def GiRaFFE_NRPy_Afield_flux(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
gammaDD = ixp.declarerank2("gammaDD", "sym01", DIM=3)
betaU = ixp.declarerank1("betaU", DIM=3)
alpha = sp.sympify("alpha")
for flux_dirn in range(3):
chsp.find_cmax_cmin(flux_dirn, gammaDD, betaU, alpha)
Ccode_kernel = outputC([chsp.cmax, chsp.cmin], ["cmax", "cmin"],
"returnstring",
params="outCverbose=False,CSE_sorting=none")
Ccode_kernel = Ccode_kernel.replace("cmax",
"*cmax").replace("cmin", "*cmin")
Ccode_kernel = Ccode_kernel.replace("betaU0", "betaUi").replace(
"betaU1", "betaUi").replace("betaU2", "betaUi")
with open(
os.path.join(Ccodesdir,
"compute_cmax_cmin_dirn" + str(flux_dirn) + ".h"),
"w") as file:
file.write(Ccode_kernel)
with open(os.path.join(Ccodesdir, name + ".h"), "w") as file:
file.write(body)
def GiRaFFE_NRPy_Source_Terms(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(
os.path.join(Ccodesdir,
"Lorenz_psi6phi_rhs__add_gauge_terms_to_A_i_rhs.h"),
"w") as file:
file.write(body)
def enforce_detgammabar_eq_detgammahat__generate_symbolic_expressions_and_C_code():
######################################
# START: GENERATE SYMBOLIC EXPRESSIONS
# Store original finite-differencing order:
FD_order_orig = par.parval_from_str("finite_difference::FD_CENTDERIVS_ORDER")
# Set new finite-differencing order:
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER", FD_order)
# Enable rfm_precompute infrastructure, which results in
# BSSN RHSs that are free of transcendental functions,
# even in curvilinear coordinates, so long as
# ConformalFactor is set to "W" (default).
cmd.mkdir(os.path.join(outdir,"rfm_files/"))
par.set_parval_from_str("reference_metric::enable_rfm_precompute","True")
par.set_parval_from_str("reference_metric::rfm_precompute_Ccode_outdir",os.path.join(outdir,"rfm_files/"))
import BSSN.Enforce_Detgammabar_Constraint as EGC
enforce_detg_constraint_symb_expressions = EGC.Enforce_Detgammabar_Constraint_symb_expressions()
# Now that we are finished with all the rfm hatted
# quantities in generic precomputed functional
# form, let's restore them to their closed-
# form expressions.
par.set_parval_from_str("reference_metric::enable_rfm_precompute","False") # Reset to False to disable rfm_precompute.
rfm.ref_metric__hatted_quantities()
# Restore original finite-differencing order:
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER", FD_order)
# END: GENERATE SYMBOLIC EXPRESSIONS
######################################
start = time.time()
print("Generating optimized C code (FD_order="+str(FD_order)+") for gamma constraint. May take a while, depending on CoordSystem.")
enforce_gammadet_string = fin.FD_outputC("returnstring", enforce_detg_constraint_symb_expressions,
params="outCverbose=False,preindent=0,includebraces=False")
with open(os.path.join(outdir,"enforcedetgammabar_constraint_FD_order_"+str(FD_order)+".h"), "w") as file:
file.write(lp.loop(["i2","i1","i0"],["0", "0", "0"],
["cctk_lsh[2]","cctk_lsh[1]","cctk_lsh[0]"],
["1","1","1"],
["#pragma omp parallel for",
"#include \"rfm_files/rfm_struct__read2.h\"",
"#include \"rfm_files/rfm_struct__read1.h\""],"",
"#include \"rfm_files/rfm_struct__read0.h\"\n"+enforce_gammadet_string))
end = time.time()
print("Finished gamma constraint C codegen (FD_order="+str(FD_order)+") in " + str(end - start) + " seconds.")
def GiRaFFE_NRPy_FCVAL(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(
os.path.join(Ccodesdir, "interpolate_metric_gfs_to_cell_faces.h"),
"w") as file:
file.write(prefunc)
desc = "Interpolate metric gridfunctions to cell faces"
name = "interpolate_metric_gfs_to_cell_faces"
interp_Cfunc = outCfunction(
outfile="returnstring",
desc=desc,
name=name,
params=
"const paramstruct *params,REAL *auxevol_gfs,const int flux_dirn",
preloop=""" int in_gf,out_gf;
REAL Qm2,Qm1,Qp0,Qp1;
""",
body=""" for(int gf = 0;gf < num_metric_gfs;gf++) {
in_gf = metric_gfs_list[gf];
out_gf = metric_gfs_face_list[gf];
for (int i2 = 2;i2 < Nxx_plus_2NGHOSTS2-1;i2++) {
for (int i1 = 2;i1 < Nxx_plus_2NGHOSTS1-1;i1++) {
for (int i0 = 2;i0 < Nxx_plus_2NGHOSTS0-1;i0++) {
Qm2 = auxevol_gfs[IDX4S(in_gf,i0-2*kronecker_delta[flux_dirn][0],i1-2*kronecker_delta[flux_dirn][1],i2-2*kronecker_delta[flux_dirn][2])];
Qm1 = auxevol_gfs[IDX4S(in_gf,i0-kronecker_delta[flux_dirn][0],i1-kronecker_delta[flux_dirn][1],i2-kronecker_delta[flux_dirn][2])];
Qp0 = auxevol_gfs[IDX4S(in_gf,i0,i1,i2)];
Qp1 = auxevol_gfs[IDX4S(in_gf,i0+kronecker_delta[flux_dirn][0],i1+kronecker_delta[flux_dirn][1],i2+kronecker_delta[flux_dirn][2])];
auxevol_gfs[IDX4S(out_gf,i0,i1,i2)] = COMPUTE_FCVAL(Qm2,Qm1,Qp0,Qp1);
}
}
}
}
""",
rel_path_to_Cparams=os.path.join("../"))
with open(
os.path.join(Ccodesdir, "interpolate_metric_gfs_to_cell_faces.h"),
"a") as file:
file.write(interp_Cfunc)
def BaikalETK_C_kernels_codegen_onepart(NRPyDir=os.path.join(".."),
params="WhichPart=BSSN_RHSs,ThornName=Baikal,FD_order=4,enable_stress_energy_source_terms=True"):
# Set default parameters
WhichPart = "BSSN_RHSs"
ThornName = "Baikal"
FD_order = 4
enable_stress_energy_source_terms = True
LapseCondition = "OnePlusLog" # Set the standard 1+log lapse condition
# Set the standard, second-order advecting-shift, Gamma-driving shift condition:
ShiftCondition = "GammaDriving2ndOrder_NoCovariant"
# Default Kreiss-Oliger dissipation strength
default_KO_strength = 0.1
import re
if params != "":
params2 = re.sub("^,","",params)
params = params2.strip()
splitstring = re.split("=|,", params)
if len(splitstring) % 2 != 0:
print("outputC: Invalid params string: "+params)
sys.exit(1)
parnm = []
value = []
for i in range(int(len(splitstring)/2)):
parnm.append(splitstring[2*i])
value.append(splitstring[2*i+1])
for i in range(len(parnm)):
parnm.append(splitstring[2*i])
value.append(splitstring[2*i+1])
for i in range(len(parnm)):
if parnm[i] == "WhichPart":
WhichPart = value[i]
elif parnm[i] == "WhichParamSet":
WhichParamSet = int(value[i])
elif parnm[i] == "ThornName":
ThornName = value[i]
elif parnm[i] == "FD_order":
FD_order = int(value[i])
elif parnm[i] == "enable_stress_energy_source_terms":
enable_stress_energy_source_terms = False
if value[i] == "True":
enable_stress_energy_source_terms = True
elif parnm[i] == "LapseCondition":
LapseCondition = value[i]
elif parnm[i] == "ShiftCondition":
ShiftCondition = value[i]
elif parnm[i] == "default_KO_strength":
default_KO_strength = float(value[i])
else:
print("BaikalETK Error: Could not parse input param: "+parnm[i])
sys.exit(1)
# Create directory for BaikalETK thorn & subdirectories in case they don't exist.
outrootdir = ThornName
cmd.mkdir(os.path.join(outrootdir))
outdir = os.path.join(outrootdir,"src") # Main C code output directory
# Copy SIMD/SIMD_intrinsics.h to $outdir/SIMD/SIMD_intrinsics.h
cmd.mkdir(os.path.join(outdir,"SIMD"))
shutil.copy(os.path.join(NRPyDir,"SIMD/")+"SIMD_intrinsics.h",os.path.join(outdir,"SIMD/"))
# Set spatial dimension (must be 3 for BSSN)
par.set_parval_from_str("grid::DIM",3)
# Step 2: Set some core parameters, including CoordSystem MoL timestepping algorithm,
# FD order, floating point precision, and CFL factor:
# Choices are: Spherical, SinhSpherical, SinhSphericalv2, Cylindrical, SinhCylindrical,
# SymTP, SinhSymTP
# NOTE: Only CoordSystem == Cartesian makes sense here; new
# boundary conditions are needed within the ETK for
# Spherical, etc. coordinates.
CoordSystem = "Cartesian"
par.set_parval_from_str("reference_metric::CoordSystem",CoordSystem)
rfm.reference_metric() # Create ReU, ReDD needed for rescaling B-L initial data, generating BSSN RHSs, etc.
REAL = "CCTK_REAL" # Set REAL to CCTK_REAL, the ETK data type for
# floating point precision (typically `double`)
# Set the gridfunction memory access type to ETK-like, so that finite_difference
# knows how to read and write gridfunctions from/to memory.
par.set_parval_from_str("grid::GridFuncMemAccess","ETK")
par.set_parval_from_str("BSSN.BSSN_gauge_RHSs::ShiftEvolutionOption", ShiftCondition)
par.set_parval_from_str("BSSN.BSSN_gauge_RHSs::LapseEvolutionOption", LapseCondition)
T4UU = None
if enable_stress_energy_source_terms == True:
registered_already = False
for i in range(len(gri.glb_gridfcs_list)):
if gri.glb_gridfcs_list[i].name == "T4UU00":
registered_already = True
if not registered_already:
T4UU = ixp.register_gridfunctions_for_single_rank2("AUXEVOL","T4UU","sym01",DIM=4)
else:
T4UU = ixp.declarerank2("T4UU","sym01",DIM=4)
# Register the BSSN constraints (Hamiltonian & momentum constraints) as gridfunctions.
registered_already = False
for i in range(len(gri.glb_gridfcs_list)):
if gri.glb_gridfcs_list[i].name == "H":
registered_already = True
if not registered_already:
H = gri.register_gridfunctions("AUX","H")
MU = ixp.register_gridfunctions_for_single_rank1("AUX", "MU")
def BSSN_RHSs_Ricci__generate_symbolic_expressions():
######################################
# START: GENERATE SYMBOLIC EXPRESSIONS
# Store original finite-differencing order:
FD_order_orig = par.parval_from_str("finite_difference::FD_CENTDERIVS_ORDER")
# Set new finite-differencing order:
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER", FD_order)
print("Generating symbolic expressions for BSSN RHSs and Ricci tensor...")
start = time.time()
# Enable rfm_precompute infrastructure, which results in
# BSSN RHSs that are free of transcendental functions,
# even in curvilinear coordinates, so long as
# ConformalFactor is set to "W" (default).
cmd.mkdir(os.path.join(outdir,"rfm_files/"))
par.set_parval_from_str("reference_metric::enable_rfm_precompute","True")
par.set_parval_from_str("reference_metric::rfm_precompute_Ccode_outdir",os.path.join(outdir,"rfm_files/"))
# Evaluate BSSN + BSSN gauge RHSs with rfm_precompute enabled:
import BSSN.BSSN_quantities as Bq
par.set_parval_from_str("BSSN.BSSN_quantities::LeaveRicciSymbolic","True")
rhs.BSSN_RHSs()
if T4UU != None:
import BSSN.BSSN_stress_energy_source_terms as Bsest
Bsest.BSSN_source_terms_for_BSSN_RHSs(T4UU)
rhs.trK_rhs += Bsest.sourceterm_trK_rhs
for i in range(3):
# Needed for Gamma-driving shift RHSs:
rhs.Lambdabar_rhsU[i] += Bsest.sourceterm_Lambdabar_rhsU[i]
# Needed for BSSN RHSs:
rhs.lambda_rhsU[i] += Bsest.sourceterm_lambda_rhsU[i]
for j in range(3):
rhs.a_rhsDD[i][j] += Bsest.sourceterm_a_rhsDD[i][j]
gaugerhs.BSSN_gauge_RHSs()
# Add Kreiss-Oliger dissipation to the BSSN RHSs:
thismodule = "KO_Dissipation"
diss_strength = par.Cparameters("REAL", thismodule, "diss_strength", default_KO_strength)
alpha_dKOD = ixp.declarerank1("alpha_dKOD")
cf_dKOD = ixp.declarerank1("cf_dKOD")
trK_dKOD = ixp.declarerank1("trK_dKOD")
betU_dKOD = ixp.declarerank2("betU_dKOD","nosym")
vetU_dKOD = ixp.declarerank2("vetU_dKOD","nosym")
lambdaU_dKOD = ixp.declarerank2("lambdaU_dKOD","nosym")
aDD_dKOD = ixp.declarerank3("aDD_dKOD","sym01")
hDD_dKOD = ixp.declarerank3("hDD_dKOD","sym01")
for k in range(3):
gaugerhs.alpha_rhs += diss_strength*alpha_dKOD[k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
rhs.cf_rhs += diss_strength* cf_dKOD[k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
rhs.trK_rhs += diss_strength* trK_dKOD[k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
for i in range(3):
if "2ndOrder" in ShiftCondition:
gaugerhs.bet_rhsU[i] += diss_strength* betU_dKOD[i][k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
gaugerhs.vet_rhsU[i] += diss_strength* vetU_dKOD[i][k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
rhs.lambda_rhsU[i] += diss_strength*lambdaU_dKOD[i][k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
for j in range(3):
rhs.a_rhsDD[i][j] += diss_strength*aDD_dKOD[i][j][k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
rhs.h_rhsDD[i][j] += diss_strength*hDD_dKOD[i][j][k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
# We use betaU as our upwinding control vector:
Bq.BSSN_basic_tensors()
betaU = Bq.betaU
# Next compute Ricci tensor
par.set_parval_from_str("BSSN.BSSN_quantities::LeaveRicciSymbolic","False")
Bq.RicciBar__gammabarDD_dHatD__DGammaUDD__DGammaU()
# Now that we are finished with all the rfm hatted
# quantities in generic precomputed functional
# form, let's restore them to their closed-
# form expressions.
par.set_parval_from_str("reference_metric::enable_rfm_precompute","False") # Reset to False to disable rfm_precompute.
rfm.ref_metric__hatted_quantities()
end = time.time()
print("Finished BSSN symbolic expressions in "+str(end-start)+" seconds.")
# Restore original finite-differencing order:
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER", FD_order)
# END: GENERATE SYMBOLIC EXPRESSIONS
######################################
BSSN_RHSs_SymbExpressions = [lhrh(lhs=gri.gfaccess("rhs_gfs","aDD00"), rhs=rhs.a_rhsDD[0][0]),
lhrh(lhs=gri.gfaccess("rhs_gfs","aDD01"), rhs=rhs.a_rhsDD[0][1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","aDD02"), rhs=rhs.a_rhsDD[0][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","aDD11"), rhs=rhs.a_rhsDD[1][1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","aDD12"), rhs=rhs.a_rhsDD[1][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","aDD22"), rhs=rhs.a_rhsDD[2][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","alpha"), rhs=gaugerhs.alpha_rhs),
lhrh(lhs=gri.gfaccess("rhs_gfs","betU0"), rhs=gaugerhs.bet_rhsU[0]),
lhrh(lhs=gri.gfaccess("rhs_gfs","betU1"), rhs=gaugerhs.bet_rhsU[1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","betU2"), rhs=gaugerhs.bet_rhsU[2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","cf"), rhs=rhs.cf_rhs),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD00"), rhs=rhs.h_rhsDD[0][0]),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD01") ,rhs=rhs.h_rhsDD[0][1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD02"), rhs=rhs.h_rhsDD[0][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD11"), rhs=rhs.h_rhsDD[1][1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD12"), rhs=rhs.h_rhsDD[1][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD22"), rhs=rhs.h_rhsDD[2][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","lambdaU0"),rhs=rhs.lambda_rhsU[0]),
lhrh(lhs=gri.gfaccess("rhs_gfs","lambdaU1"),rhs=rhs.lambda_rhsU[1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","lambdaU2"),rhs=rhs.lambda_rhsU[2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","trK"), rhs=rhs.trK_rhs),
lhrh(lhs=gri.gfaccess("rhs_gfs","vetU0"), rhs=gaugerhs.vet_rhsU[0]),
lhrh(lhs=gri.gfaccess("rhs_gfs","vetU1"), rhs=gaugerhs.vet_rhsU[1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","vetU2"), rhs=gaugerhs.vet_rhsU[2]) ]
Ricci_SymbExpressions = [lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD00"),rhs=Bq.RbarDD[0][0]),
lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD01"),rhs=Bq.RbarDD[0][1]),
lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD02"),rhs=Bq.RbarDD[0][2]),
lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD11"),rhs=Bq.RbarDD[1][1]),
lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD12"),rhs=Bq.RbarDD[1][2]),
lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD22"),rhs=Bq.RbarDD[2][2])]
return [betaU,BSSN_RHSs_SymbExpressions,Ricci_SymbExpressions]
def BSSN_RHSs_Ricci__generate_Ccode(which_expressions, all_RHSs_Ricci_exprs_list):
betaU = all_RHSs_Ricci_exprs_list[0]
BSSN_RHSs_SymbExpressions = all_RHSs_Ricci_exprs_list[1]
Ricci_SymbExpressions = all_RHSs_Ricci_exprs_list[2]
if(which_expressions == "BSSN_RHSs"):
print("Generating C code for BSSN RHSs (FD_order="+str(FD_order)+",Tmunu="+str(enable_stress_energy_source_terms)+") in "+par.parval_from_str("reference_metric::CoordSystem")+" coordinates.")
start = time.time()
BSSN_RHSs_string = fin.FD_outputC("returnstring",BSSN_RHSs_SymbExpressions,
params="outCverbose=False,SIMD_enable=True",
upwindcontrolvec=betaU)
with open(os.path.join(outdir,"BSSN_RHSs_FD_order_"+str(FD_order)+"_enable_Tmunu_"+str(enable_stress_energy_source_terms)+".h"), "w") as file:
file.write(lp.loop(["i2","i1","i0"],["cctk_nghostzones[2]","cctk_nghostzones[1]","cctk_nghostzones[0]"],
["cctk_lsh[2]-cctk_nghostzones[2]","cctk_lsh[1]-cctk_nghostzones[1]","cctk_lsh[0]-cctk_nghostzones[0]"],
["1","1","SIMD_width"],
["#pragma omp parallel for",
"#include \"rfm_files/rfm_struct__SIMD_outer_read2.h\"",
"#include \"rfm_files/rfm_struct__SIMD_outer_read1.h\""],"",
"#include \"rfm_files/rfm_struct__SIMD_inner_read0.h\"\n"+BSSN_RHSs_string))
end = time.time()
print("Finished BSSN_RHS C codegen (FD_order="+str(FD_order)+",Tmunu="+str(enable_stress_energy_source_terms)+") in " + str(end - start) + " seconds.")
elif(which_expressions == "Ricci"):
print("Generating C code for Ricci tensor (FD_order="+str(FD_order)+") in "+par.parval_from_str("reference_metric::CoordSystem")+" coordinates.")
start = time.time()
Ricci_string = fin.FD_outputC("returnstring", Ricci_SymbExpressions,
params="outCverbose=False,SIMD_enable=True")
with open(os.path.join(outdir,"BSSN_Ricci_FD_order_"+str(FD_order)+".h"), "w") as file:
file.write(lp.loop(["i2","i1","i0"],["cctk_nghostzones[2]","cctk_nghostzones[1]","cctk_nghostzones[0]"],
["cctk_lsh[2]-cctk_nghostzones[2]","cctk_lsh[1]-cctk_nghostzones[1]","cctk_lsh[0]-cctk_nghostzones[0]"],
["1","1","SIMD_width"],
["#pragma omp parallel for",
"#include \"rfm_files/rfm_struct__SIMD_outer_read2.h\"",
"#include \"rfm_files/rfm_struct__SIMD_outer_read1.h\""],"",
"#include \"rfm_files/rfm_struct__SIMD_inner_read0.h\"\n"+Ricci_string))
end = time.time()
print("Finished Ricci C codegen (FD_order="+str(FD_order)+") in " + str(end - start) + " seconds.")
else:
print("Error: unexpected argument, "+str(which_expressions)+" to BSSN_RHSs_Ricci__generate_Ccode()")
def BSSN_constraints__generate_symbolic_expressions_and_C_code():
######################################
# START: GENERATE SYMBOLIC EXPRESSIONS
######################################
# START: GENERATE SYMBOLIC EXPRESSIONS
# Store original finite-differencing order:
FD_order_orig = par.parval_from_str("finite_difference::FD_CENTDERIVS_ORDER")
# Set new finite-differencing order:
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER", FD_order)
# Define the Hamiltonian constraint and output the optimized C code.
import BSSN.BSSN_constraints as bssncon
bssncon.BSSN_constraints(add_T4UUmunu_source_terms=False)
if T4UU != None:
import BSSN.BSSN_stress_energy_source_terms as Bsest
Bsest.BSSN_source_terms_for_BSSN_RHSs(T4UU)
Bsest.BSSN_source_terms_for_BSSN_constraints(T4UU)
bssncon.H += Bsest.sourceterm_H
for i in range(3):
bssncon.MU[i] += Bsest.sourceterm_MU[i]
# Restore original finite-differencing order:
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER", FD_order)
# END: GENERATE SYMBOLIC EXPRESSIONS
######################################
start = time.time()
print("Generating optimized C code for Ham. & mom. constraints. May take a while, depending on CoordSystem.")
Ham_mom_string = fin.FD_outputC("returnstring",
[lhrh(lhs=gri.gfaccess("aux_gfs", "H"), rhs=bssncon.H),
lhrh(lhs=gri.gfaccess("aux_gfs", "MU0"), rhs=bssncon.MU[0]),
lhrh(lhs=gri.gfaccess("aux_gfs", "MU1"), rhs=bssncon.MU[1]),
lhrh(lhs=gri.gfaccess("aux_gfs", "MU2"), rhs=bssncon.MU[2])],
params="outCverbose=False")
with open(os.path.join(outdir,"BSSN_constraints_FD_order_"+str(FD_order)+"_enable_Tmunu_"+str(enable_stress_energy_source_terms)+".h"), "w") as file:
file.write(lp.loop(["i2","i1","i0"],["cctk_nghostzones[2]","cctk_nghostzones[1]","cctk_nghostzones[0]"],
["cctk_lsh[2]-cctk_nghostzones[2]","cctk_lsh[1]-cctk_nghostzones[1]","cctk_lsh[0]-cctk_nghostzones[0]"],
["1","1","1"],["#pragma omp parallel for","",""], "", Ham_mom_string))
end = time.time()
print("Finished Hamiltonian & momentum constraint C codegen (FD_order="+str(FD_order)+",Tmunu="+str(enable_stress_energy_source_terms)+") in " + str(end - start) + " seconds.")
def enforce_detgammabar_eq_detgammahat__generate_symbolic_expressions_and_C_code():
######################################
# START: GENERATE SYMBOLIC EXPRESSIONS
# Store original finite-differencing order:
FD_order_orig = par.parval_from_str("finite_difference::FD_CENTDERIVS_ORDER")
# Set new finite-differencing order:
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER", FD_order)
# Enable rfm_precompute infrastructure, which results in
# BSSN RHSs that are free of transcendental functions,
# even in curvilinear coordinates, so long as
# ConformalFactor is set to "W" (default).
cmd.mkdir(os.path.join(outdir,"rfm_files/"))
par.set_parval_from_str("reference_metric::enable_rfm_precompute","True")
par.set_parval_from_str("reference_metric::rfm_precompute_Ccode_outdir",os.path.join(outdir,"rfm_files/"))
import BSSN.Enforce_Detgammabar_Constraint as EGC
enforce_detg_constraint_symb_expressions = EGC.Enforce_Detgammabar_Constraint_symb_expressions()
# Now that we are finished with all the rfm hatted
# quantities in generic precomputed functional
# form, let's restore them to their closed-
# form expressions.
par.set_parval_from_str("reference_metric::enable_rfm_precompute","False") # Reset to False to disable rfm_precompute.
rfm.ref_metric__hatted_quantities()
# Restore original finite-differencing order:
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER", FD_order)
# END: GENERATE SYMBOLIC EXPRESSIONS
######################################
start = time.time()
print("Generating optimized C code (FD_order="+str(FD_order)+") for gamma constraint. May take a while, depending on CoordSystem.")
enforce_gammadet_string = fin.FD_outputC("returnstring", enforce_detg_constraint_symb_expressions,
params="outCverbose=False,preindent=0,includebraces=False")
with open(os.path.join(outdir,"enforcedetgammabar_constraint_FD_order_"+str(FD_order)+".h"), "w") as file:
file.write(lp.loop(["i2","i1","i0"],["0", "0", "0"],
["cctk_lsh[2]","cctk_lsh[1]","cctk_lsh[0]"],
["1","1","1"],
["#pragma omp parallel for",
"#include \"rfm_files/rfm_struct__read2.h\"",
"#include \"rfm_files/rfm_struct__read1.h\""],"",
"#include \"rfm_files/rfm_struct__read0.h\"\n"+enforce_gammadet_string))
end = time.time()
print("Finished gamma constraint C codegen (FD_order="+str(FD_order)+") in " + str(end - start) + " seconds.")
if WhichPart=="BSSN_RHSs":
BSSN_RHSs_Ricci__generate_Ccode("BSSN_RHSs", BSSN_RHSs_Ricci__generate_symbolic_expressions())
elif WhichPart=="Ricci":
BSSN_RHSs_Ricci__generate_Ccode("Ricci", BSSN_RHSs_Ricci__generate_symbolic_expressions())
elif WhichPart=="BSSN_constraints":
BSSN_constraints__generate_symbolic_expressions_and_C_code()
elif WhichPart=="detgammabar_constraint":
enforce_detgammabar_eq_detgammahat__generate_symbolic_expressions_and_C_code()
else:
print("Error: WhichPart = "+WhichPart+" is not recognized.")
sys.exit(1)
# Store all NRPy+ environment variables to an output string so NRPy+ environment from within this subprocess can be easily restored
import pickle
# https://www.pythonforthelab.com/blog/storing-binary-data-and-serializing/
outstr = []
outstr.append(pickle.dumps(len(gri.glb_gridfcs_list)))
for lst in gri.glb_gridfcs_list:
outstr.append(pickle.dumps(lst.gftype))
outstr.append(pickle.dumps(lst.name))
outstr.append(pickle.dumps(lst.rank))
outstr.append(pickle.dumps(lst.DIM))
outstr.append(pickle.dumps(len(par.glb_params_list)))
for lst in par.glb_params_list:
outstr.append(pickle.dumps(lst.type))
outstr.append(pickle.dumps(lst.module))
outstr.append(pickle.dumps(lst.parname))
outstr.append(pickle.dumps(lst.defaultval))
outstr.append(pickle.dumps(len(par.glb_Cparams_list)))
for lst in par.glb_Cparams_list:
outstr.append(pickle.dumps(lst.type))
outstr.append(pickle.dumps(lst.module))
outstr.append(pickle.dumps(lst.parname))
outstr.append(pickle.dumps(lst.defaultval))
return outstr
def BSSN_RHSs_Ricci__generate_symbolic_expressions():
######################################
# START: GENERATE SYMBOLIC EXPRESSIONS
# Store original finite-differencing order:
FD_order_orig = par.parval_from_str("finite_difference::FD_CENTDERIVS_ORDER")
# Set new finite-differencing order:
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER", FD_order)
print("Generating symbolic expressions for BSSN RHSs and Ricci tensor...")
start = time.time()
# Enable rfm_precompute infrastructure, which results in
# BSSN RHSs that are free of transcendental functions,
# even in curvilinear coordinates, so long as
# ConformalFactor is set to "W" (default).
cmd.mkdir(os.path.join(outdir,"rfm_files/"))
par.set_parval_from_str("reference_metric::enable_rfm_precompute","True")
par.set_parval_from_str("reference_metric::rfm_precompute_Ccode_outdir",os.path.join(outdir,"rfm_files/"))
# Evaluate BSSN + BSSN gauge RHSs with rfm_precompute enabled:
import BSSN.BSSN_quantities as Bq
par.set_parval_from_str("BSSN.BSSN_quantities::LeaveRicciSymbolic","True")
rhs.BSSN_RHSs()
if T4UU != None:
import BSSN.BSSN_stress_energy_source_terms as Bsest
Bsest.BSSN_source_terms_for_BSSN_RHSs(T4UU)
rhs.trK_rhs += Bsest.sourceterm_trK_rhs
for i in range(3):
# Needed for Gamma-driving shift RHSs:
rhs.Lambdabar_rhsU[i] += Bsest.sourceterm_Lambdabar_rhsU[i]
# Needed for BSSN RHSs:
rhs.lambda_rhsU[i] += Bsest.sourceterm_lambda_rhsU[i]
for j in range(3):
rhs.a_rhsDD[i][j] += Bsest.sourceterm_a_rhsDD[i][j]
gaugerhs.BSSN_gauge_RHSs()
# Add Kreiss-Oliger dissipation to the BSSN RHSs:
thismodule = "KO_Dissipation"
diss_strength = par.Cparameters("REAL", thismodule, "diss_strength", default_KO_strength)
alpha_dKOD = ixp.declarerank1("alpha_dKOD")
cf_dKOD = ixp.declarerank1("cf_dKOD")
trK_dKOD = ixp.declarerank1("trK_dKOD")
betU_dKOD = ixp.declarerank2("betU_dKOD","nosym")
vetU_dKOD = ixp.declarerank2("vetU_dKOD","nosym")
lambdaU_dKOD = ixp.declarerank2("lambdaU_dKOD","nosym")
aDD_dKOD = ixp.declarerank3("aDD_dKOD","sym01")
hDD_dKOD = ixp.declarerank3("hDD_dKOD","sym01")
for k in range(3):
gaugerhs.alpha_rhs += diss_strength*alpha_dKOD[k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
rhs.cf_rhs += diss_strength* cf_dKOD[k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
rhs.trK_rhs += diss_strength* trK_dKOD[k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
for i in range(3):
if "2ndOrder" in ShiftCondition:
gaugerhs.bet_rhsU[i] += diss_strength* betU_dKOD[i][k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
gaugerhs.vet_rhsU[i] += diss_strength* vetU_dKOD[i][k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
rhs.lambda_rhsU[i] += diss_strength*lambdaU_dKOD[i][k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
for j in range(3):
rhs.a_rhsDD[i][j] += diss_strength*aDD_dKOD[i][j][k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
rhs.h_rhsDD[i][j] += diss_strength*hDD_dKOD[i][j][k]*rfm.ReU[k] # ReU[k] = 1/scalefactor_orthog_funcform[k]
# We use betaU as our upwinding control vector:
Bq.BSSN_basic_tensors()
betaU = Bq.betaU
# Next compute Ricci tensor
par.set_parval_from_str("BSSN.BSSN_quantities::LeaveRicciSymbolic","False")
Bq.RicciBar__gammabarDD_dHatD__DGammaUDD__DGammaU()
# Now that we are finished with all the rfm hatted
# quantities in generic precomputed functional
# form, let's restore them to their closed-
# form expressions.
par.set_parval_from_str("reference_metric::enable_rfm_precompute","False") # Reset to False to disable rfm_precompute.
rfm.ref_metric__hatted_quantities()
end = time.time()
print("Finished BSSN symbolic expressions in "+str(end-start)+" seconds.")
# Restore original finite-differencing order:
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER", FD_order)
# END: GENERATE SYMBOLIC EXPRESSIONS
######################################
BSSN_RHSs_SymbExpressions = [lhrh(lhs=gri.gfaccess("rhs_gfs","aDD00"), rhs=rhs.a_rhsDD[0][0]),
lhrh(lhs=gri.gfaccess("rhs_gfs","aDD01"), rhs=rhs.a_rhsDD[0][1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","aDD02"), rhs=rhs.a_rhsDD[0][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","aDD11"), rhs=rhs.a_rhsDD[1][1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","aDD12"), rhs=rhs.a_rhsDD[1][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","aDD22"), rhs=rhs.a_rhsDD[2][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","alpha"), rhs=gaugerhs.alpha_rhs),
lhrh(lhs=gri.gfaccess("rhs_gfs","betU0"), rhs=gaugerhs.bet_rhsU[0]),
lhrh(lhs=gri.gfaccess("rhs_gfs","betU1"), rhs=gaugerhs.bet_rhsU[1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","betU2"), rhs=gaugerhs.bet_rhsU[2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","cf"), rhs=rhs.cf_rhs),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD00"), rhs=rhs.h_rhsDD[0][0]),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD01") ,rhs=rhs.h_rhsDD[0][1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD02"), rhs=rhs.h_rhsDD[0][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD11"), rhs=rhs.h_rhsDD[1][1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD12"), rhs=rhs.h_rhsDD[1][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","hDD22"), rhs=rhs.h_rhsDD[2][2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","lambdaU0"),rhs=rhs.lambda_rhsU[0]),
lhrh(lhs=gri.gfaccess("rhs_gfs","lambdaU1"),rhs=rhs.lambda_rhsU[1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","lambdaU2"),rhs=rhs.lambda_rhsU[2]),
lhrh(lhs=gri.gfaccess("rhs_gfs","trK"), rhs=rhs.trK_rhs),
lhrh(lhs=gri.gfaccess("rhs_gfs","vetU0"), rhs=gaugerhs.vet_rhsU[0]),
lhrh(lhs=gri.gfaccess("rhs_gfs","vetU1"), rhs=gaugerhs.vet_rhsU[1]),
lhrh(lhs=gri.gfaccess("rhs_gfs","vetU2"), rhs=gaugerhs.vet_rhsU[2]) ]
Ricci_SymbExpressions = [lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD00"),rhs=Bq.RbarDD[0][0]),
lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD01"),rhs=Bq.RbarDD[0][1]),
lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD02"),rhs=Bq.RbarDD[0][2]),
lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD11"),rhs=Bq.RbarDD[1][1]),
lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD12"),rhs=Bq.RbarDD[1][2]),
lhrh(lhs=gri.gfaccess("auxevol_gfs","RbarDD22"),rhs=Bq.RbarDD[2][2])]
return [betaU,BSSN_RHSs_SymbExpressions,Ricci_SymbExpressions]
import os, sys # Python module: used for system and OS specific commands
import sympy as sp # Python module: used for symbolic expressions
# Register NRPy+ root directory to the path
nrpy_dir_path = os.path.join("..", "..")
if nrpy_dir_path not in sys.path:
sys.path.append(nrpy_dir_path)
# Load NRPy+ modules
from outputC import * # NRPy+ module: used to output sympy expressions to C
import indexedexp as ixp # NRPy+ module: used to generate indexed expressions (e.g. g_{\mu\nu})
import cmdline_helper as cmd # NRPy+ module: used for command line features
# Create the NRPy+ header file directory, if it doesn't already exist
IGM_src_dir_path = os.path.join("..", "src")
cmd.mkdir(os.path.join(IGM_src_dir_path, "NRPy_generated_headers"))
NRPy_headers_dir_path = os.path.join(IGM_src_dir_path,
"NRPy_generated_headers")
# Set up a neat function to output the expressions to NRPy+ generated files
def NRPy_IGM_write_to_file(filepath,
filename,
contents,
precontents="",
postcontents=""):
with open(filepath, "w") as file:
file.write("""
/* .-----------------------------------------------------------------------.
