Commits by Projects
NetworkManager
c4a86eba52e4
de9848aa66d2
50403cccee28
3751cceeec2b
de5d07392da4
b913e1d641e9
Use stack allocation18660604aae1
Do not rely on glib implementation4cdb84188154
e93abf0552cf
bac7a2821f3a
60cc501a665a
e7f3ccf7cdb5
9ecdba316cf8
Build the message without libnl objecte9321713a90a
Avoid string copy75d694db9b7d
53c69b1d6ec2
e17cd1d742c4
Avoid string copyf17a20c568c7
fedf7ca3034e
c3d98a3df643
d0330bfa73da
Also memoizationbafd0d557dd1
f8b74e19ea1e
Memoization when using new API5d1610b347ab
bb273c0881b3
1fb3d5d7944e
e3d2fc861b86
Fixing branch condition to avoid heavy computationb58481b31ec5
4417b8bf3eef
46b452eb5a1c
640736ff6505
142d83b01973
470bcefa5f25
c40acf334e90
Use uniond7b1a911d913
79458a558bdf
9440eefb6dc4
04c70c76bccf
2b43ce3a9406
Passenum
into the affected function0861f47a1c5d
Data structure59f2c0fb3e9a
Data structuref99723eda551
Data structure38273a88712c
Data structure (inheritance -> plain class)6a42e18d53ea
Data structure36d7a3cf21ab
Data structure22edeb5b691b
Data structure4be4a3c21f63
Data structure01a20015e0bf
Async32dfa563d1ea
Async1a070f6a44a0
Static checking instead of dynamic checkingbaa0d95c95c4
Static checking instead of dynamic checkingb957403efd53
Refactor to iterate the list at most twiceba1cc6a28803
Refactor to iterate the list at most twice62a4f2652f6a
broadcastAll
instead ofnotifySingle
bfe66fb8f410
Static checking instead of dynamic checking4dd6ab8f4b47
Algorithm92712d3b31b5
Allow compiler optimization7e79e913ad11
Faster fallbackee76b0979ff0
Algorithm9f79ae685d1b
6ce7b7df96e5
Algorithm7860ef959ae4
Compiler584a06e4e8e0
Algorithm
pulseaudio
c82e4913e866
a388b909f3a0
Use more efficient mapping when possible555388ebba9a
Use more efficient mapping when possible9c4dcffca5c7
Avoid waking the process up prematurely8fa81a93c984
fe455ae0137c
Split the function for path selection95b64804ab9d
Split the function for path selection78df02dba61d
192c3aaef835
Reduce the buffer to be sentf4ab8bd83530
e92e8b11f1f3
556cdfa1902d
Access the member directly1e4e586150b7
Initialize hardware mix path
grep
b408bcee1e81
Usememchr2()
to speed upmemchr_kwset()
, while inacexec_trans()
usememchr_kwset()
when possible6e319a818ed7
C won't check the rhs of a logic and operator if the lhs is false. Therefore in this patch we can skip the slow old libc implementation ofmbrtowc()
0ffd7d26dc8f
Posing fewer condition on fast pathf07b1b95dfc4
minimize the usage ofmemchr2()
296f3b994d59
Avoid memory allocation290ca116c917
Addtry_fgrep_pattern()
to do fast conversion of keys2a45b2fbe2ca
Avoid redundant memory allocationd67b7488d4ba
Avoid redundant memory allocation724ac5564bc8
d1160ec6d239
Make unibyte locale encode errors to avoid heavy weight error reportingf6de00f6cec3
Skip the hole in the file (Solaris only)9fa500407137
/16f6616acaa4
Skipping error bytes and quickly matching empty strings before entering heavypcre_exec()
a0404f7e4266
a0404f7e4266
7600fde5bcf5
c7c8bcdefe7b
bb3d1acbea88
Add fast-path for single character matchingf6603c4e1e04
5700656ed2ec
Treat "[x-x]" case as "[x]"d4c8de2f3e4f
8a33cdeba4ba
Initialization within certain condition14892aa6e0c2
Avoid usingmemchr()
when possiblefdf9fadd0316
4fa6f48b5732
aaefbfa486df
Skip easy bytesb342369b8f6d
Avoid mem allocation until neededd6c8415e9c8c
1e166fdaa699
Avoid duplicated operations on the same characterad468bbe3df0
523e085b6a0e
Add a loop to retry dfa superset immediately after a match maximize the usage of the fastpath for multiline matches3255bc58e8fb
