mirrors/emacs
1
Fork 0
mirror of git://git.sv.gnu.org/emacs.git synced 2026-01-06 03:40:56 -08:00
Code Issues Projects Releases Wiki Activity
5639 commits 243 branches 194 tags 712 MiB
Commit graph

75 commits

Author SHA1 Message Date
Gareth Rees
981a266642 New public functions mps_pool_total_size and mps_pool_free_size. 
Old (undocumented) functions mps_{mv,mvff,mvt}_size and mps_{mv,mvff,mvt}_free_size are now macros for the new public functions.
New pool methods PoolTotalSize and PoolFreeSize, with implementations for public pool classes except SNC.
Coverage of the new functions in apss and mpmss for the manual pool classes, and in finaltest for the automatic pool classes.

Copied from Perforce
 Change: 186118
 ServerID: perforce.ravenbrook.com
 2014-05-15 17:01:59 +01:00
Gareth Rees
cf427ca23c Merge branch/2014-04-08/align into branch/2014-05-15/size. 
Copied from Perforce
 Change: 186116
 ServerID: perforce.ravenbrook.com
 2014-05-15 14:54:36 +01:00
Gareth Rees
6980f345f1 Add note about potential for speed improvement, as requested by rb. 
Copied from Perforce
 Change: 186093
 ServerID: perforce.ravenbrook.com
 2014-05-13 18:04:59 +01:00
Gareth Rees
9b98c26ff3 Describe the block pool. 
Copied from Perforce
 Change: 186091
 ServerID: perforce.ravenbrook.com
 2014-05-13 17:49:35 +01:00
Gareth Rees
7cd4759316 Catch-up merge from branch/2014-03-30/addrset to branch/2014-04-15/mvffnoseg. 
Copied from Perforce
 Change: 186082
 ServerID: perforce.ravenbrook.com
 2014-05-13 16:27:17 +01:00
Gareth Rees
34cc9d49e4 Catch-up merge from master sources @186077 to branch/2014-03-30/addrset. 
Copied from Perforce
 Change: 186081
 ServerID: perforce.ravenbrook.com
 2014-05-13 16:21:39 +01:00
Gareth Rees
f4a4f1e250 Merge branch/2014-02-23/mvff-tune into branch/2014-04-15/mvffnoseg. 
Copied from Perforce
 Change: 186079
 ServerID: perforce.ravenbrook.com
 2014-05-13 16:10:51 +01:00
Gareth Rees
7620c6fcd1 Fix rash build on os x by adding the unused attribute to functions declared static that are not called in the rash variety. 
Copied from Perforce
 Change: 185915
 ServerID: perforce.ravenbrook.com
 2014-05-01 12:18:00 +01:00
David Lovemore
70c0cb9a94 Rename bool(v) to boolof(v) to fix clash with the windows header file windows.h 
Copied from Perforce
 Change: 185841
 ServerID: perforce.ravenbrook.com
 2014-04-28 13:14:04 +01:00
Gareth Rees
4b49fc7d57 Mvff no longer uses segments, but instead maintains the address ranges it has acquired from the arena in a cbs. 
Copied from Perforce
 Change: 185574
 ServerID: perforce.ravenbrook.com
 2014-04-15 17:52:14 +01:00
Gareth Rees
87b3880405 New generic function landsize returns the total size of ranges in a land (if the land supports it). implement it for all land classes. 
The MVFF pool class doesn't have to maintain its free size any more: it can just call LandSize.
Move re-entrancy protection from CBS to Land. This allows us to remove some CBS functions. (But requires some adjustment in failoverDelete.)
In MVFF, do more checking of mvff->total.