* | This file was generated by NRPy+ for IllinoisGRMHD, as documented in: |
* | Tutorial-IllinoisGRMHD__NRPyfied_IGM_expressions.ipynb |
def GiRaFFE_NRPy_A2B(outdir, gammaDD, AD, BU):
cmd.mkdir(outdir)
# Set spatial dimension (must be 3 for BSSN)
DIM = 3
par.set_parval_from_str("grid::DIM", DIM)
# Compute the sqrt of the three metric determinant.
import GRHD.equations as gh
gh.compute_sqrtgammaDET(gammaDD)
# Import the Levi-Civita symbol and build the corresponding tensor.
# We already have a handy function to define the Levi-Civita symbol in indexedexp.py
LeviCivitaUUU = ixp.LeviCivitaTensorUUU_dim3_rank3(gh.sqrtgammaDET)
AD_dD = ixp.declarerank2("AD_dD", "nosym")
BU = ixp.zerorank1()
for i in range(DIM):
for j in range(DIM):
for k in range(DIM):
BU[i] += LeviCivitaUUU[i][j][k] * AD_dD[k][j]
# Write the code to compute derivatives with shifted stencils as needed.
with open(os.path.join(outdir, "driver_AtoB.h"), "w") as file:
file.write(prefunc)
# Now, we'll also write some more auxiliary functions to handle the order-lowering method for A2B
with open(os.path.join(outdir, "driver_AtoB.h"), "a") as file:
file.write("""REAL relative_error(REAL a, REAL b) {
if((a+b)!=0.0) {
return 2.0*fabs(a-b)/fabs(a+b);
}
else {
return 0.0;
}
}
#define M2 0
#define M1 1
#define P0 2
#define P1 3
#define P2 4
#define CN4 0
#define CN2 1
#define UP2 2
#define DN2 3
#define UP1 4
#define DN1 5
void compute_Bx_pointwise(REAL *Bx, const REAL invdy, const REAL *Ay, const REAL invdz, const REAL *Az) {
REAL dz_Ay,dy_Az;
dz_Ay = invdz*((Ay[P1]-Ay[M1])*2.0/3.0 - (Ay[P2]-Ay[M2])/12.0);
dy_Az = invdy*((Az[P1]-Az[M1])*2.0/3.0 - (Az[P2]-Az[M2])/12.0);
Bx[CN4] = dy_Az - dz_Ay;
dz_Ay = invdz*(Ay[P1]-Ay[M1])/2.0;
dy_Az = invdy*(Az[P1]-Az[M1])/2.0;
Bx[CN2] = dy_Az - dz_Ay;
dz_Ay = invdz*(-1.5*Ay[P0]+2.0*Ay[P1]-0.5*Ay[P2]);
dy_Az = invdy*(-1.5*Az[P0]+2.0*Az[P1]-0.5*Az[P2]);
Bx[UP2] = dy_Az - dz_Ay;
dz_Ay = invdz*(1.5*Ay[P0]-2.0*Ay[M1]+0.5*Ay[M2]);
dy_Az = invdy*(1.5*Az[P0]-2.0*Az[M1]+0.5*Az[M2]);
Bx[DN2] = dy_Az - dz_Ay;
dz_Ay = invdz*(Ay[P1]-Ay[P0]);
dy_Az = invdy*(Az[P1]-Az[P0]);
Bx[UP1] = dy_Az - dz_Ay;
dz_Ay = invdz*(Ay[P0]-Ay[M1]);
dy_Az = invdy*(Az[P0]-Az[M1]);
Bx[DN1] = dy_Az - dz_Ay;
}
#define TOLERANCE_A2B 1.0e-4
REAL find_accepted_Bx_order(REAL *Bx) {
REAL accepted_val = Bx[CN4];
REAL Rel_error_o2_vs_o4 = relative_error(Bx[CN2],Bx[CN4]);
REAL Rel_error_oCN2_vs_oDN2 = relative_error(Bx[CN2],Bx[DN2]);
REAL Rel_error_oCN2_vs_oUP2 = relative_error(Bx[CN2],Bx[UP2]);
if(Rel_error_o2_vs_o4 > TOLERANCE_A2B) {
accepted_val = Bx[CN2];
if(Rel_error_o2_vs_o4 > Rel_error_oCN2_vs_oDN2 || Rel_error_o2_vs_o4 > Rel_error_oCN2_vs_oUP2) {
// Should we use AND or OR in if statement?
if(relative_error(Bx[UP2],Bx[UP1]) < relative_error(Bx[DN2],Bx[DN1])) {
accepted_val = Bx[UP2];
}
else {
accepted_val = Bx[DN2];
}
}
}
return accepted_val;
}
""")
# order_lowering_body = """REAL AD0_1[5],AD0_2[5],AD1_2[5],AD1_0[5],AD2_0[5],AD2_1[5];
# const double gammaDD00 = auxevol_gfs[IDX4S(GAMMADD00GF, i0,i1,i2)];
# const double gammaDD01 = auxevol_gfs[IDX4S(GAMMADD01GF, i0,i1,i2)];
# const double gammaDD02 = auxevol_gfs[IDX4S(GAMMADD02GF, i0,i1,i2)];
# const double gammaDD11 = auxevol_gfs[IDX4S(GAMMADD11GF, i0,i1,i2)];
# const double gammaDD12 = auxevol_gfs[IDX4S(GAMMADD12GF, i0,i1,i2)];
# const double gammaDD22 = auxevol_gfs[IDX4S(GAMMADD22GF, i0,i1,i2)];
# AD0_2[M2] = in_gfs[IDX4S(AD0GF, i0,i1,i2-2)];
# AD0_2[M1] = in_gfs[IDX4S(AD0GF, i0,i1,i2-1)];
# AD0_1[M2] = in_gfs[IDX4S(AD0GF, i0,i1-2,i2)];
# AD0_1[M1] = in_gfs[IDX4S(AD0GF, i0,i1-1,i2)];
# AD0_1[P0] = AD0_2[P0] = in_gfs[IDX4S(AD0GF, i0,i1,i2)];
# AD0_1[P1] = in_gfs[IDX4S(AD0GF, i0,i1+1,i2)];
# AD0_1[P2] = in_gfs[IDX4S(AD0GF, i0,i1+2,i2)];
# AD0_2[P1] = in_gfs[IDX4S(AD0GF, i0,i1,i2+1)];
# AD0_2[P2] = in_gfs[IDX4S(AD0GF, i0,i1,i2+2)];
# AD1_2[M2] = in_gfs[IDX4S(AD1GF, i0,i1,i2-2)];
# AD1_2[M1] = in_gfs[IDX4S(AD1GF, i0,i1,i2-1)];
# AD1_0[M2] = in_gfs[IDX4S(AD1GF, i0-2,i1,i2)];
# AD1_0[M1] = in_gfs[IDX4S(AD1GF, i0-1,i1,i2)];
# AD1_2[P0] = AD1_0[P0] = in_gfs[IDX4S(AD1GF, i0,i1,i2)];
# AD1_0[P1] = in_gfs[IDX4S(AD1GF, i0+1,i1,i2)];
# AD1_0[P2] = in_gfs[IDX4S(AD1GF, i0+2,i1,i2)];
# AD1_2[P1] = in_gfs[IDX4S(AD1GF, i0,i1,i2+1)];
# AD1_2[P2] = in_gfs[IDX4S(AD1GF, i0,i1,i2+2)];
# AD2_1[M2] = in_gfs[IDX4S(AD2GF, i0,i1-2,i2)];
# AD2_1[M1] = in_gfs[IDX4S(AD2GF, i0,i1-1,i2)];
# AD2_0[M2] = in_gfs[IDX4S(AD2GF, i0-2,i1,i2)];
# AD2_0[M1] = in_gfs[IDX4S(AD2GF, i0-1,i1,i2)];
# AD2_0[P0] = AD2_1[P0] = in_gfs[IDX4S(AD2GF, i0,i1,i2)];
# AD2_0[P1] = in_gfs[IDX4S(AD2GF, i0+1,i1,i2)];
# AD2_0[P2] = in_gfs[IDX4S(AD2GF, i0+2,i1,i2)];
# AD2_1[P1] = in_gfs[IDX4S(AD2GF, i0,i1+1,i2)];
# AD2_1[P2] = in_gfs[IDX4S(AD2GF, i0,i1+2,i2)];
# const double invsqrtg = 1.0/sqrt(gammaDD00*gammaDD11*gammaDD22
# - gammaDD00*gammaDD12*gammaDD12
# + 2*gammaDD01*gammaDD02*gammaDD12
# - gammaDD11*gammaDD02*gammaDD02
# - gammaDD22*gammaDD01*gammaDD01);
# REAL BU0[4],BU1[4],BU2[4];
# compute_Bx_pointwise(BU0,invdx2,AD1_2,invdx1,AD2_1);
# compute_Bx_pointwise(BU1,invdx0,AD2_0,invdx2,AD0_2);
# compute_Bx_pointwise(BU2,invdx1,AD0_1,invdx0,AD1_0);
# auxevol_gfs[IDX4S(BU0GF, i0,i1,i2)] = find_accepted_Bx_order(BU0)*invsqrtg;
# auxevol_gfs[IDX4S(BU1GF, i0,i1,i2)] = find_accepted_Bx_order(BU1)*invsqrtg;
# auxevol_gfs[IDX4S(BU2GF, i0,i1,i2)] = find_accepted_Bx_order(BU2)*invsqrtg;
# """
# Here, we'll use the outCfunction() function to output a function that will compute the magnetic field
# on the interior. Then, we'll add postloop code to handle the ghostzones.
desc = "Compute the magnetic field from the vector potential everywhere, including ghostzones"
name = "driver_A_to_B"
driver_Ccode = outCfunction(
outfile="returnstring",
desc=desc,
name=name,
params=
"const paramstruct *restrict params,REAL *restrict in_gfs,REAL *restrict auxevol_gfs",
body=fin.FD_outputC("returnstring", [
lhrh(lhs=gri.gfaccess("out_gfs", "BU0"), rhs=BU[0]),
lhrh(lhs=gri.gfaccess("out_gfs", "BU1"), rhs=BU[1]),
lhrh(lhs=gri.gfaccess("out_gfs", "BU2"), rhs=BU[2])
]),
# body = order_lowering_body,
postloop="""
int imin[3] = { NGHOSTS_A2B, NGHOSTS_A2B, NGHOSTS_A2B };
int imax[3] = { NGHOSTS+Nxx0, NGHOSTS+Nxx1, NGHOSTS+Nxx2 };
// Now, we loop over the ghostzones to calculate the magnetic field there.
for(int which_gz = 0; which_gz < NGHOSTS_A2B; which_gz++) {
// After updating each face, adjust imin[] and imax[]
// to reflect the newly-updated face extents.
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0]-1,imin[0], imin[1],imax[1], imin[2],imax[2]); imin[0]--;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imax[0],imax[0]+1, imin[1],imax[1], imin[2],imax[2]); imax[0]++;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imin[1]-1,imin[1], imin[2],imax[2]); imin[1]--;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imax[1],imax[1]+1, imin[2],imax[2]); imax[1]++;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imin[1],imax[1], imin[2]-1,imin[2]); imin[2]--;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imin[1],imax[1], imax[2],imax[2]+1); imax[2]++;
}
""",
loopopts="InteriorPoints",
rel_path_to_Cparams=os.path.join("../")).replace(
"= NGHOSTS", "= NGHOSTS_A2B").replace(
"NGHOSTS+Nxx0", "Nxx_plus_2NGHOSTS0-NGHOSTS_A2B").replace(
"NGHOSTS+Nxx1", "Nxx_plus_2NGHOSTS1-NGHOSTS_A2B").replace(
"NGHOSTS+Nxx2", "Nxx_plus_2NGHOSTS2-NGHOSTS_A2B")
with open(os.path.join(outdir, "driver_AtoB.h"), "a") as file:
file.write(driver_Ccode)
def GiRaFFE_NRPy_BCs(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(os.path.join(Ccodesdir,name),"w") as file:
file.write(body)
def Set_up_CurviBoundaryConditions(Ccodesdir,
verbose=True,
Cparamspath=os.path.join("../"),
enable_copy_of_static_Ccodes=True,
BoundaryCondition="QuadraticExtrapolation"):
# Step P0: Check that Ccodesdir is not the same as CurviBoundaryConditions/boundary_conditions,
# to prevent trusted versions of these C codes from becoming contaminated.
if os.path.join(Ccodesdir) == os.path.join("CurviBoundaryConditions",
"boundary_conditions"):
print(
"Error: Tried to output boundary conditions C code into CurviBoundaryConditions/boundary_conditions,"
" which is not allowed, to prevent trusted versions of these C codes from becoming contaminated."
)
sys.exit(1)
# Step P1: Create the C codes output directory & copy static CurviBC files
# from CurviBoundaryConditions/boundary_conditions to Ccodesdir/
if enable_copy_of_static_Ccodes:
cmd.mkdir(os.path.join(Ccodesdir))
# Choosing boundary condition drivers with in NRPy+
# - current options are Quadratic Polynomial Extrapolation for any coordinate system,
# and the Sommerfeld boundary condition for only cartesian coordinates
if str(BoundaryCondition) == "QuadraticExtrapolation":
for file in [
"apply_bcs_curvilinear.h", "BCs_data_structs.h",
"bcstruct_freemem.h", "CurviBC_include_Cfunctions.h",
"driver_bcstruct.h", "set_bcstruct.h",
"set_up__bc_gz_map_and_parity_condns.h"
]:
shutil.copy(
os.path.join("CurviBoundaryConditions",
"boundary_conditions", file),
os.path.join(Ccodesdir))
with open(os.path.join(Ccodesdir, "CurviBC_include_Cfunctions.h"),
"a") as file:
file.write("\n#include \"apply_bcs_curvilinear.h\"")
elif str(BoundaryCondition) == "Sommerfeld":
for file in [
"apply_bcs_sommerfeld.h", "BCs_data_structs.h",
"bcstruct_freemem.h", "CurviBC_include_Cfunctions.h",
"driver_bcstruct.h", "set_bcstruct.h",
"set_up__bc_gz_map_and_parity_condns.h"
]:
shutil.copy(
os.path.join("CurviBoundaryConditions",
"boundary_conditions", file),
os.path.join(Ccodesdir))
with open(os.path.join(Ccodesdir, "CurviBC_include_Cfunctions.h"),
"a") as file:
file.write("\n#include \"apply_bcs_sommerfeld.h\"")
elif str(BoundaryCondition) == "QuadraticExtrapolation&Sommerfeld":
for file in [
"apply_bcs_curvilinear.h", "apply_bcs_sommerfeld.h",
"BCs_data_structs.h", "bcstruct_freemem.h",
"CurviBC_include_Cfunctions.h", "driver_bcstruct.h",
"set_bcstruct.h", "set_up__bc_gz_map_and_parity_condns.h"
]:
shutil.copy(
os.path.join("CurviBoundaryConditions",
"boundary_conditions", file),
os.path.join(Ccodesdir))
with open(os.path.join(Ccodesdir, "CurviBC_include_Cfunctions.h"),
"a") as file:
file.write("\n#include \"apply_bcs_sommerfeld.h\"" +
"\n#include \"apply_bcs_curvilinear.h\"")
else:
print(
"ERROR: Only Quadratic Polynomial Extrapolation (QuadraticExtrapolation) and Sommerfeld boundary conditions are currently supported\n"
)
sys.exit(1)
# Step P2: Output correct #include for set_Cparameters.h to
# Ccodesdir/boundary_conditions/RELATIVE_PATH__set_Cparameters.h
with open(os.path.join(Ccodesdir, "RELATIVE_PATH__set_Cparameters.h"),
"w") as file:
file.write(
"#include \"" + Cparamspath + "/set_Cparameters.h\"\n"
) # #include's may include forward slashes for paths, even in Windows.
# Step 0: Set up reference metric in case it hasn't already been set up.
# (Doing it twice hurts nothing).
rfm.reference_metric()
# Step 1: Set unit-vector dot products (=parity) for each of the 10 parity condition types
parity = ixp.zerorank1(DIM=10)
UnitVectors_inner = ixp.zerorank2()
xx0_inbounds, xx1_inbounds, xx2_inbounds = sp.symbols(
"xx0_inbounds xx1_inbounds xx2_inbounds", real=True)
for i in range(3):
for j in range(3):
UnitVectors_inner[i][j] = rfm.UnitVectors[i][j].subs(
rfm.xx[0],
xx0_inbounds).subs(rfm.xx[1],
xx1_inbounds).subs(rfm.xx[2], xx2_inbounds)
# Type 0: scalar
parity[0] = sp.sympify(1)
# Type 1: i0-direction vector or one-form
# Type 2: i1-direction vector or one-form
# Type 3: i2-direction vector or one-form
for i in range(3):
for Type in range(1, 4):
parity[Type] += rfm.UnitVectors[Type -
1][i] * UnitVectors_inner[Type -
1][i]
# Type 4: i0i0-direction rank-2 tensor
# parity[4] = parity[1]*parity[1]
# Type 5: i0i1-direction rank-2 tensor
# Type 6: i0i2-direction rank-2 tensor
# Type 7: i1i1-direction rank-2 tensor
# Type 8: i1i2-direction rank-2 tensor
# Type 9: i2i2-direction rank-2 tensor
count = 4
for i in range(3):
for j in range(i, 3):
parity[count] = parity[i + 1] * parity[j + 1]
count = count + 1
lhs_strings = []
for i in range(10):
lhs_strings.append("parity[" + str(i) + "]")
outputC(
parity, lhs_strings,
os.path.join(Ccodesdir, "parity_conditions_symbolic_dot_products.h"))
# Step 2.a: Generate Ccodesdir/gridfunction_defines.h file,
# containing human-readable gridfunction aliases
evolved_variables_list, auxiliary_variables_list, auxevol_variables_list = gri.output__gridfunction_defines_h__return_gf_lists(
Ccodesdir)
# Step 2.b: set the parity conditions on all gridfunctions in gf_list,
# based on how many digits are at the end of their names
def set_parity_types(list_of_gf_names):
parity_type = []
for name in list_of_gf_names:
for gf in gri.glb_gridfcs_list:
if gf.name == name:
parity_type__orig_len = len(parity_type)
if gf.DIM < 3 or gf.DIM > 4:
print(
"Error: Cannot currently specify parity conditions on gridfunctions with DIM<3 or >4."
)
sys.exit(1)
if gf.rank == 0:
parity_type.append(0)
elif gf.rank == 1:
if gf.DIM == 3:
parity_type.append(
int(gf.name[-1]) + 1
) # = 1 for e.g., beta^0; = 2 for e.g., beta^1, etc.
elif gf.DIM == 4:
parity_type.append(
int(gf.name[-1])
) # = 0 for e.g., b4^0; = 1 for e.g., beta^1, etc.
elif gf.rank == 2:
if gf.DIM == 3:
# element of a list; a[-2] the
# second-to-last element, etc.
idx0 = gf.name[-2]
idx1 = gf.name[-1]
if idx0 == "0" and idx1 == "0":
parity_type.append(4)
elif (idx0 == "0"
and idx1 == "1") or (idx0 == "1"
and idx1 == "0"):
parity_type.append(5)
elif (idx0 == "0"
and idx1 == "2") or (idx0 == "2"
and idx1 == "0"):
parity_type.append(6)
elif idx0 == "1" and idx1 == "1":
parity_type.append(7)
elif (idx0 == "1"
and idx1 == "2") or (idx0 == "2"
and idx1 == "1"):
parity_type.append(8)
elif idx0 == "2" and idx1 == "2":
parity_type.append(9)
elif gf.DIM == 4:
idx0 = gf.name[-2]
idx1 = gf.name[-1]
# g4DD00 = g_{tt} : parity type = 0
# g4DD01 = g_{tx} : parity type = 1
# g4DD02 = g_{ty} : parity type = 2
# g4DD0a = g_{ta} : parity type = a
if idx0 == "0":
parity_type.append(int(idx1))
elif idx1 == "0":
parity_type.append(int(idx0))
if idx0 == "1" and idx1 == "1":
parity_type.append(4)
elif (idx0 == "1"
and idx1 == "2") or (idx0 == "2"
and idx1 == "1"):
parity_type.append(5)
elif (idx0 == "1"
and idx1 == "3") or (idx0 == "3"
and idx1 == "1"):
parity_type.append(6)
elif idx0 == "2" and idx1 == "2":
parity_type.append(7)
elif (idx0 == "2"
and idx1 == "3") or (idx0 == "3"
and idx1 == "2"):
parity_type.append(8)
elif idx0 == "3" and idx1 == "3":
parity_type.append(9)
if len(parity_type) == parity_type__orig_len:
print(
"Error: Could not figure out parity type for " +
gf.gftype + " gridfunction: " + gf.name, gf.DIM,
gf.name[-2], gf.name[-1], gf.rank)
sys.exit(1)
if len(parity_type) != len(list_of_gf_names):
print(
"Error: For some reason the length of the parity types list did not match the length of the gf list."
)
sys.exit(1)
return parity_type
evol_parity_type = set_parity_types(evolved_variables_list)
aux_parity_type = set_parity_types(auxiliary_variables_list)
auxevol_parity_type = set_parity_types(auxevol_variables_list)
# Step 2.c: Output all gridfunctions to Ccodesdir+"/gridfunction_defines.h"
# ... then append to the file the parity type for each gridfunction.
with open(os.path.join(Ccodesdir, "gridfunction_defines.h"), "a") as file:
file.write("\n\n/* PARITY TYPES FOR ALL GRIDFUNCTIONS.\n")
file.write(
" SEE \"Tutorial-Start_to_Finish-Curvilinear_BCs.ipynb\" FOR DEFINITIONS. */\n"
)
if len(evolved_variables_list) > 0:
file.write("const int8_t evol_gf_parity[" +
str(len(evolved_variables_list)) + "] = { ")
for i in range(len(evolved_variables_list) - 1):
file.write(str(evol_parity_type[i]) + ", ")
file.write(
str(evol_parity_type[len(evolved_variables_list) - 1]) +
" };\n")
if len(auxiliary_variables_list) > 0:
file.write("const int8_t aux_gf_parity[" +
str(len(auxiliary_variables_list)) + "] = { ")
for i in range(len(auxiliary_variables_list) - 1):
file.write(str(aux_parity_type[i]) + ", ")
file.write(
str(aux_parity_type[len(auxiliary_variables_list) - 1]) +
" };\n")
if len(auxevol_variables_list) > 0:
file.write("const int8_t auxevol_gf_parity[" +
str(len(auxevol_variables_list)) + "] = { ")
for i in range(len(auxevol_variables_list) - 1):
file.write(str(auxevol_parity_type[i]) + ", ")
file.write(
str(auxevol_parity_type[len(auxevol_variables_list) - 1]) +
" };\n")
if verbose == True:
import textwrap
wrapper = textwrap.TextWrapper(initial_indent="",
subsequent_indent=" ",
width=75)
def print_parity_list(gf_type, variable_names, parity_types):
outstr = ""
if len(variable_names) != 0:
outstr += gf_type + " parity: ( "
for i in range(len(variable_names)):
outstr += variable_names[i] + ":" + str(parity_types[i])
if i != len(variable_names) - 1:
outstr += ", "
outstr += " )"
print(wrapper.fill(outstr))
print_parity_list("Evolved", evolved_variables_list, evol_parity_type)
print_parity_list("Auxiliary", auxiliary_variables_list,
aux_parity_type)
print_parity_list("AuxEvol", auxevol_variables_list,
auxevol_parity_type)
# Step 3: Find the Eigen-Coordinate and set up the Eigen-Coordinate's reference metric:
CoordSystem_orig = par.parval_from_str("reference_metric::CoordSystem")
par.set_parval_from_str("reference_metric::CoordSystem",
rfm.get_EigenCoord())
rfm.reference_metric()
# Step 4: Output C code for the Eigen-Coordinate mapping from xx->Cartesian:
rfm.xxCart_h("EigenCoord_xxCart",
os.path.join(Cparamspath, "set_Cparameters.h"),
os.path.join(Ccodesdir, "EigenCoord_xxCart.h"))
# Step 5: Output the Eigen-Coordinate mapping from Cartesian->xx:
# Step 5.a: Sanity check: First make sure that rfm.Cart_to_xx has been set. Error out if not!
if rfm.Cart_to_xx[0] == 0 or rfm.Cart_to_xx[1] == 0 or rfm.Cart_to_xx[
2] == 0:
print(
"ERROR: rfm.Cart_to_xx[], which maps Cartesian -> xx, has not been set for"
)
print(" reference_metric::CoordSystem = " +
par.parval_from_str("reference_metric::CoordSystem"))
print(
" Boundary conditions in curvilinear coordinates REQUIRE this be set."
)
sys.exit(1)
# Step 5.b: Output C code for the Eigen-Coordinate mapping from Cartesian->xx:
outputC([rfm.Cart_to_xx[0], rfm.Cart_to_xx[1], rfm.Cart_to_xx[2]], [
"Cart_to_xx0_inbounds", "Cart_to_xx1_inbounds", "Cart_to_xx2_inbounds"
], os.path.join(Ccodesdir, "EigenCoord_Cart_to_xx.h"))
# Step 6: Restore reference_metric::CoordSystem back to the original CoordSystem
par.set_parval_from_str("reference_metric::CoordSystem", CoordSystem_orig)
rfm.reference_metric()
def GiRaFFE_NRPy_Main_Driver_generate_all(out_dir):
cmd.mkdir(out_dir)
gammaDD = ixp.register_gridfunctions_for_single_rank2("AUXEVOL","gammaDD","sym01",DIM=3)
betaU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL","betaU",DIM=3)
alpha = gri.register_gridfunctions("AUXEVOL","alpha")
AD = ixp.register_gridfunctions_for_single_rank1("EVOL","AD")
BU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL","BU")
ValenciavU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL","ValenciavU")
psi6Phi = gri.register_gridfunctions("EVOL","psi6Phi")
StildeD = ixp.register_gridfunctions_for_single_rank1("EVOL","StildeD")
PhievolParenU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL","PhievolParenU",DIM=3)
AevolParen = gri.register_gridfunctions("AUXEVOL","AevolParen")
GRHD.compute_sqrtgammaDET(gammaDD)
GRFFE.compute_AD_source_term_parenthetical_for_FD(GRHD.sqrtgammaDET,betaU,alpha,psi6Phi,AD)
GRFFE.compute_psi6Phi_rhs_parenthetical(gammaDD,GRHD.sqrtgammaDET,betaU,alpha,AD,psi6Phi)
parens_to_print = [\
lhrh(lhs=gri.gfaccess("auxevol_gfs","AevolParen"),rhs=GRFFE.AevolParen),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","PhievolParenU0"),rhs=GRFFE.PhievolParenU[0]),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","PhievolParenU1"),rhs=GRFFE.PhievolParenU[1]),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","PhievolParenU2"),rhs=GRFFE.PhievolParenU[2]),\
]
subdir = "RHSs"
cmd.mkdir(os.path.join(out_dir, subdir))
desc = "Calculate quantities to be finite-differenced for the GRFFE RHSs"
name = "calculate_parentheticals_for_RHSs"
outCfunction(
outfile = os.path.join(out_dir,subdir,name+".h"), desc=desc, name=name,
params ="const paramstruct *restrict params,const REAL *restrict in_gfs,REAL *restrict auxevol_gfs",
body = fin.FD_outputC("returnstring",parens_to_print,params="outCverbose=False").replace("IDX4","IDX4S"),
loopopts ="AllPoints",
rel_path_for_Cparams=os.path.join("../"))
xi_damping = par.Cparameters("REAL",thismodule,"xi_damping",0.1)
GRFFE.compute_psi6Phi_rhs_damping_term(alpha,psi6Phi,xi_damping)
AevolParen_dD = ixp.declarerank1("AevolParen_dD",DIM=3)
PhievolParenU_dD = ixp.declarerank2("PhievolParenU_dD","nosym",DIM=3)
A_rhsD = ixp.zerorank1()
psi6Phi_rhs = GRFFE.psi6Phi_damping
for i in range(3):
A_rhsD[i] += -AevolParen_dD[i]
psi6Phi_rhs += -PhievolParenU_dD[i][i]
# Add Kreiss-Oliger dissipation to the GRFFE RHSs:
psi6Phi_dKOD = ixp.declarerank1("psi6Phi_dKOD")
AD_dKOD = ixp.declarerank2("AD_dKOD","nosym")
for i in range(3):
psi6Phi_rhs += diss_strength*psi6Phi_dKOD[i]*rfm.ReU[i] # ReU[i] = 1/scalefactor_orthog_funcform[i]
for j in range(3):
A_rhsD[j] += diss_strength*AD_dKOD[j][i]*rfm.ReU[i] # ReU[i] = 1/scalefactor_orthog_funcform[i]
RHSs_to_print = [\
lhrh(lhs=gri.gfaccess("rhs_gfs","AD0"),rhs=A_rhsD[0]),\
lhrh(lhs=gri.gfaccess("rhs_gfs","AD1"),rhs=A_rhsD[1]),\
lhrh(lhs=gri.gfaccess("rhs_gfs","AD2"),rhs=A_rhsD[2]),\
lhrh(lhs=gri.gfaccess("rhs_gfs","psi6Phi"),rhs=psi6Phi_rhs),\
]
desc = "Calculate AD gauge term and psi6Phi RHSs"
name = "calculate_AD_gauge_psi6Phi_RHSs"
source_Ccode = outCfunction(
outfile = "returnstring", desc=desc, name=name,
params ="const paramstruct *params,const REAL *in_gfs,const REAL *auxevol_gfs,REAL *rhs_gfs",
body = fin.FD_outputC("returnstring",RHSs_to_print,params="outCverbose=False").replace("IDX4","IDX4S"),
loopopts ="InteriorPoints",
rel_path_for_Cparams=os.path.join("../")).replace("=NGHOSTS","=NGHOSTS_A2B").replace("NGHOSTS+Nxx0","Nxx_plus_2NGHOSTS0-NGHOSTS_A2B").replace("NGHOSTS+Nxx1","Nxx_plus_2NGHOSTS1-NGHOSTS_A2B").replace("NGHOSTS+Nxx2","Nxx_plus_2NGHOSTS2-NGHOSTS_A2B")
# Note the above .replace() functions. These serve to expand the loop range into the ghostzones, since
# the second-order FD needs fewer than some other algorithms we use do.
with open(os.path.join(out_dir,subdir,name+".h"),"w") as file:
file.write(source_Ccode)
# Declare all the Cparameters we will need
metricderivDDD = ixp.declarerank3("metricderivDDD","sym01",DIM=3)
shiftderivUD = ixp.declarerank2("shiftderivUD","nosym",DIM=3)
lapsederivD = ixp.declarerank1("lapsederivD",DIM=3)
general_access = """const REAL gammaDD00 = auxevol_gfs[IDX4S(GAMMADD00GF,i0,i1,i2)];
const REAL gammaDD01 = auxevol_gfs[IDX4S(GAMMADD01GF,i0,i1,i2)];
const REAL gammaDD02 = auxevol_gfs[IDX4S(GAMMADD02GF,i0,i1,i2)];
const REAL gammaDD11 = auxevol_gfs[IDX4S(GAMMADD11GF,i0,i1,i2)];
const REAL gammaDD12 = auxevol_gfs[IDX4S(GAMMADD12GF,i0,i1,i2)];
const REAL gammaDD22 = auxevol_gfs[IDX4S(GAMMADD22GF,i0,i1,i2)];
const REAL betaU0 = auxevol_gfs[IDX4S(BETAU0GF,i0,i1,i2)];
const REAL betaU1 = auxevol_gfs[IDX4S(BETAU1GF,i0,i1,i2)];
const REAL betaU2 = auxevol_gfs[IDX4S(BETAU2GF,i0,i1,i2)];
const REAL alpha = auxevol_gfs[IDX4S(ALPHAGF,i0,i1,i2)];
const REAL ValenciavU0 = auxevol_gfs[IDX4S(VALENCIAVU0GF,i0,i1,i2)];
const REAL ValenciavU1 = auxevol_gfs[IDX4S(VALENCIAVU1GF,i0,i1,i2)];
const REAL ValenciavU2 = auxevol_gfs[IDX4S(VALENCIAVU2GF,i0,i1,i2)];
const REAL BU0 = auxevol_gfs[IDX4S(BU0GF,i0,i1,i2)];
const REAL BU1 = auxevol_gfs[IDX4S(BU1GF,i0,i1,i2)];
const REAL BU2 = auxevol_gfs[IDX4S(BU2GF,i0,i1,i2)];
"""
metric_deriv_access = ixp.zerorank1(DIM=3)
metric_deriv_access[0] = """const REAL metricderivDDD000 = (auxevol_gfs[IDX4S(GAMMA_FACEDD00GF,i0+1,i1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD00GF,i0,i1,i2)])/dxx0;
const REAL metricderivDDD010 = (auxevol_gfs[IDX4S(GAMMA_FACEDD01GF,i0+1,i1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD01GF,i0,i1,i2)])/dxx0;
const REAL metricderivDDD020 = (auxevol_gfs[IDX4S(GAMMA_FACEDD02GF,i0+1,i1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD02GF,i0,i1,i2)])/dxx0;
const REAL metricderivDDD110 = (auxevol_gfs[IDX4S(GAMMA_FACEDD11GF,i0+1,i1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD11GF,i0,i1,i2)])/dxx0;
const REAL metricderivDDD120 = (auxevol_gfs[IDX4S(GAMMA_FACEDD12GF,i0+1,i1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD12GF,i0,i1,i2)])/dxx0;
const REAL metricderivDDD220 = (auxevol_gfs[IDX4S(GAMMA_FACEDD22GF,i0+1,i1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD22GF,i0,i1,i2)])/dxx0;
const REAL shiftderivUD00 = (auxevol_gfs[IDX4S(BETA_FACEU0GF,i0+1,i1,i2)]-auxevol_gfs[IDX4S(BETA_FACEU0GF,i0,i1,i2)])/dxx0;
const REAL shiftderivUD10 = (auxevol_gfs[IDX4S(BETA_FACEU1GF,i0+1,i1,i2)]-auxevol_gfs[IDX4S(BETA_FACEU1GF,i0,i1,i2)])/dxx0;
const REAL shiftderivUD20 = (auxevol_gfs[IDX4S(BETA_FACEU2GF,i0+1,i1,i2)]-auxevol_gfs[IDX4S(BETA_FACEU2GF,i0,i1,i2)])/dxx0;
const REAL lapsederivD0 = (auxevol_gfs[IDX4S(ALPHA_FACEGF,i0+1,i1,i2)]-auxevol_gfs[IDX4S(ALPHA_FACEGF,i0,i1,i2)])/dxx0;
REAL Stilde_rhsD0;
"""
metric_deriv_access[1] = """const REAL metricderivDDD001 = (auxevol_gfs[IDX4S(GAMMA_FACEDD00GF,i0,i1+1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD00GF,i0,i1,i2)])/dxx1;
const REAL metricderivDDD011 = (auxevol_gfs[IDX4S(GAMMA_FACEDD01GF,i0,i1+1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD01GF,i0,i1,i2)])/dxx1;
const REAL metricderivDDD021 = (auxevol_gfs[IDX4S(GAMMA_FACEDD02GF,i0,i1+1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD02GF,i0,i1,i2)])/dxx1;
const REAL metricderivDDD111 = (auxevol_gfs[IDX4S(GAMMA_FACEDD11GF,i0,i1+1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD11GF,i0,i1,i2)])/dxx1;
const REAL metricderivDDD121 = (auxevol_gfs[IDX4S(GAMMA_FACEDD12GF,i0,i1+1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD12GF,i0,i1,i2)])/dxx1;
const REAL metricderivDDD221 = (auxevol_gfs[IDX4S(GAMMA_FACEDD22GF,i0,i1+1,i2)]-auxevol_gfs[IDX4S(GAMMA_FACEDD22GF,i0,i1,i2)])/dxx1;
const REAL shiftderivUD01 = (auxevol_gfs[IDX4S(BETA_FACEU0GF,i0,i1+1,i2)]-auxevol_gfs[IDX4S(BETA_FACEU0GF,i0,i1,i2)])/dxx1;
const REAL shiftderivUD11 = (auxevol_gfs[IDX4S(BETA_FACEU1GF,i0,i1+1,i2)]-auxevol_gfs[IDX4S(BETA_FACEU1GF,i0,i1,i2)])/dxx1;
const REAL shiftderivUD21 = (auxevol_gfs[IDX4S(BETA_FACEU2GF,i0,i1+1,i2)]-auxevol_gfs[IDX4S(BETA_FACEU2GF,i0,i1,i2)])/dxx1;
const REAL lapsederivD1 = (auxevol_gfs[IDX4S(ALPHA_FACEGF,i0,i1+1,i2)]-auxevol_gfs[IDX4S(ALPHA_FACEGF,i0,i1,i2)])/dxx1;
REAL Stilde_rhsD1;
"""
metric_deriv_access[2] = """const REAL metricderivDDD002 = (auxevol_gfs[IDX4S(GAMMA_FACEDD00GF,i0,i1,i2+1)]-auxevol_gfs[IDX4S(GAMMA_FACEDD00GF,i0,i1,i2)])/dxx2;
const REAL metricderivDDD012 = (auxevol_gfs[IDX4S(GAMMA_FACEDD01GF,i0,i1,i2+1)]-auxevol_gfs[IDX4S(GAMMA_FACEDD01GF,i0,i1,i2)])/dxx2;
const REAL metricderivDDD022 = (auxevol_gfs[IDX4S(GAMMA_FACEDD02GF,i0,i1,i2+1)]-auxevol_gfs[IDX4S(GAMMA_FACEDD02GF,i0,i1,i2)])/dxx2;
const REAL metricderivDDD112 = (auxevol_gfs[IDX4S(GAMMA_FACEDD11GF,i0,i1,i2+1)]-auxevol_gfs[IDX4S(GAMMA_FACEDD11GF,i0,i1,i2)])/dxx2;
const REAL metricderivDDD122 = (auxevol_gfs[IDX4S(GAMMA_FACEDD12GF,i0,i1,i2+1)]-auxevol_gfs[IDX4S(GAMMA_FACEDD12GF,i0,i1,i2)])/dxx2;
const REAL metricderivDDD222 = (auxevol_gfs[IDX4S(GAMMA_FACEDD22GF,i0,i1,i2+1)]-auxevol_gfs[IDX4S(GAMMA_FACEDD22GF,i0,i1,i2)])/dxx2;
const REAL shiftderivUD02 = (auxevol_gfs[IDX4S(BETA_FACEU0GF,i0,i1,i2+1)]-auxevol_gfs[IDX4S(BETA_FACEU0GF,i0,i1,i2)])/dxx2;
const REAL shiftderivUD12 = (auxevol_gfs[IDX4S(BETA_FACEU1GF,i0,i1,i2+1)]-auxevol_gfs[IDX4S(BETA_FACEU1GF,i0,i1,i2)])/dxx2;
const REAL shiftderivUD22 = (auxevol_gfs[IDX4S(BETA_FACEU2GF,i0,i1,i2+1)]-auxevol_gfs[IDX4S(BETA_FACEU2GF,i0,i1,i2)])/dxx2;
const REAL lapsederivD2 = (auxevol_gfs[IDX4S(ALPHA_FACEGF,i0,i1,i2+1)]-auxevol_gfs[IDX4S(ALPHA_FACEGF,i0,i1,i2)])/dxx2;
REAL Stilde_rhsD2;
"""
write_final_quantity = ixp.zerorank1(DIM=3)
write_final_quantity[0] = """rhs_gfs[IDX4S(STILDED0GF,i0,i1,i2)] += Stilde_rhsD0;
"""
write_final_quantity[1] = """rhs_gfs[IDX4S(STILDED1GF,i0,i1,i2)] += Stilde_rhsD1;
"""
write_final_quantity[2] = """rhs_gfs[IDX4S(STILDED2GF,i0,i1,i2)] += Stilde_rhsD2;
"""
# Declare this symbol:
sqrt4pi = par.Cparameters("REAL",thismodule,"sqrt4pi","sqrt(4.0*M_PI)")
# We need to rerun a few of these functions with the reset lists to make sure these functions
# don't cheat by using analytic expressions
GRHD.u4U_in_terms_of_ValenciavU__rescale_ValenciavU_by_applying_speed_limit(alpha, betaU, gammaDD, ValenciavU)
GRFFE.compute_smallb4U(gammaDD, betaU, alpha, GRHD.u4U_ito_ValenciavU, BU, sqrt4pi)
GRFFE.compute_smallbsquared(gammaDD, betaU, alpha, GRFFE.smallb4U)
GRFFE.compute_TEM4UU(gammaDD,betaU,alpha, GRFFE.smallb4U, GRFFE.smallbsquared,GRHD.u4U_ito_ValenciavU)
GRHD.compute_g4DD_zerotimederiv_dD(gammaDD,betaU,alpha, metricderivDDD,shiftderivUD,lapsederivD)
GRHD.compute_S_tilde_source_termD(alpha, GRHD.sqrtgammaDET,GRHD.g4DD_zerotimederiv_dD, GRFFE.TEM4UU)
for i in range(3):
desc = "Adds the source term to StildeD"+str(i)+"."
