X-Git-Url: http://git.sourceforge.jp/view?a=blobdiff_plain;f=gcc%2Fcfgcleanup.c;h=203963e02ccfb6764a8bbf0e27d6fbd5f9f4e265;hb=3992540609daf52db67593a2f4d2c95a537b991d;hp=8443451a42f6299841394fb244aa36cae3b5207b;hpb=4d2e5d528864f897011ae82014889955d1558c97;p=pf3gnuchains%2Fgcc-fork.git diff --git a/gcc/cfgcleanup.c b/gcc/cfgcleanup.c index 8443451a42f..203963e02cc 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 + 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: @@ -53,25 +53,35 @@ 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 *, struct equiv_info *); +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); 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); static int mentions_nonequal_regs (rtx *, void *); +static void merge_memattrs (rtx, rtx); /* Set flags for newly created block. */ @@ -131,12 +141,12 @@ try_simplify_condjump (basic_block cbranch_block) /* If we are partitioning hot/cold basic blocks, we don't want to mess up unconditional or indirect jumps that cross between hot - and cold sections. + 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 - must be left untouched (they are required to make it safely across - partition boundaries). See the comments at the top of + 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. */ if (BB_PARTITION (jump_block) != BB_PARTITION (jump_dest_block) @@ -189,7 +199,7 @@ mark_effect (rtx exp, regset nonequal) switch (GET_CODE (exp)) { /* In case we do clobber the register, mark it as equal, as we know the - value is dead so it don't have to match. */ + value is dead so it don't have to match. */ case CLOBBER: if (REG_P (XEXP (exp, 0))) { @@ -256,7 +266,7 @@ 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; @@ -376,11 +386,6 @@ 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; @@ -410,12 +415,12 @@ try_forward_edges (int mode, basic_block b) /* If we are partitioning hot/cold basic blocks, we don't want to mess up unconditional or indirect jumps that cross between hot - and cold sections. - + 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 - partition boundaries). See the comments at the top of + ust 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. */ if (find_reg_note (BB_END (b), REG_CROSSING_JUMP, NULL_RTX)) @@ -427,13 +432,13 @@ try_forward_edges (int mode, basic_block b) int counter; 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. - Still handle fallthru edges, as we can succeed to forward fallthru - edge to the same place as the branch edge of conditional branch - and turn conditional branch to an unconditional branch. */ + Still handle fallthru edges, as we can succeed to forward fallthru + edge to the same place as the branch edge of conditional branch + and turn conditional branch to an unconditional branch. */ if (e->flags & EDGE_COMPLEX) { ei_next (&ei); @@ -447,8 +452,8 @@ try_forward_edges (int mode, basic_block b) up jumps that cross between hot/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 must be left untouched (they are required to make it safely + to 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. */ @@ -461,10 +466,10 @@ 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) + && !