portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-11-04 05:54:32 +00:00
Code Releases Activity

Merge branch 'develop' of https://github.com/paboyle/Grid into develop

Browse Source
This commit is contained in:
Guido Cossu
2018-05-02 14:29:25 +01:00
parent b458326744 6f6c5c549a
commit e568c24d1d
6 changed files with 1245 additions and 485 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
configure.ac
View File

@@ -340,7 +340,7 @@ case ${ac_PRECISION} in
esac esac
###################### Shared memory allocation technique under MPI3 ###################### Shared memory allocation technique under MPI3
AC_ARG_ENABLE([shm],[AC_HELP_STRING([--enable-shm=shmopen|hugetlbfs|shmnone], AC_ARG_ENABLE([shm],[AC_HELP_STRING([--enable-shm=shmopen|shmget|hugetlbfs|shmnone],
[Select SHM allocation technique])],[ac_SHM=${enable_shm}],[ac_SHM=shmopen]) [Select SHM allocation technique])],[ac_SHM=${enable_shm}],[ac_SHM=shmopen])
case ${ac_SHM} in case ${ac_SHM} in
@@ -349,6 +349,10 @@ case ${ac_SHM} in
AC_DEFINE([GRID_MPI3_SHMOPEN],[1],[GRID_MPI3_SHMOPEN] ) AC_DEFINE([GRID_MPI3_SHMOPEN],[1],[GRID_MPI3_SHMOPEN] )
;; ;;
shmget)
AC_DEFINE([GRID_MPI3_SHMGET],[1],[GRID_MPI3_SHMGET] )
;;
shmnone) shmnone)
AC_DEFINE([GRID_MPI3_SHM_NONE],[1],[GRID_MPI3_SHM_NONE] ) AC_DEFINE([GRID_MPI3_SHM_NONE],[1],[GRID_MPI3_SHM_NONE] )
;; ;;
@@ -366,7 +370,7 @@ esac
AC_ARG_ENABLE([shmpath],[AC_HELP_STRING([--enable-shmpath=path], AC_ARG_ENABLE([shmpath],[AC_HELP_STRING([--enable-shmpath=path],
[Select SHM mmap base path for hugetlbfs])], [Select SHM mmap base path for hugetlbfs])],
[ac_SHMPATH=${enable_shmpath}], [ac_SHMPATH=${enable_shmpath}],
[ac_SHMPATH=/var/lib/hugetlbfs/pagesize-2MB/]) [ac_SHMPATH=/var/lib/hugetlbfs/global/pagesize-2MB/])
AC_DEFINE_UNQUOTED([GRID_SHM_PATH],["$ac_SHMPATH"],[Path to a hugetlbfs filesystem for MMAPing]) AC_DEFINE_UNQUOTED([GRID_SHM_PATH],["$ac_SHMPATH"],[Path to a hugetlbfs filesystem for MMAPing])
############### communication type selection ############### communication type selection

215
lib/communicator/SharedMemoryMPI.cc
View File

@@ -114,19 +114,151 @@ void GlobalSharedMemory::Init(Grid_MPI_Comm comm)
assert(WorldNode!=-1); assert(WorldNode!=-1);
_ShmSetup=1; _ShmSetup=1;
} }
// Gray encode support
void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm) int BinaryToGray (int binary) {
int gray = (binary>>1)^binary;
return gray;
}
int Log2Size(int TwoToPower,int MAXLOG2)
{ {
////////////////////////////////////////////////////////////////
// Assert power of two shm_size.
////////////////////////////////////////////////////////////////
int log2size = -1; int log2size = -1;
for(int i=0;i<=MAXLOG2RANKSPERNODE;i++){ for(int i=0;i<=MAXLOG2;i++){
if ( (0x1<<i) == WorldShmSize ) { if ( (0x1<<i) == TwoToPower ) {
log2size = i; log2size = i;
break; break;
} }
} }
return log2size;
}
void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,Grid_MPI_Comm & optimal_comm)
{
#undef HYPERCUBE
#ifdef HYPERCUBE
////////////////////////////////////////////////////////////////
// Assert power of two shm_size.
////////////////////////////////////////////////////////////////
int log2size = Log2Size(WorldShmSize,MAXLOG2RANKSPERNODE);
assert(log2size != -1);
////////////////////////////////////////////////////////////////
// Identify the hypercube coordinate of this node using hostname
////////////////////////////////////////////////////////////////
// n runs 0...7 9...16 18...25 27...34 (8*4) 5 bits
// i runs 0..7 3 bits
// r runs 0..3 2 bits
// 2^10 = 1024 nodes
const int maxhdim = 10;
std::vector<int> HyperCubeCoords(maxhdim,0);
std::vector<int> RootHyperCubeCoords(maxhdim,0);
int R;
int I;
int N;
const int namelen = _POSIX_HOST_NAME_MAX;
char name[namelen];
// Parse ICE-XA hostname to get hypercube location
gethostname(name,namelen);
int nscan = sscanf(name,"r%di%dn%d",&R,&I,&N) ;
assert(nscan==3);
int nlo = N%9;
int nhi = N/9;
uint32_t hypercoor = (R<<8)|(I<<5)|(nhi<<3)|nlo ;
uint32_t rootcoor = hypercoor;
//////////////////////////////////////////////////////////////////
// Print debug info
//////////////////////////////////////////////////////////////////
for(int d=0;d<maxhdim;d++){
HyperCubeCoords[d] = (hypercoor>>d)&0x1;
}
std::string hname(name);
std::cout << "hostname "<<hname<<std::endl;
std::cout << "R " << R << " I " << I << " N "<< N<<
<< " hypercoor 0x"<<std::hex<<hypercoor<<std::dec<<std::endl;
//////////////////////////////////////////////////////////////////
// broadcast node 0's base coordinate for this partition.
//////////////////////////////////////////////////////////////////
MPI_Bcast(&rootcoor, sizeof(rootcoor), MPI_BYTE, 0, WorldComm);
hypercoor=hypercoor-rootcoor;
assert(hypercoor<WorldSize);
assert(hypercoor>=0);
//////////////////////////////////////
// Printing
//////////////////////////////////////
for(int d=0;d<maxhdim;d++){
HyperCubeCoords[d] = (hypercoor>>d)&0x1;
}
////////////////////////////////////////////////////////////////
// Identify subblock of ranks on node spreading across dims
// in a maximally symmetrical way
////////////////////////////////////////////////////////////////
int ndimension = processors.size();
std::vector<int> processor_coor(ndimension);
std::vector<int> WorldDims = processors; std::vector<int> ShmDims (ndimension,1); std::vector<int> NodeDims (ndimension);
std::vector<int> ShmCoor (ndimension); std::vector<int> NodeCoor (ndimension); std::vector<int> WorldCoor(ndimension);
std::vector<int> HyperCoor(ndimension);
int dim = 0;
for(int l2=0;l2<log2size;l2++){
while ( (WorldDims[dim] / ShmDims[dim]) <= 1 ) dim=(dim+1)%ndimension;
ShmDims[dim]*=2;
dim=(dim+1)%ndimension;
}
////////////////////////////////////////////////////////////////
// Establish torus of processes and nodes with sub-blockings
////////////////////////////////////////////////////////////////
for(int d=0;d<ndimension;d++){
NodeDims[d] = WorldDims[d]/ShmDims[d];
}
////////////////////////////////////////////////////////////////
// Map Hcube according to physical lattice
// must partition. Loop over dims and find out who would join.
