Fix TAB-completion + .gdb_index slowness (generalize filename_seen_cache)
Tab completion when debugging a program binary that uses GDB index is
surprisingly much slower than when GDB uses psymtabs instead. Around
1.5x/3x slower. That's surprising, because the whole point of GDB
index is to speed things up...
For example, with:
set pagination off
set $count = 0
while $count < 400
complete b string_prin # matches gdb's string_printf
printf "count = %d\n", $count
set $count = $count + 1
end
$ time ./gdb --batch -q ./gdb-with-index -ex "source script.cmd"
real 0m11.042s
user 0m10.920s
sys 0m0.042s
$ time ./gdb --batch -q ./gdb-without-index -ex "source script.cmd"
real 0m4.635s
user 0m4.590s
sys 0m0.037s
Same but with:
- complete b string_prin
+ complete b zzzzzz
to exercise the no-matches worst case, master currently gets you
something like:
with index without index
real 0m11.971s 0m8.413s
user 0m11.912s 0m8.355s
sys 0m0.035s 0m0.035s
Running gdb under perf shows 80% spent inside
maybe_add_partial_symtab_filename, and 20% spent in the lbasename
inside that.
The problem that tab completion walks over all compunit symtabs, and
for each, walks the contained file symtabs. And there a huge number
of file symtabs (each included system header, etc.) that appear in
each compunit symtab's file symtab list. As in, when debugging GDB, I
have 367381 symtabs iterated, when of those only 5371 filenames are
unique...
This was a regression from the earlier (nice) split of symtabs in
compunit symtabs + file symtabs.
The fix here is to add a cache of unique filenames per objfile so that
the walk / uniquing is only done once. There's already a abstraction
for this in symtab.c; this patch moves that code out to a separate
file and C++ifies it bit.
This makes the worst-case scenario above consistently drop to ~2.5s
(1.5s for the "string_prin" hit case), making it over 3.3x times
faster than psymtabs in this use case (7x in the "string_prin" hit
case).
gdb/ChangeLog:
2017-07-17 Pedro Alves <palves@redhat.com>
* Makefile.in (COMMON_OBS): Add filename-seen-cache.o.
* dwarf2read.c: Include "filename-seen-cache.h".
* dwarf2read.c (dwarf2_per_objfile) <filenames_cache>: New field.
(dw2_map_symbol_filenames): Build and use a filenames_seen_cache.
* filename-seen-cache.c: New file.
* filename-seen-cache.h: New file.
* symtab.c: Include "filename-seen-cache.h".
(struct filename_seen_cache, INITIAL_FILENAME_SEEN_CACHE_SIZE)
(create_filename_seen_cache, clear_filename_seen_cache)
(delete_filename_seen_cache, filename_seen): Delete, parts moved
to filename-seen-cache.h/filename-seen-cache.c.
(output_source_filename, sources_info)
(maybe_add_partial_symtab_filename)
(make_source_files_completion_list): Adjust to use
filename_seen_cache.
2017-07-17 18:28:33 +08:00
|
|
|
/* Filename-seen cache for the GNU debugger, GDB.
|
|
|
|
|
|
2020-01-01 14:20:01 +08:00
|
|
|
Copyright (C) 1986-2020 Free Software Foundation, Inc.
|
Fix TAB-completion + .gdb_index slowness (generalize filename_seen_cache)
Tab completion when debugging a program binary that uses GDB index is
surprisingly much slower than when GDB uses psymtabs instead. Around
1.5x/3x slower. That's surprising, because the whole point of GDB
index is to speed things up...
For example, with:
set pagination off
set $count = 0
while $count < 400
complete b string_prin # matches gdb's string_printf
printf "count = %d\n", $count
set $count = $count + 1
end
$ time ./gdb --batch -q ./gdb-with-index -ex "source script.cmd"
real 0m11.042s
user 0m10.920s
sys 0m0.042s
$ time ./gdb --batch -q ./gdb-without-index -ex "source script.cmd"
real 0m4.635s
user 0m4.590s
sys 0m0.037s
Same but with:
- complete b string_prin
+ complete b zzzzzz
to exercise the no-matches worst case, master currently gets you
something like:
with index without index
real 0m11.971s 0m8.413s
user 0m11.912s 0m8.355s
sys 0m0.035s 0m0.035s
Running gdb under perf shows 80% spent inside
maybe_add_partial_symtab_filename, and 20% spent in the lbasename
inside that.