(single_succ_edge (target)->flags & EDGE_CROSSING) && single_succ (target) != EXIT_BLOCK_PTR) { /* Bypass trivial infinite loops. */ @@ -477,12 +482,11 @@ try_forward_edges (int mode, basic_block b) 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) - threaded_edges = xmalloc (sizeof (*threaded_edges) - * n_basic_blocks); + threaded_edges = XNEWVEC (edge, n_basic_blocks); else { int i; @@ -514,39 +518,6 @@ try_forward_edges (int mode, basic_block b) if (!new_target) break; - /* Avoid killing of loop pre-headers, as it is the place loop - optimizer wants to hoist code to. - - For fallthru forwarders, the LOOP_BEG note must appear between - the header of block and CODE_LABEL of the loop, for non forwarders - it must appear before the JUMP_INSN. */ - if ((mode & CLEANUP_PRE_LOOP) && optimize && flag_loop_optimize) - { - rtx insn = (EDGE_SUCC (target, 0)->flags & EDGE_FALLTHRU - ? BB_HEAD (target) : prev_nonnote_insn (BB_END (target))); - - if (!NOTE_P (insn)) - insn = NEXT_INSN (insn); - - for (; insn && !LABEL_P (insn) && !INSN_P (insn); - insn = NEXT_INSN (insn)) - if (NOTE_P (insn) - && NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_BEG) - break; - - if (insn && NOTE_P (insn)) - break; - - /* Do not clean up branches to just past the end of a loop - at this time; it can mess up the loop optimizer's - recognition of some patterns. */ - - insn = PREV_INSN (BB_HEAD (target)); - if (insn && NOTE_P (insn) - && NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_END) - break; - } - counter++; target = new_target; threaded |= new_target_threaded; @@ -653,16 +624,15 @@ 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 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 - must be left untouched (they are required to make it safely across - partition boundaries). See the comments at the top of + 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. */ if (BB_PARTITION (a) != BB_PARTITION (b)) @@ -672,20 +642,10 @@ 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", @@ -709,16 +669,15 @@ 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 - and cold sections. - + 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 - must be left untouched (they are required to make it safely across - partition boundaries). See the comments at the top of + 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. */ if (BB_PARTITION (a) != BB_PARTITION (b)) @@ -739,16 +698,6 @@ 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)); @@ -783,18 +732,16 @@ merge_blocks_move (edge e, basic_block b, basic_block c, int mode) /* If we are partitioning hot/cold basic blocks, we don't want to mess up unconditional or indirect jumps that cross between hot - and cold sections. - + 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 - must be left untouched (they are required to make it safely across - partition boundaries). See the comments at the top of + 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. */ 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) @@ -820,7 +767,7 @@ merge_blocks_move (edge e, basic_block b, basic_block c, int mode) edge_iterator ei; /* Avoid overactive code motion, as the forwarder blocks should be - eliminated by edge redirection instead. One exception might have + eliminated by edge redirection instead. One exception might have been if B is a forwarder block and C has no fallthru edge, but that should be cleaned up by bb-reorder instead. */ if (FORWARDER_BLOCK_P (b) || FORWARDER_BLOCK_P (c)) @@ -852,7 +799,7 @@ merge_blocks_move (edge e, basic_block b, basic_block c, int mode) if (! c_has_outgoing_fallthru) { merge_blocks_move_successor_nojumps (b, c); - return next == ENTRY_BLOCK_PTR ? next->next_bb : next; + return next == ENTRY_BLOCK_PTR ? next->next_bb : next; } /* If B does not have an incoming fallthru, then it can be moved @@ -878,148 +825,324 @@ merge_blocks_move (edge e, basic_block b, basic_block c, int mode) return NULL; } -/* Return true iff the condbranches at the end of BB1 and BB2 match. */ -bool -condjump_equiv_p (struct equiv_info *info, bool call_init) + +/* Removes