阻塞赋值和非阻塞赋值深度解析——仿真事件的调度
发布者:jackzhang
时间:2015-09-07 13:57:08
来源:EETOP acgoal 的 BLOG
下面有一段verilog代码和仿真文件,用的是VCS仿真和编译工具。我们来研究一下不同的驱动赋值方式对仿真结果的影响。下面我把我做的例子和大家分享一下。
设计源代码如下:
`timescale 1ns/1psmodule counter (data_out0, data_out1,clk,rst_n, data_in0, data_in1);output [3:0] data_out0;output [3:0] data_out1;input [3:0] data_in0;input [3:0] data_in1;input clk;input rst_n;reg [3:0] data_out0;reg [3:0] data_out1;always @(posedge clk or negedge rst_n) begin if(!rst_n) begin data_out0 <= 4'd0; end else begin data_out0 <= data_in0; endendalways @(posedge clk) begin if(!rst_n) begin data_out1 <= 4'd0; end else begin data_out1 <= data_in1; endendendmodule
这里有两段,分别表示带异步复位和带同步复位的。测试程序如下:
`timescale 1ns/1psmodule tb_top;reg clk;reg rst_n;wire [3:0] data_out0;wire [3:0] data_out1;reg [3:0] data_in0;reg [3:0] data_in1;counter
u_counter0(.clk(clk), .rst_n(rst_n), .data_out0(data_out0),
.data_out1(data_out1), .data_in0(data_in0), .data_in1(data_in1));initial begin clk=1'b0;endalways begin #3 clk=~clk;endinitial begin rst_n = 1'b1; #15 rst_n = 1'b0; #180 rst_n = 1'b1; #200 $finish;endinitial begin data_in0 <= 0; wait(!rst_n); wait(rst_n); @(posedge clk); data_in0 <= 1; @(posedge clk); data_in0 <= 2; @(posedge clk); data_in0 <= 4; @(posedge clk); data_in0 <= 8; @(posedge clk); data_in0 <= 0;endinitial begin data_in1 <= 0; wait(!rst_n); wait(rst_n); @(posedge clk); data_in1 <= 1; @(posedge clk); data_in1 <= 2; @(posedge clk); data_in1 <= 4; @(posedge clk); data_in1 <= 8; @(posedge clk); data_in1 <= 0;endinitial begin $vcdpluson();endendmodule注意这里红色的部分,data0和data1的驱动都用非阻塞赋值。得到的仿真波形图如下:下面我把非阻塞赋值改成阻塞赋值,改动部分如下:initial begin data_in0 = 0; wait(!rst_n); wait(rst_n); @(posedge clk); data_in0 = 1; @(posedge clk); data_in0 = 2; @(posedge clk); data_in0 = 4; @(posedge clk); data_in0 = 8; @(posedge clk); data_in0 = 0;endinitial begin data_in1 = 0; wait(!rst_n); wait(rst_n); @(posedge clk); data_in1 = 1; @(posedge clk); data_in1 = 2; @(posedge clk); data_in1 = 4; @(posedge clk); data_in1 = 8; @(posedge clk); data_in1 = 0;end这样的代码,仿真波形图如下:和上面的不一样,为什么呢,为什么这次的flop行为没有了呢?直接给结论吧。根据verilog/system verilog的standard。我这里以system verilog的standard为例,STD 1800-2009版本,第四章“Scheduling Semantics”,这一章关于仿真的调度描述。
simulation
的event大体分为:Active event -> Inactive event -> NBA event ->
Overserved event -> Reactive -> Re-Inactive -> Re-NBA
一般的,阻塞赋值发生在active/inactive event,而NBA( Non-blocking assignment update)从字面意思就能看出来是做非阻塞赋值的。
再
结合上面的例子,如果data_in0是阻塞赋值出去的话,因为例子里的dut是在时钟上升延来的时候进行非阻塞赋值(data_out0 <=
data_in0),这句话属于NBA event,而阻塞赋值data_in0 = 1 (或者2,4,8)是Active
event,他比NBA要先执行,所以我们看到data_out0在当前的cycle就被采样到了。这就是后面一个波形图的样子,看起来就像flop没有
用一样。实际上是我们driver写的有问题。
如
果data_in0在testbench里面用阻塞赋值就是 data_in0 <= 1 (或者2,4,8,见例子),那么data_out0
<= data_in0, 和data_in0 <= 1同属于NBA
event。这样子这两句话同时执行,这样就和实际电路中的flop行为一致了(就是并行执行)。