X-Git-Url: http://git.sourceforge.jp/view?a=blobdiff_plain;f=gcc%2Fcfgcleanup.c;h=75f56519a26ca79ef847ea11b7d861c6e61f99ea;hb=cdabd5b29ab4814aabf463f9597e96859fff31ce;hp=0e13f7a07cf83430b08b0cf2160efef9820d2fef;hpb=554f270788f1a960804f94da4b28faed8fcf50b2;p=pf3gnuchains%2Fgcc-fork.git diff --git a/gcc/cfgcleanup.c b/gcc/cfgcleanup.c index 0e13f7a07cf..75f56519a26 100644 --- a/gcc/cfgcleanup.c +++ b/gcc/cfgcleanup.c @@ -1,12 +1,13 @@ /* Control flow optimization code for GNU compiler. Copyright (C) 1987, 1988, 1992, 1993, 1994, 1995, 1996, 1997, 1998, - 1999, 2000, 2001, 2002, 2003, 2004, 2005 Free Software Foundation, Inc. + 1999, 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) any later +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 ANY @@ -15,9 +16,8 @@ FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public 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 +. */ /* This file contains optimizer of the control flow. The main entry point is cleanup_cfg. Following optimizations are performed: @@ -53,15 +53,21 @@ Software Foundation, 51 Franklin Street, Fifth Floor, Boston, MA #include "tree-pass.h" #include "cfgloop.h" #include "expr.h" +#include "df.h" +#include "dce.h" +#include "dbgcnt.h" #define FORWARDER_BLOCK_P(BB) ((BB)->flags & BB_FORWARDER_BLOCK) /* Set to true when we are running first pass of try_optimize_cfg loop. */ static bool first_pass; + +/* Set to true if crossjumps occured in the latest run of try_optimize_cfg. */ +static bool crossjumps_occured; + static bool try_crossjump_to_edge (int, edge, edge); static bool try_crossjump_bb (int, basic_block); static bool outgoing_edges_match (int, basic_block, basic_block); -static int flow_find_cross_jump (int, basic_block, basic_block, rtx *, rtx *); static bool old_insns_match_p (int, rtx, rtx); static void merge_blocks_move_predecessor_nojumps (basic_block, basic_block); @@ -69,7 +75,7 @@ static void merge_blocks_move_successor_nojumps (basic_block, basic_block); static bool try_optimize_cfg (int); static bool try_simplify_condjump (basic_block); static bool try_forward_edges (int, basic_block); -static edge thread_jump (int, edge, basic_block); +static edge thread_jump (edge, basic_block); static bool mark_effect (rtx, bitmap); static void notice_new_block (basic_block); static void update_forwarder_flag (basic_block); @@ -259,7 +265,7 @@ mentions_nonequal_regs (rtx *x, void *data) if exist, NULL otherwise. */ static edge -thread_jump (int mode, edge e, basic_block b) +thread_jump (edge e, basic_block b) { rtx set1, set2, cond1, cond2, insn; enum rtx_code code1, code2, reversed_code2; @@ -330,7 +336,7 @@ thread_jump (int mode, edge e, basic_block b) return NULL; } - cselib_init (false); + cselib_init (0); /* First process all values computed in the source basic block. */ for (insn = NEXT_INSN (BB_HEAD (e->src)); @@ -379,11 +385,6 @@ thread_jump (int mode, edge e, basic_block b) if (for_each_rtx (&cond2, mentions_nonequal_regs, nonequal)) goto failed_exit; - /* In case liveness information is available, we need to prove equivalence - only of the live values. */ - if (mode & CLEANUP_UPDATE_LIFE) - AND_REG_SET (nonequal, b->il.rtl->global_live_at_end); - EXECUTE_IF_SET_IN_REG_SET (nonequal, 0, i, rsi) goto failed_exit; @@ -416,8 +417,8 @@ try_forward_edges (int mode, basic_block b) and cold sections. Basic block partitioning may result in some jumps that appear to - be optimizable (or blocks that appear to be mergeable), but which really m - ust be left untouched (they are required to make it safely across + be optimizable (or blocks that appear to be mergeable), but which really + must be left untouched (they are required to make it safely across partition boundaries). See the comments at the top of bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */ @@ -427,10 +428,10 @@ try_forward_edges (int mode, basic_block b) for (ei = ei_start (b->succs); (e = ei_safe_edge (ei)); ) { basic_block target, first; - int counter; + int counter, goto_locus; bool threaded = false; int nthreaded_edges = 0; - bool may_thread = first_pass | (b->flags & BB_DIRTY); + bool may_thread = first_pass | df_get_bb_dirty (b); /* Skip complex edges because we don't know how to update them. @@ -445,6 +446,7 @@ try_forward_edges (int mode, basic_block b) target = first = e->dest; counter = NUM_FIXED_BLOCKS; + goto_locus = e->goto_locus; /* If we are partitioning hot/cold basic_blocks, we don't want to mess up jumps that cross between hot/cold sections. @@ -464,7 +466,7 @@ try_forward_edges (int mode, basic_block b) { basic_block new_target = NULL; bool new_target_threaded = false; - may_thread |= target->flags & BB_DIRTY; + may_thread |= df_get_bb_dirty (target); if (FORWARDER_BLOCK_P (target) && !