Closed kedaitinh12 closed 8 months ago
Try with the attached DirectML.dll.
same error
Can you share RealESRGAN_x4plus_anime_6B.onnx
?
Thanks.
Are you sure that the vs version is running without specifying tilesize and with the same gpu selected?
yes, sample preview work + script + GPU usage in the image below
You're using a wrapper. Can you try with core.ort.Model(....)
?
I use Hybrid so I know about this app. If I add title size the script will like this
Please redownload runtime files from Releases and follow the updated guide about the runtime files.
I do like you but both Cuda and DML don't work now LoadPlugin("C:\Encode Tools\MeGUI-2944-64\tools\ffms\ffms2.dll") FFImageSource("C:\Users\ADMIN\Downloads\LCFome7.jpeg") z_ConvertFormat(pixel_type="rgbps", colorspace_op="709:709:709:l=>rgb:709:709:f") mlrt_ort(network_path="C:/Program Files (x86)/AviSynth+/plugins64+/models/RealESRGAN/RealESRGAN_x4plus_anime_6B.onnx", builtin=false, fp16=true, provider="cuda", device=0) propSet("_FieldBased",0) deep_resize(1920,1080,grain=0) z_ConvertFormat(pixel_type="yuv420p8", colorspace_op="rgb:709:709:f=>709:709:709:l") LoadPlugin("C:\Encode Tools\MeGUI-2944-64\tools\ffms\ffms2.dll") FFImageSource("C:\Users\ADMIN\Downloads\LCFome7.jpeg") z_ConvertFormat(pixel_type="rgbps", colorspace_op="709:709:709:l=>rgb:709:709:f") mlrt_ort(network_path="C:/Program Files (x86)/AviSynth+/plugins64+/models/RealESRGAN/RealESRGAN_x4plus_anime_6B.onnx", builtin=false, fp16=true, provider="dml", device=1) propSet("_FieldBased",0) deep_resize(1920,1080,grain=0) z_ConvertFormat(pixel_type="yuv420p8", colorspace_op="rgb:709:709:f=>709:709:709:l") My mlrt_ort_loader.avsi LoadDLL(".\ort_dll\cublasLt64_12.dll") LoadDLL(".\ort_dll\cudart64_12.dll") LoadDLL(".\ort_dll\cublas64_12.dll") LoadDLL(".\ort_dll\cufft64_11.dll") LoadDLL(".\ort_dll\cudnn_ops_infer64_8.dll") LoadDLL(".\ort_dll\cudnn_cnn_infer64_8.dll") LoadDLL(".\ort_dll\cudnn_adv_infer64_8.dll") LoadDLL(".\ort_dll\DirectML.dll") LoadDLL(".\ort_dll\onnxruntime.dll") LoadPlugin("mlrt_ort.dll")
Is there error if tilesize_...
is used?
Is there error if
tilesize_...
is used?