////////////////////////////////////////////////////////////////
int hcoor = hypercoor;
for(int d=0;d<ndimension;d++){
int bits = Log2Size(NodeDims[d],MAXLOG2RANKSPERNODE);
int msk = (0x1<<bits)-1;
HyperCoor[d]=hcoor & msk;
HyperCoor[d]=BinaryToGray(HyperCoor[d]); // Space filling curve magic
hcoor = hcoor >> bits;
}
////////////////////////////////////////////////////////////////
// Check processor counts match
////////////////////////////////////////////////////////////////
int Nprocessors=1;
for(int i=0;i<ndimension;i++){
Nprocessors*=processors[i];
}
assert(WorldSize==Nprocessors);
////////////////////////////////////////////////////////////////
// Establish mapping between lexico physics coord and WorldRank
////////////////////////////////////////////////////////////////
int rank;
Lexicographic::CoorFromIndexReversed(NodeCoor,WorldNode ,NodeDims);
for(int d=0;d<ndimension;d++) NodeCoor[d]=HyperCoor[d];
Lexicographic::CoorFromIndexReversed(ShmCoor ,WorldShmRank,ShmDims);
for(int d=0;d<ndimension;d++) WorldCoor[d] = NodeCoor[d]*ShmDims[d]+ShmCoor[d];
Lexicographic::IndexFromCoorReversed(WorldCoor,rank,WorldDims);
/////////////////////////////////////////////////////////////////
// Build the new communicator
/////////////////////////////////////////////////////////////////
int ierr= MPI_Comm_split(WorldComm,0,rank,&optimal_comm);
assert(ierr==0);
#else
////////////////////////////////////////////////////////////////
// Assert power of two shm_size.
////////////////////////////////////////////////////////////////
int log2size = Log2Size(WorldShmSize,MAXLOG2RANKSPERNODE);
assert(log2size != -1); assert(log2size != -1);
//////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////
@@ -175,7 +307,69 @@ void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,
///////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////
int ierr= MPI_Comm_split(WorldComm,0,rank,&optimal_comm); int ierr= MPI_Comm_split(WorldComm,0,rank,&optimal_comm);
assert(ierr==0); assert(ierr==0);
#endif
} }
////////////////////////////////////////////////////////////////////////////////////////////
// SHMGET
////////////////////////////////////////////////////////////////////////////////////////////
#ifdef GRID_MPI3_SHMGET
void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
{
std::cout << "SharedMemoryAllocate "<< bytes<< " shmget implementation "<<std::endl;
assert(_ShmSetup==1);
assert(_ShmAlloc==0);
//////////////////////////////////////////////////////////////////////////////////////////////////////////
// allocate the shared windows for our group
//////////////////////////////////////////////////////////////////////////////////////////////////////////
MPI_Barrier(WorldShmComm);
WorldShmCommBufs.resize(WorldShmSize);
std::vector<int> shmids(WorldShmSize);
if ( WorldShmRank == 0 ) {
for(int r=0;r<WorldShmSize;r++){
size_t size = bytes;
key_t key = IPC_PRIVATE;
int flags = IPC_CREAT | SHM_R | SHM_W;
#ifdef SHM_HUGETLB
if (Hugepages) flags|=SHM_HUGETLB;
#endif
if ((shmids[r]= shmget(key,size, flags)) ==-1) {
int errsv = errno;
printf("Errno %d\n",errsv);
printf("key %d\n",key);
printf("size %lld\n",size);
printf("flags %d\n",flags);
perror("shmget");
exit(1);
}
}
}
MPI_Barrier(WorldShmComm);
MPI_Bcast(&shmids[0],WorldShmSize*sizeof(int),MPI_BYTE,0,WorldShmComm);
MPI_Barrier(WorldShmComm);
for(int r=0;r<WorldShmSize;r++){
WorldShmCommBufs[r] = (uint64_t *)shmat(shmids[r], NULL,0);
if (WorldShmCommBufs[r] == (uint64_t *)-1) {
perror("Shared memory attach failure");
shmctl(shmids[r], IPC_RMID, NULL);
exit(2);
}
}
MPI_Barrier(WorldShmComm);
///////////////////////////////////
// Mark for clean up
///////////////////////////////////
for(int r=0;r<WorldShmSize;r++){
shmctl(shmids[r], IPC_RMID,(struct shmid_ds *)NULL);
}
MPI_Barrier(WorldShmComm);
_ShmAlloc=1;
_ShmAllocBytes = bytes;
}
#endif
//////////////////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////////////////
// Hugetlbfs mapping intended // Hugetlbfs mapping intended
@@ -183,7 +377,7 @@ void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,
#ifdef GRID_MPI3_SHMMMAP #ifdef GRID_MPI3_SHMMMAP
void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags) void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
{ {
std::cout << "SharedMemoryAllocate "<< bytes<< " MMAP implementation "<<std::endl; std::cout << "SharedMemoryAllocate "<< bytes<< " MMAP implementation "<< GRID_SHM_PATH <<std::endl;
assert(_ShmSetup==1); assert(_ShmSetup==1);
assert(_ShmAlloc==0); assert(_ShmAlloc==0);
////////////////////////////////////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -193,7 +387,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
WorldShmCommBufs.resize(WorldShmSize); WorldShmCommBufs.resize(WorldShmSize);
//////////////////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////////////////
// Hugetlbf and others map filesystems as mappable huge pages // Hugetlbfs and others map filesystems as mappable huge pages
//////////////////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////////////////
char shm_name [NAME_MAX]; char shm_name [NAME_MAX];
for(int r=0;r<WorldShmSize;r++){ for(int r=0;r<WorldShmSize;r++){
@@ -344,6 +538,9 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
} }
#endif #endif
//////////////////////////////////////////////////////// ////////////////////////////////////////////////////////
// Global shared functionality finished // Global shared functionality finished
// Now move to per communicator functionality // Now move to per communicator functionality

614
lib/qcd/action/fermion/WilsonKernelsHand.cc
View File

@@ -30,181 +30,60 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define REGISTER #define REGISTER
#define LOAD_CHIMU_BODY(F) \ #define LOAD_CHIMU \
Chimu_00=ref(F)(0)(0); \ {const SiteSpinor & ref (in._odata[offset]); \
Chimu_01=ref(F)(0)(1); \ Chimu_00=ref()(0)(0);\
Chimu_02=ref(F)(0)(2); \ Chimu_01=ref()(0)(1);\
Chimu_10=ref(F)(1)(0); \ Chimu_02=ref()(0)(2);\
Chimu_11=ref(F)(1)(1); \ Chimu_10=ref()(1)(0);\
Chimu_12=ref(F)(1)(2); \ Chimu_11=ref()(1)(1);\
Chimu_20=ref(F)(2)(0); \ Chimu_12=ref()(1)(2);\
Chimu_21=ref(F)(2)(1); \ Chimu_20=ref()(2)(0);\
Chimu_22=ref(F)(2)(2); \ Chimu_21=ref()(2)(1);\
Chimu_30=ref(F)(3)(0); \ Chimu_22=ref()(2)(2);\
Chimu_31=ref(F)(3)(1); \ Chimu_30=ref()(3)(0);\
Chimu_32=ref(F)(3)(2) Chimu_31=ref()(3)(1);\
Chimu_32=ref()(3)(2);}
#define LOAD_CHIMU(DIR,F,PERM) \ #define LOAD_CHI\
{ const SiteSpinor & ref (in._odata[offset]); LOAD_CHIMU_BODY(F); } {const SiteHalfSpinor &ref(buf[offset]); \
Chi_00 = ref()(0)(0);\
#define LOAD_CHI_BODY(F) \ Chi_01 = ref()(0)(1);\
Chi_00 = ref(F)(0)(0);\ Chi_02 = ref()(0)(2);\
