Edward Lam

> OIIO has the ability to optionally use TBB, but its support is not quite 100% yet -- when enabled, all the parallel_for family of functions will use TBB, but...

Hi folks, unfortunately, I've mostly run out of time to work on this for the moment. This is what I had so far: ``` class TBBThreadProvider : public IlmThread::ThreadPoolProvider {...

~Have we tested using this on Windows? I'm missing how an OpenVDB DLL user can link against the instantiations if they're no longer declared as dllimport.~ EDIT: Ah I missed...

I do. For debug builds, I only _sometimes_ want malloc replacement. Most of the time I link without tbbmalloc_proxy in order to take advantage of the extra runtime checks provided...

Yes, see my last comment above.

Yes, this is still relevant because SRW locks are best for performance on Windows. In the CRT, std::mutex itself uses SRW locks underneath.

@Carandiru0 Not following. The gist of this issue is that we're leaving performance on the floor that's easy to have.

Thank you for the thorough answer! For completeness, I should mention this came about as I was trying to decipher what/why/how OpenUSD was hanging for me here: https://github.com/PixarAnimationStudios/OpenUSD/blob/ceab7e6541c02c9376a03de84bb1174e95e7264e/pxr/base/work/dispatcher.cpp#L55 I haven't...

I note that PR #1780 merely says "more" thread-safe. But it doesn't mean that it's now actually thread-safe? If that is the case, is more work planned here?