(single_succ_edge (target)->flags & EDGE_CROSSING) @@ -474,13 +476,35 @@ try_forward_edges (int mode, basic_block b) new_target = single_succ (target); if (target == new_target) counter = n_basic_blocks; + else if (!optimize) + { + /* When not optimizing, ensure that edges or forwarder + blocks with different locus are not optimized out. */ + int locus = single_succ_edge (target)->goto_locus; + + if (locus && goto_locus && !locator_eq (locus, goto_locus)) + counter = n_basic_blocks; + else if (locus) + goto_locus = locus; + + if (INSN_P (BB_END (target))) + { + locus = INSN_LOCATOR (BB_END (target)); + + if (locus && goto_locus + && !locator_eq (locus, goto_locus)) + counter = n_basic_blocks; + else if (locus) + goto_locus = locus; + } + } } /* Allow to thread only over one edge at time to simplify updating of probabilities. */ else if ((mode & CLEANUP_THREADING) && may_thread) { - edge t = thread_jump (mode, e, target); + edge t = thread_jump (e, target); if (t) { if (!threaded_edges) @@ -537,6 +561,8 @@ try_forward_edges (int mode, basic_block b) int edge_frequency; int n = 0; + e->goto_locus = goto_locus; + /* Don't force if target is exit block. */ if (threaded && target != EXIT_BLOCK_PTR) { @@ -622,7 +648,6 @@ static void merge_blocks_move_predecessor_nojumps (basic_block a, basic_block b) { rtx barrier; - bool only_notes; /* If we are partitioning hot/cold basic blocks, we don't want to mess up unconditional or indirect jumps that cross between hot @@ -641,20 +666,10 @@ merge_blocks_move_predecessor_nojumps (basic_block a, basic_block b) gcc_assert (BARRIER_P (barrier)); delete_insn (barrier); - /* Move block and loop notes out of the chain so that we do not - disturb their order. - - ??? A better solution would be to squeeze out all the non-nested notes - and adjust the block trees appropriately. Even better would be to have - a tighter connection between block trees and rtl so that this is not - necessary. */ - only_notes = squeeze_notes (&BB_HEAD (a), &BB_END (a)); - gcc_assert (!only_notes); - /* Scramble the insn chain. */ if (BB_END (a) != PREV_INSN (BB_HEAD (b))) reorder_insns_nobb (BB_HEAD (a), BB_END (a), PREV_INSN (BB_HEAD (b))); - a->flags |= BB_DIRTY; + df_set_bb_dirty (a); if (dump_file) fprintf (dump_file, "Moved block %d before %d and merged.\n", @@ -678,7 +693,6 @@ merge_blocks_move_successor_nojumps (basic_block a, basic_block b) { rtx barrier, real_b_end; rtx label, table; - bool only_notes; /* If we are partitioning hot/cold basic blocks, we don't want to mess up unconditional or indirect jumps that cross between hot @@ -708,16 +722,6 @@ merge_blocks_move_successor_nojumps (basic_block a, basic_block b) if (barrier && BARRIER_P (barrier)) delete_insn (barrier); - /* Move block and loop notes out of the chain so that we do not - disturb their order. - - ??? A better solution would be to squeeze out all the non-nested notes - and adjust the block trees appropriately. Even better would be to have - a tighter connection between block trees and rtl so that this is not - necessary. */ - only_notes = squeeze_notes (&BB_HEAD (b), &BB_END (b)); - gcc_assert (!only_notes); - /* Scramble the insn chain. */ reorder_insns_nobb (BB_HEAD (b), BB_END (b), BB_END (a)); @@ -763,8 +767,6 @@ merge_blocks_move (edge e, basic_block b, basic_block c, int mode) if (BB_PARTITION (b) != BB_PARTITION (c)) return NULL; - - /* If B has a fallthru edge to C, no need to move anything. */ if (e->flags & EDGE_FALLTHRU) { @@ -950,6 +952,11 @@ old_insns_match_p (int mode ATTRIBUTE_UNUSED, rtx i1, rtx i2) if (GET_CODE (i1) != GET_CODE (i2)) return false; + /* __builtin_unreachable() may lead to empty blocks (ending with + NOTE_INSN_BASIC_BLOCK). They may be crossjumped. */ + if (NOTE_INSN_BASIC_BLOCK_P (i1) && NOTE_INSN_BASIC_BLOCK_P (i2)) + return true; + p1 = PATTERN (i1); p2 = PATTERN (i2); @@ -964,13 +971,27 @@ old_insns_match_p (int mode ATTRIBUTE_UNUSED, rtx i1, rtx i2) be filled that clobbers a parameter expected by the subroutine. ??? We take the simple route for now and assume that if they're - equal, they were constructed identically. */ + equal, they were constructed identically. + + Also check for identical exception regions. */ - if (CALL_P (i1) - && (!rtx_equal_p (CALL_INSN_FUNCTION_USAGE (i1), + if (CALL_P (i1)) + { + /* Ensure the same EH region. */ + rtx n1 = find_reg_note (i1, REG_EH_REGION, 0); + rtx n2 = find_reg_note (i2, REG_EH_REGION, 