[PATCH] D27960: Use exact vector capacities to store DWARF line tables

[Simon Que via Phabricator via llvm-commits]

sque added a comment.

In https://reviews.llvm.org/D27960#628350, @sque wrote:

> In https://reviews.llvm.org/D27960#628119, @mehdi_amini wrote:
>
> > Can you provide the instructions (and inputs if needed) to reproduce?
>
>
> I'm using non-public code, but if you have some binary that can read in debug info using DWARFContext, it would suffice. As for the input binary, I used a large 4 GB executable, but I see a similar difference between https://reviews.llvm.org/B1 and https://reviews.llvm.org/B2 with a 200 MB binary.


I decided to try this with llvm-symbolizer, built from the LLVM source repo. Max RSS in kB:
A: 343280
https://reviews.llvm.org/B1: 318584
https://reviews.llvm.org/B2: 318508

Nominal capacity reduction is 32705 kB, which is about the difference between A and https://reviews.llvm.org/B1/https://reviews.llvm.org/B2. There is no significant difference between https://reviews.llvm.org/B1 and https://reviews.llvm.org/B2.

I suspect the discrepancy we saw earlier between https://reviews.llvm.org/B1 and https://reviews.llvm.org/B2 was due to a quirk of the allocator we used. But what exactly that quirk is, I don't know.

I can try the dry run tomorrow.


https://reviews.llvm.org/D27960