On Tue, May 25, 2010 at 01:06:50PM +0800, Xiang, Haihao wrote: > This interface is the same as drm_intel_bo_alloc except the allocated > size isn't rounded up, so it bypasses the cache bucket. > > The size of the BO created by drm_intel_bo_alloc for a 1920x800,4:2:0 YUV > planar surface is 4M, it is about 2.2M if using drm_intel_bo_alloc_direct.
Have you tried either Chris Wilson's or my fair-lru eviction patches? They should help quite a lot, especially for many big objects of exactly the same size like libva uses. Cause even 2*23*4M = 184 should decently fit into a 256M aperture - which is what I have here for my ilk machine. Dunno whether that 23 for the number of reference framse is correct, but it's certainly an upper bound. Yours, Daniel -- Daniel Vetter Mail: dan...@ffwll.ch M bile: +41 (0)79 365 57 48 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx
