351 Commits

Author SHA1 Message Date
Justin Viiret
ac858cd47c rose: build a separate delay rebuild matcher 2017-04-26 14:46:48 +10:00
Justin Viiret
5706acf5c0 role aliasing: bail if new graph is not implementable 2017-04-26 14:46:46 +10:00
Alex Coyte
bbd64f98ae allow streams to marked as exhausted in more cases
At stream boundaries, we can mark streams as exhausted if there are no
groups active and there are no other ways to report matches. This allows us
to stop maintaining the history buffer on subsequent stream writes.
Previously, streams were only marked as exhausted if a pure highlander case
reported all patterns or the outfix in a sole outfix case died.
2017-04-26 14:44:53 +10:00
Alex Coyte
c32d7d51d9 remove ng_rose 2017-04-26 14:44:49 +10:00
Alex Coyte
d89cf2f699 tidy 2017-04-26 14:44:29 +10:00
Alex Coyte
7767651b59 shift all early_dfa creation logic to ng_violet/ng_rose 2017-04-26 14:44:29 +10:00
Alex Coyte
9d35e2ad36 allow outfixes to be converted to prefixes of the eod event literal 2017-04-26 14:44:03 +10:00
Alex Coyte
512c049493 shift early_dfa construction earlier 2017-04-26 14:44:03 +10:00
Alex Coyte
caa46201f0 add information on the runtime impl to rose.txt 2017-04-26 14:44:03 +10:00
Justin Viiret
c6b2563df6 rose: delete literal_info requires_explode flag 2017-04-26 14:43:28 +10:00
Justin Viiret
f964801923 rose: explode mixed-case literals early 2017-04-26 14:43:28 +10:00
Justin Viiret
f307956584 rose: do not combine fragments which squash groups 2017-04-26 14:41:30 +10:00
Justin Viiret
c31c1c366e filterLits: better debug output 2017-04-26 14:41:30 +10:00
Justin Viiret
988ea6b4e1 rose: role aliasing improvements
These changes improve the performance of the role aliasing passes on
very large cases and fix a couple of small errors in the left and right
merge passes as well.
2017-04-26 14:41:30 +10:00
Justin Viiret
1584da9025 rose: don't accel noodle engines 2017-04-26 14:41:29 +10:00
Justin Viiret
68a35ff3b8 rose: give longer literals to accel analysis 2017-04-26 14:41:29 +10:00
Justin Viiret
2fda8c0b20 hwlm: move accel from hwlm build to rose 2017-04-26 14:41:29 +10:00
Justin Viiret
abcac24b47 rose: add more program dump code 2017-04-26 14:41:29 +10:00
Justin Viiret
eb14792a63 rose: group final ids by fragment 2017-04-26 14:41:29 +10:00
Justin Viiret
07a6b6510c rose/hwlm: limit literals to eight bytes
Rework HWLM to work over literals of eight bytes ("medium length"),
doing confirm in the Rose interpreter.
2017-04-26 14:41:29 +10:00
Justin Viiret
5c9c540424 rose: fix up comments referring to CHECK_LITERAL
This instruction is now called CHECK_LONG_LIT.
2017-04-26 14:41:29 +10:00
Justin Viiret
5061b76901 rose: mark RoseInstrCheckLongLit ctor explit 2017-04-26 14:41:29 +10:00
Justin Viiret
8f1b3c89fa rose: remove no-longer-used convertBadLeaves pass 2017-04-26 14:41:29 +10:00
Matthew Barr
bc2f336d9d Work around for deficiency in C++11/14/17 standard
As explained to us by STL at Microsoft (the author of their
vector), there is a hole in the standard wrt the vector copy
constructor, which always exists even if it won't compile.
2017-04-26 14:41:29 +10:00
Justin Viiret
68bdc800fc dump: render literals as regexes (with comments) 2017-04-26 14:41:29 +10:00
Alex Coyte
f9324febde Ensure the queue structure is initialised in roseEnginesEod(). 2017-03-01 13:05:10 +11:00
