// i represents a "half index" into an even or odd "half lattice".
// when oddBit={0,1} the half lattice is {even,odd}.
//
// the displacements, such as dx, refer to the full lattice coordinates.
//
// neighborIndex() takes a "half index", displaces it, and returns the
// new "half index", which can be an index into either the even or odd lattices.
// displacements of magnitude one always interchange odd and even lattices.
//
int neighborIndex(int i, int oddBit, int dx4, int dx3, int dx2, int dx1) {
int X = fullLatticeIndex(i, oddBit);
int x4 = X/(L3*L2*L1);
int x3 = (X/(L2*L1)) % L3;
int x2 = (X/L1) % L2;
int x1 = X % L1;
// assert (oddBit == (x+y+z+t)%2);
x4 = (x4+dx4+L4) % L4;
x3 = (x3+dx3+L3) % L3;
x2 = (x2+dx2+L2) % L2;
x1 = (x1+dx1+L1) % L1;
return (x4*(L3*L2*L1) + x3*(L2*L1) + x2*(L1) + x1) / 2;
}
int link_sanity_check_internal_12(Float* link, int dir, int ga_idx, QudaGaugeParam* gaugeParam, int oddBit)
{
//printf("link sanity check is called\n");
int ret =0;
Float refc_buf[6];
Float* refc = &refc_buf[0];
memset((void*)refc, 0, sizeof(refc_buf));
Float* a = link;
Float* b = link + 6;
Float* c = link + 12;
accumulateConjugateProduct(refc + 0*2, a + 1*2, b + 2*2, +1);
accumulateConjugateProduct(refc + 0*2, a + 2*2, b + 1*2, -1);
accumulateConjugateProduct(refc + 1*2, a + 2*2, b + 0*2, +1);
accumulateConjugateProduct(refc + 1*2, a + 0*2, b + 2*2, -1);
accumulateConjugateProduct(refc + 2*2, a + 0*2, b + 1*2, +1);
accumulateConjugateProduct(refc + 2*2, a + 1*2, b + 0*2, -1);
int X1h=gaugeParam->X[0]/2;
int X1 =gaugeParam->X[0];
int X2 =gaugeParam->X[1];
int X3 =gaugeParam->X[2];
int X4 =gaugeParam->X[3];
double t_boundary = (gaugeParam->t_boundary ==QUDA_ANTI_PERIODIC_T)? -1.0:1.0;
double u0 = gaugeParam->tadpole_coeff;
double coff= -u0*u0*24;
//coff = (dir < 6) ? coff : ( (ga_idx >= (X4-3)*X1h*X2*X3 )? t_boundary : 1);
//float u0 = (dir < 6) ? gaugeParam->anisotropy : ( (ga_idx >= (X4-3)*X1h*X2*X3 )? t_boundary : 1);
#if 1
{
int index = fullLatticeIndex(ga_idx, oddBit);
int i4 = index /(X3*X2*X1);
int i3 = (index - i4*(X3*X2*X1))/(X2*X1);
int i2 = (index - i4*(X3*X2*X1) - i3*(X2*X1))/X1;
int i1 = index - i4*(X3*X2*X1) - i3*(X2*X1) - i2*X1;
if (dir == 0) {
if (i4 % 2 == 1){
coff *= -1;
}
}
if (dir == 2){
if ((i1+i4) % 2 == 1){
coff *= -1;
}
}
if (dir == 4){
if ( (i4+i1+i2) % 2 == 1){
coff *= -1;
}
}
if (dir == 6){
if (ga_idx >= (X4-3)*X1h*X2*X3 ){
coff *= -1;
}
}
//printf("local ga_idx =%d, index=%d, i4,3,2,1 =%d %d %d %d\n", ga_idx, index, i4, i3, i2,i1);
}
#endif
refc[0]*=coff; refc[1]*=coff; refc[2]*=coff; refc[3]*=coff; refc[4]*=coff; refc[5]*=coff;
double delta = 0.0001;
int i;
for (i =0;i < 6; i++){
double diff = refc[i] - c[i];
double absdiff = diff > 0? diff: (-diff);