dfafast
defines how end value for the search is set960ad317db21
Avoid some copy operations5cb49d2f375f
aa1f107a1439
Complicated optimization, where in the case of UTF8 the first byte may cause the slow build of DFA. Optimizing by early scan & cache192d61e2828e
Skip unnecessary scans79d7918c889d
Shortened previous word character matching search path225362ac3967
Set flag to avoid data format check and make unknown format optimization working55a0c73874bc
Invoke exec function with different arguments as some arguments cause the function slow561ffc0f203e
Change the sequence of strings to be matched and checked against42b14d981975
d795d6d194a5
Efficientlseek
can be used to shorten the search process by skipping holes in the filecd493ae8c13f
29b36e655aca
Use fastmap when "-i" is not specified61e0661e1b08
b58925630c90
Cache the states ofANYCHAR
b75ce6f7c611
28f6eed8db83
7c9d39eab21e
Adding superset of an dfa and use it before doing the full match of the pattern. If the string prematches the superset, it must match the original dfa as well.41ac7a996473
Const1a45e78b8759
Adding pure attribute for better compiler optimization256a4b494fe1
Process the sequence of the default case in bulk7eb003e1625a
inlining functionsed0c7d559baf
inlining functionsb639643840ef
Change fgrep and egrep to shell script67822cff1f9f
Split function call for compiler optimizationbc0732da3a2c
/671698472dea
Removing optimization that is rarely executed but still wasting time on branching954a3eb3d15f
Use ternary operatora3d42de483a2
Use Aho's algorithm with efficient trie building function from the Commentz-Walter algorithm757381e58d66
Avoid duplicated matching section97318f5e59a1
Convertgrep -i foobar file
intogrep [fF][oO][oO][bB][aA][rR]
to speed up matching4fa1971d98c7
4c57448135eb
Adding newmemchr2
api and use it
rsyslog
59e58b5ec8f7
Use a new API instead of polling the existing filer descriptorsb640a84242e8
usepoll()
instead ofselect()
edb81c0283bb
d3ef5a7bde0e
Use single string buffer to avoid dual buffer extra copying (while make memory accesses more cache efficient)b9dcd4e43426
4193dc495d9e
a000bd661e04
fa8ff7601de0
39b35a32b4a4
Avoidstrcmp()
in some cases123143d0f1f5
Avoidisdigit()
bb9650eb583f
eb97d25219a2
49d1203b3582
Avoid using slowiscntrl()
262d05999029
Build the string withoutsnprintf()
8339d54ddcbc
d748ba250ea9
f3106c538f04
7619ad2c20af
4af7e0616618
69e1be41ac23
Set a hash for strings to speed up following processing80d0ad708bfe
Early conversion of string for faster processing6c52f29d593a
Reduce the size of buffer of message submission33fe6e484c8b
eb5c7a041990
62161ae32d46
023becabd706
f75395172eac
440aaff11460
0d71694fb3cb
855b68e1be65
beed1bda6969
90e1478f11b7
Fast access01de5afe83d4
Add const for compiler optimization567469f94ad5
9273b4bb4dcb
Change mem layout for CPU cache hitb76e48711fc0
Timer5ed3beb5698e
make connection parallelea1860deb070
d741c3e218ac
Change of API used752d32f6005a
Change of API used447f92cc260b
Change of API used
lvm2
ada5733c5652
'Unblocking' clear55ca8043d4bb
2d1dbb9eddd9
217f3f8741c1
c086dfadc389
6dd0bd0255d5
e2c7e0ad1196
8e9d5d12ae3a
Avoid collocate whenpvmove()
598c82fc077c
3b604e5c8e50
9a060948242e
65d0089c6477
Check whether the major device number is in the list instead of iterating throughsysfs
f65dd341a561
Get VG metadata when it's synchronous with the data on disk2c4e8254de59
9712995edd60
99237f0908d8
18dfbbb150e0
6a41286c012e
4dc81848034f
0912cf67aae2