The problem that tab completion walks over all compunit symtabs, and
for each, walks the contained file symtabs. And there a huge number
of file symtabs (each included system header, etc.) that appear in
each compunit symtab's file symtab list. As in, when debugging GDB, I
have 367381 symtabs iterated, when of those only 5371 filenames are
unique...
This was a regression from the earlier (nice) split of symtabs in
compunit symtabs + file symtabs.
The fix here is to add a cache of unique filenames per objfile so that
the walk / uniquing is only done once. There's already a abstraction
for this in symtab.c; this patch moves that code out to a separate
file and C++ifies it bit.
This makes the worst-case scenario above consistently drop to ~2.5s
(1.5s for the "string_prin" hit case), making it over 3.3x times
faster than psymtabs in this use case (7x in the "string_prin" hit
case).
gdb/ChangeLog:
2017-07-17 Pedro Alves <palves@redhat.com>
* Makefile.in (COMMON_OBS): Add filename-seen-cache.o.
* dwarf2read.c: Include "filename-seen-cache.h".
* dwarf2read.c (dwarf2_per_objfile) <filenames_cache>: New field.
(dw2_map_symbol_filenames): Build and use a filenames_seen_cache.
* filename-seen-cache.c: New file.
* filename-seen-cache.h: New file.
* symtab.c: Include "filename-seen-cache.h".
(struct filename_seen_cache, INITIAL_FILENAME_SEEN_CACHE_SIZE)
(create_filename_seen_cache, clear_filename_seen_cache)
(delete_filename_seen_cache, filename_seen): Delete, parts moved
to filename-seen-cache.h/filename-seen-cache.c.
(output_source_filename, sources_info)
(maybe_add_partial_symtab_filename)
(make_source_files_completion_list): Adjust to use
filename_seen_cache.
2017-07-17 18:28:33 +08:00
|
|
|
|
|
|
|
|
This file is part of GDB.
|
|
|
|
|
|
|
|
|
|
This program is free software; you can redistribute it and/or modify
|
|
|
|
|
it under the terms of the GNU General Public License as published by
|
|
|
|
|
the Free Software Foundation; either version 3 of the License, or
|
|
|
|
|
(at your option) any later version.
|
|
|
|
|
|
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
|
GNU General Public License for more details.
|
|
|
|
|
|
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
|
|
|
along with this program. If not, see <http://www.gnu.org/licenses/>. */
|
|
|
|
|
|
2018-03-27 03:31:10 +08:00
|
|
|
#ifndef FILENAME_SEEN_CACHE_H
|
|
|
|
|
#define FILENAME_SEEN_CACHE_H
|
|
|
|
|
|
2019-04-03 10:04:24 +08:00
|
|
|
#include "defs.h"
|
Rename common to gdbsupport
This is the next patch in the ongoing series to move gdbsever to the
top level.
This patch just renames the "common" directory. The idea is to do
this move in two parts: first rename the directory (this patch), then
move the directory to the top. This approach makes the patches a bit
more tractable.
I chose the name "gdbsupport" for the directory. However, as this
patch was largely written by sed, we could pick a new name without too
much difficulty.
Tested by the buildbot.
gdb/ChangeLog
2019-07-09 Tom Tromey <tom@tromey.com>
* contrib/ari/gdb_ari.sh: Change common to gdbsupport.
* configure: Rebuild.
* configure.ac: Change common to gdbsupport.
* gdbsupport: Rename from common.
* acinclude.m4: Change common to gdbsupport.
* Makefile.in (CONFIG_SRC_SUBDIR, COMMON_SFILES)
(HFILES_NO_SRCDIR, stamp-version, ALLDEPFILES): Change common to
gdbsupport.
