the Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA
02111-1307, USA. */
-/* Instruction scheduling pass.
-
- This pass implements list scheduling within basic blocks. It is
- run twice: (1) after flow analysis, but before register allocation,
- and (2) after register allocation.
-
- The first run performs interblock scheduling, moving insns between
- different blocks in the same "region", and the second runs only
- basic block scheduling.
-
- Interblock motions performed are useful motions and speculative
- motions, including speculative loads. Motions requiring code
- duplication are not supported. The identification of motion type
- and the check for validity of speculative motions requires
- construction and analysis of the function's control flow graph.
- The scheduler works as follows:
+/* Instruction scheduling pass. This file, along with sched-deps.c,
+ contains the generic parts. The actual entry point is found for
+ the normal instruction scheduling pass is found in sched-rgn.c.
We compute insn priorities based on data dependencies. Flow
analysis only creates a fraction of the data-dependencies we must
this pass. Notes that refer to the starting and ending of
exception regions are also carefully retained by this pass. All
other NOTE insns are grouped in their same relative order at the
- beginning of basic blocks and regions that have been scheduled.
-
- The main entry point for this pass is schedule_insns(), called for
- each function. The work of the scheduler is organized in three
- levels: (1) function level: insns are subject to splitting,
- control-flow-graph is constructed, regions are computed (after
- reload, each region is of one block), (2) region level: control
- flow graph attributes required for interblock scheduling are
- computed (dominators, reachability, etc.), data dependences and
- priorities are computed, and (3) block level: insns in the block
- are actually scheduled. */
+ beginning of basic blocks and regions that have been scheduled. */
\f
#include "config.h"
#include "system.h"
N=3: rtl at abort point, control-flow, regions info.
N=5: dependences info. */
-#define MAX_RGN_BLOCKS 10
-#define MAX_RGN_INSNS 100
-
static int sched_verbose_param = 0;
-static int sched_verbose = 0;
-
-/* nr_inter/spec counts interblock/speculative motion for the function. */
-static int nr_inter, nr_spec;
+int sched_verbose = 0;
/* Debugging file. All printouts are sent to dump, which is always set,
either to stderr, or to the dump listing file (-dRS). */
struct haifa_insn_data *h_i_d;
-#define INSN_PRIORITY(INSN) (h_i_d[INSN_UID (INSN)].priority)
-#define INSN_COST(INSN) (h_i_d[INSN_UID (INSN)].cost)
-#define INSN_UNIT(INSN) (h_i_d[INSN_UID (INSN)].units)
-#define INSN_REG_WEIGHT(INSN) (h_i_d[INSN_UID (INSN)].reg_weight)
-
-#define INSN_BLOCKAGE(INSN) (h_i_d[INSN_UID (INSN)].blockage)
-#define UNIT_BITS 5
-#define BLOCKAGE_MASK ((1 << BLOCKAGE_BITS) - 1)
-#define ENCODE_BLOCKAGE(U, R) \
- (((U) << BLOCKAGE_BITS \
- | MIN_BLOCKAGE_COST (R)) << BLOCKAGE_BITS \
- | MAX_BLOCKAGE_COST (R))
-#define UNIT_BLOCKED(B) ((B) >> (2 * BLOCKAGE_BITS))
-#define BLOCKAGE_RANGE(B) \
- (((((B) >> BLOCKAGE_BITS) & BLOCKAGE_MASK) << (HOST_BITS_PER_INT / 2)) \
- | ((B) & BLOCKAGE_MASK))
-
-/* Encodings of the `<name>_unit_blockage_range' function. */
-#define MIN_BLOCKAGE_COST(R) ((R) >> (HOST_BITS_PER_INT / 2))
-#define MAX_BLOCKAGE_COST(R) ((R) & ((1 << (HOST_BITS_PER_INT / 2)) - 1))
-
#define DONE_PRIORITY -1
#define MAX_PRIORITY 0x7fffffff
#define TAIL_PRIORITY 0x7ffffffe
#define DONE_PRIORITY_P(INSN) (INSN_PRIORITY (INSN) < 0)
#define LOW_PRIORITY_P(INSN) ((INSN_PRIORITY (INSN) & 0x7f000000) == 0)
