bwa_perf/fmt_index.cpp

#include <iostream>
#include <stdint.h>
#include <stdlib.h>
#include <vector>
#include <sys/time.h>
#include <string>
#include <stdio.h>
#include <algorithm>
#include <string.h>
#include <immintrin.h>
#include <sstream>

#include "util.h"
#include "bwt.h"
#include "fmt_index.h"
using namespace std;

double t1 = 0, t2 = 0, t3 = 0, t4 = 0, t5 = 0, t6 = 0, t7 = 0, t8 = 0, t9 = 0, t10 = 0,
	   t11 = 0, t12 = 0, t13 = 0, t14 = 0;

long f1 = 0, f2 = 0, f3 = 0, f4 = 0, f5 = 0;

const static char BASE[4] = {'A', 'C', 'G', 'T'};

// 求反向互补序列
string calc_reverse_seq(string &seq)
{
	string rseq(seq.size(), '0');
	for (size_t i = 0; i < seq.size(); ++i)
	{
		if (seq[i] == 'A')
			rseq[i] = 'T';
		else if (seq[i] == 'C')
			rseq[i] = 'G';
		else if (seq[i] == 'G')
			rseq[i] = 'C';
		else if (seq[i] == 'T')
			rseq[i] = 'A';
	}
	std::reverse(rseq.begin(), rseq.end());
	return rseq;
}

// 打印32位整型数据中包含的pre-bwt：bwt
void print_base_uint32(uint32_t p)
{
	for (int i = 30; i > 0; i -= 4)
	{
		int b1 = p >> i & 3;
		int b2 = p >> (i - 2) & 3;
		cout << BASE[b1] << BASE[b2] << endl;
	}
}

// 随机生成长度为len的序列
string generate_rand_seq(int len)
{
	string seq(len, 'A');
	for (int i = 0; i < len; ++i)
	{
		seq[i] = BASE[rand() % 4];
	}
	return seq;
}
// 创建bwt矩阵
void create_bwt_mtx(string &seq)
{
	bwtint_t seq_len = seq.size() + 1;
	string sarr[seq_len];
	sarr[0] = seq + '$';
	for (bwtint_t i = 1; i < seq_len; ++i)
	{
		sarr[i] = sarr[0].substr(i) + sarr[0].substr(0, i);
	}
	std::sort(sarr, sarr + seq_len);

	// print bwt matrix
	// for (int i = 0; i < seq_len; ++i)
	//{
	//	// cout << i << ' ' << sarr[i] << endl;
	//	cout << sarr[i] << endl;
	//}
	// cout << "bwt string" << endl;
	// for (int i = 0; i < seq_len; ++i)
	// {
	// 	cout << sarr[i].back();
	// }
	// cout << endl;
	// cout << "pre bwt string" << endl;
	// for (int i = 0; i < seq_len; ++i)
	// {
	// 	cout << sarr[i][seq_len - 2];
	// }
	// cout << endl;
}

// 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;
		}
	}
}

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);
}

FMTIndex *restore_fmt(const char *fn)
{
	FMTIndex *fmt;
	fmt = (FMTIndex *)calloc(1, sizeof(FMTIndex));
	FILE *fp = fopen(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);
	fread(&fmt->primary, sizeof(bwtint_t), 1, fp);
	fread(&fmt->sec_primary, sizeof(bwtint_t), 1, fp);
	fread(&fmt->sec_bcp, sizeof(uint8_t), 1, fp);
	fread(&fmt->first_base, sizeof(uint8_t), 1, fp);
	fread(&fmt->last_base, sizeof(uint8_t), 1, fp);
	fread(fmt->L2 + 1, sizeof(bwtint_t), 4, fp);
	fread_fix(fp, fmt->bwt_size << 2, fmt->bwt);
	fmt->seq_len = fmt->L2[4];
	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;
}

// 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;
	cout << (FMT_OCC_INTERVAL / FMT_MID_INTERVAL - 1) << ' ' << last_seq_len << ' ' << (last_seq_len + FMT_MID_INTERVAL - 1) / FMT_MID_INTERVAL - 1 << endl;
	cout << "mn_occ: " << mn_occ << ' ' << mn_occ * 4 << endl;
	i = 0;
	//cout << ((i + 16 & FMT_MID_INTV_MASK) == 0) << endl;
	// exit(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

