In stage2: ./xgcc -B./ -B/home/dave/opt/gnu/gcc/gcc-3.5.0/hppa-linux/bin/ -isystem /home/da ve/opt/gnu/gcc/gcc-3.5.0/hppa-linux/include -isystem /home/dave/opt/gnu/gcc/gcc- 3.5.0/hppa-linux/sys-include -L/home/dave/gnu/gcc-3.5/objdir/gcc/../ld -O2 -DIN_ GCC -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wold-s tyle-definition -isystem ./include -I. -I. -I../../gcc/gcc -I../../gcc/gcc/. - I../../gcc/gcc/../include -I../../gcc/gcc/../libcpp/include -I../../gcc/gcc/../l ibbanshee/libcompat -I../../gcc/gcc/../libbanshee -I../../gcc/gcc/../libbanshee/ points-to -g0 -finhibit-size-directive -fno-inline-functions -fno-exceptions - fno-zero-initialized-in-bss -fno-unit-at-a-time \ -c ../../gcc/gcc/crtstuff.c -DCRT_BEGIN \ -o crtbegin.o <built-in>:1: internal compiler error: in ten_to_ptwo, at real.c:2012
I wonder if the following patch caused it (it has a wrong date on it too): 2004-05-25 Paolo Bonzini <bonzini@gnu.org> * combine.c (gen_binary): Remove. (known_cond, simplify_shift_const find_split_point, combine_simplify_rtx, simplify_if_then_else, simplify_set, simplify_logical, expand_field_assignment, extract_left_shift, force_to_mode, if_then_else_cond, apply_distributive_law, simplify_and_const_int, simplify_shift_const, gen_lowpart_for_combine, simplify_comparison, reversed_comparison): Replace with simplify_gen_binary, simplify_gen_relational or distribute_and_simplify_rtx. (distribute_and_simplify_rtx): New function. * simplify-rtx.c (simplify_binary_operation): Use nonzero_bits to simplify ANDs where we are turning off bits already known to be off in OP0.
Subject: Re: [3.5 Regression] <built-in>:1: internal com > I wonder if the following patch caused it (it has a wrong date on it too): > 2004-05-25 Paolo Bonzini <bonzini@gnu.org> > > * combine.c (gen_binary): Remove. Possibly. My last success full build was LAST_UPDATED: Wed May 26 21:40:20 UTC 2004. This patch was applied on 27 May 2004 08:29:19 -0000. Dave
Subject: Re: [3.5 Regression] <built-in>:1: internal compiler error: in ten_to_ptwo, at real.c:2012 Please attach a preprocessed testcase. In the meanwhile, I reverted the patch. Paolo
Paolo, the preprocessed source will not help you as this is being compiled with stage2.
Subject: Re: [3.5 Regression] <built-in>:1: internal compiler error: in ten_to_ptwo, at real.c:2012 > Paolo, the preprocessed source will not help you as this is being compiled with stage2. Oh, right, so it's a miscompilation and not an ice-on-valid. Anyway, I've reverted the patch. I hope that Eric Botcazou can provide me with a testcase because the cleanup in that patch is quite desirable. I suspect the bug is in the only non-trivial change in the patch, the factoring of the new function distribute_and_simplify_rtx. In particular, it may not have shown on i686-pc-linux-gnu because it has no AND-with-complement instruction (and on PPC because of some combiner bogosity?!?). Paolo
Subject: Re: [3.5 Regression] <built-in>:1: internal com > Subject: Re: [3.5 Regression] <built-in>:1: internal > compiler error: in ten_to_ptwo, at real.c:2012 > > Please attach a preprocessed testcase. In the meanwhile, I reverted the > patch. I did confirm last night that that the suspected patch introduced the build ICE. Dave
Fixed by patch reversion.