11 static int g_is_by_qname = 0;
13 static int strnum_cmp(const char *_a, const char *_b)
15 const unsigned char *a = (const unsigned char*)_a, *b = (const unsigned char*)_b;
16 const unsigned char *pa = a, *pb = b;
18 if (isdigit(*pa) && isdigit(*pb)) {
19 while (*pa == '0') ++pa;
20 while (*pb == '0') ++pb;
21 while (isdigit(*pa) && isdigit(*pb) && *pa == *pb) ++pa, ++pb;
22 if (isdigit(*pa) && isdigit(*pb)) {
24 while (isdigit(pa[i]) && isdigit(pb[i])) ++i;
25 return isdigit(pa[i])? 1 : isdigit(pb[i])? -1 : (int)*pa - (int)*pb;
26 } else if (isdigit(*pa)) return 1;
27 else if (isdigit(*pb)) return -1;
28 else if (pa - a != pb - b) return pa - a < pb - b? 1 : -1;
30 if (*pa != *pb) return (int)*pa - (int)*pb;
34 return *pa? 1 : *pb? -1 : 0;
37 #define HEAP_EMPTY 0xffffffffffffffffull
45 #define __pos_cmp(a, b) ((a).pos > (b).pos || ((a).pos == (b).pos && ((a).i > (b).i || ((a).i == (b).i && (a).idx > (b).idx))))
47 static inline int heap_lt(const heap1_t a, const heap1_t b)
51 if (a.b == 0 || b.b == 0) return a.b == 0? 1 : 0;
52 t = strnum_cmp(bam1_qname(a.b), bam1_qname(b.b));
53 return (t > 0 || (t == 0 && __pos_cmp(a, b)));
54 } else return __pos_cmp(a, b);
57 KSORT_INIT(heap, heap1_t, heap_lt)
59 static void swap_header_targets(bam_header_t *h1, bam_header_t *h2)
62 t.n_targets = h1->n_targets, h1->n_targets = h2->n_targets, h2->n_targets = t.n_targets;
63 t.target_name = h1->target_name, h1->target_name = h2->target_name, h2->target_name = t.target_name;
64 t.target_len = h1->target_len, h1->target_len = h2->target_len, h2->target_len = t.target_len;
67 static void swap_header_text(bam_header_t *h1, bam_header_t *h2)
71 tempi = h1->l_text, h1->l_text = h2->l_text, h2->l_text = tempi;
72 temps = h1->text, h1->text = h2->text, h2->text = temps;
76 #define MERGE_UNCOMP 2
77 #define MERGE_LEVEL1 4
81 @abstract Merge multiple sorted BAM.
82 @param is_by_qname whether to sort by query name
83 @param out output BAM file name
84 @param headers name of SAM file from which to copy '@' header lines,
85 or NULL to copy them from the first file to be merged
86 @param n number of files to be merged
87 @param fn names of files to be merged
89 @discussion Padding information may NOT correctly maintained. This
90 function is NOT thread safe.
92 int bam_merge_core2(int by_qname, const char *out, const char *headers, int n, char * const *fn, int flag, const char *reg, int n_threads)
96 bam_header_t *hout = 0;
97 bam_header_t *hheaders = NULL;
98 int i, j, *RG_len = 0;
101 bam_iter_t *iter = 0;
104 tamFile fpheaders = sam_open(headers);
105 if (fpheaders == 0) {
106 const char *message = strerror(errno);
107 fprintf(stderr, "[bam_merge_core] cannot open '%s': %s\n", headers, message);
110 hheaders = sam_header_read(fpheaders);
111 sam_close(fpheaders);
114 g_is_by_qname = by_qname;
115 fp = (bamFile*)calloc(n, sizeof(bamFile));
116 heap = (heap1_t*)calloc(n, sizeof(heap1_t));
117 iter = (bam_iter_t*)calloc(n, sizeof(bam_iter_t));
119 if (flag & MERGE_RG) {
120 RG = (char**)calloc(n, sizeof(void*));
121 RG_len = (int*)calloc(n, sizeof(int));
122 for (i = 0; i != n; ++i) {
123 int l = strlen(fn[i]);
124 const char *s = fn[i];
125 if (l > 4 && strcmp(s + l - 4, ".bam") == 0) l -= 4;
126 for (j = l - 1; j >= 0; --j) if (s[j] == '/') break;
128 RG[i] = calloc(l + 1, 1);
130 strncpy(RG[i], s + j, l);
134 for (i = 0; i != n; ++i) {
136 fp[i] = bam_open(fn[i], "r");
139 fprintf(stderr, "[bam_merge_core] fail to open file %s\n", fn[i]);
