github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-06-28 16:41:01 +03:00
Code Issues Packages Projects Releases Wiki Activity
2,143 Commits 9 Branches 39 Tags
Commit Graph

87 Commits

Author SHA1 Message Date
ypicchi-arm
5145b6d2ab
Fix noodle sve2 off by one (#313) 
* Revert "Fix noodle SVE2 off by one bug"

This patch was fixing the bug when it happens at the end of the buffer
but it wasn't fixing it when we do scanDoubleOnce before the main loop

The next patch fix this bug for both case instead

This reverts commit 48dd0e5ff0bc1995d62461c92cfb76d44d1d0105.

* Fix noodle spurious match with \0 chars for SVE2

When sve2's noodle process a non full vector (before the main loop or
at the end of it), a fake \0 was being parsed, trigerring a match for
pattern that ended with \0. This patch fix this.

Signed-off-by: Yoan Picchi <yoan.picchi@arm.com>

---------

Signed-off-by: Yoan Picchi <yoan.picchi@arm.com>
 2024-08-29 13:49:29 +03:00
ypicchi-arm
aa4bc24439
Fix noodle SVE2 off by one bug (#309) 
By using svmatch on 16 bit lanes with a 8 bit predicate, we end up
including an undefined character in the pattern checks. The inactive
lane after load contains an undefined value, usually \0. Patterns
using \0 as the last character would then match this spurious
character, returning a match beyond the buffer's end. The fix checks
for such matches and rejects them.

Signed-off-by: Yoan Picchi <yoan.picchi@arm.com>
 2024-08-05 09:42:56 +03:00
Konstantinos Margaritis
c837925087
Fix/Suppress remaining Cppcheck warnings (#291) 
Fix/suppress the following cppcheck warnings:

* arithOperationsOnVoidPointer
* uninitMember
* const*
* shadowVariable
* assignmentIntegerToAddress
* containerOutOfBounds
* pointer-related warnings in Ragel source
* missingOverride
* memleak
* knownConditionTrueFalse
* noExplicitConstructor
* invalidPrintfArgType_sint
* useStlAlgorithm
* cstyleCast
* clarifyCondition
* VSX-related cstyleCast
* unsignedLessThanZero 

Furthermore, we added a suppression list to be used, which also includes the following:
* missingIncludeSystem
* missingInclude
* unmatchedSuppression
 2024-05-27 12:23:02 +03:00
Konstantinos Margaritis
cebc6541c1
Part 5 of C-style cast cppcheck (#289) 
Fixes some cstyleCasts part 5

closes some: #252
 2024-05-24 23:24:58 +03:00
Yoan Picchi
938c026256 Speed up truffle with 256b TBL instructions 
256b wide SVE vectors allow some simplification of truffle.
Up to 40% speedup on graviton3. Going from 12500 MB/s to 17000 MB/s
onhe microbenchmark.
SVE2 also offer this capability for 128b vector with a speedup around
25% compared to normal SVE

Add unit tests and benchmark for this wide variant

Signed-off-by: Yoan Picchi <yoan.picchi@arm.com>
 2024-05-22 16:13:53 +00:00
gtsoul-tech
94eff4aa60 cstylecasts and suppressions 2024-05-22 10:16:56 +03:00
gtsoul-tech
e111684bc2 fix cStyleCasts 2024-05-20 14:54:35 +03:00
Konstantinos Margaritis
e819cb1100 Fix C-style casts 2024-05-16 12:03:42 +03:00
Konstantinos Margaritis
22166ed948 Fix remaining marked as done const* cppcheck warnings 2024-05-15 10:52:31 +03:00
gtsoul-tech
94b17ecaf2 noExplicitConstructor 2024-05-10 10:07:47 +03:00
Konstantinos Margaritis
0d2f9ccbaa Fix 'unqualified call to std::move' errors in clang 15+ 2023-10-03 20:24:39 +03:00
Danila Kutenin
49eb18ee4f Optimize vectorscan for aarch64 by using shrn instruction 
This optimization is based on the thread
https://twitter.com/Danlark1/status/1539344279268691970 and uses
shift right and narrow by 4 instruction https://developer.arm.com/documentation/ddi0596/2020-12/SIMD-FP-Instructions/SHRN--SHRN2--Shift-Right-Narrow--immediate--

To achieve that, I needed to redesign a little movemask into comparemask
and have an additional step towards mask iteration. Our benchmarks
showed 10-15% improvement on average for long matches.
 2022-06-26 22:55:45 +00:00
Danila Kutenin
9af996b936 Fix all ASAN issues in vectorscan 2022-02-18 17:14:51 +00:00
Konstantinos Margaritis
81fba99f3a fix SVE2 build after the changes 2021-11-25 18:48:24 +02:00
Konstantinos Margaritis
210295a702 remove vermicelli.h and replace it with vermicelli.hpp 2021-11-02 22:30:53 +02:00
Konstantinos Margaritis
bc1a1127cf add new include file 2021-11-01 16:28:50 +00:00
Konstantinos Margaritis
713aaef799 move casemask helper functions to separate header 2021-11-01 16:05:43 +00:00
Konstantinos Margaritis
4e044d4142 Add missing copyright info from tampered files 2021-10-12 11:51:35 +03:00
George Wort
df926ef62f Implement new Vermicelli16 acceleration functions using SVE2. 
The scheme utilises the MATCH and NMATCH instructions to
scan for 16 characters at the same rate as vermicelli
scans for one.

