This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Fix PR 53743 and other -freorder-blocks-and-partition failures
- From: Teresa Johnson <tejohnson at google dot com>
- To: Steven Bosscher <stevenb dot gcc at gmail dot com>
- Cc: David Li <davidxl at google dot com>, Diego Novillo <dnovillo at google dot com>, Jan Hubicka <hubicka at ucw dot cz>, "gcc-patches at gcc dot gnu dot org" <gcc-patches at gcc dot gnu dot org>
- Date: Wed, 22 May 2013 14:05:39 -0700
- Subject: Re: Fix PR 53743 and other -freorder-blocks-and-partition failures
- References: <20130520025501 dot E9F99805D7 at tjsboxrox dot mtv dot corp dot google dot com> <CABu31nM46y++4yr=SVTHmpr1=REJPtTFjyWZ_ybgvWk64BsDJg at mail dot gmail dot com> <CAAe5K+V1LkiVQxOUjoqL6R1EFgpDMpPy5jvHzYOGT3JnX68=0w at mail dot gmail dot com>
Revised patch included below. The spacing of my pasted in patch text
looks funky again, let me know if you want the patch as an attachment
instead.
I addressed all of Steven's comments, except for the suggestion to use
gcc_assert
instead of error() in verify_hot_cold_block_grouping() to keep this consistent
with the rest of the verify_flow_info subroutines (let me know if this is ok).
The other main changes:
(1) Added several test cases (cloned from the torture subdirectories,
where I manually
built/ran with FDO and -freorder-blocks-and-partition with both the
current trunk and
my fixed trunk compiler, and was able to expose some failures I fixed.
(2) Changed existing tree-prof tests that used
-freorder-blocks-and-partition to be
built with -O2 instead of -O, so that partitioning actually kicks in.
(3) Fixed a couple of failures in the new
verify_hot_cold_block_grouping() checks
exposed by the torture tests I ran manually with splitting (2 of the
tests cloned
to tree-prof in this patch). One was in computed goto where we were
too aggressive
about cloning crossing edges, and the other was in rtl_split_edge
called from the "stack"
pass which was not correctly inserting the new bb in the correct partition since
bb layout is complete at that point.
Re-tested on x86_64-unknown-linux-gnu with bootstrap and profiledbootstrap
builds and regression testing. Re-built/ran cpu2006int with profile
feedback and -freorder-blocks-and-partition enabled.
Ok for trunk?
Thanks!
Teresa
2013-05-22 Teresa Johnson <tejohnson@google.com>
* ifcvt.c (find_if_case_1): Replace BB_COPY_PARTITION with assert
as this is now done by redirect_edge_and_branch_force.
* function.c (thread_prologue_and_epilogue_insns): Insert new bb after
barriers, and fix interaction with splitting.
* emit-rtl.c (try_split): Copy REG_CROSSING_JUMP notes.
* cfgcleanup.c (try_forward_edges): Fix early return value to properly
reflect changes made in the routine.
* bb-reorder.c (emit_barrier_after_bb): Move to cfgrtl.c.
(fix_up_fall_thru_edges): Remove incorrect check for bb layout order
since this is called in cfglayout mode, and replace partition fixup
with assert as that is now done by force_nonfallthru_and_redirect.
(add_reg_crossing_jump_notes): Handle the fact that some jumps may
already be marked with region crossing note.
(insert_section_boundary_note): Make non-static, gate on flag
has_bb_partition, rewrite to also check for multiple partitions.
(rest_of_handle_reorder_blocks): Remove call to
insert_section_boundary_note, now done later during free_cfg.
(duplicate_computed_gotos): Don't duplicate partition crossing edge.
* bb-reorder.h (insert_section_boundary_note): Declare.
* Makefile.in (cfgrtl.o): Depend on bb-reorder.h
* cfgrtl.c (rest_of_pass_free_cfg): If partitions exist
invoke insert_section_boundary_note.
(try_redirect_by_replacing_jump): Remove unnecessary
check for region crossing note.
(fixup_partition_crossing): New function.
(rtl_redirect_edge_and_branch): Fixup partition boundaries.
(emit_barrier_after_bb): Move here from bb-reorder.c, handle insertion
in non-cfglayout mode.
(force_nonfallthru_and_redirect): Fixup partition boundaries,
remove old code that tried to do this. Emit barrier correctly
when we are in cfglayout mode.
(last_bb_in_partition): New function.
(rtl_split_edge): Correctly fixup partition boundaries.
(commit_one_edge_insertion): Remove old code that tried to
fixup region crossing edge since this is now handled in
split_block, and set up insertion point correctly since
block may now end in a jump.
(verify_hot_cold_block_grouping): Guard against checking when not in
linearized RTL mode.
(rtl_verify_edges): Add checks for incorrect/missing REG_CROSSING_JUMP
notes.
(rtl_verify_flow_info_1): Move verify_hot_cold_block_grouping to
rtl_verify_flow_info, so not called in cfglayout mode.
(rtl_verify_flow_info): Move verify_hot_cold_block_grouping here.
(fixup_reorder_chain): Remove old code that attempted to fixup region
crossing note as this is now handled in force_nonfallthru_and_redirect.
(duplicate_insn_chain): Don't duplicate switch section notes.
(rtl_can_remove_branch_p): Remove unnecessary check for region crossing
note.
* basic-block.h (emit_barrier_after_bb): Declare.
* testsuite/gcc.dg/tree-prof/va-arg-pack-1.c: Cloned from c-torture, made
into -freorder-blocks-and-partition test.
* testsuite/gcc.dg/tree-prof/comp-goto-1.c: Ditto.
* testsuite/gcc.dg/tree-prof/20041218-1.c: Ditto.
* testsuite/gcc.dg/tree-prof/pr52027.c: Use -O2.
* testsuite/gcc.dg/tree-prof/pr50907.c: Ditto.
* testsuite/gcc.dg/tree-prof/pr45354.c: Ditto.
* testsuite/g++.dg/tree-prof/partition2.C: Ditto.
* testsuite/g++.dg/tree-prof/partition3.C: Ditto.
