Seismic Unix coding rules
Seismic Unix has remained a stable, easy to install, and transportable package by following a few simple rules.
The first of these is that changing the SU header definitions is forbidden. The header structure looks deceptively simple, but contains a hidden trap, that it is possible to easily introduce a gap in the header that would render the SU data made under a new system mutually incompatible with that made by other versions of SU.
Suppose, however, you need another header field? The way to approach this is to use one of the less commonly used fields, define a keyword field so that the user may change this.
One of the stumbling blocks of older versions of SU was the use of subroutines written in different languages, such as Fortran, made porting the code quite difficult. Expert programmers likely will scoff at this statement, but remember, SU is designed so that non-expert users may easily install, use, and expand this package.
Many users have wondered why SU does not depend on such widely available packages such as FFTW (Fastest Fourier Transform in the West) or GSL (the GNU scientific library). The basic reason is that SU predates many of the available open source packages that users have available. In this way, SU has always been developed from the point of view of being completely self-contained.
The reality is that the more items that a user must fetch to install your code, the less likely that they will want to use it. You will also be at the mercy of whims of third party developers who may decide arbitrarily to change their code.
There are a number of “open” packages, such as Numerical Recipes, that are commercial packages, but are not really open source. It is forbidden to introduce such code into the SU collection.
The SU materials are released under a Berkeley-style open source license. This choice avoids the viral aspect of the GNU Public License. Commercial developers may think twice about making use of GPL Licensed items, because of the requirement that improvements must be returned to the originators of code, and that derivative code automatically inherits the GPL License.
The limitation is that while code released under a Berkeley License may be absorbed into a GPL software collection, the reverse is not true. You may not include GPL licensed code in the SU distribution.
If you need to pass a variable to a subroutine then pass it as an argument to the subroutine, do not define it as a global variable.
Make sure there are Notes: in the Self Documentation. Document each declared variable. Put technical information describing the algorithm being used. Remember that “future you” will not remember the things that “present you” knows.
Document subroutines as though they will be clipped out and put in a library—because if a subroutine is useful, it will be separated from its parent program.
Pretend that the code will not run without the documentation.
An example program is sufiler.c which is located in $CWPROOT/src/su/main/filters
Proceeding block by block in the code,
The copyright statement
/* Copyright (c) Colorado School of Mines, 2011.*/ /* All rights reserved. */
The RCS (revision control system) version label
/* SUFILTER: $Revision: 1.23 $ ; $Date: 2011/11/12 00:09:00 $ */
The include files that define the SU data type
#include "su.h" #include "segy.h"
The begin selfdoc flag which allows updated to capture the self documentation for tools such as suname
/*********************** self documentation **********************/
The self documentation
char *sdoc[] = {
" ",
" SUFILTER - applies a zero-phase, sine-squared tapered filter ",
" ",
" sufilter <stdin >stdout [optional parameters] ",
" ",
" Required parameters: ",
" if dt is not set in header, then dt is mandatory ",
" ",
" Optional parameters: ",
" f=f1,f2,... array of filter frequencies(HZ) ",
" amps=a1,a2,... array of filter amplitudes ",
" dt = (from header) time sampling interval (sec) ",
" verbose=0 =1 for advisory messages ",
" ",
" Defaults:f=.10*(nyquist),.15*(nyquist),.45*(nyquist),.50*(nyquist) ",
" (nyquist calculated internally) ",
" amps=0.,1.,...,1.,0. trapezoid-like bandpass filter ",
" ",
" Examples of filters: ",
" Bandpass: sufilter <data f=10,20,40,50 | ... ",
" Bandreject: sufilter <data f=10,20,30,40 amps=1.,0.,0.,1. | .. ",
" Lowpass: sufilter <data f=10,20,40,50 amps=1.,1.,0.,0. | ... ",
" Highpass: sufilter <data f=10,20,40,50 amps=0.,0.,1.,1. | ... ",
" Notch: sufilter <data f=10,12.5,35,50,60 amps=1.,.5,0.,.5,1. |.. ",
NULL};
The attribution block, authors, technical references, and header fields and other technical details go here
/* Credits: * CWP: John Stockwell, Jack Cohen * CENPET: Werner M. Heigl - added well log support * * Possible optimization: Do assignments instead of crmuls where * filter is 0.0. * * Trace header fields accessed: ns, dt, d1 */
The end self documentation flag, this tells updatedoc that this is the end of the documentation
/**************** end self doc ***********************************/
Function prototypes
/* Prototype of function used internally */
void polygonalFilter(float *f, float *amps,
int npoly, int nfft, float dt, float *filter);
internally defined constants
#define PIBY2 1.57079632679490 #define FRAC0 0.10 /* Ratio of default f1 to Nyquist */ #define FRAC1 0.15 /* Ratio of default f2 to Nyquist */ #define FRAC2 0.45 /* Ratio of default f3 to Nyquist */ #define FRAC3 0.50 /* Ratio of default f4 to Nyquist */ #define LOOKFAC 2 /* Look ahead factor for npfao */ #define PFA_MAX 720720 /* Largest allowed nfft */
Declaration of SEGY type objects in global space.
