mirror of
https://github.com/paboyle/Grid.git
synced 2025-04-03 02:35:55 +01:00
Fixed conflicts. PLEASE avoid making wholesale cosmetic only changes, this created
a HUGE amount of difficult to resolve and understand conflicts . Wholesale formatting, reordering functions etc... in a central file like Tensor_class or Grid_vector_types while others are also editing without making substantial functionality changes creates pain.
This commit is contained in:
paboyle
parent
f4dd5062d7
commit
4908b77d46
266
configure
vendored
266
configure
vendored
@ -626,6 +626,13 @@ ac_subst_vars='am__EXEEXT_FALSE
|
||||
am__EXEEXT_TRUE
|
||||
LTLIBOBJS
|
||||
LIBOBJS
|
||||
enable_latex_docs
|
||||
enable_html_docs
|
||||
enable_dot
|
||||
DOXYGEN_DOC_FALSE
|
||||
DOXYGEN_DOC_TRUE
|
||||
DOT
|
||||
DOXYGEN
|
||||
USE_LAPACK_LIB_FALSE
|
||||
USE_LAPACK_LIB_TRUE
|
||||
USE_LAPACK_FALSE
|
||||
@ -727,7 +734,6 @@ infodir
|
||||
docdir
|
||||
oldincludedir
|
||||
includedir
|
||||
runstatedir
|
||||
localstatedir
|
||||
sharedstatedir
|
||||
sysconfdir
|
||||
@ -760,6 +766,10 @@ enable_rng
|
||||
enable_timers
|
||||
enable_chroma
|
||||
enable_lapack
|
||||
enable_doxygen
|
||||
enable_dot
|
||||
enable_html_docs
|
||||
enable_latex_docs
|
||||
'
|
||||
ac_precious_vars='build_alias
|
||||
host_alias
|
||||
@ -811,7 +821,6 @@ datadir='${datarootdir}'
|
||||
sysconfdir='${prefix}/etc'
|
||||
sharedstatedir='${prefix}/com'
|
||||
localstatedir='${prefix}/var'
|
||||
runstatedir='${localstatedir}/run'
|
||||
includedir='${prefix}/include'
|
||||
oldincludedir='/usr/include'
|
||||
docdir='${datarootdir}/doc/${PACKAGE_TARNAME}'
|
||||
@ -1064,15 +1073,6 @@ do
|
||||
| -silent | --silent | --silen | --sile | --sil)
|
||||
silent=yes ;;
|
||||
|
||||
-runstatedir | --runstatedir | --runstatedi | --runstated \
|
||||
| --runstate | --runstat | --runsta | --runst | --runs \
|
||||
| --run | --ru | --r)
|
||||
ac_prev=runstatedir ;;
|
||||
-runstatedir=* | --runstatedir=* | --runstatedi=* | --runstated=* \
|
||||
| --runstate=* | --runstat=* | --runsta=* | --runst=* | --runs=* \
|
||||
| --run=* | --ru=* | --r=*)
|
||||
runstatedir=$ac_optarg ;;
|
||||
|
||||
-sbindir | --sbindir | --sbindi | --sbind | --sbin | --sbi | --sb)
|
||||
ac_prev=sbindir ;;
