esa-kernel/diff_map.py at master · pyxyzc/esa-kernel · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
import numpy as np
import torch
import random

def diff_two_map(map1, map2):
    keys2 = map2.keys()
    values2 = map2.values()
    keys2_set = set(keys2)
    values2_set = set(values2)
    diff_map = {}
    updated_map = {}
    for k1, v1 in map1.items():
        if k1 in keys2 and v1 in values2:
            updated_map[k1] = v1
            keys2_set.remove(k1)
            values2_set.remove(v1)
    for k2, v2 in zip(keys2_set, values2_set):
        diff_map[k2] = v2
        updated_map[k2] = v2
    return updated_map, diff_map

def diff_two_map_simple(keys, old_values, new_values):
    remain_keys = []
    remain_new_values = []
    n = len(old_values)
    remain_flags = [True for _ in range(n)]
    for i, e in enumerate(new_values):
        index = np.where(old_values == e)[0]
        if index.shape[0] == 0:
            remain_new_values.append(e)
        else:
            remain_flags[index[0]] = False

    for i, e in enumerate(remain_flags):
        if e:
            remain_keys.append(keys[i])
    # print('remain_keys:',remain_keys)
    # print('remain_new_values:',remain_new_values)


map_size = 100
map1 = {}
for i in range(map_size):
    map1[i] = i * 3 + 1
map2 = {}
map1_values = list(map1.values())
random.shuffle(map1_values)
overlap = 30
for i in range(map_size):
    if i < overlap:
        map2[i] = map1_values[i]
    else:
        map2[i] = i + i *i + map_size * 2

keys = np.array(list(map1.keys()))
old_values = np.array(list(map1.values()))
new_values = np.array(list(map2.values()))

import time
times = []
for _ in range(100):
    begin = time.perf_counter_ns()
    # res1 =diff_two_map(map1, map2)
    diff_two_map_simple(keys, old_values, new_values)
    dura = time.perf_counter_ns() - begin
    times.append(dura)
times = times[10:]
print(f'simple time: {sum(times)/len(times) / 1e6}ms')


def diff_two_map_np(keys, old_values, new_values):
    equal = new_values[:, None] == old_values[None, :]

    # diff = new_values[:, None] - old_values[None, :] # N(new_values) * N(keys)
    # equal = (diff < 1e-3)


    # remain_keys = keys[~np.any(equal, axis=0)]
    # remain_new_values = new_values[~np.any(equal, axis=1)]

    rows = np.sum(equal, 1)
    cols = np.sum(equal, 0)
    # remain_new_values = new_values[np.where(rows == 0)]
    # remain_keys = keys[np.where(cols == 0)]

    remain_new_values = new_values[rows == 0]
    remain_keys = keys[cols == 0]

    return remain_keys, remain_new_values

def diff_two_map_torch(keys, old_values, new_values):
    equal = (new_values[:, None] == old_values[None, :])
    # diff = new_values[:, None] - old_values[None, :] # N(new_values) * N(keys)
    # equal = (diff < 1e-3)
    rows = torch.sum(equal, 1)
    cols = torch.sum(equal, 0)
    remain_new_values = new_values[rows == 0]
    remain_keys = keys[cols == 0]
    return remain_keys, remain_new_values


times = []
for _ in range(100):
    begin = time.perf_counter_ns()
    res2 = diff_two_map_np(keys, old_values, new_values)
    dura = time.perf_counter_ns() - begin
    # print("np dura", dura/1e6)
    times.append(dura)
times = times[10:]
print(f'time: {sum(times)/len(times) / 1e6}ms')


keys = torch.tensor(keys).cuda().to(torch.int32)
old_values = torch.tensor(old_values).cuda().to(torch.int32)
new_values = torch.tensor(new_values).cuda().to(torch.int32)

times = []
for _ in range(100):
    begin = time.perf_counter_ns()
    res3 = diff_two_map_torch(keys, old_values, new_values)
    torch.cuda.synchronize()
    dura = time.perf_counter_ns() - begin
    times.append(dura)
times = times[10:]
print(f'torch time: {sum(times)/len(times) / 1e6}ms')

import diff_map as diff_lib
times = []
for _ in range(100):
    begin = time.perf_counter_ns()
    unkept_keys, unkept_vals = diff_lib.filter_unkept(keys, old_values, new_values)
    torch.cuda.synchronize()
    dura = time.perf_counter_ns() - begin
    times.append(dura)
times = times[10:]
print(f'time: {sum(times)/len(times) / 1e6}ms')