It seems, that if I try to run an INT8 graph that has tf.transpose op in it, I get seg

<div class="snippet-clipboard-content notranslate position-relative overflow-auto" data-snippet-clip

tf.transpose segfault about tim-vx HOT 4 CLOSED

verisilicon commented on August 29, 2024

tf.transpose segfault

from tim-vx.

Comments (4)

sunshinemyson commented on August 29, 2024 1

Thanks. I think i know where is the problem. Let me ask the developer to fix this issue.
Root cause: we didn't add layout inference for transpose op yet.

from tim-vx.

sunshinemyson commented on August 29, 2024

Could you share backtrace for the segfault?

from tim-vx.

bkovalenkocomp commented on August 29, 2024

...
2021-06-30 12:56:04.580070: I tensorflow/lite/delegates/vx-delegate/op_map.cc:405] Creating Dequantize op
2021-06-30 12:56:04.580137: I tensorflow/lite/delegates/vx-delegate/op_map.cc:405] Creating Dequantize op
2021-06-30 12:56:04.580203: I tensorflow/lite/delegates/vx-delegate/op_map.cc:405] Creating Dequantize op
2021-06-30 12:56:04.580270: I tensorflow/lite/delegates/vx-delegate/op_map.cc:405] Creating Dequantize op
2021-06-30 12:56:04.580335: I tensorflow/lite/delegates/vx-delegate/op_map.cc:405] Creating Dequantize op
2021-06-30 12:56:04.580402: I tensorflow/lite/delegates/vx-delegate/op_map.cc:405] Creating Dequantize op
2021-06-30 12:56:04.580431: I tensorflow/lite/delegates/vx-delegate/delegate_main.cc:570] Verifying graph
W [HandleLayoutInfer:257]Op 19: Default layout inference pass.
E [GetMapedTensor:140]Tensor has not beed inserted in tensor map.
E [GetMapedTensor:140]Tensor has not beed inserted in tensor map.
E [GetMapedTensor:140]Tensor has not beed inserted in tensor map.
E [GetMapedTensor:140]Tensor has not beed inserted in tensor map.
E [GetMapedTensor:140]Tensor has not beed inserted in tensor map.
E [GetMapedTensor:140]Tensor has not beed inserted in tensor map.
E [GetMapedTensor:140]Tensor has not beed inserted in tensor map.
E [GetMapedTensor:140]Tensor has not beed inserted in tensor map.
E [GetMapedTensor:140]Tensor has not beed inserted in tensor map.
E [GetMapedTensor:140]Tensor has not beed inserted in tensor map.
E [vsi_nn_SetupGraph:670]CHECK STATUS(-10:The supplied parameter information does not match the kernel contract.)
2021-06-30 12:56:04.588468: I tensorflow/lite/delegates/vx-delegate/delegate_main.cc:579] Verified graph
2021-06-30 12:56:04.588552: I tensorflow/lite/delegates/vx-delegate/delegate_main.cc:585] Copying input 595:data:0
2021-06-30 12:56:04.589129: I tensorflow/lite/delegates/vx-delegate/delegate_main.cc:598] Invoking graph
2021-06-30 12:56:04.589595: I tensorflow/lite/delegates/vx-delegate/delegate_main.cc:605] Copying output 596:transpose_6:0

Thread 1 "benchmark_model" received signal SIGSEGV, Segmentation fault.
0x000000000060efbc in ?? ()

here I have transpose at the beginning and at the end of the graph, to mimic NCHW

from tim-vx.

sunshinemyson commented on August 29, 2024

Hi，

We fixed issue with transpose in the layout inference framework. Please try with latest commit. Thanks

from tim-vx.

Recommend Projects