sw_perf/align_avx2_u8.c

#include <stdint.h>
#include <stdlib.h>

#include "align.h"
#include "utils.h"

kswr_t align_avx2_u8(byte_mem_t *bmem, kswq_avx2_t *q, int tlen, const uint8_t *target, int _o_del, int _e_del, int _o_ins,
                     int _e_ins, int xtra)  // the first gap costs -(_o+_e)
{
    int slen, i, m_b, n_b, te = -1, gmax = 0, minsc, endsc;
    uint64_t *b;
    __m256i zero, oe_del, e_del, oe_ins, e_ins, shift, *H0, *H1, *E, *Hmax;
    kswr_t r;

#define __max_32(ret, xx)                                         \
    do {                                                          \
        uint8_t *maxVal = (uint8_t *)&(xx);                       \
        (xx) = _mm256_max_epu8((xx), _mm256_srli_si256((xx), 8)); \
        (xx) = _mm256_max_epu8((xx), _mm256_srli_si256((xx), 4)); \
        (xx) = _mm256_max_epu8((xx), _mm256_srli_si256((xx), 2)); \
        (xx) = _mm256_max_epu8((xx), _mm256_srli_si256((xx), 1)); \
        (ret) = MAX(maxVal[0], maxVal[16]);                       \
    } while (0)

// Given entries with arbitrary values, return whether they are all 0x00
#define allzero_32(xx) (_mm256_movemask_epi8(_mm256_cmpeq_epi8((xx), zero)) == 0xffffffff)

#define avx2_slli_wrap(xx, imm)                \
    do {                                       \
        uint8_t *arr = (uint8_t *)&(xx);       \
        uint8_t val = arr[15];                 \
        (xx) = _mm256_slli_si256((xx), (imm)); \
        arr[16] = val;                         \
    } while (0)

