Mercurial > hg > CbC > CbC_gcc
diff gcc/reorg.c @ 111:04ced10e8804
gcc 7
| author | kono |
|---|---|
| date | Fri, 27 Oct 2017 22:46:09 +0900 |
| parents | f6334be47118 |
| children | 84e7813d76e9 |
line wrap: on
line diff
--- a/gcc/reorg.c Sun Aug 21 07:07:55 2011 +0900 +++ b/gcc/reorg.c Fri Oct 27 22:46:09 2017 +0900 @@ -1,7 +1,5 @@ /* Perform instruction reorganizations for delay slot filling. - Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, - 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010 - Free Software Foundation, Inc. + Copyright (C) 1992-2017 Free Software Foundation, Inc. Contributed by Richard Kenner (kenner@vlsi1.ultra.nyu.edu). Hacked by Michael Tiemann (tiemann@cygnus.com). @@ -100,52 +98,67 @@ delay slot. In that case, we point each insn at the other with REG_CC_USER and REG_CC_SETTER notes. Note that these restrictions affect very few machines because most RISC machines with delay slots will not use CC0 - (the RT is the only known exception at this point). - - Not yet implemented: - - The Acorn Risc Machine can conditionally execute most insns, so - it is profitable to move single insns into a position to execute - based on the condition code of the previous insn. - - The HP-PA can conditionally nullify insns, providing a similar - effect to the ARM, differing mostly in which insn is "in charge". */ + (the RT is the only known exception at this point). */ #include "config.h" #include "system.h" #include "coretypes.h" -#include "tm.h" -#include "diagnostic-core.h" +#include "backend.h" +#include "target.h" #include "rtl.h" +#include "tree.h" +#include "predict.h" +#include "memmodel.h" #include "tm_p.h" -#include "expr.h" -#include "function.h" +#include "expmed.h" #include "insn-config.h" -#include "conditions.h" -#include "hard-reg-set.h" -#include "basic-block.h" -#include "regs.h" +#include "emit-rtl.h" #include "recog.h" -#include "flags.h" -#include "output.h" -#include "obstack.h" #include "insn-attr.h" #include "resource.h" -#include "except.h" #include "params.h" -#include "timevar.h" -#include "target.h" #include "tree-pass.h" -#ifdef DELAY_SLOTS - -#ifndef ANNUL_IFTRUE_SLOTS -#define eligible_for_annul_true(INSN, SLOTS, TRIAL, FLAGS) 0 -#endif -#ifndef ANNUL_IFFALSE_SLOTS -#define eligible_for_annul_false(INSN, SLOTS, TRIAL, FLAGS) 0 -#endif - + +/* First, some functions that were used before GCC got a control flow graph. + These functions are now only used here in reorg.c, and have therefore + been moved here to avoid inadvertent misuse elsewhere in the compiler. */ + +/* Return the last label to mark the same position as LABEL. Return LABEL + itself if it is null or any return rtx. */ + +static rtx +skip_consecutive_labels (rtx label_or_return) +{ + rtx_insn *insn; + + if (label_or_return && ANY_RETURN_P (label_or_return)) + return label_or_return; + + rtx_insn *label = as_a <rtx_insn *> (label_or_return); + + for (insn = label; insn != 0 && !INSN_P (insn); insn = NEXT_INSN (insn)) + if (LABEL_P (insn)) + label = insn; + + return label; +} + +/* INSN uses CC0 and is being moved into a delay slot. Set up REG_CC_SETTER + and REG_CC_USER notes so we can find it. */ + +static void +link_cc0_insns (rtx_insn *insn) +{ + rtx user = next_nonnote_insn (insn); + + if (NONJUMP_INSN_P (user) && GET_CODE (PATTERN (user)) == SEQUENCE) + user = XVECEXP (PATTERN (user), 0, 0); + + add_reg_note (user, REG_CC_SETTER, insn); + add_reg_note (insn, REG_CC_USER, user); +} + /* Insns which have delay slots that have not yet been filled. */ static struct obstack unfilled_slots_obstack; @@ -156,13 +169,16 @@ should be recomputed at each use. */ #define unfilled_slots_base \ - ((rtx *) obstack_base (&unfilled_slots_obstack)) + ((rtx_insn **) obstack_base (&unfilled_slots_obstack)) #define unfilled_slots_next \ - ((rtx *) obstack_next_free (&unfilled_slots_obstack)) - -/* Points to the label before the end of the function. */ -static rtx end_of_function_label; + ((rtx_insn **) obstack_next_free (&unfilled_slots_obstack)) + +/* Points to the label before the end of the function, or before a + return insn. */ +static rtx_code_label *function_return_label; +/* Likewise for a simple_return. */ +static rtx_code_label *function_simple_return_label; /* Mapping between INSN_UID's and position in the code since INSN_UID's do not always monotonically increase. */ @@ -171,61 +187,81 @@ /* Highest valid index in `uid_to_ruid'. */ static int max_uid; -static int stop_search_p (rtx, int); +static int stop_search_p (rtx_insn *, int); static int resource_conflicts_p (struct resources *, struct resources *); static int insn_references_resource_p (rtx, struct resources *, bool); static int insn_sets_resource_p (rtx, struct resources *, bool); -static rtx find_end_label (void); -static rtx emit_delay_sequence (rtx, rtx, int); -static rtx add_to_delay_list (rtx, rtx); -static rtx delete_from_delay_slot (rtx); -static void delete_scheduled_jump (rtx); +static rtx_code_label *find_end_label (rtx); +static rtx_insn *emit_delay_sequence (rtx_insn *, const vec<rtx_insn *> &, + int); +static void add_to_delay_list (rtx_insn *, vec<rtx_insn *> *); +static rtx_insn *delete_from_delay_slot (rtx_insn *); +static void delete_scheduled_jump (rtx_insn *); static void note_delay_statistics (int, int); -#if defined(ANNUL_IFFALSE_SLOTS) || defined(ANNUL_IFTRUE_SLOTS) -static rtx optimize_skip (rtx); -#endif -static int get_jump_flags (rtx, rtx); -static int rare_destination (rtx); -static int mostly_true_jump (rtx, rtx); -static rtx get_branch_condition (rtx, rtx); -static int condition_dominates_p (rtx, rtx); -static int redirect_with_delay_slots_safe_p (rtx, rtx, rtx); -static int redirect_with_delay_list_safe_p (rtx, rtx, rtx); -static int check_annul_list_true_false (int, rtx); -static rtx steal_delay_list_from_target (rtx, rtx, rtx, rtx, - struct resources *, - struct resources *, - struct resources *, - int, int *, int *, rtx *); -static rtx steal_delay_list_from_fallthrough (rtx, rtx, rtx, rtx, - struct resources *, - struct resources *, - struct resources *, - int, int *, int *); -static void try_merge_delay_insns (rtx, rtx); -static rtx redundant_insn (rtx, rtx, rtx); +static int get_jump_flags (const rtx_insn *, rtx); +static int mostly_true_jump (rtx); +static rtx get_branch_condition (const rtx_insn *, rtx); +static int condition_dominates_p (rtx, const rtx_insn *); +static int redirect_with_delay_slots_safe_p (rtx_insn *, rtx, rtx); +static int redirect_with_delay_list_safe_p (rtx_insn *, rtx, + const vec<rtx_insn *> &); +static int check_annul_list_true_false (int, const vec<rtx_insn *> &); +static void steal_delay_list_from_target (rtx_insn *, rtx, rtx_sequence *, + vec<rtx_insn *> *, + struct resources *, + struct resources *, + struct resources *, + int, int *, int *, + rtx *); +static void steal_delay_list_from_fallthrough (rtx_insn *, rtx, rtx_sequence *, + vec<rtx_insn *> *, + struct resources *, + struct resources *, + struct resources *, + int, int *, int *); +static void try_merge_delay_insns (rtx_insn *, rtx_insn *); +static rtx_insn *redundant_insn (rtx, rtx_insn *, const vec<rtx_insn *> &); static int own_thread_p (rtx, rtx, int); -static void update_block (rtx, rtx); -static int reorg_redirect_jump (rtx, rtx); -static void update_reg_dead_notes (rtx, rtx); -static void fix_reg_dead_note (rtx, rtx); -static void update_reg_unused_notes (rtx, rtx); +static void update_block (rtx_insn *, rtx_insn *); +static int reorg_redirect_jump (rtx_jump_insn *, rtx); +static void update_reg_dead_notes (rtx_insn *, rtx_insn *); +static void fix_reg_dead_note (rtx_insn *, rtx); +static void update_reg_unused_notes (rtx_insn *, rtx); static void fill_simple_delay_slots (int); -static rtx fill_slots_from_thread (rtx, rtx, rtx, rtx, - int, int, int, int, - int *, rtx); +static void fill_slots_from_thread (rtx_jump_insn *, rtx, rtx, rtx, + int, int, int, int, + int *, vec<rtx_insn *> *); static void fill_eager_delay_slots (void); -static void relax_delay_slots (rtx); -#ifdef HAVE_return -static void make_return_insns (rtx); -#endif +static void relax_delay_slots (rtx_insn *); +static void make_return_insns (rtx_insn *); + +/* A wrapper around next_active_insn which takes care to return ret_rtx + unchanged. */ + +static rtx +first_active_target_insn (rtx insn) +{ + if (ANY_RETURN_P (insn)) + return insn; + return next_active_insn (as_a <rtx_insn *> (insn)); +} + +/* Return true iff INSN is a simplejump, or any kind of return insn. */ + +static bool +simplejump_or_return_p (rtx insn) +{ + return (JUMP_P (insn) + && (simplejump_p (as_a <rtx_insn *> (insn)) + || ANY_RETURN_P (PATTERN (insn)))); +} /* Return TRUE if this insn should stop the search for insn to fill delay slots. LABELS_P indicates that labels should terminate the search. In all cases, jumps terminate the search. */ static int -stop_search_p (rtx insn, int labels_p) +stop_search_p (rtx_insn *insn, int labels_p) { if (insn == 0) return 1; @@ -268,22 +304,10 @@ resource_conflicts_p (struct resources *res1, struct resources *res2) { if ((res1->cc && res2->cc) || (res1->memory && res2->memory) - || (res1->unch_memory && res2->unch_memory) || res1->volatil || res2->volatil) return 1; -#ifdef HARD_REG_SET - return (res1->regs & res2->regs) != HARD_CONST (0); -#else - { - int i; - - for (i = 0; i < HARD_REG_SET_LONGS; i++) - if ((res1->regs[i] & res2->regs[i]) != 0) - return 1; - return 0; - } -#endif + return hard_reg_set_intersect_p (res1->regs, res2->regs); } /* Return TRUE if any resource marked in RES, a `struct resources', is @@ -335,23 +359,34 @@ ??? There may be a problem with the current implementation. Suppose we start with a bare RETURN insn and call find_end_label. It may set - end_of_function_label just before the RETURN. Suppose the machinery + function_return_label just before the RETURN. Suppose the machinery is able to fill the delay slot of the RETURN insn afterwards. Then - end_of_function_label is no longer valid according to the property + function_return_label is no longer valid according to the property described above and find_end_label will still return it unmodified. Note that this is probably mitigated by the following observation: - once end_of_function_label is made, it is very likely the target of + once function_return_label is made, it is very likely the target of a jump, so filling the delay slot of the RETURN will be much more - difficult. */ - -static rtx -find_end_label (void) + difficult. + KIND is either simple_return_rtx or ret_rtx, indicating which type of + return we're looking for. */ + +static rtx_code_label * +find_end_label (rtx kind) { - rtx insn; + rtx_insn *insn; + rtx_code_label **plabel; + + if (kind == ret_rtx) + plabel = &function_return_label; + else + { + gcc_assert (kind == simple_return_rtx); + plabel = &function_simple_return_label; + } /* If we found one previously, return it. */ - if (end_of_function_label) - return end_of_function_label; + if (*plabel) + return *plabel; /* Otherwise, see if there is a label at the end of the function. If there is, it must be that RETURN insns aren't needed, so that is our return @@ -366,174 +401,127 @@ /* When a target threads its epilogue we might already have a suitable return insn. If so put a label before it for the - end_of_function_label. */ + function_return_label. */ if (BARRIER_P (insn) && JUMP_P (PREV_INSN (insn)) - && GET_CODE (PATTERN (PREV_INSN (insn))) == RETURN) + && PATTERN (PREV_INSN (insn)) == kind) { - rtx temp = PREV_INSN (PREV_INSN (insn)); - end_of_function_label = gen_label_rtx (); - LABEL_NUSES (end_of_function_label) = 0; - - /* Put the label before an USE insns that may precede the RETURN insn. */ + rtx_insn *temp = PREV_INSN (PREV_INSN (insn)); + rtx_code_label *label = gen_label_rtx (); + LABEL_NUSES (label) = 0; + + /* Put the label before any USE insns that may precede the RETURN + insn. */ while (GET_CODE (temp) == USE) temp = PREV_INSN (temp); - emit_label_after (end_of_function_label, temp); + emit_label_after (label, temp); + *plabel = label; } else if (LABEL_P (insn)) - end_of_function_label = insn; + *plabel = as_a <rtx_code_label *> (insn); else { - end_of_function_label = gen_label_rtx (); - LABEL_NUSES (end_of_function_label) = 0; + rtx_code_label *label = gen_label_rtx (); + LABEL_NUSES (label) = 0; /* If the basic block reorder pass moves the return insn to some other place try to locate it again and put our - end_of_function_label there. */ - while (insn && ! (JUMP_P (insn) - && (GET_CODE (PATTERN (insn)) == RETURN))) + function_return_label there. */ + while (insn && ! (JUMP_P (insn) && (PATTERN (insn) == kind))) insn = PREV_INSN (insn); if (insn) { insn = PREV_INSN (insn); - /* Put the label before an USE insns that may proceed the + /* Put the label before any USE insns that may precede the RETURN insn. */ while (GET_CODE (insn) == USE) insn = PREV_INSN (insn); - emit_label_after (end_of_function_label, insn); + emit_label_after (label, insn); } else { -#ifdef HAVE_epilogue - if (HAVE_epilogue -#ifdef HAVE_return - && ! HAVE_return -#endif - ) - { - /* The RETURN insn has its delay slot filled so we cannot - emit the label just before it. Since we already have - an epilogue and cannot emit a new RETURN, we cannot - emit the label at all. */ - end_of_function_label = NULL_RTX; - return end_of_function_label; - } -#endif /* HAVE_epilogue */ + if (targetm.have_epilogue () && ! targetm.have_return ()) + /* The RETURN insn has its delay slot filled so we cannot + emit the label just before it. Since we already have + an epilogue and cannot emit a new RETURN, we cannot + emit the label at all. */ + return NULL; /* Otherwise, make a new label and emit a RETURN and BARRIER, if needed. */ - emit_label (end_of_function_label); -#ifdef HAVE_return - /* We don't bother trying to create a return insn if the - epilogue has filled delay-slots; we would have to try and - move the delay-slot fillers to the delay-slots for the new - return insn or in front of the new return insn. */ - if (crtl->epilogue_delay_list == NULL - && HAVE_return) + emit_label (label); + if (targetm.have_return ()) { /* The return we make may have delay slots too. */ - rtx insn = gen_return (); - insn = emit_jump_insn (insn); + rtx_insn *pat = targetm.gen_return (); + rtx_insn *insn = emit_jump_insn (pat); + set_return_jump_label (insn); emit_barrier (); if (num_delay_slots (insn) > 0) obstack_ptr_grow (&unfilled_slots_obstack, insn); } -#endif } + *plabel = label; } /* Show one additional use for this label so it won't go away until we are done. */ - ++LABEL_NUSES (end_of_function_label); - - return end_of_function_label; + ++LABEL_NUSES (*plabel); + + return *plabel; } /* Put INSN and LIST together in a SEQUENCE rtx of LENGTH, and replace the pattern of INSN with the SEQUENCE. - Chain the insns so that NEXT_INSN of each insn in the sequence points to - the next and NEXT_INSN of the last insn in the sequence points to - the first insn after the sequence. Similarly for PREV_INSN. This makes - it easier to scan all insns. - - Returns the SEQUENCE that replaces INSN. */ - -static rtx -emit_delay_sequence (rtx insn, rtx list, int length) + Returns the insn containing the SEQUENCE that replaces INSN. */ + +static rtx_insn * +emit_delay_sequence (rtx_insn *insn, const vec<rtx_insn *> &list, int length) { - int i = 1; - rtx li; - int had_barrier = 0; - /* Allocate the rtvec to hold the insns