diff options
author | Segher Boessenkool <segher@kernel.crashing.org> | 2016-01-08 04:58:40 +0100 |
---|---|---|
committer | Segher Boessenkool <segher@gcc.gnu.org> | 2016-01-08 04:58:40 +0100 |
commit | 52ad5601bf56aba4d5da585c6a212ce3f6509af3 (patch) | |
tree | 86689541fc8ed24ff266c281f8acac6768052b12 /gcc | |
parent | 7a127fa7c463b6b7c0285b32bac52bbdfc9a2117 (diff) | |
download | gcc-52ad5601bf56aba4d5da585c6a212ce3f6509af3.zip gcc-52ad5601bf56aba4d5da585c6a212ce3f6509af3.tar.gz gcc-52ad5601bf56aba4d5da585c6a212ce3f6509af3.tar.bz2 |
shrink-wrap: Once more PRs 67778, 68634, and now 68909
If a candidate PRE cannot get the prologue because a block BB is
reachable from it, but PRE does not dominate BB, we try again with the
dominators of PRE. That "try again" needs to again consider BB though,
we aren't done with it.
PR rtl-optimization/67778
PR rtl-optimization/68634
PR rtl-optimization/68909
* shrink-wrap.c (try_shrink_wrapping): Add comment. Don't pop
block from the stack until done with it. Remove a superfluous
bitmap set. Remove a superfluous bitmap test.
From-SVN: r232148
Diffstat (limited to 'gcc')
-rw-r--r-- | gcc/ChangeLog | 9 | ||||
-rw-r--r-- | gcc/shrink-wrap.c | 23 |
2 files changed, 25 insertions, 7 deletions
diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 7331599..44a2e01 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,12 @@ +2016-01-08 Segher Boessenkool <segher@kernel.crashing.org> + + PR rtl-optimization/67778 + PR rtl-optimization/68634 + PR rtl-optimization/68909 + * shrink-wrap.c (try_shrink_wrapping): Add comment. Don't pop + block from the stack until done with it. Remove a superfluous + bitmap set. Remove a superfluous bitmap test. + 2016-01-07 Martin Sebor <msebor@redhat.com> PR c/68966 diff --git a/gcc/shrink-wrap.c b/gcc/shrink-wrap.c index e51bd36..84abd6b 100644 --- a/gcc/shrink-wrap.c +++ b/gcc/shrink-wrap.c @@ -750,9 +750,21 @@ try_shrink_wrapping (edge *entry_edge, bitmap_head *bb_with, /* If we can move PRO back without having to duplicate more blocks, do so. We do this because putting the prologue earlier is better for scheduling. + We can move back to a block PRE if every path from PRE will eventually need a prologue, that is, PRO is a post-dominator of PRE. PRE needs - to dominate every block reachable from itself. */ + to dominate every block reachable from itself. We keep in BB_TMP a + bitmap of the blocks reachable from PRE that we already found, and in + VEC a stack of those we still need to consider. + + Any block reachable from PRE is also reachable from all predecessors + of PRE, so if we find we need to move PRE back further we can leave + everything not considered so far on the stack. Any block dominated + by PRE is also dominated by all other dominators of PRE, so anything + found good for some PRE does not need to be reconsidered later. + + We don't need to update BB_WITH because none of the new blocks found + can jump to a block that does not need the prologue. */ if (pro != entry) { @@ -775,18 +787,15 @@ try_shrink_wrapping (edge *entry_edge, bitmap_head *bb_with, bool ok = true; while (!vec.is_empty ()) { - basic_block bb = vec.pop (); - bitmap_set_bit (bb_tmp, pre->index); - - if (!dominated_by_p (CDI_DOMINATORS, bb, pre)) + if (!dominated_by_p (CDI_DOMINATORS, vec.last (), pre)) { ok = false; break; } + basic_block bb = vec.pop (); FOR_EACH_EDGE (e, ei, bb->succs) - if (!bitmap_bit_p (bb_with, e->dest->index) - && bitmap_set_bit (bb_tmp, e->dest->index)) + if (bitmap_set_bit (bb_tmp, e->dest->index)) vec.quick_push (e->dest); } |