@*Intro. This is a filter that inputs the format used by {\mc SAT0},
{\mc SAT1}, etc., and outputs the well-known DIMACS format for
satisfiability problems.

DIMACS format begins with zero or more optional comment lines, indicated by
their first character `\.c'. The next line should say `\.p \.{cnf} $n$ $m$',
where $n$ is the number of variables and $m$ is the number of clauses.
Then comes a string of $m$ ``clauses,'' which are sequences of
nonzero integers of absolute value $\le n$, followed by zero.
A literal for the $k$th variable is represented by $k$; its complement
is represented by $-k$.

SAT format is more flexible, more symbolic, and more complicated; it is
explained in the programs cited above. I hacked this program from {\mc SAT3}.

@c
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "gb_flip.h"
#include <time.h>
time_t myclock;
typedef unsigned int uint; /* a convenient abbreviation */
typedef unsigned long long ullng; /* ditto */
@<Type definitions@>;
@<Global variables@>;
main (int argc, char *argv[]) {
  register uint c,h,i,j,k,l,p,q,r,level,kk,pp,qq,ll;
  @<Process the command line@>;
  @<Initialize everything@>;
  @<Input the clauses@>;
  if (verbose)
    @<Report the successful completion of the input phase@>;
  myclock=time(0);
  printf("c file created by SAT-TO-DIMACS %s",ctime(&myclock));
  @<Output the clauses@>;
}

@ @<Glob...@>=
int random_seed=0; /* seed for the random words of |gb_rand| */
int verbose=1; /* level of verbosity */
int hbits=8; /* logarithm of the number of the hash lists */
int buf_size=1024; /* must exceed the length of the longest input line */

@ On the command line one can say
\smallskip
\item{$\bullet$}
`\.v$\langle\,$integer$\,\rangle$' to enable various levels of verbose
 output on |stderr|;
\item{$\bullet$}
`\.h$\langle\,$positive integer$\,\rangle$' to adjust the hash table size;
\item{$\bullet$}
`\.b$\langle\,$positive integer$\,\rangle$' to adjust the size of the input
buffer; and/or
\item{$\bullet$}
`\.s$\langle\,$integer$\,\rangle$' to define the seed for any random numbers
that are used.

@<Process the command line@>=
for (j=argc-1,k=0;j;j--) switch (argv[j][0]) {
case 'v': k|=(sscanf(argv[j]+1,"%d",&verbose)-1);@+break;
case 'h': k|=(sscanf(argv[j]+1,"%d",&hbits)-1);@+break;
case 'b': k|=(sscanf(argv[j]+1,"%d",&buf_size)-1);@+break;
case 's': k|=(sscanf(argv[j]+1,"%d",&random_seed)-1);@+break;
default: k=1; /* unrecognized command-line option */
}
if (k || hbits<0 || hbits>30 || buf_size<=0) {
  fprintf(stderr,"Usage: %s [v<n>] [h<n>] [b<n>] [s<n>] < foo.sat\n",argv[0]);
  exit(-1);
}

@*The I/O wrapper. The following routines read the input and absorb it into
temporary data areas from which all of the ``real'' data structures
can readily be initialized. My intent is to incorporate these routines in all
of the SAT-solvers in this series. Therefore I've tried to make the code
short and simple, yet versatile enough so that almost no restrictions are
placed on the sizes of problems that can be handled. These routines are
supposed to work properly unless there are more than
$2^{32}-1=4$,294,967,295 occurrences of literals in clauses,
or more than $2^{31}-1=2$,147,483,647 variables or clauses.

In these temporary tables, each variable is represented by four things:
its unique name; its serial number; the clause number (if any) in which it has
most recently appeared; and a pointer to the previous variable (if any)
with the same hash address. Several variables at a time
are represented sequentially in small chunks of memory called ``vchunks,''
which are allocated as needed (and freed later).

