Scor­pio News

  

July–September 1988, Volume 2, Issue 3











Page 14 of 39











(In this case, gbase() is a function returning a func_ptr_ptr whose value is stored three bytes after the address pointed to at location 6 (in C code this would be : *( cast(int_ptr)( *( cast(int_ptr)6 ) + 3) ). GET_DATA is a defined constant corresponding to the vector for a particular global function.)

This is all very well if you program in C, and providing you know how your particular compiler passes arguments ete. If your compiler uses a post- linker (for instance if it generates. .REL files), then it should be possible to write some machine code to do the linking. and link it with the compiler output. Once again it will probably be necessary to know how the compiler calls functions or procedures and how the parameters are passed. If the manual does not say (and most don’t) the best way to find ott is. to write a very simple program which calls a function with arguments, compile it; and dis-assemble. it (ZSID or GEMDEBUG are good tools for this, or MBIS if you:prefer to wade through a listing). Finding your bit of code among all the:runtime support can be a pain, but there-are ways around it. Compilers which don’t use a linker often tell you where the runtimes end. If not you can be fairly certain anyway that the compiled code is the last part of the object file. If your compiler uses .REL files, then you can instruct L80 to put your bit exactly where you want it by issuing /D: and /P: switches.

Most compilers pass arguments on the stack with the first argument being lowest on the stack (which is the highest address on the Z80). Sometimes the number of arguments will be pushed on just before the call. Some compilers pop off the arguments after the call, but some expect the routine that was called to do so before returning. Some compilers use special routines at the start and/or end of a function. For instance, HiSoft C starts every function with a call to 141h (or 13Eh) where IX is pushed and then loaded with SP before moving SP down to make room for local variables, and exits from functions by jumping to 17Bh (or 178h) which loads SP from IX, pops IX and removes DE bytes of parameter data from the stack. This sort of information is essential for writing machine code to link with compiled code.

If your compiler does not generate intermediate files (i.e. if it directly generates .COM files) then this linking process is not so easy. Some compilers allow small sections of machine code to be defined directly within.a function (e.g. HiSoft use inline(...), some have mini-assemblers etc.), and although this facility is not usually suitable for any significant code it-should be possible to encode a-call to. some previously loaded resident machine code. It is simplest if the resident code is at a fixed address, but above the BDOS vector is quite easy to use. There should obviously be a jump vector table at some easily found address. The relevant jump address can, of course, be found using high level code and stored in a static variable for the low level code to read. This overcomes one difficulty,.namely-finding the function parameters. There remains one problem, which ishowto return to the main code. It may be possible to push a return address and allow the compiled function to do the work, but this can involve some rather messy code. I prefer to put the


This is an OCR’d version of the scanned page and likely contains recognition errors.











Page 14 of 39