follow dev58, 1. the problem of convergence difficulty in single-precision processing of the MG algorithm was solved by using a simple method of layer truncation, referring to PyQCU/test/pyqcu/ascend/inverse.py:mg:adaptive; 2. independent settings for the precision of each layer and the device were implemented in the multi-threaded MG algorithm, referring to PyQCU/test/test.ascend.mg-dev59-np8.log; 3. the comparison with the wilson inverse results from quda was completed, referring to PyQCU/test/test.ascend.bistabcg-dev59.ipynb; however, the environment configuration for DCU and Ascend is not yet fully complete......