max/lean4-htt
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 2
12745 commits 1 branch 0 tags 5.3 GiB
Commit graph

181 commits

Author SHA1 Message Date
Leonardo de Moura
04dfd55f94 perf(kernel/expr): fix hash consing performance problem 
The kernel support opportunistic hash consing.
At commit e63c79c81e we have enabled this feature during
tactic execution, and fixed performance problems in the Certigrad
project (by @dselsam).
However, this change produced unexpected performance problems in
other Lean files (example: @JasonGross example at issue #1646 was 10x
slower after the commit above).
After analyzing the performance logs, I conjecture the hash_consing
may produce a substantial performance overhead if the following property
doesn't hold.

- Let `cache` be the hash_consing cache, then for each `e` in `cache`,
all children of `e` are also in the `cache`.

If the property above is not true, we may have problem whenever
we add an expression `t` containing multiple copies of a big term `T`.
For example, support we insert `f T_1 T_2`, where `T_1` and `T_2` are
structurally equal but are not pointer equal. Then, if we try to insert
`f T_2 T_1`, we will have to compare the huge terms twice, and the
comparison will be proportional to the size of `T_i`.

This commit tries to address this performance problem by enforcing
the property above. This is not a perfect solution since we may keep
trying to create terms using big terms created before hash_consing
has been enabled. After this commit, the example at issue #1646
is only 1.4x slower. It didn't impact the standard library
compilation (memory nor time).

I'm using this commit is a temporary workaround. We should probably
remove the hash_consing support from the kernel, and implement it
in key places. For example, for Certigrad, we can control memory
consumption by using hash_consing only during `simp`,
`instantiate_mvars` and `elaborator` finalization procedure.

@gebner @kha Any ideas/suggestions for this hash_consing issue?
 2017-06-23 12:41:35 -07:00
Leonardo de Moura
7ffd7fea3d feat(kernel): store depth of composite terms and use it in the hash code computation 
closes #1589
 2017-05-25 16:51:02 -07:00
Sebastian Ullrich
491802409a chore(*): remove unused macro_definition_cell::pp method 2017-05-24 09:51:23 +02:00
Gabriel Ebner
42288198db fix(kernel/expr): disable caching by default 2017-03-23 08:57:56 +01:00
Leonardo de Moura
544d5c67f4 feat(CMakeLists,kernel): add TRACK_LIVE_EXPRS=ON compilation option 2017-02-28 10:56:14 -08:00
Leonardo de Moura
caf92bbcc0 feat(kernel/expr): compress expr_macro representation 2016-12-28 09:01:21 -08:00
Leonardo de Moura
5572d7f3ec perf(kernel,library/module): enable expr caching when deserializing .olean files 2016-12-16 18:33:46 -08:00
Leonardo de Moura
52ef17284a fix(kernel/expr, kernel/level): if caching is enabled during finalization, these modules would register a finalizer after finalization of the main thread 
The deleted lines were not really needed. They were added before we had
the thread finalization code.
 2016-12-01 10:16:05 -08:00
Leonardo de Moura
3c878ecd01 feat(kernel): add let-expressions to the kernel 
The frontend is still using the old "let-expression macros".
We will use the new let-expressions to implement the new tactic framework.
 2016-02-29 16:40:17 -08:00
Leonardo de Moura
2b1d734544 feat(kernel/expr): remove 'contextual' flag from binder_info 2016-02-29 12:41:43 -08:00
Leonardo de Moura
d9294fc164 chore(kernel/expr): remove dead var 2016-01-13 17:36:33 -08:00
Leonardo de Moura
930fcddace feat(kernel/expr): add get_app_args_at_most 2016-01-06 17:29:28 -08:00
Daniel Selsam
413989afd6 feat(library/blast/backward): backward chaining strategy 2015-11-18 17:48:39 -08:00
Leonardo de Moura
d5d8ac8b44 fix(kernel/expr): warning when compiling using clang on OSX 2015-11-16 18:50:34 -08:00
Leonardo de Moura
92a7c38260 feat(kernel/expr): add mk_app that takes a list of arguments 2015-11-13 14:01:15 -08:00
Leonardo de Moura
76032eea90 feat(kernel/expr): change dummy expression used in default constructor 2015-11-08 14:05:02 -08:00
Leonardo de Moura
a446c54b87 fix(tests/kernel/max_sharing): max_sharing on OSX 2015-10-03 12:12:21 -07:00
Leonardo de Moura
d324d54d21 fix(kernel/expr): add global expressions storing Prop and Type.{1} to 
hash-consing cache when max-sharing is enabled
 2015-10-02 15:04:10 -07:00
Leonardo de Moura
3cf11dac87 feat(kernel/instantiate): make sure instantiate_type_univ_params and instantiate_value_univ_params caches are reset when we enable max-sharing 2015-10-01 15:42:33 -07:00
Leonardo de Moura
45594e86c7 feat(kernel/abstract): make sure Pi/Fun cache is reset when we enable max-sharing 2015-10-01 15:28:55 -07:00
Leonardo de Moura
cb2f93ee6a fix(kernel/expr,kernel/level): performance problem 
The method unordered_set::clear takes time O(n) where n is the number of
buckets in the hashtable used to implement the set.
Moreover, it does not reduce the number of buckets to 0.
That is, it keeps `n` empty buckets.

This creates performance problems if the number of buckets gets really
huge at one point.
The tst6 at tests/kernel/expr.cpp demonstrates the problem
 2015-09-30 17:25:13 -07:00
Leonardo de Moura
d07bbe7e8c fix(kernel/expr): must take binder annotations into account in the hash-consing tables 2015-09-30 17:24:41 -07:00
Leonardo de Moura
57035d3162 feat(kernel/level,library/blast/expr): add universe level hash consing support in the kernel, simplify blast/expr even more 2015-09-30 13:31:42 -07:00
Leonardo de Moura
c5603e456a feat(kernel/expr): replace "opportunistic" caching with precise caching 
We also removed compilation option LEAN_CACHE_EXPRS
 2015-09-30 12:29:43 -07:00
Leonardo de Moura
25cbc5c154 fix(kernel/expr): remove 'cast_binder_info' 
We should put it back when we decide to implement it.
We also fix the bogus comment at src/frontends/lean/parser.cpp.

see issue #667
 2015-06-11 18:11:22 -07:00
Leonardo de Moura
ef75cac1c0 feat(kernel/expr): change the rules for inferring implicit arguments, closes #344 2014-11-25 12:54:07 -08:00
Leonardo de Moura
7516fcad97 feat(kernel/type_checker): add is_stuck method, and improve ensure_pi method, closes #261 2014-10-27 13:16:50 -07:00
Leonardo de Moura
6d64da2981 perf(kernel/expr_eq_fn): use thread local cache, and avoid memory allocation/deallocation 2014-10-17 16:44:20 -07:00
Leonardo de Moura
fe484b26f3 refactor(kernel/expr): remove dead code 2014-10-16 13:45:36 -07:00
Leonardo de Moura
814778abb1 refactor(kernel/expr): tag expressions at "creation" time 2014-10-15 13:12:09 -07:00
Leonardo de Moura
a26618e0f2 feat(frontends/lean): add '[]' notation for marking arguments where class-instance resolution should be applied 2014-10-12 13:06:00 -07:00
Leonardo de Moura
a0c37b231f feat(kernel/expr): add hash_bi function that takes binder information into account 2014-09-29 16:44:55 -07:00
Leonardo de Moura
516c0c73b9 refactor(*): remove dependency to thread_local C++11 keyword, the 
current compilers have several bugs associated with it

We use the simpler __thread (gcc and clang) and
__declspec(thread) (visual studio).
 2014-09-24 12:51:04 -07:00
Leonardo de Moura
ca1b8ca80f refactor(util/memory_pool): simplify memory_pool, it is not a template anymore 2014-09-24 10:48:32 -07:00
Leonardo de Moura
da481c3274 refactor(kernel): explicit initialization/finalization 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-09-24 10:12:28 -07:00
Leonardo de Moura
3310eb3dfc feat(frontends/lean): remove restriction on implict arguments, add new test that demonstrates the new feature 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-09-07 12:29:32 -07:00
Leonardo de Moura
0d5e346143 fix(library/expr_lt): make sure the builtin order is AC-compatible 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-08-04 15:51:10 -07:00
Leonardo de Moura
5c5dea7c8e feat(kernel/expr): add has_expr_metavar_strict 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-26 19:54:04 -07:00
Leonardo de Moura
022a151cf7 feat(kernel): add general purpose 'annotations', they are just a generalization of the 'let'-annotations 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-25 09:33:31 -07:00
Leonardo de Moura
a7c6c3e840 fix(kernel/expr): memory corruption in dealloc method 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-22 10:16:44 -07:00
Leonardo de Moura
5eaf04518b refactor(*): rename Bool to Prop 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-22 09:43:18 -07:00
Leonardo de Moura
4c6ebdeaf9 perf(util/memory_pool): use memory_pool for hierarchical names and justification objects we get a 8% performance improvement when using multiple threads 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-22 09:18:26 -07:00
Leonardo de Moura
c8b6f0c7fb refactor(util): rename fixed_size_allocator to memory_pool 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-22 07:49:40 -07:00
Leonardo de Moura
79ea7c5910 perf(kernel/expr): minimize access to system memory allocator by recycling expr_cells 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-21 18:46:47 -07:00
Leonardo de Moura
9289717169 perf(kernel/expr): inline get_free_var_range, and cache its value for local and metavars 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-17 08:51:46 +01:00
Leonardo de Moura
aae40f07e2 perf(kernel/expr): use thread local deletion buffer 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-16 08:39:03 +01:00
Leonardo de Moura
405e57eb2d refactor(kernel/formatter): add formatter_factory, and simplify formatter interface 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-10 18:32:00 +01:00
Leonardo de Moura
fc4df6a430 feat(kernel/expr): add O(1) predicates has_expr_metavar and has_univ_metavar 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-07-05 13:11:30 -07:00
Leonardo de Moura
cb000eda13 refactor(kernel): store binder_infor in local constants 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-06-30 11:37:46 -07:00
Leonardo de Moura
0adacb5191 feat(kernel): add infer implicit, and use it to infer implicit arguments of inductive datatype eliminators, and tag whether parameters should be implicit or not in introduction rules in the module inductive_cmd 
Signed-off-by: Leonardo de Moura <leonardo@microsoft.com>
 2014-06-28 13:57:36 -07:00