- corrected bugs with the inlining mode
This is a major revision: - added interface files (.lusi) in the language, that can be compiled on their own, giving an object file (.lusic) and a header file (.h) - modular code generation, from Lustre to C level included. - nice amount of code refactoring
- added missing constraint check when sub-clocking tuple expressions - added an algorithm that reuses dead or clock-disjoint variables instead of declaring/using new ones. - NOT carefully tested. Use option -O 3 if you want to give it a try...
Updated the licence info and header for each file.Moved backends in separate folders
- many bugs/limitations in lifting operators to tuples have been worked out: - typing/clock calculus/normalization now work properly - still, a bug in annot generation (this one is for Ploc !!) in file normalization, line 396 - bug corrected in subtyping...
work in progress (code optimization again)
Merged horn_traces branch
added some infrastructure to ease optimization (reusing vars)
Prepared first stage of code reorg:1. moved type def in lustrespec.ml2. moved constructor and basic functions in corelang3. Modified eexpr with prenext quantifiers
added warnings for useless variables (at verbose level 1) - exact definition of 'useless' may be further refined - display could certainly be improved
Extracted scheduling from machine code computation
Create a Step call only for functions that are not in basic lib
- bug correction (regression from previous versions !) introduced in C code generated from relational operators.
- refactorization of typing code (simpler subtyping rules) - simplification of clock calculus (may be still buggy, work in progress) no impact on unclocked programs.
typo corrected in code generation for array memories (bad parentheses)
- corrected causality bug (cf. previous commit)
liveness analysis improved. BUG found in causality wrt clocks...
- reimplemented computation of dead variables - added computation of a reuse policy (depending on types) - not yet used though, would have to change code generation in order to be useful...
added liveness analysis for reusing dead variables. Not yet used.
Added declaration/definition of stateless/stateful nodes. The 'function' keyword is for stateless nodes only, the 'node' keyword is any kind of node. Improves compilation and paves the way for more optimizations.
- work in progress for stateless/stateful status computation (to turn conditionals into merges, which yield more efficient C code)
corrected bugs in clock generalization that produced pessimistic C code (not wrong though); corrected bug with node importation policy wrt (re)declaration, (re)definition...
Merge back horn backend branch in trunk
In the middle of the coding process. Just pushing thinks
Initial copy of the horn output version. Not really working yet
Moved files to trunk in lustre_compiler