cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
danb2xilinx
Observer
Observer
459 Views
Registered: ‎04-02-2019

FFT configuration not correct

I'm trying to configure a 4 channel, radix 2 FFT for selectable length(16K max), scaled and direction. According to the documentation, the bit field should be 124 bits wide- 5+3(pad) bits for size, 4 bits for direction and 4 groups of 28 bits for scaling for a total of 124 bits. This is what is shown in the IP-> implementation tab. However, the bit field width is 136 bits on the symbol tab!

If I write to the 124 bit field width as described, it doesn't work. I took a guess and padded the 4 bit direction by 12 bits to start the scale values after 12 padded bits. It appears to work. What should this be and why is the documentation wrong?

I can't look at the vhdl for the FFT as it is encrypted. The only thing that I can see is that it takes 136 bits down to 121 bits, which is the correct number of active bits with no padding. Did I win the lottery and guess where the missing bits should go? Why doesn't the IP core, documentation and implementation agree??

BTW, this is 2019.2 and FFT core 9.1

Thanks,

Dan

Tags (1)
fft2.jpg
fft1.jpg
0 Kudos
3 Replies
danb2xilinx
Observer
Observer
380 Views
Registered: ‎04-02-2019

Well? Any response from Xilinx as to this apparent bug?

0 Kudos
nathanx
Moderator
Moderator
192 Views
Registered: ‎08-01-2007

@danb2xilinx , this is a known issue, we will fix it in the next IP core release, as this is just a wiring issue, which does not impact FFT IP functionality or QOR, so, you can just ignore extra bits in the s_axis_config as a temporary work around. Hope it helps.

0 Kudos
nathanx
Moderator
Moderator
162 Views
Registered: ‎08-01-2007

@danb2xilinx ,  I will also add that the additional bits will not lead to additional hardware because they will be removed during optimization in the implementation tools. We will publish AR# 76226 to Xilinx.com for this known issue. 

0 Kudos