As far as I understand, the motivation for LTO framework was to enable cross-file interprocedural optimizations, and for this purpose an ipa pass is divided into following three stages:
LTO byte code: The LTO object file is a regular elf file with sections containing LTO byte-code. A LTO object file contains various sections for storing command line options, symbol table, global declarations and types, function bodies in GIMPLE, ipa pass summaries, ipa references, static variable initializers and the call graph. There are couple of limitations of the byte code format: 1] It is not self descriptive, which makes it harder to debug. 2] The byte code is essentially a “serialized” version of in-memory representations, which makes it prone to break across versions.
The purpose of this project is to create a dump tool for easily analyzing LTO object files similar to readelf or objdump -d for regular ELF object files.