with runtimes files before, i can work with Cuda ver without tilesize by just putting all files into plugins64+
Yeah, same error with tilesize LoadPlugin("C:\Encode Tools\MeGUI-2944-64\tools\ffms\ffms2.dll") FFImageSource("C:\Users\ADMIN\Downloads\LCFome7.jpeg") z_ConvertFormat(pixel_type="rgbps", colorspace_op="709:709:709:l=>rgb:709:709:f") mlrt_ort(network_path="C:/Program Files (x86)/AviSynth+/plugins64+/models/RealESRGAN/RealESRGAN_x4plus_anime_6B.onnx", builtin=false, fp16=true, provider="dml", device=1, tilesize_w=width/2, tilesize_h=height/2) propSet("_FieldBased",0) deep_resize(1920,1080,grain=0) z_ConvertFormat(pixel_type="yuv420p8", colorspace_op="rgb:709:709:f=>709:709:709:l")
Try with edited mlrt_ort_loader.avsi
to
LoadDLL(".\ort_dll\DirectML.dll")
LoadDLL(".\ort_dll\onnxruntime.dll")
LoadPlugin("mlrt_ort.dll")
same error
I tried to delete mlrt_ov_loader.avsi and mlrt_ort_loader.avsi. Put all runtime files to plugins64+ and use onnxruntime_only_dml.zip, DML works with this script now (without tilesize) LoadPlugin("C:\Encode Tools\MeGUI-2944-64\tools\ffms\ffms2.dll") FFImageSource("C:\Users\ADMIN\Downloads\LCFome7.jpeg") z_ConvertFormat(pixel_type="rgbps", colorspace_op="709:709:709:l=>rgb:709:709:f") mlrt_ort(network_path="C:/Program Files (x86)/AviSynth+/plugins64+/models/RealESRGAN/RealESRGAN_x4plus_anime_6B.onnx", builtin=false, fp16=true, provider="dml", device=1) propSet("_FieldBased",0) deep_resize(1920,1080,grain=0) z_ConvertFormat(pixel_type="yuv420p8", colorspace_op="rgb:709:709:f=>709:709:709:l")
Next, I put the old onnx runtime back, and Cuda works now (deleted mlrt_ov_loader.avsi and mlrt_ort_loader.avsi. Put all runtime files to plugins64+). Openvino works with and without putting all runtime files in the plugins64+ (if without, need mlrt_ort_loader.avsi to load all runtime files)
If you have Windows shipped directml.dll
in C:\Windows\System32\directml.dll
can you rename that file to C:\Windows\System32\directml_.dll
and try onnxruntime.dll
from Release
(not onnxruntime_only_dml.zip)?
Edit: In Windows 11 check also for file C:\Windows\System32\onnxruntime.dll
. If there is such file renamed to C:\Windows\System32\onnxruntime.dll
and try onnxruntime.dll
from Release
(not onnxruntime_only_dml.zip).
If you have Windows shipped
directml.dll
inC:\Windows\System32\directml.dll
can you rename that file toC:\Windows\System32\directml_.dll
and tryonnxruntime.dll
fromRelease
(not onnxruntime_only_dml.zip)?Edit: In Windows 11 check also for file
C:\Windows\System32\onnxruntime.dll
. If there is such file renamed toC:\Windows\System32\onnxruntime.dll
and tryonnxruntime.dll
fromRelease
(not onnxruntime_only_dml.zip).
after changing those names, the error is still the same as DML if I don't use onnxruntime_only_dml.zip
Ok, thanks for the feedback.
Can you test the attached versions too?
Can you test the attached versions too?
What's m and n?
Both have the same error but the error call is from mlrt_ov, I don't know why it's mlrt_ov when I don't call this function (download the image to see cause the image has very many of text) My script: LoadPlugin("C:\Encode Tools\MeGUI-2944-64\tools\ffms\ffms2.dll") FFImageSource("C:\Users\ADMIN\Downloads\LCFome7.jpeg") z_ConvertFormat(pixel_type="rgbps", colorspace_op="709:709:709:l=>rgb:709:709:f") mlrt_ort(network_path="C:/Program Files (x86)/AviSynth+/plugins64+/models/RealESRGAN/RealESRGAN_x4plus_anime_6B.onnx", builtin=false, fp16=true, provider="dml", device=1) propSet("_FieldBased",0) deep_resize(1920,1080,grain=0) z_ConvertFormat(pixel_type="yuv420p8", colorspace_op="rgb:709:709:f=>709:709:709:l")
Filenames are just marks to know which file what build config has.
Can you test this? Also use the attached DirectML.dll with the attached onnxruntime.dll.
Filenames are just marks to know which file what build config has.
Can you test this? Also use the attached DirectML.dll with the attached onnxruntime.dll.
build.zip same error with those files and my script
Filenames are just marks to know which file what build config has. Can you test this? Also use the attached DirectML.dll with the attached onnxruntime.dll. build.zip same error with those files and my script
the same error I have
Can you test the attached versions?
still with the same error
Can you share short sample with the used script and model so I can reproduce the issue?
Sure.