@d vars_per_vchunk 341 /* preferably $(2^k-1)/3$ for some $k$ */

@<Type...@>=
typedef union {
  char ch8[8];
  uint u2[2];
  long long lng;
} octa;
typedef struct tmp_var_struct {
  octa name; /* the name (one to eight ASCII characters) */
  uint serial; /* 0 for the first variable, 1 for the second, etc. */
  int stamp; /* |m| if positively in clause |m|; |-m| if negatively there */
  struct tmp_var_struct *next; /* pointer for hash list */
} tmp_var;
@#
typedef struct vchunk_struct {
  struct vchunk_struct *prev; /* previous chunk allocated (if any) */
  tmp_var var[vars_per_vchunk];
} vchunk;

@ Each clause in the temporary tables is represented by a sequence of
one or more pointers to the |tmp_var| nodes of the literals involved.
A negated literal is indicated by adding~1 to such a pointer.
The first literal of a clause is indicated by adding~2.
Several of these pointers are represented sequentially in chunks
of memory, which are allocated as needed and freed later.

@d cells_per_chunk 511 /* preferably $2^k-1$ for some $k$ */

@<Type...@>=
typedef struct chunk_struct {
  struct chunk_struct *prev; /* previous chunk allocated (if any) */
  tmp_var *cell[cells_per_chunk];
} chunk;

@ @<Glob...@>=
char *buf; /* buffer for reading the lines (clauses) of |stdin| */
tmp_var **hash; /* heads of the hash lists */
uint hash_bits[93][8]; /* random bits for universal hash function */
vchunk *cur_vchunk; /* the vchunk currently being filled */
tmp_var *cur_tmp_var; /* current place to create new |tmp_var| entries */
tmp_var *bad_tmp_var; /* the |cur_tmp_var| when we need a new |vchunk| */
chunk *cur_chunk; /* the chunk currently being filled */
tmp_var **cur_cell; /* current place to create new elements of a clause */
tmp_var **bad_cell; /* the |cur_cell| when we need a new |chunk| */
ullng vars; /* how many distinct variables have we seen? */
ullng clauses; /* how many clauses have we seen? */
ullng nullclauses; /* how many of them were null? */
ullng cells; /* how many occurrences of literals in clauses? */

@ @<Initialize everything@>=
gb_init_rand(random_seed);
buf=(char*)malloc(buf_size*sizeof(char));
if (!buf) {
  fprintf(stderr,"Couldn't allocate the input buffer (buf_size=%d)!\n",
            buf_size);
  exit(-2);
}
hash=(tmp_var**)malloc(sizeof(tmp_var)<<hbits);
if (!hash) {
  fprintf(stderr,"Couldn't allocate %d hash list heads (hbits=%d)!\n",
           1<<hbits,hbits);
  exit(-3);
}
for (h=0;h<1<<hbits;h++) hash[h]=NULL;

@ The hash address of each variable name has $h$ bits, where $h$ is the
value of the adjustable parameter |hbits|.
Thus the average number of variables per hash list is $n/2^h$ when there
are $n$ different variables. A warning is printed if this average number
exceeds 10. (For example, if $h$ has its default value, 8, the program will
suggest that you might want to increase $h$ if your input has 2560
different variables or more.)

All the hashing takes place at the very beginning,
and the hash tables are actually recycled before any SAT-solving takes place;
therefore the setting of this parameter is by no means crucial. But I didn't
want to bother with fancy coding that would determine $h$ automatically.

@<Input the clauses@>=
while (1) {
  if (!fgets(buf,buf_size,stdin)) break;
  clauses++;
  if (buf[strlen(buf)-1]!='\n') {
    fprintf(stderr,
      "The clause on line %lld (%.20s...) is too long for me;\n",clauses,buf);
    fprintf(stderr," my buf_size is only %d!\n",buf_size);
    fprintf(stderr,"Please use the command-line option b<newsize>.\n");
    exit(-4);
  }
  @<Input the clause in |buf|@>;
}
if ((vars>>hbits)>=10) {
  fprintf(stderr,"There are %lld variables but only %d hash tables;\n",
     vars,1<<hbits);
  while ((vars>>hbits)>=10) hbits++;
  fprintf(stderr," maybe you should use command-line option h%d?\n",hbits);
}
clauses-=nullclauses;
if (clauses==0) {
  fprintf(stderr,"No clauses were input!\n");
  exit(-77);
}
if (vars>=0x80000000) {
  fprintf(stderr,"Whoa, the input had %llu variables!\n",vars);
  exit(-664);
}
if (clauses>=0x80000000) {
  fprintf(stderr,"Whoa, the input had %llu clauses!\n",clauses);
  exit(-665);
}
if (cells>=0x100000000) {
  fprintf(stderr,"Whoa, the input had %llu occurrences of literals!\n",cells);
  exit(-666);
}

