turbx 1.0.2__py3-none-any.whl

turbx/rgd_xpln_ccor.py

@@ -0,0 +1,701 @@
+import os
+import re
+import sys
+import timeit
+from concurrent.futures import ThreadPoolExecutor
+from pathlib import Path, PurePosixPath
+
+import h5py
+import numpy as np
+import psutil
+from mpi4py import MPI
+from tqdm import tqdm
+
+from .h5 import h5_print_contents
+from .signal import ccor
+from .utils import even_print, format_time_string
+
+# ======================================================================
+
+def _calc_ccor_xpln(self, **kwargs):
+    '''
+    calculate cross-correlation in [z,t] at every [x,y]
+    - designed for analyzing unsteady, thin planes in [x]
+    - Multithreaded with ThreadPoolExecutor()
+        - scipy.signal.csd() automatically tries to run multithreaded
+        - set OMP_NUM_THREADS=1 and pass 'n_threads' to as kwarg manually
+    '''
+    
+    if (self.rank==0):
+        verbose = True
+    else:
+        verbose = False
+    
+    if verbose: print('\n'+'rgd.calc_ccor_xpln()'+'\n'+72*'-')
+    t_start_func = timeit.default_timer()
+    
+    ## assert that the opened RGD has fsubtype 'unsteady' (i.e. is NOT a prime file)
+    if (self.fsubtype!='unsteady'):
+        raise ValueError
+    if not self.usingmpi:
+        raise NotImplementedError('function is not implemented for non-MPI usage')
+    
+    h5py_is_mpi_build = h5py.h5.get_config().mpi
+    if not h5py_is_mpi_build:
+        if verbose: print('h5py was not compiled for parallel usage! exiting.')
+        sys.exit(1)
+    
+    rx = kwargs.get('rx',1)
+    ry = kwargs.get('ry',1)
+    rz = kwargs.get('rz',1)
+    rt = kwargs.get('rt',1)
+    
+    # cy = kwargs.get('cy',1) ## number of subdivisions per rank [y] range
+    # if not isinstance(cy,int):
+    #     raise TypeError('cy should be an int')
+    # if (cy<1):
+    #     raise TypeError('cy should be an int')
+    
+    sy = kwargs.get('sy',1) ## number of [y] layers to read at a time
+    if not isinstance(sy,int) or (sy<1):
+        raise TypeError('sy should be a positive non-zero int')
+    
+    n_threads = kwargs.get('n_threads',1)
+    #try:
+    #    n_threads = int(os.environ.get('OMP_NUM_THREADS'))
+    #except TypeError: ## not set
+    #    n_threads = os.cpu_count()
+    
+    fn_h5_out = kwargs.get('fn_h5_out',None) ## filename for output HDF5 (.h5) file
+    
+    ## Debug Rank:Proc Affinity
+    #pp = psutil.Process()
+    #print(f"[Rank {self.rank}] sees CPUs: {pp.cpu_affinity()}  |  n_threads={n_threads}  |  OMP_NUM_THREADS={os.environ.get('OMP_NUM_THREADS')}")
+    
+    ## only distribute data across [y]
+    if (rx!=1):
+        raise AssertionError('rx!=1')
+    if (rz!=1):
+        raise AssertionError('rz!=1')
+    if (rt!=1):
+        raise AssertionError('rt!=1')
+    
+    if not isinstance(ry,int) or (ry<1):
+        raise ValueError('ry should be a positive non-zero int')
+    
+    ## check the choice of ranks per dimension
+    if (rx*ry*rz*rt != self.n_ranks):
+        raise AssertionError('rx*ry*rz*rt != self.n_ranks')
+    if (rx>self.nx):
+        raise AssertionError('rx>self.nx')
+    if (ry>self.ny):
+        raise AssertionError('ry>self.ny')
+    if (rz>self.nz):
+        raise AssertionError('rz>self.nz')
+    if (rt>self.nt):
+        raise AssertionError('rt>self.nt')
+    
+    if (self.ny%ry!=0):
+        raise ValueError('ny not divisible by ry')
+    
+    ## distribute 4D data over ranks --> here only in [y]
