stable-diffusion-webui/modules/devices.py

84 lines
2.2 KiB
Python
Raw Normal View History

2022-10-21 16:11:07 -06:00
import sys, os, shlex
import contextlib
2022-09-10 23:11:27 -06:00
import torch
from modules import errors
# has_mps is only available in nightly pytorch (for now), `getattr` for compatibility
2022-09-10 23:11:27 -06:00
has_mps = getattr(torch, 'has_mps', False)
2022-09-11 09:48:36 -06:00
cpu = torch.device("cpu")
2022-10-21 16:11:07 -06:00
def extract_device_id(args, name):
for x in range(len(args)):
if name in args[x]: return args[x+1]
return None
2022-09-11 09:48:36 -06:00
2022-09-10 23:11:27 -06:00
def get_optimal_device():
2022-09-11 09:48:36 -06:00
if torch.cuda.is_available():
from modules import shared
device_id = shared.cmd_opts.device_id
2022-10-21 16:11:07 -06:00
if device_id is not None:
cuda_device = f"cuda:{device_id}"
return torch.device(cuda_device)
else:
return torch.device("cuda")
2022-09-11 09:48:36 -06:00
if has_mps:
return torch.device("mps")
return cpu
def torch_gc():
if torch.cuda.is_available():
torch.cuda.empty_cache()
torch.cuda.ipc_collect()
def enable_tf32():
if torch.cuda.is_available():
torch.backends.cuda.matmul.allow_tf32 = True
torch.backends.cudnn.allow_tf32 = True
errors.run(enable_tf32, "Enabling TF32")
2022-09-12 11:09:32 -06:00
device = device_interrogate = device_gfpgan = device_bsrgan = device_esrgan = device_scunet = device_codeformer = None
dtype = torch.float16
2022-10-10 07:11:14 -06:00
dtype_vae = torch.float16
2022-09-12 11:09:32 -06:00
def randn(seed, shape):
# Pytorch currently doesn't handle setting randomness correctly when the metal backend is used.
if device.type == 'mps':
generator = torch.Generator(device=cpu)
generator.manual_seed(seed)
noise = torch.randn(shape, generator=generator, device=cpu).to(device)
return noise
torch.manual_seed(seed)
return torch.randn(shape, device=device)
def randn_without_seed(shape):
# Pytorch currently doesn't handle setting randomness correctly when the metal backend is used.
if device.type == 'mps':
generator = torch.Generator(device=cpu)
noise = torch.randn(shape, generator=generator, device=cpu).to(device)
return noise
return torch.randn(shape, device=device)
2022-10-10 07:11:14 -06:00
def autocast(disable=False):
from modules import shared
2022-10-10 07:11:14 -06:00
if disable:
return contextlib.nullcontext()
if dtype == torch.float32 or shared.cmd_opts.precision == "full":
return contextlib.nullcontext()
return torch.autocast("cuda")