@ @<Input the clause in |buf|@>=
for (j=k=0;;) {
  while (buf[j]==' ') j++; /* scan to nonblank */
  if (buf[j]=='\n') break;
  if (buf[j]<' ' || buf[j]>'~') {
    fprintf(stderr,"Illegal character (code #%x) in the clause on line %lld!\n",
      buf[j],clauses);
    exit(-5);
  }
  if (buf[j]=='~') i=1,j++;
  else i=0;
  @<Scan and record a variable; negate it if |i==1|@>;
}
if (k==0) {
  fprintf(stderr,"(Empty line %lld is being ignored)\n",clauses);
  nullclauses++; /* strictly speaking it would be unsatisfiable */
}
goto clause_done;
empty_clause: @<Remove all variables of the current clause@>;
clause_done: cells+=k;

@ We need a hack to insert the bit codes 1 and/or 2 into a pointer value.

@d hack_in(q,t) (tmp_var*)(t|(ullng)q)

@<Scan and record a variable; negate it if |i==1|@>=
{
  register tmp_var *p;
  if (cur_tmp_var==bad_tmp_var) @<Install a new |vchunk|@>;
  @<Put the variable name beginning at |buf[j]| in |cur_tmp_var->name|
     and compute its hash code |h|@>;
  @<Find |cur_tmp_var->name| in the hash table at |p|@>;
  if (p->stamp==clauses || p->stamp==-clauses) @<Handle a duplicate literal@>@;
  else {
    p->stamp=(i? -clauses: clauses);
    if (cur_cell==bad_cell) @<Install a new |chunk|@>;
    *cur_cell=p;
    if (i==1) *cur_cell=hack_in(*cur_cell,1);
    if (k==0) *cur_cell=hack_in(*cur_cell,2);
    cur_cell++,k++;
  }
}

@ @<Install a new |vchunk|@>=
{
  register vchunk *new_vchunk;
  new_vchunk=(vchunk*)malloc(sizeof(vchunk));
  if (!new_vchunk) {
    fprintf(stderr,"Can't allocate a new vchunk!\n");
    exit(-6);
  }
  new_vchunk->prev=cur_vchunk, cur_vchunk=new_vchunk;
  cur_tmp_var=&new_vchunk->var[0];
  bad_tmp_var=&new_vchunk->var[vars_per_vchunk];
}  

@ @<Install a new |chunk|@>=
{
  register chunk *new_chunk;
  new_chunk=(chunk*)malloc(sizeof(chunk));
  if (!new_chunk) {
    fprintf(stderr,"Can't allocate a new chunk!\n");
    exit(-7);
  }
  new_chunk->prev=cur_chunk, cur_chunk=new_chunk;
  cur_cell=&new_chunk->cell[0];
  bad_cell=&new_chunk->cell[cells_per_chunk];
}  

@ The hash code is computed via ``universal hashing,'' using the following
precomputed tables of random bits.

@<Initialize everything@>=
for (j=92;j;j--) for (k=0;k<8;k++)
  hash_bits[j][k]=gb_next_rand();

@ @<Put the variable name beginning at |buf[j]| in |cur_tmp_var->name|...@>=
cur_tmp_var->name.lng=0;
for (h=l=0;buf[j+l]>' '&&buf[j+l]<='~';l++) {
  if (l>7) {
    fprintf(stderr,
        "Variable name %.9s... in the clause on line %lld is too long!\n",
        buf+j,clauses);
    exit(-8);
  }
  h^=hash_bits[buf[j+l]-'!'][l];
  cur_tmp_var->name.ch8[l]=buf[j+l];
}
if (l==0) goto empty_clause; /* `\.\~' by itself is like `true' */
j+=l;
h&=(1<<hbits)-1;

