minimap2 0.2.21 → 0.2.24.0

data/ext/minimap2/python/mappy.pyx

@@ -0,0 +1,273 @@
+from libc.stdint cimport uint8_t, int8_t
+from libc.stdlib cimport free
+cimport cmappy
+import sys
+
+__version__ = '2.24'
+
+cmappy.mm_reset_timer()
+
+cdef class Alignment:
+	cdef int _ctg_len, _r_st, _r_en
+	cdef int _q_st, _q_en
+	cdef int _NM, _mlen, _blen
+	cdef int8_t _strand, _trans_strand
+	cdef uint8_t _mapq, _is_primary
+	cdef int _seg_id
+	cdef _ctg, _cigar, _cs, _MD # these are python objects
+
+	def __cinit__(self, ctg, cl, cs, ce, strand, qs, qe, mapq, cigar, is_primary, mlen, blen, NM, trans_strand, seg_id, cs_str, MD_str):
+		self._ctg = ctg if isinstance(ctg, str) else ctg.decode()
+		self._ctg_len, self._r_st, self._r_en = cl, cs, ce
+		self._strand, self._q_st, self._q_en = strand, qs, qe
+		self._NM, self._mlen, self._blen = NM, mlen, blen
+		self._mapq = mapq
+		self._cigar = cigar
+		self._is_primary = is_primary
+		self._trans_strand = trans_strand
+		self._seg_id = seg_id
+		self._cs = cs_str
+		self._MD = MD_str
+
+	@property
+	def ctg(self): return self._ctg
+
+	@property
+	def ctg_len(self): return self._ctg_len
+
+	@property
+	def r_st(self): return self._r_st
+
+	@property
+	def r_en(self): return self._r_en
+
+	@property
+	def strand(self): return self._strand
+
+	@property
+	def trans_strand(self): return self._trans_strand
+
+	@property
+	def blen(self): return self._blen
+
+	@property
+	def mlen(self): return self._mlen
+
+	@property
+	def NM(self): return self._NM
+
+	@property
+	def is_primary(self): return (self._is_primary != 0)
+
+	@property
+	def q_st(self): return self._q_st
+
+	@property
+	def q_en(self): return self._q_en
+
+	@property
+	def mapq(self): return self._mapq
+
+	@property
+	def cigar(self): return self._cigar
+
+	@property
+	def read_num(self): return self._seg_id + 1
+
+	@property
+	def cs(self): return self._cs
+
+	@property
+	def MD(self): return self._MD
+
+	@property
+	def cigar_str(self):
+		return "".join(map(lambda x: str(x[0]) + 'MIDNSHP=XB'[x[1]], self._cigar))
+
+	def __str__(self):
+		if self._strand > 0: strand = '+'
+		elif self._strand < 0: strand = '-'
+		else: strand = '?'
+		if self._is_primary != 0: tp = 'tp:A:P'
+		else: tp = 'tp:A:S'
+		if self._trans_strand > 0: ts = 'ts:A:+'
+		elif self._trans_strand < 0: ts = 'ts:A:-'
+		else: ts = 'ts:A:.'
