16+nv) - { -#if ((MAXJOB_SPECIAL>=1)&&(SHARP_MAXTRANS>=1)) - case 0x11: - CONCAT3(inner_loop,1,1) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x12: - CONCAT3(inner_loop,2,1) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x13: - CONCAT3(inner_loop,3,1) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x14: - CONCAT3(inner_loop,4,1) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x15: - CONCAT3(inner_loop,5,1) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x16: - CONCAT3(inner_loop,6,1) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; -#endif -#if ((MAXJOB_SPECIAL>=2)&&(SHARP_MAXTRANS>=2)) - case 0x21: - CONCAT3(inner_loop,1,2) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x22: - CONCAT3(inner_loop,2,2) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x23: - CONCAT3(inner_loop,3,2) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x24: - CONCAT3(inner_loop,4,2) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x25: - CONCAT3(inner_loop,5,2) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x26: - CONCAT3(inner_loop,6,2) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; -#endif -#if ((MAXJOB_SPECIAL>=3)&&(SHARP_MAXTRANS>=3)) - case 0x31: - CONCAT3(inner_loop,1,3) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x32: - CONCAT3(inner_loop,2,3) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x33: - CONCAT3(inner_loop,3,3) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x34: - CONCAT3(inner_loop,4,3) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x35: - CONCAT3(inner_loop,5,3) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x36: - CONCAT3(inner_loop,6,3) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; -#endif -#if ((MAXJOB_SPECIAL>=4)&&(SHARP_MAXTRANS>=4)) - case 0x41: - CONCAT3(inner_loop,1,4) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x42: - CONCAT3(inner_loop,2,4) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x43: - CONCAT3(inner_loop,3,4) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x44: - CONCAT3(inner_loop,4,4) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x45: - CONCAT3(inner_loop,5,4) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x46: - CONCAT3(inner_loop,6,4) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; -#endif -#if ((MAXJOB_SPECIAL>=5)&&(SHARP_MAXTRANS>=5)) - case 0x51: - CONCAT3(inner_loop,1,5) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x52: - CONCAT3(inner_loop,2,5) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x53: - CONCAT3(inner_loop,3,5) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x54: - CONCAT3(inner_loop,4,5) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x55: - CONCAT3(inner_loop,5,5) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x56: - CONCAT3(inner_loop,6,5) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; -#endif -#if ((MAXJOB_SPECIAL>=6)&&(SHARP_MAXTRANS>=6)) - case 0x61: - CONCAT3(inner_loop,1,6) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x62: - CONCAT3(inner_loop,2,6) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x63: - CONCAT3(inner_loop,3,6) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x64: - CONCAT3(inner_loop,4,6) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x65: - CONCAT3(inner_loop,5,6) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; - case 0x66: - CONCAT3(inner_loop,6,6) (job, ispair,cth,sth,llim,ulim,gen,mi,mlim); - return; -#endif - } - } -#if (SHARP_MAXTRANS>MAXJOB_SPECIAL) - else - { - switch (nv) - { - case 1: - CONCAT2(inner_loop,1) - (job, ispair,cth,sth,llim,ulim,gen,mi,mlim,job->ntrans); - return; - case 2: - CONCAT2(inner_loop,2) - (job, ispair,cth,sth,llim,ulim,gen,mi,mlim,job->ntrans); - return; - case 3: - CONCAT2(inner_loop,3) - (job, ispair,cth,sth,llim,ulim,gen,mi,mlim,job->ntrans); - return; - case 4: - CONCAT2(inner_loop,4) - (job, ispair,cth,sth,llim,ulim,gen,mi,mlim,job->ntrans); - return; - case 5: - CONCAT2(inner_loop,5) - (job, ispair,cth,sth,llim,ulim,gen,mi,mlim,job->ntrans); - return; - case 6: - CONCAT2(inner_loop,6) - (job, ispair,cth,sth,llim,ulim,gen,mi,mlim,job->ntrans); - return; - } - } -#endif - UTIL_FAIL("Incorrect vector parameters"); + if (!inner_loop_) assign_funcs(); + inner_loop_(job, ispair, cth, sth, llim, ulim, gen, mi, mlim); + } +int sharp_veclen(void) + { + if (!veclen_) assign_funcs(); + return veclen_(); + } +int sharp_max_nvec(int spin) + { + if (!max_nvec_) assign_funcs(); + return max_nvec_(spin); + } +const char *sharp_architecture(void) + { + if (!architecture_) assign_funcs(); + return architecture_(); } diff --git a/libsharp/sharp_core.h b/libsharp/sharp_core.h deleted file mode 100644 index 1e86488..0000000 --- a/libsharp/sharp_core.h +++ /dev/null @@ -1,50 +0,0 @@ -/* - * This file is part of libsharp. - * - * libsharp is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libsharp is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libsharp; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libsharp is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/*! \file sharp_core.h - * Interface for the computational core - * - * Copyright (C) 2012-2013 Max-Planck-Society - * \author Martin Reinecke - */ - -#ifndef PLANCK_SHARP_CORE_H -#define PLANCK_SHARP_CORE_H - -#include "sharp_internal.h" -#include "sharp_ylmgen_c.h" - -#ifdef __cplusplus -extern "C" { -#endif - -void inner_loop (sharp_job *job, const int *ispair,const double *cth, - const double *sth, int llim, int ulim, sharp_Ylmgen_C *gen, int mi, - const int *mlim); - -#ifdef __cplusplus -} -#endif - -#endif diff --git a/libsharp/sharp_core_avx.c b/libsharp/sharp_core_avx.c new file mode 100644 index 0000000..724e629 --- /dev/null +++ b/libsharp/sharp_core_avx.c @@ -0,0 +1,11 @@ +#if (!defined(__AVX__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=6) + +#define XCONCATX(a,b) a##_##b +#define XCONCATX2(a,b) XCONCATX(a,b) +#define XARCH(a) XCONCATX2(a,ARCH) + +#define ARCH avx +#pragma GCC target("avx") +#include "sharp_core_inc.c" + +#endif diff --git a/libsharp/sharp_core_avx2.c b/libsharp/sharp_core_avx2.c new file mode 100644 index 0000000..a7ab0a7 --- /dev/null +++ b/libsharp/sharp_core_avx2.c @@ -0,0 +1,11 @@ +#if (!defined(__AVX2__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=6) + +#define XCONCATX(a,b) a##_##b +#define XCONCATX2(a,b) XCONCATX(a,b) +#define XARCH(a) XCONCATX2(a,ARCH) + +#define ARCH avx2 +#pragma GCC target("avx2") +#include "sharp_core_inc.c" + +#endif diff --git a/libsharp/sharp_core_avx512f.c b/libsharp/sharp_core_avx512f.c new file mode 100644 index 0000000..7f17429 --- /dev/null +++ b/libsharp/sharp_core_avx512f.c @@ -0,0 +1,11 @@ +#if (!defined(__AVX512F__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=6) + +#define XCONCATX(a,b) a##_##b +#define XCONCATX2(a,b) XCONCATX(a,b) +#define XARCH(a) XCONCATX2(a,ARCH) + +#define ARCH avx512f +#pragma GCC target("avx512f") +#include "sharp_core_inc.c" + +#endif diff --git a/libsharp/sharp_core_fma.c b/libsharp/sharp_core_fma.c new file mode 100644 index 0000000..793151f --- /dev/null +++ b/libsharp/sharp_core_fma.c @@ -0,0 +1,11 @@ +#if (!defined(__FMA__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=6) + +#define XCONCATX(a,b) a##_##b +#define XCONCATX2(a,b) XCONCATX(a,b) +#define XARCH(a) XCONCATX2(a,ARCH) + +#define ARCH fma +#pragma GCC target("fma") +#include "sharp_core_inc.c" + +#endif diff --git a/libsharp/sharp_core_fma4.c b/libsharp/sharp_core_fma4.c new file mode 100644 index 0000000..d71de74 --- /dev/null +++ b/libsharp/sharp_core_fma4.c @@ -0,0 +1,11 @@ +#if (!defined(__FMA4__)) && defined(__GNUC__) && defined (__x86_64__) && (__GNUC__>=6) + +#define XCONCATX(a,b) a##_##b +#define XCONCATX2(a,b) XCONCATX(a,b) +#define XARCH(a) XCONCATX2(a,ARCH) + +#define ARCH fma4 +#pragma GCC target("fma4") +#include "sharp_core_inc.c" + +#endif diff --git a/libsharp/sharp_core_inc.c b/libsharp/sharp_core_inc.c index 747658c..d229a49 100644 --- a/libsharp/sharp_core_inc.c +++ b/libsharp/sharp_core_inc.c @@ -23,271 +23,1159 @@ */ /*! \file sharp_core_inc.c - * Type-dependent code for the computational core + * Computational core * - * Copyright (C) 2012 Max-Planck-Society + * Copyright (C) 2012-2019 Max-Planck-Society * \author Martin Reinecke */ -typedef struct - { Tv v[nvec]; } Tb; +#include +#include +#include +#include "sharp_vecsupport.h" +#include "sharp.h" +#include "sharp_internal.h" +#include "c_utils.h" -typedef union - { Tb b; double s[VLEN*nvec]; } Y(Tbu); +typedef complex double dcmplx; + +#define nv0 (128/VLEN) +#define nvx (64/VLEN) + +typedef Tv Tbv0[nv0]; +typedef double Tbs0[nv0*VLEN]; typedef struct - { Tb r, i; } Y(Tbri); - -typedef struct - { Tb qr, qi, ur, ui; } Y(Tbqu); - -typedef struct - { double r[VLEN*nvec], i[VLEN*nvec]; } Y(Tsri); - -typedef struct - { double qr[VLEN*nvec],qi[VLEN*nvec],ur[VLEN*nvec],ui[VLEN*nvec]; } Y(Tsqu); - -typedef union - { Y(Tbri) b; Y(Tsri)s; } Y(Tburi); - -typedef union - { Y(Tbqu) b; Y(Tsqu)s; } Y(Tbuqu); - -static inline Tb Y(Tbconst)(double val) { - Tv v=vload(val); - Tb res; - for (int i=0; iv[i],v); } +typedef struct + { + Tbs0 sth, corfac, scale, lam1, lam2, csq, p1r, p1i, p2r, p2i; + } s0data_s; -static inline Tb Y(Tbprod)(Tb a, Tb b) - { Tb r; for (int i=0; iv[i],b.v[i]); } +typedef Tv Tbvx[nvx]; +typedef double Tbsx[nvx*VLEN]; -static void Y(Tbnormalize) (Tb * restrict val, Tb * restrict scale, +typedef struct + { + Tbvx sth, cfp, cfm, scp, scm, l1p, l2p, l1m, l2m, cth, + p1pr, p1pi, p2pr, p2pi, p1mr, p1mi, p2mr, p2mi; + } sxdata_v; + +typedef struct + { + Tbsx sth, cfp, cfm, scp, scm, l1p, l2p, l1m, l2m, cth, + p1pr, p1pi, p2pr, p2pi, p1mr, p1mi, p2mr, p2mi; + } sxdata_s; + +typedef union + { + sxdata_v v; + sxdata_s s; + } sxdata_u; + +static inline void Tvnormalize (Tv * restrict val, Tv * restrict scale, double maxval) { - const Tv vfsmall=vload(sharp_fsmall), vfbig=vload(sharp_fbig); const Tv vfmin=vload(sharp_fsmall*maxval), vfmax=vload(maxval); - for (int i=0;iv[i]),vfmax); - while (vanyTrue(mask)) - { - vmuleq_mask(mask,val->v[i],vfsmall); - vaddeq_mask(mask,scale->v[i],vone); - mask = vgt(vabs(val->v[i]),vfmax); - } - mask = vand_mask(vlt(vabs(val->v[i]),vfmin),vne(val->v[i],vzero)); - while (vanyTrue(mask)) - { - vmuleq_mask(mask,val->v[i],vfbig); - vsubeq_mask(mask,scale->v[i],vone); - mask = vand_mask(vlt(vabs(val->v[i]),vfmin),vne(val->v[i],vzero)); - } + vmuleq_mask(mask,*val,vfsmall); + vaddeq_mask(mask,*scale,vone); + mask = vgt(vabs(*val),vfmax); + } + mask = vand_mask(vlt(vabs(*val),vfmin),vne(*val,vzero)); + while (vanyTrue(mask)) + { + vmuleq_mask(mask,*val,vfbig); + vsubeq_mask(mask,*scale,vone); + mask = vand_mask(vlt(vabs(*val),vfmin),vne(*val,vzero)); } } -static void Y(mypow) (Tb val, int npow, Tb * restrict resd, - Tb * restrict ress) +static void mypow(Tv val, int npow, const double * restrict powlimit, + Tv * restrict resd, Tv * restrict ress) { - Tb scale=Y(Tbconst)(0.), scaleint=Y(Tbconst)(0.), res=Y(Tbconst)(1.); - - Y(Tbnormalize)(&val,&scaleint,sharp_fbighalf); - - do + Tv vminv=vload(powlimit[npow]); + Tm mask = vlt(vabs(val),vminv); + if (!vanyTrue(mask)) // no underflows possible, use quick algoritm { - if (npow&1) + Tv res=vone; + do { - for (int i=0; i>=1); + *resd=res; + *ress=vzero; + } + else + { + Tv scale=vzero, scaleint=vzero, res=vone; + Tvnormalize(&val,&scaleint,sharp_fbighalf); + do + { + if (npow&1) { - vmuleq(res.v[i],val.v[i]); - vaddeq(scale.v[i],scaleint.v[i]); + res*=val; + scale+=scaleint; + Tvnormalize(&res,&scale,sharp_fbighalf); } - Y(Tbnormalize)(&res,&scale,sharp_fbighalf); + val*=val; + scaleint+=scaleint; + Tvnormalize(&val,&scaleint,sharp_fbighalf); } - for (int i=0; i>=1); + *resd=res; + *ress=scale; } - while(npow>>=1); - - *resd=res; - *ress=scale; } -static inline int Y(rescale) (Tb * restrict lam1, Tb * restrict lam2, - Tb * restrict scale) - { - int did_scale=0; - for (int i=0;iv[i]),vload(sharp_ftol)); - if (vanyTrue(mask)) - { - did_scale=1; - vmuleq_mask(mask,lam1->v[i],vload(sharp_fsmall)); - vmuleq_mask(mask,lam2->v[i],vload(sharp_fsmall)); - vaddeq_mask(mask,scale->v[i],vone); - } - } - return did_scale; - } - -static inline int Y(TballLt)(Tb a,double b) - { - Tv vb=vload(b); - Tm res=vlt(a.v[0],vb); - for (int i=1; im; - Tb lam_1=Y(Tbconst)(0.), lam_2, scale; - Y(mypow) (sth,l,&lam_2,&scale); - Y(Tbmuleq1) (&lam_2,(gen->m&1) ? -gen->mfac[gen->m]:gen->mfac[gen->m]); - Y(Tbnormalize)(&lam_2,&scale,sharp_ftol); + Tm mask = vgt(vabs(*v2),eps); + if (vanyTrue(mask)) + { + vmuleq_mask(mask,*v1,vload(sharp_fsmall)); + vmuleq_mask(mask,*v2,vload(sharp_fsmall)); + vaddeq_mask(mask,*s,vone); + return 1; + } + return 0; + } + +NOINLINE static void iter_to_ieee(const sharp_Ylmgen_C * restrict gen, + s0data_v * restrict d, int * restrict l_, int * restrict il_, int nv2) + { + int l=gen->m, il=0; + Tv mfac = vload((gen->m&1) ? -gen->mfac[gen->m]:gen->mfac[gen->m]); + Tv limscale=vload(sharp_limscale); + int below_limit = 1; + for (int i=0; ilam1[i]=vzero; + mypow(d->sth[i],gen->m,gen->powlimit,&d->lam2[i],&d->scale[i]); + d->lam2[i] *= mfac; + Tvnormalize(&d->lam2[i],&d->scale[i],sharp_ftol); + below_limit &= vallTrue(vlt(d->scale[i],limscale)); + } - int below_limit = Y(TballLt)(scale,sharp_limscale); while (below_limit) { - if (l+2>gen->lmax) {*l_=gen->lmax+1;return;} - Tv r0=vload(gen->rf[l].f[0]),r1=vload(gen->rf[l].f[1]); - for (int i=0; irf[l+1].f[0]); r1=vload(gen->rf[l+1].f[1]); - for (int i=0; igen->lmax) {*l_=gen->lmax+1;return;} + below_limit=1; + Tv a1=vload(gen->coef[il ].a), b1=vload(gen->coef[il ].b); + Tv a2=vload(gen->coef[il+1].a), b2=vload(gen->coef[il+1].b); + for (int i=0; ilam1[i] = (a1*d->csq[i] + b1)*d->lam2[i] + d->lam1[i]; + d->lam2[i] = (a2*d->csq[i] + b2)*d->lam1[i] + d->lam2[i]; + if (rescale(&d->lam1[i], &d->lam2[i], &d->scale[i], vload(sharp_ftol))) + below_limit &= vallTrue(vlt(d->scale[i],vload(sharp_limscale))); + } + l+=4; il+=2; } - *l_=l; *lam_1_=lam_1; *lam_2_=lam_2; *scale_=scale; + *l_=l; *il_=il; } -static inline void Y(rec_step) (Tb * restrict rxp, Tb * restrict rxm, - Tb * restrict ryp, Tb * restrict rym, const Tb cth, - const sharp_ylmgen_dbl3 fx) +NOINLINE static void alm2map_kernel(s0data_v * restrict d, + const sharp_ylmgen_dbl2 * restrict coef, const dcmplx * restrict alm, + int l, int il, int lmax, int nv2) { - Tv fx0=vload(fx.f[0]),fx1=vload(fx.f[1]),fx2=vload(fx.f[2]); - for (int i=0; iv[i] = vsub(vmul(vsub(cth.v[i],fx1),vmul(fx0,ryp->v[i])), - vmul(fx2,rxp->v[i])); - rxm->v[i] = vsub(vmul(vadd(cth.v[i],fx1),vmul(fx0,rym->v[i])), - vmul(fx2,rxm->v[i])); + for (; l<=lmax-2; il+=2, l+=4) + { + Tv ar1=vload(creal(alm[l ])), ai1=vload(cimag(alm[l ])); + Tv ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1])); + Tv ar3=vload(creal(alm[l+2])), ai3=vload(cimag(alm[l+2])); + Tv ar4=vload(creal(alm[l+3])), ai4=vload(cimag(alm[l+3])); + Tv a1=vload(coef[il ].a), b1=vload(coef[il ].b); + Tv a2=vload(coef[il+1].a), b2=vload(coef[il+1].b); + for (int i=0; ip1r[i] += d->lam2[i]*ar1; + d->p1i[i] += d->lam2[i]*ai1; + d->p2r[i] += d->lam2[i]*ar2; + d->p2i[i] += d->lam2[i]*ai2; + d->lam1[i] = (a1*d->csq[i] + b1)*d->lam2[i] + d->lam1[i]; + d->p1r[i] += d->lam1[i]*ar3; + d->p1i[i] += d->lam1[i]*ai3; + d->p2r[i] += d->lam1[i]*ar4; + d->p2i[i] += d->lam1[i]*ai4; + d->lam2[i] = (a2*d->csq[i] + b2)*d->lam1[i] + d->lam2[i]; + } + } + } + else + { + for (; l<=lmax-2; il+=2, l+=4) + { + Tv ar1=vload(creal(alm[l ])), ai1=vload(cimag(alm[l ])); + Tv ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1])); + Tv ar3=vload(creal(alm[l+2])), ai3=vload(cimag(alm[l+2])); + Tv ar4=vload(creal(alm[l+3])), ai4=vload(cimag(alm[l+3])); + Tv a1=vload(coef[il ].a), b1=vload(coef[il ].b); + Tv a2=vload(coef[il+1].a), b2=vload(coef[il+1].b); + for (int i=0; ip1r[i] += d->lam2[i]*ar1; + d->p1i[i] += d->lam2[i]*ai1; + d->p2r[i] += d->lam2[i]*ar2; + d->p2i[i] += d->lam2[i]*ai2; + d->lam1[i] = (a1*d->csq[i] + b1)*d->lam2[i] + d->lam1[i]; + d->p1r[i] += d->lam1[i]*ar3; + d->p1i[i] += d->lam1[i]*ai3; + d->p2r[i] += d->lam1[i]*ar4; + d->p2i[i] += d->lam1[i]*ai4; + d->lam2[i] = (a2*d->csq[i] + b2)*d->lam1[i] + d->lam2[i]; + } + } + } + for (; l<=lmax; ++il, l+=2) + { + Tv ar1=vload(creal(alm[l ])), ai1=vload(cimag(alm[l ])); + Tv ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1])); + Tv a=vload(coef[il].a), b=vload(coef[il].b); + for (int i=0; ip1r[i] += d->lam2[i]*ar1; + d->p1i[i] += d->lam2[i]*ai1; + d->p2r[i] += d->lam2[i]*ar2; + d->p2i[i] += d->lam2[i]*ai2; + Tv tmp = (a*d->csq[i] + b)*d->lam2[i] + d->lam1[i]; + d->lam1[i] = d->lam2[i]; + d->lam2[i] = tmp; + } } } -static void Y(iter_to_ieee_spin) (const Tb cth, const Tb sth, int *l_, - Tb * rec1p_, Tb * rec1m_, Tb * rec2p_, Tb * rec2m_, - Tb * scalep_, Tb * scalem_, const sharp_Ylmgen_C * restrict gen) +NOINLINE static void calc_alm2map (sharp_job * restrict job, + const sharp_Ylmgen_C * restrict gen, s0data_v * restrict d, int nth) { - const sharp_ylmgen_dbl3 * restrict fx = gen->fx; - Tb cth2, sth2; - for (int i=0; ilmax; + int nv2 = (nth+VLEN-1)/VLEN; + iter_to_ieee(gen, d, &l, &il, nv2); + job->opcnt += il * 4*nth; + if (l>lmax) return; + job->opcnt += (lmax+1-l) * 6*nth; + + const sharp_ylmgen_dbl2 * restrict coef = gen->coef; + const dcmplx * restrict alm=job->almtmp; + int full_ieee=1; + for (int i=0; iscale[i], &d->corfac[i], gen->cf); + full_ieee &= vallTrue(vge(d->scale[i],vload(sharp_minscale))); } - Tb ccp, ccps, ssp, ssps, csp, csps, scp, scps; - Y(mypow)(cth2,gen->cosPow,&ccp,&ccps); Y(mypow)(sth2,gen->sinPow,&ssp,&ssps); - Y(mypow)(cth2,gen->sinPow,&csp,&csps); Y(mypow)(sth2,gen->cosPow,&scp,&scps); + while((!full_ieee) && (l<=lmax)) + { + Tv ar1=vload(creal(alm[l ])), ai1=vload(cimag(alm[l ])); + Tv ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1])); + Tv a=vload(coef[il].a), b=vload(coef[il].b); + full_ieee=1; + for (int i=0; ip1r[i] += d->lam2[i]*d->corfac[i]*ar1; + d->p1i[i] += d->lam2[i]*d->corfac[i]*ai1; + d->p2r[i] += d->lam2[i]*d->corfac[i]*ar2; + d->p2i[i] += d->lam2[i]*d->corfac[i]*ai2; + Tv tmp = (a*d->csq[i] + b)*d->lam2[i] + d->lam1[i]; + d->lam1[i] = d->lam2[i]; + d->lam2[i] = tmp; + if (rescale(&d->lam1[i], &d->lam2[i], &d->scale[i], vload(sharp_ftol))) + getCorfac(d->scale[i], &d->corfac[i], gen->cf); + full_ieee &= vallTrue(vge(d->scale[i],vload(sharp_minscale))); + } + l+=2; ++il; + } + if (l>lmax) return; - Tb rec2p, rec2m, scalep, scalem; - Tb rec1p=Y(Tbconst)(0.), rec1m=Y(Tbconst)(0.); + for (int i=0; ilam1[i] *= d->corfac[i]; + d->lam2[i] *= d->corfac[i]; + } + alm2map_kernel(d, coef, alm, l, il, lmax, nv2); + } + +NOINLINE static void map2alm_kernel(s0data_v * restrict d, + const sharp_ylmgen_dbl2 * restrict coef, dcmplx * restrict alm, int l, + int il, int lmax, int nv2) + { + for (; l<=lmax-2; il+=2, l+=4) + { + Tv a1=vload(coef[il ].a), b1=vload(coef[il ].b); + Tv a2=vload(coef[il+1].a), b2=vload(coef[il+1].b); + Tv atmp1[4] = {vzero, vzero, vzero, vzero}; + Tv atmp2[4] = {vzero, vzero, vzero, vzero}; + for (int i=0; ilam2[i]*d->p1r[i]; + atmp1[1] += d->lam2[i]*d->p1i[i]; + atmp1[2] += d->lam2[i]*d->p2r[i]; + atmp1[3] += d->lam2[i]*d->p2i[i]; + d->lam1[i] = (a1*d->csq[i] + b1)*d->lam2[i] + d->lam1[i]; + atmp2[0] += d->lam1[i]*d->p1r[i]; + atmp2[1] += d->lam1[i]*d->p1i[i]; + atmp2[2] += d->lam1[i]*d->p2r[i]; + atmp2[3] += d->lam1[i]*d->p2i[i]; + d->lam2[i] = (a2*d->csq[i] + b2)*d->lam1[i] + d->lam2[i]; + } + vhsum_cmplx_special (atmp1[0], atmp1[1], atmp1[2], atmp1[3], &alm[l ]); + vhsum_cmplx_special (atmp2[0], atmp2[1], atmp2[2], atmp2[3], &alm[l+2]); + } + for (; l<=lmax; ++il, l+=2) + { + Tv a=vload(coef[il].a), b=vload(coef[il].b); + Tv atmp[4] = {vzero, vzero, vzero, vzero}; + for (int i=0; ilam2[i]*d->p1r[i]; + atmp[1] += d->lam2[i]*d->p1i[i]; + atmp[2] += d->lam2[i]*d->p2r[i]; + atmp[3] += d->lam2[i]*d->p2i[i]; + Tv tmp = (a*d->csq[i] + b)*d->lam2[i] + d->lam1[i]; + d->lam1[i] = d->lam2[i]; + d->lam2[i] = tmp; + } + vhsum_cmplx_special (atmp[0], atmp[1], atmp[2], atmp[3], &alm[l]); + } + } + +NOINLINE static void calc_map2alm (sharp_job * restrict job, + const sharp_Ylmgen_C * restrict gen, s0data_v * restrict d, int nth) + { + int l,il,lmax=gen->lmax; + int nv2 = (nth+VLEN-1)/VLEN; + iter_to_ieee(gen, d, &l, &il, nv2); + job->opcnt += il * 4*nth; + if (l>lmax) return; + job->opcnt += (lmax+1-l) * 6*nth; + + const sharp_ylmgen_dbl2 * restrict coef = gen->coef; + dcmplx * restrict alm=job->almtmp; + int full_ieee=1; + for (int i=0; iscale[i], &d->corfac[i], gen->cf); + full_ieee &= vallTrue(vge(d->scale[i],vload(sharp_minscale))); + } + + while((!full_ieee) && (l<=lmax)) + { + Tv a=vload(coef[il].a), b=vload(coef[il].b); + Tv atmp[4] = {vzero, vzero, vzero, vzero}; + full_ieee=1; + for (int i=0; ilam2[i]*d->corfac[i]*d->p1r[i]; + atmp[1] += d->lam2[i]*d->corfac[i]*d->p1i[i]; + atmp[2] += d->lam2[i]*d->corfac[i]*d->p2r[i]; + atmp[3] += d->lam2[i]*d->corfac[i]*d->p2i[i]; + Tv tmp = (a*d->csq[i] + b)*d->lam2[i] + d->lam1[i]; + d->lam1[i] = d->lam2[i]; + d->lam2[i] = tmp; + if (rescale(&d->lam1[i], &d->lam2[i], &d->scale[i], vload(sharp_ftol))) + getCorfac(d->scale[i], &d->corfac[i], gen->cf); + full_ieee &= vallTrue(vge(d->scale[i],vload(sharp_minscale))); + } + vhsum_cmplx_special (atmp[0], atmp[1], atmp[2], atmp[3], &alm[l]); + l+=2; ++il; + } + if (l>lmax) return; + + for (int i=0; ilam1[i] *= d->corfac[i]; + d->lam2[i] *= d->corfac[i]; + } + map2alm_kernel(d, coef, alm, l, il, lmax, nv2); + } + +NOINLINE static void iter_to_ieee_spin (const sharp_Ylmgen_C * restrict gen, + sxdata_v * restrict d, int * restrict l_, int nv2) + { + const sharp_ylmgen_dbl2 * restrict fx = gen->coef; Tv prefac=vload(gen->prefac[gen->m]), prescale=vload(gen->fscale[gen->m]); - for (int i=0; icth[i])*vload(0.5))); + Tv sth2=vmax(vload(1e-15),vsqrt((vone-d->cth[i])*vload(0.5))); + Tm mask=vlt(d->sth[i],vzero); + vmuleq_mask(vand_mask(mask,vlt(d->cth[i],vzero)),cth2,vload(-1.)); + vmuleq_mask(vand_mask(mask,vgt(d->cth[i],vzero)),sth2,vload(-1.)); + + Tv ccp, ccps, ssp, ssps, csp, csps, scp, scps; + mypow(cth2,gen->cosPow,gen->powlimit,&ccp,&ccps); + mypow(sth2,gen->sinPow,gen->powlimit,&ssp,&ssps); + mypow(cth2,gen->sinPow,gen->powlimit,&csp,&csps); + mypow(sth2,gen->cosPow,gen->powlimit,&scp,&scps); + + d->l1p[i] = vzero; + d->l1m[i] = vzero; + d->l2p[i] = prefac*ccp; + d->scp[i] = prescale+ccps; + d->l2m[i] = prefac*csp; + d->scm[i] = prescale+csps; + Tvnormalize(&d->l2m[i],&d->scm[i],sharp_fbighalf); + Tvnormalize(&d->l2p[i],&d->scp[i],sharp_fbighalf); + d->l2p[i] *= ssp; + d->scp[i] += ssps; + d->l2m[i] *= scp; + d->scm[i] += scps; if (gen->preMinus_p) - rec2p.v[i]=vneg(rec2p.v[i]); + d->l2p[i] = vneg(d->l2p[i]); if (gen->preMinus_m) - rec2m.v[i]=vneg(rec2m.v[i]); + d->l2m[i] = vneg(d->l2m[i]); if (gen->s&1) - rec2p.v[i]=vneg(rec2p.v[i]); + d->l2p[i] = vneg(d->l2p[i]); + + Tvnormalize(&d->l2m[i],&d->scm[i],sharp_ftol); + Tvnormalize(&d->l2p[i],&d->scp[i],sharp_ftol); + + below_limit &= vallTrue(vlt(d->scm[i],limscale)) && + vallTrue(vlt(d->scp[i],limscale)); } - Y(Tbnormalize)(&rec2m,&scalem,sharp_ftol); - Y(Tbnormalize)(&rec2p,&scalep,sharp_ftol); int l=gen->mhi; - int below_limit = Y(TballLt)(scalep,sharp_limscale) - && Y(TballLt)(scalem,sharp_limscale); while (below_limit) { if (l+2>gen->lmax) {*l_=gen->lmax+1;return;} - Y(rec_step)(&rec1p,&rec1m,&rec2p,&rec2m,cth,fx[l+1]); - Y(rec_step)(&rec2p,&rec2m,&rec1p,&rec1m,cth,fx[l+2]); - if (Y(rescale)(&rec1p,&rec2p,&scalep) | Y(rescale)(&rec1m,&rec2m,&scalem)) - below_limit = Y(TballLt)(scalep,sharp_limscale) - && Y(TballLt)(scalem,sharp_limscale); + below_limit=1; + Tv fx10=vload(fx[l+1].a),fx11=vload(fx[l+1].b); + Tv fx20=vload(fx[l+2].a),fx21=vload(fx[l+2].b); + for (int i=0; il1p[i] = (d->cth[i]*fx10 - fx11)*d->l2p[i] - d->l1p[i]; + d->l1m[i] = (d->cth[i]*fx10 + fx11)*d->l2m[i] - d->l1m[i]; + d->l2p[i] = (d->cth[i]*fx20 - fx21)*d->l1p[i] - d->l2p[i]; + d->l2m[i] = (d->cth[i]*fx20 + fx21)*d->l1m[i] - d->l2m[i]; + if (rescale(&d->l1p[i],&d->l2p[i],&d->scp[i],vload(sharp_ftol)) || + rescale(&d->l1m[i],&d->l2m[i],&d->scm[i],vload(sharp_ftol))) + below_limit &= vallTrue(vlt(d->scp[i],limscale)) && + vallTrue(vlt(d->scm[i],limscale)); + } l+=2; } *l_=l; - *rec1p_=rec1p; *rec2p_=rec2p; *scalep_=scalep; - *rec1m_=rec1m; *rec2m_=rec2m; *scalem_=scalem; + } + +NOINLINE static void alm2map_spin_kernel(sxdata_v * restrict d, + const sharp_ylmgen_dbl2 * restrict fx, const dcmplx * restrict alm, + int l, int lmax, int nv2) + { + int lsave = l; + while (l<=lmax) + { + Tv fx10=vload(fx[l+1].a),fx11=vload(fx[l+1].b); + Tv fx20=vload(fx[l+2].a),fx21=vload(fx[l+2].b); + Tv agr1=vload(creal(alm[2*l ])), agi1=vload(cimag(alm[2*l ])), + acr1=vload(creal(alm[2*l+1])), aci1=vload(cimag(alm[2*l+1])); + Tv agr2=vload(creal(alm[2*l+2])), agi2=vload(cimag(alm[2*l+2])), + acr2=vload(creal(alm[2*l+3])), aci2=vload(cimag(alm[2*l+3])); + for (int i=0; il1p[i] = (d->cth[i]*fx10 - fx11)*d->l2p[i] - d->l1p[i]; + d->p1pr[i] += agr1*d->l2p[i]; + d->p1pi[i] += agi1*d->l2p[i]; + d->p1mr[i] += acr1*d->l2p[i]; + d->p1mi[i] += aci1*d->l2p[i]; + + d->p1pr[i] += aci2*d->l1p[i]; + d->p1pi[i] -= acr2*d->l1p[i]; + d->p1mr[i] -= agi2*d->l1p[i]; + d->p1mi[i] += agr2*d->l1p[i]; + d->l2p[i] = (d->cth[i]*fx20 - fx21)*d->l1p[i] - d->l2p[i]; + } + l+=2; + } + l=lsave; + while (l<=lmax) + { + Tv fx10=vload(fx[l+1].a),fx11=vload(fx[l+1].b); + Tv fx20=vload(fx[l+2].a),fx21=vload(fx[l+2].b); + Tv agr1=vload(creal(alm[2*l ])), agi1=vload(cimag(alm[2*l ])), + acr1=vload(creal(alm[2*l+1])), aci1=vload(cimag(alm[2*l+1])); + Tv agr2=vload(creal(alm[2*l+2])), agi2=vload(cimag(alm[2*l+2])), + acr2=vload(creal(alm[2*l+3])), aci2=vload(cimag(alm[2*l+3])); + for (int i=0; il1m[i] = (d->cth[i]*fx10 + fx11)*d->l2m[i] - d->l1m[i]; + d->p2pr[i] -= aci1*d->l2m[i]; + d->p2pi[i] += acr1*d->l2m[i]; + d->p2mr[i] += agi1*d->l2m[i]; + d->p2mi[i] -= agr1*d->l2m[i]; + + d->p2pr[i] += agr2*d->l1m[i]; + d->p2pi[i] += agi2*d->l1m[i]; + d->p2mr[i] += acr2*d->l1m[i]; + d->p2mi[i] += aci2*d->l1m[i]; + d->l2m[i] = (d->cth[i]*fx20 + fx21)*d->l1m[i] - d->l2m[i]; + } + l+=2; + } + } + +NOINLINE static void calc_alm2map_spin (sharp_job * restrict job, + const sharp_Ylmgen_C * restrict gen, sxdata_v * restrict d, int nth) + { + int l,lmax=gen->lmax; + int nv2 = (nth+VLEN-1)/VLEN; + iter_to_ieee_spin(gen, d, &l, nv2); + job->opcnt += (l-gen->mhi) * 7*nth; + if (l>lmax) return; + job->opcnt += (lmax+1-l) * 23*nth; + + const sharp_ylmgen_dbl2 * restrict fx = gen->coef; + const dcmplx * restrict alm=job->almtmp; + int full_ieee=1; + for (int i=0; iscp[i], &d->cfp[i], gen->cf); + getCorfac(d->scm[i], &d->cfm[i], gen->cf); + full_ieee &= vallTrue(vge(d->scp[i],vload(sharp_minscale))) && + vallTrue(vge(d->scm[i],vload(sharp_minscale))); + } + + while((!full_ieee) && (l<=lmax)) + { + Tv fx10=vload(fx[l+1].a),fx11=vload(fx[l+1].b); + Tv fx20=vload(fx[l+2].a),fx21=vload(fx[l+2].b); + Tv agr1=vload(creal(alm[2*l ])), agi1=vload(cimag(alm[2*l ])), + acr1=vload(creal(alm[2*l+1])), aci1=vload(cimag(alm[2*l+1])); + Tv agr2=vload(creal(alm[2*l+2])), agi2=vload(cimag(alm[2*l+2])), + acr2=vload(creal(alm[2*l+3])), aci2=vload(cimag(alm[2*l+3])); + full_ieee=1; + for (int i=0; il1p[i] = (d->cth[i]*fx10 - fx11)*d->l2p[i] - d->l1p[i]; + d->l1m[i] = (d->cth[i]*fx10 + fx11)*d->l2m[i] - d->l1m[i]; + + Tv l2p=d->l2p[i]*d->cfp[i], l2m=d->l2m[i]*d->cfm[i]; + Tv l1m=d->l1m[i]*d->cfm[i], l1p=d->l1p[i]*d->cfp[i]; + + d->p1pr[i] += agr1*l2p + aci2*l1p; + d->p1pi[i] += agi1*l2p - acr2*l1p; + d->p1mr[i] += acr1*l2p - agi2*l1p; + d->p1mi[i] += aci1*l2p + agr2*l1p; + + d->p2pr[i] += agr2*l1m - aci1*l2m; + d->p2pi[i] += agi2*l1m + acr1*l2m; + d->p2mr[i] += acr2*l1m + agi1*l2m; + d->p2mi[i] += aci2*l1m - agr1*l2m; + + d->l2p[i] = (d->cth[i]*fx20 - fx21)*d->l1p[i] - d->l2p[i]; + d->l2m[i] = (d->cth[i]*fx20 + fx21)*d->l1m[i] - d->l2m[i]; + if (rescale(&d->l1p[i], &d->l2p[i], &d->scp[i], vload(sharp_ftol))) + getCorfac(d->scp[i], &d->cfp[i], gen->cf); + full_ieee &= vallTrue(vge(d->scp[i],vload(sharp_minscale))); + if (rescale(&d->l1m[i], &d->l2m[i], &d->scm[i], vload(sharp_ftol))) + getCorfac(d->scm[i], &d->cfm[i], gen->cf); + full_ieee &= vallTrue(vge(d->scm[i],vload(sharp_minscale))); + } + l+=2; + } +// if (l>lmax) return; + + for (int i=0; il1p[i] *= d->cfp[i]; + d->l2p[i] *= d->cfp[i]; + d->l1m[i] *= d->cfm[i]; + d->l2m[i] *= d->cfm[i]; + } + alm2map_spin_kernel(d, fx, alm, l, lmax, nv2); + + for (int i=0; ip1pr[i]; d->p1pr[i] -= d->p2mi[i]; d->p2mi[i] += tmp; + tmp = d->p1pi[i]; d->p1pi[i] += d->p2mr[i]; d->p2mr[i] -= tmp; + tmp = d->p1mr[i]; d->p1mr[i] += d->p2pi[i]; d->p2pi[i] -= tmp; + tmp = d->p1mi[i]; d->p1mi[i] -= d->p2pr[i]; d->p2pr[i] += tmp; + } + } + +NOINLINE static void map2alm_spin_kernel(sxdata_v * restrict d, + const sharp_ylmgen_dbl2 * restrict fx, dcmplx * restrict alm, + int l, int lmax, int nv2) + { + int lsave=l; + while (l<=lmax) + { + Tv fx10=vload(fx[l+1].a),fx11=vload(fx[l+1].b); + Tv fx20=vload(fx[l+2].a),fx21=vload(fx[l+2].b); + Tv agr1=vzero, agi1=vzero, acr1=vzero, aci1=vzero; + Tv agr2=vzero, agi2=vzero, acr2=vzero, aci2=vzero; + for (int i=0; il1p[i] = (d->cth[i]*fx10 - fx11)*d->l2p[i] - d->l1p[i]; + agr1 += d->p2mi[i]*d->l2p[i]; + agi1 -= d->p2mr[i]*d->l2p[i]; + acr1 -= d->p2pi[i]*d->l2p[i]; + aci1 += d->p2pr[i]*d->l2p[i]; + agr2 += d->p2pr[i]*d->l1p[i]; + agi2 += d->p2pi[i]*d->l1p[i]; + acr2 += d->p2mr[i]*d->l1p[i]; + aci2 += d->p2mi[i]*d->l1p[i]; + d->l2p[i] = (d->cth[i]*fx20 - fx21)*d->l1p[i] - d->l2p[i]; + } + vhsum_cmplx_special (agr1,agi1,acr1,aci1,&alm[2*l]); + vhsum_cmplx_special (agr2,agi2,acr2,aci2,&alm[2*l+2]); + l+=2; + } + l=lsave; + while (l<=lmax) + { + Tv fx10=vload(fx[l+1].a),fx11=vload(fx[l+1].b); + Tv fx20=vload(fx[l+2].a),fx21=vload(fx[l+2].b); + Tv agr1=vzero, agi1=vzero, acr1=vzero, aci1=vzero; + Tv agr2=vzero, agi2=vzero, acr2=vzero, aci2=vzero; + for (int i=0; il1m[i] = (d->cth[i]*fx10 + fx11)*d->l2m[i] - d->l1m[i]; + agr1 += d->p1pr[i]*d->l2m[i]; + agi1 += d->p1pi[i]*d->l2m[i]; + acr1 += d->p1mr[i]*d->l2m[i]; + aci1 += d->p1mi[i]*d->l2m[i]; + agr2 -= d->p1mi[i]*d->l1m[i]; + agi2 += d->p1mr[i]*d->l1m[i]; + acr2 += d->p1pi[i]*d->l1m[i]; + aci2 -= d->p1pr[i]*d->l1m[i]; + d->l2m[i] = (d->cth[i]*fx20 + fx21)*d->l1m[i] - d->l2m[i]; + } + vhsum_cmplx_special (agr1,agi1,acr1,aci1,&alm[2*l]); + vhsum_cmplx_special (agr2,agi2,acr2,aci2,&alm[2*l+2]); + l+=2; + } + } + +NOINLINE static void calc_map2alm_spin (sharp_job * restrict job, + const sharp_Ylmgen_C * restrict gen, sxdata_v * restrict d, int nth) + { + int l,lmax=gen->lmax; + int nv2 = (nth+VLEN-1)/VLEN; + iter_to_ieee_spin(gen, d, &l, nv2); + job->opcnt += (l-gen->mhi) * 7*nth; + if (l>lmax) return; + job->opcnt += (lmax+1-l) * 23*nth; + + const sharp_ylmgen_dbl2 * restrict fx = gen->coef; + dcmplx * restrict alm=job->almtmp; + int full_ieee=1; + for (int i=0; iscp[i], &d->cfp[i], gen->cf); + getCorfac(d->scm[i], &d->cfm[i], gen->cf); + full_ieee &= vallTrue(vge(d->scp[i],vload(sharp_minscale))) && + vallTrue(vge(d->scm[i],vload(sharp_minscale))); + } + for (int i=0; ip1pr[i]; d->p1pr[i] -= d->p2mi[i]; d->p2mi[i] += tmp; + tmp = d->p1pi[i]; d->p1pi[i] += d->p2mr[i]; d->p2mr[i] -= tmp; + tmp = d->p1mr[i]; d->p1mr[i] += d->p2pi[i]; d->p2pi[i] -= tmp; + tmp = d->p1mi[i]; d->p1mi[i] -= d->p2pr[i]; d->p2pr[i] += tmp; + } + + while((!full_ieee) && (l<=lmax)) + { + Tv fx10=vload(fx[l+1].a),fx11=vload(fx[l+1].b); + Tv fx20=vload(fx[l+2].a),fx21=vload(fx[l+2].b); + Tv agr1=vzero, agi1=vzero, acr1=vzero, aci1=vzero; + Tv agr2=vzero, agi2=vzero, acr2=vzero, aci2=vzero; + full_ieee=1; + for (int i=0; il1p[i] = (d->cth[i]*fx10 - fx11)*d->l2p[i] - d->l1p[i]; + d->l1m[i] = (d->cth[i]*fx10 + fx11)*d->l2m[i] - d->l1m[i]; + Tv l2p = d->l2p[i]*d->cfp[i], l2m = d->l2m[i]*d->cfm[i]; + Tv l1p = d->l1p[i]*d->cfp[i], l1m = d->l1m[i]*d->cfm[i]; + agr1 += d->p1pr[i]*l2m + d->p2mi[i]*l2p; + agi1 += d->p1pi[i]*l2m - d->p2mr[i]*l2p; + acr1 += d->p1mr[i]*l2m - d->p2pi[i]*l2p; + aci1 += d->p1mi[i]*l2m + d->p2pr[i]*l2p; + agr2 += d->p2pr[i]*l1p - d->p1mi[i]*l1m; + agi2 += d->p2pi[i]*l1p + d->p1mr[i]*l1m; + acr2 += d->p2mr[i]*l1p + d->p1pi[i]*l1m; + aci2 += d->p2mi[i]*l1p - d->p1pr[i]*l1m; + + d->l2p[i] = (d->cth[i]*fx20 - fx21)*d->l1p[i] - d->l2p[i]; + d->l2m[i] = (d->cth[i]*fx20 + fx21)*d->l1m[i] - d->l2m[i]; + if (rescale(&d->l1p[i], &d->l2p[i], &d->scp[i], vload(sharp_ftol))) + getCorfac(d->scp[i], &d->cfp[i], gen->cf); + full_ieee &= vallTrue(vge(d->scp[i],vload(sharp_minscale))); + if (rescale(&d->l1m[i], &d->l2m[i], &d->scm[i], vload(sharp_ftol))) + getCorfac(d->scm[i], &d->cfm[i], gen->cf); + full_ieee &= vallTrue(vge(d->scm[i],vload(sharp_minscale))); + } + vhsum_cmplx_special (agr1,agi1,acr1,aci1,&alm[2*l]); + vhsum_cmplx_special (agr2,agi2,acr2,aci2,&alm[2*l+2]); + l+=2; + } + if (l>lmax) return; + + for (int i=0; il1p[i] *= d->cfp[i]; + d->l2p[i] *= d->cfp[i]; + d->l1m[i] *= d->cfm[i]; + d->l2m[i] *= d->cfm[i]; + } + map2alm_spin_kernel(d, fx, alm, l, lmax, nv2); + } + + +NOINLINE static void alm2map_deriv1_kernel(sxdata_v * restrict d, + const sharp_ylmgen_dbl2 * restrict fx, const dcmplx * restrict alm, + int l, int lmax, int nv2) + { + while (l<=lmax) + { + Tv fx10=vload(fx[l+1].a),fx11=vload(fx[l+1].b); + Tv fx20=vload(fx[l+2].a),fx21=vload(fx[l+2].b); + Tv ar1=vload(creal(alm[l ])), ai1=vload(cimag(alm[l ])), + ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1])); + for (int i=0; il1p[i] = (d->cth[i]*fx10 - fx11)*d->l2p[i] - d->l1p[i]; + d->l1m[i] = (d->cth[i]*fx10 + fx11)*d->l2m[i] - d->l1m[i]; + Tv lw=d->l2p[i]+d->l2m[i]; + d->p1pr[i] += ar1*lw; + d->p1pi[i] += ai1*lw; + Tv lx=d->l2m[i]-d->l2p[i]; + d->p2mr[i] += ai1*lx; + d->p2mi[i] -= ar1*lx; + lw=d->l1p[i]+d->l1m[i]; + d->p2pr[i] += ar2*lw; + d->p2pi[i] += ai2*lw; + lx=d->l1m[i]-d->l1p[i]; + d->p1mr[i] += ai2*lx; + d->p1mi[i] -= ar2*lx; + d->l2p[i] = (d->cth[i]*fx20 - fx21)*d->l1p[i] - d->l2p[i]; + d->l2m[i] = (d->cth[i]*fx20 + fx21)*d->l1m[i] - d->l2m[i]; + } + l+=2; + } + } + +NOINLINE static void calc_alm2map_deriv1(sharp_job * restrict job, + const sharp_Ylmgen_C * restrict gen, sxdata_v * restrict d, int nth) + { + int l,lmax=gen->lmax; + int nv2 = (nth+VLEN-1)/VLEN; + iter_to_ieee_spin(gen, d, &l, nv2); + job->opcnt += (l-gen->mhi) * 7*nth; + if (l>lmax) return; + job->opcnt += (lmax+1-l) * 17*nth; + + const sharp_ylmgen_dbl2 * restrict fx = gen->coef; + const dcmplx * restrict alm=job->almtmp; + int full_ieee=1; + for (int i=0; iscp[i], &d->cfp[i], gen->cf); + getCorfac(d->scm[i], &d->cfm[i], gen->cf); + full_ieee &= vallTrue(vge(d->scp[i],vload(sharp_minscale))) && + vallTrue(vge(d->scm[i],vload(sharp_minscale))); + } + + while((!full_ieee) && (l<=lmax)) + { + Tv fx10=vload(fx[l+1].a),fx11=vload(fx[l+1].b); + Tv fx20=vload(fx[l+2].a),fx21=vload(fx[l+2].b); + Tv ar1=vload(creal(alm[l ])), ai1=vload(cimag(alm[l ])), + ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1])); + full_ieee=1; + for (int i=0; il1p[i] = (d->cth[i]*fx10 - fx11)*d->l2p[i] - d->l1p[i]; + d->l1m[i] = (d->cth[i]*fx10 + fx11)*d->l2m[i] - d->l1m[i]; + Tv lw=d->l2p[i]*d->cfp[i]+d->l2m[i]*d->cfm[i]; + d->p1pr[i] += ar1*lw; + d->p1pi[i] += ai1*lw; + Tv lx=d->l2m[i]*d->cfm[i]-d->l2p[i]*d->cfp[i]; + d->p2mr[i] += ai1*lx; + d->p2mi[i] -= ar1*lx; + lw=d->l1p[i]*d->cfp[i]+d->l1m[i]*d->cfm[i]; + d->p2pr[i] += ar2*lw; + d->p2pi[i] += ai2*lw; + lx=d->l1m[i]*d->cfm[i]-d->l1p[i]*d->cfp[i]; + d->p1mr[i] += ai2*lx; + d->p1mi[i] -= ar2*lx; + d->l2p[i] = (d->cth[i]*fx20 - fx21)*d->l1p[i] - d->l2p[i]; + d->l2m[i] = (d->cth[i]*fx20 + fx21)*d->l1m[i] - d->l2m[i]; + if (rescale(&d->l1p[i], &d->l2p[i], &d->scp[i], vload(sharp_ftol))) + { + getCorfac(d->scp[i], &d->cfp[i], gen->cf); + full_ieee &= vallTrue(vge(d->scp[i],vload(sharp_minscale))); + } + if (rescale(&d->l1m[i], &d->l2m[i], &d->scm[i], vload(sharp_ftol))) + { + getCorfac(d->scm[i], &d->cfm[i], gen->cf); + full_ieee &= vallTrue(vge(d->scm[i],vload(sharp_minscale))); + } + } + l+=2; + } + if (l>lmax) return; + + for (int i=0; il1p[i] *= d->cfp[i]; + d->l2p[i] *= d->cfp[i]; + d->l1m[i] *= d->cfm[i]; + d->l2m[i] *= d->cfm[i]; + } + alm2map_deriv1_kernel(d, fx, alm, l, lmax, nv2); + } + + +#define VZERO(var) do { memset(&(var),0,sizeof(var)); } while(0) + +NOINLINE static void inner_loop_a2m(sharp_job *job, const int *ispair, + const double *cth_, const double *sth_, int llim, int ulim, + sharp_Ylmgen_C *gen, int mi, const int *mlim) + { + const int m = job->ainfo->mval[mi]; + sharp_Ylmgen_prepare (gen, m); + + switch (job->type) + { + case SHARP_ALM2MAP: + case SHARP_ALM2MAP_DERIV1: + { + if (job->spin==0) + { + //adjust the a_lm for the new algorithm + dcmplx * restrict alm=job->almtmp; + for (int il=0, l=gen->m; l<=gen->lmax; ++il,l+=2) + { + dcmplx al = alm[l]; + dcmplx al1 = (l+1>gen->lmax) ? 0. : alm[l+1]; + dcmplx al2 = (l+2>gen->lmax) ? 0. : alm[l+2]; + alm[l ] = gen->alpha[il]*(gen->eps[l+1]*al + gen->eps[l+2]*al2); + alm[l+1] = gen->alpha[il]*al1; + } + + const int nval=nv0*VLEN; + int ith=0; + int itgt[nval]; + while (ith=m) + { + itgt[nth] = ith; + d.s.csq[nth]=cth_[ith]*cth_[ith]; + d.s.sth[nth]=sth_[ith]; + ++nth; + } + else + { + int phas_idx = ith*job->s_th + mi*job->s_m; + job->phase[phas_idx] = job->phase[phas_idx+1] = 0; + } + ++ith; + } + if (nth>0) + { + int i2=((nth+VLEN-1)/VLEN)*VLEN; + for (int i=nth; is_th + mi*job->s_m; + complex double r1 = d.s.p1r[i] + d.s.p1i[i]*_Complex_I, + r2 = d.s.p2r[i] + d.s.p2i[i]*_Complex_I; + job->phase[phas_idx] = r1+r2; + if (ispair[tgt]) + job->phase[phas_idx+1] = r1-r2; + } + } + } + } + else + { + //adjust the a_lm for the new algorithm + if (job->nalm==2) + for (int l=gen->mhi; l<=gen->lmax+1; ++l) + { + job->almtmp[2*l ]*=gen->alpha[l]; + job->almtmp[2*l+1]*=gen->alpha[l]; + } + else + for (int l=gen->mhi; l<=gen->lmax+1; ++l) + job->almtmp[l]*=gen->alpha[l]; + + const int nval=nvx*VLEN; + int ith=0; + int itgt[nval]; + while (ith=m) + { + itgt[nth] = ith; + d.s.cth[nth]=cth_[ith]; d.s.sth[nth]=sth_[ith]; + ++nth; + } + else + { + int phas_idx = ith*job->s_th + mi*job->s_m; + job->phase[phas_idx ] = job->phase[phas_idx+1] = 0; + job->phase[phas_idx+2] = job->phase[phas_idx+3] = 0; + } + ++ith; + } + if (nth>0) + { + int i2=((nth+VLEN-1)/VLEN)*VLEN; + for (int i=nth; itype==SHARP_ALM2MAP) ? + calc_alm2map_spin (job, gen, &d.v, nth) : + calc_alm2map_deriv1(job, gen, &d.v, nth); + for (int i=0; is_th + mi*job->s_m; + complex double q1 = d.s.p1pr[i] + d.s.p1pi[i]*_Complex_I, + q2 = d.s.p2pr[i] + d.s.p2pi[i]*_Complex_I, + u1 = d.s.p1mr[i] + d.s.p1mi[i]*_Complex_I, + u2 = d.s.p2mr[i] + d.s.p2mi[i]*_Complex_I; + job->phase[phas_idx ] = q1+q2; + job->phase[phas_idx+2] = u1+u2; + if (ispair[tgt]) + { + dcmplx *phQ = &(job->phase[phas_idx+1]), + *phU = &(job->phase[phas_idx+3]); + *phQ = q1-q2; + *phU = u1-u2; + if ((gen->mhi-gen->m+gen->s)&1) + { *phQ=-(*phQ); *phU=-(*phU); } + } + } + } + } + } + break; + } + default: + { + UTIL_FAIL("must not happen"); + break; + } + } + } + +NOINLINE static void inner_loop_m2a(sharp_job *job, const int *ispair, + const double *cth_, const double *sth_, int llim, int ulim, + sharp_Ylmgen_C *gen, int mi, const int *mlim) + { + const int m = job->ainfo->mval[mi]; + sharp_Ylmgen_prepare (gen, m); + + switch (job->type) + { + case SHARP_MAP2ALM: + { + if (job->spin==0) + { + const int nval=nv0*VLEN; + int ith=0; + while (ith=m) + { + d.s.csq[nth]=cth_[ith]*cth_[ith]; d.s.sth[nth]=sth_[ith]; + int phas_idx = ith*job->s_th + mi*job->s_m; + dcmplx ph1=job->phase[phas_idx]; + dcmplx ph2=ispair[ith] ? job->phase[phas_idx+1] : 0.; + d.s.p1r[nth]=creal(ph1+ph2); d.s.p1i[nth]=cimag(ph1+ph2); + d.s.p2r[nth]=creal(ph1-ph2); d.s.p2i[nth]=cimag(ph1-ph2); + //adjust for new algorithm + d.s.p2r[nth]*=cth_[ith]; + d.s.p2i[nth]*=cth_[ith]; + ++nth; + } + ++ith; + } + if (nth>0) + { + int i2=((nth+VLEN-1)/VLEN)*VLEN; + for (int i=nth; ialmtmp; + dcmplx alm2 = 0.; + double alold=0; + for (int il=0, l=gen->m; l<=gen->lmax; ++il,l+=2) + { + dcmplx al = alm[l]; + dcmplx al1 = (l+1>gen->lmax) ? 0. : alm[l+1]; + alm[l ] = gen->alpha[il]*gen->eps[l+1]*al + alold*gen->eps[l]*alm2; + alm[l+1] = gen->alpha[il]*al1; + alm2=al; + alold=gen->alpha[il]; + } + } + else + { + const int nval=nvx*VLEN; + int ith=0; + while (ith=m) + { + d.s.cth[nth]=cth_[ith]; d.s.sth[nth]=sth_[ith]; + int phas_idx = ith*job->s_th + mi*job->s_m; + dcmplx p1Q=job->phase[phas_idx], + p1U=job->phase[phas_idx+2], + p2Q=ispair[ith] ? job->phase[phas_idx+1]:0., + p2U=ispair[ith] ? job->phase[phas_idx+3]:0.; + if ((gen->mhi-gen->m+gen->s)&1) + { p2Q=-p2Q; p2U=-p2U; } + d.s.p1pr[nth]=creal(p1Q+p2Q); d.s.p1pi[nth]=cimag(p1Q+p2Q); + d.s.p1mr[nth]=creal(p1U+p2U); d.s.p1mi[nth]=cimag(p1U+p2U); + d.s.p2pr[nth]=creal(p1Q-p2Q); d.s.p2pi[nth]=cimag(p1Q-p2Q); + d.s.p2mr[nth]=creal(p1U-p2U); d.s.p2mi[nth]=cimag(p1U-p2U); + ++nth; + } + ++ith; + } + if (nth>0) + { + int i2=((nth+VLEN-1)/VLEN)*VLEN; + for (int i=nth; imhi; l<=gen->lmax; ++l) + { + job->almtmp[2*l ]*=gen->alpha[l]; + job->almtmp[2*l+1]*=gen->alpha[l]; + } + } + break; + } + default: + { + UTIL_FAIL("must not happen"); + break; + } + } + } + +void XARCH(inner_loop) (sharp_job *job, const int *ispair, + const double *cth_, const double *sth_, int llim, int ulim, + sharp_Ylmgen_C *gen, int mi, const int *mlim); +void XARCH(inner_loop) (sharp_job *job, const int *ispair, + const double *cth_, const double *sth_, int llim, int ulim, + sharp_Ylmgen_C *gen, int mi, const int *mlim) + { + (job->type==SHARP_MAP2ALM) ? + inner_loop_m2a(job,ispair,cth_,sth_,llim,ulim,gen,mi,mlim) : + inner_loop_a2m(job,ispair,cth_,sth_,llim,ulim,gen,mi,mlim); + } + +#undef VZERO + +int XARCH(sharp_veclen)(void); +int XARCH(sharp_veclen)(void) + { + return VLEN; + } + +int XARCH(sharp_max_nvec)(int spin); +int XARCH(sharp_max_nvec)(int spin) + { + return (spin==0) ? nv0 : nvx; + } + +#define xstr(a) str(a) +#define str(a) #a +const char *XARCH(sharp_architecture)(void); +const char *XARCH(sharp_architecture)(void) + { + return xstr(ARCH); } diff --git a/libsharp/sharp_core_inc2.c b/libsharp/sharp_core_inc2.c deleted file mode 100644 index 5c9b4ab..0000000 --- a/libsharp/sharp_core_inc2.c +++ /dev/null @@ -1,803 +0,0 @@ -/* - * This file is part of libsharp. - * - * libsharp is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libsharp is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libsharp; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libsharp is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/*! \file sharp_core_inc2.c - * Type-dependent code for the computational core - * - * Copyright (C) 2012-2013 Max-Planck-Society - * \author Martin Reinecke - */ - -static void Z(alm2map_kernel) (const Tb cth, Y(Tbri) * restrict p1, - Y(Tbri) * restrict p2, Tb lam_1, Tb lam_2, - const sharp_ylmgen_dbl2 * restrict rf, const dcmplx * restrict alm, - int l, int lmax NJ1) - { -if (njobs>1) - { - while (llmax; - Tb lam_1,lam_2,scale; - Y(iter_to_ieee) (sth,cth,&l,&lam_1,&lam_2,&scale,gen); - job->opcnt += (l-gen->m) * 4*VLEN*nvec; - if (l>lmax) return; - job->opcnt += (lmax+1-l) * (4+4*njobs)*VLEN*nvec; - - Tb corfac; - Y(getCorfac)(scale,&corfac,gen->cf); - const sharp_ylmgen_dbl2 * restrict rf = gen->rf; - const dcmplx * restrict alm=job->almtmp; - int full_ieee = Y(TballGe)(scale,sharp_minscale); - while (!full_ieee) - { - for (int j=0; jlmax) break; - Tv r0=vload(rf[l-1].f[0]),r1=vload(rf[l-1].f[1]); - for (int i=0; ilmax) break; - r0=vload(rf[l-1].f[0]); r1=vload(rf[l-1].f[1]); - for (int i=0; icf); - full_ieee = Y(TballGe)(scale,sharp_minscale); - } - } - if (l>lmax) return; - - Y(Tbmuleq)(&lam_1,corfac); Y(Tbmuleq)(&lam_2,corfac); - Z(alm2map_kernel) (cth, p1, p2, lam_1, lam_2, rf, alm, l, lmax NJ2); - } - -static void Z(calc_map2alm) (const Tb cth, const Tb sth, - const sharp_Ylmgen_C *gen, sharp_job *job, const Y(Tbri) * restrict p1, - const Y(Tbri) * restrict p2 NJ1) - { - int lmax=gen->lmax; - Tb lam_1,lam_2,scale; - int l=gen->m; - Y(iter_to_ieee) (sth,cth,&l,&lam_1,&lam_2,&scale,gen); - job->opcnt += (l-gen->m) * 4*VLEN*nvec; - if (l>lmax) return; - job->opcnt += (lmax+1-l) * (4+4*njobs)*VLEN*nvec; - - const sharp_ylmgen_dbl2 * restrict rf = gen->rf; - Tb corfac; - Y(getCorfac)(scale,&corfac,gen->cf); - dcmplx * restrict alm=job->almtmp; - int full_ieee = Y(TballGe)(scale,sharp_minscale); - while (!full_ieee) - { - for (int j=0; jlmax) return; - Tv r0=vload(rf[l-1].f[0]),r1=vload(rf[l-1].f[1]); - for (int i=0; ilmax) return; - r0=vload(rf[l-1].f[0]); r1=vload(rf[l-1].f[1]); - for (int i=0; icf); - full_ieee = Y(TballGe)(scale,sharp_minscale); - } - } - - Y(Tbmuleq)(&lam_1,corfac); Y(Tbmuleq)(&lam_2,corfac); - Z(map2alm_kernel) (cth, p1, p2, lam_1, lam_2, rf, alm, l, lmax NJ2); - } - -static inline void Z(saddstep) (Y(Tbqu) * restrict px, Y(Tbqu) * restrict py, - const Tb rxp, const Tb rxm, const dcmplx * restrict alm NJ1) - { - for (int j=0; jv[i],rxm->v[i]); - vfmaeq(agr,px[j].qr.v[i],lw); - vfmaeq(agi,px[j].qi.v[i],lw); - vfmaeq(acr,px[j].ur.v[i],lw); - vfmaeq(aci,px[j].ui.v[i],lw); - } - for (int i=0; iv[i],rxp->v[i]); - vfmseq(agr,py[j].ui.v[i],lx); - vfmaeq(agi,py[j].ur.v[i],lx); - vfmaeq(acr,py[j].qi.v[i],lx); - vfmseq(aci,py[j].qr.v[i],lx); - } - vhsum_cmplx2(agr,agi,acr,aci,&alm[2*j],&alm[2*j+1]); - } - } - -static void Z(alm2map_spin_kernel) (Tb cth, Y(Tbqu) * restrict p1, - Y(Tbqu) * restrict p2, Tb rec1p, Tb rec1m, Tb rec2p, Tb rec2m, - const sharp_ylmgen_dbl3 * restrict fx, const dcmplx * restrict alm, int l, - int lmax NJ1) - { - while (llmax; - Tb rec1p, rec1m, rec2p, rec2m, scalem, scalep; - Y(iter_to_ieee_spin) - (cth,sth,&l,&rec1p,&rec1m,&rec2p,&rec2m,&scalep,&scalem,gen); - job->opcnt += (l-gen->m) * 10*VLEN*nvec; - if (l>lmax) return; - job->opcnt += (lmax+1-l) * (12+16*njobs)*VLEN*nvec; - - const sharp_ylmgen_dbl3 * restrict fx = gen->fx; - Tb corfacp,corfacm; - Y(getCorfac)(scalep,&corfacp,gen->cf); - Y(getCorfac)(scalem,&corfacm,gen->cf); - const dcmplx * restrict alm=job->almtmp; - int full_ieee = Y(TballGe)(scalep,sharp_minscale) - && Y(TballGe)(scalem,sharp_minscale); - while (!full_ieee) - { - Z(saddstep)(p1, p2, Y(Tbprod)(rec2p,corfacp), Y(Tbprod)(rec2m,corfacm), - &alm[2*njobs*l] NJ2); - if (++l>lmax) break; - Y(rec_step)(&rec1p,&rec1m,&rec2p,&rec2m,cth,fx[l]); - Z(saddstep)(p2, p1, Y(Tbprod)(rec1p,corfacp), Y(Tbprod)(rec1m,corfacm), - &alm[2*njobs*l] NJ2); - if (++l>lmax) break; - Y(rec_step)(&rec2p,&rec2m,&rec1p,&rec1m,cth,fx[l]); - if (Y(rescale)(&rec1p,&rec2p,&scalep) | Y(rescale)(&rec1m,&rec2m,&scalem)) - { - Y(getCorfac)(scalep,&corfacp,gen->cf); - Y(getCorfac)(scalem,&corfacm,gen->cf); - full_ieee = Y(TballGe)(scalep,sharp_minscale) - && Y(TballGe)(scalem,sharp_minscale); - } - } - - if (l>lmax) return; - - Y(Tbmuleq)(&rec1p,corfacp); Y(Tbmuleq)(&rec2p,corfacp); - Y(Tbmuleq)(&rec1m,corfacm); Y(Tbmuleq)(&rec2m,corfacm); - Z(alm2map_spin_kernel) (cth, p1, p2, rec1p, rec1m, rec2p, rec2m, fx, alm, l, - lmax NJ2); - } - -static void Z(calc_map2alm_spin) (Tb cth, Tb sth, - const sharp_Ylmgen_C * restrict gen, sharp_job *job, - const Y(Tbqu) * restrict p1, const Y(Tbqu) * restrict p2 NJ1) - { - int l, lmax=gen->lmax; - Tb rec1p, rec1m, rec2p, rec2m, scalem, scalep; - Y(iter_to_ieee_spin) - (cth,sth,&l,&rec1p,&rec1m,&rec2p,&rec2m,&scalep,&scalem,gen); - job->opcnt += (l-gen->m) * 10*VLEN*nvec; - if (l>lmax) return; - job->opcnt += (lmax+1-l) * (12+16*njobs)*VLEN*nvec; - - const sharp_ylmgen_dbl3 * restrict fx = gen->fx; - Tb corfacp,corfacm; - Y(getCorfac)(scalep,&corfacp,gen->cf); - Y(getCorfac)(scalem,&corfacm,gen->cf); - dcmplx * restrict alm=job->almtmp; - int full_ieee = Y(TballGe)(scalep,sharp_minscale) - && Y(TballGe)(scalem,sharp_minscale); - while (!full_ieee) - { - Tb t1=Y(Tbprod)(rec2p,corfacp), t2=Y(Tbprod)(rec2m,corfacm); - Z(saddstep2)(p1, p2, &t1, &t2, &alm[2*njobs*l] NJ2); - if (++l>lmax) return; - Y(rec_step)(&rec1p,&rec1m,&rec2p,&rec2m,cth,fx[l]); - t1=Y(Tbprod)(rec1p,corfacp); t2=Y(Tbprod)(rec1m,corfacm); - Z(saddstep2)(p2, p1, &t1, &t2, &alm[2*njobs*l] NJ2); - if (++l>lmax) return; - Y(rec_step)(&rec2p,&rec2m,&rec1p,&rec1m,cth,fx[l]); - if (Y(rescale)(&rec1p,&rec2p,&scalep) | Y(rescale)(&rec1m,&rec2m,&scalem)) - { - Y(getCorfac)(scalep,&corfacp,gen->cf); - Y(getCorfac)(scalem,&corfacm,gen->cf); - full_ieee = Y(TballGe)(scalep,sharp_minscale) - && Y(TballGe)(scalem,sharp_minscale); - } - } - - Y(Tbmuleq)(&rec1p,corfacp); Y(Tbmuleq)(&rec2p,corfacp); - Y(Tbmuleq)(&rec1m,corfacm); Y(Tbmuleq)(&rec2m,corfacm); - Z(map2alm_spin_kernel)(cth,p1,p2,rec1p,rec1m,rec2p,rec2m,fx,alm,l,lmax NJ2); - } - -static inline void Z(saddstep_d) (Y(Tbqu) * restrict px, Y(Tbqu) * restrict py, - const Tb rxp, const Tb rxm, const dcmplx * restrict alm NJ1) - { - for (int j=0; jlmax; - Tb rec1p, rec1m, rec2p, rec2m, scalem, scalep; - Y(iter_to_ieee_spin) - (cth,sth,&l,&rec1p,&rec1m,&rec2p,&rec2m,&scalep,&scalem,gen); - job->opcnt += (l-gen->m) * 10*VLEN*nvec; - if (l>lmax) return; - job->opcnt += (lmax+1-l) * (12+8*njobs)*VLEN*nvec; - - const sharp_ylmgen_dbl3 * restrict fx = gen->fx; - Tb corfacp,corfacm; - Y(getCorfac)(scalep,&corfacp,gen->cf); - Y(getCorfac)(scalem,&corfacm,gen->cf); - const dcmplx * restrict alm=job->almtmp; - int full_ieee = Y(TballGe)(scalep,sharp_minscale) - && Y(TballGe)(scalem,sharp_minscale); - while (!full_ieee) - { - Z(saddstep_d)(p1, p2, Y(Tbprod)(rec2p,corfacp), Y(Tbprod)(rec2m,corfacm), - &alm[njobs*l] NJ2); - if (++l>lmax) break; - Y(rec_step)(&rec1p,&rec1m,&rec2p,&rec2m,cth,fx[l]); - Z(saddstep_d)(p2, p1, Y(Tbprod)(rec1p,corfacp), Y(Tbprod)(rec1m,corfacm), - &alm[njobs*l] NJ2); - if (++l>lmax) break; - Y(rec_step)(&rec2p,&rec2m,&rec1p,&rec1m,cth,fx[l]); - if (Y(rescale)(&rec1p,&rec2p,&scalep) | Y(rescale)(&rec1m,&rec2m,&scalem)) - { - Y(getCorfac)(scalep,&corfacp,gen->cf); - Y(getCorfac)(scalem,&corfacm,gen->cf); - full_ieee = Y(TballGe)(scalep,sharp_minscale) - && Y(TballGe)(scalem,sharp_minscale); - } - } - - if (l>lmax) return; - - Y(Tbmuleq)(&rec1p,corfacp); Y(Tbmuleq)(&rec2p,corfacp); - Y(Tbmuleq)(&rec1m,corfacm); Y(Tbmuleq)(&rec2m,corfacm); - Z(alm2map_deriv1_kernel) (cth, p1, p2, rec1p, rec1m, rec2p, rec2m, fx, alm, l, - lmax NJ2); - } - - -#define VZERO(var) do { memset(&(var),0,sizeof(var)); } while(0) - -static void Z(inner_loop) (sharp_job *job, const int *ispair, - const double *cth_, const double *sth_, int llim, int ulim, - sharp_Ylmgen_C *gen, int mi, const int *mlim NJ1) - { - const int nval=nvec*VLEN; - const int m = job->ainfo->mval[mi]; - sharp_Ylmgen_prepare (gen, m); - - switch (job->type) - { - case SHARP_ALM2MAP: - case SHARP_ALM2MAP_DERIV1: - { - if (job->spin==0) - { - for (int ith=0; ith=ulim-llim) itot=ulim-llim-1; - if (mlim[itot]>=m) skip=0; - cth.s[i]=cth_[itot]; sth.s[i]=sth_[itot]; - } - if (!skip) - Z(calc_alm2map) (cth.b,sth.b,gen,job,&p1[0].b,&p2[0].b NJ2); - - for (int i=0; is_th + mi*job->s_m + 2*j; - complex double r1 = p1[j].s.r[i] + p1[j].s.i[i]*_Complex_I, - r2 = p2[j].s.r[i] + p2[j].s.i[i]*_Complex_I; - job->phase[phas_idx] = r1+r2; - if (ispair[itot]) - job->phase[phas_idx+1] = r1-r2; - } - } - } - } - } - else - { - for (int ith=0; ith=ulim-llim) itot=ulim-llim-1; - if (mlim[itot]>=m) skip=0; - cth.s[i]=cth_[itot]; sth.s[i]=sth_[itot]; - } - if (!skip) - (job->type==SHARP_ALM2MAP) ? - Z(calc_alm2map_spin ) - (cth.b,sth.b,gen,job,&p1[0].b,&p2[0].b NJ2) : - Z(calc_alm2map_deriv1) - (cth.b,sth.b,gen,job,&p1[0].b,&p2[0].b NJ2); - - for (int i=0; is_th + mi*job->s_m + 4*j; - complex double q1 = p1[j].s.qr[i] + p1[j].s.qi[i]*_Complex_I, - q2 = p2[j].s.qr[i] + p2[j].s.qi[i]*_Complex_I, - u1 = p1[j].s.ur[i] + p1[j].s.ui[i]*_Complex_I, - u2 = p2[j].s.ur[i] + p2[j].s.ui[i]*_Complex_I; - job->phase[phas_idx] = q1+q2; - job->phase[phas_idx+2] = u1+u2; - if (ispair[itot]) - { - dcmplx *phQ = &(job->phase[phas_idx+1]), - *phU = &(job->phase[phas_idx+3]); - *phQ = q1-q2; - *phU = u1-u2; - if ((gen->mhi-gen->m+gen->s)&1) - { *phQ=-(*phQ); *phU=-(*phU); } - } - } - } - } - } - } - break; - } - case SHARP_MAP2ALM: - { - if (job->spin==0) - { - for (int ith=0; ith=ulim-llim) itot=ulim-llim-1; - if (mlim[itot]>=m) skip=0; - cth.s[i]=cth_[itot]; sth.s[i]=sth_[itot]; - if ((i+ith=m)) - { - for (int j=0; js_th + mi*job->s_m + 2*j; - dcmplx ph1=job->phase[phas_idx]; - dcmplx ph2=ispair[itot] ? job->phase[phas_idx+1] : 0.; - p1[j].s.r[i]=creal(ph1+ph2); p1[j].s.i[i]=cimag(ph1+ph2); - p2[j].s.r[i]=creal(ph1-ph2); p2[j].s.i[i]=cimag(ph1-ph2); - } - } - } - if (!skip) - Z(calc_map2alm)(cth.b,sth.b,gen,job,&p1[0].b,&p2[0].b NJ2); - } - } - else - { - for (int ith=0; ith=ulim-llim) itot=ulim-llim-1; - if (mlim[itot]>=m) skip=0; - cth.s[i]=cth_[itot]; sth.s[i]=sth_[itot]; - if (i+iths_th + mi*job->s_m + 4*j; - dcmplx p1Q=job->phase[phas_idx], - p1U=job->phase[phas_idx+2], - p2Q=ispair[itot] ? job->phase[phas_idx+1]:0., - p2U=ispair[itot] ? job->phase[phas_idx+3]:0.; - if ((gen->mhi-gen->m+gen->s)&1) - { p2Q=-p2Q; p2U=-p2U; } - p1[j].s.qr[i]=creal(p1Q+p2Q); p1[j].s.qi[i]=cimag(p1Q+p2Q); - p1[j].s.ur[i]=creal(p1U+p2U); p1[j].s.ui[i]=cimag(p1U+p2U); - p2[j].s.qr[i]=creal(p1Q-p2Q); p2[j].s.qi[i]=cimag(p1Q-p2Q); - p2[j].s.ur[i]=creal(p1U-p2U); p2[j].s.ui[i]=cimag(p1U-p2U); - } - } - } - if (!skip) - Z(calc_map2alm_spin) (cth.b,sth.b,gen,job,&p1[0].b,&p2[0].b NJ2); - } - } - break; - } - default: - { - UTIL_FAIL("must not happen"); - break; - } - } - } - -#undef VZERO diff --git a/libsharp/sharp_core_inchelper.c b/libsharp/sharp_core_inchelper.c deleted file mode 100644 index 89d79cd..0000000 --- a/libsharp/sharp_core_inchelper.c +++ /dev/null @@ -1,70 +0,0 @@ -#define Tb CONCAT2(Tb,nvec) -#define Y(arg) CONCAT2(arg,nvec) -#include "sharp_core_inc.c" - -#if (SHARP_MAXTRANS>MAXJOB_SPECIAL) -#define NJ1 , int njobs -#define NJ2 , njobs -#define Z(arg) CONCAT2(arg,nvec) -#include "sharp_core_inc2.c" -#undef Z -#undef NJ1 -#undef NJ2 -#endif - -#define NJ1 -#define NJ2 - -#if ((MAXJOB_SPECIAL>=1)&&(SHARP_MAXTRANS>=1)) -#define njobs 1 -#define Z(arg) CONCAT3(arg,nvec,njobs) -#include "sharp_core_inc2.c" -#undef Z -#undef njobs -#endif - -#if ((MAXJOB_SPECIAL>=2)&&(SHARP_MAXTRANS>=2)) -#define njobs 2 -#define Z(arg) CONCAT3(arg,nvec,njobs) -#include "sharp_core_inc2.c" -#undef Z -#undef njobs -#endif - -#if ((MAXJOB_SPECIAL>=3)&&(SHARP_MAXTRANS>=3)) -#define njobs 3 -#define Z(arg) CONCAT3(arg,nvec,njobs) -#include "sharp_core_inc2.c" -#undef Z -#undef njobs -#endif - -#if ((MAXJOB_SPECIAL>=4)&&(SHARP_MAXTRANS>=4)) -#define njobs 4 -#define Z(arg) CONCAT3(arg,nvec,njobs) -#include "sharp_core_inc2.c" -#undef Z -#undef njobs -#endif - -#if ((MAXJOB_SPECIAL>=5)&&(SHARP_MAXTRANS>=5)) -#define njobs 5 -#define Z(arg) CONCAT3(arg,nvec,njobs) -#include "sharp_core_inc2.c" -#undef Z -#undef njobs -#endif - -#if ((MAXJOB_SPECIAL>=6)&&(SHARP_MAXTRANS>=6)) -#define njobs 6 -#define Z(arg) CONCAT3(arg,nvec,njobs) -#include "sharp_core_inc2.c" -#undef Z -#undef njobs -#endif - -#undef NJ1 -#undef NJ2 - -#undef Y -#undef Tb diff --git a/libsharp/sharp_cxx.h b/libsharp/sharp_cxx.h index f8b2365..b2415e0 100644 --- a/libsharp/sharp_cxx.h +++ b/libsharp/sharp_cxx.h @@ -25,14 +25,15 @@ /*! \file sharp_cxx.h * Spherical transform library * - * Copyright (C) 2012-2015 Max-Planck-Society + * Copyright (C) 2012-2019 Max-Planck-Society * \author Martin Reinecke */ #ifndef PLANCK_SHARP_CXX_H #define PLANCK_SHARP_CXX_H -#include "sharp_lowlevel.h" +#include +#include "sharp.h" #include "sharp_geomhelpers.h" #include "sharp_almhelpers.h" @@ -47,8 +48,8 @@ class sharp_base : ainfo(0), ginfo(0) {} ~sharp_base() { - sharp_destroy_geom_info(ginfo); - sharp_destroy_alm_info(ainfo); + if (ginfo) sharp_destroy_geom_info(ginfo); + if (ainfo) sharp_destroy_alm_info(ainfo); } void set_general_geometry (int nrings, const int *nph, const ptrdiff_t *ofs, @@ -107,47 +108,115 @@ template class sharp_cxxjob: public sharp_base private: static void *conv (T *ptr) { return reinterpret_cast(ptr); } + static void *conv (std::complex *ptr) + { return reinterpret_cast(ptr); } static void *conv (const T *ptr) { return const_cast(reinterpret_cast(ptr)); } + static void *conv (const std::complex *ptr) + { return const_cast(reinterpret_cast(ptr)); } public: - void alm2map (const T *alm, T *map, bool add) + void alm2map (const T *alm, T *map, bool add) const { void *aptr=conv(alm), *mptr=conv(map); int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); - sharp_execute (SHARP_ALM2MAP, 0, &aptr, &mptr, ginfo, ainfo, 1, - flags,0,0); + sharp_execute (SHARP_ALM2MAP, 0, &aptr, &mptr, ginfo, ainfo, flags, 0, 0); } - void alm2map_spin (const T *alm1, const T *alm2, T *map1, T *map2, - int spin, bool add) + void alm2map (const std::complex *alm, T *map, bool add) const + { + void *aptr=conv(alm), *mptr=conv(map); + int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); + sharp_execute (SHARP_ALM2MAP, 0, &aptr, &mptr, ginfo, ainfo, flags, 0, 0); + } + void alm2map_spin (const T *alm1, const T *alm2, + T *map1, T *map2, int spin, bool add) const { void *aptr[2], *mptr[2]; aptr[0]=conv(alm1); aptr[1]=conv(alm2); mptr[0]=conv(map1); mptr[1]=conv(map2); int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); - sharp_execute (SHARP_ALM2MAP,spin,aptr,mptr,ginfo,ainfo,1,flags,0,0); + sharp_execute (SHARP_ALM2MAP,spin,aptr,mptr,ginfo,ainfo,flags, 0, 0); } - void alm2map_der1 (const T *alm, T *map1, T *map2, bool add) + void alm2map_spin (const std::complex *alm1, const std::complex *alm2, + T *map1, T *map2, int spin, bool add) const + { + void *aptr[2], *mptr[2]; + aptr[0]=conv(alm1); aptr[1]=conv(alm2); + mptr[0]=conv(map1); mptr[1]=conv(map2); + int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); + sharp_execute (SHARP_ALM2MAP, spin, aptr, mptr, ginfo, ainfo, flags,0,0); + } + void alm2map_der1 (const T *alm, T *map1, T *map2, bool add) const { void *aptr=conv(alm), *mptr[2]; mptr[0]=conv(map1); mptr[1]=conv(map2); int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); - sharp_execute (SHARP_ALM2MAP_DERIV1,1,&aptr,mptr,ginfo,ainfo,1,flags,0,0); + sharp_execute (SHARP_ALM2MAP_DERIV1,1,&aptr,mptr,ginfo,ainfo,flags,0,0); } - void map2alm (const T *map, T *alm, bool add) + void alm2map_der1 (const std::complex *alm, T *map1, T *map2, bool add) + const + { + void *aptr=conv(alm), *mptr[2]; + mptr[0]=conv(map1); mptr[1]=conv(map2); + int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); + sharp_execute (SHARP_ALM2MAP_DERIV1,1,&aptr,mptr,ginfo,ainfo,flags,0,0); + } + void alm2map_adjoint (const T *map, T *alm, bool add) const { void *aptr=conv(alm), *mptr=conv(map); int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); - sharp_execute (SHARP_MAP2ALM,0,&aptr,&mptr,ginfo,ainfo,1,flags,0,0); + sharp_execute (SHARP_Yt,0,&aptr,&mptr,ginfo,ainfo,flags,0,0); } - void map2alm_spin (const T *map1, const T *map2, T *alm1, T *alm2, - int spin, bool add) + void alm2map_adjoint (const T *map, std::complex *alm, bool add) const + { + void *aptr=conv(alm), *mptr=conv(map); + int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); + sharp_execute (SHARP_Yt,0,&aptr,&mptr,ginfo,ainfo,flags,0,0); + } + void alm2map_spin_adjoint (const T *map1, const T *map2, T *alm1, T *alm2, + int spin, bool add) const { void *aptr[2], *mptr[2]; aptr[0]=conv(alm1); aptr[1]=conv(alm2); mptr[0]=conv(map1); mptr[1]=conv(map2); int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); - sharp_execute (SHARP_MAP2ALM,spin,aptr,mptr,ginfo,ainfo,1,flags,0,0); + sharp_execute (SHARP_Yt,spin,aptr,mptr,ginfo,ainfo,flags,0,0); + } + void alm2map_spin_adjoint (const T *map1, const T *map2, + std::complex *alm1, std::complex *alm2, int spin, bool add) const + { + alm2map_spin_adjoint (map1, map2, reinterpret_cast(alm1), + reinterpret_cast(alm2), spin, add); + } + void map2alm (const T *map, T *alm, bool add) const + { + void *aptr=conv(alm), *mptr=conv(map); + int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); + sharp_execute (SHARP_MAP2ALM,0,&aptr,&mptr,ginfo,ainfo,flags,0,0); + } + void map2alm (const T *map, std::complex *alm, bool add) const + { + void *aptr=conv(alm), *mptr=conv(map); + int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); + sharp_execute (SHARP_MAP2ALM,0,&aptr,&mptr,ginfo,ainfo,flags,0,0); + } + void map2alm_spin (const T *map1, const T *map2, T *alm1, T *alm2, + int spin, bool add) const + { + void *aptr[2], *mptr[2]; + aptr[0]=conv(alm1); aptr[1]=conv(alm2); + mptr[0]=conv(map1); mptr[1]=conv(map2); + int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); + sharp_execute (SHARP_MAP2ALM,spin,aptr,mptr,ginfo,ainfo,flags,0,0); + } + void map2alm_spin (const T *map1, const T *map2, std::complex *alm1, + std::complex *alm2, int spin, bool add) const + { + void *aptr[2], *mptr[2]; + aptr[0]=conv(alm1); aptr[1]=conv(alm2); + mptr[0]=conv(map1); mptr[1]=conv(map2); + int flags=cxxjobhelper__::val | (add ? SHARP_ADD : 0); + sharp_execute (SHARP_MAP2ALM,spin,aptr,mptr,ginfo,ainfo,flags,0,0); } }; diff --git a/libsharp/sharp_geomhelpers.c b/libsharp/sharp_geomhelpers.c index dbb44e0..0f6af39 100644 --- a/libsharp/sharp_geomhelpers.c +++ b/libsharp/sharp_geomhelpers.c @@ -25,17 +25,15 @@ /*! \file sharp_geomhelpers.c * Spherical transform library * - * Copyright (C) 2006-2012 Max-Planck-Society
    - * Copyright (C) 2007-2008 Pavel Holoborodko (for gauss_legendre_tbl) - * \author Martin Reinecke \author Pavel Holoborodko + * Copyright (C) 2006-2018 Max-Planck-Society + * \author Martin Reinecke */ #include #include "sharp_geomhelpers.h" #include "sharp_legendre_roots.h" #include "c_utils.h" -#include "ls_fft.h" -#include +#include "pocketfft/pocketfft.h" void sharp_make_subset_healpix_geom_info (int nside, int stride, int nrings, const int *rings, const double *weight, sharp_geom_info **geom_info) @@ -161,9 +159,9 @@ void sharp_make_fejer1_geom_info (int nrings, int ppring, double phi0, weight[2*k ]=2./(1.-4.*k*k)*sin((k*pi)/nrings); } if ((nrings&1)==0) weight[nrings-1]=0.; - real_plan plan = make_real_plan(nrings); - real_plan_backward_fftpack(plan,weight); - kill_real_plan(plan); + rfft_plan plan = make_rfft_plan(nrings); + rfft_backward(plan,weight,1.); + destroy_rfft_plan(plan); for (int m=0; m<(nrings+1)/2; ++m) { @@ -208,9 +206,9 @@ void sharp_make_cc_geom_info (int nrings, int ppring, double phi0, for (int k=1; k<=(n/2-1); ++k) weight[2*k-1]=2./(1.-4.*k*k) + dw; weight[2*(n/2)-1]=(n-3.)/(2*(n/2)-1) -1. -dw*((2-(n&1))*n-1); - real_plan plan = make_real_plan(n); - real_plan_backward_fftpack(plan,weight); - kill_real_plan(plan); + rfft_plan plan = make_rfft_plan(n); + rfft_backward(plan,weight,1.); + destroy_rfft_plan(plan); weight[n]=weight[0]; for (int m=0; m<(nrings+1)/2; ++m) @@ -256,9 +254,9 @@ void sharp_make_fejer2_geom_info (int nrings, int ppring, double phi0, for (int k=1; k<=(n/2-1); ++k) weight[2*k-1]=2./(1.-4.*k*k); weight[2*(n/2)-1]=(n-3.)/(2*(n/2)-1) -1.; - real_plan plan = make_real_plan(n); - real_plan_backward_fftpack(plan,weight); - kill_real_plan(plan); + rfft_plan plan = make_rfft_plan(n); + rfft_backward(plan,weight,1.); + destroy_rfft_plan(plan); for (int m=0; m #include "sharp.h" +#include "sharp_ylmgen_c.h" #define SHARP_MAXTRANS 100 @@ -55,12 +57,17 @@ typedef struct const sharp_geom_info *ginfo; const sharp_alm_info *ainfo; double time; - int ntrans; unsigned long long opcnt; } sharp_job; -int sharp_get_nv_max (void); -int sharp_nv_oracle (sharp_jobtype type, int spin, int ntrans); int sharp_get_mlim (int lmax, int spin, double sth, double cth); +void inner_loop (sharp_job *job, const int *ispair,const double *cth, + const double *sth, int llim, int ulim, sharp_Ylmgen_C *gen, int mi, + const int *mlim); + +int sharp_veclen(void); +int sharp_max_nvec(int spin); +const char *sharp_architecture(void); + #endif diff --git a/libsharp/sharp_legendre.c b/libsharp/sharp_legendre.c deleted file mode 100644 index 24d69b6..0000000 --- a/libsharp/sharp_legendre.c +++ /dev/null @@ -1,1319 +0,0 @@ -/* DO NOT EDIT. md5sum of source: a8c5c18a7a19c378187dbf461d12eb5c *//* - - NOTE NOTE NOTE - - This file is edited in sharp_legendre.c.in which is then preprocessed. - Do not make manual modifications to sharp_legendre.c. - - NOTE NOTE NOTE - -*/ - - -/* - * This file is part of libsharp. - * - * Redistribution and use in source and binary forms, with or without - * met: - * - * 1. Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * - * 2. Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution. - * - * 3. Neither the name of the copyright holder nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -/*! \file sharp_legendre.c.in - * - * Copyright (C) 2015 University of Oslo - * \author Dag Sverre Seljebotn - */ - -#ifndef NO_LEGENDRE -#if (VLEN==8) -#error This code is not tested with MIC; please compile with -DNO_LEGENDRE -/* ...or test it (it probably works) and remove this check */ -#endif - -#ifndef SHARP_LEGENDRE_CS -#define SHARP_LEGENDRE_CS 4 -#endif - -#define MAX_CS 6 -#if (SHARP_LEGENDRE_CS > MAX_CS) -#error (SHARP_LEGENDRE_CS > MAX_CS) -#endif - -#include "sharp_legendre.h" -#include "sharp_vecsupport.h" - -#include - - - -static void legendre_transform_vec1(double *recfacs, double *bl, ptrdiff_t lmax, - double xarr[(1) * VLEN], - double out[(1) * VLEN]) { - - Tv P_0, Pm1_0, Pm2_0, x0, y0; - - Tv W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu(xarr + 0 * VLEN); - Pm1_0 = vload(1.0); - P_0 = x0; - b = vload(*bl); - y0 = vmul(Pm1_0, b); - - - b = vload(*(bl + 1)); - - vfmaeq(y0, P_0, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload(*(bl + l)); - R = vload(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul(x0, Pm1_0); - W2 = W1; - W2 = vsub(W2, Pm2_0); - P_0 = W1; - vfmaeq(P_0, W2, R); - vfmaeq(y0, P_0, b); - - - } - - vstoreu(out + 0 * VLEN, y0); - -} - -static void legendre_transform_vec2(double *recfacs, double *bl, ptrdiff_t lmax, - double xarr[(2) * VLEN], - double out[(2) * VLEN]) { - - Tv P_0, Pm1_0, Pm2_0, x0, y0; - - Tv P_1, Pm1_1, Pm2_1, x1, y1; - - Tv W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu(xarr + 0 * VLEN); - Pm1_0 = vload(1.0); - P_0 = x0; - b = vload(*bl); - y0 = vmul(Pm1_0, b); - - x1 = vloadu(xarr + 1 * VLEN); - Pm1_1 = vload(1.0); - P_1 = x1; - b = vload(*bl); - y1 = vmul(Pm1_1, b); - - - b = vload(*(bl + 1)); - - vfmaeq(y0, P_0, b); - - vfmaeq(y1, P_1, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload(*(bl + l)); - R = vload(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul(x0, Pm1_0); - W2 = W1; - W2 = vsub(W2, Pm2_0); - P_0 = W1; - vfmaeq(P_0, W2, R); - vfmaeq(y0, P_0, b); - - Pm2_1 = Pm1_1; Pm1_1 = P_1; - W1 = vmul(x1, Pm1_1); - W2 = W1; - W2 = vsub(W2, Pm2_1); - P_1 = W1; - vfmaeq(P_1, W2, R); - vfmaeq(y1, P_1, b); - - - } - - vstoreu(out + 0 * VLEN, y0); - - vstoreu(out + 1 * VLEN, y1); - -} - -static void legendre_transform_vec3(double *recfacs, double *bl, ptrdiff_t lmax, - double xarr[(3) * VLEN], - double out[(3) * VLEN]) { - - Tv P_0, Pm1_0, Pm2_0, x0, y0; - - Tv P_1, Pm1_1, Pm2_1, x1, y1; - - Tv P_2, Pm1_2, Pm2_2, x2, y2; - - Tv W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu(xarr + 0 * VLEN); - Pm1_0 = vload(1.0); - P_0 = x0; - b = vload(*bl); - y0 = vmul(Pm1_0, b); - - x1 = vloadu(xarr + 1 * VLEN); - Pm1_1 = vload(1.0); - P_1 = x1; - b = vload(*bl); - y1 = vmul(Pm1_1, b); - - x2 = vloadu(xarr + 2 * VLEN); - Pm1_2 = vload(1.0); - P_2 = x2; - b = vload(*bl); - y2 = vmul(Pm1_2, b); - - - b = vload(*(bl + 1)); - - vfmaeq(y0, P_0, b); - - vfmaeq(y1, P_1, b); - - vfmaeq(y2, P_2, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload(*(bl + l)); - R = vload(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul(x0, Pm1_0); - W2 = W1; - W2 = vsub(W2, Pm2_0); - P_0 = W1; - vfmaeq(P_0, W2, R); - vfmaeq(y0, P_0, b); - - Pm2_1 = Pm1_1; Pm1_1 = P_1; - W1 = vmul(x1, Pm1_1); - W2 = W1; - W2 = vsub(W2, Pm2_1); - P_1 = W1; - vfmaeq(P_1, W2, R); - vfmaeq(y1, P_1, b); - - Pm2_2 = Pm1_2; Pm1_2 = P_2; - W1 = vmul(x2, Pm1_2); - W2 = W1; - W2 = vsub(W2, Pm2_2); - P_2 = W1; - vfmaeq(P_2, W2, R); - vfmaeq(y2, P_2, b); - - - } - - vstoreu(out + 0 * VLEN, y0); - - vstoreu(out + 1 * VLEN, y1); - - vstoreu(out + 2 * VLEN, y2); - -} - -static void legendre_transform_vec4(double *recfacs, double *bl, ptrdiff_t lmax, - double xarr[(4) * VLEN], - double out[(4) * VLEN]) { - - Tv P_0, Pm1_0, Pm2_0, x0, y0; - - Tv P_1, Pm1_1, Pm2_1, x1, y1; - - Tv P_2, Pm1_2, Pm2_2, x2, y2; - - Tv P_3, Pm1_3, Pm2_3, x3, y3; - - Tv W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu(xarr + 0 * VLEN); - Pm1_0 = vload(1.0); - P_0 = x0; - b = vload(*bl); - y0 = vmul(Pm1_0, b); - - x1 = vloadu(xarr + 1 * VLEN); - Pm1_1 = vload(1.0); - P_1 = x1; - b = vload(*bl); - y1 = vmul(Pm1_1, b); - - x2 = vloadu(xarr + 2 * VLEN); - Pm1_2 = vload(1.0); - P_2 = x2; - b = vload(*bl); - y2 = vmul(Pm1_2, b); - - x3 = vloadu(xarr + 3 * VLEN); - Pm1_3 = vload(1.0); - P_3 = x3; - b = vload(*bl); - y3 = vmul(Pm1_3, b); - - - b = vload(*(bl + 1)); - - vfmaeq(y0, P_0, b); - - vfmaeq(y1, P_1, b); - - vfmaeq(y2, P_2, b); - - vfmaeq(y3, P_3, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload(*(bl + l)); - R = vload(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul(x0, Pm1_0); - W2 = W1; - W2 = vsub(W2, Pm2_0); - P_0 = W1; - vfmaeq(P_0, W2, R); - vfmaeq(y0, P_0, b); - - Pm2_1 = Pm1_1; Pm1_1 = P_1; - W1 = vmul(x1, Pm1_1); - W2 = W1; - W2 = vsub(W2, Pm2_1); - P_1 = W1; - vfmaeq(P_1, W2, R); - vfmaeq(y1, P_1, b); - - Pm2_2 = Pm1_2; Pm1_2 = P_2; - W1 = vmul(x2, Pm1_2); - W2 = W1; - W2 = vsub(W2, Pm2_2); - P_2 = W1; - vfmaeq(P_2, W2, R); - vfmaeq(y2, P_2, b); - - Pm2_3 = Pm1_3; Pm1_3 = P_3; - W1 = vmul(x3, Pm1_3); - W2 = W1; - W2 = vsub(W2, Pm2_3); - P_3 = W1; - vfmaeq(P_3, W2, R); - vfmaeq(y3, P_3, b); - - - } - - vstoreu(out + 0 * VLEN, y0); - - vstoreu(out + 1 * VLEN, y1); - - vstoreu(out + 2 * VLEN, y2); - - vstoreu(out + 3 * VLEN, y3); - -} - -static void legendre_transform_vec5(double *recfacs, double *bl, ptrdiff_t lmax, - double xarr[(5) * VLEN], - double out[(5) * VLEN]) { - - Tv P_0, Pm1_0, Pm2_0, x0, y0; - - Tv P_1, Pm1_1, Pm2_1, x1, y1; - - Tv P_2, Pm1_2, Pm2_2, x2, y2; - - Tv P_3, Pm1_3, Pm2_3, x3, y3; - - Tv P_4, Pm1_4, Pm2_4, x4, y4; - - Tv W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu(xarr + 0 * VLEN); - Pm1_0 = vload(1.0); - P_0 = x0; - b = vload(*bl); - y0 = vmul(Pm1_0, b); - - x1 = vloadu(xarr + 1 * VLEN); - Pm1_1 = vload(1.0); - P_1 = x1; - b = vload(*bl); - y1 = vmul(Pm1_1, b); - - x2 = vloadu(xarr + 2 * VLEN); - Pm1_2 = vload(1.0); - P_2 = x2; - b = vload(*bl); - y2 = vmul(Pm1_2, b); - - x3 = vloadu(xarr + 3 * VLEN); - Pm1_3 = vload(1.0); - P_3 = x3; - b = vload(*bl); - y3 = vmul(Pm1_3, b); - - x4 = vloadu(xarr + 4 * VLEN); - Pm1_4 = vload(1.0); - P_4 = x4; - b = vload(*bl); - y4 = vmul(Pm1_4, b); - - - b = vload(*(bl + 1)); - - vfmaeq(y0, P_0, b); - - vfmaeq(y1, P_1, b); - - vfmaeq(y2, P_2, b); - - vfmaeq(y3, P_3, b); - - vfmaeq(y4, P_4, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload(*(bl + l)); - R = vload(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul(x0, Pm1_0); - W2 = W1; - W2 = vsub(W2, Pm2_0); - P_0 = W1; - vfmaeq(P_0, W2, R); - vfmaeq(y0, P_0, b); - - Pm2_1 = Pm1_1; Pm1_1 = P_1; - W1 = vmul(x1, Pm1_1); - W2 = W1; - W2 = vsub(W2, Pm2_1); - P_1 = W1; - vfmaeq(P_1, W2, R); - vfmaeq(y1, P_1, b); - - Pm2_2 = Pm1_2; Pm1_2 = P_2; - W1 = vmul(x2, Pm1_2); - W2 = W1; - W2 = vsub(W2, Pm2_2); - P_2 = W1; - vfmaeq(P_2, W2, R); - vfmaeq(y2, P_2, b); - - Pm2_3 = Pm1_3; Pm1_3 = P_3; - W1 = vmul(x3, Pm1_3); - W2 = W1; - W2 = vsub(W2, Pm2_3); - P_3 = W1; - vfmaeq(P_3, W2, R); - vfmaeq(y3, P_3, b); - - Pm2_4 = Pm1_4; Pm1_4 = P_4; - W1 = vmul(x4, Pm1_4); - W2 = W1; - W2 = vsub(W2, Pm2_4); - P_4 = W1; - vfmaeq(P_4, W2, R); - vfmaeq(y4, P_4, b); - - - } - - vstoreu(out + 0 * VLEN, y0); - - vstoreu(out + 1 * VLEN, y1); - - vstoreu(out + 2 * VLEN, y2); - - vstoreu(out + 3 * VLEN, y3); - - vstoreu(out + 4 * VLEN, y4); - -} - -static void legendre_transform_vec6(double *recfacs, double *bl, ptrdiff_t lmax, - double xarr[(6) * VLEN], - double out[(6) * VLEN]) { - - Tv P_0, Pm1_0, Pm2_0, x0, y0; - - Tv P_1, Pm1_1, Pm2_1, x1, y1; - - Tv P_2, Pm1_2, Pm2_2, x2, y2; - - Tv P_3, Pm1_3, Pm2_3, x3, y3; - - Tv P_4, Pm1_4, Pm2_4, x4, y4; - - Tv P_5, Pm1_5, Pm2_5, x5, y5; - - Tv W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu(xarr + 0 * VLEN); - Pm1_0 = vload(1.0); - P_0 = x0; - b = vload(*bl); - y0 = vmul(Pm1_0, b); - - x1 = vloadu(xarr + 1 * VLEN); - Pm1_1 = vload(1.0); - P_1 = x1; - b = vload(*bl); - y1 = vmul(Pm1_1, b); - - x2 = vloadu(xarr + 2 * VLEN); - Pm1_2 = vload(1.0); - P_2 = x2; - b = vload(*bl); - y2 = vmul(Pm1_2, b); - - x3 = vloadu(xarr + 3 * VLEN); - Pm1_3 = vload(1.0); - P_3 = x3; - b = vload(*bl); - y3 = vmul(Pm1_3, b); - - x4 = vloadu(xarr + 4 * VLEN); - Pm1_4 = vload(1.0); - P_4 = x4; - b = vload(*bl); - y4 = vmul(Pm1_4, b); - - x5 = vloadu(xarr + 5 * VLEN); - Pm1_5 = vload(1.0); - P_5 = x5; - b = vload(*bl); - y5 = vmul(Pm1_5, b); - - - b = vload(*(bl + 1)); - - vfmaeq(y0, P_0, b); - - vfmaeq(y1, P_1, b); - - vfmaeq(y2, P_2, b); - - vfmaeq(y3, P_3, b); - - vfmaeq(y4, P_4, b); - - vfmaeq(y5, P_5, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload(*(bl + l)); - R = vload(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul(x0, Pm1_0); - W2 = W1; - W2 = vsub(W2, Pm2_0); - P_0 = W1; - vfmaeq(P_0, W2, R); - vfmaeq(y0, P_0, b); - - Pm2_1 = Pm1_1; Pm1_1 = P_1; - W1 = vmul(x1, Pm1_1); - W2 = W1; - W2 = vsub(W2, Pm2_1); - P_1 = W1; - vfmaeq(P_1, W2, R); - vfmaeq(y1, P_1, b); - - Pm2_2 = Pm1_2; Pm1_2 = P_2; - W1 = vmul(x2, Pm1_2); - W2 = W1; - W2 = vsub(W2, Pm2_2); - P_2 = W1; - vfmaeq(P_2, W2, R); - vfmaeq(y2, P_2, b); - - Pm2_3 = Pm1_3; Pm1_3 = P_3; - W1 = vmul(x3, Pm1_3); - W2 = W1; - W2 = vsub(W2, Pm2_3); - P_3 = W1; - vfmaeq(P_3, W2, R); - vfmaeq(y3, P_3, b); - - Pm2_4 = Pm1_4; Pm1_4 = P_4; - W1 = vmul(x4, Pm1_4); - W2 = W1; - W2 = vsub(W2, Pm2_4); - P_4 = W1; - vfmaeq(P_4, W2, R); - vfmaeq(y4, P_4, b); - - Pm2_5 = Pm1_5; Pm1_5 = P_5; - W1 = vmul(x5, Pm1_5); - W2 = W1; - W2 = vsub(W2, Pm2_5); - P_5 = W1; - vfmaeq(P_5, W2, R); - vfmaeq(y5, P_5, b); - - - } - - vstoreu(out + 0 * VLEN, y0); - - vstoreu(out + 1 * VLEN, y1); - - vstoreu(out + 2 * VLEN, y2); - - vstoreu(out + 3 * VLEN, y3); - - vstoreu(out + 4 * VLEN, y4); - - vstoreu(out + 5 * VLEN, y5); - -} - - - -static void legendre_transform_vec1_s(float *recfacs, float *bl, ptrdiff_t lmax, - float xarr[(1) * VLEN_s], - float out[(1) * VLEN_s]) { - - Tv_s P_0, Pm1_0, Pm2_0, x0, y0; - - Tv_s W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu_s(xarr + 0 * VLEN_s); - Pm1_0 = vload_s(1.0); - P_0 = x0; - b = vload_s(*bl); - y0 = vmul_s(Pm1_0, b); - - - b = vload_s(*(bl + 1)); - - vfmaeq_s(y0, P_0, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload_s(*(bl + l)); - R = vload_s(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul_s(x0, Pm1_0); - W2 = W1; - W2 = vsub_s(W2, Pm2_0); - P_0 = W1; - vfmaeq_s(P_0, W2, R); - vfmaeq_s(y0, P_0, b); - - - } - - vstoreu_s(out + 0 * VLEN_s, y0); - -} - -static void legendre_transform_vec2_s(float *recfacs, float *bl, ptrdiff_t lmax, - float xarr[(2) * VLEN_s], - float out[(2) * VLEN_s]) { - - Tv_s P_0, Pm1_0, Pm2_0, x0, y0; - - Tv_s P_1, Pm1_1, Pm2_1, x1, y1; - - Tv_s W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu_s(xarr + 0 * VLEN_s); - Pm1_0 = vload_s(1.0); - P_0 = x0; - b = vload_s(*bl); - y0 = vmul_s(Pm1_0, b); - - x1 = vloadu_s(xarr + 1 * VLEN_s); - Pm1_1 = vload_s(1.0); - P_1 = x1; - b = vload_s(*bl); - y1 = vmul_s(Pm1_1, b); - - - b = vload_s(*(bl + 1)); - - vfmaeq_s(y0, P_0, b); - - vfmaeq_s(y1, P_1, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload_s(*(bl + l)); - R = vload_s(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul_s(x0, Pm1_0); - W2 = W1; - W2 = vsub_s(W2, Pm2_0); - P_0 = W1; - vfmaeq_s(P_0, W2, R); - vfmaeq_s(y0, P_0, b); - - Pm2_1 = Pm1_1; Pm1_1 = P_1; - W1 = vmul_s(x1, Pm1_1); - W2 = W1; - W2 = vsub_s(W2, Pm2_1); - P_1 = W1; - vfmaeq_s(P_1, W2, R); - vfmaeq_s(y1, P_1, b); - - - } - - vstoreu_s(out + 0 * VLEN_s, y0); - - vstoreu_s(out + 1 * VLEN_s, y1); - -} - -static void legendre_transform_vec3_s(float *recfacs, float *bl, ptrdiff_t lmax, - float xarr[(3) * VLEN_s], - float out[(3) * VLEN_s]) { - - Tv_s P_0, Pm1_0, Pm2_0, x0, y0; - - Tv_s P_1, Pm1_1, Pm2_1, x1, y1; - - Tv_s P_2, Pm1_2, Pm2_2, x2, y2; - - Tv_s W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu_s(xarr + 0 * VLEN_s); - Pm1_0 = vload_s(1.0); - P_0 = x0; - b = vload_s(*bl); - y0 = vmul_s(Pm1_0, b); - - x1 = vloadu_s(xarr + 1 * VLEN_s); - Pm1_1 = vload_s(1.0); - P_1 = x1; - b = vload_s(*bl); - y1 = vmul_s(Pm1_1, b); - - x2 = vloadu_s(xarr + 2 * VLEN_s); - Pm1_2 = vload_s(1.0); - P_2 = x2; - b = vload_s(*bl); - y2 = vmul_s(Pm1_2, b); - - - b = vload_s(*(bl + 1)); - - vfmaeq_s(y0, P_0, b); - - vfmaeq_s(y1, P_1, b); - - vfmaeq_s(y2, P_2, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload_s(*(bl + l)); - R = vload_s(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul_s(x0, Pm1_0); - W2 = W1; - W2 = vsub_s(W2, Pm2_0); - P_0 = W1; - vfmaeq_s(P_0, W2, R); - vfmaeq_s(y0, P_0, b); - - Pm2_1 = Pm1_1; Pm1_1 = P_1; - W1 = vmul_s(x1, Pm1_1); - W2 = W1; - W2 = vsub_s(W2, Pm2_1); - P_1 = W1; - vfmaeq_s(P_1, W2, R); - vfmaeq_s(y1, P_1, b); - - Pm2_2 = Pm1_2; Pm1_2 = P_2; - W1 = vmul_s(x2, Pm1_2); - W2 = W1; - W2 = vsub_s(W2, Pm2_2); - P_2 = W1; - vfmaeq_s(P_2, W2, R); - vfmaeq_s(y2, P_2, b); - - - } - - vstoreu_s(out + 0 * VLEN_s, y0); - - vstoreu_s(out + 1 * VLEN_s, y1); - - vstoreu_s(out + 2 * VLEN_s, y2); - -} - -static void legendre_transform_vec4_s(float *recfacs, float *bl, ptrdiff_t lmax, - float xarr[(4) * VLEN_s], - float out[(4) * VLEN_s]) { - - Tv_s P_0, Pm1_0, Pm2_0, x0, y0; - - Tv_s P_1, Pm1_1, Pm2_1, x1, y1; - - Tv_s P_2, Pm1_2, Pm2_2, x2, y2; - - Tv_s P_3, Pm1_3, Pm2_3, x3, y3; - - Tv_s W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu_s(xarr + 0 * VLEN_s); - Pm1_0 = vload_s(1.0); - P_0 = x0; - b = vload_s(*bl); - y0 = vmul_s(Pm1_0, b); - - x1 = vloadu_s(xarr + 1 * VLEN_s); - Pm1_1 = vload_s(1.0); - P_1 = x1; - b = vload_s(*bl); - y1 = vmul_s(Pm1_1, b); - - x2 = vloadu_s(xarr + 2 * VLEN_s); - Pm1_2 = vload_s(1.0); - P_2 = x2; - b = vload_s(*bl); - y2 = vmul_s(Pm1_2, b); - - x3 = vloadu_s(xarr + 3 * VLEN_s); - Pm1_3 = vload_s(1.0); - P_3 = x3; - b = vload_s(*bl); - y3 = vmul_s(Pm1_3, b); - - - b = vload_s(*(bl + 1)); - - vfmaeq_s(y0, P_0, b); - - vfmaeq_s(y1, P_1, b); - - vfmaeq_s(y2, P_2, b); - - vfmaeq_s(y3, P_3, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload_s(*(bl + l)); - R = vload_s(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul_s(x0, Pm1_0); - W2 = W1; - W2 = vsub_s(W2, Pm2_0); - P_0 = W1; - vfmaeq_s(P_0, W2, R); - vfmaeq_s(y0, P_0, b); - - Pm2_1 = Pm1_1; Pm1_1 = P_1; - W1 = vmul_s(x1, Pm1_1); - W2 = W1; - W2 = vsub_s(W2, Pm2_1); - P_1 = W1; - vfmaeq_s(P_1, W2, R); - vfmaeq_s(y1, P_1, b); - - Pm2_2 = Pm1_2; Pm1_2 = P_2; - W1 = vmul_s(x2, Pm1_2); - W2 = W1; - W2 = vsub_s(W2, Pm2_2); - P_2 = W1; - vfmaeq_s(P_2, W2, R); - vfmaeq_s(y2, P_2, b); - - Pm2_3 = Pm1_3; Pm1_3 = P_3; - W1 = vmul_s(x3, Pm1_3); - W2 = W1; - W2 = vsub_s(W2, Pm2_3); - P_3 = W1; - vfmaeq_s(P_3, W2, R); - vfmaeq_s(y3, P_3, b); - - - } - - vstoreu_s(out + 0 * VLEN_s, y0); - - vstoreu_s(out + 1 * VLEN_s, y1); - - vstoreu_s(out + 2 * VLEN_s, y2); - - vstoreu_s(out + 3 * VLEN_s, y3); - -} - -static void legendre_transform_vec5_s(float *recfacs, float *bl, ptrdiff_t lmax, - float xarr[(5) * VLEN_s], - float out[(5) * VLEN_s]) { - - Tv_s P_0, Pm1_0, Pm2_0, x0, y0; - - Tv_s P_1, Pm1_1, Pm2_1, x1, y1; - - Tv_s P_2, Pm1_2, Pm2_2, x2, y2; - - Tv_s P_3, Pm1_3, Pm2_3, x3, y3; - - Tv_s P_4, Pm1_4, Pm2_4, x4, y4; - - Tv_s W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu_s(xarr + 0 * VLEN_s); - Pm1_0 = vload_s(1.0); - P_0 = x0; - b = vload_s(*bl); - y0 = vmul_s(Pm1_0, b); - - x1 = vloadu_s(xarr + 1 * VLEN_s); - Pm1_1 = vload_s(1.0); - P_1 = x1; - b = vload_s(*bl); - y1 = vmul_s(Pm1_1, b); - - x2 = vloadu_s(xarr + 2 * VLEN_s); - Pm1_2 = vload_s(1.0); - P_2 = x2; - b = vload_s(*bl); - y2 = vmul_s(Pm1_2, b); - - x3 = vloadu_s(xarr + 3 * VLEN_s); - Pm1_3 = vload_s(1.0); - P_3 = x3; - b = vload_s(*bl); - y3 = vmul_s(Pm1_3, b); - - x4 = vloadu_s(xarr + 4 * VLEN_s); - Pm1_4 = vload_s(1.0); - P_4 = x4; - b = vload_s(*bl); - y4 = vmul_s(Pm1_4, b); - - - b = vload_s(*(bl + 1)); - - vfmaeq_s(y0, P_0, b); - - vfmaeq_s(y1, P_1, b); - - vfmaeq_s(y2, P_2, b); - - vfmaeq_s(y3, P_3, b); - - vfmaeq_s(y4, P_4, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload_s(*(bl + l)); - R = vload_s(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul_s(x0, Pm1_0); - W2 = W1; - W2 = vsub_s(W2, Pm2_0); - P_0 = W1; - vfmaeq_s(P_0, W2, R); - vfmaeq_s(y0, P_0, b); - - Pm2_1 = Pm1_1; Pm1_1 = P_1; - W1 = vmul_s(x1, Pm1_1); - W2 = W1; - W2 = vsub_s(W2, Pm2_1); - P_1 = W1; - vfmaeq_s(P_1, W2, R); - vfmaeq_s(y1, P_1, b); - - Pm2_2 = Pm1_2; Pm1_2 = P_2; - W1 = vmul_s(x2, Pm1_2); - W2 = W1; - W2 = vsub_s(W2, Pm2_2); - P_2 = W1; - vfmaeq_s(P_2, W2, R); - vfmaeq_s(y2, P_2, b); - - Pm2_3 = Pm1_3; Pm1_3 = P_3; - W1 = vmul_s(x3, Pm1_3); - W2 = W1; - W2 = vsub_s(W2, Pm2_3); - P_3 = W1; - vfmaeq_s(P_3, W2, R); - vfmaeq_s(y3, P_3, b); - - Pm2_4 = Pm1_4; Pm1_4 = P_4; - W1 = vmul_s(x4, Pm1_4); - W2 = W1; - W2 = vsub_s(W2, Pm2_4); - P_4 = W1; - vfmaeq_s(P_4, W2, R); - vfmaeq_s(y4, P_4, b); - - - } - - vstoreu_s(out + 0 * VLEN_s, y0); - - vstoreu_s(out + 1 * VLEN_s, y1); - - vstoreu_s(out + 2 * VLEN_s, y2); - - vstoreu_s(out + 3 * VLEN_s, y3); - - vstoreu_s(out + 4 * VLEN_s, y4); - -} - -static void legendre_transform_vec6_s(float *recfacs, float *bl, ptrdiff_t lmax, - float xarr[(6) * VLEN_s], - float out[(6) * VLEN_s]) { - - Tv_s P_0, Pm1_0, Pm2_0, x0, y0; - - Tv_s P_1, Pm1_1, Pm2_1, x1, y1; - - Tv_s P_2, Pm1_2, Pm2_2, x2, y2; - - Tv_s P_3, Pm1_3, Pm2_3, x3, y3; - - Tv_s P_4, Pm1_4, Pm2_4, x4, y4; - - Tv_s P_5, Pm1_5, Pm2_5, x5, y5; - - Tv_s W1, W2, b, R; - ptrdiff_t l; - - - x0 = vloadu_s(xarr + 0 * VLEN_s); - Pm1_0 = vload_s(1.0); - P_0 = x0; - b = vload_s(*bl); - y0 = vmul_s(Pm1_0, b); - - x1 = vloadu_s(xarr + 1 * VLEN_s); - Pm1_1 = vload_s(1.0); - P_1 = x1; - b = vload_s(*bl); - y1 = vmul_s(Pm1_1, b); - - x2 = vloadu_s(xarr + 2 * VLEN_s); - Pm1_2 = vload_s(1.0); - P_2 = x2; - b = vload_s(*bl); - y2 = vmul_s(Pm1_2, b); - - x3 = vloadu_s(xarr + 3 * VLEN_s); - Pm1_3 = vload_s(1.0); - P_3 = x3; - b = vload_s(*bl); - y3 = vmul_s(Pm1_3, b); - - x4 = vloadu_s(xarr + 4 * VLEN_s); - Pm1_4 = vload_s(1.0); - P_4 = x4; - b = vload_s(*bl); - y4 = vmul_s(Pm1_4, b); - - x5 = vloadu_s(xarr + 5 * VLEN_s); - Pm1_5 = vload_s(1.0); - P_5 = x5; - b = vload_s(*bl); - y5 = vmul_s(Pm1_5, b); - - - b = vload_s(*(bl + 1)); - - vfmaeq_s(y0, P_0, b); - - vfmaeq_s(y1, P_1, b); - - vfmaeq_s(y2, P_2, b); - - vfmaeq_s(y3, P_3, b); - - vfmaeq_s(y4, P_4, b); - - vfmaeq_s(y5, P_5, b); - - - for (l = 2; l <= lmax; ++l) { - b = vload_s(*(bl + l)); - R = vload_s(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - - Pm2_0 = Pm1_0; Pm1_0 = P_0; - W1 = vmul_s(x0, Pm1_0); - W2 = W1; - W2 = vsub_s(W2, Pm2_0); - P_0 = W1; - vfmaeq_s(P_0, W2, R); - vfmaeq_s(y0, P_0, b); - - Pm2_1 = Pm1_1; Pm1_1 = P_1; - W1 = vmul_s(x1, Pm1_1); - W2 = W1; - W2 = vsub_s(W2, Pm2_1); - P_1 = W1; - vfmaeq_s(P_1, W2, R); - vfmaeq_s(y1, P_1, b); - - Pm2_2 = Pm1_2; Pm1_2 = P_2; - W1 = vmul_s(x2, Pm1_2); - W2 = W1; - W2 = vsub_s(W2, Pm2_2); - P_2 = W1; - vfmaeq_s(P_2, W2, R); - vfmaeq_s(y2, P_2, b); - - Pm2_3 = Pm1_3; Pm1_3 = P_3; - W1 = vmul_s(x3, Pm1_3); - W2 = W1; - W2 = vsub_s(W2, Pm2_3); - P_3 = W1; - vfmaeq_s(P_3, W2, R); - vfmaeq_s(y3, P_3, b); - - Pm2_4 = Pm1_4; Pm1_4 = P_4; - W1 = vmul_s(x4, Pm1_4); - W2 = W1; - W2 = vsub_s(W2, Pm2_4); - P_4 = W1; - vfmaeq_s(P_4, W2, R); - vfmaeq_s(y4, P_4, b); - - Pm2_5 = Pm1_5; Pm1_5 = P_5; - W1 = vmul_s(x5, Pm1_5); - W2 = W1; - W2 = vsub_s(W2, Pm2_5); - P_5 = W1; - vfmaeq_s(P_5, W2, R); - vfmaeq_s(y5, P_5, b); - - - } - - vstoreu_s(out + 0 * VLEN_s, y0); - - vstoreu_s(out + 1 * VLEN_s, y1); - - vstoreu_s(out + 2 * VLEN_s, y2); - - vstoreu_s(out + 3 * VLEN_s, y3); - - vstoreu_s(out + 4 * VLEN_s, y4); - - vstoreu_s(out + 5 * VLEN_s, y5); - -} - - - - - -void sharp_legendre_transform_recfac(double *r, ptrdiff_t lmax) { - /* (l - 1) / l, for l >= 2 */ - ptrdiff_t l; - r[0] = 0; - r[1] = 1; - for (l = 2; l <= lmax; ++l) { - r[l] = (double)(l - 1) / (double)l; - } -} - -void sharp_legendre_transform_recfac_s(float *r, ptrdiff_t lmax) { - /* (l - 1) / l, for l >= 2 */ - ptrdiff_t l; - r[0] = 0; - r[1] = 1; - for (l = 2; l <= lmax; ++l) { - r[l] = (float)(l - 1) / (float)l; - } -} - - -/* - Compute sum_l b_l P_l(x_i) for all i. - */ - -#define LEN (SHARP_LEGENDRE_CS * VLEN) -#define LEN_s (SHARP_LEGENDRE_CS * VLEN_s) - - -void sharp_legendre_transform(double *bl, - double *recfac, - ptrdiff_t lmax, - double *x, double *out, ptrdiff_t nx) { - double xchunk[MAX_CS * VLEN], outchunk[MAX_CS * LEN]; - int compute_recfac; - ptrdiff_t i, j, len; - - compute_recfac = (recfac == NULL); - if (compute_recfac) { - recfac = malloc(sizeof(double) * (lmax + 1)); - sharp_legendre_transform_recfac(recfac, lmax); - } - - for (j = 0; j != LEN; ++j) xchunk[j] = 0; - - for (i = 0; i < nx; i += LEN) { - len = (i + (LEN) <= nx) ? (LEN) : (nx - i); - for (j = 0; j != len; ++j) xchunk[j] = x[i + j]; - switch ((len + VLEN - 1) / VLEN) { - case 6: legendre_transform_vec6(recfac, bl, lmax, xchunk, outchunk); break; - case 5: legendre_transform_vec5(recfac, bl, lmax, xchunk, outchunk); break; - case 4: legendre_transform_vec4(recfac, bl, lmax, xchunk, outchunk); break; - case 3: legendre_transform_vec3(recfac, bl, lmax, xchunk, outchunk); break; - case 2: legendre_transform_vec2(recfac, bl, lmax, xchunk, outchunk); break; - case 1: - case 0: - legendre_transform_vec1(recfac, bl, lmax, xchunk, outchunk); break; - } - for (j = 0; j != len; ++j) out[i + j] = outchunk[j]; - } - if (compute_recfac) { - free(recfac); - } -} - -void sharp_legendre_transform_s(float *bl, - float *recfac, - ptrdiff_t lmax, - float *x, float *out, ptrdiff_t nx) { - float xchunk[MAX_CS * VLEN_s], outchunk[MAX_CS * LEN_s]; - int compute_recfac; - ptrdiff_t i, j, len; - - compute_recfac = (recfac == NULL); - if (compute_recfac) { - recfac = malloc(sizeof(float) * (lmax + 1)); - sharp_legendre_transform_recfac_s(recfac, lmax); - } - - for (j = 0; j != LEN_s; ++j) xchunk[j] = 0; - - for (i = 0; i < nx; i += LEN_s) { - len = (i + (LEN_s) <= nx) ? (LEN_s) : (nx - i); - for (j = 0; j != len; ++j) xchunk[j] = x[i + j]; - switch ((len + VLEN_s - 1) / VLEN_s) { - case 6: legendre_transform_vec6_s(recfac, bl, lmax, xchunk, outchunk); break; - case 5: legendre_transform_vec5_s(recfac, bl, lmax, xchunk, outchunk); break; - case 4: legendre_transform_vec4_s(recfac, bl, lmax, xchunk, outchunk); break; - case 3: legendre_transform_vec3_s(recfac, bl, lmax, xchunk, outchunk); break; - case 2: legendre_transform_vec2_s(recfac, bl, lmax, xchunk, outchunk); break; - case 1: - case 0: - legendre_transform_vec1_s(recfac, bl, lmax, xchunk, outchunk); break; - } - for (j = 0; j != len; ++j) out[i + j] = outchunk[j]; - } - if (compute_recfac) { - free(recfac); - } -} - - -#endif \ No newline at end of file diff --git a/libsharp/sharp_legendre.c.in b/libsharp/sharp_legendre.c.in deleted file mode 100644 index cd65012..0000000 --- a/libsharp/sharp_legendre.c.in +++ /dev/null @@ -1,176 +0,0 @@ -/* - - NOTE NOTE NOTE - - This file is edited in sharp_legendre.c.in which is then preprocessed. - Do not make manual modifications to sharp_legendre.c. - - NOTE NOTE NOTE - -*/ - - -/* - * This file is part of libsharp. - * - * Redistribution and use in source and binary forms, with or without - * met: - * - * 1. Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * - * 2. Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution. - * - * 3. Neither the name of the copyright holder nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -/*! \file sharp_legendre.c.in - * - * Copyright (C) 2015 University of Oslo - * \author Dag Sverre Seljebotn - */ - -#ifndef NO_LEGENDRE -#if (VLEN==8) -#error This code is not tested with MIC; please compile with -DNO_LEGENDRE -/* ...or test it (it probably works) and remove this check */ -#endif - -#ifndef SHARP_LEGENDRE_CS -#define SHARP_LEGENDRE_CS 4 -#endif - -#define MAX_CS 6 -#if (SHARP_LEGENDRE_CS > MAX_CS) -#error (SHARP_LEGENDRE_CS > MAX_CS) -#endif - -#include "sharp_legendre.h" -#include "sharp_vecsupport.h" - -#include - -/*{ for scalar, T in [("double", ""), ("float", "_s")] }*/ -/*{ for cs in range(1, 7) }*/ -static void legendre_transform_vec{{cs}}{{T}}({{scalar}} *recfacs, {{scalar}} *bl, ptrdiff_t lmax, - {{scalar}} xarr[({{cs}}) * VLEN{{T}}], - {{scalar}} out[({{cs}}) * VLEN{{T}}]) { - /*{ for i in range(cs) }*/ - Tv{{T}} P_{{i}}, Pm1_{{i}}, Pm2_{{i}}, x{{i}}, y{{i}}; - /*{ endfor }*/ - Tv{{T}} W1, W2, b, R; - ptrdiff_t l; - - /*{ for i in range(cs) }*/ - x{{i}} = vloadu{{T}}(xarr + {{i}} * VLEN{{T}}); - Pm1_{{i}} = vload{{T}}(1.0); - P_{{i}} = x{{i}}; - b = vload{{T}}(*bl); - y{{i}} = vmul{{T}}(Pm1_{{i}}, b); - /*{ endfor }*/ - - b = vload{{T}}(*(bl + 1)); - /*{ for i in range(cs) }*/ - vfmaeq{{T}}(y{{i}}, P_{{i}}, b); - /*{ endfor }*/ - - for (l = 2; l <= lmax; ++l) { - b = vload{{T}}(*(bl + l)); - R = vload{{T}}(*(recfacs + l)); - - /* - P = x * Pm1 + recfacs[l] * (x * Pm1 - Pm2) - */ - /*{ for i in range(cs) }*/ - Pm2_{{i}} = Pm1_{{i}}; Pm1_{{i}} = P_{{i}}; - W1 = vmul{{T}}(x{{i}}, Pm1_{{i}}); - W2 = W1; - W2 = vsub{{T}}(W2, Pm2_{{i}}); - P_{{i}} = W1; - vfmaeq{{T}}(P_{{i}}, W2, R); - vfmaeq{{T}}(y{{i}}, P_{{i}}, b); - /*{ endfor }*/ - - } - /*{ for i in range(cs) }*/ - vstoreu{{T}}(out + {{i}} * VLEN{{T}}, y{{i}}); - /*{ endfor }*/ -} -/*{ endfor }*/ -/*{ endfor }*/ - - -/*{ for scalar, T in [("double", ""), ("float", "_s")] }*/ -void sharp_legendre_transform_recfac{{T}}({{scalar}} *r, ptrdiff_t lmax) { - /* (l - 1) / l, for l >= 2 */ - ptrdiff_t l; - r[0] = 0; - r[1] = 1; - for (l = 2; l <= lmax; ++l) { - r[l] = ({{scalar}})(l - 1) / ({{scalar}})l; - } -} -/*{ endfor }*/ - -/* - Compute sum_l b_l P_l(x_i) for all i. - */ - -#define LEN (SHARP_LEGENDRE_CS * VLEN) -#define LEN_s (SHARP_LEGENDRE_CS * VLEN_s) - -/*{ for scalar, T in [("double", ""), ("float", "_s")] }*/ -void sharp_legendre_transform{{T}}({{scalar}} *bl, - {{scalar}} *recfac, - ptrdiff_t lmax, - {{scalar}} *x, {{scalar}} *out, ptrdiff_t nx) { - {{scalar}} xchunk[MAX_CS * VLEN{{T}}], outchunk[MAX_CS * LEN{{T}}]; - int compute_recfac; - ptrdiff_t i, j, len; - - compute_recfac = (recfac == NULL); - if (compute_recfac) { - recfac = malloc(sizeof({{scalar}}) * (lmax + 1)); - sharp_legendre_transform_recfac{{T}}(recfac, lmax); - } - - for (j = 0; j != LEN{{T}}; ++j) xchunk[j] = 0; - - for (i = 0; i < nx; i += LEN{{T}}) { - len = (i + (LEN{{T}}) <= nx) ? (LEN{{T}}) : (nx - i); - for (j = 0; j != len; ++j) xchunk[j] = x[i + j]; - switch ((len + VLEN{{T}} - 1) / VLEN{{T}}) { - case 6: legendre_transform_vec6{{T}}(recfac, bl, lmax, xchunk, outchunk); break; - case 5: legendre_transform_vec5{{T}}(recfac, bl, lmax, xchunk, outchunk); break; - case 4: legendre_transform_vec4{{T}}(recfac, bl, lmax, xchunk, outchunk); break; - case 3: legendre_transform_vec3{{T}}(recfac, bl, lmax, xchunk, outchunk); break; - case 2: legendre_transform_vec2{{T}}(recfac, bl, lmax, xchunk, outchunk); break; - case 1: - case 0: - legendre_transform_vec1{{T}}(recfac, bl, lmax, xchunk, outchunk); break; - } - for (j = 0; j != len; ++j) out[i + j] = outchunk[j]; - } - if (compute_recfac) { - free(recfac); - } -} -/*{ endfor }*/ - -#endif diff --git a/libsharp/sharp_legendre.h b/libsharp/sharp_legendre.h deleted file mode 100644 index cfd8aee..0000000 --- a/libsharp/sharp_legendre.h +++ /dev/null @@ -1,62 +0,0 @@ -/* - * This file is part of libsharp. - * - * Redistribution and use in source and binary forms, with or without - * met: - * - * 1. Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * - * 2. Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution. - * - * 3. Neither the name of the copyright holder nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -/*! \file sharp_legendre.h - * Interface for the Legendre transform parts of the spherical transform library. - * - * Copyright (C) 2015 University of Oslo - * \author Dag Sverre Seljebotn - */ - -#ifndef SHARP_LEGENDRE_H -#define SHARP_LEGENDRE_H - -#include - -#ifdef __cplusplus -extern "C" { -#endif - -#ifndef NO_LEGENDRE - -void sharp_legendre_transform(double *bl, double *recfac, ptrdiff_t lmax, double *x, - double *out, ptrdiff_t nx); -void sharp_legendre_transform_s(float *bl, float *recfac, ptrdiff_t lmax, float *x, - float *out, ptrdiff_t nx); -void sharp_legendre_transform_recfac(double *r, ptrdiff_t lmax); -void sharp_legendre_transform_recfac_s(float *r, ptrdiff_t lmax); - -#endif - -#ifdef __cplusplus -} -#endif - -#endif diff --git a/libsharp/sharp_legendre_roots.h b/libsharp/sharp_legendre_roots.h index 2a056b2..19f8ec8 100644 --- a/libsharp/sharp_legendre_roots.h +++ b/libsharp/sharp_legendre_roots.h @@ -24,7 +24,7 @@ /*! \file sharp_legendre_roots.h * - * Copyright (C) 2006-2012 Max-Planck-Society + * Copyright (C) 2006-2019 Max-Planck-Society * \author Martin Reinecke */ diff --git a/libsharp/sharp_legendre_table.c b/libsharp/sharp_legendre_table.c deleted file mode 100644 index 7fef085..0000000 --- a/libsharp/sharp_legendre_table.c +++ /dev/null @@ -1,1467 +0,0 @@ -/* - -This file originated as a concatenation of files from libpsht. Further refactoring -could be carried out to make the code use libsharp conventions instead for SSE etc.; - -*/ - - -/* -sse_utils.h -*/ -/* - * This file is part of libc_utils. - * - * libc_utils is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libc_utils is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libc_utils; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libc_utils is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/*! \file sse_utils.h - * SSE/SSE2/SSE3-related functionality - * - * Copyright (C) 2010,2011 Max-Planck-Society - * \author Martin Reinecke - */ - - -#if (defined(__SSE__)) - -#include - - -typedef __m128 v4sf; /* vector of 4 floats (SSE1) */ - -typedef union { - float f[4]; - v4sf v; -} V4SF; - -static inline v4sf build_v4sf (float a, float b, float c, float d) - { return _mm_set_ps(d,c,b,a); } -static inline void read_v4sf (v4sf v, float *a, float *b, float *c, float *d) - { - V4SF tmp; - tmp.v = v; - if (a) *a=tmp.f[0]; - if (b) *b=tmp.f[1]; - if (c) *c=tmp.f[2]; - if (d) *d=tmp.f[3]; - } - - -#endif - -#if (defined(__SSE2__)) - -#include - - -typedef __m128d v2df; /* vector of 2 doubles (SSE2) */ - -typedef union { - double d[2]; - v2df v; -} V2DF; - -typedef struct { - v2df a,b; -} v2df2; -typedef struct { - V2DF a,b; -} V2DF2; - -#define V2DF_SIGNMASK _mm_set1_pd(-0.0) - -static inline v2df build_v2df (double a, double b) - { return _mm_set_pd(b,a); } -static inline void read_v2df (v2df v, double *a, double *b) - { _mm_store_sd(a,v); _mm_storeh_pd(b,v); } - -static inline int v2df_any_gt (v2df a, v2df b) - { - return (_mm_movemask_pd(_mm_cmpgt_pd(_mm_andnot_pd(V2DF_SIGNMASK,a),b))!=0); - } -static inline int v2df_all_ge (v2df a, v2df b) - { - return (_mm_movemask_pd(_mm_cmplt_pd(_mm_andnot_pd(V2DF_SIGNMASK,a),b))==0); - } -static inline V2DF to_V2DF (v2df x) - { V2DF X; X.v=x; return X; } -static inline V2DF2 to_V2DF2 (v2df2 x) - { V2DF2 X; X.a.v=x.a; X.b.v=x.b; return X; } -static inline v2df2 to_v2df2 (V2DF2 X) - { v2df2 x; x.a=X.a.v; x.b=X.b.v; return x; } -static inline v2df2 zero_v2df2(void) - { v2df2 x; x.a=x.b=_mm_setzero_pd(); return x; } - - -#endif - -#if (defined(__SSE3__)) - -#include - -#endif - - - -/* -ylmgen_c.h -*/ -/* - * This file is part of libpsht. - * - * libpsht is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libpsht is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libpsht; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libpsht is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/*! \file ylmgen_c.h - * Code for efficient calculation of Y_lm(phi=0,theta) - * - * Copyright (C) 2005-2011 Max-Planck-Society - * \author Martin Reinecke - */ - -typedef double ylmgen_dbl2[2]; -typedef double ylmgen_dbl3[3]; - -typedef struct - { - double cth_crit; - int mdist_crit; - /* members depending on m and m' */ - int s, m, mlo, mhi, cosPow, sinPow; - long double prefactor; - ylmgen_dbl3 *fx; - int preMinus_p, preMinus_m; - } sylmgen_d; - -typedef struct - { - double fsmall, fbig, eps, cth_crit; - int lmax, mmax, smax, m_cur, ith, nth, m_crit, spinrec; - /*! The index of the first non-negligible Y_lm value. */ - int *firstl; - double *cf, *mfac, *t1fac, *t2fac, *th, *cth, *sth, *logsth; - ylmgen_dbl2 *recfac; - double *lamfact; - /*! Points to an array of size [0..lmax] containing the Y_lm values. */ - double *ylm; - /*! Points to an array of size [0..lmax] containing the lambda_w - and lambda_x values for spin>0 transforms. */ - ylmgen_dbl2 **lambda_wx; - long double *logsum, *lc05, *ls05; - double *flm1, *flm2, *xl; - - sylmgen_d **sylm; - - int *lwx_uptodate; - int ylm_uptodate; - -#ifdef __SSE2__ - int ith1, ith2; - /*! Points to an array of size [0..lmax] containing the Y_lm values. */ - v2df *ylm_sse2; - /*! Points to an array of size [0..lmax] containing the lambda_w - and lambda_x values for spin>0 transforms. */ - v2df2 **lambda_wx_sse2; - int *lwx_uptodate_sse2; - int ylm_uptodate_sse2; -#endif - - int recfac_uptodate, lamfact_uptodate; - } Ylmgen_C; - -/*! Creates a generator which will calculate Y_lm(theta,phi=0) - up to \a l=l_max and \a m=m_max. It may regard Y_lm whose absolute - magnitude is smaller than \a epsilon as zero. If \a spinrec is nonzero, - the spin-1 and spin-2 Y_lm will be calculated by recursion from the spin-0 - ones, otherwise Wigner d matrix elements will be used. */ -static void Ylmgen_init (Ylmgen_C *gen, int l_max, int m_max, int s_max, int spinrec, - double epsilon); - -/*! Passes am array \a theta of \a nth colatitudes that will be used in - subsequent calls. The individual angles will be referenced by their - index in the array, starting with 0. - \note The array can be freed or reused after the call. */ -static void Ylmgen_set_theta (Ylmgen_C *gen, const double *theta, int nth); - -/*! Deallocates a generator previously initialised by Ylmgen_init(). */ -static void Ylmgen_destroy (Ylmgen_C *gen); - -/*! Prepares the object for the calculation at \a theta and \a m. */ -static void Ylmgen_prepare (Ylmgen_C *gen, int ith, int m); - -/*! Recalculates (if necessary) the Y_lm values. */ -static void Ylmgen_recalc_Ylm (Ylmgen_C *gen); -/*! Recalculates (if necessary) the lambda_w and lambda_x values for spin >0 - transforms. */ -/*static void Ylmgen_recalc_lambda_wx (Ylmgen_C *gen, int spin);*/ - -#ifdef __SSE2__ -/*! Prepares the object for the calculation at \a theta, \a theta2 and \a m. */ -static void Ylmgen_prepare_sse2 (Ylmgen_C *gen, int ith1, int ith2, int m); - -/*! Recalculates (if necessary) the Y_lm values. */ -static void Ylmgen_recalc_Ylm_sse2 (Ylmgen_C *gen); -/*! Recalculates (if necessary) the lambda_w and lambda_x values for spin >0 - transforms. */ -/*static void Ylmgen_recalc_lambda_wx_sse2 (Ylmgen_C *gen, int spin);*/ -#endif - -/*! Returns a pointer to an array with lmax+1 entries containing normalisation - factors that must be applied to Y_lm values computed for \a spin with the - given \a spinrec flag. The array must be deallocated (using free()) by the - user. */ -/*static double *Ylmgen_get_norm (int lmax, int spin, int spinrec);*/ - - -/* -ylmgen_c.c -*/ - -/* - * This file is part of libpsht. - * - * libpsht is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libpsht is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libpsht; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libpsht is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/* - * Code for efficient calculation of Y_lm(theta,phi=0) - * - * Copyright (C) 2005-2011 Max-Planck-Society - * Author: Martin Reinecke - */ - -#include -#include -#include "c_utils.h" - -enum { large_exponent2=90, minscale=-4, maxscale=11 }; - -/*static void sylmgen_init (sylmgen_d *gen, const Ylmgen_C *ygen, int spin) - { - int i; - UTIL_ASSERT(spin>=1,"incorrect spin"); - gen->s=spin; - gen->m=gen->mlo=gen->mhi=-1234567890; - ALLOC(gen->fx,ylmgen_dbl3,ygen->lmax+2); - - for (i=0; ilmax+2; ++i) - gen->fx[i][0]=gen->fx[i][1]=gen->fx[i][2]=0.; - - gen->cth_crit = 2.; - gen->mdist_crit = ygen->lmax+1; - }*/ - -static void sylmgen_destroy (sylmgen_d *gen) - { DEALLOC(gen->fx); } - -/*static void sylmgen_prepare (sylmgen_d *gen, const Ylmgen_C *ygen, int m_) - { - int mlo_, mhi_, ms_similar, l; - - if (m_==gen->m) return; - UTIL_ASSERT(m_>=0,"incorrect m"); - - mlo_=m_; mhi_=gen->s; - if (mhi_mhi==mhi_) && (gen->mlo==mlo_)); - - gen->m=m_; - gen->mlo = mlo_; gen->mhi = mhi_; - - if (!ms_similar) - { - for (l=gen->mhi; llmax; ++l) - { - double t = ygen->flm1[l+gen->m]*ygen->flm1[l-gen->m] - *ygen->flm1[l+gen->s]*ygen->flm1[l-gen->s]; - double lt = 2*l+1; - double l1 = l+1; - gen->fx[l+1][0]=l1*lt*t; - gen->fx[l+1][1]=gen->m*gen->s*ygen->xl[l]*ygen->xl[l+1]; - t = ygen->flm2[l+gen->m]*ygen->flm2[l-gen->m] - *ygen->flm2[l+gen->s]*ygen->flm2[l-gen->s]; - gen->fx[l+1][2]=t*l1*ygen->xl[l]; - } - gen->prefactor = 0.5L*(ygen->logsum[2*gen->mhi] - -ygen->logsum[gen->mhi+gen->mlo]-ygen->logsum[gen->mhi-gen->mlo]); - } - - gen->preMinus_p = gen->preMinus_m = 0; - if (gen->mhi==gen->m) - { - gen->cosPow = gen->mhi+gen->s; gen->sinPow = gen->mhi-gen->s; - gen->preMinus_p = gen->preMinus_m = ((gen->mhi-gen->s)&1); - } - else - { - gen->cosPow = gen->mhi+gen->m; gen->sinPow = gen->mhi-gen->m; - gen->preMinus_m = ((gen->mhi+gen->m)&1); - } - }*/ - -/*static void sylmgen_recalc (sylmgen_d *gen, const Ylmgen_C *ygen, int ith, - ylmgen_dbl2 *res, int *firstl) - { - const double ln2 = 0.6931471805599453094172321214581766; - const double inv_ln2 = 1.4426950408889634073599246810018921; - int l=gen->mhi; - int lmax = ygen->lmax; - ylmgen_dbl3 *fy = gen->fx; - const double fsmall = ygen->fsmall, fbig = ygen->fbig, eps = ygen->eps; - const double cth = ygen->cth[ith]; - long double logvalp = inv_ln2*(gen->prefactor - + ygen->lc05[ith]*gen->cosPow + ygen->ls05[ith]*gen->sinPow); - long double logvalm = inv_ln2*(gen->prefactor - + ygen->lc05[ith]*gen->sinPow + ygen->ls05[ith]*gen->cosPow); - int scalep = (int)(logvalp/large_exponent2)-minscale; - int scalem = (int)(logvalm/large_exponent2)-minscale; - double rec1p=0., rec1m=0.; - double rec2p = exp(ln2*(double)(logvalp-(scalep+minscale)*large_exponent2)); - double rec2m = exp(ln2*(double)(logvalm-(scalem+minscale)*large_exponent2)); - double corfacp,corfacm; - double tp,tm; - - if ((abs(gen->m-gen->s)>=gen->mdist_crit)&&(fabs(cth)>=gen->cth_crit)) - { *firstl=ygen->lmax+1; return; } - - if (gen->preMinus_p) - rec2p=-rec2p; - if (gen->preMinus_m) - rec2m=-rec2m; - if (gen->s&1) - rec2p=-rec2p; - - / iterate until we reach the realm of IEEE numbers / - while((scalem<0)&&(scalep<0)) - { - if (++l>lmax) break; - rec1p = (cth - fy[l][1])*fy[l][0]*rec2p - fy[l][2]*rec1p; - rec1m = (cth + fy[l][1])*fy[l][0]*rec2m - fy[l][2]*rec1m; - if (++l>lmax) break; - rec2p = (cth - fy[l][1])*fy[l][0]*rec1p - fy[l][2]*rec2p; - rec2m = (cth + fy[l][1])*fy[l][0]*rec1m - fy[l][2]*rec2m; - - while (fabs(rec2p)>fbig) - { rec1p *= fsmall; rec2p *= fsmall; ++scalep; } - while (fabs(rec2m)>fbig) - { rec1m *= fsmall; rec2m *= fsmall; ++scalem; } - } - - corfacp = (scalep<0) ? 0. : ygen->cf[scalep]; - corfacm = (scalem<0) ? 0. : ygen->cf[scalem]; - - if (l<=lmax) - { - while (1) - { - if ((fabs(rec2p*corfacp)>eps) || (fabs(rec2m*corfacm)>eps)) - break; - if (++l>lmax) break; - rec1p = (cth - fy[l][1])*fy[l][0]*rec2p - fy[l][2]*rec1p; - rec1m = (cth + fy[l][1])*fy[l][0]*rec2m - fy[l][2]*rec1m; - if ((fabs(rec1p*corfacp)>eps) || (fabs(rec1m*corfacm)>eps)) - { SWAP(rec1p,rec2p,double); SWAP(rec1m,rec2m,double); break; } - if (++l>lmax) break; - rec2p = (cth - fy[l][1])*fy[l][0]*rec1p - fy[l][2]*rec2p; - rec2m = (cth + fy[l][1])*fy[l][0]*rec1m - fy[l][2]*rec2m; - - if ((fabs(rec2p)>fbig) || (fabs(rec2m)>fbig)) - { - while (fabs(rec2p)>fbig) - { rec1p *= fsmall; rec2p *= fsmall; ++scalep; } - while (fabs(rec2m)>fbig) - { rec1m *= fsmall; rec2m *= fsmall; ++scalem; } - corfacp = (scalep<0) ? 0. : ygen->cf[scalep]; - corfacm = (scalem<0) ? 0. : ygen->cf[scalem]; - } - } - } - - *firstl=l; - if (l>lmax) - { - gen->mdist_crit=abs(gen->m-gen->s); - gen->cth_crit= fabs(cth); - return; - } - - tp = rec2p*corfacp; tm = rec2m*corfacm; - res[l][0]=tp+tm; - res[l][1]=tm-tp; - - while (1) - { - if ((fabs(tp)>eps) && (fabs(tm)>eps)) - break; - if (++l>lmax) break; - rec1p = (cth - fy[l][1])*fy[l][0]*rec2p - fy[l][2]*rec1p; - rec1m = (cth + fy[l][1])*fy[l][0]*rec2m - fy[l][2]*rec1m; - tp=rec1p*corfacp; tm=rec1m*corfacm; - res[l][0]=tp+tm; res[l][1]=tm-tp; - if ((fabs(tp)>eps) && (fabs(tm)>eps)) - { SWAP(rec1p,rec2p,double); SWAP(rec1m,rec2m,double); break; } - if (++l>lmax) break; - rec2p = (cth - fy[l][1])*fy[l][0]*rec1p - fy[l][2]*rec2p; - rec2m = (cth + fy[l][1])*fy[l][0]*rec1m - fy[l][2]*rec2m; - tp=rec2p*corfacp; tm=rec2m*corfacm; - res[l][0]=tp+tm; res[l][1]=tm-tp; - - if ((fabs(rec2p)>fbig) || (fabs(rec2m)>fbig)) - { - while (fabs(rec2p)>fbig) - { rec1p *= fsmall; rec2p *= fsmall; ++scalep; } - while (fabs(rec2m)>fbig) - { rec1m *= fsmall; rec2m *= fsmall; ++scalem; } - corfacp = (scalep<0) ? 0. : ygen->cf[scalep]; - corfacm = (scalem<0) ? 0. : ygen->cf[scalem]; - } - } - - rec1p *= corfacp; rec2p *= corfacp; - rec1m *= corfacm; rec2m *= corfacm; - - for (;llmax) break; - rec1p = (cth - fy[l][1])*fy[l][0]*rec2p - fy[l][2]*rec1p; - rec1m = (cth + fy[l][1])*fy[l][0]*rec2m - fy[l][2]*rec1m; - res[l][0] = rec1p+rec1m; res[l][1] = rec1m-rec1p; - if (++l>lmax) break; - rec2p = (cth - fy[l][1])*fy[l][0]*rec1p - fy[l][2]*rec2p; - rec2m = (cth + fy[l][1])*fy[l][0]*rec1m - fy[l][2]*rec2m; - res[l][0] = rec2p+rec2m; res[l][1] = rec2m-rec2p; - } - }*/ - -#ifdef __SSE2__ - -#define ADVANCE(L,ap,am,bp,bm) \ - { \ - v2df f0=_mm_set1_pd(fy[L][0]), \ - f1=_mm_set1_pd(fy[L][1]), \ - f2=_mm_set1_pd(fy[L][2]); \ - ap = _mm_sub_pd( \ - _mm_mul_pd(_mm_sub_pd(cth,f1),_mm_mul_pd(f0,bp)), \ - _mm_mul_pd(f2,ap)); \ - am = _mm_sub_pd( \ - _mm_mul_pd(_mm_add_pd(cth,f1),_mm_mul_pd(f0,bm)), \ - _mm_mul_pd(f2,am)); \ - } - -#define RENORMX(r1,r2,corf,sca,scb) \ - do \ - { \ - double rec1a, rec1b, rec2a, rec2b, corfaca, corfacb; \ - read_v2df (r1, &rec1a, &rec1b); read_v2df (r2, &rec2a, &rec2b); \ - read_v2df (corf, &corfaca, &corfacb); \ - while (fabs(rec2a)>fbig) \ - { \ - rec1a*=fsmall; rec2a*=fsmall; ++sca; \ - corfaca = (sca<0) ? 0. : ygen->cf[sca]; \ - } \ - while (fabs(rec2b)>fbig) \ - { \ - rec1b*=fsmall; rec2b*=fsmall; ++scb; \ - corfacb = (scb<0) ? 0. : ygen->cf[scb]; \ - } \ - r1=build_v2df(rec1a,rec1b); r2=build_v2df(rec2a,rec2b); \ - corf=build_v2df(corfaca,corfacb); \ - } \ - while(0) - -#define RENORM \ - RENORMX(rec1p,rec2p,corfacp,scale1p,scale2p); \ - RENORMX(rec1m,rec2m,corfacm,scale1m,scale2m) - -/*static void sylmgen_recalc_sse2 (sylmgen_d *gen, const Ylmgen_C *ygen, - int ith1, int ith2, v2df2 *res, int *firstl) - { - const double ln2 = 0.6931471805599453094172321214581766; - const double inv_ln2 = 1.4426950408889634073599246810018921; - int l=gen->mhi; - int lmax = ygen->lmax; - ylmgen_dbl3 *fy = gen->fx; - const double fbig=ygen->fbig, fsmall=ygen->fsmall; - v2df eps2 = build_v2df(ygen->eps,ygen->eps); - const double cth1=ygen->cth[ith1], cth2=ygen->cth[ith2]; - v2df cth = build_v2df(cth1,cth2); - long double - logval1p = inv_ln2*(gen->prefactor - + ygen->lc05[ith1]*gen->cosPow + ygen->ls05[ith1]*gen->sinPow), - logval2p = inv_ln2*(gen->prefactor - + ygen->lc05[ith2]*gen->cosPow + ygen->ls05[ith2]*gen->sinPow), - logval1m = inv_ln2*(gen->prefactor - + ygen->lc05[ith1]*gen->sinPow + ygen->ls05[ith1]*gen->cosPow), - logval2m = inv_ln2*(gen->prefactor - + ygen->lc05[ith2]*gen->sinPow + ygen->ls05[ith2]*gen->cosPow); - - int scale1p = (int)(logval1p/large_exponent2)-minscale, - scale2p = (int)(logval2p/large_exponent2)-minscale, - scale1m = (int)(logval1m/large_exponent2)-minscale, - scale2m = (int)(logval2m/large_exponent2)-minscale; - - v2df rec1p =_mm_setzero_pd(), rec1m=_mm_setzero_pd(); - v2df rec2p = build_v2df( - exp(ln2*(double)(logval1p-(scale1p+minscale)*large_exponent2)), - exp(ln2*(double)(logval2p-(scale2p+minscale)*large_exponent2))); - v2df rec2m = build_v2df( - exp(ln2*(double)(logval1m-(scale1m+minscale)*large_exponent2)), - exp(ln2*(double)(logval2m-(scale2m+minscale)*large_exponent2))); - v2df corfacp=build_v2df((scale1p<0) ? 0. : ygen->cf[scale1p], - (scale2p<0) ? 0. : ygen->cf[scale2p]), - corfacm=build_v2df((scale1m<0) ? 0. : ygen->cf[scale1m], - (scale2m<0) ? 0. : ygen->cf[scale2m]); - v2df tp,tm; - - if ((abs(gen->m-gen->s)>=gen->mdist_crit) - &&(fabs(ygen->cth[ith1])>=gen->cth_crit) - &&(fabs(ygen->cth[ith2])>=gen->cth_crit)) - { *firstl=ygen->lmax+1; return; } - - if (gen->preMinus_p) - rec2p = _mm_xor_pd (rec2p,V2DF_SIGNMASK); / negate / - if (gen->preMinus_m) - rec2m = _mm_xor_pd (rec2m,V2DF_SIGNMASK); / negate / - if (gen->s&1) - rec2p = _mm_xor_pd (rec2p,V2DF_SIGNMASK); / negate / - - / iterate until we reach the realm of IEEE numbers / - while((scale1m<0)&&(scale1p<0)&&(scale2m<0)&&(scale2p<0)) - { - if (++l>lmax) break; - ADVANCE (l,rec1p,rec1m,rec2p,rec2m) - if (++l>lmax) break; - ADVANCE (l,rec2p,rec2m,rec1p,rec1m) - - RENORM; - } - - if (l<=lmax) - { - while (1) - { - if (v2df_any_gt(_mm_mul_pd(rec2p,corfacp),eps2) || - v2df_any_gt(_mm_mul_pd(rec2m,corfacm),eps2)) - break; - if (++l>lmax) break; - ADVANCE (l,rec1p,rec1m,rec2p,rec2m) - if (v2df_any_gt(_mm_mul_pd(rec1p,corfacp),eps2) || - v2df_any_gt(_mm_mul_pd(rec1m,corfacm),eps2)) - { SWAP(rec1p,rec2p,v2df); SWAP(rec1m,rec2m,v2df); break; } - if (++l>lmax) break; - ADVANCE (l,rec2p,rec2m,rec1p,rec1m) - - RENORM; - } - } - - *firstl=l; - if (l>lmax) - { - gen->mdist_crit=abs(gen->m-gen->s); - gen->cth_crit= (fabs(cth1)lmax) break; - ADVANCE(l,rec1p,rec1m,rec2p,rec2m) - tp=_mm_mul_pd(rec1p,corfacp); tm=_mm_mul_pd(rec1m,corfacm); - res[l].a=_mm_add_pd(tp,tm); res[l].b=_mm_sub_pd(tm,tp); - if (v2df_all_ge(tp,eps2) && v2df_all_ge(tm,eps2)) - { SWAP(rec1p,rec2p,v2df); SWAP(rec1m,rec2m,v2df); break; } - if (++l>lmax) break; - ADVANCE (l,rec2p,rec2m,rec1p,rec1m) - tp=_mm_mul_pd(rec2p,corfacp); tm=_mm_mul_pd(rec2m,corfacm); - res[l].a=_mm_add_pd(tp,tm); res[l].b=_mm_sub_pd(tm,tp); - - RENORM; - } - - rec1p = _mm_mul_pd(rec1p,corfacp); rec2p = _mm_mul_pd(rec2p,corfacp); - rec1m = _mm_mul_pd(rec1m,corfacm); rec2m = _mm_mul_pd(rec2m,corfacm); - - for (;llmax) break; - ADVANCE(l,rec1p,rec1m,rec2p,rec2m) - res[l].a=_mm_add_pd(rec1p,rec1m); res[l].b=_mm_sub_pd(rec1m,rec1p); - if (++l>lmax) break; - ADVANCE(l,rec2p,rec2m,rec1p,rec1m) - res[l].a=_mm_add_pd(rec2p,rec2m); res[l].b=_mm_sub_pd(rec2m,rec2p); - } - } -*/ -#endif - -static void Ylmgen_init (Ylmgen_C *gen, int l_max, int m_max, int s_max, int spinrec, - double epsilon) - { - int m; - const double inv_sqrt4pi = 0.2820947917738781434740397257803862929220; - const double inv_ln2 = 1.4426950408889634073599246810018921; - - gen->fsmall = ldexp(1.,-large_exponent2); - gen->fbig = ldexp(1., large_exponent2); - gen->eps = epsilon; - gen->cth_crit = 2.; - gen->ith = -1; - gen->nth = 0; - gen->lmax = l_max; - gen->mmax = m_max; - gen->smax = s_max; - gen->spinrec = spinrec; - gen->m_cur = -1; - gen->m_crit = gen->mmax+1; - gen->firstl = RALLOC(int,gen->smax+1); - for (m=0; m<=gen->smax; ++m) gen->firstl[m]=-1; - gen->cf = RALLOC(double,maxscale-minscale+1); - for (m=0; m<(maxscale-minscale+1); ++m) - gen->cf[m] = ldexp(1.,(m+minscale)*large_exponent2); - gen->recfac = RALLOC(ylmgen_dbl2,gen->lmax+1); - gen->mfac = RALLOC(double,gen->mmax+1); - gen->mfac[0] = 1; - for (m=1; m<=gen->mmax; ++m) - gen->mfac[m] = gen->mfac[m-1]*sqrt((2*m+1.)