					// 输出调试信息
					// cout << "mtx line: " << cur_mtx_line << ' ' << c[0] << ' ' << c[1] << ' ' << c[2] << ' ' << c[3] << ' ';
					// for (m = 0; m < 16; ++m)
					// 	cout << c2[m] << ' ';
					// cout << endl;
				}
				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;
	// cout << "n occ: " << n_occ << endl;
	cout << "size: " << k << '\t' << fmt->bwt_size << endl;
	// exit(0);
	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
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);
	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);
	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;

#ifdef FMT_MID_INTERVAL // 加入了middle checkpoint
	// 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;
	}
#if FMT_MID_INTERVAL == 16 // middle check point interval等于16时候，只需要判断一下是不是要计算两个uint32表示的碱基序列
	if ((mk & FMT_MID_INTV_MASK) >> 3)
	{
		x += __fmt_occ_e2_aux2(fmt, ti, *p);
		++p;
	}
#elif FMT_MID_INTERVAL > 16 // 该地址是bwt和pre_bwt字符串数据的首地址
	uint32_t *end = p + ((k >> 3) - ((k & ~FMT_MID_INTV_MASK) >> 3));
	for (; p < end; ++p)
	{
		x += __fmt_occ_e2_aux2(fmt, ti, *p);
	}
#endif
#else // 没有加入middle check point interval
#if FMT_OCC_INTERVAL > 16
	uint32_t *end = p + ((k >> 3) - ((k & ~FMT_OCC_INTV_MASK) >> 3));
	// p = end - (end - p) / 4;
	for (; p < end; ++p)
	{
		x += __fmt_occ_e2_aux2(fmt, ti, *p);
	}
#else // FMT_OCC_INTERVAL等于16的时候，只需要判断一次就可以
	if ((k & FMT_OCC_INTV_MASK) >> 3)
	{
		x += __fmt_occ_e2_aux2(fmt, ti, *p);
		++p;
	}
#endif
#endif
	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
void fmt_extend2(const FMTIndex *fmt, bwtintv_t *ik, bwtintv_t *ok, int is_back, int b1, int b2)
{
	bwtint_t tk[4], tl[4], first_pos;
	// 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[b2] + 1 + tk[3];
	ok->x[2] = tl[3] - tk[3];
	// 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];
	// 第二次正向扩展
	first_pos = fmt->L2[b1] + 1 + tk[1];
	ok->x[is_back] = ok->x[is_back] + (first_pos <= fmt->primary && first_pos + tl[1] - tk[1] - 1 >= fmt->primary) + tl[2] - tk[2];
}

// 扩展一个碱基
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];
}

// 利用fmt搜索seed，完整搜索，只需要单向搜索
bwtintv_t fmt_search(FMTIndex *fmt, const string &q)
{
	bwtintv_t ik;
	bwtintv_t ok;
	int i, c1, c2, x = 0;
	int qlen = (int)q.size();

	fmt_set_intv(fmt, bval(q[x]), ik);
	ik.info = x + 1;

	// 每次扩展两个碱基
	for (i = x + 1; i + 1 < qlen; i += 2)
	{
		if (bval(q[i]) < 4 && bval(q[i + 1]) < 4)
		{
			c1 = cbval(q[i]);
			c2 = cbval(q[i + 1]);
			fmt_extend2(fmt, &ik, &ok, 0, c1, c2);
			ik = ok;
			ik.info = i + 1;
		}
		else
		{
			break;
		}
	}
	if (i < qlen && bval(q[i]) < 4)
	{ // 最后一次扩展
		c1 = cbval(q[i]);
		fmt_extend1(fmt, &ik, &ok, 0, c1);
		ik = ok;
		ik.info = i + 1;
	}
	return ik;
}

uint32_t str2bit(string &str)
{
	uint32_t pac = 0;
	for (int i = 0; i < 16; ++i)
		pac = (pac << 2) | bval(str[i]);
	return pac;
}