-#define INSN_REF_COUNT(INSN) (h_i_d[INSN_UID (INSN)].ref_count)
#define LINE_NOTE(INSN) (h_i_d[INSN_UID (INSN)].line_note)
#define INSN_TICK(INSN) (h_i_d[INSN_UID (INSN)].tick)
-#define FED_BY_SPEC_LOAD(insn) (h_i_d[INSN_UID (insn)].fed_by_spec_load)
-#define IS_LOAD_INSN(insn) (h_i_d[INSN_UID (insn)].is_load_insn)
/* Vector indexed by basic block number giving the starting line-number
for each basic block. */
static void schedule_unit PARAMS ((int, rtx, int));
static int actual_hazard PARAMS ((int, rtx, int, int));
static int potential_hazard PARAMS ((int, rtx, int));
-static int insn_cost PARAMS ((rtx, rtx, rtx));
static int priority PARAMS ((rtx));
-static void free_pending_lists PARAMS ((void));
static int rank_for_schedule PARAMS ((const PTR, const PTR));
static void swap_sort PARAMS ((rtx *, int));
static void queue_insn PARAMS ((rtx, int));
static void schedule_insn PARAMS ((rtx, struct ready_list *, int));
static void find_insn_reg_weight PARAMS ((int));
-static void schedule_block PARAMS ((int, int));
-static int insn_issue_delay PARAMS ((rtx));
static void adjust_priority PARAMS ((rtx));
-/* Control flow graph edges are kept in circular lists. */
-typedef struct
-{
- int from_block;
- int to_block;
- int next_in;
- int next_out;
-}
-haifa_edge;
-static haifa_edge *edge_table;
-
-#define NEXT_IN(edge) (edge_table[edge].next_in)
-#define NEXT_OUT(edge) (edge_table[edge].next_out)
-#define FROM_BLOCK(edge) (edge_table[edge].from_block)
-#define TO_BLOCK(edge) (edge_table[edge].to_block)
-
-/* Number of edges in the control flow graph. (In fact, larger than
- that by 1, since edge 0 is unused.) */
-static int nr_edges;
-
-/* Circular list of incoming/outgoing edges of a block. */
-static int *in_edges;
-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));
-
-/* A region is the main entity for interblock scheduling: insns
- are allowed to move between blocks in the same region, along
- control flow graph edges, in the 'up' direction. */
-typedef struct
-{
- int rgn_nr_blocks; /* Number of blocks in region. */
- int rgn_blocks; /* cblocks in the region (actually index in rgn_bb_table). */
-}
-region;
-
-/* Number of regions in the procedure. */
-static int nr_regions;
-
-/* Table of region descriptions. */
-static region *rgn_table;
-
-/* Array of lists of regions' blocks. */
-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. */
-static int *block_to_bb;
-
-/* The number of the region containing a block. */
-static int *containing_rgn;
-
-#define RGN_NR_BLOCKS(rgn) (rgn_table[rgn].rgn_nr_blocks)
-#define RGN_BLOCKS(rgn) (rgn_table[rgn].rgn_blocks)
-#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 *));
-
-extern void debug_live PARAMS ((int, int));
-
-/* Blocks of the current region being scheduled. */
-static int current_nr_blocks;
-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. */
- int nr_members; /* The number of members of the bit list. */
-}
-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 *));
-
-/* Target info declarations.
-
- The block currently being scheduled is referred to as the "target" block,
- while other blocks in the region from which insns can be moved to the
- target are called "source" blocks. The candidate structure holds info
- about such sources: are they valid? Speculative? Etc. */
-typedef bitlst bblst;
-typedef struct
-{
- char is_valid;
- char is_speculative;
- int src_prob;
- bblst split_bbs;
- bblst update_bbs;
-}
-candidate;
-
-static candidate *candidate_table;
-
-/* A speculative motion requires checking live information on the path
- from 'source' to 'target'. The split blocks are those to be checked.
