diff options
| author | Luc Van Oostenryck <luc.vanoostenryck@gmail.com> | 2017-05-13 21:53:43 +0200 |
|---|---|---|
| committer | Luc Van Oostenryck <luc.vanoostenryck@gmail.com> | 2017-05-19 05:27:51 +0200 |
| commit | 8c24aa51b70cf1fbbb5ff8fbd90f6bdc482d7e2f (patch) | |
| tree | 9aaa19e9c91e3a3f54c5ac62847003726d61fa3d /flow.c | |
| parent | e35efe330c6ae7d154197c29b127560d569016d0 (diff) | |
| download | sparse-dev-8c24aa51b70cf1fbbb5ff8fbd90f6bdc482d7e2f.tar.gz | |
fix: kill old branch in insert_branch()
insert_branch() is called after an optimization has detected
an opportunity to convert a switch or a conditional branch
into an unconditional branch (for example because the condition
is a constant). It does this by removing the BB's last instruction
and then allocate a new unconditional branch which is added at
the end of the BB. The old instruction is simply discarded.
Since the discarded instruction is one with a condition we must
insure that the associated usage is also removed (for example,
by calling kill_instruction()).
But currently kill_instruction() is called, just after the call
to insert_branch(), only at a single place. The 4 other places where
insert_branch() is called do nothing with the removed instruction
and it's condition's usage. As consequence, instructions that are
dead are not removed since it still wrongly has an user.
Fix this by adding a call to kill_instruction() in insert_branch()
itself (and make the function description more exact).
Signed-off-by: Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
Diffstat (limited to 'flow.c')
| -rw-r--r-- | flow.c | 1 |
1 files changed, 0 insertions, 1 deletions
@@ -226,7 +226,6 @@ try_to_rewrite_target: if (bb_list_size(target->parents) != 1) return retval; insert_branch(target, insn, final); - kill_instruction(insn); return 1; } |