void
ImprovedFastGaussTransform::Evaluate()
{
compute_C();
for(int j=0; j < M; j++)
{
pG[j]=0.0;
int target_base=j*d;
for(int k=0; k<K; k++){
int center_base=k*d;
double target_center_distance_square=0.0;
for(int i=0; i<d; i++){
dy[i]=py[target_base+i]-pcc[center_base+i];
target_center_distance_square += dy[i]*dy[i];
if (target_center_distance_square > ry_square[k]) break;
}
if (target_center_distance_square <= ry_square[k]){
compute_target_center_monomials();
double g=exp(-target_center_distance_square/h_square);
for(int alpha=0; alpha<p_max_total; alpha++){
pG[j]+=(C[k*p_max_total+alpha]*g*target_center_monomials[alpha]);
}
}
}
}
}
Subarray::Subarray(const DynamicParameter & dp_, bool is_fa_):
dp(dp_), num_rows(dp.num_r_subarray), num_cols(dp.num_c_subarray),
num_cols_fa_cam(dp.tag_num_c_subarray), num_cols_fa_ram(dp.data_num_c_subarray),
cell(dp.cell), cam_cell(dp.cam_cell), is_fa(is_fa_)
{
//num_cols=7;
//cout<<"num_cols ="<< num_cols <<endl;
if (!(is_fa || dp.pure_cam))
{
num_cols +=(g_ip->add_ecc_b_ ? (int)ceil(num_cols / num_bits_per_ecc_b_) : 0); // ECC overhead
uint32_t ram_num_cells_wl_stitching =
(dp.ram_cell_tech_type == lp_dram) ? dram_num_cells_wl_stitching_ :
(dp.ram_cell_tech_type == comm_dram) ? comm_dram_num_cells_wl_stitching_ : sram_num_cells_wl_stitching_;
area.h = cell.h * num_rows;
area.w = cell.w * num_cols +
ceil(num_cols / ram_num_cells_wl_stitching) * g_tp.ram_wl_stitching_overhead_; // stitching overhead
}
else //cam fa
{
//should not add dummy row here since the dummy row do not need decoder
if (is_fa)// fully associative cache
{
num_cols_fa_cam += g_ip->add_ecc_b_ ? (int)ceil(num_cols_fa_cam / num_bits_per_ecc_b_) : 0;
num_cols_fa_ram += (g_ip->add_ecc_b_ ? (int)ceil(num_cols_fa_ram / num_bits_per_ecc_b_) : 0);
num_cols = num_cols_fa_cam + num_cols_fa_ram;
}
else
{
num_cols_fa_cam += g_ip->add_ecc_b_ ? (int)ceil(num_cols_fa_cam / num_bits_per_ecc_b_) : 0;
num_cols_fa_ram = 0;
num_cols = num_cols_fa_cam;
}
area.h = cam_cell.h * (num_rows + 1);//height of subarray is decided by CAM array. blank space in sram array are filled with dummy cells
area.w = cam_cell.w * num_cols_fa_cam + cell.w * num_cols_fa_ram
+ ceil((num_cols_fa_cam + num_cols_fa_ram) / sram_num_cells_wl_stitching_)*g_tp.ram_wl_stitching_overhead_
+ 16*g_tp.wire_local.pitch //the overhead for the NAND gate to connect the two halves
+ 128*g_tp.wire_local.pitch;//the overhead for the drivers from matchline to wordline of RAM
}
assert(area.h>0);
assert(area.w>0);
compute_C();
}
//Compute the Potential
double SCFPotentialEval(double R,double Z, double phi,
double t,
struct potentialArg * potentialArgs)
{
double * args= potentialArgs->args;
//Get args
double a = *args++;
int isNonAxi = (int)*args++;
int N = *args++;
int L = *args++;
int M = *args++;
double* Acos = args;
double* Asin;
if (isNonAxi==1) //LCOV_EXCL_START
{
Asin = args + N*L*M;
} //LCOV_EXCL_STOP
//convert R,Z to r, theta
double r;
double theta;
cyl_to_spher(R, Z,&r, &theta);
double xi;
calculateXi(r, a, &xi);
//Compute the gegenbauer polynomials and its derivative.
