X-Git-Url: http://git.sourceforge.jp/view?p=pf3gnuchains%2Fgcc-fork.git;a=blobdiff_plain;f=gcc%2Fmodulo-sched.c;h=9179de504aa7d220849fad0c9280b43932883644;hp=7e9f6aa1a693cb60bd695ecdb66d74d335e64dc0;hb=0f7f01cfb1510f422a8b27e58bfed9e345116b32;hpb=7bfc9b7cc90d5fc45847f176852078b1ddd504b1 diff --git a/gcc/modulo-sched.c b/gcc/modulo-sched.c index 7e9f6aa1a69..9179de504aa 100644 --- a/gcc/modulo-sched.c +++ b/gcc/modulo-sched.c @@ -1,5 +1,5 @@ /* Swing Modulo Scheduling implementation. - Copyright (C) 2004, 2005, 2006, 2007 + Copyright (C) 2004, 2005, 2006, 2007, 2008, 2009, 2010 Free Software Foundation, Inc. Contributed by Ayal Zaks and Mustafa Hagog @@ -24,7 +24,7 @@ along with GCC; see the file COPYING3. If not see #include "system.h" #include "coretypes.h" #include "tm.h" -#include "toplev.h" +#include "diagnostic-core.h" #include "rtl.h" #include "tm_p.h" #include "hard-reg-set.h" @@ -34,7 +34,6 @@ along with GCC; see the file COPYING3. If not see #include "insn-config.h" #include "insn-attr.h" #include "except.h" -#include "toplev.h" #include "recog.h" #include "sched-int.h" #include "target.h" @@ -47,6 +46,8 @@ along with GCC; see the file COPYING3. If not see #include "ddg.h" #include "timevar.h" #include "tree-pass.h" +#include "dbgcnt.h" +#include "df.h" #ifdef INSN_SCHEDULING @@ -95,7 +96,7 @@ along with GCC; see the file COPYING3. If not see Currently SMS relies on the do-loop pattern to recognize such loops, where (1) the control part comprises of all insns defining and/or using a certain 'count' register and (2) the loop count can be - adjusted by modifying this register prior to the loop. + adjusted by modifying this register prior to the loop. TODO: Rely on cfgloop analysis instead. */ /* This page defines partial-schedule structures and functions for @@ -167,39 +168,33 @@ struct undo_replace_buff_elem }; - + static partial_schedule_ptr create_partial_schedule (int ii, ddg_ptr, int history); static void free_partial_schedule (partial_schedule_ptr); static void reset_partial_schedule (partial_schedule_ptr, int new_ii); void print_partial_schedule (partial_schedule_ptr, FILE *); +static void verify_partial_schedule (partial_schedule_ptr, sbitmap); static ps_insn_ptr ps_add_node_check_conflicts (partial_schedule_ptr, ddg_node_ptr node, int cycle, sbitmap must_precede, sbitmap must_follow); static void rotate_partial_schedule (partial_schedule_ptr, int); void set_row_column_for_ps (partial_schedule_ptr); -static bool ps_unschedule_node (partial_schedule_ptr, ddg_node_ptr ); +static void ps_insert_empty_row (partial_schedule_ptr, int, sbitmap); +static int compute_split_row (sbitmap, int, int, int, ddg_node_ptr); /* This page defines constants and structures for the modulo scheduling driver. */ -/* As in haifa-sched.c: */ -/* issue_rate is the number of insns that can be scheduled in the same - machine cycle. It can be defined in the config/mach/mach.h file, - otherwise we set it to 1. */ - -static int issue_rate; - -static int sms_order_nodes (ddg_ptr, int, int * result); +static int sms_order_nodes (ddg_ptr, int, int *, int *); static void set_node_sched_params (ddg_ptr); static partial_schedule_ptr sms_schedule_by_order (ddg_ptr, int, int, int *); -static void permute_partial_schedule (partial_schedule_ptr ps, rtx last); -static void generate_prolog_epilog (partial_schedule_ptr, struct loop *loop, +static void permute_partial_schedule (partial_schedule_ptr, rtx); +static void generate_prolog_epilog (partial_schedule_ptr, struct loop *, rtx, rtx); -static void duplicate_insns_of_cycles (partial_schedule_ptr ps, - int from_stage, int to_stage, - int is_prolog, rtx count_reg); +static void duplicate_insns_of_cycles (partial_schedule_ptr, + int, int, int, rtx); #define SCHED_ASAP(x) (((node_sched_params_ptr)(x)->aux.info)->asap) #define SCHED_TIME(x) (((node_sched_params_ptr)(x)->aux.info)->time) @@ -240,7 +235,7 @@ typedef struct node_sched_params code in order to use sched_analyze() for computing the dependencies. They are used when initializing the sched_info structure. */ static const char * -sms_print_insn (rtx insn, int aligned ATTRIBUTE_UNUSED) +sms_print_insn (const_rtx insn, int aligned ATTRIBUTE_UNUSED) { static char tmp[80]; @@ -256,7 +251,17 @@ compute_jump_reg_dependencies (rtx insn ATTRIBUTE_UNUSED, { } -static struct sched_info sms_sched_info = +static struct common_sched_info_def sms_common_sched_info; + +static struct sched_deps_info_def sms_sched_deps_info = + { + compute_jump_reg_dependencies, + NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, + NULL, + 0, 0, 0 + }; + +static struct haifa_sched_info sms_sched_info = { NULL, NULL, @@ -265,16 +270,15 @@ static struct sched_info sms_sched_info = NULL, sms_print_insn, NULL, - compute_jump_reg_dependencies, + NULL, /* insn_finishes_block_p */ NULL, NULL, NULL, NULL, - 0, 0, 0, + 0, 0, - NULL, NULL, NULL, NULL, NULL, + NULL, NULL, NULL, 0 }; - /* Given HEAD and TAIL which are the first and last insns in a loop; return the register which controls the loop. Return zero if it has more than one occurrence in the loop besides the control part or the @@ -283,8 +287,7 @@ static rtx doloop_register_get (rtx head ATTRIBUTE_UNUSED, rtx tail ATTRIBUTE_UNUSED) { #ifdef HAVE_doloop_end - rtx reg, condition, insn; - bool found = false; + rtx reg, condition, insn, first_insn_not_to_check; if (!JUMP_P (tail)) return NULL_RTX; @@ -306,25 +309,23 @@ doloop_register_get (rtx head ATTRIBUTE_UNUSED, rtx tail ATTRIBUTE_UNUSED) until the decrement. We assume the control part consists of either a single (parallel) branch-on-count or a (non-parallel) branch immediately preceded by a single (decrement) insn. */ - for (insn = head; insn != PREV_INSN (tail); insn = NEXT_INSN (insn)) - if ((found = reg_mentioned_p (reg, insn)) == true) - break; - if (found) - { - if (dump_file) - fprintf (dump_file, "SMS count_reg found outside control\n"); + first_insn_not_to_check = (GET_CODE (PATTERN (tail)) == PARALLEL ? tail + : PREV_INSN (tail)); - return NULL_RTX; - } - /* One last check in case the do-loop pattern is parallel. */ - if (GET_CODE (PATTERN (tail)) == PARALLEL) - if (reg_mentioned_p (reg, PREV_INSN (tail))) + for (insn = head; insn != first_insn_not_to_check; insn = NEXT_INSN (insn)) + if (reg_mentioned_p (reg, insn)) { if (dump_file) - fprintf (dump_file, "SMS count_reg found outside control\n"); + { + fprintf (dump_file, "SMS count_reg found "); + print_rtl_single (dump_file, reg); + fprintf (dump_file, " outside control in insn:\n"); + print_rtl_single (dump_file, insn); + } return NULL_RTX; } + return reg; #else return NULL_RTX; @@ -348,12 +349,12 @@ const_iteration_count (rtx count_reg, basic_block pre_header, get_ebb_head_tail (pre_header, pre_header, &head, &tail); for (insn = tail; insn != PREV_INSN (head); insn = PREV_INSN (insn)) - if (INSN_P (insn) && single_set (insn) && + if (NONDEBUG_INSN_P (insn) && single_set (insn) && rtx_equal_p (count_reg, SET_DEST (single_set (insn)))) { rtx pat = single_set (insn); - if (GET_CODE (SET_SRC (pat)) == CONST_INT) + if (CONST_INT_P (SET_SRC (pat))) { *count = INTVAL (SET_SRC (pat)); return insn; @@ -371,7 +372,10 @@ const_iteration_count (rtx count_reg, basic_block pre_header, static int res_MII (ddg_ptr g) { - return (g->num_nodes / issue_rate); + if (targetm.sched.sms_res_mii) + return targetm.sched.sms_res_mii (g); + + return ((g->num_nodes - g->num_debug) / issue_rate); } @@ -504,7 +508,9 @@ generate_reg_moves (partial_schedule_ptr ps, bool rescan) /* Now generate the reg_moves, attaching relevant uses to them. */ SCHED_NREG_MOVES (u) = nreg_moves; old_reg = prev_reg = copy_rtx (SET_DEST (single_set (u->insn))); - last_reg_move = u->insn; + /* Insert the reg-moves right before the notes which precede + the insn they relates to. */ + last_reg_move = u->first_note; for (i_reg_move = 0; i_reg_move < nreg_moves; i_reg_move++) { @@ -568,23 +574,27 @@ free_undo_replace_buff (struct undo_replace_buff_elem *reg_move_replaces) static void normalize_sched_times (partial_schedule_ptr ps) { - int i; - ddg_ptr g = ps->g; + int row; int amount = PS_MIN_CYCLE (ps); int ii = ps->ii; + ps_insn_ptr crr_insn; - /* Don't include the closing branch assuming that it is the last node. */ - for (i = 0; i < g->num_nodes - 1; i++) - { - ddg_node_ptr u = &g->nodes[i]; - int normalized_time = SCHED_TIME (u) - amount; - - gcc_assert (normalized_time >= 0); - - SCHED_TIME (u) = normalized_time; - SCHED_ROW (u) = normalized_time % ii; - SCHED_STAGE (u) = normalized_time / ii; - } + for (row = 0; row < ii; row++) + for (crr_insn = ps->rows[row]; crr_insn; crr_insn = crr_insn->next_in_row) + { + ddg_node_ptr u = crr_insn->node; + int normalized_time = SCHED_TIME (u) - amount; + + if (dump_file) + fprintf (dump_file, "crr_insn->node=%d, crr_insn->cycle=%d,\ + min_cycle=%d\n", crr_insn->node->cuid, SCHED_TIME + (u), ps->min_cycle); + gcc_assert (SCHED_TIME (u) >= ps->min_cycle); + gcc_assert (SCHED_TIME (u) <= ps->max_cycle); + SCHED_TIME (u) = normalized_time; + SCHED_ROW (u) = normalized_time % ii; + SCHED_STAGE (u) = normalized_time / ii; + } } /* Set SCHED_COLUMN of each node according to its position in PS. */ @@ -695,7 +705,7 @@ generate_prolog_epilog (partial_schedule_ptr ps, struct loop *loop, int i; int last_stage = PS_STAGE_COUNT (ps) - 1; edge e; - + /* Generate the prolog, inserting its insns on the loop-entry edge. */ start_sequence (); @@ -717,7 +727,7 @@ generate_prolog_epilog (partial_schedule_ptr ps, struct loop *loop, for (i = 0; i < last_stage; i++) duplicate_insns_of_cycles (ps, 0, i, 1, count_reg); - + /* Put the prolog on the entry edge. */ e = loop_preheader_edge (loop); split_edge_and_insert (e, get_insns ()); @@ -729,7 +739,7 @@ generate_prolog_epilog (partial_schedule_ptr ps, struct loop *loop, for (i = 0; i < last_stage; i++) duplicate_insns_of_cycles (ps, i + 1, last_stage, 0, count_reg); - + /* Put the epilogue on the exit edge. */ gcc_assert (single_exit (loop)); e = single_exit (loop); @@ -759,7 +769,7 @@ loop_single_full_bb_p (struct loop *loop) for (; head != NEXT_INSN (tail); head = NEXT_INSN (head)) { if (NOTE_P (head) || LABEL_P (head) - || (INSN_P (head) && JUMP_P (head))) + || (INSN_P (head) && (DEBUG_INSN_P (head) || JUMP_P (head)))) continue; empty_bb = false; break; @@ -788,14 +798,18 @@ loop_canon_p (struct loop *loop) { if (loop->inner || !loop_outer (loop)) + { + if (dump_file) + fprintf (dump_file, "SMS loop inner or !loop_outer\n"); return false; + } if (!single_exit (loop)) { if (dump_file) { rtx insn = BB_END (loop->header); - + fprintf (dump_file, "SMS loop many exits "); fprintf (dump_file, " %s %d (file, line)\n", insn_file (insn), insn_line (insn)); @@ -808,7 +822,7 @@ loop_canon_p (struct loop *loop) if (dump_file) { rtx insn = BB_END (loop->header); - + fprintf (dump_file, "SMS loop many BBs. "); fprintf (dump_file, " %s %d (file, line)\n", insn_file (insn), insn_line (insn)); @@ -844,6 +858,19 @@ canon_loop (struct loop *loop) } } +/* Setup infos. */ +static void +setup_sched_infos (void) +{ + memcpy (&sms_common_sched_info, &haifa_common_sched_info, + sizeof (sms_common_sched_info)); + sms_common_sched_info.sched_pass_id = SCHED_SMS_PASS; + common_sched_info = &sms_common_sched_info; + + sched_deps_info = &sms_sched_deps_info; + current_sched_info = &sms_sched_info; +} + /* Probability in % that the sms-ed loop rolls enough so that optimized version may be entered. Just a guess. */ #define PROB_SMS_ENOUGH_ITERATIONS 80 @@ -856,11 +883,10 @@ canon_loop (struct loop *loop) static void sms_schedule (void) { - static int passes = 0; rtx