- After a speculative motion, live information should be modified in
- the 'update' blocks.
-
- Lists of split and update blocks for each candidate of the current
- target are in array bblst_table. */
-static int *bblst_table, bblst_size, bblst_last;
-
-#define IS_VALID(src) ( candidate_table[src].is_valid )
-#define IS_SPECULATIVE(src) ( candidate_table[src].is_speculative )
-#define SRC_PROB(src) ( candidate_table[src].src_prob )
-
-/* The bb being currently scheduled. */
-static int target_bb;
-
-/* List of edges. */
-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;
-
-/* Dominators array: dom[i] contains the bbset of dominators of
- bb i in the region. */
-static bbset *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) )
-
-/* Probability: Prob[i] is a float in [0, 1] which is the probability
- of bb i relative to the region entry. */
-static float *prob;
-
-/* The probability of bb_src, relative to bb_trg. Note, that while the
- 'prob[bb]' is a float in [0, 1], this macro returns an integer
- in [0, 100]. */
-#define GET_SRC_PROB(bb_src, bb_trg) ((int) (100.0 * (prob[bb_src] / \
- prob[bb_trg])))
-
-/* Bit-set of edges, where bit i stands for edge i. */
-typedef bitset edgeset;
-
-/* Number of edges in the region. */
-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;
-#define EDGE_TO_BIT(edge) (edge_to_bit[edge])
-
-/* The split edges of a source bb is different for each target
- bb. In order to compute this efficiently, the 'potential-split edges'
- are computed for each bb prior to scheduling a region. This is actually
- the split edges of each bb relative to the region entry.
-
- pot_split[bb] is the set of potential split edges of bb. */
-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));
-
-#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(). */
-#define MIN_DIFF_PRIORITY 2
-#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));
-
/* Notes handling mechanism:
=========================
Generally, NOTES are saved before scheduling and restored after scheduling.
static rtx unlink_other_notes PARAMS ((rtx, rtx));
static rtx unlink_line_notes PARAMS ((rtx, rtx));
-static void rm_line_notes PARAMS ((int));
-static void save_line_notes PARAMS ((int));
-static void restore_line_notes PARAMS ((int));
-static void rm_redundant_line_notes PARAMS ((void));
-static void rm_other_notes PARAMS ((rtx, rtx));
static rtx reemit_notes PARAMS ((rtx, rtx));
-static int no_real_insns_p PARAMS ((rtx, rtx));
-static void get_block_head_tail PARAMS ((int, rtx *, rtx *));
-static void get_bb_head_tail PARAMS ((int, rtx *, rtx *));
-
-static void ready_add PARAMS ((struct ready_list *, rtx));
static rtx *ready_lastpos PARAMS ((struct ready_list *));
static void ready_sort PARAMS ((struct ready_list *));
static rtx ready_remove_first PARAMS ((struct ready_list *));
static void queue_to_ready PARAMS ((struct ready_list *));
static void debug_ready_list PARAMS ((struct ready_list *));
-void debug_reg_vector PARAMS ((regset));
static rtx move_insn1 PARAMS ((rtx, rtx));
static rtx move_insn PARAMS ((rtx, rtx));
-static int set_priorities PARAMS ((int));
-static void init_regions PARAMS ((void));
-static void sched_init PARAMS ((FILE *));
-static void schedule_region PARAMS ((int));
-static void propagate_deps PARAMS ((int, struct deps *, int));
#endif /* INSN_SCHEDULING */
\f
}
#else
-/* Computation of memory dependencies. */
-
-/* Data structures for the computation of data dependences in a regions. We
- keep one mem_deps structure for every basic block. Before analyzing the
- data dependences for a bb, its variables are initialized as a function of
- the variables of its predecessors. When the analysis for a bb completes,
- we save the contents to the corresponding bb_mem_deps[bb] variable. */
-
-static struct deps *bb_deps;
-
/* Pointer to the last instruction scheduled. Used by rank_for_schedule,
so that insns independent of the last scheduled insn will be preferred
over dependent instructions. */
static rtx last_scheduled_insn;
-/* Functions for construction of the control flow graph. */
-
-/* Return 1 if control flow graph should not be constructed, 0 otherwise.
