>>-----Original Message----- >>From: Chris Wilson [mailto:[email protected]] >>Sent: 2010年10月26日 16:38 >>To: Zou, Nanhai; Zou, Nanhai; [email protected] >>Subject: RE: [Intel-gfx] [PATCH 1/2] split render engine batch buffer and BLT >>engine >> >>On Tue, 26 Oct 2010 16:31:47 +0800, "Zou, Nanhai" <[email protected]> >>wrote: >>> Even if we can implement the batch buffer in a modal way. >>> I think it is not the best usage model. >>> Render engine and BLT engine are separate engines on gen6+. >>> For them to run one by one will not maximum the GPU usage. >> >>What usage pattern are you designing for? The most common pattern I've >>seen is for short intermixed operations on a connected set of buffers. So >>just what is the average length of the concurrent batch buffers (with >>correct inter-ring flushing) versus a much simpler modal approach? >> >>And do you have any x11perf figures yet? >>-Chris
I don't have x11perf result yet. We have also discussed about the cache coherence design on gen6+. There are 4 kinds of ring on gen6+ used by different engines, also now CPU and GPU share the last level cache. So we may need a better cache synchronize mechanism. Thanks Zou Nan hai >> >>-- >>Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