* aarch64-tdep.c, ada-lang.c, ada-lang.h, agent.c, alloc.c,
amd64-darwin-tdep.c, amd64-dicos-tdep.c, amd64-fbsd-nat.c,
amd64-fbsd-tdep.c, amd64-linux-nat.c, amd64-linux-tdep.c,
amd64-nbsd-tdep.c, amd64-obsd-tdep.c, amd64-sol2-tdep.c,
amd64-tdep.c, amd64-windows-tdep.c, arch-utils.c,
arch/aarch64-insn.c, arch/aarch64.c, arch/aarch64.h, arch/amd64.c,
arch/amd64.h, arch/arm-get-next-pcs.c, arch/arm-linux.c,
arch/arm.c, arch/i386.c, arch/i386.h, arch/ppc-linux-common.c,
arch/riscv.c, arch/riscv.h, arch/tic6x.c, arm-tdep.c, auto-load.c,
auxv.c, ax-gdb.c, ax-general.c, ax.h, breakpoint.c, breakpoint.h,
btrace.c, btrace.h, build-id.c, build-id.h, c-lang.h, charset.c,
charset.h, cli/cli-cmds.c, cli/cli-cmds.h, cli/cli-decode.c,
cli/cli-dump.c, cli/cli-option.h, cli/cli-script.c,
coff-pe-read.c, command.h, compile/compile-c-support.c,
compile/compile-c.h, compile/compile-cplus-symbols.c,
compile/compile-cplus-types.c, compile/compile-cplus.h,
compile/compile-loc2c.c, compile/compile.c, completer.c,
completer.h, contrib/ari/gdb_ari.sh, corefile.c, corelow.c,
cp-support.c, cp-support.h, cp-valprint.c, csky-tdep.c, ctf.c,
darwin-nat.c, debug.c, defs.h, disasm-selftests.c, disasm.c,
disasm.h, dtrace-probe.c, dwarf-index-cache.c,
dwarf-index-cache.h, dwarf-index-write.c, dwarf2-frame.c,
dwarf2expr.c, dwarf2loc.c, dwarf2read.c, event-loop.c,
event-top.c, exceptions.c, exec.c, extension.h, fbsd-nat.c,
features/aarch64-core.c, features/aarch64-fpu.c,
features/aarch64-pauth.c, features/aarch64-sve.c,
features/i386/32bit-avx.c, features/i386/32bit-avx512.c,
features/i386/32bit-core.c, features/i386/32bit-linux.c,
features/i386/32bit-mpx.c, features/i386/32bit-pkeys.c,
features/i386/32bit-segments.c, features/i386/32bit-sse.c,
features/i386/64bit-avx.c, features/i386/64bit-avx512.c,
features/i386/64bit-core.c, features/i386/64bit-linux.c,
features/i386/64bit-mpx.c, features/i386/64bit-pkeys.c,
features/i386/64bit-segments.c, features/i386/64bit-sse.c,
features/i386/x32-core.c, features/riscv/32bit-cpu.c,
features/riscv/32bit-csr.c, features/riscv/32bit-fpu.c,
features/riscv/64bit-cpu.c, features/riscv/64bit-csr.c,
features/riscv/64bit-fpu.c, features/tic6x-c6xp.c,
features/tic6x-core.c, features/tic6x-gp.c, filename-seen-cache.h,
findcmd.c, findvar.c, fork-child.c, gcore.c, gdb_bfd.c, gdb_bfd.h,
gdb_proc_service.h, gdb_regex.c, gdb_select.h, gdb_usleep.c,
gdbarch-selftests.c, gdbthread.h, gdbtypes.h, gnu-nat.c,
go32-nat.c, guile/guile.c, guile/scm-ports.c,
guile/scm-safe-call.c, guile/scm-type.c, i386-fbsd-nat.c,
i386-fbsd-tdep.c, i386-go32-tdep.c, i386-linux-nat.c,
i386-linux-tdep.c, i386-tdep.c, i387-tdep.c,
ia64-libunwind-tdep.c, ia64-linux-nat.c, inf-child.c,
inf-ptrace.c, infcall.c, infcall.h, infcmd.c, inferior-iter.h,
inferior.c, inferior.h, inflow.c, inflow.h, infrun.c, infrun.h,
inline-frame.c, language.h, linespec.c, linux-fork.c, linux-nat.c,
linux-tdep.c, linux-thread-db.c, location.c, machoread.c,