96626f64fa49
1ff7e214e0cf
420af27f088a
hash table -> radix tree719261a33a34
Read the first part of kmsg instead of the entire buffer0dc73f7dbd49
Constdd5d865020ac
Fix hang by adding timeout5577f2f4f0a5
|| instead of |bfeabea63178
Avoid multiple metadata check, but the code still double checks576dd1fc4125
772834e40aef
llvm
6c8d9bce0325
1610bcf8d924
bf0b5f5f3305
Static header selection549896a73688
ffc893deb710
95d021ba685c
092e5e75661f
e4f44dfa7398
22e14e2efdc3
04a303b8214f
Avoid copy16bb45c5c8e9
b0edfb8160e3
2638e45e83af
Specialization of a generic type to avoid buffer creationb6b25740b763
Unmerge GEP to reduce live values and register pressure16cba6923a59
c0f82e4a5178
1a8392e4835e
7e9986fa34a9
1e0b73ce8bf5
Remove unreachable blocks before optimization9f299abc0567
Drop unused tablebaf868b9b8d1
b6f73f8f0bee
UseSmallMapVector
f28cb39e4ca0
Use maps for efficient iteration427afb91d382
FnSet
->FnTree
e0714cfc0fec
Const vs non-constc4fbd3fd6268
Array accesses faster thanswitch case
8c955ea858b0
UsingFoldingSet
c5374d2ba55e
Iterating to avoid hashing/lookupsf4c3a5601aef
ab1f4ef9a2b7
Pass choosing algorithm3d019d384a40
Algoa9fbbb48bcc5
9ece8eb94564
Partial specialization of the more efficient implementation1ecf0813db3b
git
252d079cbd27
1a2e1a76ec2c
9a48a615b47d
4e38e9b1d05f
0cacebf099dd
5ebd9472a490
f23092f19e73
163ee5e63523
176841f0c9b4
af952dac7a69
a6eec1263883
e5494631ed94
9bc4222746a0
7eb6e10c9d7f
6df5762db354
dbbc93b221c6
8d6a7e9a1984
4e1d1a2eea25
2dacf26d0985
0e87b85683df
f35650dff6a4
c08e4d5b5cfa
e70a3030e747
3bc7a05b1a78
990a4fea9698
Add fast path for non-wild match12cd81743dc4
832258da969f
Limit buggy slow part in certain cases9831e92bfa83
Remove the slow APIb7099a06e8ff
1974f4791af2
Performance improvement when tracing is disabled6cbc478d83b5
2523c4be855a
3ecc7040eff2
ee7fd70edfc5
3446a59b3950
7a40a95eb4f7
0952ca8a959e
Fastfail6bf3b813486b
Skip diff large files645c432d6196
6a36e1e7bb64
31471ba21ee2
c6807a40dcd2
9ec726a4120b
adcc94a0aa70
Add cases for fast-forward4c30d50402c1
108f530385e9
9a9f243f6426
ca598d5f2ab9
ad635e82d600
da470981defc
5c9f9bf313e8
38ccaf93bbf5
df08eb357dd7
846df809bc67
f5b2dec1657e
41a078c60b82
a849735bfbf1
abb4bb83845d
b4da37380b77
d8febde370aa
Use tree structure0f9e62e0847c
Use bitmap to avoid traversal9a414486d9f0
Move the found item to the head of the collision list3fb9d685dbff
Change the type of parametersc44a4c650c66
82d0a8c05af8
62db5247790f
Work in plain C instead of shell script144e4cf7092e
7e6ac6e4391c
Algorithm36f0f344e71b
Repack files for better performancea7ddaa8eacb4
???d0da003d5b1e
5d19e8138d5f
Update promisor to do repack on these filesd5ca1ab395d4
6f92e5ff3cdc
???7b64d42d2220
Add string interning
clang
e86ee1a213d2
d0d4d66cd1f7
Use tooling instance instead of invoking the compiler externally9ac438062906
Quicker accesses to some datadf5f80f8a34e
4ed69c25ab95
Provides fast yet inaccurate declaration in libclang0a7a42e99096
baa7ca114299
ceadeb9d7861
951094baa05a
55b2d89ef0e2
Use a unified function instead of fast/slow path3c02b7312f41
5a04f9fc2b00
85fb4f5d7576
Fail fast instead of rollback562d45ce29e8
575c8ea65c2a
174cfde2479e
Use more efficient APIf20a280b3ba3
d0d4d66cd1f7
Do not start the frontend from driver26e65cec83b8
fb5f718cf1ca
Usebzero()
when possible4a18c3b25a2d
Only scan the entry list3ed5b235b114
9d33e30aba42
89aa3ede8e97
Fail fast instead of rollback72a418772e26
f78232dcbf73
007983114f44
7cea14876025
aa945900d543
3ee4cc4acfa6
41748f7d3ae3