name = "calculate_StildeD"+str(i)+"_source_term"
outCfunction(
outfile = os.path.join(out_dir,subdir,name+".h"), desc=desc, name=name,
params ="const paramstruct *params,const REAL *auxevol_gfs, REAL *rhs_gfs",
body = general_access \
+metric_deriv_access[i]\
+outputC(GRHD.S_tilde_source_termD[i],"Stilde_rhsD"+str(i),"returnstring",params="outCverbose=False").replace("IDX4","IDX4S")\
+write_final_quantity[i],
loopopts ="InteriorPoints",
rel_path_for_Cparams=os.path.join("../"))
subdir = "FCVAL"
cmd.mkdir(os.path.join(out_dir, subdir))
FCVAL.GiRaFFE_NRPy_FCVAL(os.path.join(out_dir,subdir))
subdir = "PPM"
cmd.mkdir(os.path.join(out_dir, subdir))
PPM.GiRaFFE_NRPy_PPM(os.path.join(out_dir,subdir))
# We will pass values of the gridfunction on the cell faces into the function. This requires us
# to declare them as C parameters in NRPy+. We will denote this with the _face infix/suffix.
alpha_face = gri.register_gridfunctions("AUXEVOL","alpha_face")
gamma_faceDD = ixp.register_gridfunctions_for_single_rank2("AUXEVOL","gamma_faceDD","sym01")
beta_faceU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL","beta_faceU")
# We'll need some more gridfunctions, now, to represent the reconstructions of BU and ValenciavU
# on the right and left faces
Valenciav_rU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL","Valenciav_rU",DIM=3)
B_rU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL","B_rU",DIM=3)
Valenciav_lU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL","Valenciav_lU",DIM=3)
B_lU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL","B_lU",DIM=3)
Memory_Read = """const double alpha_face = auxevol_gfs[IDX4S(ALPHA_FACEGF, i0,i1,i2)];
const double gamma_faceDD00 = auxevol_gfs[IDX4S(GAMMA_FACEDD00GF, i0,i1,i2)];
const double gamma_faceDD01 = auxevol_gfs[IDX4S(GAMMA_FACEDD01GF, i0,i1,i2)];
const double gamma_faceDD02 = auxevol_gfs[IDX4S(GAMMA_FACEDD02GF, i0,i1,i2)];
const double gamma_faceDD11 = auxevol_gfs[IDX4S(GAMMA_FACEDD11GF, i0,i1,i2)];
const double gamma_faceDD12 = auxevol_gfs[IDX4S(GAMMA_FACEDD12GF, i0,i1,i2)];
const double gamma_faceDD22 = auxevol_gfs[IDX4S(GAMMA_FACEDD22GF, i0,i1,i2)];
const double beta_faceU0 = auxevol_gfs[IDX4S(BETA_FACEU0GF, i0,i1,i2)];
const double beta_faceU1 = auxevol_gfs[IDX4S(BETA_FACEU1GF, i0,i1,i2)];
const double beta_faceU2 = auxevol_gfs[IDX4S(BETA_FACEU2GF, i0,i1,i2)];
const double Valenciav_rU0 = auxevol_gfs[IDX4S(VALENCIAV_RU0GF, i0,i1,i2)];
const double Valenciav_rU1 = auxevol_gfs[IDX4S(VALENCIAV_RU1GF, i0,i1,i2)];
const double Valenciav_rU2 = auxevol_gfs[IDX4S(VALENCIAV_RU2GF, i0,i1,i2)];
const double B_rU0 = auxevol_gfs[IDX4S(B_RU0GF, i0,i1,i2)];
const double B_rU1 = auxevol_gfs[IDX4S(B_RU1GF, i0,i1,i2)];
const double B_rU2 = auxevol_gfs[IDX4S(B_RU2GF, i0,i1,i2)];
const double Valenciav_lU0 = auxevol_gfs[IDX4S(VALENCIAV_LU0GF, i0,i1,i2)];
const double Valenciav_lU1 = auxevol_gfs[IDX4S(VALENCIAV_LU1GF, i0,i1,i2)];
const double Valenciav_lU2 = auxevol_gfs[IDX4S(VALENCIAV_LU2GF, i0,i1,i2)];
const double B_lU0 = auxevol_gfs[IDX4S(B_LU0GF, i0,i1,i2)];
const double B_lU1 = auxevol_gfs[IDX4S(B_LU1GF, i0,i1,i2)];
const double B_lU2 = auxevol_gfs[IDX4S(B_LU2GF, i0,i1,i2)];
REAL A_rhsD0 = 0; REAL A_rhsD1 = 0; REAL A_rhsD2 = 0;
"""
Memory_Write = """rhs_gfs[IDX4S(AD0GF,i0,i1,i2)] += A_rhsD0;
rhs_gfs[IDX4S(AD1GF,i0,i1,i2)] += A_rhsD1;
rhs_gfs[IDX4S(AD2GF,i0,i1,i2)] += A_rhsD2;
"""
indices = ["i0","i1","i2"]
indicesp1 = ["i0+1","i1+1","i2+1"]
subdir = "RHSs"
for flux_dirn in range(3):
Af.calculate_E_i_flux(flux_dirn,True,alpha_face,gamma_faceDD,beta_faceU,\
Valenciav_rU,B_rU,Valenciav_lU,B_lU)
E_field_to_print = [\
sp.Rational(1,4)*Af.E_fluxD[(flux_dirn+1)%3],
sp.Rational(1,4)*Af.E_fluxD[(flux_dirn+2)%3],
]
E_field_names = [\
"A_rhsD"+str((flux_dirn+1)%3),
"A_rhsD"+str((flux_dirn+2)%3),
]
desc = "Calculate the electric flux on the left face in direction " + str(flux_dirn) + "."
name = "calculate_E_field_D" + str(flux_dirn) + "_right"
outCfunction(
outfile = os.path.join(out_dir,subdir,name+".h"), desc=desc, name=name,
params ="const paramstruct *params,const REAL *auxevol_gfs,REAL *rhs_gfs",
body = Memory_Read \
+outputC(E_field_to_print,E_field_names,"returnstring",params="outCverbose=False").replace("IDX4","IDX4S")\
+Memory_Write,
loopopts ="InteriorPoints",
rel_path_for_Cparams=os.path.join("../"))
desc = "Calculate the electric flux on the left face in direction " + str(flux_dirn) + "."
name = "calculate_E_field_D" + str(flux_dirn) + "_left"
outCfunction(
outfile = os.path.join(out_dir,subdir,name+".h"), desc=desc, name=name,
params ="const paramstruct *params,const REAL *auxevol_gfs,REAL *rhs_gfs",
body = Memory_Read.replace(indices[flux_dirn],indicesp1[flux_dirn]) \
+outputC(E_field_to_print,E_field_names,"returnstring",params="outCverbose=False").replace("IDX4","IDX4S")\
+Memory_Write,
loopopts ="InteriorPoints",
rel_path_for_Cparams=os.path.join("../"))
Memory_Read = """const double alpha_face = auxevol_gfs[IDX4S(ALPHA_FACEGF, i0,i1,i2)];
const double gamma_faceDD00 = auxevol_gfs[IDX4S(GAMMA_FACEDD00GF, i0,i1,i2)];
const double gamma_faceDD01 = auxevol_gfs[IDX4S(GAMMA_FACEDD01GF, i0,i1,i2)];
const double gamma_faceDD02 = auxevol_gfs[IDX4S(GAMMA_FACEDD02GF, i0,i1,i2)];
const double gamma_faceDD11 = auxevol_gfs[IDX4S(GAMMA_FACEDD11GF, i0,i1,i2)];
const double gamma_faceDD12 = auxevol_gfs[IDX4S(GAMMA_FACEDD12GF, i0,i1,i2)];
const double gamma_faceDD22 = auxevol_gfs[IDX4S(GAMMA_FACEDD22GF, i0,i1,i2)];
const double beta_faceU0 = auxevol_gfs[IDX4S(BETA_FACEU0GF, i0,i1,i2)];
const double beta_faceU1 = auxevol_gfs[IDX4S(BETA_FACEU1GF, i0,i1,i2)];
const double beta_faceU2 = auxevol_gfs[IDX4S(BETA_FACEU2GF, i0,i1,i2)];
const double Valenciav_rU0 = auxevol_gfs[IDX4S(VALENCIAV_RU0GF, i0,i1,i2)];
const double Valenciav_rU1 = auxevol_gfs[IDX4S(VALENCIAV_RU1GF, i0,i1,i2)];
const double Valenciav_rU2 = auxevol_gfs[IDX4S(VALENCIAV_RU2GF, i0,i1,i2)];
const double B_rU0 = auxevol_gfs[IDX4S(B_RU0GF, i0,i1,i2)];
const double B_rU1 = auxevol_gfs[IDX4S(B_RU1GF, i0,i1,i2)];
const double B_rU2 = auxevol_gfs[IDX4S(B_RU2GF, i0,i1,i2)];
const double Valenciav_lU0 = auxevol_gfs[IDX4S(VALENCIAV_LU0GF, i0,i1,i2)];
const double Valenciav_lU1 = auxevol_gfs[IDX4S(VALENCIAV_LU1GF, i0,i1,i2)];
const double Valenciav_lU2 = auxevol_gfs[IDX4S(VALENCIAV_LU2GF, i0,i1,i2)];
const double B_lU0 = auxevol_gfs[IDX4S(B_LU0GF, i0,i1,i2)];
const double B_lU1 = auxevol_gfs[IDX4S(B_LU1GF, i0,i1,i2)];
const double B_lU2 = auxevol_gfs[IDX4S(B_LU2GF, i0,i1,i2)];
REAL Stilde_fluxD0 = 0; REAL Stilde_fluxD1 = 0; REAL Stilde_fluxD2 = 0;
"""
Memory_Write = """rhs_gfs[IDX4S(STILDED0GF, i0, i1, i2)] += invdx0*Stilde_fluxD0;
rhs_gfs[IDX4S(STILDED1GF, i0, i1, i2)] += invdx0*Stilde_fluxD1;
rhs_gfs[IDX4S(STILDED2GF, i0, i1, i2)] += invdx0*Stilde_fluxD2;
"""
indices = ["i0","i1","i2"]
indicesp1 = ["i0+1","i1+1","i2+1"]
assignment = "+="
assignmentp1 = "-="
invdx = ["invdx0","invdx1","invdx2"]
for flux_dirn in range(3):
Sf.calculate_Stilde_flux(flux_dirn,True,alpha_face,gamma_faceDD,beta_faceU,\
Valenciav_rU,B_rU,Valenciav_lU,B_lU,sqrt4pi)
Stilde_flux_to_print = [\
Sf.Stilde_fluxD[0],\
Sf.Stilde_fluxD[1],\
Sf.Stilde_fluxD[2],\
]
Stilde_flux_names = [\
"Stilde_fluxD0",\
"Stilde_fluxD1",\
"Stilde_fluxD2",\
]
desc = "Compute the flux of all 3 components of tilde{S}_i on the right face in the " + str(flux_dirn) + "."
name = "calculate_Stilde_flux_D" + str(flux_dirn) + "_right"
outCfunction(
outfile = os.path.join(out_dir,subdir,name+".h"), desc=desc, name=name,
params ="const paramstruct *params,const REAL *auxevol_gfs,REAL *rhs_gfs",
body = Memory_Read \
+outputC(Stilde_flux_to_print,Stilde_flux_names,"returnstring",params="outCverbose=False").replace("IDX4","IDX4S")\
+Memory_Write.replace(invdx[0],invdx[flux_dirn]),
loopopts ="InteriorPoints",
rel_path_for_Cparams=os.path.join("../"))
desc = "Compute the flux of all 3 components of tilde{S}_i on the left face in the " + str(flux_dirn) + "."
name = "calculate_Stilde_flux_D" + str(flux_dirn) + "_left"
outCfunction(
outfile = os.path.join(out_dir,subdir,name+".h"), desc=desc, name=name,
params ="const paramstruct *params,const REAL *auxevol_gfs,REAL *rhs_gfs",
body = Memory_Read.replace(indices[flux_dirn],indicesp1[flux_dirn]) \
+outputC(Stilde_flux_to_print,Stilde_flux_names,"returnstring",params="outCverbose=False").replace("IDX4","IDX4S")\
+Memory_Write.replace(invdx[0],invdx[flux_dirn]).replace(assignment,assignmentp1),
loopopts ="InteriorPoints",
rel_path_for_Cparams=os.path.join("../"))
subdir = "boundary_conditions"
cmd.mkdir(os.path.join(out_dir,subdir))
BC.GiRaFFE_NRPy_BCs(os.path.join(out_dir,subdir))
subdir = "A2B"
cmd.mkdir(os.path.join(out_dir,subdir))
A2B.GiRaFFE_NRPy_A2B(os.path.join(out_dir,subdir),gammaDD,AD,BU)
C2P_P2C.GiRaFFE_NRPy_C2P(StildeD,BU,gammaDD,betaU,alpha)
values_to_print = [\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD0"),rhs=C2P_P2C.outStildeD[0]),\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD1"),rhs=C2P_P2C.outStildeD[1]),\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD2"),rhs=C2P_P2C.outStildeD[2]),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","ValenciavU0"),rhs=C2P_P2C.ValenciavU[0]),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","ValenciavU1"),rhs=C2P_P2C.ValenciavU[1]),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","ValenciavU2"),rhs=C2P_P2C.ValenciavU[2])\
]
subdir = "C2P"
cmd.mkdir(os.path.join(out_dir,subdir))
desc = "Apply fixes to \tilde{S}_i and recompute the velocity to match with current sheet prescription."
name = "GiRaFFE_NRPy_cons_to_prims"
outCfunction(
outfile = os.path.join(out_dir,subdir,name+".h"), desc=desc, name=name,
params ="const paramstruct *params,REAL *xx[3],REAL *auxevol_gfs,REAL *in_gfs",
body = fin.FD_outputC("returnstring",values_to_print,params="outCverbose=False").replace("IDX4","IDX4S"),
loopopts ="AllPoints,Read_xxs",
rel_path_for_Cparams=os.path.join("../"))
C2P_P2C.GiRaFFE_NRPy_P2C(gammaDD,betaU,alpha, ValenciavU,BU, sqrt4pi)
values_to_print = [\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD0"),rhs=C2P_P2C.StildeD[0]),\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD1"),rhs=C2P_P2C.StildeD[1]),\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD2"),rhs=C2P_P2C.StildeD[2]),\
]
desc = "Recompute StildeD after current sheet fix to Valencia 3-velocity to ensure consistency between conservative & primitive variables."
name = "GiRaFFE_NRPy_prims_to_cons"
outCfunction(
outfile = os.path.join(out_dir,subdir,name+".h"), desc=desc, name=name,
params ="const paramstruct *params,REAL *auxevol_gfs,REAL *in_gfs",
body = fin.FD_outputC("returnstring",values_to_print,params="outCverbose=False").replace("IDX4","IDX4S"),
loopopts ="AllPoints",
rel_path_for_Cparams=os.path.join("../"))
# Write out the main driver itself:
with open(os.path.join(out_dir,"GiRaFFE_NRPy_Main_Driver.h"),"w") as file:
file.write("""// Structure to track ghostzones for PPM:
typedef struct __gf_and_gz_struct__ {
REAL *gf;
int gz_lo[4],gz_hi[4];
} gf_and_gz_struct;
// Some additional constants needed for PPM:
const int VX=0,VY=1,VZ=2,BX=3,BY=4,BZ=5;
const int NUM_RECONSTRUCT_GFS = 6;
// Include ALL functions needed for evolution
#include "RHSs/calculate_parentheticals_for_RHSs.h"
#include "RHSs/calculate_AD_gauge_psi6Phi_RHSs.h"
#include "PPM/reconstruct_set_of_prims_PPM_GRFFE_NRPy.c"
#include "FCVAL/interpolate_metric_gfs_to_cell_faces.h"
#include "RHSs/calculate_StildeD0_source_term.h"
#include "RHSs/calculate_StildeD1_source_term.h"
#include "RHSs/calculate_StildeD2_source_term.h"
// #include "RHSs/calculate_E_field_D0_right.h"
// #include "RHSs/calculate_E_field_D0_left.h"
// #include "RHSs/calculate_E_field_D1_right.h"
// #include "RHSs/calculate_E_field_D1_left.h"
// #include "RHSs/calculate_E_field_D2_right.h"
// #include "RHSs/calculate_E_field_D2_left.h"
#include "../calculate_E_field_flat_all_in_one.h"
#include "RHSs/calculate_Stilde_flux_D0_right.h"
#include "RHSs/calculate_Stilde_flux_D0_left.h"
#include "RHSs/calculate_Stilde_flux_D1_right.h"
#include "RHSs/calculate_Stilde_flux_D1_left.h"
#include "RHSs/calculate_Stilde_flux_D2_right.h"
#include "RHSs/calculate_Stilde_flux_D2_left.h"
#include "boundary_conditions/GiRaFFE_boundary_conditions.h"
#include "A2B/driver_AtoB.h"
#include "C2P/GiRaFFE_NRPy_cons_to_prims.h"
#include "C2P/GiRaFFE_NRPy_prims_to_cons.h"
void override_BU_with_old_GiRaFFE(const paramstruct *restrict params,REAL *restrict auxevol_gfs,const int n) {
#include "set_Cparameters.h"
char filename[100];
sprintf(filename,"BU0_override-%08d.bin",n);
FILE *out2D = fopen(filename, "rb");
fread(auxevol_gfs+BU0GF*Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,
sizeof(double),Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,out2D);
fclose(out2D);
sprintf(filename,"BU1_override-%08d.bin",n);
out2D = fopen(filename, "rb");
fread(auxevol_gfs+BU1GF*Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,
sizeof(double),Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,out2D);
fclose(out2D);
sprintf(filename,"BU2_override-%08d.bin",n);
out2D = fopen(filename, "rb");
fread(auxevol_gfs+BU2GF*Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,
sizeof(double),Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,out2D);
fclose(out2D);
}
void GiRaFFE_NRPy_RHSs(const paramstruct *restrict params,REAL *restrict auxevol_gfs,const REAL *restrict in_gfs,REAL *restrict rhs_gfs) {
#include "set_Cparameters.h"
// First thing's first: initialize the RHSs to zero!
#pragma omp parallel for
for(int ii=0;ii<Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2*NUM_EVOL_GFS;ii++) {
rhs_gfs[ii] = 0.0;
}
// Next calculate the easier source terms that don't require flux directions
// This will also reset the RHSs for each gf at each new timestep.
calculate_parentheticals_for_RHSs(params,in_gfs,auxevol_gfs);
calculate_AD_gauge_psi6Phi_RHSs(params,in_gfs,auxevol_gfs,rhs_gfs);
// Now, we set up a bunch of structs of pointers to properly guide the PPM algorithm.
// They also count the number of ghostzones available.
gf_and_gz_struct in_prims[NUM_RECONSTRUCT_GFS], out_prims_r[NUM_RECONSTRUCT_GFS], out_prims_l[NUM_RECONSTRUCT_GFS];
int which_prims_to_reconstruct[NUM_RECONSTRUCT_GFS],num_prims_to_reconstruct;
const int Nxxp2NG012 = Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2;
REAL *temporary = auxevol_gfs + Nxxp2NG012*AEVOLPARENGF; //We're not using this anymore
// This sets pointers to the portion of auxevol_gfs containing the relevant gridfunction.
int ww=0;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAVU0GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU0GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU0GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAVU1GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU1GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU1GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAVU2GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU2GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU2GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BU0GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*B_RU0GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*B_LU0GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BU1GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*B_RU1GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*B_LU1GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BU2GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*B_RU2GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*B_LU2GF;
ww++;
// Prims are defined AT ALL GRIDPOINTS, so we set the # of ghostzones to zero:
for(int i=0;i<NUM_RECONSTRUCT_GFS;i++) for(int j=1;j<=3;j++) { in_prims[i].gz_lo[j]=0; in_prims[i].gz_hi[j]=0; }
// Left/right variables are not yet defined, yet we set the # of gz's to zero by default:
for(int i=0;i<NUM_RECONSTRUCT_GFS;i++) for(int j=1;j<=3;j++) { out_prims_r[i].gz_lo[j]=0; out_prims_r[i].gz_hi[j]=0; }
for(int i=0;i<NUM_RECONSTRUCT_GFS;i++) for(int j=1;j<=3;j++) { out_prims_l[i].gz_lo[j]=0; out_prims_l[i].gz_hi[j]=0; }
ww=0;
which_prims_to_reconstruct[ww]=VX; ww++;
which_prims_to_reconstruct[ww]=VY; ww++;
which_prims_to_reconstruct[ww]=VZ; ww++;
which_prims_to_reconstruct[ww]=BX; ww++;
which_prims_to_reconstruct[ww]=BY; ww++;
which_prims_to_reconstruct[ww]=BZ; ww++;
num_prims_to_reconstruct=ww;
// In each direction, perform the PPM reconstruction procedure.
// Then, add the fluxes to the RHS as appropriate.
for(int flux_dirn=0;flux_dirn<3;flux_dirn++) {
// In each direction, interpolate the metric gfs (gamma,beta,alpha) to cell faces.
interpolate_metric_gfs_to_cell_faces(params,auxevol_gfs,flux_dirn+1);
// Then, reconstruct the primitive variables on the cell faces.
// This function is housed in the file: "reconstruct_set_of_prims_PPM_GRFFE_NRPy.c"
reconstruct_set_of_prims_PPM_GRFFE_NRPy(params, auxevol_gfs, flux_dirn+1, num_prims_to_reconstruct,
which_prims_to_reconstruct, in_prims, out_prims_r, out_prims_l, temporary);
// For example, if flux_dirn==0, then at gamma_faceDD00(i,j,k) represents gamma_{xx}
// at (i-1/2,j,k), Valenciav_lU0(i,j,k) is the x-component of the velocity at (i-1/2-epsilon,j,k),
// and Valenciav_rU0(i,j,k) is the x-component of the velocity at (i-1/2+epsilon,j,k).
if(flux_dirn==0) {
// Next, we calculate the source term for StildeD. Again, this also resets the rhs_gfs array at
// each new timestep.
calculate_StildeD0_source_term(params,auxevol_gfs,rhs_gfs);
// Now, compute the electric field on each face of a cell and add it to the RHSs as appropriate
//calculate_E_field_D0_right(params,auxevol_gfs,rhs_gfs);
//calculate_E_field_D0_left(params,auxevol_gfs,rhs_gfs);
calculate_E_field_flat_all_in_one(params,auxevol_gfs,rhs_gfs,flux_dirn);
// Finally, we calculate the flux of StildeD and add the appropriate finite-differences
// to the RHSs.
calculate_Stilde_flux_D0_right(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_flux_D0_left(params,auxevol_gfs,rhs_gfs);
}
else if(flux_dirn==1) {
calculate_StildeD1_source_term(params,auxevol_gfs,rhs_gfs);
//calculate_E_field_D1_right(params,auxevol_gfs,rhs_gfs);
//calculate_E_field_D1_left(params,auxevol_gfs,rhs_gfs);
calculate_E_field_flat_all_in_one(params,auxevol_gfs,rhs_gfs,flux_dirn);
calculate_Stilde_flux_D1_right(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_flux_D1_left(params,auxevol_gfs,rhs_gfs);
}
else {
calculate_StildeD2_source_term(params,auxevol_gfs,rhs_gfs);
//calculate_E_field_D2_right(params,auxevol_gfs,rhs_gfs);
//calculate_E_field_D2_left(params,auxevol_gfs,rhs_gfs);
calculate_E_field_flat_all_in_one(params,auxevol_gfs,rhs_gfs,flux_dirn);
calculate_Stilde_flux_D2_right(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_flux_D2_left(params,auxevol_gfs,rhs_gfs);
}
}
}
void GiRaFFE_NRPy_post_step(const paramstruct *restrict params,REAL *xx[3],REAL *restrict auxevol_gfs,REAL *restrict evol_gfs,const int n) {
// First, apply BCs to AD and psi6Phi. Then calculate BU from AD
apply_bcs_potential(params,evol_gfs);
driver_A_to_B(params,evol_gfs,auxevol_gfs);
//override_BU_with_old_GiRaFFE(params,auxevol_gfs,n);
// Apply fixes to StildeD, then recompute the velocity at the new timestep.
// Apply the current sheet prescription to the velocities
GiRaFFE_NRPy_cons_to_prims(params,xx,auxevol_gfs,evol_gfs);
// Then, recompute StildeD to be consistent with the new velocities
//GiRaFFE_NRPy_prims_to_cons(params,auxevol_gfs,evol_gfs);
// Finally, apply outflow boundary conditions to the velocities.
apply_bcs_velocity(params,auxevol_gfs);
}
""")
def GiRaFFE_NRPy_A2B(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(os.path.join(Ccodesdir, "compute_B_and_Bstagger_from_A.h"),
"w") as file:
file.write(body)
paramslist.sort() # Sort the list alphabetically.
###############################
###############################
# Step 2: Generate all C-code kernels for Baikal and BaikalVacuum,
# in parallel if supported by this OS.
nrpy_dir_path = os.path.join(".")
if nrpy_dir_path not in sys.path:
sys.path.append(nrpy_dir_path)
# Create all output directories if they do not yet exist
import cmdline_helper as cmd # NRPy+: Multi-platform Python command-line interface
for ThornName in ["Baikal", "BaikalVacuum"]:
outrootdir = ThornName
cmd.mkdir(os.path.join(outrootdir))
outdir = os.path.join(outrootdir, "src") # Main C code output directory
# Copy SIMD/SIMD_intrinsics.h to $outdir/SIMD/SIMD_intrinsics.h, replacing
# the line "#define REAL_SIMD_ARRAY REAL" with "#define REAL_SIMD_ARRAY CCTK_REAL"
# (since REAL is undefined in the ETK, but CCTK_REAL takes its place)
cmd.mkdir(os.path.join(outdir, "SIMD"))
import fileinput
f = fileinput.input(
os.path.join(nrpy_dir_path, "SIMD", "SIMD_intrinsics.h"))
with open(os.path.join(outdir, "SIMD", "SIMD_intrinsics.h"),
"w") as outfile:
for line in f:
outfile.write(
line.replace("#define REAL_SIMD_ARRAY REAL",
"#define REAL_SIMD_ARRAY CCTK_REAL"))
def GiRaFFE_NRPy_Main_Driver_generate_all(out_dir):
cmd.mkdir(out_dir)
gammaDD = ixp.register_gridfunctions_for_single_rank2("AUXEVOL",
"gammaDD",
"sym01",
DIM=3)
betaU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"betaU",
DIM=3)
alpha = gri.register_gridfunctions("AUXEVOL", "alpha")
ixp.register_gridfunctions_for_single_rank1("EVOL", "AD")
BU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL", "BU")
ixp.register_gridfunctions_for_single_rank1("AUXEVOL", "BstaggerU")
ValenciavU = ixp.register_gridfunctions_for_single_rank1(
"AUXEVOL", "ValenciavU")
gri.register_gridfunctions("EVOL", "psi6Phi")
StildeD = ixp.register_gridfunctions_for_single_rank1("EVOL", "StildeD")
gri.register_gridfunctions("AUXEVOL", "psi6_temp")
gri.register_gridfunctions("AUXEVOL", "psi6center")
gri.register_gridfunctions("AUXEVOL", "cmax_x")
gri.register_gridfunctions("AUXEVOL", "cmin_x")
gri.register_gridfunctions("AUXEVOL", "cmax_y")
gri.register_gridfunctions("AUXEVOL", "cmin_y")
gri.register_gridfunctions("AUXEVOL", "cmax_z")
gri.register_gridfunctions("AUXEVOL", "cmin_z")
subdir = "RHSs"
stgsrc.GiRaFFE_NRPy_Source_Terms(os.path.join(out_dir, subdir))
# Declare this symbol:
sqrt4pi = par.Cparameters("REAL", thismodule, "sqrt4pi", "sqrt(4.0*M_PI)")
cmd.mkdir(os.path.join(out_dir, subdir))
source.write_out_functions_for_StildeD_source_term(
os.path.join(out_dir, subdir), outCparams, gammaDD, betaU, alpha,
ValenciavU, BU, sqrt4pi)
subdir = "FCVAL"
cmd.mkdir(os.path.join(out_dir, subdir))
FCVAL.GiRaFFE_NRPy_FCVAL(os.path.join(out_dir, subdir))
subdir = "PPM"
cmd.mkdir(os.path.join(out_dir, subdir))
PPM.GiRaFFE_NRPy_PPM(os.path.join(out_dir, subdir))
# We will pass values of the gridfunction on the cell faces into the function. This requires us
# to declare them as C parameters in NRPy+. We will denote this with the _face infix/suffix.
alpha_face = gri.register_gridfunctions("AUXEVOL", "alpha_face")
gamma_faceDD = ixp.register_gridfunctions_for_single_rank2(
"AUXEVOL", "gamma_faceDD", "sym01")
beta_faceU = ixp.register_gridfunctions_for_single_rank1(
"AUXEVOL", "beta_faceU")
# We'll need some more gridfunctions, now, to represent the reconstructions of BU and ValenciavU
# on the right and left faces
Valenciav_rU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"Valenciav_rU",
DIM=3)
B_rU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"B_rU",
DIM=3)
Valenciav_lU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"Valenciav_lU",
DIM=3)
B_lU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"B_lU",
DIM=3)
ixp.register_gridfunctions_for_single_rank1("AUXEVOL", "Stilde_flux_HLLED")
ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"Valenciav_rrU",
DIM=3)
ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"Valenciav_rlU",
DIM=3)
ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"Valenciav_lrU",
DIM=3)
ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"Valenciav_llU",
DIM=3)
ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"Bstagger_rU",
DIM=3)
ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"Bstagger_lU",
DIM=3)
subdir = "RHSs"
Af.GiRaFFE_NRPy_Afield_flux(os.path.join(out_dir, subdir))
Sf.generate_C_code_for_Stilde_flux(os.path.join(out_dir, subdir),
True,
alpha_face,
gamma_faceDD,
beta_faceU,
Valenciav_rU,
B_rU,
Valenciav_lU,
B_lU,
sqrt4pi,
write_cmax_cmin=True)
subdir = "boundary_conditions"
cmd.mkdir(os.path.join(out_dir, subdir))
BC.GiRaFFE_NRPy_BCs(os.path.join(out_dir, subdir))
subdir = "A2B"
cmd.mkdir(os.path.join(out_dir, subdir))
A2B.GiRaFFE_NRPy_A2B(os.path.join(out_dir, subdir))
C2P_P2C.GiRaFFE_NRPy_C2P(StildeD, BU, gammaDD, betaU, alpha)
values_to_print = [
lhrh(lhs=gri.gfaccess("in_gfs", "StildeD0"),
rhs=C2P_P2C.outStildeD[0]),
lhrh(lhs=gri.gfaccess("in_gfs", "StildeD1"),
rhs=C2P_P2C.outStildeD[1]),
lhrh(lhs=gri.gfaccess("in_gfs", "StildeD2"),
rhs=C2P_P2C.outStildeD[2]),
lhrh(lhs=gri.gfaccess("auxevol_gfs", "ValenciavU0"),
rhs=C2P_P2C.ValenciavU[0]),
lhrh(lhs=gri.gfaccess("auxevol_gfs", "ValenciavU1"),
rhs=C2P_P2C.ValenciavU[1]),
lhrh(lhs=gri.gfaccess("auxevol_gfs", "ValenciavU2"),
rhs=C2P_P2C.ValenciavU[2])
]
subdir = "C2P"
cmd.mkdir(os.path.join(out_dir, subdir))
desc = "Apply fixes to \tilde{S}_i and recompute the velocity to match with current sheet prescription."
name = "GiRaFFE_NRPy_cons_to_prims"
outCfunction(
outfile=os.path.join(out_dir, subdir, name + ".h"),
desc=desc,
name=name,
params=
"const paramstruct *params,REAL *xx[3],REAL *auxevol_gfs,REAL *in_gfs",
body=fin.FD_outputC("returnstring", values_to_print,
params=outCparams),
loopopts="AllPoints,Read_xxs",
rel_path_to_Cparams=os.path.join("../"))
C2P_P2C.GiRaFFE_NRPy_P2C(gammaDD, betaU, alpha, ValenciavU, BU, sqrt4pi)
values_to_print = [
lhrh(lhs=gri.gfaccess("in_gfs", "StildeD0"), rhs=C2P_P2C.StildeD[0]),
lhrh(lhs=gri.gfaccess("in_gfs", "StildeD1"), rhs=C2P_P2C.StildeD[1]),
lhrh(lhs=gri.gfaccess("in_gfs", "StildeD2"), rhs=C2P_P2C.StildeD[2]),
]
desc = "Recompute StildeD after current sheet fix to Valencia 3-velocity to ensure consistency between conservative & primitive variables."
name = "GiRaFFE_NRPy_prims_to_cons"
outCfunction(
outfile=os.path.join(out_dir, subdir, name + ".h"),
desc=desc,
name=name,
params="const paramstruct *params,REAL *auxevol_gfs,REAL *in_gfs",
body=fin.FD_outputC("returnstring", values_to_print,
params=outCparams),
loopopts="AllPoints",
rel_path_to_Cparams=os.path.join("../"))
# Write out the main driver itself:
with open(os.path.join(out_dir, "GiRaFFE_NRPy_Main_Driver.h"),
"w") as file:
file.write(r"""// Structure to track ghostzones for PPM:
typedef struct __gf_and_gz_struct__ {
REAL *gf;
int gz_lo[4],gz_hi[4];
} gf_and_gz_struct;
// Some additional constants needed for PPM:
static const int VX=0,VY=1,VZ=2,
BX_CENTER=3,BY_CENTER=4,BZ_CENTER=5,BX_STAGGER=6,BY_STAGGER=7,BZ_STAGGER=8,
VXR=9,VYR=10,VZR=11,VXL=12,VYL=13,VZL=14; //<-- Be _sure_ to define MAXNUMVARS appropriately!
const int NUM_RECONSTRUCT_GFS = 15;
#define WORKAROUND_ENABLED
// Include ALL functions needed for evolution
#include "PPM/reconstruct_set_of_prims_PPM_GRFFE_NRPy.c"
#include "FCVAL/interpolate_metric_gfs_to_cell_faces.h"
#include "RHSs/calculate_StildeD0_source_term.h"
#include "RHSs/calculate_StildeD1_source_term.h"
#include "RHSs/calculate_StildeD2_source_term.h"
#include "RHSs/Lorenz_psi6phi_rhs__add_gauge_terms_to_A_i_rhs.h"
#include "RHSs/A_i_rhs_no_gauge_terms.h"
#include "A2B/compute_B_and_Bstagger_from_A.h"
#include "RHSs/calculate_Stilde_flux_D0.h"
#include "RHSs/calculate_Stilde_flux_D1.h"
#include "RHSs/calculate_Stilde_flux_D2.h"
#include "RHSs/calculate_Stilde_rhsD.h"
#include "boundary_conditions/GiRaFFE_boundary_conditions.h"
#include "C2P/GiRaFFE_NRPy_cons_to_prims.h"
#include "C2P/GiRaFFE_NRPy_prims_to_cons.h"
void workaround_Valencia_to_Drift_velocity(const paramstruct *params, REAL *vU0, const REAL *alpha, const REAL *betaU0, const REAL flux_dirn) {
#include "set_Cparameters.h"
// Converts Valencia 3-velocities to Drift 3-velocities for testing. The variable argument
// vu0 is any Valencia 3-velocity component or reconstruction thereof.
#pragma omp parallel for
for (int i2 = 2*(flux_dirn==3);i2 < Nxx_plus_2NGHOSTS2-1*(flux_dirn==3);i2++) for (int i1 = 2*(flux_dirn==2);i1 < Nxx_plus_2NGHOSTS1-1*(flux_dirn==2);i1++) for (int i0 = 2*(flux_dirn==1);i0 < Nxx_plus_2NGHOSTS0-1*(flux_dirn==1);i0++) {
int ii = IDX3S(i0,i1,i2);
// Here, we modify the velocity in place.
vU0[ii] = alpha[ii]*vU0[ii]-betaU0[ii];
}
}
void workaround_Drift_to_Valencia_velocity(const paramstruct *params, REAL *vU0, const REAL *alpha, const REAL *betaU0, const REAL flux_dirn) {
#include "set_Cparameters.h"
// Converts Drift 3-velocities to Valencia 3-velocities for testing. The variable argument
// vu0 is any drift (i.e. IllinoisGRMHD's definition) 3-velocity component or reconstruction thereof.
#pragma omp parallel for
for (int i2 = 2*(flux_dirn==3);i2 < Nxx_plus_2NGHOSTS2-1*(flux_dirn==3);i2++) for (int i1 = 2*(flux_dirn==2);i1 < Nxx_plus_2NGHOSTS1-1*(flux_dirn==2);i1++) for (int i0 = 2*(flux_dirn==1);i0 < Nxx_plus_2NGHOSTS0-1*(flux_dirn==1);i0++) {
int ii = IDX3S(i0,i1,i2);
// Here, we modify the velocity in place.
vU0[ii] = (vU0[ii]+betaU0[ii])/alpha[ii];
}
}
void GiRaFFE_NRPy_RHSs(const paramstruct *restrict params,REAL *restrict auxevol_gfs,REAL *restrict in_gfs,REAL *restrict rhs_gfs) {
#include "set_Cparameters.h"
// First thing's first: initialize the RHSs to zero!