0); + + if (!n1 && n2) + return false; + + if (n1 && (!n2 || XEXP (n1, 0) != XEXP (n2, 0))) + return false; + + if (!rtx_equal_p (CALL_INSN_FUNCTION_USAGE (i1), CALL_INSN_FUNCTION_USAGE (i2)) - || SIBLING_CALL_P (i1) != SIBLING_CALL_P (i2))) - return false; + || SIBLING_CALL_P (i1) != SIBLING_CALL_P (i2)) + return false; + } #ifdef STACK_REGS /* If cross_jump_death_matters is not 0, the insn's mode @@ -997,12 +1018,8 @@ old_insns_match_p (int mode ATTRIBUTE_UNUSED, rtx i1, rtx i2) if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0))) SET_HARD_REG_BIT (i2_regset, REGNO (XEXP (note, 0))); - GO_IF_HARD_REG_EQUAL (i1_regset, i2_regset, done); - - return false; - - done: - ; + if (!hard_reg_set_equal_p (i1_regset, i2_regset)) + return false; } #endif @@ -1010,43 +1027,32 @@ old_insns_match_p (int mode ATTRIBUTE_UNUSED, rtx i1, rtx i2) ? rtx_renumbered_equal_p (p1, p2) : rtx_equal_p (p1, p2)) return true; - /* Do not do EQUIV substitution after reload. First, we're undoing the - work of reload_cse. Second, we may be undoing the work of the post- - reload splitting pass. */ - /* ??? Possibly add a new phase switch variable that can be used by - targets to disallow the troublesome insns after splitting. */ - if (!reload_completed) - { - /* The following code helps take care of G++ cleanups. */ - rtx equiv1 = find_reg_equal_equiv_note (i1); - rtx equiv2 = find_reg_equal_equiv_note (i2); - - if (equiv1 && equiv2 - /* If the equivalences are not to a constant, they may - reference pseudos that no longer exist, so we can't - use them. */ - && (! reload_completed - || (CONSTANT_P (XEXP (equiv1, 0)) - && rtx_equal_p (XEXP (equiv1, 0), XEXP (equiv2, 0))))) - { - rtx s1 = single_set (i1); - rtx s2 = single_set (i2); - if (s1 != 0 && s2 != 0 - && rtx_renumbered_equal_p (SET_DEST (s1), SET_DEST (s2))) - { - validate_change (i1, &SET_SRC (s1), XEXP (equiv1, 0), 1); - validate_change (i2, &SET_SRC (s2), XEXP (equiv2, 0), 1); - if (! rtx_renumbered_equal_p (p1, p2)) - cancel_changes (0); - else if (apply_change_group ()) - return true; - } - } - } - return false; } +/* When comparing insns I1 and I2 in flow_find_cross_jump or + flow_find_head_matching_sequence, ensure the notes match. */ + +static void +merge_notes (rtx i1, rtx i2) +{ + /* If the merged insns have different REG_EQUAL notes, then + remove them. */ + rtx equiv1 = find_reg_equal_equiv_note (i1); + rtx equiv2 = find_reg_equal_equiv_note (i2); + + if (equiv1 && !equiv2) + remove_note (i1, equiv1); + else if (!equiv1 && equiv2) + remove_note (i2, equiv2); + else if (equiv1 && equiv2 + && !rtx_equal_p (XEXP (equiv1, 0), XEXP (equiv2, 0))) + { + remove_note (i1, equiv1); + remove_note (i2, equiv2); + } +} + /* Look through the insns at the end of BB1 and BB2 and find the longest sequence that are equivalent. Store the first insns for that sequence in *F1 and *F2 and return the sequence length. @@ -1054,9 +1060,8 @@ old_insns_match_p (int mode ATTRIBUTE_UNUSED, rtx i1, rtx i2) To simplify callers of this function, if the blocks match exactly, store the head of the blocks in *F1 and *F2. */ -static int -flow_find_cross_jump (int mode ATTRIBUTE_UNUSED, basic_block bb1, - basic_block bb2, rtx *f1, rtx *f2) +int +flow_find_cross_jump (basic_block bb1, basic_block bb2, rtx *f1, rtx *f2) { rtx i1, i2, last1, last2, afterlast1, afterlast2; int ninsns = 0; @@ -1087,16 +1092,16 @@ flow_find_cross_jump (int mode ATTRIBUTE_UNUSED, basic_block bb1, while (true) { /* Ignore notes. */ - while (!INSN_P (i1) && i1 != BB_HEAD (bb1)) + while (!NONDEBUG_INSN_P (i1) && i1 != BB_HEAD (bb1)) i1 = PREV_INSN (i1); - while (!INSN_P (i2) && i2 != BB_HEAD (bb2)) + while (!NONDEBUG_INSN_P (i2) && i2 != BB_HEAD (bb2)) i2 = PREV_INSN (i2); if (i1 == BB_HEAD (bb1) || i2 == BB_HEAD (bb2)) break; - if (!old_insns_match_p (mode, i1, i2)) + if (!old_insns_match_p (0, i1, i2)) break; merge_memattrs (i1, i2); @@ -1104,21 +1109,7 @@ flow_find_cross_jump (int mode ATTRIBUTE_UNUSED, basic_block bb1, /* Don't begin a cross-jump with a NOTE insn. */ if (INSN_P (i1)) { - /* If the merged insns have different REG_EQUAL notes, then - remove them. */ - rtx equiv1 = find_reg_equal_equiv_note (i1); - rtx equiv2 = find_reg_equal_equiv_note (i2); - - if (equiv1 && !equiv2) - remove_note (i1, equiv1); - else if (!equiv1 && equiv2) - remove_note (i2, equiv2); - else if (equiv1 && equiv2 - && !rtx_equal_p (XEXP (equiv1, 0), XEXP (equiv2, 0))) - { - remove_note (i1, equiv1); - remove_note (i2, equiv2); - } + merge_notes (i1, i2); afterlast1 = last1, afterlast2 = last2; last1 = i1, last2 = i2; @@ -1141,13 +1132,13 @@ flow_find_cross_jump (int mode ATTRIBUTE_UNUSED, basic_block bb1, Two, it keeps line number notes as matched as may be. */ if (ninsns) { - while (last1 != BB_HEAD (bb1) && !INSN_P (PREV_INSN (last1))) + while (last1 != BB_HEAD (bb1) && !NONDEBUG_INSN_P (PREV_INSN (last1))) last1 = PREV_INSN (last1); if (last1 != BB_HEAD (bb1) && LABEL_P (PREV_INSN (last1))) last1 = PREV_INSN (last1); - while (last2 != BB_HEAD (bb2) && !INSN_P (PREV_INSN (last2))) + while (last2 != BB_HEAD (bb2) && !NONDEBUG_INSN_P (PREV_INSN (last2))) last2 = PREV_INSN (last2); if (last2 != BB_HEAD (bb2) && LABEL_P (PREV_INSN (last2))) @@ -1160,132 +1151,95 @@ flow_find_cross_jump (int mode ATTRIBUTE_UNUSED, basic_block bb1, return ninsns; } -/* Return true iff the condbranches at the end of BB1 and BB2 match. */ -bool -condjump_equiv_p (struct equiv_info *info, bool call_init) +/* Like flow_find_cross_jump, except start looking for a matching sequence from + the head of the two blocks. Do not include jumps at the end. + If STOP_AFTER is nonzero, stop after finding that many matching + instructions. */ + +int +flow_find_head_matching_sequence (basic_block bb1, basic_block bb2, rtx *f1, + rtx *f2, int stop_after) { - basic_block bb1 = info->x_block; - basic_block bb2 = info->y_block; - edge b1 = BRANCH_EDGE (bb1); - edge b2 = BRANCH_EDGE (bb2); - edge f1 = FALLTHRU_EDGE (bb1); - edge f2 = FALLTHRU_EDGE (bb2); - bool reverse, match; - rtx set1, set2, cond1, cond2; - rtx src1, src2; - enum rtx_code code1, code2; - - /* Get around possible forwarders on fallthru edges. Other cases - should be optimized out already. */ - if (FORWARDER_BLOCK_P (f1->dest)) - f1 = single_succ_edge (f1->dest); - - if (FORWARDER_BLOCK_P (f2->dest)) - f2 = single_succ_edge (f2->dest); - - /* To simplify use of this function, return false if there are - unneeded forwarder blocks. These will get eliminated later - during cleanup_cfg. */ - if (FORWARDER_BLOCK_P (f1->dest) - || FORWARDER_BLOCK_P (f2->dest) - || FORWARDER_BLOCK_P (b1->dest) - || FORWARDER_BLOCK_P (b2->dest)) - return false; + rtx i1, i2, last1, last2, beforelast1, beforelast2; + int ninsns = 0; + edge e; + edge_iterator ei; + int nehedges1 = 0, nehedges2 = 0; - if (f1->dest == f2->dest && b1->dest == b2->dest) - reverse = false; - else if (f1->dest == b2->dest && b1->dest == f2->dest) - reverse = true; - else - return false; + FOR_EACH_EDGE (e, ei, bb1->succs) + if (e->flags & EDGE_EH) + nehedges1++; + FOR_EACH_EDGE (e, ei, bb2->succs) + if (e->flags & EDGE_EH) + nehedges2++; - set1 = pc_set (BB_END (bb1)); - set2 = pc_set (BB_END (bb2)); - if ((XEXP (SET_SRC (set1), 1) == pc_rtx) - != (XEXP (SET_SRC (set2), 1) == pc_rtx)) - reverse = !reverse; - - src1 = SET_SRC (set1); - src2 = SET_SRC (set2); - cond1 = XEXP (src1, 0); - cond2 = XEXP (src2, 0); - code1 = GET_CODE (cond1); - if (reverse) - code2 = reversed_comparison_code (cond2, BB_END (bb2)); - else - code2 = GET_CODE (cond2); + i1 = BB_HEAD (bb1); + i2 = BB_HEAD (bb2); + last1 = beforelast1 = last2 = beforelast2 = NULL_RTX; - if (code2 == UNKNOWN) - return false; - - if (call_init && !struct_equiv_init (STRUCT_EQUIV_START | info->mode, info)) - gcc_unreachable (); - /* Make the sources of the pc sets unreadable so that when we call - insns_match_p it won't process them. - The death_notes_match_p from insns_match_p won't see the local registers - used for the pc set, but that could only cause missed optimizations when - there are actually condjumps that use stack registers. */ - SET_SRC (set1) = pc_rtx; - SET_SRC (set2) = pc_rtx; - /* Verify codes and operands match. */ - if (code1 == code2) + while (true) { - match = (insns_match_p (BB_END (bb1), BB_END (bb2), info) - && rtx_equiv_p (&XEXP (cond1, 0), XEXP (cond2, 0), 1, info) - && rtx_equiv_p (&XEXP (cond1, 1), XEXP (cond2, 1), 1, info)); - } - else if (code1 == swap_condition (code2)) - { - match = (insns_match_p (BB_END (bb1), BB_END (bb2), info) - && rtx_equiv_p (&XEXP (cond1, 1), XEXP (cond2, 0), 1, info) - && rtx_equiv_p (&XEXP (cond1, 0), XEXP (cond2, 1), 1, info)); + /* Ignore notes. */ + while (!NONDEBUG_INSN_P (i1) && i1 != BB_END (bb1)) + i1 = NEXT_INSN (i1); - } - else - match = false; - SET_SRC (set1) = src1; - SET_SRC (set2) = src2; - match &= verify_changes (0); - - /* If we return true, we will join the blocks. Which means that - we will only have one branch prediction bit to work with. Thus - we require the existing branches to have probabilities that are - roughly similar. */ - if (match - && !optimize_size - && maybe_hot_bb_p (bb1) - && maybe_hot_bb_p (bb2)) - { - int prob2; + while (!NONDEBUG_INSN_P (i2) && i2 != BB_END (bb2)) + i2 = NEXT_INSN (i2); - if (b1->dest == b2->dest) - prob2 = b2->probability; - else - /* Do not use f2 probability as f2 may be forwarded. */ - prob2 = REG_BR_PROB_BASE - b2->probability; + if (NOTE_P (i1) || NOTE_P (i2) + || JUMP_P (i1) || JUMP_P (i2)) + break; - /* Fail if the difference in probabilities is greater than 50%. - This rules out two well-predicted branches with opposite - outcomes. */ - if (abs (b1->probability - prob2) > REG_BR_PROB_BASE / 2) + /* A sanity check to make sure we're not merging insns with different + effects on EH. If only one of them ends a basic block, it shouldn't + have an EH edge; if both end a basic block, there should be the same + number of EH edges. */ + if ((i1 == BB_END (bb1) && i2 != BB_END (bb2) + && nehedges1 > 0) + || (i2 == BB_END (bb2) && i1 != BB_END (bb1) + && nehedges2 > 0) + || (i1 == BB_END (bb1) && i2 == BB_END (bb2) + && nehedges1 != nehedges2)) + break; + + if (!old_insns_match_p (0, i1, i2)) + break; + + merge_memattrs (i1, i2); + + /* Don't begin a cross-jump with a NOTE insn. */ + if (INSN_P (i1)) { - if (dump_file) - fprintf (dump_file, - "Outcomes of branch in bb %i and %i differ too much (%i %i)\n", - bb1->index, bb2->index, b1->probability, prob2); + merge_notes (i1, i2); - match = false; + beforelast1 = last1, beforelast2 = last2; + last1 = i1, last2 = i2; + ninsns++; } + + if (i1 == BB_END (bb1) || i2 == BB_END (bb2) + || (stop_after > 0 && ninsns == stop_after)) + break; + + i1 = NEXT_INSN (i1); + i2 = NEXT_INSN (i2); } - if (dump_file && match) - fprintf (dump_file, "Conditionals in bb %i and %i match.\n", - bb1->index, bb2->index); +#ifdef HAVE_cc0 + /* Don't allow a compare to be shared by cross-jumping unless the insn + after the compare is also shared. */ + if (ninsns && reg_mentioned_p (cc0_rtx, last1) && sets_cc0_p (last1)) + last1 = beforelast1, last2 = beforelast2, ninsns--; +#endif + + if (ninsns) + { + *f1 = last1; + *f2 = last2; + } - if (!match) - cancel_changes (0); - return match; + return ninsns; } /* Return true iff outgoing edges of BB1 and BB2 match, together with @@ -1389,9 +1343,8 @@ outgoing_edges_match (int mode, basic_block bb1, basic_block bb2) we require the existing branches to have probabilities that are roughly similar. */ if (match - && !optimize_size - && maybe_hot_bb_p (bb1) - && maybe_hot_bb_p (bb2)) + && optimize_bb_for_speed_p (bb1) + && optimize_bb_for_speed_p (bb2)) { int prob2; @@ -1617,7 +1570,7 @@ try_crossjump_to_edge (int mode, edge e1, edge e2) partition boundaries). See the comments at the top of bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */ - if (flag_reorder_blocks_and_partition && no_new_pseudos) + if (flag_reorder_blocks_and_partition && reload_completed) return false; /* Search backward through forwarder blocks. We don't need to worry @@ -1657,7 +1610,7 @@ try_crossjump_to_edge (int mode, edge e1, edge e2) return false; /* ... and part the second. */ - nmatch = flow_find_cross_jump (mode, src1, src2, &newpos1, &newpos2); + nmatch = flow_find_cross_jump (src1, src2, &newpos1, &newpos2); /* Don't proceed with the crossjump unless we found a sufficient number of matching instructions or the 'from' block was totally matched @@ -1716,8 +1669,12 @@ try_crossjump_to_edge (int mode, edge e1, edge e2) /* Skip possible basic block header. */ if (LABEL_P (newpos2)) newpos2 = NEXT_INSN (newpos2); + while (DEBUG_INSN_P (newpos2)) + newpos2 = NEXT_INSN (newpos2); if (NOTE_P (newpos2)) newpos2 = NEXT_INSN (newpos2); + while (DEBUG_INSN_P (newpos2)) + newpos2 = NEXT_INSN (newpos2); } if (dump_file) @@ -1731,10 +1688,8 @@ try_crossjump_to_edge (int mode, edge e1, edge e2) "Cross jumping from bb %i to bb %i; %i common insns\n", src1->index, src2->index, nmatch); - redirect_to->count += src1->count; - redirect_to->frequency += src1->frequency; /* We may have some registers visible through the block. */ - redirect_to->flags |= BB_DIRTY; + df_set_bb_dirty (redirect_to); /* Recompute the frequencies and counts of outgoing edges. */ FOR_EACH_EDGE (s, ei, redirect_to->succs) @@ -1789,6 +1744,14 @@ try_crossjump_to_edge (int mode, edge e1, edge e2) / (redirect_to->frequency + src1->frequency)); } + /* Adjust count and frequency for the block. An earlier jump + threading pass may have left the profile in an inconsistent + state (see update_bb_profile_for_threading) so we must be + prepared for overflows. */ + redirect_to->count += src1->count; + redirect_to->frequency += src1->frequency; + if (redirect_to->frequency > BB_FREQ_MAX) + redirect_to->frequency = BB_FREQ_MAX; update_br_prob_note (redirect_to); /* Edit SRC1 to go to REDIRECT_TO at NEWPOS1. */ @@ -1797,7 +1760,13 @@ try_crossjump_to_edge (int mode, edge e1, edge e2) if (LABEL_P (newpos1)) newpos1 = NEXT_INSN (newpos1); - if (NOTE_P (newpos1)) + while (DEBUG_INSN_P (newpos1)) + newpos1 = NEXT_INSN (newpos1); + + if (NOTE_INSN_BASIC_BLOCK_P (newpos1)) + newpos1 = NEXT_INSN (newpos1); + + while (DEBUG_INSN_P (newpos1)) newpos1 = NEXT_INSN (newpos1); redirect_from = split_block (src1, PREV_INSN (newpos1))->src; @@ -1832,7 +1801,7 @@ try_crossjump_bb (int mode, basic_block bb) /* Don't crossjump if this block ends in a computed jump, unless we are optimizing for size. */ - if (!optimize_size + if (optimize_bb_for_size_p (bb) && bb != EXIT_BLOCK_PTR && computed_jump_p (BB_END (bb))) return false; @@ -1864,7 +1833,10 @@ try_crossjump_bb (int mode, basic_block bb) FOR_EACH_EDGE (e, ei, bb->preds) { if (e->flags & EDGE_FALLTHRU) - fallthru = e; + { + fallthru = e; + break; + } } changed = false; @@ -1873,7 +1845,8 @@ try_crossjump_bb (int mode, basic_block bb) e = EDGE_PRED (ev, ix); ix++; - /* As noted above, first try with the fallthru predecessor. */ + /* As noted above, first try with the fallthru predecessor (or, a + fallthru predecessor if we are in cfglayout mode). */ if (fallthru) { /* Don't combine the fallthru edge into anything else. @@ -1883,8 +1856,8 @@ try_crossjump_bb (int mode, basic_block bb) /* If nothing changed since the last attempt, there is nothing we can do. */ if (!first_pass - && (!(e->src->flags & BB_DIRTY) - && !(fallthru->src->flags & BB_DIRTY))) + && (!(df_get_bb_dirty (e->src)) + && !(df_get_bb_dirty (fallthru->src)))) continue; if (try_crossjump_to_edge (mode, e, fallthru)) @@ -1933,8 +1906,8 @@ try_crossjump_bb (int mode, basic_block bb) /* If nothing changed since the last attempt, there is nothing we can do. */ if (!first_pass - && (!(e->src->flags & BB_DIRTY) - && !(e2->src->flags & BB_DIRTY))) + && (!(df_get_bb_dirty (e->src)) + && !(df_get_bb_dirty (e2->src)))) continue; if (try_crossjump_to_edge (mode, e, e2)) @@ -1947,9 +1920,30 @@ try_crossjump_bb (int mode, basic_block bb) } } + if (changed) + crossjumps_occured = true; + return changed; } +/* Return true if BB contains just bb note, or bb note followed + by only DEBUG_INSNs. */ + +static bool +trivially_empty_bb_p (basic_block bb) +{ + rtx insn = BB_END (bb); + + while (1) + { + if (insn == BB_HEAD (bb)) + return true; + if (!DEBUG_INSN_P (insn)) + return false; + insn = PREV_INSN (insn); + } +} + /* Do simple CFG optimizations - basic block merging, simplifying of jump instructions etc. Return nonzero if changes were made. */ @@ -1961,12 +1955,11 @@ try_optimize_cfg (int mode) int iterations = 0; basic_block bb, b, next; - if (mode & CLEANUP_CROSSJUMP) - add_noreturn_fake_exit_edges (); - - if (mode & (CLEANUP_UPDATE_LIFE | CLEANUP_CROSSJUMP | CLEANUP_THREADING)) + if (mode & (CLEANUP_CROSSJUMP | CLEANUP_THREADING)) clear_bb_flags (); + crossjumps_occured = false; + FOR_EACH_BB (bb) update_forwarder_flag (bb); @@ -1992,18 +1985,45 @@ try_optimize_cfg (int mode) edge s; bool changed_here = false; - /* Delete trivially dead basic blocks. */ - while (EDGE_COUNT (b->preds) == 0) + /* Delete trivially dead basic blocks. This is either + blocks with no predecessors, or empty blocks with no + successors. However if the empty block with no + successors is the successor of the ENTRY_BLOCK, it is + kept. This ensures that the ENTRY_BLOCK will have a + successor which is a precondition for many RTL + passes. Empty blocks may result from expanding + __builtin_unreachable (). */ + if (EDGE_COUNT (b->preds) == 0 + || (EDGE_COUNT (b->succs) == 0 + && trivially_empty_bb_p (b) + && single_succ_edge (ENTRY_BLOCK_PTR)->dest != b)) { c = b->prev_bb; - if (dump_file) - fprintf (dump_file, "Deleting block %i.\n", - b->index); - + if ((mode & CLEANUP_CFGLAYOUT) + && EDGE_COUNT (b->preds) > 0 + && b->il.rtl->footer + && BARRIER_P (b->il.rtl->footer)) + { + edge e; + edge_iterator ei; + + FOR_EACH_EDGE (e, ei, b->preds) + if (e->flags & EDGE_FALLTHRU) + { + if (e->src->il.rtl->footer == NULL) + { + e->src->il.rtl->footer = b->il.rtl->footer; + b->il.rtl->footer = NULL; + } + break; + } + } delete_basic_block (b); if (!