Age | Commit message (Collapse) | Author |
|
|
|
|
|
|
|
|
|
|
|
|
|
also some minor cleanups
|
|
The caller can then free the token array immediately.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
It's quite a big struct & we examine it a lot (too much). Reduce
the impact of this by just looking at the active part where possible.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
We shouldn't be hashing with keys that have uninitialized memory.
|
|
|
|
Change SSE_SWIZZLES #define to 0 to disable it.
|
|
The assert was in fact over-sensitive, but emitting the extra flags
is pretty trivial & we may as well just do it whenever we know the
pipeline is running.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
When flushing/rendering, some stages (like AA line/point) need to set
pipe/driver state. Those driver functions often call draw_flush().
That leads to recursion.
Use new draw->suspend_flush flag to explicitly prevent that in the key places.
Remove the draw->vcache_flushing field.
Reuse draw->flushing as a debug/assertion var.
|
|
|
|
|
|
|
|
checks to code creating the cache.
|
|
|
|
|
|
|
|
It seems we get hash collisions fairly easily and the code as it was didn't
deal with that properly.
I think we need a simpler hashing interface...
|
|
|
|
|
|
Fixes broken polygon stipple, aaline, aapoint stages
|
|
|