PS: I should have mentioned that in my test case speedup of the filter
loop was a factor of 37.5.
The speedup number I quoted, x 6, refers to execution of the whole
program. Execution time is once again dominated (about 75%) by time
spent in sfrsd, as desired.
-- Joe
On 10/27/2015 12:00 PM, Joe Taylor wrote:
> Hi Steve,
>
> Just to let you know:
>
> In r6011 I have changed the filter loop in subtract65.f90 from a
> time-domain convolution to a frequency-domain multiplication. In my
> test case the speedup was a factor greater than 6. Two-pass decoding of
> JT65 is now acceptably fast. I think we have a definite winner!
>
> -- Joe, K1JT
------------------------------------------------------------------------------
_______________________________________________
wsjt-devel mailing list
wsjt-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/wsjt-devel