macrotab.h, main.c, maint.c, maint.h, memattr.c, memrange.h,
mi/mi-cmd-break.h, mi/mi-cmd-env.c, mi/mi-cmd-stack.c,
mi/mi-cmd-var.c, mi/mi-interp.c, mi/mi-main.c, mi/mi-parse.h,
minsyms.c, mips-linux-tdep.c, namespace.h,
nat/aarch64-linux-hw-point.c, nat/aarch64-linux-hw-point.h,
nat/aarch64-linux.c, nat/aarch64-sve-linux-ptrace.c,
nat/amd64-linux-siginfo.c, nat/fork-inferior.c,
nat/linux-btrace.c, nat/linux-btrace.h, nat/linux-namespaces.c,
nat/linux-nat.h, nat/linux-osdata.c, nat/linux-personality.c,
nat/linux-procfs.c, nat/linux-ptrace.c, nat/linux-ptrace.h,
nat/linux-waitpid.c, nat/mips-linux-watch.c,
nat/mips-linux-watch.h, nat/ppc-linux.c, nat/x86-dregs.c,
nat/x86-dregs.h, nat/x86-linux-dregs.c, nat/x86-linux.c,
nto-procfs.c, nto-tdep.c, objfile-flags.h, objfiles.c, objfiles.h,
obsd-nat.c, observable.h, osdata.c, p-valprint.c, parse.c,
parser-defs.h, ppc-linux-nat.c, printcmd.c, probe.c, proc-api.c,
procfs.c, producer.c, progspace.h, psymtab.h,
python/py-framefilter.c, python/py-inferior.c, python/py-ref.h,
python/py-type.c, python/python.c, record-btrace.c, record-full.c,
record.c, record.h, regcache-dump.c, regcache.c, regcache.h,
remote-fileio.c, remote-fileio.h, remote-sim.c, remote.c,
riscv-tdep.c, rs6000-aix-tdep.c, rust-exp.y, s12z-tdep.c,
selftest-arch.c, ser-base.c, ser-event.c, ser-pipe.c, ser-tcp.c,
ser-unix.c, skip.c, solib-aix.c, solib-target.c, solib.c,
source-cache.c, source.c, source.h, sparc-nat.c, spu-linux-nat.c,
stack.c, stap-probe.c, symfile-add-flags.h, symfile.c, symfile.h,
symtab.c, symtab.h, target-descriptions.c, target-descriptions.h,
target-memory.c, target.c, target.h, target/waitstatus.c,
target/waitstatus.h, thread-iter.h, thread.c, tilegx-tdep.c,
top.c, top.h, tracefile-tfile.c, tracefile.c, tracepoint.c,
tracepoint.h, tui/tui-io.c, ui-file.c, ui-out.h,
unittests/array-view-selftests.c,
unittests/child-path-selftests.c, unittests/cli-utils-selftests.c,
unittests/common-utils-selftests.c,
unittests/copy_bitwise-selftests.c, unittests/environ-selftests.c,
unittests/format_pieces-selftests.c,
unittests/function-view-selftests.c,
unittests/lookup_name_info-selftests.c,
unittests/memory-map-selftests.c, unittests/memrange-selftests.c,
unittests/mkdir-recursive-selftests.c,
unittests/observable-selftests.c,
unittests/offset-type-selftests.c, unittests/optional-selftests.c,
unittests/parse-connection-spec-selftests.c,
unittests/ptid-selftests.c, unittests/rsp-low-selftests.c,
unittests/scoped_fd-selftests.c,
unittests/scoped_mmap-selftests.c,
unittests/scoped_restore-selftests.c,
unittests/string_view-selftests.c, unittests/style-selftests.c,
unittests/tracepoint-selftests.c, unittests/unpack-selftests.c,
unittests/utils-selftests.c, unittests/xml-utils-selftests.c,
utils.c, utils.h, valarith.c, valops.c, valprint.c, value.c,
value.h, varobj.c, varobj.h, windows-nat.c, x86-linux-nat.c,
xml-support.c, xml-support.h, xml-tdesc.h, xstormy16-tdep.c,
xtensa-linux-nat.c, dwarf2read.h: Change common to gdbsupport.