and the SEQUENCE. */ rtvec seqv = rtvec_alloc (length + 1); rtx seq = gen_rtx_SEQUENCE (VOIDmode, seqv); - rtx seq_insn = make_insn_raw (seq); - rtx first = get_insns (); - rtx last = get_last_insn (); - - /* Make a copy of the insn having delay slots. */ - rtx delay_insn = copy_rtx (insn); - - /* If INSN is followed by a BARRIER, delete the BARRIER since it will only - confuse further processing. Update LAST in case it was the last insn. - We will put the BARRIER back in later. */ - if (NEXT_INSN (insn) && BARRIER_P (NEXT_INSN (insn))) - { - delete_related_insns (NEXT_INSN (insn)); - last = get_last_insn (); - had_barrier = 1; - } - - /* Splice our SEQUENCE into the insn stream where INSN used to be. */ - NEXT_INSN (seq_insn) = NEXT_INSN (insn); - PREV_INSN (seq_insn) = PREV_INSN (insn); - - if (insn != last) - PREV_INSN (NEXT_INSN (seq_insn)) = seq_insn; - - if (insn != first) - NEXT_INSN (PREV_INSN (seq_insn)) = seq_insn; - - /* Note the calls to set_new_first_and_last_insn must occur after - SEQ_INSN has been completely spliced into the insn stream. - - Otherwise CUR_INSN_UID will get set to an incorrect value because - set_new_first_and_last_insn will not find SEQ_INSN in the chain. */ - if (insn == last) - set_new_first_and_last_insn (first, seq_insn); - - if (insn == first) - set_new_first_and_last_insn (seq_insn, last); + rtx_insn *seq_insn = make_insn_raw (seq); + + /* If DELAY_INSN has a location, use it for SEQ_INSN. If DELAY_INSN does + not have a location, but one of the delayed insns does, we pick up a + location from there later. */ + INSN_LOCATION (seq_insn) = INSN_LOCATION (insn); + + /* Unlink INSN from the insn chain, so that we can put it into + the SEQUENCE. Remember where we want to emit SEQUENCE in AFTER. */ + rtx_insn *after = PREV_INSN (insn); + remove_insn (insn); + SET_NEXT_INSN (insn) = SET_PREV_INSN (insn) = NULL; /* Build our SEQUENCE and rebuild the insn chain. */ - XVECEXP (seq, 0, 0) = delay_insn; - INSN_DELETED_P (delay_insn) = 0; - PREV_INSN (delay_insn) = PREV_INSN (seq_insn); - - INSN_LOCATOR (seq_insn) = INSN_LOCATOR (delay_insn); - - for (li = list; li; li = XEXP (li, 1), i++) + start_sequence (); + XVECEXP (seq, 0, 0) = emit_insn (insn); + + unsigned int delay_insns = list.length (); + gcc_assert (delay_insns == (unsigned int) length); + for (unsigned int i = 0; i < delay_insns; i++) { - rtx tem = XEXP (li, 0); + rtx_insn *tem = list[i]; rtx note, next; /* Show that this copy of the insn isn't deleted. */ - INSN_DELETED_P (tem) = 0; - - XVECEXP (seq, 0, i) = tem; - PREV_INSN (tem) = XVECEXP (seq, 0, i - 1); - NEXT_INSN (XVECEXP (seq, 0, i - 1)) = tem; + tem->set_undeleted (); + + /* Unlink insn from its original place, and re-emit it into + the sequence. */ + SET_NEXT_INSN (tem) = SET_PREV_INSN (tem) = NULL; + XVECEXP (seq, 0, i + 1) = emit_insn (tem); /* SPARC assembler, for instance, emit warning when debug info is output into the delay slot. */ - if (INSN_LOCATOR (tem) && !INSN_LOCATOR (seq_insn)) - INSN_LOCATOR (seq_insn) = INSN_LOCATOR (tem); - INSN_LOCATOR (tem) = 0; + if (INSN_LOCATION (tem) && !INSN_LOCATION (seq_insn)) + INSN_LOCATION (seq_insn) = INSN_LOCATION (tem); + INSN_LOCATION (tem) = 0; for (note = REG_NOTES (tem); note; note = next) { @@ -558,28 +546,10 @@ } } } - - NEXT_INSN (XVECEXP (seq, 0, length)) = NEXT_INSN (seq_insn); - - /* If the previous insn is a SEQUENCE, update the NEXT_INSN pointer on the - last insn in that SEQUENCE to point to us. Similarly for the first - insn in the following insn if it is a SEQUENCE. */ - - if (PREV_INSN (seq_insn) && NONJUMP_INSN_P (PREV_INSN (seq_insn)) - && GET_CODE (PATTERN (PREV_INSN (seq_insn))) == SEQUENCE) - NEXT_INSN (XVECEXP (PATTERN (PREV_INSN (seq_insn)), 0, - XVECLEN (PATTERN (PREV_INSN (seq_insn)), 0) - 1)) - = seq_insn; - - if (NEXT_INSN (seq_insn) && NONJUMP_INSN_P (NEXT_INSN (seq_insn)) - && GET_CODE (PATTERN (NEXT_INSN (seq_insn))) == SEQUENCE) - PREV_INSN (XVECEXP (PATTERN (NEXT_INSN (seq_insn)), 0, 0)) = seq_insn; - - /* If there used to be a BARRIER, put it back. */ - if (had_barrier) - emit_barrier_after (seq_insn); - - gcc_assert (i == length + 1); + end_sequence (); + + /* Splice our SEQUENCE into the insn stream where INSN used to be. */ + add_insn_after (seq_insn, after, NULL); return seq_insn; } @@ -587,34 +557,23 @@ /* Add INSN to DELAY_LIST and return the head of the new list. The list must be in the order in which the insns are to be executed. */ -static rtx -add_to_delay_list (rtx insn, rtx delay_list) +static void +add_to_delay_list (rtx_insn *insn, vec<rtx_insn *> *delay_list) { - /* If we have an empty list, just make a new list element. If - INSN has its block number recorded, clear it since we may + /* If INSN has its block number recorded, clear it since we may be moving the insn to a new block. */ - - if (delay_list == 0) - { clear_hashed_info_for_insn (insn); - return gen_rtx_INSN_LIST (VOIDmode, insn, NULL_RTX); - } - - /* Otherwise this must be an INSN_LIST. Add INSN to the end of the - list. */ - XEXP (delay_list, 1) = add_to_delay_list (insn, XEXP (delay_list, 1)); - - return delay_list; + delay_list->safe_push (insn); } /* Delete INSN from the delay slot of the insn that it is in, which may produce an insn with no delay slots. Return the new insn. */ -static rtx -delete_from_delay_slot (rtx insn) +static rtx_insn * +delete_from_delay_slot (rtx_insn *insn) { - rtx trial, seq_insn, seq, prev; - rtx delay_list = 0; + rtx_insn *trial, *seq_insn, *prev; + rtx_sequence *seq; int i; int had_barrier = 0; @@ -628,22 +587,23 @@ ; seq_insn = PREV_INSN (NEXT_INSN (trial)); - seq = PATTERN (seq_insn); + seq = as_a <rtx_sequence *> (PATTERN (seq_insn)); if (NEXT_INSN (seq_insn) && BARRIER_P (NEXT_INSN (seq_insn))) had_barrier = 1; /* Create a delay list consisting of all the insns other than the one we are deleting (unless we were the only one). */ - if (XVECLEN (seq, 0) > 2) - for (i = 1; i < XVECLEN (seq, 0); i++) - if (XVECEXP (seq, 0, i) != insn) - delay_list = add_to_delay_list (XVECEXP (seq, 0, i), delay_list); + auto_vec<rtx_insn *, 5> delay_list; + if (seq->len () > 2) + for (i = 1; i < seq->len (); i++) + if (seq->insn (i) != insn) + add_to_delay_list (seq->insn (i), &delay_list); /* Delete the old SEQUENCE, re-emit the insn that used to have the delay list, and rebuild the delay list if non-empty. */ prev = PREV_INSN (seq_insn); - trial = XVECEXP (seq, 0, 0); + trial = seq->insn (0); delete_related_insns (seq_insn); add_insn_after (trial, prev, NULL); @@ -653,9 +613,9 @@ /* If there are any delay insns, remit them. Otherwise clear the annul flag. */ - if (delay_list) + if (!delay_list.is_empty ()) trial = emit_delay_sequence (trial, delay_list, XVECLEN (seq, 0) - 2); - else if (INSN_P (trial)) + else if (JUMP_P (trial)) INSN_ANNULLED_BRANCH_P (trial) = 0; INSN_FROM_TARGET_P (insn) = 0; @@ -670,7 +630,7 @@ the insn that sets CC0 for it and delete it too. */ static void -delete_scheduled_jump (rtx insn) +delete_scheduled_jump (rtx_insn *insn) { /* Delete the insn that sets cc0 for us. On machines without cc0, we could delete the insn that sets the condition code, but it is hard to find it. @@ -678,8 +638,7 @@ be other insns that became dead anyway, which we wouldn't know to delete. */ -#ifdef HAVE_cc0 - if (reg_mentioned_p (cc0_rtx, insn)) + if (HAVE_cc0 && reg_mentioned_p (cc0_rtx, insn)) { rtx note = find_reg_note (insn, REG_CC_SETTER, NULL_RTX); @@ -690,14 +649,14 @@ { if (! FIND_REG_INC_NOTE (XEXP (note, 0), NULL_RTX) && sets_cc0_p (PATTERN (XEXP (note, 0))) == 1) - delete_from_delay_slot (XEXP (note, 0)); + delete_from_delay_slot (as_a <rtx_insn *> (XEXP (note, 0))); } else { /* The insn setting CC0 is our previous insn, but it may be in a delay slot. It will be the last insn in the delay slot, if it is. */ - rtx trial = previous_insn (insn); + rtx_insn *trial = previous_insn (insn); if (NOTE_P (trial)) trial = prev_nonnote_insn (trial); if (sets_cc0_p (PATTERN (trial)) != 1 @@ -709,7 +668,6 @@ delete_from_delay_slot (trial); } } -#endif delete_related_insns (insn); } @@ -735,8 +693,6 @@ num_filled_delays[index][slots_filled][reorg_pass_number]++; } -#if defined(ANNUL_IFFALSE_SLOTS) || defined(ANNUL_IFTRUE_SLOTS) - /* Optimize the following cases: 1. When a conditional branch skips over only one instruction, @@ -769,12 +725,11 @@ This should be expanded to skip over N insns, where N is the number of delay slots required. */ -static rtx -optimize_skip (rtx insn) +static void +optimize_skip (rtx_jump_insn *insn, vec<rtx_insn *> *delay_list) { - rtx trial = next_nonnote_insn (insn); - rtx next_trial = next_active_insn (trial); - rtx delay_list = 0; + rtx_insn *trial = next_nonnote_insn (insn); + rtx_insn *next_trial = next_active_insn (trial); int flags; flags = get_jump_flags (insn, JUMP_LABEL (insn)); @@ -785,8 +740,9 @@ || recog_memoized (trial) < 0 || (! eligible_for_annul_false (insn, 0, trial, flags) && ! eligible_for_annul_true (insn, 0, trial, flags)) + || RTX_FRAME_RELATED_P (trial) || can_throw_internal (trial)) - return 0; + return; /* There are two cases where we are just executing one insn (we assume here that a branch requires only one insn; this should be generalized @@ -794,23 +750,20 @@ we have one insn followed by a branch to the same label we branch to. In both of these cases, inverting the jump and annulling the delay slot give the same effect in fewer insns. */ - if ((next_trial == next_active_insn (JUMP_LABEL (insn)) - && ! (next_trial == 0 && crtl->epilogue_delay_list != 0)) + if (next_trial == next_active_insn (JUMP_LABEL_AS_INSN (insn)) || (next_trial != 0 - && JUMP_P (next_trial) - && JUMP_LABEL (insn) == JUMP_LABEL (next_trial) - && (simplejump_p (next_trial) - || GET_CODE (PATTERN (next_trial)) == RETURN))) + && simplejump_or_return_p (next_trial) + && JUMP_LABEL (insn) == JUMP_LABEL (next_trial))) { if (eligible_for_annul_false (insn, 0, trial, flags)) { if (invert_jump (insn, JUMP_LABEL (insn), 1)) INSN_FROM_TARGET_P (trial) = 1; else if (! eligible_for_annul_true (insn, 0, trial, flags)) - return 0; + return; } - delay_list = add_to_delay_list (trial, NULL_RTX); + add_to_delay_list (trial, delay_list); next_trial = next_active_insn (trial); update_block (trial, trial); delete_related_insns (trial); @@ -819,13 +772,11 @@ branch, thread our jump to the target of that branch. Don't change this into a RETURN here, because it may not accept what we have in the delay slot. We'll fix this up later. */ - if (next_trial && JUMP_P (next_trial) - && (simplejump_p (next_trial) - || GET_CODE (PATTERN (next_trial)) == RETURN)) + if (next_trial && simplejump_or_return_p (next_trial)) { rtx target_label = JUMP_LABEL (next_trial); - if (target_label == 0) - target_label = find_end_label (); + if (ANY_RETURN_P (target_label)) + target_label = find_end_label (target_label); if (target_label) { @@ -841,10 +792,7 @@ INSN_ANNULLED_BRANCH_P (insn) = 1; } - - return delay_list; } -#endif /* Encode and return branch direction and prediction information for INSN assuming it will jump to LABEL. @@ -853,7 +801,7 @@ are predicted as very likely taken. */ static int -get_jump_flags (rtx insn, rtx label) +get_jump_flags (const rtx_insn *insn, rtx label) { int flags; @@ -861,12 +809,12 @@ be INSNs, CALL_INSNs, or JUMP_INSNs. Only JUMP_INSNs have branch direction information, and only if they are conditional jumps. - If LABEL is zero, then there is no way to determine the branch + If LABEL is a return, then there is no way to determine the branch direction. */ if (JUMP_P (insn) && (condjump_p (insn) || condjump_in_parallel_p (insn)) + && !ANY_RETURN_P (label) && INSN_UID (insn) <= max_uid - && label != 0 && INSN_UID (label) <= max_uid) flags = (uid_to_ruid[INSN_UID (label)] > uid_to_ruid[INSN_UID (insn)]) @@ -875,110 +823,25 @@ else flags = 0; - /* If insn is a conditional branch call mostly_true_jump to get - determine the branch prediction. - - Non conditional branches are predicted as very likely taken. */ - if (JUMP_P (insn) - && (condjump_p (insn) || condjump_in_parallel_p (insn))) - { - int prediction; - - prediction = mostly_true_jump (insn, get_branch_condition (insn, label)); - switch (prediction) - { - case 2: - flags |= (ATTR_FLAG_very_likely | ATTR_FLAG_likely); - break; - case 1: - flags |= ATTR_FLAG_likely; - break; - case 0: - flags |= ATTR_FLAG_unlikely; - break; - case -1: - flags |= (ATTR_FLAG_very_unlikely | ATTR_FLAG_unlikely); - break; - - default: - gcc_unreachable (); - } - } - else - flags |= (ATTR_FLAG_very_likely | ATTR_FLAG_likely); - return flags; } -/* Return 1 if INSN is a destination that will be branched to rarely (the - return point of a function); return 2 if DEST will be branched to very - rarely (a call to a function that doesn't return). Otherwise, - return 0. */ - -static int -rare_destination (rtx insn) -{ - int jump_count = 0; - rtx next; - - for (; insn; insn = next) - { - if (NONJUMP_INSN_P (insn) && GET_CODE (PATTERN (insn)) == SEQUENCE) - insn = XVECEXP (PATTERN (insn), 0, 0); - - next = NEXT_INSN (insn); - - switch (GET_CODE (insn)) - { - case CODE_LABEL: - return 0; - case BARRIER: - /* A BARRIER can either be after a JUMP_INSN or a CALL_INSN. We - don't scan past JUMP_INSNs, so any barrier we find here must - have been after a CALL_INSN and hence mean the call doesn't - return. */ - return 2; - case JUMP_INSN: - if (GET_CODE (PATTERN (insn)) == RETURN) - return 1; - else if (simplejump_p (insn) - && jump_count++ < 10) - next = JUMP_LABEL (insn); - else - return 0; - - default: - break; - } - } - - /* If we got here it means we hit the end of the function. So this - is an unlikely destination. */ - - return 1; -} - /* Return truth value of the statement that this branch is mostly taken. If we think that the branch is extremely likely to be taken, we return 2. If the branch is slightly more likely to be taken, return 1. If the branch is slightly less likely to be taken, - return 0 and if the branch is highly unlikely to be taken, return -1. - - CONDITION, if nonzero, is the condition that JUMP_INSN is testing. */ + return 0 and if the branch is highly unlikely to be taken, return -1. */ static int -mostly_true_jump (rtx jump_insn, rtx condition) +mostly_true_jump (rtx jump_insn) { - rtx target_label = JUMP_LABEL (jump_insn); - rtx note; - int rare_dest, rare_fallthrough; - /* If branch probabilities are available, then use that number since it always gives a correct answer. */ - note = find_reg_note (jump_insn, REG_BR_PROB, 0); + rtx note = find_reg_note (jump_insn, REG_BR_PROB, 0); if (note) { - int prob = INTVAL (XEXP (note, 0)); + int prob = profile_probability::from_reg_br_prob_note (XINT (note, 0)) + .to_reg_br_prob_base (); if (prob >= REG_BR_PROB_BASE * 9 / 10) return 2; @@ -990,37 +853,9 @@ return -1; } - /* Look at the relative rarities of the fallthrough and destination. If - they differ, we can predict the branch that way. */ - rare_dest = rare_destination (target_label); - rare_fallthrough = rare_destination (NEXT_INSN (jump_insn)); - - switch (rare_fallthrough - rare_dest) - { - case -2: - return -1; - case -1: - return 0; - case 0: - break; - case 1: - return 1; - case 2: - return 2; - } - - /* If we couldn't figure out what this jump was, assume it won't be - taken. This should be rare. */ - if (condition == 0) + /* If there is no note, assume branches are not taken. + This should be rare. */ return 0; - - /* Predict backward branches