brpc 在开启usercode_in_pthread性能比不开启性能好很多，想知道原因

怎么才能不开启--usercode_in_pthread，性能跟usercode_in_pthread一样，程序内部只有btread-tls，没有其他的

Start by reproducing the performance comparison with --usercode_in_pthread enabled and disabled in a program using only pthread TLS. Trace the relevant brpc threading and user-code execution paths to identify the cause of the difference; done means documenting the cause and a concrete way to achieve comparable performance without the flag.

Written by the indexing model from the issue text.