On Tue, 2021-06-22 at 15:44 -0400, Paul Koning via cctalk wrote: > The Electrologica ALGOL compilers used somewhat similar mixtures of > pseudocode and machine code.
The IBM 1401 Fortran compiler 1401-FO-050 (subset of Fortran II) generated machine code for integer operations, and bytecode for floating point. The machine didn't have any floating-point hardware. The bytecode was smaller than subroutine calls would have been. Floating-point arithmetic was rather well done. Up to 20 decimal digits, with two guard digits. Only significant defect was that rounding wasn't symmetric. Gary Mokotoff did a very good job. I have the source code if anybody wants it.