+		a = [str(self._q_st), str(self._q_en), strand, self._ctg, str(self._ctg_len), str(self._r_st), str(self._r_en),
+			str(self._mlen), str(self._blen), str(self._mapq), tp, ts, "cg:Z:" + self.cigar_str]
+		if self._cs != "": a.append("cs:Z:" + self._cs)
+		return "\t".join(a)
+
+cdef class ThreadBuffer:
+	cdef cmappy.mm_tbuf_t *_b
+
+	def __cinit__(self):
+		self._b = cmappy.mm_tbuf_init()
+
+	def __dealloc__(self):
+		cmappy.mm_tbuf_destroy(self._b)
+
+cdef class Aligner:
+	cdef cmappy.mm_idx_t *_idx
+	cdef cmappy.mm_idxopt_t idx_opt
+	cdef cmappy.mm_mapopt_t map_opt
+
+	def __cinit__(self, fn_idx_in=None, preset=None, k=None, w=None, min_cnt=None, min_chain_score=None, min_dp_score=None, bw=None, best_n=None, n_threads=3, fn_idx_out=None, max_frag_len=None, extra_flags=None, seq=None, scoring=None):
+		self._idx = NULL
+		cmappy.mm_set_opt(NULL, &self.idx_opt, &self.map_opt) # set the default options
+		if preset is not None:
+			cmappy.mm_set_opt(str.encode(preset), &self.idx_opt, &self.map_opt) # apply preset
+		self.map_opt.flag |= 4 # always perform alignment
+		self.idx_opt.batch_size = 0x7fffffffffffffffL # always build a uni-part index
+		if k is not None: self.idx_opt.k = k
+		if w is not None: self.idx_opt.w = w
+		if min_cnt is not None: self.map_opt.min_cnt = min_cnt
+		if min_chain_score is not None: self.map_opt.min_chain_score = min_chain_score
+		if min_dp_score is not None: self.map_opt.min_dp_max = min_dp_score
+		if bw is not None: self.map_opt.bw = bw
+		if best_n is not None: self.map_opt.best_n = best_n
+		if max_frag_len is not None: self.map_opt.max_frag_len = max_frag_len
+		if extra_flags is not None: self.map_opt.flag |= extra_flags
+		if scoring is not None and len(scoring) >= 4:
+			self.map_opt.a, self.map_opt.b = scoring[0], scoring[1]
+			self.map_opt.q, self.map_opt.e = scoring[2], scoring[3]
+			self.map_opt.q2, self.map_opt.e2 = self.map_opt.q, self.map_opt.e
+			if len(scoring) >= 6:
+				self.map_opt.q2, self.map_opt.e2 = scoring[4], scoring[5]
+				if len(scoring) >= 7:
+					self.map_opt.sc_ambi = scoring[6]
+
+		cdef cmappy.mm_idx_reader_t *r;
+
+		if seq is None:
+			if fn_idx_out is None:
+				r = cmappy.mm_idx_reader_open(str.encode(fn_idx_in), &self.idx_opt, NULL)
+			else:
+				r = cmappy.mm_idx_reader_open(str.encode(fn_idx_in), &self.idx_opt, str.encode(fn_idx_out))
+			if r is not NULL:
+				self._idx = cmappy.mm_idx_reader_read(r, n_threads) # NB: ONLY read the first part
+				cmappy.mm_idx_reader_close(r)
+				cmappy.mm_mapopt_update(&self.map_opt, self._idx)
+				cmappy.mm_idx_index_name(self._idx)
+		else:
+			self._idx = cmappy.mappy_idx_seq(self.idx_opt.w, self.idx_opt.k, self.idx_opt.flag&1, self.idx_opt.bucket_bits, str.encode(seq), len(seq))
+			cmappy.mm_mapopt_update(&self.map_opt, self._idx)
+			self.map_opt.mid_occ = 1000 # don't filter high-occ seeds
+
+	def __dealloc__(self):
+		if self._idx is not NULL:
+			cmappy.mm_idx_destroy(self._idx)
+
+	def __bool__(self):
+		return (self._idx != NULL)
+
+	def map(self, seq, seq2=None, buf=None, cs=False, MD=False, max_frag_len=None, extra_flags=None):
+		cdef cmappy.mm_reg1_t *regs
+		cdef cmappy.mm_hitpy_t h
+		cdef ThreadBuffer b
+		cdef int n_regs
+		cdef char *cs_str = NULL
+		cdef int l_cs_str, m_cs_str = 0
+		cdef void *km
+		cdef cmappy.mm_mapopt_t map_opt
