File: ksw.c

package info (click to toggle)
fermi-lite 0.1-5
  • links: PTS, VCS
  • area: main
  • in suites: bullseye, buster, sid
  • size: 652 kB
  • sloc: ansic: 5,157; makefile: 63; sh: 13
file content (352 lines) | stat: -rw-r--r-- 12,208 bytes parent folder | download | duplicates (4)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
/* The MIT License

   Copyright (c) 2011 by Attractive Chaos <attractor@live.co.uk>

   Permission is hereby granted, free of charge, to any person obtaining
   a copy of this software and associated documentation files (the
   "Software"), to deal in the Software without restriction, including
   without limitation the rights to use, copy, modify, merge, publish,
   distribute, sublicense, and/or sell copies of the Software, and to
   permit persons to whom the Software is furnished to do so, subject to
   the following conditions:

   The above copyright notice and this permission notice shall be
   included in all copies or substantial portions of the Software.

   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
   EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
   MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
   NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
   BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
   ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
   CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
   SOFTWARE.
*/

#include <stdlib.h>
#include <stdint.h>
#include <emmintrin.h>
#include "ksw.h"

#ifdef __GNUC__
#define LIKELY(x) __builtin_expect((x),1)
#define UNLIKELY(x) __builtin_expect((x),0)
#else
#define LIKELY(x) (x)
#define UNLIKELY(x) (x)
#endif

const kswr_t g_defr = { 0, -1, -1, -1, -1, -1, -1 };

struct _kswq_t {
	int qlen, slen;
	uint8_t shift, mdiff, max, size;
	__m128i *qp, *H0, *H1, *E, *Hmax;
};

/**
 * Initialize the query data structure
 *
 * @param size   Number of bytes used to store a score; valid valures are 1 or 2
 * @param qlen   Length of the query sequence
 * @param query  Query sequence
 * @param m      Size of the alphabet
 * @param mat    Scoring matrix in a one-dimension array
 *
 * @return       Query data structure
 */
kswq_t *ksw_qinit(int size, int qlen, const uint8_t *query, int m, const int8_t *mat)
{
	kswq_t *q;
	int slen, a, tmp, p;

	size = size > 1? 2 : 1;
	p = 8 * (3 - size); // # values per __m128i
	slen = (qlen + p - 1) / p; // segmented length
	q = (kswq_t*)malloc(sizeof(kswq_t) + 256 + 16 * slen * (m + 4)); // a single block of memory
	q->qp = (__m128i*)(((size_t)q + sizeof(kswq_t) + 15) >> 4 << 4); // align memory
	q->H0 = q->qp + slen * m;
	q->H1 = q->H0 + slen;
	q->E  = q->H1 + slen;
	q->Hmax = q->E + slen;
	q->slen = slen; q->qlen = qlen; q->size = size;
	// compute shift
	tmp = m * m;
	for (a = 0, q->shift = 127, q->mdiff = 0; a < tmp; ++a) { // find the minimum and maximum score
		if (mat[a] < (int8_t)q->shift) q->shift = mat[a];
		if (mat[a] > (int8_t)q->mdiff) q->mdiff = mat[a];
	}
	q->max = q->mdiff;
	q->shift = 256 - q->shift; // NB: q->shift is uint8_t
	q->mdiff += q->shift; // this is the difference between the min and max scores
	// An example: p=8, qlen=19, slen=3 and segmentation:
	//  {{0,3,6,9,12,15,18,-1},{1,4,7,10,13,16,-1,-1},{2,5,8,11,14,17,-1,-1}}
	if (size == 1) {
		int8_t *t = (int8_t*)q->qp;
		for (a = 0; a < m; ++a) {
			int i, k, nlen = slen * p;
			const int8_t *ma = mat + a * m;
			for (i = 0; i < slen; ++i)
				for (k = i; k < nlen; k += slen) // p iterations
					*t++ = (k >= qlen? 0 : ma[query[k]]) + q->shift;
		}
	} else {
		int16_t *t = (int16_t*)q->qp;
		for (a = 0; a < m; ++a) {
			int i, k, nlen = slen * p;
			const int8_t *ma = mat + a * m;
			for (i = 0; i < slen; ++i)
				for (k = i; k < nlen; k += slen) // p iterations
					*t++ = (k >= qlen? 0 : ma[query[k]]);
		}
	}
	return q;
}

kswr_t ksw_u8(kswq_t *q, int tlen, const uint8_t *target, int _gapo, int _gape, int xtra) // the first gap costs -(_o+_e)
{
	int slen, i, m_b, n_b, te = -1, gmax = 0, minsc, endsc;
	uint64_t *b;
	__m128i zero, gapoe, gape, shift, *H0, *H1, *E, *Hmax;
	kswr_t r;

