在FPGA中找到大矩阵的行列式是否可行？

嗨，
该应用程序的问题是需要存储和随机访问大量数据。
因此，使用CPU和一些RAM的顺序算法的使用已经非常有效和简单。
这可以在效率上打败，FPGA在很大程度上取决于应该实现的算法。
使用特殊缓冲区和智能内存控制机制，应该可以创建一些小的处理元素，这些元素由少量数据组成，然后可以同时处理。
在您的情况下，我认为大矩阵的行列式可以通过组合大矩阵（2x2或3x3）的小部分的下限来计算。
对于这些，可以在FPGA中实现许多处理元件。
但是数据流的控制机制将变得非常复杂。
这不是一项简单的任务，但是如果有一些时间来制定一个使用并行处理的有用算法，可能会产生一个高性能的解决方案。
有一个很好的综合 
Eilert



以下为原文

Hi,
the problem with this application is that a big number of data needs to be stored and randomly accessed.
So the usage of a sequential algorithm using a CPU and some RAM is already quite efficient and simple.
 
Wether this can be beaten in efficiency with an FPGA depends heavily on the algorithm that shall be implemented.
Using special buffers and an intelligent memory control mechanism it should be possible to create a number of small processing elements that are fed with small chunks of data which then can be processed concurrently.
 
In your case, I think the determinant of a large matrix can be calculated by combining the dederminants of small sections of the large matrix (2x2 or 3x3). For these a number of processing elements can be implemented in the FPGA. But the controll mechanism for the data flow will become quite complex.
 
It won't be a simple task, but given some time to work out a useful algorithm that uses parallel processing might lead to a performant solution.
 
Have a nice synthesis
  Eilert

好的，谢谢你的信息。
问候，
Koyel



以下为原文

okay thanks for the information.
 
Regards,
Koyel

同意。
使用小数组可以更容易。
为此，请查找oartitioned矩阵（或块矩阵）。
没有必要重新发明轮子;
几十年来，线性代数一直在为这类事做出很好的算法。
如果您的矩阵具有任何类型的对称性，重复值或者有点对角线，那么这种方法可以很好地工作。



以下为原文

Agreed. It would be much easier to work with small arrays of numbers. To do this look up oartitioned matrices (or block matrices). There's no need to reinvent the wheel; linear algebra has been making nice algorithms for this sort of thing for decades. This works paricuarly well if your matrix has any sort of symetry, repetitive values, or is somewhat diagonal.

我遇到了这个问题，我需要用7个512x512矩阵进行操作。
但是我找不到任何处理FPGA上的大矩阵的例子，这对我来说似乎很奇怪，因为我确信我不是第一个尝试这样做的人。
没有必要重新发明轮子;
几十年来，线性代数一直在为这类事做出很好的算法。
@ geco.kevin你可以发布一些链接吗？
因为正如我所说，我在FPGA中找不到任何大型阵列/矩阵运算的例子。
还有一个问题。
您是否认为我可以使用此算法作为每个块矩阵的基础，我仍然可以减少与软件相比的运行时间？



以下为原文

I'm having this problem as well as I need to do operations with seven 512x512 matrices.
 
However I can't find any examples that deal with big matrices on an FPGA and that seems strange to me as I'm sure I'm not the first one to try to do this.
 
There's no need to reinvent the wheel; linear algebra has been making nice algorithms for this sort of thing for decades. 
@geco.kevin can you post some links? Because as I said, I can't find any example of big arrays/matrices operation in FPGAs.
 
One more question. Do you think that I can use this algorithm as my basis for each block matrix and I will still be able to reduce my run-time comparing to software?