In days gone by, you had books for z80 and 6800 that gave the basic algorithms implemented in the particular processor.

If you can find one of the "Lance Levanthal" books, you will see the general techniques. You just have to adapt for the 8051 architecture. There are probably app notes from several 8051 manufacturers.

But the best method is to obtain the source code for a well-written 8051 application. I suspect the SDCC support library code is written in assembler. Or at worst disassemble an 8051 BASIC interpreter. This will have fairly size-efficient float routines.

And you can always just examine the ASM that a Compiler generates. But this model will probably be more suited to the needs of the compiler than raw efficiency.

David.