Index: ifcvt.c
===================================================================
--- ifcvt.c (revision 199014)
+++ ifcvt.c (working copy)
@@ -3905,10 +3905,9 @@ find_if_case_1 (basic_block test_bb, edge then_edg
if (new_bb)
{
df_bb_replace (then_bb_index, new_bb);
- /* Since the fallthru edge was redirected from test_bb to new_bb,
- we need to ensure that new_bb is in the same partition as
- test bb (you can not fall through across section boundaries). */
- BB_COPY_PARTITION (new_bb, test_bb);
+ /* This should have been done above via force_nonfallthru_and_redirect
+ (possibly called from redirect_edge_and_branch_force). */
+ gcc_checking_assert (BB_PARTITION (new_bb) == BB_PARTITION (test_bb));
}
num_true_changes++;
Index: function.c
===================================================================
--- function.c (revision 199014)
+++ function.c (working copy)
@@ -6270,8 +6270,10 @@ thread_prologue_and_epilogue_insns (void)
break;
if (e)
{
- copy_bb = create_basic_block (NEXT_INSN (BB_END (e->src)),
- NULL_RTX, e->src);
+ /* Make sure we insert after any barriers. */
+ rtx end = get_last_bb_insn (e->src);
+ copy_bb = create_basic_block (NEXT_INSN (end),
+ NULL_RTX, e->src);
BB_COPY_PARTITION (copy_bb, e->src);
}
else
@@ -6538,7 +6540,7 @@ epilogue_done:
basic_block simple_return_block_cold = NULL;
edge pending_edge_hot = NULL;
edge pending_edge_cold = NULL;
- basic_block exit_pred = EXIT_BLOCK_PTR->prev_bb;
+ basic_block exit_pred;
int i;
gcc_assert (entry_edge != orig_entry_edge);
@@ -6566,6 +6568,12 @@ epilogue_done:
else
pending_edge_cold = e;
}
+
+ /* Save a pointer to the exit's predecessor BB for use in
+ inserting new BBs at the end of the function. Do this
+ after the call to split_block above which may split
+ the original exit pred. */
+ exit_pred = EXIT_BLOCK_PTR->prev_bb;
FOR_EACH_VEC_ELT (unconverted_simple_returns, i, e)
{
Index: emit-rtl.c
===================================================================
--- emit-rtl.c (revision 199014)
+++ emit-rtl.c (working copy)
@@ -3574,6 +3574,7 @@ try_split (rtx pat, rtx trial, int last)
break;
case REG_NON_LOCAL_GOTO:
+ case REG_CROSSING_JUMP:
for (insn = insn_last; insn != NULL_RTX; insn = PREV_INSN (insn))
{
if (JUMP_P (insn))
Index: cfgcleanup.c
===================================================================
--- cfgcleanup.c (revision 199014)
+++ cfgcleanup.c (working copy)
@@ -456,7 +456,7 @@ try_forward_edges (int mode, basic_block b)
if (first != EXIT_BLOCK_PTR
&& find_reg_note (BB_END (first), REG_CROSSING_JUMP, NULL_RTX))
- return false;
+ return changed;
while (counter < n_basic_blocks)
{
Index: bb-reorder.c
===================================================================
--- bb-reorder.c (revision 199014)
+++ bb-reorder.c (working copy)
@@ -1380,15 +1380,6 @@ get_uncond_jump_length (void)
return length;
}
-/* Emit a barrier into the footer of BB. */
-
-static void
-emit_barrier_after_bb (basic_block bb)
-{
- rtx barrier = emit_barrier_after (BB_END (bb));
- BB_FOOTER (bb) = unlink_insn_chain (barrier, barrier);
-}
-
/* The landing pad OLD_LP, in block OLD_BB, has edges from both partitions.
Duplicate the landing pad and split the edges so that no EH edge
crosses partitions. */
@@ -1720,8 +1711,7 @@ fix_up_fall_thru_edges (void)
(i.e. fix it so the fall through does not cross and
the cond jump does). */
- if (!cond_jump_crosses
- && cur_bb->aux == cond_jump->dest)
+ if (!cond_jump_crosses)
{
/* Find label in fall_thru block. We've already added
any missing labels, so there must be one. */
@@ -1765,10 +1755,10 @@ fix_up_fall_thru_edges (void)
new_bb->aux = cur_bb->aux;
cur_bb->aux = new_bb;
- /* Make sure new fall-through bb is in same
- partition as bb it's falling through from. */
+ /* This is done by force_nonfallthru_and_redirect. */
+ gcc_assert (BB_PARTITION (new_bb)
+ == BB_PARTITION (cur_bb));
- BB_COPY_PARTITION (new_bb, cur_bb);
single_succ_edge (new_bb)->flags |= EDGE_CROSSING;
}
else
@@ -2064,7 +2054,10 @@ add_reg_crossing_jump_notes (void)
FOR_EACH_BB (bb)
FOR_EACH_EDGE (e, ei, bb->succs)
if ((e->flags & EDGE_CROSSING)
- && JUMP_P (BB_END (e->src)))
+ && JUMP_P (BB_END (e->src))
+ /* Some notes were added during fix_up_fall_thru_edges, via
+ force_nonfallthru_and_redirect. */
+ && !find_reg_note (BB_END (e->src), REG_CROSSING_JUMP, NULL_RTX))
add_reg_note (BB_END (e->src), REG_CROSSING_JUMP, NULL_RTX);
}
@@ -2133,25 +2126,39 @@ reorder_basic_blocks (void)
encountering this note will make the compiler switch between the
hot and cold text sections. */
-static void
+void
insert_section_boundary_note (void)
{
basic_block bb;
- int first_partition = 0;
+ int err = 0;
+ bool switched_sections = false;
+ int current_partition = 0;
- if (!flag_reorder_blocks_and_partition)
+ if (!crtl->has_bb_partition)
return;
FOR_EACH_BB (bb)
{
- if (!first_partition)
- first_partition = BB_PARTITION (bb);
- if (BB_PARTITION (bb) != first_partition)
+ if (!current_partition)
+ current_partition = BB_PARTITION (bb);
+ if (BB_PARTITION (bb) != current_partition)
{
- emit_note_before (NOTE_INSN_SWITCH_TEXT_SECTIONS, BB_HEAD (bb));
- break;
+ if (switched_sections)
+ {
+ error ("multiple hot/cold transitions found (bb %i)",
+ bb->index);
+ err = 1;
+ }
+ else
+ {
+ switched_sections = true;
+ emit_note_before (NOTE_INSN_SWITCH_TEXT_SECTIONS, BB_HEAD (bb));
+ current_partition = BB_PARTITION (bb);
+ }
}
}
+
+ gcc_assert(!err);
}
static bool
@@ -2180,8 +2187,6 @@ rest_of_handle_reorder_blocks (void)
bb->aux = bb->next_bb;
cfg_layout_finalize ();
- /* Add NOTE_INSN_SWITCH_TEXT_SECTIONS notes. */
- insert_section_boundary_note ();
return 0;
}
@@ -2315,6 +2320,11 @@ duplicate_computed_gotos (void)
if (!bitmap_bit_p (candidates, single_succ (bb)->index))
continue;
+ /* Don't duplicate a partition crossing edge, which requires difficult
+ fixup. */
+ if (find_reg_note (BB_END (bb), REG_CROSSING_JUMP, NULL_RTX))
+ continue;
+
new_bb = duplicate_block (single_succ (bb), single_succ_edge (bb), bb);
new_bb->aux = bb->aux;
bb->aux = new_bb;
Index: bb-reorder.h
===================================================================
--- bb-reorder.h (revision 199014)
+++ bb-reorder.h (working copy)
@@ -35,4 +35,6 @@ extern struct target_bb_reorder *this_target_bb_re
extern int get_uncond_jump_length (void);
+extern void insert_section_boundary_note (void);
