pytorch / glow

Compiler for Neural Network hardware accelerators

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

[Bug] error on Split operation

luyaor opened this issue · comments

ONNX standard allows that Split operation is able to split a dimension to any subpart (as long as their sum is equal to the dimension value). But when compiles following models with ONNX, the compiler fails with errors.

image

Error message:

 model-compiler: /root/dev/glow/lib/Graph/Graph.cpp:1750: void glow::Function::createSplit(llvm::StringRef, glow::NodeValue, glow::unsigned_t, glow::unsigned_t, llvm::ArrayRef<dim_t>, std::vector<SliceNode *> &): Assertion `inDims[axis] % outputNum == 0 && "Dimension to split must be divisible by outputs number."\' failed.
./bin/model-compiler[0xcbde7f]
./bin/model-compiler[0xcbc2a2]
./bin/model-compiler[0xcbe288]
/lib/x86_64-linux-gnu/libpthread.so.0(+0x12980)[0x7fc96a277980]
/lib/x86_64-linux-gnu/libc.so.6(gsignal+0xc7)[0x7fc967d28e87]
/lib/x86_64-linux-gnu/libc.so.6(abort+0x141)[0x7fc967d2a7f1]
/lib/x86_64-linux-gnu/libc.so.6(+0x303fa)[0x7fc967d1a3fa]
/lib/x86_64-linux-gnu/libc.so.6(+0x30472)[0x7fc967d1a472]
./bin/model-compiler[0x5df414]
./bin/model-compiler[0xae2fb4]
./bin/model-compiler[0xaa7b3e]
./bin/model-compiler[0xa70cbb]
./bin/model-compiler[0xa94cae]
./bin/model-compiler[0x8fe46f]
./bin/model-compiler[0xa7712c]
./bin/model-compiler[0xa7cbfa]
./bin/model-compiler[0xa7db03]
./bin/model-compiler[0xa7d6fe]
./bin/model-compiler[0x4d010a]
./bin/model-compiler[0x521510]
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xe7)[0x7fc967d0bc87]
./bin/model-compiler[0x4cee5a]

The model file and run script are as follows:

g16.onnx.zip

./bin/model-compiler -model ./g3.onnx -emit-bundle ./mybundle --backend=CPU