xapp1167与TRD14.4 关系



xapp1167与TRD14.4 关系

xapp1167与TRD14.4 都是学习zynq的最佳参考设计。都是基于ZC702，而不是zedboard的，如果要用zedboard那么必须修改硬件配置，因为在702板子中HDMI用的是软核，而zedboard中用的芯片，另外还有网卡等等都不一样配置也不一样。

1：xapp1167   :http://xgoogle.xilinx.com/search?getfields=*&numgm=5&filter=0&proxystylesheet=xilinx_c&client=xilinx_c&site=EntireSite_c&btnG=Google+Search&ie=UTF-8&oe=UTF-8&output=xml_no_dtd&requiredfields=-Archived%3Atrue&sort=date%3AD%3AL%3Ad1&q=XAPP1167.ZIP&lang2search=

2：TRD14 :http://www.wiki.xilinx.com/Zynq+Base+TRD+14.3

3：还有一个比较好的资料是workshop:http://china.xilinx.com/support/university/professors.html

TRD

TRD中包含5个文件，其中HW_platform、 zynq_fsbl、 standalone_bsp，外加uboot.elf 用于BOOT.BIN生成，sobel_qt用于界面应用程序，sobel_cmd包含了软件实现sobel代码实现和VDMA的驱动程序等等，只不过没有在LINUX环境中编译，用的是SDK中编译环境，其实编译器还是LINUX编译链。

TRD是用了sobel的FPGA硬件IP核与软件实现sobel算法比较，如下图是sobel的硬件核输入输出是AXI-Stream流，配合一个VDMA（HP接口）用作视频流与DDR的数据交换

以下是TRD中软件实现sobel的函数  img_filters.c

/*****************************************************************************
 *
 * @file img_filters.h
 *
 * Implementation of software Sobel filter.
 *
 ******************************************************************************/

#include <stdio.h>
#include <malloc.h>
#include "img_filters.h"

static ZNQ_U8 line_buffer[3][1920];		// considering max width as 1920 TODO: change this in more professional way.

unsigned char rgb2y(ZNQ_U8 R, ZNQ_U8 G, ZNQ_U8 B)
{
	unsigned char y;
	y = ((66 * R + 129 * G + 25 * B + 128) >> 8) + 16;
	return y;
}

void img_process(ZNQ_S32 *rgb_data_in, ZNQ_S32 *rgb_data_out, int height, int width, int stride)
{

	short x_weight;
	short y_weight;
	short edge_weight;
	ZNQ_U8 edge_val;

	ZNQ_U8 in_R, in_G, in_B;

	ZNQ_INT pix_i = 0, pix_j = 0;

	/* Clearing 0th row : Sobel filter is not applied on 0th ROW: to have consistent data flow */
	for(pix_j = 0; pix_j < width; pix_j++)
	{
		rgb_data_out[pix_j] = 0;
	}
	rgb_data_out = rgb_data_out + stride;

	/* convert initial(i-1 & i) 2-rows of RGB pixel data into line buffers after conv. into Y(luminance) */

	for(pix_j = 0; pix_j < (width + 1); pix_j++)
	{
		//line_buffer[0][pix_j]  = (((ZNQ_INT)(in_pix[0][pix_j].R + in_pix[0][pix_j].G + in_pix[0][pix_j].B) + 64) >> 2) ;
		in_B = ((*rgb_data_in) & 0x000000FF);
		in_G = (((*rgb_data_in) >>8 )& 0x000000FF);
		in_R = (((*rgb_data_in) >>16)& 0x000000FF);
		line_buffer[0][pix_j]  = rgb2y(in_R, in_G, in_B) ;

		rgb_data_in++;
	}
	rgb_data_in = rgb_data_in + stride - width - 1;	/*  In-case Stride is different from WIDTH of the row */

	for(pix_j = 0; pix_j < (width + 1); pix_j++)	/* 1st row */
	{
		/* line_buffer[1][pix_j]  = (((ZNQ_INT)(in_pix[1][pix_j].R + in_pix[1][pix_j].G + in_pix[1][pix_j].B) + 64) >> 2) ; */
		in_B = ((*rgb_data_in) & 0x000000FF);
		in_G = (((*rgb_data_in) >>8 )& 0x000000FF);
		in_R = (((*rgb_data_in) >>16)& 0x000000FF);
		line_buffer[1][pix_j]  = rgb2y(in_R, in_G, in_B) ;

		rgb_data_in++;
	}
	rgb_data_in = rgb_data_in + stride - width - 1;	/*  In-case Stride is different from WIDTH of the row */
	/* ++ went 1beyond so (-1) */

	for(pix_i = 2; pix_i < height; pix_i++)
	{

		/* copy (pix_i) row into line_buffer, i.e: line_buffer[2] <-- row[pix_i] */
		for(pix_j = 0; pix_j < (width + 1); pix_j++)
		{
			/* line_buffer[2][pix_j]  = (((ZNQ_INT)(in_pix[pix_i][pix_j].R + in_pix[pix_i][pix_j].G + in_pix[pix_i][pix_j].B) + 64) >> 2) ;	*/ //new pixel...rgb2y(temp1);
			in_B = ((*rgb_data_in) & 0x000000FF);
			in_G = (((*rgb_data_in) >>8 )& 0x000000FF);
			in_R = (((*rgb_data_in) >>16)& 0x000000FF);
			line_buffer[2][pix_j]  = rgb2y(in_R, in_G, in_B) ;

			rgb_data_in++;
		}
		rgb_data_in = rgb_data_in + stride - width - 1;	/*  In-case Stride is different from WIDTH of the row */

		*rgb_data_out = 0; // making first pixel zero.

		/* compute Sobel filtering over (pix_i - 1) pixels, i.e: sobel_filtering(&line_buffer[1][0])*/
		for(pix_j = 1; pix_j < (width + 1); pix_j++)
		{
			x_weight =  line_buffer[0][pix_j-1]*(-1) + line_buffer[0][pix_j+1]	 +   line_buffer[1][pix_j-1]*(-2) + line_buffer[1][pix_j+1]*(2) + line_buffer[2][pix_j-1]*(-1) + line_buffer[2][pix_j+1] ;
			//	x_weight =		 window[0][0]*(-1) +		    window[0][2]		 +	      window[1][0]*(-2)		  +		 window[1][2]*(2)		+      window[2][0]*(-1)	   +	  window[2][2] ;
			y_weight =  line_buffer[0][pix_j-1]	   + line_buffer[0][pix_j]*(2) +   line_buffer[0][pix_j+1] + line_buffer[2][pix_j-1]*(-1)  + line_buffer[2][pix_j]*(-2)   + line_buffer[2][pix_j+1]*(-1) ;
			//	y_weight =	 	 window[0][0]		   +	  window[0][1]*(2)	   +        window[0][2]	   +	  window[2][0]*(-1)		   +      window[2][1]*(-2)		  +		 window[2][2]*(-1) ;

			edge_weight = ABS(x_weight) + ABS(y_weight);

			//Added RGB fix-06-dec-12
			if(edge_weight < 255)
				edge_val = (255-(unsigned char)(edge_weight));
			else
				edge_val = 0;

			if(edge_val > 200)
				edge_val = 255;
			else if(edge_val < 100)
				edge_val = 0;

			//out_pix2_ptr = &(out_pix2[0][0]);
			// out_pix2[pix_i-1][pix_j].R = out_pix2[pix_i-1][pix_j].G = out_pix2[pix_i-1][pix_j].B = edge_val;		// write the edge value into RGB
			//////////////////////////////////////////////
			{
				rgb_data_out++;	/* col starts from 1 NOT 0...???*be careful here */
				*rgb_data_out = (edge_val) | (edge_val<<8) | (edge_val<<16) | 0xFF000000;
			}

		}	/* --Sobel_filtering(row[pix_i - 1]) complete */
		rgb_data_out = rgb_data_out + stride - width;	/*  In-case Stride is different from WIDTH of the row */

		/* copy the history of pixel data */
		/* row[pix_i - 2] <-- row[pix_i - 1], i.e: line_buf[0] <-- line_buf[1] */
		/* row[pix_i - 1] <-- row[pix_i], i.e: line_buf[1] <-- line_buf[2] */
		for(pix_j = 0; pix_j < (width /* PBJ no need of '+ 1' */); pix_j++)
		{
			line_buffer[0][pix_j] = line_buffer[1][pix_j] ;
			line_buffer[1][pix_j] = line_buffer[2][pix_j] ;
		}
	}
	// clearing the last line
	for(pix_j=0; pix_j < width; pix_j++)
	{
		rgb_data_out[pix_j] = 0;
	}
}

XAPP1167

Accelerating OpenCV Applications with Zynq-7000 All Programmable  SoC using Vivado HLS Video Libraries

xapp1167主要是用HLS实现sobel算法，在HLS中用的OPENCV库编写，然后生成pcore加入EDK中，

XAPP1167包含

4. DESIGN FILE HIERARCHY

=========================

The directory structure underneath this top-level folder is described

below:

|-- boot_image/

|   |-- bootgen.sh              # Script to build BOOT.bin

|   |-- bootimage.bif           # Include file names to create boot image

|   |-- system.bit              # Hardware bitstream to configure the FPGA

|   |-- u-boot.elf              # u-boot Boot Loader executable

|   |-- zynq_fsbl.elf           # Zynq First stage boot loader executable

|

|-- hw/

|   |-- xps_proj/               # XPS hardware project for TRD design

|

|-- opencv_install/

|   |-- arm_Linux/              # OpenCV libraries and header files for ARM

|

|-- sd_image/

|   |-- BOOT.bin                # Zynq Boot Image, which is created from

|   |                             FSBL, hardware bitstream and u-boot

|   |-- devicetree.dtb          # Linux device tree binary, loaded into

|   |                             memory by u-boot (1080p resolution)

