[PATCH] D113073: [lld-macho] Cache library paths from findLibrary

[Vy Nguyen via Phabricator via llvm-commits]

oontvoo added a comment.

In D113073#3120527 <https://reviews.llvm.org/D113073#3120527>, @smeenai wrote:

> 

.... [cropped]

>> 
>
> This is interesting. LLD has been at least 2x faster for us than ld64 (and also a bit faster than zld, which is a public ld64 fork with speed improvements), and I believe it was 3 to 4x faster for @thakis. You have some impressive ld64 improvements :)
>
> It's interesting that you have the large gap between parsing input files and the next trace. I haven't seen that on our end yet.

This is likely because for this particular app (actually also quite true for most of our apps), the input set is huge - it seems LLD pends most of its time blocking on disk-read.

By contrast,  ld64  parses all input files up front in parallel. (same for archives - it all loads all members that it knows will be needed thanks to -ObjC /-all_load/-force_load).

(at least, some good news is, as shown below,  LLD's cpu time is less than that of  LD64 ;)  - so it's efficient - just not as parallel as it could be)

  SYSTEM CPU time: 
  x ./LD64_res2.txt
  + ./LLD_released_res2.txt
      N           Min           Max        Median           Avg        Stddev
  x   5          1.84          2.01           1.9         1.916   0.065802736
  +   5          3.28          3.36          3.29         3.312   0.035637059
  Difference at 95.0% confidence
      1.396 +/- 0.0771735
      72.8601% +/- 4.02785%
      (Student's t, pooled s = 0.052915)
  USER CPU time: 
  x ./LD64_res2.txt
  + ./LLD_released_res2.txt
      N           Min           Max        Median           Avg        Stddev
  x   5         12.04         12.28         12.06        12.124    0.10644247
  +   5          5.51          5.66          5.54         5.562   0.060991803
  Difference at 95.0% confidence
      -6.562 +/- 0.126515
      -54.1241% +/- 1.04351%
      (Student's t, pooled s = 0.0867468)
  WALL time: 
  x ./LD64_res2.txt
  + ./LLD_released_res2.txt
      N           Min           Max        Median           Avg        Stddev
  x   5         12.42         12.59         12.54        12.516   0.064265076
  +   5         15.01         15.26         15.21        15.184    0.10139033
  Difference at 95.0% confidence
      2.668 +/- 0.123796
      21.3167% +/- 0.989101%
      (Student's t, pooled s = 0.0848823)



> I spent some time looking at the performance for loading inputs. Most of the time was going into parsing symbols and relocations, as expected.

do you have stats on how much time is spent in parsing  relative to reloc handling?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113073/new/

https://reviews.llvm.org/D113073