#define GEN_BWT_IDX 0
#define GEN_FMT_IDX 0
#define CMP_FMT_BWT_TIME 1
#define CMP_FMT_BWT_RESULT 1

// argv[1] 应该是索引的前缀
int main_fmtidx(int argc, char **argv)
{
	string prefix = argv[1];
	string bwt_str = prefix + ".bwt.str";
	string bwt_idx, fmt_idx;
	bwt_t *bwt;
	FMTIndex *fmt;
	ostringstream oss_bwt, oss_fmt;
	oss_bwt << '.' << OCC_INTERVAL;
	oss_fmt << '.' << FMT_OCC_INTERVAL;
#ifdef FMT_MID_INTERVAL
	oss_fmt << '.' << FMT_MID_INTERVAL;
#endif

	bwt_idx = prefix + oss_bwt.str() + ".bwt";
	fmt_idx = prefix + oss_fmt.str() + ".fmt";

	cout << bwt_idx << endl;
	cout << fmt_idx << endl;

// 生成或读取bwt索引文件
	double time_read_bwt = realtime();
#if GEN_BWT_IDX
	bwt = restore_bwt(bwt_str.c_str());
	create_interval_occ_bwt(bwt);
	dump_bwt(bwt_idx.c_str(), bwt);
#else
	bwt = restore_bwt(bwt_idx.c_str());
#endif
	time_read_bwt = realtime() - time_read_bwt;
	cout << "[time gen/read bwt:] " << time_read_bwt << "s" << endl;

// 生成或读取fmt索引文件
	double time_read_fmt = realtime();
#if GEN_FMT_IDX
	fmt = create_fmt_from_bwt(bwt);
	dump_fmt(fmt_idx.c_str(), fmt);
#else
	fmt = restore_fmt(fmt_idx.c_str());
#endif
	time_read_fmt = realtime() - time_read_fmt;
	cout << "[time gen/read fmt:] " << time_read_fmt << "s" << endl;

// 生成随机序列进行测试
	int seq_size = 10000000;
	// int seq_size = 1;
	int seq_len = 11;
	vector<string> seq_arr(seq_size);
	// seq_arr[0] = "GCGATACTAAGA";
	// seq_arr[0] = "GAGAGCTGTTCCCGTGTTTTCCATGGTTT";
#if 1
	srand(time(NULL));
	double time_gen_seq = realtime();
	for (int i = 0; i < (int)seq_arr.size(); ++i)
		seq_arr[i] = generate_rand_seq(seq_len);
	time_gen_seq = realtime() - time_gen_seq;
	cout << "[time gen seq:] " << time_gen_seq << "s" << endl;
#endif

// 对比bwt和fmt搜索的时间
#if CMP_FMT_BWT_TIME
	double time_bwt_search = realtime();
	for (int i = 0; i < (int)seq_arr.size(); ++i)
		bwt_search(bwt, seq_arr[i]);
	time_bwt_search = realtime() - time_bwt_search;
	cout << "[time bwt search:] " << time_bwt_search << "s" << endl;

	double time_fmt_search = realtime();
	for (int i = 0; i < (int)seq_arr.size(); ++i)
		fmt_search(fmt, seq_arr[i]);
	time_fmt_search = realtime() - time_fmt_search;
	cout << "[time fmt search:] " << time_fmt_search << "s" << endl;
#endif

// 对比bwt和fmt搜索的结果
#if CMP_FMT_BWT_RESULT
	double time_cmp = realtime();
	for (int i = 0; i < (int)seq_arr.size() / 100; ++i)
	{
		bwtintv_t p1 = bwt_search2(bwt, seq_arr[i]);
		bwtintv_t p2 = fmt_search(fmt, seq_arr[i]);
		if (p1.x[0] != p2.x[0] || p1.x[1] != p2.x[1] || p1.x[2] != p2.x[2])
			cout << seq_arr[i] << endl
				 << p1.x[0] << ' ' << p1.x[1] << ' ' << p1.x[2] << endl
				 << p2.x[0] << ' ' << p2.x[1] << ' ' << p2.x[2] << endl;
	}
	time_cmp = realtime() - time_cmp;
	cout << "[time compare bwt & fmt:] " << time_cmp << "s" << endl;
#endif

	return 0;
}