140 for (j = 0; j < i; ++j) bam_close(fp[j]);
141 free(fp); free(heap);
142 // FIXME: possible memory leak
145 hin = bam_header_read(fp[i]);
146 if (i == 0) { // the first BAM
148 } else { // validate multiple baf
149 int min_n_targets = hout->n_targets;
150 if (hin->n_targets < min_n_targets) min_n_targets = hin->n_targets;
152 for (j = 0; j < min_n_targets; ++j)
153 if (strcmp(hout->target_name[j], hin->target_name[j]) != 0) {
154 fprintf(stderr, "[bam_merge_core] different target sequence name: '%s' != '%s' in file '%s'\n",
155 hout->target_name[j], hin->target_name[j], fn[i]);
159 // If this input file has additional target reference sequences,
160 // add them to the headers to be output
161 if (hin->n_targets > hout->n_targets) {
162 swap_header_targets(hout, hin);
163 // FIXME Possibly we should also create @SQ text headers
164 // for the newly added reference sequences
167 bam_header_destroy(hin);
172 // If the text headers to be swapped in include any @SQ headers,
173 // check that they are consistent with the existing binary list
174 // of reference information.
175 if (hheaders->n_targets > 0) {
176 if (hout->n_targets != hheaders->n_targets) {
177 fprintf(stderr, "[bam_merge_core] number of @SQ headers in '%s' differs from number of target sequences\n", headers);
180 for (j = 0; j < hout->n_targets; ++j)
181 if (strcmp(hout->target_name[j], hheaders->target_name[j]) != 0) {
182 fprintf(stderr, "[bam_merge_core] @SQ header '%s' in '%s' differs from target sequence\n", hheaders->target_name[j], headers);
187 swap_header_text(hout, hheaders);
188 bam_header_destroy(hheaders);
193 if (bam_parse_region(hout, reg, &tid, &beg, &end) < 0) {
194 fprintf(stderr, "[%s] Malformated region string or undefined reference name\n", __func__);
197 for (i = 0; i < n; ++i) {
199 idx = bam_index_load(fn[i]);
200 iter[i] = bam_iter_query(idx, tid, beg, end);
201 bam_index_destroy(idx);
205 for (i = 0; i < n; ++i) {
206 heap1_t *h = heap + i;
208 h->b = (bam1_t*)calloc(1, sizeof(bam1_t));
209 if (bam_iter_read(fp[i], iter[i], h->b) >= 0) {
210 h->pos = ((uint64_t)h->b->core.tid<<32) | (uint32_t)((int32_t)h->b->core.pos+1)<<1 | bam1_strand(h->b);
213 else h->pos = HEAP_EMPTY;
215 if (flag & MERGE_UNCOMP) fpout = strcmp(out, "-")? bam_open(out, "wu") : bam_dopen(fileno(stdout), "wu");
216 else if (flag & MERGE_LEVEL1) fpout = strcmp(out, "-")? bam_open(out, "w1") : bam_dopen(fileno(stdout), "w1");
217 else fpout = strcmp(out, "-")? bam_open(out, "w") : bam_dopen(fileno(stdout), "w");
219 fprintf(stderr, "[%s] fail to create the output file.\n", __func__);
222 bam_header_write(fpout, hout);
223 bam_header_destroy(hout);
224 if (!(flag & MERGE_UNCOMP)) bgzf_mt(fpout, n_threads, 256);
226 ks_heapmake(heap, n, heap);
227 while (heap->pos != HEAP_EMPTY) {
229 if (flag & MERGE_RG) {
230 uint8_t *rg = bam_aux_get(b, "RG");
231 if (rg) bam_aux_del(b, rg);
232 bam_aux_append(b, "RG", 'Z', RG_len[heap->i] + 1, (uint8_t*)RG[heap->i]);
234 bam_write1_core(fpout, &b->core, b->data_len, b->data);
235 if ((j = bam_iter_read(fp[heap->i], iter[heap->i], b)) >= 0) {
236 heap->pos = ((uint64_t)b->core.tid<<32) | (uint32_t)((int)b->core.pos+1)<<1 | bam1_strand(b);
238 } else if (j == -1) {
239 heap->pos = HEAP_EMPTY;
240 free(heap->b->data); free(heap->b);
242 } else fprintf(stderr, "[bam_merge_core] '%s' is truncated. Continue anyway.\n", fn[heap->i]);
243 ks_heapadjust(heap, 0, n, heap);
246 if (flag & MERGE_RG) {
247 for (i = 0; i != n; ++i) free(RG[i]);
248 free(RG); free(RG_len);
