X-Git-Url: http://git.sourceforge.jp/view?a=blobdiff_plain;f=gcc%2Fsched-rgn.c;h=b8e474d0b448b11bb7d83c4d03a85ec5afd274d8;hb=1732796ab54789ea1a9cf74cde208aa685a4548d;hp=7df09977bd519d67749f8d31b288b838fd549a5f;hpb=f660683a337c9a50c0949ed79f0d43eceb19fbeb;p=pf3gnuchains%2Fgcc-fork.git diff --git a/gcc/sched-rgn.c b/gcc/sched-rgn.c index 7df09977bd5..b8e474d0b44 100644 --- a/gcc/sched-rgn.c +++ b/gcc/sched-rgn.c @@ -1,6 +1,6 @@ /* Instruction scheduling pass. Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998, - 1999, 2000, 2001 Free Software Foundation, Inc. + 1999, 2000, 2001, 2002, 2003 Free Software Foundation, Inc. Contributed by Michael Tiemann (tiemann@cygnus.com) Enhanced by, and currently maintained by, Jim Wilson (wilson@cygnus.com) @@ -47,6 +47,8 @@ Software Foundation, 59 Temple Place - Suite 330, Boston, MA #include "config.h" #include "system.h" +#include "coretypes.h" +#include "tm.h" #include "toplev.h" #include "rtl.h" #include "tm_p.h" @@ -60,7 +62,20 @@ Software Foundation, 59 Temple Place - Suite 330, Boston, MA #include "except.h" #include "toplev.h" #include "recog.h" +#include "cfglayout.h" #include "sched-int.h" +#include "target.h" + +/* Define when we want to do count REG_DEAD notes before and after scheduling + for sanity checking. We can't do that when conditional execution is used, + as REG_DEAD exist only for unconditional deaths. */ + +#if !defined (HAVE_conditional_execution) && defined (ENABLE_CHECKING) +#define CHECK_DEAD_NOTES 1 +#else +#define CHECK_DEAD_NOTES 0 +#endif + #ifdef INSN_SCHEDULING /* Some accessor macros for h_i_d members only used within this file. */ @@ -101,9 +116,9 @@ static int *out_edges; #define IN_EDGES(block) (in_edges[block]) #define OUT_EDGES(block) (out_edges[block]) -static int is_cfg_nonregular PARAMS ((void)); -static int build_control_flow PARAMS ((struct edge_list *)); -static void new_edge PARAMS ((int, int)); +static int is_cfg_nonregular (void); +static int build_control_flow (struct edge_list *); +static void new_edge (int, int); /* A region is the main entity for interblock scheduling: insns are allowed to move between blocks in the same region, along @@ -127,7 +142,7 @@ static int *rgn_bb_table; /* Topological order of blocks in the region (if b2 is reachable from b1, block_to_bb[b2] > block_to_bb[b1]). Note: A basic block is always referred to by either block or b, while its topological - order name (in the region) is refered to by bb. */ + order name (in the region) is referred to by bb. */ static int *block_to_bb; /* The number of the region containing a block. */ @@ -138,12 +153,12 @@ static int *containing_rgn; #define BLOCK_TO_BB(block) (block_to_bb[block]) #define CONTAINING_RGN(block) (containing_rgn[block]) -void debug_regions PARAMS ((void)); -static void find_single_block_region PARAMS ((void)); -static void find_rgns PARAMS ((struct edge_list *, sbitmap *)); -static int too_large PARAMS ((int, int *, int *)); +void debug_regions (void); +static void find_single_block_region (void); +static void find_rgns (struct edge_list *); +static int too_large (int, int *, int *); -extern void debug_live PARAMS ((int, int)); +extern void debug_live (int, int); /* Blocks of the current region being scheduled. */ static int current_nr_blocks; @@ -152,10 +167,6 @@ static int current_blocks; /* The mapping from bb to block. */ #define BB_TO_BLOCK(bb) (rgn_bb_table[current_blocks + (bb)]) -/* Bit vectors and bitset operations are needed for computations on - the control flow graph. */ - -typedef unsigned HOST_WIDE_INT *bitset; typedef struct { int *first_member; /* Pointer to the list start in bitlst_table. */ @@ -164,11 +175,9 @@ typedef struct bitlst; static int bitlst_table_last; -static int bitlst_table_size; static int *bitlst_table; -static char bitset_member PARAMS ((bitset, int, int)); -static void extract_bitlst PARAMS ((bitset, int, int, bitlst *)); +static void extract_bitlst (sbitmap, bitlst *); /* Target info declarations. @@ -209,27 +218,21 @@ static int target_bb; typedef bitlst edgelst; /* Target info functions. */ -static void split_edges PARAMS ((int, int, edgelst *)); -static void compute_trg_info PARAMS ((int)); -void debug_candidate PARAMS ((int)); -void debug_candidates PARAMS ((int)); - -/* Bit-set of bbs, where bit 'i' stands for bb 'i'. */ -typedef bitset bbset; - -/* Number of words of the bbset. */ -static int bbset_size; +static void split_edges (int, int, edgelst *); +static void compute_trg_info (int); +void debug_candidate (int); +void debug_candidates (int); -/* Dominators array: dom[i] contains the bbset of dominators of +/* Dominators array: dom[i] contains the sbitmap of dominators of bb i in the region. */ -static bbset *dom; +static sbitmap *dom; /* bb 0 is the only region entry. */ #define IS_RGN_ENTRY(bb) (!bb) /* Is bb_src dominated by bb_trg. */ #define IS_DOMINATED(bb_src, bb_trg) \ -( bitset_member (dom[bb_src], bb_trg, bbset_size) ) +( TEST_BIT (dom[bb_src], bb_trg) ) /* Probability: Prob[i] is a float in [0, 1] which is the probability of bb i relative to the region entry. */ @@ -242,7 +245,7 @@ static float *prob; prob[bb_trg]))) /* Bit-set of edges, where bit i stands for edge i. */ -typedef bitset edgeset; +typedef sbitmap edgeset; /* Number of edges in the region. */ static int rgn_nr_edges; @@ -250,11 +253,6 @@ static int rgn_nr_edges; /* Array of size rgn_nr_edges. */ static int *rgn_edges; -/* Number of words in an edgeset. */ -static int edgeset_size; - -/* Number of bits in an edgeset. */ -static int edgeset_bitsize; /* Mapping from each edge in the graph to its number in the rgn. */ static int *edge_to_bit; @@ -271,41 +269,40 @@ static edgeset *pot_split; /* For every bb, a set of its ancestor edges. */ static edgeset *ancestor_edges; -static void compute_dom_prob_ps PARAMS ((int)); +static void compute_dom_prob_ps (int); -#define ABS_VALUE(x) (((x)<0)?(-(x)):(x)) #define INSN_PROBABILITY(INSN) (SRC_PROB (BLOCK_TO_BB (BLOCK_NUM (INSN)))) #define IS_SPECULATIVE_INSN(INSN) (IS_SPECULATIVE (BLOCK_TO_BB (BLOCK_NUM (INSN)))) #define INSN_BB(INSN) (BLOCK_TO_BB (BLOCK_NUM (INSN))) /* Parameters affecting the decision of rank_for_schedule(). - ??? Nope. But MIN_PROBABILITY is used in copmute_trg_info. */ -#define MIN_DIFF_PRIORITY 2 + ??? Nope. But MIN_PROBABILITY is used in compute_trg_info. */ #define MIN_PROBABILITY 40 -#define MIN_PROB_DIFF 10 /* Speculative scheduling functions. */ -static int check_live_1 PARAMS ((int, rtx)); -static void update_live_1 PARAMS ((int, rtx)); -static int check_live PARAMS ((rtx, int)); -static void update_live PARAMS ((rtx, int)); -static void set_spec_fed PARAMS ((rtx)); -static int is_pfree PARAMS ((rtx, int, int)); -static int find_conditional_protection PARAMS ((rtx, int)); -static int is_conditionally_protected PARAMS ((rtx, int, int)); -static int may_trap_exp PARAMS ((rtx, int)); -static int haifa_classify_insn PARAMS ((rtx)); -static int is_prisky PARAMS ((rtx, int, int)); -static int is_exception_free PARAMS ((rtx, int, int)); - -static void add_branch_dependences PARAMS ((rtx, rtx)); -static void compute_block_backward_dependences PARAMS ((int)); -void debug_dependencies PARAMS ((void)); - -static void init_regions PARAMS ((void)); -static void schedule_region PARAMS ((int)); -static void propagate_deps PARAMS ((int, struct deps *)); -static void free_pending_lists PARAMS ((void)); +static int check_live_1 (int, rtx); +static void update_live_1 (int, rtx); +static int check_live (rtx, int); +static void update_live (rtx, int); +static void set_spec_fed (rtx); +static int is_pfree (rtx, int, int); +static int find_conditional_protection (rtx, int); +static int is_conditionally_protected (rtx, int, int); +static int is_prisky (rtx, int, int); +static int is_exception_free (rtx, int, int); + +static bool sets_likely_spilled (rtx); +static void sets_likely_spilled_1 (rtx, rtx, void *); +static void add_branch_dependences (rtx, rtx); +static void compute_block_backward_dependences (int); +void debug_dependencies (void); + +static void init_regions (void); +static void schedule_region (int); +static rtx concat_INSN_LIST (rtx, rtx); +static void concat_insn_mem_list (rtx, rtx, rtx *, rtx *); +static void propagate_deps (int, struct deps *); +static void free_pending_lists (void); /* Functions for construction of the control flow graph. */ @@ -316,9 +313,9 @@ static void free_pending_lists PARAMS ((void)); have nonlocal gotos. */ static int -is_cfg_nonregular () +is_cfg_nonregular (void) { - int b; + basic_block b; rtx insn; RTX_CODE code; @@ -339,14 +336,14 @@ is_cfg_nonregular () /* If we have exception handlers, then we consider the cfg not well structured. ?!? We should be able to handle this now that flow.c computes an accurate cfg for EH. */ - if (exception_handler_labels) + if (current_function_has_exception_handlers ()) return 1; /* If we have non-jumping insns which refer to labels, then we consider the cfg not well structured. */ /* Check for labels referred to other thn by jumps. */ - for (b = 0; b < n_basic_blocks; b++) - for (insn = BLOCK_HEAD (b);; insn = NEXT_INSN (insn)) + FOR_EACH_BB (b) + for (insn = BB_HEAD (b); ; insn = NEXT_INSN (insn)) { code = GET_CODE (insn); if (GET_RTX_CLASS (code) == 'i' && code != JUMP_INSN) @@ -360,7 +357,7 @@ is_cfg_nonregular () return 1; } - if (insn == BLOCK_END (b)) + if (insn == BB_END (b)) break; } @@ -377,10 +374,10 @@ is_cfg_nonregular () prevent cross block scheduling. */ static int -build_control_flow (edge_list) - struct edge_list *edge_list; +build_control_flow (struct edge_list *edge_list) { int i, unreachable, num_edges; + basic_block b; /* This already accounts for entry/exit edges. */ num_edges = NUM_EDGES (edge_list); @@ -392,10 +389,8 @@ build_control_flow (edge_list) test is redundant with the one in find_rgns, but it's much cheaper to go ahead and catch the trivial case here. */ unreachable = 0; - for (i = 0; i < n_basic_blocks; i++) + FOR_EACH_BB (b) { - basic_block b = BASIC_BLOCK (i); - if (b->pred == NULL || (b->pred->src == b && b->pred->pred_next == NULL)) @@ -403,9 +398,9 @@ build_control_flow (edge_list) } /* ??? We can kill these soon. */ - in_edges = (int *) xcalloc (n_basic_blocks, sizeof (int)); - out_edges = (int *) xcalloc (n_basic_blocks, sizeof (int)); - edge_table = (haifa_edge *) xcalloc (num_edges, sizeof (haifa_edge)); + in_edges = xcalloc (last_basic_block, sizeof (int)); + out_edges = xcalloc (last_basic_block, sizeof (int)); + edge_table = xcalloc (num_edges, sizeof (haifa_edge)); nr_edges = 0; for (i = 0; i < num_edges; i++) @@ -430,8 +425,7 @@ build_control_flow (edge_list) integer lists. */ static void -new_edge (source, target) - int source, target; +new_edge (int source, int target) { int e, next_edge; int curr_edge, fst_edge; @@ -482,81 +476,12 @@ new_edge (source, target) } } -/* BITSET macros for operations on the control flow graph. */ - -/* Compute bitwise union of two bitsets. */ -#define BITSET_UNION(set1, set2, len) \ -do { register bitset tp = set1, sp = set2; \ - register int i; \ - for (i = 0; i < len; i++) \ - *(tp++) |= *(sp++); } while (0) - -/* Compute bitwise intersection of two bitsets. */ -#define BITSET_INTER(set1, set2, len) \ -do { register bitset tp = set1, sp = set2; \ - register int i; \ - for (i = 0; i < len; i++) \ - *(tp++) &= *(sp++); } while (0) - -/* Compute bitwise difference of two bitsets. */ -#define BITSET_DIFFER(set1, set2, len) \ -do { register bitset tp = set1, sp = set2; \ - register int i; \ - for (i = 0; i < len; i++) \ - *(tp++) &= ~*(sp++); } while (0) - -/* Inverts every bit of bitset 'set'. */ -#define BITSET_INVERT(set, len) \ -do { register bitset tmpset = set; \ - register int i; \ - for (i = 0; i < len; i++, tmpset++) \ - *tmpset = ~*tmpset; } while (0) - -/* Turn on the index'th bit in bitset set. */ -#define BITSET_ADD(set, index, len) \ -{ \ - if (index >= HOST_BITS_PER_WIDE_INT * len) \ - abort (); \ - else \ - set[index/HOST_BITS_PER_WIDE_INT] |= \ - ((unsigned HOST_WIDE_INT) 1) << (index % HOST_BITS_PER_WIDE_INT); \ -} - -/* Turn off the index'th bit in set. */ -#define BITSET_REMOVE(set, index, len) \ -{ \ - if (index >= HOST_BITS_PER_WIDE_INT * len) \ - abort (); \ - else \ - set[index/HOST_BITS_PER_WIDE_INT] &= \ - ~(((unsigned HOST_WIDE_INT) 1) << (index % HOST_BITS_PER_WIDE_INT)); \ -} - -/* Check if the index'th bit in bitset set is on. */ - -static char -bitset_member (set, index, len) - bitset set; - int index, len; -{ - if (index >= HOST_BITS_PER_WIDE_INT * len) - abort (); - return ((set[index / HOST_BITS_PER_WIDE_INT] & - ((unsigned HOST_WIDE_INT) 1) << (index % HOST_BITS_PER_WIDE_INT)) - ? 1 : 0); -} - /* Translate a bit-set SET to a list BL of the bit-set members. */ static void -extract_bitlst (set, len, bitlen, bl) - bitset set; - int len; - int bitlen; - bitlst *bl; +extract_bitlst (sbitmap set, bitlst *bl) { - int i, j, offset; - unsigned HOST_WIDE_INT word; + int i; /* bblst table space is reused in each call to extract_bitlst. */ bitlst_table_last = 0; @@ -565,24 +490,11 @@ extract_bitlst (set, len, bitlen, bl) bl->nr_members = 0; /* Iterate over each word in the bitset. */ - for (i = 0; i < len; i++) - { - word = set[i]; - offset = i * HOST_BITS_PER_WIDE_INT; - - /* Iterate over each bit in the word, but do not - go beyond the end of the defined bits. */ - for (j = 0; offset < bitlen && word; j++) - { - if (word & 1) - { - bitlst_table[bitlst_table_last++] = offset; - (bl->nr_members)++; - } - word >>= 1; - ++offset; - } - } + EXECUTE_IF_SET_IN_SBITMAP (set, 0, i, + { + bitlst_table[bitlst_table_last++] = i; + (bl->nr_members)++; + }); } @@ -591,7 +503,7 @@ extract_bitlst (set, len, bitlen, bl) /* Print the regions, for debugging purposes. Callable from debugger. */ void -debug_regions () +debug_regions (void) { int rgn, bb; @@ -621,19 +533,21 @@ debug_regions () scheduling. */ static void -find_single_block_region () +find_single_block_region (void) { - int i; + basic_block bb; + + nr_regions = 0; - for (i = 0; i < n_basic_blocks; i++) + FOR_EACH_BB (bb) { - rgn_bb_table[i] = i; - RGN_NR_BLOCKS (i) = 1; - RGN_BLOCKS (i) = i; - CONTAINING_RGN (i) = i; - BLOCK_TO_BB (i) = 0; + rgn_bb_table[nr_regions] = bb->index; + RGN_NR_BLOCKS (nr_regions) = 1; + RGN_BLOCKS (nr_regions) = nr_regions; + CONTAINING_RGN (bb->index) = nr_regions; + BLOCK_TO_BB (bb->index) = 0; + nr_regions++; } - nr_regions = n_basic_blocks; } /* Update number of blocks and the estimate for number of insns @@ -641,12 +555,11 @@ find_single_block_region () scheduling (compile time considerations), otherwise return 0. */ static int -too_large (block, num_bbs, num_insns) - int block, *num_bbs, *num_insns; +too_large (int block, int *num_bbs, int *num_insns) { (*num_bbs)++; - (*num_insns) += (INSN_LUID (BLOCK_END (block)) - - INSN_LUID (BLOCK_HEAD (block))); + (*num_insns) += (INSN_LUID (BB_END (BASIC_BLOCK (block))) - + INSN_LUID (BB_HEAD (BASIC_BLOCK (block)))); if ((*num_bbs > MAX_RGN_BLOCKS) || (*num_insns > MAX_RGN_INSNS)) return 1; else @@ -656,17 +569,17 @@ too_large (block, num_bbs, num_insns) /* Update_loop_relations(blk, hdr): Check if the loop headed by max_hdr[blk] is still an inner loop. Put in max_hdr[blk] the header of the most inner loop containing blk. */ -#define UPDATE_LOOP_RELATIONS(blk, hdr) \ -{ \ - if (max_hdr[blk] == -1) \ - max_hdr[blk] = hdr; \ - else if (dfs_nr[max_hdr[blk]] > dfs_nr[hdr]) \ - RESET_BIT (inner, hdr); \ - else if (dfs_nr[max_hdr[blk]] < dfs_nr[hdr]) \ - { \ - RESET_BIT (inner,max_hdr[blk]); \ - max_hdr[blk] = hdr; \ - } \ +#define UPDATE_LOOP_RELATIONS(blk, hdr) \ +{ \ + if (max_hdr[blk] == -1) \ + max_hdr[blk] = hdr; \ + else if (dfs_nr[max_hdr[blk]] > dfs_nr[hdr]) \ + RESET_BIT (inner, hdr); \ + else if (dfs_nr[max_hdr[blk]] < dfs_nr[hdr]) \ + { \ + RESET_BIT (inner,max_hdr[blk]); \ + max_hdr[blk] = hdr; \ + } \ } /* Find regions for interblock scheduling. @@ -700,16 +613,16 @@ too_large (block, num_bbs, num_insns) of edge tables. That would simplify it somewhat. */ static void -find_rgns (edge_list, dom) - struct edge_list *edge_list; - sbitmap *dom; +find_rgns (struct edge_list *edge_list) { int *max_hdr, *dfs_nr, *stack, *degree; char no_loops = 1; int node, child, loop_head, i, head, tail; - int count = 0, sp, idx = 0, current_edge = out_edges[0]; + int count = 0, sp, idx = 0; + int current_edge = out_edges[ENTRY_BLOCK_PTR->succ->dest->index]; int num_bbs, num_insns, unreachable; int too_large_failure; + basic_block bb; /* Note if an edge has been passed. */ sbitmap passed; @@ -717,7 +630,7 @@ find_rgns (edge_list, dom) /* Note if a block is a natural loop header. */ sbitmap header; - /* Note if a block is an natural inner loop header. */ + /* Note if a block is a natural inner loop header. */ sbitmap inner; /* Note if a block is in the block queue. */ @@ -738,26 +651,26 @@ find_rgns (edge_list, dom) STACK, SP and DFS_NR are only used during the first traversal. */ /* Allocate and initialize variables for the first traversal. */ - max_hdr = (int *) xmalloc (n_basic_blocks * sizeof (int)); - dfs_nr = (int *) xcalloc (n_basic_blocks, sizeof (int)); - stack = (int *) xmalloc (nr_edges * sizeof (int)); + max_hdr = xmalloc (last_basic_block * sizeof (int)); + dfs_nr = xcalloc (last_basic_block, sizeof (int)); + stack = xmalloc (nr_edges * sizeof (int)); - inner = sbitmap_alloc (n_basic_blocks); + inner = sbitmap_alloc (last_basic_block); sbitmap_ones (inner); - header = sbitmap_alloc (n_basic_blocks); + header = sbitmap_alloc (last_basic_block); sbitmap_zero (header); passed = sbitmap_alloc (nr_edges); sbitmap_zero (passed); - in_queue = sbitmap_alloc (n_basic_blocks); + in_queue = sbitmap_alloc (last_basic_block); sbitmap_zero (in_queue); - in_stack = sbitmap_alloc (n_basic_blocks); + in_stack = sbitmap_alloc (last_basic_block); sbitmap_zero (in_stack); - for (i = 0; i < n_basic_blocks; i++) + for (i = 0; i < last_basic_block; i++) max_hdr[i] = -1; /* DFS traversal to find inner loops in the cfg. */ @@ -851,8 +764,8 @@ find_rgns (edge_list, dom) the entry node by placing a nonzero value in dfs_nr. Thus if dfs_nr is zero for any block, then it must be unreachable. */ unreachable = 0; - for (i = 0; i < n_basic_blocks; i++) - if (dfs_nr[i] == 0) + FOR_EACH_BB (bb) + if (dfs_nr[bb->index] == 0) { unreachable = 1; break; @@ -862,8 +775,8 @@ find_rgns (edge_list, dom) to