/* Copyright (C) 1994 Aladdin Enterprises. All rights reserved. This file is part of Aladdin Ghostscript. Aladdin Ghostscript is distributed with NO WARRANTY OF ANY KIND. No author or distributor accepts any responsibility for the consequences of using it, or for whether it serves any particular purpose or works at all, unless he or she says so in writing. Refer to the Aladdin Ghostscript Free Public License (the "License") for full details. Every copy of Aladdin Ghostscript must include a copy of the License, normally in a plain ASCII text file named PUBLIC. The License grants you the right to copy, modify and redistribute Aladdin Ghostscript, but only under certain conditions described in the License. Among other things, the License requires that the copyright notice and this notice be preserved on all copies. */ /* sbwbs.c */ /* Burrows/Wheeler block sorting compression filters */ #include "stdio_.h" #include "memory_.h" #include /* for qsort */ #include "gdebug.h" #include "strimpl.h" #include "sfilter.h" #include "sbwbs.h" /* ------ Move-to-front (MTF) encode and decode ------ */ private_st_MTF_state(); #define ss ((stream_MTF_state *)st) /* Initialize */ private int s_MTF_init(stream_state *st) { int i; for ( i = 0; i < 256; i++ ) ss->prev[i] = (byte)i; return 0; } /* Encode a buffer */ private int s_MTFE_process(stream_state *st, stream_cursor_read *pr, stream_cursor_write *pw, bool last) { register const byte *p = pr->ptr; register byte *q = pw->ptr; const byte *rlimit = pr->limit; uint count = rlimit - p; uint wcount = pw->limit - q; int status = (count < wcount ? 0 : (rlimit = p + wcount, 1)); while ( p < rlimit ) { byte b = *++p; int i; byte prev = b, repl; for ( i = 0; (repl = ss->prev[i]) != b; i++ ) ss->prev[i] = prev, prev = repl; ss->prev[i] = prev; *++q = (byte)i; } pr->ptr = p; pw->ptr = q; return status; } /* Stream template */ const stream_template s_MTFE_template = { &st_MTF_state, s_MTF_init, s_MTFE_process, 1, 1 }; /* Decode a buffer */ private int s_MTFD_process(stream_state *st, stream_cursor_read *pr, stream_cursor_write *pw, bool last) { register const byte *p = pr->ptr; register byte *q = pw->ptr; const byte *rlimit = pr->limit; uint count = rlimit - p; uint wcount = pw->limit - q; int status = (count < wcount ? 0 : (rlimit = p + wcount, 1)); /* Cache the first few entries in local variables. */ byte v0 = ss->prev[0], v1 = ss->prev[1], v2 = ss->prev[2], v3 = ss->prev[3]; while ( p < rlimit ) { byte first; /* Zeros far outnumber all other bytes in the BWBS */ /* code; check for them first. */ if ( *++p == 0 ) { *++q = v0; continue; } switch ( *p ) { default: { uint b = *p; *++q = first = ss->prev[b]; for ( ; b >= 8; b -= 4 ) { byte *bp = &ss->prev[b]; bp[0] = bp[-1]; bp[-1] = bp[-2]; bp[-2] = bp[-3]; bp[-3] = bp[-4]; } switch ( b ) { case 7: ss->prev[7] = ss->prev[6]; case 6: ss->prev[6] = ss->prev[5]; case 5: ss->prev[5] = ss->prev[4]; /* case 4: ; */ } } ss->prev[4] = v3; v3 = v2, v2 = v1, v1 = v0, v0 = first; break; case 3: *++q = first = v3; v3 = v2, v2 = v1, v1 = v0, v0 = first; break; case 2: *++q = first = v2; v2 = v1, v1 = v0, v0 = first; break; case 1: *++q = first = v1; v1 = v0, v0 = first; break; } } ss->prev[0] = v0; ss->prev[1] = v1; ss->prev[2] = v2; ss->prev[3] = v3; pr->ptr = p; pw->ptr = q; return status; } /* Stream template */ const stream_template s_MTFD_template = { &st_MTF_state, s_MTF_init, s_MTFD_process, 1, 1 }; #undef ss /* ------ Common code for streams that buffer a block ------ */ private_st_buffered_state(); #define ss ((stream_buffered_state *)st) /* Initialize */ private int s_buffered_no_block_init(stream_state *st) { ss->buffer = 0; ss->filling = true; ss->bpos = 0; return 0; } private int s_buffered_block_init(stream_state *st) { s_buffered_no_block_init(st); ss->buffer = gs_alloc_bytes(st->memory, ss->BlockSize, "buffer"); if ( ss->buffer == 0 ) return ERRC; /****** WRONG ******/ return 0; } /* Continue filling the buffer if needed. */ /* Return 0 if the buffer isn't full yet, 1 if it is full or if */ /* we reached the end of input data. */ /* In the latter case, also set filling = false. */ /* Note that this procedure doesn't take pw as an argument. */ private int s_buffered_process(stream_state *st, stream_cursor_read *pr, bool last) { register const byte *p = pr->ptr; const byte *rlimit = pr->limit; uint count = rlimit - p; uint left = ss->bsize - ss->bpos; if ( !ss->filling ) return 1; if ( left < count ) count = left; if_debug3('w', "[w]buffering %d bytes to position %d, last = %s\n", count, ss->bpos, (last ? "true" : "false")); memcpy(ss->buffer + ss->bpos, p + 1, count); pr->ptr = p += count; ss->bpos += count; if ( ss->bpos == ss->bsize || (p == rlimit && last) ) { ss->filling = false; return 1; } return 0; } /* Release */ private void s_buffered_release(stream_state *st) { gs_free_object(st->memory, ss->buffer, "buffer"); } #undef ss /* ------ Common code for Burrows/Wheeler block sorting filters ------ */ private_st_BWBS_state(); private void s_BWBS_release(P1(stream_state *)); #define ss ((stream_BWBS_state *)st) /* Initialize */ private int bwbs_init(stream_state *st, uint osize) { int code; ss->bsize = ss->BlockSize; code = s_buffered_block_init(st); if ( code != 0 ) return code; ss->offsets = (void *)gs_alloc_bytes(st->memory, osize, "BWBlockSort offsets"); if ( ss->offsets == 0 ) { s_BWBS_release(st); return ERRC; /****** WRONG ******/ } ss->I = -1; /* haven't read I yet */ return 0; } /* Release the filter. */ private void s_BWBS_release(stream_state *st) { gs_free_object(st->memory, ss->offsets, "BWBlockSort offsets"); s_buffered_release(st); } /* ------ BWBlockSortEncode filter ------ */ /* Initialize */ private int s_BWBSE_init(stream_state *st) { return bwbs_init(st, ss->BlockSize * sizeof(int)); } /* Compare two rotated strings for sorting. */ private stream_BWBS_state *bwbs_compare_ss; private int bwbs_compare_rotations(const void *p1, const void *p2) { const byte *buffer = bwbs_compare_ss->buffer; const byte *s1 = buffer + *(const int *)p1; const byte *s2 = buffer + *(const int *)p2; const byte *start1; const byte *end; int swap; if ( *s1 != *s2 ) return (*s1 < *s2 ? -1 : 1); if ( s1 < s2 ) swap = 1; else { const byte *t = s1; s1 = s2; s2 = t; swap = -1; } start1 = s1; end = buffer + bwbs_compare_ss->N; for ( s1++, s2++; s2 < end; s1++, s2++ ) if ( *s1 != *s2 ) return (*s1 < *s2 ? -swap : swap); s2 = buffer; for ( ; s1 < end; s1++, s2++ ) if ( *s1 != *s2 ) return (*s1 < *s2 ? -swap : swap); s1 = buffer; for ( ; s1 < start1; s1++, s2++ ) if ( *s1 != *s2 ) return (*s1 < *s2 ? -swap : swap); return 0; } /* Sort the strings. */ private void bwbse_sort(const byte *buffer, uint *indices, int N) { offsets_full Cs; #define C Cs.v /* Sort the strings. We start with a radix sort. */ uint sum = 0, j, ch; memset(C, 0, sizeof(Cs)); for ( j = 0; j < N; j++ ) C[buffer[j]]++; for ( ch = 0; ch <= 255; ch++ ) { sum += C[ch]; C[ch] = sum - C[ch]; } for ( j = 0; j < N; j++ ) indices[C[buffer[j]]++] = j; /* Now C[ch] = the number of strings that start */ /* with a character less than or equal to ch. */ sum = 0; /* qsort each bucket produced by the radix sort. */ for ( ch = 0; ch <= 255; sum = C[ch], ch++ ) qsort(indices + sum, C[ch] - sum, sizeof(*indices), bwbs_compare_rotations); #undef C } /* Encode a buffer */ private int s_BWBSE_process(stream_state *st, stream_cursor_read *pr, stream_cursor_write *pw, bool last) { register byte *q = pw->ptr; byte *wlimit = pw->limit; uint wcount = wlimit - q; uint *indices = ss->offsets; if ( ss->filling ) { int status, j, N; byte *buffer = ss->buffer; if ( wcount < sizeof(int) * 2 ) return 1; /* Continue filling the buffer. */ status = s_buffered_process(st, pr, last); if ( !status ) return 0; ss->N = N = ss->bpos; /* We reverse the string before encoding it, */ /* so it will come out of the decoder correctly. */ for ( j = N / 2 - 1; j >= 0; j-- ) { byte *p0 = &buffer[j]; byte *p1 = &buffer[N - 1 - j]; byte b = *p0; *p0 = *p1; *p1 = b; } /* Save st in a static, because that's the only way */ /* we can pass it to the comparison procedure (ugh). */ bwbs_compare_ss = ss; /* Sort the strings. */ bwbse_sort(buffer, indices, N); /* Find the unrotated string. */ for ( j = 0; j < N; j++ ) if ( indices[j] == 0 ) { ss->I = j; break; } for ( j = sizeof(int); --j >= 0; ) *++q = (byte)(N >> (j * 8)); for ( j = sizeof(int); --j >= 0; ) *++q = (byte)(ss->I >> (j * 8)); ss->bpos = 0; } /* We're reading out of the buffer, writing the permuted string. */ while ( q < wlimit && ss->bpos < ss->N ) { int i = indices[ss->bpos++]; *++q = ss->buffer[(i == 0 ? ss->N - 1 : i - 1)]; } if ( ss->bpos == ss->N ) { ss->filling = true; ss->bpos = 0; } pw->ptr = q; if ( q == wlimit ) return 1; return 0; } /* Stream template */ const stream_template s_BWBSE_template = { &st_BWBS_state, s_BWBSE_init, s_BWBSE_process, sizeof(int) * 2, 1, s_BWBS_release }; /* ------ BWBlockSortDecode filter ------ */ #define SHORT_OFFSETS #ifdef SHORT_OFFSETS /* * Letting S[0..N-1] be the data block before depermutation, we need * a table P[0..N-1] that maps the index i to O(S[i],i), where O(c,i) is * the number of occurrences of c in S before position i. * We observe that for a fixed c, O(c,i) is monotonic with i, * and falls in the range 0 .. i; consequently, if 0 <= i <= j, * 0 <= O(c,j) - O(c,i) <= j - i. Proceeding from this observation, * rather than allocate an entire int to each entry of P, * we construct three tables as follows: * P2[k,c] = O(c,k*65536) for k = 0 .. (N-1)/65536; * each entry requires an int. * P1[k,c] = O(c,k*4096) - P2[k/16,c] for k = 0 .. (N-1)/4096; * each entry falls in the range 0 .. 15*4096 and hence * requires 16 bits. * P0[i] = O(S[i],i) - P1[i/4096,S[i]] for i = 0 .. N-1; * each entry falls in the range 0 .. 4095 and hence * requires 12 bits. * Since the value we need in the decompression loop is actually * P[i] + C[S[i]], where C[c] is the sum of O(0,N) ... O(c-1,N), * we add C[c] into P2[k,c] for all k. */ /*typedef struct { uint v[256]; } offsets_full;*/ /* in sbwbs.h */ typedef struct { bits16 v[256]; } offsets_4k; #if arch_sizeof_int > 2 # define ceil_64k(n) (((n) + 0xffff) >> 16) #else # define ceil_64k(n) 1 #endif #define ceil_4k(n) (((n) + 0xfff) >> 12) #define offset_space(bsize)\ (ceil_64k(bsize) * sizeof(offsets_full) +\ ceil_4k(bsize) * sizeof(offsets_4k) +\ ((bsize + 1) >> 1) * 3) #else /* !SHORT_OFFSETS */ #define offset_space(bsize)\ (bsize * sizeof(int)) #endif /* (!)SHORT_OFFSETS */ /* Initialize */ private int s_BWBSD_init(stream_state *st) { uint bsize = ss->BlockSize; return bwbs_init(st, offset_space(bsize)); } /* Construct the decoding tables. */ #ifdef SHORT_OFFSETS private void bwbsd_construct_offsets(stream_BWBS_state *sst, offsets_full *po64k, offsets_4k *po4k, byte *po1, int N) { offsets_full Cs; #define C Cs.v uint i1; byte *b = sst->buffer; offsets_full *p2 = po64k - 1; offsets_4k *p1 = po4k; byte *p0 = po1; memset(C, 0, sizeof(Cs)); for ( i1 = 0; i1 < ceil_4k(N); i1++, p1++ ) { int j; if ( !