完善资料让更多小伙伴认识你,还能领取20积分哦, 立即完善>
在我的设计中,我需要延迟一些数据(32位宽)一段时间。
我一直在使用SRL16而不是FFs,它可以很好地延迟超过几个周期。 我还注意到,使用DSP48E1片很容易将数据延迟到48位宽,延迟三个周期,因此只使用1个DSP48而不是16个LUT或96个FF。 我的设计本来并没有真正使用很多DSP,所以这似乎是一种明智的方法(在我的情况下,我认为它们是“丰富的”),但是我的一部分认为使用这么强大的组件来做一些如此微不足道的事情是疯狂的! 我即将开始尝试,但想知道这个话题是否有一些智慧? 谢谢, 李。 以上来自于谷歌翻译 以下为原文 In my design I have a need to delay some data (32bits wide) for a number of cycles. I have been using SRL16's rather than FFs for this and it works really nicely for delays more than a few cycles. I've also noticed that it's pretty easy to use a DSP48E1 slice to delay data up to 48bits wide by three cycles and hence only use 1 DSP48 rather than 16 LUTs or 96 FFs. My design otherwise doesn't really use many DSP's so this seems like a sensible approach (in my case I would argue they are "abundant"), however part of me is thinking its crazy to use such a powerful component for something so trivial! I am about to start experimenting but wondered if there is some wisdom on this topic? Thanks, Lee. |
|
相关推荐
5个回答
|
|
硬件是有用的!
如果通过使用(丰富的)DSP48,您可以释放一些使用频繁的切片逻辑,然后继续前进并执行此操作。 将这样一个强大的组件用于如此简单的事情可能会很疯狂,但另一种方法是让LUT和FF完成所有工作时,强大的组件绝对不做任何事情。 在原帖中查看解决方案 以上来自于谷歌翻译 以下为原文 The hardware is there to be used! If by using a (plentiful) DSP48 you can free up some heavily-used slice logic, then go right ahead and do that. It may be crazy to use such a powerful component for something so simple, but the alternative is to leave the powerful component doing absolutely nothing while the LUTs and FFs do all the work. View solution in original post |
|
|
|
硬件是有用的!
如果通过使用(丰富的)DSP48,您可以释放一些使用频繁的切片逻辑,然后继续前进并执行此操作。 将这样一个强大的组件用于如此简单的事情可能会很疯狂,但另一种方法是让LUT和FF完成所有工作时,强大的组件绝对不做任何事情。 以上来自于谷歌翻译 以下为原文 The hardware is there to be used! If by using a (plentiful) DSP48 you can free up some heavily-used slice logic, then go right ahead and do that. It may be crazy to use such a powerful component for something so simple, but the alternative is to leave the powerful component doing absolutely nothing while the LUTs and FFs do all the work. |
|
|
|
什么@ u4223374说的......
- 如果提供的信息有用,请将答案标记为“接受为解决方案”。给予您认为有用且回复的帖子。 以上来自于谷歌翻译 以下为原文 what @u4223374 said ... - Please mark the Answer as "Accept as solution" if information provided is helpful. Give Kudos to a post which you think is helpful and reply oriented. |
|
|
|
一般来说,我同意其他答案。
使用DSP48(或块RAM)有一个可能的缺点,那就是时序。 虽然DSP48本身非常快,但它们并不像部件中的切片那么丰富。 因此,您最终可能会有更长的路由距离到达最近的DSP48,而最近的SRL(LUTM)将非常接近。 在大型设计中,这种额外的布线距离可能使得更难以满足时序要求。 如果您考虑使用DSP48保存切片,我假设您的设备已满。 如果没有,那么未使用的DSP与未使用的LUT没有什么特别之处,因此使用DSP单元没有太大的优势。 - Gabor 以上来自于谷歌翻译 以下为原文 Generally I agree with the other answers. There is one possible down side to using DSP48 (or block RAM for that matter) and that is timing. While the DSP48 itself is quite fast, they are not as abundant as slices in the part. Therefore it is possible that you will end up with some longer routing distances to reach the nearest DSP48 while the nearest SRL (LUTM) would be quite close. In a large design this extra routing distance could make it harder to meet timing. I assume that your device is getting full if you consider using DSP48 to save slices. If not, then there's nothing special about unused DSP's vs unused LUT's so there's no great advantage to using the DSP units. -- Gabor |
|
|
|
感谢所有的回复 - 时间影响很有意思。
我做了一些测试,最终选择了一个可配置的解决方案,所以我可以选择带有参数的DSP,SRL或基于FF的延迟,因为(目前)我不受空间限制,但我真的不认为我' 我将把DSP用于其他任何事情......但我还不确定,因为我还有很多工作要做。 我必须说使用DSP进行延迟具有许多优点,通过一些巧妙地使用级联输入和输出(以及一些限制),您可以获得每个DSP在1个周期和4个周期之间的延迟的良好时序,以及 使用它们作为输入多路复用器,因此可以节省相当多的LUT。 例如,我使用2个DSP作为2路32位输入多路复用器,具有5个周期的延迟......无需其他逻辑。 李。 以上来自于谷歌翻译 以下为原文 Thanks for all the replies -- the timing implications are interesting. I've done some testing and have ultimately gone for a configurable solution so I can choose either DSP, SRL or FF based delays with a parameter, as (at the moment) I'm not space constrained, but I really don't think I'll be using the DSPs for anything else .. but I don't know for sure yet as I still have a lot to do. I must say that using the DSPs for delays has a number of advantages, with some clever use of the cascade inputs and outputs (and with some restrictions) you can get good timing on delays between 1 cycle and 4 cycles for each DSP, and also use them as an input mux so there can be quite a LUT saving. For example I'm using 2 DSPs as a 2-way 32 bit input multiplexer with a 5 cycle delay ... with no other logic needed. Lee. |
|
|
|
只有小组成员才能发言,加入小组>>
2371 浏览 7 评论
2790 浏览 4 评论
Spartan 3-AN时钟和VHDL让ISE合成时出现错误该怎么办?
2257 浏览 9 评论
3331 浏览 0 评论
如何在RTL或xilinx spartan fpga的约束文件中插入1.56ns延迟缓冲区?
2420 浏览 15 评论
有输入,但是LVDS_25的FPGA内部接收不到数据,为什么?
744浏览 1评论
请问vc707的电源线是如何连接的,我这边可能出现了缺失元件的情况导致无法供电
531浏览 1评论
求一块XILINX开发板KC705,VC707,KC105和KCU1500
349浏览 1评论
749浏览 0评论
1950浏览 0评论
小黑屋| 手机版| Archiver| 电子发烧友 ( 湘ICP备2023018690号 )
GMT+8, 2024-11-16 21:26 , Processed in 1.036042 second(s), Total 54, Slave 47 queries .
Powered by 电子发烧友网
© 2015 bbs.elecfans.com
关注我们的微信
下载发烧友APP
电子发烧友观察
版权所有 © 湖南华秋数字科技有限公司
电子发烧友 (电路图) 湘公网安备 43011202000918 号 电信与信息服务业务经营许可证:合字B2-20210191 工商网监 湘ICP备2023018690号