usually take, forward branches usually not. If - we don't know whether this is forward or backward, assume the branch - will be taken, since most are. */ - return (target_label == 0 || INSN_UID (jump_insn) > max_uid - || INSN_UID (target_label) > max_uid - || (uid_to_ruid[INSN_UID (jump_insn)] - > uid_to_ruid[INSN_UID (target_label)])); } /* Return the condition under which INSN will branch to TARGET. If TARGET @@ -1029,7 +864,7 @@ type of jump, or it doesn't go to TARGET, return 0. */ static rtx -get_branch_condition (rtx insn, rtx target) +get_branch_condition (const rtx_insn *insn, rtx target) { rtx pat = PATTERN (insn); rtx src; @@ -1037,28 +872,28 @@ if (condjump_in_parallel_p (insn)) pat = XVECEXP (pat, 0, 0); - if (GET_CODE (pat) == RETURN) - return target == 0 ? const_true_rtx : 0; - - else if (GET_CODE (pat) != SET || SET_DEST (pat) != pc_rtx) + if (ANY_RETURN_P (pat) && pat == target) + return const_true_rtx; + + if (GET_CODE (pat) != SET || SET_DEST (pat) != pc_rtx) return 0; src = SET_SRC (pat); - if (GET_CODE (src) == LABEL_REF && XEXP (src, 0) == target) + if (GET_CODE (src) == LABEL_REF && label_ref_label (src) == target) return const_true_rtx; else if (GET_CODE (src) == IF_THEN_ELSE - && ((target == 0 && GET_CODE (XEXP (src, 1)) == RETURN) - || (GET_CODE (XEXP (src, 1)) == LABEL_REF - && XEXP (XEXP (src, 1), 0) == target)) - && XEXP (src, 2) == pc_rtx) + && XEXP (src, 2) == pc_rtx + && ((GET_CODE (XEXP (src, 1)) == LABEL_REF + && label_ref_label (XEXP (src, 1)) == target) + || (ANY_RETURN_P (XEXP (src, 1)) && XEXP (src, 1) == target))) return XEXP (src, 0); else if (GET_CODE (src) == IF_THEN_ELSE - && ((target == 0 && GET_CODE (XEXP (src, 2)) == RETURN) - || (GET_CODE (XEXP (src, 2)) == LABEL_REF - && XEXP (XEXP (src, 2), 0) == target)) - && XEXP (src, 1) == pc_rtx) + && XEXP (src, 1) == pc_rtx + && ((GET_CODE (XEXP (src, 2)) == LABEL_REF + && label_ref_label (XEXP (src, 2)) == target) + || (ANY_RETURN_P (XEXP (src, 2)) && XEXP (src, 2) == target))) { enum rtx_code rev; rev = reversed_comparison_code (XEXP (src, 0), insn); @@ -1075,7 +910,7 @@ INSN, i.e., if INSN will always branch if CONDITION is true. */ static int -condition_dominates_p (rtx condition, rtx insn) +condition_dominates_p (rtx condition, const rtx_insn *insn) { rtx other_condition = get_branch_condition (insn, JUMP_LABEL (insn)); enum rtx_code code = GET_CODE (condition); @@ -1101,66 +936,64 @@ any insns already in the delay slot of JUMP. */ static int -redirect_with_delay_slots_safe_p (rtx jump, rtx newlabel, rtx seq) +redirect_with_delay_slots_safe_p (rtx_insn *jump, rtx newlabel, rtx seq) { int flags, i; - rtx pat = PATTERN (seq); + rtx_sequence *pat = as_a <rtx_sequence *> (PATTERN (seq)); /* Make sure all the delay slots of this jump would still be valid after threading the jump. If they are still valid, then return nonzero. */ flags = get_jump_flags (jump, newlabel); - for (i = 1; i < XVECLEN (pat, 0); i++) + for (i = 1; i < pat->len (); i++) if (! ( -#ifdef ANNUL_IFFALSE_SLOTS +#if ANNUL_IFFALSE_SLOTS (INSN_ANNULLED_BRANCH_P (jump) - && INSN_FROM_TARGET_P (XVECEXP (pat, 0, i))) - ? eligible_for_annul_false (jump, i - 1, - XVECEXP (pat, 0, i), flags) : + && INSN_FROM_TARGET_P (pat->insn (i))) + ? eligible_for_annul_false (jump, i - 1, pat->insn (i), flags) : #endif -#ifdef ANNUL_IFTRUE_SLOTS +#if ANNUL_IFTRUE_SLOTS (INSN_ANNULLED_BRANCH_P (jump) && ! INSN_FROM_TARGET_P (XVECEXP (pat, 0, i))) - ? eligible_for_annul_true (jump, i - 1, - XVECEXP (pat, 0, i), flags) : + ? eligible_for_annul_true (jump, i - 1, pat->insn (i), flags) : #endif - eligible_for_delay (jump, i - 1, XVECEXP (pat, 0, i), flags))) + eligible_for_delay (jump, i - 1, pat->insn (i), flags))) break; - return (i == XVECLEN (pat, 0)); + return (i == pat->len ()); } /* Return nonzero if redirecting JUMP to NEWLABEL does not invalidate any insns we wish to place in the delay slot of JUMP. */ static int -redirect_with_delay_list_safe_p (rtx jump, rtx newlabel, rtx delay_list) +redirect_with_delay_list_safe_p (rtx_insn *jump, rtx newlabel, + const vec<rtx_insn *> &delay_list) { - int flags, i; - rtx li; - /* Make sure all the insns in DELAY_LIST would still be valid after threading the jump. If they are still valid, then return nonzero. */ - flags = get_jump_flags (jump, newlabel); - for (li = delay_list, i = 0; li; li = XEXP (li, 1), i++) + int flags = get_jump_flags (jump, newlabel); + unsigned int delay_insns = delay_list.length (); + unsigned int i = 0; + for (; i < delay_insns; i++) if (! ( -#ifdef ANNUL_IFFALSE_SLOTS +#if ANNUL_IFFALSE_SLOTS (INSN_ANNULLED_BRANCH_P (jump) - && INSN_FROM_TARGET_P (XEXP (li, 0))) - ? eligible_for_annul_false (jump, i, XEXP (li, 0), flags) : + && INSN_FROM_TARGET_P (delay_list[i])) + ? eligible_for_annul_false (jump, i, delay_list[i], flags) : #endif -#ifdef ANNUL_IFTRUE_SLOTS +#if ANNUL_IFTRUE_SLOTS (INSN_ANNULLED_BRANCH_P (jump) - && ! INSN_FROM_TARGET_P (XEXP (li, 0))) - ? eligible_for_annul_true (jump, i, XEXP (li, 0), flags) : + && ! INSN_FROM_TARGET_P (delay_list[i])) + ? eligible_for_annul_true (jump, i, delay_list[i], flags) : #endif - eligible_for_delay (jump, i, XEXP (li, 0), flags))) + eligible_for_delay (jump, i, delay_list[i], flags))) break; - return (li == NULL); + return i == delay_insns; } /* DELAY_LIST is a list of insns that have already been placed into delay @@ -1168,21 +1001,15 @@ If not, return 0; otherwise return 1. */ static int -check_annul_list_true_false (int annul_true_p, rtx delay_list) +check_annul_list_true_false (int annul_true_p, + const vec<rtx_insn *> &delay_list) { - rtx temp; - - if (delay_list) - { - for (temp = delay_list; temp; temp = XEXP (temp, 1)) - { - rtx trial = XEXP (temp, 0); - - if ((annul_true_p && INSN_FROM_TARGET_P (trial)) - || (!annul_true_p && !INSN_FROM_TARGET_P (trial))) - return 0; - } - } + rtx_insn *trial; + unsigned int i; + FOR_EACH_VEC_ELT (delay_list, i, trial) + if ((annul_true_p && INSN_FROM_TARGET_P (trial)) + || (!annul_true_p && !INSN_FROM_TARGET_P (trial))) + return 0; return 1; } @@ -1207,22 +1034,22 @@ PNEW_THREAD points to a location that is to receive the place at which execution should continue. */ -static rtx -steal_delay_list_from_target (rtx insn, rtx condition, rtx seq, - rtx delay_list, struct resources *sets, +static void +steal_delay_list_from_target (rtx_insn *insn, rtx condition, rtx_sequence *seq, + vec<rtx_insn *> *delay_list, resources *sets, struct resources *needed, struct resources *other_needed, int slots_to_fill, int *pslots_filled, int *pannul_p, rtx *pnew_thread) { - rtx temp; int slots_remaining = slots_to_fill - *pslots_filled; int total_slots_filled = *pslots_filled; - rtx new_delay_list = 0; + auto_vec<rtx_insn *, 5> new_delay_list; int must_annul = *pannul_p; int used_annul = 0; int i; struct resources cc_set; + bool *redundant; /* We can't do anything if there are more delay slots in SEQ than we can handle, or if we don't know that it will be a taken branch. @@ -1241,84 +1068,94 @@ will effect the direction of the jump in the sequence. */ CLEAR_RESOURCE (&cc_set); - for (temp = delay_list; temp; temp = XEXP (temp, 1)) + + rtx_insn *trial; + FOR_EACH_VEC_ELT (*delay_list, i, trial) { - rtx trial = XEXP (temp, 0); - mark_set_resources (trial, &cc_set, 0, MARK_SRC_DEST_CALL); - if (insn_references_resource_p (XVECEXP (seq , 0, 0), &cc_set, false)) - return delay_list; + if (insn_references_resource_p (seq->insn (0), &cc_set, false)) + return; } if (XVECLEN (seq, 0) - 1 > slots_remaining - || ! condition_dominates_p (condition, XVECEXP (seq, 0, 0)) - || ! single_set (XVECEXP (seq, 0, 0))) - return delay_list; - -#ifdef MD_CAN_REDIRECT_BRANCH + || ! condition_dominates_p (condition, seq->insn (0)) + || ! single_set (seq->insn (0))) + return; + /* On some targets, branches with delay slots can have a limited displacement. Give the back end a chance to tell us we can't do this. */ - if (! MD_CAN_REDIRECT_BRANCH (insn, XVECEXP (seq, 0, 0))) - return delay_list; -#endif - - for (i = 1; i < XVECLEN (seq, 0); i++) + if (! targetm.can_follow_jump (insn, seq->insn (0))) + return; + + redundant = XALLOCAVEC (bool, XVECLEN (seq, 0)); + for (i = 1; i < seq->len (); i++) { - rtx trial = XVECEXP (seq, 0, i); + rtx_insn *trial = seq->insn (i); int flags; if (insn_references_resource_p (trial, sets, false) || insn_sets_resource_p (trial, needed, false) || insn_sets_resource_p (trial, sets, false) -#ifdef HAVE_cc0 /* If TRIAL sets CC0, we can't copy it, so we can't steal this delay list. */ - || find_reg_note (trial, REG_CC_USER, NULL_RTX) -#endif + || (HAVE_cc0 && find_reg_note (trial, REG_CC_USER, NULL_RTX)) /* If TRIAL is from the fallthrough code of an annulled branch insn in SEQ, we cannot use it. */ - || (INSN_ANNULLED_BRANCH_P (XVECEXP (seq, 0, 0)) + || (INSN_ANNULLED_BRANCH_P (seq->insn (0)) && ! INSN_FROM_TARGET_P (trial))) - return delay_list; + return; /* If this insn was already done (usually in a previous delay slot), pretend we put it in our delay slot. */ - if (redundant_insn (trial, insn, new_delay_list)) + redundant[i] = redundant_insn (trial, insn, new_delay_list); + if (redundant[i]) continue; /* We will end up re-vectoring this branch, so compute flags based on jumping to the new label. */ - flags = get_jump_flags (insn, JUMP_LABEL (XVECEXP (seq, 0, 0))); + flags = get_jump_flags (insn, JUMP_LABEL (seq->insn (0))); if (! must_annul && ((condition == const_true_rtx || (! insn_sets_resource_p (trial, other_needed, false) && ! may_trap_or_fault_p (PATTERN (trial))))) ? eligible_for_delay (insn, total_slots_filled, trial, flags) - : (must_annul || (delay_list == NULL && new_delay_list == NULL)) + : (must_annul || (delay_list->is_empty () && new_delay_list.is_empty ())) && (must_annul = 1, - check_annul_list_true_false (0, delay_list) + check_annul_list_true_false (0, *delay_list) && check_annul_list_true_false (0, new_delay_list) && eligible_for_annul_false (insn, total_slots_filled, trial, flags))) { if (must_annul) - used_annul = 1; - temp = copy_rtx (trial); + { + /* Frame related instructions cannot go into annulled delay + slots, it messes up the dwarf info. */ + if (RTX_FRAME_RELATED_P (trial)) + return; + used_annul = 1; + } + rtx_insn *temp = copy_delay_slot_insn (trial); INSN_FROM_TARGET_P (temp) = 1; - new_delay_list = add_to_delay_list (temp, new_delay_list); + add_to_delay_list (temp, &new_delay_list); total_slots_filled++; if (--slots_remaining == 0) break; } else - return delay_list; + return; } + /* Record the effect of the instructions that were redundant and which + we therefore decided not to copy. */ + for (i = 1; i < seq->len (); i++) + if (redundant[i]) + update_block (seq->insn (i), insn); + /* Show the place to which we will be branching. */ - *pnew_thread = next_active_insn (JUMP_LABEL (XVECEXP (seq, 0, 0))); + *pnew_thread = first_active_target_insn (JUMP_LABEL (seq->insn (0))); /* Add any new insns to the delay list and update the count of the number of slots filled. */ @@ -1326,13 +1163,9 @@ if (used_annul) *pannul_p = 1; - if (delay_list == 0) - return new_delay_list; - - for (temp = new_delay_list; temp; temp = XEXP (temp, 1)) - delay_list = add_to_delay_list (XEXP (temp, 0), delay_list); - - return delay_list; + rtx_insn *temp; + FOR_EACH_VEC_ELT (new_delay_list, i, temp) + add_to_delay_list (temp, delay_list); } /* Similar to steal_delay_list_from_target except that SEQ is on the @@ -1340,9 +1173,11 @@ of SEQ is an unconditional branch. In that case we steal its delay slot for INSN since unconditional branches are much easier to fill. */ -static rtx -steal_delay_list_from_fallthrough (rtx insn, rtx condition, rtx seq, - rtx delay_list, struct resources *sets, +static void +steal_delay_list_from_fallthrough (rtx_insn *insn, rtx condition, + rtx_sequence *seq, + vec<rtx_insn *> *delay_list, + struct resources *sets, struct resources *needed, struct resources *other_needed, int slots_to_fill, int *pslots_filled, @@ -1358,29 +1193,26 @@ /* We can't do anything if SEQ's delay insn isn't an unconditional branch. */ - if (! simplejump_p (XVECEXP (seq, 0, 0)) - && GET_CODE (PATTERN (XVECEXP (seq, 0, 0))) != RETURN) - return delay_list; - - for (i = 1; i < XVECLEN (seq, 0); i++) + if (! simplejump_or_return_p (seq->insn (0))) + return; + + for (i = 1; i < seq->len (); i++) { - rtx trial = XVECEXP (seq, 0, i); + rtx_insn *trial = seq->insn (i); /* If TRIAL sets CC0, stealing it will move it too far from the use of CC0. */ if (insn_references_resource_p (trial, sets, false) || insn_sets_resource_p (trial, needed, false) || insn_sets_resource_p (trial, sets, false) -#ifdef HAVE_cc0 - || sets_cc0_p (PATTERN (trial)) -#endif - ) + || (HAVE_cc0 && sets_cc0_p (PATTERN (trial)))) break; /* If this insn was already done, we don't need it. */ - if (redundant_insn (trial, insn, delay_list)) + if (redundant_insn (trial, insn, *delay_list)) { + update_block (trial, insn); delete_from_delay_slot (trial); continue; } @@ -1390,14 +1222,14 @@ || (! insn_sets_resource_p (trial, other_needed, false) && ! may_trap_or_fault_p (PATTERN (trial))))) ? eligible_for_delay (insn, *pslots_filled, trial, flags) - : (must_annul || delay_list == NULL) && (must_annul = 1, - check_annul_list_true_false (1, delay_list) + : (must_annul || delay_list->is_empty ()) && (must_annul = 1, + check_annul_list_true_false (1, *delay_list) && eligible_for_annul_true (insn, *pslots_filled, trial, flags))) { if (must_annul) used_annul = 1; delete_from_delay_slot (trial); - delay_list = add_to_delay_list (trial, delay_list); + add_to_delay_list (trial, delay_list); if (++(*pslots_filled) == slots_to_fill) break; @@ -1408,7 +1240,6 @@ if (used_annul) *pannul_p = 1; - return delay_list; } /* Try merging insns starting at THREAD which match exactly the insns in @@ -1421,17 +1252,16 @@ we delete the merged insn. */ static void -try_merge_delay_insns (rtx insn, rtx thread) +try_merge_delay_insns (rtx_insn *insn, rtx_insn *thread) { - rtx trial, next_trial; - rtx delay_insn = XVECEXP (PATTERN (insn), 0, 0); - int annul_p = INSN_ANNULLED_BRANCH_P (delay_insn); + rtx_insn *trial, *next_trial; + rtx_insn *delay_insn = as_a <rtx_insn *> (XVECEXP (PATTERN (insn), 0, 0)); + int annul_p = JUMP_P (delay_insn) && INSN_ANNULLED_BRANCH_P (delay_insn); int slot_number = 1; int num_slots = XVECLEN (PATTERN (insn), 0); rtx next_to_match = XVECEXP (PATTERN (insn), 0, slot_number); - struct resources set, needed; - rtx merged_insns = 0; - int i; + struct resources set, needed, modified; + auto_vec<std::pair<rtx_insn *, bool>, 10> merged_insns; int flags; flags = get_jump_flags (delay_insn, JUMP_LABEL (delay_insn)); @@ -1446,7 +1276,7 @@ will essentially disable this optimization. This method is somewhat of a kludge, but I don't see a better way.) */ if (! annul_p) - for (i = 1 ; i < num_slots; i++) + for (int i = 1; i < num_slots; i++) if (XVECEXP (PATTERN (insn), 0, i)) mark_referenced_resources (XVECEXP (PATTERN (insn), 0, i), &needed, true); @@ -1464,10 +1294,8 @@ continue; if (GET_CODE (next_to_match) == GET_CODE (trial) -#ifdef HAVE_cc0 /* We can't share an insn that sets cc0. */ - && ! sets_cc0_p (pat) -#endif + && (!HAVE_cc0 || ! sets_cc0_p (pat)) && ! insn_references_resource_p (trial, &set, true) && ! insn_sets_resource_p (trial, &set, true) && ! insn_sets_resource_p (trial, &needed, true) @@ -1492,7 +1320,7 @@ INSN_FROM_TARGET_P (next_to_match) = 0; } else - merged_insns = gen_rtx_INSN_LIST (VOIDmode, trial, merged_insns); + merged_insns.safe_push (std::pair<rtx_insn *, bool> (trial, false)); if (++slot_number == num_slots) break; @@ -1509,41 +1337,56 @@ if (slot_number != num_slots && trial && NONJUMP_INSN_P (trial) && GET_CODE (PATTERN (trial)) == SEQUENCE - && ! INSN_ANNULLED_BRANCH_P (XVECEXP (PATTERN (trial), 0, 0))) + && !