hold degree counts. */ degree = dfs_nr; - for (i = 0; i < n_basic_blocks; i++) - degree[i] = 0; + FOR_EACH_BB (bb) + degree[bb->index] = 0; for (i = 0; i < num_edges; i++) { edge e = INDEX_EDGE (edge_list, i); @@ -881,19 +794,19 @@ find_rgns (edge_list, dom) if (no_loops) SET_BIT (header, 0); - /* Second travsersal:find reducible inner loops and topologically sort + /* Second traversal:find reducible inner loops and topologically sort block of each region. */ - queue = (int *) xmalloc (n_basic_blocks * sizeof (int)); + queue = xmalloc (n_basic_blocks * sizeof (int)); /* Find blocks which are inner loop headers. We still have non-reducible loops to consider at this point. */ - for (i = 0; i < n_basic_blocks; i++) + FOR_EACH_BB (bb) { - if (TEST_BIT (header, i) && TEST_BIT (inner, i)) + if (TEST_BIT (header, bb->index) && TEST_BIT (inner, bb->index)) { edge e; - int j; + basic_block jbb; /* Now check that the loop is reducible. We do this separate from finding inner loops so that we do not find a reducible @@ -906,15 +819,15 @@ find_rgns (edge_list, dom) If there exists a block that is not dominated by the loop header, then the block is reachable from outside the loop and thus the loop is not a natural loop. */ - for (j = 0; j < n_basic_blocks; j++) + FOR_EACH_BB (jbb) { /* First identify blocks in the loop, except for the loop entry block. */ - if (i == max_hdr[j] && i != j) + if (bb->index == max_hdr[jbb->index] && bb != jbb) { /* Now verify that the block is dominated by the loop header. */ - if (!TEST_BIT (dom[j], i)) + if (!dominated_by_p (CDI_DOMINATORS, jbb, bb)) break; } } @@ -922,25 +835,25 @@ find_rgns (edge_list, dom) /* If we exited the loop early, then I is the header of a non-reducible loop and we should quit processing it now. */ - if (j != n_basic_blocks) + if (jbb != EXIT_BLOCK_PTR) continue; /* I is a header of an inner loop, or block 0 in a subroutine with no loops at all. */ head = tail = -1; too_large_failure = 0; - loop_head = max_hdr[i]; + loop_head = max_hdr[bb->index]; /* Decrease degree of all I's successors for topological ordering. */ - for (e = BASIC_BLOCK (i)->succ; e; e = e->succ_next) + for (e = bb->succ; e; e = e->succ_next) if (e->dest != EXIT_BLOCK_PTR) --degree[e->dest->index]; /* Estimate # insns, and count # blocks in the region. */ num_bbs = 1; - num_insns = (INSN_LUID (BLOCK_END (i)) - - INSN_LUID (BLOCK_HEAD (i))); + num_insns = (INSN_LUID (BB_END (bb)) + - INSN_LUID (BB_HEAD (bb))); /* Find all loop latches (blocks with back edges to the loop header) or all the leaf blocks in the cfg has no loops. @@ -948,17 +861,17 @@ find_rgns (edge_list, dom) Place those blocks into the queue. */ if (no_loops) { - for (j = 0; j < n_basic_blocks; j++) + FOR_EACH_BB (jbb) /* Leaf nodes have only a single successor which must be EXIT_BLOCK. */ - if (BASIC_BLOCK (j)->succ - && BASIC_BLOCK (j)->succ->dest == EXIT_BLOCK_PTR - && BASIC_BLOCK (j)->succ->succ_next == NULL) + if (jbb->succ + && jbb->succ->dest == EXIT_BLOCK_PTR + && jbb->succ->succ_next == NULL) { - queue[++tail] = j; - SET_BIT (in_queue, j); + queue[++tail] = jbb->index; + SET_BIT (in_queue, jbb->index); - if (too_large (j, &num_bbs, &num_insns)) + if (too_large (jbb->index, &num_bbs, &num_insns)) { too_large_failure = 1; break; @@ -969,14 +882,14 @@ find_rgns (edge_list, dom) { edge e; - for (e = BASIC_BLOCK (i)->pred; e; e = e->pred_next) + for (e = bb->pred; e; e = e->pred_next) { if (e->src == ENTRY_BLOCK_PTR) continue; node = e->src->index; - if (max_hdr[node] == loop_head && node != i) + if (max_hdr[node] == loop_head && node != bb->index) { /* This is a loop latch. */ queue[++tail] = node; @@ -1038,7 +951,7 @@ find_rgns (edge_list, dom) tail = -1; break; } - else if (!TEST_BIT (in_queue, node) && node != i) + else if (!TEST_BIT (in_queue, node) && node != bb->index) { queue[++tail] = node; SET_BIT (in_queue, node); @@ -1055,12 +968,12 @@ find_rgns (edge_list, dom) if (tail >= 0 && !too_large_failure) { /* Place the loop header into list of region blocks. */ - degree[i] = -1; - rgn_bb_table[idx] = i; + degree[bb->index] = -1; + rgn_bb_table[idx] = bb->index; RGN_NR_BLOCKS (nr_regions) = num_bbs; RGN_BLOCKS (nr_regions) = idx++; - CONTAINING_RGN (i) = nr_regions; - BLOCK_TO_BB (i) = count = 0; + CONTAINING_RGN (bb->index) = nr_regions; + BLOCK_TO_BB (bb->index) = count = 0; /* Remove blocks from queue[] when their in degree becomes zero. Repeat until no blocks are left on the @@ -1099,24 +1012,24 @@ find_rgns (edge_list, dom) /* Any block that did not end up in a region is placed into a region by itself. */ - for (i = 0; i < n_basic_blocks; i++) - if (degree[i] >= 0) + FOR_EACH_BB (bb) + if (degree[bb->index] >= 0) { - rgn_bb_table[idx] = i; + rgn_bb_table[idx] = bb->index; RGN_NR_BLOCKS (nr_regions) = 1; RGN_BLOCKS (nr_regions) = idx++; - CONTAINING_RGN (i) = nr_regions++; - BLOCK_TO_BB (i) = 0; + CONTAINING_RGN (bb->index) = nr_regions++; + BLOCK_TO_BB (bb->index) = 0; } free (max_hdr); free (dfs_nr); free (stack); - free (passed); - free (header); - free (inner); - free (in_queue); - free (in_stack); + sbitmap_free (passed); + sbitmap_free (header); + sbitmap_free (inner); + sbitmap_free (in_queue); + sbitmap_free (in_stack); } /* Functions for regions scheduling information. */ @@ -1125,8 +1038,7 @@ find_rgns (edge_list, dom) Assume that these values were already computed for bb's predecessors. */ static void -compute_dom_prob_ps (bb) - int bb; +compute_dom_prob_ps (int bb) { int nxt_in_edge, fst_in_edge, pred; int fst_out_edge, nxt_out_edge, nr_out_edges, nr_rgn_out_edges; @@ -1134,34 +1046,32 @@ compute_dom_prob_ps (bb) prob[bb] = 0.0; if (IS_RGN_ENTRY (bb)) { - BITSET_ADD (dom[bb], 0, bbset_size); + SET_BIT (dom[bb], 0); prob[bb] = 1.0; return; } fst_in_edge = nxt_in_edge = IN_EDGES (BB_TO_BLOCK (bb)); - /* Intialize dom[bb] to '111..1'. */ - BITSET_INVERT (dom[bb], bbset_size); + /* Initialize dom[bb] to '111..1'. */ + sbitmap_ones (dom[bb]); do { pred = FROM_BLOCK (nxt_in_edge); - BITSET_INTER (dom[bb], dom[BLOCK_TO_BB (pred)], bbset_size); - - BITSET_UNION (ancestor_edges[bb], ancestor_edges[BLOCK_TO_BB (pred)], - edgeset_size); + sbitmap_a_and_b (dom[bb], dom[bb], dom[BLOCK_TO_BB (pred)]); + sbitmap_a_or_b (ancestor_edges[bb], ancestor_edges[bb], ancestor_edges[BLOCK_TO_BB (pred)]); - BITSET_ADD (ancestor_edges[bb], EDGE_TO_BIT (nxt_in_edge), edgeset_size); + SET_BIT (ancestor_edges[bb], EDGE_TO_BIT (nxt_in_edge)); nr_out_edges = 1; nr_rgn_out_edges = 0; fst_out_edge = OUT_EDGES (pred); nxt_out_edge = NEXT_OUT (fst_out_edge); - BITSET_UNION (pot_split[bb], pot_split[BLOCK_TO_BB (pred)], - edgeset_size); - BITSET_ADD (pot_split[bb], EDGE_TO_BIT (fst_out_edge), edgeset_size); + sbitmap_a_or_b (pot_split[bb], pot_split[bb], pot_split[BLOCK_TO_BB (pred)]); + + SET_BIT (pot_split[bb], EDGE_TO_BIT (fst_out_edge)); /* The successor doesn't belong in the region? */ if (CONTAINING_RGN (TO_BLOCK (fst_out_edge)) != @@ -1175,7 +1085,7 @@ compute_dom_prob_ps (bb) if (CONTAINING_RGN (TO_BLOCK (nxt_out_edge)) != CONTAINING_RGN (BB_TO_BLOCK (bb))) ++nr_rgn_out_edges; - BITSET_ADD (pot_split[bb], EDGE_TO_BIT (nxt_out_edge), edgeset_size); + SET_BIT (pot_split[bb], EDGE_TO_BIT (nxt_out_edge)); nxt_out_edge = NEXT_OUT (nxt_out_edge); } @@ -1192,8 +1102,8 @@ compute_dom_prob_ps (bb) } while (fst_in_edge != nxt_in_edge); - BITSET_ADD (dom[bb], bb, bbset_size); - BITSET_DIFFER (pot_split[bb], ancestor_edges[bb], edgeset_size); + SET_BIT (dom[bb], bb); + sbitmap_difference (pot_split[bb], pot_split[bb], ancestor_edges[bb]); if (sched_verbose >= 2) fprintf (sched_dump, ";; bb_prob(%d, %d) = %3d\n", bb, BB_TO_BLOCK (bb), @@ -1206,19 +1116,14 @@ compute_dom_prob_ps (bb) Note that bb_trg dominates bb_src. */ static void -split_edges (bb_src, bb_trg, bl) - int bb_src; - int bb_trg; - edgelst *bl; +split_edges (int bb_src, int bb_trg, edgelst *bl) { - int es = edgeset_size; - edgeset src = (edgeset) xcalloc (es, sizeof (HOST_WIDE_INT)); - - while (es--) - src[es] = (pot_split[bb_src])[es]; - BITSET_DIFFER (src, pot_split[bb_trg], edgeset_size); - extract_bitlst (src, edgeset_size, edgeset_bitsize, bl); - free (src); + sbitmap src = sbitmap_alloc (pot_split[bb_src]->n_bits); + sbitmap_copy (src, pot_split[bb_src]); + + sbitmap_difference (src, src, pot_split[bb_trg]); + extract_bitlst (src, bl); + sbitmap_free (src); } /* Find the valid candidate-source-blocks for the target block TRG, compute @@ -1226,10 +1131,9 @@ split_edges (bb_src, bb_trg, bl) For speculative sources, compute their update-blocks and split-blocks. */ static void -compute_trg_info (trg) - int trg; +compute_trg_info (int trg) { - register candidate *sp; + candidate *sp; edgelst el; int check_block, update_idx; int i, j, k, fst_edge, nxt_edge; @@ -1278,8 +1182,8 @@ compute_trg_info (trg) add the TO block to the update block list. This list can end up with a lot of duplicates. We need to weed them out to avoid overrunning the end of the bblst_table. */ - update_blocks = (char *) alloca (n_basic_blocks); - memset (update_blocks, 0, n_basic_blocks); + update_blocks = alloca (last_basic_block); + memset (update_blocks, 0, last_basic_block); update_idx = 0; for (j = 0; j < el.nr_members; j++) @@ -1325,8 +1229,7 @@ compute_trg_info (trg) /* Print candidates info, for debugging purposes. Callable from debugger. */ void -debug_candidate (i) - int i; +debug_candidate (int i) { if (!candidate_table[i].is_valid) return; @@ -1363,8 +1266,7 @@ debug_candidate (i) /* Print candidates info, for debugging purposes. Callable from debugger. */ void -debug_candidates (trg) - int trg; +debug_candidates (int trg) { int i; @@ -1374,19 +1276,17 @@ debug_candidates (trg) debug_candidate (i); } -/* Functions for speculative scheduing. */ +/* Functions for speculative scheduling. */ /* Return 0 if x is a set of a register alive in the beginning of one of the split-blocks of src, otherwise return 1. */ static int -check_live_1 (src, x) - int src; - rtx x; +check_live_1 (int src, rtx x) { - register int i; - register int regno; - register rtx reg = SET_DEST (x); + int i; + int regno; + rtx reg = SET_DEST (x); if (reg == 0) return 1; @@ -1398,7 +1298,7 @@ check_live_1 (src, x) if (GET_CODE (reg) == PARALLEL) { - register int i; + int i; for (i = XVECLEN (reg, 0) - 1; i >= 0; i--) if (XEXP (XVECEXP (reg, 0, i), 0) != 0) @@ -1440,7 +1340,7 @@ check_live_1 (src, x) } else { - /* Check for psuedo registers. */ + /* Check for pseudo registers. */ for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++) { int b = candidate_table[src].split_bbs.first_member[i]; @@ -1460,13 +1360,11 @@ check_live_1 (src, x) of every update-block of src. */ static void -update_live_1 (src, x) - int src; - rtx x; +update_live_1 (int src, rtx x) { - register int i; - register int regno; - register rtx reg = SET_DEST (x); + int i; + int regno; + rtx reg = SET_DEST (x); if (reg == 0) return; @@ -1478,7 +1376,7 @@ update_live_1 (src, x) if (GET_CODE (reg) == PARALLEL) { - register int i; + int i; for (i = XVECLEN (reg, 0) - 1; i >= 0; i--) if (XEXP (XVECEXP (reg, 0, i), 0) != 0) @@ -1528,9 +1426,7 @@ update_live_1 (src, x) ready-list or before the scheduling. */ static int -check_live (insn, src) - rtx insn; - int src; +check_live (rtx insn, int src) { /* Find the registers set by instruction. */ if (GET_CODE (PATTERN (insn)) == SET @@ -1555,9 +1451,7 @@ check_live (insn, src) block src to trg. */ static void -update_live (insn, src) - rtx insn; - int src; +update_live (rtx insn, int src) { /* Find the registers set by instruction. */ if (GET_CODE (PATTERN (insn)) == SET @@ -1573,97 +1467,17 @@ update_live (insn, src) } } -/* Exception Free Loads: - - We define five classes of speculative loads: IFREE, IRISKY, - PFREE, PRISKY, and MFREE. - - IFREE loads are loads that are proved to be exception-free, just - by examining the load insn. Examples for such loads are loads - from TOC and loads of global data. - - IRISKY loads are loads that are proved to be exception-risky, - just by examining the load insn. Examples for such loads are - volatile loads and loads from shared memory. - - PFREE loads are loads for which we can prove, by examining other - insns, that they are exception-free. Currently, this class consists - of loads for which we are able to find a "similar load", either in - the target block, or, if only one split-block exists, in that split - block. Load2 is similar to load1 if both have same single base - register. We identify only part of the similar loads, by finding - an insn upon which both load1 and load2 have a DEF-USE dependence. - - PRISKY loads are loads for which we can prove, by examining other - insns, that they are exception-risky. Currently we have two proofs for - such loads. The first proof detects loads that are probably guarded by a - test on the memory address. This proof is based on the - backward and forward data dependence information for the region. - Let load-insn be the examined load. - Load-insn is PRISKY iff ALL the following hold: - - - insn1 is not in the same block as load-insn - - there is a DEF-USE dependence chain (insn1, ..., load-insn) - - test-insn is either a compare or a branch, not in the same block - as load-insn - - load-insn is reachable from test-insn - - there is a DEF-USE dependence chain (insn1, ..., test-insn) - - This proof might fail when the compare and the load are fed - by an insn not in the region. To solve this, we will add to this - group all loads that have no input DEF-USE dependence. - - The second proof detects loads that are directly or indirectly - fed by a speculative load. This proof is affected by the - scheduling process. We will use the flag fed_by_spec_load. - Initially, all insns have this flag reset. After a speculative - motion of an insn, if insn is either a load, or marked as - fed_by_spec_load, we will also mark as fed_by_spec_load every - insn1 for which a DEF-USE dependence (insn, insn1) exists. A - load which is fed_by_spec_load is also PRISKY. - - MFREE (maybe-free) loads are all the remaining loads. They may be - exception-free, but we cannot prove it. - - Now, all loads in IFREE and PFREE classes are considered - exception-free, while all loads in IRISKY and PRISKY classes are - considered exception-risky. As for loads in the MFREE class, - these are considered either exception-free or exception-risky, - depending on whether we are pessimistic or optimistic. We have - to take the pessimistic approach to assure the safety of - speculative scheduling, but we can take the optimistic approach - by invoking the -fsched_spec_load_dangerous option. */ - -enum INSN_TRAP_CLASS -{ - TRAP_FREE = 0, IFREE = 1, PFREE_CANDIDATE = 2, - PRISKY_CANDIDATE = 3, IRISKY = 4, TRAP_RISKY = 5 -}; - -#define WORST_CLASS(class1, class2) \ -((class1 > class2) ? class1 : class2) - -/* Non-zero if block bb_to is equal to, or reachable from block bb_from. */ +/* Nonzero if block bb_to is equal to, or reachable from block bb_from. */ #define IS_REACHABLE(bb_from, bb_to) \ -(bb_from == bb_to \ + (bb_from == bb_to \ || IS_RGN_ENTRY (bb_from) \ - || (bitset_member (ancestor_edges[bb_to], \ - EDGE_TO_BIT (IN_EDGES (BB_TO_BLOCK (bb_from))), \ - edgeset_size))) - -/* Non-zero iff the address is comprised from at most 1 register. */ -#define CONST_BASED_ADDRESS_P(x) \ - (GET_CODE (x) == REG \ - || ((GET_CODE (x) == PLUS || GET_CODE (x) == MINUS \ - || (GET_CODE (x) == LO_SUM)) \ - && (GET_CODE (XEXP (x, 0)) == CONST_INT \ - || GET_CODE (XEXP (x, 1)) == CONST_INT))) + || (TEST_BIT (ancestor_edges[bb_to], \ + EDGE_TO_BIT (IN_EDGES (BB_TO_BLOCK (bb_from)))))) /* Turns on the fed_by_spec_load flag for insns fed by load_insn. */ static void -set_spec_fed (load_insn) - rtx load_insn; +set_spec_fed (rtx load_insn) { rtx link; @@ -1676,9 +1490,7 @@ set_spec_fed (load_insn) branch depending on insn, that guards the speculative load. */ static int -find_conditional_protection (insn, load_insn_bb) - rtx insn; - int load_insn_bb; +find_conditional_protection (rtx insn, int load_insn_bb) { rtx link; @@ -1713,9 +1525,7 @@ find_conditional_protection (insn, load_insn_bb) Locate the branch by following INSN_DEPEND from insn1. */ static int -is_conditionally_protected (load_insn, bb_src, bb_trg) - rtx load_insn; - int bb_src, bb_trg; +is_conditionally_protected (rtx load_insn, int bb_src, int bb_trg) { rtx link; @@ -1765,12 +1575,10 @@ is_conditionally_protected (load_insn, bb_src, bb_trg) load2 anyhow. */ static int -is_pfree (load_insn, bb_src, bb_trg) - rtx load_insn; - int bb_src, bb_trg; +is_pfree (rtx load_insn, int bb_src, int bb_trg) { rtx back_link; - register candidate *candp = candidate_table + bb_src; + candidate *candp = candidate_table + bb_src; if (candp->split_bbs.nr_members != 1) /* Must have exactly one escape block. */ @@ -1813,168 +1621,12 @@ is_pfree (load_insn, bb_src, bb_trg) return 0; } /* is_pfree */ -/* Returns a class that insn with GET_DEST(insn)=x may belong to, - as found by analyzing insn's expression. */ - -static int -may_trap_exp (x, is_store) - rtx x; - int is_store; -{ - enum rtx_code code; - - if (x == 0) - return TRAP_FREE; - code = GET_CODE (x); - if (is_store) - { - if (code == MEM) - return TRAP_RISKY; - else - return TRAP_FREE; - } - if (code == MEM) - { - /* The insn uses memory: a volatile load. */ - if (MEM_VOLATILE_P (x)) - return IRISKY; - /* An exception-free load. */ - if (!may_trap_p (x)) - return IFREE; - /* A load with 1 base register, to be further checked. */ - if (CONST_BASED_ADDRESS_P (XEXP (x, 0))) - return PFREE_CANDIDATE; - /* No info on the load, to be further checked. */ - return PRISKY_CANDIDATE; - } - else - { - const char *fmt; - int i, insn_class = TRAP_FREE; - - /* Neither store nor load, check if it may cause a trap. */ - if (may_trap_p (x)) - return TRAP_RISKY; - /* Recursive step: walk the insn... */ - fmt = GET_RTX_FORMAT (code); - for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--) - { - if (fmt[i] == 'e') - { - int tmp_class = may_trap_exp (XEXP (x, i), is_store); - insn_class = WORST_CLASS (insn_class, tmp_class); - } - else if (fmt[i] == 'E') - { - int j; - for (j = 0; j < XVECLEN (x, i); j++) - { - int tmp_class = may_trap_exp (XVECEXP (x, i, j), is_store); - insn_class = WORST_CLASS (insn_class, tmp_class); - if (insn_class == TRAP_RISKY || insn_class == IRISKY) - break; - } - } - if (insn_class == TRAP_RISKY || insn_class == IRISKY) - break; - } - return insn_class; - } -} - -/* Classifies insn for the purpose of verifying that it can be - moved speculatively, by examining it's patterns, returning: - TRAP_RISKY: store, or risky non-load insn (e.g. division by variable). - TRAP_FREE: non-load insn. - IFREE: load from a globaly safe location. - IRISKY: volatile load. - PFREE_CANDIDATE, PRISKY_CANDIDATE: load that need to be checked for - being either PFREE or PRISKY. */ - -static int -haifa_classify_insn (insn) - rtx insn; -{ - rtx pat = PATTERN (insn); - int tmp_class = TRAP_FREE; - int insn_class = TRAP_FREE; - enum rtx_code code; - - if (GET_CODE (pat) == PARALLEL) - { - int i, len = XVECLEN (pat, 0); - - for (i = len - 1; i >= 0; i--) - { - code = GET_CODE (XVECEXP (pat, 0, i)); - switch (code) - { - case CLOBBER: - /* Test if it is a 'store'. */ - tmp_class = may_trap_exp (XEXP (XVECEXP (pat, 0, i), 0), 1); - break; - case SET: - /* Test if it is a store. */ - tmp_class = may_trap_exp (SET_DEST (XVECEXP (pat, 0, i)), 1); - if (tmp_class == TRAP_RISKY) - break; - /* Test if it is a load. */ - tmp_class - = WORST_CLASS (tmp_class, - may_trap_exp (SET_SRC (XVECEXP (pat, 0, i)), - 0)); - break; - case COND_EXEC: - case TRAP_IF: - tmp_class = TRAP_RISKY; - break; - default: - ; - } - insn_class = WORST_CLASS (insn_class, tmp_class); - if (insn_class == TRAP_RISKY || insn_class == IRISKY) - break; - } - } - else - { - code = GET_CODE (pat); - switch (code) - { - case CLOBBER: - /* Test if it is a 'store'. */ - tmp_class = may_trap_exp (XEXP (pat, 0), 1); - break; - case SET: - /* Test if it is a store. */ - tmp_class = may_trap_exp (SET_DEST (pat), 1); - if (tmp_class == TRAP_RISKY) - break; - /* Test if it is a load. */ - tmp_class = - WORST_CLASS (tmp_class, - may_trap_exp (SET_SRC (pat), 0)); - break; - case COND_EXEC: - case TRAP_IF: - tmp_class = TRAP_RISKY; - break; - default:; - } - insn_class = tmp_class; - } - - return insn_class; -} - /* Return 1 if load_insn is prisky (i.e. if load_insn is fed by a load moved speculatively, or if load_insn is protected by a compare on load_insn's address). */ static int -is_prisky (load_insn, bb_src, bb_trg) - rtx load_insn; - int bb_src, bb_trg; +is_prisky (rtx load_insn, int bb_src, int bb_trg) { if (FED_BY_SPEC_LOAD (load_insn)) return 1; @@ -1994,9 +1646,7 @@ is_prisky (load_insn, bb_src, bb_trg) and 0 otherwise. */ static int -is_exception_free (insn, bb_src, bb_trg) - rtx insn; - int bb_src, bb_trg; +is_exception_free (rtx insn, int bb_src, int bb_trg) { int insn_class = haifa_classify_insn (insn); @@ -2045,19 +1695,19 @@ static int sched_n_insns; static int last_was_jump; /* Implementations of the sched_info functions for region scheduling. */ -static void init_ready_list PARAMS ((struct ready_list *)); -static int can_schedule_ready_p PARAMS ((rtx)); -static int new_ready PARAMS ((rtx)); -static int schedule_more_p PARAMS ((void)); -static const char *rgn_print_insn PARAMS ((rtx, int)); -static int rgn_rank PARAMS ((rtx, rtx)); -static int contributes_to_priority PARAMS ((rtx, rtx)); -static void compute_jump_reg_dependencies PARAMS ((rtx, regset)); +static void init_ready_list (struct ready_list *); +static int can_schedule_ready_p (rtx); +static int new_ready (rtx); +static int schedule_more_p (void); +static const char *rgn_print_insn (rtx, int); +static int rgn_rank (rtx, rtx); +static int contributes_to_priority (rtx, rtx); +static void compute_jump_reg_dependencies (rtx, regset, regset, regset); /* Return nonzero if there are more insns that should be scheduled. */ static int -schedule_more_p () +schedule_more_p (void) { return ! last_was_jump && sched_target_n_insns < target_n_insns; } @@ -2066,8 +1716,7 @@ schedule_more_p () once before scheduling a set of insns. */ static void -init_ready_list (ready) - struct ready_list *ready; +init_ready_list (struct ready_list *ready) { rtx prev_head = current_sched_info->prev_head; rtx next_tail = current_sched_info->next_tail; @@ -2086,8 +1735,7 @@ init_ready_list (ready) /* Prepare current target block info. */ if (current_nr_blocks > 1) { - candidate_table = (candidate *) xmalloc (current_nr_blocks - * sizeof (candidate)); + candidate_table = xmalloc (current_nr_blocks * sizeof (candidate)); bblst_last = 0; /* bblst_table holds split blocks and update blocks for each block after @@ -2095,11 +1743,10 @@ init_ready_list (ready) the TO blocks of region edges, so there can be at most rgn_nr_edges of them. */ bblst_size = (current_nr_blocks - target_bb) * rgn_nr_edges; - bblst_table = (int *) xmalloc (bblst_size * sizeof (int)); + bblst_table = xmalloc (bblst_size * sizeof (int)); bitlst_table_last = 0; - bitlst_table_size = rgn_nr_edges; - bitlst_table = (int *) xmalloc (rgn_nr_edges * sizeof (int)); + bitlst_table = xmalloc (rgn_nr_edges * sizeof (int)); compute_trg_info (target_bb); } @@ -2108,17 +1755,15 @@ init_ready_list (ready) Count number of insns in the target block being scheduled. */ for (insn = NEXT_INSN (prev_head); insn != next_tail; insn = NEXT_INSN (insn)) { - rtx next; - - if (! INSN_P (insn)) - continue; - next = NEXT_INSN (insn); + if (INSN_DEP_COUNT (insn) == 0) + { + ready_add (ready, insn); - if (INSN_DEP_COUNT (insn) == 0 - && (SCHED_GROUP_P (next) == 0 || ! INSN_P (next))) - ready_add (ready, insn); - if (!(SCHED_GROUP_P (insn))) - target_n_insns++; + if (targetm.sched.adjust_priority) + INSN_PRIORITY (insn) = + (*targetm.sched.adjust_priority) (insn, INSN_PRIORITY (insn)); + } + target_n_insns++; } /* Add to ready list all 'ready' insns in valid source blocks. @@ -2142,22 +1787,24 @@ init_ready_list (ready) if (!CANT_MOVE (insn) && (!IS_SPECULATIVE_INSN (insn) - || (insn_issue_delay (insn) <= 3 + || ((((!targetm.sched.use_dfa_pipeline_interface + || !(*targetm.sched.use_dfa_pipeline_interface) ()) + && insn_issue_delay (insn) <= 3) + || (targetm.sched.use_dfa_pipeline_interface + && (*targetm.sched.use_dfa_pipeline_interface) () + && (recog_memoized (insn) < 0 + || min_insn_conflict_delay (curr_state, + insn, insn) <= 3))) && check_live (insn, bb_src) && is_exception_free (insn, bb_src, target_bb)))) - { - rtx next; - - /* Note that we havn't squirrled away the notes for - blocks other than the current. So if this is a - speculative insn, NEXT might otherwise be a note. */ - next = next_nonnote_insn (insn); - if (INSN_DEP_COUNT (insn) == 0 - && (! next - || SCHED_GROUP_P (next) == 0 - || ! INSN_P (next))) - ready_add (ready, insn); - } + if (INSN_DEP_COUNT (insn) == 0) + { + ready_add (ready, insn); + + if (targetm.sched.adjust_priority) + INSN_PRIORITY (insn) = + (*targetm.sched.adjust_priority) (insn, INSN_PRIORITY (insn)); + } } } } @@ -2166,8 +1813,7 @@ init_ready_list (ready) insn can be scheduled, nonzero if we should silently discard it. */ static int -can_schedule_ready_p (insn) - rtx insn; +can_schedule_ready_p (rtx insn) { if (GET_CODE (insn) == JUMP_INSN) last_was_jump = 1; @@ -2175,7 +1821,6 @@ can_schedule_ready_p (insn) /* An interblock motion? */ if (INSN_BB (insn) != target_bb) { - rtx temp; basic_block b1; if (IS_SPECULATIVE_INSN (insn)) @@ -2192,39 +1837,30 @@ can_schedule_ready_p (insn) } nr_inter++; - /* Find the beginning of the scheduling group. */ - /* ??? Ought to update basic block here, but later bits of - schedule_block assumes the original insn block is - still intact. */ - - temp = insn; - while (SCHED_GROUP_P (temp)) - temp = PREV_INSN (temp); - - /* Update source block boundaries. */ - b1 = BLOCK_FOR_INSN (temp); - if (temp == b1->head && insn == b1->end) + /* Update source block boundaries. */ + b1 = BLOCK_FOR_INSN (insn); + if (insn == BB_HEAD (b1) && insn == BB_END (b1)) { /* We moved all the insns in the basic block. Emit a note after the last insn and update the begin/end boundaries to point to the note. */ rtx note = emit_note_after (NOTE_INSN_DELETED, insn); - b1->head = note; - b1->end = note; + BB_HEAD (b1) = note; + BB_END (b1) = note; } - else if (insn == b1->end) + else if (insn == BB_END (b1)) { /* We took insns from the end of the basic block, so update the end of block boundary so that it points to the first insn we did not move. */ - b1->end = PREV_INSN (temp); + BB_END (b1) = PREV_INSN (insn); } - else if (temp == b1->head) + else if (insn == BB_HEAD (b1)) { /* We took insns from the start of the basic block, so update the start of block boundary so that it points to the first insn we did not move. */ - b1->head = NEXT_INSN (insn); + BB_HEAD (b1) = NEXT_INSN (insn); } } else @@ -2241,8 +1877,7 @@ can_schedule_ready_p (insn) if it should be moved to the ready list or the queue, or zero if we should silently discard it. */ static int -new_ready (next) - rtx next; +new_ready (rtx next) { /* For speculative insns, before inserting to ready/queue, check live, exception-free, and issue-delay. */ @@ -2250,7 +1885,15 @@ new_ready (next) && (!IS_VALID (INSN_BB (next)) || CANT_MOVE (next) || (IS_SPECULATIVE_INSN (next) - && (insn_issue_delay (next) > 3 + && (0 + || (targetm.sched.use_dfa_pipeline_interface + && (*targetm.sched.use_dfa_pipeline_interface) () + && recog_memoized (next) >= 0 + && min_insn_conflict_delay (curr_state, next, + next) > 3) + || ((!targetm.sched.use_dfa_pipeline_interface + || !