#pragma omp parallel for
for(int ii=0;ii<Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2*NUM_EVOL_GFS;ii++) {
rhs_gfs[ii] = 0.0;
}
// Now, we set up a bunch of structs of pointers to properly guide the PPM algorithm.
// They also count the number of ghostzones available.
gf_and_gz_struct in_prims[NUM_RECONSTRUCT_GFS], out_prims_r[NUM_RECONSTRUCT_GFS], out_prims_l[NUM_RECONSTRUCT_GFS];
int which_prims_to_reconstruct[NUM_RECONSTRUCT_GFS],num_prims_to_reconstruct;
const int Nxxp2NG012 = Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2;
REAL *temporary = auxevol_gfs + Nxxp2NG012*PSI6_TEMPGF; // Using dedicated temporary variables for the staggered grid
REAL *psi6center = auxevol_gfs + Nxxp2NG012*PSI6CENTERGF; // Because the prescription requires more acrobatics.
// This sets pointers to the portion of auxevol_gfs containing the relevant gridfunction.
int ww=0;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAVU0GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU0GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU0GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAVU1GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU1GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU1GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAVU2GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU2GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU2GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BU0GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*B_RU0GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*B_LU0GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BU1GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*B_RU1GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*B_LU1GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BU2GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*B_RU2GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*B_LU2GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BSTAGGERU0GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*BSTAGGER_RU0GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*BSTAGGER_LU0GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BSTAGGERU1GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*BSTAGGER_RU1GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*BSTAGGER_LU1GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BSTAGGERU2GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*BSTAGGER_RU2GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*BSTAGGER_LU2GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU0GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RRU0GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RLU0GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU1GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RRU1GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RLU1GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU2GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RRU2GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RLU2GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU0GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LRU0GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LLU0GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU1GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LRU1GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LLU1GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU2GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LRU2GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LLU2GF;
ww++;
// Prims are defined AT ALL GRIDPOINTS, so we set the # of ghostzones to zero:
for(int i=0;i<NUM_RECONSTRUCT_GFS;i++) for(int j=1;j<=3;j++) { in_prims[i].gz_lo[j]=0; in_prims[i].gz_hi[j]=0; }
// Left/right variables are not yet defined, yet we set the # of gz's to zero by default:
for(int i=0;i<NUM_RECONSTRUCT_GFS;i++) for(int j=1;j<=3;j++) { out_prims_r[i].gz_lo[j]=0; out_prims_r[i].gz_hi[j]=0; }
for(int i=0;i<NUM_RECONSTRUCT_GFS;i++) for(int j=1;j<=3;j++) { out_prims_l[i].gz_lo[j]=0; out_prims_l[i].gz_hi[j]=0; }
int flux_dirn;
flux_dirn=0;
interpolate_metric_gfs_to_cell_faces(params,auxevol_gfs,flux_dirn+1);
// ftilde = 0 in GRFFE, since P=rho=0.
/* There are two stories going on here:
* 1) Computation of \partial_x on RHS of \partial_t {mhd_st_{x,y,z}},
* via PPM reconstruction onto (i-1/2,j,k), so that
* \partial_y F = [ F(i+1/2,j,k) - F(i-1/2,j,k) ] / dx
* 2) Computation of \partial_t A_i, where A_i are *staggered* gridfunctions,
* where A_x is defined at (i,j+1/2,k+1/2), A_y at (i+1/2,j,k+1/2), etc.
* Ai_rhs = \partial_t A_i = \epsilon_{ijk} \psi^{6} (v^j B^k - v^j B^k),
* where \epsilon_{ijk} is the flat-space antisymmetric operator.
* 2A) Az_rhs is defined at (i+1/2,j+1/2,k), and it depends on {Bx,By,vx,vy},
* so the trick is to reconstruct {Bx,By,vx,vy} cleverly to get to these
* staggered points. For example:
* 2Aa) vx and vy are at (i,j,k), and we reconstruct them to (i-1/2,j,k) below. After
* this, we'll reconstruct again in the y-dir'n to get {vx,vy} at (i-1/2,j-1/2,k)
* 2Ab) By_stagger is at (i,j+1/2,k), and we reconstruct below to (i-1/2,j+1/2,k). */
ww=0;
which_prims_to_reconstruct[ww]=VX; ww++;
which_prims_to_reconstruct[ww]=VY; ww++;
which_prims_to_reconstruct[ww]=VZ; ww++;
//which_prims_to_reconstruct[ww]=BX_CENTER; ww++;
which_prims_to_reconstruct[ww]=BY_CENTER; ww++;
which_prims_to_reconstruct[ww]=BZ_CENTER; ww++;
which_prims_to_reconstruct[ww]=BY_STAGGER;ww++;
num_prims_to_reconstruct=ww;
// This function is housed in the file: "reconstruct_set_of_prims_PPM_GRFFE.C"
reconstruct_set_of_prims_PPM_GRFFE_NRPy(params, auxevol_gfs, flux_dirn+1, num_prims_to_reconstruct,
which_prims_to_reconstruct, in_prims, out_prims_r, out_prims_l, temporary);
//Right and left face values of BI_CENTER are used in GRFFE__S_i__flux computation (first to compute b^a).
// Instead of reconstructing, we simply set B^x face values to be consistent with BX_STAGGER.
#pragma omp parallel for
for(int k=0;k<Nxx_plus_2NGHOSTS2;k++) for(int j=0;j<Nxx_plus_2NGHOSTS1;j++) for(int i=0;i<Nxx_plus_2NGHOSTS0;i++) {
const int index=IDX3S(i,j,k), indexim1=IDX3S(i-1+(i==0),j,k); /* indexim1=0 when i=0 */
out_prims_r[BX_CENTER].gf[index]=out_prims_l[BX_CENTER].gf[index]=in_prims[BX_STAGGER].gf[indexim1]; }
// Then add fluxes to RHS for hydro variables {vx,vy,vz}:
// This function is housed in the file: "add_fluxes_and_source_terms_to_hydro_rhss.C"
calculate_StildeD0_source_term(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_flux_D0(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_rhsD(flux_dirn+1,params,auxevol_gfs,rhs_gfs);
// Note that we have already reconstructed vx and vy along the x-direction,
// at (i-1/2,j,k). That result is stored in v{x,y}{r,l}. Bx_stagger data
// are defined at (i+1/2,j,k).
// Next goal: reconstruct Bx, vx and vy at (i+1/2,j+1/2,k).
flux_dirn=1;
// ftilde = 0 in GRFFE, since P=rho=0.
// in_prims[{VXR,VXL,VYR,VYL}].gz_{lo,hi} ghostzones are set to all zeros, which
// is incorrect. We fix this below.
// [Note that this is a cheap operation, copying only 8 integers and a pointer.]
in_prims[VXR]=out_prims_r[VX];
in_prims[VXL]=out_prims_l[VX];
in_prims[VYR]=out_prims_r[VY];
in_prims[VYL]=out_prims_l[VY];
/* There are two stories going on here:
* 1) Computation of \partial_y on RHS of \partial_t {mhd_st_{x,y,z}},
* via PPM reconstruction onto (i,j-1/2,k), so that
* \partial_y F = [ F(i,j+1/2,k) - F(i,j-1/2,k) ] / dy
* 2) Computation of \partial_t A_i, where A_i are *staggered* gridfunctions,
* where A_x is defined at (i,j+1/2,k+1/2), A_y at (i+1/2,j,k+1/2), etc.
* Ai_rhs = \partial_t A_i = \epsilon_{ijk} \psi^{6} (v^j B^k - v^j B^k),
* where \epsilon_{ijk} is the flat-space antisymmetric operator.
* 2A) Az_rhs is defined at (i+1/2,j+1/2,k), and it depends on {Bx,By,vx,vy},
* so the trick is to reconstruct {Bx,By,vx,vy} cleverly to get to these
* staggered points. For example:
* 2Aa) VXR = [right-face of vx reconstructed along x-direction above] is at (i-1/2,j,k),
* and we reconstruct it to (i-1/2,j-1/2,k) below. Similarly for {VXL,VYR,VYL}
* 2Ab) Bx_stagger is at (i+1/2,j,k), and we reconstruct to (i+1/2,j-1/2,k) below
* 2Ac) By_stagger is at (i-1/2,j+1/2,k) already for Az_rhs, from the previous step.
* 2B) Ax_rhs is defined at (i,j+1/2,k+1/2), and it depends on {By,Bz,vy,vz}.
* Again the trick is to reconstruct these onto these staggered points.
* 2Ba) Bz_stagger is at (i,j,k+1/2), and we reconstruct to (i,j-1/2,k+1/2) below */
ww=0;
// NOTE! The order of variable reconstruction is important here,
// as we don't want to overwrite {vxr,vxl,vyr,vyl}!
which_prims_to_reconstruct[ww]=VXR; ww++;
which_prims_to_reconstruct[ww]=VYR; ww++;
which_prims_to_reconstruct[ww]=VXL; ww++;
which_prims_to_reconstruct[ww]=VYL; ww++;
num_prims_to_reconstruct=ww;
#ifdef WORKAROUND_ENABLED
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU0GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU0GF,flux_dirn+1);
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU1GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU1GF,flux_dirn+1);
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU0GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU0GF,flux_dirn+1);
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU1GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU1GF,flux_dirn+1);
#endif /*WORKAROUND_ENABLED*/
// This function is housed in the file: "reconstruct_set_of_prims_PPM_GRFFE.C"
reconstruct_set_of_prims_PPM_GRFFE_NRPy(params, auxevol_gfs, flux_dirn+1, num_prims_to_reconstruct,
which_prims_to_reconstruct, in_prims, out_prims_r, out_prims_l, temporary);
#ifdef WORKAROUND_ENABLED
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU0GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU0GF,flux_dirn+1);
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU1GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU1GF,flux_dirn+1);
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU0GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU0GF,flux_dirn+1);
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU1GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU1GF,flux_dirn+1);
#endif /*WORKAROUND_ENABLED*/
interpolate_metric_gfs_to_cell_faces(params,auxevol_gfs,flux_dirn+1);
ww=0;
// Reconstruct other primitives last!
which_prims_to_reconstruct[ww]=VX; ww++;
which_prims_to_reconstruct[ww]=VY; ww++;
which_prims_to_reconstruct[ww]=VZ; ww++;
which_prims_to_reconstruct[ww]=BX_CENTER; ww++;
//which_prims_to_reconstruct[ww]=BY_CENTER; ww++;
which_prims_to_reconstruct[ww]=BZ_CENTER; ww++;
which_prims_to_reconstruct[ww]=BX_STAGGER;ww++;
which_prims_to_reconstruct[ww]=BZ_STAGGER;ww++;
num_prims_to_reconstruct=ww;
// This function is housed in the file: "reconstruct_set_of_prims_PPM_GRFFE.C"
reconstruct_set_of_prims_PPM_GRFFE_NRPy(params, auxevol_gfs, flux_dirn+1, num_prims_to_reconstruct,
which_prims_to_reconstruct, in_prims, out_prims_r, out_prims_l, temporary);
//Right and left face values of BI_CENTER are used in GRFFE__S_i__flux computation (first to compute b^a).
// Instead of reconstructing, we simply set B^y face values to be consistent with BY_STAGGER.
#pragma omp parallel for
for(int k=0;k<Nxx_plus_2NGHOSTS2;k++) for(int j=0;j<Nxx_plus_2NGHOSTS1;j++) for(int i=0;i<Nxx_plus_2NGHOSTS0;i++) {
const int index=IDX3S(i,j,k), indexjm1=IDX3S(i,j-1+(j==0),k); /* indexjm1=0 when j=0 */
out_prims_r[BY_CENTER].gf[index]=out_prims_l[BY_CENTER].gf[index]=in_prims[BY_STAGGER].gf[indexjm1]; }
// Then add fluxes to RHS for hydro variables {vx,vy,vz}:
// This function is housed in the file: "add_fluxes_and_source_terms_to_hydro_rhss.C"
calculate_StildeD1_source_term(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_flux_D1(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_rhsD(flux_dirn+1,params,auxevol_gfs,rhs_gfs);
/*****************************************
* COMPUTING RHS OF A_z, BOOKKEEPING NOTE:
* We want to compute
* \partial_t A_z - [gauge terms] = \psi^{6} (v^x B^y - v^y B^x).
* A_z is defined at (i+1/2,j+1/2,k).
* ==========================
* Where defined | Variables
* (i-1/2,j-1/2,k)| {vxrr,vxrl,vxlr,vxll,vyrr,vyrl,vylr,vyll}
* (i+1/2,j-1/2,k)| {Bx_stagger_r,Bx_stagger_l} (see Table 1 in arXiv:1007.2848)
* (i-1/2,j+1/2,k)| {By_stagger_r,By_stagger_l} (see Table 1 in arXiv:1007.2848)
* (i,j,k) | {phi}
* ==========================
******************************************/
// Interpolates to i+1/2
#define IPH(METRICm1,METRICp0,METRICp1,METRICp2) (-0.0625*((METRICm1) + (METRICp2)) + 0.5625*((METRICp0) + (METRICp1)))
// Next compute sqrt(gamma)=psi^6 at (i+1/2,j+1/2,k):
// To do so, we first compute the sqrt of the metric determinant at all points:
#pragma omp parallel for
for(int k=0;k<Nxx_plus_2NGHOSTS2;k++) for(int j=0;j<Nxx_plus_2NGHOSTS1;j++) for(int i=0;i<Nxx_plus_2NGHOSTS0;i++) {
const int index=IDX3S(i,j,k);
const REAL gxx = auxevol_gfs[IDX4ptS(GAMMADD00GF,index)];
const REAL gxy = auxevol_gfs[IDX4ptS(GAMMADD01GF,index)];
const REAL gxz = auxevol_gfs[IDX4ptS(GAMMADD02GF,index)];
const REAL gyy = auxevol_gfs[IDX4ptS(GAMMADD11GF,index)];
const REAL gyz = auxevol_gfs[IDX4ptS(GAMMADD12GF,index)];
const REAL gzz = auxevol_gfs[IDX4ptS(GAMMADD22GF,index)];
psi6center[index] = sqrt( gxx*gyy*gzz
- gxx*gyz*gyz
+2*gxy*gxz*gyz
- gyy*gxz*gxz
- gzz*gxy*gxy );
}
#pragma omp parallel for
for(int k=0;k<Nxx_plus_2NGHOSTS2;k++) for(int j=1;j<Nxx_plus_2NGHOSTS1-2;j++) for(int i=1;i<Nxx_plus_2NGHOSTS0-2;i++) {
temporary[IDX3S(i,j,k)]=
IPH(IPH(psi6center[IDX3S(i-1,j-1,k)],psi6center[IDX3S(i,j-1,k)],psi6center[IDX3S(i+1,j-1,k)],psi6center[IDX3S(i+2,j-1,k)]),
IPH(psi6center[IDX3S(i-1,j ,k)],psi6center[IDX3S(i,j ,k)],psi6center[IDX3S(i+1,j ,k)],psi6center[IDX3S(i+2,j ,k)]),
IPH(psi6center[IDX3S(i-1,j+1,k)],psi6center[IDX3S(i,j+1,k)],psi6center[IDX3S(i+1,j+1,k)],psi6center[IDX3S(i+2,j+1,k)]),
IPH(psi6center[IDX3S(i-1,j+2,k)],psi6center[IDX3S(i,j+2,k)],psi6center[IDX3S(i+1,j+2,k)],psi6center[IDX3S(i+2,j+2,k)]));
}
int A_directionz=3;
A_i_rhs_no_gauge_terms(A_directionz,params,out_prims_r,out_prims_l,temporary,
auxevol_gfs+Nxxp2NG012*CMAX_XGF,
auxevol_gfs+Nxxp2NG012*CMIN_XGF,
auxevol_gfs+Nxxp2NG012*CMAX_YGF,
auxevol_gfs+Nxxp2NG012*CMIN_YGF,
rhs_gfs+Nxxp2NG012*AD2GF);
// in_prims[{VYR,VYL,VZR,VZL}].gz_{lo,hi} ghostzones are not correct, so we fix
// this below.
// [Note that this is a cheap operation, copying only 8 integers and a pointer.]
in_prims[VYR]=out_prims_r[VY];
in_prims[VYL]=out_prims_l[VY];
in_prims[VZR]=out_prims_r[VZ];
in_prims[VZL]=out_prims_l[VZ];
flux_dirn=2;
// ftilde = 0 in GRFFE, since P=rho=0.
/* There are two stories going on here:
* 1) Single reconstruction to (i,j,k-1/2) for {vx,vy,vz,Bx,By,Bz} to compute
* z-dir'n advection terms in \partial_t {mhd_st_{x,y,z}} at (i,j,k)
* 2) Multiple reconstructions for *staggered* gridfunctions A_i:
* \partial_t A_i = \epsilon_{ijk} \psi^{6} (v^j B^k - v^j B^k),
* where \epsilon_{ijk} is the flat-space antisymmetric operator.
* 2A) Ax_rhs is defined at (i,j+1/2,k+1/2), depends on v{y,z} and B{y,z}
* 2Aa) v{y,z}{r,l} are at (i,j-1/2,k), so we reconstruct here to (i,j-1/2,k-1/2)
* 2Ab) Bz_stagger{r,l} are at (i,j-1/2,k+1/2) already.
* 2Ac) By_stagger is at (i,j+1/2,k), and below we reconstruct its value at (i,j+1/2,k-1/2)
* 2B) Ay_rhs is defined at (i+1/2,j,k+1/2), depends on v{z,x} and B{z,x}.
* 2Ba) v{x,z} are reconstructed to (i,j,k-1/2). Later we'll reconstruct again to (i-1/2,j,k-1/2).
* 2Bb) Bz_stagger is at (i,j,k+1/2). Later we will reconstruct to (i-1/2,j,k+1/2).
* 2Bc) Bx_stagger is at (i+1/2,j,k), and below we reconstruct its value at (i+1/2,j,k-1/2)
*/
ww=0;
// NOTE! The order of variable reconstruction is important here,
// as we don't want to overwrite {vxr,vxl,vyr,vyl}!
which_prims_to_reconstruct[ww]=VYR; ww++;
which_prims_to_reconstruct[ww]=VZR; ww++;
which_prims_to_reconstruct[ww]=VYL; ww++;
which_prims_to_reconstruct[ww]=VZL; ww++;
num_prims_to_reconstruct=ww;
#ifdef WORKAROUND_ENABLED
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU1GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU1GF,flux_dirn+1);
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU2GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU2GF,flux_dirn+1);
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU1GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU1GF,flux_dirn+1);
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU2GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU2GF,flux_dirn+1);
#endif /*WORKAROUND_ENABLED*/
// This function is housed in the file: "reconstruct_set_of_prims_PPM_GRFFE.C"
reconstruct_set_of_prims_PPM_GRFFE_NRPy(params, auxevol_gfs, flux_dirn+1, num_prims_to_reconstruct,
which_prims_to_reconstruct, in_prims, out_prims_r, out_prims_l, temporary);
#ifdef WORKAROUND_ENABLED
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU1GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU1GF,flux_dirn+1);
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU2GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU2GF,flux_dirn+1);
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU1GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU1GF,flux_dirn+1);
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU2GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU2GF,flux_dirn+1);
#endif /*WORKAROUND_ENABLED*/
interpolate_metric_gfs_to_cell_faces(params,auxevol_gfs,flux_dirn+1);
// Reconstruct other primitives last!
ww=0;
which_prims_to_reconstruct[ww]=VX; ww++;
which_prims_to_reconstruct[ww]=VY; ww++;
which_prims_to_reconstruct[ww]=VZ; ww++;
which_prims_to_reconstruct[ww]=BX_CENTER; ww++;
which_prims_to_reconstruct[ww]=BY_CENTER; ww++;
//which_prims_to_reconstruct[ww]=BZ_CENTER; ww++;
which_prims_to_reconstruct[ww]=BX_STAGGER; ww++;
which_prims_to_reconstruct[ww]=BY_STAGGER; ww++;
num_prims_to_reconstruct=ww;
// This function is housed in the file: "reconstruct_set_of_prims_PPM_GRFFE.C"
reconstruct_set_of_prims_PPM_GRFFE_NRPy(params, auxevol_gfs, flux_dirn+1, num_prims_to_reconstruct,
which_prims_to_reconstruct, in_prims, out_prims_r, out_prims_l, temporary);
//Right and left face values of BI_CENTER are used in GRFFE__S_i__flux computation (first to compute b^a).
// Instead of reconstructing, we simply set B^z face values to be consistent with BZ_STAGGER.
#pragma omp parallel for
for(int k=0;k<Nxx_plus_2NGHOSTS2;k++) for(int j=0;j<Nxx_plus_2NGHOSTS1;j++) for(int i=0;i<Nxx_plus_2NGHOSTS0;i++) {
const int index=IDX3S(i,j,k), indexkm1=IDX3S(i,j,k-1+(k==0)); /* indexkm1=0 when k=0 */
out_prims_r[BZ_CENTER].gf[index]=out_prims_l[BZ_CENTER].gf[index]=in_prims[BZ_STAGGER].gf[indexkm1]; }
// Then add fluxes to RHS for hydro variables {vx,vy,vz}:
// This function is housed in the file: "add_fluxes_and_source_terms_to_hydro_rhss.C"
calculate_StildeD2_source_term(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_flux_D2(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_rhsD(flux_dirn+1,params,auxevol_gfs,rhs_gfs);
// in_prims[{VYR,VYL,VZR,VZL}].gz_{lo,hi} ghostzones are not set correcty.
// We fix this below.
// [Note that this is a cheap operation, copying only 8 integers and a pointer.]
in_prims[VXR]=out_prims_r[VX];
in_prims[VZR]=out_prims_r[VZ];
in_prims[VXL]=out_prims_l[VX];
in_prims[VZL]=out_prims_l[VZ];
// FIXME: lines above seem to be inconsistent with lines below.... Possible bug, not major enough to affect evolutions though.
in_prims[VZR].gz_lo[1]=in_prims[VZR].gz_hi[1]=0;
in_prims[VXR].gz_lo[1]=in_prims[VXR].gz_hi[1]=0;
in_prims[VZL].gz_lo[1]=in_prims[VZL].gz_hi[1]=0;
in_prims[VXL].gz_lo[1]=in_prims[VXL].gz_hi[1]=0;
/*****************************************
* COMPUTING RHS OF A_x, BOOKKEEPING NOTE:
* We want to compute
* \partial_t A_x - [gauge terms] = \psi^{6} (v^y B^z - v^z B^y).
* A_x is defined at (i,j+1/2,k+1/2).
* ==========================
* Where defined | Variables
* (i,j-1/2,k-1/2)| {vyrr,vyrl,vylr,vyll,vzrr,vzrl,vzlr,vzll}
* (i,j+1/2,k-1/2)| {By_stagger_r,By_stagger_l} (see Table 1 in arXiv:1007.2848)
* (i,j-1/2,k+1/2)| {Bz_stagger_r,Bz_stagger_l} (see Table 1 in arXiv:1007.2848)
* (i,j,k) | {phi}
* ==========================
******************************************/
// Next compute phi at (i,j+1/2,k+1/2):
#pragma omp parallel for
for(int k=1;k<Nxx_plus_2NGHOSTS2-2;k++) for(int j=1;j<Nxx_plus_2NGHOSTS1-2;j++) for(int i=0;i<Nxx_plus_2NGHOSTS0;i++) {
temporary[IDX3S(i,j,k)]=
IPH(IPH(psi6center[IDX3S(i,j-1,k-1)],psi6center[IDX3S(i,j,k-1)],psi6center[IDX3S(i,j+1,k-1)],psi6center[IDX3S(i,j+2,k-1)]),
IPH(psi6center[IDX3S(i,j-1,k )],psi6center[IDX3S(i,j,k )],psi6center[IDX3S(i,j+1,k )],psi6center[IDX3S(i,j+2,k )]),
IPH(psi6center[IDX3S(i,j-1,k+1)],psi6center[IDX3S(i,j,k+1)],psi6center[IDX3S(i,j+1,k+1)],psi6center[IDX3S(i,j+2,k+1)]),
IPH(psi6center[IDX3S(i,j-1,k+2)],psi6center[IDX3S(i,j,k+2)],psi6center[IDX3S(i,j+1,k+2)],psi6center[IDX3S(i,j+2,k+2)]));
}
int A_directionx=1;
A_i_rhs_no_gauge_terms(A_directionx,params,out_prims_r,out_prims_l,temporary,
auxevol_gfs+Nxxp2NG012*CMAX_YGF,
auxevol_gfs+Nxxp2NG012*CMIN_YGF,
auxevol_gfs+Nxxp2NG012*CMAX_ZGF,
auxevol_gfs+Nxxp2NG012*CMIN_ZGF,
rhs_gfs+Nxxp2NG012*AD0GF);
// We reprise flux_dirn=0 to finish up computations of Ai_rhs's!
flux_dirn=0;
// ftilde = 0 in GRFFE, since P=rho=0.
ww=0;
// NOTE! The order of variable reconstruction is important here,
// as we don't want to overwrite {vxr,vxl,vyr,vyl}!
which_prims_to_reconstruct[ww]=VXR; ww++;
which_prims_to_reconstruct[ww]=VZR; ww++;
which_prims_to_reconstruct[ww]=VXL; ww++;
which_prims_to_reconstruct[ww]=VZL; ww++;
which_prims_to_reconstruct[ww]=BZ_STAGGER;ww++;
num_prims_to_reconstruct=ww;
#ifdef WORKAROUND_ENABLED
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU0GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU0GF,flux_dirn+1);
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU2GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU2GF,flux_dirn+1);
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU0GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU0GF,flux_dirn+1);
workaround_Valencia_to_Drift_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU2GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU2GF,flux_dirn+1);
#endif /*WORKAROUND_ENABLED*/
// This function is housed in the file: "reconstruct_set_of_prims_PPM_GRFFE.C"
reconstruct_set_of_prims_PPM_GRFFE_NRPy(params, auxevol_gfs, flux_dirn+1, num_prims_to_reconstruct,
which_prims_to_reconstruct, in_prims, out_prims_r, out_prims_l, temporary);
#ifdef WORKAROUND_ENABLED
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU0GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU0GF,flux_dirn+1);
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_RU2GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU2GF,flux_dirn+1);
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU0GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU0GF,flux_dirn+1);
workaround_Drift_to_Valencia_velocity(params,auxevol_gfs+Nxxp2NG012*VALENCIAV_LU2GF,auxevol_gfs+Nxxp2NG012*ALPHA_FACEGF,auxevol_gfs+Nxxp2NG012*BETA_FACEU2GF,flux_dirn+1);
#endif /*WORKAROUND_ENABLED*/
/*****************************************
* COMPUTING RHS OF A_y, BOOKKEEPING NOTE:
* We want to compute
* \partial_t A_y - [gauge terms] = \psi^{6} (v^z B^x - v^x B^z).
* A_y is defined at (i+1/2,j,k+1/2).
* ==========================
* Where defined | Variables
* (i-1/2,j,k-1/2)| {vyrr,vyrl,vylr,vyll,vzrr,vzrl,vzlr,vzll}
* (i+1/2,j,k-1/2)| {By_stagger_r,By_stagger_l} (see Table 1 in arXiv:1007.2848)
* (i-1/2,j,k+1/2)| {Bz_stagger_r,Bz_stagger_l} (see Table 1 in arXiv:1007.2848)
* (i,j,k) | {phi}
* ==========================
******************************************/
// Next compute phi at (i+1/2,j,k+1/2):
#pragma omp parallel for
for(int k=1;k<Nxx_plus_2NGHOSTS2-2;k++) for(int j=0;j<Nxx_plus_2NGHOSTS1;j++) for(int i=1;i<Nxx_plus_2NGHOSTS0-2;i++) {
temporary[IDX3S(i,j,k)]=
IPH(IPH(psi6center[IDX3S(i-1,j,k-1)],psi6center[IDX3S(i,j,k-1)],psi6center[IDX3S(i+1,j,k-1)],psi6center[IDX3S(i+2,j,k-1)]),
IPH(psi6center[IDX3S(i-1,j,k )],psi6center[IDX3S(i,j,k )],psi6center[IDX3S(i+1,j,k )],psi6center[IDX3S(i+2,j,k )]),
IPH(psi6center[IDX3S(i-1,j,k+1)],psi6center[IDX3S(i,j,k+1)],psi6center[IDX3S(i+1,j,k+1)],psi6center[IDX3S(i+2,j,k+1)]),
IPH(psi6center[IDX3S(i-1,j,k+2)],psi6center[IDX3S(i,j,k+2)],psi6center[IDX3S(i+1,j,k+2)],psi6center[IDX3S(i+2,j,k+2)]));
}
int A_directiony=2;
A_i_rhs_no_gauge_terms(A_directiony,params,out_prims_r,out_prims_l,temporary,
auxevol_gfs+Nxxp2NG012*CMAX_ZGF,
auxevol_gfs+Nxxp2NG012*CMIN_ZGF,
auxevol_gfs+Nxxp2NG012*CMAX_XGF,
auxevol_gfs+Nxxp2NG012*CMIN_XGF,
rhs_gfs+Nxxp2NG012*AD1GF);
// Next compute psi6phi_rhs, and add gauge terms to A_i_rhs terms!
// Note that in the following function, we don't bother with reconstruction, instead interpolating.
// We need A^i, but only have A_i. So we add gtupij to the list of input variables.
REAL *interp_vars[MAXNUMINTERP];
ww=0;
interp_vars[ww]=auxevol_gfs+Nxxp2NG012*BETAU0GF; ww++;
interp_vars[ww]=auxevol_gfs+Nxxp2NG012*BETAU1GF; ww++;
interp_vars[ww]=auxevol_gfs+Nxxp2NG012*BETAU2GF; ww++;
interp_vars[ww]=auxevol_gfs+Nxxp2NG012*GAMMADD00GF; ww++;
interp_vars[ww]=auxevol_gfs+Nxxp2NG012*GAMMADD01GF; ww++;
interp_vars[ww]=auxevol_gfs+Nxxp2NG012*GAMMADD02GF; ww++;
interp_vars[ww]=auxevol_gfs+Nxxp2NG012*GAMMADD11GF; ww++;
interp_vars[ww]=auxevol_gfs+Nxxp2NG012*GAMMADD12GF; ww++;
interp_vars[ww]=auxevol_gfs+Nxxp2NG012*GAMMADD22GF; ww++;
interp_vars[ww]=temporary;ww++;
interp_vars[ww]=auxevol_gfs+Nxxp2NG012*ALPHAGF; ww++;
interp_vars[ww]=in_gfs+Nxxp2NG012*AD0GF; ww++;
interp_vars[ww]=in_gfs+Nxxp2NG012*AD1GF; ww++;
interp_vars[ww]=in_gfs+Nxxp2NG012*AD2GF; ww++;
const int max_num_interp_variables=ww;
// if(max_num_interp_variables>MAXNUMINTERP) {CCTK_VError(VERR_DEF_PARAMS,"Error: Didn't allocate enough space for interp_vars[]."); }
// We are FINISHED with v{x,y,z}{r,l} and P{r,l} so we use these 8 gridfunctions' worth of space as temp storage.
Lorenz_psi6phi_rhs__add_gauge_terms_to_A_i_rhs(params,interp_vars,
in_gfs+Nxxp2NG012*PSI6PHIGF,
auxevol_gfs+Nxxp2NG012*VALENCIAV_RU0GF, // WARNING:
auxevol_gfs+Nxxp2NG012*VALENCIAV_RU1GF, // ALL VARIABLES
auxevol_gfs+Nxxp2NG012*VALENCIAV_RU2GF, // ON THESE LINES
auxevol_gfs+Nxxp2NG012*VALENCIAV_LU0GF, // ARE OVERWRITTEN
auxevol_gfs+Nxxp2NG012*VALENCIAV_LU1GF, // FOR TEMP STORAGE
auxevol_gfs+Nxxp2NG012*VALENCIAV_LU2GF, // .
auxevol_gfs+Nxxp2NG012*VALENCIAV_RRU0GF, // .
auxevol_gfs+Nxxp2NG012*VALENCIAV_RLU0GF, // .
rhs_gfs+Nxxp2NG012*PSI6PHIGF,
rhs_gfs+Nxxp2NG012*AD0GF,
rhs_gfs+Nxxp2NG012*AD1GF,
rhs_gfs+Nxxp2NG012*AD2GF);
/*#pragma omp parallel for
for(int k=0;k<Nxx_plus_2NGHOSTS2;k++) for(int j=0;j<Nxx_plus_2NGHOSTS1;j++) for(int i=0;i<Nxx_plus_2NGHOSTS0;i++) {
REAL x = xx[0][i];
REAL y = xx[1][j];
REAL z = xx[2][k];
if(sqrt(x*x+y*y+z*z)<min_radius_inside_of_which_conserv_to_prims_FFE_and_FFE_evolution_is_DISABLED) {
rhs_gfs[IDX4S(STILDED0GF,i,j,k)] = 0.0;
rhs_gfs[IDX4S(STILDED1GF,i,j,k)] = 0.0;
rhs_gfs[IDX4S(STILDED2GF,i,j,k)] = 0.0;
rhs_gfs[IDX4S(PSI6PHIGF,i,j,k)] = 0.0;
rhs_gfs[IDX4S(AD0GF,i,j,k)] = 0.0;
rhs_gfs[IDX4S(AD1GF,i,j,k)] = 0.0;
rhs_gfs[IDX4S(AD2GF,i,j,k)] = 0.0;
}
}*/
}
void GiRaFFE_NRPy_post_step(const paramstruct *restrict params,REAL *xx[3],REAL *restrict auxevol_gfs,REAL *restrict evol_gfs,const int n) {
#include "set_Cparameters.h"
// First, apply BCs to AD and psi6Phi. Then calculate BU from AD
apply_bcs_potential(params,evol_gfs);
const int Nxxp2NG012 = Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2;
GiRaFFE_compute_B_and_Bstagger_from_A(params,
auxevol_gfs+Nxxp2NG012*GAMMADD00GF,
auxevol_gfs+Nxxp2NG012*GAMMADD01GF,
auxevol_gfs+Nxxp2NG012*GAMMADD02GF,
auxevol_gfs+Nxxp2NG012*GAMMADD11GF,
auxevol_gfs+Nxxp2NG012*GAMMADD12GF,
auxevol_gfs+Nxxp2NG012*GAMMADD22GF,
auxevol_gfs + Nxxp2NG012*PSI6_TEMPGF, /* Temporary storage */
evol_gfs+Nxxp2NG012*AD0GF,
evol_gfs+Nxxp2NG012*AD1GF,
evol_gfs+Nxxp2NG012*AD2GF,
auxevol_gfs+Nxxp2NG012*BU0GF,
auxevol_gfs+Nxxp2NG012*BU1GF,
auxevol_gfs+Nxxp2NG012*BU2GF,
auxevol_gfs+Nxxp2NG012*BSTAGGERU0GF,
auxevol_gfs+Nxxp2NG012*BSTAGGERU1GF,
auxevol_gfs+Nxxp2NG012*BSTAGGERU2GF);
//override_BU_with_old_GiRaFFE(params,auxevol_gfs,n);
// Apply fixes to StildeD, then recompute the velocity at the new timestep.
// Apply the current sheet prescription to the velocities
GiRaFFE_NRPy_cons_to_prims(params,xx,auxevol_gfs,evol_gfs);
// Then, recompute StildeD to be consistent with the new velocities
//GiRaFFE_NRPy_prims_to_cons(params,auxevol_gfs,evol_gfs);
// Finally, apply outflow boundary conditions to the velocities.
apply_bcs_velocity(params,auxevol_gfs);
}
""")
def GiRaFFE_NRPy_Afield_flux(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(os.path.join(Ccodesdir,"A_i_rhs_no_gauge_terms.h"),"w") as file:
file.write(r"""/* Compute the part of A_i_rhs that excludes the gauge terms. I.e., we set
* A_i_rhs = \partial_t A_i = \psi^{6} (v^z B^x - v^x B^z) here.
*/
static void A_i_rhs_no_gauge_terms(const int A_dirn,const paramstruct *params,gf_and_gz_struct *out_prims_r,gf_and_gz_struct *out_prims_l,
REAL *psi6_pointer,REAL *cmax_1,REAL *cmin_1,REAL *cmax_2,REAL *cmin_2, REAL *A3_rhs) {
#include "../set_Cparameters.h"
// If A_dirn=1, then v1_offset=1 (v1=VY) and v2_offset=2 (v2=VZ)
// If A_dirn=2, then v1_offset=2 (v1=VZ) and v2_offset=0 (v2=VX)
// If A_dirn=3, then v1_offset=0 (v1=VX) and v2_offset=1 (v2=VY)
const int v1_offset = ((A_dirn-1)+1)%3, v2_offset = ((A_dirn-1)+2)%3;
const REAL *v1rr=out_prims_r[VXR+v1_offset].gf, *v2rr=out_prims_r[VXR+v2_offset].gf;
const REAL *v1rl=out_prims_l[VXR+v1_offset].gf, *v2rl=out_prims_l[VXR+v2_offset].gf;
const REAL *v1lr=out_prims_r[VXL+v1_offset].gf, *v2lr=out_prims_r[VXL+v2_offset].gf;
const REAL *v1ll=out_prims_l[VXL+v1_offset].gf, *v2ll=out_prims_l[VXL+v2_offset].gf;
const REAL *B1r=out_prims_r[BX_STAGGER+v1_offset].gf, *B1l=out_prims_l[BX_STAGGER+v1_offset].gf;
const REAL *B2r=out_prims_r[BX_STAGGER+v2_offset].gf, *B2l=out_prims_l[BX_STAGGER+v2_offset].gf;
/**** V DEPENDENCIES ****/
/* In the case of Ax_rhs, we need v{y,z}{r,l} at (i,j+1/2,k+1/2).