double C[N*L];
compute_C(xi, N, L, &C[0]);
//Compute phiTilde and its derivative
double phiTilde[L*N];
compute_phiTilde(r, a, N, L, &C[0], &phiTilde[0]);
//Compute Associated Legendre Polynomials
int M_eff = M;
int size = 0;
if (isNonAxi==0)
{
M_eff = 1;
size = L;
} else{ //LCOV_EXCL_START
size = L*L - L*(L-1)/2;
} //LCOV_EXCL_STOP
double P[size];
compute_P(cos(theta), L,M_eff, &P[0]);
double potential;
double (*PhiTilde_Pointer[1]) = {&phiTilde[0]};
double (*P_Pointer[1]) = {&P[0]};
double Constant[1] = {1.};
if (isNonAxi==1) //LCOV_EXCL_START
{
double (*Eq[1])(double, double, double, double, double, double, int) = {&computePhi};
equations e = {Eq,&PhiTilde_Pointer[0],&P_Pointer[0],&Constant[0]};
computeNonAxi(a, N, L, M,r, theta, phi, Acos, Asin, 1, e, &potential);
} //LCOV_EXCL_STOP
else
{
double (*Eq[1])(double, double, double) = {&computeAxiPhi};
axi_equations e = {Eq,&PhiTilde_Pointer[0],&P_Pointer[0],&Constant[0]};
compute(a, N, L, M,r, theta, phi, Acos, 1, e, &potential);
}
return potential;
}
//Compute the Derivatives
void computeDeriv(double R,double Z, double phi,
double t,
struct potentialArg * potentialArgs, double * F)
{
double * args= potentialArgs->args;
//Get args
double a = *args++;
int isNonAxi = (int)*args++;
int N = *args++;
int L = *args++;
int M = *args++;
double* Acos = args;
double * caching_i = (args + (isNonAxi + 1)*N*L*M);
double *Asin;
if (isNonAxi == 1)
{
Asin = args + N*L*M;
}
double *cached_type = caching_i;
double * cached_coords = (caching_i+ 1);
double * cached_values = (caching_i + 4);
if ((int)*cached_type==DERIV)
{
if (*cached_coords == R && *(cached_coords + 1) == Z && *(cached_coords + 2) == phi)
{
*F = *cached_values;
*(F + 1) = *(cached_values + 1);
*(F + 2) = *(cached_values + 2);
return;
}
}
double r;
double theta;
cyl_to_spher(R, Z, &r, &theta);
double xi;
calculateXi(r, a, &xi);
//Compute the gegenbauer polynomials and its derivative.
double C[N*L];
double dC[N*L];
double d2C[N*L];
compute_C(xi, N, L, &C[0]);
compute_dC(xi, N, L, &dC[0]);
compute_d2C(xi, N, L, &d2C[0]);
//Compute phiTilde and its derivative
double phiTilde[L*N];
compute_phiTilde(r, a, N, L, &C[0], &phiTilde[0]);
double dphiTilde[L*N];
compute_dphiTilde(r, a, N, L, &C[0], &dC[0], &dphiTilde[0]);
double d2phiTilde[L*N];
compute_d2phiTilde(r, a, N, L, &C[0], &dC[0], &d2C[0], &d2phiTilde[0]);
//Compute Associated Legendre Polynomials
int M_eff = M;
int size = 0;
if (isNonAxi==0)
{
M_eff = 1;
size = L;
} else{
size = L*L - L*(L-1)/2;
}
double P[size];
compute_P(cos(theta), L,M_eff, &P[0]);
double (*PhiTilde_Pointer[3])= {&d2phiTilde[0],&phiTilde[0],&dphiTilde[0]};
double (*P_Pointer[3]) = {&P[0], &P[0], &P[0]};
double Constant[3] = {1., 1., 1.};
if (isNonAxi==1)
{
double (*Eq[3])(double, double, double, double, double, double, int) = {&computeF_rr, &computeF_phiphi, &computeF_rphi};
equations e = {Eq,&PhiTilde_Pointer[0],&P_Pointer[0],&Constant[0]};
computeNonAxi(a, N, L, M,r, theta, phi, Acos, Asin, 3, e, F);
}
else
{
double (*Eq[3])(double, double, double) = {&computeAxiF_rr, &computeAxiF_phiphi, &computeAxiF_rphi};
axi_equations e = {Eq,&PhiTilde_Pointer[0],&P_Pointer[0],&Constant[0]};