0eaba5f9b9a6
29e517e070cc
d07865b42dcb
Precomputation3548068c22f8
e7a77727804c
ac0a256e8571
4c3cdee2a504
Not a simplecache = computation();
fix9bdffb989936
a6b00fc97669
a new class caching identifier information of each module039fc1ebe202
Complicated refactoring that causes duplicated traversals over arguments, which is then optimized by memorizing call arguments and IR argumentsccefdc8cb94e
4411b423e91d
f4f66331f0df
926f58b96ef5
953de1f092f5
Usingenum
instead of a function pointer499fc534bd02
Change sema map to extern C decls for lookupfae6a43ca3b1
Use another data structure90cdec1762c2
90ea10e5ea2f
Compute the json object without checking against raw json stringsdfa97d29ffee
/cf9fe978839a
/329e886c49d0
Create VFS overlay for replay lookupbdb673743a14
211035090970
/a5f80b2ea6d3
Shortest path findingac39f135ea0d
Unknown if it's performance relateda19dc41bd408
???3a67a19e28fc
Moving information into a container1c11b7cda0af
2f13eb116e62
Make recursive iterative0d816739a82d
Migration algorithm with performance benefits
gecko-dev
d75b9ed4bd65
04238b5088cb
5803056b7004
863402e5363e
Do not do the entire compilationc8c196a72313
c04bd7aa0e75
5f834ddfb2e1
1266c84e30ef
Lazy evaluation7bd233bc8795
d66b6c2654d3
9ea0de6975ea
5c1b11f1db48
Do not shutdown the device when there're events pending5bcb89ae72f9
Implementing faster rendering0edbe1a51235
661d20b507f4
e222ca710936
755b99f4f670
e53f74cb868d
a5dd4da99496
32dbe991d685
NewMaskSurface
facilityac0ae7443953
98cfb0e977a3
Avoid duplicated allocation761a458cda5c
Avoid repeated serialization/deserialization3a3d9ab7f782
1755aa59667c
214e03003f17
def1f6e7db52
fddb9e0b1d3b
Disable a function at all26105ef609ee
Move the rvalue reference046b03d3df31
Move the rvalue reference7fccf7772218
b106718dcb4e
0391a5005094
b6275dd517c1
ce7bb69227fd
e1f1db7272cc
Do not request remote processes when restoring input context6adaf4dfcb7f
655ade7a8b03
a7a4b9396837
Prefetching0ffa8c1cd8fa
86ed8740c135
fc3e532636e1
2c5b4e5e64a2
4c1bd947157e
Change hashtable to vector0ebdcdbf60d3
7ca3c89ebb0b
Data structure783e661627ee
59eb07a88fb7
Reduce the parameter passed into the functionc7503566b286
a2be039491cc
Merge all callbacks19d13525ebc7
Put more common case at the lhs of a logic operatordaab3fdaf71c
???8e3206382b84
Make the arguments passed through registers on x64ce294cea4266
185af4c21b68
Compiler optimization
openssl
a091e212fc55
c1b2569d234c
cfc32a1efb46
06b9ff06cc7f
Change of API to reduce the context switch latency3064b5513443
Avoid reseedingd5487a454c48
pl10bc3409459a
Static new algorithm pathcfc32a1efb46
Static new algorithm path0f3ab9a34c0a
Static new algorithm path13603583b3cc
Static new algorithm path0dd0be9408a4
Static new algorithm pathdc99b885ded3
Static new algorithm patha3ea6bf0ef70
b98530d6e09f
79dfc3ddfd80
Static new algorithm path5fc2c6896d50
e8b0dd57c0e9
afa23c46d966
c10e3f0cffb3
Avoid heavyHMAC()
function1fab06a66533
b3e02d06ba80
ae1ffe0f65c4
26b05245f0d3
0edb109f97c1
Shortened buffer length forChaCha20_ctr32()
848113a30b43
Have one call to a function instead of multiple callsa78324d95bd4
Change the field ofr
to reduce the latency ofbn_pollute()
be2c4d9bd9e8
New algorithm to speed up certain computation1d2a18dc5a3b
/a9c6edcde7f8
multiblock performance imrpovement, using larger buffer allocated7823d792d0ca
Use hash to check duplicationcb8164b05e3b
Const81f3d6323dcd
Split data for aligned and misaligned7caf122e717e
3fc7a9b96cbe
Revert3712436071c0
9d91530d2d7d
systemd
a9becdd65bb4
