Allen Akin wrote: > > Just a side comment on Ian's excellent summary... > > On Tue, Feb 12, 2002 at 11:45:58AM -0800, Ian Romanick wrote: > | - If an allocation (via the memHeap_t) fails when texture space is allocated > | (radeonUploadTexImages in lib/GL/mesa/src/drv/radeon/radeon_texmem.c), > | blocks at the end of the texList queue are freed until the allocation can > | succeed. This may be an area where the algorithm could be improved. For > | example, it might be better to find the largest free block (in the > | memHeap_t) and release memory around that block in LRU or least-often-used > | fashion until the allocation can succeed. > > I spent some time last year looking into memory management issues for > OpenGL. One of the things I noticed was that most apps render the > primitives in a scene in roughly the same order from frame to frame. > Consequently textures are referenced in roughly the same order each time > a frame is drawn: T0, T1, T2, ...; T0, T1, T2, ... and so on. > > If texture memory is overcommitted, then an LRU strategy yields a > pattern of memory management operations something like this: > > Frame 0: > Load T0 > Use T0 > Load T1 > Use T1 > Unload T0 to generate free space for T2 > Load T2 > Use T2 > Frame 1: > Unload T1 to generate free space for T0 > Load T0 > Use T0 > Unload T2 to generate free space for T1 > Load T1 > Use T1 > Unload T0 to generate free space for T2 > Load T2 > Use T2 > > All subsequent frames follow the pattern set by Frame 1. You need to > unload one or more textures each time you want to use new ones. > > On the other hand, a LIFO strategy yields a pattern like this: > > Frame 0: > Load T0 > Use T0 > Load T1 > Use T1 > Unload T1 to generate free space for T2 > Load T2 > Use T2 > Frame 1: > (T0 is still loaded) > Use T0 > Unload T2 to generate free space for T1 > Load T1 > Use T1 > Unload T1 to generate free space for T2 > Load T2 > Use T2 > > Subsequent frames follow the pattern for Frame 1. > > In this case LIFO saves one unload and one load per frame. In practice > the actual savings depend on the relative sizes of textures, the total > memory load, and memory fragmentation, of course. But in general, if > textures are re-used sequentially on each frame, LIFO is a better > strategy than LRU.
The Utah drivers detected thrashing & switched from a LRU to MRU replacement strategy. The dri drivers could/should but don't do the same... Keith _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
