#317 Issue indexing an array constraint

[Victor Lyuboslavsky]

The following code should index properly. It compiles and runs on all major simulators. (Even though every constraint solver treats it differently)

class pkt;

  rand bit one_beat;
  rand int len;
  rand bit [31:0] addr;

  constraint wr_c {
    if (!one_beat) { addr[5:0] == 0, len == 15 };
  }

endclass

[Matthew Ballance]

Hmm... I can see how this is being interpreted, but am at a bit of a loss as to how to parse it J

SVEditor's parser is a recursive-descent parser, so when it see '{' after a constraint 'if', it concludes that the 'if' is followed by a constraint block. The intent appears to be for the 'if' to be followed by a concatenation expression. Is that your interpretation?

[Victor Lyuboslavsky]

In this case, '{' could be either a constraint block or an array.

VCS treats the above constraint like this:

verilog:::
if (!one_beat) {
  { addr[5:0] == 0, len == 15 } == 2'b01;
}

[StevenAZ]

This is a tricky one. You would almost have to run a check to see if it is
a concatenation before running the actual parser to run normally which
would slow things down, or find a way to unwind the db in case things go
wrong.

Almost easier to run an IsConcatenation check, because almost anything is
possible following an open brace ( another if, constraint etc.). Would the
check have to be run on every { in a constraint block? Don't know the lrm
well enough.
On Feb 15, 2014 1:18 PM, "Victor Lyuboslavsky" vxl119@users.sf.net wrote:

In this case, '{' could be either a constraint block or an array.

VCS treats the above constraint like this:

verilog:::if (!one_beat) {
{ addr[5:0] == 0, len == 15 } == 2'b01;}

---

• [bugs:#317] http://sourceforge.net/p/sveditor/bugs/317/ Issue indexing
  an array constraint*

Status: open
Created: Sat Feb 15, 2014 04:33 AM UTC by Victor Lyuboslavsky
Last Updated: Sat Feb 15, 2014 08:04 PM UTC
Owner: nobody

The following code should index properly. It compiles and runs on all
major simulators. (Even though every constraint solver treats it
differently)

class pkt;

rand bit one_beat;
rand int len;
rand bit [31:0] addr;

constraint wr_c {
if (!one_beat) { addr[5:0] == 0, len == 15 };
}
endclass

---

Sent from sourceforge.net because you indicated interest in
https://sourceforge.net/p/sveditor/bugs/317/

To unsubscribe from further messages, please visit
https://sourceforge.net/auth/subscriptions/

[Matthew Ballance]

Okay, I have a fix that (hopefully) doesn't add too much overhead. The fix is to do an IsConcat check when entering a potential constraint_set block after if, else, and foreach. The IsConcat check spins the lexer ahead, saving tokens, until a ',' ';' or matching brace is found. If the stopping token is a ',' then it must be a concat.

Admittedly, the heuristic may be a bit crude, but it's unlikely to give false positives.

This code will be in 1.5.3

[Victor Lyuboslavsky]

I assume statements without a ',' will also work, like:

if (!one_beat) { 1'b1  };

[Matthew Ballance]

Yes, that's correct.