Chi_01 = ref(F)(0)(1);\ Chi_10 = ref()(1)(0);\
Chi_02 = ref(F)(0)(2);\ Chi_11 = ref()(1)(1);\
Chi_10 = ref(F)(1)(0);\ Chi_12 = ref()(1)(2);}
Chi_11 = ref(F)(1)(1);\
Chi_12 = ref(F)(1)(2)
#define LOAD_CHI(DIR,F,PERM) \
{const SiteHalfSpinor &ref(buf[offset]); LOAD_CHI_BODY(F); }
//G-parity implementations using in-place intrinsic ops
//1l 1h -> 1h 1l
//0l 0h , 1h 1l -> 0l 1h 0h,1l
//0h,1l -> 1l,0h
//if( (distance == 1 && !perm_will_occur) || (distance == -1 && perm_will_occur) )
//Pulled fermion through forwards face, GPBC on upper component
//Need 0= 0l 1h 1= 1l 0h
//else if( (distance == -1 && !perm) || (distance == 1 && perm) )
//Pulled fermion through backwards face, GPBC on lower component
//Need 0= 1l 0h 1= 0l 1h
//1l 1h -> 1h 1l
//0l 0h , 1h 1l -> 0l 1h 0h,1l
#define DO_TWIST_0L_1H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3) \
permute##PERM(tmp1, ref(1)(S)(C)); \
exchange##PERM(tmp2,tmp3, ref(0)(S)(C), tmp1); \
INTO = tmp2;
//0l 0h -> 0h 0l
//1l 1h, 0h 0l -> 1l 0h, 1h 0l
#define DO_TWIST_1L_0H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3) \
permute##PERM(tmp1, ref(0)(S)(C)); \
exchange##PERM(tmp2,tmp3, ref(1)(S)(C), tmp1); \
INTO = tmp2;
#define LOAD_CHI_SETUP(DIR,F) \
g = F; \
direction = st._directions[DIR]; \
distance = st._distances[DIR]; \
sl = st._grid->_simd_layout[direction]; \
inplace_twist = 0; \
if(SE->_around_the_world && this->Params.twists[DIR % 4]){ \
if(sl == 1){ \
g = (F+1) % 2; \
}else{ \
inplace_twist = 1; \
} \
}
#define LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM) \
{ const SiteSpinor &ref(in._odata[offset]); \
LOAD_CHI_SETUP(DIR,F); \
if(!inplace_twist){ \
LOAD_CHIMU_BODY(g); \
}else{ \
if( ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \
( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \
DO_TWIST_0L_1H(Chimu_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_02,0,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_10,1,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_12,1,2,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_20,2,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_21,2,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_22,2,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_30,3,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_31,3,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_32,3,2,F,PERM, U_11,U_20,U_21); \
}else{ \
DO_TWIST_1L_0H(Chimu_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_02,0,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_10,1,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_12,1,2,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_20,2,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_21,2,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_22,2,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_30,3,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_31,3,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_32,3,2,F,PERM, U_11,U_20,U_21); \
} \
} \
}
#define LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM) \
{ const SiteHalfSpinor &ref(buf[offset]); \
LOAD_CHI_SETUP(DIR,F); \
if(!inplace_twist){ \
LOAD_CHI_BODY(g); \
}else{ \
if( ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \
( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \
DO_TWIST_0L_1H(Chi_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chi_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chi_02,0,2,F,PERM, UChi_00,UChi_01,UChi_02); \
DO_TWIST_0L_1H(Chi_10,1,0,F,PERM, UChi_10,UChi_11,UChi_12); \
DO_TWIST_0L_1H(Chi_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chi_12,1,2,F,PERM, U_11,U_20,U_21); \
}else{ \
DO_TWIST_1L_0H(Chi_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chi_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chi_02,0,2,F,PERM, UChi_00,UChi_01,UChi_02); \
DO_TWIST_1L_0H(Chi_10,1,0,F,PERM, UChi_10,UChi_11,UChi_12); \
DO_TWIST_1L_0H(Chi_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chi_12,1,2,F,PERM, U_11,U_20,U_21); \
} \
} \
}
#define LOAD_CHI_GPARITY(DIR,F,PERM) LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM)
#define LOAD_CHIMU_GPARITY(DIR,F,PERM) LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM)
// To splat or not to splat depends on the implementation // To splat or not to splat depends on the implementation
#define MULT_2SPIN_BODY \ #define MULT_2SPIN(A)\
Impl::loadLinkElement(U_00,ref()(0,0)); \ {auto & ref(U._odata[sU](A)); \
Impl::loadLinkElement(U_10,ref()(1,0)); \ Impl::loadLinkElement(U_00,ref()(0,0)); \
Impl::loadLinkElement(U_20,ref()(2,0)); \ Impl::loadLinkElement(U_10,ref()(1,0)); \
Impl::loadLinkElement(U_01,ref()(0,1)); \ Impl::loadLinkElement(U_20,ref()(2,0)); \
Impl::loadLinkElement(U_11,ref()(1,1)); \ Impl::loadLinkElement(U_01,ref()(0,1)); \
Impl::loadLinkElement(U_21,ref()(2,1)); \ Impl::loadLinkElement(U_11,ref()(1,1)); \
UChi_00 = U_00*Chi_00; \ Impl::loadLinkElement(U_21,ref()(2,1)); \
UChi_10 = U_00*Chi_10; \ UChi_00 = U_00*Chi_00;\
UChi_01 = U_10*Chi_00; \ UChi_10 = U_00*Chi_10;\
UChi_11 = U_10*Chi_10; \ UChi_01 = U_10*Chi_00;\
UChi_02 = U_20*Chi_00; \ UChi_11 = U_10*Chi_10;\
UChi_12 = U_20*Chi_10; \ UChi_02 = U_20*Chi_00;\
UChi_00+= U_01*Chi_01; \ UChi_12 = U_20*Chi_10;\
UChi_10+= U_01*Chi_11; \ UChi_00+= U_01*Chi_01;\
UChi_01+= U_11*Chi_01; \ UChi_10+= U_01*Chi_11;\
UChi_11+= U_11*Chi_11; \ UChi_01+= U_11*Chi_01;\
UChi_02+= U_21*Chi_01; \ UChi_11+= U_11*Chi_11;\
UChi_12+= U_21*Chi_11; \ UChi_02+= U_21*Chi_01;\
Impl::loadLinkElement(U_00,ref()(0,2)); \ UChi_12+= U_21*Chi_11;\
Impl::loadLinkElement(U_10,ref()(1,2)); \ Impl::loadLinkElement(U_00,ref()(0,2)); \
Impl::loadLinkElement(U_20,ref()(2,2)); \ Impl::loadLinkElement(U_10,ref()(1,2)); \
UChi_00+= U_00*Chi_02; \ Impl::loadLinkElement(U_20,ref()(2,2)); \
UChi_10+= U_00*Chi_12; \ UChi_00+= U_00*Chi_02;\
UChi_01+= U_10*Chi_02; \ UChi_10+= U_00*Chi_12;\
UChi_11+= U_10*Chi_12; \ UChi_01+= U_10*Chi_02;\
UChi_02+= U_20*Chi_02; \ UChi_11+= U_10*Chi_12;\
UChi_12+= U_20*Chi_12 UChi_02+= U_20*Chi_02;\
UChi_12+= U_20*Chi_12;}
#define MULT_2SPIN(A,F) \
{auto & ref(U._odata[sU](A)); MULT_2SPIN_BODY; }
#define MULT_2SPIN_GPARITY(A,F) \
{auto & ref(U._odata[sU](F)(A)); MULT_2SPIN_BODY; }
#define PERMUTE_DIR(dir) \ #define PERMUTE_DIR(dir) \
@@ -428,87 +307,84 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
result_31-= UChi_11; \ result_31-= UChi_11; \
result_32-= UChi_12; result_32-= UChi_12;
#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ #define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON) \
SE=st.GetEntry(ptype,DIR,ss); \ SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \ offset = SE->_offset; \
