On 25.06.2009 10:32, Stephane Marchesin wrote: > On Thu, Jun 25, 2009 at 09:46, Jerome Glisse<gli...@freedesktop.org> wrote: >> On Wed, 2009-06-24 at 22:32 +0200, Roland Scheidegger wrote: >>> On 24.06.2009 20:17, Jerome Glisse wrote: >>>> I think we should let user ask at gem map ioctl time if userspace wants >>>> an surface backed mapping or not, and gem map will reply with a success >>>> or failure. So if object is in vram and there is a surface reg available >>>> it will succeed, if object is in system ram it will report to userspace >>>> that there is not automatic untiling and that userspace is on its own >>>> to untile the buffer. >>>> >>>> For the X server that the front buffer is mapped first and never >>>> unmapped, it should get a surface (assuming no other process already >>>> stole all the surface). For pixmap i think be better of not using >>>> tiling for time being (or macro tiling only benchmark below). >>>> >>>> Mesa, map/unmap things and should be able to untile on its own for >>>> front/zbuffer (we need to add texture but i am not sure it's worth >>>> it, see benchmark below). >>> I don't see benchmark with texture tiling below... >>> It definitely made some difference though when I implemented (and >>> measured...) this, though I never really worried that much about tiled >>> compressed textures, not sure micro tiled is even possible (and would >>> make sense) but macro tiled certainly should be (but IIRC I tried to >>> measure it and it didn't make much of a difference on r200 but it could >>> have changed with newer chips). >>> That said, don't forget that the performance improvement this gives is >>> chip specific, generally giving more improvement with newer chips. IIRC >>> you definitely don't want to micro tile the front buffer pre-r300. >>> >>> Roland >> Yeah i loose texture benchmark but it was very small 1-2% on quake3 >> but maybe quake3 isn't asking for much texture filtering, assuming >> filtering is the process which benefit from tiled texture. >> > > IIRC the microtiling mode will only benefit the exotic filtering modes > (anisotropic for example). Did you try this?
I am pretty sure it made quite a difference with "normal" trilinear filtering (otherwise I never would have bothered implementing it in the first place). Can't remember exactly but probably around 10% or so. Not sure if macro or micro tiling helped more, but both together were definitely accounting for more than 2% (unless using compressed textures). You are right though I think with bilinear (which q3 uses as default) there was less difference. That was on rv250 back then, and it could be different on newer chips (could depend quite a bit on if it's a chip with a lot of memory bandwidth or not too). Roland ------------------------------------------------------------------------------ -- _______________________________________________ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