+    ryl_ = np.array_split(np.arange(self.ny,dtype=np.int64),min(ry,self.ny))
+    ryl = [[b[0],b[-1]+1] for b in ryl_ ]
+    ry1,ry2 = ryl[self.rank]
+    nyr = ry2 - ry1
+    
+    ## check all [y] ranges have same size
+    for ryl_ in ryl:
+        if not (ryl_[1]-ryl_[0]==nyr):
+            raise ValueError('[y] chunks are not even in size')
+    
+    # ## [y] sub chunk range --> cyl = list of ranges in ry1:ry2
+    # ## cy is the NUMBER of chunks for the rank sub-range
+    # cyl_ = np.array_split( np.arange(ry1,ry2) , min(cy,nyr) )
+    # cyl  = [[b[0],b[-1]+1] for b in cyl_ ]
+    # 
+    # for nyc_ in [ cyl_[1]-cyl_[0] for cyl_ in cyl ]:
+    #     if (nyc_ < 1):
+    #         #raise ValueError
+    #         print(f'rank {self.rank:d}: sub-range is <1')
+    #         self.comm.Abort(1)
+    # 
+    # if 1: ## assert that [y] sub-chunk ranges are correct
+    #     
+    #     yi = np.arange(self.ny, dtype=np.int32)
+    #     
+    #     local_indices = []
+    #     for cyl_ in cyl:
+    #         cy1, cy2 = cyl_
+    #         local_indices += [ yi_ for yi_ in yi[cy1:cy2] ]
+    #     
+    #     G = self.comm.gather([ self.rank , local_indices ], root=0)
+    #     G = self.comm.bcast(G, root=0)
+    #     
+    #     all_indices = []
+    #     for G_ in G:
+    #         all_indices += G_[1]
+    #     all_indices = np.array( sorted(all_indices), dtype=np.int32 )
+    #     
+    #     if not np.array_equal( all_indices , yi ):
+    #         raise AssertionError
+    
+    if (nyr%sy!=0):
+        raise ValueError('nyr not divisible by sy')
+    
+    ## output filename : HDF5 (.h5)
+    if (fn_h5_out is None): ## automatically determine name
+        fname_path = os.path.dirname(self.fname)
+        fname_base = os.path.basename(self.fname)
+        fname_root, fname_ext = os.path.splitext(fname_base)
+        fname_root = re.findall(r'io\S+_mpi_[0-9]+', fname_root)[0]
+        fn_h5_out_base = fname_root+'_ccor.h5'
+        fn_h5_out = str(PurePosixPath(fname_path, fn_h5_out_base))
+    if (Path(fn_h5_out).suffix != '.h5'):
+        raise ValueError(f"fn_h5_out='{str(fn_h5_out)}' must end in .h5")
+    if os.path.isfile(fn_h5_out):
+        #if (os.path.getsize(fn_h5_out) > 8*1024**3):
+        #    raise ValueError(f"fn_h5_out='{str(fn_h5_out)}' exists and is >8 [GB]. exiting for your own safety.")
+        if (fn_h5_out == self.fname):
+            raise ValueError(f"fn_h5_out='{str(fn_h5_out)}' cannot be same as input filename.")
+    
+    if verbose: even_print( 'fn_h5'      , self.fname )
+    if verbose: even_print( 'fn_h5_out'  , fn_h5_out  )
+    if verbose: print(72*'-')
+    self.comm.Barrier()
+    
+    ## the data dictionary to be pickled later
+    data = {}
+    
+    ## infile
+    fsize = os.path.getsize(self.fname)/1024**3
+    if verbose: even_print(os.path.basename(self.fname),'%0.1f [GB]'%fsize)
+    if verbose: even_print('nx',f'{self.nx:d}')
+    if verbose: even_print('ny',f'{self.ny:d}')
+    if verbose: even_print('nz',f'{self.nz:d}')
+    if verbose: even_print('nt',f'{self.nt:d}')
+    if verbose: even_print('ngp',f'{self.ngp/1e6:0.1f} [M]')
+    #if verbose: even_print('cy',f'{cy:d}')
+    if verbose: even_print('sy',f'{sy:d}')
+    if verbose: even_print('n_ranks',f'{self.n_ranks:d}')
+    if verbose: even_print('n_threads',f'{n_threads:d}')
+    if verbose: print(72*'-')
+    
+    ## 0D freestream scalars
+    lchar   = self.lchar   ; data['lchar']   = lchar
