10 static int g_is_by_qname = 0;
12 static inline int strnum_cmp(const char *a, const char *b)
15 pa = (char*)a; pb = (char*)b;
17 if (isdigit(*pa) && isdigit(*pb)) {
19 ai = strtol(pa, &pa, 10);
20 bi = strtol(pb, &pb, 10);
21 if (ai != bi) return ai<bi? -1 : ai>bi? 1 : 0;
23 if (*pa != *pb) break;
28 return (pa-a) < (pb-b)? -1 : (pa-a) > (pb-b)? 1 : 0;
29 return *pa<*pb? -1 : *pa>*pb? 1 : 0;
32 #define HEAP_EMPTY 0xffffffffffffffffull
40 static inline int heap_lt(const heap1_t a, const heap1_t b)
43 int t = strnum_cmp(bam1_qname(a.b), bam1_qname(b.b));
44 return (t > 0 || (t == 0 && a.pos > b.pos));
45 } else return (a.pos > b.pos);
48 KSORT_INIT(heap, heap1_t, heap_lt)
51 @abstract Merge multiple sorted BAM.
52 @param is_by_qname whether to sort by query name
53 @param out output BAM file name
54 @param n number of files to be merged
55 @param fn names of files to be merged
57 @discussion Padding information may NOT correctly maintained. This
58 function is NOT thread safe.
60 void bam_merge_core(int by_qname, const char *out, int n, char * const *fn)
64 bam_header_t *hout = 0;
67 g_is_by_qname = by_qname;
68 fp = (bamFile*)calloc(n, sizeof(bamFile));
69 heap = (heap1_t*)calloc(n, sizeof(heap1_t));
70 for (i = 0; i != n; ++i) {
73 assert(fp[i] = bam_open(fn[i], "r"));
74 hin = bam_header_read(fp[i]);
75 if (i == 0) hout = hin;
76 else { // validate multiple baf
77 if (hout->n_targets != hin->n_targets) {
78 fprintf(stderr, "[bam_merge_core] file '%s' has different number of target sequences. Abort!\n", fn[i]);
81 for (j = 0; j < hout->n_targets; ++j) {
82 if (strcmp(hout->target_name[j], hin->target_name[j])) {
83 fprintf(stderr, "[bam_merge_core] different target sequence name: '%s' != '%s' in file '%s'. Abort!\n",
84 hout->target_name[j], hin->target_name[j], fn[i]);
87 if (hout->target_len[j] != hin->target_len[j])
88 fprintf(stderr, "[bam_merge_core] different target sequence length: %d != %d in file '%s'. Continue.\n",
89 hout->target_len[j], hin->target_len[j], fn[i]);
91 bam_header_destroy(hin);
95 h->b = (bam1_t*)calloc(1, sizeof(bam1_t));
96 if (bam_read1(fp[i], h->b) >= 0)
97 h->pos = ((uint64_t)h->b->core.tid<<32) | (uint32_t)h->b->core.pos<<1 | bam1_strand(h->b);
98 else h->pos = HEAP_EMPTY;
100 fpout = strcmp(out, "-")? bam_open(out, "w") : bam_dopen(fileno(stdout), "w");
102 bam_header_write(fpout, hout);
103 bam_header_destroy(hout);
105 ks_heapmake(heap, n, heap);
106 while (heap->pos != HEAP_EMPTY) {
108 bam_write1_core(fpout, &b->core, b->data_len, b->data);
109 if ((j = bam_read1(fp[heap->i], b)) >= 0)
110 heap->pos = ((uint64_t)b->core.tid<<32) | (uint32_t)b->core.pos<<1 | bam1_strand(b);
111 else if (j == -1) heap->pos = HEAP_EMPTY;
112 else fprintf(stderr, "[bam_merge_core] '%s' is truncated. Continue anyway.\n", fn[heap->i]);
113 ks_heapadjust(heap, 0, n, heap);
116 for (i = 0; i != n; ++i) {
118 free(heap[i].b->data);
122 free(fp); free(heap);
124 int bam_merge(int argc, char *argv[])
126 int c, is_by_qname = 0;
127 while ((c = getopt(argc, argv, "n")) >= 0) {
129 case 'n': is_by_qname = 1; break;
132 if (optind + 2 >= argc) {