Change-Id: Ie2cef904c56651e6108593c668e9b65bc001a886
 2021-10-12 11:51:34 +03:00
George Wort
c7086cb7f1 Add SVE2 support for dvermicelli 
Change-Id: I056ef15e162ab6fb1f78964321ce893f4096367e
 2021-10-12 11:51:34 +03:00
Konstantinos Margaritis
e35b88f2c8 use STL make_unique, remove wrapper header, breaks C++17 compilation 2021-10-12 11:51:34 +03:00
George Wort
9fb79ac3ec Add SVE2 support for vermicelli 
Change-Id: Ia025de53521fbaefe5fb1e4425aaf75c7d80a14e
 2021-10-12 11:51:34 +03:00
George Wort
7162446358 Remove possibly undefined behaviour from Noodle. 
Change-Id: I9a7997cea6a48927cb02b00c5dba5009bbf83850
 2021-10-12 11:51:34 +03:00
George Wort
b48ea2c1a6 Remove first check from scanDouble Noodle. 
Change-Id: I00eabb3cb06ef6a2060df52c26fa8591907a2711
 2021-10-12 11:51:34 +03:00
Konstantinos Margaritis
41ff0962c4 minor fixes 2021-10-12 11:51:34 +03:00
Konstantinos Margaritis
2753dbb3b0 rename supervector class header, use dup_*() functions names instead of set1_*(), minor fixes 2021-10-12 11:51:34 +03:00
George Wort
d1009e8830 Fix error in initial noodle double final call. 
Change-Id: Ie044988f183b47e0b2f1eed3b4bd23de75c3117d
 2021-10-12 11:51:34 +03:00
George Wort
d6df8116a5 Add SVE2 support for noodle 
Change-Id: Iacb7d1f164bdd0ba50e2e13d26fe548cf9b45a6a
 2021-10-12 11:51:34 +03:00
Konstantinos Margaritis
e215157a21 move definitions elsewhere 2021-10-12 11:51:34 +03:00
Konstantinos Margaritis
05c7c8e576 move SuperVector versions of noodleEngine scan functions to _simd.hpp file 2021-10-12 11:51:34 +03:00
Konstantinos Margaritis
c6406bebde simplify scanSingleMain() and scanDoubleMain() 2021-10-12 11:51:34 +03:00
Konstantinos Margaritis
f77837130d delete separate implementations 2021-10-12 11:51:34 +03:00
Konstantinos Margaritis
ede2b18564 add generic SIMD implementation 2021-10-12 11:51:34 +03:00
Konstantinos Margaritis
7a9a2dd0dc convert to C++ 2021-10-12 11:51:34 +03:00
Konstantinos Margaritis
831091db9e fix typo 2021-10-12 11:51:33 +03:00
Konstantinos Margaritis
ec5531a6b1 minor optimizations 2021-10-12 11:50:32 +03:00
Konstantinos Margaritis
d3ff893871 prefetch works best when addresses are 64-byte aligned 2021-10-12 11:50:32 +03:00
Konstantinos Margaritis
58cface115 optimise case handling 2021-10-12 11:50:32 +03:00
Konstantinos Margaritis
e3e101b412 simplify and make scanSingle*()/scanDouble*() more uniform 2021-10-12 11:50:32 +03:00
Konstantinos Margaritis
2f13ad0674 optimize caseMask handling 2021-10-12 11:50:32 +03:00
Konstantinos Margaritis
27bd09454f use correct function names for AVX512, fix build failure 2021-02-15 13:54:19 +02:00
Konstantinos Margaritis
9fd94e0062 use unaligned loads for short scans 2021-02-11 14:21:57 +02:00
Konstantinos Margaritis
d3e03ed88a optimize case mask AND out of the loop 2021-02-10 13:29:45 +02:00
Konstantinos Margaritis
5333467249 fix names, use own intrinsic instead of explicit _mm* ones 2020-09-23 11:51:21 +03:00
Konstantinos Margaritis
8ed5f4ac75 fix include paths for masked_move 2020-09-18 12:55:57 +03:00
Wang Xiang W
f658c4e149 Noodle: avoid an extra convert instruction 
fixes github issue #221
 2020-05-25 13:46:42 +00:00
Hong, Yang A
23e5f06594 add new Literal API for pure literal expressions: 
Design compile time api hs_compile_lit() and hs_compile_lit_multi()
to handle pure literal pattern sets. Corresponding option --literal-on
is added for hyperscan testing suites. Extended parameters and part of
flags are not supported for this api.
 2019-08-13 14:51:38 +08:00
Hong, Yang A
f68723a606 literal matching: separate path for pure literal patterns 2019-01-21 09:59:22 +08:00
Justin Viiret
af519f3190 hwlm_build: default for HWLMProto::make_small 
Silences Coverity warning.
 2017-09-18 13:29:34 +10:00
Alex Coyte
41783fe912 more comments on hwlm/fdr's start parameter 2017-08-21 11:23:41 +10:00