This is the mail archive of the 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]

[tree-ssa] patches summary

I am sending short summary of open patches and plans I am working on sorted in
priority order on how much of my other work is blocked by them.  Hope this will
help to get situation easier as I've sent quite many of patches recently.

  (I would like to make progress toward cleaning up expansion and this is very
  important for me to kill sibling call pass.  Also disabling Zdenek's
  tail recursion blocks some of his work)
     Re-enable tail call optimization
     (all unrelated problems has been solved and this patch has been re-tested)

Folding and type checking:
  (I would like to get full type checking for gimple, so types of operand match
  when they are supposed to.  This is not at all easy and fold currently poses
  fundamental problems on this thread.
  I would also like to get folding safe so fold_stmt can be called at
  each statement after gimplification and we can check that statements
  are kept in fully folded form)
     [tree-ssa,RFC/RFA?] fold-onst versus GIMPLE
     [tree-ssa] tree_ssa_useless_type_conversion_p tweek

Minor improvements:
    [tree-ssa] Avoid fold_stmt crash on return with no value
    (I would like to get fold_stmt to get called during CFG
     construction on each statement.  There are still some bugs on this
     path tought)
    Re: Should -fcross-jumping be part of -O1?

Not so minnor improvements:
    Re: Inner-loop optimization regression from 3.3 to 3.4
    (solves some of CCP pass testcases and improves SPECs)
    [tree-ssa] re-compute TREE_ADDRESSABLE during find_referenced_vars
    (improves effectivity of early dominator/dce passes at very little
    runtime cost, negative in practice)

Irreducible loops:
     [tree-ssa] Avoid jump threading from producing irreducible regions
     [tree-ssa] Irreducible loops self checking
     (contains syntactic analysis that can be used for pretty printining
     lowered gimple too)

(I am mostly done with the plans to get SSA ggc_collect-able, but little
polising is needed)
     Re: [tree-ssa]  Allow next field to point to the same field
     [tree-ssa] Put CFG into GGC memory
     [tree-ssa] Cleanup tree-optimize.c

SSA and STMT checkig and alias analsysis:
   Diego is currently working on this, thanks :)
   I do plan to add some extra checks - check that each stmt is folded
   and check of intra-procedural tree sharing problems.  More ideas are
   Once Diego is done, I will send remaining patches that are still


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