portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-14 13:57:07 +01:00
Code Releases Activity

Merge branch 'develop' into release/v0.6.0

Browse Source
This commit is contained in:
azusayamaguchi
2016-11-04 16:08:07 +00:00
parent c067051d5f 8af8b047fd
commit f7b60004f3
41 changed files with 1661 additions and 895 deletions
Download Patch File Download Diff File Expand all files Collapse all files

276
lib/FFT.h
View File

@ -30,8 +30,10 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
#define _GRID_FFT_H_
#ifdef HAVE_FFTW
#include <Grid/fftw/fftw3.h>
#include <fftw3.h>
#endif
namespace Grid {
template<class scalar> struct FFTW { };
@ -98,174 +100,198 @@ namespace Grid {
#define FFTW_BACKWARD (+1)
#endif
class FFT {
class FFT {
private:
GridCartesian *vgrid;
GridCartesian *sgrid;
int Nd;
double flops;
double flops_call;
uint64_t usec;
std::vector<int> dimensions;
std::vector<int> processors;
std::vector<int> processor_coor;
public:
static const int forward=FFTW_FORWARD;
static const int backward=FFTW_BACKWARD;
double Flops(void) {return flops;}
double MFlops(void) {return flops/usec;}
FFT ( GridCartesian * grid ) :
vgrid(grid),
Nd(grid->_ndimension),
dimensions(grid->_fdimensions),
processors(grid->_processors),
processor_coor(grid->_processor_coor)
FFT ( GridCartesian * grid ) :
vgrid(grid),
Nd(grid->_ndimension),
dimensions(grid->_fdimensions),
processors(grid->_processors),
processor_coor(grid->_processor_coor)
{
flops=0;
usec =0;
std::vector<int> layout(Nd,1);
sgrid = new GridCartesian(dimensions,layout,processors);
};
~FFT ( void) {
delete sgrid;
~FFT ( void) {
delete sgrid;
}
template<class vobj>
void FFT_dim(Lattice<vobj> &result,const Lattice<vobj> &source,int dim, int inverse){
void FFT_dim_mask(Lattice<vobj> &result,const Lattice<vobj> &source,std::vector<int> mask,int sign){
conformable(result._grid,vgrid);
conformable(source._grid,vgrid);
Lattice<vobj> tmp(vgrid);
tmp = source;
for(int d=0;d<Nd;d++){
if( mask[d] ) {
FFT_dim(result,tmp,d,sign);
tmp=result;
}
}
}
template<class vobj>
void FFT_all_dim(Lattice<vobj> &result,const Lattice<vobj> &source,int sign){
std::vector<int> mask(Nd,1);
FFT_dim_mask(result,source,mask,sign);
}
template<class vobj>
void FFT_dim(Lattice<vobj> &result,const Lattice<vobj> &source,int dim, int sign){
#ifndef HAVE_FFTW
assert(0);
#else
conformable(result._grid,vgrid);
conformable(source._grid,vgrid);
int L = vgrid->_ldimensions[dim];
int G = vgrid->_fdimensions[dim];
std::vector<int> layout(Nd,1);
std::vector<int> pencil_gd(vgrid->_fdimensions);
pencil_gd[dim] = G*processors[dim];
pencil_gd[dim] = G*processors[dim];
// Pencil global vol LxLxGxLxL per node
GridCartesian pencil_g(pencil_gd,layout,processors);
// Construct pencils
typedef typename vobj::scalar_object sobj;
typedef typename sobj::scalar_type scalar;
Lattice<sobj> pgbuf(&pencil_g);
Lattice<vobj> ssource(vgrid); ssource =source;
Lattice<sobj> pgsource(&pencil_g);
Lattice<sobj> pgresult(&pencil_g); pgresult=zero;
#ifndef HAVE_FFTW
assert(0);
#else
typedef typename FFTW<scalar>::FFTW_scalar FFTW_scalar;
typedef typename FFTW<scalar>::FFTW_plan FFTW_plan;
{
int Ncomp = sizeof(sobj)/sizeof(scalar);
int Nlow = 1;
for(int d=0;d<dim;d++){
Nlow*=vgrid->_ldimensions[d];
}
int rank = 1; /* 1d transforms */
int n[] = {G}; /* 1d transforms of length G */
int howmany = Ncomp;
int odist,idist,istride,ostride;
idist = odist = 1; /* Distance between consecutive FT's */
istride = ostride = Ncomp*Nlow; /* distance between two elements in the same FT */
int *inembed = n, *onembed = n;
int sign = FFTW_FORWARD;
if (inverse) sign = FFTW_BACKWARD;
FFTW_plan p;
{
FFTW_scalar *in = (FFTW_scalar *)&pgsource._odata[0];
FFTW_scalar *out= (FFTW_scalar *)&pgresult._odata[0];
p = FFTW<scalar>::fftw_plan_many_dft(rank,n,howmany,
in,inembed,
istride,idist,
out,onembed,
ostride, odist,
sign,FFTW_ESTIMATE);
}
std::vector<int> lcoor(Nd), gcoor(Nd);
// Barrel shift and collect global pencil
for(int p=0;p<processors[dim];p++) {
for(int idx=0;idx<sgrid->lSites();idx++) {
sgrid->LocalIndexToLocalCoor(idx,lcoor);
sobj s;
peekLocalSite(s,ssource,lcoor);
lcoor[dim]+=p*L;
pokeLocalSite(s,pgsource,lcoor);
}
ssource = Cshift(ssource,dim,L);
}
// Loop over orthog coords
int NN=pencil_g.lSites();
GridStopWatch timer;
timer.Start();
//PARALLEL_FOR_LOOP
for(int idx=0;idx<NN;idx++) {
pencil_g.LocalIndexToLocalCoor(idx,lcoor);
if ( lcoor[dim] == 0 ) { // restricts loop to plane at lcoor[dim]==0
FFTW_scalar *in = (FFTW_scalar *)&pgsource._odata[idx];
FFTW_scalar *out= (FFTW_scalar *)&pgresult._odata[idx];
FFTW<scalar>::fftw_execute_dft(p,in,out);
}
}
timer.Stop();
double add,mul,fma;
FFTW<scalar>::fftw_flops(p,&add,&mul,&fma);
flops_call = add+mul+2.0*fma;
usec += timer.useconds();
flops+= flops_call*NN;
int pc = processor_coor[dim];
for(int idx=0;idx<sgrid->lSites();idx++) {
sgrid->LocalIndexToLocalCoor(idx,lcoor);
gcoor = lcoor;
// extract the result
sobj s;
gcoor[dim] = lcoor[dim]+L*pc;
peekLocalSite(s,pgresult,gcoor);
pokeLocalSite(s,result,lcoor);
}
FFTW<scalar>::fftw_destroy_plan(p);
int Ncomp = sizeof(sobj)/sizeof(scalar);
int Nlow = 1;
for(int d=0;d<dim;d++){
Nlow*=vgrid->_ldimensions[d];
}
int rank = 1; /* 1d transforms */
int n[] = {G}; /* 1d transforms of length G */
int howmany = Ncomp;
int odist,idist,istride,ostride;
idist = odist = 1; /* Distance between consecutive FT's */
istride = ostride = Ncomp*Nlow; /* distance between two elements in the same FT */
int *inembed = n, *onembed = n;
scalar div;
if ( sign == backward ) div = 1.0/G;
else if ( sign == forward ) div = 1.0;
else assert(0);
FFTW_plan p;
{
FFTW_scalar *in = (FFTW_scalar *)&pgbuf._odata[0];
FFTW_scalar *out= (FFTW_scalar *)&pgbuf._odata[0];
p = FFTW<scalar>::fftw_plan_many_dft(rank,n,howmany,
in,inembed,
istride,idist,
out,onembed,
ostride, odist,
sign,FFTW_ESTIMATE);
}
// Barrel shift and collect global pencil
std::vector<int> lcoor(Nd), gcoor(Nd);
result = source;
for(int p=0;p<processors[dim];p++) {
PARALLEL_REGION
{
std::vector<int> cbuf(Nd);
sobj s;
PARALLEL_FOR_LOOP_INTERN
for(int idx=0;idx<sgrid->lSites();idx++) {
sgrid->LocalIndexToLocalCoor(idx,cbuf);
peekLocalSite(s,result,cbuf);
cbuf[dim]+=p*L;
pokeLocalSite(s,pgbuf,cbuf);
}
}
result = Cshift(result,dim,L);
}
// Loop over orthog coords
int NN=pencil_g.lSites();
GridStopWatch timer;
timer.Start();
PARALLEL_REGION
{
std::vector<int> cbuf(Nd);
PARALLEL_FOR_LOOP_INTERN
for(int idx=0;idx<NN;idx++) {
pencil_g.LocalIndexToLocalCoor(idx, cbuf);
if ( cbuf[dim] == 0 ) { // restricts loop to plane at lcoor[dim]==0
FFTW_scalar *in = (FFTW_scalar *)&pgbuf._odata[idx];
FFTW_scalar *out= (FFTW_scalar *)&pgbuf._odata[idx];
FFTW<scalar>::fftw_execute_dft(p,in,out);
}
}
}
timer.Stop();
// performance counting
double add,mul,fma;
FFTW<scalar>::fftw_flops(p,&add,&mul,&fma);
flops_call = add+mul+2.0*fma;
usec += timer.useconds();
flops+= flops_call*NN;
// writing out result
int pc = processor_coor[dim];
PARALLEL_REGION
{
std::vector<int> clbuf(Nd), cgbuf(Nd);
sobj s;
PARALLEL_FOR_LOOP_INTERN
for(int idx=0;idx<sgrid->lSites();idx++) {
sgrid->LocalIndexToLocalCoor(idx,clbuf);
cgbuf = clbuf;
cgbuf[dim] = clbuf[dim]+L*pc;
peekLocalSite(s,pgbuf,cgbuf);
s = s * div;
pokeLocalSite(s,result,clbuf);
}
}
// destroying plan
FFTW<scalar>::fftw_destroy_plan(p);
#endif
}
};
}
#endif