bf1d6be5b81c
8f2665a463ff
3c72b6ed4252
fd373593ba9b
aa40ff070391
539618a0ddc2
bf516294c89d
ed71f95662af
b872843c4db1
Generate fewer logsddb3706d2643
ff02f101cb7d
b4a11878f2fd
a57f7e2c828b
4fcb507a909c
244d2f07b49e
daa557208de1
4dc63c4bc785
7b943bb7e3e2
Faster API77fa610b22f3
fc9ae7178e14
6f8cbcdb27d7
fe902fa496ab
49bfc8774bf9
3bbaff3e0807
454f0f8680eb
8f1e0ad415da
fdb90ac6a6f3
ef9a3e3c2809
4d247a6cd3f6
422baca0f230
8372da448f3c
a6149b93afeb
da50b85af729
Use path as an indicator for the uid candidate. If we cannot use the uid of suggested paths, hash the path and get a random uid1086182d83d4
Test two sets linearly for equality instead of mutual superset93484b4790e2
New API that avoidsprintf()
styled formatting426bb5ddb8ff
d4cdbea5550a
Avoid waiting for the interface configuration at boot time708e25e137f8
5e5420d1a102
c79e98eadd30
Use memfds instead of shm7943f4227502
/487d37209b30
e32530cbef74
4b64536ee7fd
Alloca257f9d4a53e
2d058a87ffb2
cc25a67e2af2
bcf3ce7b3952
66a67effcc5b
acf7f253dedb
47b33c7d5284
496ae8c84b2d
00a8cf7763ec
6fe869c25179
Drop stability guarantee081f36d82de8
Elimiate spaces when encoding/decoding base64fb5c8216646e
2 msg commit for faster client conf8621b1109b0b
Set keep-free value to avoid some SSD's performance degradation when free space is less than this valueff0e7e05c990
8e060ec225b7
1930eed2a785
8580d1f73db3
Limit the number of filese12020472976
Filter the list then processing94243ef29942
df0ff1277588
/5c30a6d2b805
ae5b39588734
17c8de633faa
815b09d39b74
52b147841406
6414b7c98137
f268980d2cee
04c4d9199e04
Cache TLS/TFO written data5659958529d1
8a5cd31e5fa0
6febe75da765
Macro makes string length as constda0c630e141e
Data structure91ccab1e40a1
Add fast connectionc3950a9bbe0d
Algo6dbef3053dcb
5d67a7ae749d
Algo4b5bc5396c09
Use new compression algorithmb29ddfcb3891
Author does not know why
libgcrypt
f9371c026aad
6e669e09603e
/b6e6ace32444
a15c1def7e0f
Use optimized APIsca21a24808ef
3841b23c0ccb
656395ba4cf3
20886fdcb841
bf6d5b10cb41
49f52c67fb42
e950052bc6f5
Remove unaccelerated paths305cc878d395
Add accelerated aptah04cda6b7cc16
293e93672fda
4645f3728bb0
No need to copy all param86e5e06a97ae
Use proper sized objects so that GCC can optimize thatfe91a642c7c2
Signing parses shorter strings with the option634a85412a40
6571a6433183
c59a8ce51ceb
Use avx2 asm for faster twofish cipher computation5c418e597f0f
asm optimization2d2e5286d53e
asm optimization63ac3ba07dba
asm27747921cb1d
asm5418d9ca4c0e
asmde73a2e7237b
asm0b3ec359e227
Use intel asm for faster sha256d02958bd300d
Use intel asm for faster sha1da58a62ac1b7
Use avx asm for faster blake2saf7fc732f9a7
Use avx2 asm for faster blake2s93503c127a52
Use arm8/ce asm for faster aesa00c5b2988ce
Use aes-ni (an instruction set) asm for faster aesc9e9cb2eb6a1
Use aes-ni (an instruction set) asm for faster aesc85501af8222
b3ec0f752c92
aarch64 asm to accelerate chacha200b332c1aef03
2d4bbc0ad62c
e4eb03f56683
595251ad37bf
962b15470663
34c64eb03178
98f021961ee6
3d6334f8d94c
5d601dd57fcb
8353884bc65c
a1cc7bb15473
2857cb89c6dc
909644ef5883
2fd06e207dce
e11895da1f4a
4f46374502eb
30bd759f398f
323b1eb80ff3
297532602ed2
a39ee7555691
def7d4cad386
df629ba53a66
a5c2bbfe0db5
e4e458465b12
6fd0dd2a5f13
2e4253dc8eb5
be2238f68abc
69a6d0f9562f
e1a3931263e6
3ef21e7e1b80
f702d62d888b
b60f06f70227
Parallelize319ee14f2aab
Parallelize the computation305cc878d395
Parallelize the computation99d15543b8d9
Parallelize the computation2cb6e1f323d2
3ff9d2571c18
5a3d43485efd
