前言
生成NxN的像素矩阵是对图像进行各类滤波操作的基本前提，本文介绍一种通过bram生成3x3矩阵的方法。
程序
生成bram核
因为本文介绍的是基于bram生成的3x3像素矩阵，所以要先生成两个bram核，用于缓存前两行图像数据
在 IP catalog中选择Block Memory Generator
配置如下
注意这里选择simple dual port RAM，即伪双端口，一个端口只能写，一个端口只能读
端口A用于写入数据，注意数据的位宽要与图像位深相同，彩色通常为24位，灰度图为8位，数据深度为一行像素的长度，operating选择写优先，enable port type选择始终使能
端口B用于读取数据，这里要注意下面的primitives output register要勾选上，勾选该选项后，数据的输出会延迟一个时钟周期，用于对齐数据。
HDMI时序生成模块
这里也用到了HDMI时序生成模块，具体作用和前面文章讲的一样，一是可以做到通过同步信号简化对图像数据的管理，二是可以让测试的数据处理模块更方便的适配用HDMI显示的图像处理工程。
具体代码如下
module hdmi_tim_gen(input clk ,input rst_n ,input [23:0] data_in ,output hdmi_hs , //行同步信号output hdmi_vs , //场同步信号output hdmi_de , //数据使能output [23:0] hdmi_data , //RGB888颜色数据output reg data_req );//1280*720 分辨率时序参数
parameter H_SYNC = 11'd40; //行同步
parameter H_BACK = 11'd220; //行显示后沿
parameter H_DISP = 11'd1280; //行有效数据
parameter H_FRONT = 11'd110; //行显示前沿
parameter H_TOTAL = 11'd1650; //行扫描周期parameter V_SYNC = 11'd5; //场同步
parameter V_BACK = 11'd20; //场显示后沿
parameter V_DISP = 11'd720; //场有效数据
parameter V_FRONT = 11'd5; //场显示前沿
parameter V_TOTAL = 11'd750; //场扫描周期//reg define
reg [11:0] cnt_h;
reg [11:0] cnt_v;reg [10:0] pixel_xpos;
reg [10:0] pixel_ypos;assign hdmi_de = data_req;
assign hdmi_hs = ( cnt_h < H_SYNC ) ? 1'b0 : 1'b1; //行同步信号赋值
assign hdmi_vs = ( cnt_v < V_SYNC ) ? 1'b0 : 1'b1; //场同步信号赋值//RGB888数据输出
assign hdmi_data = hdmi_de ? data_in : 24'd0;//请求像素点颜色数据输入
always @(posedge clk or negedge rst_n) beginif(!rst_n)data_req <= 1'b0;else if(((cnt_h >= H_SYNC + H_BACK - 2'd2) && (cnt_h < H_SYNC + H_BACK + H_DISP - 2'd2))&& ((cnt_v >= V_SYNC + V_BACK) && (cnt_v < V_SYNC + V_BACK+V_DISP)))data_req <= 1'b1;elsedata_req <= 1'b0;
end//像素点x坐标
always@ (posedge clk or negedge rst_n) beginif(!rst_n)pixel_xpos <= 11'd0;else if(data_req)pixel_xpos <= cnt_h + 2'd2 - H_SYNC - H_BACK ;else pixel_xpos <= 11'd0;
end//像素点y坐标
always@ (posedge clk or negedge rst_n) beginif(!rst_n)pixel_ypos <= 11'd0;else if((cnt_v >= (V_SYNC + V_BACK)) && (cnt_v < (V_SYNC + V_BACK + V_DISP)))pixel_ypos <= cnt_v + 1'b1 - (V_SYNC + V_BACK) ;else pixel_ypos <= 11'd0;
end//行计数器对像素时钟计数
always @(posedge clk or negedge rst_n) beginif (!rst_n)cnt_h <= 11'd0;else beginif(cnt_h < H_TOTAL - 1'b1)cnt_h <= cnt_h + 1'b1;else cnt_h <= 11'd0;end
end//场计数器对行计数
