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: [lto] Stop using TREE_LIST in functions dealing with PENDING_STMT.

Kazu Hirata wrote on 06/02/06 06:13:

> The only caveat is that we have to count the number of PHI nodes first
> to figure out the size of TREE_VEC, but the cost of counting the PHI
> is probably still cheaper than the overhead associated with allocating
> multiple chunks of memory.
Probably?  Numbers please.

> Tested on x86_64-pc-linux-gnu.  OK to apply?
> Kazu Hirata
> 2006-06-02  Kazu Hirata  <>
> 	* tree-ssa.c (ssa_redirect_edge): Store TREE_VEC in
> 	(reinstall_phi_args): Blah.
> 	* tree-cfgcleanup.c (remove_forwarder_block_with_phi): Blah.

> --- 48,91 ----
>   /* Remove the corresponding arguments from the PHI nodes in E's
>      destination block and redirect it to DEST.  Return redirected edge.
> !    The list of removed arguments is stored in PENDING_STMT (e) as a
> !    TREE_VEC with PHI_RESULT in the even-th elements and PHI_ARG_DEF in
> !    the odd-th elements.  */

The rest is fine.  Could you test for performance changes?

PS: Could you send unified diffs?  I know it's not a requirement, but I find
    them much easier to read.  Thanks.

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