On Tue, 2006-02-14 at 17:42 +0100, Marcin Dalecki wrote: > On 2006-02-14, at 11:21, Rainer Emrich wrote: > > > ICE in stage2: > > > > 'tree_duplicate_sese_region': > > /SCRATCH/gcc-build/HP-UX/hppa2.0w-hp-hpux11.00/src/gcc/tree-cfg.c: > > 4430: internal > > compiler error: in do_compare_rtx_and_jump, at dojump.c:945 > > > Precisely the same problem can be observed on ppc as well. > Is it maybe related to the use of the NE as a tree_code in the > following (line 597): > > tree dummy_cond = build2 (NE, boolean_type_node, > integer_zero_node, integer_zero_node); > > inside the dom_trhead_acroess_edge() function from tree-ssa-dom.c? Highly unlikely since the dummy condition is never inserted into the IL. It's just used so that we can query the hash tables without creating lots of junk trees. > > I think is should read as NE_EXPR insead... Yes. Though in practice it doesn't matter as the tree code as well as the operands are rebuilt every time we use the dummy condition.
I'll happy pre-approve a patch to fix this little buglet. Jeff
