Failed to convert onnx to trt with trtexec

Autonomous Machines Jetson & Embedded Systems Jetson Orin NX
1

command:
/usr/src/tensorrt/bin/trtexec --onnx=stereo.onnx --workspace=8000 --saveEngine=depth.trt --fp16

part of error log:

[06/21/2023-16:14:40] [V] [TRT] Conv_223 + PWN(Clip_226) Set Tactic Name: sm80_xmma_fprop_implicit_gemm_indexed_f32f32_f32f32_f32_nchwkcrs_nchw_tilesize128x32x8_stage3_warpsize2x2x1_g1_ffma_aligna4_alignc4 Tact\
ic: 0xa9366041633a5135
[06/21/2023-16:14:40] [V] [TRT] Tactic: 0xa9366041633a5135 Time: 0.324192
[06/21/2023-16:14:40] [V] [TRT] Conv_223 + PWN(Clip_226) Set Tactic Name: sm80_xmma_fprop_implicit_gemm_f32f32_f32f32_f32_nchwkcrs_nchw_tilesize64x64x8_stage3_warpsize1x4x1_g1_ffma_t1r3s3_aligna4_alignc4 Tactic\
: 0x4727434768e46395
[06/21/2023-16:14:40] [V] [TRT] Tactic: 0x4727434768e46395 Time: 0.331346
[06/21/2023-16:14:40] [V] [TRT] Conv_223 + PWN(Clip_226) Set Tactic Name: sm80_xmma_fprop_implicit_gemm_f32f32_f32f32_f32_nchwkcrs_nchw_tilesize128x64x8_stage3_warpsize2x2x1_g1_ffma_aligna4_alignc4 Tactic: 0x12\
dbf7d94ee3696d
[06/21/2023-16:14:40] [V] [TRT] Tactic: 0x12dbf7d94ee3696d Time: 0.322405
[06/21/2023-16:14:40] [V] [TRT] Fastest Tactic: 0xa8609adc4e0ceb90 Time: 0.224914
[06/21/2023-16:14:40] [V] [TRT] --------------- Timing Runner: Conv_223 + PWN(Clip_226) (CaskFlattenConvolution)
[06/21/2023-16:14:40] [V] [TRT] CaskFlattenConvolution has no valid tactics for this config, skipping
[06/21/2023-16:14:40] [V] [TRT] >>>>>>>>>>>>>>> Chose Runner Type: CaskConvolution Tactic: 0xa8609adc4e0ceb90
[06/21/2023-16:14:40] [V] [TRT] *************** Autotuning format combination: Float(2621440,1,16384,64) long-strided -> Float(327680,1,4096,32) ***************
[06/21/2023-16:14:41] [V] [TRT] --------------- Timing Runner: Conv_223 + PWN(Clip_226) (CaskConvolution)
[06/21/2023-16:14:41] [V] [TRT] Conv_223 + PWN(Clip_226) Set Tactic Name: ampere_scudnn_128x128_relu_exp_medium_nhwc_tn_v1 Tactic: 0xd9031472c05adf51
[06/21/2023-16:14:41] [V] [TRT] Tactic: 0xd9031472c05adf51 Time: 0.519849
[06/21/2023-16:14:41] [V] [TRT] Conv_223 + PWN(Clip_226) Set Tactic Name: ampere_scudnn_128x64_sliced1x2_ldg4_relu_exp_small_nhwc_tn_v1 Tactic: 0x27b316f52c109002
[06/21/2023-16:14:41] [E] Error[1]: [resizingAllocator.cpp::deallocate::105] Error Code 1: Cuda Runtime (misaligned address)
[06/21/2023-16:14:41] [E] Error[1]: [resizingAllocator.cpp::deallocate::105] Error Code 1: Cuda Runtime (misaligned address)
[06/21/2023-16:14:41] [E] Error[1]: [resizingAllocator.cpp::deallocate::105] Error Code 1: Cuda Runtime (misaligned address)
[06/21/2023-16:14:41] [W] [TRT] Skipping tactic 0x27b316f52c109002 due to exception misaligned address
[06/21/2023-16:14:41] [W] [TRT] Skipping tactic 0x412c44dfeaf9161d due to exception misaligned address
[06/21/2023-16:14:41] [W] [TRT] Skipping tactic 0x3e2b881168d9689d due to exception misaligned address
[06/21/2023-16:14:41] [W] [TRT] Skipping tactic 0xbdfdef6b84f7ccc9 due to exception misaligned address
[06/21/2023-16:14:41] [W] [TRT] Skipping tactic 0xf90060ce8193b811 due to exception misaligned address
2

Could anyone give me some advice?

3

I was emcounter this error too. when I downgrade tensorrt version 8.5 to 8.2 it solved.
Maybe tensorrt version was not fit with cuda version.

5

Hi,

Suppose you are using JetPack 5.1.1, is that correct?

We need more information to know more about this issue.
Could you share the model with us?

Thanks.

6

Yes.
onnx model file has 30M, how can I share with you？

7

Hi,

Could you share it through some online drive?

Thanks.

8

https://pan.baidu.com/s/12itnQ5-Mh-eTP5FuCrBhtg
extraction code: hpia

9

Can you download the model ?

10

Hi,

Unfortunately, no.

It requires the account for downloading.
Could you find a DRIVE that can download directly?

Thanks.

11

It’s simple to request an account. Or is there any cloud drive without account?

14

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