the memory attributes of MEM expression + if they are not equal. */ + +void +merge_memattrs (rtx x, rtx y) +{ + int i; + int j; + enum rtx_code code; + const char *fmt; + + if (x == y) + return; + if (x == 0 || y == 0) + return; + + code = GET_CODE (x); + + if (code != GET_CODE (y)) + return; + + if (GET_MODE (x) != GET_MODE (y)) + return; + + if (code == MEM && MEM_ATTRS (x) != MEM_ATTRS (y)) + { + if (! MEM_ATTRS (x)) + MEM_ATTRS (y) = 0; + else if (! MEM_ATTRS (y)) + MEM_ATTRS (x) = 0; + else + { + rtx mem_size; + + if (MEM_ALIAS_SET (x) != MEM_ALIAS_SET (y)) + { + set_mem_alias_set (x, 0); + set_mem_alias_set (y, 0); + } + + if (! mem_expr_equal_p (MEM_EXPR (x), MEM_EXPR (y))) + { + set_mem_expr (x, 0); + set_mem_expr (y, 0); + set_mem_offset (x, 0); + set_mem_offset (y, 0); + } + else if (MEM_OFFSET (x) != MEM_OFFSET (y)) + { + set_mem_offset (x, 0); + set_mem_offset (y, 0); + } + + if (!MEM_SIZE (x)) + mem_size = NULL_RTX; + else if (!MEM_SIZE (y)) + mem_size = NULL_RTX; + else + mem_size = GEN_INT (MAX (INTVAL (MEM_SIZE (x)), + INTVAL (MEM_SIZE (y)))); + set_mem_size (x, mem_size); + set_mem_size (y, mem_size); + + set_mem_align (x, MIN (MEM_ALIGN (x), MEM_ALIGN (y))); + set_mem_align (y, MEM_ALIGN (x)); + } + } + + fmt = GET_RTX_FORMAT (code); + for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--) + { + switch (fmt[i]) + { + case 'E': + /* Two vectors must have the same length. */ + if (XVECLEN (x, i) != XVECLEN (y, i)) + return; + + for (j = 0; j < XVECLEN (x, i); j++) + merge_memattrs (XVECEXP (x, i, j), XVECEXP (y, i, j)); + + break; + + case 'e': + merge_memattrs (XEXP (x, i), XEXP (y, i)); + } + } + return; +} + + +/* Return true if I1 and I2 are equivalent and thus can be crossjumped. */ + +static bool +old_insns_match_p (int mode ATTRIBUTE_UNUSED, rtx i1, rtx i2) { - 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)) + rtx p1, p2; + + /* Verify that I1 and I2 are equivalent. */ + if (GET_CODE (i1) != GET_CODE (i2)) return false; - if (f1->dest == f2->dest && b1->dest == b2->dest) - reverse = false; - else if (f1->dest == b2->dest && b1->dest == f2->dest) - reverse = true; - else + p1 = PATTERN (i1); + p2 = PATTERN (i2); + + if (GET_CODE (p1) != GET_CODE (p2)) return false; - 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); + /* If this is a CALL_INSN, compare register usage information. + If we don't check this on stack register machines, the two + CALL_INSNs might be merged leaving reg-stack.c with mismatching + numbers of stack registers in the same basic block. + If we don't check this on machines with delay slots, a delay slot may + be filled that clobbers a parameter expected by the subroutine. - if (code2 == UNKNOWN) + ??? We take the simple route for now and assume that if they're + equal, they were constructed identically. */ + + if (CALL_P (i1) + && (!rtx_equal_p (CALL_INSN_FUNCTION_USAGE (i1), + CALL_INSN_FUNCTION_USAGE (i2)) + || SIBLING_CALL_P (i1) != SIBLING_CALL_P (i2))) 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) +#ifdef STACK_REGS + /* If cross_jump_death_matters is not 0, the insn's mode + indicates whether or not the insn contains any stack-like + regs. */ + + if ((mode & CLEANUP_POST_REGSTACK) && stack_regs_mentioned (i1)) { - 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)); + /* If register stack conversion has already been done, then + death notes must also be compared before it is certain that + the two instruction streams match. */ + + rtx note; + HARD_REG_SET i1_regset, i2_regset; + + CLEAR_HARD_REG_SET (i1_regset); + CLEAR_HARD_REG_SET (i2_regset); + for (note = REG_NOTES (i1); note; note = XEXP (note, 1)) + if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0))) + SET_HARD_REG_BIT (i1_regset, REGNO (XEXP (note, 0))); + + for (note = REG_NOTES (i2); note; note = XEXP (note, 1)) + if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0))) + SET_HARD_REG_BIT (i2_regset, REGNO (XEXP (note, 0))); + + if (!hard_reg_set_equal_p (i1_regset, i2_regset)) + return false; } - else if (code1 == swap_condition (code2)) +#endif + + if (reload_completed + ? 