insn; ddg_ptr *g_arr, g; int * node_order; - int maxii; + int maxii, max_asap; loop_iterator li; partial_schedule_ptr ps; basic_block bb = NULL; @@ -890,21 +916,19 @@ sms_schedule (void) issue_rate = 1; /* Initialize the scheduler. */ - current_sched_info = &sms_sched_info; - - /* Init Data Flow analysis, to be used in interloop dep calculation. */ - df_set_flags (DF_LR_RUN_DCE); - df_rd_add_problem (); - df_note_add_problem (); - df_chain_add_problem (DF_DU_CHAIN); - df_analyze (); - regstat_compute_calls_crossed (); - sched_init (); + setup_sched_infos (); + haifa_sched_init (); /* Allocate memory to hold the DDG array one entry for each loop. We use loop->num as index into this array. */ g_arr = XCNEWVEC (ddg_ptr, number_of_loops ()); + if (dump_file) + { + fprintf (dump_file, "\n\nSMS analysis phase\n"); + fprintf (dump_file, "===================\n\n"); + } + /* Build DDGs for all the relevant loops and hold them in G_ARR indexed by the loop index. */ FOR_EACH_LOOP (li, loop, 0) @@ -913,19 +937,32 @@ sms_schedule (void) rtx count_reg; /* For debugging. */ - if ((passes++ > MAX_SMS_LOOP_NUMBER) && (MAX_SMS_LOOP_NUMBER != -1)) + if (dbg_cnt (sms_sched_loop) == false) { if (dump_file) - fprintf (dump_file, "SMS reached MAX_PASSES... \n"); + fprintf (dump_file, "SMS reached max limit... \n"); break; } + if (dump_file) + { + rtx insn = BB_END (loop->header); + + fprintf (dump_file, "SMS loop num: %d, file: %s, line: %d\n", + loop->num, insn_file (insn), insn_line (insn)); + + } + if (! loop_canon_p (loop)) continue; if (! loop_single_full_bb_p (loop)) + { + if (dump_file) + fprintf (dump_file, "SMS not loop_single_full_bb_p\n"); continue; + } bb = loop->header; @@ -935,7 +972,7 @@ sms_schedule (void) if (single_exit (loop)->count) trip_count = latch_edge->count / single_exit (loop)->count; - /* Perfrom SMS only on loops that their average count is above threshold. */ + /* Perform SMS only on loops that their average count is above threshold. */ if ( latch_edge->count && (latch_edge->count < single_exit (loop)->count * SMS_LOOP_AVERAGE_COUNT_THRESHOLD)) @@ -966,11 +1003,15 @@ sms_schedule (void) /* Make sure this is a doloop. */ if ( !(count_reg = doloop_register_get (head, tail))) + { + if (dump_file) + fprintf (dump_file, "SMS doloop_register_get failed\n"); continue; + } /* Don't handle BBs with calls or barriers, or !single_set insns, or auto-increment insns (to avoid creating invalid reg-moves - for the auto-increment insns). + for the auto-increment insns). ??? Should handle auto-increment insns. ??? Should handle insns defining subregs. */ for (insn = head; insn != NEXT_INSN (tail); insn = NEXT_INSN (insn)) @@ -979,7 +1020,7 @@ sms_schedule (void) if (CALL_P (insn) || BARRIER_P (insn) - || (INSN_P (insn) && !JUMP_P (insn) + || (NONDEBUG_INSN_P (insn) && !JUMP_P (insn) && !single_set (insn) && GET_CODE (PATTERN (insn)) != USE) || (FIND_REG_INC_NOTE (insn, NULL_RTX) != 0) || (INSN_P (insn) && (set = single_set (insn)) @@ -997,7 +1038,7 @@ sms_schedule (void) fprintf (dump_file, "SMS loop-with-barrier\n"); else if (FIND_REG_INC_NOTE (insn, NULL_RTX) != 0) fprintf (dump_file, "SMS reg inc\n"); - else if ((INSN_P (insn) && !JUMP_P (insn) + else if ((NONDEBUG_INSN_P (insn) && !JUMP_P (insn) && !single_set (insn) && GET_CODE (PATTERN (insn)) != USE)) fprintf (dump_file, "SMS loop-with-not-single-set\n"); else @@ -1011,12 +1052,20 @@ sms_schedule (void) if (! (g = create_ddg (bb, 0))) { if (dump_file) - fprintf (dump_file, "SMS doloop\n"); + fprintf (dump_file, "SMS create_ddg failed\n"); continue; } g_arr[loop->num] = g; + if (dump_file) + fprintf (dump_file, "...OK\n"); + } + if (dump_file) + { + fprintf (dump_file, "\nSMS transformation phase\n"); + fprintf (dump_file, "=========================\n\n"); + } /* We don't want to perform SMS on new loops - created by versioning. */ FOR_EACH_LOOP (li, loop, 0) @@ -1031,7 +1080,14 @@ sms_schedule (void) continue; if (dump_file) - print_ddg (dump_file, g); + { + rtx insn = BB_END (loop->header); + + fprintf (dump_file, "SMS loop num: %d, file: %s, line: %d\n", + loop->num, insn_file (insn), insn_line (insn)); + + print_ddg (dump_file, g); + } get_ebb_head_tail (loop->header, loop->header, &head, &tail); @@ -1087,9 +1143,9 @@ sms_schedule (void) node_order = XNEWVEC (int, g->num_nodes); mii = 1; /* Need to pass some estimate of mii. */ - rec_mii = sms_order_nodes (g, mii, node_order); + rec_mii = sms_order_nodes (g, mii, node_order, &max_asap); mii = MAX (res_MII (g), rec_mii); - maxii = MAXII_FACTOR * mii; + maxii = MAX (max_asap, MAXII_FACTOR * mii); if (dump_file) fprintf (dump_file, "SMS iis %d %d %d (rec_mii, mii, maxii)\n", @@ -1101,12 +1157,14 @@ sms_schedule (void) ps = sms_schedule_by_order (g, mii, maxii, node_order); - if (ps) + if (ps){ stage_count = PS_STAGE_COUNT (ps); + gcc_assert(stage_count >= 1); + } /* Stage count of 1 means that there is no interleaving between iterations, let the scheduling passes do the job. */ - if (stage_count < 1 + if (stage_count <= 1 || (count_init && (loop_count <= stage_count)) || (flag_branch_probabilities && (trip_count <= stage_count))) { @@ -1140,14 +1198,14 @@ sms_schedule (void) the closing_branch was scheduled and should appear in the last (ii-1) row. Otherwise, we are free to schedule the branch, and we let nodes that were scheduled at the first PS_MIN_CYCLE cycle appear in the first - row; this should reduce stage_count to minimum. + row; this should reduce stage_count to minimum. TODO: Revisit the issue of scheduling the insns of the control part relative to the branch when the control part has more than one insn. */ normalize_sched_times (ps); rotate_partial_schedule (ps, PS_MIN_CYCLE (ps)); set_columns_for_ps (ps); - + canon_loop (loop); /* case the BCT count is not known , Do