#define __max_16(ret, xx) do { \
		(xx) = _mm_max_epu8((xx), _mm_srli_si128((xx), 8)); \
		(xx) = _mm_max_epu8((xx), _mm_srli_si128((xx), 4)); \
		(xx) = _mm_max_epu8((xx), _mm_srli_si128((xx), 2)); \
		(xx) = _mm_max_epu8((xx), _mm_srli_si128((xx), 1)); \
    	(ret) = _mm_extract_epi16((xx), 0) & 0x00ff; \
	} while (0)

	// initialization
	r = g_defr;
	minsc = (xtra&KSW_XSUBO)? xtra&0xffff : 0x10000;
	endsc = (xtra&KSW_XSTOP)? xtra&0xffff : 0x10000;
	m_b = n_b = 0; b = 0;
	zero = _mm_set1_epi32(0);
	gapoe = _mm_set1_epi8(_gapo + _gape);
	gape = _mm_set1_epi8(_gape);
	shift = _mm_set1_epi8(q->shift);
	H0 = q->H0; H1 = q->H1; E = q->E; Hmax = q->Hmax;
	slen = q->slen;
	for (i = 0; i < slen; ++i) {
		_mm_store_si128(E + i, zero);
		_mm_store_si128(H0 + i, zero);
		_mm_store_si128(Hmax + i, zero);
	}
	// the core loop
	for (i = 0; i < tlen; ++i) {
		int j, k, cmp, imax;
		__m128i e, h, f = zero, max = zero, *S = q->qp + target[i] * slen; // s is the 1st score vector
		h = _mm_load_si128(H0 + slen - 1); // h={2,5,8,11,14,17,-1,-1} in the above example
		h = _mm_slli_si128(h, 1); // h=H(i-1,-1); << instead of >> because x64 is little-endian
		for (j = 0; LIKELY(j < slen); ++j) {
			/* SW cells are computed in the following order:
			 *   H(i,j)   = max{H(i-1,j-1)+S(i,j), E(i,j), F(i,j)}
			 *   E(i+1,j) = max{H(i,j)-q, E(i,j)-r}
			 *   F(i,j+1) = max{H(i,j)-q, F(i,j)-r}
			 */
			// compute H'(i,j); note that at the beginning, h=H'(i-1,j-1)
			h = _mm_adds_epu8(h, _mm_load_si128(S + j));
			h = _mm_subs_epu8(h, shift); // h=H'(i-1,j-1)+S(i,j)
			e = _mm_load_si128(E + j); // e=E'(i,j)
			h = _mm_max_epu8(h, e);
			h = _mm_max_epu8(h, f); // h=H'(i,j)
			max = _mm_max_epu8(max, h); // set max
			_mm_store_si128(H1 + j, h); // save to H'(i,j)
			// now compute E'(i+1,j)
			h = _mm_subs_epu8(h, gapoe); // h=H'(i,j)-gapo
			e = _mm_subs_epu8(e, gape); // e=E'(i,j)-gape
			e = _mm_max_epu8(e, h); // e=E'(i+1,j)
			_mm_store_si128(E + j, e); // save to E'(i+1,j)
			// now compute F'(i,j+1)
			f = _mm_subs_epu8(f, gape);
			f = _mm_max_epu8(f, h);
			// get H'(i-1,j) and prepare for the next j
			h = _mm_load_si128(H0 + j); // h=H'(i-1,j)
		}
		// NB: we do not need to set E(i,j) as we disallow adjecent insertion and then deletion
		for (k = 0; LIKELY(k < 16); ++k) { // this block mimics SWPS3; NB: H(i,j) updated in the lazy-F loop cannot exceed max
			f = _mm_slli_si128(f, 1);
			for (j = 0; LIKELY(j < slen); ++j) {
				h = _mm_load_si128(H1 + j);
				h = _mm_max_epu8(h, f); // h=H'(i,j)
				_mm_store_si128(H1 + j, h);
				h = _mm_subs_epu8(h, gapoe);
				f = _mm_subs_epu8(f, gape);
				cmp = _mm_movemask_epi8(_mm_cmpeq_epi8(_mm_subs_epu8(f, h), zero));
				if (UNLIKELY(cmp == 0xffff)) goto end_loop16;
			}
		}
end_loop16:
		//int k;for (k=0;k<16;++k)printf("%d ", ((uint8_t*)&max)[k]);printf("\n");
		__max_16(imax, max); // imax is the maximum number in max
		if (imax >= minsc) { // write the b array; this condition adds branching unfornately
			if (n_b == 0 || (int32_t)b[n_b-1] + 1 != i) { // then append
				if (n_b == m_b) {
					m_b = m_b? m_b<<1 : 8;
					b = (uint64_t*)realloc(b, 8 * m_b);
				}
				b[n_b++] = (uint64_t)imax<<32 | i;
			} else if ((int)(b[n_b-1]>>32) < imax) b[n_b-1] = (uint64_t)imax<<32 | i; // modify the last
		}
		if (imax > gmax) {
			gmax = imax; te = i; // te is the end position on the target
			for (j = 0; LIKELY(j < slen); ++j) // keep the H1 vector
				_mm_store_si128(Hmax + j, _mm_load_si128(H1 + j));
			if (gmax + q->shift >= 255 || gmax >= endsc) break;
		}
		S = H1; H1 = H0; H0 = S; // swap H0 and H1
	}
	r.score = gmax + q->shift < 255? gmax : 255;
	r.te = te;
	if (r.score != 255) { // get a->qe, the end of query match; find the 2nd best score
		int max = -1, low, high, qlen = slen * 16;
		uint8_t *t = (uint8_t*)Hmax;
		for (i = 0; i < qlen; ++i, ++t)
			if ((int)*t > max) max = *t, r.qe = i / 16 + i % 16 * slen;
		//printf("%d,%d\n", max, gmax);
		if (b) {
			i = (r.score + q->max - 1) / q->max;
			low = te - i; high = te + i;
			for (i = 0; i < n_b; ++i) {
				int e = (int32_t)b[i];
				if ((e < low || e > high) && b[i]>>32 > (uint32_t)r.score2)
					r.score2 = b[i]>>32, r.te2 = e;
			}
		}
	}
	free(b);
	return r;
}

