[PATCH] D59620: [llvm-pdbutil] Add -type-ref-stats to help find unused type info

Nico Weber via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Fri Mar 22 11:50:07 PDT 2019


thakis added a comment.

Zach, are you saying this should be a postprocessing step since not writing dead information in the first place would be very slow?

In D59620#1437718 <https://reviews.llvm.org/D59620#1437718>, @aganea wrote:

> In D59620#1437234 <https://reviews.llvm.org/D59620#1437234>, @zturner wrote:
>
> > If we ever wanted more aggressive pruning of dead type information, it probably makes sense to do that as a post-processing step where we re-write the PDB, something like an `llvm-pdbutil gc` subcommand.
>
>
> @zturner You mean in terms of execution time, or in terms of code complexity in LLD?
>
> @rnk At a later stage, would it be possible to add an option to only print the referenced/unreferenced types? ie. `llvm-pdbutil dump -types=[all(default),ref,unref]`
>
> Just wanted to mention that we've just hit the 2 GB bar for PDBs in some of our production builds. So even if this patch provides a 20% reduction for PDBs, it would make a difference for us.
>  The other related issue is the symbol servers, which use CAB as a compressed file format. Unfortunately, CAB has a uncompressed file size limit of 2 GB, so we can't store them compressed anymore, which is a bit annoying to say the least.
>  I'll run your patch on some PDBs to see what numbers I come with.


We're at 1.5GB PDB size for chrome too (with split main dlls, something we might want to undo), so for Chromium PDB size reductions are useful too.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D59620/new/

https://reviews.llvm.org/D59620





More information about the llvm-commits mailing list