/* If-conversion support.
- Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007
+ Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
Free Software Foundation, Inc.
This file is part of GCC.
GCC is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by
- the Free Software Foundation; either version 2, or (at your option)
+ the Free Software Foundation; either version 3, or (at your option)
any later version.
GCC is distributed in the hope that it will be useful, but WITHOUT
License for more details.
You should have received a copy of the GNU General Public License
- along with GCC; see the file COPYING. If not, write to the Free
- Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA
- 02110-1301, USA. */
+ along with GCC; see the file COPYING3. If not see
+ <http://www.gnu.org/licenses/>. */
#include "config.h"
#include "system.h"
#include "target.h"
#include "timevar.h"
#include "tree-pass.h"
+#include "df.h"
#include "vec.h"
#include "vecprim.h"
+#include "dbgcnt.h"
-
-#ifndef HAVE_conditional_execution
-#define HAVE_conditional_execution 0
-#endif
#ifndef HAVE_conditional_move
#define HAVE_conditional_move 0
#endif
#ifndef HAVE_trap
#define HAVE_trap 0
#endif
-#ifndef HAVE_conditional_trap
-#define HAVE_conditional_trap 0
-#endif
#ifndef MAX_CONDITIONAL_EXECUTE
-#define MAX_CONDITIONAL_EXECUTE (BRANCH_COST + 1)
+#define MAX_CONDITIONAL_EXECUTE \
+ (BRANCH_COST (optimize_function_for_speed_p (cfun), false) \
+ + 1)
#endif
+#define IFCVT_MULTIPLE_DUMPS 1
+
#define NULL_BLOCK ((basic_block) NULL)
/* # of IF-THEN or IF-THEN-ELSE blocks we looked at */
execution. */
static int num_updated_if_blocks;
-/* # of changes made which require life information to be updated. */
+/* # of changes made. */
static int num_true_changes;
/* Whether conditional execution changes were made. */
static int cond_exec_changed_p;
-/* True if life data ok at present. */
-static bool life_data_ok;
-
/* Forward references. */
-static int count_bb_insns (basic_block);
-static bool cheap_bb_rtx_cost_p (basic_block, int);
+static int count_bb_insns (const_basic_block);
+static bool cheap_bb_rtx_cost_p (const_basic_block, int);
static rtx first_active_insn (basic_block);
static rtx last_active_insn (basic_block, int);
static basic_block block_fallthru (basic_block);
static int cond_exec_process_insns (ce_if_block_t *, rtx, rtx, rtx, rtx, int);
static rtx cond_exec_get_condition (rtx);
static rtx noce_get_condition (rtx, rtx *, bool);
-static int noce_operand_ok (rtx);
+static int noce_operand_ok (const_rtx);
static void merge_if_block (ce_if_block_t *);
static int find_cond_trap (basic_block, edge, edge);
static basic_block find_if_header (basic_block, int);
/* Count the number of non-jump active insns in BB. */
static int
-count_bb_insns (basic_block bb)
+count_bb_insns (const_basic_block bb)
{
int count = 0;
rtx insn = BB_HEAD (bb);
false if the cost of any instruction could not be estimated. */
static bool
-cheap_bb_rtx_cost_p (basic_block bb, int max_cost)
+cheap_bb_rtx_cost_p (const_basic_block bb, int max_cost)
{
int count = 0;
rtx insn = BB_HEAD (bb);
+ bool speed = optimize_bb_for_speed_p (bb);
while (1)
{
if (NONJUMP_INSN_P (insn))
{
- int cost = insn_rtx_cost (PATTERN (insn));
+ int cost = insn_rtx_cost (PATTERN (insn), speed);
if (cost == 0)
return false;
insn = NEXT_INSN (insn);
}
- while (NOTE_P (insn))
+ while (NOTE_P (insn) || DEBUG_INSN_P (insn))
{
if (insn == BB_END (bb))
return NULL_RTX;
while (NOTE_P (insn)
|| JUMP_P (insn)
+ || DEBUG_INSN_P (insn)
|| (skip_use_p
&& NONJUMP_INSN_P (insn)
&& GET_CODE (PATTERN (insn)) == USE))
for (insn = start; ; insn = NEXT_INSN (insn))
{
- if (NOTE_P (insn))
+ if (NOTE_P (insn) || DEBUG_INSN_P (insn))
goto insn_done;
gcc_assert(NONJUMP_INSN_P (insn) || CALL_P (insn));
from TEST_BB. For the noce transformations, we allow the symmetric
form as well. */
bool then_else_reversed;
+
+ /* Estimated cost of the particular branch instruction. */
+ int branch_cost;
};
static rtx noce_emit_store_flag (struct noce_if_info *, rtx, int, int);
build the store_flag insn directly. */
if (cond_complex)
- cond = XEXP (SET_SRC (pc_set (if_info->jump)), 0);
+ {
+ rtx set = pc_set (if_info->jump);
+ cond = XEXP (SET_SRC (set), 0);
+ if (GET_CODE (XEXP (SET_SRC (set), 2)) == LABEL_REF
+ && XEXP (XEXP (SET_SRC (set), 2), 0) == JUMP_LABEL (if_info->jump))
+ reversep = !reversep;
+ if (if_info->then_else_reversed)
+ reversep = !reversep;
+ }
if (reversep)
code = reversed_comparison_code (cond, if_info->jump);
int reversep;
rtx target, seq;
- if (GET_CODE (if_info->b) == CONST_INT
+ if (CONST_INT_P (if_info->b)
&& INTVAL (if_info->b) == STORE_FLAG_VALUE
&& if_info->a == const0_rtx)
reversep = 0;
else if (if_info->b == const0_rtx
