[Solved] tvm._ffi.base.TVMError

Questions
1

1，Cannot find config for target=llvm, workload=(‘conv2d’, (1, 3, 225, 225, ‘float32’), (32, 3, 3, 3, ‘float32’), (2, 2), (0, 0), (1, 1), ‘NCHW’, ‘float32’). A fallback configuration is used, which may bring great performance regression.
I thought it was a warning, but a warning message appeared below. It is not clear if this is related to the optimization required
WARNING:autotvm:Cannot find config for target=llvm, workload=(‘depthwise_conv2d_nchw’, (1, 1024, 9, 9, ‘float32’), (1024, 1, 3, 3, ‘float32’), (1, 1), (0, 0), (1, 1), ‘float32’). A fallback configuration is used, which may bring great performance regression.

2，tvm._ffi.base.TVMError: [12:16:31] /home/ding/tvm/src/runtime/graph/graph_runtime.h:327: Check failed: bitmask == 1|2|4|8|16 (3 vs. 31) invalid format

I guess is that this error caused the code to fail

so I need some help,Thanks

2

I don’t think these two messages are related. Cannot find config for target=llvm ... basically means TVM is trying to find an auto-tuned config for that workload, if the config does not exist, TVM will use default schedule.

3

I know why the first message comes up, is because the graph optimization Pass is 3 not 2；
But I don’t know the reason for the second message

4

I used the latest version fixed this error

5

I am using tvm through a docker container and it seems that it is already an updated version but I am still getting the following error:

tvm._ffi.base.TVMError: [12:16:31] /home/ding/tvm/src/runtime/graph/graph_runtime.h:327: Check failed: bitmask == 1|2|4|8|16 (3 vs. 31) invalid format

This error is exactly like your 2nd error @LanTn.

6

You can try git clone --recursive https://github.com/dmlc/tvm to re-pull the code. This error should not exist.

7

Hi @LanTn, Thanks. Now it is working.