|
||||
-sbindir=* | --sbindir=* | --sbindi=* | --sbind=* | --sbin=* \
|
||||
@ -1210,7 +1210,7 @@ fi
|
||||
for ac_var in exec_prefix prefix bindir sbindir libexecdir datarootdir \
|
||||
datadir sysconfdir sharedstatedir localstatedir includedir \
|
||||
oldincludedir docdir infodir htmldir dvidir pdfdir psdir \
|
||||
libdir localedir mandir runstatedir
|
||||
libdir localedir mandir
|
||||
do
|
||||
eval ac_val=\$$ac_var
|
||||
# Remove trailing slashes.
|
||||
@ -1363,7 +1363,6 @@ Fine tuning of the installation directories:
|
||||
--sysconfdir=DIR read-only single-machine data [PREFIX/etc]
|
||||
--sharedstatedir=DIR modifiable architecture-independent data [PREFIX/com]
|
||||
--localstatedir=DIR modifiable single-machine data [PREFIX/var]
|
||||
--runstatedir=DIR modifiable per-process data [LOCALSTATEDIR/run]
|
||||
--libdir=DIR object code libraries [EPREFIX/lib]
|
||||
--includedir=DIR C header files [PREFIX/include]
|
||||
--oldincludedir=DIR C header files for non-gcc [/usr/include]
|
||||
@ -1421,6 +1420,11 @@ Optional Features:
|
||||
--enable-timers=yes|no Enable system dependent high res timers
|
||||
--enable-chroma Expect chroma compiled under c++11
|
||||
--enable-lapack Enable lapack yes/no
|
||||
--enable-doxygen enable documentation generation with doxygen (auto)
|
||||
--enable-dot use 'dot' to generate graphs in doxygen (auto)
|
||||
--enable-html-docs enable HTML generation with doxygen (yes)
|
||||
--enable-latex-docs enable LaTeX documentation generation with doxygen
|
||||
(no)
|
||||
|
||||
Some influential environment variables:
|
||||
CXX C++ compiler command
|
||||
@ -3171,6 +3175,15 @@ END
|
||||
fi
|
||||
|
||||
|
||||
ac_sources="lib"
|
||||
ac_dests="include/Grid "
|
||||
while test -n "$ac_sources"; do
|
||||
set $ac_dests; ac_dest=$1; shift; ac_dests=$*
|
||||
set $ac_sources; ac_source=$1; shift; ac_sources=$*
|
||||
ac_config_links_1="$ac_config_links_1 $ac_dest:$ac_source"
|
||||
done
|
||||
ac_config_links="$ac_config_links $ac_config_links_1"
|
||||
|
||||
|
||||
ac_config_headers="$ac_config_headers lib/Config.h"
|
||||
|
||||
@ -6672,7 +6685,6 @@ else
|
||||
ac_CHROMA=no
|
||||
fi
|
||||
|
||||
|
||||
case ${ac_CHROMA} in
|
||||
yes)
|
||||
echo Enabling tests regressing to Chroma
|
||||
@ -6737,15 +6749,160 @@ fi
|
||||
###################################################################
|
||||
# Checks for doxygen support
|
||||
# if present enables the "make doxyfile" command
|
||||
#echo
|
||||
#echo Checking doxygen support
|
||||
#echo :::::::::::::::::::::::::::::::::::::::::::
|
||||
#AC_PROG_DOXYGEN
|
||||
echo
|
||||
echo Checking doxygen support
|
||||
echo :::::::::::::::::::::::::::::::::::::::::::
|
||||
|
||||
#if test -n "$DOXYGEN"
|
||||
#then
|
||||
#AC_CONFIG_FILES([docs/doxy.cfg])
|
||||
#fi
|
||||
# Check whether --enable-doxygen was given.
|
||||
if test "${enable_doxygen+set}" = set; then :
|
||||
enableval=$enable_doxygen;
|
||||
fi
|
||||
|
||||
|
||||
# Check whether --enable-dot was given.
|
||||
if test "${enable_dot+set}" = set; then :
|
||||
enableval=$enable_dot;
|
||||
fi
|
||||
|
||||
# Check whether --enable-html-docs was given.
|
||||
if test "${enable_html_docs+set}" = set; then :
|
||||
enableval=$enable_html_docs;
|
||||
else
|
||||
enable_html_docs=yes
|
||||
fi
|
||||
|
||||
# Check whether --enable-latex-docs was given.
|
||||
if test "${enable_latex_docs+set}" = set; then :
|
||||
enableval=$enable_latex_docs;
|
||||
else
|
||||
enable_latex_docs=no
|
||||
fi
|
||||
|
||||
|
||||
if test "x$enable_doxygen" = xno; then
|
||||
enable_doc=no
|
||||
else
|
||||
# Extract the first word of "doxygen", so it can be a program name with args.
|
||||
set dummy doxygen; ac_word=$2
|
||||
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
|
||||
$as_echo_n "checking for $ac_word... " >&6; }
|
||||
if ${ac_cv_prog_DOXYGEN+:} false; then :
|
||||
$as_echo_n "(cached) " >&6
|
||||
else
|
||||
if test -n "$DOXYGEN"; then
|
||||
ac_cv_prog_DOXYGEN="$DOXYGEN" # Let the user override the test.