+
+		if self._idx == NULL: return
+		map_opt = self.map_opt
+		if max_frag_len is not None: map_opt.max_frag_len = max_frag_len
+		if extra_flags is not None: map_opt.flag |= extra_flags
+
+		if self._idx is NULL: return None
+		if buf is None: b = ThreadBuffer()
+		else: b = buf
+		km = cmappy.mm_tbuf_get_km(b._b)
+
+		_seq = seq if isinstance(seq, bytes) else seq.encode()
+		if seq2 is None:
+			regs = cmappy.mm_map_aux(self._idx, _seq, NULL,  &n_regs, b._b, &map_opt)
+		else:
+			_seq2 = seq2 if isinstance(seq2, bytes) else seq2.encode()
+			regs = cmappy.mm_map_aux(self._idx, _seq, _seq2, &n_regs, b._b, &map_opt)
+
+		try:
+			i = 0
+			while i < n_regs:
+				cmappy.mm_reg2hitpy(self._idx, &regs[i], &h)
+				cigar, _cs, _MD = [], '', ''
+				for k in range(h.n_cigar32): # convert the 32-bit CIGAR encoding to Python array
+					c = h.cigar32[k]
+					cigar.append([c>>4, c&0xf])
+				if cs or MD: # generate the cs and/or the MD tag, if requested
+					if cs:
+						l_cs_str = cmappy.mm_gen_cs(km, &cs_str, &m_cs_str, self._idx, &regs[i], _seq, 1)
+						_cs = cs_str[:l_cs_str] if isinstance(cs_str, str) else cs_str[:l_cs_str].decode()
+					if MD:
+						l_cs_str = cmappy.mm_gen_MD(km, &cs_str, &m_cs_str, self._idx, &regs[i], _seq)
+						_MD = cs_str[:l_cs_str] if isinstance(cs_str, str) else cs_str[:l_cs_str].decode()
+				yield Alignment(h.ctg, h.ctg_len, h.ctg_start, h.ctg_end, h.strand, h.qry_start, h.qry_end, h.mapq, cigar, h.is_primary, h.mlen, h.blen, h.NM, h.trans_strand, h.seg_id, _cs, _MD)
+				cmappy.mm_free_reg1(&regs[i])
+				i += 1
+		finally:
+			while i < n_regs:
+				cmappy.mm_free_reg1(&regs[i])
+				i += 1
+			free(regs)
+			free(cs_str)
+
+	def seq(self, str name, int start=0, int end=0x7fffffff):
+		cdef int l
+		cdef char *s
+		if self._idx == NULL: return
+		s = cmappy.mappy_fetch_seq(self._idx, name.encode(), start, end, &l)
+		if l == 0: return None
+		r = s[:l] if isinstance(s, str) else s[:l].decode()
+		free(s)
+		return r
+
+	@property
+	def k(self): return self._idx.k
+
+	@property
+	def w(self): return self._idx.w
+
+	@property
+	def n_seq(self): return self._idx.n_seq
+
+	@property
+	def seq_names(self):
+		cdef char *p
+		if self._idx == NULL: return
+		sn = []
+		for i in range(self._idx.n_seq):
+			p = self._idx.seq[i].name
+			s = p if isinstance(p, str) else p.decode()
+			sn.append(s)
+		return sn
+
+def fastx_read(fn, read_comment=False):
+	cdef cmappy.kseq_t *ks
+	ks = cmappy.mm_fastx_open(str.encode(fn))
+	if ks is NULL: return None
+	while cmappy.kseq_read(ks) >= 0:
+		if ks.qual.l > 0: qual = ks.qual.s if isinstance(ks.qual.s, str) else ks.qual.s.decode()
+		else: qual = None
+		name = ks.name.s if isinstance(ks.name.s, str) else ks.name.s.decode()
+		seq = ks.seq.s if isinstance(ks.seq.s, str) else ks.seq.s.decode()
+		if read_comment:
+			if ks.comment.l > 0: comment = ks.comment.s if isinstance(ks.comment.s, str) else ks.comment.s.decode()
+			else: comment = None
+			yield name, seq, qual, comment
+		else:
+			yield name, seq, qual
+	cmappy.mm_fastx_close(ks)
+
+def revcomp(seq):
+	l = len(seq)
+	bseq = seq if isinstance(seq, bytes) else seq.encode()
+	cdef char *s = cmappy.mappy_revcomp(l, bseq)
+	r = s[:l] if isinstance(s, str) else s[:l].decode()
+	free(s)
+	return r
+
+def verbose(v=None):
+	if v is None: v = -1
