Re: [Tinycc-devel] TCC riscv32 port - Deferring code generation on external symbol

[grischka]

Sam Ellicott wrote:
> Example code is here
> ```
> .global _start
> _start:
> la gp, _global_pointer
> la a0, _bss_start
> la a1, _bss_end
> bgeu a0, a1, 2f
> 1: sw zero, (a0)
> addi a0, a0, 4
> bltu a0, a1, 1b
> 2: # setup the stack
> la sp, _stack_end
> call main
> ```
> For example when the `_global_pointer` symbol is encountered what is
> the correct behavior? Since the symbol is defined in a separate
> compilation unit, I'm assuming some sort of marker needs to be placed
> in the .o file to indicate what immediate value goes there at link
> time. However, I don't know what that is, or how it is generated.
> Similarly for `_bss_start`, `_bss_end`, etc.

That 'sort of marker' thing is called 'relocation' entry.

You would emit code as if it were  'la a0, 0' but put a relocation on
it that refers to the symbol.  (In tcc, using 'greloc[a]' for example).

Whether or not the symbol is defined in the current unit doesn't really
make a difference in that regard.

With riscv it's maybe two relocations as I've seen that 'la' really is
two instructions to first load the 20 HI-bits and then the 12 LO-bits.

-- gr

> Thanks!