(*targetm.sched.use_dfa_pipeline_interface) ()) + && insn_issue_delay (next) > 3) || !check_live (next, INSN_BB (next)) || !is_exception_free (next, INSN_BB (next), target_bb))))) return 0; @@ -2263,9 +1906,7 @@ new_ready (next) to be formatted so that multiple output lines will line up nicely. */ static const char * -rgn_print_insn (insn, aligned) - rtx insn; - int aligned; +rgn_print_insn (rtx insn, int aligned) { static char tmp[80]; @@ -2286,8 +1927,7 @@ rgn_print_insn (insn, aligned) is to be preferred. Zero if they are equally good. */ static int -rgn_rank (insn1, insn2) - rtx insn1, insn2; +rgn_rank (rtx insn1, rtx insn2) { /* Some comparison make sense in interblock scheduling only. */ if (INSN_BB (insn1) != INSN_BB (insn2)) @@ -2318,19 +1958,21 @@ rgn_rank (insn1, insn2) calculations. */ static int -contributes_to_priority (next, insn) - rtx next, insn; +contributes_to_priority (rtx next, rtx insn) { return BLOCK_NUM (next) == BLOCK_NUM (insn); } -/* INSN is a JUMP_INSN. Store the set of registers that must be considered - to be set by this jump in SET. */ +/* INSN is a JUMP_INSN, COND_SET is the set of registers that are + conditionally set before INSN. Store the set of registers that + must be considered as used by this jump in USED and that of + registers that must be considered as set in SET. */ static void -compute_jump_reg_dependencies (insn, set) - rtx insn ATTRIBUTE_UNUSED; - regset set ATTRIBUTE_UNUSED; +compute_jump_reg_dependencies (rtx insn ATTRIBUTE_UNUSED, + regset cond_exec ATTRIBUTE_UNUSED, + regset used ATTRIBUTE_UNUSED, + regset set ATTRIBUTE_UNUSED) { /* Nothing to do here, since we postprocess jumps in add_branch_dependences. */ @@ -2352,27 +1994,55 @@ static struct sched_info region_sched_info = NULL, NULL, NULL, NULL, - 0, 0 + 0, 0, 0 }; +/* Determine if PAT sets a CLASS_LIKELY_SPILLED_P register. */ + +static bool +sets_likely_spilled (rtx pat) +{ + bool ret = false; + note_stores (pat, sets_likely_spilled_1, &ret); + return ret; +} + +static void +sets_likely_spilled_1 (rtx x, rtx pat, void *data) +{ + bool *ret = (bool *) data; + + if (GET_CODE (pat) == SET + && REG_P (x) + && REGNO (x) < FIRST_PSEUDO_REGISTER + && CLASS_LIKELY_SPILLED_P (REGNO_REG_CLASS (REGNO (x)))) + *ret = true; +} + /* Add dependences so that branches are scheduled to run last in their block. */ static void -add_branch_dependences (head, tail) - rtx head, tail; +add_branch_dependences (rtx head, rtx tail) { rtx insn, last; - /* For all branches, calls, uses, clobbers, and cc0 setters, force them - to remain in order at the end of the block by adding dependencies and - giving the last a high priority. There may be notes present, and - prev_head may also be a note. + /* For all branches, calls, uses, clobbers, cc0 setters, and instructions + that can throw exceptions, force them to remain in order at the end of + the block by adding dependencies and giving the last a high priority. + There may be notes present, and prev_head may also be a note. Branches must obviously remain at the end. Calls should remain at the end since moving them results in worse register allocation. Uses remain - at the end to ensure proper register allocation. cc0 setters remaim - at the end because they can't be moved away from their cc0 user. */ + at the end to ensure proper register allocation. + + cc0 setters remain at the end because they can't be moved away from + their cc0 user. + + Insns setting CLASS_LIKELY_SPILLED_P registers (usually return values) + are not moved before reload because we can wind up with register + allocation failures. */ + insn = tail; last = 0; while (GET_CODE (insn) == CALL_INSN @@ -2380,16 +2050,17 @@ add_branch_dependences (head, tail) || (GET_CODE (insn) == INSN && (GET_CODE (PATTERN (insn)) == USE || GET_CODE (PATTERN (insn)) == CLOBBER + || can_throw_internal (insn) #ifdef HAVE_cc0 || sets_cc0_p (PATTERN (insn)) #endif - )) + || (!reload_completed + && sets_likely_spilled (PATTERN (insn))))) || GET_CODE (insn) == NOTE) { if (GET_CODE (insn) != NOTE) { - if (last != 0 - && !find_insn_list (insn, LOG_LINKS (last))) + if (last != 0 && !find_insn_list (insn, LOG_LINKS (last))) { add_dependence (last, insn, REG_DEP_ANTI); INSN_REF_COUNT (insn)++; @@ -2398,17 +2069,6 @@ add_branch_dependences (head, tail) CANT_MOVE (insn) = 1; last = insn; - /* Skip over insns that are part of a group. - Make each insn explicitly depend on the previous insn. - This ensures that only the group header will ever enter - the ready queue (and, when scheduled, will automatically - schedule the SCHED_GROUP_P block). */ - while (SCHED_GROUP_P (insn)) - { - rtx temp = prev_nonnote_insn (insn); - add_dependence (insn, temp, REG_DEP_ANTI); - insn = temp; - } } /* Don't overrun the bounds of the basic block. */ @@ -2430,10 +2090,6 @@ add_branch_dependences (head, tail) add_dependence (last, insn, REG_DEP_ANTI); INSN_REF_COUNT (insn) = 1; - - /* Skip over insns that are part of a group. */ - while (SCHED_GROUP_P (insn)) - insn = prev_nonnote_insn (insn); } } @@ -2445,131 +2101,126 @@ add_branch_dependences (head, tail) static struct deps *bb_deps; +/* Duplicate the INSN_LIST elements of COPY and prepend them to OLD. */ + +static rtx +concat_INSN_LIST (rtx copy, rtx old) +{ + rtx new = old; + for (; copy ; copy = XEXP (copy, 1)) + new = alloc_INSN_LIST (XEXP (copy, 0), new); + return new; +} + +static void +concat_insn_mem_list (rtx copy_insns, rtx copy_mems, rtx *old_insns_p, + rtx *old_mems_p) +{ + rtx new_insns = *old_insns_p; + rtx new_mems = *old_mems_p; + + while (copy_insns) + { + new_insns = alloc_INSN_LIST (XEXP (copy_insns, 0), new_insns); + new_mems = alloc_EXPR_LIST (VOIDmode, XEXP (copy_mems, 0), new_mems); + copy_insns = XEXP (copy_insns, 1); + copy_mems = XEXP (copy_mems, 1); + } + + *old_insns_p = new_insns; + *old_mems_p = new_mems; +} + /* After computing the dependencies for block BB, propagate the dependencies found in TMP_DEPS to the successors of the block. */ static void -propagate_deps (bb, tmp_deps) - int bb; - struct deps *tmp_deps; +propagate_deps (int bb, struct deps *pred_deps) { int b = BB_TO_BLOCK (bb); int e, first_edge; - int reg; - rtx link_insn, link_mem; - rtx u; - - /* These lists should point to the right place, for correct - freeing later. */ - bb_deps[bb].pending_read_insns = tmp_deps->pending_read_insns; - bb_deps[bb].pending_read_mems = tmp_deps->pending_read_mems; - bb_deps[bb].pending_write_insns = tmp_deps->pending_write_insns; - bb_deps[bb].pending_write_mems = tmp_deps->pending_write_mems; /* bb's structures are inherited by its successors. */ first_edge = e = OUT_EDGES (b); - if (e <= 0) - return; - - do - { - rtx x; - int b_succ = TO_BLOCK (e); - int bb_succ = BLOCK_TO_BB (b_succ); - struct deps *succ_deps = bb_deps + bb_succ; - - /* Only bbs "below" bb, in the same region, are interesting. */ - if (CONTAINING_RGN (b) != CONTAINING_RGN (b_succ) - || bb_succ <= bb) - { - e = NEXT_OUT (e); - continue; - } - - /* The reg_last lists are inherited by bb_succ. */ - EXECUTE_IF_SET_IN_REG_SET (&tmp_deps->reg_last_in_use, 0, reg, - { - struct deps_reg *tmp_deps_reg = &tmp_deps->reg_last[reg]; - struct deps_reg *succ_deps_reg = &succ_deps->reg_last[reg]; - - for (u = tmp_deps_reg->uses; u; u = XEXP (u, 1)) - if (! find_insn_list (XEXP (u, 0), succ_deps_reg->uses)) - succ_deps_reg->uses - = alloc_INSN_LIST (XEXP (u, 0), succ_deps_reg->uses); - - for (u = tmp_deps_reg->sets; u; u = XEXP (u, 1)) - if (! find_insn_list (XEXP (u, 0), succ_deps_reg->sets)) - succ_deps_reg->sets - = alloc_INSN_LIST (XEXP (u, 0), succ_deps_reg->sets); - - for (u = tmp_deps_reg->clobbers; u; u = XEXP (u, 1)) - if (! find_insn_list (XEXP (u, 0), succ_deps_reg->clobbers)) - succ_deps_reg->clobbers - = alloc_INSN_LIST (XEXP (u, 0), succ_deps_reg->clobbers); - }); - IOR_REG_SET (&succ_deps->reg_last_in_use, &tmp_deps->reg_last_in_use); - - /* Mem read/write lists are inherited by bb_succ. */ - link_insn = tmp_deps->pending_read_insns; - link_mem = tmp_deps->pending_read_mems; - while (link_insn) - { - if (!(find_insn_mem_list (XEXP (link_insn, 0), - XEXP (link_mem, 0), - succ_deps->pending_read_insns, - succ_deps->pending_read_mems))) - add_insn_mem_dependence (succ_deps, &succ_deps->pending_read_insns, - &succ_deps->pending_read_mems, - XEXP (link_insn, 0), XEXP (link_mem, 0)); - link_insn = XEXP (link_insn, 1); - link_mem = XEXP (link_mem, 1); - } + if (e > 0) + do + { + int b_succ = TO_BLOCK (e); + int bb_succ = BLOCK_TO_BB (b_succ); + struct deps *succ_deps = bb_deps + bb_succ; + int reg; + + /* Only bbs "below" bb, in the same region, are interesting. */ + if (CONTAINING_RGN (b) != CONTAINING_RGN (b_succ) + || bb_succ <= bb) + { + e = NEXT_OUT (e); + continue; + } - link_insn = tmp_deps->pending_write_insns; - link_mem = tmp_deps->pending_write_mems; - while (link_insn) - { - if (!