5
lib/Grid.h
View File

@ -77,11 +77,10 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#include <Grid/Stencil.h>
#include <Grid/Algorithms.h>
#include <Grid/parallelIO/BinaryIO.h>
#include <Grid/qcd/QCD.h>
#include <Grid/parallelIO/NerscIO.h>
#include <Grid/FFT.h>
#include <Grid/qcd/QCD.h>
#include <Grid/parallelIO/NerscIO.h>
#include <Grid/qcd/hmc/NerscCheckpointer.h>
#include <Grid/qcd/hmc/HmcRunner.h>

39
lib/Init.cc
View File

@ -44,9 +44,33 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#include <Grid.h>
#include <algorithm>
#include <iterator>
#include <cstdlib>
#include <memory>
#include <fenv.h>
#ifdef __APPLE__
static int
feenableexcept (unsigned int excepts)
{
static fenv_t fenv;
unsigned int new_excepts = excepts & FE_ALL_EXCEPT,
old_excepts; // previous masks
if ( fegetenv (&fenv) ) return -1;
old_excepts = fenv.__control & FE_ALL_EXCEPT;
// unmask
fenv.__control &= ~new_excepts;
fenv.__mxcsr &= ~(new_excepts << 7);
return ( fesetenv (&fenv) ? -1 : old_excepts );
}
#endif
namespace Grid {
//////////////////////////////////////////////////////
// Convenience functions to access stadard command line arg
// driven parallelism controls
@ -234,7 +258,7 @@ void Grid_init(int *argc,char ***argv)
std::cout<<GridLogMessage<<" --decomposition : report on default omp,mpi and simd decomposition"<<std::endl;
std::cout<<GridLogMessage<<" --debug-signals : catch sigsegv and print a blame report"<<std::endl;
std::cout<<GridLogMessage<<" --debug-stdout : print stdout from EVERY node"<<std::endl;
std::cout<<GridLogMessage<<" --timestamp : tag with millisecond resolution stamps"<<std::endl;
std::cout<<GridLogMessage<<" --notimestamp : suppress millisecond resolution stamps"<<std::endl;
std::cout<<GridLogMessage<<std::endl;
std::cout<<GridLogMessage<<"Performance:"<<std::endl;
std::cout<<GridLogMessage<<" --dslash-generic: Wilson kernel for generic Nc"<<std::endl;
@ -316,7 +340,9 @@ void Grid_init(int *argc,char ***argv)
arg= GridCmdOptionPayload(*argv,*argv+*argc,"--cacheblocking");
GridCmdOptionIntVector(arg,LebesgueOrder::Block);
}
if( GridCmdOptionExists(*argv,*argv+*argc,"--timestamp") ){
if( GridCmdOptionExists(*argv,*argv+*argc,"--notimestamp") ){
GridLogTimestamp(0);
} else {
GridLogTimestamp(1);
}
@ -390,10 +416,7 @@ void Grid_sa_signal_handler(int sig,siginfo_t *si,void * ptr)
exit(0);
return;
};
#ifdef GRID_FPE
#define _GNU_SOURCE
#include <fenv.h>
#endif
void Grid_debug_handler_init(void)
{
struct sigaction sa,osa;
@ -402,9 +425,9 @@ void Grid_debug_handler_init(void)
sa.sa_flags = SA_SIGINFO;
sigaction(SIGSEGV,&sa,NULL);
sigaction(SIGTRAP,&sa,NULL);
#ifdef GRID_FPE
feenableexcept( FE_INVALID|FE_OVERFLOW|FE_DIVBYZERO);
sigaction(SIGFPE,&sa,NULL);
#endif
}
}

1
lib/Init.h
View File

@ -54,6 +54,7 @@ namespace Grid {
void GridCmdOptionCSL(std::string str,std::vector<std::string> & vec);
void GridCmdOptionIntVector(std::string &str,std::vector<int> & vec);
void GridParseLayout(char **argv,int argc,
std::vector<int> &latt,
std::vector<int> &simd,

15
lib/Log.cc
View File

@ -31,8 +31,23 @@ directory
/* END LEGAL */
#include <Grid.h>
#include <cxxabi.h>
namespace Grid {
std::string demangle(const char* name) {
int status = -4; // some arbitrary value to eliminate the compiler warning
// enable c++11 by passing the flag -std=c++11 to g++
std::unique_ptr<char, void(*)(void*)> res {
abi::__cxa_demangle(name, NULL, NULL, &status),
std::free
};
return (status==0) ? res.get() : name ;
}
GridStopWatch Logger::StopWatch;
int Logger::timestamp;
std::ostream Logger::devnull(0);

3
lib/Log.h
View File

@ -144,6 +144,7 @@ extern GridLogger GridLogIterative ;
extern GridLogger GridLogIntegrator ;
extern Colours GridLogColours;
std::string demangle(const char* name) ;
#define _NBACKTRACE (256)
extern void * Grid_backtrace_buffer[_NBACKTRACE];
@ -162,7 +163,7 @@ std::fclose(fp); \
int symbols = backtrace (Grid_backtrace_buffer,_NBACKTRACE);\
char **strings = backtrace_symbols(Grid_backtrace_buffer,symbols);\
for (int i = 0; i < symbols; i++){\
std::fprintf (fp,"BackTrace Strings: %d %s\n",i, strings[i]); std::fflush(fp); \
std::fprintf (fp,"BackTrace Strings: %d %s\n",i, demangle(strings[i]).c_str()); std::fflush(fp); \
}\
}
#else

12
lib/Simd.h
View File

@ -237,6 +237,18 @@ namespace Grid {
stream<<">";
return stream;
}
inline std::ostream& operator<< (std::ostream& stream, const vInteger &o){
int nn=vInteger::Nsimd();
std::vector<Integer,alignedAllocator<Integer> > buf(nn);
vstore(o,&buf[0]);
stream<<"<";
for(int i=0;i<nn;i++){
stream<<buf[i];
if(i<nn-1) stream<<",";
}
stream<<">";
return stream;
}
}

11
lib/Threads.h
View File

@ -38,14 +38,19 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#ifdef GRID_OMP
#include <omp.h>
#ifdef GRID_NUMA
#define PARALLEL_FOR_LOOP _Pragma("omp parallel for schedule(static)")
#define PARALLEL_FOR_LOOP _Pragma("omp parallel for schedule(static)")
#define PARALLEL_FOR_LOOP_INTERN _Pragma("omp for schedule(static)")
#else
#define PARALLEL_FOR_LOOP _Pragma("omp parallel for schedule(runtime)")
#define PARALLEL_FOR_LOOP _Pragma("omp parallel for schedule(runtime)")
#define PARALLEL_FOR_LOOP_INTERN _Pragma("omp for schedule(runtime)")
#endif
#define PARALLEL_NESTED_LOOP2 _Pragma("omp parallel for collapse(2)")
#define PARALLEL_REGION _Pragma("omp parallel")
#else
#define PARALLEL_FOR_LOOP
#define PARALLEL_FOR_LOOP
#define PARALLEL_FOR_LOOP_INTERN
#define PARALLEL_NESTED_LOOP2
#define PARALLEL_REGION
#endif
namespace Grid {

19
lib/communicator/Communicator_mpi3_leader.cc
View File

@ -39,6 +39,10 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
///
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
#include <semaphore.h>
#include <fcntl.h>
#include <unistd.h>
#include <limits.h>
typedef sem_t *Grid_semaphore;
#define SEM_INIT(S) S = sem_open(sem_name,0,0600,0); assert ( S != SEM_FAILED );
@ -48,7 +52,6 @@ typedef sem_t *Grid_semaphore;
#include <sys/mman.h>
namespace Grid {
enum { COMMAND_ISEND, COMMAND_IRECV, COMMAND_WAITALL };
@ -91,18 +94,18 @@ public:
void SemInit(void) {
sprintf(sem_name,"/Grid_mpi3_sem_head_%d",universe_rank);
printf("SEM_NAME: %s \n",sem_name);
// printf("SEM_NAME: %s \n",sem_name);
SEM_INIT(sem_head);
sprintf(sem_name,"/Grid_mpi3_sem_tail_%d",universe_rank);
printf("SEM_NAME: %s \n",sem_name);
// printf("SEM_NAME: %s \n",sem_name);
SEM_INIT(sem_tail);
}
void SemInitExcl(void) {
sprintf(sem_name,"/Grid_mpi3_sem_head_%d",universe_rank);
printf("SEM_INIT_EXCL: %s \n",sem_name);
// printf("SEM_INIT_EXCL: %s \n",sem_name);
SEM_INIT_EXCL(sem_head);
sprintf(sem_name,"/Grid_mpi3_sem_tail_%d",universe_rank);
printf("SEM_INIT_EXCL: %s \n",sem_name);
// printf("SEM_INIT_EXCL: %s \n",sem_name);
SEM_INIT_EXCL(sem_tail);
}
void WakeUpDMA(void) {
@ -118,7 +121,7 @@ public:
SEM_WAIT(sem_tail);
};
void EventLoop (void) {
std::cout<< " Entering event loop "<<std::endl;
// std::cout<< " Entering event loop "<<std::endl;
while(1){
WaitForCommand();
// std::cout << "Getting command "<<std::endl;
@ -291,7 +294,7 @@ void MPIoffloadEngine::CommunicatorInit (MPI_Comm &communicator_world,
/////////////////////////////////////////////////////////////////////
// Split into groups that can share memory (Verticals)
/////////////////////////////////////////////////////////////////////
#define MPI_SHARED_MEM_DEBUG
#undef MPI_SHARED_MEM_DEBUG
#ifdef MPI_SHARED_MEM_DEBUG
MPI_Comm_split(communicator_universe,(UniverseRank/4),UniverseRank,&VerticalComm);
#else
@ -527,7 +530,7 @@ void Slave::Init(SlaveState * _state,MPI_Comm _squadron,int _universe_rank,int _
universe_rank=_universe_rank;
vertical_rank=_vertical_rank;
state =_state;
std::cout << "state "<<_state<<" comm "<<_squadron<<" universe_rank"<<universe_rank <<std::endl;
// std::cout << "state "<<_state<<" comm "<<_squadron<<" universe_rank"<<universe_rank <<std::endl;
state->head = state->tail = state->start = 0;
base = (uint64_t)MPIoffloadEngine::VerticalShmBufs[0];
int rank; MPI_Comm_rank(_squadron,&rank);

