- Emailed a friend.
- Went through public records to get the story on a house. (The owner from 1958 died two years ago after taking out a reverse mortgage, then her daughter got it and is selling it, possibly trying to cover what's left on the reverse mortgage.)
- Client B work.
- Condo communication, snow removal guy communication. (No way I'm shoveling 8-22" of snow on Saturday.)
- Verified that rsqrt is not the problem. The reason I was getting such different results between the two versions yesterday is that my test was flawed. My pd extension for logging only logged to something like seven significant digits. When I fed that into the C++ test harness for rsqrt, read the data from that log and so didn't get what was actually going into the pd rsqrt. After I fixed the pd block logging extension, the C++ rsqrt came up with almost exactly the same thing as the pd version.
- So now, the next problem is: what is the (significant) difference between what's coming into the rqsrt (added carrier fft results)? After a lot of shuffling things around in Google Sheets (still can't get MIN(FILTER(…)) working), I found out that the smallest value going into rqsrt on the pd is ~5.5e-9, which yields the largest result, 13229.70703. On the C++ side, the smallest input is ~3.9e-11, yielding 160226.875. Why do things get so much smaller on that side? Yet again, I do not know.