Jetson Xavier Inference Crashes on TRT8 about neo-ai-dlr

[mycpuorg]

Recommendation provided in the Issue #409 builds DLR successfully for TRT8 however, using a model compiled with Neo for a previous revision of TRT (JetPack4.4) crashes with a SEGFAULT during Inference.

Hi @mycpuorg
What is the model you were testing?

I rebuilt DLR with release-1.10.0 branch and replaced libdlr.so in the package from Neo service.
The model is able to inference with no error, the test script I used.

import dlr
import numpy as np
import os
import time

os.environ["TVM_TENSORRT_CACHE_DIR"] = "."
m=dlr.DLRModel("./compiled_model", "gpu", 0)

x = np.random.rand(1, 3, 224, 224)

for _ in range(100):
    y = m.run(x)

start = time.time()
iterations = 100
for _ in range(iterations):
    y = m.run(x)
elapse = time.time() - start
print("m elapse: ", elapse/iterations)
# m elapse:  0.003999626636505127