gdb/gdbserver/ChangeLog
2019-07-09 Tom Tromey <tom@tromey.com>
* configure: Rebuild.
* configure.ac: Change common to gdbsupport.
* acinclude.m4: Change common to gdbsupport.
* Makefile.in (SFILES, OBS, GDBREPLAY_OBS, IPA_OBJS)
(version-generated.c, gdbsupport/%-ipa.o, gdbsupport/%.o): Change
common to gdbsupport.
* ax.c, event-loop.c, fork-child.c, gdb_proc_service.h,
gdbreplay.c, gdbthread.h, hostio-errno.c, hostio.c, i387-fp.c,
inferiors.c, inferiors.h, linux-aarch64-tdesc-selftest.c,
linux-amd64-ipa.c, linux-i386-ipa.c, linux-low.c,
linux-tic6x-low.c, linux-x86-low.c, linux-x86-tdesc-selftest.c,
linux-x86-tdesc.c, lynx-i386-low.c, lynx-low.c, mem-break.h,
nto-x86-low.c, regcache.c, regcache.h, remote-utils.c, server.c,
server.h, spu-low.c, symbol.c, target.h, tdesc.c, tdesc.h,
thread-db.c, tracepoint.c, win32-i386-low.c, win32-low.c: Change
common to gdbsupport.
2019-05-06 10:29:24 +08:00
|
|
|
#include "gdbsupport/function-view.h"
|
Fix TAB-completion + .gdb_index slowness (generalize filename_seen_cache)
Tab completion when debugging a program binary that uses GDB index is
surprisingly much slower than when GDB uses psymtabs instead. Around
1.5x/3x slower. That's surprising, because the whole point of GDB
index is to speed things up...
For example, with:
set pagination off
set $count = 0
while $count < 400
complete b string_prin # matches gdb's string_printf
printf "count = %d\n", $count
set $count = $count + 1
end
$ time ./gdb --batch -q ./gdb-with-index -ex "source script.cmd"
real 0m11.042s
user 0m10.920s
sys 0m0.042s
$ time ./gdb --batch -q ./gdb-without-index -ex "source script.cmd"
real 0m4.635s
user 0m4.590s
sys 0m0.037s
Same but with:
- complete b string_prin
+ complete b zzzzzz
to exercise the no-matches worst case, master currently gets you
something like:
with index without index
real 0m11.971s 0m8.413s
user 0m11.912s 0m8.355s
sys 0m0.035s 0m0.035s
Running gdb under perf shows 80% spent inside
maybe_add_partial_symtab_filename, and 20% spent in the lbasename
inside that.
The problem that tab completion walks over all compunit symtabs, and
for each, walks the contained file symtabs. And there a huge number
of file symtabs (each included system header, etc.) that appear in
each compunit symtab's file symtab list. As in, when debugging GDB, I
have 367381 symtabs iterated, when of those only 5371 filenames are
unique...
This was a regression from the earlier (nice) split of symtabs in
compunit symtabs + file symtabs.
The fix here is to add a cache of unique filenames per objfile so that
the walk / uniquing is only done once. There's already a abstraction
for this in symtab.c; this patch moves that code out to a separate
file and C++ifies it bit.
This makes the worst-case scenario above consistently drop to ~2.5s
(1.5s for the "string_prin" hit case), making it over 3.3x times
faster than psymtabs in this use case (7x in the "string_prin" hit
case).
gdb/ChangeLog:
2017-07-17 Pedro Alves <palves@redhat.com>
* Makefile.in (COMMON_OBS): Add filename-seen-cache.o.