+
#endif
Index: Makefile.in
===================================================================
--- Makefile.in (revision 199014)
+++ Makefile.in (working copy)
@@ -3151,7 +3151,7 @@ cfgrtl.o : cfgrtl.c $(CONFIG_H) $(SYSTEM_H) corety
$(FUNCTION_H) $(EXCEPT_H) $(TM_P_H) $(INSN_ATTR_H) \
insn-config.h $(EXPR_H) \
$(CFGLOOP_H) $(OBSTACK_H) $(TARGET_H) $(TREE_H) \
- $(TREE_PASS_H) $(DF_H) $(GGC_H) $(COMMON_TARGET_H) gt-cfgrtl.h
+ $(TREE_PASS_H) $(DF_H) $(GGC_H) $(COMMON_TARGET_H) gt-cfgrtl.h bb-reorder.h
cfganal.o : cfganal.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(BASIC_BLOCK_H) \
$(TIMEVAR_H) sbitmap.h $(BITMAP_H)
cfgbuild.o : cfgbuild.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) $(RTL_H) \
Index: cfgrtl.c
===================================================================
--- cfgrtl.c (revision 199014)
+++ cfgrtl.c (working copy)
@@ -44,6 +44,7 @@ along with GCC; see the file COPYING3. If not see
#include "tree.h"
#include "hard-reg-set.h"
#include "basic-block.h"
+#include "bb-reorder.h"
#include "regs.h"
#include "flags.h"
#include "function.h"
@@ -451,6 +452,9 @@ rest_of_pass_free_cfg (void)
}
#endif
+ if (crtl->has_bb_partition)
+ insert_section_boundary_note ();
+
free_bb_for_insn ();
return 0;
}
@@ -981,8 +985,7 @@ try_redirect_by_replacing_jump (edge e, basic_bloc
partition boundaries). See the comments at the top of
bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
- if (find_reg_note (insn, REG_CROSSING_JUMP, NULL_RTX)
- || BB_PARTITION (src) != BB_PARTITION (target))
+ if (BB_PARTITION (src) != BB_PARTITION (target))
return NULL;
/* We can replace or remove a complex jump only when we have exactly
@@ -1291,6 +1294,53 @@ redirect_branch_edge (edge e, basic_block target)
return e;
}
+/* Called when edge E has been redirected to a new destination,
+ in order to update the region crossing flag on the edge and
+ jump. */
+
+static void
+fixup_partition_crossing (edge e)
+{
+ rtx note;
+
+ if (e->src == ENTRY_BLOCK_PTR || e->dest == EXIT_BLOCK_PTR)
+ return;
+ /* If we redirected an existing edge, it may already be marked
+ crossing, even though the new src is missing a reg crossing note.
+ But make sure reg crossing note doesn't already exist before
+ inserting. */
+ if (BB_PARTITION (e->src) != BB_PARTITION (e->dest))
+ {
+ e->flags |= EDGE_CROSSING;
+ note = find_reg_note (BB_END (e->src), REG_CROSSING_JUMP, NULL_RTX);
+ if (JUMP_P (BB_END (e->src))
+ && !note)
+ add_reg_note (BB_END (e->src), REG_CROSSING_JUMP, NULL_RTX);
+ }
+ else if (BB_PARTITION (e->src) == BB_PARTITION (e->dest))
+ {
+ e->flags &= ~EDGE_CROSSING;
+ /* Remove the section crossing note from jump at end of
+ src if it exists, and if no other successors are
+ still crossing. */
+ note = find_reg_note (BB_END (e->src), REG_CROSSING_JUMP, NULL_RTX);
+ if (note)
+ {
+ bool has_crossing_succ = false;
+ edge e2;
+ edge_iterator ei;
+ FOR_EACH_EDGE (e2, ei, e->src->succs)
+ {
+ has_crossing_succ |= (e2->flags & EDGE_CROSSING);
+ if (has_crossing_succ)
+ break;
+ }
+ if (!has_crossing_succ)
+ remove_note (BB_END (e->src), note);
+ }
+ }
+}
+
/* Attempt to change code to redirect edge E to TARGET. Don't do that on
expense of adding new instructions or reordering basic blocks.
@@ -1307,16 +1357,18 @@ rtl_redirect_edge_and_branch (edge e, basic_block
{
edge ret;
basic_block src = e->src;
+ basic_block dest = e->dest;
if (e->flags & (EDGE_ABNORMAL_CALL | EDGE_EH))
return NULL;
- if (e->dest == target)
+ if (dest == target)
return e;
if ((ret = try_redirect_by_replacing_jump (e, target, false)) != NULL)
{
df_set_bb_dirty (src);
+ fixup_partition_crossing (ret);
return ret;
}
@@ -1325,9 +1377,22 @@ rtl_redirect_edge_and_branch (edge e, basic_block
return NULL;
df_set_bb_dirty (src);
+ fixup_partition_crossing (ret);
return ret;
}
+/* Emit a barrier after BB, into the footer if we are in CFGLAYOUT mode. */
+
+void
+emit_barrier_after_bb (basic_block bb)
+{
+ rtx barrier = emit_barrier_after (BB_END (bb));
+ gcc_assert (current_ir_type() == IR_RTL_CFGRTL
+ || current_ir_type () == IR_RTL_CFGLAYOUT);
+ if (current_ir_type () == IR_RTL_CFGLAYOUT)
+ BB_FOOTER (bb) = unlink_insn_chain (barrier, barrier);
+}
+
/* Like force_nonfallthru below, but additionally performs redirection
Used by redirect_edge_and_branch_force. JUMP_LABEL is used only
when redirecting to the EXIT_BLOCK, it is either ret_rtx or
@@ -1492,12 +1557,6 @@ force_nonfallthru_and_redirect (edge e, basic_bloc
/* Make sure new block ends up in correct hot/cold section. */
BB_COPY_PARTITION (jump_block, e->src);
- if (flag_reorder_blocks_and_partition
- && targetm_common.have_named_sections
- && JUMP_P (BB_END (jump_block))
- && !any_condjump_p (BB_END (jump_block))
- && (EDGE_SUCC (jump_block, 0)->flags & EDGE_CROSSING))
- add_reg_note (BB_END (jump_block), REG_CROSSING_JUMP, NULL_RTX);
/* Wire edge in. */
new_edge = make_edge (e->src, jump_block, EDGE_FALLTHRU);
@@ -1508,6 +1567,10 @@ force_nonfallthru_and_redirect (edge e, basic_bloc
redirect_edge_pred (e, jump_block);
e->probability = REG_BR_PROB_BASE;
+ /* If e->src was previously region crossing, it no longer is
+ and the reg crossing note should be removed. */
+ fixup_partition_crossing (new_edge);
+
/* If asm goto has any label refs to target's label,
add also edge from asm goto bb to target. */
if (asm_goto_edge)
@@ -1559,13 +1622,16 @@ force_nonfallthru_and_redirect (edge e, basic_bloc
LABEL_NUSES (label)++;
}
- emit_barrier_after (BB_END (jump_block));
+ /* We might be in cfg layout mode, and if so, the following routine will
+ insert the barrier correctly. */
+ emit_barrier_after_bb (jump_block);
redirect_edge_succ_nodup (e, target);
if (abnormal_edge_flags)
make_edge (src, target, abnormal_edge_flags);
df_mark_solutions_dirty ();
+ fixup_partition_crossing (e);
return new_bb;
}
@@ -1654,6 +1720,21 @@ rtl_move_block_after (basic_block bb ATTRIBUTE_UNU
return false;
}
+/* Locate the last bb in the same partition as START_BB. */
+
+static basic_block
+last_bb_in_partition (basic_block start_bb)
+{
+ basic_block bb;
+ FOR_BB_BETWEEN (bb, start_bb, EXIT_BLOCK_PTR, next_bb)
+ {
+ if (BB_PARTITION (start_bb) != BB_PARTITION (bb->next_bb))
+ return bb;
+ }
+ /* Return bb before EXIT_BLOCK_PTR. */
+ return bb->prev_bb;
+}
+
/* Split a (typically critical) edge. Return the new block.