local = SE->_is_local; \ local = SE->_is_local; \
perm = SE->_permute; \ perm = SE->_permute; \
if ( local ) { \ if ( local ) { \
LOAD_CHIMU_IMPL(DIR,F,PERM); \ LOAD_CHIMU; \
PROJ; \ PROJ; \
if ( perm) { \ if ( perm) { \
PERMUTE_DIR(PERM); \ PERMUTE_DIR(PERM); \
} \ } \
} else { \ } else { \
LOAD_CHI_IMPL(DIR,F,PERM); \ LOAD_CHI; \
} \ } \
MULT_2SPIN_IMPL(DIR,F); \ MULT_2SPIN(DIR); \
RECON; RECON;
#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON) \
#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
SE=st.GetEntry(ptype,DIR,ss); \ SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \ offset = SE->_offset; \
local = SE->_is_local; \ local = SE->_is_local; \
perm = SE->_permute; \ perm = SE->_permute; \
if ( local ) { \ if ( local ) { \
LOAD_CHIMU_IMPL(DIR,F,PERM); \ LOAD_CHIMU; \
PROJ; \ PROJ; \
if ( perm) { \ if ( perm) { \
PERMUTE_DIR(PERM); \ PERMUTE_DIR(PERM); \
} \ } \
} else if ( st.same_node[DIR] ) { \ } else if ( st.same_node[DIR] ) { \
LOAD_CHI_IMPL(DIR,F,PERM); \ LOAD_CHI; \
} \ } \
if (local || st.same_node[DIR] ) { \ if (local || st.same_node[DIR] ) { \
MULT_2SPIN_IMPL(DIR,F); \ MULT_2SPIN(DIR); \
RECON; \ RECON; \
} }
#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ #define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON) \
SE=st.GetEntry(ptype,DIR,ss); \ SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \ offset = SE->_offset; \
local = SE->_is_local; \
perm = SE->_permute; \
if((!SE->_is_local)&&(!st.same_node[DIR]) ) { \ if((!SE->_is_local)&&(!st.same_node[DIR]) ) { \
LOAD_CHI_IMPL(DIR,F,PERM); \ LOAD_CHI; \
MULT_2SPIN_IMPL(DIR,F); \ MULT_2SPIN(DIR); \
RECON; \ RECON; \
nmu++; \ nmu++; \
} }
#define HAND_RESULT(ss,F) \ #define HAND_RESULT(ss) \
{ \ { \
SiteSpinor & ref (out._odata[ss]); \ SiteSpinor & ref (out._odata[ss]); \
vstream(ref(F)(0)(0),result_00); \ vstream(ref()(0)(0),result_00); \
vstream(ref(F)(0)(1),result_01); \ vstream(ref()(0)(1),result_01); \
vstream(ref(F)(0)(2),result_02); \ vstream(ref()(0)(2),result_02); \
vstream(ref(F)(1)(0),result_10); \ vstream(ref()(1)(0),result_10); \
vstream(ref(F)(1)(1),result_11); \ vstream(ref()(1)(1),result_11); \
vstream(ref(F)(1)(2),result_12); \ vstream(ref()(1)(2),result_12); \
vstream(ref(F)(2)(0),result_20); \ vstream(ref()(2)(0),result_20); \
vstream(ref(F)(2)(1),result_21); \ vstream(ref()(2)(1),result_21); \
vstream(ref(F)(2)(2),result_22); \ vstream(ref()(2)(2),result_22); \
vstream(ref(F)(3)(0),result_30); \ vstream(ref()(3)(0),result_30); \
vstream(ref(F)(3)(1),result_31); \ vstream(ref()(3)(1),result_31); \
vstream(ref(F)(3)(2),result_32); \ vstream(ref()(3)(2),result_32); \
} }
#define HAND_RESULT_EXT(ss,F) \ #define HAND_RESULT_EXT(ss) \
if (nmu){ \ if (nmu){ \
SiteSpinor & ref (out._odata[ss]); \ SiteSpinor & ref (out._odata[ss]); \
ref(F)(0)(0)+=result_00; \ ref()(0)(0)+=result_00; \
ref(F)(0)(1)+=result_01; \ ref()(0)(1)+=result_01; \
ref(F)(0)(2)+=result_02; \ ref()(0)(2)+=result_02; \
ref(F)(1)(0)+=result_10; \ ref()(1)(0)+=result_10; \
ref(F)(1)(1)+=result_11; \ ref()(1)(1)+=result_11; \
ref(F)(1)(2)+=result_12; \ ref()(1)(2)+=result_12; \
ref(F)(2)(0)+=result_20; \ ref()(2)(0)+=result_20; \
ref(F)(2)(1)+=result_21; \ ref()(2)(1)+=result_21; \
ref(F)(2)(2)+=result_22; \ ref()(2)(2)+=result_22; \
ref(F)(3)(0)+=result_30; \ ref()(3)(0)+=result_30; \
ref(F)(3)(1)+=result_31; \ ref()(3)(1)+=result_31; \
ref(F)(3)(2)+=result_32; \ ref()(3)(2)+=result_32; \
} }
@@ -587,18 +463,15 @@ WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGauge
int offset,local,perm, ptype; int offset,local,perm, ptype;
StencilEntry *SE; StencilEntry *SE;
#define HAND_DOP_SITE(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM);
HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM);
HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM);
HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM);
HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_RESULT(ss);
HAND_RESULT(ss,F)
HAND_DOP_SITE(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
} }
template<class Impl> template<class Impl>
@@ -613,18 +486,15 @@ void WilsonKernels<Impl>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,Doub
StencilEntry *SE; StencilEntry *SE;
int offset,local,perm, ptype; int offset,local,perm, ptype;
#define HAND_DOP_SITE_DAG(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON);
HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM);
HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM);
HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM);
HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM);
HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM);
HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_RESULT(ss);
HAND_RESULT(ss,F)
HAND_DOP_SITE_DAG(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
} }
template<class Impl> void template<class Impl> void
@@ -639,20 +509,16 @@ WilsonKernels<Impl>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGa
int offset,local,perm, ptype; int offset,local,perm, ptype;
StencilEntry *SE; StencilEntry *SE;
ZERO_RESULT;
#define HAND_DOP_SITE_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
ZERO_RESULT; \ HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_RESULT(ss);
HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT(ss,F)
HAND_DOP_SITE_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
} }
template<class Impl> template<class Impl>
@@ -666,20 +532,16 @@ void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,D
StencilEntry *SE; StencilEntry *SE;
int offset,local,perm, ptype; int offset,local,perm, ptype;
ZERO_RESULT;
#define HAND_DOP_SITE_DAG_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
ZERO_RESULT; \ HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_RESULT(ss);
HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT(ss,F)
HAND_DOP_SITE_DAG_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
} }
template<class Impl> void template<class Impl> void
@@ -695,20 +557,16 @@ WilsonKernels<Impl>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGa
int offset,local,perm, ptype; int offset,local,perm, ptype;
StencilEntry *SE; StencilEntry *SE;
int nmu=0; int nmu=0;
ZERO_RESULT;
#define HAND_DOP_SITE_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
ZERO_RESULT; \ HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_RESULT_EXT(ss);
HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT_EXT(ss,F)
HAND_DOP_SITE_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
} }
template<class Impl> template<class Impl>
@@ -723,193 +581,18 @@ void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,D
StencilEntry *SE; StencilEntry *SE;
