Re: [PATCH 5/7] gdb: update blockvector::lookup to handle non-contiguous blocks

["Jan Vraný" <Jan.Vrany@labware.com>]

[-- Attachment #1: Type: text/plain, Size: 1060 bytes --]

On Fri, 2026-02-20 at 09:38 -0700, Tom Tromey wrote:
> > 
> 
> Jan> TBH I have no idea. How common is this, how many ranges blocks have, nor
> Jan> how to force compiler to produce such binary. Maybe torturing the compiler
> Jan> with -O3 and PGO?
> 
> Yeah, maybe just an optimized LTO build of gdb would show it.

I tried to build with PGO but no success so I resorted to LTO build
and wrote a script to get some data:

# CUs                             1263
# CUs with addr map               1075
# blocks                          654629
# blocks with multiple ranges:    270730

Max # blocks in linear phase:     42
Avg # blocks in linear phase:     2.372987707844722
Med # blocks in linear phase:     1

Here "blocks in linear phase" means number of blocks that start
at the same address, which is the worst case.

Assuming my thinking and script is correct, in little more than 50%
cases there's only one block in linear phase, in 17% cases there are
2 blocks, in 23% cases there are 3 to 5 blocks (histogram attached).

Jan

[-- Attachment #2: blocks_in_linear_phase_gdb_lto.png --]
[-- Type: image/png, Size: 23123 bytes --]