jeltz/heptagon
1
0
Fork 0
Code Issues Pull requests Releases Activity
630 commits 2 branches 0 tags 6.9 MiB
Commit graph

60 commits

Author SHA1 Message Date
Cédric Pasteur
f45e30d207 Yet another special case for iterators 2011-06-07 10:49:36 +02:00
Cédric Pasteur
de3a61557b Improvements on Schedule_Interf 
Only take into account optimized types.
Reuse code from Interference to compute uses.
 2011-05-25 09:12:16 +02:00
Cédric Pasteur
b7ba8e7913 Remove useless case 
As targeting is only done for linear variables, 
this case is already dealt with by 
coalesce_linear_vars
 2011-05-25 09:12:15 +02:00
Cédric Pasteur
75b4fa9ddc Fixed bug in the interference of fold 2011-05-25 09:12:15 +02:00
Cédric Pasteur
2b2cba8e2d Added split operator 2011-05-25 09:12:14 +02:00
Cédric Pasteur
28b9eaa203 Fixed problems with types 2011-05-25 09:12:14 +02:00
Cédric Pasteur
c3d47f4d4b Fixed error in computation of targeting 2011-05-25 09:12:14 +02:00
Cédric Pasteur
6f32564ad5 Added a sepecial case to deal with iterators 2011-05-25 09:12:14 +02:00
Cédric Pasteur
0728f3dae7 More work on code generation 2011-05-25 09:12:13 +02:00
Cédric Pasteur
822e87605b One step closer to code generation with memalloc 2011-05-25 09:12:13 +02:00
Cédric Pasteur
3f29e8623d Interaction between linear typing and memalloc 2011-05-25 09:12:13 +02:00
Cédric Pasteur
0b9dc1fc01 Correct fix 2011-05-25 09:12:12 +02:00
Cédric Pasteur
7d2b1e5865 Fixed bug in disjoint clock 2011-05-25 09:12:12 +02:00
Cédric Pasteur
285abc48bf Fixed some bugs 2011-05-25 09:12:12 +02:00
Cédric Pasteur
dec8cb69c8 Fixed bug in computation of live vars 2011-05-25 09:12:12 +02:00
Cédric Pasteur
68e1fe1ee8 Always check if a var should be optimized 2011-05-25 09:12:12 +02:00
Cédric Pasteur
66386ddca2 Fixed some bugs 2011-05-25 09:12:12 +02:00
Cédric Pasteur
448c163181 Dsatur coloring algorithm 
It is not completely generic, as we need to know 
the difference between affinity and interference
edges.
 2011-05-25 09:12:11 +02:00
Cédric Pasteur
3f9918b570 Added memory alloc application pass 2011-05-25 09:12:11 +02:00
Cédric Pasteur
a7015a9bf4 Fix for interference 
Works on a simple program
 2011-05-25 09:12:11 +02:00
Cédric Pasteur
1059329c0e Interference compiles 2011-05-25 09:12:11 +02:00
Cédric Pasteur
7787428f34 Compile fixes 2011-05-25 09:12:10 +02:00
Cédric Pasteur
3ee0e5e7b4 First version of interference.ml 2011-05-25 09:12:10 +02:00
Cédric Pasteur
a2303ec7e1 Other compile fixes 
The java backend is disabled temporarily
 2011-04-19 09:49:00 +02:00
Leonard Gerard
c0602c6df6 etuple removed 2011-04-18 16:09:07 +02:00
Cédric Pasteur
c6aa63b444 Heptc compiles again 
AFor now expects expressions instead of static exps
 2011-04-18 15:38:42 +02:00
Cédric Pasteur
2fdf2855d3 Added a new truncated select operator 
a[>e<] returns the element in the array at index
e, a[0] if e < 0 and a[n-1] if e >= n
 2011-03-22 22:12:59 +01:00
Cédric Pasteur
b1b8e103f2 Added partial application for iterators 
For instance:
  ... = map<<n>> (f<<se>>)((t1, t1'))(t2, t3)

is translated to:
  for(int i =...)
    ... = f(t1, t1', t2[i], t3[i])
 2011-03-21 17:22:03 +01:00
Léonard Gérard
cab8bb706e backport from async. 2011-03-09 00:06:36 +01:00
Léonard Gérard
315527231c Async in Heptagon Minils Obc. 2011-01-24 16:09:27 +01:00
Léonard Gérard
2ae809c971 Fresh vars, and ident refactoring. 
Idents.enter_node should be called when entering a node, it is done automagically by the mapfold unless you call directly Hept_mapfold.node_dec.
 2010-12-15 11:26:29 +01:00
Gwenal Delaval
ed2642f847 Added controllables in every pass 2010-12-08 17:32:24 +01:00
Léonard Gérard
a22f7216f2 Added when and merge to heptagon. Need tests ! 
Clocking is still done in minils since it is way simpler.
 2010-11-01 01:05:37 +01:00
Léonard Gérard
6da5171e6a Flatten arguments fix in cloking. 2010-09-27 18:16:00 +02:00
Cédric Pasteur
df12e081ae Refactored Misc 
Created two new files: 
 - utilities/global/compiler_options.ml: contains
 the options that can be set using the cli
 - utilities/global/errors.ml: contains global
 errors definition

Misc now only contains helper functions that have 
nothing to do with the ast or the compiler.
 2010-09-15 09:38:52 +02:00
Léonard Gérard
02db2ad6b4 Changed compile flags, and cleaned a bit. 
Warn as error for partial match.
Warn for unused variables : added some TODO to check.
PS : I'll deal with callgraph which is doing things that Modules does.
 2010-09-14 09:39:02 +02:00
Cédric Pasteur
d8dffe15d8 Fixed remainings non exhaustive patterns 
Now the only shown warnings are the X type that 
we don't care about
 2010-09-13 16:02:33 +02:00
Léonard Gérard
a58c729ae2 clocking tweak. 2010-09-13 01:18:45 +02:00
Cédric Pasteur
ffff23b675 Remove Elambda 
Replace Elambda with a central repository of 
anonymous nodes. It made the AST unnecessarily 
complex.
 2010-09-06 14:03:47 +02:00
Léonard Gérard
5c09abeb18 Revet last commit a5f89876c2 and gives the right fix. 
Plus remove all the forbidden '\n', replaced with '@.' or '@\n' depending on context.
 2010-09-01 14:40:31 +02:00
Léonard Gérard
9df4f625a2 Fixed location to use formatter instead of out_channel. 2010-08-24 17:29:00 +02:00
Léonard Gérard
4b3c3ba8b5 Revert "Fixed problem in clocking" e3676d1e3c 
Fixing the actual bug :
* static_exp should not be created without type
after or during the typing pass.
 2010-08-03 22:38:42 +02:00
Léonard Gérard
418b961293 Small fixes to the compilation processes. 
*Locations should be pretty damn good now.
*Mlsc is up to date, still need a scoping pass (to be posted soon)
 2010-08-02 16:31:57 +02:00
Cédric Pasteur
8f4220e08d Made Eupdate dynamic 
Modifying an array with
 [ a with [i] = v ]
should expected a dynamic (not static) value
for i (nothing happens if i is in the wrong range).
This is the same behaviour as in Scade and it is 
useful eg to modify an array in a foldi.
 2010-07-28 12:34:07 +02:00
Cédric Pasteur
221eea1644 Added polymorphic (=) and (<>) operators 
These two operators really need to be polymorphic, 
so they cannot be in Pervasives. a <> b is translated
to not (a = b>) at parsing.
 2010-07-27 17:16:49 +02:00
Cédric Pasteur
e3676d1e3c Fixed problem in clocking 
Tuple of consts should have one clock var per 
element. Added a new function const_skeleton to
create this kind of clock.
 2010-07-26 15:34:46 +02:00
Léonard Gérard
dc9bec28bf Clock refactoring. 2010-07-23 22:13:03 +02:00
Léonard Gérard
1719e2eb36 Fix bad clock expectation for static_exps. 2010-07-23 22:05:54 +02:00
Léonard Gérard
f124bb4fd7 Refactoring Ident -> Idents ( uniform with Names etc ) 2010-07-23 19:45:19 +02:00
Léonard Gérard
57b1405731 Basic clocking port. 2010-07-15 16:20:46 +02:00