This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [dataflow] PATCH removal of more of flow.c.


Joern Rennecke wrote:

I suppose one way to reduce the cost of RU generally would be to find first scan
the insns backwards to quickly get rid of all the uses that have a dominating
definition in the same extended basic block.

P.S.: I see now that you can build use-def and def-use chains by solving RD; that should be perfectly adequate for SH -mrelax.

I'm a bit puzzled now why you need to do RU computation in the first place; I see that
RU only considers definitions as separating, while RD considers any kill of a register,
but that is not really intrisically tied to the direction in the control flow you follow;
you could do the same bitset computations that RD does, but only consider redefinitions
as impeding propagation of the original definition, in order to build the same chains that
can build with RU.



Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]