+    U_inf   = self.U_inf   ; data['U_inf']   = U_inf
+    rho_inf = self.rho_inf ; data['rho_inf'] = rho_inf
+    T_inf   = self.T_inf   ; data['T_inf']   = T_inf
+    
+    #data['M_inf'] = self.M_inf
+    data['Ma'] = self.Ma
+    data['Pr'] = self.Pr
+    
+    ## read in 1D coordinate arrays & re-dimensionalize
+    x = np.copy( self['dims/x'][()] * self.lchar )
+    y = np.copy( self['dims/y'][()] * self.lchar )
+    z = np.copy( self['dims/z'][()] * self.lchar )
+    t = np.copy( self['dims/t'][()] * self.tchar )
+    
+    nx = self.nx ; data['nx'] = nx
+    ny = self.ny ; data['ny'] = ny
+    nz = self.nz ; data['nz'] = nz
+    nt = self.nt ; data['nt'] = nt
+    
+    ## assert constant Δz
+    dz0 = np.diff(z)[0]
+    if not np.all(np.isclose(np.diff(z), dz0, rtol=1e-6)):
+        raise NotImplementedError('Δz not constant')
+    dz = np.diff(z)[0]
+    
+    ## dimensional [s]
+    dt = self.dt * self.tchar
+    np.testing.assert_allclose(dt, t[1]-t[0], rtol=1e-12, atol=1e-12)
+    
+    t_meas = self.duration * self.tchar
+    np.testing.assert_allclose(t_meas, t.max()-t.min(), rtol=1e-12, atol=1e-12)
+    
+    zrange = z.max() - z.min()
+    
+    data['x'] = x
+    data['y'] = y
+    data['z'] = z
+    
+    data['t']      = t
+    data['t_meas'] = t_meas
+    data['dt']     = dt
+    data['dz']     = dz
+    data['zrange'] = zrange
+    
+    if verbose: even_print( 'Δt/tchar'       , f'{dt/self.tchar:0.8f}' )
+    if verbose: even_print( 'Δt'             , f'{dt:0.3e} [s]'        )
+    if verbose: even_print( 'duration/tchar' , f'{self.duration:0.1f}' )
+    if verbose: even_print( 'duration'       , f'{self.duration*self.tchar:0.3e} [s]' )
+    if verbose: print(72*'-')
+    
+    ## report
+    if verbose:
+        even_print('Δt'     , f'{dt    :0.5e} [s]' )
+        even_print('t_meas' , f'{t_meas:0.5e} [s]' )
+        even_print('Δz'     , f'{dz0   :0.5e} [m]' )
+        even_print('zrange' , f'{zrange:0.5e} [m]' )
+        print(72*'-')
+    
+    ## get lags [t]
+    #lags_t,_  = ccor( np.ones(win_len,dtype=np.float32) , np.ones(win_len,dtype=np.float32), get_lags=True )
+    #n_lags_t_ = win_len*2-1
+    lags_t,_  = ccor( np.ones(nt,dtype=np.float32) , np.ones(nt,dtype=np.float32), get_lags=True )
+    n_lags_t_ = nt*2-1
+    n_lags_t  = lags_t.shape[0]
+    if (n_lags_t!=n_lags_t_):
+        raise AssertionError('check lags [t]')
+    
+    data['lags_t']   = lags_t
+    data['n_lags_t'] = n_lags_t
+    
+    if verbose:
+        even_print('n lags (Δt)' , '%i'%(n_lags_t,))
+    
+    ## get lags [z]
+    lags_z,_  = ccor( np.ones(nz,dtype=np.float32) , np.ones(nz,dtype=np.float32), get_lags=True )
+    n_lags_z_ = nz*2-1
+    n_lags_z  = lags_z.shape[0]
+    if (n_lags_z!=n_lags_z_):
+        raise AssertionError('check lags [z]')
+    
+    data['lags_z']   = lags_z
+    data['n_lags_z'] = n_lags_z
+    
+    if verbose:
+        even_print('n lags (Δz)' , '%i'%(n_lags_z,))
+    
+    # ===
+    
+    ## cross-correlation pairs
+    ## [ str:var1, str:var2, bool:do_density_weighting]
+    ccor_combis = [
+    [ 'u' , 'v' , True  ], ## [ ρ·u″ , ρ·v″ ]
+    [ 'u' , 'u' , True  ], ## [ ρ·u″ , ρ·u″ ]
+    [ 'v' , 'v' , True  ], ## [ ρ·v″ , ρ·v″ ]
+    [ 'w' , 'w' , True  ], ## [ ρ·w″ , ρ·w″ ]
+    [ 'u' , 'v' , False ], ## [ u′   , v′   ]
+    [ 'u' , 'u' , False ], ## [ u′   , u′   ]
+    [ 'v' , 'v' , False ], ## [ v′   , v′   ]
+    [ 'w' , 'w' , False ], ## [ w′   , w′   ]
+    ]
+    
+    ## generate cross-correlation scalar names
+    scalars = []
+    for ccor_combi in ccor_combis:
+        s1,s2,do_density_weighting = ccor_combi
+        if do_density_weighting:
+            scalars.append(f'r{s1}II_r{s2}II')
+        else:
+            scalars.append(f'{s1}I_{s2}I')
+    
+    ## generate avg scalar names
+    scalars_Re_avg = []
+    scalars_Fv_avg = []
+    for ccor_combi in ccor_combis:
+        s1,s2,do_density_weighting = ccor_combi
+        if do_density_weighting and ('rho' not in scalars_Re_avg):
+            scalars_Re_avg.append('rho')
+        if do_density_weighting:
+            if (s1 not in scalars_Fv_avg):
+                scalars_Fv_avg.append(s1)
+            if (s2 not in scalars_Fv_avg):
+                scalars_Fv_avg.append(s2)
+        else:
+            if (s1 not in scalars_Re_avg):
+                scalars_Re_avg.append(s1)
+            if (s2 not in scalars_Re_avg):
+                scalars_Re_avg.append(s2)
+    
+    ## numpy formatted arrays: buffers for PSD & other data (rank-local)
+    Rz         = np.zeros(shape=(nyr,n_lags_z ) , dtype={'names':scalars        , 'formats':[ np.dtype(np.float64) for s in scalars        ]})
+    Rt         = np.zeros(shape=(nyr,n_lags_t ) , dtype={'names':scalars        , 'formats':[ np.dtype(np.float64) for s in scalars        ]})
+    covariance = np.zeros(shape=(nyr,         ) , dtype={'names':scalars        , 'formats':[ np.dtype(np.float64) for s in scalars        ]})
+    avg_Re     = np.zeros(shape=(nyr,         ) , dtype={'names':scalars_Re_avg , 'formats':[ np.dtype(np.float64) for s in scalars_Re_avg ]})
+    avg_Fv     = np.zeros(shape=(nyr,         ) , dtype={'names':scalars_Fv_avg , 'formats':[ np.dtype(np.float64) for s in scalars_Fv_avg ]})
+    
+    if verbose:
+        even_print('n cross-correlation scalar combinations' , f'{len(ccor_combis):d}')
+    
+    # ==============================================================
+    # check memory
+    # ==============================================================
+    
+    hostname = MPI.Get_processor_name()
+    mem_free_gb = psutil.virtual_memory().free / 1024**3
+    G = self.comm.gather([ self.rank , hostname , mem_free_gb ], root=0)
+    G = self.comm.bcast(G, root=0)
+    
+    host_mem = {}
+    for rank, host, mem in G:
+        if host not in host_mem or mem < host_mem[host]:
+            host_mem[host] = mem
+    total_free = sum(host_mem.values())
+    
+    if verbose:
+        print(72*'-')
+        for key,value in host_mem.items():
+            even_print(f'RAM free {key}', f'{int(np.floor(value)):d} [GB]')
+        even_print('RAM free (local,min)', f'{int(np.floor(min(host_mem.values()))):d} [GB]')
+        even_print('RAM free (global)', f'{int(np.floor(total_free)):d} [GB]')
+    
+    shape_read = (nx,sy,nz,nt) ## local
+    if verbose: even_print('read shape (local)', f'[{nx:d},{sy:d},{nz:d},{nt:d}]')
+    data_gb = np.dtype(np.float64).itemsize * np.prod(shape_read) / 1024**3
+    if verbose: even_print('read size (global)', f'{int(np.ceil(data_gb*ry)):d} [GB]')
+    
+    if verbose: even_print('read size (global) ×6', f'{int(np.ceil(data_gb*ry*6)):d} [GB]')
+    ram_usage_est = data_gb*ry*6/total_free
+    if verbose: even_print('RAM usage estimate', f'{100*ram_usage_est:0.1f} [%]')
+    
+    self.comm.Barrier()
+    if (ram_usage_est>0.80):
+        print('RAM consumption might be too high. exiting.')
