795 lines
31 KiB
C
795 lines
31 KiB
C
/*
|
||
Description: 通过fmt-idx数据结构对seed过程进行加速（fm-index twice search in one time）
|
||
|
||
Copyright : All right reserved by ICT
|
||
|
||
Author : Zhang Zhonghai
|
||
Date : 2023/12/24
|
||
*/
|
||
|
||
#include <stdio.h>
|
||
#include <stdlib.h>
|
||
#include <string.h>
|
||
|
||
#include "fmt_idx.h"
|
||
#include "utils.h"
|
||
#include "bntseq.h"
|
||
#include "kvec.h"
|
||
|
||
const static char BASE[4] = {'A', 'C', 'G', 'T'};
|
||
|
||
// 生成所有KMER_LEN长度的序列，字符串表示
|
||
void gen_all_seq(char **seq_arr, int kmer_len)
|
||
{
|
||
uint32_t seq_up_val = (1 << (kmer_len << 1));
|
||
for (uint32_t i = 0; i < seq_up_val; ++i)
|
||
{
|
||
seq_arr[i] = (char *)malloc(kmer_len);
|
||
for (int j = kmer_len - 1; j >= 0; --j)
|
||
{
|
||
seq_arr[i][kmer_len - 1 - j] = BASE[(i >> (j << 1)) & 3];
|
||
}
|
||
}
|
||
}
|
||
|
||
// 生成occ，每个字节对应一个pattern
|
||
void fmt_gen_cnt_occ(FMTIndex *fmt)
|
||
{
|
||
// 0-8：大于a的occ，8-16：大于b的occ，16-24：b的occ
|
||
int i, a, b, ti;
|
||
uint32_t oa, ooa, ob, oob;
|
||
for (i = 0; i != 256; ++i) // 遍历单个字节的各种情况
|
||
{
|
||
for (a = 0; a < 4; ++a) // ba格式
|
||
{
|
||
oa = 0;
|
||
ooa = 0;
|
||
oa += ((i >> 4 & 3) == a) + ((i & 3) == a);
|
||
ooa += ((i >> 4 & 3) > a) + ((i & 3) > a);
|
||
for (b = 0; b < 4; ++b)
|
||
{
|
||
oob = ob = 0;
|
||
oob += ((i >> 6 & 3) > b && (i >> 4 & 3) == a) + ((i >> 2 & 3) > b && (i & 3) == a);
|
||
ob += ((i >> 6 & 3) == b && (i >> 4 & 3) == a) + ((i >> 2 & 3) == b && (i & 3) == a);
|
||
ti = a << 2 | b;
|
||
fmt->cnt_occ[ti][i] = ob << 24 | oob << 16 | oa << 8 | ooa;
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
// fmt-index的count table，4对应着bwt碱基的累积量，0,1,2,3分别对应着bwt是A,C,G,T，pre-bwt的累积量
|
||
void fmt_gen_cnt_table(uint32_t cnt_table[4][256])
|
||
{
|
||
int i, j, k;
|
||
uint32_t x = 0;
|
||
for (i = 0; i != 256; ++i) // 遍历单个字节的各种情况
|
||
{
|
||
for (k = 0; k < 4; ++k) // bwt碱基
|
||
{
|
||
x = 0; // for [A,C,G,T][A,C,G,T]
|
||
for (j = 0; j != 4; ++j) // pre-bwt碱基
|
||
x |= (((i >> 6 & 3) == j && (i >> 4 & 3) == k) + ((i >> 2 & 3) == j && (i & 3) == k)) << (j << 3);
|
||
cnt_table[k][i] = x;
|
||
}
|
||
}
|
||
}
|
||
|
||
// 将fmt结构数据写入到二进制文件
|
||
void dump_fmt(const char *fn, const FMTIndex *fmt)
|
||
{
|
||
FILE *fp;
|
||
fp = xopen(fn, "wb");
|
||
err_fwrite(&fmt->primary, sizeof(bwtint_t), 1, fp);
|
||
err_fwrite(&fmt->sec_primary, sizeof(bwtint_t), 1, fp);
|
||
err_fwrite(&fmt->sec_bcp, sizeof(uint8_t), 1, fp);
|
||
err_fwrite(&fmt->first_base, sizeof(uint8_t), 1, fp);
|
||
err_fwrite(&fmt->last_base, sizeof(uint8_t), 1, fp);
|
||
err_fwrite(fmt->L2 + 1, sizeof(bwtint_t), 4, fp);
|
||
err_fwrite(fmt->bwt, 4, fmt->bwt_size, fp);
|
||
err_fflush(fp);
|
||
err_fclose(fp);
|
||
}
|
||
|
||
// 从文件中读取fmt结构数据
|
||
FMTIndex *fmt_restore_fmt(const char *fn)
|
||
{
|
||
FMTIndex *fmt;
|
||
fmt = (FMTIndex *)calloc(1, sizeof(FMTIndex));
|
||
FILE *fp = xopen(fn, "rb");
|
||
|
||
fseek(fp, 0, SEEK_END);
|
||
fmt->bwt_size = (ftell(fp) - sizeof(bwtint_t) * 6 - 3) >> 2; // 以32位word为单位计算的size
|
||
fmt->bwt = (uint32_t *)calloc(fmt->bwt_size, 4);
|
||
fseek(fp, 0, SEEK_SET);
|
||
err_fread_noeof(&fmt->primary, sizeof(bwtint_t), 1, fp);
|
||
err_fread_noeof(&fmt->sec_primary, sizeof(bwtint_t), 1, fp);
|
||
err_fread_noeof(&fmt->sec_bcp, sizeof(uint8_t), 1, fp);
|
||
err_fread_noeof(&fmt->first_base, sizeof(uint8_t), 1, fp);
|
||
err_fread_noeof(&fmt->last_base, sizeof(uint8_t), 1, fp);