segy tr;
The main
int
main(int argc, char **argv)
{
Declarations
register float *rt; /* real trace */
register complex *ct; /* complex transformed trace */
float *filter; /* filter array */
float *f; /* array of filter frequencies */
int npoly; /* .... sizes of f and intfr */
float *amps; /* array of amplitude values */
int namps; /* .... size of amps */
int icount,ifs,iamps; /* loop counting variables */
float dt; /* sample spacing */
float nyq; /* nyquist frequency */
int nt; /* number of points on input trace */
int nfft; /* number of points for fft trace */
int nf; /* number of frequencies (incl Nyq) */
int verbose; /* flag to get advisory messages */
cwp_Bool seismic; /* is this seismic data? */
Initialize the command line, or output selfdoc
/* Initialize */
initargs(argc, argv);
requestdoc(1);
Get parameters from the first trace and from the command line
/* Get info from first trace */
if (!getparint("verbose", &verbose)) verbose=0;
if (!gettr(&tr)) err("can't get first trace");
seismic = ISSEISMIC(tr.trid);
if (seismic) {
if (verbose) warn("input is seismic data, trid=%d",tr.trid);
dt = ((double) tr.dt)/1000000.0;
}
else {
if (verbose) warn("input is not seismic data, trid=%d",tr.trid);
dt = tr.d1;
}
/* error trapping so that the user can have a default value of dt */
if (!(dt || getparfloat("dt", &dt))) {
dt = .004;
if (verbose) {
warn("neither dt nor d1 are set, nor is dt getparred!");
warn("assuming .004 sec sampling!");
}
}
nt = tr.ns;
nyq = 0.5/dt;
read trace by trace and perform the filtering
/* Set up FFT parameters */
nfft = npfaro(nt, LOOKFAC * nt);
if (nfft >= SU_NFLTS || nfft >= PFA_MAX)
err("Padded nt=%d -- too big", nfft);
nf = nfft/2 + 1;
/* Get frequencies that define the filter */
if ((npoly = countparval("f"))!=0) {
f = ealloc1float(npoly);
getparfloat("f",f);
} else {
npoly = 4;
f = ealloc1float(npoly);
f[0] = FRAC0 * nyq;
f[1] = FRAC1 * nyq;
f[2] = FRAC2 * nyq;
f[3] = FRAC3 * nyq;
}
/* Check f values */
if(npoly < 2) warn("Only %d value defining filter",npoly);
for(ifs=0; ifs < npoly-1; ++ifs)
if(f[ifs] < 0.0 || f[ifs] > f[ifs+1])
err("Bad filter parameters");
/* Get filter amplitude values*/
if ((namps = countparval("amps"))!=0) {
amps = ealloc1float(namps);
getparfloat("amps",amps);
} else {
namps = npoly;
amps = ealloc1float(namps);
/* default is a trapezoidal bandpass filter */
for(iamps=0; iamps<namps; ++iamps)
amps[iamps]=1.;
amps[0]=0.; amps[namps-1]=0.;
}
if (!(namps==npoly))
err("number of f values must = number of amps values");
/* Check amps values */
for(iamps = 0, icount=0; iamps < namps ; ++iamps) {
if( amps[iamps] > 0. ) ++icount;
if( amps[iamps] < 0.) err("amp values must be positive");
}
if (icount==0) err("All amps values are zero");
for(iamps = 0, icount=0; iamps < namps-1 ; ++iamps) {
if(!(amps[iamps]==amps[iamps+1])) ++icount;
}
if (icount==0) warn("All amps values are the same");
Dynamic allocation which hides “malloc” from the user
/* Allocate fft arrays */
rt = ealloc1float(nfft);
ct = ealloc1complex(nf);
filter = ealloc1float(nf);
/* Build the polygonal filter filter[]*/
polygonalFilter(f,amps,npoly,nfft,dt,filter);
note that most of the time we loop over input traces, these are in “do – while” loops