412
lib/fftw/fftw3.h
View File

@ -1,412 +0,0 @@
/*
* Copyright (c) 2003, 2007-14 Matteo Frigo
* Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
*
* The following statement of license applies *only* to this header file,
* and *not* to the other files distributed with FFTW or derived therefrom:
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.
*/
/***************************** NOTE TO USERS *********************************
*
* THIS IS A HEADER FILE, NOT A MANUAL
*
* If you want to know how to use FFTW, please read the manual,
* online at http://www.fftw.org/doc/ and also included with FFTW.
* For a quick start, see the manual's tutorial section.
*
* (Reading header files to learn how to use a library is a habit
* stemming from code lacking a proper manual. Arguably, it's a
* *bad* habit in most cases, because header files can contain
* interfaces that are not part of the public, stable API.)
*
****************************************************************************/
#ifndef FFTW3_H
#define FFTW3_H
#include <stdio.h>
#ifdef __cplusplus
extern "C"
{
#endif /* __cplusplus */
/* If <complex.h> is included, use the C99 complex type. Otherwise
define a type bit-compatible with C99 complex */
#if !defined(FFTW_NO_Complex) && defined(_Complex_I) && defined(complex) && defined(I)
# define FFTW_DEFINE_COMPLEX(R, C) typedef R _Complex C
#else
# define FFTW_DEFINE_COMPLEX(R, C) typedef R C[2]
#endif
#define FFTW_CONCAT(prefix, name) prefix ## name
#define FFTW_MANGLE_DOUBLE(name) FFTW_CONCAT(fftw_, name)
#define FFTW_MANGLE_FLOAT(name) FFTW_CONCAT(fftwf_, name)
#define FFTW_MANGLE_LONG_DOUBLE(name) FFTW_CONCAT(fftwl_, name)
#define FFTW_MANGLE_QUAD(name) FFTW_CONCAT(fftwq_, name)
/* IMPORTANT: for Windows compilers, you should add a line
#define FFTW_DLL
here and in kernel/ifftw.h if you are compiling/using FFTW as a
DLL, in order to do the proper importing/exporting, or
alternatively compile with -DFFTW_DLL or the equivalent
command-line flag. This is not necessary under MinGW/Cygwin, where
libtool does the imports/exports automatically. */
#if defined(FFTW_DLL) && (defined(_WIN32) || defined(__WIN32__))
/* annoying Windows syntax for shared-library declarations */
# if defined(COMPILING_FFTW) /* defined in api.h when compiling FFTW */
# define FFTW_EXTERN extern __declspec(dllexport)
# else /* user is calling FFTW; import symbol */
# define FFTW_EXTERN extern __declspec(dllimport)
# endif
#else
# define FFTW_EXTERN extern
#endif
enum fftw_r2r_kind_do_not_use_me {
FFTW_R2HC=0, FFTW_HC2R=1, FFTW_DHT=2,
FFTW_REDFT00=3, FFTW_REDFT01=4, FFTW_REDFT10=5, FFTW_REDFT11=6,
FFTW_RODFT00=7, FFTW_RODFT01=8, FFTW_RODFT10=9, FFTW_RODFT11=10
};
struct fftw_iodim_do_not_use_me {
int n; /* dimension size */
int is; /* input stride */
int os; /* output stride */
};
#include <stddef.h> /* for ptrdiff_t */
struct fftw_iodim64_do_not_use_me {
ptrdiff_t n; /* dimension size */
ptrdiff_t is; /* input stride */
ptrdiff_t os; /* output stride */
};
typedef void (*fftw_write_char_func_do_not_use_me)(char c, void *);
typedef int (*fftw_read_char_func_do_not_use_me)(void *);
/*
huge second-order macro that defines prototypes for all API
functions. We expand this macro for each supported precision
X: name-mangling macro
R: real data type
C: complex data type
*/
#define FFTW_DEFINE_API(X, R, C) \
\
FFTW_DEFINE_COMPLEX(R, C); \
\
typedef struct X(plan_s) *X(plan); \
\
typedef struct fftw_iodim_do_not_use_me X(iodim); \
typedef struct fftw_iodim64_do_not_use_me X(iodim64); \
\
typedef enum fftw_r2r_kind_do_not_use_me X(r2r_kind); \
\
typedef fftw_write_char_func_do_not_use_me X(write_char_func); \
typedef fftw_read_char_func_do_not_use_me X(read_char_func); \
\
FFTW_EXTERN void X(execute)(const X(plan) p); \
\
FFTW_EXTERN X(plan) X(plan_dft)(int rank, const int *n, \
C *in, C *out, int sign, unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_dft_1d)(int n, C *in, C *out, int sign, \
unsigned flags); \
FFTW_EXTERN X(plan) X(plan_dft_2d)(int n0, int n1, \
C *in, C *out, int sign, unsigned flags); \
FFTW_EXTERN X(plan) X(plan_dft_3d)(int n0, int n1, int n2, \
C *in, C *out, int sign, unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_many_dft)(int rank, const int *n, \
int howmany, \
C *in, const int *inembed, \
int istride, int idist, \
C *out, const int *onembed, \
int ostride, int odist, \
int sign, unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_guru_dft)(int rank, const X(iodim) *dims, \
int howmany_rank, \
const X(iodim) *howmany_dims, \
C *in, C *out, \
int sign, unsigned flags); \
FFTW_EXTERN X(plan) X(plan_guru_split_dft)(int rank, const X(iodim) *dims, \
int howmany_rank, \
const X(iodim) *howmany_dims, \
R *ri, R *ii, R *ro, R *io, \
unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_guru64_dft)(int rank, \
const X(iodim64) *dims, \
int howmany_rank, \
const X(iodim64) *howmany_dims, \
C *in, C *out, \
int sign, unsigned flags); \
FFTW_EXTERN X(plan) X(plan_guru64_split_dft)(int rank, \
const X(iodim64) *dims, \
int howmany_rank, \
const X(iodim64) *howmany_dims, \
R *ri, R *ii, R *ro, R *io, \
unsigned flags); \
\
FFTW_EXTERN void X(execute_dft)(const X(plan) p, C *in, C *out); \
FFTW_EXTERN void X(execute_split_dft)(const X(plan) p, R *ri, R *ii, \
R *ro, R *io); \
\
FFTW_EXTERN X(plan) X(plan_many_dft_r2c)(int rank, const int *n, \
int howmany, \
R *in, const int *inembed, \
int istride, int idist, \
C *out, const int *onembed, \
int ostride, int odist, \
unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_dft_r2c)(int rank, const int *n, \
R *in, C *out, unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_dft_r2c_1d)(int n,R *in,C *out,unsigned flags); \
FFTW_EXTERN X(plan) X(plan_dft_r2c_2d)(int n0, int n1, \
R *in, C *out, unsigned flags); \
FFTW_EXTERN X(plan) X(plan_dft_r2c_3d)(int n0, int n1, \
int n2, \
R *in, C *out, unsigned flags); \
\
\
FFTW_EXTERN X(plan) X(plan_many_dft_c2r)(int rank, const int *n, \
int howmany, \
C *in, const int *inembed, \
int istride, int idist, \
R *out, const int *onembed, \
int ostride, int odist, \
unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_dft_c2r)(int rank, const int *n, \
C *in, R *out, unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_dft_c2r_1d)(int n,C *in,R *out,unsigned flags); \
FFTW_EXTERN X(plan) X(plan_dft_c2r_2d)(int n0, int n1, \
C *in, R *out, unsigned flags); \
FFTW_EXTERN X(plan) X(plan_dft_c2r_3d)(int n0, int n1, \
int n2, \
C *in, R *out, unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_guru_dft_r2c)(int rank, const X(iodim) *dims, \
int howmany_rank, \
const X(iodim) *howmany_dims, \
R *in, C *out, \
unsigned flags); \
FFTW_EXTERN X(plan) X(plan_guru_dft_c2r)(int rank, const X(iodim) *dims, \
int howmany_rank, \
const X(iodim) *howmany_dims, \
C *in, R *out, \
unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_guru_split_dft_r2c)( \
int rank, const X(iodim) *dims, \
int howmany_rank, \
const X(iodim) *howmany_dims, \
R *in, R *ro, R *io, \
unsigned flags); \
FFTW_EXTERN X(plan) X(plan_guru_split_dft_c2r)( \
int rank, const X(iodim) *dims, \
int howmany_rank, \
const X(iodim) *howmany_dims, \
R *ri, R *ii, R *out, \
unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_guru64_dft_r2c)(int rank, \
const X(iodim64) *dims, \
int howmany_rank, \
const X(iodim64) *howmany_dims, \
R *in, C *out, \
unsigned flags); \
FFTW_EXTERN X(plan) X(plan_guru64_dft_c2r)(int rank, \
const X(iodim64) *dims, \
int howmany_rank, \
const X(iodim64) *howmany_dims, \
C *in, R *out, \
unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_guru64_split_dft_r2c)( \
int rank, const X(iodim64) *dims, \
int howmany_rank, \
const X(iodim64) *howmany_dims, \
R *in, R *ro, R *io, \
unsigned flags); \
FFTW_EXTERN X(plan) X(plan_guru64_split_dft_c2r)( \
int rank, const X(iodim64) *dims, \
int howmany_rank, \
const X(iodim64) *howmany_dims, \
R *ri, R *ii, R *out, \
unsigned flags); \
\
FFTW_EXTERN void X(execute_dft_r2c)(const X(plan) p, R *in, C *out); \
FFTW_EXTERN void X(execute_dft_c2r)(const X(plan) p, C *in, R *out); \
\
FFTW_EXTERN void X(execute_split_dft_r2c)(const X(plan) p, \
R *in, R *ro, R *io); \
FFTW_EXTERN void X(execute_split_dft_c2r)(const X(plan) p, \
R *ri, R *ii, R *out); \
\
FFTW_EXTERN X(plan) X(plan_many_r2r)(int rank, const int *n, \
int howmany, \
R *in, const int *inembed, \
int istride, int idist, \
R *out, const int *onembed, \
int ostride, int odist, \
const X(r2r_kind) *kind, unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_r2r)(int rank, const int *n, R *in, R *out, \
const X(r2r_kind) *kind, unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_r2r_1d)(int n, R *in, R *out, \
X(r2r_kind) kind, unsigned flags); \
FFTW_EXTERN X(plan) X(plan_r2r_2d)(int n0, int n1, R *in, R *out, \
X(r2r_kind) kind0, X(r2r_kind) kind1, \
unsigned flags); \
FFTW_EXTERN X(plan) X(plan_r2r_3d)(int n0, int n1, int n2, \
R *in, R *out, X(r2r_kind) kind0, \
X(r2r_kind) kind1, X(r2r_kind) kind2, \
unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_guru_r2r)(int rank, const X(iodim) *dims, \
int howmany_rank, \
const X(iodim) *howmany_dims, \
R *in, R *out, \
const X(r2r_kind) *kind, unsigned flags); \
\
FFTW_EXTERN X(plan) X(plan_guru64_r2r)(int rank, const X(iodim64) *dims, \
int howmany_rank, \
const X(iodim64) *howmany_dims, \
R *in, R *out, \
const X(r2r_kind) *kind, unsigned flags); \
\
FFTW_EXTERN void X(execute_r2r)(const X(plan) p, R *in, R *out); \
\
FFTW_EXTERN void X(destroy_plan)(X(plan) p); \
FFTW_EXTERN void X(forget_wisdom)(void); \
FFTW_EXTERN void X(cleanup)(void); \
\
FFTW_EXTERN void X(set_timelimit)(double t); \
\
FFTW_EXTERN void X(plan_with_nthreads)(int nthreads); \
FFTW_EXTERN int X(init_threads)(void); \
FFTW_EXTERN void X(cleanup_threads)(void); \
\
FFTW_EXTERN int X(export_wisdom_to_filename)(const char *filename); \
FFTW_EXTERN void X(export_wisdom_to_file)(FILE *output_file); \
FFTW_EXTERN char *X(export_wisdom_to_string)(void); \
FFTW_EXTERN void X(export_wisdom)(X(write_char_func) write_char, \
void *data); \
FFTW_EXTERN int X(import_system_wisdom)(void); \
FFTW_EXTERN int X(import_wisdom_from_filename)(const char *filename); \
FFTW_EXTERN int X(import_wisdom_from_file)(FILE *input_file); \
FFTW_EXTERN int X(import_wisdom_from_string)(const char *input_string); \
FFTW_EXTERN int X(import_wisdom)(X(read_char_func) read_char, void *data); \
\
FFTW_EXTERN void X(fprint_plan)(const X(plan) p, FILE *output_file); \
FFTW_EXTERN void X(print_plan)(const X(plan) p); \
FFTW_EXTERN char *X(sprint_plan)(const X(plan) p); \
\
FFTW_EXTERN void *X(malloc)(size_t n); \
FFTW_EXTERN R *X(alloc_real)(size_t n); \
FFTW_EXTERN C *X(alloc_complex)(size_t n); \
FFTW_EXTERN void X(free)(void *p); \
\
FFTW_EXTERN void X(flops)(const X(plan) p, \
double *add, double *mul, double *fmas); \
FFTW_EXTERN double X(estimate_cost)(const X(plan) p); \
FFTW_EXTERN double X(cost)(const X(plan) p); \
\
FFTW_EXTERN int X(alignment_of)(R *p); \
FFTW_EXTERN const char X(version)[]; \
FFTW_EXTERN const char X(cc)[]; \
FFTW_EXTERN const char X(codelet_optim)[];
/* end of FFTW_DEFINE_API macro */
FFTW_DEFINE_API(FFTW_MANGLE_DOUBLE, double, fftw_complex)
FFTW_DEFINE_API(FFTW_MANGLE_FLOAT, float, fftwf_complex)
FFTW_DEFINE_API(FFTW_MANGLE_LONG_DOUBLE, long double, fftwl_complex)
/* __float128 (quad precision) is a gcc extension on i386, x86_64, and ia64
for gcc >= 4.6 (compiled in FFTW with --enable-quad-precision) */
#if (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6)) \
&& !(defined(__ICC) || defined(__INTEL_COMPILER)) \
&& (defined(__i386__) || defined(__x86_64__) || defined(__ia64__))
# if !defined(FFTW_NO_Complex) && defined(_Complex_I) && defined(complex) && defined(I)
/* note: __float128 is a typedef, which is not supported with the _Complex
keyword in gcc, so instead we use this ugly __attribute__ version.
However, we can't simply pass the __attribute__ version to
FFTW_DEFINE_API because the __attribute__ confuses gcc in pointer
types. Hence redefining FFTW_DEFINE_COMPLEX. Ugh. */
# undef FFTW_DEFINE_COMPLEX
# define FFTW_DEFINE_COMPLEX(R, C) typedef _Complex float __attribute__((mode(TC))) C
# endif
FFTW_DEFINE_API(FFTW_MANGLE_QUAD, __float128, fftwq_complex)
#endif
#define FFTW_FORWARD (-1)
#define FFTW_BACKWARD (+1)
#define FFTW_NO_TIMELIMIT (-1.0)
/* documented flags */
#define FFTW_MEASURE (0U)
#define FFTW_DESTROY_INPUT (1U << 0)
#define FFTW_UNALIGNED (1U << 1)
#define FFTW_CONSERVE_MEMORY (1U << 2)
#define FFTW_EXHAUSTIVE (1U << 3) /* NO_EXHAUSTIVE is default */
#define FFTW_PRESERVE_INPUT (1U << 4) /* cancels FFTW_DESTROY_INPUT */
#define FFTW_PATIENT (1U << 5) /* IMPATIENT is default */
#define FFTW_ESTIMATE (1U << 6)
#define FFTW_WISDOM_ONLY (1U << 21)
/* undocumented beyond-guru flags */
#define FFTW_ESTIMATE_PATIENT (1U << 7)
#define FFTW_BELIEVE_PCOST (1U << 8)
#define FFTW_NO_DFT_R2HC (1U << 9)
#define FFTW_NO_NONTHREADED (1U << 10)
#define FFTW_NO_BUFFERING (1U << 11)
#define FFTW_NO_INDIRECT_OP (1U << 12)
#define FFTW_ALLOW_LARGE_GENERIC (1U << 13) /* NO_LARGE_GENERIC is default */
#define FFTW_NO_RANK_SPLITS (1U << 14)
#define FFTW_NO_VRANK_SPLITS (1U << 15)
#define FFTW_NO_VRECURSE (1U << 16)
#define FFTW_NO_SIMD (1U << 17)
#define FFTW_NO_SLOW (1U << 18)
#define FFTW_NO_FIXED_RADIX_LARGE_N (1U << 19)
#define FFTW_ALLOW_PRUNING (1U << 20)
#ifdef __cplusplus
} /* extern "C" */
#endif /* __cplusplus */
#endif /* FFTW3_H */