|
||
err_fread_noeof(fmt->L2 + 1, sizeof(bwtint_t), 4, fp);
|
||
fread_fix(fp, fmt->bwt_size << 2, fmt->bwt);
|
||
fmt->seq_len = fmt->L2[4];
|
||
err_fclose(fp);
|
||
fmt_gen_cnt_occ(fmt); // 字节所能表示的各种碱基组合中，各个碱基的累积数量
|
||
return fmt;
|
||
}
|
||
|
||
// 将kmer hash数据写入到文件
|
||
void fmt_dump_kmer_idx(const char *fn, const KmerHash *kh)
|
||
{
|
||
FILE *fp;
|
||
fp = xopen(fn, "wb");
|
||
err_fwrite(kh->ke10, 1, (1 << (10 << 1)) * sizeof(KmerEntryArr), fp);
|
||
err_fwrite(kh->ke11, 1, (1 << (11 << 1)) * sizeof(KmerEntry), fp);
|
||
err_fwrite(kh->ke12, 1, (1 << (12 << 1)) * sizeof(KmerEntry), fp);
|
||
err_fwrite(kh->ke13, 1, (1 << (13 << 1)) * sizeof(KmerEntry), fp);
|
||
err_fwrite(kh->ke14, 1, (1 << (14 << 1)) * sizeof(KmerEntry), fp);
|
||
err_fflush(fp);
|
||
err_fclose(fp);
|
||
}
|
||
|
||
// 从文件中读取kmer hash信息
|
||
KmerHash fmt_restore_kmer_idx(const char *fn)
|
||
{
|
||
FILE *fp = xopen(fn, "rb");
|
||
KmerHash khash;
|
||
KmerHash *kh = &khash;
|
||
int len = 1 << (10 << 1);
|
||
kh->ke10 = (KmerEntryArr *)malloc(len * sizeof(KmerEntryArr));
|
||
fread_fix(fp, len * sizeof(KmerEntryArr), kh->ke10);
|
||
len = 1 << (11 << 1);
|
||
kh->ke11 = (KmerEntry *)malloc(len * sizeof(KmerEntry));
|
||
fread_fix(fp, len * sizeof(KmerEntry), kh->ke11);
|
||
len = 1 << (12 << 1);
|
||
kh->ke12 = (KmerEntry *)malloc(len * sizeof(KmerEntry));
|
||
fread_fix(fp, len * sizeof(KmerEntry), kh->ke12);
|
||
len = 1 << (13 << 1);
|
||
kh->ke13 = (KmerEntry *)malloc(len * sizeof(KmerEntry));
|
||
fread_fix(fp, len * sizeof(KmerEntry), kh->ke13);
|
||
len = 1 << (14 << 1);
|
||
kh->ke14 = (KmerEntry *)malloc(len * sizeof(KmerEntry));
|
||
fread_fix(fp, len * sizeof(KmerEntry), kh->ke14);
|
||
err_fclose(fp);
|
||
return khash;
|
||
}
|
||
|
||
// 读取sa数据
|
||
void fmt_restore_sa(const char *fn, FMTIndex *fmt)
|
||
{
|
||
char skipped[256];
|
||
FILE *fp;
|
||
bwtint_t primary;
|
||
fp = xopen(fn, "rb");
|
||
err_fread_noeof(&primary, sizeof(bwtint_t), 1, fp);
|
||
xassert(primary == fmt->primary, "SA-BWT inconsistency: primary is not the same.");
|
||
err_fread_noeof(skipped, sizeof(bwtint_t), 4, fp); // skip
|
||
err_fread_noeof(&fmt->sa_intv, sizeof(bwtint_t), 1, fp);
|
||
err_fread_noeof(&primary, sizeof(bwtint_t), 1, fp);
|
||
xassert(primary == fmt->seq_len, "SA-BWT inconsistency: seq_len is not the same.");
|
||
fmt->n_sa = (fmt->seq_len + fmt->sa_intv) / fmt->sa_intv;
|
||
fmt->sa = (uint8_t *)malloc(SA_BYTES(fmt->n_sa));
|
||
fread_fix(fp, SA_BYTES(fmt->n_sa), fmt->sa);
|
||
err_fclose(fp);
|
||
}
|
||
|
||
// 根据interval-bwt创建fmt-index
|
||
FMTIndex *create_fmt_from_bwt(bwt_t *bwt)
|
||
{
|
||
// FILE *fmt_out = fopen("fmt.txt", "w");
|
||
FMTIndex *fmt = (FMTIndex *)calloc(1, sizeof(FMTIndex));
|
||
fmt_gen_cnt_occ(fmt);
|
||
|
||
bwtint_t i, j, k, m, n, n_occ, cnt[4], cnt2[4];
|
||
uint32_t c[4], c2[16]; /*c用来保存原来的bwt碱基串的累积值，c2用来保存pre-bwt和bwt碱基对的累计值，如AA..TT*/
|
||
uint32_t *buf; /* 计算之后变成fmt结构中bwt部分 */
|
||
|
||
#ifdef FMT_MID_INTERVAL
|
||
// 加入中间的check point
|
||
uint32_t mc[4] = {0};
|
||
uint32_t cnt_table[4][256]; // 4对应原来的cnt_table，0,1,2,3,分别对应该碱基的扩展
|
||
fmt_gen_cnt_table(cnt_table);
|
||
#endif
|
||
|
||
fmt->seq_len = bwt->seq_len; // bwt碱基序列的长度，不包含$字符，也就是该长度比bwt matrix长度少1
|
||
for (i = 0; i < 5; ++i)
|
||
fmt->L2[i] = bwt->L2[i]; // 每个碱基的总累积值
|
||
fmt->primary = bwt->primary; // $在末尾的行，在bwt matrix行中的排序位置
|
||
|
||
n_occ = (bwt->seq_len + FMT_OCC_INTERVAL - 1) / FMT_OCC_INTERVAL + 1; // check point 个数
|
||
fmt->bwt_size = (fmt->seq_len * 2 + 15) >> 4; // 要保存最后两列碱基
|
||
fmt->bwt_size += n_occ * 20; // A,C,G,T和AA,AC.....TG,TT共20个
|
||
|
||
#ifdef FMT_MID_INTERVAL
|
||
uint32_t s1;
|
||
bwtint_t mn_occ = (bwt->seq_len >> FMT_OCC_INTV_SHIFT) * (FMT_OCC_INTERVAL / FMT_MID_INTERVAL - 1);
|
||
bwtint_t last_seq_len = bwt->seq_len % FMT_OCC_INTERVAL;
|
||
mn_occ += (last_seq_len + FMT_MID_INTERVAL - 1) / FMT_MID_INTERVAL - 1;
|
||
fmt->bwt_size += mn_occ * 4;
|
||
i = 0;
|
||
#endif
|
||
|
||
buf = (uint32_t *)calloc(fmt->bwt_size, 4); // 开辟计算fmt用到的缓存
|
||
c[0] = c[1] = c[2] = c[3] = 0;
|
||
// 首行的c2，应该是对应的ACGT对应的行，减去1的occ
|
||
for (i = 0; i < 4; ++i)
|
||
{
|
||
bwtint_t before_first_line = fmt->L2[i];
|
||
bwt_occ4(bwt, before_first_line, cnt);
|
||
for (j = i * 4, k = 0; k < 4; ++j, ++k)
|
||
c2[j] = cnt[k];
|
||
}
|
||
// k表示buf存储的偏移量
|
||
for (i = k = 0; i < bwt->seq_len; ++i)
|
||
{
|
||
// 记录occ
|
||
if (i % FMT_OCC_INTERVAL == 0)
|
||
{
|
||
memcpy(buf + k, c, sizeof(uint32_t) * 4); // bwt str中各个碱基的occ
|
||
k += 4;
|
||
memcpy(buf + k, c2, sizeof(uint32_t) * 16); // pre-bwt:bwt碱基对的occ
|
||
k += 16;
|
||
#ifdef FMT_MID_INTERVAL
|
||
mc[0] = mc[1] = mc[2] = mc[3] = 0;
|
||
#endif
|
||
}
|
||
// 每个32位整数保存8个倒数第二列碱基（pre-bwt）和8个倒数第一列(bwt)碱基
|
||
if (i % 16 == 0) // 每个32位整数可以包含16个碱基，每次需要处理16个碱基，也就是间隔最小可以设置为16
|
||
{
|
||
uint32_t pre_bwt_16_seq = 0; // 16个pre-bwt碱基串
|
||
uint32_t *bwt_addr = bwt_occ_intv(bwt, i) + 4; // 这里加4还是加8要看保存occ的是是uint32还是uint64，bwt字符串i对应的基准行，因为原始的bwt-cp（check point）包含由4个uint32_t(8个uint32_t)组成的occ信息
|
||
int offset = (i % OCC_INTERVAL) / 16; // 每OCC_INTERVAL个碱基共享同一个基准地址，每16个碱基共用一个uint32整型，因此需要偏移量来获取当前碱基串的首地址
|
||
uint32_t bwt_16_seq = *(bwt_addr + offset); // 待处理的当前16个碱基串的首地址
|
||
for (j = 0; j < 16; ++j) // 对于bwt碱基串，一个一个碱基分别处理
|
||
{
|
||
bwtint_t cur_str_line = i + j; // 当前碱基在bwt str中的行排序
|
||
if (cur_str_line < bwt->seq_len) // 当前碱基行不应超出bwt str总碱基长度（bwt str长度比bwt matrix长度少1，因为bwt str不包含$）
|
||
{
|
||
uint8_t bwt_base = bwt_B0(bwt, cur_str_line); // 对应行的bwt的碱基
|
||
// 先求出该碱基对应在第一列的行（对应的bwt matrix行）
|
||
bwtint_t cur_mtx_line = cur_str_line;
|
||
if (cur_str_line >= bwt->primary) // 因为bwt序列里除去了$符号，所以，超过$所在行之后，对应的seq位置应该加一，才是真正对应bwt matrix的行
|
||
cur_mtx_line += 1;
|
||
bwt_occ4(bwt, cur_mtx_line, cnt); // 获取原来bwt-checkpoint中的occ值
|
||
for (m = 0; m < 4; ++m)
|
||
c[m] = (uint32_t)cnt[m]; // 碱基m在cur_bwt_mtx_line(包含)之前的累积值，直接拷贝原bwt中的occ即可
|
||
|
||
cnt[bwt_base] -= 1; // 得到cur_bwt_mtx_line(不包含)之前的累积量，即bwt_occ4(bwt, cur_bwt_mtx_line-1, cnt)
|
||
bwtint_t bwt_base_mtx_line = bwt->L2[bwt_base] + 1 + cnt[bwt_base]; // bwt_base对应的bwt matrix行（LF变换）
|
||
|
||
bwt_occ4(bwt, bwt_base_mtx_line, cnt2); // 计算bwt_base_mtx_line之前的occ
|
||
for (n = 0; n < 4; ++n)
|
||
{
|
||
int c2_idx = bwt_base << 2 | n; // bwt base放在前边
|
||
c2[c2_idx] = (uint32_t)cnt2[n]; // pre-bwt:bwt 碱基对的累计值
|
||
}
|
||
bwtint_t bwt_base_str_line = bwt_base_mtx_line; // bwt-str中对应的行排序
|
||
if (bwt_base_str_line >= bwt->primary) // base_line表示在bwt str中的位置，所以超出$为最尾所在行之后，要减掉1
|
||
bwt_base_str_line -= 1; // bwt碱基序列行（不包含$）
|
||
uint32_t pre_bwt_base = bwt_B0(bwt, bwt_base_str_line); // bwt列碱基对应的前一个碱基pre-bwt
|
||
// 此时，bwt_base对应的bwt matrix首行，是$排在最尾的行，说明bwt_base就是序列的第一个碱基，
|
||
// 此时计算出来的pre_bwt_base就是primary前一行的bwt base，以此来代替$字符，在后续的计算过程中需要考虑
|
||
if (bwt_base_mtx_line == bwt->primary)