+        self.comm.Abort(1)
+    
+    # ==============================================================
+    # main loop
+    # ==============================================================
+    
+    if verbose:
+        progress_bar = tqdm(
+            #total=len(ccor_combis)*cy,
+            total=len(ccor_combis)*(nyr//sy),
+            ncols=100,
+            desc='ccor',
+            leave=True,
+            file=sys.stdout,
+            mininterval=0.1,
+            smoothing=0.,
+            #bar_format="\033[B{l_bar}{bar}| {n}/{total} [{percentage:.1f}%] {elapsed}/{remaining}\033[A\n\b",
+            bar_format="{l_bar}{bar}| {n}/{total} [{percentage:.1f}%] {elapsed}/{remaining}",
+            ascii="░█",
+            colour='#FF6600',
+            )
+    
+    for cci,cc in enumerate(ccor_combis): ## ccor pairs
+        
+        if verbose: tqdm.write(72*'-')
+        
+        scalar_L, scalar_R, do_density_weighting = cc
+        
+        if do_density_weighting:
+            msg = f'ccor[ρ·{scalar_L}″,ρ·{scalar_R}″]'
+        else:
+            msg = f'ccor[{scalar_L}′,{scalar_R}′]'
+        if verbose:
+            tqdm.write(even_print('computing',msg,s=True,))
+        
+        dset_L   = self[f'data/{scalar_L}']
+        dset_R   = self[f'data/{scalar_R}']
+        dset_rho = self['data/rho']
+        
+        scalar = scalars[cci]
+        
+        ## assert scalar name
+        if do_density_weighting:
+            if (f'r{scalar_L}II_r{scalar_R}II' != scalar ):
+                raise ValueError
+        else:
+            if (f'{scalar_L}I_{scalar_R}I' != scalar ):
+                raise ValueError
+        
+        # ## [y] loop outer (chunks within rank)
+        # for cyl_ in cyl:
+        #     cy1, cy2 = cyl_
+        #     nyc = cy2 - cy1
+        
+        for ci in range(nyr//sy):
+            cy1 = ry1 + ci*sy
+            cy2 = cy1 + sy
+            nyc = cy2 - cy1
+            
+            self.comm.Barrier()
+            t_start = timeit.default_timer()
+            
+            ## read data L
+            n_scalars_read = 1 ## initialize
+            scalar_str = scalar_L
+            with dset_L.collective:
+                data_L = np.copy( dset_L[:,:,cy1:cy2,:].T ).astype(np.float64)
+            
+            ## read data R (if != data L)
+            if (scalar_L==scalar_R):
+                data_R = np.copy( data_L )
+            else:
+                n_scalars_read += 1
+                scalar_str += f',{scalar_R}'
+                with dset_R.collective:
+                    data_R = np.copy( dset_R[:,:,cy1:cy2,:].T ).astype(np.float64)
+            
+            ## read ρ
+            if do_density_weighting:
+                n_scalars_read += 1
+                scalar_str += ',ρ'
+                with dset_rho.collective:
+                    rho = np.copy( dset_rho[:,:,cy1:cy2,:].T ).astype(np.float64)
+            else:
+                rho = None
+            
+            self.comm.Barrier()
+            t_delta = timeit.default_timer() - t_start
+            data_gb = n_scalars_read * ( self.nx * ry * (cy2-cy1) * self.nz * self.nt * dset_L.dtype.itemsize ) / 1024**3
+            if verbose:
+                tqdm.write(even_print(f'read: {scalar_str}', '%0.3f [GB]  %0.3f [s]  %0.3f [GB/s]'%(data_gb,t_delta,(data_gb/t_delta)), s=True))
+            
+            ## data_L and data_R should be [nx,nyc,nz,nt] where nyc is the chunk [y] range
+            if ( data_L.shape != (nx,nyc,nz,nt) ) or ( data_R.shape != (nx,nyc,nz,nt) ):
+                print(f'rank {self.rank:d}: shape violation')
+                self.comm.Abort(1)
+            if (rho is not None) and ( rho.shape != (nx,nyc,nz,nt) ):
+                print(f'rank {self.rank:d}: shape violation')
+                self.comm.Abort(1)
+            
+            # === redimensionalize
+            
+            if scalar_L in ['u','v','w',]:
+                data_L *= U_inf
+            else:
+                raise ValueError
+            
+            if scalar_R in ['u','v','w',]:
+                data_R *= U_inf
+            else:
+                raise ValueError
+            
+            if (rho is not None): ## i.e. if do_density_weighting
