Terence Parr
|
80560efcf8
|
wow. i think i got threaded working. rm readwrite lock, sync add add to shared context and add dfa state
|
2012-07-31 19:39:31 -07:00 |
|
Terence Parr
|
3d2cbe087d
|
refactoring
|
2012-07-31 18:54:00 -07:00 |
|
Terence Parr
|
abb268416e
|
Merge branch 'master' into main
|
2012-07-31 18:34:33 -07:00 |
|
Terence Parr
|
b8f7acdeac
|
refactor to create DFAState target, fill it in, then add to DFA only after it's complete. Combine some helper methods. simpler.
|
2012-07-31 18:26:39 -07:00 |
|
Terence Parr
|
fedb7c105b
|
tweaks
|
2012-07-30 22:24:12 -07:00 |
|
Terence Parr
|
85ea296632
|
added D2 as dup temp of D state. almost ready to change position of addDFAState
|
2012-07-30 22:15:09 -07:00 |
Sam Harwell
|
8c820269e4
|
Fix IntegerList.binarySearch
|
2012-07-30 19:24:20 -05:00 |
|
Terence Parr
|
c893f2af08
|
Pulled Sam's new IntegerList
|
2012-07-30 15:20:43 -07:00 |
|
Sam Harwell
|
dcdcad8feb
|
Add IntegerList.addAll overload for Collection<Integer>, add contains, sort, hashCode, equals, toString.
|
2012-07-30 15:17:02 -05:00 |
|
Sam Harwell
|
3bf99d6d88
|
Add IntervalSet.toIntegerList(), use IntegerStack to hold mode stack in Lexer
|
2012-07-30 15:13:26 -05:00 |
|
Sam Harwell
|
fa62570737
|
Add IntegerList and IntegerStack, behaves like ArrayList<Integer> and ArrayDeque<Integer> (mostly) except doesn't require boxing
|
2012-07-30 15:13:26 -05:00 |
|
Terence Parr
|
c0ece0bd09
|
add read/write lock to lexer.
|
2012-07-29 17:21:10 -07:00 |
|
Terence Parr
|
8836e13954
|
use ReentrantReadWriteLock to sync. works much better but still occasional illegal monitor state :(
|
2012-07-29 17:01:41 -07:00 |
|
Terence Parr
|
f5f6715a07
|
back to core sync around creating DFA in decisionToDFA. not thread-safe for dfa update.
|
2012-07-29 14:33:18 -07:00 |
|
Terence Parr
|
9bcfc83678
|
adding full LL prediction in DFAState from ctx -> predicted alt. Drops from 17s to 12s and costs only 4M more ram. woot!
|
2012-07-29 13:30:18 -07:00 |
|
Terence Parr
|
79594a942e
|
synchronized lexer. just as fast now to build new parser/lexer each iter in TestJavaLR
|
2012-07-29 10:48:08 -07:00 |
|
Terence Parr
|
ceab49e3ce
|
rm for now
|
2012-07-29 10:12:09 -07:00 |
|
Terence Parr
|
e3e739dfc7
|
The lexer and parser ATN simulators' adaptivePredict now synchronize on the specific DFA of the decision to be simulated. This should prevent a lot of contention that would occur if we synchronize the entire adaptivePredict method. When the individual DFA are created, we also synchronize on the shared DFA[] table quickly to create a DFA and insert it into the array. Code generation modified to have _decisionToDFA generated at the top of both the parser and the lexer. Simulators created now with the recognizer, ATN, DFA[]. Not sure the LexerInterp/ParserInterp work but pushing ahead anyway for the moment.
|
2012-07-29 09:49:35 -07:00 |
|
Terence Parr
|
8c8752a7b2
|
print file names if > 1 file
|
2012-07-28 18:33:21 -07:00 |
|
Terence Parr
|
3f26d42d53
|
add Sam's getConflictingAlts potential bug fix as comment
|
2012-07-28 18:33:05 -07:00 |
|
Terence Parr
|
68275eb998
|
rm singleton set; didn't help.
|
2012-07-28 13:23:52 -07:00 |
|
Terence Parr
|
ee233f7dd3
|
add -SLL option, allow multiple input files, reuse same parser/lexer.
|
2012-07-28 13:07:51 -07:00 |
|
Terence Parr
|
75fd3264ff
|
adding new equivalence class support code and new SingletonSet
|
2012-07-28 11:53:40 -07:00 |
|
Terence Parr
|
680df17280
|
rm transient context cache for closure computations; slight speed improvement maybe. less complex.
