Re: Two observations using GDB 13 snapshot

[Luis Machado <luis.machado@arm.com>]

On 12/27/22 18:00, Eli Zaretskii wrote:
>> From: Tom Tromey <tom@tromey.com>
>> Cc: Tom Tromey <tom@tromey.com>,  gdb-patches@sourceware.org,
>>    luis.machado@arm.com
>> Date: Tue, 27 Dec 2022 10:19:24 -0700
>>
>> Eli> All I'm interested in is to help you and others understand what could
>> Eli> be the reason for slow reading of debug info at startup, so that it
>> Eli> could be sped up at some point.
>>
>> Ok.  I re-read the thread and I see that at the beginning you didn't
>> mention -readnow.  So, assuming that the -readnow stuff was a
>> distraction...
> 
> Right.  When it was brought up, I thought -readnow was indicative of
> what I see, but then you said it doesn't use the new DWARF reader
> code, so it is not relevant.

Sorry I brought -readnow up. I wasn't aware that mode didn't take advantage of the new DWARF reader. So it ended up
causing some confusion.

> 
>> The new indexer should not be significantly slower even when
>> single-threaded.  On my machine, if I "maint set worker-threads 0" and
>> then use a -O2 build of gdb, it's about as fast as the system /bin/gdb
>> (which uses the old code).
>>
>> Old:
>>
>>      (gdb) file ./gdb
>>      [...]
>>      Command execution time: 2.408843 (cpu), 2.130013 (wall)
>>
>> New:
>>
>>      (gdb) file ./gdb
>>      [...]
>>      Command execution time: 2.858222 (cpu), 2.865426 (wall)
>>
>> This is only slightly worse.
> 
> My numbers are very different:
> 
>   . with GDB-12.1:                 elapsed: 8.39 sec,  CPU: 6.25 sec
>   . with gdb-13.0.50.20221217-git: elapsed: 135.8 sec, CPU: 130.5 sec
> 
> No significant difference in memory consumption: 453MB for GDB 12.1,
> 441MB for GDB 13.
> 
> (Interestingly, when reading symbols from Emacs on the same system,
> the times are almost identical: 2.593 for GDB 12 vs 2.671 for GDB 13.)
> 
> Other details which might be important:
> 
>    GCC 9.2.0
>    Binutils 2.39
>    GDB compilation switches: -O2 -gdwarf-4 -g3
> 
> And one more thing: this is a native 32-bit Windows build of GDB, so
> it has DWARF2 info in PE-COFF file, not in ELF.  Maybe this could
> explain the difference?  Or maybe the C++ code is a factor?
> 
> Thanks.