issues related to CRAVE randmoization

[Tonyli]

Hi Friends,

I have known about uvm-sc and crave for a while. Recently we're looking for integrating CRAVE into our testbench and did some experiment. However, we met some issues and I'm not sure it's the real issue or we were not using it correctly.

 

1.  not ture randomization

code:

randv<sc_uint<16>> b,c,d;

 

ALU4(rand_obj* parent =0) : rand_obj(praent), op(this), b(this), c(this), d(this) {

constraint(b()<10 && b()>0);

constraint(dist(c(),distribution<short>::simple_range(0,20)));

constraint(d()<c());

}

 

Result for mutiple times:

B: 8 2 4 4 2 2 4 8 4 2 4 4 2 8 2 4 2 2 4 8 4 2 4 4 2 8 2 4 2 2 4 4 2 8 2 4 2 2 8 4 4 8 2 2 4 2 4 4 8

C: 5 5 2 7 18 11 12 4 19 17 17 19 16 16 8 14 15 16 16 16 14 15 2 6 6 1 2 10 10 1 1 1 8 7 16 16 4 13 17

D: 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 1 0

 

It seems b is not B  & D were not turely randomized.  We have checked the seed is different and we tried different solver, the issue still exist.

 

Besides, we tried vector/array randomization,  also seeing some issues similar with above, the result is not truly randomized.

 

 

This is the most blocking issue. We also have some other concerns to consult but the answer to this issue is more important as we may not be able to use it.  Thanks everyone and looking forward to some answers.

[Eyck]

Which solver(s) did you configure and use?

[thilo.voertler]

Hi Tonyli,

this is a known issue with crave. The results seems to be correct for each run, but the randomization should be more uniform. There have been some updates to crave over the last year so please use the latest version available from https://github.com/accellera-official/crave/. What you can do to improve the quality of the distribution is the prev() statement to filter out previous results in a constraint. Also feel free to report this issue and a testcase to the crave github page https://github.com/accellera-official/crave/.

 

[Tonyli]

19 hours ago, Eyck said:

Which solver(s) did you configure and use?

Hi Eyck, 

We tried CUDD/Z3 sovlers, but didn't try other optional solvers.

[Tonyli]

19 hours ago, thilo.voertler said:

Hi Tonyli,

this is a known issue with crave. The results seems to be correct for each run, but the randomization should be more uniform. There have been some updates to crave over the last year so please use the latest version available from https://github.com/accellera-official/crave/. What you can do to improve the quality of the distribution is the prev() statement to filter out previous results in a constraint. Also feel free to report this issue and a testcase to the crave github page https://github.com/accellera-official/crave/.

 

Thanks for replying. We tried y()!=y(prev), it didn't help a lot, the result won't be same with previous one but was still distributed in limited numbers.  Besides, it seems dist with "simple_range" works well, but it could not replace dist in every situation, eg “constraint(d()<c())” .