[convolutionRunner.cpp::executeConv::458] Error Code 1: Cudnn (CUDNN_STATUS_EXECUTION_FAILED)

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
1

• Hardware Platform (Jetson )
• DeepStream Version 6.0 Cuda 10.2
**• JetPack Version 4.6
**• TensorRT Version 8.0.1
**• Issue Type( bugs)

[convolutionRunner.cpp::executeConv::458] Error Code 1: Cudnn (CUDNN_STATUS_EXECUTION_FAILED)
ERROR: Infer context enqueue buffer failed, nvinfer error:NVDSINFER_TENSORRT_ERROR

According to my understanding, this error is because the versions of Cudnn and Cuda do not match each other. But it’s unreasonable that I installed jetpack 4.6 via SD card image and Cuda 10.2 and Cudnn 8.2.1 are already attached, so why does the above error occur?

4

Hi,

Could you run the cuDNN sample to make sure it can work correctly?
Thanks.

7

I can run successfully.
log.txt (4.4 KB)

8

Hi,

How do you install Deepstream?
And could you share which command or sample hits this issue?

Thanks.

9

The way i approach with deepstream.

Source:
note.txt (7.1 KB)
This was the error before I encountered the cuDNN error above. I tried to create an engine class to facilitate inference with my direction.
Error log at: in function NvDsInferContext_QueueInputBatch
ERROR: [TRT]: 1: [convolutionRunner.cpp::executeConv::458] Error Code 1: Cudnn (CUDNN_STATUS_EXECUTION_FAILED)
ERROR: Failed to enqueue trt inference batch

10

Can you support me about issue?

11

Hi,

Sorry, would you mind summarizing how you installed the Deepstream?
Do you use a custom version or the same package from the same JetPack?

Thanks.

12

Thanks for reply. I resolve problem with sample code in gstdexample in gst-plugins. For jetson the way get data for inference difference with X86.
ifdef aarch64
/* To use the converted buffer in CUDA, create an EGLImage and then use
* CUDA-EGL interop APIs */
if (USE_EGLIMAGE) {
if (NvBufSurfaceMapEglImage (dsexample->inter_buf, 0) !=0 ) {
goto error;
}

  /* dsexample->inter_buf->surfaceList[0].mappedAddr.eglImage
  * Use interop APIs cuGraphicsEGLRegisterImage and
  * cuGraphicsResourceGetMappedEglFrame to access the buffer in CUDA */

  /* Destroy the EGLImage */
  NvBufSurfaceUnMapEglImage (dsexample->inter_buf, 0);
}

endif

14

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