测试不同数据

测试不同内存开辟策略的影响
2023-08-11 00:36:14 +08:00 · 2023-08-11 00:35:43 +08:00
5 changed files with 267 additions and 5 deletions
--- a/.gitignore
+++ b/.gitignore
@ -1,5 +1,6 @@
 *.[oa]
 sw_perf
+sw_perf_discrete
 test
 test64
 .*.swp
--- a/.vscode/launch.json
+++ b/.vscode/launch.json
@ -14,6 +14,17 @@
                "all"
            ],
            "cwd": "${workspaceFolder}", // 当前工作路径：当前文件所在的工作空间
+        },
+        {
+            "name": "sw-perf-discrete",
+            "preLaunchTask": "Build",
+            "type": "cppdbg",
+            "request": "launch",
+            "program": "${workspaceRoot}/sw_perf_discrete",
+            "args": [
+                "all"
+            ],
+            "cwd": "${workspaceFolder}", // 当前工作路径：当前文件所在的工作空间
        }
    ]
 }
--- a/8
+++ b/8
@ -4,6 +4,7 @@ CFLAGS=		-Wall -Wno-unused-function -O2 -mavx2
 DFLAGS=		-DSHOW_PERF
 OBJS=		ksw_normal.o ksw_avx2.o ksw_cuda.o ksw_avx2_u8.o
 PROG=		sw_perf
+PROG2= 		sw_perf_discrete
 INCLUDES=	
 LIBS=
 SUBDIRS=	.
@ -17,13 +18,16 @@ endif
 .c.o:
 		$(CC) -c $(CFLAGS) $(DFLAGS) $(INCLUDES) $(CPPFLAGS) $< -o $@

-all:$(PROG)
+all:$(PROG) $(PROG2)

 sw_perf:$(OBJS) main.o
 		$(CC) $(CFLAGS) $(LDFLAGS) $(OBJS) main.o -o $@ -L. $(LIBS)

+$(PROG2):$(OBJS) discrete_mem_main.o
+		$(CC) $(CFLAGS) $(LDFLAGS) $(OBJS) main.o -o $@ -L. $(LIBS)
+
 clean:
-		rm -f *.o a.out $(PROG) *~ *.a
+		rm -f *.o a.out $(PROG) $(PROG2) *~ *.a

