X-Git-Url: https://git.donarmstrong.com/?a=blobdiff_plain;f=bam_sort.c;h=2c2b83d20d41047d3680a243762d0f1b95a0509d;hb=bd005db58f70b94ea727d5bc9c8ae09d1f392e53;hp=1817eb4e43f02e11b4310d7dc25355b6a3090dfc;hpb=8de1cc9006455976f86ddc862bcc5bff146494fe;p=samtools.git diff --git a/bam_sort.c b/bam_sort.c index 1817eb4..2c2b83d 100644 --- a/bam_sort.c +++ b/bam_sort.c @@ -1,6 +1,7 @@ #include #include #include +#include #include #include #include @@ -9,46 +10,60 @@ static int g_is_by_qname = 0; -static inline int strnum_cmp(const char *a, const char *b) +static int strnum_cmp(const char *_a, const char *_b) { - char *pa, *pb; - pa = (char*)a; pb = (char*)b; + const unsigned char *a = (const unsigned char*)_a, *b = (const unsigned char*)_b; + const unsigned char *pa = a, *pb = b; while (*pa && *pb) { if (isdigit(*pa) && isdigit(*pb)) { - long ai, bi; - ai = strtol(pa, &pa, 10); - bi = strtol(pb, &pb, 10); - if (ai != bi) return aibi? 1 : 0; + while (*pa == '0') ++pa; + while (*pb == '0') ++pb; + while (isdigit(*pa) && isdigit(*pb) && *pa == *pb) ++pa, ++pb; + if (isdigit(*pa) && isdigit(*pb)) { + int i = 0; + while (isdigit(pa[i]) && isdigit(pb[i])) ++i; + return isdigit(pa[i])? 1 : isdigit(pb[i])? -1 : (int)*pa - (int)*pb; + } else if (isdigit(*pa)) return 1; + else if (isdigit(*pb)) return -1; + else if (pa - a != pb - b) return pa - a < pb - b? 1 : -1; } else { - if (*pa != *pb) break; + if (*pa != *pb) return (int)*pa - (int)*pb; ++pa; ++pb; } } - if (*pa == *pb) - return (pa-a) < (pb-b)? -1 : (pa-a) > (pb-b)? 1 : 0; - return *pa<*pb? -1 : *pa>*pb? 1 : 0; + return *pa? 1 : *pb? -1 : 0; } #define HEAP_EMPTY 0xffffffffffffffffull typedef struct { int i; - uint64_t pos; + uint64_t pos, idx; bam1_t *b; } heap1_t; +#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)))) + static inline int heap_lt(const heap1_t a, const heap1_t b) { if (g_is_by_qname) { int t; if (a.b == 0 || b.b == 0) return a.b == 0? 1 : 0; t = strnum_cmp(bam1_qname(a.b), bam1_qname(b.b)); - return (t > 0 || (t == 0 && a.pos > b.pos)); - } else return (a.pos > b.pos); + return (t > 0 || (t == 0 && __pos_cmp(a, b))); + } else return __pos_cmp(a, b); } KSORT_INIT(heap, heap1_t, heap_lt) +static void swap_header_targets(bam_header_t *h1, bam_header_t *h2) +{ + bam_header_t t; + t.n_targets = h1->n_targets, h1->n_targets = h2->n_targets, h2->n_targets = t.n_targets; + t.target_name = h1->target_name, h1->target_name = h2->target_name, h2->target_name = t.target_name; + t.target_len = h1->target_len, h1->target_len = h2->target_len, h2->target_len = t.target_len; +} + static void swap_header_text(bam_header_t *h1, bam_header_t *h2) { int tempi; @@ -57,6 +72,11 @@ static void swap_header_text(bam_header_t *h1, bam_header_t *h2) temps = h1->text, h1->text = h2->text, h2->text = temps; } +#define MERGE_RG 1 +#define MERGE_UNCOMP 2 +#define MERGE_LEVEL1 4 +#define MERGE_FORCE 8 + /*! @abstract Merge multiple sorted BAM. @param is_by_qname whether to sort by query name @@ -69,29 +89,50 @@ static void swap_header_text(bam_header_t *h1, bam_header_t *h2) @discussion Padding information may NOT correctly