public inbox for gdb-patches@sourceware.org
 help / color / mirror / Atom feed
From: Tom Tromey <tom@tromey.com>
To: gdb-patches@sourceware.org
Subject: [PATCH 00/30] Rewrite the DWARF "partial" reader
Date: Wed, 25 Aug 2021 20:19:07 -0600	[thread overview]
Message-ID: <20210826021937.1490292-1-tom@tromey.com> (raw)

This series removes the DWARF partial symbol reader, replacing it with
a new indexer which is significantly -- about an order of magnitude --
faster.

The new indexer is roughly based on the DWARF .debug_names approach.
That is, rather than construct full names for all interesting DIEs,
it instead records the name given in the DWARF.

Note, though, that this series does not actually update gdb's
.debug_names writer to be correct.  The new index is constructed so
that this should be easy to do, but the series is long enough, and
complicated enough, that I did not want to add this on.  (It's not
enough to update the writer -- the reader must also be updated, and
one must decide how much backward compatibility to keep.)

Why write a new indexer, rather than simply fix the psymtab reader?  I
have tried that multiple times over the years, without success.  I
believe that parallelism is the key to increased performance, and this
is just too hard to retrofit onto the existing reader.

This series starts with a number of refactorings.  I've already landed
a number of these, but the ones that remain are specific to this
series, so I didn't try to submit them separately.

The new code is also broken into separate patches for simpler review.

Some of the performance gains are actually just an illusion -- some
work is moved to the background, and if the main part of gdb needs the
information, it will once again appear to be slow.  However, this is
still an improvement for some common situations, for example simply
starting gdb on a large program.  My preferred long-term fix here
would be to replace DWARF with something that's less terrible for
reading; but failing that we can maybe eke out some small incremental
gains through further micro-optimization, or by farming out this
scanning to something like debuginfod.

I've regression tested each patch in this series.  Only the patch to
enable the new indexer is intended to affect operation at all.

Tom



             reply	other threads:[~2021-08-26  2:19 UTC|newest]

Thread overview: 48+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-08-26  2:19 Tom Tromey [this message]
2021-08-26  2:19 ` [PATCH 01/30] Introduce make_unique_xstrndup Tom Tromey
2021-08-26  2:19 ` [PATCH 02/30] Split create_addrmap_from_aranges Tom Tromey
2021-08-26  2:19 ` [PATCH 03/30] Add dwarf2_per_cu_data::addresses_seen Tom Tromey
2021-08-26  2:19 ` [PATCH 04/30] Refactor dwarf2_get_pc_bounds Tom Tromey
2021-08-26  2:19 ` [PATCH 05/30] Allow ada_decode not to decode operators Tom Tromey
2021-08-26  2:19 ` [PATCH 06/30] Let skip_one_die not skip children Tom Tromey
2021-08-26  2:19 ` [PATCH 07/30] Add name splitting Tom Tromey
2021-08-26  2:19 ` [PATCH 08/30] Add new overload of dwarf5_djb_hash Tom Tromey
2021-08-26  2:19 ` [PATCH 09/30] Refactor build_type_psymtabs_reader Tom Tromey
2021-08-26  2:19 ` [PATCH 10/30] Add batching parameter to parallel_for_each Tom Tromey
2021-08-26  2:19 ` [PATCH 11/30] Return vector of results from parallel_for_each Tom Tromey
2021-08-27  6:20   ` Tom de Vries
2021-08-28 19:20     ` Tom Tromey
2021-08-26  2:19 ` [PATCH 12/30] Introduce DWARF abbrev cache Tom Tromey
2021-08-26  2:19 ` [PATCH 13/30] Statically examine abbrev properties Tom Tromey
2021-09-06 22:31   ` Lancelot SIX
2021-11-04 18:00     ` Tom Tromey
2021-08-26  2:19 ` [PATCH 14/30] Update skip_one_die for new " Tom Tromey
2021-08-26  2:19 ` [PATCH 15/30] Introduce the new DWARF index class Tom Tromey
2021-09-09 23:32   ` Lancelot SIX
2021-11-04 18:03     ` Tom Tromey
2021-08-26  2:19 ` [PATCH 16/30] The new DWARF indexer Tom Tromey
2021-08-26  2:19 ` [PATCH 17/30] Implement quick_symbol_functions for cooked DWARF index Tom Tromey
2021-08-26  2:19 ` [PATCH 18/30] Wire in the new DWARF indexer Tom Tromey
2021-08-26  2:19 ` [PATCH 19/30] Pre-read DWARF section data Tom Tromey
2021-08-26  2:19 ` [PATCH 20/30] Parallelize DWARF indexing Tom Tromey
2021-08-26  2:19 ` [PATCH 21/30] "Finalize" the DWARF index in the background Tom Tromey
2021-08-26  2:19 ` [PATCH 22/30] Rename write_psymtabs_to_index Tom Tromey
2021-08-26  2:19 ` [PATCH 23/30] Change the key type in psym_index_map Tom Tromey
2021-08-26  2:19 ` [PATCH 24/30] Change parameters to write_address_map Tom Tromey
2021-08-26  2:19 ` [PATCH 25/30] Genericize addrmap handling in the DWARF index writer Tom Tromey
2021-08-26  2:19 ` [PATCH 26/30] Adapt .gdb_index writer to new DWARF scanner Tom Tromey
2021-08-26  2:19 ` [PATCH 27/30] Adapt .debug_names " Tom Tromey
2021-08-26  2:19 ` [PATCH 28/30] Enable the new DWARF indexer Tom Tromey
2021-08-26  2:19 ` [PATCH 29/30] Delete DWARF psymtab code Tom Tromey
2021-08-26  2:19 ` [PATCH 30/30] Remove dwarf2_per_cu_data::v Tom Tromey
2021-08-26 20:32 ` [PATCH 00/30] Rewrite the DWARF "partial" reader Tom de Vries
2021-08-26 21:29   ` Tom Tromey
2021-08-27  7:31     ` Tom de Vries
2021-08-30 15:04       ` Tom Tromey
2021-09-06 19:46         ` Tom Tromey
2021-09-07 10:58           ` Tom de Vries
2021-09-07 12:16             ` Tom de Vries
2021-10-29 23:06               ` Tom Tromey
2021-09-09 19:00 ` Wei-min Pan
2021-09-11 21:08   ` Tom Tromey
2021-09-13 16:50     ` Weimin Pan

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20210826021937.1490292-1-tom@tromey.com \
    --to=tom@tromey.com \
    --cc=gdb-patches@sourceware.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).