|
||
{
|
||
// 计算sec_bcp
|
||
fmt->sec_bcp = pre_bwt_base << 2 | bwt_base; // 因为把$当成A处理了
|
||
fmt->sec_primary = cur_mtx_line; // pre-bwt base为$的行排序（bwt-matrix行）
|
||
fmt->first_base = bwt_base; // 原始序列第一个碱基
|
||
fmt->last_base = pre_bwt_base; // 计算后替代$字符的碱基（应该是primary行上边一行对应的bwt base）
|
||
}
|
||
// 暂存 pre-bwt碱基序列
|
||
pre_bwt_16_seq = pre_bwt_16_seq | (pre_bwt_base << (15 - j) * 2); // 序列靠前的碱基排在uint32_t数据中的高位
|
||
}
|
||
else
|
||
break;
|
||
}
|
||
// 保存bwt和pre_bwt
|
||
uint32_t pre_and_bwt_seq = 0;
|
||
uint32_t pre_and_bwt_seq_2 = 0;
|
||
for (m = 0; m < 8; ++m)
|
||
{
|
||
int lshift_bit = 30 - 2 * m;
|
||
pre_and_bwt_seq |= (((pre_bwt_16_seq & (3 << lshift_bit)) >> (m * 2)) | ((bwt_16_seq & (3 << lshift_bit)) >> ((m * 2) + 2)));
|
||
}
|
||
buf[k++] = pre_and_bwt_seq;
|
||
|
||
if (j > 8)
|
||
{
|
||
for (m = 8; m > 0; --m)
|
||
{
|
||
int lshift_bit = 2 * m - 2;
|
||
pre_and_bwt_seq_2 |= (((pre_bwt_16_seq & (3 << lshift_bit)) << (m * 2)) | ((bwt_16_seq & (3 << lshift_bit)) << (m * 2 - 2)));
|
||
}
|
||
|
||
#ifdef FMT_MID_INTERVAL // 计算前边8+8个碱基的mid interval occ
|
||
s1 = pre_and_bwt_seq;
|
||
for (m = 0; m < 4; ++m)
|
||
mc[m] += cnt_table[m][s1 & 0xff] + cnt_table[m][s1 >> 8 & 0xff] + cnt_table[m][s1 >> 16 & 0xff] + cnt_table[m][s1 >> 24];
|
||
#endif
|
||
|
||
#if FMT_MID_INTERVAL == 8 // 如果mid interval是8的话，这里要保存一次
|
||
for (m = 0; m < 4; ++m)
|
||
buf[k++] = mc[m];
|
||
#endif
|
||
|
||
buf[k++] = pre_and_bwt_seq_2;
|
||
|
||
#ifdef FMT_MID_INTERVAL
|
||
s1 = pre_and_bwt_seq_2;
|
||
for (m = 0; m < 4; ++m)
|
||
mc[m] += cnt_table[m][s1 & 0xff] + cnt_table[m][s1 >> 8 & 0xff] + cnt_table[m][s1 >> 16 & 0xff] + cnt_table[m][s1 >> 24];
|
||
|
||
if ((i + 16) % FMT_OCC_INTERVAL != 0 && j == 16 && ((i + 16) & FMT_MID_INTV_MASK) == 0)
|
||
for (m = 0; m < 4; ++m)
|
||
buf[k++] = mc[m];
|
||
#endif
|
||
}
|
||
}
|
||
}
|
||
// the last element
|
||
memcpy(buf + k, c, sizeof(uint32_t) * 4);
|
||
k += 4;
|
||
memcpy(buf + k, c2, sizeof(uint32_t) * 16);
|
||
k += 16;
|
||
xassert(k == fmt->bwt_size, "inconsistent bwt_size");
|
||
// update fmt
|
||
fmt->bwt = buf;
|
||
return fmt;
|
||
}
|
||
|
||
// 扩展两个个碱基，计算bwt base为b的pre-bwt str中各个碱基的occ
|
||
inline void fmt_e2_occ(const FMTIndex *fmt, bwtint_t k, int b1, int b2, bwtint_t cnt[4])
|
||
{
|
||
uint32_t x = 0;
|
||
uint32_t *p, *q, tmp;
|
||
bwtint_t str_line = k, cp_line = k & (~FMT_OCC_INTV_MASK); // cp = check point
|
||
int i, ti = b1 << 2 | b2;
|
||
cnt[0] = 0;
|
||
cnt[1] = 0;
|
||
cnt[2] = 0;
|
||
if (k == (bwtint_t)(-1))
|
||
{
|
||
p = fmt->bwt + 4 + b1 * 4;
|
||
for (i = b2 + 1; i < 4; ++i)
|
||
cnt[2] += p[i];
|
||
cnt[3] = p[b2];
|
||
return;
|
||
}
|
||
k -= (k >= fmt->primary); // k由bwt矩阵对应的行转换成bwt字符串对应的行（去掉了$，所以大于$的行，都减掉1）
|
||
p = fmt_occ_intv(fmt, k);
|
||
// fprintf(stderr, "k: %ld\n", k);
|
||
for (i = b1 + 1; i < 4; ++i)
|
||
cnt[0] += p[i]; // 大于b1的碱基的occ之和
|
||
cnt[1] = p[b1]; // b1的occ
|
||
q = p + 4 + b1 * 4;
|
||
for (i = b2 + 1; i < 4; ++i)
|
||
cnt[2] += q[i]; // 大于b2的occ之和
|
||
cnt[3] = q[b2]; // b2的occ
|
||
p += 20;
|
||
|
||
// 使用mid interval信息
|
||
int mk = k % FMT_OCC_INTERVAL;
|
||
int n_mintv = mk >> FMT_MID_INTV_SHIFT;
|
||
if (n_mintv > 0) // 至少超过了第一个mid interval
|
||
{
|
||
p += n_mintv * (4 + (FMT_MID_INTERVAL >> 3)) - 4; // 对应的mid interval check point的首地址，即A C G T的局部累积量
|
||
q = p + b1;
|
||
for (i = b1 + 1; i < 4; ++i)
|
||
x += p[i]; // 大于b1的碱基的occ之和