SetMemoryMax(8192)
##Load source
DGSource("D:\Encodes\The Guts\01\01.dgi")
#Detint
AnimeIVTC(mode=1,ifade=true,nnedi3pel=true,stabilize=true,aapel=4,smooth=-1,tradius=3,ediandnn=3,degrain=3,threads=12)
#AnimeIVTC(mode=2,bbob=4,ifade=true,nnedi3pel=true,stabilize=true,aapel=4,smooth=-1,tradius=3,ediandnn=3,degrain=3,threads=12,cach
ConvertBits(32)
ConvertToPlanarRGB()
mlrt_ort(network_path="C:\Program Files (x86)\AviSynth+\plugins64\models\onnx\1x_Dotzilla_Compact_80k_net_g.onnx",builtin=false,provider="dml")
ConvertBits(8).ConvertToYV12()
#Crop(6,0,-6,0)
#Fixer(2,0,0,0,chroma=true)
blackmanresizeMT(1416,1080,taps=8,threads=12)
Thanks. What is the resolution of the clip?
Thanks. What is the resolution of the clip?
720x480
Can you test the following script:
BlankClip(pixel_type="rgbps") # If still error test with BlankClip(pixel_type="rgbps", width=240, height=120)
mlrt_ort(network_path="C:\Program Files (x86)\AviSynth+\plugins64\models\onnx\1x_Dotzilla_Compact_80k_net_g.onnx",builtin=false,provider="dml")
the tragedy continues my friend
Rename C:\Windows\System32\directml.dll
to C:\Windows\System32\directml_.dll
and try again.
now avspmod closes by itself
Use the attached mlrt_ort.dll
. Make sure DirectML.dll
and onnxruntime.dll
are in same folder.
Rename C:\Windows\System32\onnxruntime.dll
to C:\Windows\System32\onnxruntime_.dll
and try again.
same mistake
Remove mlrt_ort_loader.avsi
if you have it.
Remove all onnxruntime.dll
, DirectML.dll
, onnxruntime_xxx.dll
and all cuda runtime files from your plugins folder.
Put the attached DirectML.dll
and onnxruntime.dll
next to downloaded mlrt_ort.dll
test version.
it was fun
Thanks for the testing. What's the performance? Better than mlrt_ncnn
?
yes and a little faster also uses less vram
I saw now that you asked in the other closed topic, which windows do I use
I use Windows 11 latest version
These Windows versions have onnxruntime.dll
and DirectML.dll
in System32
and they can be loaded instead of the correct one.
@SangenBR, can you test the attached onnxruntime.dll
and DirectML.dll
? Just replace the existing ones with the attached version.
@SangenBR, can you test the attached
onnxruntime.dll
andDirectML.dll
? Just replace the existing ones with the attached version.
It doesn't work, the same error as above
Ok.
What works for me and this one from the attachment
I would say it is 10 to 20% compared to mlrt_ncnn
plugins64.zip
I try with parameter device=1 (cause with device=0, the provider=mlt always chooses my iGPU), my dGPU starts with using vram but i meet this error
My script: LoadPlugin("C:\Encode Tools\MeGUI-2944-64\tools\ffms\ffms2.dll") FFImageSource("C:\Users\ADMIN\Downloads\LCFome7.jpeg") z_ConvertFormat(pixel_type="rgbps", colorspace_op="709:709:709:l=>rgb:709:709:f") mlrt_ort(network_path="C:/Program Files (x86)/AviSynth+/plugins64+/models/RealESRGAN/RealESRGAN_x4plus_anime_6B.onnx", builtin=false, fp16=true, provider="dml", device=1) propSet("_FieldBased",0) deep_resize(1920,1080,grain=0) z_ConvertFormat(pixel_type="yuv420p8", colorspace_op="rgb:709:709:f=>709:709:709:l")
when i change to mlrt_ort(network_path="C:/Program Files (x86)/AviSynth+/plugins64+/models/RealESRGAN/RealESRGAN_x4plus_anime_6B.onnx", builtin=false, fp16=true, provider="dml", device=1, tilesize_w=width/2, tilesize_h=height/2), the script works so I think it relates my dedicated GPU memory not enough but I can launch the script even use shared GPU memory in Vapoursynth version