X-Git-Url: http://git.sourceforge.jp/view?a=blobdiff_plain;f=gcc%2Fifcvt.c;h=8d6b885f342e8bd4b6199e41be6fe17eadf6ba38;hb=2a7259ea89c2e8fc0cd15f87a22b0dd9458bdd7d;hp=6fca205a4ef659486f24051a77716cc9590c05f1;hpb=ad4583d978292a368c345f43dc85ede2c0e0e498;p=pf3gnuchains%2Fgcc-fork.git diff --git a/gcc/ifcvt.c b/gcc/ifcvt.c index 6fca205a4ef..8d6b885f342 100644 --- a/gcc/ifcvt.c +++ b/gcc/ifcvt.c @@ -1,12 +1,12 @@ /* If-conversion support. - Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007 + Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2010 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 @@ -15,9 +15,8 @@ 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 + . */ #include "config.h" #include "system.h" @@ -34,7 +33,6 @@ #include "hard-reg-set.h" #include "basic-block.h" #include "expr.h" -#include "real.h" #include "output.h" #include "optabs.h" #include "toplev.h" @@ -43,13 +41,11 @@ #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 @@ -62,14 +58,15 @@ #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 */ @@ -79,25 +76,22 @@ static int num_possible_if_blocks; 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); @@ -115,7 +109,7 @@ static rtx block_has_only_trap (basic_block); /* 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); @@ -138,16 +132,17 @@ count_bb_insns (basic_block 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; @@ -195,7 +190,7 @@ first_active_insn (basic_block bb) insn = NEXT_INSN (insn); } - while (NOTE_P (insn)) + while (NOTE_P (insn) || DEBUG_INSN_P (insn)) { if (insn == BB_END (bb)) return NULL_RTX; @@ -218,6 +213,7 @@ last_active_insn (basic_block bb, int skip_use_p) while (NOTE_P (insn) || JUMP_P (insn) + || DEBUG_INSN_P (insn) || (skip_use_p && NONJUMP_INSN_P (insn) && GET_CODE (PATTERN (insn)) == USE)) @@ -270,7 +266,7 @@ cond_exec_process_insns (ce_if_block_t *ce_info ATTRIBUTE_UNUSED, 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)); @@ -388,7 +384,11 @@ cond_exec_process_if_block (ce_if_block_t * ce_info, rtx false_expr; /* test for then block insns */ rtx true_prob_val; /* probability of else block */ rtx false_prob_val; /* probability of then block */ - int n_insns; + rtx then_last_head = NULL_RTX; /* Last match at the head of THEN */ + rtx else_last_head = NULL_RTX; /* Last match at the head of ELSE */ + rtx then_first_tail = NULL_RTX; /* First match at the tail of THEN */ + rtx else_first_tail = NULL_RTX; /* First match at the tail of ELSE */ + int then_n_insns, else_n_insns, n_insns; enum rtx_code false_code; /* If test is comprised of && or || elements, and we've failed at handling @@ -421,15 +421,78 @@ cond_exec_process_if_block (ce_if_block_t * ce_info, number of insns and see if it is small enough to convert. */ then_start = first_active_insn (then_bb); then_end = last_active_insn (then_bb, TRUE); - n_insns = ce_info->num_then_insns = count_bb_insns (then_bb); + then_n_insns = ce_info->num_then_insns = count_bb_insns (then_bb); + n_insns = then_n_insns; max = MAX_CONDITIONAL_EXECUTE; if (else_bb) { + int n_matching; + max *= 2; else_start = first_active_insn (else_bb); else_end = last_active_insn (else_bb, TRUE); - n_insns += ce_info->num_else_insns = count_bb_insns (else_bb); + else_n_insns = ce_info->num_else_insns = count_bb_insns (else_bb); + n_insns += else_n_insns; + + /* Look for matching sequences at the head and tail of the two blocks, + and limit the range of insns to be converted if possible. */ + n_matching = flow_find_cross_jump (then_bb, else_bb, + &then_first_tail, &else_first_tail); + if (then_first_tail == BB_HEAD (then_bb)) + then_start = then_end = NULL_RTX; + if (else_first_tail == BB_HEAD (else_bb)) + else_start = else_end = NULL_RTX; + + if (n_matching > 0) + { + if (then_end) + then_end = prev_active_insn (then_first_tail); + if (else_end) + else_end = prev_active_insn (else_first_tail); + n_insns -= 2 * n_matching; + } + + if (then_start && else_start) + { + int longest_match = MIN (then_n_insns - n_matching, + else_n_insns - n_matching); + n_matching + = flow_find_head_matching_sequence (then_bb, else_bb, + &then_last_head, + &else_last_head, + longest_match); + + if (n_matching > 0) + { + rtx insn; + + /* We won't pass the insns in the head sequence to + cond_exec_process_insns, so we need to test them here + to make sure that they