Re: MPS codegen

[Gerd Möllmann]

Gerd Möllmann <gerd.moellmann@gmail.com> writes:

> Eli Zaretskii <eliz@gnu.org> writes:
>
>>> From: Gerd Möllmann <gerd.moellmann@gmail.com>
>>> Cc: Eli Zaretskii <eliz@gnu.org>,  emacs-devel@gnu.org
>>> Date: Thu, 13 Jun 2024 21:37:41 +0200
>>> 
>>> Maybe one could reserve + commit a suitable block upfront, maybe as one
>>> big IGC_OBJ_PAD, and use read(2) or so to read the .pdmp into that
>>> block.
>>
>> Wouldn't that slow down startup?  pdumper uses memory-mapped files to
>> make that as fast as possible, and for a good reason: at the time
>> pdumper was introduced people objected to have startup slow down even
>> by a few hundreds of milliseconds.
>
> I don't know. Did pdumper do many reads that were replaced with one
> mmap? This one could do one large read instead of one mmap.

Found this benchmarking code:

  https://github.com/david-slatinek/c-read-vs.-mmap

It comes to the conclusion that, for 32MB files, mmap is 17% faster than
a sequence of reads of 16Kb each (2000 calls). Hm.