This repository was archived by the owner on Nov 17, 2023. It is now read-only.
This repository was archived by the owner on Nov 17, 2023. It is now read-only.
Race-condition and crash with SymbolBlock on GPU #18765
Description
Description
Severe Bug with nn.SymbolBlock when ctx=mx.gpu(0), cpu is OK.
Error Message
malloc or free or Segmentation fault error may appears randomly
/home/xxxxxx/anaconda3/envs/solo/lib/python3.7/site-packages/mxnet/gluon/block.py:1517: UserWarning: Cannot decide type for the following arguments. Consider providing them as input:
data: None
input_sym_arg_type = in_param.infer_type()[0]
[17:15:59] src/operator/nn/./cudnn/./cudnn_algoreg-inl.h:97: Running performance tests to find the best convolution algorithm, this can take a while... (set the environment variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to disable)
[(1, 256, 56, 56), (1, 512, 28, 28), (1, 1024, 14, 14), (1, 2048, 7, 7)]
malloc(): unsorted double linked list corrupted
[1] 87116 abort (core dumped) python symbolblockbug.py
/home/xxxxxx/anaconda3/envs/solo/lib/python3.7/site-packages/mxnet/gluon/block.py:1517: UserWarning: Cannot decide type for the following arguments. Consider providing them as input:
data: None
input_sym_arg_type = in_param.infer_type()[0]
[17:21:29] src/operator/nn/./cudnn/./cudnn_algoreg-inl.h:97: Running performance tests to find the best convolution algorithm, this can take a while... (set the environment variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to disable)
[(1, 256, 56, 56), (1, 512, 28, 28), (1, 1024, 14, 14), (1, 2048, 7, 7)]
Segmentation fault: 11
/home/xxxxxx/anaconda3/envs/solo/lib/python3.7/site-packages/mxnet/gluon/block.py:1517: UserWarning: Cannot decide type for the following arguments. Consider providing them as input:
data: None
input_sym_arg_type = in_param.infer_type()[0]
[17:23:24] src/operator/nn/./cudnn/./cudnn_algoreg-inl.h:97: Running performance tests to find the best convolution algorithm, this can take a while... (set the environment variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to disable)
[(1, 256, 56, 56), (1, 512, 28, 28), (1, 1024, 14, 14), (1, 2048, 7, 7)]
malloc_consolidate(): invalid chunk size
[1] 87701 abort (core dumped) python symbolblockbug.py
To Reproduce
import mxnet as mx
from mxnet import gluon
from mxnet.gluon import nn
import gluoncv as gcv
class NetEncoder(nn.SymbolBlock):
def __init__(self, **kwargs):
base_network = gcv.model_zoo.resnet50_v1(pretrained=False)
outputs = ['stage1_activation2', 'stage2_activation3', 'stage3_activation5',
'stage4_activation2']
inputs, outputs, params = gcv.nn.feature._parse_network(
base_network, outputs, ['data'], pretrained=False, ctx=mx.cpu(), **kwargs)
super(NetEncoder, self).__init__(outputs, inputs, params=params)
class Foo(nn.HybridBlock):
def __init__(self):
super(Foo, self).__init__()
self.features = NetEncoder()
def hybrid_forward(self, F, x):
y = self.features(x)
return y
a = mx.nd.random.uniform(shape=(1,3,224,224), ctx=mx.gpu(0))
f = Foo()
f.collect_params().initialize()
f.hybridize()
f.collect_params().reset_ctx(mx.gpu(0))
b = f(a)
print([x.shape for x in b])Environment
- mxnet_cu102-1.7.0b20200719-py2.py3-none-manylinux2014_x86_64
- mxnet 2.0 master in April