Mercurial > hg > xemacs-beta
view src/gccache-gtk.c @ 4921:17362f371cc2
add more byte-code assertions and better failure output
-------------------- ChangeLog entries follow: --------------------
src/ChangeLog addition:
2010-02-03 Ben Wing <ben@xemacs.org>
* alloc.c (Fmake_byte_code):
* bytecode.h:
* lisp.h:
* lread.c:
* lread.c (readevalloop):
* lread.c (Fread):
* lread.c (Fread_from_string):
* lread.c (read_list_conser):
* lread.c (read_list):
* lread.c (vars_of_lread):
* symbols.c:
* symbols.c (Fdefine_function):
Turn on the "compiled-function annotation hack". Implement it
properly by hooking into Fdefalias(). Note in the docstring to
`defalias' that we do this. Remove some old broken code and
change code that implemented the old kludgy way of hooking into
the Lisp reader into bracketed by `#ifdef
COMPILED_FUNCTION_ANNOTATION_HACK_OLD_WAY', which is not enabled.
Also enable byte-code metering when DEBUG_XEMACS -- this is a form
of profiling for computing histograms of which sequences of two
bytecodes are used most often.
* bytecode-ops.h:
* bytecode-ops.h (OPCODE):
New file. Extract out all the opcodes and declare them using
OPCODE(), a bit like frame slots and such. This way the file can
be included multiple times if necessary to iterate multiple times
over the byte opcodes.
* bytecode.c:
* bytecode.c (NUM_REMEMBERED_BYTE_OPS):
* bytecode.c (OPCODE):
* bytecode.c (assert_failed_with_remembered_ops):
* bytecode.c (READ_UINT_2):
* bytecode.c (READ_INT_1):
* bytecode.c (READ_INT_2):
* bytecode.c (PEEK_INT_1):
* bytecode.c (PEEK_INT_2):
* bytecode.c (JUMP_RELATIVE):
* bytecode.c (JUMP_NEXT):
* bytecode.c (PUSH):
* bytecode.c (POP_WITH_MULTIPLE_VALUES):
* bytecode.c (DISCARD):
* bytecode.c (UNUSED):
* bytecode.c (optimize_byte_code):
* bytecode.c (optimize_compiled_function):
* bytecode.c (Fbyte_code):
* bytecode.c (vars_of_bytecode):
* bytecode.c (init_opcode_table_multi_op):
* bytecode.c (reinit_vars_of_bytecode):
* emacs.c (main_1):
* eval.c (funcall_compiled_function):
* symsinit.h:
Any time we change either the instruction pointer or the stack
pointer, assert that we're going to move it to a valid location.
This should catch failures right when they occur rather than
sometime later. This requires that we pass in another couple of
parameters into some functions (only with error-checking enabled,
see below).
Also keep track, using a circular queue, of the last 100 byte
opcodes seen, and when we hit an assert failure during byte-code
execution, output the contents of the queue in a nice readable
fashion. This requires that bytecode-ops.h be included a second
time so that a table mapping opcodes to the name of their operation
can be constructed. This table is constructed in new function
reinit_vars_of_bytecode().
Everything in the last two paras happens only when
ERROR_CHECK_BYTE_CODE.
Add some longish comments describing how the arrays that hold the
stack and instructions, and the pointers used to access them, work.
* gc.c:
Import some code from my `latest-fix' workspace to mark the
staticpro's in order from lowest to highest, rather than highest to
lowest, so it's easier to debug when something goes wrong.
* lisp.h (abort_with_message): Renamed from abort_with_msg().
* symbols.c (defsymbol_massage_name_1):
* symbols.c (defsymbol_nodump):
* symbols.c (defsymbol):
* symbols.c (defkeyword):
* symeval.h (DEFVAR_SYMVAL_FWD_OBJECT):
Make the various calls to staticpro() instead call staticpro_1(),
passing in the name of the C var being staticpro'ed, so that it
shows up in staticpro_names. Otherwise staticpro_names just has
1000+ copies of the word `location'.
author | Ben Wing <ben@xemacs.org> |
---|---|
date | Wed, 03 Feb 2010 08:01:55 -0600 |
parents | b3ce27ca7647 |
children | 16112448d484 |
line wrap: on
line source
/* Efficient caching of Gtk GCs (graphics contexts). Copyright (C) 1993 Free Software Foundation, Inc. Copyright (C) 1994, 1995 Board of Trustees, University of Illinois. This file is part of XEmacs. XEmacs 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 2, or (at your option) any later version. XEmacs 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 XEmacs; see the file COPYING. If not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ /* Synched up with: Not in FSF. */ /* XEmacs uses a lot of different display attributes; for example, assume that only four fonts are in use (normal, bold, italic, and bold-italic). Then assume that one stipple or background is used for text selections, and another is used for highlighting mousable regions. That makes 16 GCs already. Add in the fact that another GC may be needed to display the text cursor in any of those regions, and you've got 32. Add in more fonts, and it keeps increasing exponentially. We used to keep these GCs in a cache of merged (fully qualified) faces. However, a lot of other code in xterm.c used XChangeGC of existing GCs, which is kind of slow and kind of random. Also, managing the face cache was tricky because it was hard to know when a face was no longer visible on the frame -- we had to mark all frames as garbaged whenever a face was changed, which caused an unpleasant amount of flicker (since faces are created/destroyed (= changed) whenever a frame is created/destroyed. So this code maintains a cache at the GC level instead of at the face level. There is an upper limit on the size of the cache, after which we will stop creating GCs and start reusing them (reusing the least-recently- used ones first). So if faces get changed, their GCs will eventually be recycled. Also more sharing of GCs is possible. This code uses hashtables. It could be that, if the cache size is small enough, a linear search might be faster; but I doubt it, since we need `equal' comparisons, not `eq', and I expect that the optimal cache size will be ~100. Written by jwz, 14 jun 93 Hacked by William Perry, apr 2000 */ #include <config.h> #include "lisp.h" #include "hash.h" #include "gccache-gtk.h" #define GC_CACHE_SIZE 100 #define GCCACHE_HASH struct gcv_and_mask { GdkGCValues gcv; unsigned long mask; /* contains a GdkGCValuesMask bitmask. */ }; struct gc_cache_cell { GdkGC *gc; struct gcv_and_mask gcvm; struct gc_cache_cell *prev, *next; }; struct gc_cache { GdkWindow *window; /* used only as arg to XCreateGC */ int size; struct gc_cache_cell *head; struct gc_cache_cell *tail; #ifdef GCCACHE_HASH struct hash_table * table; #endif int create_count; int delete_count; }; #ifdef GCCACHE_HASH static Hashcode gc_cache_hash (const void *arg) { const struct gcv_and_mask *gcvm = (const struct gcv_and_mask *) arg; unsigned long *longs = (unsigned long *) &gcvm->gcv; Hashcode hash = gcvm->mask; unsigned int i; /* This could look at the mask and only use the used slots in the hash code. That would win in that we wouldn't have to initialize every slot of the gcv when calling gc_cache_lookup. But we need the hash function to be as fast as possible; some timings should be done. */ for (i = 0; i < (sizeof (GdkGCValues) / sizeof (unsigned long)); i++) hash = (hash << 1) ^ *longs++; return hash; } #endif /* GCCACHE_HASH */ static int gc_cache_eql (const void *arg1, const void *arg2) { /* See comment in gc_cache_hash */ const struct gcv_and_mask *gcvm1 = (const struct gcv_and_mask *) arg1; const struct gcv_and_mask *gcvm2 = (const struct gcv_and_mask *) arg2; return !memcmp(&gcvm1->gcv, &gcvm2->gcv, sizeof(gcvm1->gcv)) && gcvm1->mask == gcvm2->mask; } struct gc_cache * make_gc_cache (GtkWidget *widget) { struct gc_cache *cache = xnew (struct gc_cache); cache->window = widget->window; cache->size = 0; cache->head = cache->tail = 0; cache->create_count = cache->delete_count = 0; #ifdef GCCACHE_HASH cache->table = make_general_hash_table (GC_CACHE_SIZE, gc_cache_hash, gc_cache_eql); #endif return cache; } void free_gc_cache (struct gc_cache *cache) { struct gc_cache_cell *rest, *next; rest = cache->head; while (rest) { gdk_gc_destroy(rest->gc); next = rest->next; xfree (rest, struct gc_cache_cell *); rest = next; } #ifdef GCCACHE_HASH free_hash_table (cache->table); #endif xfree (cache, struct gc_cache *); } GdkGC * gc_cache_lookup (struct gc_cache *cache, GdkGCValues *gcv, unsigned long mask) { struct gc_cache_cell *cell, *next, *prev; struct gcv_and_mask gcvm; if ((!!cache->head) != (!!cache->tail)) ABORT (); if (cache->head && (cache->head->prev || cache->tail->next)) ABORT (); /* Gdk does not have the equivalent of 'None' for the clip_mask, so we need to check it carefully, or gdk_gc_new_with_values will coredump */ if ((mask & GDK_GC_CLIP_MASK) && !gcv->clip_mask) { mask &= ~GDK_GC_CLIP_MASK; } gcvm.mask = mask; gcvm.gcv = *gcv; /* this copies... */ #ifdef GCCACHE_HASH if (gethash (&gcvm, cache->table, (const void **) &cell)) #else /* !GCCACHE_HASH */ cell = cache->tail; /* start at the end (most recently used) */ while (cell) { if (gc_cache_eql (&gcvm, &cell->gcvm)) break; else cell = cell->prev; } /* #### This whole file needs some serious overhauling. */ if (!(mask | GDK_GC_TILE) && cell->gcvm.gcv.tile) cell = 0; else if (!(mask | GDK_GC_STIPPLE) && cell->gcvm.gcv.stipple) cell = 0; if (cell) #endif /* !GCCACHE_HASH */ { /* Found a cell. Move this cell to the end of the list, so that it will be less likely to be collected than a cell that was accessed less recently. */ if (cell == cache->tail) return cell->gc; next = cell->next; prev = cell->prev; if (prev) prev->next = next; if (next) next->prev = prev; if (cache->head == cell) cache->head = next; cell->next = 0; cell->prev = cache->tail; cache->tail->next = cell; cache->tail = cell; if (cache->head == cell) ABORT (); if (cell->next) ABORT (); if (cache->head->prev) ABORT (); if (cache->tail->next) ABORT (); return cell->gc; } /* else, cache miss. */ if (cache->size == GC_CACHE_SIZE) /* Reuse the first cell on the list (least-recently-used). Remove it from the list, and unhash it from the table. */ { cell = cache->head; cache->head = cell->next; cache->head->prev = 0; if (cache->tail == cell) cache->tail = 0; /* only one */ gdk_gc_destroy (cell->gc); cache->delete_count++; #ifdef GCCACHE_HASH remhash (&cell->gcvm, cache->table); #endif } else if (cache->size > GC_CACHE_SIZE) ABORT (); else { /* Allocate a new cell (don't put it in the list or table yet). */ cell = xnew (struct gc_cache_cell); cache->size++; } /* Now we've got a cell (new or reused). Fill it in. */ memcpy (&cell->gcvm.gcv, gcv, sizeof (GdkGCValues)); cell->gcvm.mask = mask; /* Put the cell on the end of the list. */ cell->next = 0; cell->prev = cache->tail; if (cache->tail) cache->tail->next = cell; cache->tail = cell; if (! cache->head) cache->head = cell; cache->create_count++; #ifdef GCCACHE_HASH /* Hash it in the table */ puthash (&cell->gcvm, cell, cache->table); #endif /* Now make and return the GC. */ cell->gc = gdk_gc_new_with_values (cache->window, gcv, (GdkGCValuesMask) mask); /* debug */ assert (cell->gc == gc_cache_lookup (cache, gcv, mask)); return cell->gc; }