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]

Re: [patch] move phiopt, ssa-dce and ssa-dom prototypes.

Andrew MacLeod <> wrote:
>Handle a few more prototypes in tree-flow.h
>There were only 2 routines exported from tree-ssa-phiopts, and neither 
>really belonged there.
>* I moved nonfreeing_call_p() to gimple.c since it is gimple dependent.
>blocks_in_phiopt_order() returns basic blocks in an order that 
>guarantees any single predecessor is visited before its successor.
>looking around, i think it really belongs in cfganal.c, so I moved it 
>there... unless you can think of somewhere better.  I also renamed it
>be more generally descriptive: single_pred_before_succ_order...  for 
>lack of anything better.
>tree-ssa-dom.h needed to be created for a few prototypes.
>The only 2 exports from tree-ssa-dce.c were 
>mark_virtual_operand_for_renaming and 
>mark_virtual_phi_result_for_renaming. I moved them to tree-into-ssa.c 
>which already has mark_virtual_operands_for_renaming ().   Names are 
>confusing a bit I find... and do we even need these routines? Aren't
>virtual uses based off the one root variable? so isn't all that walking
>and SET_USE-ing a waste of time?  Or is that still needed
>by the incremental bits? I never really paid attention to how that 
>worked :-)

It can be a bit tricky sometimes, but yes, another cleanup is on my long todo list.

>Bootstrapped on x86_64-unknown-linux-gnu, and currently running 
>regressions.  Assuming it passes, OK?


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