Yeah, that looks like an typo, weird how it never occurred on CI.

Codecov Report

Merging #1204 (ccb8511) into master (27c87a6) will not change coverage.
The diff coverage is 0.00%.

@@           Coverage Diff           @@
##           master    #1204   +/-   ##
=======================================
  Coverage   80.29%   80.29%           
=======================================
  Files         119      119           
  Lines        8306     8306           
=======================================
  Hits         6669     6669           
  Misses       1637     1637           

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 27c87a6...ccb8511. Read the comment docs.

Not familiar with the test suite here but if its helpful the MWE I developed to track this down looked something like:

x = cu(rand(128,128,2))
plan = plan_rfft(x, (1,2))
n = norm(plan * x)
@sync @async plan * x # call plan from other task hence screwing up workarea
for i=1:10000
    @test norm(plan * x) ≈ n
end

and it did take alot of iterations for the memory corruption to reliably screw up the result, and even sometimes for some reason I didn't understand if wasn't triggered, so it was fairly "silent". Anyway, might be overkill for the tests here but that's what I did fwiw.

Yeah, the test suite takes too long already...