Alex Coyte
734eb2ce62 we can only trim lookarounds based on information common to all literals 2017-01-17 11:39:17 +11:00
Matthew Barr
2214296b7f Convert compile-time code to not require SIMD 2016-12-14 15:29:01 +11:00
Justin Viiret
e271781d95 multibit, fatbit: make _size build-time only
This commit makes mmbit_size() and fatbit_size compile-time only, and
adds a resource limit for very large multibits.
2016-12-14 15:28:54 +11:00
Alex Coyte
e51b6d23b9 introduce Sheng-McClellan hybrid 2016-12-14 15:27:18 +11:00
Justin Viiret
ef99ae108f rose_build_merge: correctly merge NFA outfixes
We were not doing our bookkeeping properly for merges where the number
of NFAs was greater than the batch size of 200.
2016-12-02 11:34:52 +11:00
Matthew Barr
99e14df117 Fix combine2x128 2016-12-02 11:33:48 +11:00
Alex Coyte
8ff7a3cdbb correct dump filenames of som rev engines 2016-12-02 11:33:44 +11:00
Alex Coyte
32c826e9c6 have single dump function per engine 2016-12-02 11:32:36 +11:00
Alex Coyte
71ff480b77 nfa_api: remove subtype from dispatch 2016-12-02 11:32:28 +11:00
Alex Coyte
530d84c6f3 allow edge_descriptors to be created from pair<edge_descriptor, bool> 2016-12-02 11:32:20 +11:00
Alex Coyte
e1e9010cac Introduce custom adjacency-list based graph 2016-12-02 11:31:33 +11:00
Justin Viiret
29472c7b71 rose_dump: remove stray newline 2016-12-02 11:28:20 +11:00
Justin Viiret
91a7ce1cda getData256(): data needs to be 32-byte aligned 2016-12-02 11:28:16 +11:00
Alex Coyte
47f53f63a7 simple pass to pick up paths redundant with those from cyclic's succs 2016-12-02 11:24:30 +11:00
Justin Viiret
8cadba0bdd rose: call loadLongLiteralState() earlier
The ll_buf, ll_buf_nocase buffers must be initialised before anyh path
that could lead to storeLongLiteralState().
2016-12-02 11:23:14 +11:00
Justin Viiret
b9650d4fd0 rose: don't unconditionally init ll_buf etc
This is only necessary (and already always done) if there is a long
literal table.
2016-12-02 11:22:27 +11:00
Alex Coyte
592ce06eeb Create combo tops for trigger limexes 2016-12-02 11:22:23 +11:00
Alex Coyte
445cf987a8 remove unused includes 2016-10-28 14:52:56 +11:00
Justin Viiret
8869dee643 rose: simplify long lit table, add bloom filter
Replaces the original long lit hash table (used in streaming mode) with a
smaller, simpler linear probing approach. Adds a bloom filter in front
of it to reduce time spent on false positives.

Sizing of both the hash table and bloom filter are done based on max
load.
2016-10-28 14:52:45 +11:00
Justin Viiret
68bf473e2e fdr: move long literal handling into Rose
Move the hash table used for long literal support in streaming mode from
FDR to Rose, and introduce new instructions CHECK_LONG_LIT and
CHECK_LONG_LIT_NOCASE for doing literal confirm for long literals.

This simplifies FDR confirm, and guarantees that HWLM matchers will only
be used for literals < 256 bytes long.
2016-10-28 14:52:26 +11:00
Alex Coyte
648a3c4824 UE-3025: There is no need to prune tops from non-triggered graphs 2016-10-28 14:52:01 +11:00
Alex Coyte
c94899dd44 allow sets of tops on edges 2016-10-28 14:51:46 +11:00
Matthew Barr
707fe675ea Operator precedence matters 2016-10-28 14:48:46 +11:00
Matthew Barr
7849b9d611 MSVC prefers the attrib at the beginning 2016-10-28 14:48:31 +11:00