int offset,local,perm, ptype; int offset,local,perm, ptype;
int nmu=0; int nmu=0;
ZERO_RESULT;
#define HAND_DOP_SITE_DAG_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
ZERO_RESULT; \ HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ HAND_RESULT_EXT(ss);
HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT_EXT(ss,F)
HAND_DOP_SITE_DAG_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
} }
////////////////////////////////////////////////
// Specialise Gparity to simple implementation
////////////////////////////////////////////////
#define HAND_SPECIALISE_EMPTY(IMPL) \
template<> void \
WilsonKernels<IMPL>::HandDhopSite(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
template<> void \
WilsonKernels<IMPL>::HandDhopSiteDag(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
template<> void \
WilsonKernels<IMPL>::HandDhopSiteInt(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
template<> void \
WilsonKernels<IMPL>::HandDhopSiteExt(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
template<> void \
WilsonKernels<IMPL>::HandDhopSiteDagInt(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
template<> void \
WilsonKernels<IMPL>::HandDhopSiteDagExt(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
#define HAND_SPECIALISE_GPARITY(IMPL) \
template<> void \
WilsonKernels<IMPL>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
StencilEntry *SE; \
HAND_DOP_SITE(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> \
void WilsonKernels<IMPL>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
StencilEntry *SE; \
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
HAND_DOP_SITE_DAG(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE_DAG(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> void \
WilsonKernels<IMPL>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
StencilEntry *SE; \
HAND_DOP_SITE_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> \
void WilsonKernels<IMPL>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
StencilEntry *SE; \
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
HAND_DOP_SITE_DAG_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE_DAG_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> void \
WilsonKernels<IMPL>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
StencilEntry *SE; \
int nmu=0; \
HAND_DOP_SITE_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
nmu = 0; \
HAND_DOP_SITE_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
template<> \
void WilsonKernels<IMPL>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
StencilEntry *SE; \
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
int nmu=0; \
HAND_DOP_SITE_DAG_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
nmu = 0; \
HAND_DOP_SITE_DAG_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
}
HAND_SPECIALISE_GPARITY(GparityWilsonImplF);
HAND_SPECIALISE_GPARITY(GparityWilsonImplD);
HAND_SPECIALISE_GPARITY(GparityWilsonImplFH);
HAND_SPECIALISE_GPARITY(GparityWilsonImplDF);
////////////// Wilson ; uses this implementation ///////////////////// ////////////// Wilson ; uses this implementation /////////////////////
#define INSTANTIATE_THEM(A) \ #define INSTANTIATE_THEM(A) \
@@ -930,8 +613,6 @@ INSTANTIATE_THEM(WilsonImplF);
INSTANTIATE_THEM(WilsonImplD); INSTANTIATE_THEM(WilsonImplD);
INSTANTIATE_THEM(ZWilsonImplF); INSTANTIATE_THEM(ZWilsonImplF);
INSTANTIATE_THEM(ZWilsonImplD); INSTANTIATE_THEM(ZWilsonImplD);
INSTANTIATE_THEM(GparityWilsonImplF);
INSTANTIATE_THEM(GparityWilsonImplD);
INSTANTIATE_THEM(DomainWallVec5dImplF); INSTANTIATE_THEM(DomainWallVec5dImplF);
INSTANTIATE_THEM(DomainWallVec5dImplD); INSTANTIATE_THEM(DomainWallVec5dImplD);
INSTANTIATE_THEM(ZDomainWallVec5dImplF); INSTANTIATE_THEM(ZDomainWallVec5dImplF);
@@ -940,12 +621,11 @@ INSTANTIATE_THEM(WilsonImplFH);
INSTANTIATE_THEM(WilsonImplDF); INSTANTIATE_THEM(WilsonImplDF);
INSTANTIATE_THEM(ZWilsonImplFH); INSTANTIATE_THEM(ZWilsonImplFH);
INSTANTIATE_THEM(ZWilsonImplDF); INSTANTIATE_THEM(ZWilsonImplDF);
INSTANTIATE_THEM(GparityWilsonImplFH);
INSTANTIATE_THEM(GparityWilsonImplDF);
INSTANTIATE_THEM(DomainWallVec5dImplFH); INSTANTIATE_THEM(DomainWallVec5dImplFH);
INSTANTIATE_THEM(DomainWallVec5dImplDF); INSTANTIATE_THEM(DomainWallVec5dImplDF);
INSTANTIATE_THEM(ZDomainWallVec5dImplFH); INSTANTIATE_THEM(ZDomainWallVec5dImplFH);
INSTANTIATE_THEM(ZDomainWallVec5dImplDF); INSTANTIATE_THEM(ZDomainWallVec5dImplDF);
INSTANTIATE_THEM(WilsonTwoIndexAntiSymmetricImplF); INSTANTIATE_THEM(WilsonTwoIndexAntiSymmetricImplF);
INSTANTIATE_THEM(WilsonTwoIndexAntiSymmetricImplD); INSTANTIATE_THEM(WilsonTwoIndexAntiSymmetricImplD);
}} }}

878
lib/qcd/action/fermion/WilsonKernelsHandGparity.cc Normal file
View File

@@ -0,0 +1,878 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/fermion/WilsonKernelsHand.cc
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
This program 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.
This program 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 this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/qcd/action/fermion/FermionCore.h>
#define REGISTER
#define LOAD_CHIMU_BODY(F) \
Chimu_00=ref(F)(0)(0); \
Chimu_01=ref(F)(0)(1); \
Chimu_02=ref(F)(0)(2); \
Chimu_10=ref(F)(1)(0); \
Chimu_11=ref(F)(1)(1); \
Chimu_12=ref(F)(1)(2); \
Chimu_20=ref(F)(2)(0); \
Chimu_21=ref(F)(2)(1); \
Chimu_22=ref(F)(2)(2); \
Chimu_30=ref(F)(3)(0); \
Chimu_31=ref(F)(3)(1); \
Chimu_32=ref(F)(3)(2)
#define LOAD_CHIMU(DIR,F,PERM) \
{ const SiteSpinor & ref (in._odata[offset]); LOAD_CHIMU_BODY(F); }
#define LOAD_CHI_BODY(F) \
Chi_00 = ref(F)(0)(0);\
Chi_01 = ref(F)(0)(1);\
Chi_02 = ref(F)(0)(2);\
Chi_10 = ref(F)(1)(0);\
Chi_11 = ref(F)(1)(1);\
Chi_12 = ref(F)(1)(2)
#define LOAD_CHI(DIR,F,PERM) \
{const SiteHalfSpinor &ref(buf[offset]); LOAD_CHI_BODY(F); }
//G-parity implementations using in-place intrinsic ops
//1l 1h -> 1h 1l
//0l 0h , 1h 1l -> 0l 1h 0h,1l
//0h,1l -> 1l,0h
//if( (distance == 1 && !perm_will_occur) || (distance == -1 && perm_will_occur) )
//Pulled fermion through forwards face, GPBC on upper component
//Need 0= 0l 1h 1= 1l 0h
//else if( (distance == -1 && !perm) || (distance == 1 && perm) )
//Pulled fermion through backwards face, GPBC on lower component
//Need 0= 1l 0h 1= 0l 1h
//1l 1h -> 1h 1l
//0l 0h , 1h 1l -> 0l 1h 0h,1l
#define DO_TWIST_0L_1H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3) \
permute##PERM(tmp1, ref(1)(S)(C)); \
exchange##PERM(tmp2,tmp3, ref(0)(S)(C), tmp1); \
INTO = tmp2;
//0l 0h -> 0h 0l
//1l 1h, 0h 0l -> 1l 0h, 1h 0l
#define DO_TWIST_1L_0H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3) \
permute##PERM(tmp1, ref(0)(S)(C)); \