 depend:
 	( LC_ALL=C ; export LC_ALL; makedepend -Y -- $(CFLAGS) $(DFLAGS) $(CPPFLAGS) -- *.c )
--- a/discrete_mem_main.c
+++ b/discrete_mem_main.c
@ -0,0 +1,243 @@
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <stdint.h>
+#include <assert.h>
+#include "sys/time.h"
+
+#define SW_NORMAL 0
+#define SW_AVX2 1
+#define SW_CUDA 2
+#define SW_ALL 3
+
+#define BLOCK_BUF_SIZE 1048576
+#define READ_BUF_SIZE 2048
+#define SEQ_BUF_SIZE (BLOCK_BUF_SIZE + READ_BUF_SIZE)
+
+// 将文件读取到离散的内存里，看一下对sw性能的影响
+
+#ifdef SHOW_PERF
+// 用来调试，计算感兴趣部分的运行时间
+// 获取当前毫秒数
+int64_t get_mseconds()
+{
+    struct timeval tv;
+    gettimeofday(&tv, NULL);
+    return (int64_t)1000 * (tv.tv_sec + ((1e-6) * tv.tv_usec));
+}
+
+int64_t time_sw_normal = 0,
+        time_sw_avx2 = 0,
+        time_sw_avx2_u8 = 0;
+
+#endif
+
+extern int ksw_normal(int qlen, const uint8_t *query, int tlen, const uint8_t *target, int m, const int8_t *mat, int o_del, int e_del, int o_ins, int e_ins, int w, int end_bonus, int zdrop, int h0, int *_qle, int *_tle, int *_gtle, int *_gscore, int *_max_off);
+extern int ksw_avx2(int qlen, const uint8_t *query, int tlen, const uint8_t *target, int is_left, int m, const int8_t *mat, int o_del, int e_del,
+                    int o_ins, int e_ins, int a, int b, int w, int end_bonus, int zdrop, int h0, int *_qle, int *_tle, int *_gtle, int *_gscore, int *_max_off);
+extern int ksw_avx2_u8(int qlen, const uint8_t *query, int tlen, const uint8_t *target, int is_left, int m, const int8_t *mat, int o_del, int e_del,
+                       int o_ins, int e_ins, int a, int b, int w, int end_bonus, int zdrop, int h0, int *_qle, int *_tle, int *_gtle, int *_gscore, int *_max_off);
+
+/*
+ * 包含一个参数，用来区分调用那个sw算法
+ * 参数为 normal/avx2/cuda
+ */
+// 程序执行入口
+int main(int argc, char *argv[])
+{
+
+    // 初始化一些全局参数
+    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 max_off[2];
+    int qle, tle, gtle, gscore;
+
+    // 读取测试数据
+    char *query_arr = (char **)malloc(SEQ_BUF_SIZE);
+    char *target_arr = (char **)malloc(SEQ_BUF_SIZE);
+    int *info_buf = (int *)malloc(SEQ_BUF_SIZE);
+    int **info_arr = (int **)malloc(SEQ_BUF_SIZE);
+    FILE *query_f = 0, *target_f = 0, *info_f = 0;
+    // const char *qf_path = "/public/home/zzh/data/sw/q_s.fa";
+    // const char *tf_path = "/public/home/zzh/data/sw/t_s.fa";
+    // const char *if_path = "/public/home/zzh/data/sw/i_s.txt";
+    const char *qf_path = "/public/home/zzh/data/sw/q_m.fa";
+    const char *tf_path = "/public/home/zzh/data/sw/t_m.fa";
+    const char *if_path = "/public/home/zzh/data/sw/i_m.txt";
+    // const char *qf_path = "/public/home/zzh/data/sw/q_m.fa";
+    // const char *tf_path = "/public/home/zzh/data/sw/t_m.fa";
+    // const char *if_path = "/public/home/zzh/data/sw/i_m.txt";
+    query_f = fopen(qf_path, "r");
+    target_f = fopen(tf_path, "r");
+    info_f = fopen(if_path, "r");
+
+    // 每次读取一定量的数据，然后执行，直到处理完所有数据
+    int total_line_num = 0; // 目前处理的总的数据行数
+    int block_line_num = 0; // 当前循环包含的数据行数
+    int i, j;
+    // const int max_read = READ_BUF_SIZE; // 每次最多读取的字符
+    char read_buf[READ_BUF_SIZE]; // 读文件缓存
+    // int ret_code = 0;
+
+    // 初始化info_arr数组
+    i = 0;
+    j = 0;
+    while (1)
+    {
+        if (j > BLOCK_BUF_SIZE)
+            break;
+        info_arr[i] = &info_buf[j];
+        i += 1;
+        j += 3;
+    }
+
+    int score_normal = 0, score_avx2 = 0, score_avx2_u8 = 0;
+
+    while (!feof(target_f))
+    {
+        block_line_num = 0;
+        // target序列一般占用存储最多，先读取target，看一个buf能读多少行，query和info就按照这个行数来读
+        int cur_read_size = 0;
+        while (!feof(target_f) && cur_read_size < BLOCK_BUF_SIZE)
+        {
+            if (fgets(read_buf, READ_BUF_SIZE, target_f) == NULL)
+                break;
+            const int line_size = strlen(read_buf);
+            target_arr[block_line_num] = (char *)malloc(line_size);
+            assert(line_size < READ_BUF_SIZE);
+            ++block_line_num;
+            ++total_line_num;
+            strncpy(target_arr[block_line_num], read_buf, line_size);
+            cur_read_size += line_size;
+            // fprintf(stderr, "%d %d \n", line_size, cur_read_size);
+        }
+
+        // 读query
+        cur_read_size = 0;
+        for (i = 0; i < block_line_num; ++i)
+        {
+            if (fgets(read_buf, READ_BUF_SIZE, query_f) == NULL)
+                break;
+            const int line_size = strlen(read_buf);
+            assert(line_size < READ_BUF_SIZE);
+            query_arr[i] = (char *)malloc(line_size);
+            strncpy(query_arr[i], read_buf, line_size);
+            cur_read_size += line_size;
+        }
+
+        // 读info
+        cur_read_size = 0;
+        for (i = 0; i < block_line_num; ++i)
+        {
+            if (fgets(read_buf, READ_BUF_SIZE, info_f) == NULL)
+                break;
+            const int line_size = strlen(read_buf);
+            assert(line_size < READ_BUF_SIZE);