(mode & CLEANUP_CFGLAYOUT)) changed = true; - b = c; + /* Avoid trying to remove ENTRY_BLOCK_PTR. */ + b = (c == ENTRY_BLOCK_PTR ? c->next_bb : c); + continue; } /* Remove code labels no longer used. */ @@ -2024,15 +2044,17 @@ try_optimize_cfg (int mode) { rtx label = BB_HEAD (b); - delete_insn_chain (label, label); - /* In the case label is undeletable, move it after the + delete_insn_chain (label, label, false); + /* If the case label is undeletable, move it after the BASIC_BLOCK note. */ - if (NOTE_LINE_NUMBER (BB_HEAD (b)) == NOTE_INSN_DELETED_LABEL) + if (NOTE_KIND (BB_HEAD (b)) == NOTE_INSN_DELETED_LABEL) { rtx bb_note = NEXT_INSN (BB_HEAD (b)); reorder_insns_nobb (label, label, bb_note); BB_HEAD (b) = bb_note; + if (BB_END (b) == bb_note) + BB_END (b) = label; } if (dump_file) fprintf (dump_file, "Deleted label in block %i.\n", @@ -2061,6 +2083,7 @@ try_optimize_cfg (int mode) delete_basic_block (b); changed = true; b = c; + continue; } if (single_succ_p (b) @@ -2153,9 +2176,6 @@ try_optimize_cfg (int mode) while (changed); } - if (mode & CLEANUP_CROSSJUMP) - remove_fake_exit_edges (); - FOR_ALL_BB (b) b->flags &= ~(BB_FORWARDER_BLOCK | BB_NONTHREADABLE_BLOCK); @@ -2168,20 +2188,64 @@ bool delete_unreachable_blocks (void) { bool changed = false; - basic_block b, next_bb; + basic_block b, prev_bb; find_unreachable_blocks (); - /* Delete all unreachable basic blocks. */ - - for (b = ENTRY_BLOCK_PTR->next_bb; b != EXIT_BLOCK_PTR; b = next_bb) + /* When we're in GIMPLE mode and there may be debug insns, we should + delete blocks in reverse dominator order, so as to get a chance + to substitute all released DEFs into debug stmts. If we don't + have dominators information, walking blocks backward gets us a + better chance of retaining most debug information than + otherwise. */ + if (MAY_HAVE_DEBUG_STMTS && current_ir_type () == IR_GIMPLE + && dom_info_available_p (CDI_DOMINATORS)) { - next_bb = b->next_bb; + for (b = EXIT_BLOCK_PTR->prev_bb; b != ENTRY_BLOCK_PTR; b = prev_bb) + { + prev_bb = b->prev_bb; + + if (!(b->flags & BB_REACHABLE)) + { + /* Speed up the removal of blocks that don't dominate + others. Walking backwards, this should be the common + case. */ + if (!first_dom_son (CDI_DOMINATORS, b)) + delete_basic_block (b); + else + { + VEC (basic_block, heap) *h + = get_all_dominated_blocks (CDI_DOMINATORS, b); + + while (VEC_length (basic_block, h)) + { + b = VEC_pop (basic_block, h); - if (!(b->flags & BB_REACHABLE)) + prev_bb = b->prev_bb; + + gcc_assert (!(b->flags & BB_REACHABLE)); + + delete_basic_block (b); + } + + VEC_free (basic_block, heap, h); + } + + changed = true; + } + } + } + else + { + for (b = EXIT_BLOCK_PTR->prev_bb; b != ENTRY_BLOCK_PTR; b = prev_bb) { - delete_basic_block (b); - changed = true; + prev_bb = b->prev_bb; + + if (!(b->flags & BB_REACHABLE)) + { + delete_basic_block (b); + changed = true; + } } } @@ -2190,30 +2254,44 @@ delete_unreachable_blocks (void) return changed; } -/* Merges sequential blocks if possible. */ - -bool -merge_seq_blocks (void) +/* Delete any jump tables never referenced. We can't delete them at the + time of removing tablejump insn as they are referenced by the preceding + insns computing the destination, so we delay deleting and garbagecollect + them once life information is computed. */ +void +delete_dead_jumptables (void) { basic_block bb; - bool changed = false; - for (bb = ENTRY_BLOCK_PTR->next_bb; bb != EXIT_BLOCK_PTR; ) + /* A dead jump table does not belong to any basic block. Scan insns + between two adjacent basic blocks. */ + FOR_EACH_BB (bb) { - if (single_succ_p (bb) - && can_merge_blocks_p (bb, single_succ (bb))) + rtx insn, next; + + for (insn = NEXT_INSN (BB_END (bb)); + insn && !NOTE_INSN_BASIC_BLOCK_P (insn); + insn = next) { - /* Merge the blocks and retry. */ - merge_blocks (bb, single_succ (bb)); - changed = true; - continue; - } + next = NEXT_INSN (insn); + if (LABEL_P (insn) + && LABEL_NUSES (insn) == LABEL_PRESERVE_P (insn) + && JUMP_TABLE_DATA_P (next)) + { + rtx label = insn, jump = next; - bb = bb->next_bb; - } + if (dump_file) + fprintf (dump_file, "Dead jumptable %i removed\n", + INSN_UID (insn)); - return changed; + next = NEXT_INSN (next); + delete_insn (jump); + delete_insn (label); + } + } + } } + /* Tidy the CFG by deleting unreachable code and whatnot. */ @@ -2222,47 +2300,75 @@ cleanup_cfg (int mode) { bool changed = false; + /* Set the cfglayout mode flag here. We could update all the callers + but that is just inconvenient, especially given that we eventually + want to have cfglayout mode as the default. */ + if (current_ir_type () == IR_RTL_CFGLAYOUT) + mode |= CLEANUP_CFGLAYOUT; + timevar_push (TV_CLEANUP_CFG); if (delete_unreachable_blocks ()) { changed = true; /* We've possibly created trivially dead code. Cleanup it right now to introduce more opportunities for try_optimize_cfg. */ - if (!