Atari BASIC - Advanced Techniques - Embedded Machine Language

String variables can hold any of the 256 characters available in the ATASCII character set and thus each byte of memory reserved for a string variable can hold any number from 0 to 255, including the characters 34₁₀ (22₁₆, "quote") and 155₁₀ (9B₁₆, "ENTER"), although these are tricky to type in. Short relocatable 6502 machine language routines can be converted to ATASCII characters and stored in the string variable. The machine language routine can be called as a function with the USR command specifying the address of the string variable as the location in memory to execute. For example, if the machine language code is stored in a string named ROUTINE$ it can be called with parameters as ANSWER=USR(ADR(ROUTINE$),VAR1,VAR2). Parameters are pushed onto the hardware stack (in Page 1) as 16-bit integers; variables are pushed as their addresses. The return value is in the A and X registers and interpreted by BASIC as a 16-bit integer; it cannot be pushed to the stack as there is no concept of a stack frame, and for the same reason there is no concept of a void return, but typically if the machine code subroutine does not return anything useful, the value – just whatever happens to be in those registers at the time – is just ignored.

These routines have to use relocatable machine code: that is to say, they cannot use instructions like JMP or JSR that use absolute addresses, except to well-known addresses in the OS and so forth; they can only use branch instructions such as BCC (branch if carry clear) which jump backwards or forwards by roughly 128₁₀ (80₁₆) because the strings could be moved in memory. For this reason page 6 (0600₁₆–06FF₁₆), a page of memory not used by BASIC or the operating system, is very popular for storing small routines; but of course one runs the danger that another routine may also wish to be stored there.

On the 6502, relocation is not trivial. These days we expect programs to sit pretty much anywhere in memory; the loader and processor collaborate to make that happen. But microprocessors of that era did not do that. The 6502 was especially hindered by having very few indirection instructions, and those it had were asymmetric: the X and Y registers indirect in different directions. This leads either to rather clumsy code that is forever moving stuff between registers, or clever but obtuse code that keeps them where they need to be even if it would seem more obvious to stick something else there. The 6502 instruction set is small enough that, over a short time, programmers can model the entire processor in their heads, even down to knowing how many cycles each instruction takes, and then start making clever tricks.

As well as using machine code for advanced functions, fairly trivial USR routines are sometimes used simply to gain access to functions in the Atari OS that have not been provided through Atari BASIC: for example block serialization to and from devices (Atari BASIC only lets it be done byte by byte, with GET and PUT, which takes far longer for just shuffling back and forth through the OS layers than actually writing the one byte of data), or for reading and writing blocks of memory (the PEEK and POKE commands were also unnecessarily slow because of the numeric problems described above).

Machine code can also be stored as numbers in DATA statements. After character strings, DATA statements are the next most efficient for storage since the data values are stored as a string of characters as they appear in the code. This method is sometimes used for very short routines where size isn't important but ease of use is (no special loaders or clever typing routines are required), or for one-off programs that then write out the resulting block of bytes (probably stored in a string) is written out as a program that can be read in later byte-for-byte.