* dwarf2read.c: Include "filename-seen-cache.h".
* dwarf2read.c (dwarf2_per_objfile) <filenames_cache>: New field.
(dw2_map_symbol_filenames): Build and use a filenames_seen_cache.
* filename-seen-cache.c: New file.
* filename-seen-cache.h: New file.
* symtab.c: Include "filename-seen-cache.h".
(struct filename_seen_cache, INITIAL_FILENAME_SEEN_CACHE_SIZE)
(create_filename_seen_cache, clear_filename_seen_cache)
(delete_filename_seen_cache, filename_seen): Delete, parts moved
to filename-seen-cache.h/filename-seen-cache.c.
(output_source_filename, sources_info)
(maybe_add_partial_symtab_filename)
(make_source_files_completion_list): Adjust to use
filename_seen_cache.
2017-07-17 18:28:33 +08:00
|
|
|
|
|
|
|
|
/* Cache to watch for file names already seen. */
|
|
|
|
|
|
|
|
|
|
class filename_seen_cache
|
|
|
|
|
{
|
|
|
|
|
public:
|
|
|
|
|
filename_seen_cache ();
|
|
|
|
|
|
2017-09-19 17:10:03 +08:00
|
|
|
DISABLE_COPY_AND_ASSIGN (filename_seen_cache);
|
Fix TAB-completion + .gdb_index slowness (generalize filename_seen_cache)
Tab completion when debugging a program binary that uses GDB index is
surprisingly much slower than when GDB uses psymtabs instead. Around
1.5x/3x slower. That's surprising, because the whole point of GDB
index is to speed things up...
For example, with:
set pagination off
set $count = 0
while $count < 400
complete b string_prin # matches gdb's string_printf
printf "count = %d\n", $count
set $count = $count + 1
end
$ time ./gdb --batch -q ./gdb-with-index -ex "source script.cmd"
real 0m11.042s
user 0m10.920s
sys 0m0.042s
$ time ./gdb --batch -q ./gdb-without-index -ex "source script.cmd"
real 0m4.635s
user 0m4.590s
sys 0m0.037s
Same but with:
- complete b string_prin
+ complete b zzzzzz
to exercise the no-matches worst case, master currently gets you
something like:
with index without index
real 0m11.971s 0m8.413s
user 0m11.912s 0m8.355s
sys 0m0.035s 0m0.035s
Running gdb under perf shows 80% spent inside
maybe_add_partial_symtab_filename, and 20% spent in the lbasename
inside that.
The problem that tab completion walks over all compunit symtabs, and
for each, walks the contained file symtabs. And there a huge number
of file symtabs (each included system header, etc.) that appear in
each compunit symtab's file symtab list. As in, when debugging GDB, I
have 367381 symtabs iterated, when of those only 5371 filenames are
unique...
This was a regression from the earlier (nice) split of symtabs in
compunit symtabs + file symtabs.
The fix here is to add a cache of unique filenames per objfile so that
the walk / uniquing is only done once. There's already a abstraction
for this in symtab.c; this patch moves that code out to a separate
file and C++ifies it bit.
This makes the worst-case scenario above consistently drop to ~2.5s
(1.5s for the "string_prin" hit case), making it over 3.3x times
faster than psymtabs in this use case (7x in the "string_prin" hit
case).
gdb/ChangeLog:
2017-07-17 Pedro Alves <palves@redhat.com>
* Makefile.in (COMMON_OBS): Add filename-seen-cache.o.
* dwarf2read.c: Include "filename-seen-cache.h".
* dwarf2read.c (dwarf2_per_objfile) <filenames_cache>: New field.
(dw2_map_symbol_filenames): Build and use a filenames_seen_cache.
* filename-seen-cache.c: New file.
* filename-seen-cache.h: New file.
* symtab.c: Include "filename-seen-cache.h".
