引言
前面我们介绍了流水线的写法(http://blog.csdn.net/rill_zhen/article/details/45980039),流水线是数字设计中很常用的一种设计方法,可以提高运行频率,提高吞吐量。
如果组合逻辑延迟较大,一个时钟周期完成不了时,除了插入寄存器将组合逻辑拆分成流水线外,还可以采用multi-cycle的方式。
multi-cycle的工作机制很简单,从给定输入之后,等待多个周期之后,再去采样输出结果。
本小节我们将通过一个小实验来说明multi-cycle的具体RTL实现。
1,功能介绍
假设有某个模块,其计算量很大,以致延迟较大,一个周期完成不了,需要3个cycle才行。
假设时钟周期是10,这个模块的运算分为“加法,左移,减法”三个操作,分别用时7,8,9。
2,RTL实现
如果在数字设计时,遇到上述模块描述的情况时,可以考虑multi-cycle实现。
具体RTL如下:mc.v
其中关于状态机的写法,我们之前有专门介绍,如有疑问,请参考(http://blog.csdn.net/rill_zhen/article/details/39585367)。
/*
* multi-cycle example
* Rill 2015-05-29
*/
module Mmulti_cycle
(
input clk,
input rst_n,
input en_i,
input [7:0] data_i,
output en_o,
output [7:0] data_o,
output idle
);
//===================
// control path, fsm
//===================
localparam S_IDLE = 4'd0;
localparam S_CYCLE1 = 4'd1;
localparam S_CYCLE2 = 4'd2;
localparam S_CYCLE3 = 4'd3;
reg [3:0] c_state_r;
wire [3:0] n_state;
wire state_changed;
wire cs_idle = (c_state_r == S_IDLE);
wire cs_cycle1 = (c_state_r == S_CYCLE1);
wire cs_cycle2 = (c_state_r == S_CYCLE2);
wire cs_cycle3 = (c_state_r == S_CYCLE3);
wire ns_idle = cs_cycle3;
wire ns_cycle1 = cs_idle & en_i;
wire ns_cycle2 = cs_cycle1;
wire ns_cycle3 = cs_cycle2;
assign state_changed = ns_idle | ns_cycle1 | ns_cycle2 | ns_cycle3;
assign n_state = ( {4{ns_idle}} & S_IDLE
| {4{ns_cycle1}} & S_CYCLE1
| {4{ns_cycle2}} & S_CYCLE2
| {4{ns_cycle3}} & S_CYCLE3
);
always @(posedge clk)
if(~rst_n)
c_state_r <= S_IDLE;
else
c_state_r <= n_state;
//=================
// data path,calc
//=================
wire [7:0] data1;
wire [7:0] data2;
wire [7:0] data3;
assign #7 data1 = data_i + 1'b1;
assign #8 data2 = data1 << 1'b1;
assign #9 data3 = data2 - 1'b1;
assign en_o = cs_cycle3;
assign data_o = data3;
assign idle = cs_idle;
endmodule
3,testbench
具体multi-cycle模块是如何工作的呢,我们需要写个简单的TB验证一下:tb.v
/*
* multi-cycle example
* Rill 2015-05-29
*/
module Ttb;
reg clk;
reg rst_n;
reg en_i_r;
reg [7:0] data_i_r;
wire en_o;
wire [7:0] data_o;
wire idle;
Mmulti_cycle mc0
(
.clk (clk),
.rst_n (rst_n),
.en_i (en_i_r),
.data_i (data_i_r),
.en_o (en_o),
.data_o (data_o),
.idle (idle)
);
initial
begin
clk = 1'b0;
rst_n = 1'b0;
en_i_r = 1'b0;
data_i_r = 8'b0;
fork
forever #5 clk = ~clk;
join_none
repeat(10) @(posedge clk);
rst_n = 1'b1;
repeat(10) @(posedge clk);
@(posedge clk);
en_i_r = 1'b1;
data_i_r = 8'h1;
repeat(10) @(posedge clk);
$finish();
end
endmodule
4,nc脚本和vflist
mc.sh:
#! /bin/bash
#
# mc.sh
# usage: ./mc.sh c/w/r
# Rill create 2014-09-03
#
TOP_MODULE=Ttb
export SRC_DIR=$(pwd)
tcl_file=run.tcl
if [ $# != 1 ];then
echo "args must be c/w/r"
exit 0
fi
if [ $1 == "c" ]; then
echo "compile lib..."
ncvlog -f ./vflist -sv -update -LINEDEBUG;
#ncelab -delay_mode zero -access +rwc -timescale 1ns/10ps ${TOP_MODULE}
ncelab -delay_mode distribute -access +rwc -timescale 1ns/10ps ${TOP_MODULE}
exit 0
fi
if [ -e ${tcl_file} ];then
rm ${tcl_file} -f
fi
touch ${tcl_file}
if [ $1 == "w" ];then
echo "open wave..."
echo "database -open waves -into waves.shm -default;" >> ${tcl_file}
echo "probe -shm -variable -all -depth all;" >> ${tcl_file}
echo "run" >> ${tcl_file}
echo "exit" >> ${tcl_file}
fi
if [ $1 == "w" -o $1 == "r" ];then
echo "sim start..."
ncsim ${TOP_MODULE} -input ${tcl_file}
fi
echo "$(date) sim done!"
vflist:
-incdir ${SRC_DIR}
${SRC_DIR}/mc.v
${SRC_DIR}/tb.v
5,验证
运行mc.sh c; mc.sh w即可得到仿真波形:
通过波形可以看出,mc模块在经过3个cycle之后输出了运算结果3。
6,小结
pipeline和multi-cycle是处理长延迟逻辑常用的两种方式,我们都介绍过了。
Enjoy!
来源:CSDN
作者:Rill
链接:https://blog.csdn.net/rill_zhen/article/details/46240487