Block coverage closing implicit case defaults

Hi, when i extract a block coverage i have realized that automatic case defaults are created and they are %0 coverage. Do I have any option to exclude code block from line coverage or close automatic implicit case default creation in line coverage report.

Thanks

Find more posts tagged with

DSim

Accepted answers

David_Altair

> Can you confirm me about last comment I posted here.

Your understanding is correct: neither of them is available today. That said, I am going to bring up the problem to an attention of our R&D.

All comments

David_Altair

> Do I have any option to exclude code block from line coverage ...

I do not understand your question. If you are asking:

Do I have a way to instruct DSim to exclude any implicit default statement from line coverage?

As of today, it is not possible. That is, any occurrence of an implicit default statement is included in line coverage.

That said, if you create a feature request, and enough people upvote it, we may put it on our roadmap.

--------------------------------------------------------------------------------
> Do I have any option to ... close automatic implicit case default creation in line coverage report

I do not understand your question. If you are asking:

How do I close line coverage for an implicit default statement?

Let's take a look at the following code snippet:

logic [15:0] data;
logic [9:0] result;

case (data)
    16'd0: result = 10'b0111111111;
    16'd1: result = 10'b1011111111;
    16'd2: result = 10'b1101111111;
    16'd3: result = 10'b1110111111;
    16'd4: result = 10'b1111011111;
    16'd5: result = 10'b1111101111;
    16'd6: result = 10'b1111110111;
    16'd7: result = 10'b1111111011;
    16'd8: result = 10'b1111111101;
    16'd9: result = 10'b1111111110;
    // here is an implicit default statement
endcase

It is enough to exercise any data value outside of the [16'd0:16'd9] range.

Deniz Güzel

Hi David,

Sorry for misunderstanding. What I am asking was "Even though a designer design a 'case' which covers all the possible values DSIM creates implicit case defaults to these cases and of course they are impossible the cover. I setup a test environment for giving you right info all the integral types which can be 2 state or 4 state and enumareted type cases is imposible the cover unless a designer makes for example 'FETCH' state in default which is highly unaccurate. To sum up, It is impossible the take %100 percentage line coverage from a case whom case variable type can be nearly everything unless 'default' construct is used for possible case variable expectation even if this is a common lint problem.

Some vendors have different approaches to these issue. So that's why I asked, is there an run_time option, pragma, directives or compile flags to control either implicit case creation or excluding a line from line coverage simply giving comment pragmas and as far as I understood neither of them is available today.

I am closing the case.

Thank you

Deniz Güzel

Can you confirm me about last comment I posted here.

Shaun_Altair

Hi Deniz. I have moved this post to the more appropriate forum, DSim -> Coverage.

David_Altair

Hi Deniz,

I understand the problem that you are facing. Do you mind to share a code snippet, so that I may reproduce the problem on our end.

Deniz Güzel

Hi David,

Simpliest code snippet I can share with you related with issue is that :

module test_dut (input bit i_clk, 
                 output bit [1:0] state);

  bit [1:0] state_local;

  always @(posedge i_clk) begin
    case (state_local)
      'b00 : state_local = 'h1;
      'b01 : state_local = 'h2;
      'b10 : state_local = 'h3;
      'b11 : state_local = 'h0;
    endcase
  end

  assign state = state_local;

endmodule

module test_tb ();

  bit i_clk;
  bit [1:0] state;

  always #20 i_clk = ~i_clk;

  test_dut dut (.i_clk(i_clk),
                .state(state)); 

  initial begin
    #500ns;
    $finish;
  end

endmodule

Here there is no way to cover implicit default case state and get %100 line coverage from test_dut because there is no other possible value for 2 bit 2 state "bit" type value.

Expecting designers to put one of the possible states in the default case like :

case (state_local)
      'b00 : state_local = 'h1;
      'b01 : state_local = 'h2;
      'b10 : state_local = 'h3;
      default : state_local = 'h0;
endcase

is generally not feasible. And putting explicit default as :

case (state_local)
      'b00 : state_local = 'h1;
      'b01 : state_local = 'h2;
      'b10 : state_local = 'h3;
      'b11 : state_local = 'h0;
      default : state_local = 'h0;
endcase

Changes nothing as expected.

So basically any design which does not uses possible default values never gets %100 coverage but that is a "wrong negative" if there is literally no other possible way using values "0", "1", "x", "z".

David_Altair

> Can you confirm me about last comment I posted here.

Your understanding is correct: neither of them is available today. That said, I am going to bring up the problem to an attention of our R&D.

Deniz Güzel

Hi David,

Thank you for clarification

Charles_Dawson

While the example in question is simple, most examples will not be so. In order to determine that all possible values are covered by the case statements, the simulator would have to do an evaluation of the solution space of the equation in the case statement. I do not know how feasible it would be to statically evaluate the solution space and determine if there are possible values which would need a default statement. Generally, it is bad coding practice to not have a default for a case statement. What we need is to have the ability exclude lines from coverage consideration.

Deniz Güzel

Hi Charles,

I definetly agree with you which is about not being cost efficent to create implicit case defaults if and only if given case parameter is not totally covered and if case parameter is totally covered then do not create implicit defaults. Also about not using explicit defaults is bad coding practice which ends up as lint problem. However, I have suggested ideas to come up with a solution. One of them is exluding lines from line coverage using pragmas as you said before and also I have suggested that controlling creation of implicit defaults and implicit elses for all over that run maybe using a run flag without identifying whether implicit defaults would be possible or not. Maybe you might also want to consider that idea because of being more feasible and cost efficent rather then identify every branch in coverage scope.

With that said, thank you all for time.