Check that the cache itself is idle. Previously it would start during
transforms and long pyramid rendering ops and toss writes and large
seeks into the tile cache while it was potentially under heavy pressure.
Also increase its flush rate more to be more in line with modern system
and editing needs.
Although the clean-first strategy gives fast light-load performance,
it also degrades catastrophically under moderate cache pressure. LRU is
not as efficient under light load, but degrades more gracefully under
moderate and heavy load.
Add more profiling In order to analyze efficiency and behavior of the
tile cache. Profiling includes run-time indication of idle swapper
activity.
Also, fixes to original tile profiling code that appears to have
bitrotted some; it was no longer tracking changes made to tile memory
outside of tile.c.
2009-01-17 Michael Natterer <mitch@gimp.org>
* all files with a GPL header and all COPYING files:
Change licence to GPLv3 (and to LGPLv3 for libgimp).
Cleaned up some copyright headers and regenerated the parsers in
the ImageMap plugin.
svn path=/trunk/; revision=27913
2008-10-08 Sven Neumann <sven@gimp.org>
* app/base/tile-cache.c: use a GMutex instead of a GStaticMutex
as the latter needs API that causes compiler warnings about
dereferencing of type-punned pointers.
svn path=/trunk/; revision=27170
2007-06-13 Sven Neumann <sven@gimp.org>
* app/base/base.c (base_exit): call tile_cache_exit() before
tile_swap_exit(). tile_cache_exit() may use the tile swap in
pathological cases.
* app/base/tile-cache.c (tile_cache_exit): warn if the tile cache
is not empty.
svn path=/trunk/; revision=22772
2006-09-07 Sven Neumann <sven@gimp.org>
* app/base/tile-cache.c
* app/base/tile-manager.c
* app/base/tile-private.h
* app/base/tile.c: removed tile locking macros that were only used
for the threaded tile swapper.
2006-09-07 Sven Neumann <sven@gimp.org>
* configure.in: removed --enable_threaded_tile_swapper option.
* app/base/tile-cache.c
* app/base/tile-private.h
* app/base/tile-swap.c
* app/base/tile.c: removed threaded tile swapper code which has
been sitting here unused for too long.
2006-08-24 Sven Neumann <sven@gimp.org>
* INSTALL: mention that multi-processor support is on by
default.
* app/base/tile-cache.c: fixed compilation if multi-processor
support is disabled (bug #352609).
2006-08-16 Sven Neumann <sven@gimp.org>
* app/base/tile-cache.c: actually enable tile cache locking and
added a missing lock in tile_idle_preswap(). Should fix bug
#346923.
2005-09-29 Sven Neumann <sven@gimp.org>
* app/base/tile-cache.c
* app/base/tile-manager.c
* app/base/tile-private.h
* app/base/tile-swap.c
* app/base/tile.c: keep the data size of a tile in the Tile struct
instead of recalculating it over and over again. Adds an overhead
of 4 bytes per tile but speeds up the tile cache significantly.
2005-02-13 Sven Neumann <sven@gimp.org>
* configure.in
* app/base/tile-cache.c
* app/base/tile-private.h
* app/base/tile-swap.c
* app/base/tile.c: in an attempt to safe this code from bit-rot,
ported the experimental threaded tile-swapper to gthread.
2004-06-13 Sven Neumann <sven@gimp.org>
* app/paint-funcs/paint-funcs.c (shapeburst_region): don't call
tile_ewidth() three times from the inner loop.
* app/base/tile-manager.c (tile_manager_get): don't call
tile_size() twice on the same tile.
* app/base/tile-private.h: added tile_size_inline(), an inline
version of the tile_size() function.
* app/base/tile-cache.c
* app/base/tile-manager.c
* app/base/tile-swap.c
* app/base/tile.c: use tile_size_inline() from inside the tile
subsystem.
2004-02-18 Sven Neumann <sven@gimp.org>
* tile-cache.c
* tile-private.h
* tile.[ch]: removed trailing whitespace, added some newlines,
let tile_is_valid() return a gboolean instead of a gint.
