------- Comment #21 from jakub at gcc dot gnu dot org 2010-09-02 16:16 -------
Created an attachment (id=21670)
The last valgrind error is my fault, we don't need a BARRIER there, any
instruction that flushes the queued ones will do.
So, either we do something like in this patch, or dwarf2out.c would need to
record the size of barrier_args_size and for BARRIERs with higher INSN_UID just
use offset 0 (i.e. args_size before offset -= args_size;). But the latter
would be more code and uglier.