+                rho *= rho_inf
+            
+            # === compute mean-removed data
+            
+            ## avg(□) or avg(ρ·□)/avg(ρ) in [t]
+            if do_density_weighting:
+                rho_avg     = np.mean(        rho , axis=3, dtype=np.float64, keepdims=True) ## [x,y,z,1]
+                data_L_avg  = np.mean( rho*data_L , axis=3, dtype=np.float64, keepdims=True) ## [x,y,z,1]
+                data_L_avg /= rho_avg
+                data_R_avg  = np.mean( rho*data_R , axis=3, dtype=np.float64, keepdims=True) ## [x,y,z,1]
+                data_R_avg /= rho_avg
+            else:
+                data_L_avg = np.mean( data_L , axis=3, dtype=np.float64, keepdims=True) ## [x,y,z,1]
+                data_R_avg = np.mean( data_R , axis=3, dtype=np.float64, keepdims=True) ## [x,y,z,1]
+            
+            ## Reynolds prime □′ or Favre prime □″
+            data_L -= data_L_avg
+            data_R -= data_R_avg
+            
+            ## assert stationarity / definition averaging
+            ## avg(□′)==0 or avg(ρ·□″)==0
+            if do_density_weighting:
+                a_ = np.mean(rho*data_L, axis=3, dtype=np.float64, keepdims=True)
+                b_ = np.mean(rho*data_R, axis=3, dtype=np.float64, keepdims=True)
+            else:
+                a_ = np.mean(data_L, axis=3, dtype=np.float64, keepdims=True)
+                b_ = np.mean(data_R, axis=3, dtype=np.float64, keepdims=True)
+            if not np.allclose( a_, np.zeros_like(a_), atol=1e-6 ) or not np.allclose( b_, np.zeros_like(b_), atol=1e-6 ):
+                print(f'rank {self.rank:d}: avg(□′)!=0 or avg(ρ·□″)!=0')
+                self.comm.Abort(1)
+            
+            ## covariance: <□′·□′> OR <ρ□″·ρ□″> --> note that this is NOT the typical Favre <ρ·□″□″>
+            if do_density_weighting:
+                covariance_ = np.mean( rho*data_L * rho*data_R , axis=3 , dtype=np.float64, keepdims=True)
+            else:
+                covariance_ = np.mean( data_L*data_R , axis=3 , dtype=np.float64, keepdims=True)
+            
+            ## write this chunk/scalar's covariance to covariance buffer
+            ## avg over [x,z] : [x,y,z,1] --> [y]
+            yiA = cy1 - ry1
+            yiB = cy2 - ry1
+            covariance[scalar][yiA:yiB] = np.squeeze( np.mean( covariance_ , axis=(0,2,3) , dtype=np.float64) )
+            
+            ## write (rank-local) 1D [y] averages
+            if do_density_weighting:
+                avg_Fv[scalar_L][yiA:yiB] = np.squeeze( np.mean( data_L_avg , axis=(0,2,3) , dtype=np.float64) )
+                avg_Fv[scalar_R][yiA:yiB] = np.squeeze( np.mean( data_R_avg , axis=(0,2,3) , dtype=np.float64) )
+                avg_Re['rho'][yiA:yiB]    = np.squeeze( np.mean( rho_avg    , axis=(0,2,3) , dtype=np.float64) )
+            else:
+                avg_Re[scalar_L][yiA:yiB] = np.squeeze( np.mean( data_L_avg , axis=(0,2,3) , dtype=np.float64) )
+                avg_Re[scalar_R][yiA:yiB] = np.squeeze( np.mean( data_R_avg , axis=(0,2,3) , dtype=np.float64) )
+            
+            # ===============================================================================
+            # At this point you have 4D [x,y,z,t] [□′,□′] or [ρ·□″,ρ·□″] data
+            # ===============================================================================
+            
+            def __ccor_z_thread_kernel(xi,ti,yii,do_density_weighting):
+                if do_density_weighting:
+                    uL = rho[xi,yii,:,ti] * data_L[xi,yii,:,ti]
+                    uR = rho[xi,yii,:,ti] * data_R[xi,yii,:,ti]
+                else:
+                    uL = data_L[xi,yii,:,ti]
+                    uR = data_R[xi,yii,:,ti]
+                return xi,ti,ccor(uL,uR)
+            
+            def __ccor_t_thread_kernel(xi,zi,yii,do_density_weighting):
+                if do_density_weighting:
+                    uL = rho[xi,yii,zi,:] * data_L[xi,yii,zi,:]
+                    uR = rho[xi,yii,zi,:] * data_R[xi,yii,zi,:]
+                else:
+                    uL = data_L[xi,yii,zi,:]
+                    uR = data_R[xi,yii,zi,:]
+                return xi,zi,ccor(uL,uR)
+            