|
2012-07-28 10:51:07 -07:00 |
|
Terence Parr
|
eda95f7478
|
DFA cache on again; dropped from 25 to 16M!!!
|
2012-07-28 10:34:34 -07:00 |
|
Terence Parr
|
8a301f59f8
|
try making smaller config list.
|
2012-07-28 10:29:04 -07:00 |
|
Terence Parr
|
2ee5a2f3ae
|
release entire ConfigHashSet upon readonly. sames some mem and is faster.
|
2012-07-27 22:38:43 -07:00 |
|
Terence Parr
|
7ff5544573
|
wipe out config lookup in ATNConfigSet when it goes readonly.
|
2012-07-27 22:20:23 -07:00 |
|
Terence Parr
|
54fd28c29d
|
add() was wrong
|
2012-07-27 21:03:41 -07:00 |
|
Terence Parr
|
8279c3da11
|
turn on local context cache.
|
2012-07-27 20:53:10 -07:00 |
|
Terence Parr
|
14372f2515
|
only 14 unit tests fail.
|
2012-07-27 17:38:02 -07:00 |
|
Terence Parr
|
b8f072e231
|
rebuilt new hashtable as T[][]
|
2012-07-27 17:18:33 -07:00 |
|
Terence Parr
|
b843d6cf59
|
initial new hash table impl.
|
2012-07-27 14:51:07 -07:00 |
|
Terence Parr
|
b83c90af2f
|
play with atnconfigset fields
|
2012-07-27 14:50:36 -07:00 |
|
Terence Parr
|
9445d1fdd6
|
update comment
|
2012-07-27 10:38:43 -07:00 |
|
Terence Parr
|
26165c5d54
|
atnconfigsets can be readonly now. no copying to put into new DFA state, set readonly instead. made atnconfigset.equals() more accurate. add -notree option to test java lr
|
2012-07-27 10:34:59 -07:00 |
|
Terence Parr
|
093af0edce
|
update comment
|
2012-07-26 18:26:17 -07:00 |
|
Terence Parr
|
6322ff3abd
|
add comment
|
2012-07-26 18:17:05 -07:00 |
|
Terence Parr
|
0e24a66f07
|
reduce memory usage by creating LexerATNConfig a subclass with the lexer action index. saves about 2M on java.* with javalr.
|
2012-07-26 17:51:02 -07:00 |
|
Terence Parr
|
3c7b4c2a33
|
big cleanup.
|
2012-07-26 17:28:10 -07:00 |
|
Terence Parr
|
6b8e6cf0a7
|
reduces a few equals calls.
|
2012-07-26 10:58:11 -07:00 |
|
Terence Parr
|
faafe8dfa2
|
no local/shared predctx caches. remove spurious array creation.
|
2012-07-26 10:35:52 -07:00 |
|
Terence Parr
|
86a4f26596
|
still has dfa and local predict ctx cache, but doesn't help much, and takes time. Eval preds as we pass them in full ctx mode; no sempred ctx in configs then. removed LANDMINE!!!!!!!!! some lame code that uses fields like _startIndex; fix that. added bail option to testjavalr. 18s to parse java.* with javalr full ctx now.
|
2012-07-25 18:47:34 -07:00 |
|
Terence Parr
|
05de714987
|
add chk for null contextCache.
|
2012-07-25 10:58:04 -07:00 |
|
Terence Parr
|
3774c5752b
|
add more tracking var
|
2012-07-25 10:30:19 -07:00 |
|
Terence Parr
|
3cec976f7e
|
Don't pursue the closure if there is just one state or it's predicting unique alt. Fixed up computeReachSet() to do this. Dropped from 6s to 4.5s for TestJavaLR -SLL on java.* with 45M max footprint.
|
2012-07-24 18:45:23 -07:00 |
|
Terence Parr
|
2495ce6b20
|
add SLL option to the ATN simulator and use in TestJavaLR.
|
2012-07-24 17:42:47 -07:00 |
|
Terence Parr
|
2ae4d4eebd
|
add ctx cache to merge op. now i get optimal graph it seems. doesn't seem longer for java.* with JavaLR. still landmine in Pattern.java, Foo.java. Made toDOTString sort predctx nodes by id for repeatable tests. updated unit tests.
|
2012-07-24 16:26:01 -07:00 |
|
Terence Parr
|
ac4f00524e
|
add ctx cache.
|
2012-07-24 14:19:43 -07:00 |
|
Terence Parr
|
9627652b67
|
tweak to compile
|
2012-07-24 12:20:56 -07:00 |