if (absdiff > delta){
printf("ERROR: sanity check failed for link\n");
display_link_internal(link);
printf("refc = (%.10f,%.10f) (%.10f,%.10f) (%.10f,%.10f)\n",
refc[0], refc[1], refc[2], refc[3], refc[4], refc[5]);
printf("dir=%d, ga_idx=%d, coff=%f, t_boundary=%f\n",dir, ga_idx,coff, t_boundary);
printf("X=%d %d %d %d, X1h=%d\n", gaugeParam->X[0], X2, X3, X4, X1h);
return -1;
}
}
return ret;
}
int link_sanity_check_internal_8(Float* link, int dir, int ga_idx, QudaGaugeParam* gaugeParam, int oddBit)
{
int ret =0;
Float ref_link_buf[18];
Float* ref = & ref_link_buf[0];
memset(ref, 0, sizeof(ref_link_buf));
ref[0] = atan2(link[1], link[0]);
ref[1] = atan2(link[13], link[12]);
for (int i=2; i<7; i++) {
ref[i] = link[i];
}
int X1h=gaugeParam->X[0]/2;
int X2 =gaugeParam->X[1];
int X3 =gaugeParam->X[2];
int X4 =gaugeParam->X[3];
double t_boundary = (gaugeParam->t_boundary ==QUDA_ANTI_PERIODIC_T)? -1.0:1.0;
// First reconstruct first row
Float row_sum = 0.0;
row_sum += ref[2]*ref[2];
row_sum += ref[3]*ref[3];
row_sum += ref[4]*ref[4];
row_sum += ref[5]*ref[5];
#define SMALL_NUM 1e-24
row_sum = (row_sum != 0)?row_sum: SMALL_NUM;
#if 1
Float u0= -gaugeParam->tadpole_coeff*gaugeParam->tadpole_coeff*24;
{
int X1h=gaugeParam->X[0]/2;
int X1 =gaugeParam->X[0];
int X2 =gaugeParam->X[1];
int X3 =gaugeParam->X[2];
int X4 =gaugeParam->X[3];
int index = fullLatticeIndex(ga_idx, oddBit);
int i4 = index /(X3*X2*X1);
int i3 = (index - i4*(X3*X2*X1))/(X2*X1);
int i2 = (index - i4*(X3*X2*X1) - i3*(X2*X1))/X1;
int i1 = index - i4*(X3*X2*X1) - i3*(X2*X1) - i2*X1;
if (dir == 0) {
if (i4 % 2 == 1){
u0 *= -1;
}
}
if (dir == 1){
if ((i1+i4) % 2 == 1){
u0 *= -1;
}
}
if (dir == 2){
if ( (i4+i1+i2) % 2 == 1){
u0 *= -1;
}
}
if (dir == 3){
if (ga_idx >= (X4-3)*X1h*X2*X3 ){
u0 *= -1;
}
}
//printf("local ga_idx =%d, index=%d, i4,3,2,1 =%d %d %d %d\n", ga_idx, index, i4, i3, i2,i1);
}
#endif
Float U00_mag = sqrt( (1.f/(u0*u0) - row_sum)>0? (1.f/(u0*u0)-row_sum):0);
ref[14] = ref[0];
ref[15] = ref[1];
ref[0] = U00_mag * cos(ref[14]);
ref[1] = U00_mag * sin(ref[14]);
Float column_sum = 0.0;
for (int i=0; i<2; i++) column_sum += ref[i]*ref[i];
for (int i=6; i<8; i++) column_sum += ref[i]*ref[i];
Float U20_mag = sqrt( (1.f/(u0*u0) - column_sum) > 0? (1.f/(u0*u0)-column_sum) : 0);
ref[12] = U20_mag * cos(ref[15]);
ref[13] = U20_mag * sin(ref[15]);
// First column now restored
// finally reconstruct last elements from SU(2) rotation
Float r_inv2 = 1.0/(u0*row_sum);
// U11
Float A[2];
complexDotProduct(A, ref+0, ref+6);
complexConjugateProduct(ref+8, ref+12, ref+4);
accumulateComplexProduct(ref+8, A, ref+2, u0);
ref[8] *= -r_inv2;
ref[9] *= -r_inv2;
// U12
complexConjugateProduct(ref+10, ref+12, ref+2);
