Parametrized UVM Verification Environment

[MarcelA]

Hi all

 

I'm trying to build a generlc verification environment for specific modules of mine, that essentially only differ (besides their actual implementation) by the number of inputs/outputs of a certain interface standard and the data_size of each of these interfaces.

 

This leads to the point, that I like to have an environment, where I can set the number of interfaces and for each of these interfaces the data_size.

 

Unfortunately this simple setup of non-dynamic pre-compile settings is getting me in a lot of trouble.

 

(1) Sequence_Item

 

class input_item #(int unsigned data_size) extends uvm_sequence_item;
    `uvm_object_param_utils(input_item)
    
    rand logic[data_size-1:0] data;

   .......

 

The first question that comes to mind writing this code is the following:

 

Is it possible to factory overwrite a param. class with another specilisation of that same class (e.g. define a driver with input_item<32> and then factory overwrite it with input_item<42>)?

Otherwise a item_base class would be necessary that than would be extended by this class.

 

 

(2) Analysis Ports

class env extends uvm_env;
 ....

    // need N = number of interfaces analysis ports between each monitor and scoreboard
    uvm_analysis_port #(input_item#(17)) ap_1;
    uvm_analysis_port #(input_item#(12)) ap_2;
    .....
    uvm_analysis_port #(input_item#(42)) ap_N;


....
    

 

Due to the different input_items all ports are of a different type. Therefore it is not possible to create an array of N=number_of_interfaces, which leads to this not being possible to implement. Furthermore the analysis_port/export classes cannot be overwritten through the means of the factory.

 

 

(3) Virtual Sequence

 

class top_vseq_base extends uvm_sequence #(uvm_sequence_item);
 
`uvm_object_utils(top_vseq_base)
 
      uvm_sequencer #(input_item#(17)) seq_1;
      uvm_sequencer #(input_item#(12)) seq_2;
      ...
      uvm_sequencer #(input_item#(42)) seq_N;

 

In the virtual sequence I essentially run into two problems:

 

1. The first one is the same problem as in (2) of not being able to create an array of different types or having a pre-processor for-loop 

2. The other one is the fact, that I'm not able to get access to the number N . Even if they were all of the same type and I would declare a dynamic array, there is no build_phase and no way to get informations through either the config_db or "p_sequencer.variable". I could put a member variable into the virtual sequencer, but I'm not sure if it is a good idea/possible to create a dynamic array in the body method.

 

 

General Solutions so far:

 

I only see two solutions here ...

 

1. Defining a gigantic input_item with data_size of 256/512 and then cut it everywhere. But unfortunately I will be in need of an array of completely different items in the next version of this environment anyhow. The reason for that is, that I would like to group a bunch of M different interfaces into one environment, all of them running a different item. Therefore the analysis_ports would all run a different item.

 

2. Just building a code generator, in which the user sets all parameters, creating the necessary environment for the given DUT.

 

 

If you have any input, I would be glad to hear it.

 

Thanks

 

Marcel

[dflores]

Hi. Any updates on this?