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: [trans-mem] optimize read for write


> This is wrong.
> It will produce a correct but non-maximal answer.
> ANTIC is a backwards intersection problem.
> As such the initialization must be all 1 bits in order to get a
> maximal correct answer.

Are the AVAIL sets (in trans-mem.c) also wrong?  I noticed that the lcm
code with sbitmaps seeded the bits to 1's as well.

> If you are using bitmaps, your options are basically to either set
> all-1's (and ruin the sparseness), or do what we do now in tree level

Richard mentioned something about the worklist for C++ cases not being
bounded by the total number of basic blocks.  If that's the case, then
setting all-1's is out of the question.

Richard, any comments?


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