maintained. This function is NOT thread safe. */ -void bam_merge_core(int by_qname, const char *out, const char *headers, int n, char * const *fn) +int bam_merge_core(int by_qname, const char *out, const char *headers, int n, char * const *fn, + int flag, const char *reg) { bamFile fpout, *fp; heap1_t *heap; bam_header_t *hout = 0; bam_header_t *hheaders = NULL; - int i, j; + int i, j, *RG_len = 0; + uint64_t idx = 0; + char **RG = 0; + bam_iter_t *iter = 0; if (headers) { tamFile fpheaders = sam_open(headers); if (fpheaders == 0) { - fprintf(stderr, "[bam_merge_core] Cannot open file `%s'. Continue anyway.\n", headers); - } else { - hheaders = sam_header_read(fpheaders); - sam_close(fpheaders); + const char *message = strerror(errno); + fprintf(stderr, "[bam_merge_core] cannot open '%s': %s\n", headers, message); + return -1; } + hheaders = sam_header_read(fpheaders); + sam_close(fpheaders); } g_is_by_qname = by_qname; fp = (bamFile*)calloc(n, sizeof(bamFile)); heap = (heap1_t*)calloc(n, sizeof(heap1_t)); + iter = (bam_iter_t*)calloc(n, sizeof(bam_iter_t)); + // prepare RG tag + if (flag & MERGE_RG) { + RG = (char**)calloc(n, sizeof(void*)); + RG_len = (int*)calloc(n, sizeof(int)); + for (i = 0; i != n; ++i) { + int l = strlen(fn[i]); + const char *s = fn[i]; + if (l > 4 && strcmp(s + l - 4, ".bam") == 0) l -= 4; + for (j = l - 1; j >= 0; --j) if (s[j] == '/') break; + ++j; l -= j; + RG[i] = calloc(l + 1, 1); + RG_len[i] = l; + strncpy(RG[i], s + j, l); + } + } + // read the first for (i = 0; i != n; ++i) { - heap1_t *h; bam_header_t *hin; fp[i] = bam_open(fn[i], "r"); if (fp[i] == 0) { @@ -100,58 +141,100 @@ void bam_merge_core(int by_qname, const char *out, const char *headers, int n, c for (j = 0; j < i; ++j) bam_close(fp[j]); free(fp); free(heap); // FIXME: possible memory leak - return; + return -1; } hin = bam_header_read(fp[i]); - if (i == 0) { // the first SAM + if (i == 0) { // the first BAM hout = hin; - if (hheaders) { - // If the text headers to be swapped in include any @SQ headers, - // check that they are consistent with the existing binary list - // of reference information. - if (hheaders->n_targets > 0) { - if (hout->n_targets != hheaders->n_targets) - fprintf(stderr, "[bam_merge_core] number of @SQ headers in `%s' differs from number of target sequences", headers); - for (j = 0; j < hout->n_targets; ++j) - if (strcmp(hout->target_name[j], hheaders->target_name[j]) != 0) - fprintf(stderr, "[bam_merge_core] @SQ header '%s' in '%s' differs from target sequence", hheaders->target_name[j], headers); - } - swap_header_text(hout, hheaders); - bam_header_destroy(hheaders); - hheaders = NULL; - } } else { // validate multiple baf - if (hout->n_targets != hin->n_targets) { - fprintf(stderr, "[bam_merge_core] file '%s' has different number of target sequences. Abort!\n", fn[i]); - exit(1); - } - for (j = 0; j < hout->n_targets; ++j) { - if (strcmp(hout->target_name[j], hin->target_name[j])) { - fprintf(stderr, "[bam_merge_core] different target sequence name: '%s' != '%s' in file '%s'. Abort!