133 fprintf(stderr, "Usage: samtools merge [-n] <out.bam> <in1.bam> <in2.bam> [...]\n");
136 bam_merge_core(is_by_qname, argv[optind], argc - optind - 1, argv + optind + 1);
140 typedef bam1_t *bam1_p;
142 static inline int bam1_lt(const bam1_p a, const bam1_p b)
145 int t = strnum_cmp(bam1_qname(a), bam1_qname(b));
146 return (t < 0 || (t == 0 && (((uint64_t)a->core.tid<<32|a->core.pos) < ((uint64_t)b->core.tid<<32|b->core.pos))));
147 } else return (((uint64_t)a->core.tid<<32|a->core.pos) < ((uint64_t)b->core.tid<<32|b->core.pos));
149 KSORT_INIT(sort, bam1_p, bam1_lt)
151 static void sort_blocks(int n, int k, bam1_p *buf, const char *prefix, const bam_header_t *h)
156 ks_mergesort(sort, k, buf, 0);
157 name = (char*)calloc(strlen(prefix) + 20, 1);
158 if (n >= 0) sprintf(name, "%s.%.4d.bam", prefix, n);
159 else sprintf(name, "%s.bam", prefix);
160 assert(fp = bam_open(name, "w"));
162 bam_header_write(fp, h);
163 for (i = 0; i < k; ++i)
164 bam_write1_core(fp, &buf[i]->core, buf[i]->data_len, buf[i]->data);
169 @abstract Sort an unsorted BAM file based on the chromosome order
170 and the leftmost position of an alignment
172 @param is_by_qname whether to sort by query name
173 @param fn name of the file to be sorted
174 @param prefix prefix of the output and the temporary files; upon
175 sucessess, prefix.bam will be written.
176 @param max_mem approxiate maximum memory (very inaccurate)
178 @discussion It may create multiple temporary subalignment files
179 and then merge them by calling bam_merge_core(). This function is
182 void bam_sort_core(int is_by_qname, const char *fn, const char *prefix, size_t max_mem)
186 bam_header_t *header;
190 g_is_by_qname = is_by_qname;
192 fp = strcmp(fn, "-")? bam_open(fn, "r") : bam_dopen(fileno(stdin), "r");
194 header = bam_header_read(fp);
195 buf = (bam1_t**)calloc(max_mem / BAM_CORE_SIZE, sizeof(bam1_t*));
198 if (buf[k] == 0) buf[k] = (bam1_t*)calloc(1, sizeof(bam1_t));
200 if ((ret = bam_read1(fp, b)) < 0) break;
203 if (mem >= max_mem) {
204 sort_blocks(n++, k, buf, prefix, header);
209 fprintf(stderr, "[bam_sort_core] truncated file. Continue anyway.\n");
210 if (n == 0) sort_blocks(-1, k, buf, prefix, header);
213 fprintf(stderr, "[bam_sort_core] merging from %d files...\n", n+1);
214 sort_blocks(n++, k, buf, prefix, header);
215 fnout = (char*)calloc(strlen(prefix) + 20, 1);
216 sprintf(fnout, "%s.bam", prefix);
217 fns = (char**)calloc(n, sizeof(char*));
218 for (i = 0; i < n; ++i) {
219 fns[i] = (char*)calloc(strlen(prefix) + 20, 1);
220 sprintf(fns[i], "%s.%.4d.bam", prefix, i);
222 bam_merge_core(is_by_qname, fnout, n, fns);
224 for (i = 0; i < n; ++i) {
230 for (k = 0; k < max_mem / BAM_CORE_SIZE; ++k) {
237 bam_header_destroy(header);
241 int bam_sort(int argc, char *argv[])
243 size_t max_mem = 500000000;
244 int c, is_by_qname = 0;
245 while ((c = getopt(argc, argv, "nm:")) >= 0) {
247 case 'n': is_by_qname = 1; break;
248 case 'm': max_mem = atol(optarg); break;
251 if (optind + 2 > argc) {
252 fprintf(stderr, "Usage: samtools sort [-n] [-m <maxMem>] <in.bam> <out.prefix>\n");
255 bam_sort_core(is_by_qname, argv[optind], argv[optind+1], max_mem);