|
||||
else
|
||||
as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
|
||||
for as_dir in $PATH
|
||||
do
|
||||
IFS=$as_save_IFS
|
||||
test -z "$as_dir" && as_dir=.
|
||||
for ac_exec_ext in '' $ac_executable_extensions; do
|
||||
if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
|
||||
ac_cv_prog_DOXYGEN="doxygen"
|
||||
$as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
|
||||
break 2
|
||||
fi
|
||||
done
|
||||
done
|
||||
IFS=$as_save_IFS
|
||||
|
||||
fi
|
||||
fi
|
||||
DOXYGEN=$ac_cv_prog_DOXYGEN
|
||||
if test -n "$DOXYGEN"; then
|
||||
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $DOXYGEN" >&5
|
||||
$as_echo "$DOXYGEN" >&6; }
|
||||
else
|
||||
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
|
||||
$as_echo "no" >&6; }
|
||||
fi
|
||||
|
||||
|
||||
if test x$DOXYGEN = x; then
|
||||
if test "x$enable_doxygen" = xyes; then
|
||||
as_fn_error $? "could not find doxygen" "$LINENO" 5
|
||||
fi
|
||||
enable_doc=no
|
||||
else
|
||||
doxy_ver=`doxygen --version`
|
||||
doxy_major=`expr "$doxy_ver" : '\([0-9]\)\..*'`
|
||||
doxy_minor=`expr "$doxy_ver" : '[0-9]\.\([0-9]\).*'`
|
||||
if test $doxy_major -eq "1" -a $doxy_minor -ge "3" ; then
|
||||
enable_doc=yes
|
||||
# Extract the first word of "dot", so it can be a program name with args.
|
||||
set dummy dot; ac_word=$2
|
||||
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
|
||||
$as_echo_n "checking for $ac_word... " >&6; }
|
||||
if ${ac_cv_prog_DOT+:} false; then :
|
||||
$as_echo_n "(cached) " >&6
|
||||
else
|
||||
if test -n "$DOT"; then
|
||||
ac_cv_prog_DOT="$DOT" # Let the user override the test.
|
||||
else
|
||||
as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
|
||||
for as_dir in $PATH
|
||||
do
|
||||
IFS=$as_save_IFS
|
||||
test -z "$as_dir" && as_dir=.
|
||||
for ac_exec_ext in '' $ac_executable_extensions; do
|
||||
if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
|
||||
ac_cv_prog_DOT="dot"
|
||||
$as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
|
||||
break 2
|
||||
fi
|
||||
done
|
||||
done
|
||||
IFS=$as_save_IFS
|
||||
|
||||
fi
|
||||
fi
|
||||
DOT=$ac_cv_prog_DOT
|
||||
if test -n "$DOT"; then
|
||||
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $DOT" >&5
|
||||
$as_echo "$DOT" >&6; }
|
||||
else
|
||||
{ $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
|
||||
$as_echo "no" >&6; }
|
||||
fi
|
||||
|
||||
|
||||
else
|
||||
{ $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: doxygen version $doxy_ver too old, doxygen will not be used." >&5
|
||||
$as_echo "$as_me: WARNING: doxygen version $doxy_ver too old, doxygen will not be used." >&2;}
|
||||
enable_doc=no
|
||||
fi
|
||||
fi
|
||||
fi
|
||||
|
||||
if test x$enable_doc = xyes; then
|
||||
DOXYGEN_DOC_TRUE=
|
||||
DOXYGEN_DOC_FALSE='#'
