In search of the original Fortran compiler (2017) - https://news.ycombinator.com/item?id=23497895 - June 2020 (3 comments)
A 63-pass FORTRAN compiler for the IBM 1401, illustrated - https://news.ycombinator.com/item?id=9289070 - March 2015 (8 comments)
However, there were earlier languages which were at least high-level enough to support algebraic notation. Perhaps the first the compiler written by Laning and Zierler for the Whirlwind at MIT: https://en.wikipedia.org/wiki/Laning_and_Zierler_system (which was thought for a while to have been an inspiration for Fortran -- but the head of the Fortran project, John Backus, later found notes from before Laning and Zierler published).
There was one Fortran compiler from that era for the 1401 that made more than 50 'passes' to minimize memory use.
Today 'N passes for large N' is popular for experimental compilers, not to save memory, but because the passes themselves are easy to write. (At resource expense, and the danger that users will never be able to understand error messages.)
PL/I and Algol were targeting the space that C took over but had good and bad ideas that were abandoned, misunderstood, forgotten, etc. For instance the first Algol had no definition of how to do I/O which sounds terrible at first -- except today a C prog might throw away the stdlib like arduino does.
I'm not a fan of the Dragon Book, but I also don't think it held back much (although I'm not an expert on its societal impact). The older editions had very little about optimisations at all.