+	return cmappy.mm_verbose_level(v)

data/ext/minimap2/python/minimap2.py

@@ -0,0 +1,39 @@
+#!/usr/bin/env python
+
+import sys
+import getopt
+import mappy as mp
+
+def main(argv):
+	opts, args = getopt.getopt(argv[1:], "x:n:m:k:w:r:c")
+	if len(args) < 2:
+		print("Usage: minimap2.py [options] <ref.fa>|<ref.mmi> <query.fq>")
+		print("Options:")
+		print("  -x STR      preset: sr, map-pb, map-ont, asm5, asm10 or splice")
+		print("  -n INT      mininum number of minimizers")
+		print("  -m INT      mininum chaining score")
+		print("  -k INT      k-mer length")
+		print("  -w INT      minimizer window length")
+		print("  -r INT      band width")
+		print("  -c          output the cs tag")
+		sys.exit(1)
+
+	preset = min_cnt = min_sc = k = w = bw = None
+	out_cs = False
+	for opt, arg in opts:
+		if opt == '-x': preset = arg
+		elif opt == '-n': min_cnt = int(arg)
+		elif opt == '-m': min_chain_score = int(arg)
+		elif opt == '-r': bw = int(arg)
+		elif opt == '-k': k = int(arg)
+		elif opt == '-w': w = int(arg)
+		elif opt == '-c': out_cs = True
+
+	a = mp.Aligner(args[0], preset=preset, min_cnt=min_cnt, min_chain_score=min_sc, k=k, w=w, bw=bw)
+	if not a: raise Exception("ERROR: failed to load/build index file '{}'".format(args[0]))
+	for name, seq, qual in mp.fastx_read(args[1]): # read one sequence
+		for h in a.map(seq, cs=out_cs): # traverse hits
+			print('{}\t{}\t{}'.format(name, len(seq), h))
+
+if __name__ == "__main__":
+	main(sys.argv)

data/ext/minimap2/sdust.c

@@ -0,0 +1,213 @@
+#include <string.h>
+#include <stdint.h>
+#include <stdio.h>
+#include "kalloc.h"
+#include "kdq.h"
+#include "kvec.h"
+#include "sdust.h"
+
+#define SD_WLEN 3
+#define SD_WTOT (1<<(SD_WLEN<<1))
+#define SD_WMSK (SD_WTOT - 1)
+
+typedef struct {
+	int start, finish;
+	int r, l;
+} perf_intv_t;
+
+typedef kvec_t(perf_intv_t) perf_intv_v;
+typedef kvec_t(uint64_t) uint64_v;
+
+KDQ_INIT(int)
+
+#if defined(_NO_NT4_TBL) || defined(_SDUST_MAIN)
+unsigned char seq_nt4_table[256] = {
+	0, 1, 2, 3,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 0, 4, 1,  4, 4, 4, 2,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  3, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 0, 4, 1,  4, 4, 4, 2,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  3, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4, 
+	4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4,  4, 4, 4, 4
+};
+#else
+extern unsigned char seq_nt4_table[256];
+#endif
+
+struct sdust_buf_s {
+	kdq_t(int) *w;
+	perf_intv_v P; // the list of perfect intervals for the current window, sorted by descending start and then by ascending finish
+	uint64_v res;  // the result
+	void *km;      // memory pool
+};
+
+sdust_buf_t *sdust_buf_init(void *km)
+{
+	sdust_buf_t *buf;
+	buf = (sdust_buf_t*)kcalloc(km, 1, sizeof(sdust_buf_t));
+	buf->km = km;
+	buf->w = kdq_init(int, buf->km);
+	kdq_resize(int, buf->w, 8);
+	return buf;
+}
+
+void sdust_buf_destroy(sdust_buf_t *buf)
+{
+	if (buf == 0) return;
+	kdq_destroy(int, buf->w);
+	kfree(buf->km, buf->P.a); kfree(buf->km, buf->res.a); kfree(buf->km, buf);
+}
+
+static inline void shift_window(int t, kdq_t(int) *w, int T, int W, int *L, int *rw, int *rv, int *cw, int *cv)
+{
+	int s;
+	if ((int)kdq_size(w) >= W - SD_WLEN + 1) { // TODO: is this right for SD_WLEN!=3?