/(2*m)); - for (m=0; m<=gen->mmax; ++m) - gen->mfac[m] = inv_ln2*log(inv_sqrt4pi*gen->mfac[m]); - - gen->t1fac = RALLOC(double,gen->lmax+1); - for (m=0; m<=gen->lmax; ++m) - gen->t1fac[m] = sqrt(4.*(m+1)*(m+1)-1.); - gen->t2fac = RALLOC(double,gen->lmax+gen->mmax+1); - for (m=0; m<=gen->lmax+gen->mmax; ++m) - gen->t2fac[m] = 1./sqrt(m+1.); - - gen->lamfact = RALLOC(double,gen->lmax+1); - gen->ylm = RALLOC(double,gen->lmax+1); - ALLOC(gen->lambda_wx,ylmgen_dbl2 *,gen->smax+1); - for (m=0; m<=gen->smax; ++m) - gen->lambda_wx[m]=NULL; - - gen->sylm = RALLOC(sylmgen_d *,gen->smax+1); - for (m=0; m<=gen->smax; ++m) - gen->sylm[m]=NULL; - - gen->ylm_uptodate = 0; - gen->lwx_uptodate = RALLOC(int,gen->smax+1); - SET_ARRAY(gen->lwx_uptodate,0,gen->smax+1,0); - gen->recfac_uptodate = 0; - gen->lamfact_uptodate = 0; - - gen->th = gen->cth = gen->sth = gen->logsth = NULL; - -#ifdef __SSE2__ - gen->ith1 = gen->ith2 = -1; - gen->ylm_sse2 = RALLOC(v2df,gen->lmax+1); - ALLOC(gen->lambda_wx_sse2,v2df2 *,gen->smax+1); - for (m=0; m<=gen->smax; ++m) - gen->lambda_wx_sse2[m]=NULL; - gen->ylm_uptodate_sse2 = 0; - gen->lwx_uptodate_sse2 = RALLOC(int,gen->smax+1); - SET_ARRAY(gen->lwx_uptodate_sse2,0,gen->smax+1,0); -#endif - - ALLOC(gen->logsum,long double,2*gen->lmax+1); - gen->lc05 = gen->ls05 = NULL; - ALLOC(gen->flm1,double,2*gen->lmax+1); - ALLOC(gen->flm2,double,2*gen->lmax+1); - ALLOC(gen->xl,double,gen->lmax+1); - - gen->logsum[0] = 0.; - for (m=1; m<2*gen->lmax+1; ++m) - gen->logsum[m] = gen->logsum[m-1]+logl((long double)m); - for (m=0; m<2*gen->lmax+1; ++m) - { - gen->flm1[m] = sqrt(1./(m+1.)); - gen->flm2[m] = sqrt(m/(m+1.)); - } - - gen->xl[0]=0; - for (m=1; mlmax+1; ++m) gen->xl[m]=1./m; - } - -static void Ylmgen_destroy (Ylmgen_C *gen) - { - int m; - - DEALLOC(gen->firstl); - DEALLOC(gen->cf); - DEALLOC(gen->recfac); - DEALLOC(gen->mfac); - DEALLOC(gen->t1fac); - DEALLOC(gen->t2fac); - DEALLOC(gen->lamfact); - DEALLOC(gen->ylm); - DEALLOC(gen->lwx_uptodate); - for (m=0; m<=gen->smax; ++m) - DEALLOC(gen->lambda_wx[m]); - DEALLOC(gen->lambda_wx); - for (m=0; m<=gen->smax; ++m) - if (gen->sylm[m]) - { - sylmgen_destroy (gen->sylm[m]); - DEALLOC(gen->sylm[m]); - } - DEALLOC(gen->sylm); - DEALLOC(gen->th); - DEALLOC(gen->cth); - DEALLOC(gen->sth); - DEALLOC(gen->logsth); - DEALLOC(gen->logsum); - DEALLOC(gen->lc05); - DEALLOC(gen->ls05); - DEALLOC(gen->flm1); - DEALLOC(gen->flm2); - DEALLOC(gen->xl); -#ifdef __SSE2__ - DEALLOC(gen->ylm_sse2); - for (m=0; m<=gen->smax; ++m) - DEALLOC(gen->lambda_wx_sse2[m]); - DEALLOC(gen->lambda_wx_sse2); - DEALLOC(gen->lwx_uptodate_sse2); -#endif - } - -static void Ylmgen_set_theta (Ylmgen_C *gen, const double *theta, int nth) - { - const double inv_ln2 = 1.4426950408889634073599246810018921; - int m; - DEALLOC(gen->th); - DEALLOC(gen->cth); - DEALLOC(gen->sth); - DEALLOC(gen->logsth); - DEALLOC(gen->lc05); - DEALLOC(gen->ls05); - gen->th = RALLOC(double,nth); - gen->cth = RALLOC(double,nth); - gen->sth = RALLOC(double,nth); - gen->logsth = RALLOC(double,nth); - gen->lc05 = RALLOC(long double,nth); - gen->ls05 = RALLOC(long double,nth); - for (m=0; m=0.)&&(th<=pi),"bad theta angle specified"); - /* tiny adjustments to make sure cos and sin (theta/2) are positive */ - if (th==0.) th=1e-16; - if (ABSAPPROX(th,pi,1e-15)) th=pi-1e-15; - gen->th[m] = th; - gen->cth[m] = cos(th); - gen->sth[m] = sin(th); - gen->logsth[m] = inv_ln2*log(gen->sth[m]); - gen->lc05[m]=logl(cosl(0.5L*th)); - gen->ls05[m]=logl(sinl(0.5L*th)); - } - - gen->nth = nth; - gen->ith = -1; -#ifdef __SSE2__ - gen->ith1 = gen->ith2 = -1; -#endif - } - -static void Ylmgen_prepare (Ylmgen_C *gen, int ith, int m) - { - if ((ith==gen->ith) && (m==gen->m_cur)) return; - - gen->ylm_uptodate = 0; - SET_ARRAY(gen->lwx_uptodate,0,gen->smax+1,0); - - gen->ith = ith; - - if (m!=gen->m_cur) - { - gen->recfac_uptodate = 0; - gen->lamfact_uptodate = 0; - gen->m_cur = m; - } - } - -static void Ylmgen_recalc_recfac (Ylmgen_C *gen) - { - double f_old=1; - int l, m; - - if (gen->recfac_uptodate) return; - gen->recfac_uptodate = 1; - - m = gen->m_cur; - for (l=m; l<=gen->lmax; ++l) - { - gen->recfac[l][0] = gen->t1fac[l]*gen->t2fac[l+m]*gen->t2fac[l-m]; - gen->recfac[l][1] = gen->recfac[l][0]/f_old; - f_old = gen->recfac[l][0]; - } - } - -/*static void Ylmgen_recalc_lamfact (Ylmgen_C *gen) - { - int l, m; - - if (gen->lamfact_uptodate) return; - gen->lamfact_uptodate = 1; - - m = gen->m_cur; - gen->lamfact[m] = 0; - for (l=m+1; l<=gen->lmax; ++l) - gen->lamfact[l] = sqrt((2*l+1.)/(2*l-1.) * (l*l-m*m)); - }*/ - -#define RENORMALIZE_SCALAR \ - do \ - { \ - while (fabs(lam_2)>fbig) \ - { lam_1*=fsmall; lam_2*=fsmall; ++scale; } \ - corfac = (scale<0) ? 0. : gen->cf[scale]; \ - } \ - while(0) - -static void Ylmgen_recalc_Ylm (Ylmgen_C *gen) - { - const double ln2 = 0.6931471805599453094172321214581766; - - double logval,lam_1,lam_2,corfac; - double eps=gen->eps, fbig=gen->fbig, fsmall=gen->fsmall; - ylmgen_dbl2 *recfac = gen->recfac; - int lmax=gen->lmax; - int scale,l; - int m = gen->m_cur; - double cth=gen->cth[gen->ith], sth=gen->sth[gen->ith]; - double *result = gen->ylm; - - if (gen->ylm_uptodate) return; - gen->ylm_uptodate=1; - - if (((m>=gen->m_crit)&&(fabs(cth)>=gen->cth_crit)) || ((m>0)&&(sth==0))) - { gen->firstl[0]=gen->lmax+1; return; } - - Ylmgen_recalc_recfac(gen); - - logval = gen->mfac[m]; - if (m>0) logval += m*gen->logsth[gen->ith]; - scale = (int) (logval/large_exponent2)-minscale; - corfac = (scale<0) ? 0. : gen->cf[scale]; - - lam_1 = 0; - lam_2 = exp(ln2*(logval-(scale+minscale)*large_exponent2)); - if (m&1) lam_2 = -lam_2; - - l=m; - if (scale<0) - { - while (1) - { - if (++l>lmax) break; - lam_1 = cth*lam_2*recfac[l-1][0] - lam_1*recfac[l-1][1]; - if (++l>lmax) break; - lam_2 = cth*lam_1*recfac[l-1][0] - lam_2*recfac[l-1][1]; - if (fabs(lam_2)>fbig) - { - RENORMALIZE_SCALAR; - if (scale>=0) break; - } - } - } - - lam_1*=corfac; - lam_2*=corfac; - - if (l<=lmax) - { - while (1) - { - if (fabs(lam_2)>eps) break; - if (++l>lmax) break; - lam_1 = cth*lam_2*recfac[l-1][0] - lam_1*recfac[l-1][1]; - if (fabs(lam_1)>eps) - { double x=lam_1; lam_1=lam_2; lam_2=x; break; } - if (++l>lmax) break; - lam_2 = cth*lam_1*recfac[l-1][0] - lam_2*recfac[l-1][1]; - } - } - - gen->firstl[0]=l; - if (l>lmax) - { gen->m_crit=m; gen->cth_crit=fabs(cth); return; } - - for(;llmax) break; - lam_1 = cth*lam_2*recfac[l-1][0] - lam_1*recfac[l-1][1]; - result[l] = lam_1; - if (++l>lmax) break; - lam_2 = cth*lam_1*recfac[l-1][0] - lam_2*recfac[l-1][1]; - } - } - - -/* -static void Ylmgen_recalc_lambda_wx1 (Ylmgen_C *gen) - { - if (gen->lwx_uptodate[1]) return; - Ylmgen_recalc_Ylm(gen); - gen->firstl[1] = gen->firstl[0]; - if (gen->firstl[1]>gen->lmax) return; - Ylmgen_recalc_lamfact(gen); - gen->lwx_uptodate[1] = 1; - - { - double cth=gen->cth[gen->ith]; - double xsth=1./(gen->sth[gen->ith]); - double m=gen->m_cur; - double m_on_sth = m*xsth; - double lam_lm=0; - ylmgen_dbl2 *lambda_wx = gen->lambda_wx[1]; - int l; - double ell; - for (ell=l=gen->firstl[1]; l<=gen->lmax; ++l, ell+=1.) - { - double lam_lm1m=lam_lm; - lam_lm=gen->ylm[l]; - lambda_wx[l][0] = xsth*(gen->lamfact[l]*lam_lm1m - ell*cth*lam_lm); - lambda_wx[l][1] = m_on_sth*lam_lm; - } - } - } - -static void Ylmgen_recalc_lambda_wx2 (Ylmgen_C *gen) - { - if (gen->lwx_uptodate[2]) return; - Ylmgen_recalc_Ylm(gen); - gen->firstl[2] = gen->firstl[0]; - if (gen->firstl[2]>gen->lmax) return; - Ylmgen_recalc_lamfact(gen); - gen->lwx_uptodate[2] = 1; - - { - double cth=gen->cth[gen->ith]; - double sth=gen->sth[gen->ith]; - double m=gen->m_cur; - double one_on_s2 = 1./(sth*sth); - double two_on_s2 = 2*one_on_s2; - double two_c_on_s2 = cth * two_on_s2; - double m2 = m*m; - double two_m_on_s2 = m*two_on_s2; - double lam_lm=0; - ylmgen_dbl2 *lambda_wx = gen->lambda_wx[2]; - int l; - double ell; - for (ell=l=gen->firstl[2]; l<=gen->lmax; ++l, ell+=1.) - { - double lam_lm1m=lam_lm; - lam_lm=gen->ylm[l]; - { - const double t1 = lam_lm1m*gen->lamfact[l]; - const double a_w = (m2-ell)*two_on_s2 - ell*(ell-1.); - const double a_x = cth*(ell-1.)*lam_lm; - lambda_wx[l][0] = a_w*lam_lm + t1*two_c_on_s2; - lambda_wx[l][1] = two_m_on_s2 * (t1-a_x); - } - } - } - } - -void Ylmgen_recalc_lambda_wx (Ylmgen_C *gen, int spin) - { - UTIL_ASSERT ((spin>0) && (spin<=gen->smax), - "invalid spin in Ylmgen_recalc_lambda_wx"); - - if (!gen->lambda_wx[spin]) - gen->lambda_wx[spin]=RALLOC(ylmgen_dbl2,gen->lmax+1); - - if (gen->spinrec && spin==1) { Ylmgen_recalc_lambda_wx1(gen); return; } - if (gen->spinrec && spin==2) { Ylmgen_recalc_lambda_wx2(gen); return; } - - if (!gen->sylm[spin]) - { - gen->sylm[spin]=RALLOC(sylmgen_d,1); - sylmgen_init(gen->sylm[spin],gen,spin); - } - if (gen->lwx_uptodate[spin]) return; - sylmgen_prepare(gen->sylm[spin],gen,gen->m_cur); - sylmgen_recalc(gen->sylm[spin],gen,gen->ith,gen->lambda_wx[spin], - &gen->firstl[spin]); - gen->lwx_uptodate[spin] = 1; - } -*/ -#ifdef __SSE2__ - -static void Ylmgen_prepare_sse2 (Ylmgen_C *gen, int ith1, int ith2, int m) - { - if ((ith1==gen->ith1) && (ith2==gen->ith2) && (m==gen->m_cur)) return; - - gen->ylm_uptodate_sse2 = 0; - SET_ARRAY(gen->lwx_uptodate_sse2,0,gen->smax+1,0); - - gen->ith1 = ith1; gen->ith2 = ith2; - - if (m!=gen->m_cur) - { - gen->recfac_uptodate = gen->lamfact_uptodate = 0; - gen->m_cur = m; - } - } - - -#define RENORMALIZE \ - do \ - { \ - double lam1a, lam1b, lam2a, lam2b, corfaca, corfacb; \ - read_v2df (lam_1, &lam1a, &lam1b); read_v2df (lam_2, &lam2a, &lam2b); \ - read_v2df (corfac, &corfaca, &corfacb); \ - while (fabs(lam2a)>fbig) \ - { \ - lam1a*=fsmall; lam2a*=fsmall; ++scale1; \ - corfaca = (scale1<0) ? 0. : gen->cf[scale1]; \ - } \ - while (fabs(lam2b)>fbig) \ - { \ - lam1b*=fsmall; lam2b*=fsmall; ++scale2; \ - corfacb = (scale2<0) ? 0. : gen->cf[scale2]; \ - } \ - lam_1=build_v2df(lam1a,lam1b); lam_2=build_v2df(lam2a,lam2b); \ - corfac=build_v2df(corfaca,corfacb); \ - } \ - while(0) -#define GETPRE(prea,preb,lv) \ - { \ - prea=_mm_mul_pd(_mm_set1_pd(recfac[lv][0]),cth); \ - preb=_mm_set1_pd(recfac[lv][1]); \ - } -#define NEXTSTEP(prea,preb,prec,pred,reca,recb,lv) \ - { \ - preb = _mm_mul_pd(preb,reca); \ - prea = _mm_mul_pd(prea,recb); \ - prec = _mm_set1_pd(recfac[lv][0]); \ - pred = _mm_set1_pd(recfac[lv][1]); \ - reca = _mm_sub_pd(prea,preb); \ - prec = _mm_mul_pd(cth,prec); \ - } - -static void Ylmgen_recalc_Ylm_sse2 (Ylmgen_C *gen) - { - const double ln2 = 0.6931471805599453094172321214581766; - - v2df lam_1,lam_2,corfac; - double logval1,logval2; - double eps=gen->eps, fbig=gen->fbig, fsmall=gen->fsmall; - v2df eps2=build_v2df(eps,eps); - v2df fbig2=build_v2df(fbig,fbig); - ylmgen_dbl2 *recfac = gen->recfac; - int lmax=gen->lmax; - int scale1,scale2,l; - int m = gen->m_cur; - double cth1=gen->cth[gen->ith1], cth2=gen->cth[gen->ith2]; - v2df cth=build_v2df(cth1,cth2); - v2df *result = gen->ylm_sse2; - v2df pre0,pre1,pre2,pre3; - - if (gen->ylm_uptodate_sse2) return; - gen->ylm_uptodate_sse2=1; - - if ((m>=gen->m_crit)&&(fabs(cth1)>=gen->cth_crit)&&(fabs(cth2)>=gen->cth_crit)) - { gen->firstl[0]=gen->lmax+1; return; } - - Ylmgen_recalc_recfac(gen); - - logval1 = logval2 = gen->mfac[m]; - if (m>0) logval1 += m*gen->logsth[gen->ith1]; - if (m>0) logval2 += m*gen->logsth[gen->ith2]; - scale1 = (int) (logval1/large_exponent2)-minscale; - scale2 = (int) (logval2/large_exponent2)-minscale; - corfac = build_v2df((scale1<0) ? 0. : gen->cf[scale1], - (scale2<0) ? 0. : gen->cf[scale2]); - - lam_1 = _mm_setzero_pd(); - lam_2 = build_v2df(exp(ln2*(logval1-(scale1+minscale)*large_exponent2)), - exp(ln2*(logval2-(scale2+minscale)*large_exponent2))); - if (m&1) lam_2 = _mm_xor_pd (lam_2,V2DF_SIGNMASK); /* negate */ - - l=m; - if ((scale1<0) && (scale2<0)) - { - GETPRE(pre0,pre1,l) - while (1) - { - if (++l>lmax) break; - NEXTSTEP(pre0,pre1,pre2,pre3,lam_1,lam_2,l) - if (++l>lmax) break; - NEXTSTEP(pre2,pre3,pre0,pre1,lam_2,lam_1,l) - if (v2df_any_gt(lam_2,fbig2)) - { - RENORMALIZE; - if ((scale1>=0) || (scale2>=0)) break; - } - } - } - - if (l<=lmax) - { - GETPRE(pre0,pre1,l) - while (1) - { - v2df t1; - result[l]=t1=_mm_mul_pd(lam_2,corfac); - if (v2df_any_gt(t1,eps2)) - break; - if (++l>lmax) break; - NEXTSTEP(pre0,pre1,pre2,pre3,lam_1,lam_2,l) - - result[l]=t1=_mm_mul_pd(lam_1,corfac); - if (v2df_any_gt(t1,eps2)) - { v2df tmp=lam_1;lam_1=lam_2;lam_2=tmp; break; } - if (++l>lmax) break; - NEXTSTEP(pre2,pre3,pre0,pre1,lam_2,lam_1,l) - - if (v2df_any_gt(lam_2,fbig2)) - RENORMALIZE; - } - } - - gen->firstl[0]=l; - if (l>lmax) - { - gen->m_crit=m; - gen->cth_crit= (fabs(cth1)lmax) return; - NEXTSTEP(pre0,pre1,pre2,pre3,lam_1,lam_2,l) - - result[l]=t1=_mm_mul_pd(lam_1,corfac); - if (v2df_all_ge(t1,eps2)) - { v2df tmp=lam_1;lam_1=lam_2;lam_2=tmp; break; } - if (++l>lmax) return; - NEXTSTEP(pre2,pre3,pre0,pre1,lam_2,lam_1,l) - - if (v2df_any_gt(lam_2,fbig2)) - RENORMALIZE; - } - - lam_1 = _mm_mul_pd (lam_1,corfac); - lam_2 = _mm_mul_pd (lam_2,corfac); - - GETPRE(pre0,pre1,l) - for(;llmax) break; - NEXTSTEP(pre0,pre1,pre2,pre3,lam_1,lam_2,l) - result[l] = lam_1; - if (++l>lmax) break; - NEXTSTEP(pre2,pre3,pre0,pre1,lam_2,lam_1,l) - } - } - -/*static void Ylmgen_recalc_lambda_wx1_sse2 (Ylmgen_C *gen) - { - if (gen->lwx_uptodate_sse2[1]) return; - Ylmgen_recalc_Ylm_sse2(gen); - gen->firstl[1] = gen->firstl[0]; - if (gen->firstl[1]>gen->lmax) return; - Ylmgen_recalc_lamfact(gen); - gen->lwx_uptodate_sse2[1] = 1; - - { - v2df cth=build_v2df(gen->cth[gen->ith1],gen->cth[gen->ith2]); - v2df xsth=build_v2df(1./gen->sth[gen->ith1],1./gen->sth[gen->ith2]); - v2df m=build_v2df(gen->m_cur,gen->m_cur); - v2df m_on_sth = _mm_mul_pd(m,xsth); - v2df lam_lm=_mm_setzero_pd(); - v2df2 *lambda_wx = gen->lambda_wx_sse2[1]; - int l; - v2df ell=build_v2df(gen->firstl[1],gen->firstl[1]); - v2df uno=_mm_set1_pd(1.); - for (l=gen->firstl[1]; l<=gen->lmax; ++l, ell=_mm_add_pd(ell,uno)) - { - v2df lamfact=_mm_load1_pd(&gen->lamfact[l]); - v2df lam_lm1m=lam_lm; - lam_lm=gen->ylm_sse2[l]; - lambda_wx[l].a = _mm_mul_pd(xsth,_mm_sub_pd(_mm_mul_pd(lamfact,lam_lm1m), - _mm_mul_pd(_mm_mul_pd(ell,cth),lam_lm))); - lambda_wx[l].b = _mm_mul_pd(m_on_sth,lam_lm); - } - } - } - -static void Ylmgen_recalc_lambda_wx2_sse2 (Ylmgen_C *gen) - { - if (gen->lwx_uptodate_sse2[2]) return; - Ylmgen_recalc_Ylm_sse2(gen); - gen->firstl[2] = gen->firstl[0]; - if (gen->firstl[2]>gen->lmax) return; - Ylmgen_recalc_lamfact(gen); - gen->lwx_uptodate_sse2[2] = 1; - - { - v2df cth=build_v2df(gen->cth[gen->ith1],gen->cth[gen->ith2]); - v2df sth=build_v2df(gen->sth[gen->ith1],gen->sth[gen->ith2]); - v2df m=build_v2df(gen->m_cur,gen->m_cur); - v2df uno=_mm_set1_pd(1.); - v2df one_on_s2 = _mm_div_pd(uno,_mm_mul_pd(sth,sth)); - v2df two_on_s2 = _mm_mul_pd(_mm_set1_pd(2.),one_on_s2); - v2df two_c_on_s2 = _mm_mul_pd(cth,two_on_s2); - v2df m2 = _mm_mul_pd(m,m); - v2df two_m_on_s2 = _mm_mul_pd(m,two_on_s2); - v2df lam_lm=_mm_setzero_pd(); - v2df2 *lambda_wx = gen->lambda_wx_sse2[2]; - int l; - v2df ell=build_v2df(gen->firstl[2],gen->firstl[2]); - for (l=gen->firstl[2]; l<=gen->lmax; ++l, ell=_mm_add_pd(ell,uno)) - { - v2df lamfact=_mm_load1_pd(&gen->lamfact[l]); - v2df lam_lm1m=lam_lm; - lam_lm=gen->ylm_sse2[l]; - { - const v2df t1 = _mm_mul_pd(lam_lm1m,lamfact); - const v2df ellm1 = _mm_sub_pd(ell,uno); - const v2df a_w = _mm_sub_pd - (_mm_mul_pd(_mm_sub_pd(m2,ell),two_on_s2),_mm_mul_pd(ell,ellm1)); - const v2df a_x = _mm_mul_pd(_mm_mul_pd(cth,ellm1),lam_lm); - lambda_wx[l].a = - _mm_add_pd(_mm_mul_pd(a_w,lam_lm),_mm_mul_pd(t1,two_c_on_s2)); - lambda_wx[l].b = _mm_mul_pd(two_m_on_s2,_mm_sub_pd(t1,a_x)); - } - } - } - }*/ - -/*static void Ylmgen_recalc_lambda_wx_sse2 (Ylmgen_C *gen, int spin) - { - UTIL_ASSERT ((spin>0) && (spin<=gen->smax), - "invalid spin in Ylmgen_recalc_lambda_wx_sse2"); - - if (!gen->lambda_wx_sse2[spin]) - gen->lambda_wx_sse2[spin]=RALLOC(v2df2,gen->lmax+1); - - if (gen->spinrec && spin==1) { Ylmgen_recalc_lambda_wx1_sse2(gen); return; } - if (gen->spinrec && spin==2) { Ylmgen_recalc_lambda_wx2_sse2(gen); return; } - - if (!gen->sylm[spin]) - { - gen->sylm[spin]=RALLOC(sylmgen_d,1); - sylmgen_init(gen->sylm[spin],gen,spin); - } - if (gen->lwx_uptodate_sse2[spin]) return; - sylmgen_prepare(gen->sylm[spin],gen,gen->m_cur); - sylmgen_recalc_sse2(gen->sylm[spin],gen,gen->ith1,gen->ith2, - gen->lambda_wx_sse2[spin],&gen->firstl[spin]); - gen->lwx_uptodate_sse2[spin] = 1; - }*/ - -#endif /* __SSE2__ */ - -/* -double *Ylmgen_get_norm (int lmax, int spin, int spinrec) - { - const double pi = 3.141592653589793238462643383279502884197; - double *res=RALLOC(double,lmax+1); - int l; - double spinsign; - / sign convention for H=1 (LensPix paper) / -#if 1 - spinsign = (spin>0) ? -1.0 : 1.0; -#else - spinsign = 1.0; -#endif - - if (spin==0) - { - for (l=0; l<=lmax; ++l) - res[l]=1.; - return res; - } - - if ((!spinrec) || (spin>=3)) - { - spinsign = (spin&1) ? -spinsign : spinsign; - for (l=0; l<=lmax; ++l) - res[l] = (l - -void sharp_normalized_associated_legendre_table( - ptrdiff_t m, - int spin, - ptrdiff_t lmax, - ptrdiff_t ntheta, - double *theta, - ptrdiff_t theta_stride, - ptrdiff_t l_stride, - ptrdiff_t spin_stride, - double *out -) { - if (spin != 0) UTIL_FAIL ("sharp_normalized_associated_legendre_table: only spin=0 has been implemented so far"); - - Ylmgen_C ctx; - ptrdiff_t itheta, l, lmin; - - Ylmgen_init(&ctx, lmax, lmax, 0, 0, 1e-300); - Ylmgen_set_theta(&ctx, theta, ntheta); - - itheta = 0; - #ifdef __SSE2__ - for (; itheta < ntheta - 1; itheta += 2) { - Ylmgen_prepare_sse2(&ctx, itheta, itheta + 1, m); - Ylmgen_recalc_Ylm_sse2(&ctx); - lmin = IMIN(*ctx.firstl, lmax + 1); - for (l = m; l < lmin; ++l) { - out[itheta * theta_stride + (l - m) * l_stride + spin * spin_stride] = 0; - out[(itheta + 1) * theta_stride + (l - m) * l_stride + spin * spin_stride] = 0; - } - for (l = IMAX(lmin, m); l <= lmax; ++l) { - double v1, v2; - read_v2df(ctx.ylm_sse2[l], &v1, &v2); - out[itheta * theta_stride + (l - m) * l_stride + spin * spin_stride] = v1; - out[(itheta + 1) * theta_stride + (l - m) * l_stride + spin * spin_stride] = v2; - } - } - #endif - for (; itheta < ntheta; itheta += 1) { - Ylmgen_prepare(&ctx, itheta, m); - Ylmgen_recalc_Ylm(&ctx); - lmin = IMIN(*ctx.firstl, lmax + 1); - for (l = m; l < lmin; ++l) { - out[itheta * theta_stride + (l - m) * l_stride + spin * spin_stride] = 0; - } - for (l = IMAX(lmin, m); l <= lmax; ++l) { - out[itheta * theta_stride + (l - m) * l_stride + spin * spin_stride] = ctx.ylm[l]; - } - } - Ylmgen_destroy(&ctx); -} diff --git a/libsharp/sharp_legendre_table.h b/libsharp/sharp_legendre_table.h deleted file mode 100644 index de4cdd6..0000000 --- a/libsharp/sharp_legendre_table.h +++ /dev/null @@ -1,97 +0,0 @@ -/* - * This file is part of libsharp. - * - * Redistribution and use in source and binary forms, with or without - * met: - * - * 1. Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * - * 2. Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in the - * documentation and/or other materials provided with the distribution. - * - * 3. Neither the name of the copyright holder nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -/*! \file sharp_legendre_table.h - * Interface for computing tables of the normalized associated Legendre transform - * - * Copyright (C) 2017 Dag Sverre Seljebotn - * \author Dag Sverre Seljebotn - * - * Note: This code was mainly copied from libpsht; only a small high-level wrapper added - */ - -#ifndef SHARP_LEGENDRE_TABLE_H -#define SHARP_LEGENDRE_TABLE_H - -#include - -#ifdef __cplusplus -extern "C" { -#endif - -#ifndef NO_LEGENDRE_TABLE - - -/*! Returns a table of the normalized associated Legendre polynomials. m is a single - fixed argument and a table for multiple l and cos(theta) is provided. - (Internally, sin(theta) is also used for part of the computation, making theta - the most convenient argument.) - - NOTE: Support for spin-weighted Legendre functions is on the TODO-list. Only spin=0 - is supported now. - - \param m The m-value to compute a table for; must be >= 0 - \param spin The spin parameter; pass 0 for the regular associated Legendre functions. - NOTE: This is present for future compatability, currently only 0 is supported. - \param lmax A table will be provided for l = m .. lmax - \param ntheta How many theta values to evaluate for - \param theta Contiguous 1D array of theta values - \param theta_stride See below - \param l_stride See below - \param spin_stride See below. "ispin" will always be 0 if spin==0, or 0 for positive spin - and 1 for the corresponding negative spin otherwise. - \param out Contiguous 3D array that will receive the output. Each output entry - is assigned to out[itheta * theta_stride + (l - m) * l_stride + ispin * spin_stride]. - */ -void sharp_normalized_associated_legendre_table( - ptrdiff_t m, - int spin, - ptrdiff_t lmax, - ptrdiff_t ntheta, - /* contiguous 1D array of theta values to compute for, - contains ntheta values */ - double *theta, - /* contiguous 2D array, in "theta-major ordering". Has `ntheta` - rows and `ncols` columns. Indexed as out[itheta * ncols + (l - m)]. - If `ncols > lmax - m` then those entries are not accessed. - */ - ptrdiff_t theta_stride, - ptrdiff_t l_stride, - ptrdiff_t spin_stride, - double *out -); - -#endif - -#ifdef __cplusplus -} -#endif - -#endif diff --git a/libsharp/sharp_lowlevel.h b/libsharp/sharp_lowlevel.h deleted file mode 100644 index d9aa01b..0000000 --- a/libsharp/sharp_lowlevel.h +++ /dev/null @@ -1,272 +0,0 @@ -/* - * This file is part of libsharp. - * - * libsharp is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libsharp is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libsharp; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libsharp is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/*! \file sharp_lowlevel.h - * Low-level, portable interface for the spherical transform library. - * - * Copyright (C) 2012-2013 Max-Planck-Society - * \author Martin Reinecke \author Dag Sverre Seljebotn - */ - -#ifndef PLANCK_SHARP_LOWLEVEL_H -#define PLANCK_SHARP_LOWLEVEL_H - -#include - -#ifdef __cplusplus -extern "C" { -#endif - -/*! \internal - Helper type containing information about a single ring. */ -typedef struct - { - double theta, phi0, weight, cth, sth; - ptrdiff_t ofs; - int nph, stride; - } sharp_ringinfo; - -/*! \internal - Helper type containing information about a pair of rings with colatitudes - symmetric around the equator. */ -typedef struct - { - sharp_ringinfo r1,r2; - } sharp_ringpair; - -/*! \internal - Type holding all required information about a map geometry. */ -typedef struct - { - sharp_ringpair *pair; - int npairs, nphmax; - } sharp_geom_info; - -/*! \defgroup almgroup Helpers for dealing with a_lm */ -/*! \{ */ - -/*! \internal - Helper type for index calculation in a_lm arrays. */ -typedef struct - { - /*! Maximum \a l index of the array */ - int lmax; - /*! Number of different \a m values in this object */ - int nm; - /*! Array with \a nm entries containing the individual m values */ - int *mval; - /*! Combination of flags from sharp_almflags */ - int flags; - /*! Array with \a nm entries containing the (hypothetical) indices of - the coefficients with quantum numbers 0,\a mval[i] */ - ptrdiff_t *mvstart; - /*! Stride between a_lm and a_(l+1),m */ - ptrdiff_t stride; - } sharp_alm_info; - -/*! alm_info flags */ -typedef enum { SHARP_PACKED = 1, - /*!< m=0-coefficients are packed so that the (zero) imaginary part is - not present. mvstart is in units of *real* float/double for all - m; stride is in units of reals for m=0 and complex for m!=0 */ - SHARP_REAL_HARMONICS = 1<<6 - /*!< Use the real spherical harmonic convention. For - m==0, the alm are treated exactly the same as in - the complex case. For m!=0, alm[i] represent a - pair (+abs(m), -abs(m)) instead of (real, imag), - and the coefficients are scaled by a factor of - sqrt(2) relative to the complex case. In other - words, (sqrt(.5) * alm[i]) recovers the - corresponding complex coefficient (when accessed - as complex). - */ - } sharp_almflags; - - - -/*! Creates an a_lm data structure from the following parameters: - \param lmax maximum \a l quantum number (>=0) - \param mmax maximum \a m quantum number (0<= \a mmax <= \a lmax) - \param stride the stride between entries with identical \a m, and \a l - differing by 1. - \param mstart the index of the (hypothetical) coefficient with the - quantum numbers 0,\a m. Must have \a mmax+1 entries. - \param alm_info will hold a pointer to the newly created data structure - */ -void sharp_make_alm_info (int lmax, int mmax, int stride, - const ptrdiff_t *mstart, sharp_alm_info **alm_info); -/*! Creates an a_lm data structure which from the following parameters: - \param lmax maximum \a l quantum number (\a >=0) - \param nm number of different \a m (\a 0<=nm<=lmax+1) - \param stride the stride between entries with identical \a m, and \a l - differing by 1. - \param mval array with \a nm entries containing the individual m values - \param mvstart array with \a nm entries containing the (hypothetical) - indices of the coefficients with the quantum numbers 0,\a mval[i] - \param flags a combination of sharp_almflags (pass 0 unless you know you need this) - \param alm_info will hold a pointer to the newly created data structure - */ -void sharp_make_general_alm_info (int lmax, int nm, int stride, const int *mval, - const ptrdiff_t *mvstart, int flags, sharp_alm_info **alm_info); -/*! Returns the index of the coefficient with quantum numbers \a l, - \a mval[mi]. - \note for a \a sharp_alm_info generated by sharp_make_alm_info() this is - the index for the coefficient with the quantum numbers \a l, \a mi. */ -ptrdiff_t sharp_alm_index (const sharp_alm_info *self, int l, int mi); -/*! Returns the number of alm coefficients described by \a self. If the SHARP_PACKED - flag is set, this is number of "real" coeffecients (for m < 0 and m >= 0), - otherwise it is the number of complex coefficients (with m>=0). */ -ptrdiff_t sharp_alm_count(const sharp_alm_info *self); -/*! Deallocates the a_lm info object. */ -void sharp_destroy_alm_info (sharp_alm_info *info); - -/*! \} */ - -/*! \defgroup geominfogroup Functions for dealing with geometry information */ -/*! \{ */ - -/*! Creates a geometry information from a set of ring descriptions. - All arrays passed to this function must have \a nrings elements. - \param nrings the number of rings in the map - \param nph the number of pixels in each ring - \param ofs the index of the first pixel in each ring in the map array - \param stride the stride between consecutive pixels - \param phi0 the azimuth (in radians) of the first pixel in each ring - \param theta the colatitude (in radians) of each ring - \param wgt the pixel weight to be used for the ring in map2alm - and adjoint map2alm transforms. - Pass NULL to use 1.0 as weight for all rings. - \param geom_info will hold a pointer to the newly created data structure - */ -void sharp_make_geom_info (int nrings, const int *nph, const ptrdiff_t *ofs, - const int *stride, const double *phi0, const double *theta, - const double *wgt, sharp_geom_info **geom_info); - -/*! Counts the number of grid points needed for (the local part of) a map described - by \a info. - */ -ptrdiff_t sharp_map_size(const sharp_geom_info *info); - -/*! Deallocates the geometry information in \a info. */ -void sharp_destroy_geom_info (sharp_geom_info *info); - -/*! \} */ - -/*! \defgroup lowlevelgroup Low-level libsharp SHT interface */ -/*! \{ */ - -/*! Enumeration of SHARP job types. */ -typedef enum { SHARP_YtW=0, /*!