compute(a, N, L, M,r, theta, phi, Acos, 3, e, F);
}
//Caching
*cached_type = (double)DERIV;
* cached_coords = R;
* (cached_coords + 1) = Z;
* (cached_coords + 2) = phi;
* (cached_values) = *F;
* (cached_values + 1) = *(F + 1);
* (cached_values + 2) = *(F + 2);
}
static void compute_csas(ConsensusSA *csa)
{
unsigned long i, sa_i, sa_i_size = 0, sa_prime, sa_prime_size;
GtArray *splice_form;
GtBittab **C, **left, **right, **L, **R, *U_i, *SA_i, *SA_prime;
#ifndef NDEBUG
unsigned long u_i_size, u_i_minus_1_size;
gt_assert(csa && csa->set_of_sas);
#endif
/* init sets */
C = gt_malloc(sizeof (GtBittab*) * csa->number_of_sas);
left = gt_malloc(sizeof (GtBittab*) * csa->number_of_sas);
right = gt_malloc(sizeof (GtBittab*) * csa->number_of_sas);
L = gt_malloc(sizeof (GtBittab*) * csa->number_of_sas);
R = gt_malloc(sizeof (GtBittab*) * csa->number_of_sas);
for (i = 0; i < csa->number_of_sas; i++) {
C[i] = gt_bittab_new(csa->number_of_sas);
left[i] = gt_bittab_new(csa->number_of_sas);
right[i] = gt_bittab_new(csa->number_of_sas);
L[i] = gt_bittab_new(csa->number_of_sas);
R[i] = gt_bittab_new(csa->number_of_sas);
}
U_i = gt_bittab_new(csa->number_of_sas);
SA_i = gt_bittab_new(csa->number_of_sas);
SA_prime = gt_bittab_new(csa->number_of_sas);
splice_form = gt_array_new(sizeof (unsigned long));
/* compute sets */
compute_C(C, csa);
compute_left(left, csa);
compute_right(right, csa);
compute_L(L, C, left, csa->number_of_sas);
compute_R(R, C, right, csa->number_of_sas);
/* U_0 = SA */
for (i = 0; i < csa->number_of_sas; i++)
gt_bittab_set_bit(U_i, i);
#ifndef NDEBUG
/* preparation for assertion below */
u_i_minus_1_size = gt_bittab_count_set_bits(U_i);
#endif
while (gt_bittab_is_true(U_i)) {
sa_i = GT_UNDEF_ULONG;
for (sa_prime = gt_bittab_get_first_bitnum(U_i);
sa_prime != gt_bittab_get_last_bitnum(U_i);
sa_prime = gt_bittab_get_next_bitnum(U_i, sa_prime)) {
if (sa_i == GT_UNDEF_ULONG) {
sa_i = sa_prime;
gt_bittab_or(SA_i, L[sa_i], R[sa_i]);
sa_i_size = gt_bittab_count_set_bits(SA_i);
}
else {
gt_bittab_or(SA_prime, L[sa_prime], R[sa_prime]);
sa_prime_size = gt_bittab_count_set_bits(SA_prime);
if (sa_prime_size > sa_i_size) {
sa_i = sa_prime;
sa_i_size = sa_prime_size;
gt_bittab_equal(SA_i, SA_prime);
}
}
}
/* make sure the computed splice form is maximal w.r.t. to compatibility */
gt_assert(splice_form_is_valid(SA_i, csa));
/* process splice form */
if (csa->process_splice_form) {
gt_array_reset(splice_form);
gt_bittab_get_all_bitnums(SA_i, splice_form);
csa->process_splice_form(splice_form, csa->set_of_sas, csa->number_of_sas,
csa->size_of_sa, csa->userdata);
}
/* U_i = U_i-1 \ SA_i */
gt_bittab_nand(U_i, U_i, SA_i);
#ifndef NDEBUG
/* ensure that |U_i| < |U_i-1| */
u_i_size = gt_bittab_count_set_bits(U_i);
gt_assert(u_i_size < u_i_minus_1_size);
u_i_minus_1_size = u_i_size;
#endif
}
/* free sets */
for (i = 0; i < csa->number_of_sas; i++) {
gt_bittab_delete(C[i]);
gt_bittab_delete(left[i]);
gt_bittab_delete(right[i]);
gt_bittab_delete(L[i]);
gt_bittab_delete(R[i]);
}
gt_free(C);
gt_free(left);
gt_free(right);
gt_free(L);
gt_free(R);
gt_bittab_delete(U_i);
gt_bittab_delete(SA_i);
gt_bittab_delete(SA_prime);
gt_array_delete(splice_form);
}
