public inbox for gdb-patches@sourceware.org
 help / color / mirror / Atom feed
From: Tom Tromey <tom@tromey.com>
To: gdb-patches@sourceware.org
Cc: Tom Tromey <tom@tromey.com>
Subject: [PATCH v2 23/32] "Finalize" the DWARF index in the background
Date: Thu,  4 Nov 2021 12:08:58 -0600	[thread overview]
Message-ID: <20211104180907.2360627-24-tom@tromey.com> (raw)
In-Reply-To: <20211104180907.2360627-1-tom@tromey.com>

After scanning the CUs, the DWARF indexer merges all the data into a
single vector, canonicalizing C++ names as it proceeds.  While not
necessarily single-threaded, this process is currently done in just
one thread, to keep memory costs lower.

However, this work is all done without reference to any data outside
of the indexes.  This patch improves the apparent performance of GDB
by moving it to the background.  All uses of the index are then made
to wait for this process to complete.

In our ongoing example, this reduces the scanning time on gdb itself
to 0.173937 (wall).  Recall that before this patch, the time was
0.668923; and psymbol reader does this in 1.598869.  That is, at the
end of this series, we see about a 10x speedup.
---
 gdb/dwarf2/cooked-index.c | 10 ++++++++--
 gdb/dwarf2/cooked-index.h | 17 +++++++++++++++++
 2 files changed, 25 insertions(+), 2 deletions(-)

diff --git a/gdb/dwarf2/cooked-index.c b/gdb/dwarf2/cooked-index.c
index bb10b884a43..7e114c4edd2 100644
--- a/gdb/dwarf2/cooked-index.c
+++ b/gdb/dwarf2/cooked-index.c
@@ -117,9 +117,13 @@ cooked_index::add (sect_offset die_offset, enum dwarf_tag tag,
 }
 
 cooked_index_vector::cooked_index_vector (vec_type &&vec)
-  : m_vector (std::move (vec))
+  : m_vector (std::move (vec)),
+    m_future (gdb::thread_pool::g_thread_pool->post_task
+	      ([this] ()
+	      {
+		finalize ();
+	      }))
 {
-  finalize ();
 }
 
 /* See cooked-index.h.  */
@@ -152,6 +156,8 @@ cooked_index_vector::get_addrmaps ()
 cooked_index_vector::range
 cooked_index_vector::find (gdb::string_view name, bool completing)
 {
+  m_future.wait ();
+
   auto lower = std::lower_bound (m_entries.begin (), m_entries.end (),
 				 name,
 				 [=] (const cooked_index_entry *entry,
diff --git a/gdb/dwarf2/cooked-index.h b/gdb/dwarf2/cooked-index.h
index a4bfc3eb183..365150b8692 100644
--- a/gdb/dwarf2/cooked-index.h
+++ b/gdb/dwarf2/cooked-index.h
@@ -254,6 +254,17 @@ class cooked_index_vector
   explicit cooked_index_vector (vec_type &&vec);
   DISABLE_COPY_AND_ASSIGN (cooked_index_vector);
 
+  ~cooked_index_vector ()
+  {
+    /* The 'finalize' method may be run in a different thread.  If
+       this object is destroyed before this completes, then the method
+       will end up writing to freed memory.  Waiting for this to
+       complete avoids this problem; and the cost seems ignorable
+       because creating and immediately destroying the debug info is a
+       relatively rare thing to do.  */
+    m_future.wait ();
+  }
+
   /* A simple range over part of m_entries.  */
   typedef iterator_range<std::vector<cooked_index_entry *>::iterator> range;
 
@@ -265,6 +276,7 @@ class cooked_index_vector
   /* Return a range of all the entries.  */
   range all_entries ()
   {
+    m_future.wait ();
     return { m_entries.begin (), m_entries.end () };
   }
 
@@ -305,6 +317,11 @@ class cooked_index_vector
 
   /* Storage for canonical names.  */
   std::vector<gdb::unique_xmalloc_ptr<char>> m_names;
+
+  /* A future that tracks when the 'finalize' method is done.  Note
+     that the 'get' method is never called on this future, only
+     'wait'.  */
+  std::future<void> m_future;
 };
 
 #endif /* GDB_DWARF2_COOKED_INDEX_H */
-- 
2.31.1


  parent reply	other threads:[~2021-11-04 18:10 UTC|newest]