kswr_t ksw_i16(kswq_t *q, int tlen, const uint8_t *target, int _gapo, int _gape, int xtra) // the first gap costs -(_o+_e)
{
	int slen, i, m_b, n_b, te = -1, gmax = 0, minsc, endsc;
	uint64_t *b;
	__m128i zero, gapoe, gape, *H0, *H1, *E, *Hmax;
	kswr_t r;

#define __max_8(ret, xx) do { \
		(xx) = _mm_max_epi16((xx), _mm_srli_si128((xx), 8)); \
		(xx) = _mm_max_epi16((xx), _mm_srli_si128((xx), 4)); \
		(xx) = _mm_max_epi16((xx), _mm_srli_si128((xx), 2)); \
    	(ret) = _mm_extract_epi16((xx), 0); \
	} while (0)

	// initialization
	r = g_defr;
	minsc = (xtra&KSW_XSUBO)? xtra&0xffff : 0x10000;
	endsc = (xtra&KSW_XSTOP)? xtra&0xffff : 0x10000;
	m_b = n_b = 0; b = 0;
	zero = _mm_set1_epi32(0);
	gapoe = _mm_set1_epi16(_gapo + _gape);
	gape = _mm_set1_epi16(_gape);
	H0 = q->H0; H1 = q->H1; E = q->E; Hmax = q->Hmax;
	slen = q->slen;
	for (i = 0; i < slen; ++i) {
		_mm_store_si128(E + i, zero);
		_mm_store_si128(H0 + i, zero);
		_mm_store_si128(Hmax + i, zero);
	}
	// the core loop
	for (i = 0; i < tlen; ++i) {
		int j, k, imax;
		__m128i e, h, f = zero, max = zero, *S = q->qp + target[i] * slen; // s is the 1st score vector
		h = _mm_load_si128(H0 + slen - 1); // h={2,5,8,11,14,17,-1,-1} in the above example
		h = _mm_slli_si128(h, 2);
		for (j = 0; LIKELY(j < slen); ++j) {
			h = _mm_adds_epi16(h, *S++);
			e = _mm_load_si128(E + j);
			h = _mm_max_epi16(h, e);
			h = _mm_max_epi16(h, f);
			max = _mm_max_epi16(max, h);
			_mm_store_si128(H1 + j, h);
			h = _mm_subs_epu16(h, gapoe);
			e = _mm_subs_epu16(e, gape);
			e = _mm_max_epi16(e, h);
			_mm_store_si128(E + j, e);
			f = _mm_subs_epu16(f, gape);
			f = _mm_max_epi16(f, h);
			h = _mm_load_si128(H0 + j);
		}
		for (k = 0; LIKELY(k < 16); ++k) {
			f = _mm_slli_si128(f, 2);
			for (j = 0; LIKELY(j < slen); ++j) {
				h = _mm_load_si128(H1 + j);
				h = _mm_max_epi16(h, f);
				_mm_store_si128(H1 + j, h);
				h = _mm_subs_epu16(h, gapoe);
				f = _mm_subs_epu16(f, gape);
				if(UNLIKELY(!_mm_movemask_epi8(_mm_cmpgt_epi16(f, h)))) goto end_loop8;
			}
		}
end_loop8:
		__max_8(imax, max);
