[Autotuning]"Cannot find config for target" occuring at compilation with autotuning log file

Questions
#1

Hello,

I have an log file of a model (which I did not autotune myself), when I compile the model applying it I get some warnings like

Cannot find config for target=llvm --system-lib -device=arm_cpu -target=arm -mcpu=cortex-a72 -mattr=+neon,+vfp4 -mfloat-abi=soft, workload=('conv2d', (1, 16, 1024, 1024, 'float32'), (64, 16, 3, 3, 'float32'), (1, 1), (1, 1), (1, 1), 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
Cannot find config for target=llvm --system-lib -device=arm_cpu -target=arm -mcpu=cortex-a72 -mattr=+neon,+vfp4 -mfloat-abi=soft, workload=('conv2d', (1, 16, 2048, 2048, 'float32'), (3, 16, 3, 3, 'float32'), (1, 1), (1, 1), (1, 1), 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.

Why do I still get fallback configurations when compiling after autotuning?
How can I get these tuned as well?

#2

Are these two tasks extracted while tuning?

#3

I guess so, although as I mentioned I did not do the autotuning myself.
I am a bit confused about what is a task now.

#4

I think the tasks have been extracted using autotvm.tasks.exctract_from_program with ops=(relay.op.nn/conv2d)
I can’t find it in the documentation though, only in the examples.

#5

Each task represents a workload of conv2d. You can print tsk.workload to see whether these two missing conv2d workloads extracted.