The edge must not be abnormal.
@@ -1664,7 +1745,7 @@ rtl_move_block_after (basic_block bb ATTRIBUTE_UNU
static basic_block
rtl_split_edge (edge edge_in)
{
- basic_block bb;
+ basic_block bb, new_bb;
rtx before;
/* Abnormal edges cannot be split. */
@@ -1696,13 +1777,50 @@ rtl_split_edge (edge edge_in)
}
else
{
- bb = create_basic_block (before, NULL, edge_in->dest->prev_bb);
- /* ??? Why not edge_in->dest->prev_bb here? */
- BB_COPY_PARTITION (bb, edge_in->dest);
+ if (edge_in->src == ENTRY_BLOCK_PTR)
+ {
+ bb = create_basic_block (before, NULL, edge_in->dest->prev_bb);
+ BB_COPY_PARTITION (bb, edge_in->dest);
+ }
+ else
+ {
+ basic_block after = edge_in->dest->prev_bb;
+ /* If this is post-bb reordering, and the edge crosses a partition
+ boundary, the new block needs to be inserted in the bb chain
+ at the end of the src partition (since we put the new bb into
+ that partition, see below). Otherwise we may end up creating
+ an extra partition crossing in the chain, which is illegal.
+ It can't go after the src, because src may have a fall-through
+ to a different block. */
+ if (crtl->bb_reorder_complete
+ && (edge_in->flags & EDGE_CROSSING))
+ {
+ after = last_bb_in_partition (edge_in->src);
+ before = NEXT_INSN (BB_END (after));
+ /* The instruction following the last bb in partition should
+ be a barrier, since it cannot end in a fall-through. */
+ gcc_checking_assert (BARRIER_P (before));
+ before = NEXT_INSN (before);
+ }
+ bb = create_basic_block (before, NULL, after);
+ /* Put the split bb into the src partition, to avoid creating
+ a situation where a cold bb dominates a hot bb, in the case
+ where src is cold and dest is hot. The src will dominate
+ the new bb (whereas it might not have dominated dest). */
+ BB_COPY_PARTITION (bb, edge_in->src);
+ }
}
make_single_succ_edge (bb, edge_in->dest, EDGE_FALLTHRU);
+ /* Can't allow a region crossing edge to be fallthrough. */
+ if (BB_PARTITION (bb) != BB_PARTITION (edge_in->dest)
+ && edge_in->dest != EXIT_BLOCK_PTR)
+ {
+ new_bb = force_nonfallthru (single_succ_edge (bb));
+ gcc_assert (!new_bb);
+ }
+
/* For non-fallthru edges, we must adjust the predecessor's
jump instruction to target our new block. */
if ((edge_in->flags & EDGE_FALLTHRU) == 0)
@@ -1815,17 +1933,13 @@ commit_one_edge_insertion (edge e)
else
{
bb = split_edge (e);
- after = BB_END (bb);
- if (flag_reorder_blocks_and_partition
- && targetm_common.have_named_sections
- && e->src != ENTRY_BLOCK_PTR
- && BB_PARTITION (e->src) == BB_COLD_PARTITION
- && !(e->flags & EDGE_CROSSING)
- && JUMP_P (after)
- && !any_condjump_p (after)
- && (single_succ_edge (bb)->flags & EDGE_CROSSING))
- add_reg_note (after, REG_CROSSING_JUMP, NULL_RTX);
+ /* If E crossed a partition boundary, we needed to make bb end in
+ a region-crossing jump, even though it was originally fallthru. */
+ if (JUMP_P (BB_END (bb)))
+ before = BB_END (bb);
+ else
+ after = BB_END (bb);
}
/* Now that we've found the spot, do the insertion. */
@@ -2071,7 +2185,11 @@ verify_hot_cold_block_grouping (void)
bool switched_sections = false;
int current_partition = BB_UNPARTITIONED;
- if (!crtl->bb_reorder_complete)
+ /* Even after bb reordering is complete, we go into cfglayout mode
+ again (in compgoto). Ensure we don't call this before going back
+ into linearized RTL when any layout fixes would have been committed. */
+ if (!crtl->bb_reorder_complete
+ || current_ir_type() != IR_RTL_CFGRTL)
return err;
FOR_EACH_BB (bb)
@@ -2116,6 +2234,7 @@ rtl_verify_edges (void)
edge e, fallthru = NULL;
edge_iterator ei;
rtx note;
+ bool has_crossing_edge = false;
if (JUMP_P (BB_END (bb))
&& (note = find_reg_note (BB_END (bb), REG_BR_PROB, NULL_RTX))
@@ -2141,6 +2260,7 @@ rtl_verify_edges (void)
is_crossing = (BB_PARTITION (e->src) != BB_PARTITION (e->dest)
&& e->src != ENTRY_BLOCK_PTR
&& e->dest != EXIT_BLOCK_PTR);
+ has_crossing_edge |= is_crossing;
if (e->flags & EDGE_CROSSING)
{
if (!is_crossing)
@@ -2160,6 +2280,13 @@ rtl_verify_edges (void)
e->src->index);
err = 1;
}
+ if (JUMP_P (BB_END (bb))
+ && !find_reg_note (BB_END (bb), REG_CROSSING_JUMP, NULL_RTX))
+ {
+ error ("No region crossing jump at section boundary in bb %i",
+ bb->index);
+ err = 1;
+ }
}
else if (is_crossing)
{
@@ -2188,6 +2315,15 @@ rtl_verify_edges (void)
n_abnormal++;
}
+ if (!has_crossing_edge
+ && find_reg_note (BB_END (bb), REG_CROSSING_JUMP, NULL_RTX))
+ {
+ print_rtl_with_bb (stderr, get_insns (), TDF_RTL |
TDF_BLOCKS | TDF_DETAILS);
+ error ("Region crossing jump across same section in bb %i",
+ bb->index);
+ err = 1;
+ }
+
if (n_eh && !find_reg_note (BB_END (bb), REG_EH_REGION, NULL_RTX))
{
error ("missing REG_EH_REGION note at the end of bb %i", bb->index);
@@ -2395,8 +2531,6 @@ rtl_verify_flow_info_1 (void)
err |= rtl_verify_edges ();
- err |= verify_hot_cold_block_grouping();
-
return err;
}
@@ -2642,6 +2776,8 @@ rtl_verify_flow_info (void)
err |= rtl_verify_bb_layout ();
+ err |= verify_hot_cold_block_grouping ();
+
return err;
}
@@ -3343,7 +3479,7 @@ fixup_reorder_chain (void)
edge e_fall, e_taken, e;
rtx bb_end_insn;
rtx ret_label = NULL_RTX;
- basic_block nb, src_bb;
+ basic_block nb;
edge_iterator ei;
if (EDGE_COUNT (bb->succs) == 0)
@@ -3478,7 +3614,6 @@ fixup_reorder_chain (void)
/* We got here if we need to add a new jump insn.
