I have started using "armnn" library (neural network library for arm processors) I'm running a tflite model on an Arm Mali-T760 with gpu acceleration (with the cpuRef or cpuAcc optimization the network seems to run without any problems). I run the tflite network using "EnqueueWorkload" function, the function works and gives good results for a while but then gets stuck and its thread is blocked (like some kind of infinity loop or deadlock).
It gerts stuck on: output->Execute(); in LoadedNetwork.cpp file.
I've tried to run the "dmesg mali" command and when the thread gets stuck I get following results:
[ 50.671951] INFO: task mali_vinstr_ser:954 blocked for more than 10 seconds.
[ 50.672004] mali_vinstr_ser D ffffff8008084f98 0 954 2 0x00000000
[ 60.671738] INFO: task mali_vinstr_ser:954 blocked for more than 10 seconds.
[ 60.671796] mali_vinstr_ser D ffffff8008084f98 0 954 2 0x00000000
[ 70.671819] INFO: task mali_vinstr_ser:954 blocked for more than 10 seconds.
[ 70.671877] mali_vinstr_ser D ffffff8008084f98 0 954 2 0x00000000
[ 80.671713] INFO: task mali_vinstr_ser:954 blocked for more than 10 seconds.
[ 80.671763] mali_vinstr_ser D ffffff8008084f98 0 954 2 0x00000000
[ 90.671771] INFO: task mali_vinstr_ser:954 blocked for more than 10 seconds.
[ 90.671823] mali_vinstr_ser D ffffff8008084f98 0 954 2 0x00000000
[ 100.671744] INFO: task mali_vinstr_ser:954 blocked for more than 10 seconds.
[ 100.671805] mali_vinstr_ser D ffffff8008084f98 0 954 2 0x00000000
[ 110.671714] INFO: task mali_vinstr_ser:954 blocked for more than 10 seconds.
[ 110.671785] mali_vinstr_ser D ffffff8008084f98 0 954 2 0x00000000
[ 120.671815] INFO: task mali_vinstr_ser:954 blocked for more than 10 seconds.
[ 120.671874] mali_vinstr_ser D ffffff8008084f98 0 954 2 0x00000000
[ 130.671732] INFO: task mali_vinstr_ser:954 blocked for more than 10 seconds.
[ 130.671795] mali_vinstr_ser D ffffff8008084f98 0 954 2 0x00000000
[ 140.671715] INFO: task mali_vinstr_ser:954 blocked for more than 10 seconds.
[ 140.671770] mali_vinstr_ser D ffffff8008084f98 0 954 2 0x00000000
But when the thread runs normally I get no results running this command.
I'm looking for any hints about the problematic layer\or poblematic inputs I'm giving to the network.
Thanks in advance, Dor
User contributions licensed under CC BY-SA 3.0