floating point exception while convert onnx -> tensorrt model for ppocv3_rec model!

[Cute-Yang]

请提供下述完整信息以便快速定位问题/Please provide the following information to quickly locate the problem

• 系统环境/System Environment：centos7 /tensorrt 8.2 cuda11.0 cudnn8.1
• 版本号/Version：Paddle：2.6 PaddleOC: 问题相关组件/Related components：
• 运行指令/Command Code：trtexec --onnx=model.onnx --saveEngine=model.plan --workspace=1024 --verbose
• model.onnx是使用paddle2onnx转换得到的序列化文件
• 完整报错/Complete Error Message： [11/04/2022-17:46:56] [V] [TRT] p2o.Conv.30 + p2o.BatchNormalization.26 Set Tactic Name: volta_scudnn_128x32_sliced1x4_ldg4_relu_exp_medium_nhwc_tn_v1 Tactic: -423878181466897819 [11/04/2022-17:46:56] [V] [TRT] Tactic: -423878181466897819 Time: 0.03336 [11/04/2022-17:46:56] [V] [TRT] Fastest Tactic: -8010679767156598961 Time: 0.033088 [11/04/2022-17:46:56] [V] [TRT] >>>>>>>>>>>>>>> Chose Runner Type: CaskConvolution Tactic: -8010679767156598961 [11/04/2022-17:46:56] [V] [TRT] =============== Computing costs for [11/04/2022-17:46:56] [V] [TRT] Autotuning format combination: Float(1536,3,3,1) -> Float(1536,3,1,1) [11/04/2022-17:46:56] [V] [TRT] --------------- Timing Runner: shuffle_after_batch_norm_53.tmp_3 (Shuffle) [11/04/2022-17:46:56] [V] [TRT] Tactic: 0 Time: 0.004488 [11/04/2022-17:46:56] [V] [TRT] Tactic: 1 Time: 0.037028 [11/04/2022-17:46:56] [V] [TRT] Fastest Tactic: 0 Time: 0.004488 [11/04/2022-17:46:56] [V] [TRT] >>>>>>>>>>>>>>> Chose Runner Type: Shuffle Tactic: 0 [11/04/2022-17:46:56] [V] [TRT] Autotuning format combination: Float(1536,1,1536,512) -> Float(1536,1,512,512) [11/04/2022-17:46:56] [V] [TRT] --------------- Timing Runner: shuffle_after_batch_norm_53.tmp_3 (Shuffle) [11/04/2022-17:46:56] [V] [TRT] Tactic: 0 Time: 0.005064 [11/04/2022-17:46:56] [V] [TRT] Tactic: 1 Time: 0.014648 [11/04/2022-17:46:56] [V] [TRT] Fastest Tactic: 0 Time: 0.005064 [11/04/2022-17:46:56] [V] [TRT] >>>>>>>>>>>>>>> Chose Runner Type: Shuffle Tactic: 0 [11/04/2022-17:46:56] [V] [TRT] Autotuning format combination: Float(48,3:32,3,1) -> Float(48,3:32,1,1) [11/04/2022-17:46:56] [V] [TRT] --------------- Timing Runner: shuffle_after_batch_norm_53.tmp_3 (Shuffle) [11/04/2022-17:46:56] [V] [TRT] Tactic: 0 Time: 0.004496 [11/04/2022-17:46:56] [V] [TRT] Tactic: 1 Time: 0.014648 [11/04/2022-17:46:56] [V] [TRT] Fastest Tactic: 0 Time: 0.004496 [11/04/2022-17:46:56] [V] [TRT] >>>>>>>>>>>>>>> Chose Runner Type: Shuffle Tactic: 0 [11/04/2022-17:46:56] [V] [TRT] =============== Computing costs for [11/04/2022-17:46:56] [V] [TRT] Autotuning format combination: Float(1536,3,1,1) -> Float(1536,3,1,1) [11/04/2022-17:46:56] [V] [TRT] Autotuning format combination: Float(1536,1,512,512) -> Float(1536,1,512,512) [11/04/2022-17:46:56] [V] [TRT] Autotuning format combination: Float(48,3:32,1,1) -> Float(48,3:32,1,1) [11/04/2022-17:46:56] [V] [TRT] Autotuning format combination: Float(1:4,3,1,1) -> Float(1:4,3,1,1) [11/04/2022-17:46:56] [V] [TRT] =============== Computing costs for [11/04/2022-17:46:56] [V] [TRT] Autotuning format combination: Float(1536,3,1,1) -> Float(0,0,0,1) [1] 3291024 floating point exception (core dumped) /data/home/sunyyao/lazydog_build/TensorRT-8.2.1.8/bin/trtexec

[QianLingjun]

请问你解决了吗？我是onnx转rknn时候出现这个错误，onnx验证是可以直接进行预测的，没有问题。转模型时候，出现问题了。

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.

[Cute-Yang]

• floating point exception 这个识别阶段的模型具有三个动态维度，batch_size,heigh,width，指定一个动态范围之后，就可以转成功了，后面我是直接使用tensorRT的api搭建模型的，就不会出现这种奇怪的问题了，你看下你那个是否也需要指定动态范围