On Mon, Apr 27, 2015 at 3:19 AM, Deepthi Nandakumar <deep...@multicorewareinc.com> wrote: > Thanks, I like initAqQPs since it gets rid of the messy section, but I was > also hoping we could avoid passing qp around for the compress* functions. > > Since cuGeom now has cuIdx - the analysis functions can just look up the > appropriate QP in m_aqQp?
only if we have initAqQPs() fill in QPs for the CUs below the DQP depth, otherwise they wouldn't know what QP they are "inheriting" from the larger block. > In fact, with some interesting calculations on > cuGeom->absPartIdx and depth, we dont even need cuIdx? it would be tricky, but I believe possible. I'm not sure it would be more efficient, though. There seem to be two possible end-goals: 1 - remove the m_aqQP[] array and calculate CU QP on demand in the current locations where it is changing QP. This requires keeping the "int qp" argument to the compress functions. 2 - keep m_aqQP[] and fill in QPs for all CUs, not just those above the delta QP depth limit, then set QP and lambda at the top of each compress function. The "int qp" arguments go away all other things being equal, I prefer #1 for its simplicity -- Steve Borho _______________________________________________ x265-devel mailing list x265-devel@videolan.org https://mailman.videolan.org/listinfo/x265-devel