exchange##PERM(tmp2,tmp3, ref(1)(S)(C), tmp1); \
INTO = tmp2;
#define LOAD_CHI_SETUP(DIR,F) \
g = F; \
direction = st._directions[DIR]; \
distance = st._distances[DIR]; \
sl = st._grid->_simd_layout[direction]; \
inplace_twist = 0; \
if(SE->_around_the_world && this->Params.twists[DIR % 4]){ \
if(sl == 1){ \
g = (F+1) % 2; \
}else{ \
inplace_twist = 1; \
} \
}
#define LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM) \
{ const SiteSpinor &ref(in._odata[offset]); \
LOAD_CHI_SETUP(DIR,F); \
if(!inplace_twist){ \
LOAD_CHIMU_BODY(g); \
}else{ \
if( ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \
( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \
DO_TWIST_0L_1H(Chimu_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_02,0,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_10,1,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_12,1,2,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_20,2,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_21,2,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_22,2,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_30,3,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chimu_31,3,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chimu_32,3,2,F,PERM, U_11,U_20,U_21); \
}else{ \
DO_TWIST_1L_0H(Chimu_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_02,0,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_10,1,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_12,1,2,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_20,2,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_21,2,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_22,2,2,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_30,3,0,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chimu_31,3,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chimu_32,3,2,F,PERM, U_11,U_20,U_21); \
} \
} \
}
#define LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM) \
{ const SiteHalfSpinor &ref(buf[offset]); \
LOAD_CHI_SETUP(DIR,F); \
if(!inplace_twist){ \
LOAD_CHI_BODY(g); \
}else{ \
if( ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \
( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \
DO_TWIST_0L_1H(Chi_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chi_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_0L_1H(Chi_02,0,2,F,PERM, UChi_00,UChi_01,UChi_02); \
DO_TWIST_0L_1H(Chi_10,1,0,F,PERM, UChi_10,UChi_11,UChi_12); \
DO_TWIST_0L_1H(Chi_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_0L_1H(Chi_12,1,2,F,PERM, U_11,U_20,U_21); \
}else{ \
DO_TWIST_1L_0H(Chi_00,0,0,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chi_01,0,1,F,PERM, U_11,U_20,U_21); \
DO_TWIST_1L_0H(Chi_02,0,2,F,PERM, UChi_00,UChi_01,UChi_02); \
DO_TWIST_1L_0H(Chi_10,1,0,F,PERM, UChi_10,UChi_11,UChi_12); \
DO_TWIST_1L_0H(Chi_11,1,1,F,PERM, U_00,U_01,U_10); \
DO_TWIST_1L_0H(Chi_12,1,2,F,PERM, U_11,U_20,U_21); \
} \
} \
}
#define LOAD_CHI_GPARITY(DIR,F,PERM) LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM)
#define LOAD_CHIMU_GPARITY(DIR,F,PERM) LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM)
// To splat or not to splat depends on the implementation
#define MULT_2SPIN_BODY \
Impl::loadLinkElement(U_00,ref()(0,0)); \
Impl::loadLinkElement(U_10,ref()(1,0)); \
Impl::loadLinkElement(U_20,ref()(2,0)); \
Impl::loadLinkElement(U_01,ref()(0,1)); \
Impl::loadLinkElement(U_11,ref()(1,1)); \
Impl::loadLinkElement(U_21,ref()(2,1)); \
UChi_00 = U_00*Chi_00; \
UChi_10 = U_00*Chi_10; \
UChi_01 = U_10*Chi_00; \
UChi_11 = U_10*Chi_10; \
UChi_02 = U_20*Chi_00; \
UChi_12 = U_20*Chi_10; \
UChi_00+= U_01*Chi_01; \
UChi_10+= U_01*Chi_11; \
UChi_01+= U_11*Chi_01; \
UChi_11+= U_11*Chi_11; \
UChi_02+= U_21*Chi_01; \
UChi_12+= U_21*Chi_11; \
Impl::loadLinkElement(U_00,ref()(0,2)); \
Impl::loadLinkElement(U_10,ref()(1,2)); \
Impl::loadLinkElement(U_20,ref()(2,2)); \
UChi_00+= U_00*Chi_02; \
UChi_10+= U_00*Chi_12; \
UChi_01+= U_10*Chi_02; \
UChi_11+= U_10*Chi_12; \
UChi_02+= U_20*Chi_02; \
UChi_12+= U_20*Chi_12
#define MULT_2SPIN(A,F) \
{auto & ref(U._odata[sU](A)); MULT_2SPIN_BODY; }
#define MULT_2SPIN_GPARITY(A,F) \
{auto & ref(U._odata[sU](F)(A)); MULT_2SPIN_BODY; }
#define PERMUTE_DIR(dir) \
permute##dir(Chi_00,Chi_00);\
permute##dir(Chi_01,Chi_01);\
permute##dir(Chi_02,Chi_02);\
permute##dir(Chi_10,Chi_10);\
permute##dir(Chi_11,Chi_11);\
permute##dir(Chi_12,Chi_12);
// hspin(0)=fspin(0)+timesI(fspin(3));
// hspin(1)=fspin(1)+timesI(fspin(2));
#define XP_PROJ \
Chi_00 = Chimu_00+timesI(Chimu_30);\
Chi_01 = Chimu_01+timesI(Chimu_31);\
Chi_02 = Chimu_02+timesI(Chimu_32);\
Chi_10 = Chimu_10+timesI(Chimu_20);\
Chi_11 = Chimu_11+timesI(Chimu_21);\
Chi_12 = Chimu_12+timesI(Chimu_22);
#define YP_PROJ \
Chi_00 = Chimu_00-Chimu_30;\
Chi_01 = Chimu_01-Chimu_31;\
Chi_02 = Chimu_02-Chimu_32;\
Chi_10 = Chimu_10+Chimu_20;\
Chi_11 = Chimu_11+Chimu_21;\
Chi_12 = Chimu_12+Chimu_22;
#define ZP_PROJ \
Chi_00 = Chimu_00+timesI(Chimu_20); \
Chi_01 = Chimu_01+timesI(Chimu_21); \
Chi_02 = Chimu_02+timesI(Chimu_22); \
Chi_10 = Chimu_10-timesI(Chimu_30); \
Chi_11 = Chimu_11-timesI(Chimu_31); \
Chi_12 = Chimu_12-timesI(Chimu_32);
#define TP_PROJ \
Chi_00 = Chimu_00+Chimu_20; \
Chi_01 = Chimu_01+Chimu_21; \
Chi_02 = Chimu_02+Chimu_22; \
Chi_10 = Chimu_10+Chimu_30; \
Chi_11 = Chimu_11+Chimu_31; \
Chi_12 = Chimu_12+Chimu_32;
// hspin(0)=fspin(0)-timesI(fspin(3));
// hspin(1)=fspin(1)-timesI(fspin(2));
#define XM_PROJ \
Chi_00 = Chimu_00-timesI(Chimu_30);\
Chi_01 = Chimu_01-timesI(Chimu_31);\
Chi_02 = Chimu_02-timesI(Chimu_32);\
Chi_10 = Chimu_10-timesI(Chimu_20);\
Chi_11 = Chimu_11-timesI(Chimu_21);\
Chi_12 = Chimu_12-timesI(Chimu_22);
#define YM_PROJ \
Chi_00 = Chimu_00+Chimu_30;\
Chi_01 = Chimu_01+Chimu_31;\
Chi_02 = Chimu_02+Chimu_32;\
Chi_10 = Chimu_10-Chimu_20;\
Chi_11 = Chimu_11-Chimu_21;\
Chi_12 = Chimu_12-Chimu_22;
#define ZM_PROJ \
Chi_00 = Chimu_00-timesI(Chimu_20); \
Chi_01 = Chimu_01-timesI(Chimu_21); \
Chi_02 = Chimu_02-timesI(Chimu_22); \
Chi_10 = Chimu_10+timesI(Chimu_30); \
Chi_11 = Chimu_11+timesI(Chimu_31); \
Chi_12 = Chimu_12+timesI(Chimu_32);
#define TM_PROJ \
Chi_00 = Chimu_00-Chimu_20; \
Chi_01 = Chimu_01-Chimu_21; \
Chi_02 = Chimu_02-Chimu_22; \
Chi_10 = Chimu_10-Chimu_30; \
Chi_11 = Chimu_11-Chimu_31; \
Chi_12 = Chimu_12-Chimu_32;
// fspin(0)=hspin(0);
// fspin(1)=hspin(1);
// fspin(2)=timesMinusI(hspin(1));
// fspin(3)=timesMinusI(hspin(0));
#define XP_RECON\
result_00 = UChi_00;\
result_01 = UChi_01;\
result_02 = UChi_02;\
result_10 = UChi_10;\
result_11 = UChi_11;\
result_12 = UChi_12;\
result_20 = timesMinusI(UChi_10);\
result_21 = timesMinusI(UChi_11);\
result_22 = timesMinusI(UChi_12);\
result_30 = timesMinusI(UChi_00);\