* However, v{y,z}{r,l}{r,l} are defined at (i,j-1/2,k-1/2), so
* v{y,z}{r,l} at (i,j+1/2,k+1/2) is stored at v{y,z}{r,l}{r,l}(i,j+1,k+1).
* In the case of Ay_rhs, we need v{x,z}{r,l} at (i+1/2,j,k+1/2).
* However, v{x,z}{r,l}{r,l} are defined at (i-1/2,j,k-1/2), so
* v{x,z}{r,l} at (i+1/2,j,k+1/2) is stored at v{x,z}{r,l}{r,l}(i+1,j,k+1).
* In the case of Az_rhs, we need v{x,y}{r,l} at (i+1/2,j+1/2,k).
* However, v{x,y}{r,l}{r,l} are defined at (i-1/2,j-1/2,k), so
* v{x,y}{r,l} at (i+1/2,j+1/2,k) is stored at v{x,y}{r,l}{r,l}(i+1,j+1,k). */
static const int vs_ijk_offset[4][3] = { {0,0,0} , {0,1,1} , {1,0,1} , {1,1,0} }; // Note that vs_ijk_offset[0] is UNUSED; we choose a 1-offset for convenience.
/**** B DEPENDENCIES ****/
/* In the case of Ax_rhs, we need B{y,z}{r,l} at (i,j+1/2,k+1/2).
* However, By_stagger{r,l} is defined at (i,j+1/2,k-1/2), and
* Bz_stagger{r,l} is defined at (i,j-1/2,k+1/2), so
* By_stagger{r,l} at (i,j+1/2,k+1/2) is stored at By_stagger{r,l}(i,j,k+1), and
* Bz_stagger{r,l} at (i,j+1/2,k+1/2) is stored at Bz_stagger{r,l}(i,j+1,k).
* In the case of Ay_rhs, we need B{z,x}_stagger{r,l} at (i+1/2,j,k+1/2).
* However, Bz_stagger{r,l} is defined at (i-1/2,j,k+1/2), and
* Bx_stagger{r,l} is defined at (i+1/2,j,k-1/2), so
* Bz_stagger{r,l} at (i+1/2,j,k+1/2) is stored at Bz_stagger{r,l}(i+1,j,k), and
* Bx_stagger{r,l} at (i+1/2,j,k+1/2) is stored at Bx_stagger{r,l}(i,j,k+1).
* In the case of Az_rhs, we need B{x,y}_stagger{r,l} at (i+1/2,j+1/2,k).
* However, Bx_stagger{r,l} is defined at (i+1/2,j-1/2,k), and
* By_stagger{r,l} is defined at (i-1/2,j+1/2,k), so
* Bx_stagger{r,l} at (i+1/2,j+1/2,k) is stored at Bx_stagger{r,l}(i,j+1,k), and
* By_stagger{r,l} at (i+1/2,j+1/2,k) is stored at By_stagger{r,l}(i+1,j,k).
*/
static const int B1_ijk_offset[4][3] = { {0,0,0} , {0,0,1} , {1,0,0} , {0,1,0} }; // Note that B1_ijk_offset[0] is UNUSED; we choose a 1-offset for convenience.
static const int B2_ijk_offset[4][3] = { {0,0,0} , {0,1,0} , {0,0,1} , {1,0,0} }; // Note that B2_ijk_offset[0] is UNUSED; we choose a 1-offset for convenience.
#pragma omp parallel for
for(int k=NGHOSTS;k<Nxx_plus_2NGHOSTS2-NGHOSTS;k++) for(int j=NGHOSTS;j<Nxx_plus_2NGHOSTS1-NGHOSTS;j++) for(int i=NGHOSTS;i<Nxx_plus_2NGHOSTS0-NGHOSTS;i++) {
const int index=IDX3S(i,j,k);
// The following lines set the indices appropriately. See justification in exorbitant comments above.
const int index_v =IDX3S(i+vs_ijk_offset[A_dirn][0],j+vs_ijk_offset[A_dirn][1],k+vs_ijk_offset[A_dirn][2]);
const int index_B1=IDX3S(i+B1_ijk_offset[A_dirn][0],j+B1_ijk_offset[A_dirn][1],k+B1_ijk_offset[A_dirn][2]);
const int index_B2=IDX3S(i+B2_ijk_offset[A_dirn][0],j+B2_ijk_offset[A_dirn][1],k+B2_ijk_offset[A_dirn][2]);
// Stores 1/sqrt(gamma)==exp(6 phi) at (i+1/2,j+1/2,k) for Az, (i+1/2,j,k+1/2) for Ay, and (i,j+1/2,k+1/2) for Az.
const REAL psi6_interped=psi6_pointer[index];
const REAL B1lL = B1l[index_B1];
const REAL B1rL = B1r[index_B1];
const REAL B2lL = B2l[index_B2];
const REAL B2rL = B2r[index_B2];
const REAL A3_rhs_rr = psi6_interped*(v1rr[index_v]*B2rL - v2rr[index_v]*B1rL);
const REAL A3_rhs_rl = psi6_interped*(v1rl[index_v]*B2rL - v2rl[index_v]*B1lL);
const REAL A3_rhs_lr = psi6_interped*(v1lr[index_v]*B2lL - v2lr[index_v]*B1rL);
const REAL A3_rhs_ll = psi6_interped*(v1ll[index_v]*B2lL - v2ll[index_v]*B1lL);
// All variables for the A_i_rhs computation are now at the appropriate staggered point,
// so it's time to compute the HLL flux!
// Note that with PPM, cmin and cmax are defined between ijk=3 and ijk<cctk_lsh[]-2 for all directions.
const REAL cmax_1L = cmax_1[index_B2];
const REAL cmin_1L = cmin_1[index_B2];
const REAL cmax_2L = cmax_2[index_B1];
const REAL cmin_2L = cmin_2[index_B1];
const REAL B1tilder_minus_B1tildel = psi6_interped*( B1rL - B1lL );
const REAL B2tilder_minus_B2tildel = psi6_interped*( B2rL - B2lL );
/*---------------------------
* Implement 2D HLL flux
* [see Del Zanna, Bucciantini & Londrillo A&A 400, 397 (2003), Eq. (44)]
*
* Note that cmax/cmin (\alpha^{\pm} as defined in that paper) is at a slightly DIFFERENT
* point than that described in the Del Zanna et al paper (e.g., (i+1/2,j,k) instead of
* (i+1/2,j+1/2,k) for F3). Yuk Tung Liu discussed this point with M. Shibata,
* who found that the effect is negligible.
---------------------------*/
A3_rhs[index] = (cmax_1L*cmax_2L*A3_rhs_ll + cmax_1L*cmin_2L*A3_rhs_lr +
cmin_1L*cmax_2L*A3_rhs_rl + cmin_1L*cmin_2L*A3_rhs_rr)
/( (cmax_1L+cmin_1L)*(cmax_2L+cmin_2L) )
- cmax_1L*cmin_1L*(B2tilder_minus_B2tildel)/(cmax_1L+cmin_1L)
+ cmax_2L*cmin_2L*(B1tilder_minus_B1tildel)/(cmax_2L+cmin_2L);
}
}
""")
from outputC import outCfunction, lhrh, add_to_Cfunction_dict, outC_function_outdir_dict, outC_function_dict, outC_function_prototype_dict, outC_function_master_list, outC_function_element # NRPy+: Core C code output module
import finite_difference as fin # NRPy+: Finite difference C code generation module
import NRPy_param_funcs as par # NRPy+: Parameter interface
import grid as gri # NRPy+: Functions having to do with numerical grids
import loop as lp # NRPy+: Generate C code loops
import indexedexp as ixp # NRPy+: Symbolic indexed expression (e.g., tensors, vectors, etc.) support
import reference_metric as rfm # NRPy+: Reference metric support
import cmdline_helper as cmd # NRPy+: Multi-platform Python command-line interface
thismodule = __name__
par.set_parval_from_str("finite_difference::FD_CENTDERIVS_ORDER",2)
out_dir = os.path.join("GiRaFFE_standalone_Ccodes")
cmd.mkdir(out_dir)
CoordSystem = "Cartesian"
par.set_parval_from_str("reference_metric::CoordSystem",CoordSystem)
rfm.reference_metric() # Create ReU, ReDD needed for rescaling B-L initial data, generating BSSN RHSs, etc.
outCparams = "outCverbose=False,CSE_sorting=none"
xi_damping = par.Cparameters("REAL",thismodule,"xi_damping",0.1)
# Default Kreiss-Oliger dissipation strength
default_KO_strength = 0.1
diss_strength = par.Cparameters("REAL", thismodule, "diss_strength", default_KO_strength)
import GRHD.equations as GRHD # NRPy+: Generate general relativistic hydrodynamics equations
def GiRaFFE_NRPy_Afield_flux(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(os.path.join(Ccodesdir, "A_i_rhs_no_gauge_terms.h"),
"w") as file:
file.write(body)
def GiRaFFE_NRPy_Main_Driver_generate_all(out_dir):
cmd.mkdir(out_dir)
gammaDD = ixp.register_gridfunctions_for_single_rank2("AUXEVOL",
"gammaDD",
"sym01",
DIM=3)
betaU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"betaU",
DIM=3)
alpha = gri.register_gridfunctions("AUXEVOL", "alpha")
AD = ixp.register_gridfunctions_for_single_rank1("EVOL", "AD")
BU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL", "BU")
ValenciavU = ixp.register_gridfunctions_for_single_rank1(
"AUXEVOL", "ValenciavU")
psi6Phi = gri.register_gridfunctions("EVOL", "psi6Phi")
StildeD = ixp.register_gridfunctions_for_single_rank1("EVOL", "StildeD")
ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"PhievolParenU",
DIM=3)
gri.register_gridfunctions("AUXEVOL", "AevolParen")
# Declare this symbol:
sqrt4pi = par.Cparameters("REAL", thismodule, "sqrt4pi", "sqrt(4.0*M_PI)")
GRHD.compute_sqrtgammaDET(gammaDD)
GRFFE.compute_AD_source_term_operand_for_FD(GRHD.sqrtgammaDET, betaU,
alpha, psi6Phi, AD)
GRFFE.compute_psi6Phi_rhs_flux_term_operand(gammaDD, GRHD.sqrtgammaDET,
betaU, alpha, AD, psi6Phi)
parens_to_print = [\
lhrh(lhs=gri.gfaccess("auxevol_gfs","AevolParen"),rhs=GRFFE.AevolParen),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","PhievolParenU0"),rhs=GRFFE.PhievolParenU[0]),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","PhievolParenU1"),rhs=GRFFE.PhievolParenU[1]),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","PhievolParenU2"),rhs=GRFFE.PhievolParenU[2]),\
]
subdir = "RHSs"
cmd.mkdir(os.path.join(out_dir, subdir))
desc = "Calculate quantities to be finite-differenced for the GRFFE RHSs"
name = "calculate_AD_gauge_term_psi6Phi_flux_term_for_RHSs"
outCfunction(
outfile=os.path.join(out_dir, subdir, name + ".h"),
desc=desc,
name=name,
params=
"const paramstruct *restrict params,const REAL *restrict in_gfs,REAL *restrict auxevol_gfs",
body=fin.FD_outputC("returnstring", parens_to_print,
params=outCparams).replace("IDX4", "IDX4S"),
loopopts="AllPoints",
rel_path_for_Cparams=os.path.join("../"))
xi_damping = par.Cparameters("REAL", thismodule, "xi_damping", 0.1)
GRFFE.compute_psi6Phi_rhs_damping_term(alpha, psi6Phi, xi_damping)
AevolParen_dD = ixp.declarerank1("AevolParen_dD", DIM=3)
PhievolParenU_dD = ixp.declarerank2("PhievolParenU_dD", "nosym", DIM=3)
A_rhsD = ixp.zerorank1()
psi6Phi_rhs = GRFFE.psi6Phi_damping
for i in range(3):
A_rhsD[i] += -AevolParen_dD[i]
psi6Phi_rhs += -PhievolParenU_dD[i][i]
# Add Kreiss-Oliger dissipation to the GRFFE RHSs:
# psi6Phi_dKOD = ixp.declarerank1("psi6Phi_dKOD")
# AD_dKOD = ixp.declarerank2("AD_dKOD","nosym")
# for i in range(3):
# psi6Phi_rhs += diss_strength*psi6Phi_dKOD[i]*rfm.ReU[i] # ReU[i] = 1/scalefactor_orthog_funcform[i]
# for j in range(3):
# A_rhsD[j] += diss_strength*AD_dKOD[j][i]*rfm.ReU[i] # ReU[i] = 1/scalefactor_orthog_funcform[i]
RHSs_to_print = [\
lhrh(lhs=gri.gfaccess("rhs_gfs","AD0"),rhs=A_rhsD[0]),\
lhrh(lhs=gri.gfaccess("rhs_gfs","AD1"),rhs=A_rhsD[1]),\
lhrh(lhs=gri.gfaccess("rhs_gfs","AD2"),rhs=A_rhsD[2]),\
lhrh(lhs=gri.gfaccess("rhs_gfs","psi6Phi"),rhs=psi6Phi_rhs),\
]
desc = "Calculate AD gauge term and psi6Phi RHSs"
name = "calculate_AD_gauge_psi6Phi_RHSs"
source_Ccode = outCfunction(
outfile="returnstring",
desc=desc,
name=name,
params=
"const paramstruct *params,const REAL *in_gfs,const REAL *auxevol_gfs,REAL *rhs_gfs",
body=fin.FD_outputC("returnstring", RHSs_to_print,
params=outCparams).replace("IDX4", "IDX4S"),
loopopts="InteriorPoints",
rel_path_for_Cparams=os.path.join("../")).replace(
"= NGHOSTS", "= NGHOSTS_A2B").replace(
"NGHOSTS+Nxx0", "Nxx_plus_2NGHOSTS0-NGHOSTS_A2B").replace(
"NGHOSTS+Nxx1", "Nxx_plus_2NGHOSTS1-NGHOSTS_A2B").replace(
"NGHOSTS+Nxx2", "Nxx_plus_2NGHOSTS2-NGHOSTS_A2B")
# Note the above .replace() functions. These serve to expand the loop range into the ghostzones, since
# the second-order FD needs fewer than some other algorithms we use do.
with open(os.path.join(out_dir, subdir, name + ".h"), "w") as file:
file.write(source_Ccode)
source.write_out_functions_for_StildeD_source_term(
os.path.join(out_dir, subdir), outCparams, gammaDD, betaU, alpha,
ValenciavU, BU, sqrt4pi)
subdir = "FCVAL"
cmd.mkdir(os.path.join(out_dir, subdir))
FCVAL.GiRaFFE_NRPy_FCVAL(os.path.join(out_dir, subdir))
subdir = "PPM"
cmd.mkdir(os.path.join(out_dir, subdir))
PPM.GiRaFFE_NRPy_PPM(os.path.join(out_dir, subdir))
# We will pass values of the gridfunction on the cell faces into the function. This requires us
# to declare them as C parameters in NRPy+. We will denote this with the _face infix/suffix.
alpha_face = gri.register_gridfunctions("AUXEVOL", "alpha_face")
gamma_faceDD = ixp.register_gridfunctions_for_single_rank2(
"AUXEVOL", "gamma_faceDD", "sym01")
beta_faceU = ixp.register_gridfunctions_for_single_rank1(
"AUXEVOL", "beta_faceU")
# We'll need some more gridfunctions, now, to represent the reconstructions of BU and ValenciavU
# on the right and left faces
Valenciav_rU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"Valenciav_rU",
DIM=3)
B_rU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"B_rU",
DIM=3)
Valenciav_lU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"Valenciav_lU",
DIM=3)
B_lU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL",
"B_lU",
DIM=3)
subdir = "RHSs"
Af.GiRaFFE_NRPy_Afield_flux(os.path.join(out_dir, subdir))
Sf.generate_C_code_for_Stilde_flux(os.path.join(out_dir,
subdir), True, alpha_face,
gamma_faceDD, beta_faceU, Valenciav_rU,
B_rU, Valenciav_lU, B_lU, sqrt4pi)
subdir = "boundary_conditions"
cmd.mkdir(os.path.join(out_dir, subdir))
BC.GiRaFFE_NRPy_BCs(os.path.join(out_dir, subdir))
subdir = "A2B"
cmd.mkdir(os.path.join(out_dir, subdir))
A2B.GiRaFFE_NRPy_A2B(os.path.join(out_dir, subdir), gammaDD, AD, BU)
C2P_P2C.GiRaFFE_NRPy_C2P(StildeD, BU, gammaDD, betaU, alpha)
values_to_print = [\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD0"),rhs=C2P_P2C.outStildeD[0]),\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD1"),rhs=C2P_P2C.outStildeD[1]),\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD2"),rhs=C2P_P2C.outStildeD[2]),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","ValenciavU0"),rhs=C2P_P2C.ValenciavU[0]),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","ValenciavU1"),rhs=C2P_P2C.ValenciavU[1]),\
lhrh(lhs=gri.gfaccess("auxevol_gfs","ValenciavU2"),rhs=C2P_P2C.ValenciavU[2])\
]
subdir = "C2P"
cmd.mkdir(os.path.join(out_dir, subdir))
desc = "Apply fixes to \tilde{S}_i and recompute the velocity to match with current sheet prescription."
name = "GiRaFFE_NRPy_cons_to_prims"
outCfunction(
outfile=os.path.join(out_dir, subdir, name + ".h"),
desc=desc,
name=name,
params=
"const paramstruct *params,REAL *xx[3],REAL *auxevol_gfs,REAL *in_gfs",
body=fin.FD_outputC("returnstring", values_to_print,
params=outCparams).replace("IDX4", "IDX4S"),
loopopts="AllPoints,Read_xxs",
rel_path_for_Cparams=os.path.join("../"))
C2P_P2C.GiRaFFE_NRPy_P2C(gammaDD, betaU, alpha, ValenciavU, BU, sqrt4pi)
values_to_print = [\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD0"),rhs=C2P_P2C.StildeD[0]),\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD1"),rhs=C2P_P2C.StildeD[1]),\
lhrh(lhs=gri.gfaccess("in_gfs","StildeD2"),rhs=C2P_P2C.StildeD[2]),\
]
desc = "Recompute StildeD after current sheet fix to Valencia 3-velocity to ensure consistency between conservative & primitive variables."
name = "GiRaFFE_NRPy_prims_to_cons"
outCfunction(
outfile=os.path.join(out_dir, subdir, name + ".h"),
desc=desc,
name=name,
params="const paramstruct *params,REAL *auxevol_gfs,REAL *in_gfs",
body=fin.FD_outputC("returnstring", values_to_print,
params=outCparams).replace("IDX4", "IDX4S"),
loopopts="AllPoints",
rel_path_for_Cparams=os.path.join("../"))
# Write out the main driver itself:
with open(os.path.join(out_dir, "GiRaFFE_NRPy_Main_Driver.h"),
"w") as file:
file.write(r"""// Structure to track ghostzones for PPM:
typedef struct __gf_and_gz_struct__ {
REAL *gf;
int gz_lo[4],gz_hi[4];
} gf_and_gz_struct;
// Some additional constants needed for PPM:
const int VX=0,VY=1,VZ=2,BX=3,BY=4,BZ=5;
const int NUM_RECONSTRUCT_GFS = 6;
// Include ALL functions needed for evolution
#include "RHSs/calculate_AD_gauge_term_psi6Phi_flux_term_for_RHSs.h"
#include "RHSs/calculate_AD_gauge_psi6Phi_RHSs.h"
#include "PPM/reconstruct_set_of_prims_PPM_GRFFE_NRPy.c"
#include "FCVAL/interpolate_metric_gfs_to_cell_faces.h"
#include "RHSs/calculate_StildeD0_source_term.h"
#include "RHSs/calculate_StildeD1_source_term.h"
#include "RHSs/calculate_StildeD2_source_term.h"
#include "../calculate_E_field_flat_all_in_one.h"
#include "RHSs/calculate_Stilde_flux_D0.h"
#include "RHSs/calculate_Stilde_flux_D1.h"
#include "RHSs/calculate_Stilde_flux_D2.h"
#include "boundary_conditions/GiRaFFE_boundary_conditions.h"
#include "A2B/driver_AtoB.h"
#include "C2P/GiRaFFE_NRPy_cons_to_prims.h"
#include "C2P/GiRaFFE_NRPy_prims_to_cons.h"
void override_BU_with_old_GiRaFFE(const paramstruct *restrict params,REAL *restrict auxevol_gfs,const int n) {
#include "set_Cparameters.h"
char filename[100];
sprintf(filename,"BU0_override-%08d.bin",n);
FILE *out2D = fopen(filename, "rb");
fread(auxevol_gfs+BU0GF*Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,
sizeof(double),Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,out2D);
fclose(out2D);
sprintf(filename,"BU1_override-%08d.bin",n);
out2D = fopen(filename, "rb");
fread(auxevol_gfs+BU1GF*Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,
sizeof(double),Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,out2D);
fclose(out2D);
sprintf(filename,"BU2_override-%08d.bin",n);
out2D = fopen(filename, "rb");
fread(auxevol_gfs+BU2GF*Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,
sizeof(double),Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2,out2D);
fclose(out2D);
}
void GiRaFFE_NRPy_RHSs(const paramstruct *restrict params,REAL *restrict auxevol_gfs,const REAL *restrict in_gfs,REAL *restrict rhs_gfs) {
#include "set_Cparameters.h"
// First thing's first: initialize the RHSs to zero!
#pragma omp parallel for
for(int ii=0;ii<Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2*NUM_EVOL_GFS;ii++) {
rhs_gfs[ii] = 0.0;
}
// Next calculate the easier source terms that don't require flux directions
// This will also reset the RHSs for each gf at each new timestep.
calculate_AD_gauge_term_psi6Phi_flux_term_for_RHSs(params,in_gfs,auxevol_gfs);
calculate_AD_gauge_psi6Phi_RHSs(params,in_gfs,auxevol_gfs,rhs_gfs);
// Now, we set up a bunch of structs of pointers to properly guide the PPM algorithm.
// They also count the number of ghostzones available.
gf_and_gz_struct in_prims[NUM_RECONSTRUCT_GFS], out_prims_r[NUM_RECONSTRUCT_GFS], out_prims_l[NUM_RECONSTRUCT_GFS];
int which_prims_to_reconstruct[NUM_RECONSTRUCT_GFS],num_prims_to_reconstruct;
const int Nxxp2NG012 = Nxx_plus_2NGHOSTS0*Nxx_plus_2NGHOSTS1*Nxx_plus_2NGHOSTS2;
REAL *temporary = auxevol_gfs + Nxxp2NG012*AEVOLPARENGF; //We're not using this anymore
// This sets pointers to the portion of auxevol_gfs containing the relevant gridfunction.
int ww=0;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAVU0GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU0GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU0GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAVU1GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU1GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU1GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAVU2GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_RU2GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*VALENCIAV_LU2GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BU0GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*B_RU0GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*B_LU0GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BU1GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*B_RU1GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*B_LU1GF;
ww++;
in_prims[ww].gf = auxevol_gfs + Nxxp2NG012*BU2GF;
out_prims_r[ww].gf = auxevol_gfs + Nxxp2NG012*B_RU2GF;
out_prims_l[ww].gf = auxevol_gfs + Nxxp2NG012*B_LU2GF;
ww++;
// Prims are defined AT ALL GRIDPOINTS, so we set the # of ghostzones to zero:
for(int i=0;i<NUM_RECONSTRUCT_GFS;i++) for(int j=1;j<=3;j++) { in_prims[i].gz_lo[j]=0; in_prims[i].gz_hi[j]=0; }
// Left/right variables are not yet defined, yet we set the # of gz's to zero by default:
for(int i=0;i<NUM_RECONSTRUCT_GFS;i++) for(int j=1;j<=3;j++) { out_prims_r[i].gz_lo[j]=0; out_prims_r[i].gz_hi[j]=0; }
for(int i=0;i<NUM_RECONSTRUCT_GFS;i++) for(int j=1;j<=3;j++) { out_prims_l[i].gz_lo[j]=0; out_prims_l[i].gz_hi[j]=0; }
ww=0;
which_prims_to_reconstruct[ww]=VX; ww++;
which_prims_to_reconstruct[ww]=VY; ww++;
which_prims_to_reconstruct[ww]=VZ; ww++;
which_prims_to_reconstruct[ww]=BX; ww++;
which_prims_to_reconstruct[ww]=BY; ww++;
which_prims_to_reconstruct[ww]=BZ; ww++;
num_prims_to_reconstruct=ww;
// In each direction, perform the PPM reconstruction procedure.
// Then, add the fluxes to the RHS as appropriate.
for(int flux_dirn=0;flux_dirn<3;flux_dirn++) {
// In each direction, interpolate the metric gfs (gamma,beta,alpha) to cell faces.
interpolate_metric_gfs_to_cell_faces(params,auxevol_gfs,flux_dirn+1);
// Then, reconstruct the primitive variables on the cell faces.
// This function is housed in the file: "reconstruct_set_of_prims_PPM_GRFFE_NRPy.c"
reconstruct_set_of_prims_PPM_GRFFE_NRPy(params, auxevol_gfs, flux_dirn+1, num_prims_to_reconstruct,
which_prims_to_reconstruct, in_prims, out_prims_r, out_prims_l, temporary);
// For example, if flux_dirn==0, then at gamma_faceDD00(i,j,k) represents gamma_{xx}
// at (i-1/2,j,k), Valenciav_lU0(i,j,k) is the x-component of the velocity at (i-1/2-epsilon,j,k),
// and Valenciav_rU0(i,j,k) is the x-component of the velocity at (i-1/2+epsilon,j,k).
if(flux_dirn==0) {
// Next, we calculate the source term for StildeD. Again, this also resets the rhs_gfs array at
// each new timestep.
calculate_StildeD0_source_term(params,auxevol_gfs,rhs_gfs);
// Now, compute the electric field on each face of a cell and add it to the RHSs as appropriate
//calculate_E_field_D0_right(params,auxevol_gfs,rhs_gfs);
//calculate_E_field_D0_left(params,auxevol_gfs,rhs_gfs);
// Finally, we calculate the flux of StildeD and add the appropriate finite-differences
// to the RHSs.
calculate_Stilde_flux_D0(params,auxevol_gfs,rhs_gfs);
}
else if(flux_dirn==1) {
calculate_StildeD1_source_term(params,auxevol_gfs,rhs_gfs);
//calculate_E_field_D1_right(params,auxevol_gfs,rhs_gfs);
//calculate_E_field_D1_left(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_flux_D1(params,auxevol_gfs,rhs_gfs);
}
else {
calculate_StildeD2_source_term(params,auxevol_gfs,rhs_gfs);
//calculate_E_field_D2_right(params,auxevol_gfs,rhs_gfs);
//calculate_E_field_D2_left(params,auxevol_gfs,rhs_gfs);
calculate_Stilde_flux_D2(params,auxevol_gfs,rhs_gfs);
}
for(int count=0;count<=1;count++) {
// This function is written to be general, using notation that matches the forward permutation added to AD2,
// i.e., [F_HLL^x(B^y)]_z corresponding to flux_dirn=0, count=1.
// The SIGN parameter is necessary because
// -E_z(x_i,y_j,z_k) = 0.25 ( [F_HLL^x(B^y)]_z(i+1/2,j,k)+[F_HLL^x(B^y)]_z(i-1/2,j,k)
// -[F_HLL^y(B^x)]_z(i,j+1/2,k)-[F_HLL^y(B^x)]_z(i,j-1/2,k) )
// Note the negative signs on the reversed permutation terms!
// By cyclically permuting with flux_dirn, we
// get contributions to the other components, and by incrementing count, we get the backward permutations:
// Let's suppose flux_dirn = 0. Then we will need to update Ay (count=0) and Az (count=1):
// flux_dirn=count=0 -> AD0GF+(flux_dirn+1+count)%3 = AD0GF + (0+1+0)%3=AD1GF <- Updating Ay!
// (flux_dirn)%3 = (0)%3 = 0 Vx
// (flux_dirn-count+2)%3 = (0-0+2)%3 = 2 Vz . Inputs Vx, Vz -> SIGN = -1 ; 2.0*((REAL)count)-1.0=-1 check!
// flux_dirn=0,count=1 -> AD0GF+(flux_dirn+1+count)%3 = AD0GF + (0+1+1)%3=AD2GF <- Updating Az!
// (flux_dirn)%3 = (0)%3 = 0 Vx
// (flux_dirn-count+2)%3 = (0-1+2)%3 = 1 Vy . Inputs Vx, Vy -> SIGN = +1 ; 2.0*((REAL)count)-1.0=2-1=+1 check!
// Let's suppose flux_dirn = 1. Then we will need to update Az (count=0) and Ax (count=1):
// flux_dirn=1,count=0 -> AD0GF+(flux_dirn+1+count)%3 = AD0GF + (1+1+0)%3=AD2GF <- Updating Az!
// (flux_dirn)%3 = (1)%3 = 1 Vy
// (flux_dirn-count+2)%3 = (1-0+2)%3 = 0 Vx . Inputs Vy, Vx -> SIGN = -1 ; 2.0*((REAL)count)-1.0=-1 check!
// flux_dirn=count=1 -> AD0GF+(flux_dirn+1+count)%3 = AD0GF + (1+1+1)%3=AD0GF <- Updating Ax!
// (flux_dirn)%3 = (1)%3 = 1 Vy
// (flux_dirn-count+2)%3 = (1-1+2)%3 = 2 Vz . Inputs Vy, Vz -> SIGN = +1 ; 2.0*((REAL)count)-1.0=2-1=+1 check!
// Let's suppose flux_dirn = 2. Then we will need to update Ax (count=0) and Ay (count=1):
// flux_dirn=2,count=0 -> AD0GF+(flux_dirn+1+count)%3 = AD0GF + (2+1+0)%3=AD0GF <- Updating Ax!
// (flux_dirn)%3 = (2)%3 = 2 Vz
// (flux_dirn-count+2)%3 = (2-0+2)%3 = 1 Vy . Inputs Vz, Vy -> SIGN = -1 ; 2.0*((REAL)count)-1.0=-1 check!
// flux_dirn=2,count=1 -> AD0GF+(flux_dirn+1+count)%3 = AD0GF + (2+1+1)%3=AD1GF <- Updating Ay!
// (flux_dirn)%3 = (2)%3 = 2 Vz
// (flux_dirn-count+2)%3 = (2-1+2)%3 = 0 Vx . Inputs Vz, Vx -> SIGN = +1 ; 2.0*((REAL)count)-1.0=2-1=+1 check!
calculate_E_field_flat_all_in_one(params,
&auxevol_gfs[IDX4ptS(VALENCIAV_RU0GF+(flux_dirn)%3, 0)],&auxevol_gfs[IDX4ptS(VALENCIAV_RU0GF+(flux_dirn-count+2)%3, 0)],
&auxevol_gfs[IDX4ptS(VALENCIAV_LU0GF+(flux_dirn)%3, 0)],&auxevol_gfs[IDX4ptS(VALENCIAV_LU0GF+(flux_dirn-count+2)%3, 0)],
&auxevol_gfs[IDX4ptS(B_RU0GF +(flux_dirn)%3, 0)],&auxevol_gfs[IDX4ptS(B_RU0GF +(flux_dirn-count+2)%3, 0)],
&auxevol_gfs[IDX4ptS(B_LU0GF +(flux_dirn)%3, 0)],&auxevol_gfs[IDX4ptS(B_LU0GF +(flux_dirn-count+2)%3, 0)],
&auxevol_gfs[IDX4ptS(B_RU0GF +(flux_dirn-count+2)%3, 0)],
&auxevol_gfs[IDX4ptS(B_LU0GF +(flux_dirn-count+2)%3, 0)],
&rhs_gfs[IDX4ptS(AD0GF+(flux_dirn+1+count)%3,0)], 2.0*((REAL)count)-1.0, flux_dirn);
}
}
}
void GiRaFFE_NRPy_post_step(const paramstruct *restrict params,REAL *xx[3],REAL *restrict auxevol_gfs,REAL *restrict evol_gfs,const int n) {
// First, apply BCs to AD and psi6Phi. Then calculate BU from AD
apply_bcs_potential(params,evol_gfs);
driver_A_to_B(params,evol_gfs,auxevol_gfs);
//override_BU_with_old_GiRaFFE(params,auxevol_gfs,n);
// Apply fixes to StildeD, then recompute the velocity at the new timestep.
// Apply the current sheet prescription to the velocities
GiRaFFE_NRPy_cons_to_prims(params,xx,auxevol_gfs,evol_gfs);
// Then, recompute StildeD to be consistent with the new velocities
//GiRaFFE_NRPy_prims_to_cons(params,auxevol_gfs,evol_gfs);
// Finally, apply outflow boundary conditions to the velocities.
apply_bcs_velocity(params,auxevol_gfs);
}
""")
def construct_Makefile_from_outC_function_dict(Ccodesrootdir, exec_name, uses_free_parameters_h=False,
compiler_opt_option="fastdebug", addl_CFLAGS=None,
addl_libraries=None, mkdir_Ccodesrootdir=True, use_make=True, CC="gcc"):
if "main" not in outC_function_dict:
print("construct_Makefile_from_outC_function_dict() error: C codes will not compile if main() function not defined!")
print(" Make sure that the main() function registered to outC_function_dict has name \"main\".")
sys.exit(1)
if not os.path.isdir(Ccodesrootdir):
if not mkdir_Ccodesrootdir:
print("Error (in construct_Makefile_from_outC_function_dict): Directory \"" + Ccodesrootdir + "\" does not exist.")
sys.exit(1)
else:
import cmdline_helper as cmd
cmd.mkdir(Ccodesrootdir)
Makefile_list_of_files = []
def add_to_Makefile(Ccodesrootdir, path_and_file):
Makefile_list_of_files.append(path_and_file)
return os.path.join(Ccodesrootdir, path_and_file)
for key, item in outC_function_dict.items():
# Convention: Output all C files ending in _gridN into the gridN/ subdirectory.
if "grid" in key.split("_")[-1] and not "grids" in key:
subdir = key.split("_")[-1]
with open(add_to_Makefile(Ccodesrootdir, os.path.join(subdir, key+".c")), "w") as file:
file.write(item)
elif outC_function_outdir_dict[key] != "default":
subdir = outC_function_outdir_dict[key]
with open(add_to_Makefile(Ccodesrootdir, os.path.join(subdir, key+".c")), "w") as file:
file.write(item)
else:
with open(add_to_Makefile(Ccodesrootdir, os.path.join(key+".c")), "w") as file:
file.write(item)
CFLAGS = " -O2 -march=native -g -fopenmp -Wall -Wno-unused-variable"
DEBUGCFLAGS = " -O2 -g -Wall -Wno-unused-variable -Wno-unknown-pragmas" # OpenMP requires -fopenmp, and when disabling
# -fopenmp, unknown pragma warnings appear.
# -Wunknown-pragmas silences these warnings
FASTCFLAGS = " -O2 -march=native -fopenmp -Wall -Wno-unused-variable"
if CC == "gcc":
CFLAGS += " -std=gnu99"
DEBUGCFLAGS += " -std=gnu99"
FASTCFLAGS += " -std=gnu99"
CHOSEN_CFLAGS = CFLAGS
if compiler_opt_option == "debug":
CHOSEN_CFLAGS = DEBUGCFLAGS
elif compiler_opt_option == "fast":
CHOSEN_CFLAGS = FASTCFLAGS
if addl_CFLAGS is not None:
if not isinstance(addl_CFLAGS, list):
print("Error: construct_Makefile_from_outC_function_dict(): addl_CFLAGS must be a list!")
sys.exit(1)
for FLAG in addl_CFLAGS:
CHOSEN_CFLAGS += " "+FLAG
all_str = exec_name + " "
dep_list = []
compile_list = []
for c_file in Makefile_list_of_files:
object_file = c_file.replace(".c", ".o")
all_str += " " + object_file
addl_headers = ""
if uses_free_parameters_h:
if c_file == "main.c":
addl_headers += " free_parameters.h"
dep_list.append(object_file + ": " + c_file + addl_headers)
compile_list.append("\t$(CC) $(CFLAGS) -c " + c_file + " -o " + object_file)
linked_libraries = " -lm"
if addl_libraries is not None:
if not isinstance(addl_libraries, list):
print("Error: construct_Makefile_from_outC_function_dict(): addl_libraries must be a list!")
sys.exit(1)
for lib in addl_libraries:
linked_libraries += " " + lib
if use_make:
with open(os.path.join(Ccodesrootdir, "Makefile"), "w") as Makefile:
Makefile.write("""CC = """ + CC + """
CFLAGS = """ + CHOSEN_CFLAGS + """
#CFLAGS = """ + CFLAGS + """
#CFLAGS = """ + DEBUGCFLAGS + """
#CFLAGS = """ + FASTCFLAGS + "\n")
Makefile.write("all: " + all_str + "\n")
for idx, dep in enumerate(dep_list):
Makefile.write(dep + "\n")
Makefile.write(compile_list[idx] + "\n\n")
Makefile.write(exec_name + ": " + all_str.replace(exec_name, "") + "\n")
Makefile.write("\t$(CC) $(CFLAGS) main.c " + all_str.replace(exec_name, "").replace("main.o", "") + " -o " + exec_name + linked_libraries + "\n")
Makefile.write("\nclean:\n\trm -f *.o */*.o *~ */*~ ./#* *.txt *.dat *.avi *.png " + exec_name + "\n")
else:
with open(os.path.join(Ccodesrootdir, "backup_script_nomake.sh"), "w") as backup:
for compile_line in compile_list:
backup.write(compile_line.replace("$(CC)", CC).replace("$(CFLAGS)", CFLAGS).replace("\t", "") + "\n")
backup.write(CC + " " + CFLAGS + " main.c " + all_str.replace(exec_name, "").replace("main.o", "") + " -o " + exec_name + linked_libraries + "\n")
os.chmod(os.path.join(Ccodesrootdir, "backup_script_nomake.sh"), stat.S_IRWXU)
def GiRaFFE_NRPy_A2B(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(os.path.join(Ccodesdir, "compute_B_and_Bstagger_from_A.h"),
"w") as file:
file.write("""#define LOOP_DEFINE_SIMPLE \\
_Pragma("omp parallel for") \\
for(int k=0;k<Nxx_plus_2NGHOSTS2;k++) \\
for(int j=0;j<Nxx_plus_2NGHOSTS1;j++) \\
for(int i=0;i<Nxx_plus_2NGHOSTS0;i++)
void GiRaFFE_compute_B_and_Bstagger_from_A(const paramstruct *params,
const REAL *gxx, const REAL *gxy, const REAL *gxz, const REAL *gyy, const REAL *gyz,const REAL *gzz,
REAL *psi3_bssn, const REAL *Ax, const REAL *Ay, const REAL *Az,
REAL *Bx, REAL *By, REAL *Bz, REAL *Bx_stagger, REAL *By_stagger, REAL *Bz_stagger) {
#include "../set_Cparameters.h"
LOOP_DEFINE_SIMPLE {
const int index=IDX3S(i,j,k);
psi3_bssn[index] = sqrt(sqrt( gxx[index]*gyy[index]*gzz[index]
- gxx[index]*gyz[index]*gyz[index]
+2*gxy[index]*gxz[index]*gyz[index]
- gyy[index]*gxz[index]*gxz[index]
- gzz[index]*gxy[index]*gxy[index]));
}
LOOP_DEFINE_SIMPLE {
// Look Mom, no if() statements!
const int shiftedim1 = (i-1)*(i!=0); // This way, i=0 yields shiftedim1=0 and shiftedi=1, used below for our COPY boundary condition.
const int shiftedi = shiftedim1+1;
const int shiftedjm1 = (j-1)*(j!=0);
const int shiftedj = shiftedjm1+1;
const int shiftedkm1 = (k-1)*(k!=0);
const int shiftedk = shiftedkm1+1;
int index,indexim1,indexjm1,indexkm1;
const int actual_index = IDX3S(i,j,k);
const REAL Psim3 = 1.0/psi3_bssn[actual_index];
// For the lower boundaries, the following applies a "copy"
// boundary condition on Bi_stagger where needed.
