Please sign in to comment.
Work around an IORef contention problem that slows thread-delay.
This fixes a huge performance regression that I introduced in 82af25a that affected the following command line: benchmarks/thread-delay -n 10000 The regression increased the run time of that program from about 0.2 seconds to a consistent 20 seconds. The good run time was erratic, sometimes spiking up to 16 seconds or so, but the bad run time was very consistent. I believe that the underlying cause of the problem is http://hackage.haskell.org/trac/ghc/ticket/3838
- Loading branch information...