result_31 = timesMinusI(UChi_01);\
result_32 = timesMinusI(UChi_02);
#define XP_RECON_ACCUM\
result_00+=UChi_00;\
result_01+=UChi_01;\
result_02+=UChi_02;\
result_10+=UChi_10;\
result_11+=UChi_11;\
result_12+=UChi_12;\
result_20-=timesI(UChi_10);\
result_21-=timesI(UChi_11);\
result_22-=timesI(UChi_12);\
result_30-=timesI(UChi_00);\
result_31-=timesI(UChi_01);\
result_32-=timesI(UChi_02);
#define XM_RECON\
result_00 = UChi_00;\
result_01 = UChi_01;\
result_02 = UChi_02;\
result_10 = UChi_10;\
result_11 = UChi_11;\
result_12 = UChi_12;\
result_20 = timesI(UChi_10);\
result_21 = timesI(UChi_11);\
result_22 = timesI(UChi_12);\
result_30 = timesI(UChi_00);\
result_31 = timesI(UChi_01);\
result_32 = timesI(UChi_02);
#define XM_RECON_ACCUM\
result_00+= UChi_00;\
result_01+= UChi_01;\
result_02+= UChi_02;\
result_10+= UChi_10;\
result_11+= UChi_11;\
result_12+= UChi_12;\
result_20+= timesI(UChi_10);\
result_21+= timesI(UChi_11);\
result_22+= timesI(UChi_12);\
result_30+= timesI(UChi_00);\
result_31+= timesI(UChi_01);\
result_32+= timesI(UChi_02);
#define YP_RECON_ACCUM\
result_00+= UChi_00;\
result_01+= UChi_01;\
result_02+= UChi_02;\
result_10+= UChi_10;\
result_11+= UChi_11;\
result_12+= UChi_12;\
result_20+= UChi_10;\
result_21+= UChi_11;\
result_22+= UChi_12;\
result_30-= UChi_00;\
result_31-= UChi_01;\
result_32-= UChi_02;
#define YM_RECON_ACCUM\
result_00+= UChi_00;\
result_01+= UChi_01;\
result_02+= UChi_02;\
result_10+= UChi_10;\
result_11+= UChi_11;\
result_12+= UChi_12;\
result_20-= UChi_10;\
result_21-= UChi_11;\
result_22-= UChi_12;\
result_30+= UChi_00;\
result_31+= UChi_01;\
result_32+= UChi_02;
#define ZP_RECON_ACCUM\
result_00+= UChi_00;\
result_01+= UChi_01;\
result_02+= UChi_02;\
result_10+= UChi_10;\
result_11+= UChi_11;\
result_12+= UChi_12;\
result_20-= timesI(UChi_00); \
result_21-= timesI(UChi_01); \
result_22-= timesI(UChi_02); \
result_30+= timesI(UChi_10); \
result_31+= timesI(UChi_11); \
result_32+= timesI(UChi_12);
#define ZM_RECON_ACCUM\
result_00+= UChi_00;\
result_01+= UChi_01;\
result_02+= UChi_02;\
result_10+= UChi_10;\
result_11+= UChi_11;\
result_12+= UChi_12;\
result_20+= timesI(UChi_00); \
result_21+= timesI(UChi_01); \
result_22+= timesI(UChi_02); \
result_30-= timesI(UChi_10); \
result_31-= timesI(UChi_11); \
result_32-= timesI(UChi_12);
#define TP_RECON_ACCUM\
result_00+= UChi_00;\
result_01+= UChi_01;\
result_02+= UChi_02;\
result_10+= UChi_10;\
result_11+= UChi_11;\
result_12+= UChi_12;\
result_20+= UChi_00; \
result_21+= UChi_01; \
result_22+= UChi_02; \
result_30+= UChi_10; \
result_31+= UChi_11; \
result_32+= UChi_12;
#define TM_RECON_ACCUM\
result_00+= UChi_00;\
result_01+= UChi_01;\
result_02+= UChi_02;\
result_10+= UChi_10;\
result_11+= UChi_11;\
result_12+= UChi_12;\
result_20-= UChi_00; \
result_21-= UChi_01; \
result_22-= UChi_02; \
result_30-= UChi_10; \
result_31-= UChi_11; \
result_32-= UChi_12;
#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \
local = SE->_is_local; \
perm = SE->_permute; \
if ( local ) { \
LOAD_CHIMU_IMPL(DIR,F,PERM); \
PROJ; \
if ( perm) { \
PERMUTE_DIR(PERM); \
} \
} else { \
LOAD_CHI_IMPL(DIR,F,PERM); \
} \
MULT_2SPIN_IMPL(DIR,F); \
RECON;
#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \
local = SE->_is_local; \
perm = SE->_permute; \
if ( local ) { \
LOAD_CHIMU_IMPL(DIR,F,PERM); \
PROJ; \
if ( perm) { \
PERMUTE_DIR(PERM); \
} \
} else if ( st.same_node[DIR] ) { \
LOAD_CHI_IMPL(DIR,F,PERM); \
} \
if (local || st.same_node[DIR] ) { \
MULT_2SPIN_IMPL(DIR,F); \
RECON; \
}
#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \
local = SE->_is_local; \
perm = SE->_permute; \
if((!SE->_is_local)&&(!st.same_node[DIR]) ) { \
LOAD_CHI_IMPL(DIR,F,PERM); \
MULT_2SPIN_IMPL(DIR,F); \
RECON; \
nmu++; \
}
#define HAND_RESULT(ss,F) \
{ \
SiteSpinor & ref (out._odata[ss]); \
vstream(ref(F)(0)(0),result_00); \
vstream(ref(F)(0)(1),result_01); \
vstream(ref(F)(0)(2),result_02); \
vstream(ref(F)(1)(0),result_10); \
vstream(ref(F)(1)(1),result_11); \
vstream(ref(F)(1)(2),result_12); \
vstream(ref(F)(2)(0),result_20); \
vstream(ref(F)(2)(1),result_21); \
vstream(ref(F)(2)(2),result_22); \
vstream(ref(F)(3)(0),result_30); \
vstream(ref(F)(3)(1),result_31); \
vstream(ref(F)(3)(2),result_32); \
}
#define HAND_RESULT_EXT(ss,F) \
if (nmu){ \
SiteSpinor & ref (out._odata[ss]); \
ref(F)(0)(0)+=result_00; \
ref(F)(0)(1)+=result_01; \
ref(F)(0)(2)+=result_02; \
ref(F)(1)(0)+=result_10; \
ref(F)(1)(1)+=result_11; \
ref(F)(1)(2)+=result_12; \
ref(F)(2)(0)+=result_20; \
ref(F)(2)(1)+=result_21; \
ref(F)(2)(2)+=result_22; \
ref(F)(3)(0)+=result_30; \
ref(F)(3)(1)+=result_31; \
ref(F)(3)(2)+=result_32; \
}
#define HAND_DECLARATIONS(a) \
Simd result_00; \
Simd result_01; \
Simd result_02; \
Simd result_10; \
Simd result_11; \
Simd result_12; \
Simd result_20; \
Simd result_21; \
Simd result_22; \
Simd result_30; \
Simd result_31; \
Simd result_32; \
Simd Chi_00; \
Simd Chi_01; \
Simd Chi_02; \
Simd Chi_10; \
Simd Chi_11; \
Simd Chi_12; \
Simd UChi_00; \
Simd UChi_01; \
Simd UChi_02; \
Simd UChi_10; \
Simd UChi_11; \
Simd UChi_12; \
Simd U_00; \
Simd U_10; \
Simd U_20; \
Simd U_01; \
Simd U_11; \
Simd U_21;
#define ZERO_RESULT \
result_00=zero; \
result_01=zero; \
result_02=zero; \
result_10=zero; \
result_11=zero; \
result_12=zero; \
result_20=zero; \
result_21=zero; \
result_22=zero; \
result_30=zero; \
result_31=zero; \
result_32=zero;
#define Chimu_00 Chi_00
#define Chimu_01 Chi_01
#define Chimu_02 Chi_02
#define Chimu_10 Chi_10
#define Chimu_11 Chi_11
#define Chimu_12 Chi_12
#define Chimu_20 UChi_00
#define Chimu_21 UChi_01
#define Chimu_22 UChi_02
#define Chimu_30 UChi_10
#define Chimu_31 UChi_11
#define Chimu_32 UChi_12
namespace Grid {
namespace QCD {
template<class Impl> void
WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
HAND_DECLARATIONS(ignore);
int offset,local,perm, ptype;
StencilEntry *SE;
#define HAND_DOP_SITE(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT(ss,F)
HAND_DOP_SITE(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
template<class Impl>
void WilsonKernels<Impl>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
HAND_DECLARATIONS(ignore);
StencilEntry *SE;
int offset,local,perm, ptype;
#define HAND_DOP_SITE_DAG(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT(ss,F)
HAND_DOP_SITE_DAG(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
template<class Impl> void
WilsonKernels<Impl>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
HAND_DECLARATIONS(ignore);
int offset,local,perm, ptype;
StencilEntry *SE;
#define HAND_DOP_SITE_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
ZERO_RESULT; \