-
- We decide not to build the control flow graph if there is possibly more
- than one entry to the function, if computed branches exist, of if we
- have nonlocal gotos. */
-
-static int
-is_cfg_nonregular ()
-{
- int b;
- rtx insn;
- RTX_CODE code;
-
- /* If we have a label that could be the target of a nonlocal goto, then
- the cfg is not well structured. */
- if (nonlocal_goto_handler_labels)
- return 1;
-
- /* If we have any forced labels, then the cfg is not well structured. */
- if (forced_labels)
- return 1;
-
- /* If this function has a computed jump, then we consider the cfg
- not well structured. */
- if (current_function_has_computed_jump)
- return 1;
-
- /* 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)
- 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))
- {
- code = GET_CODE (insn);
- if (GET_RTX_CLASS (code) == 'i')
- {
- rtx note;
-
- for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
- if (REG_NOTE_KIND (note) == REG_LABEL)
- return 1;
- }
-
- if (insn == BLOCK_END (b))
- break;
- }
-
- /* All the tests passed. Consider the cfg well structured. */
- return 0;
-}
-
-/* Build the control flow graph and set nr_edges.
-
- Instead of trying to build a cfg ourselves, we rely on flow to
- do it for us. Stamp out useless code (and bug) duplication.
-
- Return nonzero if an irregularity in the cfg is found which would
- prevent cross block scheduling. */
+/* Compute the function units used by INSN. This caches the value
+ returned by function_units_used. A function unit is encoded as the
+ unit number if the value is non-negative and the compliment of a
+ mask if the value is negative. A function unit index is the
+ non-negative encoding. */
-static int
-build_control_flow (edge_list)
- struct edge_list *edge_list;
+HAIFA_INLINE int
+insn_unit (insn)
+ rtx insn;
{
- int i, unreachable, num_edges;
-
- /* This already accounts for entry/exit edges. */
- num_edges = NUM_EDGES (edge_list);
-
- /* Unreachable loops with more than one basic block are detected
- during the DFS traversal in find_rgns.
-
- Unreachable loops with a single block are detected here. This
- 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++)
- {
- basic_block b = BASIC_BLOCK (i);
-
- if (b->pred == NULL
- || (b->pred->src == b
- && b->pred->pred_next == NULL))
- unreachable = 1;
- }
-
- /* ??? 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));
+ register int unit = INSN_UNIT (insn);
- nr_edges = 0;
- for (i = 0; i < num_edges; i++)
+ if (unit == 0)
{
- edge e = INDEX_EDGE (edge_list, i);
-
- if (e->dest != EXIT_BLOCK_PTR
- && e->src != ENTRY_BLOCK_PTR)
- new_edge (e->src->index, e->dest->index);
- }
-
- /* Increment by 1, since edge 0 is unused. */
- nr_edges++;
-
- return unreachable;
-}
-
-/* Record an edge in the control flow graph from SOURCE to TARGET.
-
- In theory, this is redundant with the s_succs computed above, but
- we have not converted all of haifa to use information from the
- integer lists. */
-
-static void
-new_edge (source, target)
- int source, target;
-{
- int e, next_edge;
- int curr_edge, fst_edge;
+ recog_memoized (insn);
- /* Check for duplicates. */
- fst_edge = curr_edge = OUT_EDGES (source);
- while (curr_edge)
- {
- if (FROM_BLOCK (curr_edge) == source
- && TO_BLOCK (curr_edge) == target)
+ /* A USE insn, or something else we don't need to understand.
