[PATCH] D49493: [DebugInfo] Reduce debug_str_offsets section size

[David Blaikie via Phabricator via llvm-commits]

dblaikie added a subscriber: labath.
dblaikie added a comment.

Anyone thought about whether it'd be worth making the indexed V unindexed
case work implicitly? If we were willing to add another pointer to
DwarfStringPoolEntryRef (doubling its size, admittedly) - back to the
string pool itself - then we could index any string where the
DwarfStringPoolEntryRef was asked for an index, and otherwise skip that?
Rather than having to declare up-front whether an index was required.

(if we wanted to get super-fancy, maybe we could emit indexes as assembly
expressions ((str_label/str_begin_label)/word_size) then you wouldn't have
to increase the size of DwarfStringPoolEntryRef, because the absolute index
wouldn't have to be known ahead of time - it'd be computed by the assembler
instead - which is just moving the work around, of course)


Repository:
  rL LLVM

https://reviews.llvm.org/D49493