// E.g., Bx_stagger(i,jmin,k) = Bx_stagger(i,jmin+1,k)
// We find the copy BC works better than extrapolation.
// For the upper boundaries, we do the following copy:
// E.g., Psi(imax+1,j,k)=Psi(imax,j,k)
/**************/
/* Bx_stagger */
/**************/
index = IDX3S(i,shiftedj,shiftedk);
indexjm1 = IDX3S(i,shiftedjm1,shiftedk);
indexkm1 = IDX3S(i,shiftedj,shiftedkm1);
// Set Bx_stagger = \partial_y A_z - partial_z A_y
// "Grid" Ax(i,j,k) is actually Ax(i,j+1/2,k+1/2)
// "Grid" Ay(i,j,k) is actually Ay(i+1/2,j,k+1/2)
// "Grid" Az(i,j,k) is actually Ay(i+1/2,j+1/2,k)
// Therefore, the 2nd order derivative \partial_z A_y at (i+1/2,j,k) is:
// ["Grid" Ay(i,j,k) - "Grid" Ay(i,j,k-1)]/dZ
Bx_stagger[actual_index] = (Az[index]-Az[indexjm1])*invdx1 - (Ay[index]-Ay[indexkm1])*invdx2;
// Now multiply Bx_stagger by 1/sqrt(gamma(i+1/2,j,k)]) = 1/sqrt(1/2 [gamma + gamma_ip1]) = exp(-6 x 1/2 [phi + phi_ip1] )
const int imax_minus_i = (Nxx_plus_2NGHOSTS0-1)-i;
const int indexip1jk = IDX3S(i + ( (imax_minus_i > 0) - (0 > imax_minus_i) ),j,k);
Bx_stagger[actual_index] *= Psim3/psi3_bssn[indexip1jk];
/**************/
/* By_stagger */
/**************/
index = IDX3S(shiftedi,j,shiftedk);
indexim1 = IDX3S(shiftedim1,j,shiftedk);
indexkm1 = IDX3S(shiftedi,j,shiftedkm1);
// Set By_stagger = \partial_z A_x - \partial_x A_z
By_stagger[actual_index] = (Ax[index]-Ax[indexkm1])*invdx2 - (Az[index]-Az[indexim1])*invdx0;
// Now multiply By_stagger by 1/sqrt(gamma(i,j+1/2,k)]) = 1/sqrt(1/2 [gamma + gamma_jp1]) = exp(-6 x 1/2 [phi + phi_jp1] )
const int jmax_minus_j = (Nxx_plus_2NGHOSTS1-1)-j;
const int indexijp1k = IDX3S(i,j + ( (jmax_minus_j > 0) - (0 > jmax_minus_j) ),k);
By_stagger[actual_index] *= Psim3/psi3_bssn[indexijp1k];
/**************/
/* Bz_stagger */
/**************/
index = IDX3S(shiftedi,shiftedj,k);
indexim1 = IDX3S(shiftedim1,shiftedj,k);
indexjm1 = IDX3S(shiftedi,shiftedjm1,k);
// Set Bz_stagger = \partial_x A_y - \partial_y A_x
Bz_stagger[actual_index] = (Ay[index]-Ay[indexim1])*invdx0 - (Ax[index]-Ax[indexjm1])*invdx1;
// Now multiply Bz_stagger by 1/sqrt(gamma(i,j,k+1/2)]) = 1/sqrt(1/2 [gamma + gamma_kp1]) = exp(-6 x 1/2 [phi + phi_kp1] )
const int kmax_minus_k = (Nxx_plus_2NGHOSTS2-1)-k;
const int indexijkp1 = IDX3S(i,j,k + ( (kmax_minus_k > 0) - (0 > kmax_minus_k) ));
Bz_stagger[actual_index] *= Psim3/psi3_bssn[indexijkp1];
}
LOOP_DEFINE_SIMPLE {
// Look Mom, no if() statements!
const int shiftedim1 = (i-1)*(i!=0); // This way, i=0 yields shiftedim1=0 and shiftedi=1, used below for our COPY boundary condition.
const int shiftedi = shiftedim1+1;
const int shiftedjm1 = (j-1)*(j!=0);
const int shiftedj = shiftedjm1+1;
const int shiftedkm1 = (k-1)*(k!=0);
const int shiftedk = shiftedkm1+1;
int index,indexim1,indexjm1,indexkm1;
const int actual_index = IDX3S(i,j,k);
// For the lower boundaries, the following applies a "copy"
// boundary condition on Bi and Bi_stagger where needed.
// E.g., Bx(imin,j,k) = Bx(imin+1,j,k)
// We find the copy BC works better than extrapolation.
/******/
/* Bx */
/******/
index = IDX3S(shiftedi,j,k);
indexim1 = IDX3S(shiftedim1,j,k);
// Set Bx = 0.5 ( Bx_stagger + Bx_stagger_im1 )
// "Grid" Bx_stagger(i,j,k) is actually Bx_stagger(i+1/2,j,k)
Bx[actual_index] = 0.5 * ( Bx_stagger[index] + Bx_stagger[indexim1] );
/******/
/* By */
/******/
index = IDX3S(i,shiftedj,k);
indexjm1 = IDX3S(i,shiftedjm1,k);
// Set By = 0.5 ( By_stagger + By_stagger_im1 )
// "Grid" By_stagger(i,j,k) is actually By_stagger(i,j+1/2,k)
By[actual_index] = 0.5 * ( By_stagger[index] + By_stagger[indexjm1] );
/******/
/* Bz */
/******/
index = IDX3S(i,j,shiftedk);
indexkm1 = IDX3S(i,j,shiftedkm1);
// Set Bz = 0.5 ( Bz_stagger + Bz_stagger_im1 )
// "Grid" Bz_stagger(i,j,k) is actually Bz_stagger(i,j+1/2,k)
Bz[actual_index] = 0.5 * ( Bz_stagger[index] + Bz_stagger[indexkm1] );
}
}
""")
def GiRaFFE_NRPy_Afield_flux(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
gammaDD = ixp.declarerank2("gammaDD", "sym01", DIM=3)
betaU = ixp.declarerank1("betaU", DIM=3)
alpha = sp.sympify("alpha")
for flux_dirn in range(3):
chsp.find_cmax_cmin(flux_dirn, gammaDD, betaU, alpha)
Ccode_kernel = outputC([chsp.cmax, chsp.cmin], ["cmax", "cmin"],
"returnstring",
params="outCverbose=False,CSE_sorting=none")
Ccode_kernel = Ccode_kernel.replace("cmax",
"*cmax").replace("cmin", "*cmin")
Ccode_kernel = Ccode_kernel.replace("betaU0", "betaUi").replace(
"betaU1", "betaUi").replace("betaU2", "betaUi")
with open(
os.path.join(Ccodesdir,
"compute_cmax_cmin_dirn" + str(flux_dirn) + ".h"),
"w") as file:
file.write(Ccode_kernel)
with open(os.path.join(Ccodesdir, "calculate_E_field_flat_all_in_one.h"),
"w") as file:
file.write(
r"""void find_cmax_cmin(const REAL gammaDD00, const REAL gammaDD01, const REAL gammaDD02,
const REAL gammaDD11, const REAL gammaDD12, const REAL gammaDD22,
const REAL betaUi, const REAL alpha, const int flux_dirn,
REAL *cmax, REAL *cmin) {
switch(flux_dirn) {
case 0:
#include "compute_cmax_cmin_dirn0.h"
break;
case 1:
#include "compute_cmax_cmin_dirn1.h"
break;
case 2:
#include "compute_cmax_cmin_dirn2.h"
break;
default:
printf("Invalid parameter flux_dirn!"); *cmax = 1.0/0.0; *cmin = 1.0/0.0;
break;
}
}
REAL HLLE_solve(REAL F0B1_r, REAL F0B1_l, REAL U_r, REAL U_l, REAL cmin, REAL cmax) {
// Eq. 3.15 of https://epubs.siam.org/doi/abs/10.1137/1025002?journalCode=siread
// F_HLLE = (c_min F_R + c_max F_L - c_min c_max (U_R-U_L)) / (c_min + c_max)
return (cmin*F0B1_r + cmax*F0B1_l - cmin*cmax*(U_r-U_l)) / (cmin+cmax);
}
/*
Calculate the electric flux on both faces in the input direction.
The input count is an integer that is either 0 or 1. If it is 0, this implies
that the components are input in order of a backwards permutation and the final
results will need to be multiplied by -1.0. If it is 1, then the permutation is forwards.
*/
void calculate_E_field_flat_all_in_one(const paramstruct *params,
const REAL *Vr0,const REAL *Vr1,
const REAL *Vl0,const REAL *Vl1,
const REAL *Br0,const REAL *Br1,
const REAL *Bl0,const REAL *Bl1,
const REAL *Brflux_dirn,
const REAL *Blflux_dirn,
const REAL *gamma_faceDD00, const REAL *gamma_faceDD01, const REAL *gamma_faceDD02,
const REAL *gamma_faceDD11, const REAL *gamma_faceDD12, const REAL *gamma_faceDD22,
const REAL *beta_faceU0, const REAL *beta_faceU1, const REAL *alpha_face,
REAL *A2_rhs,const REAL SIGN,const int flux_dirn) {
// This function is written to be generic and compute the contribution for all three AD RHSs.
// However, for convenience, the notation used in the function itself is for the contribution
// to AD2, specifically the [F_HLL^x(B^y)]_z term, with reconstructions in the x direction. This
// corresponds to flux_dirn=0 and count=1 (which corresponds to SIGN=+1.0).
// Thus, Az(i,j,k) += 0.25 ( [F_HLL^x(B^y)]_z(i+1/2,j,k)+[F_HLL^x(B^y)]_z(i-1/2,j,k)) are solved here.
// The other terms are computed by cyclically permuting the indices when calling this function.
#include "../set_Cparameters.h"
#pragma omp parallel for
for(int i2=NGHOSTS; i2<NGHOSTS+Nxx2; i2++) {
for(int i1=NGHOSTS; i1<NGHOSTS+Nxx1; i1++) {
for(int i0=NGHOSTS; i0<NGHOSTS+Nxx0; i0++) {
// First, we set the index from which we will read memory. indexp1 is incremented by
// one point in the direction of reconstruction. These correspond to the faces at at
// i-1/2 and i+1/2, respectively.
// Now, we read in memory. We need the x and y components of velocity and magnetic field on both
// the left and right sides of the interface at *both* faces.
// Here, the point (i0,i1,i2) corresponds to the point (i-1/2,j,k)
const int index = IDX3S(i0,i1,i2);
const double alpha = alpha_face[index];
const double betaU0 = beta_faceU0[index];
const double betaU1 = beta_faceU1[index];
const double v_rU0 = alpha*Vr0[index]-betaU0;
const double v_rU1 = alpha*Vr1[index]-betaU1;
const double B_rU0 = Br0[index];
const double B_rU1 = Br1[index];
const double B_rflux_dirn = Brflux_dirn[index];
const double v_lU0 = alpha*Vl0[index]-betaU0;
const double v_lU1 = alpha*Vl1[index]-betaU1;
const double B_lU0 = Bl0[index];
const double B_lU1 = Bl1[index];
const double B_lflux_dirn = Blflux_dirn[index];
// We will also need need the square root of the metric determinant here at this point:
const REAL gxx = gamma_faceDD00[index];
const REAL gxy = gamma_faceDD01[index];
const REAL gxz = gamma_faceDD02[index];
const REAL gyy = gamma_faceDD11[index];
const REAL gyz = gamma_faceDD12[index];
const REAL gzz = gamma_faceDD22[index];
const REAL sqrtgammaDET = sqrt( gxx*gyy*gzz
- gxx*gyz*gyz
+2*gxy*gxz*gyz
- gyy*gxz*gxz
- gzz*gxy*gxy );
// *******************************
// REPEAT ABOVE, but at i+1, which corresponds to point (i+1/2,j,k)
// Recall that the documentation here assumes flux_dirn==0, but the
// algorithm is generalized so that any flux_dirn or velocity/magnetic
// field component can be computed via permuting the inputs into this
// function.
const int indexp1 = IDX3S(i0+(flux_dirn==0),i1+(flux_dirn==1),i2+(flux_dirn==2));
const double alpha_p1 = alpha_face[indexp1];
const double betaU0_p1 = beta_faceU0[indexp1];
const double betaU1_p1 = beta_faceU1[indexp1];
const double v_rU0_p1 = alpha_p1*Vr0[indexp1]-betaU0_p1;
const double v_rU1_p1 = alpha_p1*Vr1[indexp1]-betaU1_p1;
const double B_rU0_p1 = Br0[indexp1];
const double B_rU1_p1 = Br1[indexp1];
const double B_rflux_dirn_p1 = Brflux_dirn[indexp1];
const double v_lU0_p1 = alpha_p1*Vl0[indexp1]-betaU0_p1;
const double v_lU1_p1 = alpha_p1*Vl1[indexp1]-betaU1_p1;
const double B_lU0_p1 = Bl0[indexp1];
const double B_lU1_p1 = Bl1[indexp1];
const double B_lflux_dirn_p1 = Blflux_dirn[indexp1];
// We will also need need the square root of the metric determinant here at this point:
const REAL gxx_p1 = gamma_faceDD00[indexp1];
const REAL gxy_p1 = gamma_faceDD01[indexp1];
const REAL gxz_p1 = gamma_faceDD02[indexp1];
const REAL gyy_p1 = gamma_faceDD11[indexp1];
const REAL gyz_p1 = gamma_faceDD12[indexp1];
const REAL gzz_p1 = gamma_faceDD22[indexp1];
const REAL sqrtgammaDET_p1 = sqrt( gxx_p1*gyy_p1*gzz_p1
- gxx_p1*gyz_p1*gyz_p1
+2*gxy_p1*gxz_p1*gyz_p1
- gyy_p1*gxz_p1*gxz_p1
- gzz_p1*gxy_p1*gxy_p1 );
// *******************************
// DEBUGGING:
// if(flux_dirn==0 && SIGN>0 && i1==Nxx_plus_2NGHOSTS1/2 && i2==Nxx_plus_2NGHOSTS2/2) {
// printf("index=%d & indexp1=%d\n",index,indexp1);
// }
// Since we are computing A_z, the relevant equation here is:
// -E_z(x_i,y_j,z_k) = 0.25 ( [F_HLL^x(B^y)]_z(i+1/2,j,k)+[F_HLL^x(B^y)]_z(i-1/2,j,k)
// -[F_HLL^y(B^x)]_z(i,j+1/2,k)-[F_HLL^y(B^x)]_z(i,j-1/2,k) )
// We will construct the above sum one half at a time, first with SIGN=+1, which
// corresponds to flux_dirn = 0, count=1, and
// takes care of the terms:
// [F_HLL^x(B^y)]_z(i+1/2,j,k)+[F_HLL^x(B^y)]_z(i-1/2,j,k)
// ( Note that we will repeat the above with flux_dirn = 1, count = 0, with SIGN=-1
// AND with the input components switched (x->y,y->x) so that we get the term
// -[F_HLL^y(B^x)]_z(i,j+1/2,k)-[F_HLL^y(B^x)]_z(i,j-1/2,k)
// thus completing the above sum. )
// Here, [F_HLL^i(B^j)]_k = (v^i B^j - v^j B^i) in general.
// Calculate the flux vector on each face for each component of the E-field:
// The F(B) terms are as Eq. 6 in Giacomazzo: https://arxiv.org/pdf/1009.2468.pdf
// [F^i(B^j)]_k = \sqrt{\gamma} (v^i B^j - v^j B^i)
// Therefore since we want [F_HLL^x(B^y)]_z,
// we will code (v^x B^y - v^y B^x) on both left and right faces.
const REAL F0B1_r = sqrtgammaDET*(v_rU0*B_rU1 - v_rU1*B_rU0);
const REAL F0B1_l = sqrtgammaDET*(v_lU0*B_lU1 - v_lU1*B_lU0);
// Compute the state vector for these terms:
const REAL U_r = B_rflux_dirn;
const REAL U_l = B_lflux_dirn;
REAL cmin,cmax;
// Basic HLLE solver:
find_cmax_cmin(gxx,gxy,gxz,
gyy,gyz,gzz,
betaU0,alpha,flux_dirn,
&cmax, &cmin);
const REAL FHLL_0B1 = HLLE_solve(F0B1_r, F0B1_l, U_r, U_l, cmin, cmax);
// ************************************
// ************************************
// REPEAT ABOVE, but at point i+1
// Calculate the flux vector on each face for each component of the E-field:
const REAL F0B1_r_p1 = sqrtgammaDET_p1*(v_rU0_p1*B_rU1_p1 - v_rU1_p1*B_rU0_p1);
const REAL F0B1_l_p1 = sqrtgammaDET_p1*(v_lU0_p1*B_lU1_p1 - v_lU1_p1*B_lU0_p1);
// Compute the state vector for this flux direction
const REAL U_r_p1 = B_rflux_dirn_p1;
const REAL U_l_p1 = B_lflux_dirn_p1;
//const REAL U_r_p1 = B_rU1_p1;
//const REAL U_l_p1 = B_lU1_p1;
// Basic HLLE solver, but at the next point:
find_cmax_cmin(gxx_p1,gxy_p1,gxz_p1,
gyy_p1,gyz_p1,gzz_p1,
betaU0_p1,alpha_p1,flux_dirn,
&cmax, &cmin);
const REAL FHLL_0B1p1 = HLLE_solve(F0B1_r_p1, F0B1_l_p1, U_r_p1, U_l_p1, cmin, cmax);
// ************************************
// ************************************
// With the Riemann problem solved, we add the contributions to the RHSs:
// -E_z(x_i,y_j,z_k) &= 0.25 ( [F_HLL^x(B^y)]_z(i+1/2,j,k)+[F_HLL^x(B^y)]_z(i-1/2,j,k)
// -[F_HLL^y(B^x)]_z(i,j+1/2,k)-[F_HLL^y(B^x)]_z(i,j-1/2,k) )
// (Eq. 11 in https://arxiv.org/pdf/1009.2468.pdf)
// This code, as written, solves the first two terms for flux_dirn=0. Calling this function for count=0
// and flux_dirn=1 flips x for y to solve the latter two, switching to SIGN=-1 as well.
// Here, we finally add together the output of the HLLE solver at i-1/2 and i+1/2
// We also multiply by the SIGN dictated by the order of the input vectors and divide by 4.
A2_rhs[index] += SIGN*0.25*(FHLL_0B1 + FHLL_0B1p1);
// flux dirn = 0 ===================> i-1/2 i+1/2
// Eq 11 in Giacomazzo:
// -FxBy(avg over i-1/2 and i+1/2) + FyBx(avg over j-1/2 and j+1/2)
// Eq 6 in Giacomazzo:
// FxBy = vxBy - vyBx
// ->
// FHLL_0B1 = vyBx - vxBy
} // END LOOP: for(int i0=NGHOSTS; i0<NGHOSTS+Nxx0; i0++)
} // END LOOP: for(int i1=NGHOSTS; i1<NGHOSTS+Nxx1; i1++)
} // END LOOP: for(int i2=NGHOSTS; i2<NGHOSTS+Nxx2; i2++)
}
""")
def GiRaFFE_NRPy_BCs(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(os.path.join(Ccodesdir,"GiRaFFE_boundary_conditions.h"),"w") as file:
file.write("""// Currently, we're using basic Cartesian boundary conditions, pending fixes by Zach.
// Part P8a: Declare boundary condition FACE_UPDATE macro,
// which updates a single face of the 3D grid cube
// using quadratic polynomial extrapolation.
// Basic extrapolation boundary conditions
#define FACE_UPDATE(which_gf, i0min,i0max, i1min,i1max, i2min,i2max, FACEX0,FACEX1,FACEX2) \\
for(int i2=i2min;i2<i2max;i2++) for(int i1=i1min;i1<i1max;i1++) for(int i0=i0min;i0<i0max;i0++) { \\
gfs[IDX4S(which_gf,i0,i1,i2)] = \\
+2.0*gfs[IDX4S(which_gf,i0+1*FACEX0,i1+1*FACEX1,i2+1*FACEX2)] \\
-1.0*gfs[IDX4S(which_gf,i0+2*FACEX0,i1+2*FACEX1,i2+2*FACEX2)]; \\
}
// +1.0*gfs[IDX4S(which_gf,i0+3*FACEX0,i1+3*FACEX1,i2+3*FACEX2)]; \\
// Basic Copy boundary conditions
#define FACE_UPDATE_COPY(which_gf, i0min,i0max, i1min,i1max, i2min,i2max, FACEX0,FACEX1,FACEX2) \\
for(int i2=i2min;i2<i2max;i2++) for(int i1=i1min;i1<i1max;i1++) for(int i0=i0min;i0<i0max;i0++) { \\
gfs[IDX4S(which_gf,i0,i1,i2)] = gfs[IDX4S(which_gf,i0+1*FACEX0,i1+1*FACEX1,i2+1*FACEX2)]; \\
}
// Part P8b: Boundary condition driver routine: Apply BCs to all six
// boundary faces of the cube, filling in the innermost
// ghost zone first, and moving outward.
const int MAXFACE = -1;
const int NUL = +0;
const int MINFACE = +1;
// This macro acts differently in that it acts on an entire 3-vector of gfs, instead of 1.
// which_gf_0 corresponds to the zeroth component of that vector. The if statements only
// evaluate true if the velocity is directed inwards on the face in consideration.
#define FACE_UPDATE_OUTFLOW(which_gf, i0min,i0max, i1min,i1max, i2min,i2max, FACEX0,FACEX1,FACEX2) \\
for(int i2=i2min;i2<i2max;i2++) for(int i1=i1min;i1<i1max;i1++) for(int i0=i0min;i0<i0max;i0++) { \\
aux_gfs[IDX4S(which_gf_0,i0,i1,i2)] = \\
aux_gfs[IDX4S(which_gf_0,i0+FACEX0,i1+FACEX1,i2+FACEX2)]; \\
aux_gfs[IDX4S(which_gf_0+1,i0,i1,i2)] = \\
aux_gfs[IDX4S(which_gf_0+1,i0+FACEX0,i1+FACEX1,i2+FACEX2)]; \\
aux_gfs[IDX4S(which_gf_0+2,i0,i1,i2)] = \\
aux_gfs[IDX4S(which_gf_0+2,i0+FACEX0,i1+FACEX1,i2+FACEX2)]; \\
}
/* if(FACEX0*aux_gfs[IDX4S(which_gf_0+0,i0,i1,i2)] > 0.0) { \\
aux_gfs[IDX4S(which_gf_0+0,i0,i1,i2)] = 0.0; \\
} \\
if(FACEX1*aux_gfs[IDX4S(which_gf_0+1,i0,i1,i2)] > 0.0) { \\
aux_gfs[IDX4S(which_gf_0+1,i0,i1,i2)] = 0.0; \\
} \\
if(FACEX2*aux_gfs[IDX4S(which_gf_0+2,i0,i1,i2)] > 0.0) { \\
aux_gfs[IDX4S(which_gf_0+2,i0,i1,i2)] = 0.0; \\
} \\
*/
void apply_bcs_potential(const paramstruct *restrict params,REAL *gfs) {
#include "../set_Cparameters.h"
// First, we apply extrapolation boundary conditions to AD
#pragma omp parallel for
for(int which_gf=0;which_gf<NUM_EVOL_GFS;which_gf++) {
if(which_gf < STILDED0GF || which_gf > STILDED2GF) {
int imin[3] = { NGHOSTS, NGHOSTS, NGHOSTS };
int imax[3] = { Nxx_plus_2NGHOSTS0-NGHOSTS, Nxx_plus_2NGHOSTS1-NGHOSTS, Nxx_plus_2NGHOSTS2-NGHOSTS };
for(int which_gz = 0; which_gz < NGHOSTS; which_gz++) {
// After updating each face, adjust imin[] and imax[]
// to reflect the newly-updated face extents.
FACE_UPDATE(which_gf, imin[0]-1,imin[0], imin[1],imax[1], imin[2],imax[2], MINFACE,NUL,NUL); imin[0]--;
FACE_UPDATE(which_gf, imax[0],imax[0]+1, imin[1],imax[1], imin[2],imax[2], MAXFACE,NUL,NUL); imax[0]++;
FACE_UPDATE(which_gf, imin[0],imax[0], imin[1]-1,imin[1], imin[2],imax[2], NUL,MINFACE,NUL); imin[1]--;
FACE_UPDATE(which_gf, imin[0],imax[0], imax[1],imax[1]+1, imin[2],imax[2], NUL,MAXFACE,NUL); imax[1]++;
FACE_UPDATE(which_gf, imin[0],imax[0], imin[1],imax[1], imin[2]-1,imin[2], NUL,NUL,MINFACE);
imin[2]--;
FACE_UPDATE(which_gf, imin[0],imax[0], imin[1],imax[1], imax[2],imax[2]+1, NUL,NUL,MAXFACE);
imax[2]++;
}
}
}
// Then, we apply copy boundary conditions to StildeD and psi6Phi
/*#pragma omp parallel for
for(int which_gf=3;which_gf<NUM_EVOL_GFS;which_gf++) {
int imin[3] = { NGHOSTS, NGHOSTS, NGHOSTS };
int imax[3] = { Nxx_plus_2NGHOSTS0-NGHOSTS, Nxx_plus_2NGHOSTS1-NGHOSTS, Nxx_plus_2NGHOSTS2-NGHOSTS };
for(int which_gz = 0; which_gz < NGHOSTS; which_gz++) {
// After updating each face, adjust imin[] and imax[]
// to reflect the newly-updated face extents.
FACE_UPDATE_COPY(which_gf, imin[0]-1,imin[0], imin[1],imax[1], imin[2],imax[2], MINFACE,NUL,NUL); imin[0]--;
FACE_UPDATE_COPY(which_gf, imax[0],imax[0]+1, imin[1],imax[1], imin[2],imax[2], MAXFACE,NUL,NUL); imax[0]++;
FACE_UPDATE_COPY(which_gf, imin[0],imax[0], imin[1]-1,imin[1], imin[2],imax[2], NUL,MINFACE,NUL); imin[1]--;
FACE_UPDATE_COPY(which_gf, imin[0],imax[0], imax[1],imax[1]+1, imin[2],imax[2], NUL,MAXFACE,NUL); imax[1]++;
FACE_UPDATE_COPY(which_gf, imin[0],imax[0], imin[1],imax[1], imin[2]-1,imin[2], NUL,NUL,MINFACE); imin[2]--;
FACE_UPDATE_COPY(which_gf, imin[0],imax[0], imin[1],imax[1], imax[2],imax[2]+1, NUL,NUL,MAXFACE); imax[2]++;
}
}*/
}
void apply_bcs_velocity(const paramstruct *restrict params,REAL *aux_gfs) {
#include "../set_Cparameters.h"
// Apply outflow/copy boundary conditions to ValenciavU by passing VALENCIAVU0 as which_gf_0
// for(int which_gf=VALENCIAVU0GF;which_gf<=VALENCIAVU2GF;which_gf++) {
const int which_gf_0 = VALENCIAVU0GF;
int imin[3] = { NGHOSTS, NGHOSTS, NGHOSTS };
int imax[3] = { Nxx_plus_2NGHOSTS0-NGHOSTS, Nxx_plus_2NGHOSTS1-NGHOSTS, Nxx_plus_2NGHOSTS2-NGHOSTS };
for(int which_gz = 0; which_gz < NGHOSTS; which_gz++) {
FACE_UPDATE_OUTFLOW(which_gf_0, imin[0]-1,imin[0], imin[1],imax[1], imin[2],imax[2], MINFACE,NUL,NUL); imin[0]--;
FACE_UPDATE_OUTFLOW(which_gf_0, imax[0],imax[0]+1, imin[1],imax[1], imin[2],imax[2], MAXFACE,NUL,NUL); imax[0]++;
FACE_UPDATE_OUTFLOW(which_gf_0, imin[0],imax[0], imin[1]-1,imin[1], imin[2],imax[2], NUL,MINFACE,NUL); imin[1]--;
FACE_UPDATE_OUTFLOW(which_gf_0, imin[0],imax[0], imax[1],imax[1]+1, imin[2],imax[2], NUL,MAXFACE,NUL); imax[1]++;
FACE_UPDATE_OUTFLOW(which_gf_0, imin[0],imax[0], imin[1],imax[1], imin[2]-1,imin[2], NUL,NUL,MINFACE);
imin[2]--;
FACE_UPDATE_OUTFLOW(which_gf_0, imin[0],imax[0], imin[1],imax[1], imax[2],imax[2]+1, NUL,NUL,MAXFACE);
imax[2]++;
}
// }
}
/*// A supplement to the boundary conditions for debugging. This will overwrite data with exact conditions
void FACE_UPDATE_EXACT(const paramstruct *restrict params,REAL *restrict xx[3],
const int n, const REAL dt,REAL *out_gfs,REAL *aux_gfs,
const int i0min,const int i0max, const int i1min,const int i1max, const int i2min,const int i2max,
const int FACEX0,const int FACEX1,const int FACEX2) {
#include "../set_Cparameters.h"
for(int i2=i2min;i2<i2max;i2++) for(int i1=i1min;i1<i1max;i1++) for(int i0=i0min;i0<i0max;i0++) {
REAL xx0 = xx[0][i0]-n*dt;
REAL xx1 = xx[1][i1];
REAL xx2 = xx[2][i2];
if(xx0<=lbound) {
#include "../GiRaFFEfood_A_v_1D_tests_left.h"
}
else if (xx0<rbound) {
#include "../GiRaFFEfood_A_v_1D_tests_center.h"
}
else {
#include "../GiRaFFEfood_A_v_1D_tests_right.h"
}
out_gfs[IDX4S(PSI6PHIGF, i0,i1,i2)] = 0.0;
}
}
void apply_bcs_EXACT(const paramstruct *restrict params,REAL *restrict xx[3],
const int n, const REAL dt,
REAL *out_gfs,REAL *aux_gfs) {
#include "../set_Cparameters.h"
int imin[3] = { NGHOSTS, NGHOSTS, NGHOSTS };
int imax[3] = { Nxx_plus_2NGHOSTS0-NGHOSTS, Nxx_plus_2NGHOSTS1-NGHOSTS, Nxx_plus_2NGHOSTS2-NGHOSTS };
for(int which_gz = 0; which_gz < NGHOSTS; which_gz++) {
// After updating each face, adjust imin[] and imax[]
// to reflect the newly-updated face extents.
// Right now, we only want to update the xmin and xmax faces with the exact data.
FACE_UPDATE_EXACT(Nxx,Nxx_plus_2NGHOSTS,xx,n,dt,out_gfs,aux_gfs,imin[0]-1,imin[0], imin[1],imax[1], imin[2],imax[2], MINFACE,NUL,NUL);
imin[0]--;
FACE_UPDATE_EXACT(Nxx,Nxx_plus_2NGHOSTS,xx,n,dt,out_gfs,aux_gfs,imax[0],imax[0]+1, imin[1],imax[1], imin[2],imax[2], MAXFACE,NUL,NUL);
imax[0]++;
FACE_UPDATE_EXACT(Nxx,Nxx_plus_2NGHOSTS,xx,n,dt,out_gfs,aux_gfs,imin[0],imax[0], imin[1]-1,imin[1], imin[2],imax[2], NUL,MINFACE,NUL);
imin[1]--;
FACE_UPDATE_EXACT(Nxx,Nxx_plus_2NGHOSTS,xx,n,dt,out_gfs,aux_gfs,imin[0],imax[0], imax[1],imax[1]+1, imin[2],imax[2], NUL,MAXFACE,NUL);
imax[1]++;
FACE_UPDATE_EXACT(Nxx,Nxx_plus_2NGHOSTS,xx,n,dt,out_gfs,aux_gfs,imin[0],imax[0], imin[1],imax[1], imin[2]-1,imin[2], NUL,NUL,MINFACE);
imin[2]--;
FACE_UPDATE_EXACT(Nxx,Nxx_plus_2NGHOSTS,xx,n,dt,out_gfs,aux_gfs,imin[0],imax[0], imin[1],imax[1], imax[2],imax[2]+1, NUL,NUL,MAXFACE);
imax[2]++;
}
}
// A supplement to the boundary conditions for debugging. This will overwrite data with exact conditions
void FACE_UPDATE_EXACT_StildeD(const paramstruct *restrict params,REAL *restrict xx[3],
REAL *out_gfs,REAL *out_gfs_exact,
const int i0min,const int i0max, const int i1min,const int i1max, const int i2min,const int i2max,
const int FACEX0,const int FACEX1,const int FACEX2) {
#include "../set_Cparameters.h"
// This is currently modified to calculate more exact boundary conditions for StildeD. Rename if it works.
for(int i2=i2min;i2<i2max;i2++) for(int i1=i1min;i1<i1max;i1++) for(int i0=i0min;i0<i0max;i0++) {
#include "../GiRaFFEfood_NRPy_Stilde.h"
}
idx = IDX3(i0,i1,i2);
out_gfs[IDX4ptS(STILDED0GF,idx)] = out_gfs_exact[IDX4ptS(STILDED0GF,idx)];
out_gfs[IDX4ptS(STILDED1GF,idx)] = out_gfs_exact[IDX4ptS(STILDED1GF,idx)];
out_gfs[IDX4ptS(STILDED2GF,idx)] = out_gfs_exact[IDX4ptS(STILDED2GF,idx)];
}
void apply_bcs_EXACT_StildeD(const paramstruct *restrict params,REAL *restrict xx[3],
REAL *out_gfs,REAL *out_gfs_exact) {
#include "../set_Cparameters.h"
int imin[3] = { NGHOSTS, NGHOSTS, NGHOSTS };
int imax[3] = { Nxx_plus_2NGHOSTS0-NGHOSTS, Nxx_plus_2NGHOSTS1-NGHOSTS, Nxx_plus_2NGHOSTS2-NGHOSTS };
for(int which_gz = 0; which_gz < NGHOSTS; which_gz++) {
// After updating each face, adjust imin[] and imax[]
// to reflect the newly-updated face extents.
// Right now, we only want to update the xmin and xmax faces with the exact data.