+            sscanf(read_buf, "%d %d %d\n", &info_arr[i][0], &info_arr[i][1], &info_arr[i][2]);
+            cur_read_size += line_size;
+            // fprintf(stderr, "%-8d%-8d%-8d\n", info_arr[i][0], info_arr[i][1], info_arr[i][2]);
+            // fprintf(stderr, "%s\n", read_buf);
+        }
+
+        // 性能测试
+
+        // 普通 sw
+        int cur_query_pos = 0;
+        int cur_target_pos = 0;
+        for (i = 0; i < block_line_num; ++i)
+        {
+#ifdef SHOW_PERF
+            int64_t start_time = get_mseconds();
+#endif
+            score_normal += ksw_normal(
+                info_arr[i][0],
+                (uint8_t *)query_arr[i],
+                info_arr[i][1],
+                (uint8_t *)target_arr[i],
+                5, mat, 6, 1, 6, 1, 100, 5, 100,
+                info_arr[i][2],
+                &qle, &tle, &gtle, &gscore, &max_off[0]);
+#ifdef SHOW_PERF
+            time_sw_normal += get_mseconds() - start_time;
+#endif
+            // 更新query和target位置信息
+            cur_query_pos += info_arr[i][0];
+            cur_target_pos += info_arr[i][1];
+            // fprintf(stderr, "%d %d %d %d %d %d %d\n", score_normal, qle, tle, gtle, gscore, max_off[0], max_off[1]);
+        }
+
+        // avx2 sw
+        cur_query_pos = 0;
+        cur_target_pos = 0;
+        for (i = 0; i < block_line_num; ++i)
+        {
+#ifdef SHOW_PERF
+            int64_t start_time = get_mseconds();
+#endif
+            score_avx2 += ksw_avx2(
+                info_arr[i][0],
+                (uint8_t *)query_arr[i],
+                info_arr[i][1],
+                (uint8_t *)target_arr[i],
+                0, 5, mat, 6, 1, 6, 1,
+                1, 4,
+                100, 5, 100,
+                info_arr[i][2],
+                &qle, &tle, &gtle, &gscore, &max_off[0]);
+#ifdef SHOW_PERF
+            time_sw_avx2 += get_mseconds() - start_time;
+#endif
+            // 更新query和target位置信息
+            cur_query_pos += info_arr[i][0];
+            cur_target_pos += info_arr[i][1];
+            // fprintf(stderr, "%d %d %d %d %d %d %d\n", score_avx2, qle, tle, gtle, gscore, max_off[0], max_off[1]);
+        }
+
+        // avx2 u8 sw
+        cur_query_pos = 0;
+        cur_target_pos = 0;
+        for (i = 0; i < block_line_num; ++i)
+        {
+#ifdef SHOW_PERF
+            int64_t start_time = get_mseconds();
+#endif
+            score_avx2_u8 += ksw_avx2_u8(
+                info_arr[i][0],
+                (uint8_t *)query_arr[i],
+                info_arr[i][1],
+                (uint8_t *)target_arr[i],
+                0, 5, mat, 6, 1, 6, 1,
+                1, 4,
+                100, 5, 100,
+                info_arr[i][2],
+                &qle, &tle, &gtle, &gscore, &max_off[0]);
+#ifdef SHOW_PERF
+            time_sw_avx2_u8 += get_mseconds() - start_time;
+#endif
+            // 更新query和target位置信息
+            cur_query_pos += info_arr[i][0];
+            cur_target_pos += info_arr[i][1];
+            // fprintf(stderr, "%d %d %d %d %d %d %d\n", score_normal, qle, tle, gtle, gscore, max_off[0], max_off[1]);
+        }
+
+        // fprintf(stderr, "%d %d \n", block_line_num, total_line_num);
+    }
+
+    // fprintf(stderr, "%d \n", score_normal);
+
+#ifdef SHOW_PERF
+    fprintf(stderr, "time_sw_normal: %f s; score: %d\n", time_sw_normal / 1000.0, score_normal);
+    fprintf(stderr, "time_sw_avx2: %f s; score: %d\n", time_sw_avx2 / 1000.0, score_avx2);
+    fprintf(stderr, "time_sw_avx2_u8: %f s; score: %d\n", time_sw_avx2_u8 / 1000.0, score_avx2_u8);
+#endif
+
+    if (query_f != 0)
+        fclose(query_f);
+    if (target_f != 0)
+        fclose(target_f);
+    if (info_f != 0)
+        fclose(info_f);
+}
--- a/main.c
+++ b/main.c
@ -85,12 +85,15 @@ int main(int argc, char *argv[])
    // const char *qf_path = "/public/home/zzh/data/sw/q_s.fa";
    // const char *tf_path = "/public/home/zzh/data/sw/t_s.fa";
    // const char *if_path = "/public/home/zzh/data/sw/i_s.txt";
-    const char *qf_path = "/public/home/zzh/data/sw/q_m.fa";
-    const char *tf_path = "/public/home/zzh/data/sw/t_m.fa";
-    const char *if_path = "/public/home/zzh/data/sw/i_m.txt";
    // const char *qf_path = "/public/home/zzh/data/sw/q_m.fa";
    // const char *tf_path = "/public/home/zzh/data/sw/t_m.fa";
    // const char *if_path = "/public/home/zzh/data/sw/i_m.txt";
+    // const char *qf_path = "/public/home/zzh/data/sw/q_l.fa";
+    // const char *tf_path = "/public/home/zzh/data/sw/t_l.fa";
+    // const char *if_path = "/public/home/zzh/data/sw/i_l.txt";
+    const char *qf_path = "/public/home/zzh/data/sw/query.fa";
+    const char *tf_path = "/public/home/zzh/data/sw/target.fa";
+    const char *if_path = "/public/home/zzh/data/sw/info.txt";
    query_f = fopen(qf_path, "r");
    target_f = fopen(tf_path, "r");
    info_f = fopen(if_path, "r");
Author	SHA1	Message	Date
zzh	ede75a66fa	测试不同数据	2023-08-11 00:36:14 +08:00
zzh	5a41d5e207	测试不同内存开辟策略的影响	2023-08-11 00:35:43 +08:00