+		s = *kdq_shift(int, w);
+		*rw -= --cw[s];
+		if (*L > (int)kdq_size(w))
+			--*L, *rv -= --cv[s];
+	}
+	kdq_push(int, w, t);
+	++*L;
+	*rw += cw[t]++;
+	*rv += cv[t]++;
+	if (cv[t] * 10 > T<<1) {
+		do {
+			s = kdq_at(w, kdq_size(w) - *L);
+			*rv -= --cv[s];
+			--*L;
+		} while (s != t);
+	}
+}
+
+static inline void save_masked_regions(void *km, uint64_v *res, perf_intv_v *P, int start)
+{
+	int i, saved = 0;
+	perf_intv_t *p;
+	if (P->n == 0 || P->a[P->n - 1].start >= start) return;
+	p = &P->a[P->n - 1];
+	if (res->n) {
+		int s = res->a[res->n - 1]>>32, f = (uint32_t)res->a[res->n - 1];
+		if (p->start <= f) // if overlapping with or adjacent to the previous interval
+			saved = 1, res->a[res->n - 1] = (uint64_t)s<<32 | (f > p->finish? f : p->finish);
+	}
+	if (!saved) kv_push(uint64_t, km, *res, (uint64_t)p->start<<32|p->finish);
+	for (i = P->n - 1; i >= 0 && P->a[i].start < start; --i); // remove perfect intervals that have falled out of the window
+	P->n = i + 1;
+}
+
+static void find_perfect(void *km, perf_intv_v *P, const kdq_t(int) *w, int T, int start, int L, int rv, const int *cv)
+{
+	int c[SD_WTOT], r = rv, i, max_r = 0, max_l = 0;
+	memcpy(c, cv, SD_WTOT * sizeof(int));
+	for (i = (long)kdq_size(w) - L - 1; i >= 0; --i) {
+		int j, t = kdq_at(w, i), new_r, new_l;
+		r += c[t]++;
+		new_r = r, new_l = kdq_size(w) - i - 1;
+		if (new_r * 10 > T * new_l) {
+			for (j = 0; j < (int)P->n && P->a[j].start >= i + start; ++j) { // find insertion position
+				perf_intv_t *p = &P->a[j];
+				if (max_r == 0 || p->r * max_l > max_r * p->l)
+					max_r = p->r, max_l = p->l;
+			}
+			if (max_r == 0 || new_r * max_l >= max_r * new_l) { // then insert
+				max_r = new_r, max_l = new_l;
+				if (P->n == P->m) kv_resize(perf_intv_t, km, *P, P->n + 1);
+				memmove(&P->a[j+1], &P->a[j], (P->n - j) * sizeof(perf_intv_t)); // make room
+				++P->n;
+				P->a[j].start = i + start, P->a[j].finish = kdq_size(w) + (SD_WLEN - 1) + start;
+				P->a[j].r = new_r, P->a[j].l = new_l;
+			}
+		}
+	}
+}
+
+const uint64_t *sdust_core(const uint8_t *seq, int l_seq, int T, int W, int *n, sdust_buf_t *buf)
+{
+	int rv = 0, rw = 0, L = 0, cv[SD_WTOT], cw[SD_WTOT];
+	int i, start, l; // _start_: start of the current window; _l_: length of a contiguous A/C/G/T (sub)sequence
+	unsigned t; // current word
+
+	buf->P.n = buf->res.n = 0;
+	buf->w->front = buf->w->count = 0;
+	memset(cv, 0, SD_WTOT * sizeof(int));
+	memset(cw, 0, SD_WTOT * sizeof(int));
+	if (l_seq < 0) l_seq = strlen((const char*)seq);
+	for (i = l = t = 0; i <= l_seq; ++i) {
+		int b = i < l_seq? seq_nt4_table[seq[i]] : 4;
+		if (b < 4) { // an A/C/G/T base
+			++l, t = (t<<2 | b) & SD_WMSK;
+			if (l >= SD_WLEN) { // we have seen a word
+				start = (l - W > 0? l - W : 0) + (i + 1 - l); // set the start of the current window
+				save_masked_regions(buf->km, &buf->res, &buf->P, start); // save intervals falling out of the current window?
