James Golick method cache patches #4

haileys · 2013-07-13T08:22:02Z

This pull request merges in the @jamesgolick hierarchical method cache patches.

I've fixed a bug in rb_const_set and rb_const_remove where these functions failed to invalidate the global state version (which is what constant lookups are cached against)

This isn't a comprehensive fix, but it'll do in the interim.

This reverts commit 2d9a08f.

This is the beginning of a basic gc profiler that has less overhead and memory footprint than GC::Profiler. More to come.

@josephruscio

Thanks @josephruscio for the feedback.

because otherwise the ruby_core_source gem doesn't work

Everything is now under RubyVM::MethodCache: - RubyVM::MethodCache.invalidations - RubyVM::MethodCache.hits - RubyVM::MethodCache.misses - RubyVM::MethodCache.miss_time (milliseconds)

Set RubyVM::MethodCache.invalidation_log to something that responds to

…lementation :-)

…class hierarchy to expire the method cache from Object

…its subclasses are unreachable now anyway

…dules

…mind it

Conflicts: common.mk

…hes.

…k-method-cache-patches Conflicts: gc.c version.h

James Golick method cache patches

tmm1 · 2013-07-13T08:33:59Z

💥

jamesgolick added 30 commits November 18, 2012 23:14

reduce the cost of GC::Profiler.total_time

2d9a08f

This isn't a comprehensive fix, but it'll do in the interim.

use clock_gettime if available. thanks @tmm1

4c5995a

Revert "reduce the cost of GC::Profiler.total_time"

5175b83

This reverts commit 2d9a08f.

Add GC::BasicProfiler

576cba1

This is the beginning of a basic gc profiler that has less overhead and memory footprint than GC::Profiler. More to come.

make gc_profile_enable_get correctly return a ruby boolean

380f070

zero GC::BasicProfiler stats on disable

e2730ac

Thanks @josephruscio for the feedback.

copy and paste comment failure

eaa3546

ghetto: switch the patchlevel to a real one

1ecab79

because otherwise the ruby_core_source gem doesn't work

expose method cache invalidation count

c9a8a18

this VALUE isn't being used as a pointer, so it needs converting

9c10912

Add more method cache stats.

82c00d2

Everything is now under RubyVM::MethodCache: - RubyVM::MethodCache.invalidations - RubyVM::MethodCache.hits - RubyVM::MethodCache.misses - RubyVM::MethodCache.miss_time (milliseconds)

Add a way to trace method cache invalidations.

6bb0c67

Set RubyVM::MethodCache.invalidation_log to something that responds to

Fix incompatible pointer assignment.

094f2f4

Implement per-class method caches.

0fbecb8

remove stuff about the global method cache

02859e8

the very very basics of subclass tracking

4e1b2e9

rename that

4fc3635

turns out we don't need both links I don't think

909a08d

I think this will do what we really want

c8c74a9

subclass tracking. thanks to charlie sommerville for a lot of the imp…

4668e93

…lementation :-)

detach from superclass subclasslist on gc

f9ec752

couple fixes for detach from superclass subclass list

ae4843e

try not to clobber the entire method cache

d8119fd

make sure that the inline caches aren't out of sync

dcffa00

Fix a bunch of bugs and segfaults

1af25fa

refactor subclass list additions

11d9700

reorder this to avoid dat segfault

a8208e5

there's something wrong with this free but I can't figure out what

b8122f9

instrument the cost of invalidating method cache trees

f916d9a

add a second, global sequence to avoid having to traverse the entire …

69f6af2

…class hierarchy to expire the method cache from Object

jamesgolick and others added 26 commits April 9, 2013 18:39

move the subclasses table on to the classext

9f0f1c2

remove * as per tmm1

b21954a

try zeroing the subclasses first

7d0aacb

zero the subclasses of ICLASSes as well

a415894

fix subclasses /included_in methods

08800a9

gotta return ST_CONTINUE from here or everything gets shredded

3c9f01f

we don't need to clear a method cache tree if we're gc'ing a class - …

0050081

…its subclasses are unreachable now anyway

if we're sweeping an rb_method_entry_t, we've already cleared the cache

35e7224

if we're looking at a module, zero the klass of the ICLASS

5d1c2ee

make sure that we correctly add to the subclasses list for all the mo…

e9fab17

…dules

make sure that we remove an iclass from its module's subclasses

450c2f2

properly recurse down the class hirearchy

83ce9c8

we're not actually using the klass in this function anymore, so never…

12e8032

…mind it

avoid calling rb_clear_cache_by_class an extra time here

a99987f

switch to using linked lists for subclass tracking again

7e54b39

sp_ar: include sparse array

614330e

Conflicts: common.mk

[experimental] switch to using funny-falcon's sa_table for method cac…

f83cf0c

…hes.

fix rb_class_subclasses to reflect the switch back to linked lists

dbf0ece

switch both sequences to uint64_t so they're 64 bits on all platforms

f480e5e

Merge remote-tracking branch 'jamesgolick/klasscache' into jamesgolic…

3894f3b

…k-method-cache-patches Conflicts: gc.c version.h

remove statistics

c875e5c

move vm sequence/method cache stuff into vm.c

90b963c

invalidate vm state version on const_set and const_remove

561b1cf

expose current global_state_version to ruby land

06de5fc

remove unnecessary invalidations

fb08ad4

split out cache into method state version and constant state version

da97620

haileys pushed a commit that referenced this pull request Jul 13, 2013

Merge pull request #4 from github/jamesgolick-method-cache-patches

d151f4e

James Golick method cache patches

haileys merged commit d151f4e into github-p231 Jul 13, 2013

haileys deleted the jamesgolick-method-cache-patches branch July 13, 2013 08:31

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

James Golick method cache patches #4

James Golick method cache patches #4

haileys commented Jul 13, 2013

tmm1 commented Jul 13, 2013

James Golick method cache patches #4

James Golick method cache patches #4

Conversation

haileys commented Jul 13, 2013

tmm1 commented Jul 13, 2013