		if (imax >= minsc) {
			if (n_b == 0 || (int32_t)b[n_b-1] + 1 != i) {
				if (n_b == m_b) {
					m_b = m_b? m_b<<1 : 8;
					b = (uint64_t*)realloc(b, 8 * m_b);
				}
				b[n_b++] = (uint64_t)imax<<32 | i;
			} else if ((int)(b[n_b-1]>>32) < imax) b[n_b-1] = (uint64_t)imax<<32 | i; // modify the last
		}
		if (imax > gmax) {
			gmax = imax; te = i;
			for (j = 0; LIKELY(j < slen); ++j)
				_mm_store_si128(Hmax + j, _mm_load_si128(H1 + j));
			if (gmax >= endsc) break;
		}
		S = H1; H1 = H0; H0 = S;
	}
	r.score = gmax; r.te = te;
	{
		int max = -1, low, high, qlen = slen * 8;
		uint16_t *t = (uint16_t*)Hmax;
		for (i = 0, r.qe = -1; i < qlen; ++i, ++t)
			if ((int)*t > max) max = *t, r.qe = i / 8 + i % 8 * slen;
		if (b) {
			i = (r.score + q->max - 1) / q->max;
			low = te - i; high = te + i;
			for (i = 0; i < n_b; ++i) {
				int e = (int32_t)b[i];
				if ((e < low || e > high) && b[i]>>32 > (uint32_t)r.score2)
					r.score2 = b[i]>>32, r.te2 = e;
			}
		}
	}
	free(b);
	return r;
}

static void revseq(int l, uint8_t *s)
{
	int i, t;
	for (i = 0; i < l>>1; ++i)
		t = s[i], s[i] = s[l - 1 - i], s[l - 1 - i] = t;
}

kswr_t ksw_align(int qlen, uint8_t *query, int tlen, uint8_t *target, int m, const int8_t *mat, int gapo, int gape, int xtra, kswq_t **qry)
{
	int size;
	kswq_t *q;
	kswr_t r, rr;
	kswr_t (*func)(kswq_t*, int, const uint8_t*, int, int, int);

	q = (qry && *qry)? *qry : ksw_qinit((xtra&KSW_XBYTE)? 1 : 2, qlen, query, m, mat);
	if (qry && *qry == 0) *qry = q;
	func = q->size == 2? ksw_i16 : ksw_u8;
	size = q->size;
	r = func(q, tlen, target, gapo, gape, xtra);
	if (qry == 0) free(q);
	if ((xtra&KSW_XSTART) == 0 || ((xtra&KSW_XSUBO) && r.score < (xtra&0xffff))) return r;
	revseq(r.qe + 1, query); revseq(r.te + 1, target); // +1 because qe/te points to the exact end, not the position after the end
	q = ksw_qinit(size, r.qe + 1, query, m, mat);
	rr = func(q, tlen, target, gapo, gape, KSW_XSTOP | r.score);
	revseq(r.qe + 1, query); revseq(r.te + 1, target);
	free(q);
	if (r.score == rr.score)
		r.tb = r.te - rr.te, r.qb = r.qe - rr.qe;
	return r;
}