- && GET_CODE (if_info->a) == CONST_INT
+ && CONST_INT_P (if_info->a)
&& INTVAL (if_info->a) == STORE_FLAG_VALUE
&& (reversed_comparison_code (if_info->cond, if_info->jump)
!= UNKNOWN))
int normalize, can_reverse;
enum machine_mode mode;
- if (GET_CODE (if_info->a) == CONST_INT
- && GET_CODE (if_info->b) == CONST_INT)
+ if (CONST_INT_P (if_info->a)
+ && CONST_INT_P (if_info->b))
{
mode = GET_MODE (if_info->x);
ifalse = INTVAL (if_info->a);
normalize = 0;
else if (ifalse == 0 && exact_log2 (itrue) >= 0
&& (STORE_FLAG_VALUE == 1
- || BRANCH_COST >= 2))
+ || if_info->branch_cost >= 2))
normalize = 1;
else if (itrue == 0 && exact_log2 (ifalse) >= 0 && can_reverse
- && (STORE_FLAG_VALUE == 1 || BRANCH_COST >= 2))
+ && (STORE_FLAG_VALUE == 1 || if_info->branch_cost >= 2))
normalize = 1, reversep = 1;
else if (itrue == -1
&& (STORE_FLAG_VALUE == -1
- || BRANCH_COST >= 2))
+ || if_info->branch_cost >= 2))
normalize = -1;
else if (ifalse == -1 && can_reverse
- && (STORE_FLAG_VALUE == -1 || BRANCH_COST >= 2))
+ && (STORE_FLAG_VALUE == -1 || if_info->branch_cost >= 2))
normalize = -1, reversep = 1;
- else if ((BRANCH_COST >= 2 && STORE_FLAG_VALUE == -1)
- || BRANCH_COST >= 3)
+ else if ((if_info->branch_cost >= 2 && STORE_FLAG_VALUE == -1)
+ || if_info->branch_cost >= 3)
normalize = -1;
else
return FALSE;
/* If that fails, construct conditional increment or decrement using
setcc. */
- if (BRANCH_COST >= 2
+ if (if_info->branch_cost >= 2
&& (XEXP (if_info->a, 1) == const1_rtx
|| XEXP (if_info->a, 1) == constm1_rtx))
{
int reversep;
reversep = 0;
- if ((BRANCH_COST >= 2
+ if ((if_info->branch_cost >= 2
|| STORE_FLAG_VALUE == -1)
&& ((if_info->a == const0_rtx
&& rtx_equal_p (if_info->b, if_info->x))
/* ??? FIXME: Magic number 5. */
if (cse_not_expected
&& MEM_P (a) && MEM_P (b)
- && BRANCH_COST >= 5)
+ && MEM_ADDR_SPACE (a) == MEM_ADDR_SPACE (b)
+ && if_info->branch_cost >= 5)
{
+ enum machine_mode address_mode
+ = targetm.addr_space.address_mode (MEM_ADDR_SPACE (a));
+
a = XEXP (a, 0);
b = XEXP (b, 0);
- x = gen_reg_rtx (Pmode);
+ x = gen_reg_rtx (address_mode);
is_mem = 1;
}
if insn_rtx_cost can't be estimated. */
if (insn_a)
{
- insn_cost = insn_rtx_cost (PATTERN (insn_a));
- if (insn_cost == 0 || insn_cost > COSTS_N_INSNS (BRANCH_COST))
+ insn_cost = insn_rtx_cost (PATTERN (insn_a),
+ optimize_bb_for_speed_p (BLOCK_FOR_INSN (insn_a)));
+ if (insn_cost == 0 || insn_cost > COSTS_N_INSNS (if_info->branch_cost))
return FALSE;
}
else
if (insn_b)
{
- insn_cost += insn_rtx_cost (PATTERN (insn_b));
- if (insn_cost == 0 || insn_cost > COSTS_N_INSNS (BRANCH_COST))
+ insn_cost += insn_rtx_cost (PATTERN (insn_b),
+ optimize_bb_for_speed_p (BLOCK_FOR_INSN (insn_b)));
+ if (insn_cost == 0 || insn_cost > COSTS_N_INSNS (if_info->branch_cost))
return FALSE;
}
set_mem_align (tmp,
MIN (MEM_ALIGN (if_info->a), MEM_ALIGN (if_info->b)));
+ gcc_assert (MEM_ADDR_SPACE (if_info->a) == MEM_ADDR_SPACE (if_info->b));
+ set_mem_addr_space (tmp, MEM_ADDR_SPACE (if_info->a));
+
noce_emit_move_insn (if_info->x, tmp);
}
else if (target != x)
make equivalent types of changes) to get the constants we need
if they're off by one in the right direction. */
- if (GET_CODE (target) == CONST_INT)
+ if (CONST_INT_P (target))
{
enum rtx_code code = GET_CODE (if_info->cond);
rtx op_a = XEXP (if_info->cond, 0);
/* First, look to see if we put a constant in a register. */
prev_insn = prev_nonnote_insn (if_info->cond_earliest);
if (prev_insn
+ && BLOCK_FOR_INSN (prev_insn)
+ == BLOCK_FOR_INSN (if_info->cond_earliest)
&& INSN_P (prev_insn)
&& GET_CODE (PATTERN (prev_insn)) == SET)
{
rtx src = find_reg_equal_equiv_note (prev_insn);
if (!src)
src = SET_SRC (PATTERN (prev_insn));
- if (GET_CODE (src) == CONST_INT)
+ if (CONST_INT_P (src))
{
if (rtx_equal_p (op_a, SET_DEST (PATTERN (prev_insn))))
op_a = src;
else if (rtx_equal_p (op_b, SET_DEST (PATTERN (prev_insn))))
op_b = src;
- if (GET_CODE (op_a) == CONST_INT)
+ if (CONST_INT_P (op_a))
{
rtx tmp = op_a;
op_a = op_b;
/* Now, look to see if we can get the right constant by
adjusting the conditional. */
- if (GET_CODE (op_b) == CONST_INT)
+ if (CONST_INT_P (op_b))
{
HOST_WIDE_INT desired_val = INTVAL (target);
HOST_WIDE_INT actual_val = INTVAL (op_b);
return TRUE;
}
-/* Convert "if (a < 0) x = -a; else x = a;" to "x = abs(a);", etc. */
+/* Convert "if (a < 0) x = -a; else x = a;" to "x = abs(a);",
+ "if (a < 0) x = ~a; else x = a;" to "x = one_cmpl_abs(a);",
+ etc. */
static int
noce_try_abs (struct noce_if_info *if_info)
{
rtx cond, earliest, target, seq, a, b, c;
int negate;
+ bool one_cmpl = false;
+
+ /* Reject modes with signed zeros. */
+ if (HONOR_SIGNED_ZEROS (GET_MODE (if_info->x)))
+ return FALSE;
/* Recognize A and B as constituting an ABS or NABS. The canonical
