2D fft in fpga

[jiangliang]

I have learned that we can performe two group of 1D FFT to archive 2D FFT:

1. do 1D FFT on each row (real to complex)
2. do 1D FFT on each column resulting from (1) (complex to complex)

Also, there are 1D FFT function in LV FPGA, my question is how can we do this in a better way, say I have to performe a 512*512 FFT, I'll have to put it in say 512 block of BRAM, and write them in row, and fetch them in column? It seems to me just kinda unreasonable to have so many BRAM and operate them in a kind messy way.

Anybody have better ideas? I considered DRAM as well, but it seems there would be large latch there, maybe the overall performance will be lower then using BRAM? I'm not sure, enlighten me, please