-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_prefetcher.py
91 lines (80 loc) · 3.04 KB
/
data_prefetcher.py
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
"""
Based off PyTorch/NVIDIA AMP data preloader.
https://github.com/NVIDIA/apex/blob/master/examples/imagenet/main_amp.py
"""
import torch
class data_prefetcher():
"""Prefetches data to GPU from a tf.loader"""
def __init__(self, loader, length,
permute_channel: bool,
stream_prefetch: bool=True):
"""
loader: A tf.data loader
length: The length of the loader (None if infinite, but don't do that)
permute_channel: If channel should be permuted on GPU
"""
self.base_loader = loader
self.loader = iter(loader)
self.stream = torch.cuda.Stream()
self.length = length
self.permute_channel = permute_channel
self.stream_prefetch = stream_prefetch
self.idx = 0
self.preload()
def __len__(self):
return self.length
def query_idle_stream(self) -> bool:
"""
Returns true if stream is not active with work (idle)
"""
return self.stream.query()
def preload(self):
try:
self.next_input, self.next_target = next(self.loader)
self.next_input = self.next_input._numpy()
self.next_target = self.next_target._numpy()
self.next_input_py = torch.from_numpy(self.next_input).pin_memory()
self.next_target_py = torch.from_numpy(self.next_target).pin_memory()
except StopIteration:
self.next_input = None
self.next_target = None
return
if self.stream_prefetch:
with torch.cuda.stream(self.stream):
self.next_input = self.next_input_py.cuda(non_blocking=True)
self.next_target = self.next_target_py.cuda(non_blocking=True)
if self.permute_channel:
self.next_input = self.next_input.permute(0, 3, 1, 2)
else:
self.next_input = self.next_input_py
self.next_target = self.next_target_py
if self.permute_channel:
self.next_input = self.next_input.permute(0, 3, 1, 2)
def reset(self):
self.idx = 0
self.loader = iter(self.base_loader)
self.stream = torch.cuda.Stream()
self.preload()
def next(self):
if self.stream_prefetch:
torch.cuda.current_stream().wait_stream(self.stream)
curr_input = self.next_input
curr_target = self.next_target
if self.stream_prefetch:
if curr_input is not None:
curr_input.record_stream(torch.cuda.current_stream())
if curr_target is not None:
curr_target.record_stream(torch.cuda.current_stream())
self.preload()
self.idx += 1
return curr_input, curr_target
def __iter__(self):
self.idx = 0
return self
def __next__(self):
if self.length is not None and self.idx >= self.length:
raise StopIteration
input, target = self.next()
if input is None or target is None:
raise StopIteration
return input, target