accumulateComplexProduct(ref+10, A, ref+4, -u0);
ref[10] *= r_inv2;
ref[11] *= r_inv2;
// U21
complexDotProduct(A, ref+0, ref+12);
complexConjugateProduct(ref+14, ref+6, ref+4);
accumulateComplexProduct(ref+14, A, ref+2, -u0);
ref[14] *= r_inv2;
ref[15] *= r_inv2;
// U12
complexConjugateProduct(ref+16, ref+6, ref+2);
accumulateComplexProduct(ref+16, A, ref+4, u0);
ref[16] *= -r_inv2;
ref[17] *= -r_inv2;
double delta = 0.0001;
int i;
for (i =0;i < 18; i++){
double diff = ref[i] - link[i];
double absdiff = diff > 0? diff: (-diff);
if ( (ref[i] != ref[i]) || (absdiff > delta)){
printf("ERROR: sanity check failed for link\n");
display_link_internal(link);
printf("reconstructed link is\n");
display_link_internal(ref);
printf("dir=%d, ga_idx=%d, u0=%f, t_boundary=%f\n",dir, ga_idx, u0, t_boundary);
printf("X=%d %d %d %d, X1h=%d\n", gaugeParam->X[0], X2, X3, X4, X1h);
return -1;
}
}
return ret;
}
int site_link_sanity_check_internal_12(Float* link, int dir, int ga_idx, QudaGaugeParam* gaugeParam, int oddBit)
{
int ret =0;
Float refc_buf[6];
Float* refc = &refc_buf[0];
memset((void*)refc, 0, sizeof(refc_buf));
Float* a = link;
Float* b = link + 6;
Float* c = link + 12;
accumulateConjugateProduct(refc + 0*2, a + 1*2, b + 2*2, +1);
accumulateConjugateProduct(refc + 0*2, a + 2*2, b + 1*2, -1);
accumulateConjugateProduct(refc + 1*2, a + 2*2, b + 0*2, +1);
accumulateConjugateProduct(refc + 1*2, a + 0*2, b + 2*2, -1);
accumulateConjugateProduct(refc + 2*2, a + 0*2, b + 1*2, +1);
accumulateConjugateProduct(refc + 2*2, a + 1*2, b + 0*2, -1);
int X1h=gaugeParam->X[0]/2;
int X1 =gaugeParam->X[0];
int X2 =gaugeParam->X[1];
int X3 =gaugeParam->X[2];
int X4 =gaugeParam->X[3];
#if 1
double coeff= 1.0;
{
int index = fullLatticeIndex(ga_idx, oddBit);
int i4 = index /(X3*X2*X1);
int i3 = (index - i4*(X3*X2*X1))/(X2*X1);
int i2 = (index - i4*(X3*X2*X1) - i3*(X2*X1))/X1;
int i1 = index - i4*(X3*X2*X1) - i3*(X2*X1) - i2*X1;
if (dir == XUP) {
if (i4 % 2 == 1){
coeff *= -1;
}
}
if (dir == YUP){
if ((i1+i4) % 2 == 1){
coeff *= -1;
}
}
if (dir == ZUP){
if ( (i4+i1+i2) % 2 == 1){
coeff *= -1;
}
}
if (dir == TUP){
if ((commCoords(3) == commDim(3) -1) && i4 == (X4-1) ){
coeff *= -1;
}
}
}
refc[0]*=coeff; refc[1]*=coeff; refc[2]*=coeff; refc[3]*=coeff; refc[4]*=coeff; refc[5]*=coeff;
#endif
double delta = 0.0001;
int i;
for (i =0;i < 6; i++){
double diff = refc[i] - c[i];
double absdiff = diff > 0? diff: (-diff);
if (absdiff > delta){
printf("ERROR: sanity check failed for site link\n");
display_link_internal(link);
printf("refc = (%.10f,%.10f) (%.10f,%.10f) (%.10f,%.10f)\n",
refc[0], refc[1], refc[2], refc[3], refc[4], refc[5]);
printf("X=%d %d %d %d, X1h=%d\n", gaugeParam->X[0], X2, X3, X4, X1h);
return -1;
}
}
return ret;
}