always @(posedge clk or negedge rst_n) beginif (!rst_n)cnt_v <= 11'd0;else if(cnt_h == H_TOTAL - 1'b1) beginif(cnt_v < V_TOTAL - 1'b1)cnt_v <= cnt_v + 1'b1;else cnt_v <= 11'd0;end
endendmodule生成3x3像素矩阵的顶层模块
module kernel_3x3_gen
(input clk, input rst_n, //准备要进行处理的图像数据input vs_i,input de_i,input [23:0] data_i,//矩阵化后的图像数据和控制信号output vs_o,output de_o,output reg [23:0] mat11, output reg [23:0] mat12,output reg [23:0] mat13,output reg [23:0] mat21, output reg [23:0] mat22, output reg [23:0] mat23,output reg [23:0] mat31, output reg [23:0] mat32, output reg [23:0] mat33
);//wire define
wire [23:0] row1_data; //第一行数据
wire [23:0] row2_data; //第二行数据
wire de_i_en ;//reg define
reg [23:0] row3_data; //第三行数据，即当前正在接受的数据
reg [23:0] row3_data_d0;
reg [23:0] row3_data_d1;
reg [23:0] row2_data_d0;
reg [1:0] vs_i_d;
reg [1:0] de_i_d;assign de_i_en = de_i_d[0] ;
assign vs_o = vs_i_d[1];
assign de_o = de_i_d[1] ;//当前数据放在第3行
always@(posedge clk or negedge rst_n) beginif(!rst_n)row3_data <= 0;else begin if(de_i)row3_data <= data_i ;elserow3_data <= row3_data ;end
end//用于存储列数据的RAM
line_shift u_line_shift
(.clk (clk),.de_i (de_i),.data_i (data_i), //当前行的数据.data1_o (row2_data), //前一行的数据.data2_o (row1_data) //前前一行的数据
);//将同步信号延迟两拍，用于同步化处理
always@(posedge clk or negedge rst_n) beginif(!rst_n) begin vs_i_d <= 0;de_i_d <= 0;endelse begin vs_i_d <= { vs_i_d[0], vs_i };de_i_d <= { de_i_d[0], de_i };end
endalways @(posedge clk or negedge rst_n)beginif(!rst_n)beginrow3_data_d1 <= 0;row3_data_d0 <= 0;row2_data_d0 <= 0;endelse beginrow3_data_d0 <= row3_data;row3_data_d1 <= row3_data_d0;row2_data_d0 <= row2_data;end
end//在同步处理后的控制信号下，输出图像矩阵
always@(posedge clk or negedge rst_n) beginif(!rst_n) begin {mat11, mat12, mat13} <= 0;{mat21, mat22, mat23} <= 0;{mat31, mat32, mat33} <= 0;endelse if(de_i_en) begin {mat11, mat12, mat13} <= {mat12, mat13, row1_data};{mat21, mat22, mat23} <= {mat22, mat23, row2_data_d0};{mat31, mat32, mat33} <= {mat32, mat33, row3_data_d1};endelse begin {mat11, mat12, mat13} <= 0;{mat21, mat22, mat23} <= 0;{mat31, mat32, mat33} <= 0;end
endendmodule
行移位模块
module line_shift(input clk,input de_i,input [23:0] data_i, //当前行的数据output [23:0] data1_o, //前一行的数据output [23:0] data2_o //前前一行的数据
);//reg define
reg de_i_d0;
reg de_i_d1;
reg de_i_d2;
reg [10:0] ram_rd_addr;
reg [10:0] ram_rd_addr_d0;
reg [10:0] ram_rd_addr_d1;
reg [23:0] data_i_d0;
reg [23:0] data_i_d1;
reg [23:0] data_i_d2;