\n", + int min_n_targets = hout->n_targets; + if (hin->n_targets < min_n_targets) min_n_targets = hin->n_targets; + + for (j = 0; j < min_n_targets; ++j) + if (strcmp(hout->target_name[j], hin->target_name[j]) != 0) { + fprintf(stderr, "[bam_merge_core] different target sequence name: '%s' != '%s' in file '%s'\n", hout->target_name[j], hin->target_name[j], fn[i]); - exit(1); + return -1; } + + // If this input file has additional target reference sequences, + // add them to the headers to be output + if (hin->n_targets > hout->n_targets) { + swap_header_targets(hout, hin); + // FIXME Possibly we should also create @SQ text headers + // for the newly added reference sequences } + bam_header_destroy(hin); } - h = heap + i; + } + + if (hheaders) { + // If the text headers to be swapped in include any @SQ headers, + // check that they are consistent with the existing binary list + // of reference information. + if (hheaders->n_targets > 0) { + if (hout->n_targets != hheaders->n_targets) { + fprintf(stderr, "[bam_merge_core] number of @SQ headers in '%s' differs from number of target sequences\n", headers); + if (!reg) return -1; + } + for (j = 0; j < hout->n_targets; ++j) + if (strcmp(hout->target_name[j], hheaders->target_name[j]) != 0) { + fprintf(stderr, "[bam_merge_core] @SQ header '%s' in '%s' differs from target sequence\n", hheaders->target_name[j], headers); + if (!reg) return -1; + } + } + + swap_header_text(hout, hheaders); + bam_header_destroy(hheaders); + } + + if (reg) { + int tid, beg, end; + if (bam_parse_region(hout, reg, &tid, &beg, &end) < 0) { + fprintf(stderr, "[%s] Malformated region string or undefined reference name\n", __func__); + return -1; + } + for (i = 0; i < n; ++i) { + bam_index_t *idx; + idx = bam_index_load(fn[i]); + iter[i] = bam_iter_query(idx, tid, beg, end); + bam_index_destroy(idx); + } + } + + for (i = 0; i < n; ++i) { + heap1_t *h = heap + i; h->i = i; h->b = (bam1_t*)calloc(1, sizeof(bam1_t)); - if (bam_read1(fp[i], h->b) >= 0) - h->pos = ((uint64_t)h->b->core.tid<<32) | (uint32_t)h->b->core.pos<<1 | bam1_strand(h->b); + if (bam_iter_read(fp[i], iter[i], h->b) >= 0) { + h->pos = ((uint64_t)h->b->core.tid<<32) | (uint32_t)((int32_t)h->b->core.pos+1)<<1 | bam1_strand(h->b); + h->idx = idx++; + } else h->pos = HEAP_EMPTY; } - fpout = strcmp(out, "-")? bam_open(out, "w") : bam_dopen(fileno(stdout), "w"); - assert(fpout); + if (flag & MERGE_UNCOMP) fpout = strcmp(out, "-")? bam_open(out, "wu") : bam_dopen(fileno(stdout), "wu"); + else if (flag & MERGE_LEVEL1) fpout = strcmp(out, "-")? bam_open(out, "w1") : bam_dopen(fileno(stdout), "w1"); + else fpout = strcmp(out, "-")? bam_open(out, "w") : bam_dopen(fileno(stdout), "w"); + if (fpout == 0) { + fprintf(stderr, "[%s] fail to create the output file.