< analysis */ - SHARP_MAP2ALM=SHARP_YtW, /*!< analysis */ - SHARP_Y=1, /*!< synthesis */ - SHARP_ALM2MAP=SHARP_Y, /*!< synthesis */ - SHARP_Yt=2, /*!< adjoint synthesis */ - SHARP_WY=3, /*!< adjoint analysis */ - SHARP_ALM2MAP_DERIV1=4 /*!< synthesis of first derivatives */ - } sharp_jobtype; - -/*! Job flags */ -typedef enum { SHARP_DP = 1<<4, - /*!< map and a_lm are in double precision */ - SHARP_ADD = 1<<5, - /*!< results are added to the output arrays, instead of - overwriting them */ - - /* NOTE: SHARP_REAL_HARMONICS, 1<<6, is also available in sharp_jobflags, - but its use here is deprecated in favor of having it in the sharp_alm_info */ - - SHARP_NO_FFT = 1<<7, - - SHARP_USE_WEIGHTS = 1<<20, /* internal use only */ - SHARP_NO_OPENMP = 1<<21, /* internal use only */ - SHARP_NVMAX = (1<<4)-1 /* internal use only */ - } sharp_jobflags; - -/*! Performs a libsharp SHT job. The interface deliberately does not use - the C99 "complex" data type, in order to be callable from C89 and C++. - \param type the type of SHT - \param spin the spin of the quantities to be transformed - \param alm contains pointers to the a_lm coefficients. If \a spin==0, - alm[0] points to the a_lm of the first SHT, alm[1] to those of the second - etc. If \a spin>0, alm[0] and alm[1] point to the a_lm of the first SHT, - alm[2] and alm[3] to those of the second, etc. The exact data type of \a alm - depends on whether the SHARP_DP flag is set. - \param map contains pointers to the maps. If \a spin==0, - map[0] points to the map of the first SHT, map[1] to that of the second - etc. If \a spin>0, or \a type is SHARP_ALM2MAP_DERIV1, map[0] and map[1] - point to the maps of the first SHT, map[2] and map[3] to those of the - second, etc. The exact data type of \a map depends on whether the SHARP_DP - flag is set. - \param geom_info A \c sharp_geom_info object compatible with the provided - \a map arrays. - \param alm_info A \c sharp_alm_info object compatible with the provided - \a alm arrays. All \c m values from 0 to some \c mmax<=lmax must be present - exactly once. - \param ntrans the number of simultaneous SHTs - \param flags See sharp_jobflags. In particular, if SHARP_DP is set, then - \a alm is expected to have the type "complex double **" and \a map is - expected to have the type "double **"; otherwise, the expected - types are "complex float **" and "float **", respectively. - \param time If not NULL, the wall clock time required for this SHT - (in seconds) will be written here. - \param opcnt If not NULL, a conservative estimate of the total floating point - operation count for this SHT will be written here. */ -void sharp_execute (sharp_jobtype type, int spin, void *alm, void *map, - const sharp_geom_info *geom_info, const sharp_alm_info *alm_info, int ntrans, - int flags, double *time, unsigned long long *opcnt); - -void sharp_set_chunksize_min(int new_chunksize_min); -void sharp_set_nchunks_max(int new_nchunks_max); - - -typedef enum { SHARP_ERROR_NO_MPI = 1, - /*!< libsharp not compiled with MPI support */ - } sharp_errors; - -/*! Works like sharp_execute_mpi, but is always present whether or not libsharp - is compiled with USE_MPI. This is primarily useful for wrapper code etc. - - Note that \a pcomm has the type MPI_Comm*, except we declare void* to avoid - pulling in MPI headers. I.e., the comm argument of sharp_execute_mpi - is *(MPI_Comm*)pcomm. - - Other parameters are the same as sharp_execute_mpi. - - Returns 0 if successful, or SHARP_ERROR_NO_MPI if MPI is not available - (in which case nothing is done). - */ -int sharp_execute_mpi_maybe (void *pcomm, sharp_jobtype type, int spin, - void *alm, void *map, const sharp_geom_info *geom_info, - const sharp_alm_info *alm_info, int ntrans, int flags, double *time, - unsigned long long *opcnt); - - - -/*! \} */ - -#ifdef __cplusplus -} -#endif - -#endif diff --git a/libsharp/sharp_mpi.c b/libsharp/sharp_mpi.c index a364ed4..a90ec69 100644 --- a/libsharp/sharp_mpi.c +++ b/libsharp/sharp_mpi.c @@ -25,7 +25,7 @@ /*! \file sharp_mpi.c * Functionality only needed for MPI-parallel transforms * - * Copyright (C) 2012-2013 Max-Planck-Society + * Copyright (C) 2012-2019 Max-Planck-Society * \author Martin Reinecke \author Dag Sverre Seljebotn */ @@ -101,7 +101,7 @@ static void sharp_make_mpi_info (MPI_Comm comm, const sharp_job *job, DEALLOC(theta_tmp); DEALLOC(ispair_tmp); - minfo->nph=2*job->nmaps*job->ntrans; + minfo->nph=2*job->nmaps; minfo->almcount=RALLOC(int,minfo->ntasks); minfo->almdisp=RALLOC(int,minfo->ntasks+1); @@ -184,8 +184,8 @@ static void alloc_phase_mpi (sharp_job *job, int nm, int ntheta, { ptrdiff_t phase_size = (job->type==SHARP_MAP2ALM) ? (ptrdiff_t)(nmfull)*ntheta : (ptrdiff_t)(nm)*nthetafull; - job->phase=RALLOC(dcmplx,2*job->ntrans*job->nmaps*phase_size); - job->s_m=2*job->ntrans*job->nmaps; + job->phase=RALLOC(dcmplx,2*job->nmaps*phase_size); + job->s_m=2*job->nmaps; job->s_th = job->s_m * ((job->type==SHARP_MAP2ALM) ? nmfull : nm); } @@ -315,12 +315,12 @@ static void sharp_execute_job_mpi (sharp_job *job, MPI_Comm comm) void sharp_execute_mpi (MPI_Comm comm, sharp_jobtype type, int spin, void *alm, void *map, const sharp_geom_info *geom_info, - const sharp_alm_info *alm_info, int ntrans, int flags, double *time, + const sharp_alm_info *alm_info, int flags, double *time, unsigned long long *opcnt) { sharp_job job; sharp_build_job_common (&job, type, spin, alm, map, geom_info, alm_info, - ntrans, flags); + flags); sharp_execute_job_mpi (&job, comm); if (time!=NULL) *time = job.time; @@ -331,15 +331,15 @@ void sharp_execute_mpi (MPI_Comm comm, sharp_jobtype type, int spin, without declaring it in C header as it should not be available to C code */ void sharp_execute_mpi_fortran(MPI_Fint comm, sharp_jobtype type, int spin, void *alm, void *map, const sharp_geom_info *geom_info, - const sharp_alm_info *alm_info, int ntrans, int flags, double *time, + const sharp_alm_info *alm_info, int flags, double *time, unsigned long long *opcnt); void sharp_execute_mpi_fortran(MPI_Fint comm, sharp_jobtype type, int spin, void *alm, void *map, const sharp_geom_info *geom_info, - const sharp_alm_info *alm_info, int ntrans, int flags, double *time, + const sharp_alm_info *alm_info, int flags, double *time, unsigned long long *opcnt) { sharp_execute_mpi(MPI_Comm_f2c(comm), type, spin, alm, map, geom_info, - alm_info, ntrans, flags, time, opcnt); + alm_info, flags, time, opcnt); } #endif diff --git a/libsharp/sharp_mpi.h b/libsharp/sharp_mpi.h index 1053a65..5de37bf 100644 --- a/libsharp/sharp_mpi.h +++ b/libsharp/sharp_mpi.h @@ -25,7 +25,7 @@ /*! \file sharp_mpi.h * Interface for the spherical transform library with MPI support. * - * Copyright (C) 2011,2012 Max-Planck-Society + * Copyright (C) 2011-2019 Max-Planck-Society * \author Martin Reinecke \author Dag Sverre Seljebotn */ @@ -33,28 +33,25 @@ #define PLANCK_SHARP_MPI_H #include -#include "sharp_lowlevel.h" +#include "sharp.h" #ifdef __cplusplus extern "C" { #endif /*! Performs an MPI parallel libsharp SHT job. The interface deliberately does - not use the C99 "complex" data type, in order to be callable from C. + not use the C99 "complex" data type, in order to be callable from C89 and C++. \param comm the MPI communicator to be used for this SHT \param type the type of SHT \param spin the spin of the quantities to be transformed \param alm contains pointers to the a_lm coefficients. If \a spin==0, - alm[0] points to the a_lm of the first SHT, alm[1] to those of the second - etc. If \a spin>0, alm[0] and alm[1] point to the a_lm of the first SHT, - alm[2] and alm[3] to those of the second, etc. The exact data type of \a alm + alm[0] points to the a_lm of the SHT. If \a spin>0, alm[0] and alm[1] + point to the two a_lm sets of the SHT. The exact data type of \a alm depends on whether the SHARP_DP flag is set. \param map contains pointers to the maps. If \a spin==0, - map[0] points to the map of the first SHT, map[1] to that of the second - etc. If \a spin>0, or \a type is SHARP_ALM2MAP_DERIV1, map[0] and map[1] - point to the maps of the first SHT, map[2] and map[3] to those of the - second, etc. The exact data type of \a map depends on whether the SHARP_DP - flag is set. + map[0] points to the map of the SHT. If \a spin>0, or \a type is + SHARP_ALM2MAP_DERIV1, map[0] and map[1] point to the two maps of the SHT. + The exact data type of \a map depends on whether the SHARP_DP flag is set. \param geom_info A \c sharp_geom_info object compatible with the provided \a map arrays. The total map geometry is the union of all \a geom_info objects over the participating MPI tasks. @@ -62,7 +59,6 @@ extern "C" { \a alm arrays. All \c m values from 0 to some \c mmax<=lmax must be present exactly once in the union of all \a alm_info objects over the participating MPI tasks. - \param ntrans the number of simultaneous SHTs \param flags See sharp_jobflags. In particular, if SHARP_DP is set, then \a alm is expected to have the type "complex double **" and \a map is expected to have the type "double **"; otherwise, the expected @@ -73,7 +69,7 @@ extern "C" { operation count for this SHT will be written here. */ void sharp_execute_mpi (MPI_Comm comm, sharp_jobtype type, int spin, void *alm, void *map, const sharp_geom_info *geom_info, - const sharp_alm_info *alm_info, int ntrans, int flags, double *time, + const sharp_alm_info *alm_info, int flags, double *time, unsigned long long *opcnt); #ifdef __cplusplus diff --git a/libsharp/sharp_testsuite.c b/libsharp/sharp_testsuite.c index 2b1c7af..f22a91e 100644 --- a/libsharp/sharp_testsuite.c +++ b/libsharp/sharp_testsuite.c @@ -23,8 +23,8 @@ */ /* \file sharp_testsuite.c - * - * Copyright (C) 2012-2013 Max-Planck-Society + * + * Copyright (C) 2012-2019 Max-Planck-Society * \author Martin Reinecke */ @@ -42,17 +42,68 @@ #include "sharp_geomhelpers.h" #include "sharp_almhelpers.h" #include "c_utils.h" -#include "sharp_announce.h" #include "memusage.h" -#include "sharp_vecsupport.h" + +static void OpenMP_status(void) + { +#ifndef _OPENMP + printf("OpenMP: not supported by this binary\n"); +#else + int threads = omp_get_max_threads(); + if (threads>1) + printf("OpenMP active: max. %d threads.\n",threads); + else + printf("OpenMP active, but running with 1 thread only.\n"); +#endif + } + +static void MPI_status(void) + { +#ifndef USE_MPI + printf("MPI: not supported by this binary\n"); +#else + int tasks; + MPI_Comm_size(MPI_COMM_WORLD,&tasks); + if (tasks>1) + printf("MPI active with %d tasks.\n",tasks); + else + printf("MPI active, but running with 1 task only.\n"); +#endif + } + +static void sharp_announce (const char *name) + { + size_t m, nlen=strlen(name); + printf("\n+-"); + for (m=0; mnm; ++mi) { int m=helper->mval[mi]; - int state=1234567*cnt+8912*m; // random seed + unsigned state=1234567u*(unsigned)cnt+8912u*(unsigned)m; // random seed for (int l=m;l<=helper->lmax; ++l) { if ((l=0,"lmax must not be negative"); if (*mmax<0) *mmax=lmax; UTIL_ASSERT(*mmax<=lmax,"mmax larger than lmax"); - if (mytask==0) printf ("lmax: %d, mmax: %d\n",lmax,*mmax); + verbose &= (mytask==0); + if (verbose) printf ("lmax: %d, mmax: %d\n",lmax,*mmax); sharp_make_triangular_alm_info(lmax,*mmax,1,ainfo); #ifdef USE_MPI @@ -260,14 +312,14 @@ static void get_infos (const char *gname, int lmax, int *mmax, int *gpar1, if (*gpar1<1) *gpar1=lmax/2; if (*gpar1==0) ++(*gpar1); sharp_make_healpix_geom_info (*gpar1, 1, ginfo); - if (mytask==0) printf ("HEALPix grid, nside=%d\n",*gpar1); + if (verbose) printf ("HEALPix grid, nside=%d\n",*gpar1); } else if (strcmp(gname,"gauss")==0) { if (*gpar1<1) *gpar1=lmax+1; if (*gpar2<1) *gpar2=2*(*mmax)+1; sharp_make_gauss_geom_info (*gpar1, *gpar2, 0., 1, *gpar2, ginfo); - if (mytask==0) + if (verbose) printf ("Gauss-Legendre grid, nlat=%d, nlon=%d\n",*gpar1,*gpar2); } else if (strcmp(gname,"fejer1")==0) @@ -275,21 +327,21 @@ static void get_infos (const char *gname, int lmax, int *mmax, int *gpar1, if (*gpar1<1) *gpar1=2*lmax+1; if (*gpar2<1) *gpar2=2*(*mmax)+1; sharp_make_fejer1_geom_info (*gpar1, *gpar2, 0., 1, *gpar2, ginfo); - if (mytask==0) printf ("Fejer1 grid, nlat=%d, nlon=%d\n",*gpar1,*gpar2); + if (verbose) printf ("Fejer1 grid, nlat=%d, nlon=%d\n",*gpar1,*gpar2); } else if (strcmp(gname,"fejer2")==0) { if (*gpar1<1) *gpar1=2*lmax+1; if (*gpar2<1) *gpar2=2*(*mmax)+1; sharp_make_fejer2_geom_info (*gpar1, *gpar2, 0., 1, *gpar2, ginfo); - if (mytask==0) printf ("Fejer2 grid, nlat=%d, nlon=%d\n",*gpar1,*gpar2); + if (verbose) printf ("Fejer2 grid, nlat=%d, nlon=%d\n",*gpar1,*gpar2); } else if (strcmp(gname,"cc")==0) { if (*gpar1<1) *gpar1=2*lmax+1; if (*gpar2<1) *gpar2=2*(*mmax)+1; sharp_make_cc_geom_info (*gpar1, *gpar2, 0., 1, *gpar2, ginfo); - if (mytask==0) + if (verbose) printf("Clenshaw-Curtis grid, nlat=%d, nlon=%d\n",*gpar1,*gpar2); } else if (strcmp(gname,"smallgauss")==0) @@ -319,7 +371,7 @@ static void get_infos (const char *gname, int lmax, int *mmax, int *gpar1, ofs+=pring; } } - if (mytask==0) + if (verbose) { ptrdiff_t npix=get_npix(*ginfo); printf("Small Gauss grid, nlat=%d, npix=%ld, savings=%.2f%%\n", @@ -358,97 +410,83 @@ static void check_sign_scale(void) sharp_make_triangular_alm_info(lmax,mmax,1,&alms); ptrdiff_t nalms = ((mmax+1)*(mmax+2))/2 + (mmax+1)*(lmax-mmax); - for (int ntrans=1; ntrans<10; ++ntrans) - { - double **map; - ALLOC2D(map,double,2*ntrans,npix); + double **map; + ALLOC2D(map,double,2,npix); - dcmplx **alm; - ALLOC2D(alm,dcmplx,2*ntrans,nalms); - for (int i=0; i<2*ntrans; ++i) - for (int j=0; j=9,"usage: grid lmax mmax geom1 geom2 spin ntrans"); + UTIL_ASSERT(argc>=8,"usage: grid lmax mmax geom1 geom2 spin"); int lmax=atoi(argv[3]); int mmax=atoi(argv[4]); int gpar1=atoi(argv[5]); int gpar2=atoi(argv[6]); int spin=atoi(argv[7]); - int ntrans=atoi(argv[8]); if (mytask==0) printf("Testing map analysis accuracy.\n"); - if (mytask==0) printf("spin=%d, ntrans=%d\n", spin, ntrans); + if (mytask==0) printf("spin=%d\n", spin); sharp_geom_info *ginfo; sharp_alm_info *ainfo; - get_infos (argv[2], lmax, &mmax, &gpar1, &gpar2, &ginfo, &ainfo); + get_infos (argv[2], lmax, &mmax, &gpar1, &gpar2, &ginfo, &ainfo, 1); - int ncomp = ntrans*((spin==0) ? 1 : 2); + int ncomp = (spin==0) ? 1 : 2; double t_a2m=1e30, t_m2a=1e30; unsigned long long op_a2m, op_m2a; double *err_abs,*err_rel; @@ -557,7 +598,7 @@ static void sharp_test (int argc, const char **argv) { ++nrpt; double ta2m2, tm2a2; - do_sht (ginfo, ainfo, spin, ntrans, 0, &err_abs, &err_rel, &ta2m2, &tm2a2, + do_sht (ginfo, ainfo, spin, &err_abs, &err_rel, &ta2m2, &tm2a2, &op_a2m, &op_m2a); if (ta2m2=9,"usage: grid lmax mmax geom1 geom2 spin ntrans"); - int lmax=atoi(argv[3]); - int mmax=atoi(argv[4]); - int gpar1=atoi(argv[5]); - int gpar2=atoi(argv[6]); - int spin=atoi(argv[7]); - int ntrans=atoi(argv[8]); - - if (mytask==0) printf("Testing map analysis accuracy.\n"); - if (mytask==0) printf("spin=%d, ntrans=%d\n", spin, ntrans); - - sharp_geom_info *ginfo; - sharp_alm_info *ainfo; - get_infos (argv[2], lmax, &mmax, &gpar1, &gpar2, &ginfo, &ainfo); - - double ta2m_auto=1e30, tm2a_auto=1e30, ta2m_min=1e30, tm2a_min=1e30; - unsigned long long opa2m_min=0, opm2a_min=0; - int nvmin_a2m=-1, nvmin_m2a=-1; - for (int nv=0; nv<=6; ++nv) - { - int ntries=0; - double tacc=0; - do - { - double t_a2m, t_m2a; - unsigned long long op_a2m, op_m2a; - double *err_abs,*err_rel; - do_sht (ginfo, ainfo, spin, ntrans, nv, &err_abs, &err_rel, - &t_a2m, &t_m2a, &op_a2m, &op_m2a); - - DEALLOC(err_abs); - DEALLOC(err_rel); - tacc+=t_a2m+t_m2a; - ++ntries; - if (nv==0) - { - if (t_a2m -#include "sharp_vecutil.h" + +#ifndef VLEN + +#if (defined(__AVX512F__)) +#define VLEN 8 +#elif (defined (__AVX__)) +#define VLEN 4 +#elif (defined (__SSE2__)) +#define VLEN 2 +#else +#define VLEN 1 +#endif + +#endif typedef double Ts; #if (VLEN==1) typedef double Tv; -typedef float Tv_s; typedef int Tm; -#define vadd(a,b) ((a)+(b)) -#define vadd_s(a,b) ((a)+(b)) -#define vaddeq(a,b) ((a)+=(b)) -#define vaddeq_mask(mask,a,b) if (mask) (a)+=(b); -#define vsub(a,b) ((a)-(b)) -#define vsub_s(a,b) ((a)-(b)) -#define vsubeq(a,b) ((a)-=(b)) -#define vsubeq_mask(mask,a,b) if (mask) (a)-=(b); -#define vmul(a,b) ((a)*(b)) -#define vmul_s(a,b) ((a)*(b)) -#define vmuleq(a,b) ((a)*=(b)) -#define vmuleq_mask(mask,a,b) if (mask) (a)*=(b); -#define vfmaeq(a,b,c) ((a)+=(b)*(c)) -#define vfmaeq_s(a,b,c) ((a)+=(b)*(c)) -#define vfmseq(a,b,c) ((a)-=(b)*(c)) -#define vfmaaeq(a,b,c,d,e) ((a)+=(b)*(c)+(d)*(e)) -#define vfmaseq(a,b,c,d,e) ((a)+=(b)*(c)-(d)*(e)) -#define vneg(a) (-(a)) #define vload(a) (a) -#define vload_s(a) (a) -#define vloadu(p) (*(p)) -#define vloadu_s(p) (*(p)) +#define vzero 0. +#define vone 1. + +#define vaddeq_mask(mask,a,b) if (mask) (a)+=(b); +#define vsubeq_mask(mask,a,b) if (mask) (a)-=(b); +#define vmuleq_mask(mask,a,b) if (mask) (a)*=(b); +#define vneg(a) (-(a)) #define vabs(a) fabs(a) #define vsqrt(a) sqrt(a) #define vlt(a,b) ((a)<(b)) @@ -72,16 +70,16 @@ typedef int Tm; #define vge(a,b) ((a)>=(b)) #define vne(a,b) ((a)!=(b)) #define vand_mask(a,b) ((a)&&(b)) -#define vstoreu(p, a) (*(p)=a) -#define vstoreu_s(p, a) (*(p)=a) - +#define vor_mask(a,b) ((a)||(b)) static inline Tv vmin (Tv a, Tv b) { return (ab) ? a : b; } - #define vanyTrue(a) (a) #define vallTrue(a) (a) -#define vzero 0. -#define vone 1. + +static inline void vhsum_cmplx_special (Tv a, Tv b, Tv c, Tv d, + _Complex double * restrict cc) + { cc[0] += a+_Complex_I*b; cc[1] += c+_Complex_I*d; } + #endif @@ -97,7 +95,6 @@ static inline Tv vmax (Tv a, Tv b) { return (a>b) ? a : b; } #endif typedef __m128d Tv; -typedef __m128 Tv_s; typedef __m128d Tm; #if defined(__SSE4_1__) @@ -106,110 +103,84 @@ typedef __m128d Tm; static inline Tv vblend__(Tv m, Tv a, Tv b) { return _mm_or_pd(_mm_and_pd(a,m),_mm_andnot_pd(m,b)); } #endif -#define vzero _mm_setzero_pd() -#define vone _mm_set1_pd(1.) - -#define vadd(a,b) _mm_add_pd(a,b) -#define vadd_s(a,b) _mm_add_ps(a,b) -#define vaddeq(a,b) a=_mm_add_pd(a,b) -#define vaddeq_mask(mask,a,b) a=_mm_add_pd(a,vblend__(mask,b,vzero)) -#define vsub(a,b) _mm_sub_pd(a,b) -#define vsub_s(a,b) _mm_sub_ps(a,b) -#define vsubeq(a,b) a=_mm_sub_pd(a,b) -#define vsubeq_mask(mask,a,b) a=_mm_sub_pd(a,vblend__(mask,b,vzero)) -#define vmul(a,b) _mm_mul_pd(a,b) -#define vmul_s(a,b) _mm_mul_ps(a,b) -#define vmuleq(a,b) a=_mm_mul_pd(a,b) -#define vmuleq_mask(mask,a,b) a=_mm_mul_pd(a,vblend__(mask,b,vone)) -#define vfmaeq(a,b,c) a=_mm_add_pd(a,_mm_mul_pd(b,c)) -#define vfmaeq_s(a,b,c) a=_mm_add_ps(a,_mm_mul_ps(b,c)) -#define vfmseq(a,b,c) a=_mm_sub_pd(a,_mm_mul_pd(b,c)) -#define vfmaaeq(a,b,c,d,e) \ - a=_mm_add_pd(a,_mm_add_pd(_mm_mul_pd(b,c),_mm_mul_pd(d,e))) -#define vfmaseq(a,b,c,d,e) \ - a=_mm_add_pd(a,_mm_sub_pd(_mm_mul_pd(b,c),_mm_mul_pd(d,e))) -#define vneg(a) _mm_xor_pd(_mm_set1_pd(-0.),a) #define vload(a) _mm_set1_pd(a) -#define vload_s(a) _mm_set1_ps(a) -#define vabs(a) _mm_andnot_pd(_mm_set1_pd(-0.),a) +#define vzero _mm_setzero_pd() +#define vone vload(1.) + +#define vaddeq_mask(mask,a,b) a+=vblend__(mask,b,vzero) +#define vsubeq_mask(mask,a,b) a-=vblend__(mask,b,vzero) +#define vmuleq_mask(mask,a,b) a*=vblend__(mask,b,vone) +#define vneg(a) _mm_xor_pd(vload(-0.),a) +#define vabs(a) _mm_andnot_pd(vload(-0.),a) #define vsqrt(a) _mm_sqrt_pd(a) #define vlt(a,b) _mm_cmplt_pd(a,b) #define vgt(a,b) _mm_cmpgt_pd(a,b) #define vge(a,b) _mm_cmpge_pd(a,b) #define vne(a,b) _mm_cmpneq_pd(a,b) #define vand_mask(a,b) _mm_and_pd(a,b) +#define vor_mask(a,b) _mm_or_pd(a,b) #define vmin(a,b) _mm_min_pd(a,b) #define vmax(a,b) _mm_max_pd(a,b); #define vanyTrue(a) (_mm_movemask_pd(a)!=0) #define vallTrue(a) (_mm_movemask_pd(a)==3) -#define vloadu(p) _mm_loadu_pd(p) -#define vloadu_s(p) _mm_loadu_ps(p) -#define vstoreu(p, v) _mm_storeu_pd(p, v) -#define vstoreu_s(p, v) _mm_storeu_ps(p, v) + +static inline void vhsum_cmplx_special (Tv a, Tv b, Tv c, + Tv d, _Complex double * restrict cc) + { + union {Tv v; _Complex double c; } u1, u2; +#if defined(__SSE3__) + u1.v = _mm_hadd_pd(a,b); u2.v=_mm_hadd_pd(c,d); +#else + u1.v = _mm_shuffle_pd(a,b,_MM_SHUFFLE2(0,1)) + + _mm_shuffle_pd(a,b,_MM_SHUFFLE2(1,0)); + u2.v = _mm_shuffle_pd(c,d,_MM_SHUFFLE2(0,1)) + + _mm_shuffle_pd(c,d,_MM_SHUFFLE2(1,0)); +#endif + cc[0]+=u1.c; cc[1]+=u2.c; + } #endif #if (VLEN==4) #include -#if (USE_FMA4) -#include -#endif typedef __m256d Tv; -typedef __m256 Tv_s; typedef __m256d Tm; #define vblend__(m,a,b) _mm256_blendv_pd(b,a,m) -#define vzero _mm256_setzero_pd() -#define vone _mm256_set1_pd(1.) - -#define vadd(a,b) _mm256_add_pd(a,b) -#define vadd_s(a,b) _mm256_add_ps(a,b) -#define vaddeq(a,b) a=_mm256_add_pd(a,b) -#define vaddeq_mask(mask,a,b) a=_mm256_add_pd(a,vblend__(mask,b,vzero)) -#define vsub(a,b) _mm256_sub_pd(a,b) -#define vsub_s(a,b) _mm256_sub_ps(a,b) -#define vsubeq(a,b) a=_mm256_sub_pd(a,b) -#define vsubeq_mask(mask,a,b) a=_mm256_sub_pd(a,vblend__(mask,b,vzero)) -#define vmul(a,b) _mm256_mul_pd(a,b) -#define vmul_s(a,b) _mm256_mul_ps(a,b) -#define vmuleq(a,b) a=_mm256_mul_pd(a,b) -#define vmuleq_mask(mask,a,b) a=_mm256_mul_pd(a,vblend__(mask,b,vone)) -#if (USE_FMA4) -#define vfmaeq(a,b,c) a=_mm256_macc_pd(b,c,a) -#define vfmaeq_s(a,b,c) a=_mm256_macc_ps(b,c,a) -#define vfmseq(a,b,c) a=_mm256_nmacc_pd(b,c,a) -#define vfmaaeq(a,b,c,d,e) a=_mm256_macc_pd(d,e,_mm256_macc_pd(b,c,a)) -#define vfmaseq(a,b,c,d,e) a=_mm256_nmacc_pd(d,e,_mm256_macc_pd(b,c,a)) -#else -#define vfmaeq(a,b,c) a=_mm256_add_pd(a,_mm256_mul_pd(b,c)) -#define vfmaeq_s(a,b,c) a=_mm256_add_ps(a,_mm256_mul_ps(b,c)) -#define vfmseq(a,b,c) a=_mm256_sub_pd(a,_mm256_mul_pd(b,c)) -#define vfmaaeq(a,b,c,d,e) \ - a=_mm256_add_pd(a,_mm256_add_pd(_mm256_mul_pd(b,c),_mm256_mul_pd(d,e))) -#define vfmaseq(a,b,c,d,e) \ - a=_mm256_add_pd(a,_mm256_sub_pd(_mm256_mul_pd(b,c),_mm256_mul_pd(d,e))) -#endif -#define vneg(a) _mm256_xor_pd(_mm256_set1_pd(-0.),a) #define vload(a) _mm256_set1_pd(a) -#define vload_s(a) _mm256_set1_ps(a) -#define vabs(a) _mm256_andnot_pd(_mm256_set1_pd(-0.),a) +#define vzero _mm256_setzero_pd() +#define vone vload(1.) + +#define vaddeq_mask(mask,a,b) a+=vblend__(mask,b,vzero) +#define vsubeq_mask(mask,a,b) a-=vblend__(mask,b,vzero) +#define vmuleq_mask(mask,a,b) a*=vblend__(mask,b,vone) +#define vneg(a) _mm256_xor_pd(vload(-0.),a) +#define vabs(a) _mm256_andnot_pd(vload(-0.),a) #define vsqrt(a) _mm256_sqrt_pd(a) #define vlt(a,b) _mm256_cmp_pd(a,b,_CMP_LT_OQ) #define vgt(a,b) _mm256_cmp_pd(a,b,_CMP_GT_OQ) #define vge(a,b) _mm256_cmp_pd(a,b,_CMP_GE_OQ) #define vne(a,b) _mm256_cmp_pd(a,b,_CMP_NEQ_OQ) #define vand_mask(a,b) _mm256_and_pd(a,b) +#define vor_mask(a,b) _mm256_or_pd(a,b) #define vmin(a,b) _mm256_min_pd(a,b) #define vmax(a,b) _mm256_max_pd(a,b) #define vanyTrue(a) (_mm256_movemask_pd(a)!=0) #define vallTrue(a) (_mm256_movemask_pd(a)==15) -#define vloadu(p) _mm256_loadu_pd(p) -#define vloadu_s(p) _mm256_loadu_ps(p) -#define vstoreu(p, v) _mm256_storeu_pd(p, v) -#define vstoreu_s(p, v) _mm256_storeu_ps(p, v) +static inline void vhsum_cmplx_special (Tv a, Tv b, Tv c, Tv d, + _Complex double * restrict cc) + { + Tv tmp1=_mm256_hadd_pd(a,b), tmp2=_mm256_hadd_pd(c,d); + Tv tmp3=_mm256_permute2f128_pd(tmp1,tmp2,49), + tmp4=_mm256_permute2f128_pd(tmp1,tmp2,32); + tmp1=tmp3+tmp4; + union {Tv v; _Complex double c[2]; } u; + u.v=tmp1; + cc[0]+=u.c[0]; cc[1]+=u.c[1]; + } #endif @@ -220,35 +191,33 @@ typedef __m256d Tm; typedef __m512d Tv; typedef __mmask8 Tm; -#define vadd(a,b) _mm512_add_pd(a,b) -#define vaddeq(a,b) a=_mm512_add_pd(a,b) -#define vaddeq_mask(mask,a,b) a=_mm512_mask_add_pd(a,mask,a,b); -#define vsub(a,b) _mm512_sub_pd(a,b) -#define vsubeq(a,b) a=_mm512_sub_pd(a,b) -#define vsubeq_mask(mask,a,b) a=_mm512_mask_sub_pd(a,mask,a,b); -#define vmul(a,b) _mm512_mul_pd(a,b) -#define vmuleq(a,b) a=_mm512_mul_pd(a,b) -#define vmuleq_mask(mask,a,b) a=_mm512_mask_mul_pd(a,mask,a,b); -#define vfmaeq(a,b,c) a=_mm512_fmadd_pd(b,c,a) -#define vfmseq(a,b,c) a=_mm512_fnmadd_pd(b,c,a) -#define vfmaaeq(a,b,c,d,e) a=_mm512_fmadd_pd(d,e,_mm512_fmadd_pd(b,c,a)) -#define vfmaseq(a,b,c,d,e) a=_mm512_fnmadd_pd(d,e,_mm512_fmadd_pd(b,c,a)) -#define vneg(a) _mm512_mul_pd(a,_mm512_set1_pd(-1.)) #define vload(a) _mm512_set1_pd(a) -#define vabs(a) (__m512d)_mm512_andnot_epi64((__m512i)_mm512_set1_pd(-0.),(__m512i)a) +#define vzero _mm512_setzero_pd() +#define vone vload(1.) + +#define vaddeq_mask(mask,a,b) a=_mm512_mask_add_pd(a,mask,a,b); +#define vsubeq_mask(mask,a,b) a=_mm512_mask_sub_pd(a,mask,a,b); +#define vmuleq_mask(mask,a,b) a=_mm512_mask_mul_pd(a,mask,a,b); +#define vneg(a) _mm512_mul_pd(a,vload(-1.)) +#define vabs(a) (__m512d)_mm512_andnot_epi64((__m512i)vload(-0.),(__m512i)a) #define vsqrt(a) _mm512_sqrt_pd(a) -#define vlt(a,b) _mm512_cmplt_pd_mask(a,b) -#define vgt(a,b) _mm512_cmpnle_pd_mask(a,b) -#define vge(a,b) _mm512_cmpnlt_pd_mask(a,b) -#define vne(a,b) _mm512_cmpneq_pd_mask(a,b) +#define vlt(a,b) _mm512_cmp_pd_mask(a,b,_CMP_LT_OQ) +#define vgt(a,b) _mm512_cmp_pd_mask(a,b,_CMP_GT_OQ) +#define vge(a,b) _mm512_cmp_pd_mask(a,b,_CMP_GE_OQ) +#define vne(a,b) _mm512_cmp_pd_mask(a,b,_CMP_NEQ_OQ) #define vand_mask(a,b) ((a)&(b)) +#define vor_mask(a,b) ((a)|(b)) #define vmin(a,b) _mm512_min_pd(a,b) #define vmax(a,b) _mm512_max_pd(a,b) #define vanyTrue(a) (a!=0) #define vallTrue(a) (a==255) -#define vzero _mm512_setzero_pd() -#define vone _mm512_set1_pd(1.) +static inline void vhsum_cmplx_special (Tv a, Tv b, Tv c, Tv d, + _Complex double * restrict cc) + { + cc[0] += _mm512_reduce_add_pd(a)+_Complex_I*_mm512_reduce_add_pd(b); + cc[1] += _mm512_reduce_add_pd(c)+_Complex_I*_mm512_reduce_add_pd(d); + } #endif diff --git a/libsharp/sharp_vecutil.h b/libsharp/sharp_vecutil.h deleted file mode 100644 index f6161ca..0000000 --- a/libsharp/sharp_vecutil.h +++ /dev/null @@ -1,63 +0,0 @@ -/* - * This file is part of libc_utils. - * - * libc_utils is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * libc_utils is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with libc_utils; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/* - * libc_utils is being developed at the Max-Planck-Institut fuer Astrophysik - * and financially supported by the Deutsches Zentrum fuer Luft- und Raumfahrt - * (DLR). - */ - -/*! \file sharp_vecutil.h - * Functionality related to vector instruction support - * - * Copyright (C) 2012,2013 Max-Planck-Society - * \author Martin Reinecke - */ - -#ifndef SHARP_VECUTIL_H -#define SHARP_VECUTIL_H - -#ifndef VLEN - -#if (defined (__MIC__)) -#define VLEN 8 -#elif (defined (__AVX__)) -#define VLEN 4 -#elif (defined (__SSE2__)) -#define VLEN 2 -#else -#define VLEN 1 -#endif - -#endif - -#if (VLEN==1) -#define VLEN_s 1 -#else -#define VLEN_s (2*VLEN) -#endif - -#ifndef USE_FMA4 -#ifdef __FMA4__ -#define USE_FMA4 1 -#else -#define USE_FMA4 0 -#endif -#endif - -#endif diff --git a/libsharp/sharp_ylmgen_c.c b/libsharp/sharp_ylmgen_c.c index 6e8cee5..f408eea 100644 --- a/libsharp/sharp_ylmgen_c.c +++ b/libsharp/sharp_ylmgen_c.c @@ -25,7 +25,7 @@ /* * Helper code for efficient calculation of Y_lm(theta,phi=0) * - * Copyright (C) 2005-2014 Max-Planck-Society + * Copyright (C) 2005-2019 Max-Planck-Society * Author: Martin Reinecke */ @@ -59,35 +59,44 @@ void sharp_Ylmgen_init (sharp_Ylmgen_C *gen, int l_max, int m_max, int spin) gen->cf[m]=gen->cf[m+1]*sharp_fsmall; for (int m=-sharp_minscale+1; m<(sharp_maxscale-sharp_minscale+1); ++m) gen->cf[m]=gen->cf[m-1]*sharp_fbig; + gen->powlimit=RALLOC(double,m_max+spin+1); + gen->powlimit[0]=0.; + const double ln2 = 0.6931471805599453094172321214581766; + const double expo=-400*ln2; + for (int m=1; m<=m_max+spin; ++m) + gen->powlimit[m]=exp(expo/m); gen->m = -1; if (spin==0) { - gen->rf = RALLOC(sharp_ylmgen_dbl2,gen->lmax+1); gen->mfac = RALLOC(double,gen->mmax+1); gen->mfac[0] = inv_sqrt4pi; for (int m=1; m<=gen->mmax; ++m) gen->mfac[m] = gen->mfac[m-1]*sqrt((2*m+1.)