+            self.comm.Barrier()
+            t_start = timeit.default_timer()
+            
+            ## [y] loop inner (indices within chunk)
+            for yi in range(cy1,cy2):
+                
+                yii  = yi - cy1 ## chunk local
+                yiii = yi - ry1 ## rank local
+                
+                ## ccor buffers for [y] loop inner
+                R_xt = np.zeros((nx,nt,n_lags_z) , dtype=np.float64) ## [x,t] range for ccor(z)
+                R_xz = np.zeros((nx,nz,n_lags_t) , dtype=np.float64) ## [x,z] range for ccor(t)
+                
+                # ===========================================================================
+                # ccor(z) : loop over [x,t]
+                # ===========================================================================
+                
+                ## concurrent/threaded execution for ccor(z)
+                tasks = [(xi,ti,yii,do_density_weighting) for xi in range(nx) for ti in range(nt)]
+                with ThreadPoolExecutor(max_workers=n_threads) as executor:
+                    results = executor.map(lambda t: __ccor_z_thread_kernel(*t,), tasks)
+                    for xi,ti,result in results:
+                        R_xt[xi,ti,:] = result
+                
+                # for xi in range(nx):
+                #     for ti in range(nt):
+                #         
+                #         ## 1D [z] □′ or ρ·□″ vectors
+                #         if do_density_weighting:
+                #             uL = np.copy( rho[xi,yii,:,ti] * data_L[xi,yii,:,ti] )
+                #             uR = np.copy( rho[xi,yii,:,ti] * data_R[xi,yii,:,ti] )
+                #         else:
+                #             uL = np.copy( data_L[xi,yii,:,ti] )
+                #             uR = np.copy( data_R[xi,yii,:,ti] )
+                #         
+                #         R_xt[xi,ti,:] = ccor( uL , uR )
+                
+                ## avg in [x,t] & write in rank context
+                Rz[scalar][yiii,:] = np.mean(R_xt, axis=(0,1), dtype=np.float64)
+                
+                # ===========================================================================
+                # ccor(t) : loop over [x,z]
+                # ===========================================================================
+                
+                ## concurrent/threaded execution for ccor(t)
+                tasks = [(xi,zi,yii,do_density_weighting) for xi in range(nx) for zi in range(nz)]
+                with ThreadPoolExecutor(max_workers=n_threads) as executor:
+                    results = executor.map(lambda t: __ccor_t_thread_kernel(*t,), tasks)
+                    for xi,zi,result in results:
+                        R_xz[xi,zi,:] = result
+                
+                # for xi in range(nx):
+                #     for zi in range(nz):
+                #         
+                #         ## 1D [z] □′ or ρ·□″ vectors
+                #         if do_density_weighting:
+                #             uL = np.copy( rho[xi,yii,zi,:] * data_L[xi,yii,zi,:] )
+                #             uR = np.copy( rho[xi,yii,zi,:] * data_R[xi,yii,zi,:] )
+                #         else:
+                #             uL = np.copy( data_L[xi,yii,zi,:] )
+                #             uR = np.copy( data_R[xi,yii,zi,:] )
+                #         
+                #         R_xz[xi,zi,:] = ccor( uL , uR )
+                
+                ## avg in [x,z] & write in rank context
+                Rt[scalar][yiii,:] = np.mean(R_xz, axis=(0,1), dtype=np.float64)
+            
+            self.comm.Barrier()
+            t_delta = timeit.default_timer() - t_start
+            if verbose: tqdm.write(even_print(msg, format_time_string(t_delta), s=True))
+            if verbose: progress_bar.update() ## (scalar, [y] chunk) progress
+            #break ## debug
+        #break ## debug
+    if verbose: progress_bar.close()
+    self.comm.Barrier()
+    if verbose: print(72*'-')
+    
+    # ==============================================================
+    # write HDF5 (.h5) file
+    # ==============================================================
+    
+    ## overwrite outfile!
+    ## open on rank 0 and write attributes, dimensions, etc.
+    if (self.rank==0):
+        with h5py.File(fn_h5_out, 'w') as hfw:
+            
+            ## write floats,ints as top-level attributes
+            for key,val in data.items():
+                if isinstance(data[key], (int,np.int32,np.int64)):
+                    hfw.attrs[key] = val
+                elif isinstance(data[key], (float,np.float32,np.float64)):
+                    hfw.attrs[key] = val
+                elif isinstance(data[key], np.ndarray):
+                    pass
+                else:
+                    print(f'key {key} is type {str(type(data[key]))}')
+                    self.comm.Abort(1)
+            
+            ## write numpy arrays
+            hfw.create_dataset( 'dims/x'      , data=x      ) ## [m]
+            hfw.create_dataset( 'dims/y'      , data=y      ) ## [m]
+            hfw.create_dataset( 'dims/z'      , data=z      ) ## [m]
+            hfw.create_dataset( 'dims/t'      , data=t      ) ## [s]
+            hfw.create_dataset( 'dims/lags_z' , data=lags_z ) ## [m]
+            hfw.create_dataset( 'dims/lags_t' , data=lags_t ) ## [s]
+            
+            ## initialize datasets
+            for scalar in scalars:
+                hfw.create_dataset( f'covariance/{scalar}'  , shape=(ny,)         , dtype=np.float64, chunks=None         , data=np.full((ny,),0.,np.float64)         )
+                hfw.create_dataset( f'Rz/{scalar}'          , shape=(ny,n_lags_z) , dtype=np.float64, chunks=(1,n_lags_z) , data=np.full((ny,n_lags_z),0.,np.float64) )
+                hfw.create_dataset( f'Rt/{scalar}'          , shape=(ny,n_lags_t) , dtype=np.float64, chunks=(1,n_lags_t) , data=np.full((ny,n_lags_t),0.,np.float64) )
+            
+            ## initialize datasets 1D [y] mean
+            for scalar in avg_Re.dtype.names:
+                hfw.create_dataset( f'avg/Re/{scalar}', shape=(ny,), dtype=np.float64, chunks=None, data=np.full((ny,),0.,np.float64) )
+            for scalar in avg_Fv.dtype.names:
+                hfw.create_dataset( f'avg/Fv/{scalar}', shape=(ny,), dtype=np.float64, chunks=None, data=np.full((ny,),0.,np.float64) )
+    
+    self.comm.Barrier()
+    
+    with h5py.File(fn_h5_out, 'a', driver='mpio', comm=self.comm) as hfw:
+        
+        ## collectively write covariance,Rz,Rt
+        for scalar in scalars:
+            dset = hfw[f'covariance/{scalar}']
+            with dset.collective:
+                dset[ry1:ry2] = covariance[scalar][:]
+            dset = hfw[f'Rz/{scalar}']
+            with dset.collective:
+                dset[ry1:ry2,:] = Rz[scalar][:,:]
+            dset = hfw[f'Rt/{scalar}']
+            with dset.collective:
+                dset[ry1:ry2,:] = Rt[scalar][:,:]
+        
+        ## collectively write 1D [y] avgs (Reynolds,Favre)
+        for scalar in avg_Re.dtype.names:
+            dset = hfw[f'avg/Re/{scalar}']
+            with dset.collective:
+                dset[ry1:ry2] = avg_Re[scalar][:]
+        for scalar in avg_Fv.dtype.names:
+            dset = hfw[f'avg/Fv/{scalar}']
+            with dset.collective:
+                dset[ry1:ry2] = avg_Fv[scalar][:]
+    
+    ## report file contents
+    self.comm.Barrier()
+    if (self.rank==0):
+        even_print( os.path.basename(fn_h5_out) , f'{(os.path.getsize(fn_h5_out)/1024**2):0.1f} [MB]' )
+        print(72*'-')
+        with h5py.File(fn_h5_out,'r') as hfr:
+            h5_print_contents(hfr)
+    self.comm.Barrier()
+    
+    if verbose: print(72*'-')
+    if verbose: print('total time : rgd.calc_ccor_xpln() : %s'%format_time_string((timeit.default_timer() - t_start_func)))
+    if verbose: print(72*'-')
+    return