reg [23:0] data1_o_d0;//在数据到来时，RAM的读地址累加
always@(posedge clk)beginif(de_i)ram_rd_addr <= ram_rd_addr + 1 ; elseram_rd_addr <= 0 ;
end//将数据使能延迟两拍
always@(posedge clk) beginde_i_d0 <= de_i;de_i_d1 <= de_i_d0;de_i_d2 <= de_i_d1;
end//将RAM地址延迟2拍
always@(posedge clk ) beginram_rd_addr_d0 <= ram_rd_addr;ram_rd_addr_d1 <= ram_rd_addr_d0;
end//输入数据延迟3拍送入RAM
always@(posedge clk)begindata_i_d0 <= data_i;data_i_d1 <= data_i_d0;data_i_d2 <= data_i_d1;
end//用于存储前一行图像的RAM
blk_mem_gen_0 u_ram_1024x8_0(.clka (clk),.wea (de_i_d2),.addra (ram_rd_addr_d1), //在延迟的第三个时钟周期，当前行的数据写入RAM0.dina (data_i_d2),.clkb (clk),.addrb (ram_rd_addr), .doutb (data1_o) //延迟一个时钟周期，输出RAM0中前一行图像的数据
);//寄存前一行图像的数据
always@(posedge clk)begindata1_o_d0 <= data1_o;
end//用于存储前前一行图像的RAM
blk_mem_gen_0 u_ram_1024x8_1(.clka (clk),.wea (de_i_d1),.addra (ram_rd_addr_d0),.dina (data1_o_d0), //在延迟的第二个时钟周期，将前一行图像的数据写入RAM1.clkb (clk),.addrb (ram_rd_addr),.doutb (data2_o) //延迟一个时钟周期，输出RAM1中前前一行图像的数据
);endmodule
仿真模块
`timescale 1ns/1nsmodule pic_tb();reg clk,rst_n ;reg [23:0] data_in ;
wire hdmi_hs,hdmi_vs,hdmi_de ;
wire [23:0] hdmi_data ;
wire data_req ;reg vs_i,de_i ;
wire vs_o,de_o ;
wire [23:0] mat11, mat12, mat13 ;
wire [23:0] mat21, mat22, mat23 ;
wire [23:0] mat31, mat32, mat33 ;
//延迟1clk，与data同步,hdmi时序中，data比de延迟了一个时钟周期
always @(posedge clk)beginvs_i <= hdmi_vs;de_i <= hdmi_de;
endinitial beginclk = 1;rst_n = 0;#20 rst_n = 1;
end
always #10 clk = ~clk;reg [23:0] img[0:1280*720-1];
reg [31:0] addr;
initial begin$readmemh("D:/pic/img2txt.txt",img);
endalways @(posedge clk or negedge rst_n)beginif(!rst_n)beginaddr <= 0 ;data_in <= 0 ;endelse if(data_req) begindata_in <= img[addr];addr <= addr + 1;if(addr == (1280*720-1))addr <= 0;end
endhdmi_tim_gen u_hdmi_tim_gen(.clk (clk), .rst_n (rst_n),//input.data_in (data_in),//output.hdmi_hs (hdmi_hs),.hdmi_vs (hdmi_vs),.hdmi_de (hdmi_de),.hdmi_data (hdmi_data),.data_req (data_req)
);kernel_3x3_gen u_kernel_3x3_gen(.clk (clk), .rst_n (rst_n),//预处理灰度数据.vs_i (vs_i),.de_i (de_i), .data_i (hdmi_data),//输出3x3矩阵.vs_o (vs_o),.de_o (de_o),.mat11 (mat11), .mat12 (mat12), .mat13 (mat13),.mat21 (mat21), .mat22 (mat22), .mat23 (mat23),.mat31 (mat31), .mat32 (mat32), .mat33 (mat33)
);endmodule整体架构
仿真结果
截取部分数据结果
mat31、mat32、mat33是第一行数据（最先输入的那一行），mat11、mat12、mat13是第三行数据（最后输入的那一行），可以看见数据的移位满足像素矩阵的要求。