|   |-- init.sh                 # Script to set up environment, sourced at the

|   |                             end of the boot process

|   |-- opencv_lib.img          # Pre-compiled OpenCV libraries required to

|   |                             run the application

|   |-- run.sh                  # Wrapper script for application

|   |-- uramdisk.image.gz       # Ramdisk image, loaded into memory by u-boot

|   |-- uImage                  # Linux kernel image, loaded into memory by

|   |                             u-boot

|   |-- video_library_cmd       # Cmd line based Linux application executable

|

|-- sw/

|   |-- demo/                   # Demo design for OpenCV-acceleration

|   |-- fast-corner/            # Corner-detection design for OpenCV-acceleration

|   |-- share/                  # Shared source files for ARM Linux application

|

|-- README.txt                  # This file

|

|-- Video_Library_Windows.bat   # The batch file to start TRD on Windows

To enable easily building the reference designs, a Makefile is also provided

with the following targets:

‘make elf‘       -- build sw (video_library_cmd)

‘make sim‘       -- build and run csim test

‘make core‘      -- synthesis and export pcore

‘make bitstream‘ -- generate bistream (system.bit)

‘make boot‘      -- generate boot image (BOOT.bin)

‘make all‘       -- build sw and hw, generate sd_image

Running ‘make all‘ from the Linux command line or the Vivado HLS Command Prompt under Windows that launched by provided batch file rebuilds the entire design and generates an SD card image that can be run. Detailed instructions to Build Hardware
and Software for this TRD are provided in XAPP1167.

我们用vivado HLS command 窗口进去E:\vivado_project\xapp1167_ise\sw\demo

make all

然后在目录中生成所有的硬软件源文件，包括EDK，HLS。。工程

最后这句话我们终于明白了：

XAPP1167是详细解释了怎么实现TRD的软硬件过程，而TRD仅仅是个操作步骤流程图说明，没有带任何具体实现

HLS OPENCV 实现sobel

***************************************************************************/
#include "top.h"

void image_filter(AXI_STREAM& INPUT_STREAM, AXI_STREAM& OUTPUT_STREAM, int rows, int cols) {
    //Create AXI streaming interfaces for the core
#pragma HLS INTERFACE axis port=INPUT_STREAM
#pragma HLS INTERFACE axis port=OUTPUT_STREAM

#pragma HLS RESOURCE core=AXI_SLAVE variable=rows metadata="-bus_bundle CONTROL_BUS"
#pragma HLS RESOURCE core=AXI_SLAVE variable=cols metadata="-bus_bundle CONTROL_BUS"
#pragma HLS RESOURCE core=AXI_SLAVE variable=return metadata="-bus_bundle CONTROL_BUS"

#pragma HLS INTERFACE ap_stable port=rows
#pragma HLS INTERFACE ap_stable port=cols

    RGB_IMAGE img_0(rows, cols);
    RGB_IMAGE img_1(rows, cols);
    RGB_IMAGE img_2(rows, cols);
    RGB_IMAGE img_3(rows, cols);
    RGB_IMAGE img_4(rows, cols);
    RGB_IMAGE img_5(rows, cols);
    RGB_PIXEL pix(50, 50, 50);
#pragma HLS dataflow
    hls::AXIvideo2Mat(INPUT_STREAM, img_0);
    hls::Sobel<1,0,3>(img_0, img_1);
    hls::SubS(img_1, pix, img_2);
    hls::Scale(img_2, img_3, 2, 0);
    hls::Erode(img_3, img_4);
    hls::Dilate(img_4, img_5);
    hls::Mat2AXIvideo(img_5, OUTPUT_STREAM);
}

参考

==============

XAPP1167 - Accelerating OpenCV Applications with Zynq-7000 All Programmable SoC using Vivado HLS Video Libraries

UG902 - Vivado Design Suite User Guide: High-Level Synthesis

US873 - Zynq Concepts, Tools and Techniques

UG798 - Xilinx Design Tools: Installation and Licensing Guide

UG926 - Zynq-7000 ZC702 Evaluation Kit Getting Started Guide

UG850 - ZC702 Evaluation Board User Guide

http://www.wiki.xilinx.com/Zynq+Base+TRD+14.5



时间： 2024-11-03 01:25:09

xapp1167与TRD14.4 关系的相关文章

Hibernate的七种映射关系之七种关联映射（二）

继续上篇博客 七.Hibernate双向一对多关联映射:让多的一端来维护关系. 主要是解决一对多单向关联的缺陷,而不是需求驱动的. 1.在Student.java实体类里添加Classes引用.private Classes classes; 2.Student.hbm.xml里添加many-to-one标签:<many-to-one name="classes" column="classesid"/>.Classes.hbm.xml在例子(六)里的那