2
lib/lattice/Lattice_ET.h
View File

@ -261,6 +261,7 @@ GridUnopClass(UnaryExp, exp(a));
GridBinOpClass(BinaryAdd, lhs + rhs);
GridBinOpClass(BinarySub, lhs - rhs);
GridBinOpClass(BinaryMul, lhs *rhs);
GridBinOpClass(BinaryDiv, lhs /rhs);
GridBinOpClass(BinaryAnd, lhs &rhs);
GridBinOpClass(BinaryOr, lhs | rhs);
@ -385,6 +386,7 @@ GRID_DEF_UNOP(exp, UnaryExp);
GRID_DEF_BINOP(operator+, BinaryAdd);
GRID_DEF_BINOP(operator-, BinarySub);
GRID_DEF_BINOP(operator*, BinaryMul);
GRID_DEF_BINOP(operator/, BinaryDiv);
GRID_DEF_BINOP(operator&, BinaryAnd);
GRID_DEF_BINOP(operator|, BinaryOr);

11
lib/lattice/Lattice_base.h
View File

@ -300,17 +300,6 @@ PARALLEL_FOR_LOOP
*this = (*this)+r;
return *this;
}
strong_inline friend Lattice<vobj> operator / (const Lattice<vobj> &lhs,const Lattice<vobj> &rhs){
conformable(lhs,rhs);
Lattice<vobj> ret(lhs._grid);
PARALLEL_FOR_LOOP
for(int ss=0;ss<lhs._grid->oSites();ss++){
ret._odata[ss] = lhs._odata[ss]*pow(rhs._odata[ss],-1.0);
}
return ret;
};
}; // class Lattice
template<class vobj> std::ostream& operator<< (std::ostream& stream, const Lattice<vobj> &o){

2
lib/lattice/Lattice_rng.h
View File

@ -294,7 +294,7 @@ namespace Grid {
int rank,o_idx,i_idx;
_grid->GlobalIndexToGlobalCoor(gidx,gcoor);
_grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor);
int l_idx=generator_idx(o_idx,i_idx);
const int num_rand_seed=16;

2
lib/parallelIO/BinaryIO.h
View File

@ -457,7 +457,7 @@ class BinaryIO {
// available (how short sighted is that?)
//////////////////////////////////////////////////////////
Umu = zero;
static uint32_t csum=0;
static uint32_t csum; csum=0;
fobj fileObj;
static sobj siteObj; // Static to place in symmetric region for SHMEM

24
lib/qcd/action/fermion/CayleyFermion5D.cc
View File

@ -50,6 +50,30 @@ namespace QCD {
mass(_mass)
{ }
template<class Impl>
void CayleyFermion5D<Impl>::Dminus(const FermionField &psi, FermionField &chi)
{
int Ls=this->Ls;
FermionField tmp(psi._grid);
this->DW(psi,tmp,DaggerNo);
for(int s=0;s<Ls;s++){
axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],tmp,s,s);// chi = (1-c[s] D_W) psi
}
}
template<class Impl>
void CayleyFermion5D<Impl>::DminusDag(const FermionField &psi, FermionField &chi)
{
int Ls=this->Ls;
FermionField tmp(psi._grid);
this->DW(psi,tmp,DaggerYes);
for(int s=0;s<Ls;s++){
axpby_ssp(chi,Coeff_t(1.0),psi,-cs[s],tmp,s,s);// chi = (1-c[s] D_W) psi
}
}
template<class Impl>
void CayleyFermion5D<Impl>::M5D (const FermionField &psi, FermionField &chi)
{

4
lib/qcd/action/fermion/CayleyFermion5D.h
View File

@ -56,6 +56,9 @@ namespace Grid {
virtual void M5D (const FermionField &psi, FermionField &chi);
virtual void M5Ddag(const FermionField &psi, FermionField &chi);
virtual void Dminus(const FermionField &psi, FermionField &chi);
virtual void DminusDag(const FermionField &psi, FermionField &chi);
/////////////////////////////////////////////////////
// Instantiate different versions depending on Impl
/////////////////////////////////////////////////////
@ -117,6 +120,7 @@ namespace Grid {
GridRedBlackCartesian &FourDimRedBlackGrid,
RealD _mass,RealD _M5,const ImplParams &p= ImplParams());
protected:
void SetCoefficientsZolotarev(RealD zolohi,Approx::zolotarev_data *zdata,RealD b,RealD c);
void SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD b,RealD c);

5
lib/qcd/action/fermion/DomainWallFermion.h
View File

@ -42,6 +42,10 @@ namespace Grid {
INHERIT_IMPL_TYPES(Impl);
public:
void MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _m) {
this->MomentumSpacePropagatorHt(out,in,_m);
};
virtual void Instantiatable(void) {};
// Constructors
DomainWallFermion(GaugeField &_Umu,
@ -51,6 +55,7 @@ namespace Grid {
GridRedBlackCartesian &FourDimRedBlackGrid,
RealD _mass,RealD _M5,const ImplParams &p= ImplParams()) :
CayleyFermion5D<Impl>(_Umu,
FiveDimGrid,
FiveDimRedBlackGrid,

14
lib/qcd/action/fermion/FermionOperator.h
View File

@ -91,6 +91,20 @@ namespace Grid {
virtual void Mdiag (const FermionField &in, FermionField &out) { Mooee(in,out);}; // Same as Mooee applied to both CB's
virtual void Mdir (const FermionField &in, FermionField &out,int dir,int disp)=0; // case by case Wilson, Clover, Cayley, ContFrac, PartFrac
virtual void MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _m) { assert(0);};
virtual void FreePropagator(const FermionField &in,FermionField &out,RealD mass) {
FFT theFFT((GridCartesian *) in._grid);
FermionField in_k(in._grid);
FermionField prop_k(in._grid);
theFFT.FFT_all_dim(in_k,in,FFT::forward);
this->MomentumSpacePropagator(prop_k,in_k,mass);
theFFT.FFT_all_dim(out,prop_k,FFT::backward);
};
///////////////////////////////////////////////
// Updates gauge field during HMC
///////////////////////////////////////////////

6
lib/qcd/action/fermion/OverlapWilsonCayleyTanhFermion.h
View File

@ -42,7 +42,11 @@ namespace Grid {
INHERIT_IMPL_TYPES(Impl);
public:
// Constructors
void MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _m) {
this->MomentumSpacePropagatorHw(out,in,_m);
};
// Constructors
OverlapWilsonCayleyTanhFermion(GaugeField &_Umu,
GridCartesian &FiveDimGrid,
GridRedBlackCartesian &FiveDimRedBlackGrid,

100
lib/qcd/action/fermion/WilsonFermion.cc
View File

@ -101,6 +101,7 @@ void WilsonFermion<Impl>::Meooe(const FermionField &in, FermionField &out) {
DhopOE(in, out, DaggerNo);
}
}
template <class Impl>
void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
if (in.checkerboard == Odd) {
@ -109,32 +110,87 @@ void WilsonFermion<Impl>::MeooeDag(const FermionField &in, FermionField &out) {
DhopOE(in, out, DaggerYes);
}
}
template <class Impl>
void WilsonFermion<Impl>::Mooee(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
typename FermionField::scalar_type scal(4.0 + mass);
out = scal * in;
}
template <class Impl>
void WilsonFermion<Impl>::Mooee(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
typename FermionField::scalar_type scal(4.0 + mass);
out = scal * in;
}
template <class Impl>
void WilsonFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
Mooee(in, out);
}
template <class Impl>
void WilsonFermion<Impl>::MooeeDag(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
Mooee(in, out);
}
template<class Impl>
void WilsonFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
out = (1.0/(4.0+mass))*in;
}
template<class Impl>
void WilsonFermion<Impl>::MooeeInvDag(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
MooeeInv(in,out);
}
template <class Impl>
void WilsonFermion<Impl>::MooeeInv(const FermionField &in, FermionField &out) {
out.checkerboard = in.checkerboard;
out = (1.0 / (4.0 + mass)) * in;
}
template<class Impl>
void WilsonFermion<Impl>::MomentumSpacePropagator(FermionField &out, const FermionField &in,RealD _m) {
template <class Impl>
void WilsonFermion<Impl>::MooeeInvDag(const FermionField &in,
FermionField &out) {
out.checkerboard = in.checkerboard;
MooeeInv(in, out);
}
// what type LatticeComplex
conformable(_grid,out._grid);
typedef typename FermionField::vector_type vector_type;
typedef typename FermionField::scalar_type ScalComplex;
typedef Lattice<iSinglet<vector_type> > LatComplex;
Gamma::GammaMatrix Gmu [] = {
Gamma::GammaX,
Gamma::GammaY,
Gamma::GammaZ,
Gamma::GammaT
};
std::vector<int> latt_size = _grid->_fdimensions;
FermionField num (_grid); num = zero;
LatComplex wilson(_grid); wilson= zero;
LatComplex one (_grid); one = ScalComplex(1.0,0.0);
LatComplex denom(_grid); denom= zero;
LatComplex kmu(_grid);
ScalComplex ci(0.0,1.0);
// momphase = n * 2pi / L
for(int mu=0;mu<Nd;mu++) {
LatticeCoordinate(kmu,mu);
RealD TwoPiL = M_PI * 2.0/ latt_size[mu];
kmu = TwoPiL * kmu;
wilson = wilson + 2.0*sin(kmu*0.5)*sin(kmu*0.5); // Wilson term
num = num - sin(kmu)*ci*(Gamma(Gmu[mu])*in); // derivative term
denom=denom + sin(kmu)*sin(kmu);
}
wilson = wilson + _m; // 2 sin^2 k/2 + m
num = num + wilson*in; // -i gmu sin k + 2 sin^2 k/2 + m
denom= denom+wilson*wilson; // sin^2 k + (2 sin^2 k/2 + m)^2
denom= one/denom;
out = num*denom; // [ -i gmu sin k + 2 sin^2 k/2 + m] / [ sin^2 k + (2 sin^2 k/2 + m)^2 ]
}
///////////////////////////////////
// Internal

11
lib/qcd/action/fermion/WilsonFermion.h
View File

@ -78,16 +78,15 @@ class WilsonFermion : public WilsonKernels<Impl>, public WilsonFermionStatic {
virtual void MooeeInv(const FermionField &in, FermionField &out);
virtual void MooeeInvDag(const FermionField &in, FermionField &out);
virtual void MomentumSpacePropagator(FermionField &out,const FermionField &in,RealD _mass) ;
////////////////////////
// Derivative interface
////////////////////////
// Interface calls an internal routine
void DhopDeriv(GaugeField &mat, const FermionField &U, const FermionField &V,
int dag);
void DhopDerivOE(GaugeField &mat, const FermionField &U,
const FermionField &V, int dag);
void DhopDerivEO(GaugeField &mat, const FermionField &U,
const FermionField &V, int dag);
void DhopDeriv(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
///////////////////////////////////////////////////////////////
// non-hermitian hopping term; half cb or both

142
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@ -482,6 +482,148 @@ void WilsonFermion5D<Impl>::DW(const FermionField &in, FermionField &out,int dag
axpy(out,4.0-M5,in,out);
}
template<class Impl>
void WilsonFermion5D<Impl>::MomentumSpacePropagatorHt(FermionField &out,const FermionField &in, RealD mass)
{
// what type LatticeComplex
GridBase *_grid = _FourDimGrid;
conformable(_grid,out._grid);
typedef typename FermionField::vector_type vector_type;
typedef typename FermionField::scalar_type ScalComplex;
typedef iSinglet<ScalComplex> Tcomplex;
typedef Lattice<iSinglet<vector_type> > LatComplex;
Gamma::GammaMatrix Gmu [] = {
Gamma::GammaX,
Gamma::GammaY,
Gamma::GammaZ,
Gamma::GammaT
};
std::vector<int> latt_size = _grid->_fdimensions;
FermionField num (_grid); num = zero;
LatComplex sk(_grid); sk = zero;
LatComplex sk2(_grid); sk2= zero;
LatComplex W(_grid); W= zero;
LatComplex a(_grid); a= zero;
LatComplex one (_grid); one = ScalComplex(1.0,0.0);
LatComplex denom(_grid); denom= zero;
LatComplex cosha(_grid);
LatComplex kmu(_grid);
LatComplex Wea(_grid);
LatComplex Wema(_grid);
ScalComplex ci(0.0,1.0);
for(int mu=0;mu<Nd;mu++) {
LatticeCoordinate(kmu,mu);
RealD TwoPiL = M_PI * 2.0/ latt_size[mu];
kmu = TwoPiL * kmu;
sk2 = sk2 + 2.0*sin(kmu*0.5)*sin(kmu*0.5);
sk = sk + sin(kmu) *sin(kmu);
num = num - sin(kmu)*ci*(Gamma(Gmu[mu])*in);
}
W = one - M5 + sk2;
////////////////////////////////////////////
// Cosh alpha -> alpha
////////////////////////////////////////////
cosha = (one + W*W + sk) / (W*2.0);
// FIXME Need a Lattice acosh
for(int idx=0;idx<_grid->lSites();idx++){
std::vector<int> lcoor(Nd);
Tcomplex cc;
RealD sgn;
_grid->LocalIndexToLocalCoor(idx,lcoor);
peekLocalSite(cc,cosha,lcoor);
assert((double)real(cc)>=1.0);
assert(fabs((double)imag(cc))<=1.0e-15);
cc = ScalComplex(::acosh(real(cc)),0.0);
pokeLocalSite(cc,a,lcoor);
}
Wea = ( exp( a) * W );
Wema= ( exp(-a) * W );
num = num + ( one - Wema ) * mass * in;
denom= ( Wea - one ) + mass*mass * (one - Wema);
out = num/denom;
}
template<class Impl>
void WilsonFermion5D<Impl>::MomentumSpacePropagatorHw(FermionField &out,const FermionField &in,RealD mass)
{
Gamma::GammaMatrix Gmu [] = {
Gamma::GammaX,
Gamma::GammaY,
Gamma::GammaZ,
Gamma::GammaT
};
GridBase *_grid = _FourDimGrid;
conformable(_grid,out._grid);
typedef typename FermionField::vector_type vector_type;
typedef typename FermionField::scalar_type ScalComplex;
typedef Lattice<iSinglet<vector_type> > LatComplex;
std::vector<int> latt_size = _grid->_fdimensions;
LatComplex sk(_grid); sk = zero;
LatComplex sk2(_grid); sk2= zero;
LatComplex w_k(_grid); w_k= zero;
LatComplex b_k(_grid); b_k= zero;
LatComplex one (_grid); one = ScalComplex(1.0,0.0);
FermionField num (_grid); num = zero;
LatComplex denom(_grid); denom= zero;
LatComplex kmu(_grid);
ScalComplex ci(0.0,1.0);
for(int mu=0;mu<Nd;mu++) {
LatticeCoordinate(kmu,mu);
RealD TwoPiL = M_PI * 2.0/ latt_size[mu];
kmu = TwoPiL * kmu;
sk2 = sk2 + 2.0*sin(kmu*0.5)*sin(kmu*0.5);
sk = sk + sin(kmu)*sin(kmu);
num = num - sin(kmu)*ci*(Gamma(Gmu[mu])*in);
}
num = num + mass * in ;
b_k = sk2 - M5;
w_k = sqrt(sk + b_k*b_k);
denom= ( w_k + b_k + mass*mass) ;
denom= one/denom;
out = num*denom;
}
FermOpTemplateInstantiate(WilsonFermion5D);
GparityFermOpTemplateInstantiate(WilsonFermion5D);