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT(ss,F)
HAND_DOP_SITE_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
template<class Impl>
void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
HAND_DECLARATIONS(ignore);
StencilEntry *SE;
int offset,local,perm, ptype;
#define HAND_DOP_SITE_DAG_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
ZERO_RESULT; \
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT(ss,F)
HAND_DOP_SITE_DAG_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
template<class Impl> void
WilsonKernels<Impl>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
HAND_DECLARATIONS(ignore);
int offset,local,perm, ptype;
StencilEntry *SE;
int nmu=0;
#define HAND_DOP_SITE_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
ZERO_RESULT; \
HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT_EXT(ss,F)
HAND_DOP_SITE_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
template<class Impl>
void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
HAND_DECLARATIONS(ignore);
StencilEntry *SE;
int offset,local,perm, ptype;
int nmu=0;
#define HAND_DOP_SITE_DAG_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \
ZERO_RESULT; \
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \
HAND_RESULT_EXT(ss,F)
HAND_DOP_SITE_DAG_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN);
}
#define HAND_SPECIALISE_GPARITY(IMPL) \
template<> void \
WilsonKernels<IMPL>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
StencilEntry *SE; \
HAND_DOP_SITE(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> \
void WilsonKernels<IMPL>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
StencilEntry *SE; \
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
HAND_DOP_SITE_DAG(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE_DAG(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> void \
WilsonKernels<IMPL>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
StencilEntry *SE; \
HAND_DOP_SITE_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> \
void WilsonKernels<IMPL>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
StencilEntry *SE; \
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
HAND_DOP_SITE_DAG_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
HAND_DOP_SITE_DAG_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
\
template<> void \
WilsonKernels<IMPL>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
StencilEntry *SE; \
int nmu=0; \
HAND_DOP_SITE_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
nmu = 0; \
HAND_DOP_SITE_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
} \
template<> \
void WilsonKernels<IMPL>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out) \
{ \
typedef IMPL Impl; \
typedef typename Simd::scalar_type S; \
typedef typename Simd::vector_type V; \
\
HAND_DECLARATIONS(ignore); \
\
StencilEntry *SE; \
int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \
int nmu=0; \
HAND_DOP_SITE_DAG_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
nmu = 0; \
HAND_DOP_SITE_DAG_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \
}
HAND_SPECIALISE_GPARITY(GparityWilsonImplF);
HAND_SPECIALISE_GPARITY(GparityWilsonImplD);
HAND_SPECIALISE_GPARITY(GparityWilsonImplFH);
HAND_SPECIALISE_GPARITY(GparityWilsonImplDF);
////////////// Wilson ; uses this implementation /////////////////////
#define INSTANTIATE_THEM(A) \
template void WilsonKernels<A>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
int ss,int sU,const FermionField &in, FermionField &out); \
template void WilsonKernels<A>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out);\
template void WilsonKernels<A>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
int ss,int sU,const FermionField &in, FermionField &out); \
template void WilsonKernels<A>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out); \
template void WilsonKernels<A>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
int ss,int sU,const FermionField &in, FermionField &out); \
template void WilsonKernels<A>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out);
INSTANTIATE_THEM(GparityWilsonImplF);
INSTANTIATE_THEM(GparityWilsonImplD);
INSTANTIATE_THEM(GparityWilsonImplFH);
INSTANTIATE_THEM(GparityWilsonImplDF);
}}

11
lib/stencil/Stencil.h
View File

@@ -66,6 +66,8 @@ void Gather_plane_simple_table (std::vector<std::pair<int,int> >& table,const La
parallel_for(int i=0;i<num;i++){ parallel_for(int i=0;i<num;i++){
compress.Compress(&buffer[off],table[i].first,rhs._odata[so+table[i].second]); compress.Compress(&buffer[off],table[i].first,rhs._odata[so+table[i].second]);
} }
// Further optimisatoin: i) streaming store the result
// ii) software prefetch the first element of the next table entry
} }
/////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////
@@ -505,25 +507,24 @@ class CartesianStencil { // Stencil runs along coordinate axes only; NO diagonal
template<class decompressor> template<class decompressor>
void CommsMerge(decompressor decompress,std::vector<Merge> &mm,std::vector<Decompress> &dd) { void CommsMerge(decompressor decompress,std::vector<Merge> &mm,std::vector<Decompress> &dd) {
mergetime-=usecond();
for(int i=0;i<mm.size();i++){ for(int i=0;i<mm.size();i++){
mergetime-=usecond();
parallel_for(int o=0;o<mm[i].buffer_size/2;o++){ parallel_for(int o=0;o<mm[i].buffer_size/2;o++){
decompress.Exchange(mm[i].mpointer, decompress.Exchange(mm[i].mpointer,
mm[i].vpointers[0], mm[i].vpointers[0],
mm[i].vpointers[1], mm[i].vpointers[1],
mm[i].type,o); mm[i].type,o);
} }
mergetime+=usecond();
} }
mergetime+=usecond();
decompresstime-=usecond();
for(int i=0;i<dd.size();i++){ for(int i=0;i<dd.size();i++){
decompresstime-=usecond();
parallel_for(int o=0;o<dd[i].buffer_size;o++){ parallel_for(int o=0;o<dd[i].buffer_size;o++){
decompress.Decompress(dd[i].kernel_p,dd[i].mpi_p,o); decompress.Decompress(dd[i].kernel_p,dd[i].mpi_p,o);
} }
decompresstime+=usecond();
} }
decompresstime+=usecond();
} }
//////////////////////////////////////// ////////////////////////////////////////
// Set up routines // Set up routines

2
lib/threads/Threads.h
View File

@@ -40,7 +40,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#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)") #define PARALLEL_FOR_LOOP_INTERN _Pragma("omp for schedule(static)")
#define PARALLEL_NESTED_LOOP2 _Pragma("omp parallel for schedule(static) collapse(2)") #define PARALLEL_NESTED_LOOP2 _Pragma("omp parallel for collapse(2)")
#define PARALLEL_REGION _Pragma("omp parallel") #define PARALLEL_REGION _Pragma("omp parallel")
#define PARALLEL_CRITICAL _Pragma("omp critical") #define PARALLEL_CRITICAL _Pragma("omp critical")
#else #else