[Lldb-commits] [PATCH] D62634: Improve DWARF parsing and accessing by 1% to 2%

Greg Clayton via Phabricator via lldb-commits lldb-commits at lists.llvm.org
Thu May 30 08:08:04 PDT 2019

clayborg added a comment.

In D62634#1522631 <https://reviews.llvm.org/D62634#1522631>, @labath wrote:

> BTW, how do you measure these things?
> When I tried to benchmark something I got a difference of at least 5% on what should be identical runs. I'd have to make hundreds of runs in order for a 1% difference in average to show up as statistically significant..

I quit all programs on my macOS machine and run the same script 5 times in a row. I get quite consistent results that are plus or minus 0.1 seconds usually. So I try to ensure nothing else is running. I have a dwarf script that loads 100 or so files with debug info (all .so files from a directory) and then sets a breakpoint by name and by file and line where the file is a header file, to ensure all line tables get checked.

  $ ~/Documents/src/lldb/svn/lldb/build/Release/lldb -o 'command script import dwarfperf.py' -o 'dwarfperf -d ~/Documents/143642775' -o q

This will load all files from "~/Documents/143642775" and then set breakpoints and it returns a time. These are all files with no accelerator tables to ensure the DWARF must be manually indexed.



More information about the lldb-commits mailing list