\n", __func__); + return -1; + } bam_header_write(fpout, hout); bam_header_destroy(hout); ks_heapmake(heap, n, heap); while (heap->pos != HEAP_EMPTY) { bam1_t *b = heap->b; + if (flag & MERGE_RG) { + uint8_t *rg = bam_aux_get(b, "RG"); + if (rg) bam_aux_del(b, rg); + bam_aux_append(b, "RG", 'Z', RG_len[heap->i] + 1, (uint8_t*)RG[heap->i]); + } bam_write1_core(fpout, &b->core, b->data_len, b->data); - if ((j = bam_read1(fp[heap->i], b)) >= 0) { - heap->pos = ((uint64_t)b->core.tid<<32) | (uint32_t)b->core.pos<<1 | bam1_strand(b); + if ((j = bam_iter_read(fp[heap->i], iter[heap->i], b)) >= 0) { + heap->pos = ((uint64_t)b->core.tid<<32) | (uint32_t)((int)b->core.pos+1)<<1 | bam1_strand(b); + heap->idx = idx++; } else if (j == -1) { heap->pos = HEAP_EMPTY; free(heap->b->data); free(heap->b); @@ -160,34 +243,62 @@ void bam_merge_core(int by_qname, const char *out, const char *headers, int n, c ks_heapadjust(heap, 0, n, heap); } - for (i = 0; i != n; ++i) bam_close(fp[i]); + if (flag & MERGE_RG) { + for (i = 0; i != n; ++i) free(RG[i]); + free(RG); free(RG_len); + } + for (i = 0; i != n; ++i) { + bam_iter_destroy(iter[i]); + bam_close(fp[i]); + } bam_close(fpout); - free(fp); free(heap); + free(fp); free(heap); free(iter); + return 0; } + int bam_merge(int argc, char *argv[]) { - int c, is_by_qname = 0; - char *fn_headers = NULL; + int c, is_by_qname = 0, flag = 0, ret = 0; + char *fn_headers = NULL, *reg = 0; - while ((c = getopt(argc, argv, "h:n")) >= 0) { + while ((c = getopt(argc, argv, "h:nru1R:f")) >= 0) { switch (c) { + case 'r': flag |= MERGE_RG; break; + case 'f': flag |= MERGE_FORCE; break; case 'h': fn_headers = strdup(optarg); break; case 'n': is_by_qname = 1; break; + case '1': flag |= MERGE_LEVEL1; break; + case 'u': flag |= MERGE_UNCOMP; break; + case 'R': reg = strdup(optarg); break; } } if (optind + 2 >= argc) { fprintf(stderr, "\n"); - fprintf(stderr, "Usage: samtools merge [-n] [-h inh.sam] [...]\n\n"); + fprintf(stderr, "Usage: samtools merge [-nr] [-h inh.sam] [...]\n\n"); fprintf(stderr, "Options: -n sort by read names\n"); + fprintf(stderr, " -r attach RG tag (inferred from file names)\n"); + fprintf(stderr, " -u uncompressed BAM output\n"); + fprintf(stderr, " -f overwrite the output BAM if exist\n"); + fprintf(stderr, " -1 compress level 1\n"); + fprintf(stderr, " -R STR merge file in the specified region STR [all]\n"); fprintf(stderr, " -h FILE copy the header in FILE to [in1.bam]\n\n"); fprintf(stderr, "Note: Samtools' merge does not reconstruct the @RG dictionary in the header. Users\n"); fprintf(stderr, " must provide the correct header with -h, or uses Picard which properly maintains\n"); fprintf(stderr, " the header dictionary in merging.\n\n"); return 1; } - bam_merge_core(is_by_qname, argv[optind], fn_headers, argc - optind - 1, argv + optind + 1); + if (!(flag & MERGE_FORCE) && strcmp(argv[optind], "-")) { + FILE *fp = fopen(argv[optind], "rb"); + if (fp != NULL) { + fclose(fp); + fprintf(stderr, "[%s] File '%s' exists. Please apply '-f' to overwrite. Abort.\n", __func__, argv[optind]); + return 1; + } + } + if (bam_merge_core(is_by_qname, argv[optind], fn_headers, argc - optind - 1, argv + optind + 1, flag, reg) < 0) ret = 1; + free(reg); free(fn_headers); - return 0; + return ret; } typedef bam1_t *bam1_p; @@ -196,21 +307,26 @@ static inline int bam1_lt(const bam1_p a, const bam1_p b) { if (g_is_by_qname) { int t = strnum_cmp(bam1_qname(a), bam1_qname(b)); - return (t < 0 || (t == 0 && (((uint64_t)a->core.tid<<32|a->core.pos) < ((uint64_t)b->core.tid<<32|b->core.pos)))); - } else return (((uint64_t)a->core.tid<<32|a->core.pos) < ((uint64_t)b->core.tid<<32|b->core.pos)); + 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))))); + } else return (((uint64_t)a->core.tid<<32|(a->core.pos+1)) < ((uint64_t)b->core.tid<<32|(b->core.pos+1))); } KSORT_INIT(sort, bam1_p, bam1_lt) static void sort_blocks(int n, int k, bam1_p *buf, const char *prefix, const bam_header_t *h, int is_stdout) { - char *name; + char *name, mode[3]; int i; bamFile fp; ks_mergesort(sort, k, buf, 0); name = (char*)calloc(strlen(prefix) + 20, 1); - if (n >= 0) sprintf(name, "%s.%.4d.bam", prefix, n); - else sprintf(name, "%s.bam", prefix); - fp = is_stdout? bam_dopen(fileno(stdout), "w") : bam_open(name, "w"); + if (n >= 0) { + sprintf(name, "%s.%.4d.bam", prefix, n); + strcpy(mode, "w1"); + } else { + sprintf(name, "%s.bam", prefix); + strcpy(mode, "w"); + } + fp = is_stdout? bam_dopen(fileno(stdout), mode) : bam_open(name, mode); if (fp == 0) { fprintf(stderr, "[sort_blocks] fail to create file %s.\n", name); free(name); @@ -263,7 +379,7 @@ void bam_sort_core_ext(int is_by_qname, const char *fn, const char *prefix, size mem += ret; ++k; if (mem >= max_mem) { - sort_blocks(n++, k, buf, prefix, header, is_stdout); + sort_blocks(n++, k, buf, prefix, header, 0); mem = 0; k = 0; } } @@ -273,7 +389,7 @@ void bam_sort_core_ext(int is_by_qname, const char *fn, const char *prefix, size else { // then merge char **fns, *fnout; fprintf(stderr, "[bam_sort_core] merging from %d files...\n", n+1); - sort_blocks(n++, k, buf, prefix, header, is_stdout); + sort_blocks(n++, k, buf, prefix, header, 0); fnout = (char*)calloc(strlen(prefix) + 20, 1); if (is_stdout) sprintf(fnout, "-"); else sprintf(fnout, "%s.bam", prefix); @@ -282,7 +398,7 @@ void bam_sort_core_ext(int is_by_qname, const char *fn, const char *prefix, size fns[i] = (char*)calloc(strlen(prefix) + 20, 1); sprintf(fns[i], "%s.%.4d.bam", prefix, i); } - bam_merge_core(is_by_qname, fnout, 0, n, fns); + bam_merge_core(is_by_qname, fnout, 0, n, fns, 0, 0); free(fnout); for (i = 0; i < n; ++i) { unlink(fns[i]); @@ -306,6 +422,31 @@ void bam_sort_core(int is_by_qname, const char *fn, const char *prefix, size_t m bam_sort_core_ext(is_by_qname, fn, prefix, max_mem, 0); } + +size_t bam_sort_get_max_mem(char *max_mem_string) +{ + char c; + size_t max_mem; + size_t multiplier=1; + c=max_mem_string[strlen(max_mem_string)-1]; + switch(c) { + case 'G': + multiplier*=1024; + case 'M': + multiplier*=1024; + case 'K': + multiplier*=1024; + case 'B': + max_mem_string[strlen(max_mem_string)-1]='\0'; + break; + default: + break; + } + max_mem = multiplier * atol(max_mem_string); + // max_mem should be checked that it was not zero after atol! + return max_mem; +} + int bam_sort(int argc, char *argv[]) { size_t max_mem = 500000000; @@ -314,7 +455,7 @@ int bam_sort(int argc, char *argv[]) switch (c) { case 'o': is_stdout = 1; break; case 'n': is_by_qname = 1; break; - case 'm': max_mem = atol(optarg); break; + case 'm': max_mem = bam_sort_get_max_mem(optarg); break; } } if (optind + 2 > argc) {