(mode & (CLEANUP_NO_INSN_DEL | CLEANUP_UPDATE_LIFE)) + if (!(mode & (CLEANUP_NO_INSN_DEL)) && !reload_completed) - delete_trivially_dead_insns (get_insns(), max_reg_num ()); + delete_trivially_dead_insns (get_insns (), max_reg_num ()); } compact_blocks (); + /* To tail-merge blocks ending in the same noreturn function (e.g. + a call to abort) we have to insert fake edges to exit. Do this + here once. The fake edges do not interfere with any other CFG + cleanups. */ + if (mode & CLEANUP_CROSSJUMP) + add_noreturn_fake_exit_edges (); + + if (!dbg_cnt (cfg_cleanup)) + return changed; + while (try_optimize_cfg (mode)) { delete_unreachable_blocks (), changed = true; - if (mode & CLEANUP_UPDATE_LIFE) + if (!(mode & CLEANUP_NO_INSN_DEL)) { - /* Cleaning up CFG introduces more opportunities for dead code - removal that in turn may introduce more opportunities for - cleaning up the CFG. */ - if (!update_life_info_in_dirty_blocks (UPDATE_LIFE_GLOBAL_RM_NOTES, - PROP_DEATH_NOTES - | PROP_SCAN_DEAD_CODE - | PROP_KILL_DEAD_CODE - | ((mode & CLEANUP_LOG_LINKS) - ? PROP_LOG_LINKS : 0))) - break; - } - else if (!(mode & CLEANUP_NO_INSN_DEL) - && (mode & CLEANUP_EXPENSIVE) - && !reload_completed) - { - if (!delete_trivially_dead_insns (get_insns(), max_reg_num ())) + /* Try to remove some trivially dead insns when doing an expensive + cleanup. But delete_trivially_dead_insns doesn't work after + reload (it only handles pseudos) and run_fast_dce is too costly + to run in every iteration. + + For effective cross jumping, we really want to run a fast DCE to + clean up any dead conditions, or they get in the way of performing + useful tail merges. + + Other transformations in cleanup_cfg are not so sensitive to dead + code, so delete_trivially_dead_insns or even doing nothing at all + is good enough. */ + if ((mode & CLEANUP_EXPENSIVE) && !reload_completed + && !delete_trivially_dead_insns (get_insns (), max_reg_num ())) break; + else if ((mode & CLEANUP_CROSSJUMP) + && crossjumps_occured) + run_fast_dce (); } else break; - delete_dead_jumptables (); } + if (mode & CLEANUP_CROSSJUMP) + remove_fake_exit_edges (); + + /* Don't call delete_dead_jumptables in cfglayout mode, because + that function assumes that jump tables are in the insns stream. + But we also don't _have_ to delete dead jumptables in cfglayout + mode because we shouldn't even be looking at things that are + not in a basic block. Dead jumptables are cleaned up when + going out of cfglayout mode. */ + if (!(mode & CLEANUP_CFGLAYOUT)) + delete_dead_jumptables (); + timevar_pop (TV_CLEANUP_CFG); return changed; @@ -2271,15 +2377,15 @@ cleanup_cfg (int mode) static unsigned int rest_of_handle_jump (void) { - delete_unreachable_blocks (); - - if (cfun->tail_call_emit) + if (crtl->tail_call_emit) fixup_tail_calls (); return 0; } -struct tree_opt_pass pass_jump = +struct rtl_opt_pass pass_jump = { + { + RTL_PASS, "sibling", /* name */ NULL, /* gate */ rest_of_handle_jump, /* execute */ @@ -2291,44 +2397,27 @@ struct tree_opt_pass pass_jump = 0, /* properties_provided */ 0, /* properties_destroyed */ TODO_ggc_collect, /* todo_flags_start */ - TODO_dump_func | TODO_verify_flow, /* todo_flags_finish */ - 'i' /* letter */ + } }; static unsigned int rest_of_handle_jump2 (void) { - /* Turn NOTE_INSN_EXPECTED_VALUE into REG_BR_PROB. Do this - before jump optimization switches branch directions. */ - if (flag_guess_branch_prob) - expected_value_to_br_prob (); - delete_trivially_dead_insns (get_insns (), max_reg_num ()); - reg_scan (get_insns (), max_reg_num ()); if (dump_file) dump_flow_info (dump_file, dump_flags); cleanup_cfg ((optimize ? CLEANUP_EXPENSIVE : 0) | (flag_thread_jumps ? CLEANUP_THREADING : 0)); - - purge_line_number_notes (); - - if (optimize) - cleanup_cfg (CLEANUP_EXPENSIVE); - - /* Jump optimization, and the removal of NULL pointer checks, may - have reduced the number of instructions substantially. CSE, and - future passes, allocate arrays whose dimensions involve the - maximum instruction UID, so if we can reduce the maximum UID - we'll save big on memory. */ - renumber_insns (); return 0; } -struct tree_opt_pass pass_jump2 = +struct rtl_opt_pass pass_jump2 = { + { + RTL_PASS, "jump", /* name */ NULL, /* gate */ rest_of_handle_jump2, /* execute */ @@ -2340,8 +2429,8 @@ struct tree_opt_pass pass_jump2 = 0, /* properties_provided */ 0, /* properties_destroyed */ TODO_ggc_collect, /* todo_flags_start */ - TODO_dump_func, /* todo_flags_finish */ - 'j' /* letter */ + TODO_dump_func | TODO_verify_rtl_sharing,/* todo_flags_finish */ + } };