(JUMP_P (XVECEXP (PATTERN (trial), 0, 0)) + && INSN_ANNULLED_BRANCH_P (XVECEXP (PATTERN (trial), 0, 0)))) { - rtx pat = PATTERN (trial); + rtx_sequence *pat = as_a <rtx_sequence *> (PATTERN (trial)); rtx filled_insn = XVECEXP (pat, 0, 0); /* Account for resources set/needed by the filled insn. */ mark_set_resources (filled_insn, &set, 0, MARK_SRC_DEST_CALL); mark_referenced_resources (filled_insn, &needed, true); - for (i = 1; i < XVECLEN (pat, 0); i++) + for (int i = 1; i < pat->len (); i++) { - rtx dtrial = XVECEXP (pat, 0, i); - + rtx_insn *dtrial = pat->insn (i); + + CLEAR_RESOURCE (&modified); + /* Account for resources set by the insn following NEXT_TO_MATCH + inside INSN's delay list. */ + for (int j = 1; slot_number + j < num_slots; j++) + mark_set_resources (XVECEXP (PATTERN (insn), 0, slot_number + j), + &modified, 0, MARK_SRC_DEST_CALL); + /* Account for resources set by the insn before DTRIAL and inside + TRIAL's delay list. */ + for (int j = 1; j < i; j++) + mark_set_resources (XVECEXP (pat, 0, j), + &modified, 0, MARK_SRC_DEST_CALL); if (! insn_references_resource_p (dtrial, &set, true) && ! insn_sets_resource_p (dtrial, &set, true) && ! insn_sets_resource_p (dtrial, &needed, true) -#ifdef HAVE_cc0 - && ! sets_cc0_p (PATTERN (dtrial)) -#endif + && (!HAVE_cc0 || ! sets_cc0_p (PATTERN (dtrial))) && rtx_equal_p (PATTERN (next_to_match), PATTERN (dtrial)) + /* Check that DTRIAL and NEXT_TO_MATCH does not reference a + resource modified between them (only dtrial is checked because + next_to_match and dtrial shall to be equal in order to hit + this line) */ + && ! insn_references_resource_p (dtrial, &modified, true) && eligible_for_delay (delay_insn, slot_number - 1, dtrial, flags)) { if (! annul_p) { - rtx new_rtx; + rtx_insn *new_rtx; update_block (dtrial, thread); new_rtx = delete_from_delay_slot (dtrial); - if (INSN_DELETED_P (thread)) + if (thread->deleted ()) thread = new_rtx; INSN_FROM_TARGET_P (next_to_match) = 0; } else - merged_insns = gen_rtx_INSN_LIST (SImode, dtrial, - merged_insns); + merged_insns.safe_push (std::pair<rtx_insn *, bool> (dtrial, + true)); if (++slot_number == num_slots) break; @@ -1567,27 +1410,24 @@ target. */ if (slot_number == num_slots && annul_p) { - for (; merged_insns; merged_insns = XEXP (merged_insns, 1)) - { - if (GET_MODE (merged_insns) == SImode) - { - rtx new_rtx; - - update_block (XEXP (merged_insns, 0), thread); - new_rtx = delete_from_delay_slot (XEXP (merged_insns, 0)); - if (INSN_DELETED_P (thread)) - thread = new_rtx; - } - else - { - update_block (XEXP (merged_insns, 0), thread); - delete_related_insns (XEXP (merged_insns, 0)); - } - } + unsigned int len = merged_insns.length (); + for (unsigned int i = len - 1; i < len; i--) + if (merged_insns[i].second) + { + update_block (merged_insns[i].first, thread); + rtx_insn *new_rtx = delete_from_delay_slot (merged_insns[i].first); + if (thread->deleted ()) + thread = new_rtx; + } + else + { + update_block (merged_insns[i].first, thread); + delete_related_insns (merged_insns[i].first); + } INSN_ANNULLED_BRANCH_P (delay_insn) = 0; - for (i = 0; i < XVECLEN (PATTERN (insn), 0); i++) + for (int i = 0; i < XVECLEN (PATTERN (insn), 0); i++) INSN_FROM_TARGET_P (XVECEXP (PATTERN (insn), 0, i)) = 0; } } @@ -1611,12 +1451,13 @@ redundant insn, but the cost of splitting seems greater than the possible gain in rare cases. */ -static rtx -redundant_insn (rtx insn, rtx target, rtx delay_list) +static rtx_insn * +redundant_insn (rtx insn, rtx_insn *target, const vec<rtx_insn *> &delay_list) { rtx target_main = target; rtx ipat = PATTERN (insn); - rtx trial, pat; + rtx_insn *trial; + rtx pat; struct resources needed, set; int i; unsigned insns_to_search; @@ -1632,10 +1473,13 @@ trial && insns_to_search > 0; trial = PREV_INSN (trial)) { - if (LABEL_P (trial)) + /* (use (insn))s can come immediately after a barrier if the + label that used to precede them has been deleted as dead. + See delete_related_insns. */ + if (LABEL_P (trial) || BARRIER_P (trial)) return 0; - if (!NONDEBUG_INSN_P (trial)) + if (!INSN_P (trial)) continue; --insns_to_search; @@ -1643,33 +1487,29 @@ if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER) continue; - if (GET_CODE (pat) == SEQUENCE) + if (rtx_sequence *seq = dyn_cast <rtx_sequence *> (pat)) { /* Stop for a CALL and its delay slots because it is difficult to track its resource needs correctly. */ - if (CALL_P (XVECEXP (pat, 0, 0))) + if (CALL_P (seq->element (0))) return 0; /* Stop for an INSN or JUMP_INSN with delayed effects and its delay slots because it is difficult to track its resource needs correctly. */ -#ifdef INSN_SETS_ARE_DELAYED - if (INSN_SETS_ARE_DELAYED (XVECEXP (pat, 0, 0))) + if (INSN_SETS_ARE_DELAYED (seq->insn (0))) return 0; -#endif - -#ifdef INSN_REFERENCES_ARE_DELAYED - if (INSN_REFERENCES_ARE_DELAYED (XVECEXP (pat, 0, 0))) + + if (INSN_REFERENCES_ARE_DELAYED (seq->insn (0))) return 0; -#endif /* See if any of the insns in the delay slot match, updating resource requirements as we go. */ - for (i = XVECLEN (pat, 0) - 1; i > 0; i--) - if (GET_CODE (XVECEXP (pat, 0, i)) == GET_CODE (insn) - && rtx_equal_p (PATTERN (XVECEXP (pat, 0, i)), ipat) - && ! find_reg_note (XVECEXP (pat, 0, i), REG_UNUSED, NULL_RTX)) + for (i = seq->len () - 1; i > 0; i--) + if (GET_CODE (seq->element (i)) == GET_CODE (insn) + && rtx_equal_p (PATTERN (seq->element (i)), ipat) + && ! find_reg_note (seq->element (i), REG_UNUSED, NULL_RTX)) break; /* If found a match, exit this loop early. */ @@ -1699,9 +1539,7 @@ target_main = XVECEXP (PATTERN (target), 0, 0); if (resource_conflicts_p (&needed, &set) -#ifdef HAVE_cc0 - || reg_mentioned_p (cc0_rtx, ipat) -#endif + || (HAVE_cc0 && reg_mentioned_p (cc0_rtx, ipat)) /* The insn requiring the delay may not set anything needed or set by INSN. */ || insn_sets_resource_p (target_main, &needed, true) @@ -1711,18 +1549,16 @@ /* Insns we pass may not set either NEEDED or SET, so merge them for simpler tests. */ needed.memory |= set.memory; - needed.unch_memory |= set.unch_memory; IOR_HARD_REG_SET (needed.regs, set.regs); /* This insn isn't redundant if it conflicts with an insn that either is or will be in a delay slot of TARGET. */ - while (delay_list) - { - if (insn_sets_resource_p (XEXP (delay_list, 0), &needed, true)) - return 0; - delay_list = XEXP (delay_list, 1); - } + unsigned int j; + rtx_insn *temp; + FOR_EACH_VEC_ELT (delay_list, j, temp) + if (insn_sets_resource_p (temp, &needed, true)) + return 0; if (NONJUMP_INSN_P (target) && GET_CODE (PATTERN (target)) == SEQUENCE) for (i = 1; i < XVECLEN (PATTERN (target), 0); i++) @@ -1738,7 +1574,7 @@ trial && !LABEL_P (trial) && insns_to_search > 0; trial = PREV_INSN (trial)) { - if (!NONDEBUG_INSN_P (trial)) + if (!INSN_P (trial)) continue; --insns_to_search; @@ -1746,37 +1582,38 @@ if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER) continue; - if (GET_CODE (pat) == SEQUENCE) + if (rtx_sequence *seq = dyn_cast <rtx_sequence *> (pat)) { + bool annul_p = false; + rtx_insn *control = seq->insn (0); + /* If this is a CALL_INSN and its delay slots, it is hard to track the resource needs properly, so give up. */ - if (CALL_P (XVECEXP (pat, 0, 0))) + if (CALL_P (control)) return 0; /* If this is an INSN or JUMP_INSN with delayed effects, it is hard to track the resource needs properly, so give up. */ -#ifdef INSN_SETS_ARE_DELAYED - if (INSN_SETS_ARE_DELAYED (XVECEXP (pat, 0, 0))) + if (INSN_SETS_ARE_DELAYED (control)) return 0; -#endif - -#ifdef INSN_REFERENCES_ARE_DELAYED - if (INSN_REFERENCES_ARE_DELAYED (XVECEXP (pat, 0, 0))) + + if (INSN_REFERENCES_ARE_DELAYED (control)) return 0; -#endif + + if (JUMP_P (control)) + annul_p = INSN_ANNULLED_BRANCH_P (control); /* See if any of the insns in the delay slot match, updating resource requirements as we go. */ - for (i = XVECLEN (pat, 0) - 1; i > 0; i--) + for (i = seq->len () - 1; i > 0; i--) { - rtx candidate = XVECEXP (pat, 0, i); + rtx_insn *candidate = seq->insn (i); /* If an insn will be annulled if the branch is false, it isn't considered as a possible duplicate insn. */ if (rtx_equal_p (PATTERN (candidate), ipat) - && ! (INSN_ANNULLED_BRANCH_P (XVECEXP (pat, 0, 0)) - && INSN_FROM_TARGET_P (candidate))) + && ! (annul_p && INSN_FROM_TARGET_P (candidate))) { /* Show that this insn will be used in the sequel. */ INSN_FROM_TARGET_P (candidate) = 0; @@ -1785,15 +1622,14 @@ /* Unless this is an annulled insn from the target of a branch, we must stop if it sets anything needed or set by INSN. */ - if ((! INSN_ANNULLED_BRANCH_P (XVECEXP (pat, 0, 0)) - || ! INSN_FROM_TARGET_P (candidate)) + if ((!annul_p || !INSN_FROM_TARGET_P (candidate)) && insn_sets_resource_p (candidate, &needed, true)) return 0; } /* If the insn requiring the delay slot conflicts with INSN, we must stop. */ - if (insn_sets_resource_p (XVECEXP (pat, 0, 0), &needed, true)) + if (insn_sets_resource_p (control, &needed, true)) return 0; } else @@ -1823,17 +1659,20 @@ static int own_thread_p (rtx thread, rtx label, int allow_fallthrough) { - rtx active_insn; - rtx insn; + rtx_insn *active_insn; + rtx_insn *insn; /* We don't own the function end. */ - if (thread == 0) + if (thread == 0 || ANY_RETURN_P (thread)) return 0; - /* Get the first active insn, or THREAD, if it is an active insn. */ - active_insn = next_active_insn (PREV_INSN (thread)); - - for (insn = thread; insn != active_insn; insn = NEXT_INSN (insn)) + /* We have a non-NULL insn. */ + rtx_insn *thread_insn = as_a <rtx_insn *> (thread); + + /* Get the first active insn, or THREAD_INSN, if it is an active insn. */ + active_insn = next_active_insn (PREV_INSN (thread_insn)); + + for (insn = thread_insn; insn != active_insn; insn = NEXT_INSN (insn)) if (LABEL_P (insn) && (insn != label || LABEL_NUSES (insn) != 1)) return 0; @@ -1842,7 +1681,7 @@ return 1; /* Ensure that we reach a BARRIER before any insn or label. */ - for (insn = prev_nonnote_insn (thread); + for (insn = prev_nonnote_insn (thread_insn); insn == 0 || !BARRIER_P (insn); insn = prev_nonnote_insn (insn)) if (insn == 0 @@ -1856,27 +1695,20 @@ } /* Called when INSN is being moved from a location near the target of a jump. - We leave a marker of the form (use (INSN)) immediately in front - of WHERE for mark_target_live_regs. These markers will be deleted when - reorg finishes. + We leave a marker of the form (use (INSN)) immediately in front of WHERE + for mark_target_live_regs. These markers will be deleted at the end. We used to try to update the live status of registers if WHERE is at the start of a basic block, but that can't work since we may remove a BARRIER in relax_delay_slots. */ static void -update_block (rtx insn, rtx where) +update_block (rtx_insn *insn, rtx_insn *where) { - /* Ignore if this was in a delay slot and it came from the target of - a branch. */ - if (INSN_FROM_TARGET_P (insn)) - return; - emit_insn_before (gen_rtx_USE (VOIDmode, insn), where); /* INSN might be making a value live in a block where it didn't use to be. So recompute liveness information for this block. */ - incr_ticks_for_insn (insn); } @@ -1884,7 +1716,7 @@ the basic block containing the jump. */ static int -reorg_redirect_jump (rtx jump, rtx nlabel) +reorg_redirect_jump (rtx_jump_insn *jump, rtx nlabel) { incr_ticks_for_insn (jump); return redirect_jump (jump, nlabel, 1); @@ -1902,9 +1734,10 @@ is dead because it sees a REG_DEAD note immediately before a CODE_LABEL. */ static void -update_reg_dead_notes (rtx insn, rtx delayed_insn) +update_reg_dead_notes (rtx_insn *insn, rtx_insn *delayed_insn) { - rtx p, link, next; + rtx link, next; + rtx_insn *p; for (p = next_nonnote_insn (insn); p != delayed_insn; p = next_nonnote_insn (p)) @@ -1935,9 +1768,10 @@ confused into thinking the register is dead. */ static void -fix_reg_dead_note (rtx start_insn, rtx stop_insn) +fix_reg_dead_note (rtx_insn *start_insn, rtx stop_insn) { - rtx p, link, next; + rtx link, next; + rtx_insn *p; for (p = next_nonnote_insn (start_insn); p != stop_insn; p = next_nonnote_insn (p)) @@ -1965,7 +1799,7 @@ does. */ static void -update_reg_unused_notes (rtx insn, rtx redundant_insn) +update_reg_unused_notes (rtx_insn *insn, rtx redundant_insn) { rtx link, next; @@ -1983,12 +1817,17 @@ } } -/* Return the label before INSN, or put a new label there. */ - -static rtx -get_label_before (rtx insn) +static vec <rtx> sibling_labels; + +/* Return the label before INSN, or put a new label there. If SIBLING is + non-zero, it is another label associated with the new label (if any), + typically the former target of the jump that will be redirected to + the new label. */ + +static rtx_insn * +get_label_before (rtx_insn *insn, rtx sibling) { - rtx label; + rtx_insn *label; /* Find an existing label at this point or make a new one if there is none. */ @@ -1996,11 +1835,16 @@ if (label == 0 || !LABEL_P (label)) { - rtx prev = PREV_INSN (insn); + rtx_insn *prev = PREV_INSN (insn); label = gen_label_rtx (); emit_label_after (label, prev); LABEL_NUSES (label) = 0; + if (sibling) + { + sibling_labels.safe_push (label); + sibling_labels.safe_push (sibling); + } } return label; } @@ -2022,12 +1866,13 @@ static void fill_simple_delay_slots (int non_jumps_p) { - rtx insn, pat, trial, next_trial; + rtx_insn *insn, *trial, *next_trial; + rtx pat; int i; int num_unfilled_slots = unfilled_slots_next - unfilled_slots_base; struct resources needed, set; int slots_to_fill, slots_filled; - rtx delay_list; + auto_vec<rtx_insn *, 5> delay_list; for (i = 0; i < num_unfilled_slots; i++) { @@ -2037,7 +1882,7 @@ insn = unfilled_slots_base[i]; if (insn == 0 - || INSN_DELETED_P (insn) + || insn->deleted () || (NONJUMP_INSN_P (insn) && GET_CODE (PATTERN (insn)) == SEQUENCE) || (JUMP_P (insn) && non_jumps_p) @@ -2083,7 +1928,7 @@ CALL_INSNs. */ slots_filled = 0; - delay_list = 0; + delay_list.truncate (0); if (JUMP_P (insn)) flags = get_jump_flags (insn, JUMP_LABEL (insn)); @@ -2097,9 +1942,9 @@ && no_labels_between_p (insn, trial) && ! can_throw_internal (trial)) { - rtx *tmp; + rtx_insn **tmp; slots_filled++; - delay_list = add_to_delay_list (trial, delay_list); + add_to_delay_list (trial, &delay_list); /* TRIAL may have had its delay slot filled, then unfilled. When the delay slot is unfilled, TRIAL is placed back on the unfilled @@ -2116,12 +1961,12 @@ if (*tmp == trial) *tmp = 0; { - rtx next = NEXT_INSN (trial); - rtx prev = PREV_INSN (trial); + rtx_insn *next = NEXT_INSN (trial); + rtx_insn *prev = PREV_INSN (trial); if (prev) - NEXT_INSN (prev) = next; + SET_NEXT_INSN (prev) = next; if (next) - PREV_INSN (next) = prev; + SET_PREV_INSN (next) = prev; } } @@ -2139,9 +1984,24 @@ if (slots_filled < slots_to_fill) { + /* If the flags register is dead after the insn, then we want to be + able to accept a candidate that clobbers it. For this purpose, + we need to filter the flags register during life analysis, so + that it doesn't create RAW and WAW dependencies, while still + creating the necessary WAR dependencies. */ + bool filter_flags + = (slots_to_fill == 1 + && targetm.flags_regnum != INVALID_REGNUM + && find_regno_note (insn, REG_DEAD, targetm.flags_regnum)); + struct resources fset; CLEAR_RESOURCE (&needed); CLEAR_RESOURCE (&set); mark_set_resources (insn, &set, 0, MARK_SRC_DEST); + if (filter_flags) + { + CLEAR_RESOURCE (&fset); + mark_set_resources (insn, &fset, 0, MARK_SRC_DEST); + } mark_referenced_resources (insn, &needed, false); for (trial = prev_nonnote_insn (insn); ! stop_search_p (trial, 1); @@ -2152,19 +2012,19 @@ /* This must be an INSN or CALL_INSN. */ pat = PATTERN (trial); - /* USE and CLOBBER at this level was just for flow; ignore it. */ + /* Stand-alone USE and CLOBBER are just for flow. */ if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER) continue; /* Check for resource conflict first, to avoid unnecessary splitting. */ if (! insn_references_resource_p (trial, &set, true) - && ! insn_sets_resource_p (trial, &set, true) + && ! insn_sets_resource_p (trial, + filter_flags ? &fset : &set, + true) && ! insn_sets_resource_p (trial, &needed, true) -#ifdef HAVE_cc0 /* Can't separate set of cc0 from its use. */ - && ! (reg_mentioned_p (cc0_rtx, pat) && ! sets_cc0_p (pat)) -#endif + && (!HAVE_cc0 || ! (reg_mentioned_p (cc0_rtx, pat) && ! sets_cc0_p (pat))) && ! can_throw_internal (trial)) { trial = try_split (pat, trial, 1); @@ -2177,8 +2037,7 @@ tail, of the list. */ update_reg_dead_notes (trial, insn); - delay_list = gen_rtx_INSN_LIST (VOIDmode, - trial, delay_list); + delay_list.safe_insert (0, trial); update_block (trial, trial); delete_related_insns (trial); if (slots_to_fill == ++slots_filled) @@ -2188,6 +2047,18 @@ } mark_set_resources (trial, &set, 0, MARK_SRC_DEST_CALL); + if (filter_flags) + { + mark_set_resources (trial, &fset, 0, MARK_SRC_DEST_CALL); + /* If the flags register is set, then it doesn't create RAW + dependencies any longer and it also doesn't create WAW + dependencies since it's dead after the original insn. */ + if (TEST_HARD_REG_BIT (fset.regs, targetm.flags_regnum)) + { + CLEAR_HARD_REG_BIT (needed.regs, targetm.flags_regnum); + CLEAR_HARD_REG_BIT (fset.regs, targetm.flags_regnum); + } + } mark_referenced_resources (trial, &needed, true); } } @@ -2195,17 +2066,17 @@ /* If all needed slots haven't been filled, we come here. */ /* Try to optimize case of jumping around a single insn. */ -#if defined(ANNUL_IFFALSE_SLOTS) || defined(ANNUL_IFTRUE_SLOTS) - if (slots_filled != slots_to_fill - && delay_list == 0 + if ((ANNUL_IFTRUE_SLOTS || ANNUL_IFFALSE_SLOTS) + && slots_filled != slots_to_fill + && delay_list.is_empty () && JUMP_P (insn) - && (condjump_p (insn) || condjump_in_parallel_p (insn))) + && (condjump_p (insn) || condjump_in_parallel_p (insn)) + && !ANY_RETURN_P (JUMP_LABEL (insn))) { - delay_list = optimize_skip (insn); - if (delay_list) + optimize_skip (as_a <rtx_jump_insn *> (insn), &delay_list); + if (!delay_list.is_empty ()) slots_filled += 1; } -#endif /* Try to get insns from beyond the insn needing the delay slot. These insns can neither set or reference resources set in insns being @@ -2242,102 +2113,76 @@ Presumably, we should also check to see if we could get back to this function via `setjmp'. */ && ! can_throw_internal (insn) - && (!JUMP_P (insn) - || ((condjump_p (insn) || condjump_in_parallel_p (insn)) - && ! simplejump_p (insn) - && JUMP_LABEL (insn) != 0))) + && !JUMP_P (insn)) { - /* Invariant: If insn is a JUMP_INSN, the insn's jump - label. Otherwise, zero. */ - rtx target = 0; int maybe_never = 0; rtx pat, trial_delay; CLEAR_RESOURCE (&needed); CLEAR_RESOURCE (&set); + mark_set_resources (insn, &set, 0, MARK_SRC_DEST_CALL); + mark_referenced_resources (insn, &needed, true); if (CALL_P (insn)) - { - mark_set_resources (insn, &set, 0, MARK_SRC_DEST_CALL); - mark_referenced_resources (insn, &needed, true); - maybe_never = 1; - } - else + maybe_never = 1; + + for (trial = next_nonnote_insn (insn); !stop_search_p (trial, 1); + trial = next_trial) { - mark_set_resources (insn, &set, 0, MARK_SRC_DEST_CALL); - mark_referenced_resources (insn, &needed, true); - if (JUMP_P (insn)) - target = JUMP_LABEL (insn); - } - - if (target == 0) - for (trial = next_nonnote_insn (insn); trial; trial = next_trial) - { - next_trial = next_nonnote_insn (trial); - - if (LABEL_P (trial) - || BARRIER_P (trial)) - break; - - /* We must have an INSN, JUMP_INSN, or CALL_INSN. */ - pat = PATTERN (trial); - - /* Stand-alone USE and CLOBBER are just for flow. */ - if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER) + next_trial = next_nonnote_insn (trial); + + /* This must be an INSN or CALL_INSN. */ + pat = PATTERN (trial); + + /* Stand-alone USE and CLOBBER are just for flow. */ + if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER) + continue; + + /* If this already has filled delay slots, get the insn needing + the delay slots. */ + if (GET_CODE (pat) == SEQUENCE) + trial_delay = XVECEXP (pat, 0, 0); + else + trial_delay = trial; + + /* Stop our search when seeing a jump. */ + if (JUMP_P (trial_delay)) + break; + + /* See if we have a resource problem before we try to split. */ + if (GET_CODE (pat) != SEQUENCE + && ! insn_references_resource_p (trial, &set, true) + && ! insn_sets_resource_p (trial, &set, true) + && ! insn_sets_resource_p (trial, &needed, true) + && (!HAVE_cc0 && ! (reg_mentioned_p (cc0_rtx, pat) && ! sets_cc0_p (pat))) + && ! (maybe_never && may_trap_or_fault_p (pat)) + && (trial = try_split (pat, trial, 0)) + && eligible_for_delay (insn, slots_filled, trial, flags) + && ! can_throw_internal (trial)) + { + next_trial = next_nonnote_insn (trial); + add_to_delay_list (trial, &delay_list); + if (HAVE_cc0 && reg_mentioned_p (cc0_rtx, pat)) + link_cc0_insns (trial); + + delete_related_insns (trial); + if (slots_to_fill == ++slots_filled) + break; continue; - - /* If this already has filled delay slots, get the insn needing - the delay slots. */ - if (GET_CODE (pat) == SEQUENCE) - trial_delay = XVECEXP (pat, 0, 0); - else - trial_delay = trial; - - /* Stop our search when seeing an unconditional jump. */ - if (JUMP_P (trial_delay)) - break; - - /* See if we have a resource problem before we try to - split. */ - if (GET_CODE (pat) != SEQUENCE - && ! insn_references_resource_p (trial, &set, true) - && ! insn_sets_resource_p (trial, &set, true) - && ! insn_sets_resource_p (trial, &needed, true) -#ifdef HAVE_cc0 - && ! (reg_mentioned_p (cc0_rtx, pat) && ! sets_cc0_p (pat)) -#endif - && ! (maybe_never && may_trap_or_fault_p (pat)) - && (trial = try_split (pat, trial, 0)) - && eligible_for_delay (insn, slots_filled, trial, flags) - && ! can_throw_internal(trial)) - { - next_trial = next_nonnote_insn (trial); - delay_list = add_to_delay_list (trial, delay_list); - -#ifdef HAVE_cc0 - if (reg_mentioned_p (cc0_rtx, pat)) - link_cc0_insns (trial); -#endif - - delete_related_insns (trial); - if (slots_to_fill == ++slots_filled) - break; - continue; - } - - mark_set_resources (trial, &set, 0, MARK_SRC_DEST_CALL); - mark_referenced_resources (trial, &needed, true); - - /* Ensure we don't put insns between the setting of cc and the - comparison by moving a setting of cc into an earlier delay - slot since these insns could clobber the condition code. */ - set.cc = 1; - - /* If this is a call or jump, we might not get here. */ - if (CALL_P (trial_delay) - || JUMP_P (trial_delay)) - maybe_never = 1; - } + } + + mark_set_resources (trial, &set, 0, MARK_SRC_DEST_CALL); + mark_referenced_resources (trial, &needed, true); + + /* Ensure we don't put insns between the setting of cc and the + comparison by moving a setting of cc into an earlier delay + slot since these insns could clobber the condition code. */ + set.cc = 1; + + /* If this is a call, we might not get here. */ + if (CALL_P (trial_delay)) + maybe_never = 1; + } /* If there are slots left to fill and our search was stopped by an unconditional branch, try the insn at the branch target. We can @@ -2346,19 +2191,16 @@ Don't do this if the insn at the branch target is a branch. */ if (slots_to_fill != slots_filled && trial - && JUMP_P (trial) + && jump_to_label_p (trial) && simplejump_p (trial) - && (target == 0 || JUMP_LABEL (trial) == target) - && (next_trial = next_active_insn (JUMP_LABEL (trial))) != 0 + && (next_trial = next_active_insn (JUMP_LABEL_AS_INSN (trial))) != 0 && ! (NONJUMP_INSN_P (next_trial) && GET_CODE (PATTERN (next_trial)) == SEQUENCE) && !JUMP_P (next_trial) && ! insn_references_resource_p (next_trial, &set, true) && ! insn_sets_resource_p (next_trial, &set, true) && ! insn_sets_resource_p (next_trial, &needed, true) -#ifdef HAVE_cc0 - && ! reg_mentioned_p (cc0_rtx, PATTERN (next_trial)) -#endif + && (!HAVE_cc0 || ! reg_mentioned_p (cc0_rtx, PATTERN (next_trial))) && ! (maybe_never && may_trap_or_fault_p (PATTERN (next_trial))) && (next_trial = try_split (PATTERN (next_trial), next_trial, 0)) && eligible_for_delay (insn, slots_filled, next_trial, flags) @@ -2366,43 +2208,37 @@ { /* See comment in relax_delay_slots about necessity of using next_real_insn here. */ - rtx new_label = next_real_insn (next_trial); + rtx_insn *new_label = next_real_insn (next_trial); if (new_label != 0) - new_label = get_label_before (new_label); + new_label = get_label_before (new_label, JUMP_LABEL (trial)); else - new_label = find_end_label (); + new_label = find_end_label (simple_return_rtx); if (new_label) { - delay_list - = add_to_delay_list (copy_rtx (next_trial), delay_list); + add_to_delay_list (copy_delay_slot_insn (next_trial), + &delay_list); slots_filled++; - reorg_redirect_jump (trial, new_label); - - /* If we merged because we both jumped to the same place, - redirect the original insn also. */ - if (target) - reorg_redirect_jump (insn, new_label); + reorg_redirect_jump (as_a <rtx_jump_insn *> (trial), + new_label); } } } /* If this is an unconditional jump, then try to get insns from the target of the jump. */ - if (JUMP_P (insn) - && simplejump_p (insn) + rtx_jump_insn *jump_insn; + if ((jump_insn = dyn_cast <rtx_jump_insn *> (insn)) + && simplejump_p (jump_insn) && slots_filled != slots_to_fill) - delay_list - = fill_slots_from_thread (insn, const_true_rtx, - next_active_insn (JUMP_LABEL (insn)), - NULL, 1, 1, - own_thread_p (JUMP_LABEL (insn), - JUMP_LABEL (insn), 0), - slots_to_fill, &slots_filled, - delay_list); - - if (delay_list) + fill_slots_from_thread (jump_insn, const_true_rtx, + next_active_insn (JUMP_LABEL_AS_INSN (insn)), + NULL, 1, 1, own_thread_p (JUMP_LABEL (insn), + JUMP_LABEL (insn), 0), + slots_to_fill, &slots_filled, &delay_list); + + if (!delay_list.is_empty ()) unfilled_slots_base[i] = emit_delay_sequence (insn, delay_list, slots_filled); @@ -2411,134 +2247,62 @@ note_delay_statistics (slots_filled, 0); } - -#ifdef DELAY_SLOTS_FOR_EPILOGUE - /* See if the epilogue needs any delay slots. Try to fill them if so. - The only thing we can do is scan backwards from the end of the - function. If we did this in a previous pass, it is incorrect to do it - again. */ - if (crtl->epilogue_delay_list) - return; - - slots_to_fill = DELAY_SLOTS_FOR_EPILOGUE; - if (slots_to_fill == 0) - return; - - slots_filled = 0; - CLEAR_RESOURCE (&set); - - /* The frame pointer and stack pointer are needed at the beginning of - the epilogue, so instructions setting them can not be put in the - epilogue delay slot. However, everything else needed at function - end is safe, so we don't want to use end_of_function_needs here. */ - CLEAR_RESOURCE (&needed); - if (frame_pointer_needed) - { - SET_HARD_REG_BIT (needed.regs, FRAME_POINTER_REGNUM); -#if HARD_FRAME_POINTER_REGNUM != FRAME_POINTER_REGNUM - SET_HARD_REG_BIT (needed.regs, HARD_FRAME_POINTER_REGNUM); -#endif - if (! EXIT_IGNORE_STACK - || current_function_sp_is_unchanging) - SET_HARD_REG_BIT (needed.regs, STACK_POINTER_REGNUM); - } - else - SET_HARD_REG_BIT (needed.regs, STACK_POINTER_REGNUM); - -#ifdef EPILOGUE_USES - for (i = 0; i < FIRST_PSEUDO_REGISTER; i++) - { - if (EPILOGUE_USES (i)) - SET_HARD_REG_BIT (needed.regs, i); - } -#endif - - for (trial = get_last_insn (); ! stop_search_p (trial, 1); - trial = PREV_INSN (trial)) - { - if (NOTE_P (trial)) - continue; - pat = PATTERN (trial); - if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER) - continue; - - if (! insn_references_resource_p (trial, &set, true) - && ! insn_sets_resource_p (trial, &needed, true) - && ! insn_sets_resource_p (trial, &set, true) -#ifdef HAVE_cc0 - /* Don't want to mess with cc0 here. */ - && ! reg_mentioned_p (cc0_rtx, pat) -#endif - && ! can_throw_internal (trial)) - { - trial = try_split (pat, trial, 1); - if (ELIGIBLE_FOR_EPILOGUE_DELAY (trial, slots_filled)) - { - /* Here as well we are searching backward, so put the - insns we find on the head of the list. */ - - crtl->epilogue_delay_list - = gen_rtx_INSN_LIST (VOIDmode, trial, - crtl->epilogue_delay_list); - mark_end_of_function_resources (trial, true); - update_block (trial, trial); - delete_related_insns (trial); - - /* Clear deleted bit so final.c will output the insn. */ - INSN_DELETED_P (trial) = 0; - - if (slots_to_fill == ++slots_filled) - break; - continue; - } - } - - mark_set_resources (trial, &set, 0, MARK_SRC_DEST_CALL); - mark_referenced_resources (trial, &needed, true); - } - - note_delay_statistics (slots_filled, 0); -#endif } -/* Follow any unconditional jump at LABEL; +/* Follow any unconditional jump at LABEL, for the purpose of redirecting JUMP; return the ultimate label reached by any such chain of jumps. - Return null if the chain ultimately leads to a return instruction. + Return a suitable return rtx if the chain ultimately leads to a + return instruction. If LABEL is not followed by a jump, return LABEL. If the chain loops or we can't find end, return LABEL, - since that tells caller to avoid changing the insn. */ + since that tells caller to avoid changing the insn. + If the returned label is obtained by following a crossing jump, + set *CROSSING to true, otherwise set it to false. */ static rtx -follow_jumps (rtx label) +follow_jumps (rtx label, rtx_insn *jump, bool *crossing) { - rtx insn; - rtx next; - rtx value = label; + rtx_insn *insn; + rtx_insn *next; int depth; + *crossing = false; + if (ANY_RETURN_P (label)) + return label; + + rtx_insn *value = as_a <rtx_insn *> (label); + for (depth = 0; (depth < 10 && (insn = next_active_insn (value)) != 0 && JUMP_P (insn) - && ((JUMP_LABEL (insn) != 0 && any_uncondjump_p (insn) - && onlyjump_p (insn)) - || GET_CODE (PATTERN (insn)) == RETURN) + && JUMP_LABEL (insn) != NULL_RTX + && ((any_uncondjump_p (insn) && onlyjump_p (insn)) + || ANY_RETURN_P (PATTERN (insn))) && (next = NEXT_INSN (insn)) && BARRIER_P (next)); depth++) { - rtx tem; + rtx this_label_or_return = JUMP_LABEL (insn); /* If we have found a cycle, make the insn jump to itself. */ - if (JUMP_LABEL (insn) == label) + if (this_label_or_return == label) return label; - tem = next_active_insn (JUMP_LABEL (insn)); - if (tem && (GET_CODE (PATTERN (tem)) == ADDR_VEC - || GET_CODE (PATTERN (tem)) == ADDR_DIFF_VEC)) + /* Cannot follow returns and cannot look through tablejumps. */ + if (ANY_RETURN_P (this_label_or_return)) + return this_label_or_return; + + rtx_insn *this_label = as_a <rtx_insn *> (this_label_or_return); + if (NEXT_INSN (this_label) + && JUMP_TABLE_DATA_P (NEXT_INSN (this_label))) break; - value = JUMP_LABEL (insn); + if (!targetm.can_follow_jump (jump, insn)) + break; + if (!