|
||||
else
|
||||
DOXYGEN_DOC_TRUE='#'
|
||||
DOXYGEN_DOC_FALSE=
|
||||
fi
|
||||
|
||||
|
||||
if test x$DOT = x; then
|
||||
if test "x$enable_dot" = xyes; then
|
||||
as_fn_error $? "could not find dot" "$LINENO" 5
|
||||
fi
|
||||
enable_dot=no
|
||||
else
|
||||
enable_dot=yes
|
||||
fi
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
if test -n "$DOXYGEN"
|
||||
then
|
||||
ac_config_files="$ac_config_files docs/doxy.cfg"
|
||||
|
||||
fi
|
||||
|
||||
echo
|
||||
echo Creating configuration files
|
||||
@ -6756,10 +6913,24 @@ ac_config_files="$ac_config_files lib/Makefile"
|
||||
|
||||
ac_config_files="$ac_config_files tests/Makefile"
|
||||
|
||||
ac_config_files="$ac_config_files tests/IO/Makefile"
|
||||
|
||||
ac_config_files="$ac_config_files tests/core/Makefile"
|
||||
|
||||
ac_config_files="$ac_config_files tests/debug/Makefile"
|
||||
|
||||
ac_config_files="$ac_config_files tests/forces/Makefile"
|
||||
|
||||
ac_config_files="$ac_config_files tests/hmc/Makefile"
|
||||
|
||||
ac_config_files="$ac_config_files tests/solver/Makefile"
|
||||
|
||||
ac_config_files="$ac_config_files tests/qdpxx/Makefile"
|
||||
|
||||
ac_config_files="$ac_config_files benchmarks/Makefile"
|
||||
|
||||
ac_config_files="$ac_config_files prerequisites/Makefile"
|
||||
|
||||
cat >confcache <<\_ACEOF
|
||||
# This file is a shell script that caches the results of configure
|
||||
# tests run on this system so they can be shared between configure
|
||||
@ -6925,6 +7096,10 @@ if test -z "${USE_LAPACK_LIB_TRUE}" && test -z "${USE_LAPACK_LIB_FALSE}"; then
|
||||
as_fn_error $? "conditional \"USE_LAPACK_LIB\" was never defined.
|
||||
Usually this means the macro was only invoked conditionally." "$LINENO" 5
|
||||
fi
|
||||
if test -z "${DOXYGEN_DOC_TRUE}" && test -z "${DOXYGEN_DOC_FALSE}"; then
|
||||
as_fn_error $? "conditional \"DOXYGEN_DOC\" was never defined.
|
||||
Usually this means the macro was only invoked conditionally." "$LINENO" 5
|
||||
fi
|
||||
|
||||
: "${CONFIG_STATUS=./config.status}"
|
||||
ac_write_fail=0
|
||||
@ -7349,6 +7524,7 @@ cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
|
||||
# Files that config.status was made for.
|
||||
config_files="$ac_config_files"
|
||||
config_headers="$ac_config_headers"
|
||||
config_links="$ac_config_links"
|
||||
config_commands="$ac_config_commands"
|
||||
|
||||
_ACEOF
|
||||
@ -7379,6 +7555,9 @@ $config_files
|
||||
Configuration headers:
|
||||
$config_headers
|
||||
|
||||
Configuration links:
|
||||
$config_links
|
||||
|
||||
Configuration commands:
|
||||
$config_commands
|
||||
|
||||
@ -7517,13 +7696,22 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
|
||||
for ac_config_target in $ac_config_targets
|
||||
do
|
||||
case $ac_config_target in
|
||||
