Open dal2 opened 7 years ago
wow, ok. Well... there are various possible ways forward, but I reckon the easiest might be to proceed by commenting out much of the kernel, then uncommenting bit by bit, until we find the line that crashes it.
The general principle is as follows:
COCL_DUMP_CL=1
set, like:
COCL_DUMP_CL=1 /Library/Frameworks/Python.framework/Versions/3.6/bin/python3.6 /Users/user/PycharmProjects/tensorflow-test/cnn.py
/tmp/0.cl
. (Check the contents of /tmp
, there might be more than one .cl
file, let me know about this, if it's the case, please, mauye show the directory listing)DUMP
to LOAD
, and check it still runs:
COCL_LOAD_CL=1 /Library/Frameworks/Python.framework/Versions/3.6/bin/python3.6 /Users/user/PycharmProjects/tensorflow-test/cnn.py
/tmp/0.cl
file/tmp/0.cl
file in a text editorkernel
at the start of the name, like:
kernel void some_really_long_name( ... ) {
... stuff here...
}
So, basically comment out everything inside the curly brackets of this kernel fucntions, and comment out all the other functions, and the structs. You comment out by putting //
at the start of each line. Then rerun, and hopefully it wont crash with the -45 error, though it might probably not do anything useufl, since the operation will be returning the wrong values. So, it might stil crash, but later, but thats ok.
Then, gradually uncomment stuff, and try to find out which line causes the crash above.
It might take a few hours to work through this process. There's an example of my following this process ofr a bug in clBLAS, here https://github.com/clMathLibraries/clBLAS/issues/108
https://github.com/hughperkins/TensorFlow-Examples/blob/as-unit-tests/examples/3_NeuralNetworks/dynamic_rnn.py