git
/
gcc
mirror of git://gcc.gnu.org/git/gcc.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

function.c (bb_active_p): Delete. 

* function.c (bb_active_p): Delete.
	(dup_block_and_redirect, active_insn_between): New functions.
	(convert_jumps_to_returns, emit_return_for_exit): New functions,
	split out from..
	(thread_prologue_and_epilogue_insns): ..here.  Delete
	shadowing variables.  Don't do prologue register clobber tests
	when shrink wrapping already failed.  Delete all last_bb_active
	code.  Instead compute tail block candidates for duplicating
	exit path.  Remove these from antic set.  Duplicate tails when
	reached from both blocks needing a prologue/epilogue and
	blocks not needing such.
	* ifcvt.c (dead_or_predicable): Test both flag_shrink_wrap and
	HAVE_simple_return.
	* bb-reorder.c (get_uncond_jump_length): Make global.
	* bb-reorder.h (get_uncond_jump_length): Declare.
	* cfgrtl.c (rtl_create_basic_block): Comment typo fix.
	(rtl_split_edge): Likewise.  Warning fix.
	(rtl_duplicate_bb): New function.
	(rtl_cfg_hooks): Enable can_duplicate_block_p and duplicate_block.
	* Makefile.in (function.o): Update dependencies.

From-SVN: r181188
This commit is contained in:
Alan Modra 2011-11-09 11:50:14 +10:30 committed by Alan Modra
parent dce5eb107d
commit ffe1468659
7 changed files with 518 additions and 250 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
gcc/ChangeLog
View File

@ -1,3 +1,26 @@
2011-11-09 Alan Modra <amodra@gmail.com>
* function.c (bb_active_p): Delete.
(dup_block_and_redirect, active_insn_between): New functions.
(convert_jumps_to_returns, emit_return_for_exit): New functions,
split out from..
(thread_prologue_and_epilogue_insns): ..here. Delete
shadowing variables. Don't do prologue register clobber tests
when shrink wrapping already failed. Delete all last_bb_active
code. Instead compute tail block candidates for duplicating
exit path. Remove these from antic set. Duplicate tails when
reached from both blocks needing a prologue/epilogue and
blocks not needing such.
* ifcvt.c (dead_or_predicable): Test both flag_shrink_wrap and
HAVE_simple_return.
* bb-reorder.c (get_uncond_jump_length): Make global.
* bb-reorder.h (get_uncond_jump_length): Declare.
* cfgrtl.c (rtl_create_basic_block): Comment typo fix.
(rtl_split_edge): Likewise. Warning fix.
(rtl_duplicate_bb): New function.
(rtl_cfg_hooks): Enable can_duplicate_block_p and duplicate_block.
* Makefile.in (function.o): Update dependencies.
2011-11-08 Ulrich Weigand <Ulrich.Weigand@de.ibm.com> 2011-11-08 Ulrich Weigand <Ulrich.Weigand@de.ibm.com>
Georg-Johann Lay <avr@gjlay.de> Georg-Johann Lay <avr@gjlay.de>

3
gcc/Makefile.in
View File

@ -2807,7 +2807,8 @@ function.o : function.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) $(RTL_ERROR_
$(OPTABS_H) $(LIBFUNCS_H) $(REGS_H) hard-reg-set.h insn-config.h $(RECOG_H) \ $(OPTABS_H) $(LIBFUNCS_H) $(REGS_H) hard-reg-set.h insn-config.h $(RECOG_H) \
output.h $(EXCEPT_H) $(HASHTAB_H) $(GGC_H) $(TM_P_H) langhooks.h \ output.h $(EXCEPT_H) $(HASHTAB_H) $(GGC_H) $(TM_P_H) langhooks.h \
gt-function.h $(TARGET_H) $(BASIC_BLOCK_H) $(INTEGRATE_H) $(PREDICT_H) \ gt-function.h $(TARGET_H) $(BASIC_BLOCK_H) $(INTEGRATE_H) $(PREDICT_H) \
$(TREE_PASS_H) $(DF_H) $(TIMEVAR_H) vecprim.h $(COMMON_TARGET_H) $(TREE_PASS_H) $(DF_H) $(TIMEVAR_H) vecprim.h $(PARAMS_H) bb-reorder.h \
$(COMMON_TARGET_H)
statistics.o : statistics.c $(CONFIG_H) $(SYSTEM_H) coretypes.h \ statistics.o : statistics.c $(CONFIG_H) $(SYSTEM_H) coretypes.h \
$(TREE_PASS_H) $(TREE_DUMP_H) $(HASHTAB_H) statistics.h $(FUNCTION_H) $(TREE_PASS_H) $(TREE_DUMP_H) $(HASHTAB_H) statistics.h $(FUNCTION_H)
stmt.o : stmt.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) $(RTL_H) \ stmt.o : stmt.c $(CONFIG_H) $(SYSTEM_H) coretypes.h $(TM_H) $(RTL_H) \

3
gcc/bb-reorder.c
View File

@ -181,7 +181,6 @@ static fibheapkey_t bb_to_key (basic_block);
static bool better_edge_p (const_basic_block, const_edge, int, int, int, int, const_edge); static bool better_edge_p (const_basic_block, const_edge, int, int, int, int, const_edge);
static void connect_traces (int, struct trace *); static void connect_traces (int, struct trace *);
static bool copy_bb_p (const_basic_block, int); static bool copy_bb_p (const_basic_block, int);
static int get_uncond_jump_length (void);
static bool push_to_next_round_p (const_basic_block, int, int, int, gcov_type); static bool push_to_next_round_p (const_basic_block, int, int, int, gcov_type);
/* Check to see if bb should be pushed into the next round of trace /* Check to see if bb should be pushed into the next round of trace
@ -1193,7 +1192,7 @@ copy_bb_p (const_basic_block bb, int code_may_grow)
/* Return the length of unconditional jump instruction. */ /* Return the length of unconditional jump instruction. */
static int int
get_uncond_jump_length (void) get_uncond_jump_length (void)
{ {
rtx label, jump; rtx label, jump;

2
gcc/bb-reorder.h
View File

@ -34,4 +34,6 @@ extern struct target_bb_reorder *this_target_bb_reorder;
#define this_target_bb_reorder (&default_target_bb_reorder) #define this_target_bb_reorder (&default_target_bb_reorder)
#endif #endif
extern int get_uncond_jump_length (void);
#endif #endif

36
gcc/cfgrtl.c
View File

@ -322,9 +322,9 @@ create_basic_block_structure (rtx head, rtx end, rtx bb_note, basic_block after)
} }
/* Create new basic block consisting of instructions in between HEAD and END /* Create new basic block consisting of instructions in between HEAD and END
and place it to the BB chain after block AFTER. END can be NULL in to and place it to the BB chain after block AFTER. END can be NULL to
create new empty basic block before HEAD. Both END and HEAD can be NULL to create a new empty basic block before HEAD. Both END and HEAD can be
create basic block at the end of INSN chain. */ NULL to create basic block at the end of INSN chain. */
static basic_block static basic_block
rtl_create_basic_block (void *headp, void *endp, basic_block after) rtl_create_basic_block (void *headp, void *endp, basic_block after)
@ -1411,8 +1411,8 @@ rtl_split_edge (edge edge_in)
before = NULL_RTX; before = NULL_RTX;
/* If this is a fall through edge to the exit block, the blocks might be /* If this is a fall through edge to the exit block, the blocks might be
not adjacent, and the right place is the after the source. */ not adjacent, and the right place is after the source. */
if (edge_in->flags & EDGE_FALLTHRU && edge_in->dest == EXIT_BLOCK_PTR) if ((edge_in->flags & EDGE_FALLTHRU) && edge_in->dest == EXIT_BLOCK_PTR)
{ {
before = NEXT_INSN (BB_END (edge_in->src)); before = NEXT_INSN (BB_END (edge_in->src));
bb = create_basic_block (before, NULL, edge_in->src); bb = create_basic_block (before, NULL, edge_in->src);
@ -3177,6 +3177,21 @@ rtl_can_remove_branch_p (const_edge e)
return true; return true;
} }
/* We do not want to declare these functions in a header file, since they
should only be used through the cfghooks interface, and we do not want to
move them here since it would require also moving quite a lot of related
code. They are in cfglayout.c. */
extern bool cfg_layout_can_duplicate_bb_p (const_basic_block);
extern basic_block cfg_layout_duplicate_bb (basic_block);
static basic_block
rtl_duplicate_bb (basic_block bb)
{
bb = cfg_layout_duplicate_bb (bb);
bb->aux = NULL;
return bb;
}
/* Implementation of CFG manipulation for linearized RTL. */ /* Implementation of CFG manipulation for linearized RTL. */
struct cfg_hooks rtl_cfg_hooks = { struct cfg_hooks rtl_cfg_hooks = {
"rtl", "rtl",
@ -3193,8 +3208,8 @@ struct cfg_hooks rtl_cfg_hooks = {
rtl_merge_blocks, rtl_merge_blocks,
rtl_predict_edge, rtl_predict_edge,
rtl_predicted_by_p, rtl_predicted_by_p,
NULL, /* can_duplicate_block_p */ cfg_layout_can_duplicate_bb_p,
NULL, /* duplicate_block */ rtl_duplicate_bb,
rtl_split_edge, rtl_split_edge,
rtl_make_forwarder_block, rtl_make_forwarder_block,
rtl_tidy_fallthru_edge, rtl_tidy_fallthru_edge,
@ -3216,13 +3231,6 @@ struct cfg_hooks rtl_cfg_hooks = {
This representation will hopefully become the default one in future This representation will hopefully become the default one in future
version of the compiler. */ version of the compiler. */
/* We do not want to declare these functions in a header file, since they
should only be used through the cfghooks interface, and we do not want to
move them here since it would require also moving quite a lot of related
code. They are in cfglayout.c. */
extern bool cfg_layout_can_duplicate_bb_p (const_basic_block);
extern basic_block cfg_layout_duplicate_bb (basic_block);
struct cfg_hooks cfg_layout_rtl_cfg_hooks = { struct cfg_hooks cfg_layout_rtl_cfg_hooks = {
"cfglayout mode", "cfglayout mode",
rtl_verify_flow_info_1, rtl_verify_flow_info_1,

697
gcc/function.c
View File

@ -65,6 +65,8 @@ along with GCC; see the file COPYING3. If not see
#include "df.h" #include "df.h"
#include "timevar.h" #include "timevar.h"
#include "vecprim.h" #include "vecprim.h"
#include "params.h"
#include "bb-reorder.h"
/* So we can assign to cfun in this file. */ /* So we can assign to cfun in this file. */
#undef cfun #undef cfun
@ -5290,8 +5292,6 @@ requires_stack_frame_p (rtx insn, HARD_REG_SET prologue_used,
HARD_REG_SET hardregs; HARD_REG_SET hardregs;
unsigned regno; unsigned regno;
if (!INSN_P (insn) || DEBUG_INSN_P (insn))
return false;
if (CALL_P (insn)) if (CALL_P (insn))
return !SIBLING_CALL_P (insn); return !SIBLING_CALL_P (insn);
@ -5514,22 +5514,185 @@ set_return_jump_label (rtx returnjump)
JUMP_LABEL (returnjump) = ret_rtx; JUMP_LABEL (returnjump) = ret_rtx;
} }
/* Return true if BB has any active insns. */ #ifdef HAVE_simple_return
static bool /* Create a copy of BB instructions and insert at BEFORE. Redirect
bb_active_p (basic_block bb) preds of BB to COPY_BB if they don't appear in NEED_PROLOGUE. */
static void
dup_block_and_redirect (basic_block bb, basic_block copy_bb, rtx before,
bitmap_head *need_prologue)
{ {
rtx label; edge_iterator ei;
edge e;
rtx insn = BB_END (bb);
/* Test whether there are active instructions in BB. */ /* We know BB has a single successor, so there is no need to copy a
label = BB_END (bb); simple jump at the end of BB. */
while (label && !LABEL_P (label)) if (simplejump_p (insn))
insn = PREV_INSN (insn);
start_sequence ();
duplicate_insn_chain (BB_HEAD (bb), insn);
if (dump_file)
{ {
if (active_insn_p (label)) unsigned count = 0;
break; for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
label = PREV_INSN (label); if (active_insn_p (insn))
++count;
fprintf (dump_file, "Duplicating bb %d to bb %d, %u active insns.\n",
bb->index, copy_bb->index, count);
} }
return BB_HEAD (bb) != label || !LABEL_P (label); insn = get_insns ();
end_sequence ();
emit_insn_before (insn, before);
/* Redirect all the paths that need no prologue into copy_bb. */
for (ei = ei_start (bb->preds); (e = ei_safe_edge (ei)); )
if (!bitmap_bit_p (need_prologue, e->src->index))
{
redirect_edge_and_branch_force (e, copy_bb);
continue;
}
else
ei_next (&ei);
} }
#endif
#if defined (HAVE_return) || defined (HAVE_simple_return)
/* Return true if there are any active insns between HEAD and TAIL. */
static bool
active_insn_between (rtx head, rtx tail)
{
while (tail)
{
if (active_insn_p (tail))
return true;
if (tail == head)
return false;
tail = PREV_INSN (tail);
}
return false;
}
/* LAST_BB is a block that exits, and empty of active instructions.
Examine its predecessors for jumps that can be converted to
(conditional) returns. */
static VEC (edge, heap) *
convert_jumps_to_returns (basic_block last_bb, bool simple_p,
VEC (edge, heap) *unconverted ATTRIBUTE_UNUSED)
{
int i;
basic_block bb;
rtx label;
edge_iterator ei;
edge e;
VEC(basic_block,heap) *src_bbs;
src_bbs = VEC_alloc (basic_block, heap, EDGE_COUNT (last_bb->preds));
FOR_EACH_EDGE (e, ei, last_bb->preds)
if (e->src != ENTRY_BLOCK_PTR)
VEC_quick_push (basic_block, src_bbs, e->src);
label = BB_HEAD (last_bb);
FOR_EACH_VEC_ELT (basic_block, src_bbs, i, bb)
{
rtx jump = BB_END (bb);
if (!JUMP_P (jump) || JUMP_LABEL (jump) != label)
continue;
e = find_edge (bb, last_bb);
/* If we have an unconditional jump, we can replace that
with a simple return instruction. */
if (simplejump_p (jump))
{
/* The use of the return register might be present in the exit
fallthru block. Either:
- removing the use is safe, and we should remove the use in
the exit fallthru block, or
- removing the use is not safe, and we should add it here.
For now, we conservatively choose the latter. Either of the
2 helps in crossjumping. */
emit_use_return_register_into_block (bb);
emit_return_into_block (simple_p, bb);
delete_insn (jump);
}
/* If we have a conditional jump branching to the last
block, we can try to replace that with a conditional
return instruction. */
else if (condjump_p (jump))
{
rtx dest;
if (simple_p)
dest = simple_return_rtx;
else
dest = ret_rtx;
if (!redirect_jump (jump, dest, 0))
{
#ifdef HAVE_simple_return
if (simple_p)
{
if (dump_file)
fprintf (dump_file,
"Failed to redirect bb %d branch.\n", bb->index);
VEC_safe_push (edge, heap, unconverted, e);
}
#endif
continue;
}
/* See comment in simplejump_p case above. */
emit_use_return_register_into_block (bb);
/* If this block has only one successor, it both jumps
and falls through to the fallthru block, so we can't
delete the edge. */
if (single_succ_p (bb))
continue;
}
else
{
#ifdef HAVE_simple_return
if (simple_p)
{
if (dump_file)
fprintf (dump_file,
"Failed to redirect bb %d branch.\n", bb->index);
VEC_safe_push (edge, heap, unconverted, e);
}
#endif
continue;
}
/* Fix up the CFG for the successful change we just made. */
redirect_edge_succ (e, EXIT_BLOCK_PTR);
}
VEC_free (basic_block, heap, src_bbs);
return unconverted;
}
/* Emit a return insn for the exit fallthru block. */
static basic_block
emit_return_for_exit (edge exit_fallthru_edge, bool simple_p)
{
basic_block last_bb = exit_fallthru_edge->src;
if (JUMP_P (BB_END (last_bb)))
{
last_bb = split_edge (exit_fallthru_edge);
exit_fallthru_edge = single_succ_edge (last_bb);
}
emit_barrier_after (BB_END (last_bb));
emit_return_into_block (simple_p, last_bb);
exit_fallthru_edge->flags &= ~EDGE_FALLTHRU;
return last_bb;
}
#endif
/* Generate the prologue and epilogue RTL if the machine supports it. Thread /* Generate the prologue and epilogue RTL if the machine supports it. Thread
this into place with notes indicating where the prologue ends and where this into place with notes indicating where the prologue ends and where
@ -5583,20 +5746,17 @@ static void
thread_prologue_and_epilogue_insns (void) thread_prologue_and_epilogue_insns (void)
{ {
bool inserted; bool inserted;
basic_block last_bb;
bool last_bb_active ATTRIBUTE_UNUSED;
#ifdef HAVE_simple_return #ifdef HAVE_simple_return
VEC (rtx, heap) *unconverted_simple_returns = NULL; VEC (edge, heap) *unconverted_simple_returns = NULL;
basic_block simple_return_block_hot = NULL;
basic_block simple_return_block_cold = NULL;
bool nonempty_prologue; bool nonempty_prologue;
bitmap_head bb_flags;
unsigned max_grow_size;
#endif #endif
rtx returnjump ATTRIBUTE_UNUSED; rtx returnjump;
rtx seq ATTRIBUTE_UNUSED, epilogue_end ATTRIBUTE_UNUSED; rtx seq ATTRIBUTE_UNUSED, epilogue_end ATTRIBUTE_UNUSED;
rtx prologue_seq ATTRIBUTE_UNUSED, split_prologue_seq ATTRIBUTE_UNUSED; rtx prologue_seq ATTRIBUTE_UNUSED, split_prologue_seq ATTRIBUTE_UNUSED;
edge e, entry_edge, orig_entry_edge, exit_fallthru_edge; edge e, entry_edge, orig_entry_edge, exit_fallthru_edge;
edge_iterator ei; edge_iterator ei;
bitmap_head bb_flags;
df_analyze (); df_analyze ();
@ -5614,18 +5774,6 @@ thread_prologue_and_epilogue_insns (void)
entry_edge = single_succ_edge (ENTRY_BLOCK_PTR); entry_edge = single_succ_edge (ENTRY_BLOCK_PTR);
orig_entry_edge = entry_edge; orig_entry_edge = entry_edge;
exit_fallthru_edge = find_fallthru_edge (EXIT_BLOCK_PTR->preds);
if (exit_fallthru_edge != NULL)
{
last_bb = exit_fallthru_edge->src;
last_bb_active = bb_active_p (last_bb);
}
else
{
last_bb = NULL;
last_bb_active = false;
}
split_prologue_seq = NULL_RTX; split_prologue_seq = NULL_RTX;
if (flag_split_stack if (flag_split_stack
&& (lookup_attribute ("no_split_stack", DECL_ATTRIBUTES (cfun->decl)) && (lookup_attribute ("no_split_stack", DECL_ATTRIBUTES (cfun->decl))
@ -5675,9 +5823,9 @@ thread_prologue_and_epilogue_insns (void)
} }
#endif #endif
#ifdef HAVE_simple_return
bitmap_initialize (&bb_flags, &bitmap_default_obstack); bitmap_initialize (&bb_flags, &bitmap_default_obstack);
#ifdef HAVE_simple_return
/* Try to perform a kind of shrink-wrapping, making sure the /* Try to perform a kind of shrink-wrapping, making sure the
prologue/epilogue is emitted only around those parts of the prologue/epilogue is emitted only around those parts of the
function that require it. */ function that require it. */
@ -5697,11 +5845,11 @@ thread_prologue_and_epilogue_insns (void)
HARD_REG_SET prologue_clobbered, prologue_used, live_on_edge; HARD_REG_SET prologue_clobbered, prologue_used, live_on_edge;
HARD_REG_SET set_up_by_prologue; HARD_REG_SET set_up_by_prologue;
rtx p_insn; rtx p_insn;
VEC(basic_block, heap) *vec; VEC(basic_block, heap) *vec;
basic_block bb; basic_block bb;
bitmap_head bb_antic_flags; bitmap_head bb_antic_flags;
bitmap_head bb_on_list; bitmap_head bb_on_list;
bitmap_head bb_tail;
if (dump_file) if (dump_file)
fprintf (dump_file, "Attempting shrink-wrapping optimization.\n"); fprintf (dump_file, "Attempting shrink-wrapping optimization.\n");
@ -5726,14 +5874,12 @@ thread_prologue_and_epilogue_insns (void)
prepare_shrink_wrap (entry_edge->dest); prepare_shrink_wrap (entry_edge->dest);
/* That may have inserted instructions into the last block. */
if (last_bb && !last_bb_active)
last_bb_active = bb_active_p (last_bb);
bitmap_initialize (&bb_antic_flags, &bitmap_default_obstack); bitmap_initialize (&bb_antic_flags, &bitmap_default_obstack);
bitmap_initialize (&bb_on_list, &bitmap_default_obstack); bitmap_initialize (&bb_on_list, &bitmap_default_obstack);
bitmap_initialize (&bb_tail, &bitmap_default_obstack);
/* Find the set of basic blocks that require a stack frame. */ /* Find the set of basic blocks that require a stack frame,
and blocks that are too big to be duplicated. */
vec = VEC_alloc (basic_block, heap, n_basic_blocks); vec = VEC_alloc (basic_block, heap, n_basic_blocks);
@ -5747,66 +5893,80 @@ thread_prologue_and_epilogue_insns (void)
add_to_hard_reg_set (&set_up_by_prologue, Pmode, add_to_hard_reg_set (&set_up_by_prologue, Pmode,
PIC_OFFSET_TABLE_REGNUM); PIC_OFFSET_TABLE_REGNUM);
/* We don't use a different max size depending on
optimize_bb_for_speed_p because increasing shrink-wrapping
opportunities by duplicating tail blocks can actually result
in an overall decrease in code size. */
max_grow_size = get_uncond_jump_length ();
max_grow_size *= PARAM_VALUE (PARAM_MAX_GROW_COPY_BB_INSNS);
FOR_EACH_BB (bb) FOR_EACH_BB (bb)
{ {
rtx insn; rtx insn;
/* As a special case, check for jumps to the last bb that unsigned size = 0;
cannot successfully be converted to simple_returns later
on, and mark them as requiring a frame. These are FOR_BB_INSNS (bb, insn)
conditional jumps that jump to their fallthru block, so if (NONDEBUG_INSN_P (insn))
it's not a case that is expected to occur often. */ {
if (JUMP_P (BB_END (bb)) && any_condjump_p (BB_END (bb)) if (requires_stack_frame_p (insn, prologue_used,
&& single_succ_p (bb) set_up_by_prologue))
&& !last_bb_active {
&& single_succ (bb) == last_bb) if (bb == entry_edge->dest)
{ goto fail_shrinkwrap;
bitmap_set_bit (&bb_flags, bb->index); bitmap_set_bit (&bb_flags, bb->index);
VEC_quick_push (basic_block, vec, bb); VEC_quick_push (basic_block, vec, bb);
} break;
else }
FOR_BB_INSNS (bb, insn) else if (size <= max_grow_size)
if (requires_stack_frame_p (insn, prologue_used, {
set_up_by_prologue)) size += get_attr_min_length (insn);
{ if (size > max_grow_size)
bitmap_set_bit (&bb_flags, bb->index); bitmap_set_bit (&bb_on_list, bb->index);
VEC_quick_push (basic_block, vec, bb); }
break; }
}
} }
/* Blocks that really need a prologue, or are too big for tails. */
bitmap_ior_into (&bb_on_list, &bb_flags);
/* For every basic block that needs a prologue, mark all blocks /* For every basic block that needs a prologue, mark all blocks
reachable from it, so as to ensure they are also seen as reachable from it, so as to ensure they are also seen as
requiring a prologue. */ requiring a prologue. */
while (!VEC_empty (basic_block, vec)) while (!VEC_empty (basic_block, vec))
{ {
basic_block tmp_bb = VEC_pop (basic_block, vec); basic_block tmp_bb = VEC_pop (basic_block, vec);
edge e;
edge_iterator ei;
FOR_EACH_EDGE (e, ei, tmp_bb->succs) FOR_EACH_EDGE (e, ei, tmp_bb->succs)
if (e->dest != EXIT_BLOCK_PTR if (e->dest != EXIT_BLOCK_PTR
&& bitmap_set_bit (&bb_flags, e->dest->index)) && bitmap_set_bit (&bb_flags, e->dest->index))
VEC_quick_push (basic_block, vec, e->dest); VEC_quick_push (basic_block, vec, e->dest);
} }
/* If the last basic block contains only a label, we'll be able
to convert jumps to it to (potentially conditional) return /* Find the set of basic blocks that need no prologue, have a
insns later. This means we don't necessarily need a prologue single successor, can be duplicated, meet a max size
for paths reaching it. */ requirement, and go to the exit via like blocks. */
if (last_bb && optimize) VEC_quick_push (basic_block, vec, EXIT_BLOCK_PTR);
while (!VEC_empty (basic_block, vec))
{ {
if (!last_bb_active) basic_block tmp_bb = VEC_pop (basic_block, vec);
bitmap_clear_bit (&bb_flags, last_bb->index);
else if (!bitmap_bit_p (&bb_flags, last_bb->index)) FOR_EACH_EDGE (e, ei, tmp_bb->preds)
goto fail_shrinkwrap; if (single_succ_p (e->src)
&& !bitmap_bit_p (&bb_on_list, e->src->index)
&& can_duplicate_block_p (e->src)
&& bitmap_set_bit (&bb_tail, e->src->index))
VEC_quick_push (basic_block, vec, e->src);
} }
/* Now walk backwards from every block that is marked as needing /* Now walk backwards from every block that is marked as needing
a prologue to compute the bb_antic_flags bitmap. */ a prologue to compute the bb_antic_flags bitmap. Exclude
bitmap_copy (&bb_antic_flags, &bb_flags); tail blocks; They can be duplicated to be used on paths not
needing a prologue. */
bitmap_clear (&bb_on_list);
bitmap_and_compl (&bb_antic_flags, &bb_flags, &bb_tail);
FOR_EACH_BB (bb) FOR_EACH_BB (bb)
{ {
edge e; if (!bitmap_bit_p (&bb_antic_flags, bb->index))
edge_iterator ei;
if (!bitmap_bit_p (&bb_flags, bb->index))
continue; continue;
FOR_EACH_EDGE (e, ei, bb->preds) FOR_EACH_EDGE (e, ei, bb->preds)
if (!bitmap_bit_p (&bb_antic_flags, e->src->index) if (!bitmap_bit_p (&bb_antic_flags, e->src->index)
@ -5816,8 +5976,6 @@ thread_prologue_and_epilogue_insns (void)
while (!VEC_empty (basic_block, vec)) while (!VEC_empty (basic_block, vec))
{ {
basic_block tmp_bb = VEC_pop (basic_block, vec); basic_block tmp_bb = VEC_pop (basic_block, vec);
edge e;
edge_iterator ei;
bool all_set = true; bool all_set = true;
bitmap_clear_bit (&bb_on_list, tmp_bb->index); bitmap_clear_bit (&bb_on_list, tmp_bb->index);
@ -5862,28 +6020,134 @@ thread_prologue_and_epilogue_insns (void)
} }
} }
/* Test whether the prologue is known to clobber any register if (entry_edge != orig_entry_edge)
(other than FP or SP) which are live on the edge. */
CLEAR_HARD_REG_BIT (prologue_clobbered, STACK_POINTER_REGNUM);
if (frame_pointer_needed)
CLEAR_HARD_REG_BIT (prologue_clobbered, HARD_FRAME_POINTER_REGNUM);
CLEAR_HARD_REG_SET (live_on_edge);
reg_set_to_hard_reg_set (&live_on_edge,
df_get_live_in (entry_edge->dest));
if (hard_reg_set_intersect_p (live_on_edge, prologue_clobbered))
{ {
entry_edge = orig_entry_edge; /* Test whether the prologue is known to clobber any register
if (dump_file) (other than FP or SP) which are live on the edge. */
fprintf (dump_file, "Shrink-wrapping aborted due to clobber.\n"); CLEAR_HARD_REG_BIT (prologue_clobbered, STACK_POINTER_REGNUM);
if (frame_pointer_needed)
CLEAR_HARD_REG_BIT (prologue_clobbered, HARD_FRAME_POINTER_REGNUM);
CLEAR_HARD_REG_SET (live_on_edge);
reg_set_to_hard_reg_set (&live_on_edge,
df_get_live_in (entry_edge->dest));
if (hard_reg_set_intersect_p (live_on_edge, prologue_clobbered))
{
entry_edge = orig_entry_edge;
if (dump_file)
fprintf (dump_file,
"Shrink-wrapping aborted due to clobber.\n");
}
} }
else if (entry_edge != orig_entry_edge) if (entry_edge != orig_entry_edge)
{ {
crtl->shrink_wrapped = true; crtl->shrink_wrapped = true;
if (dump_file) if (dump_file)
fprintf (dump_file, "Performing shrink-wrapping.\n"); fprintf (dump_file, "Performing shrink-wrapping.\n");
/* Find tail blocks reachable from both blocks needing a
prologue and blocks not needing a prologue. */
if (!bitmap_empty_p (&bb_tail))
FOR_EACH_BB (bb)
{
bool some_pro, some_no_pro;
if (!bitmap_bit_p (&bb_tail, bb->index))
continue;
some_pro = some_no_pro = false;
FOR_EACH_EDGE (e, ei, bb->preds)
{
if (bitmap_bit_p (&bb_flags, e->src->index))
some_pro = true;
else
some_no_pro = true;
}
if (some_pro && some_no_pro)
VEC_quick_push (basic_block, vec, bb);
else
bitmap_clear_bit (&bb_tail, bb->index);
}
/* Find the head of each tail. */
while (!VEC_empty (basic_block, vec))
{
basic_block tbb = VEC_pop (basic_block, vec);
if (!bitmap_bit_p (&bb_tail, tbb->index))
continue;
while (single_succ_p (tbb))
{
tbb = single_succ (tbb);
bitmap_clear_bit (&bb_tail, tbb->index);
}
}
/* Now duplicate the tails. */
if (!bitmap_empty_p (&bb_tail))
FOR_EACH_BB_REVERSE (bb)
{
basic_block copy_bb, tbb;
rtx insert_point;
int eflags;
if (!bitmap_clear_bit (&bb_tail, bb->index))
continue;
/* Create a copy of BB, instructions and all, for
use on paths that don't need a prologue.
Ideal placement of the copy is on a fall-thru edge
or after a block that would jump to the copy. */
FOR_EACH_EDGE (e, ei, bb->preds)
if (!bitmap_bit_p (&bb_flags, e->src->index)
&& single_succ_p (e->src))
break;
if (e)
{
copy_bb = create_basic_block (NEXT_INSN (BB_END (e->src)),
NULL_RTX, e->src);
BB_COPY_PARTITION (copy_bb, e->src);
}
else
{
/* Otherwise put the copy at the end of the function. */
copy_bb = create_basic_block (NULL_RTX, NULL_RTX,
EXIT_BLOCK_PTR->prev_bb);
BB_COPY_PARTITION (copy_bb, bb);
}
insert_point = emit_note_after (NOTE_INSN_DELETED,
BB_END (copy_bb));
emit_barrier_after (BB_END (copy_bb));
tbb = bb;
while (1)
{
dup_block_and_redirect (tbb, copy_bb, insert_point,
&bb_flags);
tbb = single_succ (tbb);
if (tbb == EXIT_BLOCK_PTR)
break;
e = split_block (copy_bb, PREV_INSN (insert_point));
copy_bb = e->dest;
}
/* Quiet verify_flow_info by (ab)using EDGE_FAKE.
We have yet to add a simple_return to the tails,
as we'd like to first convert_jumps_to_returns in
case the block is no longer used after that. */
eflags = EDGE_FAKE;
if (CALL_P (PREV_INSN (insert_point))
&& SIBLING_CALL_P (PREV_INSN (insert_point)))
eflags = EDGE_SIBCALL | EDGE_ABNORMAL;
make_single_succ_edge (copy_bb, EXIT_BLOCK_PTR, eflags);
/* verify_flow_info doesn't like a note after a
sibling call. */
delete_insn (insert_point);
if (bitmap_empty_p (&bb_tail))
break;
}
} }
fail_shrinkwrap: fail_shrinkwrap:
bitmap_clear (&bb_tail);
bitmap_clear (&bb_antic_flags); bitmap_clear (&bb_antic_flags);
bitmap_clear (&bb_on_list); bitmap_clear (&bb_on_list);
VEC_free (basic_block, heap, vec); VEC_free (basic_block, heap, vec);
@ -5911,147 +6175,73 @@ thread_prologue_and_epilogue_insns (void)
rtl_profile_for_bb (EXIT_BLOCK_PTR); rtl_profile_for_bb (EXIT_BLOCK_PTR);
#ifdef HAVE_return exit_fallthru_edge = find_fallthru_edge (EXIT_BLOCK_PTR->preds);
/* If we're allowed to generate a simple return instruction, then by /* If we're allowed to generate a simple return instruction, then by
definition we don't need a full epilogue. If the last basic definition we don't need a full epilogue. If the last basic
block before the exit block does not contain active instructions, block before the exit block does not contain active instructions,
examine its predecessors and try to emit (conditional) return examine its predecessors and try to emit (conditional) return
instructions. */ instructions. */
if (optimize && !last_bb_active #ifdef HAVE_simple_return
&& (HAVE_return || entry_edge != orig_entry_edge)) if (entry_edge != orig_entry_edge)
{ {
edge_iterator ei2; if (optimize)
int i; {
basic_block bb; unsigned i, last;
rtx label;
VEC(basic_block,heap) *src_bbs;
/* convert_jumps_to_returns may add to EXIT_BLOCK_PTR->preds
(but won't remove). Stop at end of current preds. */
last = EDGE_COUNT (EXIT_BLOCK_PTR->preds);
for (i = 0; i < last; i++)
{
e = EDGE_I (EXIT_BLOCK_PTR->preds, i);
if (LABEL_P (BB_HEAD (e->src))
&& !bitmap_bit_p (&bb_flags, e->src->index)
&& !active_insn_between (BB_HEAD (e->src), BB_END (e->src)))
unconverted_simple_returns
= convert_jumps_to_returns (e->src, true,
unconverted_simple_returns);
}
}
if (exit_fallthru_edge != NULL
&& EDGE_COUNT (exit_fallthru_edge->src->preds) != 0
&& !bitmap_bit_p (&bb_flags, exit_fallthru_edge->src->index))
{
basic_block last_bb;
last_bb = emit_return_for_exit (exit_fallthru_edge, true);
returnjump = BB_END (last_bb);
exit_fallthru_edge = NULL;
}
}
#endif
#ifdef HAVE_return
if (HAVE_return)
{
if (exit_fallthru_edge == NULL) if (exit_fallthru_edge == NULL)
goto epilogue_done; goto epilogue_done;
label = BB_HEAD (last_bb);
src_bbs = VEC_alloc (basic_block, heap, EDGE_COUNT (last_bb->preds)); if (optimize)
FOR_EACH_EDGE (e, ei2, last_bb->preds)
if (e->src != ENTRY_BLOCK_PTR)
VEC_quick_push (basic_block, src_bbs, e->src);
FOR_EACH_VEC_ELT (basic_block, src_bbs, i, bb)
{ {
bool simple_p; basic_block last_bb = exit_fallthru_edge->src;
rtx jump;
e = find_edge (bb, last_bb);
jump = BB_END (bb); if (LABEL_P (BB_HEAD (last_bb))
&& !active_insn_between (BB_HEAD (last_bb), BB_END (last_bb)))
convert_jumps_to_returns (last_bb, false, NULL);
if (EDGE_COUNT (exit_fallthru_edge->src->preds) != 0)
{
last_bb = emit_return_for_exit (exit_fallthru_edge, false);
epilogue_end = returnjump = BB_END (last_bb);
#ifdef HAVE_simple_return #ifdef HAVE_simple_return
simple_p = (entry_edge != orig_entry_edge /* Emitting the return may add a basic block.
&& !bitmap_bit_p (&bb_flags, bb->index)); Fix bb_flags for the added block. */
#else if (last_bb != exit_fallthru_edge->src)
simple_p = false; bitmap_set_bit (&bb_flags, last_bb->index);
#endif #endif
goto epilogue_done;
if (!simple_p
&& (!HAVE_return || !JUMP_P (jump)
|| JUMP_LABEL (jump) != label))
continue;
/* If we have an unconditional jump, we can replace that
with a simple return instruction. */
if (!JUMP_P (jump))
{
emit_barrier_after (BB_END (bb));
emit_return_into_block (simple_p, bb);
} }
else if (simplejump_p (jump))
{
/* The use of the return register might be present in the exit
fallthru block. Either:
- removing the use is safe, and we should remove the use in
the exit fallthru block, or
- removing the use is not safe, and we should add it here.
For now, we conservatively choose the latter. Either of the
2 helps in crossjumping. */
emit_use_return_register_into_block (bb);
emit_return_into_block (simple_p, bb);
delete_insn (jump);
}
else if (condjump_p (jump) && JUMP_LABEL (jump) != label)
{
basic_block new_bb;
edge new_e;
gcc_assert (simple_p);
new_bb = split_edge (e);
emit_barrier_after (BB_END (new_bb));
emit_return_into_block (simple_p, new_bb);
#ifdef HAVE_simple_return
if (BB_PARTITION (new_bb) == BB_HOT_PARTITION)
simple_return_block_hot = new_bb;
else
simple_return_block_cold = new_bb;
#endif
new_e = single_succ_edge (new_bb);
redirect_edge_succ (new_e, EXIT_BLOCK_PTR);
continue;
}
/* If we have a conditional jump branching to the last
block, we can try to replace that with a conditional
return instruction. */
else if (condjump_p (jump))
{
rtx dest;
if (simple_p)
dest = simple_return_rtx;
else
dest = ret_rtx;
if (! redirect_jump (jump, dest, 0))
{
#ifdef HAVE_simple_return
if (simple_p)
VEC_safe_push (rtx, heap,
unconverted_simple_returns, jump);
#endif
continue;
}
/* See comment in simple_jump_p case above. */
emit_use_return_register_into_block (bb);
/* If this block has only one successor, it both jumps
and falls through to the fallthru block, so we can't
delete the edge. */
if (single_succ_p (bb))
continue;
}
else
{
#ifdef HAVE_simple_return
if (simple_p)
VEC_safe_push (rtx, heap,
unconverted_simple_returns, jump);
#endif
continue;
}
/* Fix up the CFG for the successful change we just made. */
redirect_edge_succ (e, EXIT_BLOCK_PTR);
}
VEC_free (basic_block, heap, src_bbs);
if (HAVE_return)
{
/* Emit a return insn for the exit fallthru block. Whether
this is still reachable will be determined later. */
emit_barrier_after (BB_END (last_bb));
emit_return_into_block (false, last_bb);
epilogue_end = BB_END (last_bb);
if (JUMP_P (epilogue_end))
set_return_jump_label (epilogue_end);
single_succ_edge (last_bb)->flags &= ~EDGE_FALLTHRU;
goto epilogue_done;
} }
} }
#endif #endif
@ -6171,10 +6361,13 @@ epilogue_done:
convert to conditional simple_returns, but couldn't for some convert to conditional simple_returns, but couldn't for some
reason, create a block to hold a simple_return insn and redirect reason, create a block to hold a simple_return insn and redirect
those remaining edges. */ those remaining edges. */
if (!VEC_empty (rtx, unconverted_simple_returns)) if (!VEC_empty (edge, unconverted_simple_returns))
{ {
basic_block simple_return_block_hot = NULL;
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 = EXIT_BLOCK_PTR->prev_bb;
rtx jump;
int i; int i;
gcc_assert (entry_edge != orig_entry_edge); gcc_assert (entry_edge != orig_entry_edge);
@ -6184,25 +6377,48 @@ epilogue_done:
if (returnjump != NULL_RTX if (returnjump != NULL_RTX
&& JUMP_LABEL (returnjump) == simple_return_rtx) && JUMP_LABEL (returnjump) == simple_return_rtx)
{ {
edge e = split_block (exit_fallthru_edge->src, e = split_block (BLOCK_FOR_INSN (returnjump), PREV_INSN (returnjump));
PREV_INSN (returnjump));
if (BB_PARTITION (e->src) == BB_HOT_PARTITION) if (BB_PARTITION (e->src) == BB_HOT_PARTITION)
simple_return_block_hot = e->dest; simple_return_block_hot = e->dest;
else else
simple_return_block_cold = e->dest; simple_return_block_cold = e->dest;
} }
FOR_EACH_VEC_ELT (rtx, unconverted_simple_returns, i, jump) /* Also check returns we might need to add to tail blocks. */
{ FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR->preds)
basic_block src_bb = BLOCK_FOR_INSN (jump); if (EDGE_COUNT (e->src->preds) != 0
edge e = find_edge (src_bb, last_bb); && (e->flags & EDGE_FAKE) != 0
basic_block *pdest_bb; && !bitmap_bit_p (&bb_flags, e->src->index))
{
if (BB_PARTITION (e->src) == BB_HOT_PARTITION)
pending_edge_hot = e;
else
pending_edge_cold = e;
}
if (BB_PARTITION (src_bb) == BB_HOT_PARTITION) FOR_EACH_VEC_ELT (edge, unconverted_simple_returns, i, e)
pdest_bb = &simple_return_block_hot; {
basic_block *pdest_bb;
edge pending;
if (BB_PARTITION (e->src) == BB_HOT_PARTITION)
{
pdest_bb = &simple_return_block_hot;
pending = pending_edge_hot;
}
else else
pdest_bb = &simple_return_block_cold; {
if (*pdest_bb == NULL) pdest_bb = &simple_return_block_cold;
pending = pending_edge_cold;
}
if (*pdest_bb == NULL && pending != NULL)
{
emit_return_into_block (true, pending->src);
pending->flags &= ~(EDGE_FALLTHRU | EDGE_FAKE);
*pdest_bb = pending->src;
}
else if (*pdest_bb == NULL)
{ {
basic_block bb; basic_block bb;
rtx start; rtx start;
@ -6219,7 +6435,19 @@ epilogue_done:
} }
redirect_edge_and_branch_force (e, *pdest_bb); redirect_edge_and_branch_force (e, *pdest_bb);
} }
VEC_free (rtx, heap, unconverted_simple_returns); VEC_free (edge, heap, unconverted_simple_returns);
}
if (entry_edge != orig_entry_edge)
{
FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR->preds)
if (EDGE_COUNT (e->src->preds) != 0
&& (e->flags & EDGE_FAKE) != 0
&& !bitmap_bit_p (&bb_flags, e->src->index))
{
emit_return_into_block (true, e->src);
e->flags &= ~(EDGE_FALLTHRU | EDGE_FAKE);
}
} }
#endif #endif
@ -6233,8 +6461,11 @@ epilogue_done:
if (!CALL_P (insn) if (!CALL_P (insn)
|| ! SIBLING_CALL_P (insn) || ! SIBLING_CALL_P (insn)
#ifdef HAVE_simple_return
|| (entry_edge != orig_entry_edge || (entry_edge != orig_entry_edge
&& !bitmap_bit_p (&bb_flags, bb->index))) && !bitmap_bit_p (&bb_flags, bb->index))
#endif
)
{ {
ei_next (&ei); ei_next (&ei);
continue; continue;
@ -6281,7 +6512,9 @@ epilogue_done:
} }
#endif #endif
#ifdef HAVE_simple_return
bitmap_clear (&bb_flags); bitmap_clear (&bb_flags);
#endif
/* Threading the prologue and epilogue changes the artificial refs /* Threading the prologue and epilogue changes the artificial refs
in the entry and exit blocks. */ in the entry and exit blocks. */

4
gcc/ifcvt.c
View File

@ -4167,13 +4167,14 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb,
if (NONDEBUG_INSN_P (insn)) if (NONDEBUG_INSN_P (insn))
df_simulate_find_defs (insn, merge_set); df_simulate_find_defs (insn, merge_set);
#ifdef HAVE_simple_return
/* If shrink-wrapping, disable this optimization when test_bb is /* If shrink-wrapping, disable this optimization when test_bb is
the first basic block and merge_bb exits. The idea is to not the first basic block and merge_bb exits. The idea is to not
move code setting up a return register as that may clobber a move code setting up a return register as that may clobber a
register used to pass function parameters, which then must be register used to pass function parameters, which then must be
saved in caller-saved regs. A caller-saved reg requires the saved in caller-saved regs. A caller-saved reg requires the
prologue, killing a shrink-wrap opportunity. */ prologue, killing a shrink-wrap opportunity. */
if ((flag_shrink_wrap && !epilogue_completed) if ((flag_shrink_wrap && HAVE_simple_return && !epilogue_completed)
&& ENTRY_BLOCK_PTR->next_bb == test_bb && ENTRY_BLOCK_PTR->next_bb == test_bb
&& single_succ_p (new_dest) && single_succ_p (new_dest)
&& single_succ (new_dest) == EXIT_BLOCK_PTR && single_succ (new_dest) == EXIT_BLOCK_PTR
@ -4224,6 +4225,7 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb,
} }
BITMAP_FREE (return_regs); BITMAP_FREE (return_regs);
} }
#endif
} }
no_body: no_body: