modules/devices.py

import os
import gc
import sys
import time
import contextlib
import torch
from modules.errors import log
from modules import cmd_args, shared, memstats, errors

if sys.platform == "darwin":
    from modules import mac_specific # pylint: disable=ungrouped-imports


previous_oom = 0
backup_sdpa = None
debug = os.environ.get('SD_DEVICE_DEBUG', None) is not None


def has_mps() -> bool:
    if sys.platform != "darwin":
        return False
    else:
        return mac_specific.has_mps # pylint: disable=used-before-assignment


def get_gpu_info():
    def get_driver():
        import subprocess
        if torch.cuda.is_available() and torch.version.cuda:
            try:
                result = subprocess.run('nvidia-smi --query-gpu=driver_version --format=csv,noheader', shell=True, check=False, env=os.environ, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
                version = result.stdout.decode(encoding="utf8", errors="ignore").strip()
                return version
            except Exception:
                return ''
        else:
            return ''

    def get_package_version(pkg: str):
        import pkg_resources
        spec = pkg_resources.working_set.by_key.get(pkg, None) # more reliable than importlib
        version = pkg_resources.get_distribution(pkg).version if spec is not None else ''
        return version

    if not torch.cuda.is_available():
        try:
            if shared.cmd_opts.use_openvino:
                return {
                    'device': get_openvino_device(),
                    'openvino': get_package_version("openvino"),
                }
            elif shared.cmd_opts.use_directml:
                return {
                    'device': f'{torch.cuda.get_device_name(torch.cuda.current_device())} n={torch.cuda.device_count()}',
                    'directml': get_package_version("torch-directml"),
                }
            else:
                return {}
        except Exception:
            return {}
    else:
        try:
            if hasattr(torch, "xpu") and torch.xpu.is_available():
                return {
                    'device': f'{torch.xpu.get_device_name(torch.xpu.current_device())} n={torch.xpu.device_count()}',
                    'ipex': get_package_version('intel-extension-for-pytorch'),
                }
            elif torch.version.cuda:
                return {
                    'device': f'{torch.cuda.get_device_name(torch.cuda.current_device())} n={torch.cuda.device_count()} arch={torch.cuda.get_arch_list()[-1]} cap={torch.cuda.get_device_capability(device)}',
                    'cuda': torch.version.cuda,
                    'cudnn': torch.backends.cudnn.version(),
                    'driver': get_driver(),
                }
            elif torch.version.hip:
                return {
                    'device': f'{torch.cuda.get_device_name(torch.cuda.current_device())} n={torch.cuda.device_count()}',
                    'hip': torch.version.hip,
                }
            else:
                return {
                    'device': 'unknown'
                }
        except Exception as ex:
            if debug:
                errors.display(ex, 'Device exception')
            return { 'error': ex }


def extract_device_id(args, name): # pylint: disable=redefined-outer-name
    for x in range(len(args)):
        if name in args[x]:
            return args[x + 1]
    return None


def get_cuda_device_string():
    if backend == 'ipex':
        if shared.cmd_opts.device_id is not None:
            return f"xpu:{shared.cmd_opts.device_id}"
        return "xpu"
    elif backend == 'directml' and torch.dml.is_available():
        if shared.cmd_opts.device_id is not None:
            return f"privateuseone:{shared.cmd_opts.device_id}"
        return torch.dml.get_device_string(torch.dml.default_device().index)
    else:
        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 cuda_ok or backend == 'directml':
        return get_cuda_device_string()
    if has_mps() and backend != 'openvino':
        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:
        log.debug(f'Forcing CPU for task: {task}')
        return cpu
    return get_optimal_device()


def torch_gc(force=False):
    t0 = time.time()
    mem = memstats.memory_stats()
    gpu = mem.get('gpu', {})
    ram = mem.get('ram', {})
    oom = gpu.get('oom', 0)
    if backend == "directml":
        used_gpu = round(100 * torch.cuda.memory_allocated() / (1 << 30) / gpu.get('total', 1)) if gpu.get('total', 1) > 1 else 0
    else:
        used_gpu = round(100 * gpu.get('used', 0) / gpu.get('total', 1)) if gpu.get('total', 1) > 1 else 0
    used_ram = round(100 * ram.get('used', 0) / ram.get('total', 1)) if ram.get('total', 1) > 1 else 0
    global previous_oom # pylint: disable=global-statement
    if oom > previous_oom:
        previous_oom = oom
        log.warning(f'GPU out-of-memory error: {mem}')
        force = True
    if used_gpu >= shared.opts.torch_gc_threshold or used_ram >= shared.opts.torch_gc_threshold:
        log.info(f'High memory utilization: GPU={used_gpu}% RAM={used_ram}% {mem}')
        force = True
    if not force:
        return