form is a branch around the negation, taken when the object is the
c = a; a = b; b = c;
negate = 1;
}
+ else if (GET_CODE (a) == NOT && rtx_equal_p (XEXP (a, 0), b))
+ {
+ negate = 0;
+ one_cmpl = true;
+ }
+ else if (GET_CODE (b) == NOT && rtx_equal_p (XEXP (b, 0), a))
+ {
+ c = a; a = b; b = c;
+ negate = 1;
+ one_cmpl = true;
+ }
else
return FALSE;
{
rtx set, insn = prev_nonnote_insn (earliest);
if (insn
+ && BLOCK_FOR_INSN (insn) == BLOCK_FOR_INSN (earliest)
&& (set = single_set (insn))
&& rtx_equal_p (SET_DEST (set), c))
{
}
start_sequence ();
-
- target = expand_abs_nojump (GET_MODE (if_info->x), b, if_info->x, 1);
+ if (one_cmpl)
+ target = expand_one_cmpl_abs_nojump (GET_MODE (if_info->x), b,
+ if_info->x);
+ else
+ target = expand_abs_nojump (GET_MODE (if_info->x), b, if_info->x, 1);
/* ??? It's a quandary whether cmove would be better here, especially
for integers. Perhaps combine will clean things up. */
if (target && negate)
- target = expand_simple_unop (GET_MODE (target), NEG, target, if_info->x, 0);
+ {
+ if (one_cmpl)
+ target = expand_simple_unop (GET_MODE (target), NOT, target,
+ if_info->x, 0);
+ else
+ target = expand_simple_unop (GET_MODE (target), NEG, target,
+ if_info->x, 0);
+ }
if (! target)
{
rtx cond, t, m, c, seq;
enum machine_mode mode;
enum rtx_code code;
- bool b_unconditional;
+ bool t_unconditional;
cond = if_info->cond;
code = GET_CODE (cond);
if (GET_MODE (m) != mode)
return FALSE;
- /* This is only profitable if T is cheap, or T is unconditionally
- executed/evaluated in the original insn sequence. The latter
- happens if INSN_B was taken from TEST_BB, or if there was no
- INSN_B which can happen for e.g. conditional stores to memory. */
- b_unconditional = (if_info->insn_b == NULL_RTX
- || BLOCK_FOR_INSN (if_info->insn_b) == if_info->test_bb);
- if (rtx_cost (t, SET) >= COSTS_N_INSNS (2)
- && (!b_unconditional
- || t != if_info->b))
+ /* This is only profitable if T is unconditionally executed/evaluated in the
+ original insn sequence or T is cheap. The former happens if B is the
+ non-zero (T) value and if INSN_B was taken from TEST_BB, or there was no
+ INSN_B which can happen for e.g. conditional stores to memory. For the
+ cost computation use the block TEST_BB where the evaluation will end up
+ after the transformation. */
+ t_unconditional =
+ (t == if_info->b
+ && (if_info->insn_b == NULL_RTX
+ || BLOCK_FOR_INSN (if_info->insn_b) == if_info->test_bb));
+ if (!(t_unconditional
+ || (rtx_cost (t, SET, optimize_bb_for_speed_p (if_info->test_bb))
+ < COSTS_N_INSNS (2))))
return FALSE;
start_sequence ();
if (GET_CODE (cond) == ZERO_EXTRACT)
{
if (XEXP (cond, 1) != const1_rtx
- || GET_CODE (XEXP (cond, 2)) != CONST_INT
+ || !CONST_INT_P (XEXP (cond, 2))
|| ! rtx_equal_p (x, XEXP (cond, 0)))
return FALSE;
bitnum = INTVAL (XEXP (cond, 2));
{
/* Check for "if (X & C) x = x op C". */
if (! rtx_equal_p (x, XEXP (a, 0))
- || GET_CODE (XEXP (a, 1)) != CONST_INT
+ || !CONST_INT_P (XEXP (a, 1))
|| (INTVAL (XEXP (a, 1)) & GET_MODE_MASK (mode))
!= (unsigned HOST_WIDE_INT) 1 << bitnum)
return FALSE;
{
/* Check for "if (X & C) x &= ~C". */
if (! rtx_equal_p (x, XEXP (a, 0))
- || GET_CODE (XEXP (a, 1)) != CONST_INT
+ || !CONST_INT_P (XEXP (a, 1))
|| (INTVAL (XEXP (a, 1)) & GET_MODE_MASK (mode))
!= (~((HOST_WIDE_INT) 1 << bitnum) & GET_MODE_MASK (mode)))
return FALSE;
/* Return true if OP is ok for if-then-else processing. */
static int
-noce_operand_ok (rtx op)
+noce_operand_ok (const_rtx op)
{
/* We special-case memories, so handle any of them with
no address side effects. */
/* Return true if a write into MEM may trap or fault. */
static bool
-noce_mem_write_may_trap_or_fault_p (rtx mem)
+noce_mem_write_may_trap_or_fault_p (const_rtx mem)
{
rtx addr;
addr = XEXP (addr, 1);
break;
case PLUS:
- if (GET_CODE (XEXP (addr, 1)) == CONST_INT)
+ if (CONST_INT_P (XEXP (addr, 1)))
addr = XEXP (addr, 0);
else
return false;
return false;
}
+/* Return whether we can use store speculation for MEM. TOP_BB is the
+ basic block above the conditional block where we are considering
+ doing the speculative store. We look for whether MEM is set
+ unconditionally later in the function. */
+
+static bool
+noce_can_store_speculate_p (basic_block top_bb, const_rtx mem)
+{
+ basic_block dominator;
+
+ for (dominator = get_immediate_dominator (CDI_POST_DOMINATORS, top_bb);
+ dominator != NULL;
+ dominator = get_immediate_dominator (CDI_POST_DOMINATORS, dominator))
+ {
+ rtx insn;
+
+ FOR_BB_INSNS (dominator, insn)
+ {
+ /* If we see something that might be a memory barrier, we
+ have to stop looking. Even if the MEM is set later in
+ the function, we still don't want to set it
+ unconditionally before the barrier. */
+ if (INSN_P (insn)
+ && (volatile_insn_p (PATTERN (insn))
+ || (CALL_P (insn) && (!RTL_CONST_CALL_P (insn)))))
+ return false;
+
+ if (memory_modified_in_insn_p (mem, insn))
+ return true;
+ if (modified_in_p (XEXP (mem, 0), insn))
+ return false;
+
+ }
+ }
+
+ return false;
+}
+
/* Given a simple IF-THEN-JOIN or IF-THEN-ELSE-JOIN block, attempt to convert
it without using conditional execution. Return TRUE if we were successful
at converting the block. */
else
{
insn_b = prev_nonnote_insn (if_info->cond_earliest);
+ while (insn_b && DEBUG_INSN_P (insn_b))
+ insn_b = prev_nonnote_insn (insn_b);
/* We're going to be moving the evaluation of B down from above
COND_EARLIEST to JUMP. Make sure the relevant data is still
intact. */
if (! insn_b
+ || BLOCK_FOR_INSN (insn_b) != BLOCK_FOR_INSN (if_info->cond_earliest)
|| !NONJUMP_INSN_P (insn_b)
|| (set_b = single_set (insn_b)) == NULL_RTX
|| ! rtx_equal_p (x, SET_DEST (set_b))
+ || ! noce_operand_ok (SET_SRC (set_b))
|| reg_overlap_mentioned_p (x, SET_SRC (set_b))
- || modified_between_p (SET_SRC (set_b),
- PREV_INSN (if_info->cond_earliest), jump)
+ || modified_between_p (SET_SRC (set_b), insn_b, jump)
/* Likewise with X. In particular this can happen when
noce_get_condition looks farther back in the instruction
stream than one might expect. */
|| reg_overlap_mentioned_p (x, cond)
|| reg_overlap_mentioned_p (x, a)
- || modified_between_p (x, PREV_INSN (if_info->cond_earliest), jump))
+ || modified_between_p (x, insn_b, jump))
insn_b = set_b = NULL_RTX;
}
if (GET_MODE (x) == BLKmode)
return FALSE;
- if (GET_MODE (x) == ZERO_EXTRACT
- && (GET_CODE (XEXP (x, 1)) != CONST_INT
- || GET_CODE (XEXP (x, 2)) != CONST_INT))
+ if (GET_CODE (x) == ZERO_EXTRACT
+ && (!CONST_INT_P (XEXP (x, 1))
+ || !CONST_INT_P (XEXP (x, 2))))
return FALSE;
x = gen_reg_rtx (GET_MODE (GET_CODE (x) == STRICT_LOW_PART
if (! noce_operand_ok (a) || ! noce_operand_ok (b))
return FALSE;
+ retry:
/* Set up the info block for our subroutines. */
if_info->insn_a = insn_a;
if_info->insn_b = insn_b;
goto success;
}
- /* Disallow the "if (...) x = a;" form (with an implicit "else x = x;")
- for optimizations if writing to x may trap or fault, i.e. it's a memory
- other than a static var or a stack slot, is misaligned on strict
- aligned machines or is read-only.
- If x is a read-only memory, then the program is valid only if we
- avoid the store into it. If there are stores on both the THEN and
- ELSE arms, then we can go ahead with the conversion; either the
- program is broken, or the condition is always false such that the
- other memory is selected. */
- if (!set_b && MEM_P (orig_x) && noce_mem_write_may_trap_or_fault_p (orig_x))
- return FALSE;
+ if (!set_b && MEM_P (orig_x))
+ {
+ /* Disallow the "if (...) x = a;" form (implicit "else x = x;")
+ for optimizations if writing to x may trap or fault,
+ i.e. it's a memory other than a static var or a stack slot,
+ is misaligned on strict aligned machines or is read-only. If
+ x is a read-only memory, then the program is valid only if we
+ avoid the store into it. If there are stores on both the
+ THEN and ELSE arms, then we can go ahead with the conversion;
+ either the program is broken, or the condition is always
+ false such that the other memory is selected. */
+ if (noce_mem_write_may_trap_or_fault_p (orig_x))
+ return FALSE;
+
+ /* Avoid store speculation: given "if (...) x = a" where x is a
+ MEM, we only want to do the store if x is always set
+ somewhere in the function. This avoids cases like
+ if (pthread_mutex_trylock(mutex))
+ ++global_variable;
+ where we only want global_variable to be changed if the mutex
+ is held. FIXME: This should ideally be expressed directly in
+ RTL somehow. */
+ if (!noce_can_store_speculate_p (test_bb, orig_x))
+ return FALSE;
+ }
if (noce_try_move (if_info))
goto success;
if (HAVE_conditional_move
&& noce_try_cmove (if_info))
goto success;
- if (! HAVE_conditional_execution)
+ if (! targetm.have_conditional_execution ())
{
if (noce_try_store_flag_constants (if_info))
goto success;
goto success;
}
+ if (!else_bb && set_b)
+ {
+ insn_b = set_b = NULL_RTX;
+ b = orig_x;
+ goto retry;
+ }
+
return FALSE;
success:
REGS. COND is the condition we will test. */
static int
-check_cond_move_block (basic_block bb, rtx *vals, VEC (int, heap) *regs, rtx cond)
+check_cond_move_block (basic_block bb, rtx *vals, VEC (int, heap) **regs, rtx cond)
{
rtx insn;
{
rtx set, dest, src;
- if (!INSN_P (insn) || JUMP_P (insn))
+ if (!NONDEBUG_INSN_P (insn) || JUMP_P (insn))
continue;
set = single_set (insn);
if (!set)
vals[REGNO (dest)] = src;
- VEC_safe_push (int, heap, regs, REGNO (dest));
+ VEC_safe_push (int, heap, *regs, REGNO (dest));
}
return TRUE;
rtx set, target, dest, t, e;
unsigned int regno;
- if (!INSN_P (insn) || JUMP_P (insn))
+ /* ??? Maybe emit conditional debug insn? */
+ if (!NONDEBUG_INSN_P (insn) || JUMP_P (insn))
continue;
set = single_set (insn);
gcc_assert (set && REG_P (SET_DEST (set)));
memset (else_vals, 0, size);
/* Make sure the blocks are suitable. */
- if (!check_cond_move_block (then_bb, then_vals, then_regs, cond)
- || (else_bb && !check_cond_move_block (else_bb, else_vals, else_regs, cond)))
- return FALSE;
+ if (!check_cond_move_block (then_bb, then_vals, &then_regs, cond)
+ || (else_bb && !check_cond_move_block (else_bb, else_vals, &else_regs, cond)))
+ {
+ VEC_free (int, heap, then_regs);
+ VEC_free (int, heap, else_regs);
+ return FALSE;
+ }
/* Make sure the blocks can be used together. If the same register
is set in both blocks, and is not set to a constant in both
if (!CONSTANT_P (then_vals[reg])
&& !CONSTANT_P (else_vals[reg])
&& !rtx_equal_p (then_vals[reg], else_vals[reg]))
- return FALSE;
+ {
+ VEC_free (int, heap, then_regs);
+ VEC_free (int, heap, else_regs);
+ return FALSE;
+ }
}
}
branches, since if we convert we are going to always execute
them. */
if (c > MAX_CONDITIONAL_EXECUTE)
- return FALSE;
+ {
+ VEC_free (int, heap, then_regs);
+ VEC_free (int, heap, else_regs);
+ return FALSE;
+ }
/* Try to emit the conditional moves. First do the then block,
then do anything left in the else blocks. */
then_vals, else_vals, true)))
{
end_sequence ();
+ VEC_free (int, heap, then_regs);
+ VEC_free (int, heap, else_regs);
return FALSE;
}
seq = end_ifcvt_sequence (if_info);
if (!seq)
- return FALSE;
+ {
+ VEC_free (int, heap, then_regs);
+ VEC_free (int, heap, else_regs);
+ return FALSE;
+ }
loc_insn = first_active_insn (then_bb);
if (!loc_insn)
VEC_free (int, heap, then_regs);
VEC_free (int, heap, else_regs);
-
return TRUE;
}
basic_block then_bb, else_bb, join_bb;
bool then_else_reversed = false;
rtx jump, cond;
+ rtx cond_earliest;
struct noce_if_info if_info;
/* We only ever should get here before reload. */
/* If this is not a standard conditional jump, we can't parse it. */
cond = noce_get_condition (jump,
- &if_info.cond_earliest,
+ &cond_earliest,
then_else_reversed);
if (!cond)
return FALSE;
if_info.else_bb = else_bb;
if_info.join_bb = join_bb;
if_info.cond = cond;
+ if_info.cond_earliest = cond_earliest;
if_info.jump = jump;
if_info.then_else_reversed = then_else_reversed;
+ if_info.branch_cost = BRANCH_COST (optimize_bb_for_speed_p (test_bb),
+ predictable_edge_p (then_edge));
/* Do the real work. */
/* All block merging is done into the lower block numbers. */
combo_bb = test_bb;
+ df_set_bb_dirty (test_bb);
/* Merge any basic blocks to handle && and || subtests. Each of
the blocks are on the fallthru path from the predecessor block. */
else if (EDGE_COUNT (join_bb->preds) < 2
&& join_bb != EXIT_BLOCK_PTR)
{
- /* We can merge the JOIN. */
+ /* We can merge the JOIN cleanly and update the dataflow try
+ again on this pass.*/
merge_blocks (combo_bb, join_bb);
num_true_changes++;
}
then_edge = EDGE_SUCC (test_bb, 0);
else_edge = EDGE_SUCC (test_bb, 1);
+ if (df_get_bb_dirty (then_edge->dest))
+ return NULL;
+ if (df_get_bb_dirty (else_edge->dest))
+ return NULL;
+
/* Neither edge should be abnormal. */
if ((then_edge->flags & EDGE_COMPLEX)
|| (else_edge->flags & EDGE_COMPLEX))
&& noce_find_if_block (test_bb, then_edge, else_edge, pass))
goto success;
- if (HAVE_conditional_execution && reload_completed
+ if (targetm.have_conditional_execution () && reload_completed
&& cond_exec_find_if_block (&ce_info))
goto success;
- if (HAVE_trap && HAVE_conditional_trap
+ if (HAVE_trap
+ && optab_handler (ctrap_optab, word_mode)->insn_code != CODE_FOR_nothing
&& find_cond_trap (test_bb, then_edge, else_edge))
goto success;
if (dom_info_state (CDI_POST_DOMINATORS) >= DOM_NO_FAST_QUERY
- && (! HAVE_conditional_execution || reload_completed))
+ && (! targetm.have_conditional_execution () || reload_completed))
{
if (find_if_case_1 (test_bb, then_edge, else_edge))
goto success;
success:
if (dump_file)
fprintf (dump_file, "Conversion succeeded on pass %d.\n", pass);
+ /* Set this so we continue looking. */
+ cond_exec_changed_p = TRUE;
return ce_info.test_bb;
}
if (INSN_P (insn)
&& !JUMP_P (insn)
+ && !DEBUG_INSN_P (insn)
&& GET_CODE (PATTERN (insn)) != USE
&& GET_CODE (PATTERN (insn)) != CLOBBER)
n_insns++;
/* We only ever should get here after reload,
and only if we have conditional execution. */
- gcc_assert (HAVE_conditional_execution && reload_completed);
+ gcc_assert (targetm.have_conditional_execution () && reload_completed);
/* Discover if any fall through predecessors of the current test basic block
were && tests (which jump to the else block) or || tests (which jump to
if (EDGE_COUNT (then_bb->succs) > 0
