Describe the bug
When deploying an LPRNet model to CV181X (SG2002) using any quantization mode (F16/BF16/INT8), the compiler (tpuc-opt) crashes during the address assignment phase.
Environment
- TPU-MLIR Version: v1.27-20260206
- Chip: cv181x (SG2002 / MaixCam)
- OS: Docker sophgo/tpuc_dev:v3.1
- Python: 3.10
Error Message
terminate called after throwing an instance of 'std::out_of_range'
what(): vector::_M_range_check: __n (which is 65536) >= this->size() (which is 65536)
Steps to Reproduce
- Convert the attached ONNX to MLIR (Success).
- Run model_deploy:
model_deploy.py --mlir lprnet_no_pool.mlir --quantize F16 --chip cv181x --model test.cvimodel
Observations
The crash happens at the very last step: --address-assign. It seems like a counter or array index in the backend backend-compiler is overflowing (hitting the 2^16 limit).
Attached Files
error.log.txt
Describe the bug
When deploying an LPRNet model to CV181X (SG2002) using any quantization mode (F16/BF16/INT8), the compiler (tpuc-opt) crashes during the address assignment phase.
Environment
Error Message
terminate called after throwing an instance of 'std::out_of_range'
what(): vector::_M_range_check: __n (which is 65536) >= this->size() (which is 65536)
Steps to Reproduce
model_deploy.py --mlir lprnet_no_pool.mlir --quantize F16 --chip cv181x --model test.cvimodel
Observations
The crash happens at the very last step:
--address-assign. It seems like a counter or array index in the backend backend-compiler is overflowing (hitting the 2^16 limit).Attached Files
error.log.txt