138
lib/qcd/action/fermion/WilsonFermion5D.h
View File

@ -47,68 +47,82 @@ namespace QCD {
// [DIFFERS from original CPS red black implementation parity = (x+y+z+t+s)|2 ]
////////////////////////////////////////////////////////////////////////////////
class WilsonFermion5DStatic {
public:
// S-direction is INNERMOST and takes no part in the parity.
static const std::vector<int> directions;
static const std::vector<int> displacements;
const int npoint = 8;
};
template<class Impl>
class WilsonFermion5D : public WilsonKernels<Impl>, public WilsonFermion5DStatic
{
public:
INHERIT_IMPL_TYPES(Impl);
typedef WilsonKernels<Impl> Kernels;
PmuStat stat;
void Report(void);
void ZeroCounters(void);
double DhopCalls;
double DhopCommTime;
double DhopComputeTime;
double DerivCalls;
double DerivCommTime;
double DerivComputeTime;
double DerivDhopComputeTime;
///////////////////////////////////////////////////////////////
// Implement the abstract base
///////////////////////////////////////////////////////////////
GridBase *GaugeGrid(void) { return _FourDimGrid ;}
GridBase *GaugeRedBlackGrid(void) { return _FourDimRedBlackGrid ;}
GridBase *FermionGrid(void) { return _FiveDimGrid;}
GridBase *FermionRedBlackGrid(void) { return _FiveDimRedBlackGrid;}
// full checkerboard operations; leave unimplemented as abstract for now
virtual RealD M (const FermionField &in, FermionField &out){assert(0); return 0.0;};
virtual RealD Mdag (const FermionField &in, FermionField &out){assert(0); return 0.0;};
// half checkerboard operations; leave unimplemented as abstract for now
virtual void Meooe (const FermionField &in, FermionField &out){assert(0);};
virtual void Mooee (const FermionField &in, FermionField &out){assert(0);};
virtual void MooeeInv (const FermionField &in, FermionField &out){assert(0);};
virtual void MeooeDag (const FermionField &in, FermionField &out){assert(0);};
virtual void MooeeDag (const FermionField &in, FermionField &out){assert(0);};
virtual void MooeeInvDag (const FermionField &in, FermionField &out){assert(0);};
virtual void Mdir (const FermionField &in, FermionField &out,int dir,int disp){assert(0);}; // case by case Wilson, Clover, Cayley, ContFrac, PartFrac
// These can be overridden by fancy 5d chiral action
virtual void DhopDeriv (GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
virtual void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
virtual void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
// Implement hopping term non-hermitian hopping term; half cb or both
// Implement s-diagonal DW
void DW (const FermionField &in, FermionField &out,int dag);
void Dhop (const FermionField &in, FermionField &out,int dag);
void DhopOE(const FermionField &in, FermionField &out,int dag);
void DhopEO(const FermionField &in, FermionField &out,int dag);
// add a DhopComm
////////////////////////////////////////////////////////////////////////////////
// This is the 4d red black case appropriate to support
//
// parity = (x+y+z+t)|2;
// generalised five dim fermions like mobius, zolotarev etc..
//
// i.e. even even contains fifth dim hopping term.
//
// [DIFFERS from original CPS red black implementation parity = (x+y+z+t+s)|2 ]
////////////////////////////////////////////////////////////////////////////////
class WilsonFermion5DStatic {
public:
// S-direction is INNERMOST and takes no part in the parity.
static const std::vector<int> directions;
static const std::vector<int> displacements;
const int npoint = 8;
};
template<class Impl>
class WilsonFermion5D : public WilsonKernels<Impl>, public WilsonFermion5DStatic
{
public:
INHERIT_IMPL_TYPES(Impl);
typedef WilsonKernels<Impl> Kernels;
PmuStat stat;
void Report(void);
void ZeroCounters(void);
double DhopCalls;
double DhopCommTime;
double DhopComputeTime;
double DerivCalls;
double DerivCommTime;
double DerivComputeTime;
double DerivDhopComputeTime;
///////////////////////////////////////////////////////////////
// Implement the abstract base
///////////////////////////////////////////////////////////////
GridBase *GaugeGrid(void) { return _FourDimGrid ;}
GridBase *GaugeRedBlackGrid(void) { return _FourDimRedBlackGrid ;}
GridBase *FermionGrid(void) { return _FiveDimGrid;}
GridBase *FermionRedBlackGrid(void) { return _FiveDimRedBlackGrid;}
// full checkerboard operations; leave unimplemented as abstract for now
virtual RealD M (const FermionField &in, FermionField &out){assert(0); return 0.0;};
virtual RealD Mdag (const FermionField &in, FermionField &out){assert(0); return 0.0;};
// half checkerboard operations; leave unimplemented as abstract for now
virtual void Meooe (const FermionField &in, FermionField &out){assert(0);};
virtual void Mooee (const FermionField &in, FermionField &out){assert(0);};
virtual void MooeeInv (const FermionField &in, FermionField &out){assert(0);};
virtual void MeooeDag (const FermionField &in, FermionField &out){assert(0);};
virtual void MooeeDag (const FermionField &in, FermionField &out){assert(0);};
virtual void MooeeInvDag (const FermionField &in, FermionField &out){assert(0);};
virtual void Mdir (const FermionField &in, FermionField &out,int dir,int disp){assert(0);}; // case by case Wilson, Clover, Cayley, ContFrac, PartFrac
// These can be overridden by fancy 5d chiral action
virtual void DhopDeriv (GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
virtual void DhopDerivEO(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
virtual void DhopDerivOE(GaugeField &mat,const FermionField &U,const FermionField &V,int dag);
void MomentumSpacePropagatorHt(FermionField &out,const FermionField &in,RealD mass) ;
void MomentumSpacePropagatorHw(FermionField &out,const FermionField &in,RealD mass) ;
// Implement hopping term non-hermitian hopping term; half cb or both
// Implement s-diagonal DW
void DW (const FermionField &in, FermionField &out,int dag);
void Dhop (const FermionField &in, FermionField &out,int dag);
void DhopOE(const FermionField &in, FermionField &out,int dag);
void DhopEO(const FermionField &in, FermionField &out,int dag);
// add a DhopComm
// -- suboptimal interface will presently trigger multiple comms.
void DhopDir(const FermionField &in, FermionField &out,int dir,int disp);

18
lib/qcd/action/fermion/WilsonKernels.h
View File

@ -61,14 +61,8 @@ public:
switch(Opt) {
#ifdef AVX512
case OptInlineAsm:
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
WilsonKernels<Impl>::DiracOptAsmDhopSite(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
sF++;
}
sU++;
}
break;
WilsonKernels<Impl>::DiracOptAsmDhopSite(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
break;
#endif
case OptHandUnroll:
for (int site = 0; site < Ns; site++) {
@ -115,13 +109,7 @@ public:
switch(Opt) {
#ifdef AVX512
case OptInlineAsm:
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
WilsonKernels<Impl>::DiracOptAsmDhopSiteDag(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
sF++;
}
sU++;
}
WilsonKernels<Impl>::DiracOptAsmDhopSiteDag(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
break;
#endif
case OptHandUnroll:

28
lib/qcd/utils/LinalgUtils.h
View File

@ -39,8 +39,8 @@ namespace QCD{
//on the 5d (rb4d) checkerboarded lattices
////////////////////////////////////////////////////////////////////////
template<class vobj>
void axpibg5x(Lattice<vobj> &z,const Lattice<vobj> &x,RealD a,RealD b)
template<class vobj,class Coeff>
void axpibg5x(Lattice<vobj> &z,const Lattice<vobj> &x,Coeff a,Coeff b)
{
z.checkerboard = x.checkerboard;
conformable(x,z);
@ -57,8 +57,8 @@ PARALLEL_FOR_LOOP
}
}
template<class vobj>
void axpby_ssp(Lattice<vobj> &z, RealD a,const Lattice<vobj> &x,RealD b,const Lattice<vobj> &y,int s,int sp)
template<class vobj,class Coeff>
void axpby_ssp(Lattice<vobj> &z, Coeff a,const Lattice<vobj> &x,Coeff b,const Lattice<vobj> &y,int s,int sp)
{
z.checkerboard = x.checkerboard;
conformable(x,y);
@ -72,8 +72,8 @@ PARALLEL_FOR_LOOP
}
}
template<class vobj>
void ag5xpby_ssp(Lattice<vobj> &z,RealD a,const Lattice<vobj> &x,RealD b,const Lattice<vobj> &y,int s,int sp)
template<class vobj,class Coeff>
void ag5xpby_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const Lattice<vobj> &y,int s,int sp)
{
z.checkerboard = x.checkerboard;
conformable(x,y);
@ -90,8 +90,8 @@ PARALLEL_FOR_LOOP
}
}
template<class vobj>
void axpbg5y_ssp(Lattice<vobj> &z,RealD a,const Lattice<vobj> &x,RealD b,const Lattice<vobj> &y,int s,int sp)
template<class vobj,class Coeff>
void axpbg5y_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const Lattice<vobj> &y,int s,int sp)
{
z.checkerboard = x.checkerboard;
conformable(x,y);
@ -108,8 +108,8 @@ PARALLEL_FOR_LOOP
}
}
template<class vobj>
void ag5xpbg5y_ssp(Lattice<vobj> &z,RealD a,const Lattice<vobj> &x,RealD b,const Lattice<vobj> &y,int s,int sp)
template<class vobj,class Coeff>
void ag5xpbg5y_ssp(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const Lattice<vobj> &y,int s,int sp)
{
z.checkerboard = x.checkerboard;
conformable(x,y);
@ -127,8 +127,8 @@ PARALLEL_FOR_LOOP
}
}
template<class vobj>
void axpby_ssp_pminus(Lattice<vobj> &z,RealD a,const Lattice<vobj> &x,RealD b,const Lattice<vobj> &y,int s,int sp)
template<class vobj,class Coeff>
void axpby_ssp_pminus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const Lattice<vobj> &y,int s,int sp)
{
z.checkerboard = x.checkerboard;
conformable(x,y);
@ -144,8 +144,8 @@ PARALLEL_FOR_LOOP
}
}
template<class vobj>
void axpby_ssp_pplus(Lattice<vobj> &z,RealD a,const Lattice<vobj> &x,RealD b,const Lattice<vobj> &y,int s,int sp)
template<class vobj,class Coeff>
void axpby_ssp_pplus(Lattice<vobj> &z,Coeff a,const Lattice<vobj> &x,Coeff b,const Lattice<vobj> &y,int s,int sp)
{
z.checkerboard = x.checkerboard;
conformable(x,y);

65
lib/qcd/utils/SUn.h
View File

@ -674,6 +674,37 @@ class SU {
out += la;
}
}
/*
add GaugeTrans
*/
template<typename GaugeField,typename GaugeMat>
static void GaugeTransform( GaugeField &Umu, GaugeMat &g){
GridBase *grid = Umu._grid;
conformable(grid,g._grid);
GaugeMat U(grid);
GaugeMat ag(grid); ag = adj(g);
for(int mu=0;mu<Nd;mu++){
U= PeekIndex<LorentzIndex>(Umu,mu);
U = g*U*Cshift(ag, mu, 1);
PokeIndex<LorentzIndex>(Umu,U,mu);
}
}
template<typename GaugeMat>
static void GaugeTransform( std::vector<GaugeMat> &U, GaugeMat &g){
GridBase *grid = g._grid;
GaugeMat ag(grid); ag = adj(g);
for(int mu=0;mu<Nd;mu++){
U[mu] = g*U[mu]*Cshift(ag, mu, 1);
}
}
template<typename GaugeField,typename GaugeMat>
static void RandomGaugeTransform(GridParallelRNG &pRNG, GaugeField &Umu, GaugeMat &g){
LieRandomize(pRNG,g,1.0);
GaugeTransform(Umu,g);
}
// Projects the algebra components a lattice matrix (of dimension ncol*ncol -1 )
// inverse operation: FundamentalLieAlgebraMatrix
@ -702,23 +733,33 @@ class SU {
PokeIndex<LorentzIndex>(out, Umu, mu);
}
}
static void TepidConfiguration(GridParallelRNG &pRNG,
LatticeGaugeField &out) {
LatticeMatrix Umu(out._grid);
for (int mu = 0; mu < Nd; mu++) {
LieRandomize(pRNG, Umu, 0.01);
PokeIndex<LorentzIndex>(out, Umu, mu);
template<typename GaugeField>
static void TepidConfiguration(GridParallelRNG &pRNG,GaugeField &out){
typedef typename GaugeField::vector_type vector_type;
typedef iSUnMatrix<vector_type> vMatrixType;
typedef Lattice<vMatrixType> LatticeMatrixType;
LatticeMatrixType Umu(out._grid);
for(int mu=0;mu<Nd;mu++){
LieRandomize(pRNG,Umu,0.01);
PokeIndex<LorentzIndex>(out,Umu,mu);
}
}
static void ColdConfiguration(GridParallelRNG &pRNG, LatticeGaugeField &out) {
LatticeMatrix Umu(out._grid);
Umu = 1.0;
for (int mu = 0; mu < Nd; mu++) {
PokeIndex<LorentzIndex>(out, Umu, mu);
template<typename GaugeField>
static void ColdConfiguration(GridParallelRNG &pRNG,GaugeField &out){
typedef typename GaugeField::vector_type vector_type;
typedef iSUnMatrix<vector_type> vMatrixType;
typedef Lattice<vMatrixType> LatticeMatrixType;
LatticeMatrixType Umu(out._grid);
Umu=1.0;
for(int mu=0;mu<Nd;mu++){
PokeIndex<LorentzIndex>(out,Umu,mu);
}
}
static void taProj(const LatticeMatrix &in, LatticeMatrix &out) {
template<typename LatticeMatrixType>
static void taProj( const LatticeMatrixType &in, LatticeMatrixType &out){
out = Ta(in);
}
template <typename LatticeMatrixType>

2
lib/qcd/utils/WilsonLoops.h
View File

@ -522,4 +522,4 @@ typedef WilsonLoops<PeriodicGimplR> SU3WilsonLoops;
}
}
#endif
#endif

51
lib/simd/Grid_avx.h
View File

@ -365,6 +365,18 @@ namespace Optimization {
}
};
struct Div{
// Real float
inline __m256 operator()(__m256 a, __m256 b){
return _mm256_div_ps(a,b);
}
// Real double
inline __m256d operator()(__m256d a, __m256d b){
return _mm256_div_pd(a,b);
}
};
struct Conj{
// Complex single
inline __m256 operator()(__m256 in){
@ -437,14 +449,13 @@ namespace Optimization {
};
#if defined (AVX2) || defined (AVXFMA4)
#define _mm256_alignr_epi32(ret,a,b,n) ret=(__m256) _mm256_alignr_epi8((__m256i)a,(__m256i)b,(n*4)%16)
#define _mm256_alignr_epi64(ret,a,b,n) ret=(__m256d) _mm256_alignr_epi8((__m256i)a,(__m256i)b,(n*8)%16)
#if defined (AVX2)
#define _mm256_alignr_epi32_grid(ret,a,b,n) ret=(__m256) _mm256_alignr_epi8((__m256i)a,(__m256i)b,(n*4)%16)
#define _mm256_alignr_epi64_grid(ret,a,b,n) ret=(__m256d) _mm256_alignr_epi8((__m256i)a,(__m256i)b,(n*8)%16)
#endif
#if defined (AVX1) || defined (AVXFMA)
#define _mm256_alignr_epi32(ret,a,b,n) { \
#if defined (AVX1) || defined (AVXFMA)
#define _mm256_alignr_epi32_grid(ret,a,b,n) { \
__m128 aa, bb; \
\
aa = _mm256_extractf128_ps(a,1); \
@ -458,7 +469,7 @@ namespace Optimization {
ret = _mm256_insertf128_ps(ret,aa,0); \
}
#define _mm256_alignr_epi64(ret,a,b,n) { \
#define _mm256_alignr_epi64_grid(ret,a,b,n) { \
__m128d aa, bb; \
\
aa = _mm256_extractf128_pd(a,1); \
@ -474,19 +485,6 @@ namespace Optimization {
#endif
inline std::ostream & operator << (std::ostream& stream, const __m256 a)
{
const float *p=(const float *)&a;
stream<< "{"<<p[0]<<","<<p[1]<<","<<p[2]<<","<<p[3]<<","<<p[4]<<","<<p[5]<<","<<p[6]<<","<<p[7]<<"}";
return stream;
};
inline std::ostream & operator<< (std::ostream& stream, const __m256d a)
{
const double *p=(const double *)&a;
stream<< "{"<<p[0]<<","<<p[1]<<","<<p[2]<<","<<p[3]<<"}";
return stream;
};
struct Rotate{
static inline __m256 rotate(__m256 in,int n){
@ -518,11 +516,10 @@ namespace Optimization {
__m256 tmp = Permute::Permute0(in);
__m256 ret;
if ( n > 3 ) {
_mm256_alignr_epi32(ret,in,tmp,n);
_mm256_alignr_epi32_grid(ret,in,tmp,n);
} else {
_mm256_alignr_epi32(ret,tmp,in,n);
_mm256_alignr_epi32_grid(ret,tmp,in,n);
}
// std::cout << " align epi32 n=" <<n<<" in "<<tmp<<in<<" -> "<< ret <<std::endl;
return ret;
};
@ -531,18 +528,15 @@ namespace Optimization {
__m256d tmp = Permute::Permute0(in);
__m256d ret;
if ( n > 1 ) {
_mm256_alignr_epi64(ret,in,tmp,n);
_mm256_alignr_epi64_grid(ret,in,tmp,n);
} else {
_mm256_alignr_epi64(ret,tmp,in,n);
_mm256_alignr_epi64_grid(ret,tmp,in,n);
}
// std::cout << " align epi64 n=" <<n<<" in "<<tmp<<in<<" -> "<< ret <<std::endl;
return ret;
};
};
//Complex float Reduce
template<>
inline Grid::ComplexF Reduce<Grid::ComplexF, __m256>::operator()(__m256 in){
@ -631,6 +625,7 @@ namespace Optimization {
// Arithmetic operations
typedef Optimization::Sum SumSIMD;
typedef Optimization::Sub SubSIMD;
typedef Optimization::Div DivSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::Conj ConjSIMD;

