- VertexBuffer holds permanent copy of the vertex data and is only read from
GPU memory was written outside a sb:bind()/sb:unbind() pair.
- "Orphan" the current GPU buffer on unmap() before uploading the changed data
to the GPU to avoid implicit synchronisation lags.
Instead of letting OpenGL manage the memory mapping, allocate a temporary
block of memory upon VBO:map(). Upon VBO::unmap() the block is copied to
graphics memory using glBufferSubDataARB() and then released.
Hopefully this will fix some weird error with Windows XP + old drivers, where
VBOs can be created, but the mapped memory cannot be allocated resulting in
the sprite batch drawing nothing at all.
According to the interwebs this could actually faster than using OpenGL mapped
memory, because there is no unnecessary syncing involved.
Also improved memory usage in SpriteBatch. (No need to retain full copy of
sprites in memory. Can read that data back later to save data from OpenGL
context destruction).
vrld
Bart van Strien
rude
Bill Meltsner