Canny preprocessor node does not work when using DirectML

[rabidcopy]

Windows + RX 570, 2ef459b

python main.py --dont-upcast-attention --fp16-vae --use-quad-cross-attention --normalvram --disable-smart-memory --preview-method auto --directml
Using directml with device:
Total VRAM 1024 MB, total RAM 16335 MB
Set vram state to: NORMAL_VRAM
Disabling smart memory management
Device: privateuseone
VAE dtype: torch.float16
disabling upcasting of attention
Using sub quadratic optimization for cross attention, if you have memory or speed issues try using: --use-split-cross-attention111

Import times for custom nodes:
   0.0 seconds: C:\Users\zatoc\Desktop\ComfyUI\custom_nodes\sampler_tonemap.py
   0.0 seconds: C:\Users\zatoc\Desktop\ComfyUI\custom_nodes\sampler_rescalecfg.py
   0.0 seconds: C:\Users\zatoc\Desktop\ComfyUI\custom_nodes\sampler_tonemap_rescalecfg.py
   0.0 seconds: C:\Users\zatoc\Desktop\ComfyUI\custom_nodes\sdxl_model_merging.py
   0.0 seconds: C:\Users\zatoc\Desktop\ComfyUI\custom_nodes\advanced_model_merging.py

Starting server

To see the GUI go to: http://127.0.0.1:8188
got prompt
!!! Exception during processing !!!
Traceback (most recent call last):
  File "C:\Users\zatoc\Desktop\ComfyUI\execution.py", line 152, in recursive_execute
    output_data, output_ui = get_output_data(obj, input_data_all)
  File "C:\Users\zatoc\Desktop\ComfyUI\execution.py", line 82, in get_output_data
    return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
  File "C:\Users\zatoc\Desktop\ComfyUI\execution.py", line 75, in map_node_over_list
    results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
  File "C:\Users\zatoc\Desktop\ComfyUI\comfy_extras\nodes_canny.py", line 293, in detect_edge
    output = canny(image.to(comfy.model_management.get_torch_device()).movedim(-1, 1), low_threshold, high_threshold)
  File "C:\Users\zatoc\Desktop\ComfyUI\comfy_extras\nodes_canny.py", line 248, in canny
    edges: Tensor = F.threshold(magnitude, low_threshold, 0.0)
  File "C:\Users\zatoc\Desktop\ComfyUI\venv\lib\site-packages\torch\nn\functional.py", line 1427, in _threshold
    result = _VF.threshold(input, threshold, value)
RuntimeError: The parameter is incorrect.

Downgrading just nodes_canny.py from d70b0bc to bdba394 makes it work, assumedly because it is not running on the GPU and running into this issue with DirectML.

[MythicalChu]

I'm on DirectML with a RX 580 8Gb and it's working just fine (Windows too). My ComfyUI is updated to just a few days ago, also, I'm not using any of the flags you are on start up. Maybe one of them is causing your problem?

[rabidcopy]

I'm on DirectML with a RX 580 8Gb and it's working just fine (Windows too). My ComfyUI is updated to just a few days ago, also, I'm not using any of the flags you are on start up. Maybe one of them is causing your problem?

Still throws the error when running just python main.py --directml on latest commit.

[Lorg0n]

Hmmm, weird. I have a rx570 8gb and everything works fine.

[drago87]

Having the same problem on a RX 5700 XT 8gb