Hi Boris,
The cfg is not updated during the dbr pass (you might have edges to
non-existing blocks etc.). Furthermore, the BB rules do not apply after the dbr
pass (you might have labels inside BBs and not only before the bb_note insn or
jump-insns in the middle of the BB and not at BB_END).
What you might want to do is reconstruct the cfg after the dbr pass:
1. Clear existing edges using clear_edges ()
2. Split the BBs in case there's a label/jump insns in the middle of it (using
split_block ())
3. Reconstruct edges between bbs using make_edge ()
This would create a new cfg and bb structure.
Hope this helps,
Tomer
-Original Message-
From: Richard Guenther [mailto:[EMAIL PROTECTED]
Sent: Tuesday, March 18, 2008 19:47
To: Boris Boesler
Cc: Jim Wilson; GCC
Subject: Re: Basic block infrastructure after dbr pass
On Tue, Mar 18, 2008 at 6:40 PM, Boris Boesler <[EMAIL PROTECTED]> wrote:
> Am 18.03.2008 um 16:21 schrieb Jim Wilson:
>
> > Boris Boesler wrote:
> >> The following code generators use FOR_EACH_BB[_REVERSE] in the
> >> target machine dependent reorg pass:
> >> - bfin
> >> - frv
> >> - ia64
> >> - mt
> >> - s390
> >
> > The very first thing that ia64_reorg does is
> > compute_bb_for_insn ();
>
> For a few seconds I thought you saved my day.
>
> I'm not talking about BLOCK_FOR_INSN (insn)
>
> I haven't specified my problem properly? If I traverse basic blocks
> via FOR_EACH_BB (used in compute_bb_for_insn, too) I get insns which
> are not in the insn-stream for(insn = get_insns(), insn; insn =
> NEXT_INSN(insn)) ..
>
> B1 ---> B2 --> B3
> /\ /\ /\
> / \/ \ / \
>vv vv vv
> i1 -> i2i3 -> i4 i5 -> i6
> | ^
> +-+
>
> How can I say this? In a first pass I dump each insn as a node in a
> VCG-file, each node is identified by its memory address (I used the
> UID, but insns appear multiple times if they were copied, but UID or
> address makes no difference). In a second pass I dump each basic
> block as a node with edges to the head insn and to the end insn of
> the basic block, which should be emitted in the first pass. But
> sometimes some of the basic blocks have edges to insns which do not
> exits.
>
> Possible solutions (from OK to horrific):
> 1) These insns are dead code.
> 2) Bug in my output (but I checked it quite often)
> 3) Bug in GCC
You are probably mix-matching functions for use in cfg_layout mode vs.
non-cfg_layout mode.
Richard.