*crossing) + *crossing = CROSSING_JUMP_P (jump); + value = this_label; } if (depth == 10) return label; @@ -2569,29 +2333,31 @@ case, we can only take insns from the head of the thread for our delay slot. We then adjust the jump to point after the insns we have taken. */ -static rtx -fill_slots_from_thread (rtx insn, rtx condition, rtx thread, - rtx opposite_thread, int likely, int thread_if_true, - int own_thread, int slots_to_fill, - int *pslots_filled, rtx delay_list) +static void +fill_slots_from_thread (rtx_jump_insn *insn, rtx condition, + rtx thread_or_return, rtx opposite_thread, int likely, + int thread_if_true, int own_thread, int slots_to_fill, + int *pslots_filled, vec<rtx_insn *> *delay_list) { rtx new_thread; struct resources opposite_needed, set, needed; - rtx trial; + rtx_insn *trial; int lose = 0; int must_annul = 0; int flags; /* Validate our arguments. */ - gcc_assert(condition != const_true_rtx || thread_if_true); - gcc_assert(own_thread || thread_if_true); + gcc_assert (condition != const_true_rtx || thread_if_true); + gcc_assert (own_thread || thread_if_true); flags = get_jump_flags (insn, JUMP_LABEL (insn)); /* If our thread is the end of subroutine, we can't get any delay insns from that. */ - if (thread == 0) - return delay_list; + if (thread_or_return == NULL_RTX || ANY_RETURN_P (thread_or_return)) + return; + + rtx_insn *thread = as_a <rtx_insn *> (thread_or_return); /* If this is an unconditional branch, nothing is needed at the opposite thread. Otherwise, compute what is needed there. */ @@ -2642,18 +2408,16 @@ if (! insn_references_resource_p (trial, &set, true) && ! insn_sets_resource_p (trial, &set, true) && ! insn_sets_resource_p (trial, &needed, true) -#ifdef HAVE_cc0 - && ! (reg_mentioned_p (cc0_rtx, pat) - && (! own_thread || ! sets_cc0_p (pat))) -#endif + && (!HAVE_cc0 || (! (reg_mentioned_p (cc0_rtx, pat) + && (! own_thread || ! sets_cc0_p (pat))))) && ! can_throw_internal (trial)) { - rtx prior_insn; + rtx_insn *prior_insn; /* If TRIAL is redundant with some insn before INSN, we don't actually need to add it to the delay list; we can merely pretend we did. */ - if ((prior_insn = redundant_insn (trial, insn, delay_list))) + if ((prior_insn = redundant_insn (trial, insn, *delay_list))) { fix_reg_dead_note (prior_insn, insn); if (own_thread) @@ -2679,11 +2443,14 @@ /* There are two ways we can win: If TRIAL doesn't set anything needed at the opposite thread and can't trap, or if it can - go into an annulled delay slot. */ + go into an annulled delay slot. But we want neither to copy + nor to speculate frame-related insns. */ if (!must_annul - && (condition == const_true_rtx + && ((condition == const_true_rtx + && (own_thread || !RTX_FRAME_RELATED_P (trial))) || (! insn_sets_resource_p (trial, &opposite_needed, true) - && ! may_trap_or_fault_p (pat)))) + && ! may_trap_or_fault_p (pat) + && ! RTX_FRAME_RELATED_P (trial)))) { old_trial = trial; trial = try_split (pat, trial, 0); @@ -2695,14 +2462,9 @@ if (eligible_for_delay (insn, *pslots_filled, trial, flags)) goto winner; } - else if (0 -#ifdef ANNUL_IFTRUE_SLOTS - || ! thread_if_true -#endif -#ifdef ANNUL_IFFALSE_SLOTS - || thread_if_true -#endif - ) + else if (!RTX_FRAME_RELATED_P (trial) + && ((ANNUL_IFTRUE_SLOTS && ! thread_if_true) + || (ANNUL_IFFALSE_SLOTS && thread_if_true))) { old_trial = trial; trial = try_split (pat, trial, 0); @@ -2711,21 +2473,19 @@ if (thread == old_trial) thread = trial; pat = PATTERN (trial); - if ((must_annul || delay_list == NULL) && (thread_if_true - ? check_annul_list_true_false (0, delay_list) + if ((must_annul || delay_list->is_empty ()) && (thread_if_true + ? check_annul_list_true_false (0, *delay_list) && eligible_for_annul_false (insn, *pslots_filled, trial, flags) - : check_annul_list_true_false (1, delay_list) + : check_annul_list_true_false (1, *delay_list) && eligible_for_annul_true (insn, *pslots_filled, trial, flags))) { - rtx temp; + rtx_insn *temp; must_annul = 1; winner: -#ifdef HAVE_cc0 - if (reg_mentioned_p (cc0_rtx, pat)) + if (HAVE_cc0 && reg_mentioned_p (cc0_rtx, pat)) link_cc0_insns (trial); -#endif /* If we own this thread, delete the insn. If this is the destination of a branch, show that a basic block status @@ -2760,7 +2520,7 @@ gcc_assert (REG_NOTE_KIND (note) == REG_LABEL_OPERAND); } - if (JUMP_P (trial) && JUMP_LABEL (trial)) + if (jump_to_label_p (trial)) LABEL_NUSES (JUMP_LABEL (trial))++; delete_related_insns (trial); @@ -2779,17 +2539,17 @@ gcc_assert (REG_NOTE_KIND (note) == REG_LABEL_OPERAND); } - if (JUMP_P (trial) && JUMP_LABEL (trial)) + if (jump_to_label_p (trial)) LABEL_NUSES (JUMP_LABEL (trial))--; } else new_thread = next_active_insn (trial); - temp = own_thread ? trial : copy_rtx (trial); + temp = own_thread ? trial : copy_delay_slot_insn (trial); if (thread_if_true) INSN_FROM_TARGET_P (temp) = 1; - delay_list = add_to_delay_list (temp, delay_list); + add_to_delay_list (temp, delay_list); if (slots_to_fill == ++(*pslots_filled)) { @@ -2804,13 +2564,14 @@ &set, true) && (prior_insn = redundant_insn (new_thread, insn, - delay_list))) + *delay_list))) { /* We know we do not own the thread, so no need to call update_block and delete_insn. */ fix_reg_dead_note (prior_insn, insn); update_reg_unused_notes (prior_insn, new_thread); - new_thread = next_active_insn (new_thread); + new_thread + = next_active_insn (as_a<rtx_insn *> (new_thread)); } break; } @@ -2852,7 +2613,7 @@ && REG_P (SET_DEST (pat)) && !reg_overlap_mentioned_p (SET_DEST (pat), SET_SRC (pat))) { - rtx next = next_nonnote_insn (trial); + rtx_insn *next = next_nonnote_insn (trial); if (next && NONJUMP_INSN_P (next) && GET_CODE (PATTERN (next)) != USE @@ -2870,28 +2631,26 @@ && GET_CODE (PATTERN (trial)) == SEQUENCE && JUMP_P (XVECEXP (PATTERN (trial), 0, 0))) { + rtx_sequence *sequence = as_a <rtx_sequence *> (PATTERN (trial)); /* If this is the `true' thread, we will want to follow the jump, so we can only do this if we have taken everything up to here. */ if (thread_if_true && trial == new_thread) { - delay_list - = steal_delay_list_from_target (insn, condition, PATTERN (trial), - delay_list, &set, &needed, - &opposite_needed, slots_to_fill, - pslots_filled, &must_annul, - &new_thread); + steal_delay_list_from_target (insn, condition, sequence, + delay_list, &set, &needed, + &opposite_needed, slots_to_fill, + pslots_filled, &must_annul, + &new_thread); /* If we owned the thread and are told that it branched elsewhere, make sure we own the thread at the new location. */ if (own_thread && trial != new_thread) own_thread = own_thread_p (new_thread, new_thread, 0); } else if (! thread_if_true) - delay_list - = steal_delay_list_from_fallthrough (insn, condition, - PATTERN (trial), - delay_list, &set, &needed, - &opposite_needed, slots_to_fill, - pslots_filled, &must_annul); + steal_delay_list_from_fallthrough (insn, condition, sequence, + delay_list, &set, &needed, + &opposite_needed, slots_to_fill, + pslots_filled, &must_annul); } /* If we haven't found anything for this delay slot and it is very @@ -2900,8 +2659,10 @@ depend on the destination register. If so, try to place the opposite arithmetic insn after the jump insn and put the arithmetic insn in the delay slot. If we can't do this, return. */ - if (delay_list == 0 && likely && new_thread + if (delay_list->is_empty () && likely + && new_thread && !ANY_RETURN_P (new_thread) && NONJUMP_INSN_P (new_thread) + && !RTX_FRAME_RELATED_P (new_thread) && GET_CODE (PATTERN (new_thread)) != ASM_INPUT && asm_noperands (PATTERN (new_thread)) < 0) { @@ -2909,14 +2670,16 @@ rtx dest; rtx src; - trial = new_thread; + /* We know "new_thread" is an insn due to NONJUMP_INSN_P (new_thread) + above. */ + trial = as_a <rtx_insn *> (new_thread); pat = PATTERN (trial); if (!NONJUMP_INSN_P (trial) || GET_CODE (pat) != SET || ! eligible_for_delay (insn, 0, trial, flags) || can_throw_internal (trial)) - return 0; + return; dest = SET_DEST (pat), src = SET_SRC (pat); if ((GET_CODE (src) == PLUS || GET_CODE (src) == MINUS) @@ -2928,7 +2691,7 @@ { rtx other = XEXP (src, 1); rtx new_arith; - rtx ninsn; + rtx_insn *ninsn; /* If this is a constant adjustment, use the same code with the negated constant. Otherwise, reverse the sense of the @@ -2940,14 +2703,14 @@ new_arith = gen_rtx_fmt_ee (GET_CODE (src) == PLUS ? MINUS : PLUS, GET_MODE (src), dest, other); - ninsn = emit_insn_after (gen_rtx_SET (VOIDmode, dest, new_arith), - insn); + ninsn = emit_insn_after (gen_rtx_SET (dest, new_arith), insn); if (recog_memoized (ninsn) < 0 - || (extract_insn (ninsn), ! constrain_operands (1))) + || (extract_insn (ninsn), + !constrain_operands (1, get_preferred_alternatives (ninsn)))) { delete_related_insns (ninsn); - return 0; + return; } if (own_thread) @@ -2964,16 +2727,16 @@ else new_thread = next_active_insn (trial); - ninsn = own_thread ? trial : copy_rtx (trial); + ninsn = own_thread ? trial : copy_delay_slot_insn (trial); if (thread_if_true) INSN_FROM_TARGET_P (ninsn) = 1; - delay_list = add_to_delay_list (ninsn, NULL_RTX); + add_to_delay_list (ninsn, delay_list); (*pslots_filled)++; } } - if (delay_list && must_annul) + if (!delay_list->is_empty () && must_annul) INSN_ANNULLED_BRANCH_P (insn) = 1; /* If we are to branch into the middle of this thread, find an appropriate @@ -2982,29 +2745,32 @@ if (new_thread != thread) { rtx label; + bool crossing = false; gcc_assert (thread_if_true); - if (new_thread && JUMP_P (new_thread) - && (simplejump_p (new_thread) - || GET_CODE (PATTERN (new_thread)) == RETURN) + if (new_thread && simplejump_or_return_p (new_thread) && redirect_with_delay_list_safe_p (insn, JUMP_LABEL (new_thread), - delay_list)) - new_thread = follow_jumps (JUMP_LABEL (new_thread)); - - if (new_thread == 0) - label = find_end_label (); + *delay_list)) + new_thread = follow_jumps (JUMP_LABEL (new_thread), insn, + &crossing); + + if (ANY_RETURN_P (new_thread)) + label = find_end_label (new_thread); else if (LABEL_P (new_thread)) label = new_thread; else - label = get_label_before (new_thread); + label = get_label_before (as_a <rtx_insn *> (new_thread), + JUMP_LABEL (insn)); if (label) - reorg_redirect_jump (insn, label); + { + reorg_redirect_jump (insn, label); + if (crossing) + CROSSING_JUMP_P (insn) = 1; + } } - - return delay_list; } /* Make another attempt to find insns to place in delay slots. @@ -3020,27 +2786,29 @@ static void fill_eager_delay_slots (void) { - rtx insn; + rtx_insn *insn; int i; int num_unfilled_slots = unfilled_slots_next - unfilled_slots_base; for (i = 0; i < num_unfilled_slots; i++) { rtx condition; - rtx target_label, insn_at_target, fallthrough_insn; - rtx delay_list = 0; + rtx target_label, insn_at_target; + rtx_insn *fallthrough_insn; + auto_vec<rtx_insn *, 5> delay_list; + rtx_jump_insn *jump_insn; int own_target; int own_fallthrough; int prediction, slots_to_fill, slots_filled; insn = unfilled_slots_base[i]; if (insn == 0 - || INSN_DELETED_P (insn) - || !JUMP_P (insn) - || ! (condjump_p (insn) || condjump_in_parallel_p (insn))) + || insn->deleted () + || ! (jump_insn = dyn_cast <rtx_jump_insn *> (insn)) + || ! (condjump_p (jump_insn) || condjump_in_parallel_p (jump_insn))) continue; - slots_to_fill = num_delay_slots (insn); + slots_to_fill = num_delay_slots (jump_insn); /* Some machine description have defined instructions to have delay slots only in certain circumstances which may depend on nearby insns (which change due to reorg's actions). @@ -3056,8 +2824,8 @@ continue; slots_filled = 0; - target_label = JUMP_LABEL (insn); - condition = get_branch_condition (insn, target_label); + target_label = JUMP_LABEL (jump_insn); + condition = get_branch_condition (jump_insn, target_label); if (condition == 0) continue; @@ -3066,7 +2834,7 @@ them. Then see whether the branch is likely true. We don't need to do a lot of this for unconditional branches. */ - insn_at_target = next_active_insn (target_label); + insn_at_target = first_active_target_insn (target_label); own_target = own_thread_p (target_label, target_label, 0); if (condition == const_true_rtx) @@ -3077,9 +2845,9 @@ } else { - fallthrough_insn = next_active_insn (insn); - own_fallthrough = own_thread_p (NEXT_INSN (insn), NULL_RTX, 1); - prediction = mostly_true_jump (insn, condition); + fallthrough_insn = next_active_insn (jump_insn); + own_fallthrough = own_thread_p (NEXT_INSN (jump_insn), NULL_RTX, 1); + prediction = mostly_true_jump (jump_insn); } /* If this insn is expected to branch, first try to get insns from our @@ -3088,51 +2856,42 @@ if (prediction > 0) { - delay_list - = fill_slots_from_thread (insn, condition, insn_at_target, - fallthrough_insn, prediction == 2, 1, - own_target, - slots_to_fill, &slots_filled, delay_list); - - if (delay_list == 0 && own_fallthrough) + fill_slots_from_thread (jump_insn, condition, insn_at_target, + fallthrough_insn, prediction == 2, 1, + own_target, + slots_to_fill, &slots_filled, &delay_list); + + if (delay_list.is_empty () && own_fallthrough) { /* Even though we didn't find anything for delay slots, we might have found a redundant insn which we deleted from the thread that was filled. So we have to recompute the next insn at the target. */ - target_label = JUMP_LABEL (insn); - insn_at_target = next_active_insn (target_label); - - delay_list - = fill_slots_from_thread (insn, condition, fallthrough_insn, - insn_at_target, 0, 0, - own_fallthrough, - slots_to_fill, &slots_filled, - delay_list); + target_label = JUMP_LABEL (jump_insn); + insn_at_target = first_active_target_insn (target_label); + + fill_slots_from_thread (jump_insn, condition, fallthrough_insn, + insn_at_target, 0, 0, own_fallthrough, + slots_to_fill, &slots_filled, + &delay_list); } } else { if (own_fallthrough) - delay_list - = fill_slots_from_thread (insn, condition, fallthrough_insn, - insn_at_target, 0, 0, - own_fallthrough, - slots_to_fill, &slots_filled, - delay_list); - - if (delay_list == 0) - delay_list - = fill_slots_from_thread (insn, condition, insn_at_target, - next_active_insn (insn), 0, 1, - own_target, - slots_to_fill, &slots_filled, - delay_list); + fill_slots_from_thread (jump_insn, condition, fallthrough_insn, + insn_at_target, 0, 0, own_fallthrough, + slots_to_fill, &slots_filled, &delay_list); + + if (delay_list.is_empty ()) + fill_slots_from_thread (jump_insn, condition, insn_at_target, + next_active_insn (insn), 0, 1, own_target, + slots_to_fill, &slots_filled, &delay_list); } - if (delay_list) + if (!delay_list.is_empty ()) unfilled_slots_base[i] - = emit_delay_sequence (insn, delay_list, slots_filled); + = emit_delay_sequence (jump_insn, delay_list, slots_filled); if (slots_to_fill == slots_filled) unfilled_slots_base[i] = 0; @@ -3141,16 +2900,16 @@ } } -static void delete_computation (rtx insn); +static void delete_computation (rtx_insn *insn); /* Recursively delete prior insns that compute the value (used only by INSN which the caller is deleting) stored in the register mentioned by NOTE which is a REG_DEAD note associated with INSN. */ static void -delete_prior_computation (rtx note, rtx insn) +delete_prior_computation (rtx note, rtx_insn *insn) { - rtx our_prev; + rtx_insn *our_prev; rtx reg = XEXP (note, 0); for (our_prev = prev_nonnote_insn (insn); @@ -3262,14 +3021,13 @@ delete the insn that set it. */ static void -delete_computation (rtx insn) +delete_computation (rtx_insn *insn) { rtx note, next; -#ifdef HAVE_cc0 - if (reg_referenced_p (cc0_rtx, PATTERN (insn))) + if (HAVE_cc0 && reg_referenced_p (cc0_rtx, PATTERN (insn))) { - rtx prev = prev_nonnote_insn (insn); + rtx_insn *prev = prev_nonnote_insn (insn); /* We assume that at this stage CC's are always set explicitly and always immediately before the jump that @@ -3287,7 +3045,6 @@ add_reg_note (prev, REG_UNUSED, cc0_rtx); } } -#endif for (note = REG_NOTES (insn); note; note = next) { @@ -3309,7 +3066,7 @@ if that's what the previous thing was. */ static void -delete_jump (rtx insn) +delete_jump (rtx_insn *insn) { rtx set = single_set (insn); @@ -3317,54 +3074,95 @@ delete_computation (insn); } +static rtx_insn * +label_before_next_insn (rtx_insn *x, rtx scan_limit) +{ + rtx_insn *insn = next_active_insn (x); + while (insn) + { + insn = PREV_INSN (insn); + if (insn == scan_limit || insn == NULL_RTX) + return NULL; + if (LABEL_P (insn)) + break; + } + return insn; +} + +/* Return TRUE if there is a NOTE_INSN_SWITCH_TEXT_SECTIONS note in between + BEG and END. */ + +static bool +switch_text_sections_between_p (const rtx_insn *beg, const rtx_insn *end) +{ + const rtx_insn *p; + for (p = beg; p != end; p = NEXT_INSN (p)) + if (NOTE_P (p) && NOTE_KIND (p) == NOTE_INSN_SWITCH_TEXT_SECTIONS) + return true; + return false; +} + /* Once we have tried two ways to fill a delay slot, make a pass over the code to try to improve the results and to do such things as more jump threading. */ static void -relax_delay_slots (rtx first) +relax_delay_slots (rtx_insn *first) { - rtx insn, next, pat; - rtx trial, delay_insn, target_label; + rtx_insn *insn, *next; + rtx_sequence *pat; + rtx_insn *delay_insn; + rtx target_label; /* Look at every JUMP_INSN and see if we can improve it. */ for (insn = first; insn; insn = next) { - rtx other; + rtx_insn *other; + bool crossing; next = next_active_insn (insn); /* If this is a jump insn, see if it now jumps to a jump, jumps to the next insn, or jumps to a label that is not the last of a group of consecutive labels. */ - if (JUMP_P (insn) + if (is_a <rtx_jump_insn *> (insn) && (condjump_p (insn) || condjump_in_parallel_p (insn)) - && (target_label = JUMP_LABEL (insn)) != 0) + && !ANY_RETURN_P (target_label = JUMP_LABEL (insn))) { - target_label = skip_consecutive_labels (follow_jumps (target_label)); - if (target_label == 0) - target_label = find_end_label (); - - if (target_label && next_active_insn (target_label) == next - && ! condjump_in_parallel_p (insn)) + rtx_jump_insn *jump_insn = as_a <rtx_jump_insn *> (insn); + target_label + = skip_consecutive_labels (follow_jumps (target_label, jump_insn, + &crossing)); + if (ANY_RETURN_P (target_label)) + target_label = find_end_label (target_label); + + if (target_label + && next_active_insn (as_a<rtx_insn *> (target_label)) == next + && ! condjump_in_parallel_p (jump_insn) + && ! (next && switch_text_sections_between_p (jump_insn, next))) { - delete_jump (insn); + delete_jump (jump_insn); continue; } - if (target_label && target_label != JUMP_LABEL (insn)) - reorg_redirect_jump (insn, target_label); + if (target_label && target_label != JUMP_LABEL (jump_insn)) + { + reorg_redirect_jump (jump_insn, target_label); + if (crossing) + CROSSING_JUMP_P (jump_insn) = 1; + } /* See if this jump conditionally branches around an unconditional jump. If so, invert this jump and point it to the target of the - second jump. */ - if (next && JUMP_P (next) - && any_condjump_p (insn) - && (simplejump_p (next) || GET_CODE (PATTERN (next)) == RETURN) + second jump. Check if it's possible on the target. */ + if (next && simplejump_or_return_p (next) + && any_condjump_p (jump_insn) && target_label - && next_active_insn (target_label) == next_active_insn (next) - && no_labels_between_p (insn, next)) + && (next_active_insn (as_a<rtx_insn *> (target_label)) + == next_active_insn (next)) + && no_labels_between_p (jump_insn, next) + && targetm.can_follow_jump (jump_insn, next)) { rtx label = JUMP_LABEL (next); @@ -3376,16 +3174,16 @@ invert_jump fails. */ ++LABEL_NUSES (target_label); - if (label) + if (!ANY_RETURN_P (label)) ++LABEL_NUSES (label); - if (invert_jump (insn, label, 1)) + if (invert_jump (jump_insn, label, 1)) { delete_related_insns (next); - next = insn; + next = jump_insn; } - if (label) + if (!ANY_RETURN_P (label)) --LABEL_NUSES (label); if (--LABEL_NUSES (target_label) == 0) @@ -3403,36 +3201,33 @@ Don't do this if we expect the conditional branch to be true, because we would then be making the more common case longer. */ - if (JUMP_P (insn) - && (simplejump_p (insn) || GET_CODE (PATTERN (insn)) == RETURN) + if (simplejump_or_return_p (insn) && (other = prev_active_insn (insn)) != 0 && any_condjump_p (other) && no_labels_between_p (other, insn) - && 0 > mostly_true_jump (other, - get_branch_condition (other, - JUMP_LABEL (other)))) + && 0 > mostly_true_jump (other)) { rtx other_target = JUMP_LABEL (other); target_label = JUMP_LABEL (insn); - if (invert_jump (other, target_label, 0)) - reorg_redirect_jump (insn, other_target); + if (invert_jump (as_a <rtx_jump_insn *> (other), target_label, 0)) + reorg_redirect_jump (as_a <rtx_jump_insn *> (insn), other_target); } - /* Now look only at cases where we have filled a delay slot. */ - if (!NONJUMP_INSN_P (insn) - || GET_CODE (PATTERN (insn)) != SEQUENCE) + /* Now look only at cases where we have a filled delay slot. */ + if (!NONJUMP_INSN_P (insn) || GET_CODE (PATTERN (insn)) != SEQUENCE) continue; - pat = PATTERN (insn); - delay_insn = XVECEXP (pat, 0, 0); + pat = as_a <rtx_sequence *> (PATTERN (insn)); + delay_insn = pat->insn (0); /* See if the first insn in the delay slot is redundant with some previous insn. Remove it from the delay slot if so; then set up to reprocess this insn. */ - if (redundant_insn (XVECEXP (pat, 0, 1), delay_insn, 0)) + if (redundant_insn (pat->insn (1), delay_insn, vNULL)) { - delete_from_delay_slot (XVECEXP (pat, 0, 1)); + update_block (pat->insn (1), insn); + delete_from_delay_slot (pat->insn (1)); next = prev_active_insn (next); continue; } @@ -3445,12 +3240,12 @@ Only do so if optimizing for size since this results in slower, but smaller code. */ if (optimize_function_for_size_p (cfun) - && GET_CODE (PATTERN (delay_insn)) == RETURN + && ANY_RETURN_P (PATTERN (delay_insn)) && next && JUMP_P (next) - && GET_CODE (PATTERN (next)) == RETURN) + && PATTERN (next) == PATTERN (delay_insn)) { - rtx after; + rtx_insn *after; int i; /* Delete the RETURN and just execute the delay list insns. @@ -3469,108 +3264,116 @@ for (i = 0; i < XVECLEN (pat, 0); i++) INSN_FROM_TARGET_P (XVECEXP (pat, 0, i)) = 0; - trial = PREV_INSN (insn); + rtx_insn *prev = PREV_INSN (insn); delete_related_insns (insn); gcc_assert (GET_CODE (pat) == SEQUENCE); - add_insn_after (delay_insn, trial, NULL); + add_insn_after (delay_insn, prev, NULL); after = delay_insn; - for (i = 1; i < XVECLEN (pat, 0); i++) - after = emit_copy_of_insn_after (XVECEXP (pat, 0, i), after); + for (i = 1; i < pat->len (); i++) + after = emit_copy_of_insn_after (pat->insn (i), after); delete_scheduled_jump (delay_insn); continue; } /* Now look only at the cases where we have a filled JUMP_INSN. */ - if (!JUMP_P (XVECEXP (PATTERN (insn), 0, 0)) - || ! (condjump_p (XVECEXP (PATTERN (insn), 0, 0)) - || condjump_in_parallel_p (XVECEXP (PATTERN (insn), 0, 0)))) + rtx_jump_insn *delay_jump_insn = + dyn_cast <rtx_jump_insn *> (delay_insn); + if (! delay_jump_insn || !(condjump_p (delay_jump_insn) + || condjump_in_parallel_p (delay_jump_insn))) + continue; + + target_label = JUMP_LABEL (delay_jump_insn); + if (target_label && ANY_RETURN_P (target_label)) continue; - target_label = JUMP_LABEL (delay_insn); - - if (target_label) + /* If this jump goes to another unconditional jump, thread it, but + don't convert a jump into a RETURN here. */ + rtx trial = skip_consecutive_labels (follow_jumps (target_label, + delay_jump_insn, + &crossing)); + if (ANY_RETURN_P (trial)) + trial = find_end_label (trial); + + if (trial && trial != target_label + && redirect_with_delay_slots_safe_p (delay_jump_insn, trial, insn)) { - /* If this jump goes to another unconditional jump, thread it, but - don't convert a jump into a RETURN here. */ - trial = skip_consecutive_labels (follow_jumps (target_label)); - if (trial == 0) - trial = find_end_label (); - - if (trial && trial != target_label - && redirect_with_delay_slots_safe_p (delay_insn, trial, insn)) - { - reorg_redirect_jump (delay_insn, trial); - target_label = trial; - } - - /* If the first insn at TARGET_LABEL is redundant with a previous - insn, redirect the jump to the following insn and process again. - We use next_real_insn instead of next_active_insn so we - don't skip USE-markers, or we'll end up with incorrect - liveness info. */ - trial = next_real_insn (target_label); - if (trial && GET_CODE (PATTERN (trial)) != SEQUENCE - && redundant_insn (trial, insn, 0) - && ! can_throw_internal (trial)) + reorg_redirect_jump (delay_jump_insn, trial); + target_label = trial; + if (crossing) + CROSSING_JUMP_P (delay_jump_insn) = 1; + } + + /* If the first insn at TARGET_LABEL is redundant with a previous + insn, redirect the jump to the following insn and process again. + We use next_real_insn instead of next_active_insn so we + don't skip USE-markers, or we'll end up with incorrect + liveness info. */ + trial = next_real_insn (target_label); + if (trial && GET_CODE (PATTERN (trial)) != SEQUENCE + && redundant_insn (trial, insn, vNULL) + && ! can_throw_internal (trial)) + { + /* Figure out where to emit the special USE insn so we don't + later incorrectly compute register live/death info. */ + rtx_insn *tmp = next_active_insn (as_a<rtx_insn *> (trial)); + if (tmp == 0) + tmp = find_end_label (simple_return_rtx); + + if (tmp) { - /* Figure out where to emit the special USE insn so we don't - later incorrectly compute register live/death info. */ - rtx tmp = next_active_insn (trial); - if (tmp == 0) - tmp = find_end_label (); - - if (tmp) - { - /* Insert the special USE insn and update dataflow info. */ - update_block (trial, tmp); - - /* Now emit a label before the special USE insn, and - redirect our jump to the new label. */ - target_label = get_label_before (PREV_INSN (tmp)); - reorg_redirect_jump (delay_insn, target_label); - next = insn; - continue; - } - } - - /* Similarly, if it is an unconditional jump with one insn in its - delay list and that insn is redundant, thread the jump. */ - if (trial && GET_CODE (PATTERN (trial)) == SEQUENCE - && XVECLEN (PATTERN (trial), 0) == 2 - && JUMP_P (XVECEXP (PATTERN (trial), 0, 0)) - && (simplejump_p (XVECEXP (PATTERN (trial), 0, 0)) - || GET_CODE (PATTERN (XVECEXP (PATTERN (trial), 0, 0))) == RETURN) - && redundant_insn (XVECEXP (PATTERN (trial), 0, 1), insn, 0)) - { - target_label = JUMP_LABEL (XVECEXP (PATTERN (trial), 0, 0)); - if (target_label == 0) - target_label = find_end_label (); - - if (target_label - && redirect_with_delay_slots_safe_p (delay_insn, target_label, - insn)) - { - reorg_redirect_jump (delay_insn, target_label); - next = insn; - continue; - } + /* Insert the special USE insn and update dataflow info. + We know "trial" is an insn here as it is the output of + next_real_insn () above. */ + update_block (as_a <rtx_insn *> (trial), tmp); + + /* Now emit a label before the special USE insn, and + redirect our jump to the new label. */ + target_label = get_label_before (PREV_INSN (tmp), target_label); + reorg_redirect_jump (delay_jump_insn, target_label); + next = insn; + continue; } } - if (! INSN_ANNULLED_BRANCH_P (delay_insn) - && prev_active_insn (target_label) == insn - && ! condjump_in_parallel_p (delay_insn) -#ifdef HAVE_cc0 + /* Similarly, if it is an unconditional jump with one insn in its + delay list and that insn is redundant, thread the jump. */ + rtx_sequence *trial_seq = + trial ? dyn_cast <rtx_sequence *> (PATTERN (trial)) : NULL; + if (trial_seq + && trial_seq->len () == 2 + && JUMP_P (trial_seq->insn (0)) + && simplejump_or_return_p (trial_seq->insn (0)) + && redundant_insn (trial_seq->insn (1), insn, vNULL)) + { + rtx temp_label = JUMP_LABEL (trial_seq->insn (0)); + if (ANY_RETURN_P (temp_label)) + temp_label = find_end_label (temp_label); + + if (temp_label + && redirect_with_delay_slots_safe_p (delay_jump_insn, + temp_label, insn)) + { + update_block (trial_seq->insn (1), insn); + reorg_redirect_jump (delay_jump_insn, temp_label); + next = insn; + continue; + } + } + + /* See if we have a simple (conditional) jump that is useless. */ + if (! INSN_ANNULLED_BRANCH_P (delay_jump_insn) + && ! condjump_in_parallel_p (delay_jump_insn) + && prev_active_insn (as_a<rtx_insn *> (target_label)) == insn + && ! BARRIER_P (prev_nonnote_insn (as_a<rtx_insn *> (target_label))) /* If the last insn in the delay slot sets CC0 for some insn, various code assumes that it is in a delay slot. We could put it back where it belonged and delete the register notes, but it doesn't seem worthwhile in this uncommon case. */ - && ! find_reg_note (XVECEXP (pat, 0, XVECLEN (pat, 0) - 1), - REG_CC_USER, NULL_RTX) -#endif - ) + && (!HAVE_cc0 + || ! find_reg_note (XVECEXP (pat, 0, XVECLEN (pat, 0) - 1), + REG_CC_USER, NULL_RTX))) { - rtx after; + rtx_insn *after; int i; /* All this insn does is execute its delay list and jump to the @@ -3591,14 +3394,14 @@ for (i = 0; i < XVECLEN (pat, 0); i++) INSN_FROM_TARGET_P (XVECEXP (pat, 0, i)) = 0; - trial = PREV_INSN (insn); + rtx_insn *prev = PREV_INSN (insn); delete_related_insns (insn); gcc_assert (GET_CODE (pat) == SEQUENCE); - add_insn_after (delay_insn, trial, NULL); - after = delay_insn; - for (i = 1; i < XVECLEN (pat, 0); i++) - after = emit_copy_of_insn_after (XVECEXP (pat, 0, i), after); - delete_scheduled_jump (delay_insn); + add_insn_after (delay_jump_insn, prev, NULL); + after = delay_jump_insn; + for (i = 1; i < pat->len (); i++) + after = emit_copy_of_insn_after (pat->insn (i), after); + delete_scheduled_jump (delay_jump_insn); continue; } @@ -3606,10 +3409,10 @@ identical to the one in its delay slot. In this case, we can just delete the branch and the insn in its delay slot. */ if (next && NONJUMP_INSN_P (next) - && prev_label (next_active_insn (next)) == target_label + && label_before_next_insn (next, insn) == target_label && simplejump_p (insn) && XVECLEN (pat, 0) == 2 - && rtx_equal_p (PATTERN (next), PATTERN (XVECEXP (pat, 0, 1)))) + && rtx_equal_p (PATTERN (next), PATTERN (pat->insn (1)))) { delete_related_insns (insn); continue; @@ -3620,23 +3423,24 @@ this jump and point it to the target of the second jump. We cannot do this for annulled jumps, though. Again, don't convert a jump to a RETURN here. */ - if (! INSN_ANNULLED_BRANCH_P (delay_insn) - && any_condjump_p (delay_insn) - && next && JUMP_P (next) - && (simplejump_p (next) || GET_CODE (PATTERN (next)) == RETURN) - && next_active_insn (target_label) == next_active_insn (next) + if (! INSN_ANNULLED_BRANCH_P (delay_jump_insn) + && any_condjump_p (delay_jump_insn) + && next && simplejump_or_return_p (next) + && (next_active_insn (as_a<rtx_insn *> (target_label)) + == next_active_insn (next)) && no_labels_between_p (insn, next)) { rtx label = JUMP_LABEL (next); - rtx old_label = JUMP_LABEL (delay_insn); - - if (label == 0) - label = find_end_label (); + rtx old_label = JUMP_LABEL (delay_jump_insn); + + if (ANY_RETURN_P (label)) + label = find_end_label (label); /* find_end_label can generate a new label. Check this first. */ if (label && no_labels_between_p (insn, next) - && redirect_with_delay_slots_safe_p (delay_insn, label, insn)) + && redirect_with_delay_slots_safe_p (delay_jump_insn, + label, insn)) { /* Be careful how we do this to avoid deleting code or labels that are momentarily dead. See similar optimization in @@ -3644,7 +3448,7 @@ if (old_label) ++LABEL_NUSES (old_label); - if (invert_jump (delay_insn, label, 1)) + if (invert_jump (delay_jump_insn, label, 1)) { int i; @@ -3669,12 +3473,13 @@ /* If we own the thread opposite the way this insn branches, see if we can merge its delay slots with following insns. */ - if (INSN_FROM_TARGET_P (XVECEXP (pat, 0, 1)) + if (INSN_FROM_TARGET_P (pat->insn (1)) && own_thread_p (NEXT_INSN (insn), 0, 1)) try_merge_delay_insns (insn, next); - else if (! INSN_FROM_TARGET_P (XVECEXP (pat, 0, 1)) + else if (! INSN_FROM_TARGET_P (pat->insn (1)) && own_thread_p (target_label, target_label, 0)) - try_merge_delay_insns (insn, next_active_insn (target_label)); + try_merge_delay_insns (insn, + next_active_insn (as_a<rtx_insn *> (target_label))); /* If we get here, we haven't deleted INSN. But we may have deleted NEXT, so recompute it. */ @@ -3682,43 +3487,40 @@ } } -#ifdef HAVE_return /* Look for filled jumps to the end of function label. We can try to convert them into RETURN insns if the insns in the delay slot are valid for the RETURN as well. */ static void -make_return_insns (rtx first) +make_return_insns (rtx_insn *first) { - rtx insn, jump_insn, pat; - rtx real_return_label = end_of_function_label; + rtx_insn *insn; + rtx_jump_insn *jump_insn; + rtx real_return_label = function_return_label; + rtx real_simple_return_label = function_simple_return_label; int slots, i; -#ifdef DELAY_SLOTS_FOR_EPILOGUE - /* If a previous pass filled delay slots in the epilogue, things get a - bit more complicated, as those filler insns would generally (without - data flow analysis) have to be executed after any existing branch - delay slot filler insns. It is also unknown whether such a - transformation would actually be profitable. Note that the existing - code only cares for branches with (some) filled delay slots. */ - if (crtl->epilogue_delay_list != NULL) - return; -#endif - /* See if there is a RETURN insn in the function other than the one we made for END_OF_FUNCTION_LABEL. If so, set up anything we can't change into a RETURN to jump to it. */ for (insn = first; insn; insn = NEXT_INSN (insn)) - if (JUMP_P (insn) && GET_CODE (PATTERN (insn)) == RETURN) + if (JUMP_P (insn) && ANY_RETURN_P (PATTERN (insn))) { - real_return_label = get_label_before (insn); + rtx t = get_label_before (insn, NULL_RTX); + if (PATTERN (insn) == ret_rtx) + real_return_label = t; + else + real_simple_return_label = t; break; } /* Show an extra usage of REAL_RETURN_LABEL so it won't go away if it was equal to END_OF_FUNCTION_LABEL. */ - LABEL_NUSES (real_return_label)++; + if (real_return_label) + LABEL_NUSES (real_return_label)++; + if (real_simple_return_label) + LABEL_NUSES (real_simple_return_label)++; /* Clear the list of insns to fill so we can use it. */ obstack_free (&unfilled_slots_obstack, unfilled_firstobj); @@ -3726,28 +3528,44 @@ for (insn = first; insn; insn = NEXT_INSN (insn)) { int flags; + rtx kind, real_label; /* Only look at filled JUMP_INSNs that go to the end of function label. */ - if (!NONJUMP_INSN_P (insn) - || GET_CODE (PATTERN (insn)) != SEQUENCE - || !JUMP_P (XVECEXP (PATTERN (insn), 0, 0)) - || JUMP_LABEL (XVECEXP (PATTERN (insn), 0, 0)) != end_of_function_label) + if (!NONJUMP_INSN_P (insn)) + continue; + + if (GET_CODE (PATTERN (insn)) != SEQUENCE) + continue; + + rtx_sequence *pat = as_a <rtx_sequence *> (PATTERN (insn)); + + if (!jump_to_label_p (pat->insn (0))) continue; - pat = PATTERN (insn); - jump_insn = XVECEXP (pat, 0, 0); + if (JUMP_LABEL (pat->insn (0)) == function_return_label) + { + kind = ret_rtx; + real_label = real_return_label; + } + else if (JUMP_LABEL (pat->insn (0)) == function_simple_return_label) + { + kind = simple_return_rtx; + real_label = real_simple_return_label; + } + else + continue; + + jump_insn = as_a <rtx_jump_insn *> (pat->insn (0)); /* If we can't make the jump into a RETURN, try to redirect it to the best RETURN and go on to the next insn. */ - if (! reorg_redirect_jump (jump_insn, NULL_RTX)) + if (!reorg_redirect_jump (jump_insn, kind)) { /* Make sure redirecting the jump will not invalidate the delay slot insns. */ - if (redirect_with_delay_slots_safe_p (jump_insn, - real_return_label, - insn)) - reorg_redirect_jump (jump_insn, real_return_label); + if (redirect_with_delay_slots_safe_p (jump_insn, real_label, insn)) + reorg_redirect_jump (jump_insn, real_label); continue; } @@ -3761,20 +3579,20 @@ { for (i = 1; i < XVECLEN (pat, 0); i++) if (! ( -#ifdef ANNUL_IFFALSE_SLOTS +#if ANNUL_IFFALSE_SLOTS (INSN_ANNULLED_BRANCH_P (jump_insn) - && INSN_FROM_TARGET_P (XVECEXP (pat, 0, i))) + && INSN_FROM_TARGET_P (pat->insn (i))) ? eligible_for_annul_false (jump_insn, i - 1, - XVECEXP (pat, 0, i), flags) : + pat->insn (i), flags) : #endif -#ifdef ANNUL_IFTRUE_SLOTS +#if ANNUL_IFTRUE_SLOTS (INSN_ANNULLED_BRANCH_P (jump_insn) - && ! INSN_FROM_TARGET_P (XVECEXP (pat, 0, i))) + && ! INSN_FROM_TARGET_P (pat->insn (i))) ? eligible_for_annul_true (jump_insn, i - 1, - XVECEXP (pat, 0, i), flags) : + pat->insn (i), flags) : #endif eligible_for_delay (jump_insn, i - 1, - XVECEXP (pat, 0, i), flags))) + pat->insn (i), flags))) break; } else @@ -3787,9 +3605,9 @@ RETURN, delete the SEQUENCE and output the individual insns, followed by the RETURN. Then set things up so we try to find insns for its delay slots, if it needs some. */ - if (GET_CODE (PATTERN (jump_insn)) == RETURN) + if (ANY_RETURN_P (PATTERN (jump_insn))) { - rtx prev = PREV_INSN (insn); + rtx_insn *prev = PREV_INSN (insn); delete_related_insns (insn); for (i = 1; i < XVECLEN (pat, 0); i++) @@ -3804,30 +3622,33 @@ else /* It is probably more efficient to keep this with its current delay slot as a branch to a RETURN. */ - reorg_redirect_jump (jump_insn, real_return_label); + reorg_redirect_jump (jump_insn, real_label); } /* Now delete REAL_RETURN_LABEL if we never used it. Then try to fill any new delay slots we have created. */ - if (--LABEL_NUSES (real_return_label) == 0) + if (real_return_label != NULL_RTX && --LABEL_NUSES (real_return_label) == 0) delete_related_insns (real_return_label); + if (real_simple_return_label != NULL_RTX + && --LABEL_NUSES (real_simple_return_label) == 0) + delete_related_insns (real_simple_return_label); fill_simple_delay_slots (1); fill_simple_delay_slots (0); } -#endif /* Try to find insns to place in delay slots. */ -void -dbr_schedule (rtx first) +static void +dbr_schedule (rtx_insn *first) { - rtx insn, next, epilogue_insn = 0; + rtx_insn *insn, *next, *epilogue_insn = 0; int i; + bool need_return_insns; /* If the current function has no insns other than the prologue and epilogue, then do not try to fill any delay slots. */ - if (n_basic_blocks == NUM_FIXED_BLOCKS) + if (n_basic_blocks_for_fn (cfun) == NUM_FIXED_BLOCKS) return; /* Find the highest INSN_UID and allocate and initialize our map from @@ -3856,29 +3677,30 @@ { rtx target; - INSN_ANNULLED_BRANCH_P (insn) = 0; - INSN_FROM_TARGET_P (insn) = 0; - /* Skip vector tables. We can't get attributes for them. */ if (JUMP_TABLE_DATA_P (insn)) continue; + if (JUMP_P (insn)) + INSN_ANNULLED_BRANCH_P (insn) = 0; + INSN_FROM_TARGET_P (insn) = 0; + if (num_delay_slots (insn) > 0) obstack_ptr_grow (&unfilled_slots_obstack, insn); /* Ensure all jumps go to the last of a set of consecutive labels. */ if (JUMP_P (insn) && (condjump_p (insn) || condjump_in_parallel_p (insn)) - && JUMP_LABEL (insn) != 0 + && !ANY_RETURN_P (JUMP_LABEL (insn)) && ((target = skip_consecutive_labels (JUMP_LABEL (insn))) != JUMP_LABEL (insn))) - redirect_jump (insn, target, 1); + redirect_jump (as_a <rtx_jump_insn *> (insn), target, 1); } init_resource_info (epilogue_insn); /* Show we haven't computed an end-of-function label yet. */ - end_of_function_label = 0; + function_return_label = function_simple_return_label = NULL; /* Initialize the statistics for this function. */ memset (num_insns_needing_delays, 0, sizeof num_insns_needing_delays); @@ -3893,20 +3715,26 @@ { fill_simple_delay_slots (1); fill_simple_delay_slots (0); - fill_eager_delay_slots (); + if (!targetm.no_speculation_in_delay_slots_p ()) + fill_eager_delay_slots (); relax_delay_slots (first); } /* If we made an end of function label, indicate that it is now safe to delete it by undoing our prior adjustment to LABEL_NUSES. If it is now unused, delete it. */ - if (end_of_function_label && --LABEL_NUSES (end_of_function_label) == 0) - delete_related_insns (end_of_function_label); - -#ifdef HAVE_return - if (HAVE_return && end_of_function_label != 0) + if (function_return_label && --LABEL_NUSES (function_return_label) == 0) + delete_related_insns (function_return_label); + if (function_simple_return_label + && --LABEL_NUSES (function_simple_return_label) == 0) + delete_related_insns (function_simple_return_label); + + need_return_insns = false; + need_return_insns |= targetm.have_return () && function_return_label != 0; + need_return_insns |= (targetm.have_simple_return () + && function_simple_return_label != 0); + if (need_return_insns) make_return_insns (first); -#endif /* Delete any USE insns made by update_block; subsequent passes don't need them or know how to deal with them. */ @@ -3959,17 +3787,19 @@ memset (total_annul_slots, 0, sizeof total_annul_slots); for (insn = first; insn; insn = NEXT_INSN (insn)) { - if (! INSN_DELETED_P (insn) + if (! insn->deleted () && NONJUMP_INSN_P (insn) && GET_CODE (PATTERN (insn)) != USE && GET_CODE (PATTERN (insn)) != CLOBBER) { if (GET_CODE (PATTERN (insn)) == SEQUENCE) { + rtx control; j = XVECLEN (PATTERN (insn), 0) - 1; if (j > MAX_DELAY_HISTOGRAM) j = MAX_DELAY_HISTOGRAM; - if (INSN_ANNULLED_BRANCH_P (XVECEXP (PATTERN (insn), 0, 0))) + control = XVECEXP (PATTERN (insn), 0, 0); + if (JUMP_P (control) && INSN_ANNULLED_BRANCH_P (control)) total_annul_slots[j]++; else total_delay_slots[j]++; @@ -3991,137 +3821,139 @@ } } fprintf (dump_file, "\n"); -#if defined (ANNUL_IFTRUE_SLOTS) || defined (ANNUL_IFFALSE_SLOTS) - fprintf (dump_file, ";; Reorg annuls: "); - need_comma = 0; - for (j = 0; j < MAX_DELAY_HISTOGRAM + 1; j++) + + if (ANNUL_IFTRUE_SLOTS || ANNUL_IFFALSE_SLOTS) { - if (total_annul_slots[j]) + fprintf (dump_file, ";; Reorg annuls: "); + need_comma = 0; + for (j = 0; j < MAX_DELAY_HISTOGRAM + 1; j++) { - if (need_comma) - fprintf (dump_file, ", "); - need_comma = 1; - fprintf (dump_file, "%d got %d delays", total_annul_slots[j], j); + if (total_annul_slots[j]) + { + if (need_comma) + fprintf (dump_file, ", "); + need_comma = 1; + fprintf (dump_file, "%d got %d delays", total_annul_slots[j], j); + } } + fprintf (dump_file, "\n"); } - fprintf (dump_file, "\n"); -#endif + fprintf (dump_file, "\n"); } - /* For all JUMP insns, fill in branch prediction notes, so that during - assembler output a target can set branch prediction bits in the code. - We have to do this now, as up until this point the destinations of - JUMPS can be moved around and changed, but past right here that cannot - happen. */ - for (insn = first; insn; insn = NEXT_INSN (insn)) + if (!sibling_labels.is_empty ()) { - int pred_flags; - - if (NONJUMP_INSN_P (insn)) - { - rtx pat = PATTERN (insn); - - if (GET_CODE (pat) == SEQUENCE) - insn = XVECEXP (pat, 0, 0); - } - if (!JUMP_P (insn)) - continue; - - pred_flags = get_jump_flags (insn, JUMP_LABEL (insn)); - add_reg_note (insn, REG_BR_PRED, GEN_INT (pred_flags)); + update_alignments (sibling_labels); + sibling_labels.release (); } + free_resource_info (); free (uid_to_ruid); -#ifdef DELAY_SLOTS_FOR_EPILOGUE - /* SPARC assembler, for instance, emit warning when debug info is output - into the delay slot. */ - { - rtx link; - - for (link = crtl->epilogue_delay_list; - link; - link = XEXP (link, 1)) - INSN_LOCATOR (XEXP (link, 0)) = 0; - } - -#endif crtl->dbr_scheduled_p = true; } -#endif /* DELAY_SLOTS */ -static bool -gate_handle_delay_slots (void) -{ -#ifdef DELAY_SLOTS - /* At -O0 dataflow info isn't updated after RA. */ - return optimize > 0 && flag_delayed_branch && !crtl->dbr_scheduled_p; -#else - return 0; -#endif -} - /* Run delay slot optimization. */ static unsigned int rest_of_handle_delay_slots (void) { -#ifdef DELAY_SLOTS - dbr_schedule (get_insns ()); -#endif + if (DELAY_SLOTS) + dbr_schedule (get_insns ()); + return 0; } -struct rtl_opt_pass pass_delay_slots = +namespace { + +const pass_data pass_data_delay_slots = +{ + RTL_PASS, /* type */ + "dbr", /* name */ + OPTGROUP_NONE, /* optinfo_flags */ + TV_DBR_SCHED, /* tv_id */ + 0, /* properties_required */ + 0, /* properties_provided */ + 0, /* properties_destroyed */ + 0, /* todo_flags_start */ + 0, /* todo_flags_finish */ +}; + +class pass_delay_slots : public rtl_opt_pass { - { - RTL_PASS, - "dbr", /* name */ - gate_handle_delay_slots, /* gate */ - rest_of_handle_delay_slots, /* execute */ - NULL, /* sub */ - NULL, /* next */ - 0, /* static_pass_number */ - TV_DBR_SCHED, /* tv_id */ - 0, /* properties_required */ - 0, /* properties_provided */ - 0, /* properties_destroyed */ - 0, /* todo_flags_start */ - TODO_dump_func | - TODO_ggc_collect /* todo_flags_finish */ - } -}; +public: + pass_delay_slots (gcc::context *ctxt) + : rtl_opt_pass (pass_data_delay_slots, ctxt) + {} + + /* opt_pass methods: */ + virtual bool gate (function *); + virtual unsigned int execute (function *) + { + return rest_of_handle_delay_slots (); + } + +}; // class pass_delay_slots + +bool +pass_delay_slots::gate (function *) +{ + /* At -O0 dataflow info isn't updated after RA. */ + if (DELAY_SLOTS) + return optimize > 0 && flag_delayed_branch && !crtl->dbr_scheduled_p; + + return false; +} + +} // anon namespace + +rtl_opt_pass * +make_pass_delay_slots (gcc::context *ctxt) +{ + return new pass_delay_slots (ctxt); +} /* Machine dependent reorg pass. */ -static bool -gate_handle_machine_reorg (void) + +namespace { + +const pass_data pass_data_machine_reorg = { - return targetm.machine_dependent_reorg != 0; -} - - -static unsigned int -rest_of_handle_machine_reorg (void) -{ - targetm.machine_dependent_reorg (); - return 0; -} - -struct rtl_opt_pass pass_machine_reorg = + RTL_PASS, /* type */ + "mach", /* name */ + OPTGROUP_NONE, /* optinfo_flags */ + TV_MACH_DEP, /* tv_id */ + 0, /* properties_required */ + 0, /* properties_provided */ + 0, /* properties_destroyed */ + 0, /* todo_flags_start */ + 0, /* todo_flags_finish */ +}; + +class pass_machine_reorg : public rtl_opt_pass { - { - RTL_PASS, - "mach", /* name */ - gate_handle_machine_reorg, /* gate */ - rest_of_handle_machine_reorg, /* execute */ - NULL, /* sub */ - NULL, /* next */ - 0, /* static_pass_number */ - TV_MACH_DEP, /* tv_id */ - 0, /* properties_required */ - 0, /* properties_provided */ - 0, /* properties_destroyed */ - 0, /* todo_flags_start */ - TODO_dump_func | - TODO_ggc_collect /* todo_flags_finish */ - } -}; +public: + pass_machine_reorg (gcc::context *ctxt) + : rtl_opt_pass (pass_data_machine_reorg, ctxt) + {} + + /* opt_pass methods: */ + virtual bool gate (function *) + { + return targetm.machine_dependent_reorg != 0; + } + + virtual unsigned int execute (function *) + { + targetm.machine_dependent_reorg (); + return 0; + } + +}; // class pass_machine_reorg + +} // anon namespace + +rtl_opt_pass * +make_pass_machine_reorg (gcc::context *ctxt) +{ + return new pass_machine_reorg (ctxt); +}