FACE_UPDATE_EXACT_StildeD(Nxx,Nxx_plus_2NGHOSTS,xx,out_gfs,out_gfs_exact,imin[0]-1,imin[0], imin[1],imax[1], imin[2],imax[2], MINFACE,NUL,NUL);
imin[0]--;
FACE_UPDATE_EXACT_StildeD(Nxx,Nxx_plus_2NGHOSTS,xx,out_gfs,out_gfs_exact,imax[0],imax[0]+1, imin[1],imax[1], imin[2],imax[2], MAXFACE,NUL,NUL);
imax[0]++;
//FACE_UPDATE_EXACT_StildeD(Nxx,Nxx_plus_2NGHOSTS,xx,out_gfs,out_gfs_exact,imin[0],imax[0], imin[1]-1,imin[1], imin[2],imax[2], NUL,MINFACE,NUL);
imin[1]--;
//FACE_UPDATE_EXACT_StildeD(Nxx,Nxx_plus_2NGHOSTS,xx,out_gfs,out_gfs_exact,imin[0],imax[0], imax[1],imax[1]+1, imin[2],imax[2], NUL,MAXFACE,NUL);
imax[1]++;
//FACE_UPDATE_EXACT_StildeD(Nxx,Nxx_plus_2NGHOSTS,xx,out_gfs,out_gfs_exact,imin[0],imax[0], imin[1],imax[1], imin[2]-1,imin[2], NUL,NUL,MINFACE);
imin[2]--;
//FACE_UPDATE_EXACT_StildeD(Nxx,Nxx_plus_2NGHOSTS,xx,out_gfs,out_gfs_exact,imin[0],imax[0], imin[1],imax[1], imax[2],imax[2]+1, NUL,NUL,MAXFACE);
imax[2]++;
}
}*/
""")
# Step 0: Add NRPy's directory to the path
# https://stackoverflow.com/questions/16780014/import-file-from-parent-directory
import os, sys
nrpy_dir_path = os.path.join("..")
if nrpy_dir_path not in sys.path:
sys.path.append(nrpy_dir_path)
import cmdline_helper as cmd # NRPy+: Multi-platform Python command-line interface
Ccodesdir = "GiRaFFE_standalone_Ccodes/A2B"
cmd.mkdir(os.path.join(Ccodesdir))
def GiRaFFE_NRPy_A2B(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(os.path.join(Ccodesdir, "compute_B_and_Bstagger_from_A.h"),
"w") as file:
file.write("""#define LOOP_DEFINE_SIMPLE \\
_Pragma("omp parallel for") \\
for(int k=0;k<Nxx_plus_2NGHOSTS2;k++) \\
for(int j=0;j<Nxx_plus_2NGHOSTS1;j++) \\
for(int i=0;i<Nxx_plus_2NGHOSTS0;i++)
void GiRaFFE_compute_B_and_Bstagger_from_A(const paramstruct *params,
const REAL *gxx, const REAL *gxy, const REAL *gxz, const REAL *gyy, const REAL *gyz,const REAL *gzz,
REAL *psi3_bssn, const REAL *Ax, const REAL *Ay, const REAL *Az,
REAL *Bx, REAL *By, REAL *Bz, REAL *Bx_stagger, REAL *By_stagger, REAL *Bz_stagger) {
#include "../set_Cparameters.h"
LOOP_DEFINE_SIMPLE {
const int index=IDX3S(i,j,k);
def GiRaFFE_NRPy_Source_Terms(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(
os.path.join(Ccodesdir,
"Lorenz_psi6phi_rhs__add_gauge_terms_to_A_i_rhs.h"),
"w") as file:
file.write(
"""static inline REAL avg(const REAL f[PLUS2+1][PLUS2+1][PLUS2+1],const int imin,const int imax, const int jmin,const int jmax, const int kmin,const int kmax);
#define MINUS2 0
#define MINUS1 1
#define PLUS0 2
#define PLUS1 3
#define PLUS2 4
// The "I" suffix denotes interpolation. In other words, these
// definitions are used for interpolation ONLY. The order here
// matters as well!
static const int SHIFTXI=0,SHIFTYI=1,SHIFTZI=2,GUPXXI=3,GUPXYI=4,GUPXZI=5,GUPYYI=6,GUPYZI=7,GUPZZI=8,
PSII=9,LAPM1I=10,A_XI=11,A_YI=12,A_ZI=13,LAPSE_PSI2I=14,LAPSE_OVER_PSI6I=15;
#define MAXNUMINTERP 16
// GiRaFFE_NRPy does not store the inverse metric. So, the actual inputs to the function will be
// just the metric, which we can invert in an early step. To keep things consistent, we'll label the
// components with the same labels as the inverse:
static const int GXXI=3,GXYI=4,GXZI=5,GYYI=6,GYZI=7,GZZI=8;
static void Lorenz_psi6phi_rhs__add_gauge_terms_to_A_i_rhs(const paramstruct *params,REAL **in_vars,const REAL *psi6phi,
/* TEMPS: */
REAL *shiftx_iphjphkph,REAL *shifty_iphjphkph,REAL *shiftz_iphjphkph,
REAL *alpha_iphjphkph,REAL *alpha_Phi_minus_betaj_A_j_iphjphkph,REAL *alpha_sqrtg_Ax_interp,
REAL *alpha_sqrtg_Ay_interp,REAL *alpha_sqrtg_Az_interp,
/* END TEMPS, 8 total! */
REAL *psi6phi_rhs,REAL *Ax_rhs,REAL *Ay_rhs,REAL *Az_rhs) {
#include "../set_Cparameters.h"
/* Compute
* \\partial_t psi6phi = -\\partial_j ( \\alpha \\sqrt{\\gamma} A^j - \\beta^j psi6phi)
* (Eq 13 of http://arxiv.org/pdf/1110.4633.pdf), using Lorenz gauge.
* Note that the RHS consists of a shift advection term on psi6phi and
* a term depending on the vector potential.
* psi6phi is defined at (i+1/2,j+1/2,k+1/2), but instead of reconstructing
* to compute the RHS of \\partial_t psi6phi, we instead use standard
* interpolations.
*/
// The stencil here is {-1,1},{-1,1},{-1,1} for x,y,z directions, respectively.
// Note that ALL input variables are defined at ALL gridpoints, so no
// worries about ghostzones.
#pragma omp parallel for
for(int k=1;k<Nxx_plus_2NGHOSTS2-1;k++) for(int j=1;j<Nxx_plus_2NGHOSTS1-1;j++) for(int i=1;i<Nxx_plus_2NGHOSTS0-1;i++) {
const int index=IDX3S(i,j,k);
REAL INTERP_VARS[MAXNUMINTERP][PLUS2+1][PLUS2+1][PLUS2+1];
// First compute \\partial_j \\alpha \\sqrt{\\gamma} A^j (RHS of \\partial_i psi6phi)
// FIXME: Would be much cheaper & easier to unstagger A_i, raise, then interpolate A^i.
// However, we keep it this way to be completely compatible with the original
// Illinois GRMHD thorn, called mhd_evolve.
//
//Step 1) j=x: Need to raise A_i, but to do that, we must have all variables at the same gridpoints:
// The goal is to compute \\partial_j (\\alpha \\sqrt{\\gamma} A^j) at (i+1/2,j+1/2,k+1/2)
// We do this by first interpolating (RHS1x) = (\\alpha \\sqrt{\\gamma} A^x) at
// (i,j+1/2,k+1/2)and (i+1,j+1/2,k+1/2), then taking \\partial_x (RHS1x) =
// [ RHS1x(i+1,j+1/2,k+1/2) - RHS1x(i,j+1/2,k+1/2) ]/dX.
// First bring gup's, psi, and alpha to (i,j+1/2,k+1/2):
int num_vars_to_interp;
int vars_to_interpolate[MAXNUMINTERP] = {GUPXXI,GUPXYI,GUPXZI,GUPYYI,GUPYZI,GUPZZI,LAPM1I,PSII,SHIFTXI,SHIFTYI,SHIFTZI};
num_vars_to_interp = 11;
// We may set interp_limits to be more general than we need.
int interp_limits[6] = {-1,1,-1,1,-1,1}; SET_INDEX_ARRAYS_NRPY_3DBLOCK(interp_limits);
//SET_INDEX_ARRAYS_NRPY_3DBLOCK(interp_limits);
// for(int ww=0;ww<num_vars_to_interp;ww++) {
// int whichvar=vars_to_interpolate[ww];
// // Read in variable at interp. stencil points from main memory, store in INTERP_VARS.
// for(int kk=PLUS0;kk<=PLUS1;kk++) for(int jj=PLUS0;jj<=PLUS1;jj++) for(int ii=PLUS0;ii<=PLUS1;ii++) {
// INTERP_VARS[whichvar][kk][jj][ii] = in_vars[whichvar][index_arr_3DB[kk][jj][ii]]; }
// }
// Major change here to invert the metric on the spot!
// Read in variable at interp. stencil points from main memory, store in INTERP_VARS.
for(int kk=PLUS0;kk<=PLUS1;kk++) for(int jj=PLUS0;jj<=PLUS1;jj++) for(int ii=PLUS0;ii<=PLUS1;ii++) {
// First, we will read in each component of the metric, then find the determinant.
// We write the twelfth root to psi_bssn. Then, we invert the metric and store these values.
const REAL gammaDD00 = in_vars[GXXI][index_arr_3DB[kk][jj][ii]];
const REAL gammaDD01 = in_vars[GXYI][index_arr_3DB[kk][jj][ii]];
const REAL gammaDD02 = in_vars[GXZI][index_arr_3DB[kk][jj][ii]];
const REAL gammaDD11 = in_vars[GYYI][index_arr_3DB[kk][jj][ii]];
const REAL gammaDD12 = in_vars[GYZI][index_arr_3DB[kk][jj][ii]];
const REAL gammaDD22 = in_vars[GZZI][index_arr_3DB[kk][jj][ii]];
// Generated by NRPy+:
/*
* NRPy+ Finite Difference Code Generation, Step 2 of 1: Evaluate SymPy expressions and write to main memory:
*/
const double FDPart3_0 = cbrt(gammaDD00*gammaDD11*gammaDD22 - gammaDD00*((gammaDD12)*(gammaDD12)) - ((gammaDD01)*(gammaDD01))*gammaDD22 + 2*gammaDD01*gammaDD02*gammaDD12 - ((gammaDD02)*(gammaDD02))*gammaDD11);
const double FDPart3_1 = (1.0/(FDPart3_0));
const REAL gamma_bssnDD00 = FDPart3_1*gammaDD00;
const REAL gamma_bssnDD01 = FDPart3_1*gammaDD01;
const REAL gamma_bssnDD02 = FDPart3_1*gammaDD02;
const REAL gamma_bssnDD11 = FDPart3_1*gammaDD11;
const REAL gamma_bssnDD12 = FDPart3_1*gammaDD12;
const REAL gamma_bssnDD22 = FDPart3_1*gammaDD22;
const double tmp_5 = gamma_bssnDD00*gamma_bssnDD11*gamma_bssnDD22 - gamma_bssnDD00*((gamma_bssnDD12)*(gamma_bssnDD12)) - ((gamma_bssnDD01)*(gamma_bssnDD01))*gamma_bssnDD22 + 2*gamma_bssnDD01*gamma_bssnDD02*gamma_bssnDD12 - ((gamma_bssnDD02)*(gamma_bssnDD02))*gamma_bssnDD11;
const double tmp_6 = (1.0/(tmp_5));
INTERP_VARS[GUPXXI][kk][jj][ii] = tmp_6*(gamma_bssnDD11*gamma_bssnDD22 - ((gamma_bssnDD12)*(gamma_bssnDD12)));
INTERP_VARS[GUPXYI][kk][jj][ii] = tmp_6*(-gamma_bssnDD01*gamma_bssnDD22 + gamma_bssnDD02*gamma_bssnDD12);
INTERP_VARS[GUPXZI][kk][jj][ii] = tmp_6*(gamma_bssnDD01*gamma_bssnDD12 - gamma_bssnDD02*gamma_bssnDD11);
INTERP_VARS[GUPYYI][kk][jj][ii] = tmp_6*(gamma_bssnDD00*gamma_bssnDD22 - ((gamma_bssnDD02)*(gamma_bssnDD02)));
INTERP_VARS[GUPYZI][kk][jj][ii] = tmp_6*(-gamma_bssnDD00*gamma_bssnDD12 + gamma_bssnDD01*gamma_bssnDD02);
INTERP_VARS[GUPZZI][kk][jj][ii] = tmp_6*(gamma_bssnDD00*gamma_bssnDD11 - ((gamma_bssnDD01)*(gamma_bssnDD01)));
INTERP_VARS[PSII][kk][jj][ii] = pow(FDPart3_0,1.0/4.0);
// Now, we read in the lapse function.
int whichvar=vars_to_interpolate[6];
INTERP_VARS[whichvar][kk][jj][ii] = in_vars[whichvar][index_arr_3DB[kk][jj][ii]]-1.0; // Input alpha, expect alpha-1
// Finally, we read in the shift vector into the array.
for(int ww=8;ww<num_vars_to_interp;ww++) {
int whichvar=vars_to_interpolate[ww];
INTERP_VARS[whichvar][kk][jj][ii] = in_vars[whichvar][index_arr_3DB[kk][jj][ii]];
}
}
// Next set \\alpha at (i+1/2,j+1/2,k+1/2). Will come in handy when computing damping term later.
alpha_iphjphkph[index] = avg(INTERP_VARS[LAPM1I] , PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS1)+1.0;
//A_x needs a stencil s.t. interp_limits={0,1,-1,1,-1,1}:
for(int kk=MINUS1;kk<=PLUS1;kk++) for(int jj=MINUS1;jj<=PLUS1;jj++) for(int ii=PLUS0;ii<=PLUS1;ii++) {
INTERP_VARS[A_XI][kk][jj][ii] = in_vars[A_XI][index_arr_3DB[kk][jj][ii]]; }
//A_y needs a stencil s.t. interp_limits={-1,1,0,1,-1,1}:
for(int kk=MINUS1;kk<=PLUS1;kk++) for(int jj=PLUS0;jj<=PLUS1;jj++) for(int ii=MINUS1;ii<=PLUS1;ii++) {
INTERP_VARS[A_YI][kk][jj][ii] = in_vars[A_YI][index_arr_3DB[kk][jj][ii]]; }
//A_z needs a stencil s.t. interp_limits={-1,1,-1,1,0,1}:
for(int kk=PLUS0;kk<=PLUS1;kk++) for(int jj=MINUS1;jj<=PLUS1;jj++) for(int ii=MINUS1;ii<=PLUS1;ii++) {
INTERP_VARS[A_ZI][kk][jj][ii] = in_vars[A_ZI][index_arr_3DB[kk][jj][ii]]; }
// FIRST DO A^X TERM (interpolate to (i,j+1/2,k+1/2) )
// \\alpha \\sqrt{\\gamma} A^x = \\alpha psi^6 A^x (RHS of \\partial_i psi6phi)
// Note that gupij is \\tilde{\\gamma}^{ij}, so we need to multiply by \\psi^{-4}.
const REAL gupxx_jphkph = avg(INTERP_VARS[GUPXXI], PLUS0,PLUS0, PLUS0,PLUS1, PLUS0,PLUS1);
const REAL gupxy_jphkph = avg(INTERP_VARS[GUPXYI], PLUS0,PLUS0, PLUS0,PLUS1, PLUS0,PLUS1);
const REAL gupxz_jphkph = avg(INTERP_VARS[GUPXZI], PLUS0,PLUS0, PLUS0,PLUS1, PLUS0,PLUS1);
for(int kk=PLUS0;kk<=PLUS1;kk++) for(int jj=PLUS0;jj<=PLUS1;jj++) for(int ii=PLUS0;ii<=PLUS1;ii++) {
const REAL Psi2 = INTERP_VARS[PSII][kk][jj][ii]*INTERP_VARS[PSII][kk][jj][ii];
const REAL alpha = INTERP_VARS[LAPM1I][kk][jj][ii]+1.0;
INTERP_VARS[LAPSE_PSI2I][kk][jj][ii]=alpha*Psi2;
INTERP_VARS[LAPSE_OVER_PSI6I][kk][jj][ii]=alpha/(Psi2*Psi2*Psi2);
}
const REAL lapse_Psi2_jphkph = avg(INTERP_VARS[LAPSE_PSI2I], PLUS0,PLUS0, PLUS0,PLUS1, PLUS0,PLUS1);
const REAL A_x_jphkph = avg(INTERP_VARS[A_XI], PLUS0,PLUS0, PLUS0,PLUS0, PLUS0,PLUS0); // @ (i,j+1/2,k+1/2)
const REAL A_y_jphkph = avg(INTERP_VARS[A_YI],MINUS1,PLUS0, PLUS0,PLUS1, PLUS0,PLUS0); // @ (i+1/2,j,k+1/2)
const REAL A_z_jphkph = avg(INTERP_VARS[A_ZI],MINUS1,PLUS0, PLUS0,PLUS0, PLUS0,PLUS1); // @ (i+1/2,j+1/2,k)
alpha_sqrtg_Ax_interp[index] = lapse_Psi2_jphkph*
( gupxx_jphkph*A_x_jphkph + gupxy_jphkph*A_y_jphkph + gupxz_jphkph*A_z_jphkph );
// DO A^Y TERM (interpolate to (i+1/2,j,k+1/2) )
// \\alpha \\sqrt{\\gamma} A^y = \\alpha psi^6 A^y (RHS of \\partial_i psi6phi)
// Note that gupij is \\tilde{\\gamma}^{ij}, so we need to multiply by \\psi^{-4}.
const REAL gupxy_iphkph = avg(INTERP_VARS[GUPXYI], PLUS0,PLUS1, PLUS0,PLUS0, PLUS0,PLUS1);
const REAL gupyy_iphkph = avg(INTERP_VARS[GUPYYI], PLUS0,PLUS1, PLUS0,PLUS0, PLUS0,PLUS1);
const REAL gupyz_iphkph = avg(INTERP_VARS[GUPYZI], PLUS0,PLUS1, PLUS0,PLUS0, PLUS0,PLUS1);
const REAL lapse_Psi2_iphkph = avg(INTERP_VARS[LAPSE_PSI2I], PLUS0,PLUS1, PLUS0,PLUS0, PLUS0,PLUS1);
//REAL lapse_iphkph = avg(INTERP_VARS[LAPM1I], PLUS0,PLUS1, PLUS0,PLUS0, PLUS0,PLUS1)+1.0;
//REAL psi_iphkph = avg(INTERP_VARS[PSII ], PLUS0,PLUS1, PLUS0,PLUS0, PLUS0,PLUS1);
const REAL A_x_iphkph = avg(INTERP_VARS[A_XI], PLUS0,PLUS1,MINUS1,PLUS0, PLUS0,PLUS0); // @ (i,j+1/2,k+1/2)
const REAL A_y_iphkph = avg(INTERP_VARS[A_YI], PLUS0,PLUS0, PLUS0,PLUS0, PLUS0,PLUS0); // @ (i+1/2,j,k+1/2)
const REAL A_z_iphkph = avg(INTERP_VARS[A_ZI], PLUS0,PLUS0,MINUS1,PLUS0, PLUS0,PLUS1); // @ (i+1/2,j+1/2,k)
alpha_sqrtg_Ay_interp[index] = lapse_Psi2_iphkph*
( gupxy_iphkph*A_x_iphkph + gupyy_iphkph*A_y_iphkph + gupyz_iphkph*A_z_iphkph );
// DO A^Z TERM (interpolate to (i+1/2,j+1/2,k) )
// \\alpha \\sqrt{\\gamma} A^z = \\alpha psi^6 A^z (RHS of \\partial_i psi6phi)
// Note that gupij is \\tilde{\\gamma}^{ij}, so we need to multiply by \\psi^{-4}.
const REAL gupxz_iphjph = avg(INTERP_VARS[GUPXZI], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS0);
const REAL gupyz_iphjph = avg(INTERP_VARS[GUPYZI], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS0);
const REAL gupzz_iphjph = avg(INTERP_VARS[GUPZZI], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS0);
//REAL lapse_iphjph = avg(INTERP_VARS[LAPM1I], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS0)+1.0;
//REAL psi_iphjph = avg(INTERP_VARS[PSII ], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS0);
const REAL lapse_Psi2_iphjph = avg(INTERP_VARS[LAPSE_PSI2I], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS0);
const REAL A_x_iphjph = avg(INTERP_VARS[A_XI], PLUS0,PLUS1, PLUS0,PLUS0,MINUS1,PLUS0); // @ (i,j+1/2,k+1/2)
const REAL A_y_iphjph = avg(INTERP_VARS[A_YI], PLUS0,PLUS0, PLUS0,PLUS1,MINUS1,PLUS0); // @ (i+1/2,j,k+1/2)
const REAL A_z_iphjph = avg(INTERP_VARS[A_ZI], PLUS0,PLUS0, PLUS0,PLUS0, PLUS0,PLUS0); // @ (i+1/2,j+1/2,k)
alpha_sqrtg_Az_interp[index] = lapse_Psi2_iphjph*
( gupxz_iphjph*A_x_iphjph + gupyz_iphjph*A_y_iphjph + gupzz_iphjph*A_z_iphjph );
// Next set \\alpha \\Phi - \\beta^j A_j at (i+1/2,j+1/2,k+1/2):
// We add a "L" suffix to shifti_iphjphkph to denote "LOCAL", as we set
// shifti_iphjphkph[] gridfunction below.
const REAL shiftx_iphjphkphL = avg(INTERP_VARS[SHIFTXI], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS1);
const REAL shifty_iphjphkphL = avg(INTERP_VARS[SHIFTYI], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS1);
const REAL shiftz_iphjphkphL = avg(INTERP_VARS[SHIFTZI], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS1);
const REAL lapse_over_Psi6_iphjphkphL = avg(INTERP_VARS[LAPSE_OVER_PSI6I], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS1);
//REAL psi_iphjphkph = avg(INTERP_VARS[PSII ], PLUS0,PLUS1, PLUS0,PLUS1, PLUS0,PLUS1);
//REAL psi2_iphjphkph= psi_iphjphkph*psi_iphjphkph;
//REAL psi6_iphjphkph= psi2_iphjphkph*psi2_iphjphkph*psi2_iphjphkph;
const REAL A_x_iphjphkph = avg(INTERP_VARS[A_XI], PLUS0,PLUS1, PLUS0,PLUS0, PLUS0,PLUS0); // @ (i,j+1/2,k+1/2)
const REAL A_y_iphjphkph = avg(INTERP_VARS[A_YI], PLUS0,PLUS0, PLUS0,PLUS1, PLUS0,PLUS0); // @ (i+1/2,j,k+1/2)
const REAL A_z_iphjphkph = avg(INTERP_VARS[A_ZI], PLUS0,PLUS0, PLUS0,PLUS0, PLUS0,PLUS1); // @ (i+1/2,j+1/2,k)
alpha_Phi_minus_betaj_A_j_iphjphkph[index] = psi6phi[index]*lapse_over_Psi6_iphjphkphL
- (shiftx_iphjphkphL*A_x_iphjphkph + shifty_iphjphkphL*A_y_iphjphkph + shiftz_iphjphkphL*A_z_iphjphkph);
// Finally, save shifti_iphjphkph, for \\partial_j \\beta^j psi6phi
shiftx_iphjphkph[index]=shiftx_iphjphkphL;
shifty_iphjphkph[index]=shifty_iphjphkphL;
shiftz_iphjphkph[index]=shiftz_iphjphkphL;
}
// This loop requires two additional ghostzones in every direction. Hence the following loop definition:
#pragma omp parallel for
for(int k=NGHOSTS;k<Nxx_plus_2NGHOSTS2-NGHOSTS;k++) for(int j=NGHOSTS;j<Nxx_plus_2NGHOSTS1-NGHOSTS;j++) for(int i=NGHOSTS;i<Nxx_plus_2NGHOSTS0-NGHOSTS;i++) {
const int index = IDX3S(i,j,k);
// \\partial_t A_i = [reconstructed stuff] + [gauge stuff],
// where [gauge stuff] = -\\partial_i (\\alpha \\Phi - \\beta^j A_j)
const REAL alpha_Phi_minus_betaj_A_j_iphjphkphL = alpha_Phi_minus_betaj_A_j_iphjphkph[index];
// - partial_i -> - (A_{i} - A_{i-1})/dX = (A_{i-1} - A_{i})/dX, for Ax
Ax_rhs[index] += invdx0*(alpha_Phi_minus_betaj_A_j_iphjphkph[IDX3S(i-1,j,k)] - alpha_Phi_minus_betaj_A_j_iphjphkphL);
Ay_rhs[index] += invdx1*(alpha_Phi_minus_betaj_A_j_iphjphkph[IDX3S(i,j-1,k)] - alpha_Phi_minus_betaj_A_j_iphjphkphL);
Az_rhs[index] += invdx2*(alpha_Phi_minus_betaj_A_j_iphjphkph[IDX3S(i,j,k-1)] - alpha_Phi_minus_betaj_A_j_iphjphkphL);
// \\partial_t psi6phi = [shift advection term] + \\partial_j (\\alpha \\sqrt{\\gamma} A^j)
// Here we compute [shift advection term] = \\partial_j (\\beta^j psi6phi)
// Cache misses are likely more expensive than branch mispredictions here,
// which is why we use if() statements and array lookups inside the if()'s.
REAL psi6phi_rhsL=0.0;
const REAL psi6phiL=psi6phi[index];
const REAL shiftx_iphjphkphL=shiftx_iphjphkph[index];
const REAL shifty_iphjphkphL=shifty_iphjphkph[index];
const REAL shiftz_iphjphkphL=shiftz_iphjphkph[index];
// \\partial_x (\\beta^x psi6phi) :
if(shiftx_iphjphkphL < 0.0) {
psi6phi_rhsL+=0.5*invdx0*(+ shiftx_iphjphkph[IDX3S(i-2,j,k)]*psi6phi[IDX3S(i-2,j,k)]
-4.0*shiftx_iphjphkph[IDX3S(i-1,j,k)]*psi6phi[IDX3S(i-1,j,k)]
+3.0*shiftx_iphjphkphL* psi6phiL);
} else {
psi6phi_rhsL+=0.5*invdx0*(- shiftx_iphjphkph[IDX3S(i+2,j,k)]*psi6phi[IDX3S(i+2,j,k)]
+4.0*shiftx_iphjphkph[IDX3S(i+1,j,k)]*psi6phi[IDX3S(i+1,j,k)]
-3.0*shiftx_iphjphkphL* psi6phiL);
}
// \\partial_y (\\beta^y psi6phi) :
if(shifty_iphjphkphL < 0.0) {
psi6phi_rhsL+=0.5*invdx1*(+ shifty_iphjphkph[IDX3S(i,j-2,k)]*psi6phi[IDX3S(i,j-2,k)]
-4.0*shifty_iphjphkph[IDX3S(i,j-1,k)]*psi6phi[IDX3S(i,j-1,k)]
+3.0*shifty_iphjphkphL* psi6phiL);
} else {
psi6phi_rhsL+=0.5*invdx1*(- shifty_iphjphkph[IDX3S(i,j+2,k)]*psi6phi[IDX3S(i,j+2,k)]
+4.0*shifty_iphjphkph[IDX3S(i,j+1,k)]*psi6phi[IDX3S(i,j+1,k)]
-3.0*shifty_iphjphkphL* psi6phiL);
}
// \\partial_z (\\beta^z psi6phi) :
if(shiftz_iphjphkphL < 0.0) {
psi6phi_rhsL+=0.5*invdx2*(+ shiftz_iphjphkph[IDX3S(i,j,k-2)]*psi6phi[IDX3S(i,j,k-2)]
-4.0*shiftz_iphjphkph[IDX3S(i,j,k-1)]*psi6phi[IDX3S(i,j,k-1)]
+3.0*shiftz_iphjphkphL* psi6phiL);
} else {
psi6phi_rhsL+=0.5*invdx2*(- shiftz_iphjphkph[IDX3S(i,j,k+2)]*psi6phi[IDX3S(i,j,k+2)]
+4.0*shiftz_iphjphkph[IDX3S(i,j,k+1)]*psi6phi[IDX3S(i,j,k+1)]
-3.0*shiftz_iphjphkphL* psi6phiL);
}
// Next we add \\partial_j (\\alpha \\sqrt{\\gamma} A^j) to \\partial_t psi6phi:
psi6phi_rhsL+=invdx0*(alpha_sqrtg_Ax_interp[index] - alpha_sqrtg_Ax_interp[IDX3S(i+1,j,k)])
+ invdx1*(alpha_sqrtg_Ay_interp[index] - alpha_sqrtg_Ay_interp[IDX3S(i,j+1,k)])
+ invdx2*(alpha_sqrtg_Az_interp[index] - alpha_sqrtg_Az_interp[IDX3S(i,j,k+1)]);
// *GENERALIZED* LORENZ GAUGE:
// Finally, add damping factor to \\partial_t psi6phi
//subtract lambda * alpha psi^6 Phi
psi6phi_rhsL+=-xi_damping*alpha_iphjphkph[index]*psi6phiL;
psi6phi_rhs[index] = psi6phi_rhsL;
}
}
static inline REAL avg(const REAL f[PLUS2+1][PLUS2+1][PLUS2+1],const int imin,const int imax, const int jmin,const int jmax, const int kmin,const int kmax) {
REAL retval=0.0,num_in_sum=0.0;
for(int kk=kmin;kk<=kmax;kk++) for(int jj=jmin;jj<=jmax;jj++) for(int ii=imin;ii<=imax;ii++) {
retval+=f[kk][jj][ii]; num_in_sum++;
}
return retval/num_in_sum;
}
""")
def GiRaFFE_NRPy_A2B(outdir,gammaDD,AD,BU):
cmd.mkdir(outdir)
# Set spatial dimension (must be 3 for BSSN)
DIM = 3
par.set_parval_from_str("grid::DIM",DIM)
# Compute the sqrt of the three metric determinant.
import GRHD.equations as gh
gh.compute_sqrtgammaDET(gammaDD)
# Import the Levi-Civita symbol and build the corresponding tensor.
# We already have a handy function to define the Levi-Civita symbol in WeylScalars
import WeylScal4NRPy.WeylScalars_Cartesian as weyl
LeviCivitaDDD = weyl.define_LeviCivitaSymbol_rank3()
LeviCivitaUUU = ixp.zerorank3()
for i in range(DIM):
for j in range(DIM):
for k in range(DIM):
LCijk = LeviCivitaDDD[i][j][k]
#LeviCivitaDDD[i][j][k] = LCijk * sp.sqrt(gho.gammadet)
LeviCivitaUUU[i][j][k] = LCijk / gh.sqrtgammaDET
AD_dD = ixp.declarerank2("AD_dD","nosym")
BU = ixp.zerorank1()
for i in range(DIM):
for j in range(DIM):
for k in range(DIM):
BU[i] += LeviCivitaUUU[i][j][k] * AD_dD[k][j]
# Write the code to compute derivatives with shifted stencils as needed.
with open(os.path.join(outdir,"driver_AtoB.h"),"w") as file:
file.write("""void compute_A2B_in_ghostzones(const paramstruct *restrict params,REAL *restrict in_gfs,REAL *restrict auxevol_gfs,
const int i0min,const int i0max,
const int i1min,const int i1max,
const int i2min,const int i2max) {
#include "../set_Cparameters.h"
for(int i2=i2min;i2<i2max;i2++) for(int i1=i1min;i1<i1max;i1++) for(int i0=i0min;i0<i0max;i0++) {
REAL dx_Ay,dx_Az,dy_Ax,dy_Az,dz_Ax,dz_Ay;
// Check to see if we're on the +x or -x face. If so, use a downwinded- or upwinded-stencil, respectively.
// Otherwise, use a centered stencil.
if (i0 > 0 && i0 < Nxx_plus_2NGHOSTS0-1) {
dx_Ay = invdx0*(-1.0/2.0*in_gfs[IDX4S(AD1GF, i0-1,i1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD1GF, i0+1,i1,i2)]);
dx_Az = invdx0*(-1.0/2.0*in_gfs[IDX4S(AD2GF, i0-1,i1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD2GF, i0+1,i1,i2)]);
}
else if (i0==0) {
dx_Ay = invdx0*(-3.0/2.0*in_gfs[IDX4S(AD1GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD1GF, i0+1,i1,i2)] - 1.0/2.0*in_gfs[IDX4S(AD1GF, i0+2,i1,i2)]);
dx_Az = invdx0*(-3.0/2.0*in_gfs[IDX4S(AD2GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD2GF, i0+1,i1,i2)] - 1.0/2.0*in_gfs[IDX4S(AD2GF, i0+2,i1,i2)]);
}
else {
dx_Ay = invdx0*((3.0/2.0)*in_gfs[IDX4S(AD1GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD1GF, i0-1,i1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD1GF, i0-2,i1,i2)]);
dx_Az = invdx0*((3.0/2.0)*in_gfs[IDX4S(AD2GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD2GF, i0-1,i1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD2GF, i0-2,i1,i2)]);
}
// As above, but in the y direction.
if (i1 > 0 && i1 < Nxx_plus_2NGHOSTS1-1) {
dy_Ax = invdx1*(-1.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1-1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1+1,i2)]);
dy_Az = invdx1*(-1.0/2.0*in_gfs[IDX4S(AD2GF, i0,i1-1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD2GF, i0,i1+1,i2)]);
}
else if (i1==0) {
dy_Ax = invdx1*(-3.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD0GF, i0,i1+1,i2)] - 1.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1+2,i2)]);
dy_Az = invdx1*(-3.0/2.0*in_gfs[IDX4S(AD2GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD2GF, i0,i1+1,i2)] - 1.0/2.0*in_gfs[IDX4S(AD2GF, i0,i1+2,i2)]);
}
else {
dy_Ax = invdx1*((3.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD0GF, i0,i1-1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1-2,i2)]);
dy_Az = invdx1*((3.0/2.0)*in_gfs[IDX4S(AD2GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD2GF, i0,i1-1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD2GF, i0,i1-2,i2)]);
}
// As above, but in the z direction.
if (i2 > 0 && i2 < Nxx_plus_2NGHOSTS2-1) {
dz_Ax = invdx2*(-1.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1,i2-1)] + (1.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1,i2+1)]);
dz_Ay = invdx2*(-1.0/2.0*in_gfs[IDX4S(AD1GF, i0,i1,i2-1)] + (1.0/2.0)*in_gfs[IDX4S(AD1GF, i0,i1,i2+1)]);
}
else if (i2==0) {
dz_Ax = invdx2*(-3.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD0GF, i0,i1,i2+1)] - 1.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1,i2+2)]);
dz_Ay = invdx2*(-3.0/2.0*in_gfs[IDX4S(AD1GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD1GF, i0,i1,i2+1)] - 1.0/2.0*in_gfs[IDX4S(AD1GF, i0,i1,i2+2)]);
}
else {
dz_Ax = invdx2*((3.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD0GF, i0,i1,i2-1)] + (1.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1,i2-2)]);
dz_Ay = invdx2*((3.0/2.0)*in_gfs[IDX4S(AD1GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD1GF, i0,i1,i2-1)] + (1.0/2.0)*in_gfs[IDX4S(AD1GF, i0,i1,i2-2)]);
}
// Compute the magnetic field in the normal way, using the previously calculated derivatives.
const double gammaDD00 = auxevol_gfs[IDX4S(GAMMADD00GF, i0,i1,i2)];
const double gammaDD01 = auxevol_gfs[IDX4S(GAMMADD01GF, i0,i1,i2)];
const double gammaDD02 = auxevol_gfs[IDX4S(GAMMADD02GF, i0,i1,i2)];
const double gammaDD11 = auxevol_gfs[IDX4S(GAMMADD11GF, i0,i1,i2)];
const double gammaDD12 = auxevol_gfs[IDX4S(GAMMADD12GF, i0,i1,i2)];
const double gammaDD22 = auxevol_gfs[IDX4S(GAMMADD22GF, i0,i1,i2)];
/*
* NRPy+ Finite Difference Code Generation, Step 2 of 1: Evaluate SymPy expressions and write to main memory:
*/
const double invsqrtg = 1.0/sqrt(gammaDD00*gammaDD11*gammaDD22
- gammaDD00*gammaDD12*gammaDD12
+ 2*gammaDD01*gammaDD02*gammaDD12
- gammaDD11*gammaDD02*gammaDD02
- gammaDD22*gammaDD01*gammaDD01);
auxevol_gfs[IDX4S(BU0GF, i0,i1,i2)] = (dy_Az-dz_Ay)*invsqrtg;
auxevol_gfs[IDX4S(BU1GF, i0,i1,i2)] = (dz_Ax-dx_Az)*invsqrtg;
auxevol_gfs[IDX4S(BU2GF, i0,i1,i2)] = (dx_Ay-dy_Ax)*invsqrtg;
}
}
""")
# Now, we'll also write some more auxiliary functions to handle the order-lowering method for A2B
with open(os.path.join(outdir,"driver_AtoB.h"),"a") as file:
file.write("""REAL relative_error(REAL a, REAL b) {
if((a+b)!=0.0) {
return 2.0*fabs(a-b)/fabs(a+b);
}
else {
return 0.0;
}
}
#define M2 0
#define M1 1
#define P0 2
#define P1 3
#define P2 4
#define CN4 0
#define CN2 1
#define UP2 2
#define DN2 3
#define UP1 4
#define DN1 5
void compute_Bx_pointwise(REAL *Bx, const REAL invdy, const REAL *Ay, const REAL invdz, const REAL *Az) {
REAL dz_Ay,dy_Az;
dz_Ay = invdz*((Ay[P1]-Ay[M1])*2.0/3.0 - (Ay[P2]-Ay[M2])/12.0);
dy_Az = invdy*((Az[P1]-Az[M1])*2.0/3.0 - (Az[P2]-Az[M2])/12.0);
Bx[CN4] = dy_Az - dz_Ay;
dz_Ay = invdz*(Ay[P1]-Ay[M1])/2.0;
dy_Az = invdy*(Az[P1]-Az[M1])/2.0;
Bx[CN2] = dy_Az - dz_Ay;
dz_Ay = invdz*(-1.5*Ay[P0]+2.0*Ay[P1]-0.5*Ay[P2]);
dy_Az = invdy*(-1.5*Az[P0]+2.0*Az[P1]-0.5*Az[P2]);
Bx[UP2] = dy_Az - dz_Ay;
dz_Ay = invdz*(1.5*Ay[P0]-2.0*Ay[M1]+0.5*Ay[M2]);
dy_Az = invdy*(1.5*Az[P0]-2.0*Az[M1]+0.5*Az[M2]);
Bx[DN2] = dy_Az - dz_Ay;
dz_Ay = invdz*(Ay[P1]-Ay[P0]);
dy_Az = invdy*(Az[P1]-Az[P0]);
Bx[UP1] = dy_Az - dz_Ay;
dz_Ay = invdz*(Ay[P0]-Ay[M1]);
dy_Az = invdy*(Az[P0]-Az[M1]);
Bx[DN1] = dy_Az - dz_Ay;
}
#define TOLERANCE_A2B 1.0e-4
REAL find_accepted_Bx_order(REAL *Bx) {
REAL accepted_val = Bx[CN4];
REAL Rel_error_o2_vs_o4 = relative_error(Bx[CN2],Bx[CN4]);
REAL Rel_error_oCN2_vs_oDN2 = relative_error(Bx[CN2],Bx[DN2]);
REAL Rel_error_oCN2_vs_oUP2 = relative_error(Bx[CN2],Bx[UP2]);
if(Rel_error_o2_vs_o4 > TOLERANCE_A2B) {
accepted_val = Bx[CN2];
if(Rel_error_o2_vs_o4 > Rel_error_oCN2_vs_oDN2 || Rel_error_o2_vs_o4 > Rel_error_oCN2_vs_oUP2) {
// Should we use AND or OR in if statement?