(find_insn_mem_list (XEXP (link_insn, 0), - XEXP (link_mem, 0), - succ_deps->pending_write_insns, - succ_deps->pending_write_mems))) - add_insn_mem_dependence (succ_deps, - &succ_deps->pending_write_insns, - &succ_deps->pending_write_mems, - XEXP (link_insn, 0), XEXP (link_mem, 0)); - - link_insn = XEXP (link_insn, 1); - link_mem = XEXP (link_mem, 1); - } + /* The reg_last lists are inherited by bb_succ. */ + EXECUTE_IF_SET_IN_REG_SET (&pred_deps->reg_last_in_use, 0, reg, + { + struct deps_reg *pred_rl = &pred_deps->reg_last[reg]; + struct deps_reg *succ_rl = &succ_deps->reg_last[reg]; + + succ_rl->uses = concat_INSN_LIST (pred_rl->uses, succ_rl->uses); + succ_rl->sets = concat_INSN_LIST (pred_rl->sets, succ_rl->sets); + succ_rl->clobbers = concat_INSN_LIST (pred_rl->clobbers, + succ_rl->clobbers); + succ_rl->uses_length += pred_rl->uses_length; + succ_rl->clobbers_length += pred_rl->clobbers_length; + }); + IOR_REG_SET (&succ_deps->reg_last_in_use, &pred_deps->reg_last_in_use); + + /* Mem read/write lists are inherited by bb_succ. */ + concat_insn_mem_list (pred_deps->pending_read_insns, + pred_deps->pending_read_mems, + &succ_deps->pending_read_insns, + &succ_deps->pending_read_mems); + concat_insn_mem_list (pred_deps->pending_write_insns, + pred_deps->pending_write_mems, + &succ_deps->pending_write_insns, + &succ_deps->pending_write_mems); + + succ_deps->last_pending_memory_flush + = concat_INSN_LIST (pred_deps->last_pending_memory_flush, + succ_deps->last_pending_memory_flush); + + succ_deps->pending_lists_length += pred_deps->pending_lists_length; + succ_deps->pending_flush_length += pred_deps->pending_flush_length; + + /* last_function_call is inherited by bb_succ. */ + succ_deps->last_function_call + = concat_INSN_LIST (pred_deps->last_function_call, + succ_deps->last_function_call); + + /* sched_before_next_call is inherited by bb_succ. */ + succ_deps->sched_before_next_call + = concat_INSN_LIST (pred_deps->sched_before_next_call, + succ_deps->sched_before_next_call); + + e = NEXT_OUT (e); + } + while (e != first_edge); - /* last_function_call is inherited by bb_succ. */ - for (u = tmp_deps->last_function_call; u; u = XEXP (u, 1)) - if (! find_insn_list (XEXP (u, 0), succ_deps->last_function_call)) - succ_deps->last_function_call - = alloc_INSN_LIST (XEXP (u, 0), succ_deps->last_function_call); - - /* last_pending_memory_flush is inherited by bb_succ. */ - for (u = tmp_deps->last_pending_memory_flush; u; u = XEXP (u, 1)) - if (! find_insn_list (XEXP (u, 0), - succ_deps->last_pending_memory_flush)) - succ_deps->last_pending_memory_flush - = alloc_INSN_LIST (XEXP (u, 0), - succ_deps->last_pending_memory_flush); - - /* sched_before_next_call is inherited by bb_succ. */ - x = LOG_LINKS (tmp_deps->sched_before_next_call); - for (; x; x = XEXP (x, 1)) - add_dependence (succ_deps->sched_before_next_call, - XEXP (x, 0), REG_DEP_ANTI); - - e = NEXT_OUT (e); - } - while (e != first_edge); + /* These lists should point to the right place, for correct + freeing later. */ + bb_deps[bb].pending_read_insns = pred_deps->pending_read_insns; + bb_deps[bb].pending_read_mems = pred_deps->pending_read_mems; + bb_deps[bb].pending_write_insns = pred_deps->pending_write_insns; + bb_deps[bb].pending_write_mems = pred_deps->pending_write_mems; + + /* Can't allow these to be freed twice. */ + pred_deps->pending_read_insns = 0; + pred_deps->pending_read_mems = 0; + pred_deps->pending_write_insns = 0; + pred_deps->pending_write_mems = 0; } /* Compute backward dependences inside bb. In a multiple blocks region: (1) a bb is analyzed after its predecessors, and (2) the lists in effect at the end of bb (after analyzing for bb) are inherited by - bb's successrs. + bb's successors. Specifically for reg-reg data dependences, the block insns are scanned by sched_analyze () top-to-bottom. Two lists are @@ -2584,8 +2235,7 @@ propagate_deps (bb, tmp_deps) similar, and the result is interblock dependences in the region. */ static void -compute_block_backward_dependences (bb) - int bb; +compute_block_backward_dependences (int bb) { rtx head, tail; struct deps tmp_deps; @@ -2608,7 +2258,7 @@ compute_block_backward_dependences (bb) them to the unused_*_list variables, so that they can be reused. */ static void -free_pending_lists () +free_pending_lists (void) { int bb; @@ -2624,7 +2274,7 @@ free_pending_lists () /* Print dependences for debugging, callable from debugger. */ void -debug_dependencies () +debug_dependencies (void) { int bb; @@ -2642,14 +2292,27 @@ debug_dependencies () fprintf (sched_dump, "\n;; --- Region Dependences --- b %d bb %d \n", BB_TO_BLOCK (bb), bb); - fprintf (sched_dump, ";; %7s%6s%6s%6s%6s%6s%11s%6s\n", - "insn", "code", "bb", "dep", "prio", "cost", "blockage", "units"); - fprintf (sched_dump, ";; %7s%6s%6s%6s%6s%6s%11s%6s\n", - "----", "----", "--", "---", "----", "----", "--------", "-----"); + if (targetm.sched.use_dfa_pipeline_interface + && (*targetm.sched.use_dfa_pipeline_interface) ()) + { + fprintf (sched_dump, ";; %7s%6s%6s%6s%6s%6s%14s\n", + "insn", "code", "bb", "dep", "prio", "cost", + "reservation"); + fprintf (sched_dump, ";; %7s%6s%6s%6s%6s%6s%14s\n", + "----", "----", "--", "---", "----", "----", + "-----------"); + } + else + { + fprintf (sched_dump, ";; %7s%6s%6s%6s%6s%6s%11s%6s\n", + "insn", "code", "bb", "dep", "prio", "cost", "blockage", "units"); + fprintf (sched_dump, ";; %7s%6s%6s%6s%6s%6s%11s%6s\n", + "----", "----", "--", "---", "----", "----", "--------", "-----"); + } + for (insn = head; insn != next_tail; insn = NEXT_INSN (insn)) { rtx link; - int unit, range; if (! INSN_P (insn)) { @@ -2669,22 +2332,46 @@ debug_dependencies () continue; } - unit = insn_unit (insn); - range = (unit < 0 - || function_units[unit].blockage_range_function == 0) ? 0 : - function_units[unit].blockage_range_function (insn); - fprintf (sched_dump, - ";; %s%5d%6d%6d%6d%6d%6d %3d -%3d ", - (SCHED_GROUP_P (insn) ? "+" : " "), - INSN_UID (insn), - INSN_CODE (insn), - INSN_BB (insn), - INSN_DEP_COUNT (insn), - INSN_PRIORITY (insn), - insn_cost (insn, 0, 0), - (int) MIN_BLOCKAGE_COST (range), - (int) MAX_BLOCKAGE_COST (range)); - insn_print_units (insn); + if (targetm.sched.use_dfa_pipeline_interface + && (*targetm.sched.use_dfa_pipeline_interface) ()) + { + fprintf (sched_dump, + ";; %s%5d%6d%6d%6d%6d%6d ", + (SCHED_GROUP_P (insn) ? "+" : " "), + INSN_UID (insn), + INSN_CODE (insn), + INSN_BB (insn), + INSN_DEP_COUNT (insn), + INSN_PRIORITY (insn), + insn_cost (insn, 0, 0)); + + if (recog_memoized (insn) < 0) + fprintf (sched_dump, "nothing"); + else + print_reservation (sched_dump, insn); + } + else + { + int unit = insn_unit (insn); + int range + = (unit < 0 + || function_units[unit].blockage_range_function == 0 + ? 0 + : function_units[unit].blockage_range_function (insn)); + fprintf (sched_dump, + ";; %s%5d%6d%6d%6d%6d%6d %3d -%3d ", + (SCHED_GROUP_P (insn) ? "+" : " "), + INSN_UID (insn), + INSN_CODE (insn), + INSN_BB (insn), + INSN_DEP_COUNT (insn), + INSN_PRIORITY (insn), + insn_cost (insn, 0, 0), + (int) MIN_BLOCKAGE_COST (range), + (int) MAX_BLOCKAGE_COST (range)); + insn_print_units (insn); + } + fprintf (sched_dump, "\t: "); for (link = INSN_DEPEND (insn); link; link = XEXP (link, 1)) fprintf (sched_dump, "%d ", INSN_UID (XEXP (link, 0))); @@ -2700,8 +2387,7 @@ debug_dependencies () scheduled after its flow predecessors. */ static void -schedule_region (rgn) - int rgn; +schedule_region (int rgn) { int bb; int rgn_n_insns = 0; @@ -2713,8 +2399,8 @@ schedule_region (rgn) init_deps_global (); - /* Initializations for region data dependence analyisis. */ - bb_deps = (struct deps *) xmalloc (sizeof (struct deps) * current_nr_blocks); + /* Initializations for region data dependence analysis. */ + bb_deps = xmalloc (sizeof (struct deps) * current_nr_blocks); for (bb = 0; bb < current_nr_blocks; bb++) init_deps (bb_deps + bb); @@ -2729,6 +2415,10 @@ schedule_region (rgn) get_block_head_tail (BB_TO_BLOCK (bb), &head, &tail); compute_forward_dependences (head, tail); + + if (targetm.sched.dependencies_evaluation_hook) + targetm.sched.dependencies_evaluation_hook (head, tail); + } /* Set priorities. */ @@ -2745,20 +2435,17 @@ schedule_region (rgn) { int i; - prob = (float *) xmalloc ((current_nr_blocks) * sizeof (float)); - - bbset_size = current_nr_blocks / HOST_BITS_PER_WIDE_INT + 1; - dom = (bbset *) xmalloc (current_nr_blocks * sizeof (bbset)); - for (i = 0; i < current_nr_blocks; i++) - dom[i] = (bbset) xcalloc (bbset_size, sizeof (HOST_WIDE_INT)); + prob = xmalloc ((current_nr_blocks) * sizeof (float)); + dom = sbitmap_vector_alloc (current_nr_blocks, current_nr_blocks); + sbitmap_vector_zero (dom, current_nr_blocks); /* Edge to bit. */ rgn_nr_edges = 0; - edge_to_bit = (int *) xmalloc (nr_edges * sizeof (int)); + edge_to_bit = xmalloc (nr_edges * sizeof (int)); for (i = 1; i < nr_edges; i++) if (CONTAINING_RGN (FROM_BLOCK (i)) == rgn) EDGE_TO_BIT (i) = rgn_nr_edges++; - rgn_edges = (int *) xmalloc (rgn_nr_edges * sizeof (int)); + rgn_edges = xmalloc (rgn_nr_edges * sizeof (int)); rgn_nr_edges = 0; for (i = 1; i < nr_edges; i++) @@ -2766,18 +2453,10 @@ schedule_region (rgn) rgn_edges[rgn_nr_edges++] = i; /* Split edges. */ - edgeset_size = rgn_nr_edges / HOST_BITS_PER_WIDE_INT + 1; - edgeset_bitsize = rgn_nr_edges; - pot_split = (edgeset *) xmalloc (current_nr_blocks * sizeof (edgeset)); - ancestor_edges - = (edgeset *) xmalloc (current_nr_blocks * sizeof (edgeset)); - for (i = 0; i < current_nr_blocks; i++) - { - pot_split[i] = - (edgeset) xcalloc (edgeset_size, sizeof (HOST_WIDE_INT)); - ancestor_edges[i] = - (edgeset) xcalloc (edgeset_size, sizeof (HOST_WIDE_INT)); - } + pot_split = sbitmap_vector_alloc (current_nr_blocks, rgn_nr_edges); + sbitmap_vector_zero (pot_split, current_nr_blocks); + ancestor_edges = sbitmap_vector_alloc (current_nr_blocks, rgn_nr_edges); + sbitmap_vector_zero (ancestor_edges, current_nr_blocks); /* Compute probabilities, dominators, split_edges. */ for (bb = 0; bb < current_nr_blocks; bb++) @@ -2806,7 +2485,7 @@ schedule_region (rgn) /* rm_other_notes only removes notes which are _inside_ the block---that is, it won't remove notes before the first real insn - or after the last real insn of the block. So if the first insn + or after the last real insn of the block. So if the first insn has a REG_SAVE_NOTE which would otherwise be emitted before the insn, it is redundant with the note before the start of the block, and so we have to take it out. */ @@ -2837,10 +2516,10 @@ schedule_region (rgn) sched_rgn_n_insns += sched_n_insns; /* Update target block boundaries. */ - if (head == BLOCK_HEAD (b)) - BLOCK_HEAD (b) = current_sched_info->head; - if (tail == BLOCK_END (b)) - BLOCK_END (b) = current_sched_info->tail; + if (head == BB_HEAD (BASIC_BLOCK (b))) + BB_HEAD (BASIC_BLOCK (b)) = current_sched_info->head; + if (tail == BB_END (BASIC_BLOCK (b))) + BB_END (BASIC_BLOCK (b)) = current_sched_info->tail; /* Clean up. */ if (current_nr_blocks > 1) @@ -2875,20 +2554,12 @@ schedule_region (rgn) if (current_nr_blocks > 1) { - int i; - free (prob); - for (i = 0; i < current_nr_blocks; ++i) - { - free (dom[i]); - free (pot_split[i]); - free (ancestor_edges[i]); - } - free (dom); + sbitmap_vector_free (dom); + sbitmap_vector_free (pot_split); + sbitmap_vector_free (ancestor_edges); free (edge_to_bit); free (rgn_edges); - free (pot_split); - free (ancestor_edges); } } @@ -2899,18 +2570,16 @@ static int *deaths_in_region; /* Initialize data structures for region scheduling. */ static void -init_regions () +init_regions (void) { sbitmap blocks; int rgn; nr_regions = 0; - rgn_table = (region *) xmalloc ((n_basic_blocks) * sizeof (region)); - rgn_bb_table = (int *) xmalloc ((n_basic_blocks) * sizeof (int)); - block_to_bb = (int *) xmalloc ((n_basic_blocks) * sizeof (int)); - containing_rgn = (int *) xmalloc ((n_basic_blocks) * sizeof (int)); - - blocks = sbitmap_alloc (n_basic_blocks); + rgn_table = xmalloc ((n_basic_blocks) * sizeof (region)); + rgn_bb_table = xmalloc ((n_basic_blocks) * sizeof (int)); + block_to_bb = xmalloc ((last_basic_block) * sizeof (int)); + containing_rgn = xmalloc ((last_basic_block) * sizeof (int)); /* Compute regions for scheduling. */ if (reload_completed @@ -2928,26 +2597,16 @@ init_regions () } else { - sbitmap *dom; struct edge_list *edge_list; - dom = sbitmap_vector_alloc (n_basic_blocks, n_basic_blocks); - - /* The scheduler runs after flow; therefore, we can't blindly call - back into find_basic_blocks since doing so could invalidate the - info in global_live_at_start. - - Consider a block consisting entirely of dead stores; after life - analysis it would be a block of NOTE_INSN_DELETED notes. If - we call find_basic_blocks again, then the block would be removed - entirely and invalidate our the register live information. - - We could (should?) recompute register live information. Doing - so may even be beneficial. */ + /* The scheduler runs after estimate_probabilities; therefore, we + can't blindly call back into find_basic_blocks since doing so + could invalidate the branch probability info. We could, + however, call cleanup_cfg. */ edge_list = create_edge_list (); /* Compute the dominators and post dominators. */ - calculate_dominance_info (NULL, dom, CDI_DOMINATORS); + calculate_dominance_info (CDI_DOMINATORS); /* build_control_flow will return nonzero if it detects unreachable blocks or any other irregularity with the cfg which prevents @@ -2955,7 +2614,7 @@ init_regions () if (build_control_flow (edge_list) != 0) find_single_block_region (); else - find_rgns (edge_list, dom); + find_rgns (edge_list); if (sched_verbose >= 3) debug_regions (); @@ -2965,37 +2624,42 @@ init_regions () /* For now. This will move as more and more of haifa is converted to using the cfg code in flow.c. */ - free (dom); + free_dominance_info (CDI_DOMINATORS); } } - deaths_in_region = (int *) xmalloc (sizeof (int) * nr_regions); - /* Remove all death notes from the subroutine. */ - for (rgn = 0; rgn < nr_regions; rgn++) + if (CHECK_DEAD_NOTES) { - int b; + blocks = sbitmap_alloc (last_basic_block); + deaths_in_region = xmalloc (sizeof (int) * nr_regions); + /* Remove all death notes from the subroutine. */ + for (rgn = 0; rgn < nr_regions; rgn++) + { + int b; - sbitmap_zero (blocks); - for (b = RGN_NR_BLOCKS (rgn) - 1; b >= 0; --b) - SET_BIT (blocks, rgn_bb_table[RGN_BLOCKS (rgn) + b]); + sbitmap_zero (blocks); + for (b = RGN_NR_BLOCKS (rgn) - 1; b >= 0; --b) + SET_BIT (blocks, rgn_bb_table[RGN_BLOCKS (rgn) + b]); - deaths_in_region[rgn] = count_or_remove_death_notes (blocks, 1); + deaths_in_region[rgn] = count_or_remove_death_notes (blocks, 1); + } + sbitmap_free (blocks); } - - sbitmap_free (blocks); + else + count_or_remove_death_notes (NULL, 1); } /* The one entry point in this file. DUMP_FILE is the dump file for this pass. */ void -schedule_insns (dump_file) - FILE *dump_file; +schedule_insns (FILE *dump_file) { sbitmap large_region_blocks, blocks; int rgn; int any_large_regions; + basic_block bb; /* Taking care of this degenerate case makes the rest of this code simpler. */ @@ -3010,16 +2674,16 @@ schedule_insns (dump_file) init_regions (); current_sched_info = ®ion_sched_info; - + /* Schedule every region in the subroutine. */ for (rgn = 0; rgn < nr_regions; rgn++) schedule_region (rgn); /* Update life analysis for the subroutine. Do single block regions first so that we can verify that live_at_start didn't change. Then - do all other blocks. */ + do all other blocks. */ /* ??? There is an outside possibility that update_life_info, or more - to the point propagate_block, could get called with non-zero flags + to the point propagate_block, could get called with nonzero flags more than once for one basic block. This would be kinda bad if it were to happen, since REG_INFO would be accumulated twice for the block, and we'd have twice the REG_DEAD notes. @@ -3029,46 +2693,58 @@ schedule_insns (dump_file) best way to test for this kind of thing... */ allocate_reg_life_data (); - compute_bb_for_insn (get_max_uid ()); + compute_bb_for_insn (); any_large_regions = 0; - large_region_blocks = sbitmap_alloc (n_basic_blocks); - sbitmap_ones (large_region_blocks); - - blocks = sbitmap_alloc (n_basic_blocks); - + large_region_blocks = sbitmap_alloc (last_basic_block); + sbitmap_zero (large_region_blocks); + FOR_EACH_BB (bb) + SET_BIT (large_region_blocks, bb->index); + + blocks = sbitmap_alloc (last_basic_block); + sbitmap_zero (blocks); + + /* Update life information. For regions consisting of multiple blocks + we've possibly done interblock scheduling that affects global liveness. + For regions consisting of single blocks we need to do only local + liveness. */ for (rgn = 0; rgn < nr_regions; rgn++) if (RGN_NR_BLOCKS (rgn) > 1) any_large_regions = 1; else { - sbitmap_zero (blocks); SET_BIT (blocks, rgn_bb_table[RGN_BLOCKS (rgn)]); RESET_BIT (large_region_blocks, rgn_bb_table[RGN_BLOCKS (rgn)]); - - /* Don't update reg info after reload, since that affects - regs_ever_live, which should not change after reload. */ - update_life_info (blocks, UPDATE_LIFE_LOCAL, - (reload_completed ? PROP_DEATH_NOTES - : PROP_DEATH_NOTES | PROP_REG_INFO)); - -#ifndef HAVE_conditional_execution - /* ??? REG_DEAD notes only exist for unconditional deaths. We need - a count of the conditional plus unconditional deaths for this to - work out. */ - /* In the single block case, the count of registers that died should - not have changed during the schedule. */ - if (count_or_remove_death_notes (blocks, 0) != deaths_in_region[rgn]) - abort (); -#endif } + /* Don't update reg info after reload, since that affects + regs_ever_live, which should not change after reload. */ + update_life_info (blocks, UPDATE_LIFE_LOCAL, + (reload_completed ? PROP_DEATH_NOTES + : PROP_DEATH_NOTES | PROP_REG_INFO)); if (any_large_regions) { update_life_info (large_region_blocks, UPDATE_LIFE_GLOBAL, PROP_DEATH_NOTES | PROP_REG_INFO); } + if (CHECK_DEAD_NOTES) + { + /* Verify the counts of basic block notes in single the basic block + regions. */ + for (rgn = 0; rgn < nr_regions; rgn++) + if (RGN_NR_BLOCKS (rgn) == 1) + { + sbitmap_zero (blocks); + SET_BIT (blocks, rgn_bb_table[RGN_BLOCKS (rgn)]); + + if (deaths_in_region[rgn] + != count_or_remove_death_notes (blocks, 0)) + abort (); + } + free (deaths_in_region); + } + /* Reposition the prologue and epilogue notes in case we moved the prologue/epilogue insns. */ if (reload_completed) @@ -3121,7 +2797,5 @@ schedule_insns (dump_file) sbitmap_free (blocks); sbitmap_free (large_region_blocks); - - free (deaths_in_region); } #endif