2fd83faa876d
98674fdaa30a
1d85452412b6
c030e33533fb
6deb0ccdf718
Parallelizecafadc1e4fb9
31e4b1a96a07
8d1faf567145
f365961422f1
71dda4507053
d94ec5f5f8a5
e7ab4e1a7396
d325ab5d86e6
7317fcfadf00
0bdf26eea8cd
2374753938df
Data structure8725c99ffa41
Algo4de62d806442
copy block to stack for compiler optimization4b7451d3e8e7
likely/unlikelyefa9042f82ff
Block alignmente7b941c3de9c
Compiler optimizationbf9e0b79e620
Compiler optimization925d4fb3e8f2
Compiler optimizationf3e511610363
89fa74d6b3e5
Faster implementation with macro74184c28fbe7
f7505b550dd5
06e122baa332
Algobd4bd23a2511
Algo0e9e7d72f3c9
Algo51501b638546
Algoa48d07ccadee
Algo8e1c0f9b894c
Algoba892a0a874c
37d0a1ebdc2d
Algoec2f8de409a9
Algo
linux
cb0a8d23024e
Fallback tovmalloc()
when continueous memory cannot be quickly returned48ad1abef402
Use zram basedmemset_l()
a339b351304d
Avoid weighted affinity wake when the scheduler allows using idle CPUfbd02630c6e3
Avoid heavy weight__kmem_cache_free_bulk()
6622f5cdbaf3
Avoid linear search on the corresponding device0a672f74131d
03486830c577
Avoid size alignment9e641bdcfa4e
7fdd29d02e0a
Do not lock on the entire send path tree9344de1b1c64
Avoid getting cpuvar from thread hardware mapping4e6d47206c32
Split allocation in different placesf379fdf10b12
Use idr instead of list4f72adc50682
Just return the cycles, leaving human readable formatting to call site66feed61cdf6
Fast signalingbbddca8e8fac
Avoid lockinga60a132e89cd
Turn off the carrier earlier than data sync1ed7b84b28c1
Remove duplicated copy operation06e386a1db54
Avoid irq save operation586a32ac1d33
8be2748a40cb
a3795208b9a9
Validate without linearization2889caa92321
362108b5addd
Avoid lock09772d92cd5a
Avoid instrumenting extra function call in some conditionca47e8c72ae1
a7f58b9ecfd3
9226b5b440f2
2a4c32edd39b
Fastfail support to improve I/O latencycf0523350c6f
c435ee34551e
Avoid callback lookup on the server sidebef3efbeb897
Limit the rate of read which will broadcast SMI affecting the entire systemf872f5400cc0
Avoid returning a list ofstruct pages
so that arch code won't walk the entire list20b3d54ecba5
Add the proxy layer to issue fewer commands to the hardware on non DirectLPI casesc9fefa08190f
291a00d1a70f
Tweaking an argument for fast recoveryeec7e1c16d2b
Return the correct error code to minimize retriesb83bd893f3a0
Reduce the delay9f7714d46383
Introducing newstruct
to send memory pages to video controller351a4dedb34c
a5243dabb95c
6c308fecb4d1
d409014e4fee
12d82e4b0aa6
47df3ddedd22
bcd5642607ab
856661d514e8
2ab91adae722
1ce42845f987
7ef5ca4fe41e
3568bdf0419f
759f29b62fb9
2d550dbad83c
225785aec726
0c7c1bed7e13
Use RCU8cb7c15b3254
Put the loop into the CS to avoid repeated lock acquisitioncc91210cd2f6
Use RCU5a27aa822029
Introducing kref for lock optimization97334162e4d7
Prefetch8308756f45a1
Some microoptimization including avoiding locking on an internal structuree502a2af4c35
Use radix tree instead of hash table53e617e3dd5f
Data structure relatedb4f501916ce2
Use raw CPU ptrf7d9afea6cfb
Generate more efficient mips asm codefe31fca35d6a
Div optimization3d66c6ba3f97
Use per-cpu pointerabbdb5a74cea
Removingunlikely
9def970eada8
ad6048325c78
32-bit division faster than 64-bitbdfc7cbdeef8
2671e9fc62c3
Use atomic variable by default unlessREFCOUNT_FULL
is enabled57a2af6bbc7a
20dcfe1b7df4
Align cachelinec17b0aadb7d8
Weaker form of barrierbbe08c0a43e2
Microoptimizationd70b3ef54cea
Support write-through cached memory mappingcffb78b0e0b3
545ac13892ab