Sft constraint vs weighted constraint

[SV_UVM_Infant]

Hi,

We use the below constraint wrt "unaligned_transfer".

 

//--------------------------------------------------------------------------

  constraint axi_narrow_unaligned_c {
        unaligned_transfer dist {1'b0 := 10 , 1'b1 := 90};
        soft unaligned_addr_offset == 'h0;
        solve unaligned_transfer before unaligned_addr_offset;
        (unaligned_transfer == 1'b1) -> unaligned_addr_offset inside {'h1,'h2,'h3};
           }

//--------------------------------------------------------------------------

 

There is a soft value = 0, provided for unaligned_addr_offset, and a weighted distribution allowed for  unaligned_transfer

 

What we observe is that the simulator always select "unaligned_transfer" = 0. Looks like its trying to prioritise the soft constraint. Is this expected?

 

Could you someone please point the LRM section where I can try to understand the expected simulator behaviour here?

 

Thanks

Melvin

[Mark Strickland]

Section 18.5.14 of the 2017 LRM says "when there is no solution that satisfies all active hard constraints (if any) simultaneously with a constraint defined as soft, the solver shall discard that soft constraint and find a solution that satisfies the remaining constraints".  In this case, there is a solution that satisfies all the constraints, unaligned_transfer == 0.  I am guessing you hoped that the "solve..before" would mean unaligned_transfer was first chosen sometimes to be 1, thus making the soft constraint not satisfiable.  However, 18.5.10 states "solve..before" does not change legal sets, just distributions within the legal set: "Adding this order constraint does not change the set of legal value combinations, but alters their probability of occurrence"

[SV_UVM_Infant]

Thank you Mark for the crisp answer!