    # actual gc
    collected = gc.collect() # python gc
    if cuda_ok:
        try:
            with torch.cuda.device(get_cuda_device_string()):
                torch.cuda.empty_cache() # cuda gc
                torch.cuda.ipc_collect()
        except Exception:
            pass
    t1 = time.time()
    log.debug(f'GC: collected={collected} device={torch.device(get_optimal_device_name())} {memstats.memory_stats()} time={round(t1 - t0, 2)}')


def set_cuda_sync_mode(mode):
    """
    Set the CUDA device synchronization mode: auto, spin, yield or block.
    auto: Chooses spin or yield depending on the number of available CPU cores.
    spin: Runs one CPU core per GPU at 100% to poll for completed operations.
    yield: Gives control to other threads between polling, if any are waiting.
    block: Lets the thread sleep until the GPU driver signals completion.
    """
    if mode == -1 or mode == 'none' or not cuda_ok:
        return
    try:
        import ctypes
        log.info(f'Set cuda synch: mode={mode}')
        torch.cuda.set_device(torch.device(get_optimal_device_name()))
        ctypes.CDLL('libcudart.so').cudaSetDeviceFlags({'auto': 0, 'spin': 1, 'yield': 2, 'block': 4}[mode])
    except Exception:
        pass


def test_fp16():
    if shared.cmd_opts.experimental:
        return True
    try:
        x = torch.tensor([[1.5,.0,.0,.0]]).to(device=device, dtype=torch.float16)
        layerNorm = torch.nn.LayerNorm(4, eps=0.00001, elementwise_affine=True, dtype=torch.float16, device=device)
        _y = layerNorm(x)
        return True
    except Exception as ex:
        log.warning(f'Torch FP16 test failed: Forcing FP32 operations: {ex}')
        shared.opts.cuda_dtype = 'FP32'
        shared.opts.no_half = True
        shared.opts.no_half_vae = True
        return False

def test_bf16():
    if shared.cmd_opts.experimental:
        return True
    try:
        import torch.nn.functional as F
        image = torch.randn(1, 4, 32, 32).to(device=device, dtype=torch.bfloat16)
        _out = F.interpolate(image, size=(64, 64), mode="nearest")
        return True
    except Exception:
        log.warning('Torch BF16 test failed: Fallback to FP16 operations')
        return False


def set_cuda_params():
    # log.debug('Verifying Torch settings')
    if cuda_ok:
        try:
            torch.backends.cuda.matmul.allow_tf32 = True
            torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = True
            torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction = True
        except Exception:
            pass
        if torch.backends.cudnn.is_available():
            try:
                torch.backends.cudnn.deterministic = shared.opts.cudnn_deterministic
                torch.backends.cudnn.benchmark = True
                if shared.opts.cudnn_benchmark:
                    log.debug('Torch enable cuDNN benchmark')
                    torch.backends.cudnn.benchmark_limit = 0
                torch.backends.cudnn.allow_tf32 = True
            except Exception:
                pass
    try:
        if shared.opts.cross_attention_optimization == "Scaled-Dot-Product" or shared.opts.cross_attention_optimization == "Dynamic Attention SDP":
            torch.backends.cuda.enable_flash_sdp('Flash attention' in shared.opts.sdp_options)
            torch.backends.cuda.enable_mem_efficient_sdp('Memory attention' in shared.opts.sdp_options)
            torch.backends.cuda.enable_math_sdp('Math attention' in shared.opts.sdp_options)
            if backend == "rocm":
                global backup_sdpa # pylint: disable=global-statement
                if 'Flash attention' in shared.opts.sdp_options:
                    try:
                        # https://github.com/huggingface/diffusers/discussions/7172
                        from flash_attn import flash_attn_func
                        if backup_sdpa is None:
                            backup_sdpa = torch.nn.functional.scaled_dot_product_attention
                        def sdpa_hijack(query, key, value, attn_mask=None, dropout_p=0.0, is_causal=False, scale=None):
                            if query.shape[3] <= 128 and attn_mask is None:
                                return flash_attn_func(q=query.transpose(1, 2), k=key.transpose(1, 2), v=value.transpose(1, 2), dropout_p=dropout_p, causal=is_causal, softmax_scale=scale).transpose(1, 2)
                            else:
                                return backup_sdpa(query=query, key=key, value=value, attn_mask=attn_mask, dropout_p=dropout_p, is_causal=is_causal, scale=scale)
                        torch.nn.functional.scaled_dot_product_attention = sdpa_hijack
                        shared.log.debug('ROCm Flash Attention Hijacked')
                    except Exception as err:
                        log.error(f'ROCm Flash Attention failed: {err}')
                elif backup_sdpa is not None: # Restore original SDPA
                    torch.nn.functional.scaled_dot_product_attention = backup_sdpa
    except Exception:
        pass
    if shared.cmd_opts.profile:
        shared.log.debug(f'Torch info: {torch.__config__.show()}')
    global dtype, dtype_vae, dtype_unet, unet_needs_upcast, inference_context # pylint: disable=global-statement
    if shared.opts.cuda_dtype == 'FP32':
        dtype = torch.float32
        dtype_vae = torch.float32
        dtype_unet = torch.float32
        fp16_ok = None
        bf16_ok = None
    elif shared.opts.cuda_dtype == 'BF16' or dtype == torch.bfloat16:
        fp16_ok = test_fp16()
        bf16_ok = test_bf16()
        dtype = torch.bfloat16 if bf16_ok else torch.float16
        dtype_vae = torch.bfloat16 if bf16_ok else torch.float16
        dtype_unet = torch.bfloat16 if bf16_ok else torch.float16
    elif shared.opts.cuda_dtype == 'FP16' or dtype == torch.float16:
        fp16_ok = test_fp16()
        bf16_ok = None
        dtype = torch.float16 if fp16_ok else torch.float32
        dtype_vae = torch.float16 if fp16_ok else torch.float32
        dtype_unet = torch.float16 if fp16_ok else torch.float32
    if shared.opts.no_half:
        log.info('Torch override dtype: no-half set')
        dtype = torch.float32
        dtype_vae = torch.float32
        dtype_unet = torch.float32
    if shared.opts.no_half_vae: # set dtype again as no-half-vae options take priority
        log.info('Torch override VAE dtype: no-half set')
        dtype_vae = torch.float32
    unet_needs_upcast = shared.opts.upcast_sampling
    if shared.opts.inference_mode == 'inference-mode':
        inference_context = torch.inference_mode
    elif shared.opts.inference_mode == 'none':
        inference_context = contextlib.nullcontext
    else:
        inference_context = torch.no_grad
    log_device_name = get_raw_openvino_device() if shared.cmd_opts.use_openvino else torch.device(get_optimal_device_name())
    log.debug(f'Desired Torch parameters: dtype={shared.opts.cuda_dtype} no-half={shared.opts.no_half} no-half-vae={shared.opts.no_half_vae} upscast={shared.opts.upcast_sampling}')
    log.info(f'Setting Torch parameters: device={log_device_name} dtype={dtype} vae={dtype_vae} unet={dtype_unet} context={inference_context.__name__} fp16={fp16_ok} bf16={bf16_ok} optimization={shared.opts.cross_attention_optimization}')