Copied from Perforce
 Change: 185569
 ServerID: perforce.ravenbrook.com
 2014-04-15 16:35:34 +01:00
Gareth Rees
4deafee125 Lands maintain the total size of the address ranges they maintain. (this avoids the need to do free size accounting in mvff.) 
Copied from Perforce
 Change: 185567
 ServerID: perforce.ravenbrook.com
 2014-04-15 14:23:53 +01:00
Gareth Rees
42f18f93af Catch-up merge from master/...@185556 to branch/2014-03-30/addrset. 
Copied from Perforce
 Change: 185563
 ServerID: perforce.ravenbrook.com
 2014-04-15 13:14:50 +01:00
Gareth Rees
93a79ac8a6 Add -wconversion to the options for gcc. ensure that the mps builds. 
Copied from Perforce
 Change: 185463
 ServerID: perforce.ravenbrook.com
 2014-04-11 14:25:15 +01:00
Gareth Rees
2e053caffc Alas, mvff and mvt can't round up alignment because they need to support buffered allocation. 
Copied from Perforce
 Change: 185381
 ServerID: perforce.ravenbrook.com
 2014-04-09 13:01:58 +01:00
Gareth Rees
4c9426514c Alignment is now configurable for mv pools using mps_key_align. 
MVFF and MVT support alignments down to FreelistMinimumAlignment.
For MVFF and MVT, alignments that are too small are rounded up automatically, to make it easier to write portable programs.

Copied from Perforce
 Change: 185369
 ServerID: perforce.ravenbrook.com
 2014-04-08 21:32:19 +01:00
Gareth Rees
262bb324f3 No keyword arguments needed in these cbsfastlandclass initializations. 
Copied from Perforce
 Change: 185329
 ServerID: perforce.ravenbrook.com
 2014-04-08 00:13:50 +01:00
Gareth Rees
b90f8b9bcd Use cbs subclasses to implement the fast-find and find-in-zones features. this avoids the need to control behaviour by passing booleans, and it means that there is no wasted space in the cbs block structure when these features are not used. 
Bring splay tree design up to date; add missing diagrams; move it to the "current" section of the manual.

Copied from Perforce
 Change: 185294
 ServerID: perforce.ravenbrook.com
 2014-04-07 15:36:18 +01:00
Gareth Rees
4b5d396792 Catch-up merge from master sources to branch/2014-03-30/addrset. 
Copied from Perforce
 Change: 185288
 ServerID: perforce.ravenbrook.com
 2014-04-07 15:02:09 +01:00
Gareth Rees
d3d2795fce Improve control over checking: 
1. Where Type is a pointer type with a signature, replace CHECKL(TypeCheck(val)) with CHECKD(Type, val).
2. Where Type is a pointer type with no signature, replace CHECKL(TypeCheck(val)) with CHECKD_NOSIG(Type, val).
3. Where Type is a pointer type with a signature, but the structure is not visible at point of checking, replace CHECKL(TypeCheck(val)) with CHECKD_NOSIG(Type, val). Reference <design/check/#.hidden-type>
4. Make BTCheck extern and use it where possible.
5. Replace AVER(TypeCheck(val)) with AVERT(Type, val).

Copied from Perforce
 Change: 185263
 ServerID: perforce.ravenbrook.com
 2014-04-06 22:51:05 +01:00
Gareth Rees
673eea41a0 Check classes after defining them. 
Copied from Perforce
 Change: 185228
 ServerID: perforce.ravenbrook.com
 2014-04-04 12:51:07 +01:00
Gareth Rees
86a50c3d1c Fix file-at-a-time compilation. 
Copied from Perforce
 Change: 185210
 ServerID: perforce.ravenbrook.com
 2014-04-03 15:01:53 +01:00
Gareth Rees
ce6b34aa8f Test the failover module (both always and never failing over). 
Fix result code bug in failoverInsert.
Test all result codes in fotest.
Tidy up code and documentation.

Copied from Perforce
 Change: 185207
 ServerID: perforce.ravenbrook.com
 2014-04-03 14:46:58 +01:00
Gareth Rees
0b159dc650 New module failover implements a fail-over allocator as a land class. 
Use Failover in MVT and MVFF.
Test Failover in landtest.
Implementation of LandFindInZones for Freelist (untested).
Remove signature from RangeStruct so we can embed it without a space cost.