|
||
cnt[0] += __fmt_mid_sum(x);
|
||
x = *q;
|
||
cnt[1] += __fmt_mid_sum(x); // b1的occ
|
||
for (i = 3; i > b2; --i)
|
||
cnt[2] += x >> (i << 3) & 0xff; // 大于b2的occ之和
|
||
cnt[3] += x >> (b2 << 3) & 0xff; // b2的occ
|
||
x = 0;
|
||
p += 4;
|
||
}
|
||
uint32_t *end = p + ((k >> 3) - ((k & ~FMT_MID_INTV_MASK) >> 3));
|
||
for (; p < end; ++p)
|
||
{
|
||
x += __fmt_occ_e2_aux2(fmt, ti, *p);
|
||
}
|
||
|
||
tmp = *p & ~((1U << ((~k & 7) << 2)) - 1);
|
||
x += __fmt_occ_e2_aux2(fmt, ti, tmp);
|
||
|
||
if (b1 == 0)
|
||
{
|
||
x -= (~k & 7) << 8;
|
||
if (b2 == 0)
|
||
x -= (~k & 7) << 24;
|
||
}
|
||
// 如果跨过了second_primary,那么可能需要减掉一次累积值
|
||
if (b1 == fmt->first_base && cp_line < fmt->sec_primary && str_line >= fmt->sec_primary)
|
||
{
|
||
if (b2 < fmt->last_base)
|
||
cnt[2] -= 1;
|
||
else if (b2 == fmt->last_base)
|
||
cnt[3] -= 1;
|
||
}
|
||
cnt[0] += x & 0xff;
|
||
cnt[1] += x >> 8 & 0xff;
|
||
cnt[2] += x >> 16 & 0xff;
|
||
cnt[3] += x >> 24 & 0xff;
|
||
}
|
||
|
||
// 扩展两个碱基
|
||
inline void fmt_extend2(const FMTIndex *fmt, bwtintv_t *ik, bwtintv_t *ok1, bwtintv_t *ok2, int is_back, int b1, int b2)
|
||
{
|
||
bwtint_t tk[4], tl[4];
|
||
bwtintv_t intv = {0};
|
||
// tk表示在k行之前所有各个碱基累积出现次数，tl表示在l行之前的累积
|
||
fmt_e2_occ(fmt, ik->x[!is_back] - 1, b1, b2, tk);
|
||
fmt_e2_occ(fmt, ik->x[!is_back] - 1 + ik->x[2], b1, b2, tl);
|
||
// 第一次扩展
|
||
intv.x[!is_back] = fmt->L2[b1] + 1 + tk[1];
|
||
intv.x[is_back] = ik->x[is_back] + (ik->x[!is_back] <= fmt->primary && ik->x[!is_back] + ik->x[2] - 1 >= fmt->primary) + tl[0] - tk[0];
|
||
intv.x[2] = tl[1] - tk[1];
|
||
*ok1 = intv;
|
||
// 第二次扩展
|
||
intv.x[is_back] = intv.x[is_back] + (intv.x[!is_back] <= fmt->primary && intv.x[!is_back] + intv.x[2] - 1 >= fmt->primary) + tl[2] - tk[2];
|
||
intv.x[!is_back] = fmt->L2[b2] + 1 + tk[3];
|
||
intv.x[2] = tl[3] - tk[3];
|
||
*ok2 = intv;
|
||
}
|
||
|
||
// 扩展一个碱基
|
||
inline void fmt_extend1(const FMTIndex *fmt, bwtintv_t *ik, bwtintv_t *ok, int is_back, int b1)
|
||
{
|
||
bwtint_t tk[4], tl[4];
|
||
int b2 = 3; // 如果只扩展一次，那么第二个碱基设置成T，可以减小一些计算量，如计算大于b2的累积数量
|
||
// tk表示在k行之前所有各个碱基累积出现次数，tl表示在l行之前的累积
|
||
fmt_e2_occ(fmt, ik->x[!is_back] - 1, b1, b2, tk);
|
||
fmt_e2_occ(fmt, ik->x[!is_back] - 1 + ik->x[2], b1, b2, tl);
|
||
// 这里是反向扩展
|
||
ok->x[!is_back] = fmt->L2[b1] + 1 + tk[1];
|
||
ok->x[2] = tl[1] - tk[1];
|
||
// 第一次正向扩展
|
||
ok->x[is_back] = ik->x[is_back] + (ik->x[!is_back] <= fmt->primary && ik->x[!is_back] + ik->x[2] - 1 >= fmt->primary) + tl[0] - tk[0];
|
||
}
|
||
|
||
// 序列和参考基因直接对比
|
||
static void direct_extend(const FMTIndex *fmt, int len, const uint8_t *q, int left_pos, int right_pos, bwtint_t mtx_line, bwtintv_t *mt)
|
||
{
|
||
#define PAC_BASE(pac, l) ((pac)[(l) >> 2] >> ((~(l) & 3) << 1) & 3)
|
||
#define EXTEND_BASE_LOOP(qcond, rcond, qstep, rstep) \
|
||
while (k != qcond && r != rcond) \
|
||
{ \
|
||
const int base = PAC_BASE(fmt->pac, r); \
|
||
if (q[k] != base) break; \
|
||
k += qstep; \
|
||
r += rstep; \
|
||
}
|
||
#define EXTEND_BASE_LOOP_COMP(qcond, rcond, qstep, rstep) \
|
||
while (k != qcond && r != rcond) \
|
||
{ \
|
||
const int base = 3 - PAC_BASE(fmt->pac, r); \
|
||
if (q[k] != base) break; \
|
||
k += qstep; \
|
||
r += rstep; \
|
||
}
|
||
|
||
int k;
|
||
int64_t r, rp;
|
||
mt->num_match = 1;
|
||
rp = fmt_sa(fmt, mtx_line);
|
||
r = rp >= fmt->l_pac ? (fmt->l_pac << 1) - 1 - rp : rp;
|