Hibernate的七种映射关系之七种关联映射（一）

关联映射就是将关联关系映射到数据库里,在对象模型中就是一个或多个引用. 一.Hibernate多对一关联映射:就是在"多"的一端加外键,指向"一"的一端. 比如多个学生对应一个班级,多个用户对应一个级别等等,都是多对一关系. 1."多"端实体加入引用"一"端实体的变量及getter,setter方法. 比如说多个学生对应一个班级,在学生实体类加入:private Grade grade; 2."多"端配置文

CRS和ASM有啥关系

CRS和ASM没有关系 CRS是Oracle 10gR1 RAC后推出了自身的集群软件,这个软件的名称叫做Oracle Cluster Ready Service(Oracle集群就绪服务),简称CRS ASM是Oracle 10g R2后为了简化Oracle数据库的管理而推出来的一项新功能,这是Oracle自己提供的卷管理器,主要用于替代操作系统所提供的LVM,它不仅支持单实例,同时对RAC的支持也是非常好.ASM可以自动管理磁盘组并提供有效的数据冗余功能. 总上所述,CRS是一个集群软件,只

其实你的痛苦 跟别人完全没有关系

http://foxue.qq.com/a/20151009/043881.htm 文:净界法师 诸位要知道一个事情,当你看到某一个人,你起快乐的感受,你跟他之间有善业的因缘,一定是善业的因缘.跟他没有关系,跟你你自己有关系. 佛教的根本思想都是莫向外求.其实,你今生遇到什么人.碰到什么事,你今生会有多大的快乐.你有多大的痛苦,跟别人完全没有关系,完全没有关系,别人只是一个助缘,他顶多是刺激你的业力显现出来,他只是助缘. 你看到某一个人你感到痛苦,跟他也没有关系,那是你过去生有某一方面的罪业.这

JOSNObject与JSONArray的关系

这里简单介绍下JOSNObject与JSONArray的关系. JOSNObject:json对象      用{}表示 JSONArray:json数据       用 [ ] 表示 服务器返回的json基本是这两种形式的搭配使用,他们之间可以互相嵌套使用,使用起来比较简单,不多说,上图: private void text() throws Exception{     //对象     JSONObject jsonObject1=new JSONObject();     jsonObj

PHP 方法重写override 与 抽象方法的实现之间的关系

重写由final关键字决定,但受父类的访问权限限制 实现基于继承,所以实现父类的抽象方法必须可访问到,父类抽象方法不可为private 1.父类某方法能否被子类重写与此方法的访问级别无关 public protected private对某方法内否被重写没有影响,能否被重写要看此方法是否被final修饰(final类不可被继承,final方法不可被重写) 2.但重写方法要受到访问级别的限制,即访问权限不可提升规定 (不仅仅是PHP,其他面向对象语言依然适用),访问权限只可以降低,不可以提升. 3

点圆的关系---1

输入代码: /* *Copyright (c)2015,烟台大学计算机与控制工程学院 *All rights reserved. *文件名称:sum123.cpp *作 者:林海云 *完成日期:2015年6月12日 *版 本 号:v2.0 * *问题描述:(1)先建立一个Point(点)类,包含数据成员x,y(坐标点): (2)以Point为基类,派生出一个Circle(圆)类,增加数据成员(半径),基类的成员表示圆心: (3)编写上述两类中的构造.析构函数及必要运算符重载函数(本项目主要是输入

使用maintainJ快速了解项目类关系

MaintainJ能够在很短的时间内进行debug(找出缺陷).能够方便的了解项目已有特性从而容易进行功能增强(enhance).能够展现调用耗时从而能够快速定位瓶颈调整系统提升系统性能.此外利用它还能快速的建立与实际调用一致的UML 类关系视图和调用时序视图,这也是形成文档的关键资料.而这些工作的快速完成均是由于MaintainJ是一款基于AOP实现的能够在操作具体用例(Use case)过程中时时形成调用关系.类关系以及传递的参数和响应时间等等的原因.MaintainJ能够对java web

理清javascript中prototype、__proto__、Object、Function的关系，更好地理解原型继承

本文参考了http://www.blogjava.net/heavensay/archive/2013/10/20/405440.html这篇文章,对其内容作了个简单总结,形成了几条简单的结论,让读者更容易记住prototype.__proto__.Object.Function之间的关系. 结论1:Object.prototype只是一个普通对象,它是js原型链的最顶端. (typeof Object.prototype) === object;//true Object.prototype.