(i1 & 15) ) *++p2 = Cs; for ( j = 0; j < 256; j++ ) p1->v[j] = C[j] - p2->v[j]; j = (N + 1 - (i1 << 12)) >> 1; if ( j > 4096/2 ) j = 4096/2; for ( ; j > 0; j--, b += 2, p0 += 3 ) { byte b0 = b[0]; uint d0 = C[b0]++ - (p1->v[b0] + p2->v[b0]); byte b1 = b[1]; uint d1 = C[b1]++ - (p1->v[b1] + p2->v[b1]); p0[0] = d0 >> 4; p0[1] = (byte)((d0 << 4) + (d1 >> 8)); p0[2] = (byte)d1; } } /* If the block length is odd, discount the extra byte. */ if ( N & 1 ) C[sst->buffer[N]]--; /* Compute the cumulative totals in C. */ { int sum = 0, ch; for ( ch = 0; ch <= 255; ch++ ) { sum += C[ch]; C[ch] = sum - C[ch]; } } /* Add the C values back into the 64K table, */ /* which saves an addition of C[b] in the decoding loop. */ { int i2, ch; for ( p2 = po64k, i2 = ceil_64k(N); i2 > 0; p2++, i2-- ) for ( ch = 0; ch < 256; ch++ ) p2->v[ch] += C[ch]; } #undef C } #else /* !SHORT_OFFSETS */ private void bwbsd_construct_offsets(stream_BWBS_state *sst, int *po, int N) { offsets_full Cs; #define C Cs.v uint i; byte *b = sst->buffer; int *p = po; memset(C, 0, sizeof(Cs)); for ( i = 0; i < N; i++, p++, b++ ) *p = C[*b]++; /* Compute the cumulative totals in C. */ { int sum = 0, ch; for ( ch = 0; ch <= 255; ch++ ) { sum += C[ch]; C[ch] = sum - C[ch]; } } /* Add the C values back into the offsets. */ for ( i = 0, b = sst->buffer, p = po; i < N; i++, b++, p++ ) *p += C[*b]; #undef C } #endif /* (!)SHORT_OFFSETS */ /* Decode a buffer */ private int s_BWBSD_process(stream_state *st, stream_cursor_read *pr, stream_cursor_write *pw, bool last) { register const byte *p = pr->ptr; const byte *rlimit = pr->limit; uint count = rlimit - p; register byte *q = pw->ptr; byte *wlimit = pw->limit; #ifdef SHORT_OFFSETS uint BlockSize = ss->BlockSize; offsets_full *po64k = ss->offsets; offsets_4k *po4k = (offsets_4k *)(po64k + ceil_64k(BlockSize)); byte *po1 = (byte *)(po4k + ceil_4k(BlockSize)); #else /* !SHORT_OFFSETS */ int *po = ss->offsets; #endif /* (!)SHORT_OFFSETS */ if ( ss->I < 0 ) { /* Read block parameters */ int I, N, j; if ( count < sizeof(int) * 2 ) return 0; for ( N = 0, j = 0; j < sizeof(int); j++ ) N = (N << 8) + *++p; for ( I = 0, j = 0; j < sizeof(int); j++ ) I = (I << 8) + *++p; ss->N = N; ss->I = I; if_debug2('w', "[w]N=%d I=%d\n", N, I); pr->ptr = p; if ( N < 0 || N > ss->BlockSize || I < 0 || I >= N ) return ERRC; if ( N == 0 ) return EOFC; count -= sizeof(int) * 2; ss->bpos = 0; ss->bsize = N; } if ( ss->filling ) { /* Continue filling the buffer. */ if ( !s_buffered_process(st, pr, last) ) return 0; /* Construct the inverse sort order. */ #ifdef SHORT_OFFSETS bwbsd_construct_offsets(ss, po64k, po4k, po1, ss->bsize); #else /* !SHORT_OFFSETS */ bwbsd_construct_offsets(ss, po, ss->bsize); #endif /* (!)SHORT_OFFSETS */ ss->bpos = 0; ss->i = ss->I; } /* We're reading out of the buffer. */ while ( q < wlimit && ss->bpos < ss->bsize ) { int i = ss->i; byte b = ss->buffer[i]; #ifdef SHORT_OFFSETS uint d; const byte *pd = &po1[(i >> 1) + i]; *++q = b; if ( !(i & 1) ) d = ((uint)pd[0] << 4) + (pd[1] >> 4); else d = ((pd[0] & 0xf) << 8) + pd[1]; ss->i = po64k[i >> 16].v[b] + po4k[i >> 12].v[b] + d; #else /* !SHORT_OFFSETS */ *++q = b; ss->i = po[i]; #endif /* (!)SHORT_OFFSETS */ ss->bpos++; } if ( ss->bpos == ss->bsize ) { ss->I = -1; ss->filling = true; } pw->ptr = q; if ( q == wlimit ) return 1; return 0; } #undef ss /* Stream template */ const stream_template s_BWBSD_template = { &st_BWBS_state, s_BWBSD_init, s_BWBSD_process, 1, sizeof(int) * 2, s_BWBS_release };