+ We can't pass these directly to function_units_used because it will
+ trigger a fatal error for unrecognizable insns. */
+ if (INSN_CODE (insn) < 0)
+ unit = -1;
+ else
{
- return;
+ unit = function_units_used (insn);
+ /* Increment non-negative values so we can cache zero. */
+ if (unit >= 0)
+ unit++;
}
-
- curr_edge = NEXT_OUT (curr_edge);
-
- if (fst_edge == curr_edge)
- break;
+ /* We only cache 16 bits of the result, so if the value is out of
+ range, don't cache it. */
+ if (FUNCTION_UNITS_SIZE < HOST_BITS_PER_SHORT
+ || unit >= 0
+ || (unit & ~((1 << (HOST_BITS_PER_SHORT - 1)) - 1)) == 0)
+ INSN_UNIT (insn) = unit;
}
+ return (unit > 0 ? unit - 1 : unit);
+}
- e = ++nr_edges;
+/* Compute the blockage range for executing INSN on UNIT. This caches
+ the value returned by the blockage_range_function for the unit.
+ These values are encoded in an int where the upper half gives the
+ minimum value and the lower half gives the maximum value. */
- FROM_BLOCK (e) = source;
- TO_BLOCK (e) = target;
+HAIFA_INLINE static unsigned int
+blockage_range (unit, insn)
+ int unit;
+ rtx insn;
+{
+ unsigned int blockage = INSN_BLOCKAGE (insn);
+ unsigned int range;
- if (OUT_EDGES (source))
+ if ((int) UNIT_BLOCKED (blockage) != unit + 1)
{
- next_edge = NEXT_OUT (OUT_EDGES (source));
- NEXT_OUT (OUT_EDGES (source)) = e;
- NEXT_OUT (e) = next_edge;
+ range = function_units[unit].blockage_range_function (insn);
+ /* We only cache the blockage range for one unit and then only if
+ the values fit. */
+ if (HOST_BITS_PER_INT >= UNIT_BITS + 2 * BLOCKAGE_BITS)
+ INSN_BLOCKAGE (insn) = ENCODE_BLOCKAGE (unit + 1, range);
}
else
- {
- OUT_EDGES (source) = e;
- NEXT_OUT (e) = e;
- }
+ range = BLOCKAGE_RANGE (blockage);
- if (IN_EDGES (target))
- {
- next_edge = NEXT_IN (IN_EDGES (target));
- NEXT_IN (IN_EDGES (target)) = e;
- NEXT_IN (e) = next_edge;
- }
- else
- {
- IN_EDGES (target) = e;
- NEXT_IN (e) = e;
- }
+ return range;
}
-/* 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] |= \
- 1 << (index % HOST_BITS_PER_WIDE_INT); \
-}
+/* A vector indexed by function unit instance giving the last insn to use
+ the unit. The value of the function unit instance index for unit U
+ instance I is (U + I * FUNCTION_UNITS_SIZE). */
+static rtx unit_last_insn[FUNCTION_UNITS_SIZE * MAX_MULTIPLICITY];
-/* 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] &= \
- ~(1 << (index%HOST_BITS_PER_WIDE_INT)); \
-}
+/* A vector indexed by function unit instance giving the minimum time when
+ the unit will unblock based on the maximum blockage cost. */
+static int unit_tick[FUNCTION_UNITS_SIZE * MAX_MULTIPLICITY];
+
+/* A vector indexed by function unit number giving the number of insns
+ that remain to use the unit. */
+static int unit_n_insns[FUNCTION_UNITS_SIZE];
-/* Check if the index'th bit in bitset set is on. */
+/* Access the unit_last_insn array. Used by the visualization code. */
-static char
-bitset_member (set, index, len)
- bitset set;
- int index, len;
