Dynamically resize the vertex buffer #13
Conversation
src/gc_gl.c
Outdated
| return; | ||
| if (glparamstate.imm_mode.current_numverts >= glparamstate.imm_mode.current_vertices_size) { | ||
| if (!glparamstate.imm_mode.current_vertices) return; | ||
| int new_size = glparamstate.imm_mode.current_vertices_size * 2; |
There was a problem hiding this comment.
Not sure how keen I am on doubling every time. Have you done any profiling of how many vertices you end up with at peak usage? I can see doubling when current_vertices_size is small but maybe cap the increment somewhere? Is there a benefit to starting small?
There was a problem hiding this comment.
I'm afraid it really depends on the application. I noticed the problem by running this code, and the maximum there is 162. Starting small makes sense, because we don't know how the API will be used and what is most important for the client: maybe it has objcts with a low number of vertices but high-definition textures, so it would rather save memory for that. Note that each vertex takes up quite a bit of memory (it's 12 floats, so 48 bytes).
I do agree that capping the increment makes sense. Maybe you double until we get to 256, after which we just add 256 every time. I'll do like this.
Avoid reserving memory if glBegin() is never called, and dynamically increase its size as needed. We never free the buffer memory, since it's likely that the same amount of memory will again be needed during the next frame.
mardy
force-pushed
the
dynamic-glbegin-buffer
branch
from
aa07e7d to
3b359e1
Compare
Avoid reserving memory if glBegin() is never called, and dynamically increase its size as needed. We never free the buffer memory, since it's likely that the same amount of memory will again be needed during the next frame.