File size: 4,200 Bytes
c9ea4f0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
import sys
import contextlib
from functools import lru_cache

import torch
from modules import errors, shared

if sys.platform == "darwin":
    from modules import mac_specific


def has_mps() -> bool:
    if sys.platform != "darwin":
        return False
    else:
        return mac_specific.has_mps


def get_cuda_device_string():
    if shared.cmd_opts.device_id is not None:
        return f"cuda:{shared.cmd_opts.device_id}"

    return "cuda"


def get_optimal_device_name():
    if torch.cuda.is_available():
        return get_cuda_device_string()

    if has_mps():
        return "mps"

    return "cpu"


def get_optimal_device():
    return torch.device(get_optimal_device_name())


def get_device_for(task):
    if task in shared.cmd_opts.use_cpu:
        return cpu

    return get_optimal_device()


def torch_gc():

    if torch.cuda.is_available():
        with torch.cuda.device(get_cuda_device_string()):
            torch.cuda.empty_cache()
            torch.cuda.ipc_collect()

    if has_mps():
        mac_specific.torch_mps_gc()


def enable_tf32():
    if torch.cuda.is_available():

        # enabling benchmark option seems to enable a range of cards to do fp16 when they otherwise can't
        # see https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/4407
        if any(torch.cuda.get_device_capability(devid) == (7, 5) for devid in range(0, torch.cuda.device_count())):
            torch.backends.cudnn.benchmark = True

        torch.backends.cuda.matmul.allow_tf32 = True
        torch.backends.cudnn.allow_tf32 = True


errors.run(enable_tf32, "Enabling TF32")

cpu: torch.device = torch.device("cpu")
device: torch.device = None
device_interrogate: torch.device = None
device_gfpgan: torch.device = None
device_esrgan: torch.device = None
device_codeformer: torch.device = None
dtype: torch.dtype = torch.float16
dtype_vae: torch.dtype = torch.float16
dtype_unet: torch.dtype = torch.float16
unet_needs_upcast = False


def cond_cast_unet(input):
    return input.to(dtype_unet) if unet_needs_upcast else input


def cond_cast_float(input):
    return input.float() if unet_needs_upcast else input


nv_rng = None


def autocast(disable=False):
    if disable:
        return contextlib.nullcontext()

    if dtype == torch.float32 or shared.cmd_opts.precision == "full":
        return contextlib.nullcontext()

    return torch.autocast("cuda")


def without_autocast(disable=False):
    return torch.autocast("cuda", enabled=False) if torch.is_autocast_enabled() and not disable else contextlib.nullcontext()


class NansException(Exception):
    pass


def test_for_nans(x, where):
    if shared.cmd_opts.disable_nan_check:
        return

    if not torch.all(torch.isnan(x)).item():
        return

    if where == "unet":
        message = "A tensor with all NaNs was produced in Unet."

        if not shared.cmd_opts.no_half:
            message += " This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the \"Upcast cross attention layer to float32\" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this."

    elif where == "vae":
        message = "A tensor with all NaNs was produced in VAE."

        if not shared.cmd_opts.no_half and not shared.cmd_opts.no_half_vae:
            message += " This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this."
    else:
        message = "A tensor with all NaNs was produced."

    message += " Use --disable-nan-check commandline argument to disable this check."

    raise NansException(message)


@lru_cache
def first_time_calculation():
    """
    just do any calculation with pytorch layers - the first time this is done it allocaltes about 700MB of memory and
    spends about 2.7 seconds doing that, at least wih NVidia.
    """

    x = torch.zeros((1, 1)).to(device, dtype)
    linear = torch.nn.Linear(1, 1).to(device, dtype)
    linear(x)

    x = torch.zeros((1, 1, 3, 3)).to(device, dtype)
    conv2d = torch.nn.Conv2d(1, 1, (3, 3)).to(device, dtype)
    conv2d(x)