In HLL, I have these to be parsed (front-end):


Then my back-end would generate:



If I use FASM as back-end compiler, then it is end of story. But mine is not, it will generate the final EXE by itself, by computing RVA, e.g. distance of nocond: from different JMP location, depending on how many bytes in between.

E.g.


That's JMP 103 because of adjacent memory bytes.

JMP 102 would actually points to RET itself. Sometimes different CPU instructions may have bigger size.

E.g


See, GOTO(JMP) need to skip 4 bytes for the label (104).
----
So how do I calculate the length of JMP (in bytes) for each of these conditional instruction? Do I need 2 pass, 3 pass,etc to first find out the size of code block in between as illustrated below?

e,g,


Good news is I planning to revive my Sambal Compiler to support more features....

You don't "need" to do multiple passes, but it often generates good solutions.

Otherwise, for single pass, you can keep a list of unknown targets and fill in the distance when you encounter the future address. As long as you always leave enough space in the originating jmp then it works. But if you try to reserve only a 2-byte jmp and the forward distance is more than 127 bytes then it fails, so you have to reserve a 5-byte jmp instead and sometimes have 3 extra bytes doing nothing of value.

For a multi-pass approach, I have written an article that explains what fasm does to generate correct relative values.