"$ac_config_links_1") CONFIG_LINKS="$CONFIG_LINKS $ac_config_links_1" ;;
|
||||
"lib/Config.h") CONFIG_HEADERS="$CONFIG_HEADERS lib/Config.h" ;;
|
||||
"depfiles") CONFIG_COMMANDS="$CONFIG_COMMANDS depfiles" ;;
|
||||
"docs/doxy.cfg") CONFIG_FILES="$CONFIG_FILES docs/doxy.cfg" ;;
|
||||
"Makefile") CONFIG_FILES="$CONFIG_FILES Makefile" ;;
|
||||
"lib/Makefile") CONFIG_FILES="$CONFIG_FILES lib/Makefile" ;;
|
||||
"tests/Makefile") CONFIG_FILES="$CONFIG_FILES tests/Makefile" ;;
|
||||
"tests/IO/Makefile") CONFIG_FILES="$CONFIG_FILES tests/IO/Makefile" ;;
|
||||
"tests/core/Makefile") CONFIG_FILES="$CONFIG_FILES tests/core/Makefile" ;;
|
||||
"tests/debug/Makefile") CONFIG_FILES="$CONFIG_FILES tests/debug/Makefile" ;;
|
||||
"tests/forces/Makefile") CONFIG_FILES="$CONFIG_FILES tests/forces/Makefile" ;;
|
||||
"tests/hmc/Makefile") CONFIG_FILES="$CONFIG_FILES tests/hmc/Makefile" ;;
|
||||
"tests/solver/Makefile") CONFIG_FILES="$CONFIG_FILES tests/solver/Makefile" ;;
|
||||
"tests/qdpxx/Makefile") CONFIG_FILES="$CONFIG_FILES tests/qdpxx/Makefile" ;;
|
||||
"benchmarks/Makefile") CONFIG_FILES="$CONFIG_FILES benchmarks/Makefile" ;;
|
||||
"prerequisites/Makefile") CONFIG_FILES="$CONFIG_FILES prerequisites/Makefile" ;;
|
||||
|
||||
*) as_fn_error $? "invalid argument: \`$ac_config_target'" "$LINENO" 5;;
|
||||
esac
|
||||
@ -7537,6 +7725,7 @@ done
|
||||
if $ac_need_defaults; then
|
||||
test "${CONFIG_FILES+set}" = set || CONFIG_FILES=$config_files
|
||||
test "${CONFIG_HEADERS+set}" = set || CONFIG_HEADERS=$config_headers
|
||||
test "${CONFIG_LINKS+set}" = set || CONFIG_LINKS=$config_links
|
||||
test "${CONFIG_COMMANDS+set}" = set || CONFIG_COMMANDS=$config_commands
|
||||
fi
|
||||
|
||||
@ -7834,7 +8023,7 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
|
||||
fi # test -n "$CONFIG_HEADERS"
|
||||
|
||||
|
||||
eval set X " :F $CONFIG_FILES :H $CONFIG_HEADERS :C $CONFIG_COMMANDS"
|
||||
eval set X " :F $CONFIG_FILES :H $CONFIG_HEADERS :L $CONFIG_LINKS :C $CONFIG_COMMANDS"
|
||||
shift
|
||||
for ac_tag
|
||||
do
|
||||
@ -8111,7 +8300,38 @@ $as_echo X"$_am_arg" |
|
||||
}
|
||||
s/.*/./; q'`/stamp-h$_am_stamp_count
|
||||
;;
|
||||
:L)
|
||||
#
|
||||
# CONFIG_LINK
|
||||
#
|
||||
|
||||
if test "$ac_source" = "$ac_file" && test "$srcdir" = '.'; then
|
||||
:
|
||||
else
|
||||
# Prefer the file from the source tree if names are identical.
|
||||
if test "$ac_source" = "$ac_file" || test ! -r "$ac_source"; then
|
||||
ac_source=$srcdir/$ac_source
|
||||
fi
|
||||
|
||||
{ $as_echo "$as_me:${as_lineno-$LINENO}: linking $ac_source to $ac_file" >&5
|
||||
$as_echo "$as_me: linking $ac_source to $ac_file" >&6;}
|
||||
|
||||
if test ! -r "$ac_source"; then
|
||||
as_fn_error $? "$ac_source: file not found" "$LINENO" 5
|
||||
fi
|
||||
rm -f "$ac_file"