@ @<Find |cur_tmp_var->name| in the hash table...@>=
for (p=hash[h];p;p=p->next)
  if (p->name.lng==cur_tmp_var->name.lng) break;
if (!p) { /* new variable found */
  p=cur_tmp_var++;
  p->next=hash[h], hash[h]=p;
  p->serial=vars++;
  p->stamp=0;
}

@ The most interesting aspect of the input phase is probably the ``unwinding''
that we might need to do when encountering a literal more than once
in the same clause.

@<Handle a duplicate literal@>=
{
  if ((p->stamp>0)==(i>0)) goto empty_clause;
}

@ An input line that begins with `\.{\~\ }' is silently treated as a comment.
Otherwise redundant clauses are logged, in case they were unintentional.
(One can, however, intentionally
use redundant clauses to force the order of the variables.)

@<Remove all variables of the current clause@>=
while (k) {
  @<Move |cur_cell| backward to the previous cell@>;
  k--;
}
if ((buf[0]!='~')||(buf[1]!=' '))
  fprintf(stderr,"(The clause on line %lld is always satisfied)\n",clauses);
else if (vars==0) printf("c %s",buf+2); /* retain opening comments */
nullclauses++;

@ @<Move |cur_cell| backward to the previous cell@>=
if (cur_cell>&cur_chunk->cell[0]) cur_cell--;
else {
  register chunk *old_chunk=cur_chunk;
  cur_chunk=old_chunk->prev;@+free(old_chunk);
  bad_cell=&cur_chunk->cell[cells_per_chunk];
  cur_cell=bad_cell-1;
}

@ Here I must omit `|free(old_vchunk)|' from the code that's usually
in this section, because the variable data will be used later.

@<Move |cur_tmp_var| backward to the previous temporary variable@>=
if (cur_tmp_var>&cur_vchunk->var[0]) cur_tmp_var--;
else {
  register vchunk *old_vchunk=cur_vchunk;
  cur_vchunk=old_vchunk->prev; /* and don't |free(old_vchunk)| */
  bad_tmp_var=&cur_vchunk->var[vars_per_vchunk];
  cur_tmp_var=bad_tmp_var-1;
}

@ @<Report the successful completion of the input phase@>=
fprintf(stderr,"(%lld variables, %lld clauses, %llu literals successfully read)\n",
                       vars,clauses,cells);

@*The output phase. I had to input everything first because DIMACS format
specifies the number of variables and clauses right at the beginning.

@<Output the clauses@>=
@<Show the variable names as comments@>;
printf("p cnf %lld %lld\n",vars,clauses);
@<Translate all the temporary cells into the simple DIMACS form@>;
@<Check consistency@>;

@ This section is optional, but I'm including it today while I remember
how to provide it.

@<Show the variable names as comments@>=
for (c=vars; c; c--) {
  @<Move |cur_tmp_var| back...@>;
  printf("c %.8s -> %d\n",cur_tmp_var->name.ch8,c);
}

@ @<Translate all the temporary cells into the simple DIMACS form@>=
for (c=clauses; c; c--) {
  @<Translate the cells for the literals of clause |c|@>;
  printf(" 0\n");
}

@ @d hack_out(q) (((ullng)q)&0x3)
@d hack_clean(q) ((tmp_var*)((ullng)q&-4))

@<Translate the cells for the literals of clause |c|@>=
for (i=0;i<2;j++) {
  @<Move |cur_cell| back...@>;
  i=hack_out(*cur_cell);
  p=hack_clean(*cur_cell)->serial;
  printf(" %s%d",i&1?"-":"",p+1);
}

@ @<Check consistency@>=
if (cur_cell!=&cur_chunk->cell[0] ||
     cur_chunk->prev!=NULL ||
     cur_tmp_var!=&cur_vchunk->var[0] ||
     cur_vchunk->prev!=NULL) {
  fprintf(stderr,"This can't happen (consistency check failure)!\n");
  exit(-14);
}

@*Index.