Note force_nonfallthru can delete E_FALL and thus we have to
save E_FALL->src prior to the call to force_nonfallthru. */
- src_bb = e_fall->src;
nb = force_nonfallthru_and_redirect (e_fall, e_fall->dest, ret_label);
if (nb)
{
@@ -3486,17 +3621,6 @@ fixup_reorder_chain (void)
bb->aux = nb;
/* Don't process this new block. */
bb = nb;
-
- /* Make sure new bb is tagged for correct section (same as
- fall-thru source, since you cannot fall-thru across
- section boundaries). */
- BB_COPY_PARTITION (src_bb, single_pred (bb));
- if (flag_reorder_blocks_and_partition
- && targetm_common.have_named_sections
- && JUMP_P (BB_END (bb))
- && !any_condjump_p (BB_END (bb))
- && (EDGE_SUCC (bb, 0)->flags & EDGE_CROSSING))
- add_reg_note (BB_END (bb), REG_CROSSING_JUMP, NULL_RTX);
}
}
@@ -3796,10 +3920,11 @@ duplicate_insn_chain (rtx from, rtx to)
case NOTE_INSN_FUNCTION_BEG:
/* There is always just single entry to function. */
case NOTE_INSN_BASIC_BLOCK:
+ /* We should only switch text sections once. */
+ case NOTE_INSN_SWITCH_TEXT_SECTIONS:
break;
case NOTE_INSN_EPILOGUE_BEG:
- case NOTE_INSN_SWITCH_TEXT_SECTIONS:
emit_note_copy (insn);
break;
@@ -4611,8 +4736,7 @@ rtl_can_remove_branch_p (const_edge e)
if (e->flags & (EDGE_ABNORMAL_CALL | EDGE_EH))
return false;
- if (find_reg_note (insn, REG_CROSSING_JUMP, NULL_RTX)
- || BB_PARTITION (src) != BB_PARTITION (target))
+ if (BB_PARTITION (src) != BB_PARTITION (target))
return false;
if (!onlyjump_p (insn)
Index: basic-block.h
===================================================================
--- basic-block.h (revision 199014)
+++ basic-block.h (working copy)
@@ -796,6 +796,7 @@ extern basic_block force_nonfallthru_and_redirect
extern bool contains_no_active_insn_p (const_basic_block);
extern bool forwarder_block_p (const_basic_block);
extern bool can_fallthru (basic_block, basic_block);
+extern void emit_barrier_after_bb (basic_block bb);
/* In cfgbuild.c. */
extern void find_many_sub_basic_blocks (sbitmap);
Index: testsuite/gcc.dg/tree-prof/va-arg-pack-1.c
===================================================================
--- testsuite/gcc.dg/tree-prof/va-arg-pack-1.c (revision 0)
+++ testsuite/gcc.dg/tree-prof/va-arg-pack-1.c (revision 0)
@@ -0,0 +1,145 @@
+/* __builtin_va_arg_pack () builtin tests. */
+/* { dg-require-effective-target freorder } */
+/* { dg-options "-O2 -freorder-blocks-and-partition" } */
+
+#include <stdarg.h>
+
+extern void abort (void);
+
+int v1 = 8;
+long int v2 = 3;
+void *v3 = (void *) &v2;
+struct A { char c[16]; } v4 = { "foo" };
+long double v5 = 40;
+char seen[20];
+int cnt;
+
+__attribute__ ((noinline)) int
+foo1 (int x, int y, ...)
+{
+ int i;
+ long int l;
+ void *v;
+ struct A a;
+ long double ld;
+ va_list ap;
+
+ va_start (ap, y);
+ if (x < 0 || x >= 20 || seen[x])
+ abort ();
+ seen[x] = ++cnt;
+ if (y != 6)
+ abort ();
+ i = va_arg (ap, int);
+ if (i != 5)
+ abort ();
+ switch (x)
+ {
+ case 0:
+ i = va_arg (ap, int);
+ if (i != 9 || v1 != 9)
+ abort ();
+ a = va_arg (ap, struct A);
+ if (__builtin_memcmp (a.c, v4.c, sizeof (a.c)) != 0)
+ abort ();
+ v = (void *) va_arg (ap, struct A *);
+ if (v != (void *) &v4)
+ abort ();
+ l = va_arg (ap, long int);
+ if (l != 3 || v2 != 4)
+ abort ();
+ break;
+ case 1:
+ ld = va_arg (ap, long double);
+ if (ld != 41 || v5 != ld)
+ abort ();
+ i = va_arg (ap, int);
+ if (i != 8)
+ abort ();
+ v = va_arg (ap, void *);
+ if (v != &v2)
+ abort ();
+ break;
+ case 2:
+ break;
+ default:
+ abort ();
+ }
+ va_end (ap);
+ return x;
+}
+
+__attribute__ ((noinline)) int
+foo2 (int x, int y, ...)
+{
+ long long int ll;
+ void *v;
+ struct A a, b;
+ long double ld;
+ va_list ap;
+
+ va_start (ap, y);
+ if (x < 0 || x >= 20 || seen[x])
+ abort ();
+ seen[x] = ++cnt | 64;
+ if (y != 10)
+ abort ();
+ switch (x)
+ {
+ case 11:
+ break;
+ case 12:
+ ld = va_arg (ap, long double);
+ if (ld != 41 || v5 != 40)
+ abort ();
+ a = va_arg (ap, struct A);
+ if (__builtin_memcmp (a.c, v4.c, sizeof (a.c)) != 0)
+ abort ();
+ b = va_arg (ap, struct A);
+ if (__builtin_memcmp (b.c, v4.c, sizeof (b.c)) != 0)
+ abort ();
+ v = va_arg (ap, void *);
+ if (v != &v2)
+ abort ();
+ ll = va_arg (ap, long long int);
+ if (ll != 16LL)
+ abort ();
+ break;
+ case 2:
+ break;
+ default:
+ abort ();
+ }
+ va_end (ap);
+ return x + 8;
+}
+
+__attribute__ ((noinline)) int
+foo3 (void)
+{
+ return 6;
+}
+
+extern inline __attribute__ ((always_inline, gnu_inline)) int
+bar (int x, ...)
+{
+ if (x < 10)
+ return foo1 (x, foo3 (), 5, __builtin_va_arg_pack ());
+ return foo2 (x, foo3 () + 4, __builtin_va_arg_pack ());
+}
+
+int
+main (void)
+{
+ if (bar (0, ++v1, v4, &v4, v2++) != 0)
+ abort ();
+ if (bar (1, ++v5, 8, v3) != 1)
+ abort ();
+ if (bar (2) != 2)
+ abort ();
+ if (bar (v1 + 2) != 19)
+ abort ();
+ if (bar (v1 + 3, v5--, v4, v4, v3, 16LL) != 20)
+ abort ();
+ return 0;
+}
Index: testsuite/gcc.dg/tree-prof/comp-goto-1.c
===================================================================
--- testsuite/gcc.dg/tree-prof/comp-goto-1.c (revision 0)
+++ testsuite/gcc.dg/tree-prof/comp-goto-1.c (revision 0)
@@ -0,0 +1,166 @@
+/* { dg-require-effective-target freorder } */
+/* { dg-options "-O2 -freorder-blocks-and-partition" } */
+#include <stdlib.h>
+
+#if !defined(NO_LABEL_VALUES) && (!defined(STACK_SIZE) || STACK_SIZE
>= 4000) && __INT_MAX__ >= 2147483647
+typedef unsigned int uint32;
+typedef signed int sint32;
+
+typedef uint32 reg_t;
+
+typedef unsigned long int host_addr_t;
+typedef uint32 target_addr_t;
+typedef sint32 target_saddr_t;
+
+typedef union
+{
+ struct
+ {
+ unsigned int offset:18;
+ unsigned int ignore:4;
+ unsigned int s1:8;
+ int :2;
+ signed int simm:14;
+ unsigned int s3:8;
+ unsigned int s2:8;
+ int pad2:2;
+ } f1;
+ long long ll;
+ double d;
+} insn_t;
+
+typedef struct
+{
+ target_addr_t vaddr_tag;
+ unsigned long int rigged_paddr;
+} tlb_entry_t;
+
+typedef struct
+{
+ insn_t *pc;
+ reg_t registers[256];
+ insn_t *program;
+ tlb_entry_t tlb_tab[0x100];
+} environment_t;
+
+enum operations
+{
+ LOAD32_RR,
+ METAOP_DONE
+};
+
+host_addr_t
+f ()
+{
+ abort ();
+}
+
+reg_t
+simulator_kernel (int what, environment_t *env)
+{
+ register insn_t *pc = env->pc;
+ register reg_t *regs = env->registers;
+ register insn_t insn;
+ register int s1;
+ register reg_t r2;
+ register void *base_addr = &&sim_base_addr;
+ register tlb_entry_t *tlb = env->tlb_tab;
+
+ if (what != 0)
+ {
+ int i;
+ static void *op_map[] =
+ {
+ &&L_LOAD32_RR,
+ &&L_METAOP_DONE,
+ };
+ insn_t *program = env->program;
+ for (i = 0; i < what; i++)
+ program[i].f1.offset = op_map[program[i].f1.offset] - base_addr;
+ }
+
+ sim_base_addr:;
+
+ insn = *pc++;
+ r2 = (*(reg_t *) (((char *) regs) + (insn.f1.s2 << 2)));
+ s1 = (insn.f1.s1 << 2);
+ goto *(base_addr + insn.f1.offset);
+
+ L_LOAD32_RR:
+ {
+ target_addr_t vaddr_page = r2 / 4096;
+ unsigned int x = vaddr_page % 0x100;
+ insn = *pc++;
+
+ for (;;)
+ {
+ target_addr_t tag = tlb[x].vaddr_tag;
+ host_addr_t rigged_paddr = tlb[x].rigged_paddr;
+
+ if (tag == vaddr_page)
+ {
+ *(reg_t *) (((char *) regs) + s1) = *(uint32 *) (rigged_paddr + r2);
+ r2 = *(reg_t *) (((char *) regs) + (insn.f1.s2 << 2));
+ s1 = insn.f1.s1 << 2;
+ goto *(base_addr + insn.f1.offset);
+ }
+
+ if (((target_saddr_t) tag < 0))
+ {
+ *(reg_t *) (((char *) regs) + s1) = *(uint32 *) f ();
+ r2 = *(reg_t *) (((char *) regs) + (insn.f1.s2 << 2));
+ s1 = insn.f1.s1 << 2;
+ goto *(base_addr + insn.f1.offset);
+ }
+
+ x = (x - 1) % 0x100;
+ }
+
+ L_METAOP_DONE:
+ return (*(reg_t *) (((char *) regs) + s1));
+ }
+}
+
+insn_t program[2 + 1];
+
+void *malloc ();
+
+int
+main ()
+{
+ environment_t env;
+ insn_t insn;
+ int i, res;
+ host_addr_t a_page = (host_addr_t) malloc (2 * 4096);
+ target_addr_t a_vaddr = 0x123450;
+ target_addr_t vaddr_page = a_vaddr / 4096;
+ a_page = (a_page + 4096 - 1) & -4096;
+
+ env.tlb_tab[((vaddr_page) % 0x100)].vaddr_tag = vaddr_page;
+ env.tlb_tab[((vaddr_page) % 0x100)].rigged_paddr = a_page -
vaddr_page * 4096;
+ insn.f1.offset = LOAD32_RR;
+ env.registers[0] = 0;
+ env.registers[2] = a_vaddr;
+ *(sint32 *) (a_page + a_vaddr % 4096) = 88;
+ insn.f1.s1 = 0;
+ insn.f1.s2 = 2;
+
+ for (i = 0; i < 2; i++)
+ program[i] = insn;
+
+ insn.f1.offset = METAOP_DONE;
+ insn.f1.s1 = 0;
+ program[2] = insn;
+
+ env.pc = program;
+ env.program = program;
+
+ res = simulator_kernel (2 + 1, &env);
+
+ if (res != 88)
+ abort ();
+ exit (0);
+}
+#else
+main(){ exit (0); }
+#endif
Index: testsuite/gcc.dg/tree-prof/pr52027.c
===================================================================
--- testsuite/gcc.dg/tree-prof/pr52027.c (revision 199014)
+++ testsuite/gcc.dg/tree-prof/pr52027.c (working copy)
@@ -1,6 +1,6 @@
/* PR debug/52027 */
/* { dg-require-effective-target freorder } */
-/* { dg-options "-O -freorder-blocks-and-partition -fno-reorder-functions" } */
+/* { dg-options "-O2 -freorder-blocks-and-partition
-fno-reorder-functions" } */
void
foo (int len)
Index: testsuite/gcc.dg/tree-prof/pr50907.c
===================================================================
--- testsuite/gcc.dg/tree-prof/pr50907.c (revision 199014)
+++ testsuite/gcc.dg/tree-prof/pr50907.c (working copy)
@@ -1,5 +1,5 @@
/* PR middle-end/50907 */
/* { dg-require-effective-target freorder } */
-/* { dg-options "-O -freorder-blocks-and-partition -fschedule-insns
-fselective-scheduling -fpic" { target { { powerpc*-*-* ia64-*-*
x86_64-*-* } && fpic } } } */
+/* { dg-options "-O2 -freorder-blocks-and-partition -fschedule-insns
-fselective-scheduling -fpic" { target { { powerpc*-*-* ia64-*-*
x86_64-*-* } && fpic } } } */
#include "pr45354.c"
Index: testsuite/gcc.dg/tree-prof/pr45354.c
===================================================================
--- testsuite/gcc.dg/tree-prof/pr45354.c (revision 199014)
+++ testsuite/gcc.dg/tree-prof/pr45354.c (working copy)
@@ -1,5 +1,5 @@
/* { dg-require-effective-target freorder } */
-/* { dg-options "-O -freorder-blocks-and-partition -fschedule-insns
-fselective-scheduling" { target powerpc*-*-* ia64-*-* x86_64-*-* } }
*/
+/* { dg-options "-O2 -freorder-blocks-and-partition -fschedule-insns
-fselective-scheduling" { target powerpc*-*-* ia64-*-* x86_64-*-* } }
*/
extern void abort (void);
Index: testsuite/gcc.dg/tree-prof/20041218-1.c
===================================================================
--- testsuite/gcc.dg/tree-prof/20041218-1.c (revision 0)
+++ testsuite/gcc.dg/tree-prof/20041218-1.c (revision 0)
@@ -0,0 +1,119 @@
+/* PR rtl-optimization/16968 */
+/* Testcase by Jakub Jelinek <jakub@redhat.com> */
+/* { dg-require-effective-target freorder } */
+/* { dg-options "-O2 -freorder-blocks-and-partition" } */
+
+struct T
+{
+ unsigned int b, c, *d;
+ unsigned char e;
+};
+struct S
+{
+ unsigned int a;
+ struct T f;
+};
+struct U
+{
+ struct S g, h;
+};
+struct V
+{
+ unsigned int i;
+ struct U j;
+};
+
+extern void exit (int);
+extern void abort (void);
+
+void *
+dummy1 (void *x)
+{
+ return "";
+}
+
+void *
+dummy2 (void *x, void *y)
+{
+ exit (0);
+}
+
+struct V *
+baz (unsigned int x)
+{
+ static struct V v;
+ __builtin_memset (&v, 0x55, sizeof (v));
+ return &v;
+}
+
+int
+check (void *x, struct S *y)
+{
+ if (y->a || y->f.b || y->f.c || y->f.d || y->f.e)
+ abort ();
+ return 1;
+}
+
+static struct V *
+bar (unsigned int x, void *y)
+{
+ const struct T t = { 0, 0, (void *) 0, 0 };
+ struct V *u;
+ void *v;
+ v = dummy1 (y);
+ if (!v)
+ return (void *) 0;
+
+ u = baz (sizeof (struct V));
+ u->i = x;
+ u->j.g.a = 0;
+ u->j.g.f = t;
+ u->j.h.a = 0;
+ u->j.h.f = t;
+
+ if (!check (v, &u->j.g) || !check (v, &u->j.h))
+ return (void *) 0;
+ return u;
+}
+
+int
+foo (unsigned int *x, unsigned int y, void **z)
+{
+ void *v;
+ unsigned int i, j;
+
+ *z = v = (void *) 0;
+
+ for (i = 0; i < y; i++)
+ {
+ struct V *c;
+
+ j = *x;
+
+ switch (j)
+ {
+ case 1:
+ c = bar (j, x);
+ break;
+ default:
+ c = 0;
+ break;
+ }
+ if (c)
+ v = dummy2 (v, c);
+ else
+ return 1;
+ }
+
+ *z = v;
+ return 0;
+}
+
+int
+main (void)
+{
+ unsigned int one = 1;
+ void *p;
+ foo (&one, 1, &p);
+ abort ();
+}
Index: testsuite/g++.dg/tree-prof/partition2.C
===================================================================
--- testsuite/g++.dg/tree-prof/partition2.C (revision 199014)
+++ testsuite/g++.dg/tree-prof/partition2.C (working copy)
@@ -1,6 +1,6 @@
// PR middle-end/45458
// { dg-require-effective-target freorder }
-// { dg-options "-fnon-call-exceptions -freorder-blocks-and-partition" }
+// { dg-options "-O2 -fnon-call-exceptions -freorder-blocks-and-partition" }
int
main ()
Index: testsuite/g++.dg/tree-prof/partition3.C
===================================================================
--- testsuite/g++.dg/tree-prof/partition3.C (revision 199014)
+++ testsuite/g++.dg/tree-prof/partition3.C (working copy)
@@ -1,6 +1,6 @@
// PR middle-end/45566
// { dg-require-effective-target freorder }
-// { dg-options "-O -fnon-call-exceptions -freorder-blocks-and-partition" }
+// { dg-options "-O2 -fnon-call-exceptions -freorder-blocks-and-partition" }
int k;
On Tue, May 21, 2013 at 10:17 PM, Teresa Johnson <tejohnson@google.com> wrote:
> On Mon, May 20, 2013 at 3:42 PM, Steven Bosscher <stevenb.gcc@gmail.com> wrote:
>> On Mon, May 20, 2013 at 4:55 AM, Teresa Johnson wrote:
>>
>>> * ifcvt.c (find_if_case_1): Replace BB_COPY_PARTITION with assert
>>> as this is now done by redirect_edge_and_branch_force.
>>> * function.c (thread_prologue_and_epilogue_insns): Insert new bb after
>>> barriers, and fix interaction with splitting.
>>> * emit-rtl.c (try_split): Copy REG_CROSSING_JUMP notes.
>>> * cfgcleanup.c (try_forward_edges): Fix early return value to properly
>>> reflect changes made in the routine.
>>> * bb-reorder.c (emit_barrier_after_bb): Handle insertion in
>>> non-cfglayout mode.
>>> (fix_up_fall_thru_edges): Remove incorrect check for bb layout order
>>> since this is called in cfglayout mode, and replace partition fixup
>>> with assert as that is now done by force_nonfallthru_and_redirect.
>>> (add_reg_crossing_jump_notes): Handle the fact that some jumps may
>>> already be marked with region crossing note.
>>> (insert_section_boundary_note): Make non-static, gate on flag
>>> has_bb_partition, rewrite to also check for multiple partitions.
>>> (rest_of_handle_reorder_blocks): Remove call to
>>> insert_section_boundary_note, now done later during free_cfg.
>>> * bb-reorder.h: Declare insert_section_boundary_note and
>>> emit_barrier_after_bb, which are no longer static.
>>> * Makefile.in (cfgrtl.o): Depend on bb-reorder.h
>>> * cfgrtl.c (rest_of_pass_free_cfg): If partitions exist
>>> invoke insert_section_boundary_note.
>>> (try_redirect_by_replacing_jump): Remove unnecessary
>>> check for region crossing note.
>>> (fixup_partition_crossing): New function.
>>> (rtl_redirect_edge_and_branch): Fixup partition boundaries.
>>> (force_nonfallthru_and_redirect): Fixup partition boundaries,
>>> remove old code that tried to do this. Emit barrier correctly
>>> when we are in cfglayout mode.
>>> (rtl_split_edge): Correctly fixup partition boundaries.
>>> (commit_one_edge_insertion): Remove old code that tried to
>>> fixup region crossing edge since this is now handled in
>>> split_block, and set up insertion point correctly since
>>> block may now end in a jump.
>>> (rtl_verify_edges): Add checks for incorrect/missing REG_CROSSING_JUMP
>>> notes.
>>> (fixup_reorder_chain): Remove old code that attempted to fixup region
>>> crossing note as this is now handled in force_nonfallthru_and_redirect.
>>> (duplicate_insn_chain): Don't duplicate switch section notes.
>>> (rtl_can_remove_branch_p): Remove unnecessary check for region crossing
>>> note.
>>>
>>> Index: ifcvt.c
>>> ===================================================================
>>> --- ifcvt.c (revision 199014)
>>> +++ ifcvt.c (working copy)
>>> @@ -3905,10 +3905,9 @@ find_if_case_1 (basic_block test_bb, edge then_edg
>>> if (new_bb)
>>> {
>>> df_bb_replace (then_bb_index, new_bb);
>>> - /* Since the fallthru edge was redirected from test_bb to new_bb,
>>> - we need to ensure that new_bb is in the same partition as
>>> - test bb (you can not fall through across section boundaries). */
>>> - BB_COPY_PARTITION (new_bb, test_bb);
>>> + /* This should have been done above via force_nonfallthru_and_redirect
>>> + (possibly called from redirect_edge_and_branch_force). */
>>> + gcc_assert (BB_PARTITION (new_bb) == BB_PARTITION (test_bb));
>>> }
>>
>> Nit: gcc_checking_assert() ?
>
> ok.
>
>>
>>
>>> Index: function.c
>>> ===================================================================
>>> --- function.c (revision 199014)
>>> +++ function.c (working copy)
>>> @@ -6270,8 +6270,10 @@ thread_prologue_and_epilogue_insns (void)
>>> break;
>>> if (e)
>>> {
>>> - copy_bb = create_basic_block (NEXT_INSN (BB_END (e->src)),
>>> - NULL_RTX, e->src);
>>> + /* Make sure we insert after any barriers. */
>>> + rtx end = get_last_bb_insn (e->src);
>>> + copy_bb = create_basic_block (NEXT_INSN (end),
>>> + NULL_RTX, e->src);
>>> BB_COPY_PARTITION (copy_bb, e->src);
>>> }
>>> else
>>
>> This is a nice catch. The change at first glance looks unrelated to
>> the profiling fixes, but apparently you have a test case where e is
>> not an EDGE_FALLTHRU edge? Otherwise there can't be a barrier. The
>> change looks OK to me.
>
> Yes, according to my notes this fix was from a case in hmmer where the
> edge was a crossing edge.
>
>>
>>
>>> Index: bb-reorder.c
>>> ===================================================================
>>> --- bb-reorder.c (revision 199014)
>>> +++ bb-reorder.c (working copy)
>>> @@ -1380,13 +1380,16 @@ get_uncond_jump_length (void)
>>> return length;
>>> }
>>>
>>> -/* Emit a barrier into the footer of BB. */
>>> +/* Emit a barrier after BB, into the footer if we are in CFGLAYOUT mode. */
>>>
>>> -static void
>>> +void
>>> emit_barrier_after_bb (basic_block bb)
>>> {
>>
>> I still think this function should just be moved out of bb-reorder.c
>> then. Move it to cfgrtl,c please, and prototype it in basic-block.h
>> (one day we'll have to split basic-block.h into rtl/gimple/generic
>> parts, TODO++).
>>
>> But to be honest, I still don't really understand why we emit a
>> barrier at all if we're in cfglayout mode. They're ignored, they're
>> going to be overlooked if someone looks for barriers after basic
>> blocks (nobody looks in the footer, and they should not), and
>> fixup_reorder_chain ought to put barriers where needed when coming out
>> of cfglayout mode. (TODO++ - that's how it goes for me every time I
>> look at gcc code ;-)
>
> Ok, I've moved it. Interestingly, previous to my modification to
> enable this to be called in non-cfglayout mode it was only called from
> bbpart, which is in cfglayout mode. This may be another case where
> bbpart is confused about what mode it is in? Probably worth some
> investigation/cleanup at some point as you note.
>
>>
>>
>>> {
>>> - emit_note_before (NOTE_INSN_SWITCH_TEXT_SECTIONS, BB_HEAD (bb));
>>> - break;
>>> + if (switched_sections)
>>> + {
>>> + error ("multiple hot/cold transitions found (bb %i)",
>>> + bb->index);
>>> + err = 1;
>>
>> Just gcc_assert (! switched_sections).
>
> The rest of the rtl flow verification code used error () and returned
> an err=1, so I was trying to be consistent with that. Do we want to be
> different in this routine?
>
>>
>>
>>> @@ -2180,8 +2199,6 @@ rest_of_handle_reorder_blocks (void)
>>> bb->aux = bb->next_bb;
>>> cfg_layout_finalize ();
>>>
>>> - /* Add NOTE_INSN_SWITCH_TEXT_SECTIONS notes. */
>>> - insert_section_boundary_note ();
>>> return 0;
>>> }
>>
>> ...and the titans were defeated.
>>
>>
>>> + /* Remove the region crossing note from jump at end of
>>
>> s/region/section/
>>
>> Looks good to me, overall. The only thing missing is test cases, if
>> you have any.
>
> I manually ran all of the gcc.c-torture/execute tests to collect and
> feedback fdo, and enable -freorder-blocks-and-partitions, with and
> without my fixes. This gave me a few test cases that I will add as
> fdo/partitioning test cases. A couple were still problems that I
> hadn't flushed out yet, relating to the new verification code that
> checks that we only have a single section switch in the function,
> because a couple post-bbro phases (e.g. compgoto) were not being
> careful enough.
>
> Revised patch under testing and coming tomorrow.
>
>>
>> Can you file a bug report for the debug info issues?
>
> Will do.
>
> Thanks,
> Teresa
>
>>
>> Ciao!
>> Steven
>
>
>
> --
> Teresa Johnson | Software Engineer | tejohnson@google.com | 408-460-2413
--
Teresa Johnson | Software Engineer | tejohnson@google.com | 408-460-2413