Thread overview: 53+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-11-04 18:08 [PATCH v2 00/32] Rewrite the DWARF "partial" reader Tom Tromey
2021-11-04 18:08 ` [PATCH v2 01/32] Introduce make_unique_xstrndup Tom Tromey
2021-11-05 19:20   ` Simon Marchi
2021-11-05 20:08     ` Tom Tromey
2021-11-04 18:08 ` [PATCH v2 02/32] Split create_addrmap_from_aranges Tom Tromey
2021-11-04 18:08 ` [PATCH v2 03/32] Add dwarf2_per_cu_data::addresses_seen Tom Tromey
2021-11-04 18:08 ` [PATCH v2 04/32] Refactor dwarf2_get_pc_bounds Tom Tromey
2021-11-05 19:51   ` Simon Marchi
2021-11-24 15:53     ` Tom Tromey
2021-11-04 18:08 ` [PATCH v2 05/32] Allow ada_decode not to decode operators Tom Tromey
2021-11-04 18:08 ` [PATCH v2 06/32] Let skip_one_die not skip children Tom Tromey
2021-11-04 18:08 ` [PATCH v2 07/32] Add name splitting Tom Tromey
2021-11-04 18:08 ` [PATCH v2 08/32] Add new overload of dwarf5_djb_hash Tom Tromey
2021-11-05 20:01   ` Simon Marchi
2021-11-07 17:02     ` Tom Tromey
2021-11-04 18:08 ` [PATCH v2 09/32] Refactor build_type_psymtabs_reader Tom Tromey
2021-11-04 18:08 ` [PATCH v2 10/32] Add batching parameter to parallel_for_each Tom Tromey
2021-11-04 18:08 ` [PATCH v2 11/32] Return vector of results from parallel_for_each Tom Tromey
2021-11-17 20:37   ` Lancelot SIX
2021-11-18 14:41     ` Tom Tromey
2021-11-04 18:08 ` [PATCH v2 12/32] Specialize std::hash for gdb_exception Tom Tromey
2021-11-04 18:08 ` [PATCH v2 13/32] Introduce DWARF abbrev cache Tom Tromey
2021-11-04 18:08 ` [PATCH v2 14/32] Statically examine abbrev properties Tom Tromey
2021-11-04 18:08 ` [PATCH v2 15/32] Update skip_one_die for new " Tom Tromey
2021-11-04 18:08 ` [PATCH v2 16/32] Introduce the new DWARF index class Tom Tromey
2021-11-04 18:08 ` [PATCH v2 17/32] The new DWARF indexer Tom Tromey
2021-11-04 18:08 ` [PATCH v2 18/32] Implement quick_symbol_functions for cooked DWARF index Tom Tromey
2021-11-04 18:08 ` [PATCH v2 19/32] Wire in the new DWARF indexer Tom Tromey
2021-11-04 18:08 ` [PATCH v2 20/32] Introduce thread-safe handling for complaints Tom Tromey
2021-11-04 18:08 ` [PATCH v2 21/32] Pre-read DWARF section data Tom Tromey
2021-11-04 18:08 ` [PATCH v2 22/32] Parallelize DWARF indexing Tom Tromey
2021-11-04 18:08 ` Tom Tromey [this message]
2021-11-04 18:08 ` [PATCH v2 24/32] Rename write_psymtabs_to_index Tom Tromey
2021-11-04 18:09 ` [PATCH v2 25/32] Change the key type in psym_index_map Tom Tromey
2021-11-04 18:09 ` [PATCH v2 26/32] Change parameters to write_address_map Tom Tromey
2021-11-04 18:09 ` [PATCH v2 27/32] Genericize addrmap handling in the DWARF index writer Tom Tromey
2021-11-04 18:09 ` [PATCH v2 28/32] Adapt .gdb_index writer to new DWARF scanner Tom Tromey
2021-11-04 18:09 ` [PATCH v2 29/32] Adapt .debug_names " Tom Tromey
2021-11-04 18:09 ` [PATCH v2 30/32] Enable the new DWARF indexer Tom Tromey
2021-11-04 18:09 ` [PATCH v2 31/32] Delete DWARF psymtab code Tom Tromey
2021-11-04 18:09 ` [PATCH v2 32/32] Remove dwarf2_per_cu_data::v Tom Tromey
2021-11-06 12:25 ` [PATCH v2 00/32] Rewrite the DWARF "partial" reader Tom de Vries
2021-11-11 12:23   ` Tom de Vries
2021-11-16 23:56   ` Tom Tromey
2021-11-17  9:22     ` Tom de Vries
2021-11-18 14:43     ` Tom Tromey
2021-11-22 19:59     ` Tom Tromey
2021-11-22 20:52       ` Tom de Vries
2021-11-22 22:11         ` Tom Tromey
2021-11-23  7:56           ` Tom de Vries
2021-11-23 17:00             ` Tom Tromey
2021-11-08 17:41 ` Simon Marchi
2021-11-10 19:56   ` Tom Tromey

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=20211104180907.2360627-24-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).