||
k = right_pos;
|
||
if (rp < fmt->l_pac) // 匹配到了正向链
|
||
{
|
||
// 向前继续扩展
|
||
r += right_pos - left_pos;
|
||
EXTEND_BASE_LOOP(len, fmt->l_pac, 1, 1);
|
||
mt->rm[0].qe = k;
|
||
mt->rm[0].reverse = 0;
|
||
// 向后扩展，x位置之前的碱基
|
||
r -= k - left_pos + 1;
|
||
k = left_pos - 1;
|
||
EXTEND_BASE_LOOP(-1, -1, -1, -1);
|
||
mt->rm[0].qs = k + 1;
|
||
mt->rm[0].rs = r + 1;
|
||
}
|
||
else // 匹配到了互补链
|
||
{
|
||
r -= right_pos - left_pos;
|
||
EXTEND_BASE_LOOP_COMP(len, -1, 1, -1);
|
||
mt->rm[0].qe = k;
|
||
mt->rm[0].reverse = 1;
|
||
// 扩展x之前的碱基
|
||
r += k - left_pos + 1;
|
||
k = left_pos - 1;
|
||
EXTEND_BASE_LOOP_COMP(-1, fmt->l_pac, -1, 1);
|
||
mt->rm[0].qs = k + 1;
|
||
mt->rm[0].rs = r - 1;
|
||
}
|
||
mt->info = mt->rm[0].qs;
|
||
mt->info = mt->info << 32 | mt->rm[0].qe;
|
||
mt->x[2] = 1;
|
||
}
|
||
|
||
static inline void fmt_reverse_intvs(bwtintv_v *p)
|
||
{
|
||
if (p->n > 1) {
|
||
int j;
|
||
for (j = 0; j < p->n >> 1; ++j)
|
||
{
|
||
bwtintv_t tmp = p->a[p->n - 1 - j];
|
||
p->a[p->n - 1 - j] = p->a[j];
|
||
p->a[j] = tmp;
|
||
}
|
||
}
|
||
}
|
||
|
||
// 找smem（seed）
|
||
int fmt_smem(const FMTIndex *fmt, int len, const uint8_t *q, int x, int min_intv, int min_seed_len, bwtintv_v *mem, bwtintv_v *tmpvec)
|
||
{
|
||
int i, j, ret, kmer_len;
|
||
bwtintv_t ik = {0}, ok1 = {0}, ok2 = {0};
|
||
bwtintv_t mt = {0};
|
||
bwtintv_v *curr;
|
||
uint32_t qbit = 0;
|
||
mem->n = 0;
|
||
int only_forward = x == 0 || q[x - 1] > 3;
|
||
if (q[x] > 3) return x + 1;
|
||
|
||
if (min_intv < 1) min_intv = 1; // the interval size should be at least 1
|
||
curr = tmpvec; // use the temporary vector if provided
|
||
|
||
qbit = build_forward_kmer(&q[x], len - x, HASH_KMER_LEN, &kmer_len);
|
||
bwt_kmer_get(&fmt->kmer_hash, &ik, qbit, 0); // 初始碱基位置
|
||
ik.info = x + 1;
|
||
|
||
// check change of the interval size and whether the interval size is too small to be extended further
|
||
#define CHECK_INTV_CHANGE(iv, ov, end_pos) \
|
||
if (ov.x[2] != iv.x[2]) { kv_push(bwtintv_t, *curr, iv); if (ov.x[2] < min_intv) break; } iv = ov; iv.info = end_pos
|
||
#define PUSH_VAL_AND_SKIP(iv) \
|
||
do { kv_push(bwtintv_t, *curr, iv); goto backward_search; } while(0)
|
||
|
||
// 处理kmer对应的匹配信息
|
||
if (only_forward) j = kmer_len - 1;
|
||
else j = 1;
|
||
for (curr->n = 0; j < kmer_len; ++j)
|
||
{
|
||
bwt_kmer_get(&fmt->kmer_hash, &ok1, qbit, j);
|
||
CHECK_INTV_CHANGE(ik, ok1, x + j + 1);
|
||
}
|
||
if (kmer_len != HASH_KMER_LEN) // 遇到了N或者到了序列最后
|
||
PUSH_VAL_AND_SKIP(ik);
|
||
|
||
// 扩展kmer之后的碱基
|
||
// __builtin_prefetch(fmt_occ_intv(fmt, ik.x[1] - 1), 0, 2);
|
||
// __builtin_prefetch(fmt_occ_intv(fmt, ik.x[1] - 1 + ik.x[2]), 0, 2);
|
||
for (i = (int)ik.info; i + 1 < len; i += 2)
|
||
{ // forward search
|
||
if (q[i] < 4 && q[i + 1] < 4)
|
||
{
|
||
fmt_extend2(fmt, &ik, &ok1, &ok2, 0, 3 - q[i], 3 - q[i + 1]);
|
||
__builtin_prefetch(fmt_occ_intv(fmt, ok2.x[1] - 1), 0, 2);
|
||
__builtin_prefetch(fmt_occ_intv(fmt, ok2.x[1] - 1 + ok2.x[2]), 0, 2);
|
||
CHECK_INTV_CHANGE(ik, ok1, i + 1);
|
||
CHECK_INTV_CHANGE(ik, ok2, i + 2);
|
||
|
||
#if 1 // 间隔为1的时候直接与reference比对
|
||
if (min_intv == 1 && ok2.x[2] == min_intv) // 在这里进行判断是否只有一个候选了
|
||
{
|
||
direct_extend(fmt, len, q, x, i + 2, ok2.x[0], &mt);
|
||
kv_push(bwtintv_t, *mem, mt);
|
||
ret = (uint32_t)mt.info;
|
||
if (only_forward || mt.rm[0].qs == 0 || q[mt.rm[0].qs - 1] > 3) goto fmt_smem_end;
