Skip to content

Latest commit

 

History

History
24 lines (20 loc) · 1.43 KB

FFT.md

File metadata and controls

24 lines (20 loc) · 1.43 KB

Table of small FFTs

The "cost" is estimated as 2xFMAs + 1xADDs. Excess vGPRs is #vGPRs - 4*N.

N FMAs ADDs Excess vGPRs Cost Cost/N^2 Source
2 0 4 4 1
3 6 6 1 18 2 FFT-3
4 0 16 5 16 1 FFT-4
5 12 24 4 28 1.1 FFT-5
6 12 24 11 28 0.8 FFT-6
7 10 62 13 82 1.7 FFT-7
8 4 52 9 60 0.9 FFT-8
9 20 72 17 112 1.4 FFT-9
10 24 68 13 116 1.2 FFT-10
11 110 30 33 250 2.1 FFT-11
12 24 72 13 120 0.8 FFT-12
13 156 36 25 348 2.1 FFT-13
14 20 152 13 192 1 FFT-14
15 66 102 15 234 1 FFT-15
16 28 124 13 180 0.7 FFT-16