/(2*m)); - gen->root = RALLOC(double,2*gen->lmax+5); - gen->iroot = RALLOC(double,2*gen->lmax+5); - for (int m=0; m<2*gen->lmax+5; ++m) + gen->root = RALLOC(double,2*gen->lmax+8); + gen->iroot = RALLOC(double,2*gen->lmax+8); + for (int m=0; m<2*gen->lmax+8; ++m) { gen->root[m] = sqrt(m); gen->iroot[m] = (m==0) ? 0. : 1./gen->root[m]; } + gen->eps=RALLOC(double, gen->lmax+4); + gen->alpha=RALLOC(double, gen->lmax/2+2); + gen->coef=RALLOC(sharp_ylmgen_dbl2, gen->lmax/2+2); } else { gen->m=gen->mlo=gen->mhi=-1234567890; - ALLOC(gen->fx,sharp_ylmgen_dbl3,gen->lmax+2); - for (int m=0; mlmax+2; ++m) - gen->fx[m].f[0]=gen->fx[m].f[1]=gen->fx[m].f[2]=0.; - ALLOC(gen->inv,double,gen->lmax+1); + ALLOC(gen->coef,sharp_ylmgen_dbl2,gen->lmax+3); + for (int m=0; mlmax+3; ++m) + gen->coef[m].a=gen->coef[m].b=0.; + ALLOC(gen->alpha,double,gen->lmax+3); + ALLOC(gen->inv,double,gen->lmax+2); gen->inv[0]=0; - for (int m=1; mlmax+1; ++m) gen->inv[m]=1./m; - ALLOC(gen->flm1,double,2*gen->lmax+1); - ALLOC(gen->flm2,double,2*gen->lmax+1); - for (int m=0; m<2*gen->lmax+1; ++m) + for (int m=1; mlmax+2; ++m) gen->inv[m]=1./m; + ALLOC(gen->flm1,double,2*gen->lmax+3); + ALLOC(gen->flm2,double,2*gen->lmax+3); + for (int m=0; m<2*gen->lmax+3; ++m) { gen->flm1[m] = sqrt(1./(m+1.)); gen->flm2[m] = sqrt(m/(m+1.)); @@ -124,16 +133,18 @@ void sharp_Ylmgen_init (sharp_Ylmgen_C *gen, int l_max, int m_max, int spin) void sharp_Ylmgen_destroy (sharp_Ylmgen_C *gen) { DEALLOC(gen->cf); + DEALLOC(gen->powlimit); + DEALLOC(gen->alpha); + DEALLOC(gen->coef); if (gen->s==0) { - DEALLOC(gen->rf); DEALLOC(gen->mfac); DEALLOC(gen->root); DEALLOC(gen->iroot); + DEALLOC(gen->eps); } else { - DEALLOC(gen->fx); DEALLOC(gen->prefac); DEALLOC(gen->fscale); DEALLOC(gen->flm1); @@ -150,13 +161,20 @@ void sharp_Ylmgen_prepare (sharp_Ylmgen_C *gen, int m) if (gen->s==0) { - gen->rf[m].f[0] = gen->root[2*m+3]; - gen->rf[m].f[1] = 0.; - for (int l=m+1; l<=gen->lmax; ++l) + gen->eps[m] = 0.; + for (int l=m+1; llmax+4; ++l) + gen->eps[l] = gen->root[l+m]*gen->root[l-m] + *gen->iroot[2*l+1]*gen->iroot[2*l-1]; + gen->alpha[0] = 1./gen->eps[m+1]; + gen->alpha[1] = gen->eps[m+1]/(gen->eps[m+2]*gen->eps[m+3]); + for (int il=1, l=m+2; llmax+1; ++il, l+=2) + gen->alpha[il+1]= ((il&1) ? -1 : 1) + /(gen->eps[l+2]*gen->eps[l+3]*gen->alpha[il]); + for (int il=0, l=m; llmax+2; ++il, l+=2) { - double tmp=gen->root[2*l+3]*gen->iroot[l+1+m]*gen->iroot[l+1-m]; - gen->rf[l].f[0] = tmp*gen->root[2*l+1]; - gen->rf[l].f[1] = tmp*gen->root[l+m]*gen->root[l-m]*gen->iroot[2*l-1]; + gen->coef[il].a = ((il&1) ? -1 : 1)*gen->alpha[il]*gen->alpha[il]; + double t1 = gen->eps[l+2], t2 = gen->eps[l+1]; + gen->coef[il].b = -gen->coef[il].a*(t1*t1+t2*t2); } } else @@ -169,17 +187,25 @@ void sharp_Ylmgen_prepare (sharp_Ylmgen_C *gen, int m) if (!ms_similar) { - for (int l=gen->mhi; llmax; ++l) + gen->alpha[gen->mhi] = 1.; + gen->coef[gen->mhi].a = gen->coef[gen->mhi].b = 0.; + for (int l=gen->mhi; llmax+1; ++l) { double t = gen->flm1[l+gen->m]*gen->flm1[l-gen->m] *gen->flm1[l+gen->s]*gen->flm1[l-gen->s]; double lt = 2*l+1; double l1 = l+1; - gen->fx[l+1].f[0]=l1*lt*t; - gen->fx[l+1].f[1]=gen->m*gen->s*gen->inv[l]*gen->inv[l+1]; + double flp10=l1*lt*t; + double flp11=gen->m*gen->s*gen->inv[l]*gen->inv[l+1]; t = gen->flm2[l+gen->m]*gen->flm2[l-gen->m] *gen->flm2[l+gen->s]*gen->flm2[l-gen->s]; - gen->fx[l+1].f[2]=t*l1*gen->inv[l]; + double flp12=t*l1*gen->inv[l]; + if (l>gen->mhi) + gen->alpha[l+1] = gen->alpha[l-1]*flp12; + else + gen->alpha[l+1] = 1.; + gen->coef[l+1].a = flp10*gen->alpha[l]/gen->alpha[l+1]; + gen->coef[l+1].b = flp11*gen->coef[l+1].a; } } diff --git a/libsharp/sharp_ylmgen_c.h b/libsharp/sharp_ylmgen_c.h index 3328f76..130d797 100644 --- a/libsharp/sharp_ylmgen_c.h +++ b/libsharp/sharp_ylmgen_c.h @@ -25,7 +25,7 @@ /*! \file sharp_ylmgen_c.h * Code for efficient calculation of Y_lm(phi=0,theta) * - * Copyright (C) 2005-2012 Max-Planck-Society + * Copyright (C) 2005-2019 Max-Planck-Society * \author Martin Reinecke */ @@ -41,27 +41,28 @@ static const double sharp_fbig=0x1p+800,sharp_fsmall=0x1p-800; static const double sharp_ftol=0x1p-60; static const double sharp_fbighalf=0x1p+400; -typedef struct { double f[2]; } sharp_ylmgen_dbl2; -typedef struct { double f[3]; } sharp_ylmgen_dbl3; +typedef struct { double a, b; } sharp_ylmgen_dbl2; typedef struct { /* for public use; immutable during lifetime */ int lmax, mmax, s; double *cf; + double *powlimit; /* for public use; will typically change after call to Ylmgen_prepare() */ int m; + double *alpha; + sharp_ylmgen_dbl2 *coef; + /* used if s==0 */ - double *mfac; - sharp_ylmgen_dbl2 *rf; + double *mfac, *eps; /* used if s!=0 */ int sinPow, cosPow, preMinus_p, preMinus_m; double *prefac; int *fscale; - sharp_ylmgen_dbl3 *fx; /* internal usage only */ /* used if s==0 */ diff --git a/pocketfft/pocketfft.c b/pocketfft/pocketfft.c new file mode 100644 index 0000000..de1af3e --- /dev/null +++ b/pocketfft/pocketfft.c @@ -0,0 +1,2190 @@ +/* + * This file is part of pocketfft. + * Licensed under a 3-clause BSD style license - see LICENSE.md + */ + +/* + * Main implementation file. + * + * Copyright (C) 2004-2018 Max-Planck-Society + * \author Martin Reinecke + */ + +#include +#include + +#include "pocketfft.h" + +#define RALLOC(type,num) \ + ((type *)malloc((num)*sizeof(type))) +#define DEALLOC(ptr) \ + do { free(ptr); (ptr)=NULL; } while(0) + +#define SWAP(a,b,type) \ + do { type tmp_=(a); (a)=(b); (b)=tmp_; } while(0) + +#ifdef __GNUC__ +#define NOINLINE __attribute__((noinline)) +#define WARN_UNUSED_RESULT __attribute__ ((warn_unused_result)) +#else +#define NOINLINE +#define WARN_UNUSED_RESULT +#endif + +// adapted from https://stackoverflow.com/questions/42792939/ +// CAUTION: this function only works for arguments in the range [-0.25; 0.25]! +static void my_sincosm1pi (double a, double *restrict res) + { + double s = a * a; + /* Approximate cos(pi*x)-1 for x in [-0.25,0.25] */ + double r = -1.0369917389758117e-4; + r = fma (r, s, 1.9294935641298806e-3); + r = fma (r, s, -2.5806887942825395e-2); + r = fma (r, s, 2.3533063028328211e-1); + r = fma (r, s, -1.3352627688538006e+0); + r = fma (r, s, 4.0587121264167623e+0); + r = fma (r, s, -4.9348022005446790e+0); + double c = r*s; + /* Approximate sin(pi*x) for x in [-0.25,0.25] */ + r = 4.6151442520157035e-4; + r = fma (r, s, -7.3700183130883555e-3); + r = fma (r, s, 8.2145868949323936e-2); + r = fma (r, s, -5.9926452893214921e-1); + r = fma (r, s, 2.5501640398732688e+0); + r = fma (r, s, -5.1677127800499516e+0); + s = s * a; + r = r * s; + s = fma (a, 3.1415926535897931e+0, r); + res[0] = c; + res[1] = s; + } + +NOINLINE static void calc_first_octant(size_t den, double * restrict res) + { + size_t n = (den+4)>>3; + if (n==0) return; + res[0]=1.; res[1]=0.; + if (n==1) return; + size_t l1=(size_t)sqrt(n); + for (size_t i=1; in) end = n-start; + for (size_t i=1; i>2; + size_t i=0, idx1=0, idx2=2*ndone-2; + for (; i+1>1; + double * p = res+n-1; + calc_first_octant(n<<2, p); + int i4=0, in=n, i=0; + for (; i4<=in-i4; ++i, i4+=4) // octant 0 + { + res[2*i] = p[2*i4]; res[2*i+1] = p[2*i4+1]; + } + for (; i4-in <= 0; ++i, i4+=4) // octant 1 + { + int xm = in-i4; + res[2*i] = p[2*xm+1]; res[2*i+1] = p[2*xm]; + } + for (; i4<=3*in-i4; ++i, i4+=4) // octant 2 + { + int xm = i4-in; + res[2*i] = -p[2*xm+1]; res[2*i+1] = p[2*xm]; + } + for (; i>2; + if ((n&7)==0) + res[quart] = res[quart+1] = hsqt2; + for (size_t i=2, j=2*quart-2; i>1; + if ((n&3)==0) + for (size_t i=0; i>1))<<1)==n) + { res=2; n=tmp; } + + size_t limit=(size_t)sqrt(n+0.01); + for (size_t x=3; x<=limit; x+=2) + while (((tmp=(n/x))*x)==n) + { + res=x; + n=tmp; + limit=(size_t)sqrt(n+0.01); + } + if (n>1) res=n; + + return res; + } + +NOINLINE static double cost_guess (size_t n) + { + const double lfp=1.1; // penalty for non-hardcoded larger factors + size_t ni=n; + double result=0.; + size_t tmp; + while (((tmp=(n>>1))<<1)==n) + { result+=2; n=tmp; } + + size_t limit=(size_t)sqrt(n+0.01); + for (size_t x=3; x<=limit; x+=2) + while ((tmp=(n/x))*x==n) + { + result+= (x<=5) ? x : lfp*x; // penalize larger prime factors + n=tmp; + limit=(size_t)sqrt(n+0.01); + } + if (n>1) result+=(n<=5) ? n : lfp*n; + + return result*ni; + } + +/* returns the smallest composite of 2, 3, 5, 7 and 11 which is >= n */ +NOINLINE static size_t good_size(size_t n) + { + if (n<=6) return n; + + size_t bestfac=2*n; + for (size_t f2=1; f2=n) bestfac=f235711; + return bestfac; + } + +typedef struct cmplx { + double r,i; +} cmplx; + +#define NFCT 25 +typedef struct cfftp_fctdata + { + size_t fct; + cmplx *tw, *tws; + } cfftp_fctdata; + +typedef struct cfftp_plan_i + { + size_t length, nfct; + cmplx *mem; + cfftp_fctdata fct[NFCT]; + } cfftp_plan_i; +typedef struct cfftp_plan_i * cfftp_plan; + +#define PMC(a,b,c,d) { a.r=c.r+d.r; a.i=c.i+d.i; b.r=c.r-d.r; b.i=c.i-d.i; } +#define ADDC(a,b,c) { a.r=b.r+c.r; a.i=b.i+c.i; } +#define SCALEC(a,b) { a.r*=b; a.i*=b; } +#define ROT90(a) { double tmp_=a.r; a.r=-a.i; a.i=tmp_; } +#define ROTM90(a) { double tmp_=-a.r; a.r=a.i; a.i=tmp_; } +#define CH(a,b,c) ch[(a)+ido*((b)+l1*(c))] +#define CC(a,b,c) cc[(a)+ido*((b)+cdim*(c))] +#define WA(x,i) wa[(i)-1+(x)*(ido-1)] +/* a = b*c */ +#define A_EQ_B_MUL_C(a,b,c) { a.r=b.r*c.r-b.i*c.i; a.i=b.r*c.i+b.i*c.r; } +/* a = conj(b)*c*/ +#define A_EQ_CB_MUL_C(a,b,c) { a.r=b.r*c.r+b.i*c.i; a.i=b.r*c.i-b.i*c.r; } + +#define PMSIGNC(a,b,c,d) { a.r=c.r+sign*d.r; a.i=c.i+sign*d.i; b.r=c.r-sign*d.r; b.i=c.i-sign*d.i; } +/* a = b*c */ +#define MULPMSIGNC(a,b,c) { a.r=b.r*c.r-sign*b.i*c.i; a.i=b.r*c.i+sign*b.i*c.r; } +/* a *= b */ +#define MULPMSIGNCEQ(a,b) { double xtmp=a.r; a.r=b.r*a.r-sign*b.i*a.i; a.i=b.r*a.i+sign*b.i*xtmp; } + +NOINLINE static void pass2b (size_t ido, size_t l1, const cmplx * restrict cc, + cmplx * restrict ch, const cmplx * restrict wa) + { + const size_t cdim=2; + + if (ido==1) + for (size_t k=0; kip) iwal-=ip; + cmplx xwal=wal[iwal]; + iwal+=l; if (iwal>ip) iwal-=ip; + cmplx xwal2=wal[iwal]; + for (size_t ik=0; ikip) iwal-=ip; + cmplx xwal=wal[iwal]; + for (size_t ik=0; iklength==1) return 0; + size_t len=plan->length; + size_t l1=1, nf=plan->nfct; + cmplx *ch = RALLOC(cmplx, len); + if (!ch) return -1; + cmplx *p1=c, *p2=ch; + + for(size_t k1=0; k1fct[k1].fct; + size_t l2=ip*l1; + size_t ido = len/l2; + if (ip==4) + sign>0 ? pass4b (ido, l1, p1, p2, plan->fct[k1].tw) + : pass4f (ido, l1, p1, p2, plan->fct[k1].tw); + else if(ip==2) + sign>0 ? pass2b (ido, l1, p1, p2, plan->fct[k1].tw) + : pass2f (ido, l1, p1, p2, plan->fct[k1].tw); + else if(ip==3) + sign>0 ? pass3b (ido, l1, p1, p2, plan->fct[k1].tw) + : pass3f (ido, l1, p1, p2, plan->fct[k1].tw); + else if(ip==5) + sign>0 ? pass5b (ido, l1, p1, p2, plan->fct[k1].tw) + : pass5f (ido, l1, p1, p2, plan->fct[k1].tw); + else if(ip==7) pass7 (ido, l1, p1, p2, plan->fct[k1].tw, sign); + else if(ip==11) pass11(ido, l1, p1, p2, plan->fct[k1].tw, sign); + else + { + if (passg(ido, ip, l1, p1, p2, plan->fct[k1].tw, plan->fct[k1].tws, sign)) + { DEALLOC(ch); return -1; } + SWAP(p1,p2,cmplx *); + } + SWAP(p1,p2,cmplx *); + l1=l2; + } + if (p1!=c) + { + if (fct!=1.) + for (size_t i=0; ilength; + size_t nfct=0; + while ((length%4)==0) + { if (nfct>=NFCT) return -1; plan->fct[nfct++].fct=4; length>>=2; } + if ((length%2)==0) + { + length>>=1; + // factor 2 should be at the front of the factor list + if (nfct>=NFCT) return -1; + plan->fct[nfct++].fct=2; + SWAP(plan->fct[0].fct, plan->fct[nfct-1].fct,size_t); + } + size_t maxl=(size_t)(sqrt((double)length))+1; + for (size_t divisor=3; (length>1)&&(divisor=NFCT) return -1; + plan->fct[nfct++].fct=divisor; + length/=divisor; + } + maxl=(size_t)(sqrt((double)length))+1; + } + if (length>1) plan->fct[nfct++].fct=length; + plan->nfct=nfct; + return 0; + } + +NOINLINE static size_t cfftp_twsize (cfftp_plan plan) + { + size_t twsize=0, l1=1; + for (size_t k=0; knfct; ++k) + { + size_t ip=plan->fct[k].fct, ido= plan->length/(l1*ip); + twsize+=(ip-1)*(ido-1); + if (ip>11) + twsize+=ip; + l1*=ip; + } + return twsize; + } + +NOINLINE WARN_UNUSED_RESULT static int cfftp_comp_twiddle (cfftp_plan plan) + { + size_t length=plan->length; + double *twid = RALLOC(double, 2*length); + if (!twid) return -1; + sincos_2pibyn(length, twid); + size_t l1=1; + size_t memofs=0; + for (size_t k=0; knfct; ++k) + { + size_t ip=plan->fct[k].fct, ido= length/(l1*ip); + plan->fct[k].tw=plan->mem+memofs; + memofs+=(ip-1)*(ido-1); + for (size_t j=1; jfct[k].tw[(j-1)*(ido-1)+i-1].r = twid[2*j*l1*i]; + plan->fct[k].tw[(j-1)*(ido-1)+i-1].i = twid[2*j*l1*i+1]; + } + if (ip>11) + { + plan->fct[k].tws=plan->mem+memofs; + memofs+=ip; + for (size_t j=0; jfct[k].tws[j].r = twid[2*j*l1*ido]; + plan->fct[k].tws[j].i = twid[2*j*l1*ido+1]; + } + } + l1*=ip; + } + DEALLOC(twid); + return 0; + } + +static cfftp_plan make_cfftp_plan (size_t length) + { + if (length==0) return NULL; + cfftp_plan plan = RALLOC(cfftp_plan_i,1); + if (!plan) return NULL; + plan->length=length; + plan->nfct=0; + for (size_t i=0; ifct[i]=(cfftp_fctdata){0,0,0}; + plan->mem=0; + if (length==1) return plan; + if (cfftp_factorize(plan)!=0) { DEALLOC(plan); return NULL; } + size_t tws=cfftp_twsize(plan); + plan->mem=RALLOC(cmplx,tws); + if (!plan->mem) { DEALLOC(plan); return NULL; } + if (cfftp_comp_twiddle(plan)!=0) + { DEALLOC(plan->mem); DEALLOC(plan); return NULL; } + return plan; + } + +static void destroy_cfftp_plan (cfftp_plan plan) + { + DEALLOC(plan->mem); + DEALLOC(plan); + } + +typedef struct rfftp_fctdata + { + size_t fct; + double *tw, *tws; + } rfftp_fctdata; + +typedef struct rfftp_plan_i + { + size_t length, nfct; + double *mem; + rfftp_fctdata fct[NFCT]; + } rfftp_plan_i; +typedef struct rfftp_plan_i * rfftp_plan; + +#define WA(x,i) wa[(i)+(x)*(ido-1)] +#define PM(a,b,c,d) { a=c+d; b=c-d; } +/* (a+ib) = conj(c+id) * (e+if) */ +#define MULPM(a,b,c,d,e,f) { a=c*e+d*f; b=c*f-d*e; } + +#define CC(a,b,c) cc[(a)+ido*((b)+l1*(c))] +#define CH(a,b,c) ch[(a)+ido*((b)+cdim*(c))] + +NOINLINE static void radf2 (size_t ido, size_t l1, const double * restrict cc, + double * restrict ch, const double * restrict wa) + { + const size_t cdim=2; + + for (size_t k=0; k1) + { + for (size_t j=1, jc=ip-1; j=ip) iang-=ip; + double ar1=csarr[2*iang], ai1=csarr[2*iang+1]; + iang+=l; if (iang>=ip) iang-=ip; + double ar2=csarr[2*iang], ai2=csarr[2*iang+1]; + iang+=l; if (iang>=ip) iang-=ip; + double ar3=csarr[2*iang], ai3=csarr[2*iang+1]; + iang+=l; if (iang>=ip) iang-=ip; + double ar4=csarr[2*iang], ai4=csarr[2*iang+1]; + for (size_t ik=0; ik=ip) iang-=ip; + double ar1=csarr[2*iang], ai1=csarr[2*iang+1]; + iang+=l; if (iang>=ip) iang-=ip; + double ar2=csarr[2*iang], ai2=csarr[2*iang+1]; + for (size_t ik=0; ik=ip) iang-=ip; + double ar=csarr[2*iang], ai=csarr[2*iang+1]; + for (size_t ik=0; ikip) iang-=ip; + double ar1=csarr[2*iang], ai1=csarr[2*iang+1]; + iang+=l; if(iang>ip) iang-=ip; + double ar2=csarr[2*iang], ai2=csarr[2*iang+1]; + iang+=l; if(iang>ip) iang-=ip; + double ar3=csarr[2*iang], ai3=csarr[2*iang+1]; + iang+=l; if(iang>ip) iang-=ip; + double ar4=csarr[2*iang], ai4=csarr[2*iang+1]; + for (size_t ik=0; ikip) iang-=ip; + double ar1=csarr[2*iang], ai1=csarr[2*iang+1]; + iang+=l; if(iang>ip) iang-=ip; + double ar2=csarr[2*iang], ai2=csarr[2*iang+1]; + for (size_t ik=0; ikip) iang-=ip; + double war=csarr[2*iang], wai=csarr[2*iang+1]; + for (size_t ik=0; iklength==1) return 0; + size_t n=plan->length; + size_t l1=n, nf=plan->nfct; + double *ch = RALLOC(double, n); + if (!ch) return -1; + double *p1=c, *p2=ch; + + for(size_t k1=0; k1fct[k].fct; + size_t ido=n / l1; + l1 /= ip; + if(ip==4) + radf4(ido, l1, p1, p2, plan->fct[k].tw); + else if(ip==2) + radf2(ido, l1, p1, p2, plan->fct[k].tw); + else if(ip==3) + radf3(ido, l1, p1, p2, plan->fct[k].tw); + else if(ip==5) + radf5(ido, l1, p1, p2, plan->fct[k].tw); + else + { + radfg(ido, ip, l1, p1, p2, plan->fct[k].tw, plan->fct[k].tws); + SWAP (p1,p2,double *); + } + SWAP (p1,p2,double *); + } + copy_and_norm(c,p1,n,fct); + DEALLOC(ch); + return 0; + } + +WARN_UNUSED_RESULT +static int rfftp_backward(rfftp_plan plan, double c[], double fct) + { + if (plan->length==1) return 0; + size_t n=plan->length; + size_t l1=1, nf=plan->nfct; + double *ch = RALLOC(double, n); + if (!ch) return -1; + double *p1=c, *p2=ch; + + for(size_t k=0; kfct[k].fct, + ido= n/(ip*l1); + if(ip==4) + radb4(ido, l1, p1, p2, plan->fct[k].tw); + else if(ip==2) + radb2(ido, l1, p1, p2, plan->fct[k].tw); + else if(ip==3) + radb3(ido, l1, p1, p2, plan->fct[k].tw); + else if(ip==5) + radb5(ido, l1, p1, p2, plan->fct[k].tw); + else + radbg(ido, ip, l1, p1, p2, plan->fct[k].tw, plan->fct[k].tws); + SWAP (p1,p2,double *); + l1*=ip; + } + copy_and_norm(c,p1,n,fct); + DEALLOC(ch); + return 0; + } + +WARN_UNUSED_RESULT +static int rfftp_factorize (rfftp_plan plan) + { + size_t length=plan->length; + size_t nfct=0; + while ((length%4)==0) + { if (nfct>=NFCT) return -1; plan->fct[nfct++].fct=4; length>>=2; } + if ((length%2)==0) + { + length>>=1; + // factor 2 should be at the front of the factor list + if (nfct>=NFCT) return -1; + plan->fct[nfct++].fct=2; + SWAP(plan->fct[0].fct, plan->fct[nfct-1].fct,size_t); + } + size_t maxl=(size_t)(sqrt((double)length))+1; + for (size_t divisor=3; (length>1)&&(divisor=NFCT) return -1; + plan->fct[nfct++].fct=divisor; + length/=divisor; + } + maxl=(size_t)(sqrt((double)length))+1; + } + if (length>1) plan->fct[nfct++].fct=length; + plan->nfct=nfct; + return 0; + } + +static size_t rfftp_twsize(rfftp_plan plan) + { + size_t twsize=0, l1=1; + for (size_t k=0; knfct; ++k) + { + size_t ip=plan->fct[k].fct, ido= plan->length/(l1*ip); + twsize+=(ip-1)*(ido-1); + if (ip>5) twsize+=2*ip; + l1*=ip; + } + return twsize; + return 0; + } + +WARN_UNUSED_RESULT NOINLINE static int rfftp_comp_twiddle (rfftp_plan plan) + { + size_t length=plan->length; + double *twid = RALLOC(double, 2*length); + if (!twid) return -1; + sincos_2pibyn_half(length, twid); + size_t l1=1; + double *ptr=plan->mem; + for (size_t k=0; knfct; ++k) + { + size_t ip=plan->fct[k].fct, ido=length/(l1*ip); + if (knfct-1) // last factor doesn't need twiddles + { + plan->fct[k].tw=ptr; ptr+=(ip-1)*(ido-1); + for (size_t j=1; jfct[k].tw[(j-1)*(ido-1)+2*i-2] = twid[2*j*l1*i]; + plan->fct[k].tw[(j-1)*(ido-1)+2*i-1] = twid[2*j*l1*i+1]; + } + } + if (ip>5) // special factors required by *g functions + { + plan->fct[k].tws=ptr; ptr+=2*ip; + plan->fct[k].tws[0] = 1.; + plan->fct[k].tws[1] = 0.; + for (size_t i=1; i<=(ip>>1); ++i) + { + plan->fct[k].tws[2*i ] = twid[2*i*(length/ip)]; + plan->fct[k].tws[2*i+1] = twid[2*i*(length/ip)+1]; + plan->fct[k].tws[2*(ip-i) ] = twid[2*i*(length/ip)]; + plan->fct[k].tws[2*(ip-i)+1] = -twid[2*i*(length/ip)+1]; + } + } + l1*=ip; + } + DEALLOC(twid); + return 0; + } + +NOINLINE static rfftp_plan make_rfftp_plan (size_t length) + { + if (length==0) return NULL; + rfftp_plan plan = RALLOC(rfftp_plan_i,1); + if (!plan) return NULL; + plan->length=length; + plan->nfct=0; + plan->mem=NULL; + for (size_t i=0; ifct[i]=(rfftp_fctdata){0,0,0}; + if (length==1) return plan; + if (rfftp_factorize(plan)!=0) { DEALLOC(plan); return NULL; } + size_t tws=rfftp_twsize(plan); + plan->mem=RALLOC(double,tws); + if (!plan->mem) { DEALLOC(plan); return NULL; } + if (rfftp_comp_twiddle(plan)!=0) + { DEALLOC(plan->mem); DEALLOC(plan); return NULL; } + return plan; + } + +NOINLINE static void destroy_rfftp_plan (rfftp_plan plan) + { + DEALLOC(plan->mem); + DEALLOC(plan); + } + +typedef struct fftblue_plan_i + { + size_t n, n2; + cfftp_plan plan; + double *mem; + double *bk, *bkf; + } fftblue_plan_i; +typedef struct fftblue_plan_i * fftblue_plan; + +NOINLINE static fftblue_plan make_fftblue_plan (size_t length) + { + fftblue_plan plan = RALLOC(fftblue_plan_i,1); + if (!plan) return NULL; + plan->n = length; + plan->n2 = good_size(plan->n*2-1); + plan->mem = RALLOC(double, 2*plan->n+2*plan->n2); + if (!plan->mem) { DEALLOC(plan); return NULL; } + plan->bk = plan->mem; + plan->bkf = plan->bk+2*plan->n; + +/* initialize b_k */ + double *tmp = RALLOC(double,4*plan->n); + if (!tmp) { DEALLOC(plan->mem); DEALLOC(plan); return NULL; } + sincos_2pibyn(2*plan->n,tmp); + plan->bk[0] = 1; + plan->bk[1] = 0; + + size_t coeff=0; + for (size_t m=1; mn; ++m) + { + coeff+=2*m-1; + if (coeff>=2*plan->n) coeff-=2*plan->n; + plan->bk[2*m ] = tmp[2*coeff ]; + plan->bk[2*m+1] = tmp[2*coeff+1]; + } + + /* initialize the zero-padded, Fourier transformed b_k. Add normalisation. */ + double xn2 = 1./plan->n2; + plan->bkf[0] = plan->bk[0]*xn2; + plan->bkf[1] = plan->bk[1]*xn2; + for (size_t m=2; m<2*plan->n; m+=2) + { + plan->bkf[m] = plan->bkf[2*plan->n2-m] = plan->bk[m] *xn2; + plan->bkf[m+1] = plan->bkf[2*plan->n2-m+1] = plan->bk[m+1] *xn2; + } + for (size_t m=2*plan->n;m<=(2*plan->n2-2*plan->n+1);++m) + plan->bkf[m]=0.; + plan->plan=make_cfftp_plan(plan->n2); + if (!plan->plan) + { DEALLOC(tmp); DEALLOC(plan->mem); DEALLOC(plan); return NULL; } + if (cfftp_forward(plan->plan,plan->bkf,1.)!=0) + { DEALLOC(tmp); DEALLOC(plan->mem); DEALLOC(plan); return NULL; } + DEALLOC(tmp); + + return plan; + } + +NOINLINE static void destroy_fftblue_plan (fftblue_plan plan) + { + DEALLOC(plan->mem); + destroy_cfftp_plan(plan->plan); + DEALLOC(plan); + } + +NOINLINE WARN_UNUSED_RESULT +static int fftblue_fft(fftblue_plan plan, double c[], int isign, double fct) + { + size_t n=plan->n; + size_t n2=plan->n2; + double *bk = plan->bk; + double *bkf = plan->bkf; + double *akf = RALLOC(double, 2*n2); + if (!akf) return -1; + +/* initialize a_k and FFT it */ + if (isign>0) + for (size_t m=0; m<2*n; m+=2) + { + akf[m] = c[m]*bk[m] - c[m+1]*bk[m+1]; + akf[m+1] = c[m]*bk[m+1] + c[m+1]*bk[m]; + } + else + for (size_t m=0; m<2*n; m+=2) + { + akf[m] = c[m]*bk[m] + c[m+1]*bk[m+1]; + akf[m+1] =-c[m]*bk[m+1] + c[m+1]*bk[m]; + } + for (size_t m=2*n; m<2*n2; ++m) + akf[m]=0; + + if (cfftp_forward (plan->plan,akf,fct)!=0) + { DEALLOC(akf); return -1; } + +/* do the convolution */ + if (isign>0) + for (size_t m=0; m<2*n2; m+=2) + { + double im = -akf[m]*bkf[m+1] + akf[m+1]*bkf[m]; + akf[m ] = akf[m]*bkf[m] + akf[m+1]*bkf[m+1]; + akf[m+1] = im; + } + else + for (size_t m=0; m<2*n2; m+=2) + { + double im = akf[m]*bkf[m+1] + akf[m+1]*bkf[m]; + akf[m ] = akf[m]*bkf[m] - akf[m+1]*bkf[m+1]; + akf[m+1] = im; + } + +/* inverse FFT */ + if (cfftp_backward (plan->plan,akf,1.)!=0) + { DEALLOC(akf); return -1; } + +/* multiply by b_k */ + if (isign>0) + for (size_t m=0; m<2*n; m+=2) + { + c[m] = bk[m] *akf[m] - bk[m+1]*akf[m+1]; + c[m+1] = bk[m+1]*akf[m] + bk[m] *akf[m+1]; + } + else + for (size_t m=0; m<2*n; m+=2) + { + c[m] = bk[m] *akf[m] + bk[m+1]*akf[m+1]; + c[m+1] =-bk[m+1]*akf[m] + bk[m] *akf[m+1]; + } + DEALLOC(akf); + return 0; + } + +WARN_UNUSED_RESULT +static int cfftblue_backward(fftblue_plan plan, double c[], double fct) + { return fftblue_fft(plan,c,1,fct); } + +WARN_UNUSED_RESULT +static int cfftblue_forward(fftblue_plan plan, double c[], double fct) + { return fftblue_fft(plan,c,-1,fct); } + +WARN_UNUSED_RESULT +static int rfftblue_backward(fftblue_plan plan, double c[], double fct) + { + size_t n=plan->n; + double *tmp = RALLOC(double,2*n); + if (!tmp) return -1; + tmp[0]=c[0]; + tmp[1]=0.; + memcpy (tmp+2,c+1, (n-1)*sizeof(double)); + if ((n&1)==0) tmp[n+1]=0.; + for (size_t m=2; mn; + double *tmp = RALLOC(double,2*n); + if (!tmp) return -1; + for (size_t m=0; mblueplan=0; + plan->packplan=0; + if ((length<50) || (largest_prime_factor(length)<=sqrt(length))) + { + plan->packplan=make_cfftp_plan(length); + if (!plan->packplan) { DEALLOC(plan); return NULL; } + return plan; + } + double comp1 = cost_guess(length); + double comp2 = 2*cost_guess(good_size(2*length-1)); + comp2*=1.5; /* fudge factor that appears to give good overall performance */ + if (comp2blueplan=make_fftblue_plan(length); + if (!plan->blueplan) { DEALLOC(plan); return NULL; } + } + else + { + plan->packplan=make_cfftp_plan(length); + if (!plan->packplan) { DEALLOC(plan); return NULL; } + } + return plan; + } + +void destroy_cfft_plan (cfft_plan plan) + { + if (plan->blueplan) + destroy_fftblue_plan(plan->blueplan); + if (plan->packplan) + destroy_cfftp_plan(plan->packplan); + DEALLOC(plan); + } + +WARN_UNUSED_RESULT int cfft_backward(cfft_plan plan, double c[], double fct) + { + if (plan->packplan) + return cfftp_backward(plan->packplan,c,fct); + // if (plan->blueplan) + return cfftblue_backward(plan->blueplan,c,fct); + } + +WARN_UNUSED_RESULT int cfft_forward(cfft_plan plan, double c[], double fct) + { + if (plan->packplan) + return cfftp_forward(plan->packplan,c,fct); + // if (plan->blueplan) + return cfftblue_forward(plan->blueplan,c,fct); + } + +typedef struct rfft_plan_i + { + rfftp_plan packplan; + fftblue_plan blueplan; + } rfft_plan_i; + +rfft_plan make_rfft_plan (size_t length) + { + if (length==0) return NULL; + rfft_plan plan = RALLOC(rfft_plan_i,1); + if (!plan) return NULL; + plan->blueplan=0; + plan->packplan=0; + if ((length<50) || (largest_prime_factor(length)<=sqrt(length))) + { + plan->packplan=make_rfftp_plan(length); + if (!plan->packplan) { DEALLOC(plan); return NULL; } + return plan; + } + double comp1 = 0.5*cost_guess(length); + double comp2 = 2*cost_guess(good_size(2*length-1)); + comp2*=1.5; /* fudge factor that appears to give good overall performance */ + if (comp2blueplan=make_fftblue_plan(length); + if (!plan->blueplan) { DEALLOC(plan); return NULL; } + } + else + { + plan->packplan=make_rfftp_plan(length); + if (!plan->packplan) { DEALLOC(plan); return NULL; } + } + return plan; + } + +void destroy_rfft_plan (rfft_plan plan) + { + if (plan->blueplan) + destroy_fftblue_plan(plan->blueplan); + if (plan->packplan) + destroy_rfftp_plan(plan->packplan); + DEALLOC(plan); + } + +size_t rfft_length(rfft_plan plan) + { + if (plan->packplan) return plan->packplan->length; + return plan->blueplan->n; + } + +size_t cfft_length(cfft_plan plan) + { + if (plan->packplan) return plan->packplan->length; + return plan->blueplan->n; + } + +WARN_UNUSED_RESULT int rfft_backward(rfft_plan plan, double c[], double fct) + { + if (plan->packplan) + return rfftp_backward(plan->packplan,c,fct); + else // if (plan->blueplan) + return rfftblue_backward(plan->blueplan,c,fct); + } + +WARN_UNUSED_RESULT int rfft_forward(rfft_plan plan, double c[], double fct) + { + if (plan->packplan) + return rfftp_forward(plan->packplan,c,fct); + else // if (plan->blueplan) + return rfftblue_forward(plan->blueplan,c,fct); + } diff --git a/pocketfft/pocketfft.h b/pocketfft/pocketfft.h new file mode 100644 index 0000000..9eb3985 --- /dev/null +++ b/pocketfft/pocketfft.h @@ -0,0 +1,34 @@ +/* + * This file is part of pocketfft. + * Licensed under a 3-clause BSD style license - see LICENSE.md + */ + +/*! \file pocketfft.h + * Public interface of the pocketfft library + * + * Copyright (C) 2008-2018 Max-Planck-Society + * \author Martin Reinecke + */ + +#ifndef POCKETFFT_H +#define POCKETFFT_H + +#include + +struct cfft_plan_i; +typedef struct cfft_plan_i * cfft_plan; +cfft_plan make_cfft_plan (size_t length); +void destroy_cfft_plan (cfft_plan plan); +int cfft_backward(cfft_plan plan, double c[], double fct); +int cfft_forward(cfft_plan plan, double c[], double fct); +size_t cfft_length(cfft_plan plan); + +struct rfft_plan_i; +typedef struct