250 for (i = 0; i != n; ++i) {
251 bam_iter_destroy(iter[i]);
255 free(fp); free(heap); free(iter);
259 int bam_merge_core(int by_qname, const char *out, const char *headers, int n, char * const *fn, int flag, const char *reg)
261 return bam_merge_core2(by_qname, out, headers, n, fn, flag, reg, 0);
264 int bam_merge(int argc, char *argv[])
266 int c, is_by_qname = 0, flag = 0, ret = 0, n_threads = 0;
267 char *fn_headers = NULL, *reg = 0;
269 while ((c = getopt(argc, argv, "h:nru1R:f@:")) >= 0) {
271 case 'r': flag |= MERGE_RG; break;
272 case 'f': flag |= MERGE_FORCE; break;
273 case 'h': fn_headers = strdup(optarg); break;
274 case 'n': is_by_qname = 1; break;
275 case '1': flag |= MERGE_LEVEL1; break;
276 case 'u': flag |= MERGE_UNCOMP; break;
277 case 'R': reg = strdup(optarg); break;
278 case '@': n_threads = atoi(optarg); break;
281 if (optind + 2 >= argc) {
282 fprintf(stderr, "\n");
283 fprintf(stderr, "Usage: samtools merge [-nr] [-h inh.sam] <out.bam> <in1.bam> <in2.bam> [...]\n\n");
284 fprintf(stderr, "Options: -n sort by read names\n");
285 fprintf(stderr, " -r attach RG tag (inferred from file names)\n");
286 fprintf(stderr, " -u uncompressed BAM output\n");
287 fprintf(stderr, " -f overwrite the output BAM if exist\n");
288 fprintf(stderr, " -1 compress level 1\n");
289 fprintf(stderr, " -@ INT number of BAM compression threads [0]\n");
290 fprintf(stderr, " -R STR merge file in the specified region STR [all]\n");
291 fprintf(stderr, " -h FILE copy the header in FILE to <out.bam> [in1.bam]\n\n");
292 fprintf(stderr, "Note: Samtools' merge does not reconstruct the @RG dictionary in the header. Users\n");
293 fprintf(stderr, " must provide the correct header with -h, or uses Picard which properly maintains\n");
294 fprintf(stderr, " the header dictionary in merging.\n\n");
297 if (!(flag & MERGE_FORCE) && strcmp(argv[optind], "-")) {
298 FILE *fp = fopen(argv[optind], "rb");
301 fprintf(stderr, "[%s] File '%s' exists. Please apply '-f' to overwrite. Abort.\n", __func__, argv[optind]);
305 if (bam_merge_core2(is_by_qname, argv[optind], fn_headers, argc - optind - 1, argv + optind + 1, flag, reg, n_threads) < 0) ret = 1;
311 typedef bam1_t *bam1_p;
313 static inline int bam1_lt(const bam1_p a, const bam1_p b)
316 int t = strnum_cmp(bam1_qname(a), bam1_qname(b));
317 return (t < 0 || (t == 0 && (((uint64_t)a->core.tid<<32|(a->core.pos+1)) < ((uint64_t)b->core.tid<<32|(b->core.pos+1)))));
318 } else return (((uint64_t)a->core.tid<<32|(a->core.pos+1)) < ((uint64_t)b->core.tid<<32|(b->core.pos+1)));
320 KSORT_INIT(sort, bam1_p, bam1_lt)
322 static void sort_blocks(int n, int k, bam1_p *buf, const char *prefix, const bam_header_t *h, int is_stdout)
327 ks_mergesort(sort, k, buf, 0);
328 name = (char*)calloc(strlen(prefix) + 20, 1);
330 sprintf(name, "%s.%.4d.bam", prefix, n);
333 sprintf(name, "%s.bam", prefix);
336 fp = is_stdout? bam_dopen(fileno(stdout), mode) : bam_open(name, mode);
338 fprintf(stderr, "[sort_blocks] fail to create file %s.\n", name);
340 // FIXME: possible memory leak
344 bam_header_write(fp, h);
345 for (i = 0; i < k; ++i)
346 bam_write1_core(fp, &buf[i]->core, buf[i]->data_len, buf[i]->data);
351 @abstract Sort an unsorted BAM file based on the chromosome order
352 and the leftmost position of an alignment
354 @param is_by_qname whether to sort by query name
355 @param fn name of the file to be sorted
356 @param prefix prefix of the output and the temporary files; upon
357 sucessess, prefix.bam will be written.