|
||||
|
||||
# Try a relative symlink, then a hard link, then a copy.
|
||||
case $ac_source in
|
||||
[\\/$]* | ?:[\\/]* ) ac_rel_source=$ac_source ;;
|
||||
*) ac_rel_source=$ac_top_build_prefix$ac_source ;;
|
||||
esac
|
||||
ln -s "$ac_rel_source" "$ac_file" 2>/dev/null ||
|
||||
ln "$ac_source" "$ac_file" 2>/dev/null ||
|
||||
cp -p "$ac_source" "$ac_file" ||
|
||||
as_fn_error $? "cannot link or copy $ac_source to $ac_file" "$LINENO" 5
|
||||
fi
|
||||
;;
|
||||
:C) { $as_echo "$as_me:${as_lineno-$LINENO}: executing $ac_file commands" >&5
|
||||
$as_echo "$as_me: executing $ac_file commands" >&6;}
|
||||
;;
|
||||
|
||||
@ -43,8 +43,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
||||
#include <Grid/algorithms/iterative/NormalEquations.h>
|
||||
#include <Grid/algorithms/iterative/SchurRedBlack.h>
|
||||
|
||||
#include <algorithms/iterative/ConjugateGradientMultiShift.h>
|
||||
#include <algorithms/iterative/ConjugateGradientMixedPrec.h>
|
||||
#include <Grid/algorithms/iterative/ConjugateGradientMultiShift.h>
|
||||
#include <Grid/algorithms/iterative/ConjugateGradientMixedPrec.h>
|
||||
|
||||
// Lanczos support
|
||||
#include <Grid/algorithms/iterative/MatrixUtils.h>
|
||||
|
||||
@ -128,12 +128,6 @@ public:
|
||||
//////////////////////////////////////////////////////////
|
||||
// SIMD lane addressing
|
||||
//////////////////////////////////////////////////////////
|
||||
virtual int iIndex(std::vector<int> &lcoor)
|
||||
{
|
||||
int idx=0;
|
||||
for(int d=0;d<_ndimension;d++) idx+=_istride[d]*(lcoor[d]/_rdimensions[d]);
|
||||
return idx;
|
||||
}
|
||||
inline void iCoorFromIindex(std::vector<int> &coor,int lane)
|
||||
{
|
||||
Lexicographic::CoorFromIndex(coor,lane,_simd_layout);
|
||||
|
||||
@ -500,7 +500,6 @@ namespace QCD {
|
||||
#include <Grid/qcd/hmc/integrators/Integrator.h>
|
||||
#include <Grid/qcd/hmc/integrators/Integrator_algorithm.h>
|
||||
#include <Grid/qcd/hmc/HMC.h>
|
||||
|
||||
#include <qcd/smearing/Smearing.h>
|
||||
#include <Grid/qcd/smearing/Smearing.h>
|
||||
|
||||
#endif
|
||||
|
||||
@ -1,9 +1,9 @@
|
||||
#ifndef GRID_QCD_SMEARING_H
|
||||
#define GRID_QCD_SMEARING_H
|
||||
|
||||
#include <qcd/smearing/BaseSmearing.h>
|
||||
#include <qcd/smearing/APEsmearing.h>
|
||||
#include <qcd/smearing/StoutSmearing.h>
|
||||
#include <qcd/smearing/GaugeConfiguration.h>
|
||||
#include <Grid/qcd/smearing/BaseSmearing.h>
|
||||
#include <Grid/qcd/smearing/APEsmearing.h>
|
||||
#include <Grid/qcd/smearing/StoutSmearing.h>
|
||||
#include <Grid/qcd/smearing/GaugeConfiguration.h>
|
||||
|
||||
#endif
|
||||
|
||||
@ -142,88 +142,12 @@ class Grid_simd {
|
||||
conv_t_union(){};
|
||||
} conv_t;
|
||||
|
||||
////////////////////////////////////////////////////////////////////
|
||||
// General permute; assumes vector length is same across
|
||||
// all subtypes; may not be a good assumption, but could
|
||||
// add the vector width as a template param for BG/Q for example
|
||||
////////////////////////////////////////////////////////////////////
|
||||
friend inline void permute0(Grid_simd &y,Grid_simd b){
|
||||
y.v = Optimization::Permute::Permute0(b.v);
|
||||
}
|
||||
friend inline void permute1(Grid_simd &y,Grid_simd b){
|
||||
y.v = Optimization::Permute::Permute1(b.v);
|
||||
}
|
||||
friend inline void permute2(Grid_simd &y,Grid_simd b){
|
||||
y.v = Optimization::Permute::Permute2(b.v);
|
||||
}