(struct filename_seen_cache, INITIAL_FILENAME_SEEN_CACHE_SIZE)
(create_filename_seen_cache, clear_filename_seen_cache)
(delete_filename_seen_cache, filename_seen): Delete, parts moved
to filename-seen-cache.h/filename-seen-cache.c.
(output_source_filename, sources_info)
(maybe_add_partial_symtab_filename)
(make_source_files_completion_list): Adjust to use
filename_seen_cache.
2017-07-17 18:28:33 +08:00
|
|
|
|
|
|
|
|
/* Empty the cache, but do not delete it. */
|
|
|
|
|
void clear ();
|
|
|
|
|
|
|
|
|
|
/* If FILE is not already in the table of files in CACHE, add it and
|
|
|
|
|
return false; otherwise return true.
|
|
|
|
|
|
|
|
|
|
NOTE: We don't manage space for FILE, we assume FILE lives as
|
|
|
|
|
long as the caller needs. */
|
|
|
|
|
bool seen (const char *file);
|
|
|
|
|
|
|
|
|
|
/* Traverse all cache entries, calling CALLBACK on each. The
|
|
|
|
|
filename is passed as argument to CALLBACK. */
|
|
|
|
|
void traverse (gdb::function_view<void (const char *filename)> callback)
|
|
|
|
|
{
|
|
|
|
|
auto erased_cb = [] (void **slot, void *info) -> int
|
|
|
|
|
{
|
|
|
|
|
auto filename = (const char *) *slot;
|
|
|
|
|
auto restored_cb = (decltype (callback) *) info;
|
|
|
|
|
(*restored_cb) (filename);
|
|
|
|
|
return 1;
|
|
|
|
|
};
|
|
|
|
|
|
2020-09-18 01:47:50 +08:00
|
|
|
htab_traverse_noresize (m_tab.get (), erased_cb, &callback);
|
Fix TAB-completion + .gdb_index slowness (generalize filename_seen_cache)
Tab completion when debugging a program binary that uses GDB index is
surprisingly much slower than when GDB uses psymtabs instead. Around
1.5x/3x slower. That's surprising, because the whole point of GDB
index is to speed things up...
For example, with:
set pagination off
set $count = 0
while $count < 400
complete b string_prin # matches gdb's string_printf
printf "count = %d\n", $count
set $count = $count + 1
end
$ time ./gdb --batch -q ./gdb-with-index -ex "source script.cmd"
real 0m11.042s
user 0m10.920s
sys 0m0.042s
$ time ./gdb --batch -q ./gdb-without-index -ex "source script.cmd"
real 0m4.635s
user 0m4.590s
sys 0m0.037s
Same but with:
- complete b string_prin
+ complete b zzzzzz
to exercise the no-matches worst case, master currently gets you
something like:
with index without index
real 0m11.971s 0m8.413s
user 0m11.912s 0m8.355s
sys 0m0.035s 0m0.035s
Running gdb under perf shows 80% spent inside
maybe_add_partial_symtab_filename, and 20% spent in the lbasename
inside that.
The problem that tab completion walks over all compunit symtabs, and
for each, walks the contained file symtabs. And there a huge number
of file symtabs (each included system header, etc.) that appear in
each compunit symtab's file symtab list. As in, when debugging GDB, I
have 367381 symtabs iterated, when of those only 5371 filenames are
unique...
This was a regression from the earlier (nice) split of symtabs in
compunit symtabs + file symtabs.
The fix here is to add a cache of unique filenames per objfile so that
the walk / uniquing is only done once. There's already a abstraction
for this in symtab.c; this patch moves that code out to a separate
file and C++ifies it bit.
This makes the worst-case scenario above consistently drop to ~2.5s
(1.5s for the "string_prin" hit case), making it over 3.3x times
faster than psymtabs in this use case (7x in the "string_prin" hit
case).
gdb/ChangeLog:
2017-07-17 Pedro Alves <palves@redhat.com>
* Makefile.in (COMMON_OBS): Add filename-seen-cache.o.
* dwarf2read.c: Include "filename-seen-cache.h".
* dwarf2read.c (dwarf2_per_objfile) <filenames_cache>: New field.