if(relative_error(Bx[UP2],Bx[UP1]) < relative_error(Bx[DN2],Bx[DN1])) {
accepted_val = Bx[UP2];
}
else {
accepted_val = Bx[DN2];
}
}
}
return accepted_val;
}
""")
order_lowering_body = """REAL AD0_1[5],AD0_2[5],AD1_2[5],AD1_0[5],AD2_0[5],AD2_1[5];
const double gammaDD00 = auxevol_gfs[IDX4S(GAMMADD00GF, i0,i1,i2)];
const double gammaDD01 = auxevol_gfs[IDX4S(GAMMADD01GF, i0,i1,i2)];
const double gammaDD02 = auxevol_gfs[IDX4S(GAMMADD02GF, i0,i1,i2)];
const double gammaDD11 = auxevol_gfs[IDX4S(GAMMADD11GF, i0,i1,i2)];
const double gammaDD12 = auxevol_gfs[IDX4S(GAMMADD12GF, i0,i1,i2)];
const double gammaDD22 = auxevol_gfs[IDX4S(GAMMADD22GF, i0,i1,i2)];
AD0_2[M2] = in_gfs[IDX4S(AD0GF, i0,i1,i2-2)];
AD0_2[M1] = in_gfs[IDX4S(AD0GF, i0,i1,i2-1)];
AD0_1[M2] = in_gfs[IDX4S(AD0GF, i0,i1-2,i2)];
AD0_1[M1] = in_gfs[IDX4S(AD0GF, i0,i1-1,i2)];
AD0_1[P0] = AD0_2[P0] = in_gfs[IDX4S(AD0GF, i0,i1,i2)];
AD0_1[P1] = in_gfs[IDX4S(AD0GF, i0,i1+1,i2)];
AD0_1[P2] = in_gfs[IDX4S(AD0GF, i0,i1+2,i2)];
AD0_2[P1] = in_gfs[IDX4S(AD0GF, i0,i1,i2+1)];
AD0_2[P2] = in_gfs[IDX4S(AD0GF, i0,i1,i2+2)];
AD1_2[M2] = in_gfs[IDX4S(AD1GF, i0,i1,i2-2)];
AD1_2[M1] = in_gfs[IDX4S(AD1GF, i0,i1,i2-1)];
AD1_0[M2] = in_gfs[IDX4S(AD1GF, i0-2,i1,i2)];
AD1_0[M1] = in_gfs[IDX4S(AD1GF, i0-1,i1,i2)];
AD1_2[P0] = AD1_0[P0] = in_gfs[IDX4S(AD1GF, i0,i1,i2)];
AD1_0[P1] = in_gfs[IDX4S(AD1GF, i0+1,i1,i2)];
AD1_0[P2] = in_gfs[IDX4S(AD1GF, i0+2,i1,i2)];
AD1_2[P1] = in_gfs[IDX4S(AD1GF, i0,i1,i2+1)];
AD1_2[P2] = in_gfs[IDX4S(AD1GF, i0,i1,i2+2)];
AD2_1[M2] = in_gfs[IDX4S(AD2GF, i0,i1-2,i2)];
AD2_1[M1] = in_gfs[IDX4S(AD2GF, i0,i1-1,i2)];
AD2_0[M2] = in_gfs[IDX4S(AD2GF, i0-2,i1,i2)];
AD2_0[M1] = in_gfs[IDX4S(AD2GF, i0-1,i1,i2)];
AD2_0[P0] = AD2_1[P0] = in_gfs[IDX4S(AD2GF, i0,i1,i2)];
AD2_0[P1] = in_gfs[IDX4S(AD2GF, i0+1,i1,i2)];
AD2_0[P2] = in_gfs[IDX4S(AD2GF, i0+2,i1,i2)];
AD2_1[P1] = in_gfs[IDX4S(AD2GF, i0,i1+1,i2)];
AD2_1[P2] = in_gfs[IDX4S(AD2GF, i0,i1+2,i2)];
const double invsqrtg = 1.0/sqrt(gammaDD00*gammaDD11*gammaDD22
- gammaDD00*gammaDD12*gammaDD12
+ 2*gammaDD01*gammaDD02*gammaDD12
- gammaDD11*gammaDD02*gammaDD02
- gammaDD22*gammaDD01*gammaDD01);
REAL BU0[4],BU1[4],BU2[4];
compute_Bx_pointwise(BU0,invdx2,AD1_2,invdx1,AD2_1);
compute_Bx_pointwise(BU1,invdx0,AD2_0,invdx2,AD0_2);
compute_Bx_pointwise(BU2,invdx1,AD0_1,invdx0,AD1_0);
auxevol_gfs[IDX4S(BU0GF, i0,i1,i2)] = find_accepted_Bx_order(BU0)*invsqrtg;
auxevol_gfs[IDX4S(BU1GF, i0,i1,i2)] = find_accepted_Bx_order(BU1)*invsqrtg;
auxevol_gfs[IDX4S(BU2GF, i0,i1,i2)] = find_accepted_Bx_order(BU2)*invsqrtg;
"""
# Here, we'll use the outCfunction() function to output a function that will compute the magnetic field
# on the interior. Then, we'll add postloop code to handle the ghostzones.
desc="Compute the magnetic field from the vector potential everywhere, including ghostzones"
name="driver_A_to_B"
driver_Ccode = outCfunction(
outfile = "returnstring", desc=desc, name=name,
params = "const paramstruct *restrict params,REAL *restrict in_gfs,REAL *restrict auxevol_gfs",
body = fin.FD_outputC("returnstring",[lhrh(lhs=gri.gfaccess("out_gfs","BU0"),rhs=BU[0]),\
lhrh(lhs=gri.gfaccess("out_gfs","BU1"),rhs=BU[1]),\
lhrh(lhs=gri.gfaccess("out_gfs","BU2"),rhs=BU[2])]).replace("IDX4","IDX4S"),
# body = order_lowering_body,
postloop = """
int imin[3] = { NGHOSTS_A2B, NGHOSTS_A2B, NGHOSTS_A2B };
int imax[3] = { NGHOSTS+Nxx0, NGHOSTS+Nxx1, NGHOSTS+Nxx2 };
// Now, we loop over the ghostzones to calculate the magnetic field there.
for(int which_gz = 0; which_gz < NGHOSTS_A2B; which_gz++) {
// After updating each face, adjust imin[] and imax[]
// to reflect the newly-updated face extents.
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0]-1,imin[0], imin[1],imax[1], imin[2],imax[2]); imin[0]--;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imax[0],imax[0]+1, imin[1],imax[1], imin[2],imax[2]); imax[0]++;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imin[1]-1,imin[1], imin[2],imax[2]); imin[1]--;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imax[1],imax[1]+1, imin[2],imax[2]); imax[1]++;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imin[1],imax[1], imin[2]-1,imin[2]); imin[2]--;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imin[1],imax[1], imax[2],imax[2]+1); imax[2]++;
}
""",
loopopts="InteriorPoints",
rel_path_for_Cparams=os.path.join("../")).replace("=NGHOSTS","=NGHOSTS_A2B").replace("NGHOSTS+Nxx0","Nxx_plus_2NGHOSTS0-NGHOSTS_A2B").replace("NGHOSTS+Nxx1","Nxx_plus_2NGHOSTS1-NGHOSTS_A2B").replace("NGHOSTS+Nxx2","Nxx_plus_2NGHOSTS2-NGHOSTS_A2B")
with open(os.path.join(outdir,"driver_AtoB.h"),"a") as file:
file.write(driver_Ccode)
def GiRaFFE_NRPy_FCVAL(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(os.path.join(Ccodesdir,"interpolate_metric_gfs_to_cell_faces.h"),"w") as file:
file.write("""// Side note: the following values could be used for cell averaged gfs:
// am2=-1.0/12.0, am1=7.0/12.0, a0=7.0/12.0, a1=-1.0/12.0
// However, since the metric gfs store the grid point values instead of the cell average,
// the following coefficients should be used:
// am2 = -1/16, am1 = 9/16, a0 = 9/16, a1 = -1/16
// This will yield the third-order-accurate face values at m-1/2,
// using values specified at {m-2,m-1,m,m+1}
#define AM2 -0.0625
#define AM1 0.5625
#define A0 0.5625
#define A1 -0.0625
#define COMPUTE_FCVAL(METRICm2,METRICm1,METRIC,METRICp1) (AM2*(METRICm2) + AM1*(METRICm1) + A0*(METRIC) + A1*(METRICp1))
const int metric_gfs_list[10] = {GAMMADD00GF,
GAMMADD01GF,
GAMMADD02GF,
GAMMADD11GF,
GAMMADD12GF,
GAMMADD22GF,
BETAU0GF,
BETAU1GF,
BETAU2GF,
ALPHAGF};
const int metric_gfs_face_list[10] = {GAMMA_FACEDD00GF,
GAMMA_FACEDD01GF,
GAMMA_FACEDD02GF,
GAMMA_FACEDD11GF,
GAMMA_FACEDD12GF,
GAMMA_FACEDD22GF,
BETA_FACEU0GF,
BETA_FACEU1GF,
BETA_FACEU2GF,
ALPHA_FACEGF};
const int num_metric_gfs = 10;
""")
desc = "Interpolate metric gridfunctions to cell faces"
name = "interpolate_metric_gfs_to_cell_faces"
interp_Cfunc = outCfunction(
outfile = "returnstring", desc=desc, name=name,
params ="const paramstruct *params,REAL *auxevol_gfs,const int flux_dirn",
preloop =""" int in_gf,out_gf;
REAL Qm2,Qm1,Qp0,Qp1;
""" ,
body =""" for(int gf = 0;gf < num_metric_gfs;gf++) {
in_gf = metric_gfs_list[gf];
out_gf = metric_gfs_face_list[gf];
for (int i2 = 2;i2 < Nxx_plus_2NGHOSTS2-1;i2++) {
for (int i1 = 2;i1 < Nxx_plus_2NGHOSTS1-1;i1++) {
for (int i0 = 2;i0 < Nxx_plus_2NGHOSTS0-1;i0++) {
Qm2 = auxevol_gfs[IDX4S(in_gf,i0-2*kronecker_delta[flux_dirn][0],i1-2*kronecker_delta[flux_dirn][1],i2-2*kronecker_delta[flux_dirn][2])];
Qm1 = auxevol_gfs[IDX4S(in_gf,i0-kronecker_delta[flux_dirn][0],i1-kronecker_delta[flux_dirn][1],i2-kronecker_delta[flux_dirn][2])];
Qp0 = auxevol_gfs[IDX4S(in_gf,i0,i1,i2)];
Qp1 = auxevol_gfs[IDX4S(in_gf,i0+kronecker_delta[flux_dirn][0],i1+kronecker_delta[flux_dirn][1],i2+kronecker_delta[flux_dirn][2])];
auxevol_gfs[IDX4S(out_gf,i0,i1,i2)] = COMPUTE_FCVAL(Qm2,Qm1,Qp0,Qp1);
}
}
}
}
""",
rel_path_to_Cparams=os.path.join("../"))
with open(os.path.join(Ccodesdir,"interpolate_metric_gfs_to_cell_faces.h"),"a") as file:
file.write(interp_Cfunc)
def GiRaFFE_NRPy_PPM(Ccodesdir):
cmd.mkdir(Ccodesdir)
# Write out the code to a file.
with open(
os.path.join(Ccodesdir,
"reconstruct_set_of_prims_PPM_GRFFE_NRPy.c"),
"w") as file:
file.write("""/*****************************************
* PPM Reconstruction Interface.
* Zachariah B. Etienne (2013)
*
* This version of PPM implements the standard
* Colella & Woodward PPM, but in the GRFFE
* limit, where P=rho=0. Thus, e.g., ftilde=0.
*****************************************/
#define MINUS2 0
#define MINUS1 1
#define PLUS0 2
#define PLUS1 3
#define PLUS2 4
#define MAXNUMINDICES 5
// ^^^^^^^^^^^^^ Be _sure_ to define MAXNUMINDICES appropriately!
#define MIN(a,b) ( ((a) < (b)) ? (a) : (b) )
#define MAX(a,b) ( ((a) > (b)) ? (a) : (b) )
#define SQR(x) ((x) * (x))
// FIXME: Should make this zero-offset for NRPy+ standards. Probably a wrapper function for compatibility with a minimum of other changes?
const int kronecker_delta[4][3] = { { 0,0,0 },
{ 1,0,0 },
{ 0,1,0 },
{ 0,0,1 } };
// You'll find the #define's for LOOP_DEFINE and SET_INDEX_ARRAYS_NRPY inside:
#include "loop_defines_reconstruction_NRPy.h"
static inline REAL slope_limit_NRPy(const REAL dU,const REAL dUp1);
static inline void monotonize_NRPy(const REAL U,REAL Ur,REAL Ul);
static void reconstruct_set_of_prims_PPM_GRFFE_NRPy(const paramstruct *params,REAL *auxevol_gfs,const int flux_dirn,
const int num_prims_to_reconstruct,const int *which_prims_to_reconstruct,
const gf_and_gz_struct *in_prims,gf_and_gz_struct *out_prims_r,
gf_and_gz_struct *out_prims_l,REAL *temporary) {
#include "set_Cparameters.h"
const int Nxx_plus_2NGHOSTS[3] = {Nxx_plus_2NGHOSTS0,Nxx_plus_2NGHOSTS1,Nxx_plus_2NGHOSTS2};
REAL U[NUM_RECONSTRUCT_GFS][MAXNUMINDICES],dU[NUM_RECONSTRUCT_GFS][MAXNUMINDICES],slope_lim_dU[NUM_RECONSTRUCT_GFS][MAXNUMINDICES],
Ur[NUM_RECONSTRUCT_GFS][MAXNUMINDICES],Ul[NUM_RECONSTRUCT_GFS][MAXNUMINDICES];
int ijkgz_lo_hi[4][2];
for(int ww=0;ww<num_prims_to_reconstruct;ww++) {
const int whichvar=which_prims_to_reconstruct[ww];
if(in_prims[whichvar].gz_lo[flux_dirn]!=0 || in_prims[whichvar].gz_hi[flux_dirn]!=0) {
printf("TOO MANY GZ'S! WHICHVAR=%d: %d %d %d : %d %d %d DIRECTION %d",whichvar,
in_prims[whichvar].gz_lo[1],in_prims[whichvar].gz_lo[2],in_prims[whichvar].gz_lo[3],
in_prims[whichvar].gz_hi[1],in_prims[whichvar].gz_hi[2],in_prims[whichvar].gz_hi[3],flux_dirn);
exit(0);
}
// *** LOOP 1: Interpolate to Ur and Ul, which are face values ***
// You will find that Ur depends on U at MINUS1,PLUS0, PLUS1,PLUS2, and
// Ul depends on U at MINUS2,MINUS1,PLUS0,PLUS1.
// However, we define the below loop from MINUS2 to PLUS2. Why not split
// this up and get additional points? Maybe we should. In GRMHD, the
// reason is that later on, Ur and Ul depend on ftilde, which is
// defined from MINUS2 to PLUS2, so we would lose those points anyway.
// But in GRFFE, ftilde is set to zero, so there may be a potential
// for boosting performance here.
LOOP_DEFINE(2,2, Nxx_plus_2NGHOSTS,flux_dirn, ijkgz_lo_hi,in_prims[whichvar].gz_lo,in_prims[whichvar].gz_hi) {
SET_INDEX_ARRAYS_NRPY(-2,2,flux_dirn);
/* *** LOOP 1a: READ INPUT *** */
// Read in a primitive at all gridpoints between m = MINUS2 & PLUS2, where m's direction is given by flux_dirn. Store to U.
for(int ii=MINUS2;ii<=PLUS2;ii++) U[whichvar][ii] = in_prims[whichvar].gf[index_arr[flux_dirn][ii]];
/* *** LOOP 1b: DO COMPUTATION *** */
/* First, compute simple dU = U(i) - U(i-1), where direction of i
* is given by flux_dirn, and U is a primitive variable:
* {vx,vy,vz,Bx,By,Bz}. */
// Note that for Ur and Ul at i, we must compute dU(i-1),dU(i),dU(i+1),
// and dU(i+2)
dU[whichvar][MINUS1] = U[whichvar][MINUS1]- U[whichvar][MINUS2];
dU[whichvar][PLUS0] = U[whichvar][PLUS0] - U[whichvar][MINUS1];
dU[whichvar][PLUS1] = U[whichvar][PLUS1] - U[whichvar][PLUS0];
dU[whichvar][PLUS2] = U[whichvar][PLUS2] - U[whichvar][PLUS1];
// Then, compute slope-limited dU, using MC slope limiter:
slope_lim_dU[whichvar][MINUS1]=slope_limit_NRPy(dU[whichvar][MINUS1],dU[whichvar][PLUS0]);
slope_lim_dU[whichvar][PLUS0] =slope_limit_NRPy(dU[whichvar][PLUS0], dU[whichvar][PLUS1]);
slope_lim_dU[whichvar][PLUS1] =slope_limit_NRPy(dU[whichvar][PLUS1], dU[whichvar][PLUS2]);
// Finally, compute face values Ur and Ul based on the PPM prescription
// (Eq. A1 in http://arxiv.org/pdf/astro-ph/0503420.pdf, but using standard 1/6=(1.0/6.0) coefficient)
// Ur[PLUS0] represents U(i+1/2)
// We applied a simplification to the following line: Ur=U+0.5*(U(i+1)-U) + ... = 0.5*(U(i+1)+U) + ...
Ur[whichvar][PLUS0] = 0.5*(U[whichvar][PLUS1] + U[whichvar][PLUS0] ) + (1.0/6.0)*(slope_lim_dU[whichvar][PLUS0] - slope_lim_dU[whichvar][PLUS1]);
// Ul[PLUS0] represents U(i-1/2)
// We applied a simplification to the following line: Ul=U(i-1)+0.5*(U-U(i-1)) + ... = 0.5*(U+U(i-1)) + ...
Ul[whichvar][PLUS0] = 0.5*(U[whichvar][PLUS0] + U[whichvar][MINUS1]) + (1.0/6.0)*(slope_lim_dU[whichvar][MINUS1] - slope_lim_dU[whichvar][PLUS0]);
/* *** LOOP 1c: WRITE OUTPUT *** */
// Store right face values to {vxr,vyr,vzr,Bxr,Byr,Bzr},
// and left face values to {vxl,vyl,vzl,Bxl,Byl,Bzl}
out_prims_r[whichvar].gf[index_arr[flux_dirn][PLUS0]] = Ur[whichvar][PLUS0];
out_prims_l[whichvar].gf[index_arr[flux_dirn][PLUS0]] = Ul[whichvar][PLUS0];
}
// *** LOOP 2 (REMOVED): STEEPEN RHOB. RHOB DOES NOT EXIST IN GRFFE EQUATIONS ***
}
// *** LOOP 3: FLATTEN BASED ON FTILDE AND MONOTONIZE ***
for(int ww=0;ww<num_prims_to_reconstruct;ww++) {
const int whichvar=which_prims_to_reconstruct[ww];
// ftilde() depends on P(MINUS2,MINUS1,PLUS1,PLUS2), THUS IS SET TO ZERO IN GRFFE
LOOP_DEFINE(2,2, Nxx_plus_2NGHOSTS,flux_dirn, ijkgz_lo_hi,in_prims[whichvar].gz_lo,in_prims[whichvar].gz_hi) {
SET_INDEX_ARRAYS_NRPY(0,0,flux_dirn);
U[whichvar][PLUS0] = in_prims[whichvar].gf[index_arr[flux_dirn][PLUS0]];
Ur[whichvar][PLUS0] = out_prims_r[whichvar].gf[index_arr[flux_dirn][PLUS0]];
Ul[whichvar][PLUS0] = out_prims_l[whichvar].gf[index_arr[flux_dirn][PLUS0]];
// ftilde_gf was computed in the function compute_ftilde_gf(), called before this routine
//REAL ftilde = ftilde_gf[index_arr[flux_dirn][PLUS0]];
// ...and then flatten (local operation)
Ur[whichvar][PLUS0] = Ur[whichvar][PLUS0];
Ul[whichvar][PLUS0] = Ul[whichvar][PLUS0];
// Then monotonize
monotonize_NRPy(U[whichvar][PLUS0],Ur[whichvar][PLUS0],Ul[whichvar][PLUS0]);
out_prims_r[whichvar].gf[index_arr[flux_dirn][PLUS0]] = Ur[whichvar][PLUS0];
out_prims_l[whichvar].gf[index_arr[flux_dirn][PLUS0]] = Ul[whichvar][PLUS0];
}
// Note: ftilde=0 in GRFFE. Ur depends on ftilde, which depends on points of U between MINUS2 and PLUS2
out_prims_r[whichvar].gz_lo[flux_dirn]+=2;
out_prims_r[whichvar].gz_hi[flux_dirn]+=2;
// Note: ftilde=0 in GRFFE. Ul depends on ftilde, which depends on points of U between MINUS2 and PLUS2
out_prims_l[whichvar].gz_lo[flux_dirn]+=2;
out_prims_l[whichvar].gz_hi[flux_dirn]+=2;
}
// *** LOOP 4: SHIFT Ur AND Ul ***
/* Currently face values are set so that
* a) Ur(i) represents U(i+1/2), and
* b) Ul(i) represents U(i-1/2)
* Here, we shift so that the indices are consistent:
* a) U(i-1/2+epsilon) = oldUl(i) = newUr(i)
* b) U(i-1/2-epsilon) = oldUr(i-1) = newUl(i)
* Note that this step is not strictly necessary if you keep
* track of indices when computing the flux. */
for(int ww=0;ww<num_prims_to_reconstruct;ww++) {
const int whichvar=which_prims_to_reconstruct[ww];
LOOP_DEFINE(3,2, Nxx_plus_2NGHOSTS,flux_dirn, ijkgz_lo_hi,in_prims[whichvar].gz_lo,in_prims[whichvar].gz_hi) {
SET_INDEX_ARRAYS_NRPY(-1,0,flux_dirn);
temporary[index_arr[flux_dirn][PLUS0]] = out_prims_r[whichvar].gf[index_arr[flux_dirn][MINUS1]];
}
LOOP_DEFINE(3,2, Nxx_plus_2NGHOSTS,flux_dirn, ijkgz_lo_hi,in_prims[whichvar].gz_lo,in_prims[whichvar].gz_hi) {
SET_INDEX_ARRAYS_NRPY(0,0,flux_dirn);
// Then shift so that Ur represents the gridpoint at i-1/2+epsilon,
// and Ul represents the gridpoint at i-1/2-epsilon.
// Ur(i-1/2) = Ul(i-1/2) = U(i-1/2+epsilon)
// Ul(i-1/2) = Ur(i+1/2 - 1) = U(i-1/2-epsilon)
out_prims_r[whichvar].gf[index_arr[flux_dirn][PLUS0]] = out_prims_l[whichvar].gf[index_arr[flux_dirn][PLUS0]];
out_prims_l[whichvar].gf[index_arr[flux_dirn][PLUS0]] = temporary[index_arr[flux_dirn][PLUS0]];
}
// Ul was just shifted, so we lost another ghostzone.
out_prims_l[whichvar].gz_lo[flux_dirn]+=1;
out_prims_l[whichvar].gz_hi[flux_dirn]+=0;
// As for Ur, we didn't need to get rid of another ghostzone,
// but we did ... seems wasteful!
out_prims_r[whichvar].gz_lo[flux_dirn]+=1;
out_prims_r[whichvar].gz_hi[flux_dirn]+=0;
}
}
// Set SLOPE_LIMITER_COEFF = 2.0 for MC, 1 for minmod
#define SLOPE_LIMITER_COEFF 2.0
//Eq. 60 in JOURNAL OF COMPUTATIONAL PHYSICS 123, 1-14 (1996)
// [note the factor of 2 missing in the |a_{j+1} - a_{j}| term].
// Recall that dU = U_{i} - U_{i-1}.
static inline REAL slope_limit_NRPy(const REAL dU,const REAL dUp1) {
if(dU*dUp1 > 0.0) {
//delta_m_U=0.5 * [ (u_(i+1)-u_i) + (u_i-u_(i-1)) ] = (u_(i+1) - u_(i-1))/2 <-- first derivative, second-order; this should happen most of the time (smooth flows)
const REAL delta_m_U = 0.5*(dU + dUp1);
// EXPLANATION OF BELOW LINE OF CODE.
// In short, sign_delta_a_j = sign(delta_m_U) = (0.0 < delta_m_U) - (delta_m_U < 0.0).
// If delta_m_U>0, then (0.0 < delta_m_U)==1, and (delta_m_U < 0.0)==0, so sign_delta_a_j=+1
// If delta_m_U<0, then (0.0 < delta_m_U)==0, and (delta_m_U < 0.0)==1, so sign_delta_a_j=-1
// If delta_m_U==0,then (0.0 < delta_m_U)==0, and (delta_m_U < 0.0)==0, so sign_delta_a_j=0
const int sign_delta_m_U = (0.0 < delta_m_U) - (delta_m_U < 0.0);
//Decide whether to use 2nd order derivative or first-order derivative, limiting slope.
return sign_delta_m_U*MIN(fabs(delta_m_U),MIN(SLOPE_LIMITER_COEFF*fabs(dUp1),SLOPE_LIMITER_COEFF*fabs(dU)));
}
return 0.0;
}
static inline void monotonize_NRPy(const REAL U,REAL Ur,REAL Ul) {
const REAL dU = Ur - Ul;
const REAL mU = 0.5*(Ur+Ul);
if ( (Ur-U)*(U-Ul) <= 0.0) {
Ur = U;
Ul = U;
return;
}
if ( dU*(U-mU) > (1.0/6.0)*SQR(dU)) {
Ul = 3.0*U - 2.0*Ur;
return;
}
if ( dU*(U-mU) < -(1.0/6.0)*SQR(dU)) {
Ur = 3.0*U - 2.0*Ul;
return;
}
}
""")
with open(os.path.join(Ccodesdir, "loop_defines_reconstruction_NRPy.h"),
"w") as file:
file.write("""#ifndef loop_defines_reconstruction_NRPy_H_
#define loop_defines_reconstruction_NRPy_H_
#define LOOP_DEFINE(gz_shift_lo,gz_shift_hi, ext,flux_dirn, ijkgz_lo_hi,gz_lo,gz_hi) \\
for(int rr=1;rr<=3;rr++) { \\
ijkgz_lo_hi[rr][0]= gz_lo[rr]; \\
ijkgz_lo_hi[rr][1]=ext[rr-1]-gz_hi[rr]; \\
} \\
ijkgz_lo_hi[flux_dirn][0] += gz_shift_lo; \\
ijkgz_lo_hi[flux_dirn][1] -= gz_shift_hi; \\
/* The following line is valid C99 */ \\
_Pragma("omp parallel for private(U,dU,slope_lim_dU,Ur,Ul)") \\
for(int k=ijkgz_lo_hi[3][0];k<ijkgz_lo_hi[3][1];k++) \\
for(int j=ijkgz_lo_hi[2][0];j<ijkgz_lo_hi[2][1];j++) \\
for(int i=ijkgz_lo_hi[1][0];i<ijkgz_lo_hi[1][1];i++)
// This define only sets indices.
// FIXME: benchmark with and without the if() statement.
// FIXME: try without index_arr being defined in all directions.
#define SET_INDEX_ARRAYS_NRPY(IMIN,IMAX,flux_dirn) \\
const int max_shift=(MAXNUMINDICES/2); \\
/* DEBUGGING ONLY: if(IMIN<-max_shift || IMAX>max_shift) CCTK_VError(VERR_DEF_PARAMS,"FIX MAXNUMINDICES!"); */ \\
int index_arr[4][MAXNUMINDICES]; \\
for(int idx=IMIN;idx<=IMAX;idx++) { \\
index_arr[flux_dirn][idx+max_shift]= \\
IDX3S( \\
i+idx*kronecker_delta[flux_dirn][0], \\
j+idx*kronecker_delta[flux_dirn][1], \\
k+idx*kronecker_delta[flux_dirn][2]); \\
}
#define SET_INDEX_ARRAYS_NRPY_3DBLOCK(IJKLOHI) \\
const int max_shift=(MAXNUMINDICES/2); \\
int index_arr_3DB[MAXNUMINDICES][MAXNUMINDICES][MAXNUMINDICES]; \\
for(int idx_k=IJKLOHI[4];idx_k<=IJKLOHI[5];idx_k++) for(int idx_j=IJKLOHI[2];idx_j<=IJKLOHI[3];idx_j++) for(int idx_i=IJKLOHI[0];idx_i<=IJKLOHI[1];idx_i++) { \\
index_arr_3DB[idx_k+max_shift][idx_j+max_shift][idx_i+max_shift]=CCTK_GFINDEX3D(cctkGH,i+idx_i,j+idx_j,k+idx_k); \\
}
#endif /* loop_defines_reconstruction_NRPy_H_ */
""")
def GiRaFFE_NRPy_A2B(outdir):
cmd.mkdir(outdir)
# Set spatial dimension (must be 3 for BSSN)
DIM = 3
par.set_parval_from_str("grid::DIM", DIM)
# Register the gridfunction gammadet. This determinant will be calculated separately
gammadet = gri.register_gridfunctions("AUXEVOL", "gammadet")
# Import the Levi-Civita symbol and build the corresponding tensor.
# We already have a handy function to define the Levi-Civita symbol in WeylScalars
import WeylScal4NRPy.WeylScalars_Cartesian as weyl
LeviCivitaDDD = weyl.define_LeviCivitaSymbol_rank3()
LeviCivitaUUU = ixp.zerorank3()
for i in range(DIM):
for j in range(DIM):
for k in range(DIM):
LCijk = LeviCivitaDDD[i][j][k]
#LeviCivitaDDD[i][j][k] = LCijk * sp.sqrt(gho.gammadet)
LeviCivitaUUU[i][j][k] = LCijk / sp.sqrt(gammadet)
# Step 1.a: A useful function
AD = ixp.register_gridfunctions_for_single_rank1("EVOL", "AD")
BU = ixp.register_gridfunctions_for_single_rank1("AUXEVOL", "BU")
AD_dD = ixp.declarerank2("AD_dD", "nosym")
BU = ixp.zerorank1(
) # BU is already registered as a gridfunction, but we need to zero its values and declare it in this scope.
for i in range(DIM):
for j in range(DIM):
for k in range(DIM):
BU[i] += LeviCivitaUUU[i][j][k] * AD_dD[k][j]
# Write the code to compute derivatives with shifted stencils as needed.
with open(os.path.join(outdir, "driver_AtoB.h"), "w") as file:
file.write(
"""void compute_A2B_in_ghostzones(const paramstruct *restrict params,REAL *restrict in_gfs,REAL *restrict auxevol_gfs,
const int i0min,const int i0max,
const int i1min,const int i1max,
const int i2min,const int i2max) {
#include "set_Cparameters.h"
for(int i2=i2min;i2<i2max;i2++) for(int i1=i1min;i1<i1max;i1++) for(int i0=i0min;i0<i0max;i0++) {
REAL dx_Ay,dx_Az,dy_Ax,dy_Az,dz_Ax,dz_Ay;
// Check to see if we're on the +x or -x face. If so, use a downwinded- or upwinded-stencil, respectively.
// Otherwise, use a centered stencil.
if (i0 > 0 && i0 < Nxx_plus_2NGHOSTS0-1) {
dx_Ay = invdx0*(-1.0/2.0*in_gfs[IDX4S(AD1GF, i0-1,i1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD1GF, i0+1,i1,i2)]);
dx_Az = invdx0*(-1.0/2.0*in_gfs[IDX4S(AD2GF, i0-1,i1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD2GF, i0+1,i1,i2)]);
}
else if (i0==0) {
dx_Ay = invdx0*(-3.0/2.0*in_gfs[IDX4S(AD1GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD1GF, i0+1,i1,i2)] - 1.0/2.0*in_gfs[IDX4S(AD1GF, i0+2,i1,i2)]);
dx_Az = invdx0*(-3.0/2.0*in_gfs[IDX4S(AD2GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD2GF, i0+1,i1,i2)] - 1.0/2.0*in_gfs[IDX4S(AD2GF, i0+2,i1,i2)]);
}
else {
dx_Ay = invdx0*((3.0/2.0)*in_gfs[IDX4S(AD1GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD1GF, i0-1,i1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD1GF, i0-2,i1,i2)]);
dx_Az = invdx0*((3.0/2.0)*in_gfs[IDX4S(AD2GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD2GF, i0-1,i1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD2GF, i0-2,i1,i2)]);
}
// As above, but in the y direction.
if (i1 > 0 && i1 < Nxx_plus_2NGHOSTS1-1) {
dy_Ax = invdx1*(-1.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1-1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1+1,i2)]);
dy_Az = invdx1*(-1.0/2.0*in_gfs[IDX4S(AD2GF, i0,i1-1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD2GF, i0,i1+1,i2)]);
}
else if (i1==0) {
dy_Ax = invdx1*(-3.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD0GF, i0,i1+1,i2)] - 1.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1+2,i2)]);
dy_Az = invdx1*(-3.0/2.0*in_gfs[IDX4S(AD2GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD2GF, i0,i1+1,i2)] - 1.0/2.0*in_gfs[IDX4S(AD2GF, i0,i1+2,i2)]);
}
else {
dy_Ax = invdx1*((3.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD0GF, i0,i1-1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1-2,i2)]);
dy_Az = invdx1*((3.0/2.0)*in_gfs[IDX4S(AD2GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD2GF, i0,i1-1,i2)] + (1.0/2.0)*in_gfs[IDX4S(AD2GF, i0,i1-2,i2)]);
}
// As above, but in the z direction.
if (i2 > 0 && i2 < Nxx_plus_2NGHOSTS2-1) {
dz_Ax = invdx2*(-1.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1,i2-1)] + (1.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1,i2+1)]);
dz_Ay = invdx2*(-1.0/2.0*in_gfs[IDX4S(AD1GF, i0,i1,i2-1)] + (1.0/2.0)*in_gfs[IDX4S(AD1GF, i0,i1,i2+1)]);
}
else if (i2==0) {
dz_Ax = invdx2*(-3.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD0GF, i0,i1,i2+1)] - 1.0/2.0*in_gfs[IDX4S(AD0GF, i0,i1,i2+2)]);
dz_Ay = invdx2*(-3.0/2.0*in_gfs[IDX4S(AD1GF, i0,i1,i2)] + 2*in_gfs[IDX4S(AD1GF, i0,i1,i2+1)] - 1.0/2.0*in_gfs[IDX4S(AD1GF, i0,i1,i2+2)]);
}
else {
dz_Ax = invdx2*((3.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD0GF, i0,i1,i2-1)] + (1.0/2.0)*in_gfs[IDX4S(AD0GF, i0,i1,i2-2)]);
dz_Ay = invdx2*((3.0/2.0)*in_gfs[IDX4S(AD1GF, i0,i1,i2)] - 2*in_gfs[IDX4S(AD1GF, i0,i1,i2-1)] + (1.0/2.0)*in_gfs[IDX4S(AD1GF, i0,i1,i2-2)]);
}
// Compute the magnetic field in the normal way, using the previously calculated derivatives.
const REAL sqrtgammadet = sqrt(auxevol_gfs[IDX4S(GAMMADETGF, i0,i1,i2)]);
auxevol_gfs[IDX4S(BU0GF, i0,i1,i2)] = (dy_Az-dz_Ay)/sqrtgammadet;
auxevol_gfs[IDX4S(BU1GF, i0,i1,i2)] = (dz_Ax-dx_Az)/sqrtgammadet;
auxevol_gfs[IDX4S(BU2GF, i0,i1,i2)] = (dx_Ay-dy_Ax)/sqrtgammadet;
}
}
""")
# Here, we'll use the outCfunction() function to output a function that will compute the magnetic field
# on the interior. Then, we'll add postloop code to handle the ghostzones.
desc = "Compute the magnetic field from the vector potential everywhere, including ghostzones"
name = "driver_A_to_B"
driver_Ccode = outCfunction(
outfile = "returnstring", desc=desc, name=name,
params = "const paramstruct *restrict params,REAL *restrict in_gfs,REAL *restrict auxevol_gfs",
body = fin.FD_outputC("returnstring",[lhrh(lhs=gri.gfaccess("out_gfs","BU0"),rhs=BU[0]),\
lhrh(lhs=gri.gfaccess("out_gfs","BU1"),rhs=BU[1]),\
lhrh(lhs=gri.gfaccess("out_gfs","BU2"),rhs=BU[2])],
params="outCverbose=False").replace("IDX4","IDX4S"),
postloop = """
int imin[3] = { NGHOSTS_A2B, NGHOSTS_A2B, NGHOSTS_A2B };
int imax[3] = { NGHOSTS+Nxx0, NGHOSTS+Nxx1, NGHOSTS+Nxx2 };
// Now, we loop over the ghostzones to calculate the magnetic field there.
for(int which_gz = 0; which_gz < NGHOSTS_A2B; which_gz++) {
// After updating each face, adjust imin[] and imax[]
// to reflect the newly-updated face extents.
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0]-1,imin[0], imin[1],imax[1], imin[2],imax[2]); imin[0]--;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imax[0],imax[0]+1, imin[1],imax[1], imin[2],imax[2]); imax[0]++;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imin[1]-1,imin[1], imin[2],imax[2]); imin[1]--;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imax[1],imax[1]+1, imin[2],imax[2]); imax[1]++;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imin[1],imax[1], imin[2]-1,imin[2]); imin[2]--;
compute_A2B_in_ghostzones(params,in_gfs,auxevol_gfs,imin[0],imax[0], imin[1],imax[1], imax[2],imax[2]+1); imax[2]++;
}
""",
loopopts="InteriorPoints").replace("=NGHOSTS","=NGHOSTS_A2B").replace("NGHOSTS+Nxx0","Nxx_plus_2NGHOSTS0-NGHOSTS_A2B").replace("NGHOSTS+Nxx1","Nxx_plus_2NGHOSTS1-NGHOSTS_A2B").replace("NGHOSTS+Nxx2","Nxx_plus_2NGHOSTS2-NGHOSTS_A2B")
with open(os.path.join(outdir, "driver_AtoB.h"), "a") as file:
file.write(driver_Ccode)