|
||||
friend inline void permute3(Grid_simd &y,Grid_simd b){
|
||||
y.v = Optimization::Permute::Permute3(b.v);
|
||||
}
|
||||
friend inline void permute(Grid_simd &y,Grid_simd b,int perm)
|
||||
{
|
||||
if ( perm & RotateBit ) {
|
||||
int dist = perm&0xF;
|
||||
y=rotate(b,dist);
|
||||
return;
|
||||
} else {
|
||||
switch(perm){
|
||||
case 3: permute3(y,b); break;
|
||||
case 2: permute2(y,b); break;
|
||||
case 1: permute1(y,b); break;
|
||||
case 0: permute0(y,b); break;
|
||||
default: assert(0);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
};// end of Grid_simd class definition
|
||||
Vector_type v;
|
||||
|
||||
static inline int Nsimd(void) {
|
||||
return sizeof(Vector_type) / sizeof(Scalar_type);
|
||||
}
|
||||
template <class S, class V, IfNotComplex<S> =0>
|
||||
inline void rotate( Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
|
||||
{
|
||||
nrot = nrot % Grid_simd<S,V>::Nsimd();
|
||||
// std::cout << "Rotate Real by "<<nrot<<std::endl;
|
||||
ret.v = Optimization::Rotate::rotate(b.v,nrot);
|
||||
}
|
||||
template <class S, class V, IfComplex<S> =0>
|
||||
inline void rotate(Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
|
||||
{
|
||||
nrot = nrot % Grid_simd<S,V>::Nsimd();
|
||||
// std::cout << "Rotate Complex by "<<nrot<<std::endl;
|
||||
ret.v = Optimization::Rotate::rotate(b.v,2*nrot);
|
||||
}
|
||||
|
||||
///////////////////////
|
||||
// Splat
|
||||
///////////////////////
|
||||
|
||||
// this is only for the complex version
|
||||
template <class S, class V, IfComplex<S> =0, class ABtype>
|
||||
inline void vsplat(Grid_simd<S,V> &ret,ABtype a, ABtype b){
|
||||
ret.v = binary<V>(a, b, VsplatSIMD());
|
||||
}
|
||||
|
||||
template <class S, class V>
|
||||
inline void vbroadcast(Grid_simd<S,V> &ret,const Grid_simd<S,V> &src,int lane){
|
||||
S* typepun =(S*) &src;
|
||||
vsplat(ret,typepun[lane]);
|
||||
}
|
||||
|
||||
// overload if complex
|
||||
template <class S,class V> inline void vsplat(Grid_simd<S,V> &ret, EnableIf<is_complex < S >, S> c) {
|
||||
vsplat(ret,real(c),imag(c));
|
||||
}
|
||||
|
||||
//if real fill with a, if complex fill with a in the real part (first function above)
|
||||
template <class S,class V>
|
||||
inline void vsplat(Grid_simd<S,V> &ret,NotEnableIf<is_complex< S>,S> a){
|
||||
ret.v = unary<V>(a, VsplatSIMD());
|
||||
}
|
||||
//////////////////////////
|
||||
=======
|
||||
Grid_simd &operator=(const Grid_simd &&rhs) {
|
||||
v = rhs.v;
|
||||
return *this;
|
||||
@ -251,7 +175,6 @@ class Grid_simd {
|
||||
};
|
||||
|
||||
Grid_simd(const Real a) { vsplat(*this, Scalar_type(a)); };
|
||||
>>>>>>> 8b9301a74cde2f28a321baf4bca854d950585b56
|
||||
|
||||
///////////////////////////////////////////////
|
||||
// mac, mult, sub, add, adj
|
||||
@ -484,6 +407,27 @@ inline Grid_simd<S, V> rotate(Grid_simd<S, V> b, int nrot) {
|
||||
ret.v = Optimization::Rotate::rotate(b.v, 2 * nrot);
|
||||
return ret;
|
||||
}
|
||||
template <class S, class V, IfNotComplex<S> =0>
|
||||
inline void rotate( Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
|
||||
{
|
||||
nrot = nrot % Grid_simd<S,V>::Nsimd();
|
||||
// std::cout << "Rotate Real by "<<nrot<<std::endl;
|
||||
ret.v = Optimization::Rotate::rotate(b.v,nrot);
|
||||
}
|
||||
template <class S, class V, IfComplex<S> =0>
|
||||
inline void rotate(Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
|
||||
{
|
||||
