I checked Dennis M. Ritchie's "Users' Reference to B" and found an example of implementing a B program at the bottom of the manual. It said that bc generates intermediate code suitable for ba, and then ba generates assembly code. So, I am curious about what the intermediate code generated by bc is?

I found this reference (https://www.softwarepreservation.org/projects/BCPL/cambridge/Richards-Bootstrapping_BCPL-1973.pdf) written by Martin Richards about the intermediate code of BCPL. I never saw or used B, but this is one possibility.

But it seems very different from C compilers prior to the Portable C Compiler, even though code generation was table driven (https://wolfram.schneider.org/bsd/7thEdManVol2/ctour/ctour.html).

Note, before the RTS group at Project MAC started using UNIX, we had a home-written operating system for PDP-11/45 and PDP-11/70. I got a BCPL compiler from somewhere and made some enhancements - such as direct support for external variables and routines using a linker rather than the pure BCPL global array. When RTS got access to a VAX-11/780 running VMS, I was able to modify the Sixth or Seventh edition C compiler to generate code for the VAX-11/780 and wrote enough of a compatibility library to port various C programs to VMS. All that vanished once we were able to install 4BSD on the VAX-11/780. The machine was shared by the people doing the NIL project (New Implementation of LISP) on the VAX. I don't remember the details, but this paper coauthored by Guy Steele (https://www.dreamsongs.com/Files/HOPL2-Uncut.pdf) implies that NIL shifted their efforts to a different target machine: "In 1978, Gabriel and Guy Steele set out to implement NIL [Brooks, 1982a] on the S-1 Mark IIA, a supercomputer being designed and built by the Lawrence Livermore National Laboratory [Correll, 1979; Hailpern, 1979]."