|
||
goto backward_search;
|
||
}
|
||
#endif
|
||
} else if (q[i] < 4) // q[i+1] >= 4
|
||
{
|
||
fmt_extend1(fmt, &ik, &ok1, 0, 3 - q[i]);
|
||
CHECK_INTV_CHANGE(ik, ok1, i + 1);
|
||
PUSH_VAL_AND_SKIP(ik);
|
||
}
|
||
else // q[i] >= 4
|
||
{
|
||
PUSH_VAL_AND_SKIP(ik);
|
||
}
|
||
}
|
||
for (; i == len - 1; ++i) // 扩展到了最后一个碱基
|
||
{
|
||
if (q[i] < 4)
|
||
{
|
||
fmt_extend1(fmt, &ik, &ok1, 0, 3 - q[i]);
|
||
CHECK_INTV_CHANGE(ik, ok1, i + 1);
|
||
}
|
||
else
|
||
PUSH_VAL_AND_SKIP(ik);
|
||
}
|
||
if (i == len)
|
||
kv_push(bwtintv_t, *curr, ik); // push the last interval if we reach the end
|
||
|
||
backward_search:
|
||
fmt_reverse_intvs(curr); // s.t. smaller intervals (i.e. longer matches) visited first
|
||
if (mt.num_match == 0) ret = curr->a[0].info; // this will be the returned value，扩展到的最远的位置
|
||
|
||
// 按照种子进行遍历，反向扩展
|
||
#define CHECK_ADD_MEM(pos, intv, mem) \
|
||
if (((int)((intv).info) - (pos) >= min_seed_len) && (mem->n == 0 || (pos) < mem->a[mem->n - 1].info >> 32)) { \
|
||
(intv).info |= (uint64_t)(pos) << 32; kv_push(bwtintv_t, *mem, (intv)); \
|
||
}
|
||
|
||
#define CHECK_INTV_ADD_MEM(ok, pos, intv, mem) \
|
||
if (ok.x[2] < min_intv) { CHECK_ADD_MEM(pos, intv, mem); break; }
|
||
|
||
int last_kmer_start = 0;
|
||
for (j = 0; j < curr->n; ++j)
|
||
{
|
||
bwtintv_t *p = &curr->a[j]; // 前向扩展的种子
|
||
// __builtin_prefetch(fmt_occ_intv(fmt, p->x[0] - 1), 0, 2);
|
||
// __builtin_prefetch(fmt_occ_intv(fmt, p->x[0] - 1 + p->x[2]), 0, 2);
|
||
#if 1
|
||
if (!only_forward && p->info - x < HASH_KMER_LEN) {
|
||
if (last_kmer_start && kmer_len == HASH_KMER_LEN && p->info == last_kmer_start && p->info - kmer_len > 0 && q[p->info - kmer_len] < 4)
|
||
qbit = ((qbit << 2) | (3 - q[p->info - kmer_len])) & ((1L << (kmer_len << 1)) - 1); // 创建反向kmer
|
||
else qbit = build_backward_kmer(q, p->info - 1, HASH_KMER_LEN, &kmer_len); // 创建反向kmer
|
||
last_kmer_start = p->info - 1;
|
||
i = 1;
|
||
do { bwt_kmer_get(&fmt->kmer_hash, &ik, qbit, kmer_len - i++); } while (ik.x[2] < min_intv);
|
||
if (i > 2) continue;
|
||
p->x[0] = ik.x[1]; p->x[1] = ik.x[0]; p->x[2] = ik.x[2];
|
||
i = p->info - (kmer_len - i + 3);
|
||
} else {
|
||
i = x - 1;
|
||
}
|
||
#else
|
||
i = x - 1;
|
||
#endif
|
||
for (; i > 0; i -= 2)
|
||
{
|
||
if (q[i] < 4 && q[i - 1] < 4) // 两个都可以扩展
|
||
{
|
||
fmt_extend2(fmt, p, &ok1, &ok2, 1, q[i], q[i - 1]);
|
||
__builtin_prefetch(fmt_occ_intv(fmt, ok2.x[0] - 1), 0, 2);
|
||
__builtin_prefetch(fmt_occ_intv(fmt, ok2.x[0] - 1 + ok2.x[2]), 0, 2);
|
||
CHECK_INTV_ADD_MEM(ok1, i + 1, *p, mem);
|
||
ok1.info = p->info;
|
||
CHECK_INTV_ADD_MEM(ok2, i, ok1, mem);
|
||
ok2.info = p->info; *p = ok2;
|
||
}
|
||
else if (q[i] < 4) // 只能扩展一个
|
||
{
|
||
fmt_extend1(fmt, p, &ok1, 1, q[i]);
|
||
CHECK_INTV_ADD_MEM(ok1, i + 1, *p, mem);
|
||
ok1.info = p->info;
|
||
CHECK_ADD_MEM(i, ok1, mem);
|
||
goto fmt_smem_end;
|
||
}
|
||
else
|
||
{ // 不能扩展
|
||
CHECK_ADD_MEM(i + 1, *p, mem);
|
||
goto fmt_smem_end;
|
||
}
|
||
}
|
||
for (; i == 0; --i)
|
||
{ // 扩展到了第一个碱基
|
||
if (q[i] < 4) {
|
||
fmt_extend1(fmt, p, &ok1, 1, q[i]);
|
||
CHECK_INTV_ADD_MEM(ok1, i + 1, *p, mem);
|
||
ok1.info = p->info; *p = ok1;
|
||
} else {
|
||
CHECK_ADD_MEM(i + 1, *p, mem);
|
||
goto fmt_smem_end;
|
||
}
|
||
}
|
||
if (i == -1) {
|
||
CHECK_ADD_MEM(i + 1, *p, mem);
|
||
goto fmt_smem_end;
|
||
}
|
||
}
|
||
|
||
fmt_smem_end:
|
||
