Forum: Programming
2020-07-15, 20:14
|
Replies: 43
Views: 25,120
No division is done. An example for the case...
No division is done. An example for the case where n (the FFT length) is known at compile-time to be 5*2^some_power.
double ksinpi(double k, const double n) {
const double multiplier = 935.0,...
|
Forum: Programming
2020-07-15, 15:00
|
Replies: 43
Views: 25,120
.
After lots of study, my final solution...
.
After lots of study, my final solution was based on your post. It differs in that I need sincos (2*pi*k/n) where n is known in advance and is highly composite. The advance knowledge of n lets...
|
Forum: Programming
2020-01-17, 02:06
|
Replies: 43
Views: 25,120
|
Forum: Programming
2020-01-12, 20:50
|
Replies: 43
Views: 25,120
|
Forum: Programming
2020-01-12, 05:47
|
Replies: 43
Views: 25,120
|
Forum: Programming
2020-01-11, 20:20
|
Replies: 43
Views: 25,120
|
Forum: Programming
2020-01-10, 23:25
|
Replies: 43
Views: 25,120
Yes, it is for FFT twiddles on a GPU.
Take...
Yes, it is for FFT twiddles on a GPU.
Take the awesome Radeon VII. Using the memory bandwidth from the spec sheets, reading a sin/cos twiddle from memory is 53 clocks latency if memory is...
|
Forum: Programming
2020-01-10, 04:15
|
Replies: 43
Views: 25,120
|
Forum: Programming
2020-01-09, 22:10
|
Replies: 43
Views: 25,120
|
Forum: Programming
2020-01-09, 05:28
|
Replies: 43
Views: 25,120
|
Forum: Programming
2020-01-09, 04:58
|
Replies: 43
Views: 25,120
Algorithm wanted - sine and/or cosine
I cannot locate my Knuth book. I need the fastest possible algorithm to compute sine(pi * a/b) and/or cosine(pi * a/b) on a GPU where a and b are positive 32-bit integers and a is less than half of...
|