358 @param max_mem approxiate maximum memory (very inaccurate)
360 @discussion It may create multiple temporary subalignment files
361 and then merge them by calling bam_merge_core(). This function is
364 void bam_sort_core_ext(int is_by_qname, const char *fn, const char *prefix, size_t max_mem, int is_stdout)
368 bam_header_t *header;
372 g_is_by_qname = is_by_qname;
374 fp = strcmp(fn, "-")? bam_open(fn, "r") : bam_dopen(fileno(stdin), "r");
376 fprintf(stderr, "[bam_sort_core] fail to open file %s\n", fn);
379 header = bam_header_read(fp);
380 buf = (bam1_t**)calloc(max_mem / BAM_CORE_SIZE, sizeof(bam1_t*));
383 if (buf[k] == 0) buf[k] = (bam1_t*)calloc(1, sizeof(bam1_t));
385 if ((ret = bam_read1(fp, b)) < 0) break;
388 if (mem >= max_mem) {
389 sort_blocks(n++, k, buf, prefix, header, 0);
394 fprintf(stderr, "[bam_sort_core] truncated file. Continue anyway.\n");
395 if (n == 0) sort_blocks(-1, k, buf, prefix, header, is_stdout);
398 fprintf(stderr, "[bam_sort_core] merging from %d files...\n", n+1);
399 sort_blocks(n++, k, buf, prefix, header, 0);
400 fnout = (char*)calloc(strlen(prefix) + 20, 1);
401 if (is_stdout) sprintf(fnout, "-");
402 else sprintf(fnout, "%s.bam", prefix);
403 fns = (char**)calloc(n, sizeof(char*));
404 for (i = 0; i < n; ++i) {
405 fns[i] = (char*)calloc(strlen(prefix) + 20, 1);
406 sprintf(fns[i], "%s.%.4d.bam", prefix, i);
408 bam_merge_core(is_by_qname, fnout, 0, n, fns, 0, 0);
410 for (i = 0; i < n; ++i) {
416 for (k = 0; k < max_mem / BAM_CORE_SIZE; ++k) {
423 bam_header_destroy(header);
427 void bam_sort_core(int is_by_qname, const char *fn, const char *prefix, size_t max_mem)
429 bam_sort_core_ext(is_by_qname, fn, prefix, max_mem, 0);
433 size_t bam_sort_get_max_mem(char *max_mem_string)
438 c=max_mem_string[strlen(max_mem_string)-1];
447 max_mem_string[strlen(max_mem_string)-1]='\0';
452 max_mem = multiplier * atol(max_mem_string);
453 // max_mem should be checked that it was not zero after atol!
457 int bam_sort(int argc, char *argv[])
459 size_t max_mem = 500000000;
460 int c, is_by_qname = 0, is_stdout = 0;
461 while ((c = getopt(argc, argv, "nom:")) >= 0) {
463 case 'o': is_stdout = 1; break;
464 case 'n': is_by_qname = 1; break;
465 case 'm': max_mem = bam_sort_get_max_mem(optarg); break;
468 if (optind + 2 > argc) {
469 fprintf(stderr, "Usage: samtools sort [-on] [-m <maxMem>] <in.bam> <out.prefix>\n");
472 bam_sort_core_ext(is_by_qname, argv[optind], argv[optind+1], max_mem, is_stdout);