&& (!single_succ_p (then_bb)
|| (single_succ_edge (then_bb)->flags & EDGE_COMPLEX)
- || (flow2_completed && tablejump_p (BB_END (then_bb), NULL, NULL))))
+ || (epilogue_completed && tablejump_p (BB_END (then_bb), NULL, NULL))))
return FALSE;
/* If the THEN block has no successors, conditional execution can still
&& single_succ (then_bb) == single_succ (else_bb)
&& single_pred_p (else_bb)
&& ! (single_succ_edge (else_bb)->flags & EDGE_COMPLEX)
- && ! (flow2_completed && tablejump_p (BB_END (else_bb), NULL, NULL)))
+ && ! (epilogue_completed && tablejump_p (BB_END (else_bb), NULL, NULL)))
join_bb = single_succ (else_bb);
/* Otherwise it is not an IF-THEN or IF-THEN-ELSE combination. */
}
/* Attempt to generate the conditional trap. */
- seq = gen_cond_trap (code, XEXP (cond, 0),
- XEXP (cond, 1),
+ seq = gen_cond_trap (code, copy_rtx (XEXP (cond, 0)),
+ copy_rtx (XEXP (cond, 1)),
TRAP_CODE (PATTERN (trap)));
if (seq == NULL)
return FALSE;
/* Delete the trap block if possible. */
remove_edge (trap_bb == then_bb ? then_edge : else_edge);
+ df_set_bb_dirty (test_bb);
+ df_set_bb_dirty (then_bb);
+ df_set_bb_dirty (else_bb);
+
if (EDGE_COUNT (trap_bb->preds) == 0)
{
delete_basic_block (trap_bb);
find_if_case_1 (basic_block test_bb, edge then_edge, edge else_edge)
{
basic_block then_bb = then_edge->dest;
- basic_block else_bb = else_edge->dest, new_bb;
+ basic_block else_bb = else_edge->dest;
+ basic_block new_bb;
int then_bb_index;
/* If we are partitioning hot/cold basic blocks, we don't want to
test_bb->index, then_bb->index);
/* THEN is small. */
- if (! cheap_bb_rtx_cost_p (then_bb, COSTS_N_INSNS (BRANCH_COST)))
+ if (! cheap_bb_rtx_cost_p (then_bb,
+ COSTS_N_INSNS (BRANCH_COST (optimize_bb_for_speed_p (then_edge->src),
+ predictable_edge_p (then_edge)))))
return FALSE;
/* Registers set are dead, or are predicable. */
/* Conversion went ok, including moving the insns and fixing up the
jump. Adjust the CFG to match. */
- bitmap_ior (test_bb->il.rtl->global_live_at_end,
- else_bb->il.rtl->global_live_at_start,
- then_bb->il.rtl->global_live_at_end);
-
-
/* We can avoid creating a new basic block if then_bb is immediately
followed by else_bb, i.e. deleting then_bb allows test_bb to fall
thru to else_bb. */
}
else
new_bb = redirect_edge_and_branch_force (FALLTHRU_EDGE (test_bb),
- else_bb);
+ else_bb);
+
+ df_set_bb_dirty (test_bb);
+ df_set_bb_dirty (else_bb);
then_bb_index = then_bb->index;
delete_basic_block (then_bb);
block we removed. */
if (new_bb)
{
- new_bb->index = then_bb_index;
- SET_BASIC_BLOCK (then_bb_index, 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);
}
- /* We've possibly created jump to next insn, cleanup_cfg will solve that
- later. */
num_true_changes++;
num_updated_if_blocks++;
test_bb->index, else_bb->index);
/* ELSE is small. */
- if (! cheap_bb_rtx_cost_p (else_bb, COSTS_N_INSNS (BRANCH_COST)))
+ if (! cheap_bb_rtx_cost_p (else_bb,
+ COSTS_N_INSNS (BRANCH_COST (optimize_bb_for_speed_p (else_edge->src),
+ predictable_edge_p (else_edge)))))
return FALSE;
/* Registers set are dead, or are predicable. */
/* Conversion went ok, including moving the insns and fixing up the
jump. Adjust the CFG to match. */
- bitmap_ior (test_bb->il.rtl->global_live_at_end,
- then_bb->il.rtl->global_live_at_start,
- else_bb->il.rtl->global_live_at_end);
-
+ df_set_bb_dirty (test_bb);
+ df_set_bb_dirty (then_bb);
delete_basic_block (else_bb);
num_true_changes++;
basic_block other_bb, basic_block new_dest, int reversep)
{
rtx head, end, jump, earliest = NULL_RTX, old_dest, new_label = NULL_RTX;
+ /* Number of pending changes. */
+ int n_validated_changes = 0;
jump = BB_END (test_bb);
head = BB_HEAD (merge_bb);
end = BB_END (merge_bb);
+ while (DEBUG_INSN_P (end) && end != head)
+ end = PREV_INSN (end);
+
/* If merge_bb ends with a tablejump, predicating/moving insn's
into test_bb and then deleting merge_bb will result in the jumptable
that follows merge_bb being removed along with merge_bb and then we
if (LABEL_P (head))
head = NEXT_INSN (head);
+ while (DEBUG_INSN_P (head) && head != end)
+ head = NEXT_INSN (head);
if (NOTE_P (head))
{
if (head == end)
goto no_body;
}
head = NEXT_INSN (head);
+ while (DEBUG_INSN_P (head) && head != end)
+ head = NEXT_INSN (head);
}
if (JUMP_P (end))
goto no_body;
}
end = PREV_INSN (end);
+ while (DEBUG_INSN_P (end) && end != head)
+ end = PREV_INSN (end);
}
/* Disable handling dead code by conditional execution if the machine needs
to do anything funny with the tests, etc. */
#ifndef IFCVT_MODIFY_TESTS
- if (HAVE_conditional_execution)
+ if (targetm.have_conditional_execution ())
{
/* In the conditional execution case, we have things easy. We know
the condition is reversible. We don't have to check life info