+rtx
+get_unit_last_insn (instance)
+ int instance;
{
- if (index >= HOST_BITS_PER_WIDE_INT * len)
- abort ();
- return (set[index / HOST_BITS_PER_WIDE_INT] &
- 1 << (index % HOST_BITS_PER_WIDE_INT)) ? 1 : 0;
+ return unit_last_insn[instance];
}
-/* Translate a bit-set SET to a list BL of the bit-set members. */
+/* Reset the function unit state to the null state. */
static void
-extract_bitlst (set, len, bitlen, bl)
- bitset set;
- int len;
- int bitlen;
- bitlst *bl;
+clear_units ()
{
- int i, j, offset;
- unsigned HOST_WIDE_INT word;
+ memset ((char *) unit_last_insn, 0, sizeof (unit_last_insn));
+ memset ((char *) unit_tick, 0, sizeof (unit_tick));
+ memset ((char *) unit_n_insns, 0, sizeof (unit_n_insns));
+}
- /* bblst table space is reused in each call to extract_bitlst. */
- bitlst_table_last = 0;
+/* Return the issue-delay of an insn. */
- bl->first_member = &bitlst_table[bitlst_table_last];
- bl->nr_members = 0;
+HAIFA_INLINE int
+insn_issue_delay (insn)
+ rtx insn;
+{
+ int i, delay = 0;
+ int unit = insn_unit (insn);
- /* Iterate over each word in the bitset. */
- for (i = 0; i < len; i++)
+ /* Efficiency note: in fact, we are working 'hard' to compute a
+ value that was available in md file, and is not available in
+ function_units[] structure. It would be nice to have this
+ value there, too. */
+ if (unit >= 0)
{
- 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;
- }
+ if (function_units[unit].blockage_range_function &&
+ function_units[unit].blockage_function)
+ delay = function_units[unit].blockage_function (insn, insn);
}
+ else
+ for (i = 0, unit = ~unit; unit; i++, unit >>= 1)
+ if ((unit & 1) != 0 && function_units[i].blockage_range_function
+ && function_units[i].blockage_function)
+ delay = MAX (delay, function_units[i].blockage_function (insn, insn));
+ return delay;
}
-/* Functions for the construction of regions. */
-
-/* Print the regions, for debugging purposes. Callable from debugger. */
+/* Return the actual hazard cost of executing INSN on the unit UNIT,
+ instance INSTANCE at time CLOCK if the previous actual hazard cost
+ was COST. */
-void
-debug_regions ()
+HAIFA_INLINE int
+actual_hazard_this_instance (unit, instance, insn, clock, cost)
+ int unit, instance, clock, cost;
+ rtx insn;
{
- int rgn, bb;
+ int tick = unit_tick[instance]; /* Issue time of the last issued insn. */
- fprintf (sched_dump, "\n;; ------------ REGIONS ----------\n\n");
- for (rgn = 0; rgn < nr_regions; rgn++)
+ if (tick - clock > cost)
{
- fprintf (sched_dump, ";;\trgn %d nr_blocks %d:\n", rgn,
- rgn_table[rgn].rgn_nr_blocks);
- fprintf (sched_dump, ";;\tbb/block: ");
-
- for (bb = 0; bb < rgn_table[rgn].rgn_nr_blocks; bb++)
- {
- current_blocks = RGN_BLOCKS (rgn);
+ /* The scheduler is operating forward, so unit's last insn is the
+ executing insn and INSN is the candidate insn. We want a
+ more exact measure of the blockage if we execute INSN at CLOCK
+ given when we committed the execution of the unit's last insn.