/* Main loop over traces */
do {
register int i;
/* Load trace into rt (zero-padded) */
memcpy((void *) rt, (const void *) tr.data, nt*FSIZE);
memset((void *) (rt + nt), 0 , (nfft-nt)*FSIZE);
/* FFT, filter, inverse FFT */
pfarc(1, nfft, rt, ct);
for (i = 0; i < nf; ++i) ct[i] = crmul(ct[i], filter[i]);
pfacr(-1, nfft, ct, rt);
/* Load traces back in, recall filter had nfft factor */
for (i = 0; i < nt; ++i) tr.data[i] = rt[i];
puttr(&tr);
} while (gettr(&tr));
return(CWP_Exit());
}
Definition of the subroutine that builds the filter
void polygonalFilter(float *f, float *amps, int npoly,
int nfft, float dt, float *filter)
/*************************************************************************
polygonalFilter -- polygonal filter with sin^2 tapering
**************************************************************************
Input:
f array[npoly] of frequencies defining the filter
amps array[npoly] of amplitude values
npoly size of input f and amps arrays
dt time sampling interval
nfft number of points in the fft
Output:
filter array[nfft] filter values
**************************************************************************
Notes: Filter is to be applied in the frequency domain
**************************************************************************
Author: CWP: John Stockwell 1992
*************************************************************************/
#define PIBY2 1.57079632679490
{
int *intfr; /* .... integerizations of f */
int icount,ifs; /* loop counting variables */
int taper=0; /* flag counter */
int nf; /* number of frequencies (incl Nyq) */
int nfm1; /* nf-1 */
float onfft; /* reciprocal of nfft */
float df; /* frequency spacing (from dt) */
intfr=alloc1int(npoly);
nf = nfft/2 + 1;
nfm1 = nf - 1;
onfft = 1.0 / nfft;
/* Compute array of integerized frequencies that define the filter*/
df = onfft / dt;
for(ifs=0; ifs < npoly ; ++ifs) {
intfr[ifs] = NINT(f[ifs]/df);
if (intfr[ifs] > nfm1) intfr[ifs] = nfm1;
}
/* Build filter, with scale, and taper specified by amps[] values*/
/* Do low frequency end first*/
for(icount=0; icount < intfr[0] ; ++icount)
filter[icount] = amps[0] * onfft;
/* now do the middle frequencies */
for(ifs=0 ; ifs<npoly-1 ; ++ifs){
if(amps[ifs] < amps[ifs+1]) {
++taper;
for(icount=intfr[ifs]; icount<=intfr[ifs+1]; ++icount) {
float c = PIBY2 / (intfr[ifs+1] - intfr[ifs] + 2);
float s = sin(c*(icount - intfr[ifs] + 1));
float adiff = amps[ifs+1] - amps[ifs];
filter[icount] = (amps[ifs] + adiff*s*s) * onfft;
}
} else if (amps[ifs] > amps[ifs+1]) {
++taper;
for(icount=intfr[ifs]; icount<=intfr[ifs+1]; ++icount) {
float c = PIBY2 / (intfr[ifs+1] - intfr[ifs] + 2);
float s = sin(c*(intfr[ifs+1] - icount + 1));
float adiff = amps[ifs] - amps[ifs+1];
filter[icount] = (amps[ifs+1] + adiff*s*s) * onfft;
}
} else
if(!(taper)){
for(icount=intfr[ifs]; icount <= intfr[ifs+1]; ++icount)
filter[icount] = amps[ifs] * onfft;
} else {
for(icount=intfr[ifs]+1; icount <= intfr[ifs+1]; ++icount)
filter[icount] = amps[ifs] * onfft;
}
}
/* finally do the high frequency end */
for(icount=intfr[npoly-1]+1; icount<nf; ++icount){
filter[icount] = amps[npoly-1] * onfft;
}
}