nrot = nrot % Grid_simd<S,V>::Nsimd();
|
||||
// std::cout << "Rotate Complex by "<<nrot<<std::endl;
|
||||
ret.v = Optimization::Rotate::rotate(b.v,2*nrot);
|
||||
}
|
||||
|
||||
|
||||
template <class S, class V>
|
||||
inline void vbroadcast(Grid_simd<S,V> &ret,const Grid_simd<S,V> &src,int lane){
|
||||
S* typepun =(S*) &src;
|
||||
vsplat(ret,typepun[lane]);
|
||||
}
|
||||
|
||||
///////////////////////
|
||||
// Splat
|
||||
|
||||
@ -233,7 +233,6 @@ class iVector {
|
||||
vstream(out._internal[i], in._internal[i]);
|
||||
}
|
||||
}
|
||||
<<<<<<< HEAD
|
||||
friend strong_inline void vbroadcast(iVector<vtype,N> &out,const iVector<vtype,N> &in,int lane){
|
||||
for(int i=0;i<N;i++){
|
||||
vbroadcast(out._internal[i],in._internal[i],lane);
|
||||
@ -242,13 +241,6 @@ class iVector {
|
||||
friend strong_inline void permute(iVector<vtype,N> &out,const iVector<vtype,N> &in,int permutetype){
|
||||
for(int i=0;i<N;i++){
|
||||
permute(out._internal[i],in._internal[i],permutetype);
|
||||
=======
|
||||
friend strong_inline void permute(iVector<vtype, N> &out,
|
||||
const iVector<vtype, N> &in,
|
||||
int permutetype) {
|
||||
for (int i = 0; i < N; i++) {
|
||||
permute(out._internal[i], in._internal[i], permutetype);
|
||||
>>>>>>> 8b9301a74cde2f28a321baf4bca854d950585b56
|
||||
}
|
||||
}
|
||||
friend strong_inline void rotate(iVector<vtype,N> &out,const iVector<vtype,N> &in,int rot){
|
||||
@ -346,7 +338,6 @@ class iMatrix {
|
||||
return *this;
|
||||
}
|
||||
|
||||
<<<<<<< HEAD
|
||||
friend strong_inline void zeroit(iMatrix<vtype,N> &that){
|
||||
for(int i=0;i<N;i++){
|
||||
for(int j=0;j<N;j++){
|
||||
@ -376,26 +367,6 @@ class iMatrix {
|
||||
for(int j=0;j<N;j++){
|
||||
permute(out._internal[i][j],in._internal[i][j],permutetype);
|
||||
}}
|
||||
=======
|
||||
friend strong_inline void zeroit(iMatrix<vtype, N> &that) {
|
||||
for (int i = 0; i < N; i++) {
|
||||
for (int j = 0; j < N; j++) {
|
||||
zeroit(that._internal[i][j]);
|
||||
}
|
||||
}
|
||||
}
|
||||
friend strong_inline void prefetch(iMatrix<vtype, N> &that) {
|
||||
for (int i = 0; i < N; i++)
|
||||
for (int j = 0; j < N; j++) prefetch(that._internal[i][j]);
|
||||
}
|
||||
friend strong_inline void vstream(iMatrix<vtype, N> &out,
|
||||
const iMatrix<vtype, N> &in) {
|
||||
for (int i = 0; i < N; i++) {
|
||||
for (int j = 0; j < N; j++) {
|
||||
vstream(out._internal[i][j], in._internal[i][j]);
|
||||
}
|
||||
}
|
||||
>>>>>>> 8b9301a74cde2f28a321baf4bca854d950585b56
|
||||
}
|
||||
friend strong_inline void rotate(iMatrix<vtype,N> &out,const iMatrix<vtype,N> &in,int rot){
|
||||
for(int i=0;i<N;i++){
|
||||
@ -404,16 +375,6 @@ class iMatrix {
|
||||
}}
|
||||
}
|
||||
|
||||
friend strong_inline void permute(iMatrix<vtype, N> &out,
|
||||
const iMatrix<vtype, N> &in,
|
||||
int permutetype) {
|
||||
for (int i = 0; i < N; i++) {
|
||||
for (int j = 0; j < N; j++) {
|
||||
permute(out._internal[i][j], in._internal[i][j], permutetype);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Unary negation
|
||||
friend strong_inline iMatrix<vtype, N> operator-(const iMatrix<vtype, N> &r) {
|
||||
iMatrix<vtype, N> ret;
|
||||
|
||||
@ -25,7 +25,7 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
#include <Grid.h>
|
||||
#include <Grid/Grid.h>
|
||||
|
||||
using namespace std;
|
||||
using namespace Grid;
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user