Copied from Perforce
 Change: 185196
 ServerID: perforce.ravenbrook.com
 2014-04-03 12:52:23 +01:00
Gareth Rees
b409ae89f6 Turn freelist into a land class. 
Copied from Perforce
 Change: 185155
 ServerID: perforce.ravenbrook.com
 2014-04-02 14:01:18 +01:00
Gareth Rees
5314260bc8 Avoid type puns. 
Copied from Perforce
 Change: 185151
 ServerID: perforce.ravenbrook.com
 2014-04-02 12:16:38 +01:00
Gareth Rees
1d8238419a Landiargs -> liargs for terseness and consistency. 
Copied from Perforce
 Change: 185147
 ServerID: perforce.ravenbrook.com
 2014-04-01 23:39:03 +01:00
Gareth Rees
d2cbfda452 First pass at implementation of lands (collections of address ranges). 100% boilerplate! 
Copied from Perforce
 Change: 185131
 ServerID: perforce.ravenbrook.com
 2014-04-01 19:51:55 +01:00
Richard Brooksby
887c6e3845 Tidying up interfaces to cbsinit and cbsfindinzones. 
Documenting assumptions about CBS block allocation used by the arena.

Copied from Perforce
 Change: 184593
 ServerID: perforce.ravenbrook.com
 2014-02-28 10:30:32 +00:00
Richard Brooksby
ddfec2c708 Eliminating complex zonedcbs adt and reintroducing a cbs with zone summaries. removing the default “high” segment preference, that caused the cbs trees to unbalance, and isn’t required outside of ep. considerable reduction in code and speed up, now at parity with master on test xc/release/gcbench -x 1234 amc 
Copied from Perforce
 Change: 184587
 ServerID: perforce.ravenbrook.com
 2014-02-28 00:35:05 +00:00
Richard Brooksby
f31fc5325a Instead of aggressively returning every page it can, mvff takes a parameter for the proportion of spare space to hold in its free lists before attempting to return space to the arena. 
Copied from Perforce
 Change: 184498
 ServerID: perforce.ravenbrook.com
 2014-02-24 23:04:31 +00:00
Richard Brooksby
57c9c5aac8 Finishing the mvff block pool. oops. 
Copied from Perforce
 Change: 184487
 ServerID: perforce.ravenbrook.com
 2014-02-24 21:14:40 +00:00
Richard Brooksby
5fc0d82d6a Initialising a shared inline mfs for the cbss in mvff so that mvff can be used as the arena’s bootstrap and control pool. 
Copied from Perforce
 Change: 184486
 ServerID: perforce.ravenbrook.com
 2014-02-24 21:14:40 +00:00
Richard Brooksby
05a2713826 Reforming mvff pool class to avoid using segments and to actually have a chance to return memory to the arena. 
Copied from Perforce
 Change: 184467
 ServerID: perforce.ravenbrook.com
 2014-02-23 21:00:53 +00:00
Richard Brooksby
7a73e98638 Refactored implementation of preferenced allocation policy from vm arena to general purpose arena. 
Eliminating SegPrefExpress in most places, especially where there wasn’t really any preference.
Eliminating special case knowledge about garbage collection from the arena.

Copied from Perforce
 Change: 184279
 ServerID: perforce.ravenbrook.com
 2014-01-30 18:08:10 +00:00
Richard Brooksby
8e4e0f066e Eliminating dodgy use of *segprefdefault() which could lead to implicit calls to memcpy. 
Copied from Perforce
 Change: 184217
 ServerID: perforce.ravenbrook.com
 2014-01-29 15:13:22 +00:00
Richard Brooksby
56d48bbd1d iterating over segments using the pool and segment rings, rather than scanning the tract table, which may be full of irrelevant tracts, especially when non-gc pools are large. see <https://info.ravenbrook.com/mail/2014/01/25/12-16-09/0/>. 
Copied from Perforce
 Change: 184120
 ServerID: perforce.ravenbrook.com
 2014-01-25 15:33:10 +00:00
Gareth Rees
da3afade7b Mvt takes configurable alignment via a keyword argument. 
Copied from Perforce
 Change: 182605
 ServerID: perforce.ravenbrook.com
 2013-06-07 14:45:14 +01:00
Gareth Rees
938f1945fa Encapsulate cbsfindlargest falling back to freelistfindlargest as suggested by rb in review. 
Copied from Perforce
 Change: 182502
 ServerID: perforce.ravenbrook.com
 2013-06-04 17:10:38 +01:00
Gareth Rees
2bda69bc6b New test case "fotest" for fail-over of cbs to freelist in mvff. (plus stub for mvt.) need some testing hooks in mvff and mvt for getting the cbs. 
This reveals a couple of bugs in MVFF: (i) if CBSDelete fails, this might be because the block is actually on the Freelist instead; (ii) we can't SegFree before calling FreelistDelete because the Freelist needs access to the memory in order to read its pointers.

