Hi,I’m trying External Tensor Functions on my model,and want to use cuDNN function for some convolutions. I use the ‘tvm.contrib.cudnn.conv_forward()’ but encounter an error: Check failed: e == CUDNN_STATUS_SUCCESS (3 vs. 0) : cuDNN: CUDNN_STATUS_BAD_PARAM. Is there something wrong with the cuDNN of jetson nano? Or something wrong in tvm.contrib?

CUDA version:10.2.3 cuDNN version:8.2.1 Here is my code:

import tvm
from tvm import te
import numpy as np
from tvm.contrib import cudnn
import numpy as np
data_np = np.random.uniform(size= (1,64, 128, 128)).astype(np.float32)
weight_np = np.random.uniform(size=(64, 64 , 3, 3)).astype(np.float32)
dev = tvm.cuda()
data_tvm = tvm.nd.array(data_np, device=dev)
weight_tvm = tvm.nd.array(weight_np, device=dev)
X = te.placeholder((1,64, 128, 128),dtype='float32', name='X')
W = te.placeholder((64,64, 3, 3),dtype='float32', name='W')
Y = tvm.contrib.cudnn.conv_forward(X, W, pad=0, stride=1, dilation=0, conv_mode=0, tensor_format=0, algo=-1, conv_dtype=None, groups=1)

Thank you!