sw_perf/extend.c

/*********************************************************************************************
    Description: sw extend functions in bwa-mem

    Copyright : All right reserved by NCIC.ICT

    Author : Zhang Zhonghai
    Date : 2024/04/08
***********************************************************************************************/
#include <stdio.h>
#include <stdlib.h>

#include "byte_alloc.h"
#include "utils.h"
#include "profiling.h"
#include "extend.h"
#include "debug.h"


#define EXTEND_PERFORMANCE_TEST(kernel_id, func, sp, ep)     \
    do                                                       \
    {                                                        \
        PROF_START(extend);                                  \
        int i, score;                                        \
        for (i = sp; i < ep; ++i)                            \
        {                                                    \
            score = func(                                    \
                &bmem,                                       \
                kv_A(kv_A(i_arr, i), 0), kv_A(q_arr, i).a,   \
                kv_A(kv_A(i_arr, i), 1), kv_A(t_arr, i).a,   \
                5, mat, 6, 1, 6, 1, 100, 5, 100,             \
                /*kv_A(kv_A(i_arr, i), 2)*/10,                     \
                &qle, &tle, &gtle, &gscore, &max_off[0]);    \
            score_total[kernel_id] += score;                 \
        }                                                    \
        PROF_END(gprof[kernel_prof_idx[kernel_id]], extend); \
    } while (0)

/******
 * 输入说明：三个文件，query.fa, target.fa, info.txt
 * query.fa:  每一行代表一个query序列，由ACGTN组成
 * target.fa: 每一行代表一个reference（target）序列，由ACGTN组成
 * info.txt:  每一行由三个数字组成，分别代表query序列长度，target序列长度，以及初始分数h0
 */
int main_extend(int argc, char *argv[])
{
    if (argc < 3) {
        fprintf(stderr, "Need 3 files: query, target, info.\n");
        return -1;
    }
    const char *qf_path = argv[0];
    const char *tf_path = argv[1];
    const char *if_path = argv[2];

    FILE *qfp = fopen(qf_path, "r");
    FILE *tfp = fopen(tf_path, "r");
    FILE *ifp = fopen(if_path, "r");

    buf_t read_buf = {0};
    seq_v q_arr = {0};
    seq_v t_arr = {0};
    qti_v i_arr = {0};
    uint64_t score_total[EXTEND_FUNC_NUM] = {0};
    const int kmax_row = 3000000;

    int query_read_row = read_seq(&q_arr, &read_buf, kmax_row, qfp);
    int target_read_row = read_seq(&t_arr, &read_buf, kmax_row, tfp);
    int info_read_row = read_qt_info(&i_arr, &read_buf, kmax_row, 3, ifp);
    // fprintf(stderr, "read row: %d\t%d\t%d\n", query_read_row, target_read_row, info_read_row);

    int8_t mat[25] = {1, -4, -4, -4, -1,
                      -4, 1, -4, -4, -1,
                      -4, -4, 1, -4, -1,
                      -4, -4, -4, 1, -1,
                      -1, -1, -1, -1, -1};
    int kernel_prof_idx[] = {G_EXT_SCALAR, G_EXT_AVX2_I16, G_EXT_AVX2_U8, G_EXT_AVX2_I16_SP};
    byte_mem_t bmem = {0};
    byte_mem_init_alloc(&bmem, 1024 * 1024);
    int max_off[2], qle, tle, gtle, gscore;
    int excute_lines = MIN(MIN(query_read_row, target_read_row), info_read_row);
    //open_qti_files();
    //open_debug_files();
    fprintf(stderr, "excute nums: %d\n", excute_lines);
    EXTEND_PERFORMANCE_TEST(0, extend_scalar, 0, excute_lines);
    EXTEND_PERFORMANCE_TEST(1, extend_avx2_i16, 0, excute_lines);
    EXTEND_PERFORMANCE_TEST(2, extend_avx2_u8, 0, excute_lines);
    EXTEND_PERFORMANCE_TEST(3, extend_avx2_i16_sp, 0, excute_lines);
    int i = 0; for(; i<ARRAY_SIZE(kernel_prof_idx); ++i) { gdata[kernel_prof_idx[i]] = score_total[i]; }

#ifdef SHOW_PERF
    fprintf(stderr, "[extend scalar ] time: %9.6lf s; score: %ld\n", gprof[G_EXT_SCALAR] / TIME_DIVIDE_BY, gdata[G_EXT_SCALAR]);
    fprintf(stderr, "[extend avx i16] time: %9.6lf s; score: %ld\n", gprof[G_EXT_AVX2_I16] / TIME_DIVIDE_BY, gdata[G_EXT_AVX2_I16]);
    fprintf(stderr, "[extend avx u8 ] time: %9.6lf s; score: %ld\n", gprof[G_EXT_AVX2_U8] / TIME_DIVIDE_BY, gdata[G_EXT_AVX2_U8]);
    fprintf(stderr, "[extend avx sp ] time: %9.6lf s; score: %ld\n", gprof[G_EXT_AVX2_I16_SP] / TIME_DIVIDE_BY, gdata[G_EXT_AVX2_I16_SP]);
#endif
    // close_files();
    fclose(qfp);
    fclose(tfp);
    fclose(ifp);
    return 0;
}