- if (bb != BLOCK_TO_BB (BB_TO_BLOCK (bb)))
- abort ();
+ The blockage value is given by either the unit's max blockage
+ constant, blockage range function, or blockage function. Use
+ the most exact form for the given unit. */
- fprintf (sched_dump, " %d/%d ", bb, BB_TO_BLOCK (bb));
+ if (function_units[unit].blockage_range_function)
+ {
+ if (function_units[unit].blockage_function)
+ tick += (function_units[unit].blockage_function
+ (unit_last_insn[instance], insn)
+ - function_units[unit].max_blockage);
+ else
+ tick += ((int) MAX_BLOCKAGE_COST (blockage_range (unit, insn))
+ - function_units[unit].max_blockage);
}
-
- fprintf (sched_dump, "\n\n");
+ if (tick - clock > cost)
+ cost = tick - clock;
}
+ return cost;
}
-/* Build a single block region for each basic block in the function.
- This allows for using the same code for interblock and basic block
- scheduling. */
-
-static void
-find_single_block_region ()
-{
- int i;
-
- for (i = 0; i < n_basic_blocks; i++)
- {
- rgn_bb_table[i] = i;
- RGN_NR_BLOCKS (i) = 1;
- RGN_BLOCKS (i) = i;
- CONTAINING_RGN (i) = i;
- BLOCK_TO_BB (i) = 0;
- }
- nr_regions = n_basic_blocks;
-}
-
-/* Update number of blocks and the estimate for number of insns
- in the region. Return 1 if the region is "too large" for interblock
- scheduling (compile time considerations), otherwise return 0. */
-
-static int
-too_large (block, num_bbs, num_insns)
- int block, *num_bbs, *num_insns;
-{
- (*num_bbs)++;
- (*num_insns) += (INSN_LUID (BLOCK_END (block)) -
- INSN_LUID (BLOCK_HEAD (block)));
- if ((*num_bbs > MAX_RGN_BLOCKS) || (*num_insns > MAX_RGN_INSNS))
- return 1;
- else
- return 0;
-}
-
-/* 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; \
- } \
-}
-
-/* Find regions for interblock scheduling.
-
- A region for scheduling can be:
-
- * A loop-free procedure, or
-
- * A reducible inner loop, or
-
- * A basic block not contained in any other region.
-
- ?!? In theory we could build other regions based on extended basic
- blocks or reverse extended basic blocks. Is it worth the trouble?
-
- Loop blocks that form a region are put into the region's block list
- in topological order.
-
- This procedure stores its results into the following global (ick) variables
-
- * rgn_nr
- * rgn_table
- * rgn_bb_table
- * block_to_bb
- * containing region
-
- We use dominator relationships to avoid making regions out of non-reducible
- loops.
-
- This procedure needs to be converted to work on pred/succ lists instead
- of edge tables. That would simplify it somewhat. */
-
-static void
-find_rgns (edge_list, dom)
- struct edge_list *edge_list;
- sbitmap *dom;
-{
- 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 num_bbs, num_insns, unreachable;
- int too_large_failure;
-
- /* Note if an edge has been passed. */
- sbitmap passed;
-
- /* Note if a block is a natural loop header. */
- sbitmap header;
-
- /* Note if a block is an natural inner loop header. */
- sbitmap inner;
-
- /* Note if a block is in the block queue. */
- sbitmap in_queue;
-
- /* Note if a block is in the block queue. */
- sbitmap in_stack;
-
- int num_edges = NUM_EDGES (edge_list);
-
- /* Perform a DFS traversal of the cfg. Identify loop headers, inner loops
- and a mapping from block to its loop header (if the block is contained
- in a loop, else -1).
-
- Store results in HEADER, INNER, and MAX_HDR respectively, these will
- be used as inputs to the second traversal.