* app/core/gimpimage-projection.c
* app/core/gimpimage-undo-push.c
* app/paint/gimppaintcore.c
* app/tools/gimpinktool.c: use the return value from tile_is_valid()
as a boolean.
2002-11-21 Michael Natterer <mitch@gimp.org>
* app/base/tile-cache.[ch]: added tile_cache_init() and
tile_cache_exit() as public functions.
* app/base/base.c: call them. Connect to GimpBaseConfig's
notify::tile-cache-size.
* app/core/gimpimage.c: connect to notify::transparency-type and
notify::transparency-size and call
gimp_image_invalidate_layer_previews() in the callback.
* app/display/gimpdisplay-foreach.[ch]: removed
gdisplays_expose_full().
* app/display/gimpdisplayshell-handlers.c: connect to
notify::transparency-type and notify::transparency-size and expose
the shell in the callback.
* app/display/gimpdisplayshell-render.[ch]: added render_init()
and render_exit() functions and connect to
notify::transparency-type and notify::transparency-size for
setting up the render buffers.
* app/gui/gui.c: call the new functions instead of render_setup().
Connect to notify::show-tool-tips.
* app/gui/preferences-dialog.c: copy values between the global
config object and its local clone only when they have changed.
Wrap setting of lots of values in g_object_[freeze|thaw]_notify().
Added lots of mnemonics.
* app/gui/resolution-calibrate-dialog.c: added mnemonics.
* app/widgets/gimppropwidgets.c: g_object_set() the values edited
by the gimp_prop_coordinates() much less often by remembering the
old values and setting them only when they have changed.
2002-05-28 Sven Neumann <sven@gimp.org>
* app/base/base-config.[ch]
* app/gui/user-install-dialog.c: use a gulong for the tile_cache_size.
* app/base/tile-cache.c: cleanup, no changes.
* app/config/gimpconfig-types.c (string_to_memsize): avoid overflows.
2001-12-07 Sven Neumann <sven@gimp.org>
* app/core/gimpmarshal.[ch]: removed from CVS, they are generated.
* app/base/Makefile.am
* app/base/base-enums.h: new file defining enums that are to be
registered. Used to build app/base/base-enums.c.
* app/base/base-types.h: include base-enums.h.
* tools/pdbgen/Makefile.am
* tools/pdbgen/enumcode.pl
* tools/pdbgen/enums.pl: parse the new base-enums.h file and modified
the perl voodoo so it doesn't prefix enums with GIMP_ that are already
properly namespaced.
* app/core/core-types.h: don't need to chop GIMP from enum.
* app/pdb/color_cmds.c
* app/pdb/tools_cmds.c
* libgimp/gimpenums.h
* plug-ins/script-fu/script-fu-constants.c: regenerated.
* app/config/gimpconfig-deserialize.[ch]
* app/config/gimpconfig-serialize.[ch]
* app/config/gimpconfig.[ch]: made GimpConfig an interface including
a reasonable default implementation that works on object properties.
* app/config/Makefile.am
* app/config/gimpbaseconfig.[ch]: new GimpBaseConfig using the
GimpConfig interface. Yet only used for testing from app/main.c.
* app/main.c: test the new GimpBaseConfig object.
* app/gimprc.c
* app/base/base-config.h
* app/base/*.c
* app/core/gimpdatafiles.c
* app/core/gimpdrawable-transform.c
* app/core/gimppreviewcache.c
* app/gui/preferences-dialog.c
* app/paint-funcs/paint-funcs.c
* app/xcf/xcf-seek.c: need to include glib-object.h since base-config
contains registered enums now. Follow name change of InterpolationType
to GimpInterpolationType.
2001-05-15 Michael Natterer <mitch@gimp.org>
* configure.in: new directory app/base/
* app/Makefile.am
* app/boundary.[ch]
* app/brush_scale.[ch]
* app/gimpchecks.h
* app/gimplut.[ch]
* app/pixel_processor.[ch]
* app/pixel_region.[ch]
* app/pixel_surround.[ch]
* app/temp_buf.[ch]
* app/tile.[ch]
* app/tile_cache.[ch]
* app/tile_manager.[ch]
* app/tile_manager_pvt.h
* app/tile_pvt.h
* app/tile_swap.[ch]: moved to base/
* app/base/Makefile.am
* app/base/base-types.h
* app/base/*: new directory for the sub-object pixel maniplation
and storage stuff. Does not include Gtk+ or anything outside
base/. Did some cleanup in all files.
* app/appenums.h
* app/apptypes.h
* app/core/gimpimage.h: removed types which are now in
base/base-types.h.
* app/base/base-config.[ch]
* app/gimprc.[ch]: put the config variables for base/ to their own
file so base/ doesn not have to include gimprc.h (does not yet
work, i.e. the variables are un-configurable right now)
* app/main.c: set a log handler for "Gimp-Base".
* app/paint-funcs/Makefile.am
* app/paint-funcs/paint-funcs.[ch]: removed the color hash which
maps RGB to color indices because it's a totally standalone system
which has nothing to do with the paint-funcs and introduced a
GimpImage dependency.
paint-funcs/ should be considered on the same sub-object
(glib-only) level as base/, only in a different directory.
* app/core/Makefile.am
* app/core/gimpimage-colorhash.[ch]: put the color hash here.
* app/gimage.c: don't invalidate the color hash here...
* app/core/gimpimage.c: ... but in the colormap_changed() default
inplementation. Initialize the hash in class_init().
* tools/pdbgen/Makefile.am: scan app/base/base-types.h for enums.
* tools/pdbgen/enums.pl: regenerated.
* app/[lots]
* app/core/[of]
* app/gui/[files]
* app/pdb/[all]
* app/tools/[over]
* app/widgets/[the]
* tools/pdbgen/pdb/[place]: changed #includes accordingly. And use
base_config->value instead of the stuff from gimprc.h.
2000-12-29 Michael Natterer <mitch@gimp.org>
* app/Makefile.am
* app/channel_pvt.h
* app/drawable_pvt.h
* app/gdisplayF.h
* app/gimpdrawableP.h
* app/gimpimageP.h
* app/layer_pvt.h
* app/toolsF.h: removed these files.
* app/apptypes.h
* tools/pdbgen/enums.pl: added tons of opaque typedefs and enums.
* tools/pdbgen/pdb/brush_select.pdb
* tools/pdbgen/pdb/brushes.pdb
* tools/pdbgen/pdb/channel.pdb
* tools/pdbgen/pdb/color.pdb
* tools/pdbgen/pdb/convert.pdb
* tools/pdbgen/pdb/display.pdb
* tools/pdbgen/pdb/drawable.pdb
* tools/pdbgen/pdb/fileops.pdb
* tools/pdbgen/pdb/gradient_select.pdb
* tools/pdbgen/pdb/gradients.pdb
* tools/pdbgen/pdb/help.pdb
* tools/pdbgen/pdb/image.pdb
* tools/pdbgen/pdb/layer.pdb
* tools/pdbgen/pdb/pattern_select.pdb
* tools/pdbgen/pdb/patterns.pdb
* tools/pdbgen/pdb/selection.pdb
* tools/pdbgen/pdb/tools.pdb
* app/*: chainsaw #include cleanup:
- Never (never!!) include stuff in header files except where we
need access to structures' contents (like derived objects).
- Added prototypes and proper formating in many files.
- The #include order in *all* *.c files is as follows:
#include "config.h"
#include <system stuff>
#include <gtk/gtk.h>
#include "apptypes.h"
#include "gimp stuff"
#include "libgimp stuff"
#include "libgimp/gimpintl.h"
By following this scheme we can easily see a file's dependencies
from it's #include's and can grep for the inclusion to find out
where a file is used.
* tools/pdbgen/app.pl: changed to follow the include scheme above.
* libgimp/Makefile.am
* libgimp/gimpuitypes.h: new file, included from libgimp/gimpui.h
and from app/apptypes.h.
* libgimp/gimpcolorbutton.[ch]
* libgimp/gimpdialog.[ch]
* libgimp/gimphelpui.[ch]
* libgimp/gimpparasite.[ch]
* libgimp/gimppatheditor.[ch]
* libgimp/gimpprotocol.c
* libgimp/gimpquerybox.[ch]
* libgimp/gimpsizeentry.[ch]
* libgimp/gimptypes.h
* libgimp/gimpui.h
* libgimp/gimpunit.h
* libgimp/gimpunitmenu.[ch]
* libgimp/gimpwidgets.[ch]: changed accordingly.
* plug-ins/FractalExplorer/Dialogs.c
* plug-ins/gdyntext/message_window.c
* plug-ins/imagemap/imap_default_dialog.c
* plug-ins/imagemap/imap_file.c: these files used to include
"libgimp/gimpui.h" without including "libgimp/gimp.h". This is
no longer possible because the libgimpui headers don't inlcude
"libgimp/gimpunit.h" any more.
Thu Jan 6 18:39:05 GMT 2000 Adam D. Moss <adam@gimp.org>
* app/tile_cache.c
* app/tile_swap.c: Maybe fixed tile_cache_set_size()
so that it actually works now. You can use this to set GIMP's
tile cache size at runtime without restarting. If the prefs
guys (Sven, Michael?) don't wire it in I'll look at it, but it
can't be soon.
1999-12-07 Michael Natterer <mitch@gimp.org>
* app/main.c
* app/menus.c: Wrapped the 'plugin_domains' definition and
initialization and menu_translate() with '#ifdef ENABLE_NLS'.
* app/palette.c: Removed another trailing comma from an enum.
* app/tile_cache.c (tile_idle_thread): Changed tile->swap to
tile->swap_offset because tile->swap doesn't exist. Note that I
don't understand at all what's going on there. I commited this
because (1) this way it compiles and (2) it at least _looks_ like
the right thing to do. Garry, could you check this please?
1999-12-05 Garry R. Osgood <gosgood@idt.net>
Inaugural checkin: Tile accounting
* app/tile_cache.c
* app/tile.c
cur_cache_dirty in tile_cache.c was not being fully updated. Fewer
dirty bytes were being added to this global than what were being
placed on the dirty tile queue. Deductions, however, were correct
sometimes causing cur_cache_dirty to wrap around zero and become
large positive. This skewed the number of tiles to be swapped to disk,
usually downward, but sometimes dramatically upward.
active_tile_count, (tile.c) currently a diagnostic counter, was
not being decremented.
Full patch documentation at
http://idt.net/~gosgood/gimp-patch/patch01.html#991205-0
* app/tile_swap.c (tile_swap_default_out): don't update a tile's swap
file position until the tile's actually been successfully written
out
* app/tile_cache.c (tile_cache_zorch_next): add test to make sure tile
was successfully swapped out before dropping it from the cache
--klm
Wed Oct 14 17:46:15 EDT 1998 Adrian Likins <adrian@gimp.org>
* app/*, po/de.po, de/POTFILES.in, libgimp/gimpintl.h:
Lots of ii8n stuff here and some additions to the de.po.
Applied gimp-egger-981005-1 ,gimp-egger-981006-1,
gimp-egger-981007-1, gimp-egger-981008-1,
gimp-egger-981009-1.patch, gimp-egger-981010-1.patch
* plug-in/guillotine/guillotine.c: added the coordinates
of the split images from the original image to the title.
ie foo.jpg (0,0) for the image in the topleft.
* plug-in/script-fu/scripts/neon-logo.scm,
perspective-shadow.scm, predator.scm,rendermap.scm,
ripply-anim.scm, select_to_image.scm,swirltile.scm,
xach-effect.scm: updated scripts to use new script-fu stuff
wooo boy! a big un!
in testing this, it looks like some of the po files are busted.
but the code stuff seems okay.
-adrian
Modified tile swapping to preswap dirty tiles, using a thread if
threads are enabled in configure or a gtk_timeout if they're not. The
code appears stable but could use improvement. Be careful when
working in here; there are four sets of semaphores and getting the
locking wrong will cause deadlocks or tile corruption. I'll try to
write up a document describing what's going on in here sometime soon.
--sg