fmt_reverse_intvs(mem); // s.t. sorted by the start coordinate
|
||
return ret;
|
||
}
|
||
|
||
int fmt_seed_strategy1(const FMTIndex *fmt, int len, const uint8_t *q, int x, int min_len, int max_intv, bwtintv_t *mem)
|
||
{
|
||
int i, kmer_len;
|
||
bwtintv_t ik = {0}, ok1={0}, ok2={0};
|
||
uint64_t qbit;
|
||
memset(mem, 0, sizeof(bwtintv_t));
|
||
if (q[x] > 3) return x + 1;
|
||
|
||
qbit = build_forward_kmer(&q[x], len - x, HASH_KMER_LEN, &kmer_len);
|
||
bwt_kmer_get(&fmt->kmer_hash, &ik, qbit, kmer_len-1); // 初始碱基位置
|
||
ik.info = x + kmer_len;
|
||
|
||
//fmt_set_intv(fmt, q[x], ik);
|
||
//ik.info = x + 1;
|
||
|
||
// __builtin_prefetch(fmt_occ_intv(fmt, ik.x[1] - 1), 0, 2);
|
||
// __builtin_prefetch(fmt_occ_intv(fmt, ik.x[1] - 1 + ik.x[2]), 0, 2);
|
||
|
||
#define COND_SET_RETURN(iv, ov, start_pos, end_pos, max_intv, min_len) \
|
||
if (iv.x[2] < max_intv && end_pos - start_pos >= min_len) \
|
||
{ \
|
||
(ov) = (iv); \
|
||
(ov).info = (uint64_t)start_pos << 32 | (end_pos + 1); \
|
||
return end_pos + 1; \
|
||
}
|
||
|
||
for (i = (int)ik.info; i + 1 < len; i += 2)
|
||
{ // forward search
|
||
if (q[i] < 4 && q[i + 1] < 4)
|
||
{
|
||
fmt_extend2(fmt, &ik, &ok1, &ok2, 0, 3 - q[i], 3 - q[i + 1]);
|
||
__builtin_prefetch(fmt_occ_intv(fmt, ok2.x[1] - 1), 0, 2);
|
||
__builtin_prefetch(fmt_occ_intv(fmt, ok2.x[1] - 1 + ok2.x[2]), 0, 2);
|
||
COND_SET_RETURN(ok1, *mem, x, i, max_intv, min_len);
|
||
COND_SET_RETURN(ok2, *mem, x, i + 1, max_intv, min_len);
|
||
ik = ok2;
|
||
}
|
||
else if (q[i] < 4) // q[i+1] >= 4
|
||
{
|
||
fmt_extend1(fmt, &ik, &ok1, 0, 3 - q[i]);
|
||
COND_SET_RETURN(ok1, *mem, x, i, max_intv, min_len);
|
||
return i + 2;
|
||
}
|
||
else // q[i] >= 4
|
||
{
|
||
return i + 1;
|
||
}
|
||
}
|
||
if (i == len - 1) {
|
||
fmt_extend1(fmt, &ik, &ok1, 0, 3 - q[i]);
|
||
COND_SET_RETURN(ok1, *mem, x, i, max_intv, min_len);
|
||
}
|
||
return len;
|
||
}
|
||
|
||
// 这里的k是bwt str的行
|
||
inline static void fmt_get_previous_base(const FMTIndex *fmt, bwtint_t k, uint8_t *b1, uint8_t *b2)
|
||
{
|
||
uint32_t *p;
|
||
uint8_t base2;
|
||
// 第一步，找到check point位置
|
||
p = fmt_occ_intv(fmt, k); // check point起始位置
|
||
p += 20; // bwt碱基起始位置
|
||
// 第二步，找到mid check point位置
|
||
int mk = k & FMT_OCC_INTV_MASK;
|
||
int n_mintv = mk >> FMT_MID_INTV_SHIFT;
|
||
p += n_mintv * (4 + (FMT_MID_INTERVAL >> 3)); // 跳过mid间隔的bwt碱基位置
|
||
// 第三步，找到具体的uint32_t
|
||
p += (k & FMT_MID_INTV_MASK) >> 3; // 每个uint32_t包含8个碱基（和8个倒数第二bwt碱基）
|
||
// 第四步，获取碱基
|
||
base2 = *p >> ((~(k) & 0x7) << 2) & 0xf;
|
||
*b2 = base2 >> 2 & 3;
|
||
*b1 = base2 & 3;
|
||
}
|
||
|
||
// k, k1, k2都是bwt矩阵对应的行
|
||
inline static void fmt_previous_line(const FMTIndex *fmt, bwtint_t k, bwtint_t *k1, bwtint_t *k2)
|
||
{
|
||
uint8_t b1, b2;
|
||
bwtint_t tk[4], kk;
|
||
kk = k - (k >= fmt->primary); // k由bwt矩阵对应的行转换成bwt字符串对应的行（去掉了$，所以大于$的行，都减掉1）
|
||
fmt_get_previous_base(fmt, kk, &b1, &b2);
|
||
fmt_e2_occ(fmt, k, b1, b2, tk);
|
||
*k1 = fmt->L2[b1] + tk[1];
|
||
*k2 = fmt->L2[b2] + tk[3];
|
||
}
|
||
|
||
bwtint_t fmt_sa(const FMTIndex *fmt, bwtint_t k)
|
||
{
|
||
bwtint_t sa = 0, mask = fmt->sa_intv - 1;
|
||
bwtint_t k1, k2;
|
||
while (k & mask)
|
||
{
|
||
++sa;
|
||
fmt_previous_line(fmt, k, &k1, &k2);
|
||
if (!(k1 & mask)) {
|
||
k = k1;
|
||
break;
|
||
}
|
||
++sa;
|
||
k = k2;
|
||
}
|
||
sa += bwt_get_sa(fmt->sa, k / fmt->sa_intv);
|
||
return sa;
|
||
} |