rfft_plan_i * rfft_plan; +rfft_plan make_rfft_plan (size_t length); +void destroy_rfft_plan (rfft_plan plan); +int rfft_backward(rfft_plan plan, double c[], double fct); +int rfft_forward(rfft_plan plan, double c[], double fct); +size_t rfft_length(rfft_plan plan); + +#endif diff --git a/python/fake_pyrex/Pyrex/Distutils/__init__.py b/python/fake_pyrex/Pyrex/Distutils/__init__.py deleted file mode 100644 index 51c8e16..0000000 --- a/python/fake_pyrex/Pyrex/Distutils/__init__.py +++ /dev/null @@ -1 +0,0 @@ -# work around broken setuptools monkey patching diff --git a/python/fake_pyrex/Pyrex/Distutils/build_ext.py b/python/fake_pyrex/Pyrex/Distutils/build_ext.py deleted file mode 100644 index 4f846f6..0000000 --- a/python/fake_pyrex/Pyrex/Distutils/build_ext.py +++ /dev/null @@ -1 +0,0 @@ -build_ext = "yes, it's there!" diff --git a/python/fake_pyrex/Pyrex/__init__.py b/python/fake_pyrex/Pyrex/__init__.py deleted file mode 100644 index 51c8e16..0000000 --- a/python/fake_pyrex/Pyrex/__init__.py +++ /dev/null @@ -1 +0,0 @@ -# work around broken setuptools monkey patching diff --git a/python/fake_pyrex/README b/python/fake_pyrex/README deleted file mode 100644 index cf3f3ff..0000000 --- a/python/fake_pyrex/README +++ /dev/null @@ -1,2 +0,0 @@ -This directory is here to fool setuptools into building .pyx files -even if Pyrex is not installed. See ../setup.py. \ No newline at end of file diff --git a/python/libsharp/__init__.py b/python/libsharp/__init__.py deleted file mode 100644 index dd0fa41..0000000 --- a/python/libsharp/__init__.py +++ /dev/null @@ -1 +0,0 @@ -from .libsharp import * diff --git a/python/libsharp/libsharp.pxd b/python/libsharp/libsharp.pxd deleted file mode 100644 index 27a4608..0000000 --- a/python/libsharp/libsharp.pxd +++ /dev/null @@ -1,92 +0,0 @@ -cdef extern from "sharp.h": - - void sharp_legendre_transform_s(float *bl, float *recfac, ptrdiff_t lmax, float *x, - float *out, ptrdiff_t nx) - void sharp_legendre_transform(double *bl, double *recfac, ptrdiff_t lmax, double *x, - double *out, ptrdiff_t nx) - void sharp_legendre_transform_recfac(double *r, ptrdiff_t lmax) - void sharp_legendre_transform_recfac_s(float *r, ptrdiff_t lmax) - void sharp_legendre_roots(int n, double *x, double *w) - - # sharp_lowlevel.h - ctypedef struct sharp_alm_info: - # Maximum \a l index of the array - int lmax - # Number of different \a m values in this object - int nm - # Array with \a nm entries containing the individual m values - int *mval - # Combination of flags from sharp_almflags - int flags - # Array with \a nm entries containing the (hypothetical) indices of - # the coefficients with quantum numbers 0,\a mval[i] - long *mvstart - # Stride between a_lm and a_(l+1),m - long stride - - ctypedef struct sharp_geom_info: - pass - - void sharp_make_alm_info (int lmax, int mmax, int stride, - ptrdiff_t *mvstart, sharp_alm_info **alm_info) - - void sharp_make_geom_info (int nrings, int *nph, ptrdiff_t *ofs, - int *stride, double *phi0, double *theta, - double *wgt, sharp_geom_info **geom_info) - - void sharp_destroy_alm_info(sharp_alm_info *info) - void sharp_destroy_geom_info(sharp_geom_info *info) - - ptrdiff_t sharp_map_size(sharp_geom_info *info) - ptrdiff_t sharp_alm_count(sharp_alm_info *self) - - - ctypedef enum sharp_jobtype: - SHARP_YtW - SHARP_Yt - SHARP_WY - SHARP_Y - - ctypedef enum: - SHARP_DP - SHARP_ADD - - void sharp_execute(sharp_jobtype type_, - int spin, - void *alm, - void *map, - sharp_geom_info *geom_info, - sharp_alm_info *alm_info, - int ntrans, - int flags, - double *time, - unsigned long long *opcnt) nogil - - ctypedef enum: - SHARP_ERROR_NO_MPI - - int sharp_execute_mpi_maybe (void *pcomm, sharp_jobtype type, int spin, - void *alm, void *map, sharp_geom_info *geom_info, - sharp_alm_info *alm_info, int ntrans, int flags, double *time, - unsigned long long *opcnt) nogil - - void sharp_normalized_associated_legendre_table(int m, int spin, int lmax, int ntheta, - double *theta, int theta_stride, int l_stride, int spin_stride, double *out) nogil - - -cdef extern from "sharp_geomhelpers.h": - void sharp_make_subset_healpix_geom_info( - int nside, int stride, int nrings, - int *rings, double *weight, sharp_geom_info **geom_info) - void sharp_make_gauss_geom_info( - int nrings, int nphi, double phi0, - int stride_lon, int stride_lat, sharp_geom_info **geom_info) - -cdef extern from "sharp_almhelpers.h": - void sharp_make_triangular_alm_info (int lmax, int mmax, int stride, - sharp_alm_info **alm_info) - void sharp_make_rectangular_alm_info (int lmax, int mmax, int stride, - sharp_alm_info **alm_info) - void sharp_make_mmajor_real_packed_alm_info (int lmax, int stride, - int nm, const int *ms, sharp_alm_info **alm_info) - diff --git a/python/libsharp/libsharp.pyx b/python/libsharp/libsharp.pyx deleted file mode 100644 index dfefc93..0000000 --- a/python/libsharp/libsharp.pyx +++ /dev/null @@ -1,324 +0,0 @@ -import numpy as np -cimport numpy as np -cimport cython - -__all__ = ['legendre_transform', 'legendre_roots', 'sht', 'synthesis', 'adjoint_synthesis', - 'analysis', 'adjoint_analysis', 'healpix_grid', 'triangular_order', 'rectangular_order', - 'packed_real_order', 'normalized_associated_legendre_table'] - - -def legendre_transform(x, bl, out=None): - if out is None: - out = np.empty_like(x) - if out.shape[0] == 0: - return out - elif x.dtype == np.float64: - if bl.dtype != np.float64: - bl = bl.astype(np.float64) - return _legendre_transform(x, bl, out=out) - elif x.dtype == np.float32: - if bl.dtype != np.float32: - bl = bl.astype(np.float32) - return _legendre_transform_s(x, bl, out=out) - else: - raise ValueError("unsupported dtype") - - -def _legendre_transform(double[::1] x, double[::1] bl, double[::1] out): - if out.shape[0] != x.shape[0]: - raise ValueError('x and out must have same shape') - sharp_legendre_transform(&bl[0], NULL, bl.shape[0] - 1, &x[0], &out[0], x.shape[0]) - return np.asarray(out) - - -def _legendre_transform_s(float[::1] x, float[::1] bl, float[::1] out): - if out.shape[0] != x.shape[0]: - raise ValueError('x and out must have same shape') - sharp_legendre_transform_s(&bl[0], NULL, bl.shape[0] - 1, &x[0], &out[0], x.shape[0]) - return np.asarray(out) - - -def legendre_roots(n): - x = np.empty(n, np.double) - w = np.empty(n, np.double) - cdef double[::1] x_buf = x, w_buf = w - if not (x_buf.shape[0] == w_buf.shape[0] == n): - raise AssertionError() - if n > 0: - sharp_legendre_roots(n, &x_buf[0], &w_buf[0]) - return x, w - - -JOBTYPE_TO_CONST = { - 'Y': SHARP_Y, - 'Yt': SHARP_Yt, - 'WY': SHARP_WY, - 'YtW': SHARP_YtW -} - -def sht(jobtype, geom_info ginfo, alm_info ainfo, double[:, :, ::1] input, - int spin=0, comm=None, add=False): - cdef void *comm_ptr - cdef int flags = SHARP_DP | (SHARP_ADD if add else 0) - cdef int r - cdef sharp_jobtype jobtype_i - cdef double[:, :, ::1] output_buf - cdef int ntrans = input.shape[0] - cdef int ntotcomp = ntrans * input.shape[1] - cdef int i, j - - if spin == 0 and input.shape[1] != 1: - raise ValueError('For spin == 0, we need input.shape[1] == 1') - elif spin != 0 and input.shape[1] != 2: - raise ValueError('For spin != 0, we need input.shape[1] == 2') - - - cdef size_t[::1] ptrbuf = np.empty(2 * ntotcomp, dtype=np.uintp) - cdef double **alm_ptrs = &ptrbuf[0] - cdef double **map_ptrs = &ptrbuf[ntotcomp] - - try: - jobtype_i = JOBTYPE_TO_CONST[jobtype] - except KeyError: - raise ValueError('jobtype must be one of: %s' % ', '.join(sorted(JOBTYPE_TO_CONST.keys()))) - - if jobtype_i == SHARP_Y or jobtype_i == SHARP_WY: - output = np.empty((input.shape[0], input.shape[1], ginfo.local_size()), dtype=np.float64) - output_buf = output - for i in range(input.shape[0]): - for j in range(input.shape[1]): - alm_ptrs[i * input.shape[1] + j] = &input[i, j, 0] - map_ptrs[i * input.shape[1] + j] = &output_buf[i, j, 0] - else: - output = np.empty((input.shape[0], input.shape[1], ainfo.local_size()), dtype=np.float64) - output_buf = output - for i in range(input.shape[0]): - for j in range(input.shape[1]): - alm_ptrs[i * input.shape[1] + j] = &output_buf[i, j, 0] - map_ptrs[i * input.shape[1] + j] = &input[i, j, 0] - - if comm is None: - with nogil: - sharp_execute ( - jobtype_i, - geom_info=ginfo.ginfo, alm_info=ainfo.ainfo, - spin=spin, alm=alm_ptrs, map=map_ptrs, - ntrans=ntrans, flags=flags, time=NULL, opcnt=NULL) - else: - from mpi4py import MPI - if not isinstance(comm, MPI.Comm): - raise TypeError('comm must be an mpi4py communicator') - from .libsharp_mpi import _addressof - comm_ptr = _addressof(comm) - with nogil: - r = sharp_execute_mpi_maybe ( - comm_ptr, jobtype_i, - geom_info=ginfo.ginfo, alm_info=ainfo.ainfo, - spin=spin, alm=alm_ptrs, map=map_ptrs, - ntrans=ntrans, flags=flags, time=NULL, opcnt=NULL) - if r == SHARP_ERROR_NO_MPI: - raise Exception('MPI requested, but not available') - - return output - - -def synthesis(*args, **kw): - return sht('Y', *args, **kw) - -def adjoint_synthesis(*args, **kw): - return sht('Yt', *args, **kw) - -def analysis(*args, **kw): - return sht('YtW', *args, **kw) - -def adjoint_analysis(*args, **kw): - return sht('WY', *args, **kw) - - -# -# geom_info -# -class NotInitializedError(Exception): - pass - - -cdef class geom_info: - cdef sharp_geom_info *ginfo - - def __cinit__(self, *args, **kw): - self.ginfo = NULL - - def local_size(self): - if self.ginfo == NULL: - raise NotInitializedError() - return sharp_map_size(self.ginfo) - - def __dealloc__(self): - if self.ginfo != NULL: - sharp_destroy_geom_info(self.ginfo) - self.ginfo = NULL - - -cdef class healpix_grid(geom_info): - - _weight_cache = {} # { (nside, 'T'/'Q'/'U') -> numpy array of ring weights cached from file } - - def __init__(self, int nside, stride=1, int[::1] rings=None, double[::1] weights=None): - if weights is not None and weights.shape[0] != 2 * nside: - raise ValueError('weights must have length 2 * nside') - sharp_make_subset_healpix_geom_info(nside, stride, - nrings=4 * nside - 1 if rings is None else rings.shape[0], - rings=NULL if rings is None else &rings[0], - weight=NULL if weights is None else &weights[0], - geom_info=&self.ginfo) - - @classmethod - def load_ring_weights(cls, nside, fields): - """ - Loads HEALPix ring weights from file. The environment variable - HEALPIX should be set, and this routine will look in the `data` - subdirectory. - - Parameters - ---------- - - nside: int - HEALPix nside parameter - - fields: tuple of str - Which weights to extract; pass ('T',) to only get scalar - weights back, or ('T', 'Q', 'U') to get all the weights - - Returns - ------- - - List of NumPy arrays, according to fields parameter. - - """ - import os - from astropy.io import fits - data_path = os.path.join(os.environ['HEALPIX'], 'data') - fits_field_names = { - 'T': 'TEMPERATURE WEIGHTS', - 'Q': 'Q-POLARISATION WEIGHTS', - 'U': 'U-POLARISATION WEIGHTS'} - - must_load = [field for field in fields if (nside, field) not in cls._weight_cache] - - if must_load: - hdulist = fits.open(os.path.join(data_path, 'weight_ring_n%05d.fits' % nside)) - try: - for field in must_load: - w = hdulist[1].data.field(fits_field_names[field]).ravel().astype(np.double) - w += 1 - cls._weight_cache[nside, field] = w - finally: - hdulist.close() - return [cls._weight_cache[(nside, field)].copy() for field in fields] - -# -# alm_info -# - - -cdef class alm_info: - cdef sharp_alm_info *ainfo - - def __cinit__(self, *args, **kw): - self.ainfo = NULL - - def local_size(self): - if self.ainfo == NULL: - raise NotInitializedError() - return sharp_alm_count(self.ainfo) - - def mval(self): - if self.ainfo == NULL: - raise NotInitializedError() - return np.asarray( self.ainfo.mval) - - def mvstart(self): - if self.ainfo == NULL: - raise NotInitializedError() - return np.asarray( self.ainfo.mvstart) - - def __dealloc__(self): - if self.ainfo != NULL: - sharp_destroy_alm_info(self.ainfo) - self.ainfo = NULL - - @cython.boundscheck(False) - def almxfl(self, np.ndarray[double, ndim=3, mode='c'] alm, np.ndarray[double, ndim=2, mode='c'] fl): - """Multiply Alm by a Ell based array - - - Parameters - ---------- - alm : np.ndarray - input alm, 3 dimensions = (different signal x polarizations x lm-ordering) - fl : np.ndarray - either 1 dimension, e.g. gaussian beam, or 2 dimensions e.g. a polarized beam - - Returns - ------- - None, it modifies alms in-place - - """ - cdef int mvstart = 0 - cdef bint has_multiple_beams = alm.shape[2] > 1 and fl.shape[1] > 1 - cdef int f, i_m, m, num_ells, i_l, i_signal, i_pol, i_mv - - for i_m in range(self.ainfo.nm): - m = self.ainfo.mval[i_m] - f = 1 if (m==0) else 2 - num_ells = self.ainfo.lmax + 1 - m - - if not has_multiple_beams: - for i_signal in range(alm.shape[0]): - for i_pol in range(alm.shape[1]): - for i_l in range(num_ells): - l = m + i_l - for i_mv in range(mvstart + f*i_l, mvstart + f*i_l +f): - alm[i_signal, i_pol, i_mv] *= fl[l, 0] - else: - for i_signal in range(alm.shape[0]): - for i_pol in range(alm.shape[1]): - for i_l in range(num_ells): - l = m + i_l - for i_mv in range(mvstart + f*i_l, mvstart + f*i_l +f): - alm[i_signal, i_pol, i_mv] *= fl[l, i_pol] - mvstart += f * num_ells - -cdef class triangular_order(alm_info): - def __init__(self, int lmax, mmax=None, stride=1): - mmax = mmax if mmax is not None else lmax - sharp_make_triangular_alm_info(lmax, mmax, stride, &self.ainfo) - - -cdef class rectangular_order(alm_info): - def __init__(self, int lmax, mmax=None, stride=1): - mmax = mmax if mmax is not None else lmax - sharp_make_rectangular_alm_info(lmax, mmax, stride, &self.ainfo) - - -cdef class packed_real_order(alm_info): - def __init__(self, int lmax, stride=1, int[::1] ms=None): - sharp_make_mmajor_real_packed_alm_info(lmax=lmax, stride=stride, - nm=lmax + 1 if ms is None else ms.shape[0], - ms=NULL if ms is None else &ms[0], - alm_info=&self.ainfo) - -# -# -# - -@cython.boundscheck(False) -def normalized_associated_legendre_table(int lmax, int m, theta): - cdef double[::1] theta_ = np.ascontiguousarray(theta, dtype=np.double) - out = np.zeros((theta_.shape[0], lmax - m + 1), np.double) - cdef double[:, ::1] out_ = out - if lmax < m: - raise ValueError("lmax < m") - with nogil: - sharp_normalized_associated_legendre_table(m, 0, lmax, theta_.shape[0], &theta_[0], lmax - m + 1, 1, 1, &out_[0,0]) - return out diff --git a/python/libsharp/libsharp_mpi.pyx b/python/libsharp/libsharp_mpi.pyx deleted file mode 100644 index e819a77..0000000 --- a/python/libsharp/libsharp_mpi.pyx +++ /dev/null @@ -1,17 +0,0 @@ -cdef extern from "mpi.h": - ctypedef void *MPI_Comm - -cdef extern from "Python.h": - object PyLong_FromVoidPtr(void*) - -cdef extern: - ctypedef class mpi4py.MPI.Comm [object PyMPICommObject]: - cdef MPI_Comm ob_mpi - cdef unsigned flags - -# For compatibility with mpi4py <= 1.3.1 -# Newer versions could use the MPI._addressof function -def _addressof(Comm comm): - cdef void *ptr = NULL - ptr = &comm.ob_mpi - return PyLong_FromVoidPtr(ptr) diff --git a/python/libsharp/tests/__init__.py b/python/libsharp/tests/__init__.py deleted file mode 100644 index 1bb8bf6..0000000 --- a/python/libsharp/tests/__init__.py +++ /dev/null @@ -1 +0,0 @@ -# empty diff --git a/python/libsharp/tests/test_legendre.py b/python/libsharp/tests/test_legendre.py deleted file mode 100644 index 0129b29..0000000 --- a/python/libsharp/tests/test_legendre.py +++ /dev/null @@ -1,58 +0,0 @@ -import numpy as np -from scipy.special import legendre -from scipy.special import p_roots -import libsharp - -from numpy.testing import assert_allclose - - -def check_legendre_transform(lmax, ntheta): - l = np.arange(lmax + 1) - if lmax >= 1: - sigma = -np.log(1e-3) / lmax / (lmax + 1) - bl = np.exp(-sigma*l*(l+1)) - bl *= (2 * l + 1) - else: - bl = np.asarray([1], dtype=np.double) - - theta = np.linspace(0, np.pi, ntheta, endpoint=True) - x = np.cos(theta) - - # Compute truth using scipy.special.legendre - P = np.zeros((ntheta, lmax + 1)) - for l in range(lmax + 1): - P[:, l] = legendre(l)(x) - y0 = np.dot(P, bl) - - - # double-precision - y = libsharp.legendre_transform(x, bl) - - assert_allclose(y, y0, rtol=1e-12, atol=1e-12) - - # single-precision - y32 = libsharp.legendre_transform(x.astype(np.float32), bl) - assert_allclose(y, y0, rtol=1e-5, atol=1e-5) - - -def test_legendre_transform(): - nthetas_to_try = [0, 9, 17, 19] + list(np.random.randint(500, size=20)) - for ntheta in nthetas_to_try: - for lmax in [0, 1, 2, 3, 20] + list(np.random.randint(50, size=4)): - yield check_legendre_transform, lmax, ntheta - -def check_legendre_roots(n): - xs, ws = ([], []) if n == 0 else p_roots(n) # from SciPy - xl, wl = libsharp.legendre_roots(n) - assert_allclose(xs, xl, rtol=1e-14, atol=1e-14) - assert_allclose(ws, wl, rtol=1e-14, atol=1e-14) - -def test_legendre_roots(): - """ - Test the Legendre root-finding algorithm from libsharp by comparing it with - the SciPy version. - """ - yield check_legendre_roots, 0 - yield check_legendre_roots, 1 - yield check_legendre_roots, 32 - yield check_legendre_roots, 33 diff --git a/python/libsharp/tests/test_legendre_table.py b/python/libsharp/tests/test_legendre_table.py deleted file mode 100644 index eb02df2..0000000 --- a/python/libsharp/tests/test_legendre_table.py +++ /dev/null @@ -1,36 +0,0 @@ -from __future__ import print_function -import numpy as np - -from numpy.testing import assert_almost_equal -from nose.tools import eq_, ok_ - -from libsharp import normalized_associated_legendre_table -from scipy.special import sph_harm, p_roots - -def test_compare_legendre_table_with_scipy(): - def test(theta, m, lmax): - Plm = normalized_associated_legendre_table(lmax, m, theta) - - Plm_p = sph_harm(m, np.arange(m, lmax + 1), 0, theta)[None, :] - if not np.allclose(Plm_p, Plm): - print(Plm_p) - print(Plm) - return ok_, np.allclose(Plm_p, Plm) - - yield test(np.pi/2, 0, 10) - yield test(np.pi/4, 0, 10) - yield test(3 * np.pi/4, 0, 10) - yield test(np.pi/4, 1, 4) - yield test(np.pi/4, 2, 4) - yield test(np.pi/4, 50, 50) - yield test(np.pi/2, 49, 50) - - -def test_legendre_table_wrapper_logic(): - # tests the SSE 2 logic in the high-level wrapper by using an odd number of thetas - theta = np.asarray([np.pi/2, np.pi/4, 3 * np.pi / 4]) - m = 3 - lmax = 10 - Plm = normalized_associated_legendre_table(lmax, m, theta) - assert np.allclose(Plm[1, :], normalized_associated_legendre_table(lmax, m, np.pi/4)[0, :]) - assert np.allclose(Plm[2, :], normalized_associated_legendre_table(lmax, m, 3 * np.pi/4)[0, :]) diff --git a/python/libsharp/tests/test_sht.py b/python/libsharp/tests/test_sht.py deleted file mode 100644 index 63ccf20..0000000 --- a/python/libsharp/tests/test_sht.py +++ /dev/null @@ -1,32 +0,0 @@ -import numpy as np -from numpy.testing import assert_allclose -import libsharp - -from mpi4py import MPI - - -def test_basic(): - lmax = 10 - nside = 8 - rank = MPI.COMM_WORLD.Get_rank() - ms = np.arange(rank, lmax + 1, MPI.COMM_WORLD.Get_size(), dtype=np.int32) - - order = libsharp.packed_real_order(lmax, ms=ms) - grid = libsharp.healpix_grid(nside) - - - alm = np.zeros(order.local_size()) - if rank == 0: - alm[0] = 1 - elif rank == 1: - alm[0] = 1 - - - map = libsharp.synthesis(grid, order, np.repeat(alm[None, None, :], 3, 0), comm=MPI.COMM_WORLD) - assert np.all(map[2, :] == map[1, :]) and np.all(map[1, :] == map[0, :]) - map = map[0, 0, :] - print(rank, "shape", map.shape) - print(rank, "mean", map.mean()) - -if __name__=="__main__": - test_basic() diff --git a/python/libsharp/tests/test_smoothing_noise_pol_mpi.py b/python/libsharp/tests/test_smoothing_noise_pol_mpi.py deleted file mode 100644 index 2cdff95..0000000 --- a/python/libsharp/tests/test_smoothing_noise_pol_mpi.py +++ /dev/null @@ -1,137 +0,0 @@ -# This test needs to be run with: - -# mpirun -np X python test_smoothing_noise_pol_mpi.py - -from mpi4py import MPI - -import numpy as np - -import healpy as hp - -import libsharp - -mpi = True -rank = MPI.COMM_WORLD.Get_rank() - -nside = 256 -npix = hp.nside2npix(nside) - -np.random.seed(100) -input_map = np.random.normal(size=(3, npix)) -fwhm_deg = 10 -lmax = 512 - -nrings = 4 * nside - 1 # four missing pixels - -if rank == 0: - print("total rings", nrings) - -n_mpi_processes = MPI.COMM_WORLD.Get_size() -rings_per_process = nrings // n_mpi_processes + 1 -# ring indices are 1-based - -ring_indices_emisphere = np.arange(2*nside, dtype=np.int32) + 1 -local_ring_indices = ring_indices_emisphere[rank::n_mpi_processes] - -# to improve performance, simmetric rings north/south need to be in the same rank -# therefore we use symmetry to create the full ring indexing - -if local_ring_indices[-1] == 2 * nside: - # has equator ring - local_ring_indices = np.concatenate( - [local_ring_indices[:-1], - nrings - local_ring_indices[::-1] + 1] - ) -else: - # does not have equator ring - local_ring_indices = np.concatenate( - [local_ring_indices, - nrings - local_ring_indices[::-1] + 1] - ) - -print("rank", rank, "n_rings", len(local_ring_indices)) - -if not mpi: - local_ring_indices = None -grid = libsharp.healpix_grid(nside, rings=local_ring_indices) - -# returns start index of the ring and number of pixels -startpix, ringpix, _, _, _ = hp.ringinfo(nside, local_ring_indices.astype(np.int64)) - -local_npix = grid.local_size() - -def expand_pix(startpix, ringpix, local_npix): - """Turn first pixel index and number of pixel in full array of pixels - - to be optimized with cython or numba - """ - local_pix = np.empty(local_npix, dtype=np.int64) - i = 0 - for start, num in zip(startpix, ringpix): - local_pix[i:i+num] = np.arange(start, start+num) - i += num - return local_pix - -local_pix = expand_pix(startpix, ringpix, local_npix) - -local_map = input_map[:, local_pix] - -local_hitmap = np.zeros(npix) -local_hitmap[local_pix] = 1 -hp.write_map("hitmap_{}.fits".format(rank), local_hitmap, overwrite=True) - -print("rank", rank, "npix", npix, "local_npix", local_npix, "local_map len", len(local_map), "unique pix", len(np.unique(local_pix))) - -local_m_indices = np.arange(rank, lmax + 1, MPI.COMM_WORLD.Get_size(), dtype=np.int32) -if not mpi: - local_m_indices = None - -order = libsharp.packed_real_order(lmax, ms=local_m_indices) -local_nl = order.local_size() -print("rank", rank, "local_nl", local_nl, "mval", order.mval()) - -mpi_comm = MPI.COMM_WORLD if mpi else None - -# map2alm -# maps in libsharp are 3D, 2nd dimension is IQU, 3rd is pixel - -alm_sharp_I = libsharp.analysis(grid, order, - np.ascontiguousarray(local_map[0].reshape((1, 1, -1))), - spin=0, comm=mpi_comm) -alm_sharp_P = libsharp.analysis(grid, order, - np.ascontiguousarray(local_map[1:].reshape((1, 2, -1))), - spin=2, comm=mpi_comm) - -beam = hp.gauss_beam(fwhm=np.radians(fwhm_deg), lmax=lmax, pol=True) - -print("Smooth") -# smooth in place (zonca implemented this function) -order.almxfl(alm_sharp_I, np.ascontiguousarray(beam[:, 0:1])) -order.almxfl(alm_sharp_P, np.ascontiguousarray(beam[:, (1, 2)])) - -# alm2map - -new_local_map_I = libsharp.synthesis(grid, order, alm_sharp_I, spin=0, comm=mpi_comm) -new_local_map_P = libsharp.synthesis(grid, order, alm_sharp_P, spin=2, comm=mpi_comm) - -# Transfer map to first process for writing - -local_full_map = np.zeros(input_map.shape, dtype=np.float64) -local_full_map[0, local_pix] = new_local_map_I -local_full_map[1:, local_pix] = new_local_map_P - -output_map = np.zeros(input_map.shape, dtype=np.float64) if rank == 0 else None -mpi_comm.Reduce(local_full_map, output_map, root=0, op=MPI.SUM) - -if rank == 0: - # hp.write_map("sharp_smoothed_map.fits", output_map, overwrite=True) - # hp_smoothed = hp.alm2map(hp.map2alm(input_map, lmax=lmax), nside=nside) # transform only - hp_smoothed = hp.smoothing(input_map, fwhm=np.radians(fwhm_deg), lmax=lmax) - std_diff = (hp_smoothed-output_map).std() - print("Std of difference between libsharp and healpy", std_diff) - # hp.write_map( - # "healpy_smoothed_map.fits", - # hp_smoothed, - # overwrite=True - # ) - assert std_diff < 1e-5 diff --git a/python/setup.py b/python/setup.py deleted file mode 100644 index 788d7a6..0000000 --- a/python/setup.py +++ /dev/null @@ -1,83 +0,0 @@ -#! /usr/bin/env python - -descr = """Spherical Harmionic transforms package - -Python API for the libsharp spherical harmonic transforms library -""" - -import os -import sys - -DISTNAME = 'libsharp' -DESCRIPTION = 'libsharp library for fast Spherical Harmonic Transforms' -LONG_DESCRIPTION = descr -MAINTAINER = 'Dag Sverre Seljebotn', -MAINTAINER_EMAIL = 'd.s.seljebotn@astro.uio.no', -URL = 'http://sourceforge.net/projects/libsharp/' -LICENSE = 'GPL' -DOWNLOAD_URL = "http://sourceforge.net/projects/libsharp/" -VERSION = '0.1' - -# Add our fake Pyrex at the end of the Python search path -# in order to fool setuptools into allowing compilation of -# pyx files to C files. Importing Cython.Distutils then -# makes Cython the tool of choice for this rather than -# (the possibly nonexisting) Pyrex. -project_path = os.path.split(__file__)[0] -sys.path.append(os.path.join(project_path, 'fake_pyrex')) - -from setuptools import setup, find_packages, Extension -from Cython.Build import cythonize -import numpy as np - -libsharp = os.environ.get('LIBSHARP', None) -libsharp_include = os.environ.get('LIBSHARP_INCLUDE', libsharp and os.path.join(libsharp, 'include')) -libsharp_lib = os.environ.get('LIBSHARP_LIB', libsharp and os.path.join(libsharp, 'lib')) - -if libsharp_include is None or libsharp_lib is None: - sys.stderr.write('Please set LIBSHARP environment variable to the install directly of libsharp, ' - 'this script will refer to the lib and include sub-directories. Alternatively ' - 'set LIBSHARP_INCLUDE and LIBSHARP_LIB\n') - sys.exit(1) - -if __name__ == "__main__": - setup(install_requires = ['numpy'], - packages = find_packages(), - test_suite="nose.collector", - # Well, technically zipping the package will work, but since it's - # all compiled code it'll just get unzipped again at runtime, which - # is pointless: - zip_safe = False, - name = DISTNAME, - version = VERSION, - maintainer = MAINTAINER, - maintainer_email = MAINTAINER_EMAIL, - description = DESCRIPTION, - license = LICENSE, - url = URL, - download_url = DOWNLOAD_URL, - long_description = LONG_DESCRIPTION, - classifiers = - [ 'Development Status :: 3 - Alpha', - 'Environment :: Console', - 'Intended Audience :: Developers', - 'Intended Audience :: Science/Research', - 'License :: OSI Approved :: GNU General Public License (GPL)', - 'Topic :: Scientific/Engineering'], - ext_modules = cythonize([ - Extension("libsharp.libsharp", - ["libsharp/libsharp.pyx"], - libraries=["sharp", "fftpack", "c_utils"], - include_dirs=[libsharp_include, np.get_include()], - library_dirs=[libsharp_lib], - extra_link_args=["-fopenmp"], - ), - Extension("libsharp.libsharp_mpi", - ["libsharp/libsharp_mpi.pyx"], - libraries=["sharp", "fftpack", "c_utils"], - include_dirs=[libsharp_include, np.get_include()], - library_dirs=[libsharp_lib], - extra_link_args=["-fopenmp"], - ), - ]), - ) diff --git a/runjinja.py b/runjinja.py deleted file mode 100755 index fb06737..0000000 --- a/runjinja.py +++ /dev/null @@ -1,19 +0,0 @@ -#!/usr/bin/env python - -""" -Preprocesses foo.c.in to foo.c. Reads STDIN and writes STDOUT. -""" - -import sys -import hashlib -from jinja2 import Template, Environment - -env = Environment(block_start_string='/*{', - block_end_string='}*/', - variable_start_string='{{', - variable_end_string='}}') - -extra_vars = dict(len=len) -input = sys.stdin.read() -sys.stdout.write('/* DO NOT EDIT. md5sum of source: %s */' % hashlib.md5(input.encode()).hexdigest()) -sys.stdout.write(env.from_string(input).render(**extra_vars)) diff --git a/runtest.sh b/runtest.sh new file mode 100755 index 0000000..be291ec --- /dev/null +++ b/runtest.sh @@ -0,0 +1,4 @@ +#!/bin/sh + +./sharp_testsuite acctest +