+				shift_window(t, buf->w, T, W, &L, &rw, &rv, cw, cv);
+				if (rw * 10 > L * T)
+					find_perfect(buf->km, &buf->P, buf->w, T, start, L, rv, cv);
+			}
+		} else { // N or the end of sequence; N effectively breaks input into pieces of independent sequences
+			start = (l - W + 1 > 0? l - W + 1 : 0) + (i + 1 - l);
+			while (buf->P.n) save_masked_regions(buf->km, &buf->res, &buf->P, start++); // clear up unsaved perfect intervals
+			l = t = 0;
+		}
+	}
+	*n = buf->res.n;
+	return buf->res.a;
+}
+
+uint64_t *sdust(void *km, const uint8_t *seq, int l_seq, int T, int W, int *n)
+{
+	uint64_t *ret;
+	sdust_buf_t *buf;
+	buf = sdust_buf_init(km);
+	ret = (uint64_t*)sdust_core(seq, l_seq, T, W, n, buf);
+	buf->res.a = 0;
+	sdust_buf_destroy(buf);
+	return ret;
+}
+
+#ifdef _SDUST_MAIN
+#include <zlib.h>
+#include <stdio.h>
+#include "ketopt.h"
+#include "kseq.h"
+KSEQ_INIT(gzFile, gzread)
+
+int main(int argc, char *argv[])
+{
+	gzFile fp;
+	kseq_t *ks;
+	int W = 64, T = 20, c;
+	ketopt_t o = KETOPT_INIT;
+
+	while ((c = ketopt(&o, argc, argv, 1, "w:t:", 0)) >= 0) {
+		if (c == 'w') W = atoi(o.arg);
+		else if (c == 't') T = atoi(o.arg);
+	}
+	if (o.ind == argc) {
+		fprintf(stderr, "Usage: sdust [-w %d] [-t %d] <in.fa>\n", W, T);
+		return 1;
+	}
+	fp = strcmp(argv[o.ind], "-")? gzopen(argv[o.ind], "r") : gzdopen(fileno(stdin), "r");
+	ks = kseq_init(fp);
+	while (kseq_read(ks) >= 0) {
+		uint64_t *r;
+		int i, n;
+		r = sdust(0, (uint8_t*)ks->seq.s, -1, T, W, &n);
+		for (i = 0; i < n; ++i)
+			printf("%s\t%d\t%d\n", ks->name.s, (int)(r[i]>>32), (int)r[i]);
+		free(r);
+	}
+	kseq_destroy(ks);
+	gzclose(fp);
+	return 0;
+}
+#endif

data/ext/minimap2/sdust.h

@@ -0,0 +1,25 @@
+#ifndef SDUST_H
+#define SDUST_H
+
+#include <stdint.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct sdust_buf_s;
+typedef struct sdust_buf_s sdust_buf_t;
+
+// the simple interface
+uint64_t *sdust(void *km, const uint8_t *seq, int l_seq, int T, int W, int *n);
+
+// the following interface dramatically reduce heap allocations when sdust is frequently called.
+sdust_buf_t *sdust_buf_init(void *km);
+void sdust_buf_destroy(sdust_buf_t *buf);
+const uint64_t *sdust_core(const uint8_t *seq, int l_seq, int T, int W, int *n, sdust_buf_t *buf);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

data/ext/minimap2/seed.c

@@ -0,0 +1,131 @@
+#include "mmpriv.h"
+#include "kalloc.h"
+#include "ksort.h"
+
+void mm_seed_mz_flt(void *km, mm128_v *mv, int32_t q_occ_max, float q_occ_frac)
+{
+	mm128_t *a;
+	size_t i, j, st;
+	if (mv->n <= q_occ_max || q_occ_frac <= 0.0f || q_occ_max <= 0) return;
+	KMALLOC(km, a, mv->n);
+	for (i = 0; i < mv->n; ++i)
+		a[i].x = mv->a[i].x, a[i].y = i;
+	radix_sort_128x(a, a + mv->n);
+	for (st = 0, i = 1; i <= mv->n; ++i) {
+		if (i == mv->n || a[i].x != a[st].x) {
+			int32_t cnt = i - st;
+			if (cnt > q_occ_max && cnt > mv->n * q_occ_frac)
+				for (j = st; j < i; ++j)
+					mv->a[a[j].y].x = 0;
+			st = i;
+		}
+	}
+	kfree(km, a);
+	for (i = j = 0; i < mv->n; ++i)
+		if (mv->a[i].x != 0)
+			mv->a[j++] = mv->a[i];
+	mv->n = j;
+}
+
+mm_seed_t *mm_seed_collect_all(void *km, const mm_idx_t *mi, const mm128_v *mv, int32_t *n_m_)
+{
+	mm_seed_t *m;
+	size_t i;
+	int32_t k;
+	m = (mm_seed_t*)kmalloc(km, mv->n * sizeof(mm_seed_t));
+	for (i = k = 0; i < mv->n; ++i) {