Copied from Perforce
 Change: 182500
 ServerID: perforce.ravenbrook.com
 2013-06-04 16:51:24 +01:00
Gareth Rees
6f356edfb1 In mvff, fall back from cbs to freelist if cbs fails. 
Copied from Perforce
 Change: 182434
 ServerID: perforce.ravenbrook.com
 2013-06-03 16:21:52 +01:00
Gareth Rees
db9328da7a Use range objects in the cbs interface instead of base, limit pairs. the idea is that freelist and cbs should offer similar interfaces so that the testing code can be shared. 
Copied from Perforce
 Change: 182364
 ServerID: perforce.ravenbrook.com
 2013-05-31 16:29:26 +01:00
Gareth Rees
934ae05ca3 Rename the enumeration cbsfinddelete to finddelete in anticipation of it being shared between cbs and new freelist module. 
Since ABQDisposition enumeration only has two values, it's better to use Bool.

Copied from Perforce
 Change: 182347
 ServerID: perforce.ravenbrook.com
 2013-05-31 00:09:33 +01:00
Gareth Rees
87019f4ee2 Fix review comments from <https://info.ravenbrook.com/mail/2013/05/28/22-21-11/0/>: 
* Rename RangeOverlap to RangesOverlap.
* MVFF shouldn't assume that CBSInsert and CBSDelete can't fail.
* Remove unused function mps_mvff_stat.
* Don't call range methods in RangeCheck.
* RangeInit can't fail, so return void.

Copied from Perforce
 Change: 182297
 ServerID: perforce.ravenbrook.com
 2013-05-29 12:19:19 +01:00
Gareth Rees
4c5bca4bd0 Fix review comments from rb in <https://info.ravenbrook.com/mail/2013/05/23/18-17-17/0/> 
Copied from Perforce
 Change: 182149
 ServerID: perforce.ravenbrook.com
 2013-05-24 00:48:19 +01:00
Gareth Rees
5ea309b5cc Catch-up merge from master sources to branch/2013-05-17/emergency. 
Copied from Perforce
 Change: 182048
 ServerID: perforce.ravenbrook.com
 2013-05-21 18:31:50 +01:00
Gareth Rees
1912c062b5 Use the new cbs interface. 
Copied from Perforce
 Change: 182016
 ServerID: perforce.ravenbrook.com
 2013-05-20 20:47:10 +01:00
Gareth Rees
217831cc47 Remove "emergency" free list allocator from the cbs module (it belongs in its own module) and update clients and the design accordingly. 
Copied from Perforce
 Change: 181927
 ServerID: perforce.ravenbrook.com
 2013-05-19 14:27:24 +01:00
Richard Brooksby
f876b80dff Fixing the mvff documentation to say that the keyword arguments are optional, and documenting their default values. 
Attaching the note about some keywords not affecting allocation points to those keywords with a footnote reference.
Fixing the default alignment setting to MVFF pool to MPS_PF_ALIGN.

Copied from Perforce
 Change: 181874
 ServerID: perforce.ravenbrook.com
 2013-05-16 18:12:54 +01:00
Richard Brooksby
6806f70e58 Correcting the documentation of the restrictions on the keyword arguments to the mvff pool class. 
Copied from Perforce
 Change: 181872
 ServerID: perforce.ravenbrook.com
 2013-05-16 17:40:14 +01:00