prob_val = GEN_INT (REG_BR_PROB_BASE - INTVAL (prob_val));
}
- if (! cond_exec_process_insns ((ce_if_block_t *)0, head, end, cond,
- prob_val, 0))
- goto cancel;
+ if (cond_exec_process_insns (NULL, head, end, cond, prob_val, 0)
+ && verify_changes (0))
+ n_validated_changes = num_validated_changes ();
+ else
+ cancel_changes (0);
earliest = jump;
}
- else
#endif
+ /* Try the NCE path if the CE path did not result in any changes. */
+ if (n_validated_changes == 0)
{
/* In the non-conditional execution case, we have to verify that there
are no trapping operations, no calls, no references to memory, and
that any registers modified are dead at the branch site. */
rtx insn, cond, prev;
- regset merge_set, tmp, test_live, test_set;
- struct propagate_block_info *pbi;
+ bitmap merge_set, test_live, test_set;
unsigned i, fail = 0;
bitmap_iterator bi;
{
if (CALL_P (insn))
return FALSE;
- if (INSN_P (insn))
+ if (NONDEBUG_INSN_P (insn))
{
if (may_trap_p (PATTERN (insn)))
return FALSE;
TEST_SET = set of registers set between EARLIEST and the
end of the block. */
- tmp = ALLOC_REG_SET (®_obstack);
- merge_set = ALLOC_REG_SET (®_obstack);
- test_live = ALLOC_REG_SET (®_obstack);
- test_set = ALLOC_REG_SET (®_obstack);
+ merge_set = BITMAP_ALLOC (®_obstack);
+ test_live = BITMAP_ALLOC (®_obstack);
+ test_set = BITMAP_ALLOC (®_obstack);
/* ??? bb->local_set is only valid during calculate_global_regs_live,
so we must recompute usage for MERGE_BB. Not so bad, I suppose,
expander called from noce_emit_cmove), we must resize the
array first. */
if (max_regno < max_reg_num ())
+ max_regno = max_reg_num ();
+
+ FOR_BB_INSNS (merge_bb, insn)
{
- max_regno = max_reg_num ();
- allocate_reg_info (max_regno, FALSE, FALSE);
+ if (NONDEBUG_INSN_P (insn))
+ {
+ unsigned int uid = INSN_UID (insn);
+ df_ref *def_rec;
+ for (def_rec = DF_INSN_UID_DEFS (uid); *def_rec; def_rec++)
+ {
+ df_ref def = *def_rec;
+ bitmap_set_bit (merge_set, DF_REF_REGNO (def));
+ }
+ }
}
- propagate_block (merge_bb, tmp, merge_set, merge_set, 0);
/* For small register class machines, don't lengthen lifetimes of
hard registers before reload. */
/* For TEST, we're interested in a range of insns, not a whole block.
Moreover, we're interested in the insns live from OTHER_BB. */
- COPY_REG_SET (test_live, other_bb->il.rtl->global_live_at_start);
- pbi = init_propagate_block_info (test_bb, test_live, test_set, test_set,
- 0);
-
+ /* The loop below takes the set of live registers
+ after JUMP, and calculates the live set before EARLIEST. */
+ bitmap_copy (test_live, df_get_live_in (other_bb));
+ df_simulate_initialize_backwards (test_bb, test_live);
for (insn = jump; ; insn = prev)
{
- prev = propagate_one_insn (pbi, insn);
+ if (INSN_P (insn))
+ {
+ df_simulate_find_defs (insn, test_set);
+ df_simulate_one_insn_backwards (test_bb, insn, test_live);
+ }
+ prev = PREV_INSN (insn);
if (insn == earliest)
break;
}
- free_propagate_block_info (pbi);
-
/* We can perform the transformation if
MERGE_SET & (TEST_SET | TEST_LIVE)
and
- TEST_SET & merge_bb->il.rtl->global_live_at_start
+ TEST_SET & DF_LIVE_IN (merge_bb)
are empty. */
if (bitmap_intersect_p (test_set, merge_set)
|| bitmap_intersect_p (test_live, merge_set)
- || bitmap_intersect_p (test_set,
- merge_bb->il.rtl->global_live_at_start))
+ || bitmap_intersect_p (test_set, df_get_live_in (merge_bb)))
fail = 1;
- FREE_REG_SET (tmp);
- FREE_REG_SET (merge_set);
- FREE_REG_SET (test_live);
- FREE_REG_SET (test_set);
+ BITMAP_FREE (merge_set);
+ BITMAP_FREE (test_live);
+ BITMAP_FREE (test_set);
if (fail)
return FALSE;
goto cancel;
}
- if (! apply_change_group ())
- return FALSE;
+ if (verify_changes (n_validated_changes))
+ confirm_change_group ();
+ else
+ goto cancel;
if (other_bb != new_dest)
{
if (! note)
continue;
set = single_set (insn);
- if (!set || !function_invariant_p (SET_SRC (set)))
+ if (!set || !function_invariant_p (SET_SRC (set))
+ || !function_invariant_p (XEXP (note, 0)))
remove_note (insn, note);
} while (insn != end && (insn = NEXT_INSN (insn)));
/* Main entry point for all if-conversion. */
static void
-if_convert (int x_life_data_ok)
+if_convert (void)
{
basic_block bb;
int pass;
+ if (optimize == 1)
+ {
+ df_live_add_problem ();
+ df_live_set_all_dirty ();
+ }
+
num_possible_if_blocks = 0;
num_updated_if_blocks = 0;
num_true_changes = 0;
- life_data_ok = (x_life_data_ok != 0);
-
- /* Some transformations in this pass can create new pseudos,
- if the pass runs before reload. Make sure we can do so. */
- gcc_assert (! no_new_pseudos || reload_completed);
loop_optimizer_init (AVOID_CFG_MODIFICATIONS);
- if (current_loops)
- {
- mark_loop_exit_edges ();
- loop_optimizer_finalize ();
- }
+ mark_loop_exit_edges ();
+ loop_optimizer_finalize ();