args = cmd_args.parser.parse_args()
backend = 'not set'
if args.use_openvino:
    from modules.intel.openvino import get_openvino_device
    from modules.intel.openvino import get_device as get_raw_openvino_device
    backend = 'openvino'
    if hasattr(torch, 'xpu') and torch.xpu.is_available():
        torch.xpu.is_available = lambda *args, **kwargs: False
    torch.cuda.is_available = lambda *args, **kwargs: False
elif args.use_ipex or (hasattr(torch, 'xpu') and torch.xpu.is_available()):
    backend = 'ipex'
    from modules.intel.ipex import ipex_init
    ok, e = ipex_init()
    if not ok:
        log.error(f'IPEX initialization failed: {e}')
        backend = 'cpu'
elif args.use_directml:
    backend = 'directml'
    from modules.dml import directml_init
    ok, e = directml_init()
    if not ok:
        log.error(f'DirectML initialization failed: {e}')
        backend = 'cpu'
elif torch.cuda.is_available() and torch.version.cuda:
    backend = 'cuda'
elif torch.cuda.is_available() and torch.version.hip:
    backend = 'rocm'
elif sys.platform == 'darwin':
    backend = 'mps'
else:
    backend = 'cpu'

inference_context = torch.no_grad
cuda_ok = torch.cuda.is_available()
cpu = torch.device("cpu")
device = device_interrogate = device_gfpgan = device_esrgan = device_codeformer = None
dtype = torch.float16
dtype_vae = torch.float16
dtype_unet = torch.float16
unet_needs_upcast = False
onnx = None
if args.profile:
    log.info(f'Torch build config: {torch.__config__.show()}')
# set_cuda_sync_mode('block') # none/auto/spin/yield/block


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


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


def randn(seed, shape):
    torch.manual_seed(seed)
    if backend == 'ipex':
        torch.xpu.manual_seed_all(seed)
    if device.type == 'mps':
        return torch.randn(shape, device=cpu).to(device)
    return torch.randn(shape, device=device)


def randn_without_seed(shape):
    if device.type == 'mps':
        return torch.randn(shape, device=cpu).to(device)
    return torch.randn(shape, device=device)


def autocast(disable=False):
    if disable:
        return contextlib.nullcontext()
    if dtype == torch.float32 or shared.cmd_opts.precision == "Full":
        return contextlib.nullcontext()
    if shared.cmd_opts.use_directml:
        return torch.dml.amp.autocast(dtype)
    if cuda_ok:
        return torch.autocast("cuda")
    else:
        return torch.autocast("cpu")


def without_autocast(disable=False):
    if disable:
        return contextlib.nullcontext()
    if shared.cmd_opts.use_directml:
        return torch.dml.amp.autocast(enabled=False) if torch.is_autocast_enabled() else contextlib.nullcontext() # pylint: disable=unexpected-keyword-arg
    if cuda_ok:
        return torch.autocast("cuda", enabled=False) if torch.is_autocast_enabled() else contextlib.nullcontext()
    else:
        return torch.autocast("cpu", enabled=False) if torch.is_autocast_enabled() else contextlib.nullcontext()


class NansException(Exception):
    pass


def test_for_nans(x, where):
    if shared.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.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.opts.no_half and not shared.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)