This option controls the sorting order of sections in large execution regions to minimize the distance between sections that call each other.
If the execution region contains more code than the range of a branch instruction then the linker switches to large region mode. In this mode the linker sorts according to the approximated average call depth of each section in ascending order. The linker might also place distribute veneers amongst the code sections to minimize the number of veneers.
Note
Large region mode can result in large changes to the layout of an image even when small changes are made to the input.
To disable large region mode and revert to lexical order,
use --no_largeregions. Section placement is then
predictable and image comparisons are more predictable. However
some branches might not reach the target causing the link step to
fail. If this happens you must place code/data sections explicitly
using an appropriate scatter file or write your own veneer.
Large region support enables:
average call depth sorting,
--sort=AvgCallDepthAPI sorting,
--apiveneer injection,
--veneerinject.
The following command lines are equivalent:
armlink --largeregions --no_api --no_veneerinject --sort=Lexical armlink --no_largeregions
The default is --no_largeregions. The linker
automatically switches to --largeregions if at
least one execution region contains more code than the smallest
inter-section branch. The smallest inter-section branch depends
on the code in the region and the target processor:
- 32Mb
Execution region contains only ARM instructions.
- 16Mb
Execution region contains Thumb instructions, 32-bit encoded Thumb instructions are supported.
- 4Mb
Execution region contains Thumb instructions, no 32-bit encoded Thumb instructions are supported.