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; +} + +/* 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. + + 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) +{ + rtx i1, i2, last1, last2, afterlast1, afterlast2; + int ninsns = 0; + + /* Skip simple jumps at the end of the blocks. Complex jumps still + need to be compared for equivalence, which we'll do below. */ + + i1 = BB_END (bb1); + last1 = afterlast1 = last2 = afterlast2 = NULL_RTX; + if (onlyjump_p (i1) + || (returnjump_p (i1) && !side_effects_p (PATTERN (i1)))) { - 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)); + last1 = i1; + i1 = PREV_INSN (i1); + } + i2 = BB_END (bb2); + if (onlyjump_p (i2) + || (returnjump_p (i2) && !side_effects_p (PATTERN (i2)))) + { + last2 = i2; + /* Count everything except for unconditional jump as insn. */ + if (!simplejump_p (i2) && !returnjump_p (i2) && last1) + ninsns++; + i2 = PREV_INSN (i2); } - 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)) + + while (true) { - int prob2; + /* Ignore notes. */ + while (!INSN_P (i1) && i1 != BB_HEAD (bb1)) + i1 = PREV_INSN (i1); - 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; + while (!INSN_P (i2) && i2 != BB_HEAD (bb2)) + i2 = PREV_INSN (i2); - /* 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) + if (i1 == BB_HEAD (bb1) || i2 == BB_HEAD (bb2)) + break; + + if (!old_insns_match_p (mode, 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); + /* 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); + } - match = false; + afterlast1 = last1, afterlast2 = last2; + last1 = i1, last2 = i2; + ninsns++; } + + i1 = PREV_INSN (i1); + i2 = PREV_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 the insn after a compare to be shared by + cross-jumping unless the compare is also shared. */ + if (ninsns && reg_mentioned_p (cc0_rtx, last1) && ! sets_cc0_p (last1)) + last1 = afterlast1, last2 = afterlast2, ninsns--; +#endif + + /* Include preceding notes and labels in the cross-jump. One, + this may bring us to the head of the blocks as requested above. + Two, it keeps line number notes as matched as may be. */ + if (ninsns) + { + while (last1 != BB_HEAD (bb1) && !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))) + last2 = PREV_INSN (last2); - if (!match) - cancel_changes (0); - return match; + if (last2 != BB_HEAD (bb2) && LABEL_P (PREV_INSN (last2))) + last2 = PREV_INSN (last2); + + *f1 = last1; + *f2 = last2; + } + + return ninsns; } -/* Return true iff outgoing edges of INFO->y_block and INFO->x_block match, - together with the branch instruction. This means that if we commonize the - control flow before end of the basic block, the semantic remains unchanged. - If we need to compare jumps, we set STRUCT_EQUIV_MATCH_JUMPS in *MODE, - and pass *MODE to struct_equiv_init or assign it to INFO->mode, as - appropriate. +/* Return true iff outgoing edges of BB1 and BB2 match, together with + the branch instruction. This means that if we commonize the control + flow before end of the basic block, the semantic remains unchanged. We may assume that there exists one edge with a common destination. */ static bool -outgoing_edges_match (int *mode, struct equiv_info *info) +outgoing_edges_match (int mode, basic_block bb1, basic_block bb2) { - basic_block bb1 = info->y_block; - basic_block bb2 = info->x_block; int nehedges1 = 0, nehedges2 = 0; edge fallthru1 = 0, fallthru2 = 0; edge e1, e2; @@ -1035,19 +1158,114 @@ outgoing_edges_match (int *mode, struct equiv_info *info) & (EDGE_COMPLEX | EDGE_FAKE)) == 0 && (!JUMP_P (BB_END (bb2)) || simplejump_p (BB_END (bb2)))); - *mode |= STRUCT_EQUIV_MATCH_JUMPS; /* Match conditional jumps - this may get tricky when fallthru and branch edges are crossed. */ if (EDGE_COUNT (bb1->succs) == 2 && any_condjump_p (BB_END (bb1)) && onlyjump_p (BB_END (bb1))) { + edge b1, f1, b2, f2; + bool reverse, match; + rtx set1, set2, cond1, cond2; + enum rtx_code code1, code2; + if (EDGE_COUNT (bb2->succs) != 2 || !any_condjump_p (BB_END (bb2)) || !onlyjump_p (BB_END (bb2))) return false; - info->mode = *mode; - return condjump_equiv_p (info, true); + + b1 = BRANCH_EDGE (bb1); + b2 = BRANCH_EDGE (bb2); + f1 = FALLTHRU_EDGE (bb1); + f2 = FALLTHRU_EDGE (bb2); + + /* 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; + + 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; + + 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; + + cond1 = XEXP (SET_SRC (set1), 0); + cond2 = XEXP (SET_SRC (set2), 0); + code1 = GET_CODE (cond1); + if (reverse) + code2 = reversed_comparison_code (cond2, BB_END (bb2)); + else + code2 = GET_CODE (cond2); + + if (code2 == UNKNOWN) + return false; + + /* Verify codes and operands match. */ + match = ((code1 == code2 + && rtx_renumbered_equal_p (XEXP (cond1, 0), XEXP (cond2, 0)) + && rtx_renumbered_equal_p (XEXP (cond1, 1), XEXP (cond2, 1))) + || (code1 == swap_condition (code2) + && rtx_renumbered_equal_p (XEXP (cond1, 1), + XEXP (cond2, 0)) + && rtx_renumbered_equal_p (XEXP (cond1, 0), + XEXP (cond2, 1)))); + + /* 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; + + 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; + + /* 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) + { + 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); + + return false; + } + } + + if (dump_file && match) + fprintf (dump_file, "Conditionals in bb %i and %i match.\n", + bb1->index, bb2->index); + + return match; } /* Generic case - we are seeing a computed jump, table jump or trapping @@ -1095,22 +1313,31 @@ outgoing_edges_match (int *mode, struct equiv_info *info) identical = false; } - if (identical - && struct_equiv_init (STRUCT_EQUIV_START | *mode, info)) + if (identical) { + replace_label_data rr; bool match; - /* Indicate that LABEL1 is to be replaced with LABEL2 + /* Temporarily replace references to LABEL1 with LABEL2 in BB1->END so that we could compare the instructions. */ - info->y_label = label1; - info->x_label = label2; + rr.r1 = label1; + rr.r2 = label2; + rr.update_label_nuses = false; + for_each_rtx (&BB_END (bb1), replace_label, &rr); - match = insns_match_p (BB_END (bb1), BB_END (bb2), info); + match = old_insns_match_p (mode, BB_END (bb1), BB_END (bb2)); if (dump_file && match) fprintf (dump_file, "Tablejumps in bb %i and %i match.