    // initialization
    r = g_defr;
    minsc = (xtra & KSW_XSUBO) ? xtra & 0xffff : 0x10000;
    endsc = (xtra & KSW_XSTOP) ? xtra & 0xffff : 0x10000;
    m_b = n_b = 0;
    b = 0;
    zero = _mm256_set1_epi32(0);
    oe_del = _mm256_set1_epi8(_o_del + _e_del);
    e_del = _mm256_set1_epi8(_e_del);
    oe_ins = _mm256_set1_epi8(_o_ins + _e_ins);
    e_ins = _mm256_set1_epi8(_e_ins);
    shift = _mm256_set1_epi8(q->shift);
    H0 = q->H0;
    H1 = q->H1;
    E = q->E;
    Hmax = q->Hmax;
    slen = q->slen;
    for (i = 0; i < slen; ++i) {
        _mm256_store_si256(E + i, zero);
        _mm256_store_si256(H0 + i, zero);
        _mm256_store_si256(Hmax + i, zero);
    }
    // the core loop
    for (i = 0; i < tlen; ++i) {
        int j, k, imax;
        __m256i e, h, t, f = zero, max = zero, *S = q->qp + target[i] * slen;  // s is the 1st score vector
        h = _mm256_load_si256(H0 + slen - 1);                                  // h={2,5,8,11,14,17,-1,-1} in the above example
        avx2_slli_wrap(h, 1);  // h=H(i-1,-1); << instead of >> because x64 is little-endian
        for (j = 0; LIKELY(j < slen); ++j) {
            /* SW cells are computed in the following order:
             *   H(i,j)   = max{H(i-1,j-1)+S(i,j), E(i,j), F(i,j)}
             *   E(i+1,j) = max{H(i,j)-q, E(i,j)-r}
             *   F(i,j+1) = max{H(i,j)-q, F(i,j)-r}
             */
            // compute H'(i,j); note that at the beginning, h=H'(i-1,j-1)
            h = _mm256_adds_epu8(h, _mm256_load_si256(S + j));
            h = _mm256_subs_epu8(h, shift);  // h=H'(i-1,j-1)+S(i,j)
            e = _mm256_load_si256(E + j);    // e=E'(i,j)
            h = _mm256_max_epu8(h, e);
            h = _mm256_max_epu8(h, f);      // h=H'(i,j)
            max = _mm256_max_epu8(max, h);  // set max
            _mm256_store_si256(H1 + j, h);  // save to H'(i,j)
            // now compute E'(i+1,j)
            e = _mm256_subs_epu8(e, e_del);   // e=E'(i,j) - e_del
            t = _mm256_subs_epu8(h, oe_del);  // h=H'(i,j) - o_del - e_del
            e = _mm256_max_epu8(e, t);        // e=E'(i+1,j)
            _mm256_store_si256(E + j, e);     // save to E'(i+1,j)
            // now compute F'(i,j+1)
            f = _mm256_subs_epu8(f, e_ins);
            t = _mm256_subs_epu8(h, oe_ins);  // h=H'(i,j) - o_ins - e_ins
            f = _mm256_max_epu8(f, t);
            // get H'(i-1,j) and prepare for the next j
            h = _mm256_load_si256(H0 + j);  // h=H'(i-1,j)
        }
        // NB: we do not need to set E(i,j) as we disallow adjecent insertion and then deletion
        for (k = 0; LIKELY(k < 32); ++k) {  // this block mimics SWPS3; NB: H(i,j) updated in the lazy-F loop cannot exceed max
            avx2_slli_wrap(f, 1);
            for (j = 0; LIKELY(j < slen); ++j) {
                h = _mm256_load_si256(H1 + j);
                h = _mm256_max_epu8(h, f);  // h=H'(i,j)
                _mm256_store_si256(H1 + j, h);
                h = _mm256_subs_epu8(h, oe_ins);
                f = _mm256_subs_epu8(f, e_ins);
                if (UNLIKELY(allzero_32(_mm256_subs_epu8(f, h))))
                    goto end_loop32;
            }
        }
    end_loop32:
        __max_32(imax, max);  // imax is the maximum number in max
        if (imax >= minsc) {  // write the b array; this condition adds branching unfornately
            if (n_b == 0 ||
                (int32_t)b[n_b - 1] + 1 !=
                    i) {  // then append 新的max用新的数值记录，用来记录次优分值，不能与最优分值相邻（否则就是连续匹配的）
                if (n_b == m_b) {
                    m_b = m_b ? m_b << 1 : 8;
                    b = (uint64_t *)realloc(b, 8 * m_b);
                }
                b[n_b++] = (uint64_t)imax << 32 | i;
            } else if ((int)(b[n_b - 1] >> 32) < imax)
                b[n_b - 1] = (uint64_t)imax << 32 | i;  // modify the last
        }
        if (imax > gmax) {
            gmax = imax;
            te = i;                             // te is the end position on the target
            for (j = 0; LIKELY(j < slen); ++j)  // keep the H1 vector
                _mm256_store_si256(Hmax + j, _mm256_load_si256(H1 + j));
            if (gmax + q->shift >= 255 || gmax >= endsc)
                break;  // 最大分值溢出了，结束运算
        }
        S = H1;
        H1 = H0;
        H0 = S;  // swap H0 and H1
    }
    // fprintf(gf[0], "%d\n", i);
    r.score = gmax + q->shift < 255 ? gmax : 255;
    r.te = te;
    if (r.score != 255) {  // get a->qe, the end of query match; find the 2nd best score
        int max = -1, tmp, low, high, qlen = slen * 32;
        uint8_t *t = (uint8_t *)Hmax;
        for (i = 0; i < qlen; ++i, ++t)
            if ((int)*t > max)
                max = *t, r.qe = i / 32 + i % 32 * slen;
            else if ((int)*t == max && (tmp = i / 32 + i % 32 * slen) < r.qe)
                r.qe = tmp;
        if (b) {
            i = (r.score + q->max - 1) / q->max;
            low = te - i;
            high = te + i;
            for (i = 0; i < n_b; ++i) {
                int e = (int32_t)b[i];
                if ((e < low || e > high) && (int)(b[i] >> 32) > r.score2)
                    r.score2 = b[i] >> 32, r.te2 = e;
            }
        }
    }
    free(b);
    return r;
}