12
lib/simd/Grid_avx512.h
View File

@ -240,6 +240,17 @@ namespace Optimization {
}
};
struct Div{
// Real float
inline __m512 operator()(__m512 a, __m512 b){
return _mm512_div_ps(a,b);
}
// Real double
inline __m512d operator()(__m512d a, __m512d b){
return _mm512_div_pd(a,b);
}
};
struct Conj{
// Complex single
@ -497,6 +508,7 @@ namespace Optimization {
typedef Optimization::Sum SumSIMD;
typedef Optimization::Sub SubSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::Div DivSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::Conj ConjSIMD;
typedef Optimization::TimesMinusI TimesMinusISIMD;

12
lib/simd/Grid_imci.h
View File

@ -244,6 +244,17 @@ namespace Optimization {
}
};
struct Div{
// Real float
inline __m512 operator()(__m512 a, __m512 b){
return _mm512_div_ps(a,b);
}
// Real double
inline __m512d operator()(__m512d a, __m512d b){
return _mm512_div_pd(a,b);
}
};
struct Conj{
// Complex single
@ -437,6 +448,7 @@ namespace Optimization {
// Arithmetic operations
typedef Optimization::Sum SumSIMD;
typedef Optimization::Sub SubSIMD;
typedef Optimization::Div DivSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::Conj ConjSIMD;

15
lib/simd/Grid_sse4.h
View File

@ -224,6 +224,18 @@ namespace Optimization {
}
};
struct Div{
// Real float
inline __m128 operator()(__m128 a, __m128 b){
return _mm_div_ps(a,b);
}
// Real double
inline __m128d operator()(__m128d a, __m128d b){
return _mm_div_pd(a,b);
}
};
struct Conj{
// Complex single
inline __m128 operator()(__m128 in){
@ -372,6 +384,8 @@ namespace Optimization {
}
}
//////////////////////////////////////////////////////////////////////////////////////
// Here assign types
@ -398,6 +412,7 @@ namespace Optimization {
// Arithmetic operations
typedef Optimization::Sum SumSIMD;
typedef Optimization::Sub SubSIMD;
typedef Optimization::Div DivSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::Conj ConjSIMD;

88
lib/simd/Grid_vector_types.h
View File

@ -77,38 +77,24 @@ struct RealPart<std::complex<T> > {
//////////////////////////////////////
// demote a vector to real type
//////////////////////////////////////
// type alias used to simplify the syntax of std::enable_if
template <typename T>
using Invoke = typename T::type;
template <typename Condition, typename ReturnType>
using EnableIf = Invoke<std::enable_if<Condition::value, ReturnType> >;
template <typename Condition, typename ReturnType>
using NotEnableIf = Invoke<std::enable_if<!Condition::value, ReturnType> >;
template <typename T> using Invoke = typename T::type;
template <typename Condition, typename ReturnType> using EnableIf = Invoke<std::enable_if<Condition::value, ReturnType> >;
template <typename Condition, typename ReturnType> using NotEnableIf = Invoke<std::enable_if<!Condition::value, ReturnType> >;
////////////////////////////////////////////////////////
// Check for complexity with type traits
template <typename T>
struct is_complex : public std::false_type {};
template <>
struct is_complex<std::complex<double> > : public std::true_type {};
template <>
struct is_complex<std::complex<float> > : public std::true_type {};
template <typename T> struct is_complex : public std::false_type {};
template <> struct is_complex<std::complex<double> > : public std::true_type {};
template <> struct is_complex<std::complex<float> > : public std::true_type {};
template <typename T>
using IfReal = Invoke<std::enable_if<std::is_floating_point<T>::value, int> >;
template <typename T>
using IfComplex = Invoke<std::enable_if<is_complex<T>::value, int> >;
template <typename T>
using IfInteger = Invoke<std::enable_if<std::is_integral<T>::value, int> >;
template <typename T> using IfReal = Invoke<std::enable_if<std::is_floating_point<T>::value, int> >;
template <typename T> using IfComplex = Invoke<std::enable_if<is_complex<T>::value, int> >;
template <typename T> using IfInteger = Invoke<std::enable_if<std::is_integral<T>::value, int> >;
template <typename T>
using IfNotReal =
Invoke<std::enable_if<!std::is_floating_point<T>::value, int> >;
template <typename T>
using IfNotComplex = Invoke<std::enable_if<!is_complex<T>::value, int> >;
template <typename T>
using IfNotInteger = Invoke<std::enable_if<!std::is_integral<T>::value, int> >;
template <typename T> using IfNotReal = Invoke<std::enable_if<!std::is_floating_point<T>::value, int> >;
template <typename T> using IfNotComplex = Invoke<std::enable_if<!is_complex<T>::value, int> >;
template <typename T> using IfNotInteger = Invoke<std::enable_if<!std::is_integral<T>::value, int> >;
////////////////////////////////////////////////////////
// Define the operation templates functors
@ -285,6 +271,20 @@ class Grid_simd {
return a * b;
}
//////////////////////////////////
// Divides
//////////////////////////////////
friend inline Grid_simd operator/(const Scalar_type &a, Grid_simd b) {
Grid_simd va;
vsplat(va, a);
return va / b;
}
friend inline Grid_simd operator/(Grid_simd b, const Scalar_type &a) {
Grid_simd va;
vsplat(va, a);
return b / a;
}
///////////////////////
// Unary negation
///////////////////////
@ -428,7 +428,6 @@ inline void rotate(Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
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;
@ -512,7 +511,6 @@ template <class S, class V, IfInteger<S> = 0>
inline void vfalse(Grid_simd<S, V> &ret) {
vsplat(ret, 0);
}
template <class S, class V>
inline void zeroit(Grid_simd<S, V> &z) {
vzero(z);
@ -530,7 +528,6 @@ inline void vstream(Grid_simd<S, V> &out, const Grid_simd<S, V> &in) {
typedef typename S::value_type T;
binary<void>((T *)&out.v, in.v, VstreamSIMD());
}
template <class S, class V, IfInteger<S> = 0>
inline void vstream(Grid_simd<S, V> &out, const Grid_simd<S, V> &in) {
out = in;
@ -569,6 +566,34 @@ inline Grid_simd<S, V> operator*(Grid_simd<S, V> a, Grid_simd<S, V> b) {
return ret;
};
// Distinguish between complex types and others
template <class S, class V, IfComplex<S> = 0>
inline Grid_simd<S, V> operator/(Grid_simd<S, V> a, Grid_simd<S, V> b) {
typedef Grid_simd<S, V> simd;
simd ret;
simd den;
typename simd::conv_t conv;
ret = a * conjugate(b) ;
den = b * conjugate(b) ;
auto real_den = toReal(den);
ret.v=binary<V>(ret.v, real_den.v, DivSIMD());
return ret;
};
// Real/Integer types
template <class S, class V, IfNotComplex<S> = 0>
inline Grid_simd<S, V> operator/(Grid_simd<S, V> a, Grid_simd<S, V> b) {
Grid_simd<S, V> ret;
ret.v = binary<V>(a.v, b.v, DivSIMD());
return ret;
};
///////////////////////
// Conjugate
///////////////////////
@ -582,7 +607,6 @@ template <class S, class V, IfNotComplex<S> = 0>
inline Grid_simd<S, V> conjugate(const Grid_simd<S, V> &in) {
return in; // for real objects
}
// Suppress adj for integer types... // odd; why conjugate above but not adj??
template <class S, class V, IfNotInteger<S> = 0>
inline Grid_simd<S, V> adj(const Grid_simd<S, V> &in) {
@ -596,14 +620,12 @@ template <class S, class V, IfComplex<S> = 0>
inline void timesMinusI(Grid_simd<S, V> &ret, const Grid_simd<S, V> &in) {
ret.v = binary<V>(in.v, ret.v, TimesMinusISIMD());
}
template <class S, class V, IfComplex<S> = 0>
inline Grid_simd<S, V> timesMinusI(const Grid_simd<S, V> &in) {
Grid_simd<S, V> ret;
timesMinusI(ret, in);
return ret;
}
template <class S, class V, IfNotComplex<S> = 0>
inline Grid_simd<S, V> timesMinusI(const Grid_simd<S, V> &in) {
return in;
@ -616,14 +638,12 @@ template <class S, class V, IfComplex<S> = 0>
inline void timesI(Grid_simd<S, V> &ret, const Grid_simd<S, V> &in) {
ret.v = binary<V>(in.v, ret.v, TimesISIMD());
}
template <class S, class V, IfComplex<S> = 0>
inline Grid_simd<S, V> timesI(const Grid_simd<S, V> &in) {
Grid_simd<S, V> ret;
timesI(ret, in);
return ret;
}
template <class S, class V, IfNotComplex<S> = 0>
inline Grid_simd<S, V> timesI(const Grid_simd<S, V> &in) {
return in;

30
lib/tensors/Tensor_arith_mul.h
View File

@ -126,6 +126,36 @@ iVector<rtype,N> operator * (const iVector<mtype,N>& lhs,const iScalar<vtype>& r
mult(&ret,&lhs,&rhs);
return ret;
}
//////////////////////////////////////////////////////////////////
// Divide by scalar
//////////////////////////////////////////////////////////////////
template<class rtype,class vtype> strong_inline
iScalar<rtype> operator / (const iScalar<rtype>& lhs,const iScalar<vtype>& rhs)
{
iScalar<rtype> ret;
ret._internal = lhs._internal/rhs._internal;
return ret;
}
template<class rtype,class vtype,int N> strong_inline
iVector<rtype,N> operator / (const iVector<rtype,N>& lhs,const iScalar<vtype>& rhs)
{
iVector<rtype,N> ret;
for(int i=0;i<N;i++){
ret._internal[i] = lhs._internal[i]/rhs._internal;
}
return ret;
}
template<class rtype,class vtype,int N> strong_inline
iMatrix<rtype,N> operator / (const iMatrix<rtype,N>& lhs,const iScalar<vtype>& rhs)
{
iMatrix<rtype,N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = lhs._internal[i][j]/rhs._internal;
}}
return ret;
}
//////////////////////////////////////////////////////////////////
// Glue operators to mult routines. Must resolve return type cleverly from typeof(internal)