loop-versioning */ @@ -1183,7 +1241,7 @@ sms_schedule (void) print_node_sched_params (dump_file, g->num_nodes, g); /* Generate prolog and epilog. */ generate_prolog_epilog (ps, loop, count_reg, count_init); - + free_undo_replace_buff (reg_move_replaces); } @@ -1193,11 +1251,10 @@ sms_schedule (void) free_ddg (g); } - regstat_free_calls_crossed (); free (g_arr); /* Release scheduler data, needed until now because of DFA. */ - sched_finish (); + haifa_sched_finish (); loop_optimizer_finalize (); } @@ -1277,6 +1334,9 @@ sms_schedule (void) set to 0 to save compile time. */ #define DFA_HISTORY SMS_DFA_HISTORY +/* A threshold for the number of repeated unsuccessful attempts to insert + an empty row, before we flush the partial schedule and start over. */ +#define MAX_SPLIT_NUM 10 /* Given the partial schedule PS, this function calculates and returns the cycles in which we can schedule the node with the given index I. NOTE: Here we do the backtracking in SMS, in some special cases. We have @@ -1315,20 +1375,45 @@ get_sched_window (partial_schedule_ptr ps, int *nodes_order, int i, for (e = u_node->in; e != 0; e = e->next_in) { ddg_node_ptr v_node = e->src; + + if (dump_file) + { + fprintf (dump_file, "\nProcessing edge: "); + print_ddg_edge (dump_file, e); + fprintf (dump_file, + "\nScheduling %d (%d) in psp_not_empty," + " checking p %d (%d): ", u_node->cuid, + INSN_UID (u_node->insn), v_node->cuid, INSN_UID + (v_node->insn)); + } + if (TEST_BIT (sched_nodes, v_node->cuid)) { - int node_st = SCHED_TIME (v_node) - + e->latency - (e->distance * ii); + int p_st = SCHED_TIME (v_node); + + early_start = + MAX (early_start, p_st + e->latency - (e->distance * ii)); - early_start = MAX (early_start, node_st); + if (dump_file) + fprintf (dump_file, + "pred st = %d; early_start = %d; latency: %d", + p_st, early_start, e->latency); if (e->data_type == MEM_DEP) end = MIN (end, SCHED_TIME (v_node) + ii - 1); } + else if (dump_file) + fprintf (dump_file, "the node is not scheduled\n"); } start = early_start; end = MIN (end, early_start + ii); + /* Schedule the node close to it's predecessors. */ step = 1; + + if (dump_file) + fprintf (dump_file, + "\nScheduling %d (%d) in a window (%d..%d) with step %d\n", + u_node->cuid, INSN_UID (u_node->insn), start, end, step); } else if (!psp_not_empty && pss_not_empty) @@ -1339,24 +1424,58 @@ get_sched_window (partial_schedule_ptr ps, int *nodes_order, int i, for (e = u_node->out; e != 0; e = e->next_out) { ddg_node_ptr v_node = e->dest; + + if (dump_file) + { + fprintf (dump_file, "\nProcessing edge:"); + print_ddg_edge (dump_file, e); + fprintf (dump_file, + "\nScheduling %d (%d) in pss_not_empty," + " checking s %d (%d): ", u_node->cuid, + INSN_UID (u_node->insn), v_node->cuid, INSN_UID + (v_node->insn)); + } + if (TEST_BIT (sched_nodes, v_node->cuid)) { - late_start = MIN (late_start, - SCHED_TIME (v_node) - e->latency - + (e->distance * ii)); + int s_st = SCHED_TIME (v_node); + + late_start = MIN (late_start, + s_st - e->latency + (e->distance * ii)); + + if (dump_file) + fprintf (dump_file, + "succ st = %d; late_start = %d; latency = %d", + s_st, late_start, e->latency); + if (e->data_type == MEM_DEP) end = MAX (end, SCHED_TIME (v_node) - ii + 1); + if (dump_file) + fprintf (dump_file, "end = %d\n", end); + } + else if (dump_file) + fprintf (dump_file, "the node is not scheduled\n"); + } start = late_start; end = MAX (end, late_start - ii); + /* Schedule the node close to it's successors. */ step = -1; + + if (dump_file) + fprintf (dump_file, + "\nScheduling %d (%d) in a window (%d..%d) with step %d\n", + u_node->cuid, INSN_UID (u_node->insn), start, end, step); + } else if (psp_not_empty && pss_not_empty) { int early_start = INT_MIN; int late_start = INT_MAX; + int count_preds = 0; + int count_succs = 0; start = INT_MIN; end = INT_MAX; @@ -1364,31 +1483,91 @@ get_sched_window (partial_schedule_ptr ps, int *nodes_order, int i, { ddg_node_ptr v_node = e->src; + if (dump_file) + { + fprintf (dump_file, "\nProcessing edge:"); + print_ddg_edge (dump_file, e); + fprintf (dump_file, + "\nScheduling %d (%d) in psp_pss_not_empty," + " checking p %d (%d): ", u_node->cuid, INSN_UID + (u_node->insn), v_node->cuid, INSN_UID + (v_node->insn)); + } + if (TEST_BIT (sched_nodes, v_node->cuid)) { + int p_st = SCHED_TIME (v_node); + early_start = MAX (early_start, - SCHED_TIME (v_node) + e->latency + p_st + e->latency - (e->distance * ii)); + + if (dump_file) + fprintf (dump_file, + "pred st = %d; early_start = %d; latency = %d", + p_st, early_start, e->latency); + + if (e->type == TRUE_DEP && e->data_type == REG_DEP) + count_preds++; + if (e->data_type == MEM_DEP) end = MIN (end, SCHED_TIME (v_node) + ii - 1); } + else if (dump_file) + fprintf (dump_file, "the node is not scheduled\n"); + } for (e = u_node->out; e != 0; e = e->next_out) { ddg_node_ptr v_node = e->dest; + if (dump_file) + { + fprintf (dump_file, "\nProcessing edge:"); + print_ddg_edge (dump_file, e); + fprintf (dump_file, + "\nScheduling %d (%d) in psp_pss_not_empty," + " checking s %d (%d): ", u_node->cuid, INSN_UID + (u_node->insn), v_node->cuid, INSN_UID + (v_node->insn)); + } + if (TEST_BIT (sched_nodes, v_node->cuid)) { + int s_st = SCHED_TIME (v_node); + late_start = MIN (late_start, - SCHED_TIME (v_node) - e->latency + s_st - e->latency + (e->distance * ii)); + + if (dump_file) + fprintf (dump_file, + "succ st = %d; late_start = %d; latency = %d", + s_st, late_start, e->latency); + + if (e->type == TRUE_DEP && e->data_type == REG_DEP) + count_succs++; + if (e->data_type == MEM_DEP) start = MAX (start, SCHED_TIME (v_node) - ii + 1); } + else if (dump_file) + fprintf (dump_file, "the node is not scheduled\n"); + } start = MAX (start, early_start); end = MIN (end, MIN (early_start + ii, late_start + 1)); step = 1; + /* If there are more successors than predecessors schedule the + node close to it's successors. */ + if (count_succs >= count_preds) + { + int old_start = start; + + start = end - 1; + end = old_start - 1; + step = -1; + } } else /* psp is empty && pss is empty. */ { @@ -1404,21 +1583,152 @@ get_sched_window (partial_schedule_ptr ps, int *nodes_order, int i, sbitmap_free (pss); if ((start >= end && step == 1) || (start <= end && step == -1)) + { + if (dump_file) + fprintf (dump_file, "\nEmpty window: start=%d, end=%d, step=%d\n", + start, end, step); return -1; - else + } + return 0; } +/* Calculate MUST_PRECEDE/MUST_FOLLOW bitmaps of U_NODE; which is the + node currently been scheduled. At the end of the calculation + MUST_PRECEDE/MUST_FOLLOW contains all predecessors/successors of + U_NODE which are (1) already scheduled in the first/last row of + U_NODE's scheduling window, (2) whose dependence inequality with U + becomes an equality when U is scheduled in this same row, and (3) + whose dependence latency is zero. + + The first and last rows are calculated using the following parameters: + START/END rows - The cycles that begins/ends the traversal on the window; + searching for an empty cycle to schedule U_NODE. + STEP - The direction in which we traverse the window. + II - The initiation interval. */ + +static void +calculate_must_precede_follow (ddg_node_ptr u_node, int start, int end, + int step, int ii, sbitmap sched_nodes, + sbitmap must_precede, sbitmap must_follow) +{ + ddg_edge_ptr e; + int first_cycle_in_window, last_cycle_in_window; + + gcc_assert (must_precede && must_follow); + + /* Consider the following scheduling window: + {first_cycle_in_window, first_cycle_in_window+1, ..., + last_cycle_in_window}. If step is 1 then the following will be + the order we traverse the window: {start=first_cycle_in_window, + first_cycle_in_window+1, ..., end=last_cycle_in_window+1}, + or {start=last_cycle_in_window, last_cycle_in_window-1, ..., + end=first_cycle_in_window-1} if step is -1. */ + first_cycle_in_window = (step == 1) ? start : end - step; + last_cycle_in_window = (step == 1) ? end - step : start; + + sbitmap_zero (must_precede); + sbitmap_zero (must_follow); + + if (dump_file) + fprintf (dump_file, "\nmust_precede: "); + + /* Instead of checking if: + (SMODULO (SCHED_TIME (e->src), ii) == first_row_in_window) + && ((SCHED_TIME (e->src) + e->latency - (e->distance * ii)) == + first_cycle_in_window) + && e->latency == 0 + we use the fact that latency is non-negative: + SCHED_TIME (e->src) - (e->distance * ii) <= + SCHED_TIME (e->src) + e->latency - (e->distance * ii)) <= + first_cycle_in_window + and check only if + SCHED_TIME (e->src) - (e->distance * ii) == first_cycle_in_window */ + for (e = u_node->in; e != 0; e = e->next_in) + if (TEST_BIT (sched_nodes, e->src->cuid) + && ((SCHED_TIME (e->src) - (e->distance * ii)) == + first_cycle_in_window)) + { + if (dump_file) + fprintf (dump_file, "%d ", e->src->cuid); + + SET_BIT (must_precede, e->src->cuid); + } + + if (dump_file) + fprintf (dump_file, "\nmust_follow: "); + + /* Instead of checking if: + (SMODULO (SCHED_TIME (e->dest), ii) == last_row_in_window) + && ((SCHED_TIME (e->dest) - e->latency + (e->distance * ii)) == + last_cycle_in_window) + && e->latency == 0 + we use the fact that latency is non-negative: + SCHED_TIME (e->dest) + (e->distance * ii) >= + SCHED_TIME (e->dest) - e->latency + (e->distance * ii)) >= + last_cycle_in_window + and check only if + SCHED_TIME (e->dest) + (e->distance * ii) == last_cycle_in_window */ + for (e = u_node->out; e != 0; e = e->next_out) + if (TEST_BIT (sched_nodes, e->dest->cuid) + && ((SCHED_TIME (e->dest) + (e->distance * ii)) == + last_cycle_in_window)) + { + if (dump_file) + fprintf (dump_file, "%d ", e->dest->cuid); + + SET_BIT (must_follow, e->dest->cuid); + } + + if (dump_file) + fprintf (dump_file, "\n"); +} + +/* Return 1 if U_NODE can be scheduled in CYCLE. Use the following + parameters to decide if that's possible: + PS - The partial schedule. + U - The serial number of U_NODE. + NUM_SPLITS - The number of row splits made so far. + MUST_PRECEDE - The nodes that must precede U_NODE. (only valid at + the first row of the scheduling window) + MUST_FOLLOW - The nodes that must follow U_NODE. (only valid at the + last row of the scheduling window) */ + +static bool +try_scheduling_node_in_cycle (partial_schedule_ptr ps, ddg_node_ptr u_node, + int u, int cycle, sbitmap sched_nodes, + int *num_splits, sbitmap must_precede, + sbitmap must_follow) +{ + ps_insn_ptr psi; + bool success = 0; + + verify_partial_schedule (ps, sched_nodes); + psi = ps_add_node_check_conflicts (ps, u_node, cycle, + must_precede, must_follow); + if (psi) + { + SCHED_TIME (u_node) = cycle; + SET_BIT (sched_nodes, u); + success = 1; + *num_splits = 0; + if (dump_file) + fprintf (dump_file, "Scheduled w/o split in %d\n", cycle); + + } + + return success; +} + /* This function implements the scheduling algorithm for SMS according to the above algorithm. */ static partial_schedule_ptr sms_schedule_by_order (ddg_ptr g, int mii, int maxii, int *nodes_order) { int ii = mii; - int i, c, success; - int try_again_with_larger_ii = true; + int i, c, success, num_splits = 0; + int flush_and_start_over = true; int num_nodes = g->num_nodes; - ddg_edge_ptr e; int start, end, step; /* Place together into one struct? */ sbitmap sched_nodes = sbitmap_alloc (num_nodes); sbitmap must_precede = sbitmap_alloc (num_nodes); @@ -1430,19 +1740,13 @@ sms_schedule_by_order (ddg_ptr g, int mii, int maxii, int *nodes_order) sbitmap_ones (tobe_scheduled); sbitmap_zero (sched_nodes); - while ((! sbitmap_equal (tobe_scheduled, sched_nodes) - || try_again_with_larger_ii ) && ii < maxii) + while (flush_and_start_over && (ii < maxii)) { - int j; - bool unscheduled_nodes = false; if (dump_file) fprintf (dump_file, "Starting with ii=%d\n", ii); - if (try_again_with_larger_ii) - { - try_again_with_larger_ii = false; - sbitmap_zero (sched_nodes); - } + flush_and_start_over = false; + sbitmap_zero (sched_nodes); for (i = 0; i < num_nodes; i++) { @@ -1450,7 +1754,7 @@ sms_schedule_by_order (ddg_ptr g, int mii, int maxii, int *nodes_order) ddg_node_ptr u_node = &ps->g->nodes[u]; rtx insn = u_node->insn; - if (!INSN_P (insn)) + if (!NONDEBUG_INSN_P (insn)) { RESET_BIT (tobe_scheduled, u); continue; @@ -1466,101 +1770,264 @@ sms_schedule_by_order (ddg_ptr g, int mii, int maxii, int *nodes_order) continue; /* Try to get non-empty scheduling window. */ - j = i; - while (get_sched_window (ps, nodes_order, i, sched_nodes, ii, &start, &step, &end) < 0 - && j > 0) - { - unscheduled_nodes = true; - if (TEST_BIT (NODE_PREDECESSORS (u_node), nodes_order[j - 1]) - || TEST_BIT (NODE_SUCCESSORS (u_node), nodes_order[j - 1])) - { - ps_unschedule_node (ps, &ps->g->nodes[nodes_order[j - 1]]); - RESET_BIT (sched_nodes, nodes_order [j - 1]); - } - j--; - } - if (j < 0) - { - /* ??? Try backtracking instead of immediately ii++? */ - ii++; - try_again_with_larger_ii = true; - reset_partial_schedule (ps, ii); - break; - } - /* 2. Try scheduling u in window. */ - if (dump_file) - fprintf (dump_file, - "Trying to schedule node %d in (%d .. %d) step %d\n", - u, start, end, step); - - /* use must_follow & must_precede bitmaps to determine order - of nodes within the cycle. */ - sbitmap_zero (must_precede); - sbitmap_zero (must_follow); - /* TODO: We can add an insn to the must_precede or must_follow - bitmaps only if it has tight dependence to U and they - both scheduled in the same row. The current check is less - conservative and content with the fact that both U and the - insn are scheduled in the same row. */ - for (e = u_node->in; e != 0; e = e->next_in) - if (TEST_BIT (sched_nodes, e->src->cuid) - && (SMODULO (SCHED_TIME (e->src), ii) == SMODULO (start, ii))) - SET_BIT (must_precede, e->src->cuid); - - for (e = u_node->out; e != 0; e = e->next_out) - if (TEST_BIT (sched_nodes, e->dest->cuid) - && (SMODULO (SCHED_TIME (e->dest), ii) == - SMODULO ((end - step), ii))) - SET_BIT (must_follow, e->dest->cuid); - - success = 0; - if ((step > 0 && start < end) || (step < 0 && start > end)) - for (c = start; c != end; c += step) - { - ps_insn_ptr psi; - - psi = ps_add_node_check_conflicts (ps, u_node, c, - must_precede, - must_follow); - - if (psi) - { - SCHED_TIME (u_node) = c; - SET_BIT (sched_nodes, u); - success = 1; - if (dump_file) - fprintf (dump_file, "Schedule in %d\n", c); - break; - } - } - if (!success) - { - /* ??? Try backtracking instead of immediately ii++? */ - ii++; - try_again_with_larger_ii = true; - reset_partial_schedule (ps, ii); - break; - } - if (unscheduled_nodes) - break; + success = 0; + if (get_sched_window (ps, nodes_order, i, sched_nodes, ii, &start, + &step, &end) == 0) + { + if (dump_file) + fprintf (dump_file, "\nTrying to schedule node %d \ + INSN = %d in (%d .. %d) step %d\n", u, (INSN_UID + (g->nodes[u].insn)), start, end, step); + + gcc_assert ((step > 0 && start < end) + || (step < 0 && start > end)); + + calculate_must_precede_follow (u_node, start, end, step, ii, + sched_nodes, must_precede, + must_follow); + + for (c = start; c != end; c += step) + { + sbitmap tmp_precede = NULL; + sbitmap tmp_follow = NULL; + + if (c == start) + { + if (step == 1) + tmp_precede = must_precede; + else /* step == -1. */ + tmp_follow = must_follow; + } + if (c == end - step) + { + if (step == 1) + tmp_follow = must_follow; + else /* step == -1. */ + tmp_precede = must_precede; + } + + success = + try_scheduling_node_in_cycle (ps, u_node, u, c, + sched_nodes, + &num_splits, tmp_precede, + tmp_follow); + if (success) + break; + } + + verify_partial_schedule (ps, sched_nodes); + } + if (!success) + { + int split_row; + + if (ii++ == maxii) + break; + + if (num_splits >= MAX_SPLIT_NUM) + { + num_splits = 0; + flush_and_start_over = true; + verify_partial_schedule (ps, sched_nodes); + reset_partial_schedule (ps, ii); + verify_partial_schedule (ps, sched_nodes); + break; + } + + num_splits++; + /* The scheduling window is exclusive of 'end' + whereas compute_split_window() expects an inclusive, + ordered range. */ + if (step == 1) + split_row = compute_split_row (sched_nodes, start, end - 1, + ps->ii, u_node); + else + split_row = compute_split_row (sched_nodes, end + 1, start, + ps->ii, u_node); - /* ??? If (success), check register pressure estimates. */ - } /* Continue with next node. */ - } /* While try_again_with_larger_ii. */ + ps_insert_empty_row (ps, split_row, sched_nodes); + i--; /* Go back and retry node i. */ - sbitmap_free (sched_nodes); - sbitmap_free (must_precede); - sbitmap_free (must_follow); - sbitmap_free (tobe_scheduled); + if (dump_file) + fprintf (dump_file, "num_splits=%d\n", num_splits); + } + /* ??? If (success), check register pressure estimates. */ + } /* Continue with next node. */ + } /* While flush_and_start_over. */ if (ii >= maxii) { free_partial_schedule (ps); ps = NULL; } + else + gcc_assert (sbitmap_equal (tobe_scheduled, sched_nodes)); + + sbitmap_free (sched_nodes); + sbitmap_free (must_precede); + sbitmap_free (must_follow); + sbitmap_free (tobe_scheduled); + return ps; } +/* This function inserts a new empty row into PS at the position + according to SPLITROW, keeping all already scheduled instructions + intact and updating their SCHED_TIME and cycle accordingly. */ +static void +ps_insert_empty_row (partial_schedule_ptr ps, int split_row, + sbitmap sched_nodes) +{ + ps_insn_ptr crr_insn; + ps_insn_ptr *rows_new; + int ii = ps->ii; + int new_ii = ii + 1; + int row; + + verify_partial_schedule (ps, sched_nodes); + + /* We normalize sched_time and rotate ps to have only non-negative sched + times, for simplicity of updating cycles after inserting new row. */ + split_row -= ps->min_cycle; + split_row = SMODULO (split_row, ii); + if (dump_file) + fprintf (dump_file, "split_row=%d\n", split_row); + + normalize_sched_times (ps); + rotate_partial_schedule (ps, ps->min_cycle); + + rows_new = (ps_insn_ptr *) xcalloc (new_ii, sizeof (ps_insn_ptr)); + for (row = 0; row < split_row; row++) + { + rows_new[row] = ps->rows[row]; + ps->rows[row] = NULL; + for (crr_insn = rows_new[row]; + crr_insn; crr_insn = crr_insn->next_in_row) + { + ddg_node_ptr u = crr_insn->node; + int new_time = SCHED_TIME (u) + (SCHED_TIME (u) / ii); + + SCHED_TIME (u) = new_time; + crr_insn->cycle = new_time; + SCHED_ROW (u) = new_time % new_ii; + SCHED_STAGE (u) = new_time / new_ii; + } + + } + + rows_new[split_row] = NULL; + + for (row = split_row; row < ii; row++) + { + rows_new[row + 1] = ps->rows[row]; + ps->rows[row] = NULL; + for (crr_insn = rows_new[row + 1]; + crr_insn; crr_insn = crr_insn->next_in_row) + { + ddg_node_ptr u = crr_insn->node; + int new_time = SCHED_TIME (u) + (SCHED_TIME (u) / ii) + 1; + + SCHED_TIME (u) = new_time; + crr_insn->cycle = new_time; + SCHED_ROW (u) = new_time % new_ii; + SCHED_STAGE (u) = new_time / new_ii; + } + } + + /* Updating ps. */ + ps->min_cycle = ps->min_cycle + ps->min_cycle / ii + + (SMODULO (ps->min_cycle, ii) >= split_row ? 1 : 0); + ps->max_cycle = ps->max_cycle + ps->max_cycle / ii + + (SMODULO (ps->max_cycle, ii) >= split_row ? 1 : 0); + free (ps->rows); + ps->rows = rows_new; + ps->ii = new_ii; + gcc_assert (ps->min_cycle >= 0); + + verify_partial_schedule (ps, sched_nodes); + + if (dump_file) + fprintf (dump_file, "min_cycle=%d, max_cycle=%d\n", ps->min_cycle, + ps->max_cycle); +} + +/* Given U_NODE which is the node that failed to be scheduled; LOW and + UP which are the boundaries of it's scheduling window; compute using + SCHED_NODES and II a row in the partial schedule that can be split + which will separate a critical predecessor from a critical successor + thereby expanding the window, and return it. */ +static int +compute_split_row (sbitmap sched_nodes, int low, int up, int ii, + ddg_node_ptr u_node) +{ + ddg_edge_ptr e; + int lower = INT_MIN, upper = INT_MAX; + ddg_node_ptr crit_pred = NULL; + ddg_node_ptr crit_succ = NULL; + int crit_cycle; + + for (e = u_node->in; e != 0; e = e->next_in) + { + ddg_node_ptr v_node = e->src; + + if (TEST_BIT (sched_nodes, v_node->cuid) + && (low == SCHED_TIME (v_node) + e->latency - (e->distance * ii))) + if (SCHED_TIME (v_node) > lower) + { + crit_pred = v_node; + lower = SCHED_TIME (v_node); + } + } + + if (crit_pred != NULL) + { + crit_cycle = SCHED_TIME (crit_pred) + 1; + return SMODULO (crit_cycle, ii); + } + + for (e = u_node->out; e != 0; e = e->next_out) + { + ddg_node_ptr v_node = e->dest; + if (TEST_BIT (sched_nodes, v_node->cuid) + && (up == SCHED_TIME (v_node) - e->latency + (e->distance * ii))) + if (SCHED_TIME (v_node) < upper) + { + crit_succ = v_node; + upper = SCHED_TIME (v_node); + } + } + + if (crit_succ != NULL) + { + crit_cycle = SCHED_TIME (crit_succ); + return SMODULO (crit_cycle, ii); + } + + if (dump_file) + fprintf (dump_file, "Both crit_pred and crit_succ are NULL\n"); + + return SMODULO ((low + up + 1) / 2, ii); +} + +static void +verify_partial_schedule (partial_schedule_ptr ps, sbitmap sched_nodes) +{ + int row; + ps_insn_ptr crr_insn; + + for (row = 0; row < ps->ii; row++) + for (crr_insn = ps->rows[row]; crr_insn; crr_insn = crr_insn->next_in_row) + { + ddg_node_ptr u = crr_insn->node; + + gcc_assert (TEST_BIT (sched_nodes, u->cuid)); + /* ??? Test also that all nodes of sched_nodes are in ps, perhaps by + popcount (sched_nodes) == number of insns in ps. */ + gcc_assert (SCHED_TIME (u) >= ps->min_cycle); + gcc_assert (SCHED_TIME (u) <= ps->max_cycle); + } +} + /* This page implements the algorithm for ordering the nodes of a DDG for modulo scheduling, activated through the @@ -1577,7 +2044,7 @@ typedef struct node_order_params * nopa; static void order_nodes_of_sccs (ddg_all_sccs_ptr, int * result); static int order_nodes_in_scc (ddg_ptr, sbitmap, sbitmap, int*, int); -static nopa calculate_order_params (ddg_ptr, int mii); +static nopa calculate_order_params (ddg_ptr, int, int *); static int find_max_asap (ddg_ptr, sbitmap); static int find_max_hv_min_mob (ddg_ptr, sbitmap); static int find_max_dv_min_mob (ddg_ptr, sbitmap); @@ -1600,29 +2067,37 @@ check_nodes_order (int *node_order, int num_nodes) sbitmap_zero (tmp); + if (dump_file) + fprintf (dump_file, "SMS final nodes order: \n"); + for (i = 0; i < num_nodes; i++) { int u = node_order[i]; + if (dump_file) + fprintf (dump_file, "%d ", u); gcc_assert (u < num_nodes && u >= 0 && !TEST_BIT (tmp, u)); SET_BIT (tmp, u); } + if (dump_file) + fprintf (dump_file, "\n"); + sbitmap_free (tmp); } /* Order the nodes of G for scheduling and pass the result in NODE_ORDER. Also set aux.count of each node to ASAP. - Return the recMII for the given DDG. */ + Put maximal ASAP to PMAX_ASAP. Return the recMII for the given DDG. */ static int -sms_order_nodes (ddg_ptr g, int mii, int * node_order) +sms_order_nodes (ddg_ptr g, int mii, int * node_order, int *pmax_asap) { int i; int rec_mii = 0; ddg_all_sccs_ptr sccs = create_ddg_all_sccs (g); - nopa nops = calculate_order_params (g, mii); + nopa nops = calculate_order_params (g, mii, pmax_asap); if (dump_file) print_sccs (dump_file, sccs, g); @@ -1661,7 +2136,7 @@ order_nodes_of_sccs (ddg_all_sccs_ptr all_sccs, int * node_order) sbitmap_zero (prev_sccs); sbitmap_ones (ones); - /* Perfrom the node ordering starting from the SCC with the highest recMII. + /* Perform the node ordering starting