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: [tree-ssa] DCE compile time repair.


On Fri, 2003-06-13 at 14:29, Andrew MacLeod wrote:

> Well I figure its going to be reasonably dense.. We're checking the bit
> on every basic block which has a PHI node in it...   Whats the density
> of that? And we'll be checking those block frequently... I see C++ code
> sometimes has a lot of PHI nodes :-) So quick acess is good.
> 
Hmm, could be.  I guess we'd need to measure that.  Lots of PHI nodes
doesn't necessarily mean that they're going to be lots of blocks with
PHI nodes.  It may be that we have a handful of blocks with lots of PHI
nodes in them.

> I dont think we'd notice performance either way, does a reasonably dense
> sparse bitmap take less memory than an sbitmap?
> 
There's a break even point, for sure.  I don't know what that is,
though.  We're a bit memory-hungry and it doesn't appear to me that
these bitmaps are going to be dense, but I don't have hard numbers to
back it up.  I guess we'll have to find out.


Diego.


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