\n", bb1->index, bb2->index); + /* Set the original label in BB1->END because when deleting + a block whose end is a tablejump, the tablejump referenced + from the instruction is deleted too. */ + rr.r1 = label2; + rr.r2 = label1; + for_each_rtx (&BB_END (bb1), replace_label, &rr); + return match; } } @@ -1120,8 +1347,7 @@ outgoing_edges_match (int *mode, struct equiv_info *info) /* First ensure that the instructions match. There may be many outgoing edges so this test is generally cheaper. */ - if (!struct_equiv_init (STRUCT_EQUIV_START | *mode, info) - || !insns_match_p (BB_END (bb1), BB_END (bb2), info)) + if (!old_insns_match_p (mode, BB_END (bb1), BB_END (bb2))) return false; /* Search the outgoing edges, ensure that the counts do match, find possible @@ -1133,7 +1359,7 @@ outgoing_edges_match (int *mode, struct equiv_info *info) FOR_EACH_EDGE (e1, ei, bb1->succs) { e2 = EDGE_SUCC (bb2, ei.index); - + if (e1->flags & EDGE_EH) nehedges1++; @@ -1175,11 +1401,43 @@ outgoing_edges_match (int *mode, struct equiv_info *info) return false; } - /* We don't need to match the rest of edges as above checks should be enough - to ensure that they are equivalent. */ + /* The same checks as in try_crossjump_to_edge. It is required for RTL + version of sequence abstraction. */ + FOR_EACH_EDGE (e1, ei, bb2->succs) + { + edge e2; + edge_iterator ei; + basic_block d1 = e1->dest; + + if (FORWARDER_BLOCK_P (d1)) + d1 = EDGE_SUCC (d1, 0)->dest; + + FOR_EACH_EDGE (e2, ei, bb1->succs) + { + basic_block d2 = e2->dest; + if (FORWARDER_BLOCK_P (d2)) + d2 = EDGE_SUCC (d2, 0)->dest; + if (d1 == d2) + break; + } + + if (!e2) + return false; + } + return true; } +/* Returns true if BB basic block has a preserve label. */ + +static bool +block_has_preserve_label (basic_block bb) +{ + return (bb + && block_label (bb) + && LABEL_PRESERVE_P (block_label (bb))); +} + /* E1 and E2 are edges with the same destination block. Search their predecessors for common code. If found, redirect control flow from (maybe the middle of) E1->SRC to (maybe the middle of) E2->SRC. */ @@ -1187,13 +1445,14 @@ outgoing_edges_match (int *mode, struct equiv_info *info) static bool try_crossjump_to_edge (int mode, edge e1, edge e2) { - int nmatch, i; + int nmatch; basic_block src1 = e1->src, src2 = e2->src; basic_block redirect_to, redirect_from, to_remove; + rtx newpos1, newpos2; edge s; edge_iterator ei; - struct equiv_info info; - rtx x_active, y_active; + + newpos1 = newpos2 = NULL_RTX; /* If we have partitioned hot/cold basic blocks, it is a bad idea to try this optimization. @@ -1204,7 +1463,7 @@ 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 @@ -1240,66 +1499,24 @@ try_crossjump_to_edge (int mode, edge e1, edge e2) return false; /* Look for the common insn sequence, part the first ... */ - info.x_block = src2; - info.y_block = src1; - if (!outgoing_edges_match (&mode, &info)) + if (!outgoing_edges_match (mode, src1, src2)) return false; /* ... and part the second. */ - info.input_cost = optimize_size ? COSTS_N_INSNS (1) : -1; - nmatch = struct_equiv_block_eq (STRUCT_EQUIV_START | mode, &info); + nmatch = flow_find_cross_jump (mode, 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 (such that its predecessors will hopefully be redirected and the block removed). */ - if (!nmatch) - return false; - if ((nmatch -info.cur.input_count < PARAM_VALUE (PARAM_MIN_CROSSJUMP_INSNS)) - && (info.cur.y_start != BB_HEAD (src1))) - return false; - while (info.need_rerun) - { - nmatch = struct_equiv_block_eq (STRUCT_EQUIV_RERUN | mode, &info); - if (!nmatch) - return false; - if ((nmatch -info.cur.input_count < PARAM_VALUE (PARAM_MIN_CROSSJUMP_INSNS)) - && (info.cur.y_start != BB_HEAD (src1))) - return false; - } - nmatch = struct_equiv_block_eq (STRUCT_EQUIV_FINAL | mode, &info); - if ((nmatch -info.cur.input_count < PARAM_VALUE (PARAM_MIN_CROSSJUMP_INSNS)) - && (info.cur.y_start != BB_HEAD (src1))) + if ((nmatch < PARAM_VALUE (PARAM_MIN_CROSSJUMP_INSNS)) + && (newpos1 != BB_HEAD (src1))) return false; - /* Skip possible basic block header. */ - x_active = info.cur.x_start; - if (LABEL_P (x_active)) - x_active = NEXT_INSN (x_active); - if (NOTE_P (x_active)) - x_active = NEXT_INSN (x_active); - - y_active = info.cur.y_start; - if (LABEL_P (y_active)) - y_active = NEXT_INSN (y_active); - if (NOTE_P (y_active)) - y_active = NEXT_INSN (y_active); - - /* In order for this code to become active, either we have to be called - before reload, or struct_equiv_block_eq needs to add register scavenging - code to allocate input_reg after reload. */ - if (info.input_reg) - { - emit_insn_before (gen_move_insn (info.input_reg, info.x_input), - x_active); - emit_insn_before (gen_move_insn (info.input_reg, info.y_input), - y_active); - } - - for (i = 0; i < info.cur.local_count; i++) - if (info.local_rvalue[i]) - emit_insn_before (gen_move_insn (info.x_local[i], info.y_local[i]), - y_active); + /* Avoid deleting preserve label when redirecting ABNORMAL edges. */ + if (block_has_preserve_label (e1->dest) + && (e1->flags & EDGE_ABNORMAL)) + return false; /* Here we know that the insns in the end of SRC1 which are common with SRC2 will be deleted. @@ -1335,41 +1552,33 @@ try_crossjump_to_edge (int mode, edge e1, edge e2) /* Avoid splitting if possible. We must always split when SRC2 has EH predecessor edges, or we may end up with basic blocks with both normal and EH predecessor edges. */ - if (info.cur.x_start == BB_HEAD (src2) + if (newpos2 == BB_HEAD (src2) && !(EDGE_PRED (src2, 0)->flags & EDGE_EH)) redirect_to = src2; else { - if (info.cur.x_start == BB_HEAD (src2)) + if (newpos2 == BB_HEAD (src2)) { /* Skip possible basic block header. */ - if (LABEL_P (info.cur.x_start)) - info.cur.x_start = NEXT_INSN (info.cur.x_start); - if (NOTE_P (info.cur.x_start)) - info.cur.x_start = NEXT_INSN (info.cur.x_start); + if (LABEL_P (newpos2)) + newpos2 = NEXT_INSN (newpos2); + if (NOTE_P (newpos2)) + newpos2 = NEXT_INSN (newpos2); } if (dump_file) fprintf (dump_file, "Splitting bb %i before %i insns\n", src2->index, nmatch); - redirect_to = split_block (src2, PREV_INSN (info.cur.x_start))->dest; - COPY_REG_SET (info.y_block->il.rtl->global_live_at_end, - info.x_block->il.rtl->global_live_at_end); + redirect_to = split_block (src2, PREV_INSN (newpos2))->dest; } if (dump_file) - { - fprintf (dump_file, "Cross jumping from bb %i to bb %i; %i common insns", - src1->index, src2->index, nmatch); - if (info.cur.local_count) - fprintf (dump_file, ", %i local registers", info.cur.local_count); - fprintf (dump_file, "\n"); - } + fprintf (dump_file, + "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 trought the block. */ - redirect_to->flags |= BB_DIRTY; + /* We may have some registers visible through the block. */ + df_set_bb_dirty (redirect_to); /* Recompute the frequencies and counts of outgoing edges. */ FOR_EACH_EDGE (s, ei, redirect_to->succs) @@ -1393,8 +1602,8 @@ try_crossjump_to_edge (int mode, edge e1, edge e2) s->count += s2->count; /* Take care to update possible forwarder blocks. We verified - that there is no more than one in the chain, so we can't run - into infinite loop. */ + that there is no more than one in the chain, so we can't run + into infinite loop. */ if (FORWARDER_BLOCK_P (s->dest)) { single_succ_edge (s->dest)->count += s2->count; @@ -1424,11 +1633,26 @@ 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. */ - redirect_from = split_block (src1, PREV_INSN (y_active))->src; + /* Skip possible basic block header. */ + if (LABEL_P (newpos1)) + newpos1 = NEXT_INSN (newpos1); + + if (NOTE_P (newpos1)) + newpos1 = NEXT_INSN (newpos1); + + redirect_from = split_block (src1, PREV_INSN (newpos1))->src; to_remove = single_succ (redirect_from); redirect_edge_and_branch_force (single_succ_edge (redirect_from), redirect_to); @@ -1467,16 +1691,16 @@ try_crossjump_bb (int mode, basic_block bb) /* If we are partitioning hot/cold basic blocks, we don't want to mess up unconditional or indirect jumps that cross between hot - and cold sections. - + 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 - must be left untouched (they are required to make it safely across - partition boundaries). See the comments at the top of + 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. */ - if (BB_PARTITION (EDGE_PRED (bb, 0)->src) != - BB_PARTITION (EDGE_PRED (bb, 1)->src) + if (BB_PARTITION (EDGE_PRED (bb, 0)->src) != + BB_PARTITION (EDGE_PRED (bb, 1)->src) || (EDGE_PRED (bb, 0)->flags & EDGE_CROSSING)) return false; @@ -1492,7 +1716,10 @@ 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; @@ -1501,7 +1728,8 @@ 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. @@ -1511,8 +1739,8 @@ 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)) @@ -1561,8 +1789,8 @@ 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)) @@ -1575,6 +1803,9 @@ try_crossjump_bb (int mode, basic_block bb) } } + if (changed) + crossjumps_occured = true; + return changed; } @@ -1589,12 +1820,11 @@ 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); @@ -1621,7 +1851,7 @@ try_optimize_cfg (int mode) bool changed_here = false; /* Delete trivially dead basic blocks. */ - while (EDGE_COUNT (b->preds) == 0) + if (EDGE_COUNT (b->preds) == 0) { c = b->prev_bb; if (dump_file) @@ -1631,7 +1861,9 @@ try_optimize_cfg (int mode) 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. */ @@ -1652,15 +1884,17 @@ 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", @@ -1703,7 +1937,7 @@ try_optimize_cfg (int mode) does not fit merge_blocks interface and is kept here in hope that it will become useless once more of compiler is transformed to use cfg_layout mode. */ - + if ((mode & CLEANUP_CFGLAYOUT) && can_merge_blocks_p (b, c)) { @@ -1781,9 +2015,6 @@ 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); @@ -1818,30 +2049,46 @@ 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_P (next) + && (GET_CODE (PATTERN (next)) == ADDR_VEC + || GET_CODE (PATTERN (next)) == ADDR_DIFF_VEC)) + { + 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. */ @@ -1850,66 +2097,97 @@ 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) - { - /* 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 (!(mode & CLEANUP_NO_INSN_DEL)) { - 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; } -static void +static unsigned int rest_of_handle_jump (void) { delete_unreachable_blocks (); if (cfun->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 */ + NULL, /* gate */ + rest_of_handle_jump, /* execute */ NULL, /* sub */ NULL, /* next */ 0, /* static_pass_number */ @@ -1918,48 +2196,30 @@ 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 void +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); - cleanup_cfg ((optimize ? CLEANUP_EXPENSIVE : 0) | CLEANUP_PRE_LOOP - | (flag_thread_jumps ? CLEANUP_THREADING : 0)); - - create_loop_notes (); - - purge_line_number_notes (); - - if (optimize) - cleanup_cfg (CLEANUP_EXPENSIVE | CLEANUP_PRE_LOOP); - - /* 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 (dump_file); + dump_flow_info (dump_file, dump_flags); + cleanup_cfg ((optimize ? CLEANUP_EXPENSIVE : 0) + | (flag_thread_jumps ? CLEANUP_THREADING : 0)); + 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 */ + NULL, /* gate */ + rest_of_handle_jump2, /* execute */ NULL, /* sub */ NULL, /* next */ 0, /* static_pass_number */ @@ -1968,8 +2228,8 @@ 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 */ + } };