nvm_dlaSample got failed

Autonomous Vehicles DRIVE AGX Orin DRIVE AGX Orin General
1

Please provide the following info (tick the boxes after creating this topic):
Software Version
DRIVE OS 6.0.8.1

Target Operating System
Linux

Hardware Platform
DRIVE AGX Orin Developer Kit (940-63710-0010-D00)

SDK Manager Version
other

Host Machine Version
native Ubuntu Linux 20.04 Host installed with DRIVE OS Docker Containers

Hello, I tried to run nvm_dlaSample

with using my own model but failed during NvMediaDlaAppendLoadable

.

wooseok@tegra-ubuntu:/opt/nvidia/drive-linux/samples/nvmedia/dla_sample$ ./nvm_dlaSample -v debug --mode runtime -d 0 -n 2 --loadable /usr/src/tensorrt/bin/20240717_dl10.9.0_v1.3.plan
NvMTest: >>>>>>>>>>>>>>>>>>>> Test Begin >>>>>>>>>>>>>>>>>>>> 

NvMTest: List of input arguments: 
NvMTest: Test runtime = 1 
NvMTest: Test scisync = 0 
NvMTest: Test ping = 0 
NvMTest: Test multithread = 0 
NvMTest: loadable name = /usr/src/tensorrt/bin/20240717_dl10.9.0_v1.3.plan 
NvMTest: dlaId = 0 
NvMTest: numTasks = 2 
NvMTest: verbose level = 4 
NvMTest: Test with dlaId = 0 numTasks = 2 profile name /usr/src/tensorrt/bin/20240717_dl10.9.0_v1.3.plan
NvMTest: NvSciBufModuleOpen Successful
NvMTest: NvMediaTensorNvSciBufInit Successful
NvMTest: InitNvSciBuf Successful
NvMTest: NvMediaDlaCreate Successful
NvMTest: Create Successful
NvMTest: NvMediaDlaInit Successful
NvMTest: Init Successful
NvMTest: Reading loadable file to memory = /usr/src/tensorrt/bin/20240717_dl10.9.0_v1.3.plan
NvMTest: Open filename = %s Successful
NvMTest: alloc buffer Successful
NvMTest: readFileToMemory Successful
NvMTest: NvMediaDlaLoadableCreate Successful
NvMTest: ERROR: NvMediaDlaAppendLoadable Fail
NvMTest: ERROR: AddLoadable Fail
NvMTest: ERROR: Test: Setup failure 
NvMTest: ERROR: Check loadable index argument Fail
NvMTest: NvMediaDlaDestroy Successful
NvMTest: >>>>>>>>>>>>>>>>>>>> Runtime Test >>>>>>>>>>>>>>>>>>>> 
NvMTest: >>>>>>>>>>>>>>>>>>>> Test Fail >>>>>>>>>>>>>>>>>>>> 
NvMTest: <<<<<<<<<<<<<<<<<<<< Test End <<<<<<<<<<<<<<<<<<<<<
3

Dear @wooseok.won,
Could you share the command used to build DLA loadable? Is it possible to share the details to repro?

5

Hello @SivaRamaKrishnaNV
Please find the attached log during the conversion from .onnx to .plan
trtexec_24Jul22.log (18.2 KB)

6

Dear @wooseok.won,
Could you check if all layers in model runs on DLA? Please check adding --buildDLAStandalone --fp16 --useDLACore=0 --inputIOFormats=fp16:chw16 --outputIOFormats=fp16:chw16

7

Hi @SivaRamaKrishnaNV
Please compare the attachments. trtexec with the option you gave didn’t work.
trtexec_24Jul25_trtexec_with_buildDLAStandalone.log (6.9 KB)
trtexec_24Jul25_trtexec_without_buildDLAStandalone.log (18.1 KB)
Is that the reason why NvMediaDlaAppendLoadable got failed during nvm_dlaSample?

8

Dear @wooseok.won,

Flatten_182' is not supported on DLA.

It appears the model has non DLA supported layers.

9

Dear @SivaRamaKrishnaNV,
There are two of usages in my model.
But isn’t it Relu, Flatten are the most common function?
Can you please suggest any alternative?

10

Hi @SivaRamaKrishnaNV,
Please give me some feedback.

11

Dear @wooseok.won,
Please check Deep-Learning-Accelerator-SW/operators/README.md at dla3.12.1 · NVIDIA/Deep-Learning-Accelerator-SW · GitHub if it helps. I see reLU is natively supported. Please check if flatten can be replaced with any of the supported layers so that accuracy is not effected in your model.

14

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.