free_dominance_info (CDI_DOMINATORS);
- /* Compute postdominators if we think we'll use them. */
- if (HAVE_conditional_execution || life_data_ok)
- calculate_dominance_info (CDI_POST_DOMINATORS);
+ /* Compute postdominators. */
+ calculate_dominance_info (CDI_POST_DOMINATORS);
- if (life_data_ok)
- clear_bb_flags ();
+ df_set_flags (DF_LR_RUN_DCE);
/* Go through each of the basic blocks looking for things to convert. If we
have conditional execution, we make multiple passes to allow us to handle
pass = 0;
do
{
+ df_analyze ();
+ /* Only need to do dce on the first pass. */
+ df_clear_flags (DF_LR_RUN_DCE);
cond_exec_changed_p = FALSE;
pass++;
FOR_EACH_BB (bb)
{
- basic_block new_bb;
- while ((new_bb = find_if_header (bb, pass)))
- bb = new_bb;
+ basic_block new_bb;
+ while (!df_get_bb_dirty (bb)
+ && (new_bb = find_if_header (bb, pass)) != NULL)
+ bb = new_bb;
}
#ifdef IFCVT_MULTIPLE_DUMPS
if (dump_file && cond_exec_changed_p)
- print_rtl_with_bb (dump_file, get_insns ());
+ {
+ if (dump_flags & TDF_SLIM)
+ print_rtl_slim_with_bb (dump_file, get_insns (), dump_flags);
+ else
+ print_rtl_with_bb (dump_file, get_insns ());
+ }
#endif
}
while (cond_exec_changed_p);
clear_aux_for_blocks ();
- /* Rebuild life info for basic blocks that require it. */
- if (num_true_changes && life_data_ok)
- {
- /* If we allocated new pseudos, we must resize the array for sched1. */
- if (max_regno < max_reg_num ())
- {
- max_regno = max_reg_num ();
- allocate_reg_info (max_regno, FALSE, FALSE);
- }
- update_life_info_in_dirty_blocks (UPDATE_LIFE_GLOBAL_RM_NOTES,
- PROP_DEATH_NOTES | PROP_SCAN_DEAD_CODE
- | PROP_KILL_DEAD_CODE);
- }
+ /* If we allocated new pseudos, we must resize the array for sched1. */
+ if (max_regno < max_reg_num ())
+ max_regno = max_reg_num ();
/* Write the final stats. */
if (dump_file && num_possible_if_blocks > 0)
num_true_changes);
}
+ if (optimize == 1)
+ df_remove_problem (df_live);
+
#ifdef ENABLE_CHECKING
verify_flow_info ();
#endif
static bool
gate_handle_if_conversion (void)
{
- return (optimize > 0);
+ return (optimize > 0)
+ && dbg_cnt (if_conversion);
}
/* If-conversion and CFG cleanup. */
if (dump_file)
dump_flow_info (dump_file, dump_flags);
cleanup_cfg (CLEANUP_EXPENSIVE);
- reg_scan (get_insns (), max_reg_num ());
- if_convert (0);
+ if_convert ();
}
- timevar_push (TV_JUMP);
- cleanup_cfg (CLEANUP_EXPENSIVE);
- reg_scan (get_insns (), max_reg_num ());
- timevar_pop (TV_JUMP);
+ cleanup_cfg (0);
return 0;
}
-struct tree_opt_pass pass_rtl_ifcvt =
+struct rtl_opt_pass pass_rtl_ifcvt =
{
+ {
+ RTL_PASS,
"ce1", /* name */
gate_handle_if_conversion, /* gate */
rest_of_handle_if_conversion, /* execute */
0, /* properties_provided */
0, /* properties_destroyed */
0, /* todo_flags_start */
- TODO_dump_func, /* todo_flags_finish */
- 'C' /* letter */
+ TODO_df_finish | TODO_verify_rtl_sharing |
+ TODO_dump_func /* todo_flags_finish */
+ }
};
static bool
gate_handle_if_after_combine (void)
{
- return (optimize > 0 && flag_if_conversion);
+ return optimize > 0 && flag_if_conversion
+ && dbg_cnt (if_after_combine);
}
static unsigned int
rest_of_handle_if_after_combine (void)
{
- no_new_pseudos = 0;
- if_convert (1);
- no_new_pseudos = 1;
+ if_convert ();
return 0;
}
-struct tree_opt_pass pass_if_after_combine =
+struct rtl_opt_pass pass_if_after_combine =
{
+ {
+ RTL_PASS,
"ce2", /* name */
gate_handle_if_after_combine, /* gate */
rest_of_handle_if_after_combine, /* execute */
0, /* properties_provided */
0, /* properties_destroyed */
0, /* todo_flags_start */
+ TODO_df_finish | TODO_verify_rtl_sharing |
TODO_dump_func |
- TODO_ggc_collect, /* todo_flags_finish */
- 'C' /* letter */
+ TODO_ggc_collect /* todo_flags_finish */
+ }
};
static bool
gate_handle_if_after_reload (void)
{
- return (optimize > 0);
+ return optimize > 0 && flag_if_conversion2
+ && dbg_cnt (if_after_reload);
}
static unsigned int
rest_of_handle_if_after_reload (void)
{
- /* Last attempt to optimize CFG, as scheduling, peepholing and insn
- splitting possibly introduced more crossjumping opportunities. */
- cleanup_cfg (CLEANUP_EXPENSIVE
- | CLEANUP_UPDATE_LIFE
- | (flag_crossjumping ? CLEANUP_CROSSJUMP : 0));
- if (flag_if_conversion2)
- if_convert (1);
+ if_convert ();
return 0;
}
-struct tree_opt_pass pass_if_after_reload =
+struct rtl_opt_pass pass_if_after_reload =
{
+ {
+ RTL_PASS,
"ce3", /* name */
gate_handle_if_after_reload, /* gate */
rest_of_handle_if_after_reload, /* execute */
0, /* properties_provided */
0, /* properties_destroyed */
0, /* todo_flags_start */
+ TODO_df_finish | TODO_verify_rtl_sharing |
TODO_dump_func |
- TODO_ggc_collect, /* todo_flags_finish */
- 'E' /* letter */
+ TODO_ggc_collect /* todo_flags_finish */
+ }
};
OSDN Git Service
