Announcement

Collapse
No announcement yet.

Locking and unlocking buffers not as fast as I'd expect.

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

    Locking and unlocking buffers not as fast as I'd expect.

    Anyone know why rhilock/unlockstructuredbuffer is so slow, even in release build? It takes around 1ms to lock two buffers of about 50mb, slower than the actual memcpy. In debug it's much slower.

    I'm able to retrieve data much faster directly with Opengl, I'd like to figure out why this is slow and how to improve it.

    this is an array of FVectors.

    It's really the only hiccup I'm having right now with my GPU version of my version of naive surface nets, the 2 to 3 Ms for processing a 128x128x128 volume is almost all in the lock/unlock and memcpy.

    I could leave the vertices on the gpu if I didn't need the collision data... Hm now that I think about it, I could just update the collision once a frame or every other frame, that should make this performance cost not really noticeable.

    Any ideas are welcome, thanks for the help!
Working...
X