from the SCC with the highest recMII. For each SCC order the nodes according to their ASAP/ALAP/HEIGHT etc. */ for (i = 0; i < all_sccs->num_sccs; i++) { @@ -1697,7 +2172,7 @@ order_nodes_of_sccs (ddg_all_sccs_ptr all_sccs, int * node_order) /* MII is needed if we consider backarcs (that do not close recursive cycles). */ static struct node_order_params * -calculate_order_params (ddg_ptr g, int mii ATTRIBUTE_UNUSED) +calculate_order_params (ddg_ptr g, int mii ATTRIBUTE_UNUSED, int *pmax_asap) { int u; int max_asap; @@ -1748,7 +2223,19 @@ calculate_order_params (ddg_ptr g, int mii ATTRIBUTE_UNUSED) HEIGHT (e->dest) + e->latency); } } + if (dump_file) + { + fprintf (dump_file, "\nOrder params\n"); + for (u = 0; u < num_nodes; u++) + { + ddg_node_ptr u_node = &g->nodes[u]; + fprintf (dump_file, "node %d, ASAP: %d, ALAP: %d, HEIGHT: %d\n", u, + ASAP (u_node), ALAP (u_node), HEIGHT (u_node)); + } + } + + *pmax_asap = max_asap; return node_order_params_arr; } @@ -2015,7 +2502,7 @@ print_partial_schedule (partial_schedule_ptr ps, FILE *dump) { ps_insn_ptr ps_i = ps->rows[i]; - fprintf (dump, "\n[CYCLE %d ]: ", i); + fprintf (dump, "\n[ROW %d ]: ", i); while (ps_i) { fprintf (dump, "%d, ", @@ -2098,10 +2585,10 @@ ps_insn_find_column (partial_schedule_ptr ps, ps_insn_ptr ps_i, next_ps_i; next_ps_i = next_ps_i->next_in_row) { - if (TEST_BIT (must_follow, next_ps_i->node->cuid) + if (must_follow && TEST_BIT (must_follow, next_ps_i->node->cuid) && ! first_must_follow) first_must_follow = next_ps_i; - if (TEST_BIT (must_precede, next_ps_i->node->cuid)) + if (must_precede && TEST_BIT (must_precede, next_ps_i->node->cuid)) { /* If we have already met a node that must follow, then there is no possible column. */ @@ -2135,8 +2622,8 @@ ps_insn_find_column (partial_schedule_ptr ps, ps_insn_ptr ps_i, } /* Advances the PS_INSN one column in its current row; returns false - in failure and true in success. Bit N is set in MUST_FOLLOW if - the node with cuid N must be come after the node pointed to by + in failure and true in success. Bit N is set in MUST_FOLLOW if + the node with cuid N must be come after the node pointed to by PS_I when scheduled in the same cycle. */ static int ps_insn_advance_column (partial_schedule_ptr ps, ps_insn_ptr ps_i, @@ -2158,7 +2645,7 @@ ps_insn_advance_column (partial_schedule_ptr ps, ps_insn_ptr ps_i, /* Check if next_in_row is dependent on ps_i, both having same sched times (typically ANTI_DEP). If so, ps_i cannot skip over it. */ - if (TEST_BIT (must_follow, next_node->cuid)) + if (must_follow && TEST_BIT (must_follow, next_node->cuid)) return false; /* Advance PS_I over its next_in_row in the doubly linked list. */ @@ -2184,9 +2671,9 @@ ps_insn_advance_column (partial_schedule_ptr ps, ps_insn_ptr ps_i, } /* Inserts a DDG_NODE to the given partial schedule at the given cycle. - Returns 0 if this is not possible and a PS_INSN otherwise. Bit N is - set in MUST_PRECEDE/MUST_FOLLOW if the node with cuid N must be come - before/after (respectively) the node pointed to by PS_I when scheduled + Returns 0 if this is not possible and a PS_INSN otherwise. Bit N is + set in MUST_PRECEDE/MUST_FOLLOW if the node with cuid N must be come + before/after (respectively) the node pointed to by PS_I when scheduled in the same cycle. */ static ps_insn_ptr add_node_to_ps (partial_schedule_ptr ps, ddg_node_ptr node, int cycle, @@ -2256,7 +2743,7 @@ ps_has_conflicts (partial_schedule_ptr ps, int from, int to) { rtx insn = crr_insn->node->insn; - if (!INSN_P (insn)) + if (!NONDEBUG_INSN_P (insn)) continue; /* Check if there is room for the current insn. */ @@ -2264,7 +2751,7 @@ ps_has_conflicts (partial_schedule_ptr ps, int from, int to) return true; /* Update the DFA state and return with failure if the DFA found - recource conflicts. */ + resource conflicts. */ if (state_transition (curr_state, insn) >= 0) return true; @@ -2287,8 +2774,8 @@ ps_has_conflicts (partial_schedule_ptr ps, int from, int to) /* Checks if the given node causes resource conflicts when added to PS at cycle C. If not the node is added to PS and returned; otherwise zero - is returned. Bit N is set in MUST_PRECEDE/MUST_FOLLOW if the node with - cuid N must be come before/after (respectively) the node pointed to by + is returned. Bit N is set in MUST_PRECEDE/MUST_FOLLOW if the node with + cuid N must be come before/after (respectively) the node pointed to by PS_I when scheduled in the same cycle. */ ps_insn_ptr ps_add_node_check_conflicts (partial_schedule_ptr ps, ddg_node_ptr n, @@ -2361,26 +2848,6 @@ rotate_partial_schedule (partial_schedule_ptr ps, int start_cycle) ps->min_cycle -= start_cycle; } -/* Remove the node N from the partial schedule PS; because we restart the DFA - each time we want to check for resource conflicts; this is equivalent to - unscheduling the node N. */ -static bool -ps_unschedule_node (partial_schedule_ptr ps, ddg_node_ptr n) -{ - ps_insn_ptr ps_i; - int row = SMODULO (SCHED_TIME (n), ps->ii); - - if (row < 0 || row > ps->ii) - return false; - - for (ps_i = ps->rows[row]; - ps_i && ps_i->node != n; - ps_i = ps_i->next_in_row); - if (!ps_i) - return false; - - return remove_node_from_ps (ps, ps_i); -} #endif /* INSN_SCHEDULING */ static bool @@ -2415,8 +2882,10 @@ rest_of_handle_sms (void) return 0; } -struct tree_opt_pass pass_sms = +struct rtl_opt_pass pass_sms = { + { + RTL_PASS, "sms", /* name */ gate_handle_sms, /* gate */ rest_of_handle_sms, /* execute */ @@ -2428,9 +2897,9 @@ struct tree_opt_pass pass_sms = 0, /* properties_provided */ 0, /* properties_destroyed */ TODO_dump_func, /* todo_flags_start */ - TODO_df_finish | + TODO_df_finish | TODO_verify_rtl_sharing | TODO_dump_func | - TODO_ggc_collect, /* todo_flags_finish */ - 'm' /* letter */ + TODO_ggc_collect /* todo_flags_finish */ + } };