don't clobber the condition. */ + for (insn = BB_HEAD (then_bb); + insn != NEXT_INSN (then_last_head); + insn = NEXT_INSN (insn)) + if (!LABEL_P (insn) && !NOTE_P (insn) + && !DEBUG_INSN_P (insn) + && modified_in_p (test_expr, insn)) + return FALSE; + } + + if (then_last_head == then_end) + then_start = then_end = NULL_RTX; + if (else_last_head == else_end) + else_start = else_end = NULL_RTX; + + if (n_matching > 0) + { + if (then_start) + then_start = next_active_insn (then_last_head); + if (else_start) + else_start = next_active_insn (else_last_head); + n_insns -= 2 * n_matching; + } + } } if (n_insns > max) @@ -573,7 +636,21 @@ cond_exec_process_if_block (ce_if_block_t * ce_info, fprintf (dump_file, "%d insn%s converted to conditional execution.\n", n_insns, (n_insns == 1) ? " was" : "s were"); - /* Merge the blocks! */ + /* Merge the blocks! If we had matching sequences, make sure to delete one + copy at the appropriate location first: delete the copy in the THEN branch + for a tail sequence so that the remaining one is executed last for both + branches, and delete the copy in the ELSE branch for a head sequence so + that the remaining one is executed first for both branches. */ + if (then_first_tail) + { + rtx from = then_first_tail; + if (!INSN_P (from)) + from = next_active_insn (from); + delete_insn_chain (from, BB_END (then_bb), false); + } + if (else_last_head) + delete_insn_chain (first_active_insn (else_bb), else_last_head, false); + merge_if_block (ce_info); cond_exec_changed_p = TRUE; return TRUE; @@ -627,6 +704,9 @@ struct noce_if_info 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); @@ -661,7 +741,15 @@ noce_emit_store_flag (struct noce_if_info *if_info, rtx x, int reversep, 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); @@ -895,12 +983,12 @@ noce_try_store_flag (struct noce_if_info *if_info) 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)) @@ -942,8 +1030,8 @@ noce_try_store_flag_constants (struct noce_if_info *if_info) 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); @@ -964,20 +1052,20 @@ noce_try_store_flag_constants (struct noce_if_info *if_info) 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; @@ -1108,7 +1196,7 @@ noce_try_addcc (struct noce_if_info *if_info) /* 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)) { @@ -1159,7 +1247,7 @@ noce_try_store_flag_mask (struct noce_if_info *if_info) 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)) @@ -1318,11 +1406,15 @@ noce_try_cmove_arith (struct noce_if_info *if_info) /* ??? 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; } @@ -1347,8 +1439,10 @@ noce_try_cmove_arith (struct noce_if_info *if_info) 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 @@ -1356,8 +1450,10 @@ noce_try_cmove_arith (struct noce_if_info *if_info) 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; } @@ -1469,6 +1565,9 @@ noce_try_cmove_arith (struct noce_if_info *if_info) 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) @@ -1526,7 +1625,7 @@ noce_get_alt_condition (struct noce_if_info *if_info, rtx target, 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); @@ -1536,20 +1635,22 @@ noce_get_alt_condition (struct noce_if_info *if_info, rtx target, /* 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; @@ -1561,7 +1662,7 @@ noce_get_alt_condition (struct noce_if_info *if_info, rtx target, /* 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); @@ -1730,13 +1831,20 @@ noce_try_minmax (struct noce_if_info *if_info) 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 @@ -1750,6 +1858,17 @@ noce_try_abs (struct noce_if_info *if_info) 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; @@ -1774,6 +1893,7 @@ noce_try_abs (struct noce_if_info *if_info) { 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)) { @@ -1820,13 +1940,23 @@ noce_try_abs (struct noce_if_info *if_info) } 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) { @@ -1856,7 +1986,7 @@ noce_try_sign_mask (struct noce_if_info *if_info) 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); @@ -1885,15 +2015,19 @@ noce_try_sign_mask (struct noce_if_info *if_info) 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 (); @@ -1951,7 +2085,7 @@ noce_try_bitop (struct noce_if_info *if_info) 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)); @@ -1969,7 +2103,7 @@ noce_try_bitop (struct noce_if_info *if_info) { /* 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; @@ -1995,7 +2129,7 @@ noce_try_bitop (struct noce_if_info *if_info) { /* 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; @@ -2073,7 +2207,7 @@ noce_get_condition (rtx jump, rtx *earliest, bool then_else_reversed) /* 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. */ @@ -2089,7 +2223,7 @@ noce_operand_ok (rtx op) /* 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; @@ -2120,7 +2254,7 @@ noce_mem_write_may_trap_or_fault_p (rtx mem) 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; @@ -2139,6 +2273,44 @@ noce_mem_write_may_trap_or_fault_p (rtx mem) 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. */ @@ -2196,22 +2368,25 @@ noce_process_if_block (struct noce_if_info *if_info) 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; } @@ -2229,15 +2404,15 @@ noce_process_if_block (struct noce_if_info *if_info) the lifetime of hard registers on small register class machines. */ orig_x = x; if (!REG_P (x) - || (SMALL_REGISTER_CLASSES - && REGNO (x) < FIRST_PSEUDO_REGISTER)) + || (HARD_REGISTER_P (x) + && targetm.small_register_classes_for_mode_p (GET_MODE (x)))) { 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 @@ -2248,6 +2423,7 @@ noce_process_if_block (struct noce_if_info *if_info) 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; @@ -2291,17 +2467,31 @@ noce_process_if_block (struct noce_if_info *if_info) 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; @@ -2316,7 +2506,7 @@ noce_process_if_block (struct noce_if_info *if_info) 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; @@ -2331,6 +2521,13 @@ noce_process_if_block (struct noce_if_info *if_info) goto success; } + if (!else_bb && set_b) + { + insn_b = set_b = NULL_RTX; + b = orig_x; + goto retry; + } + return FALSE; success: @@ -2383,7 +2580,8 @@ noce_process_if_block (struct noce_if_info *if_info) 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; @@ -2397,7 +2595,7 @@ check_cond_move_block (basic_block bb, rtx *vals, VEC (int, heap) *regs, rtx con { 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) @@ -2406,7 +2604,8 @@ check_cond_move_block (basic_block bb, rtx *vals, VEC (int, heap) *regs, rtx con dest = SET_DEST (set); src = SET_SRC (set); if (!REG_P (dest) - || (SMALL_REGISTER_CLASSES && HARD_REGISTER_P (dest))) + || (HARD_REGISTER_P (dest) + && targetm.small_register_classes_for_mode_p (GET_MODE (dest)))) return FALSE; if (!CONSTANT_P (src) && !register_operand (src, VOIDmode)) @@ -2444,7 +2643,7 @@ check_cond_move_block (basic_block bb, rtx *vals, VEC (int, heap) *regs, rtx con vals[REGNO (dest)] = src; - VEC_safe_push (int, heap, regs, REGNO (dest)); + VEC_safe_push (int, heap, *regs, REGNO (dest)); } return TRUE; @@ -2475,7 +2674,8 @@ cond_move_convert_if_block (struct noce_if_info *if_infop, 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))); @@ -2545,9 +2745,14 @@ cond_move_process_if_block (struct noce_if_info *if_info) 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 @@ -2568,7 +2773,11 @@ cond_move_process_if_block (struct noce_if_info *if_info) 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; + } } } @@ -2582,7 +2791,11 @@ cond_move_process_if_block (struct noce_if_info *if_info) 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. */ @@ -2594,11 +2807,17 @@ cond_move_process_if_block (struct noce_if_info *if_info) 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) @@ -2631,7 +2850,6 @@ cond_move_process_if_block (struct noce_if_info *if_info) VEC_free (int, heap, then_regs); VEC_free (int, heap, else_regs); - return TRUE; } @@ -2645,13 +2863,13 @@ cond_move_process_if_block (struct noce_if_info *if_info) Return TRUE if we were successful at converting the block. */ static int -noce_find_if_block (basic_block test_bb, - edge then_edge, edge else_edge, +noce_find_if_block (basic_block test_bb, edge then_edge, edge else_edge, int pass) { 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. */ @@ -2726,9 +2944,7 @@ noce_find_if_block (basic_block test_bb, return FALSE; /* If this is not a standard conditional jump, we can't parse it. */ - cond = noce_get_condition (jump, - &if_info.cond_earliest, - then_else_reversed); + cond = noce_get_condition (jump, &cond_earliest, then_else_reversed); if (!cond) return FALSE; @@ -2743,8 +2959,11 @@ noce_find_if_block (basic_block test_bb, 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. */ @@ -2773,6 +2992,7 @@ merge_if_block (struct ce_if_block * ce_info) /* 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. */ @@ -2847,7 +3067,8 @@ merge_if_block (struct ce_if_block * ce_info) 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++; } @@ -2887,6 +3108,11 @@ find_if_header (basic_block test_bb, int pass) 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)) @@ -2910,7 +3136,7 @@ find_if_header (basic_block test_bb, int pass) /* Otherwise this must be a multiway branch of some sort. */ return NULL; - memset (&ce_info, '\0', sizeof (ce_info)); + memset (&ce_info, 0, sizeof (ce_info)); ce_info.test_bb = test_bb; ce_info.then_bb = then_edge->dest; ce_info.else_bb = else_edge->dest; @@ -2920,20 +3146,22 @@ find_if_header (basic_block test_bb, int pass) IFCVT_INIT_EXTRA_FIELDS (&ce_info); #endif - if (! reload_completed + if (!reload_completed && noce_find_if_block (test_bb, then_edge, else_edge, pass)) goto success; - if (HAVE_conditional_execution && reload_completed + if (reload_completed + && targetm.have_conditional_execution () && 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)) + && (reload_completed || !targetm.have_conditional_execution ())) { if (find_if_case_1 (test_bb, then_edge, else_edge)) goto success; @@ -2946,6 +3174,8 @@ find_if_header (basic_block test_bb, int pass) 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; } @@ -3005,6 +3235,7 @@ block_jumps_and_fallthru_p (basic_block cur_bb, basic_block target_bb) if (INSN_P (insn) && !JUMP_P (insn) + && !DEBUG_INSN_P (insn) && GET_CODE (PATTERN (insn)) != USE && GET_CODE (PATTERN (insn)) != CLOBBER) n_insns++; @@ -3036,8 +3267,8 @@ cond_exec_find_if_block (struct ce_if_block * ce_info) ce_info->last_test_bb = test_bb; /* We only ever should get here after reload, - and only if we have conditional execution. */ - gcc_assert (HAVE_conditional_execution && reload_completed); + and if we have conditional execution. */ + gcc_assert (reload_completed && targetm.have_conditional_execution ()); /* 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 @@ -3118,7 +3349,8 @@ cond_exec_find_if_block (struct ce_if_block * ce_info) 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 @@ -3164,8 +3396,9 @@ cond_exec_find_if_block (struct ce_if_block * ce_info) else if (single_succ_p (else_bb) && 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))) + && !(single_succ_edge (else_bb)->flags & EDGE_COMPLEX) + && !(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. */ @@ -3303,8 +3536,8 @@ find_cond_trap (basic_block test_bb, edge then_edge, edge else_edge) } /* 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; @@ -3314,6 +3547,10 @@ find_cond_trap (basic_block test_bb, edge then_edge, edge else_edge) /* 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); @@ -3452,7 +3689,8 @@ static int 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 @@ -3497,7 +3735,9 @@ find_if_case_1 (basic_block test_bb, edge then_edge, edge else_edge) 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. */ @@ -3508,11 +3748,6 @@ find_if_case_1 (basic_block test_bb, edge then_edge, edge else_edge) /* 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. */ @@ -3526,7 +3761,10 @@ find_if_case_1 (basic_block test_bb, edge then_edge, edge else_edge) } 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); @@ -3535,15 +3773,12 @@ find_if_case_1 (basic_block test_bb, edge then_edge, edge else_edge) 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++; @@ -3616,7 +3851,9 @@ find_if_case_2 (basic_block test_bb, edge then_edge, edge else_edge) 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. */ @@ -3626,10 +3863,8 @@ find_if_case_2 (basic_block test_bb, edge then_edge, edge else_edge) /* 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++; @@ -3663,6 +3898,8 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, 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); @@ -3670,6 +3907,9 @@ dead_or_predicable (basic_block test_bb, basic_block merge_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 @@ -3679,6 +3919,8 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, 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) @@ -3687,6 +3929,8 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, goto no_body; } head = NEXT_INSN (head); + while (DEBUG_INSN_P (head) && head != end) + head = NEXT_INSN (head); } if (JUMP_P (end)) @@ -3697,12 +3941,14 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, 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 @@ -3731,22 +3977,24 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, 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, merge_set_noclobber, test_live, test_set; unsigned i, fail = 0; bitmap_iterator bi; @@ -3755,7 +4003,7 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, { if (CALL_P (insn)) return FALSE; - if (INSN_P (insn)) + if (NONDEBUG_INSN_P (insn)) { if (may_trap_p (PATTERN (insn))) return FALSE; @@ -3782,14 +4030,16 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, /* Collect: MERGE_SET = set of registers set in MERGE_BB + MERGE_SET_NOCLOBBER = like MERGE_SET, but only includes registers + that are really set, not just clobbered. TEST_LIVE = set of registers live at EARLIEST - TEST_SET = set of registers set between EARLIEST and the - end of the block. */ + 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); + merge_set_noclobber = 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, @@ -3798,17 +4048,23 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, 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)) + { + df_simulate_find_defs (insn, merge_set); + df_simulate_find_noclobber_defs (insn, merge_set_noclobber); + } } - propagate_block (merge_bb, tmp, merge_set, merge_set, 0); /* For small register class machines, don't lengthen lifetimes of hard registers before reload. */ - if (SMALL_REGISTER_CLASSES && ! reload_completed) + if (! reload_completed + && targetm.small_register_classes_for_mode_p (VOIDmode)) { - EXECUTE_IF_SET_IN_BITMAP (merge_set, 0, i, bi) + EXECUTE_IF_SET_IN_BITMAP (merge_set_noclobber, 0, i, bi) { if (i < FIRST_PSEUDO_REGISTER && ! fixed_regs[i] @@ -3820,35 +4076,39 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, /* 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) + MERGE_SET_NOCLOBBER & TEST_SET and - TEST_SET & merge_bb->il.rtl->global_live_at_start + MERGE_SET & TEST_LIVE) + and + TEST_SET & DF_LIVE_IN (merge_bb) are empty. */ - if (bitmap_intersect_p (test_set, merge_set) + if (bitmap_intersect_p (test_set, merge_set_noclobber) || 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_noclobber); + BITMAP_FREE (merge_set); + BITMAP_FREE (test_live); + BITMAP_FREE (test_set); if (fail) return FALSE; @@ -3869,8 +4129,10 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, 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) { @@ -3912,7 +4174,8 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, 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))); @@ -3938,34 +4201,30 @@ dead_or_predicable (basic_block test_bb, basic_block merge_bb, /* 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 @@ -3973,6 +4232,9 @@ if_convert (int x_life_data_ok) 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++; @@ -3983,14 +4245,20 @@ if_convert (int x_life_data_ok) 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); @@ -4007,19 +4275,9 @@ if_convert (int x_life_data_ok) 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) @@ -4035,6 +4293,9 @@ if_convert (int x_life_data_ok) num_true_changes); } + if (optimize == 1) + df_remove_problem (df_live); + #ifdef ENABLE_CHECKING verify_flow_info (); #endif @@ -4043,7 +4304,8 @@ if_convert (int x_life_data_ok) static bool gate_handle_if_conversion (void) { - return (optimize > 0); + return (optimize > 0) + && dbg_cnt (if_conversion); } /* If-conversion and CFG cleanup. */ @@ -4055,19 +4317,17 @@ rest_of_handle_if_conversion (void) 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 */ @@ -4079,14 +4339,16 @@ struct tree_opt_pass pass_rtl_ifcvt = 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); } @@ -4095,14 +4357,14 @@ gate_handle_if_after_combine (void) 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 */ @@ -4114,34 +4376,32 @@ struct tree_opt_pass pass_if_after_combine = 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 */ @@ -4153,7 +4413,8 @@ struct tree_opt_pass pass_if_after_reload = 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 */ + } };