Rationale: the << normal mode >> (i.e., not translating enums) was
not even accessible!

('::' vs '__')

soc2c -2cw7: use mkff+fixffx+arm.looploc to fix extern lib calls.

The bug was (in soc2c.ml::137) that, for array of instances of size
n, it was generating n+1 initialisations, causing a seg fault.

Factorize duplicated code in gen_loop_file and gen_loop_file4ogensim
along the way.

by
- remove duplication (add a profile_info function ine Lv6Verbose
- be more efficient (computing Sys.time only when necessary
- control the time-profiling printing via the -dbg flag "profile", and
  not the verbosity level

I was re-implementing an inneficient topological sort there (using a
linear list instead of logarithmic map). I win 1s on tri.ec (~8000
locals).

I did a map instead of a fold when checking for the absence of cycle.
On programs with 10 000 of equations, I go from 150s to 5s.

Also, force a workflow (via a lock file named commited) that make
sure that the content of lv6version.ml is always correct xrt shat and
git commit number.

as an opam package in the verimag opam repo.

Ditto for Misc and Compile

Also, do not set the seed when comparing compilers.

Indeed, otherwise, in some cases, rdbg or lurette migth not realise
that lus2lic exited in error.

Fix the bug detected in the previous commit where type instance were not propagated in package obtained from a model instance.

More precisely, issuing an "exit 2" is not enough for the
should_work proc in test/site.exp to state that the test fails.

That fix revealed a certain number of bad programs.
Some were not really wrong ; there were just not respecting the
naming convention used to automated the test (i.e., main node name =
file name).

It also releaved a real bug due to the fact that type
instanciatiation is not propagated during model instanciatiation,
where creates a loop between (Lic2soc.process_node and
Lic2soc.soc_of_node). Indeed it searches in the soc the table for an
abstract profile.

* la syntaxe est changee pour l'accepter
* la stucture IdSolver (compil "lazy") contient maintenant
  "all_srcs" qui permet d'acceder si besoin a n'importe quelle info source,
  notamment les params statiques des noeuds identifies par "Pack::name"

(which breaks lurettetop)

nb : Mamadou's change broke some tests (18->53). Empty body is
generated for nodes that have one eq.  I'll fix that later.

(which breaks lurettetop)

nb : Mamadou's change broke some tests (18->53). Empty body is
generated for nodes that have one eq.  I'll fix that later.

lus2lic -h shows the basic options
lus2lic -more shows more experimental options
lus2lic -dev shows experimental/internal  options

Indeed, I've missed that case and have no nonreg test for -2ch

that I should have catched then. Sigh.

only affect array concat

by grouping instances of the same type into arrays.

The rationale is to be able to generate for loops with iterators.

The previous code was working, but probably by chance.

Actually, that was the default behavior. Now the default behavior is
to generate loops for iterators.

nb : with -2ch, loops are still always inlined

Indeed, an iterator is a kind of node.

combined with -cc, actually computes the wcet.

nb : only works if wcet tools are installed (owcet.arm, arm-elf-gcc)

combined with -cc, actually computes the wcet.

nb : only works if wcet tools are installed (owcet.arm, arm-elf-gcc)

Indeed, in order to sort Soc.gao, I currently process in 3 stages ;
1. compute a total ordering of Actions.t according to the deps
2. transform Actions.t into Soc.gao (List.map)
3. factorize Soc.gao by looking at consecutive gao that have the same guard

With this change, between steps 1 and 2, I reorder the list of actions
so that group of independent actions are sorted according to their
clock. The idea is of course to provide to step 3 more opportunities
of factorization.

The result is not very convincing yet.

I've also tried to sort the actions wrt their clock before performing
the toposort. Seems not too bad.

I've kept the old behavior by default though.