(dw2_map_symbol_filenames): Build and use a filenames_seen_cache.
* filename-seen-cache.c: New file.
* filename-seen-cache.h: New file.
* symtab.c: Include "filename-seen-cache.h".
(struct filename_seen_cache, INITIAL_FILENAME_SEEN_CACHE_SIZE)
(create_filename_seen_cache, clear_filename_seen_cache)
(delete_filename_seen_cache, filename_seen): Delete, parts moved
to filename-seen-cache.h/filename-seen-cache.c.
(output_source_filename, sources_info)
(maybe_add_partial_symtab_filename)
(make_source_files_completion_list): Adjust to use
filename_seen_cache.
2017-07-17 18:28:33 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
/* Table of files seen so far. */
|
2020-09-18 01:47:50 +08:00
|
|
|
htab_up m_tab;
|
Fix TAB-completion + .gdb_index slowness (generalize filename_seen_cache)
Tab completion when debugging a program binary that uses GDB index is
surprisingly much slower than when GDB uses psymtabs instead. Around
1.5x/3x slower. That's surprising, because the whole point of GDB
index is to speed things up...
For example, with:
set pagination off
set $count = 0
while $count < 400
complete b string_prin # matches gdb's string_printf
printf "count = %d\n", $count
set $count = $count + 1
end
$ time ./gdb --batch -q ./gdb-with-index -ex "source script.cmd"
real 0m11.042s
user 0m10.920s
sys 0m0.042s
$ time ./gdb --batch -q ./gdb-without-index -ex "source script.cmd"
real 0m4.635s
user 0m4.590s
sys 0m0.037s
Same but with:
- complete b string_prin
+ complete b zzzzzz
to exercise the no-matches worst case, master currently gets you
something like:
with index without index
real 0m11.971s 0m8.413s
user 0m11.912s 0m8.355s
sys 0m0.035s 0m0.035s
Running gdb under perf shows 80% spent inside
maybe_add_partial_symtab_filename, and 20% spent in the lbasename
inside that.
The problem that tab completion walks over all compunit symtabs, and
for each, walks the contained file symtabs. And there a huge number
of file symtabs (each included system header, etc.) that appear in
each compunit symtab's file symtab list. As in, when debugging GDB, I
have 367381 symtabs iterated, when of those only 5371 filenames are
unique...
This was a regression from the earlier (nice) split of symtabs in
compunit symtabs + file symtabs.
The fix here is to add a cache of unique filenames per objfile so that
the walk / uniquing is only done once. There's already a abstraction
for this in symtab.c; this patch moves that code out to a separate
file and C++ifies it bit.
This makes the worst-case scenario above consistently drop to ~2.5s
(1.5s for the "string_prin" hit case), making it over 3.3x times
faster than psymtabs in this use case (7x in the "string_prin" hit
case).
gdb/ChangeLog:
2017-07-17 Pedro Alves <palves@redhat.com>
* Makefile.in (COMMON_OBS): Add filename-seen-cache.o.
* dwarf2read.c: Include "filename-seen-cache.h".
* dwarf2read.c (dwarf2_per_objfile) <filenames_cache>: New field.
(dw2_map_symbol_filenames): Build and use a filenames_seen_cache.
* filename-seen-cache.c: New file.
* filename-seen-cache.h: New file.
* symtab.c: Include "filename-seen-cache.h".
(struct filename_seen_cache, INITIAL_FILENAME_SEEN_CACHE_SIZE)
(create_filename_seen_cache, clear_filename_seen_cache)
(delete_filename_seen_cache, filename_seen): Delete, parts moved
to filename-seen-cache.h/filename-seen-cache.c.
(output_source_filename, sources_info)
(maybe_add_partial_symtab_filename)
(make_source_files_completion_list): Adjust to use
filename_seen_cache.
2017-07-17 18:28:33 +08:00
|
|
|
};
|
2018-03-27 03:31:10 +08:00
|
|
|
|
|
|
|
|
#endif /* FILENAME_SEEN_CACHE_H */
|
