[llvm-dev] [LLD] Linker Relaxation

[PkmX via llvm-dev]

Currently we are thinking about adding relaxation to lld as a two-pass
algorithm:

1. First we iterate over each relaxable section and present each relax
relocation to Target, which can perform modification to the byte
content. To support efficient deletion, each RelaxableSection contains
an iovec-style array of (ptr, len) tuples to keep track each segment
we want to keep, and we also need maps between original input address
/ final address after relaxation to speed up calculation.
2. Calculate updated address for each symbol / relocation taking
deleted bytes into account.

I think both of these steps can be paralleled on a per-section basis.
We loop over these two steps until no more relaxation can be performed
or the desired number of iterations is met, then in `writeTo` of each
RelaxableSection we just write out each [ptr, ptr + len) range of
bytes and perform relocations. This should ensure minimal copying of
the section contents is done.

On Wed, Jul 12, 2017 at 2:27 AM, Rui Ueyama via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
>
> By the way, since this is an optional code relaxation, we can think about it later. The first thing I would do is to add RISC-V support to lld without code shrinking relaxations, which I believe is doable by at most a few hundreds lines of code.
>

Yes. However given that LLVM currently only have very incomplete MC
support for RISC-V, what would be the way to add tests to lld which
seems depend on having a working `llvm-mc`? I suppose calling
binutils's as is out of the question.

Chih-Mao Chen (PkmX)
Software R&D, Andes Technology