Simulation Behaviour of Associative Array with 64 Bit Index - QuestaSIM


Hi All,

This post is related to unexpected Simulation Behaviour with an Associative Array having 64 Bit Index in QuestaSIM. I had faced this problem and thought many others may have faced it also. So it is worth sharing the solution over here.

NOTE :- This behaviour was observed in QuestaSIM 10.2. It may or may not be observed in the same or other versions of QuestaSIM.

SAMPLE CODE : 

Here is the sample code :

typedef struct packed {
  bit [7:0] data;
  bit        valid;
} dword_memory_st;

`define key 64
program p;
  int loop_count;
  int iteration; // Try with more than 8 to get the Errors ;)
  dword_memory_st MEMORY[bit [`key-1:0]];
  bit [`key-1:0] st = 'hffff_ffff_ffff_fffa;
  bit [`key-1:0] ft = 'hffff_ffff_ffff_ffff;
  bit [`key-1:0] temp;
  bit [`key:0] max_memory_address = (1 << `key);
  bit bounded_address_used = 0;

  initial
  begin
    void'($value$plusargs("ITE=%d", iteration));
    $display("Iteration - %0d", iteration);
    void'($value$plusargs("BOU=%b", bounded_address_used));
    $display("Bounded Address Used - %0b", bounded_address_used);
    $display("Associative Array Index Size - %0d", `key);
    $display("***** ***** ****** ****** ***** ***** ***** ***** ***** ***** ***** ***** ***** ***** ***** *****");
    forever
    begin
      for(bit[`key-1:0] i=0; i<iteration; i++)
      begin
        if(bounded_address_used)
          MEMORY[bounded_address(st+i)] = i + (loop_count*10);
        else
          MEMORY[st+i] = i + (loop_count*10);
         $display("MEMORY[%0h] exists = %0b, MEMORY[%0h] = %0h", st+i, MEMORY.exists(st+i), st+i, MEMORY[st+i]);
        if((st+i) == ft)
          $display("Rollover Happened");
      end
      loop_count++;
      if(loop_count == 3)
       break;
       $display("***** ***** ****** ****** ***** ***** ***** ***** ***** ***** ***** ***** ***** ***** ***** *****");
    end
    $display("***** ***** ****** ****** ***** ***** ***** ***** ***** ***** ***** ***** ***** ***** ***** *****");
  end

  function bit[`key-1:0] bounded_address(input bit[`key:0] addr);
    MEMORY.delete(addr);
    bounded_address = (addr % max_memory_address);
  endfunction : bounded_address
endprogram

PROBLEM :

Now are the results of this sample code :

Result with 63 Bit Index

Result with 64 Bit Index

So as shown in the image, for 63 Bit Index (Or any Index less than 64 Bits), things work perfectly & after rollover also, it works properly. But for 64 Bit Index, after rollover, tool does not allow to overwrite existing indexes (FFFF_FFFF_FFFF_FFFA - 0000_0000_0000_0000) or store new indexes (0000_0000_0000_0001 - 0000_0000_0000_0003)

SOLUTION : 

Before writing/overwriting value for any index, first delete it.

 This solution works, because of unexpected Simulator Behaviour, For 64 Bit index, after Rollover, simulator considers every index exists, but it does not let any index to be overwritten. (That's why for 64 Bit Index Result, Displays of exist method for each index, gives value 1, but while reading from that index, it gives Warning of Nonexistent Associative Array).

Function bounded_address in the sample code, does the same thing.

Here are the result of this sample code with use of bounded_address method.

Result with 63 Bit Index & bounded_address Method Usage

Result with 64 Bit Index & bounded_address Method Usage

Please feel free to comment on this post or for any suggestions. Your suggestions are always welcomed.

Comments

Post a Comment

Popular posts from this blog

SystemVerilog Event Regions

Image
Hi All, In the last post, we had discussed the Verilog Event Regions. Today we'll discuss the event regions in SystemVerilog, which are somewhat similar to the Verilog Event regions only, plus there are some additional regions as well. Similar to Verilog, in Systemverilog also the event queue is divided into 5 different regions. A SystemVerilog simulator, shall maintain some form of data structure that allows events to be dynamically scheduled, executed, and removed as the simulator advances through time. Simulation proceeds by executing and removing all events in the current simulation time slot before moving on to the next nonempty time slot, in time order. And in that given simulation timeslot, it is necessary to execute events of all of the above mentioned regions. Most important thing is that these regions encompass the Verilog Event Regions and hence the legacy Verilog code should run correctly without any modification in SystemVerilog simulator. Prepone Region
Read more

Standard Cell Based ASIC Design Flow

Image
Hi All,  Today we will discuss about ASIC Design Flow, as it is one of the most important and crucial thing to understand. ASIC Design Flow Now, we will try to get brief idea about each steps in the above image. Specification  - Specification of the system, like what is the functionality of system, what is the desired speed of the system, area constraint of the system, whether it has any power constraint or not, tolerances of various constraints, operative conditions of the system. RTL Description  - Generally after specification, the system is partitioned into various blocks and RTL description (HDL Code like Verilog or VHDL), mostly in behaviour form is written. Sometimes Higher Level Language Code like C/C++ is also written, and in that case, a High Level Synthesis Tool is used, which converts the Higher Language Code into RTL Level Code. After RTL Coding, Functional Verification is done to check the correctness of the functionality. Logic Synthesis  - Now, wh
Read more

How to implement more than 1 Analysis Implementation Ports in a Class in UVM

Hi All, In this post, we will discuss about how to include more than 1 Analysis Implementation Port in a Class in UVM. Analysis Ports are used for many purposes like Coverage Collection, Received/Transmitted Packet Checking. And sometimes, you may need to take coverage on different packets (so you will require different analysis implementation ports) and you do not want to define 2 different class, just for coverage. So if there are more than 1 (suppose 2) analysis implementation ports in a class, then obvious (and the first also !!!) question arises in the mind is that, how to implement 2 write methods in the same class. Well, in UVM, we have 2 different methods to address the problem. Method - 1 : Use "imp" Suffix defined via Macro Declare Macro `uvm_analysis_imp_decl(<Postfix>) outside of the component. This macro adds post-fix to the analysis implementation port class name and associated write method. Instantiate Analysis Implementation Port with u
Read more