Commit graph

10920 commits

Author SHA1 Message Date
Leonardo de Moura
313fd69e8c feat(library/compiler/borrowed_annotation): mark objects as owned when stored in constructors
@kha
2019-02-23 09:12:53 -08:00
Leonardo de Moura
b971db6c11 chore(boot): update 2019-02-22 16:09:48 -08:00
Leonardo de Moura
d23216e3ee fix(library/compiler/llnf): bug at explicit_rc_fn 2019-02-22 16:01:52 -08:00
Leonardo de Moura
07aa990f22 fix(library/compiler/borrowed_annotation): ensure that exported functions do not take borrowed arguments
In the future, we should automatically generate wrappers that make all
adjustments for us.
2019-02-22 15:32:30 -08:00
Leonardo de Moura
b1b75c7c2e feat(library/compiler): borrow inference procedure 2019-02-22 15:23:42 -08:00
Leonardo de Moura
ba43d355b7 feat(library/compiler): add borrowed annotation inference skeleton 2019-02-22 11:14:38 -08:00
Leonardo de Moura
6646d7d62c chore(boot): update 2019-02-22 10:19:39 -08:00
Leonardo de Moura
01922794c5 fix(library/compiler/llnf): mk_boxed_version must take care of borrowed arguments and result 2019-02-22 10:18:17 -08:00
Leonardo de Moura
20c877b277 chore(library/init/io): add inline 2019-02-22 09:55:36 -08:00
Leonardo de Moura
778e7e41f9 refactor(library/init/data/rbmap/basic): pass ins node-cell to balance1 and balance2.
The idea is to reuse the cell. The trick is like the one we used for
improving state_t. It seems to work pretty well. Now, the Lean
version is 29% slower than the C++ one.

cc @kha
2019-02-20 18:27:58 -08:00
Leonardo de Moura
835718955f refactor(library/init/data/rbmap/basic): store color in the node
@kha Now the Lean version is approx. 50% slower than the C++ version.
2019-02-20 17:52:03 -08:00
Leonardo de Moura
f91a37e686 chore(boot): update 2019-02-20 17:06:02 -08:00
Leonardo de Moura
ac9e37ed86 feat(library/compiler/llnf): we postpone the simplification of 1-reachable and all-eq cases_on expressions
Reason: the `cases_on`-expressions are used to guide
`insert_reset_reuse_fn`. The new `simp_cases` simplifier applies the
1-reachable and all-eq cases_on expression simplifications after
`insert_reset_reuse_fn` is executed.
2019-02-20 17:00:50 -08:00
Leonardo de Moura
35adf5f540 feat(library/compiler/llnf): avoid unfruitful reuse instruction replacements 2019-02-20 16:34:46 -08:00
Leonardo de Moura
a068c91e50 chore(boot): update 2019-02-20 16:20:22 -08:00
Leonardo de Moura
54a89dabb7 feat(library/compiler/llnf): new reset/reuse insertion procedure 2019-02-20 16:12:58 -08:00
Leonardo de Moura
937b947938 feat(library/compiler/llnf): decorate _cnstr instruction with inductive type name 2019-02-20 14:45:35 -08:00
Leonardo de Moura
b8cee758a5 feat(library/compiler/llnf): add push_proj_fn 2019-02-20 13:20:27 -08:00
Leonardo de Moura
2f218f7ec0 refactor(library/compiler/llnf): remove reset/reuse insertion from to_llnf_fn
`to_llnf_fn` now produces an IR similar to the \lambda_pure in our paper.
2019-02-20 11:16:29 -08:00
Leonardo de Moura
0cbf587a00 chore(boot): update 2019-02-19 22:17:57 -08:00
Leonardo de Moura
75ad042c66 feat(library/compiler/llnf): improve reuse/reset insertion
It is now closer to the procedure at reuse.txt
2019-02-19 22:14:29 -08:00
Leonardo de Moura
6276d1a7f1 feat(library/compiler/emit_cpp): avoid reset field instructions when reuse instruction is guaranteed to be executed
expr_const_folding.lean takes 3 secs now.
2019-02-19 13:44:46 -08:00
Leonardo de Moura
e2eeccdb2a chore(boot): update 2019-02-19 13:06:33 -08:00
Leonardo de Moura
fe1d17583c feat(library/compiler/emit_cpp): special support for (proj|inc)*;reset sequences
We save inc/dec operations. It improved the performance of
`expr_const_folding.lean`. On my Linux desktop
Before: 3.7 secs
After:  3.1 secs

cc @kha
2019-02-19 13:06:22 -08:00
Leonardo de Moura
b739d7f343 chore(boot): update 2019-02-18 20:53:04 -08:00
Leonardo de Moura
00aa78fffc feat(library/compiler/llnf): given y := _unbox.n x, mark x as an unboxed scalar if n < sizeof(void*) 2019-02-18 20:52:02 -08:00
Leonardo de Moura
fe4b1509ba chore(boot): update 2019-02-18 20:22:18 -08:00
Leonardo de Moura
feea8ecccd feat(library/compiler/llnf): avoid inc/dec operations on persistent objects
inc/dec operations are noop's for persistent objects.
2019-02-18 20:22:18 -08:00
Leonardo de Moura
59e9751c2b chore(boot): update 2019-02-17 11:46:17 -08:00
Leonardo de Moura
b752dd1984 fix(library/compiler/emit_cpp): mark global objects as persistent
They may be used by tasks, but they are not directly reachable from
task starting point.
2019-02-17 11:45:51 -08:00
Leonardo de Moura
8ac1a1969d fix(runtime/object): task.bind and task.map should also invoke to_mt 2019-02-17 11:11:41 -08:00
Leonardo de Moura
0a19f46c6a chore(boot): update 2019-02-17 10:53:04 -08:00
Leonardo de Moura
fca40151b2 fix(library/compiler/emit_cpp): is_shared should only be used on heap objects 2019-02-17 10:51:51 -08:00
Leonardo de Moura
827021a6c5 feat(runtime/object): given mk_task(o), mark objects reachable from o as MTHeap 2019-02-17 10:29:41 -08:00
Leonardo de Moura
d100f95469 feat(runtime/object): make dbg_trace thread safe 2019-02-17 09:50:55 -08:00
Leonardo de Moura
7623f64b5e feat(runtime,library/init/util): add some debugging helper function 2019-02-17 09:22:37 -08:00
Leonardo de Moura
75f4eb2f84 feat(library/compiler/emit_cpp): initialize task_manager at generated main function 2019-02-17 09:21:17 -08:00
Leonardo de Moura
170579c803 feat(library/init/core): task builting primitives 2019-02-17 08:45:46 -08:00
Leonardo de Moura
b7e7ca9527 chore(boot): update 2019-02-17 07:32:05 -08:00
Leonardo de Moura
243d7c08e3 feat(library/compiler/llnf): avoid inc and dec operations on (boxed) small nat literals 2019-02-17 07:31:14 -08:00
Leonardo de Moura
fbedc1d098 chore(boot): update 2019-02-17 07:18:28 -08:00
Leonardo de Moura
3c60af74ae feat(library/compiler/llnf): avoid inc and dec instructions on neutral elements 2019-02-17 07:17:54 -08:00
Leonardo de Moura
f4143c030f chore(boot): update 2019-02-17 06:53:05 -08:00
Leonardo de Moura
78440919e8 feat(library/compiler/llnf): avoid unnecessary inc x and dec x instructions
Example:
```
x = lean::box(0);
...
lean::inc(x);
...
lean::inc(x);
...
lean::dec(x);
...
```
In the example above, the `inc` and `dec` operations are unnecessary
since `x` is known to be a (boxed) scalar value. This commit fixes this.
2019-02-17 06:50:25 -08:00
Leonardo de Moura
0c1c1dd607 chore(boot): update 2019-02-17 06:26:15 -08:00
Leonardo de Moura
221704900a feat(library/compiler/llnf): avoid inc x and dec x instructions in branches where x is known to be a scalar value 2019-02-17 06:25:00 -08:00
Leonardo de Moura
9e0b28d8ce feat(library/init/data/array/basic): improve 2019-02-16 16:08:10 -08:00
Leonardo de Moura
3c73c43ab2 feat(runtime,library/init/data/array/basic): add builtin support for arrays 2019-02-16 15:27:23 -08:00
Leonardo de Moura
7ed04d3ff1 chore(boot): update 2019-02-16 12:28:41 -08:00
Leonardo de Moura
e98acc86b8 feat(library/compiler/emit_cpp): "merge" the case with most occurrences as default: 2019-02-16 12:27:25 -08:00