On Mar 19 14:13, Mark Geisert wrote: > On 3/19/2025 2:59 AM, Corinna Vinschen wrote: > > On Mar 19 08:56, Lionel Cons wrote: > > > On Tue, 18 Mar 2025 at 20:51, Corinna Vinschen > > > <corinna-cygwin-rdbxbdvo6bxqt0dzr+a...@public.gmane.org> wrote: > > > > > > > > Fodder for future Cygwin development: > > > > > > What about switching Cygwin's dlmalloc to use USE_LOCKS for better > > > multi thread performance? > > > > USE_LOCKS won't help multi threading performance. It just adds a global > > lock we already have. We could probably get rid of the locking calls > > in mm/malloc_wrapper.cc. Nice, but no help performancewise. > > > > We could try to enable MSPACES (but not ONLY_MSPACES) and redirect > > all malloc calls from user-space to per-thread mspace_malloc calls. > > That may be worth a try. > > Maybe something like > https://cygwin.com/pipermail/cygwin-patches/2020q4/010905.html > (Note that that won't apply now due to code re-arrangements since.) > > I didn't have the bandwidth to follow up on why that wasn't accepted back at > the time.
I did an MSPACES test quite a while ago as well. I had a bit of code which redirected all malloc calls in malloc_wrapper.cc to mspace_malloc calls, if the caller thread was a pthread. One mspace per thread. However, any thread can call free on memory allocated by any other thread, so per the descriptions in dlmalloc I had MSPACES, FOOTERS, USE_LOCKS and USE_SPIN_LOCKS set to 1. "Seomthing" didn't work right, but I don't remember. Corinna