+		const uint64_t *cr;
+		mm_seed_t *q;
+		mm128_t *p = &mv->a[i];
+		uint32_t q_pos = (uint32_t)p->y, q_span = p->x & 0xff;
+		int t;
+		cr = mm_idx_get(mi, p->x>>8, &t);
+		if (t == 0) continue;
+		q = &m[k++];
+		q->q_pos = q_pos, q->q_span = q_span, q->cr = cr, q->n = t, q->seg_id = p->y >> 32;
+		q->is_tandem = q->flt = 0;
+		if (i > 0 && p->x>>8 == mv->a[i - 1].x>>8) q->is_tandem = 1;
+		if (i < mv->n - 1 && p->x>>8 == mv->a[i + 1].x>>8) q->is_tandem = 1;
+	}
+	*n_m_ = k;
+	return m;
+}
+
+#define MAX_MAX_HIGH_OCC 128
+
+void mm_seed_select(int32_t n, mm_seed_t *a, int len, int max_occ, int max_max_occ, int dist)
+{ // for high-occ minimizers, choose up to max_high_occ in each high-occ streak
+	extern void ks_heapdown_uint64_t(size_t i, size_t n, uint64_t*);
+	extern void ks_heapmake_uint64_t(size_t n, uint64_t*);
+	int32_t i, last0, m;
+	uint64_t b[MAX_MAX_HIGH_OCC]; // this is to avoid a heap allocation
+
+	if (n == 0 || n == 1) return;
+	for (i = m = 0; i < n; ++i)
+		if (a[i].n > max_occ) ++m;
+	if (m == 0) return; // no high-frequency k-mers; do nothing
+	for (i = 0, last0 = -1; i <= n; ++i) {
+		if (i == n || a[i].n <= max_occ) {
+			if (i - last0 > 1) {
+				int32_t ps = last0 < 0? 0 : (uint32_t)a[last0].q_pos>>1;
+				int32_t pe = i == n? len : (uint32_t)a[i].q_pos>>1;
+				int32_t j, k, st = last0 + 1, en = i;
+				int32_t max_high_occ = (int32_t)((double)(pe - ps) / dist + .499);
+				if (max_high_occ > 0) {
+					if (max_high_occ > MAX_MAX_HIGH_OCC)
+						max_high_occ = MAX_MAX_HIGH_OCC;
+					for (j = st, k = 0; j < en && k < max_high_occ; ++j, ++k)
+						b[k] = (uint64_t)a[j].n<<32 | j;
+					ks_heapmake_uint64_t(k, b); // initialize the binomial heap
+					for (; j < en; ++j) { // if there are more, choose top max_high_occ
+						if (a[j].n < (int32_t)(b[0]>>32)) { // then update the heap
+							b[0] = (uint64_t)a[j].n<<32 | j;
+							ks_heapdown_uint64_t(0, k, b);
+						}
+					}
+					for (j = 0; j < k; ++j) a[(uint32_t)b[j]].flt = 1;
+				}
+				for (j = st; j < en; ++j) a[j].flt ^= 1;
+				for (j = st; j < en; ++j)
+					if (a[j].n > max_max_occ)
+						a[j].flt = 1;
+			}
+			last0 = i;
+		}
+	}
+}
+
+mm_seed_t *mm_collect_matches(void *km, int *_n_m, int qlen, int max_occ, int max_max_occ, int dist, const mm_idx_t *mi, const mm128_v *mv, int64_t *n_a, int *rep_len, int *n_mini_pos, uint64_t **mini_pos)
+{
+	int rep_st = 0, rep_en = 0, n_m, n_m0;
+	size_t i;
+	mm_seed_t *m;
+	*n_mini_pos = 0;
+	*mini_pos = (uint64_t*)kmalloc(km, mv->n * sizeof(uint64_t));
+	m = mm_seed_collect_all(km, mi, mv, &n_m0);
+	if (dist > 0 && max_max_occ > max_occ) {
+		mm_seed_select(n_m0, m, qlen, max_occ, max_max_occ, dist);
+	} else {
+		for (i = 0; i < n_m0; ++i)
+			if (m[i].n > max_occ)
+				m[i].flt = 1;
+	}
+	for (i = 0, n_m = 0, *rep_len = 0, *n_a = 0; i < n_m0; ++i) {
+		mm_seed_t *q = &m[i];
+		//fprintf(stderr, "X\t%d\t%d\t%d\n", q->q_pos>>1, q->n, q->flt);
+		if (q->flt) {
+			int en = (q->q_pos >> 1) + 1, st = en - q->q_span;
+			if (st > rep_en) {
+				*rep_len += rep_en - rep_st;
+				rep_st = st, rep_en = en;
+			} else rep_en = en;
+		} else {
+			*n_a += q->n;
+			(*mini_pos)[(*n_mini_pos)++] = (uint64_t)q->q_span<<32 | q->q_pos>>1;
+			m[n_m++] = *q;
+		}
+	}
+	*rep_len += rep_en - rep_st;
+	*_n_m = n_m;
+	return m;
+}