200dk 推理模型报 libgcc_s.so.1 must be installed for pthread_cancel to work

一、问题现象（附报错日志上下文）：
用msame或运行模型会报错: `libgcc_s.so.1 must be installed for pthread_cancel to work`
```
✗ ./msame --model ~/om/vol_generator.om --output out
[INFO] acl init success
[INFO] open device 0 success
[INFO] create context success
[INFO] create stream success
[INFO] get run mode success
[INFO] load model /home/HwHiAiUser/om/vol_generator.om success
[INFO] create model description success
[INFO] get input dynamic gear count success
[INFO] create model output success
out/2021122_8_28_18_743311
[ERROR] execute model failed, modelId is 1
Inference time: 62.802ms
[ERROR] model execute failed
[INFO] unload model success, model Id is 1
[INFO] destroy model input success
[ERROR] Sample process failed
[INFO] end to destroy stream
[INFO] end to destroy context
[INFO] end to reset device is 0
[INFO] end to finalize acl
libgcc_s.so.1 must be installed for pthread_cancel to work
[1]    6509 abort      ./msame --model ~/om/vol_generator.om --output out
```

二、软件版本:
 - CANN 版本:  5.0.4
 - Pytorch 版本: 1.9.1
 - 操作系统版本: Ubuntu 18.04

三、测试步骤：

转换onnx， 或直接下载 https://github.com/HardysJin/DROID-SLAM/releases/download/v0/vol_generator.onnx
```
import torch
import torch.nn as nn
import torch.nn.functional as F

class Volume_generator(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.pool = nn.AvgPool2d(2, stride=2)

def forward(self, fmap1, fmap2):
        # all pairs correlation
        batch, num, dim, ht, wd = fmap1.shape
        fmap1 = fmap1.view(batch*num, dim, ht*wd) / 4.0
        fmap2 = fmap2.view(batch*num, dim, ht*wd) / 4.0
        
        corr = torch.matmul(fmap1.transpose(1,2), fmap2).view(batch*num*ht*wd, 1, ht, wd)
        
        corr_pyramid = [corr]
        for i in range(3):
            corr_pyramid.append(F.avg_pool2d(corr_pyramid[-1], 2, stride=2))
        
        return corr_pyramid

x = torch.randn([1, 1, 128, 40, 72],dtype=torch.float32)
y = torch.randn([1, 1, 128, 40, 72],dtype=torch.float32)

torch.onnx.export(
    Volume_generator(),
    (x,y),
    'vol_generator.onnx',
    verbose=True,
    input_names=['fmap1', 'fmap2'],
    output_names=['vol1', 'vol2', 'vol3', 'vol4']
)
```

<hr>

atc转换命令，或直接下载 https://github.com/HardysJin/DROID-SLAM/releases/download/v0/vol_generator.om
```
atc  --input_shape="fmap1:1,1,128,40,72;fmap2:1,1,128,40,72" --input_fp16_nodes="fmap1;fmap2" --output="vol_generator" --soc_version=Ascend310 --framework=5 --model="vol_generator.onnx"
```

<hr>

msame测试
```
./msame --model ~/om/vol_generator.om --output out
```

Ascend/modelzoo
暂停

内容风险标识

评论 (17)

Ascend/modelzoo暂停 .gitee-modal { width: 500px !important; }

内容风险标识