-
- 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));
-
- inner = sbitmap_alloc (n_basic_blocks);
- sbitmap_ones (inner);
-
- header = sbitmap_alloc (n_basic_blocks);
- sbitmap_zero (header);
-
- passed = sbitmap_alloc (nr_edges);
- sbitmap_zero (passed);
-
- in_queue = sbitmap_alloc (n_basic_blocks);
- sbitmap_zero (in_queue);
-
- in_stack = sbitmap_alloc (n_basic_blocks);
- sbitmap_zero (in_stack);
-
- for (i = 0; i < n_basic_blocks; i++)
- max_hdr[i] = -1;
-
- /* DFS traversal to find inner loops in the cfg. */
-
- sp = -1;
- while (1)
- {
- if (current_edge == 0 || TEST_BIT (passed, current_edge))
- {
- /* We have reached a leaf node or a node that was already
- processed. Pop edges off the stack until we find
- an edge that has not yet been processed. */
- while (sp >= 0
- && (current_edge == 0 || TEST_BIT (passed, current_edge)))
- {
- /* Pop entry off the stack. */
- current_edge = stack[sp--];
- node = FROM_BLOCK (current_edge);
- child = TO_BLOCK (current_edge);
- RESET_BIT (in_stack, child);
- if (max_hdr[child] >= 0 && TEST_BIT (in_stack, max_hdr[child]))
- UPDATE_LOOP_RELATIONS (node, max_hdr[child]);
- current_edge = NEXT_OUT (current_edge);
- }
-
- /* See if have finished the DFS tree traversal. */
- if (sp < 0 && TEST_BIT (passed, current_edge))
- break;
-
- /* Nope, continue the traversal with the popped node. */
- continue;
- }
-
- /* Process a node. */
- node = FROM_BLOCK (current_edge);
- child = TO_BLOCK (current_edge);
- SET_BIT (in_stack, node);
- dfs_nr[node] = ++count;
-
- /* If the successor is in the stack, then we've found a loop.
- Mark the loop, if it is not a natural loop, then it will
- be rejected during the second traversal. */
- if (TEST_BIT (in_stack, child))
- {
- no_loops = 0;
- SET_BIT (header, child);
- UPDATE_LOOP_RELATIONS (node, child);
- SET_BIT (passed, current_edge);
- current_edge = NEXT_OUT (current_edge);
- continue;
- }
-
- /* If the child was already visited, then there is no need to visit
- it again. Just update the loop relationships and restart
- with a new edge. */
- if (dfs_nr[child])
- {
- if (max_hdr[child] >= 0 && TEST_BIT (in_stack, max_hdr[child]))
- UPDATE_LOOP_RELATIONS (node, max_hdr[child]);
- SET_BIT (passed, current_edge);
- current_edge = NEXT_OUT (current_edge);
- continue;
- }
-
- /* Push an entry on the stack and continue DFS traversal. */
- stack[++sp] = current_edge;
- SET_BIT (passed, current_edge);
- current_edge = OUT_EDGES (child);
-
- /* This is temporary until haifa is converted to use rth's new
- cfg routines which have true entry/exit blocks and the
- appropriate edges from/to those blocks.
-
- Generally we update dfs_nr for a node when we process its
- out edge. However, if the node has no out edge then we will
- not set dfs_nr for that node. This can confuse the scheduler
- into thinking that we have unreachable blocks, which in turn
- disables cross block scheduling.
-
- So, if we have a node with no out edges, go ahead and mark it
- as reachable now. */
- if (current_edge == 0)
- dfs_nr[child] = ++count;
- }
-
- /* Another check for unreachable blocks. The earlier test in
- is_cfg_nonregular only finds unreachable blocks that do not
- form a loop.
-
- The DFS traversal will mark every block that is reachable from
- 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)
- {
- unreachable = 1;
- break;
- }
-
- /* Gross. To avoid wasting memory, the second pass uses the dfs_nr array
- to hold degree counts. */
- degree = dfs_nr;
-
- for (i = 0; i < n_basic_blocks; i++)
- degree[i] = 0;
- for (i = 0; i < num_edges; i++)
- {
- edge e = INDEX_EDGE (edge_list, i);
-
- if (e->dest != EXIT_BLOCK_PTR)
- degree[e->dest->index]++;
- }
-
- /* Do not perform region scheduling if there are any unreachable
- blocks. */
- if (!unreachable)
- {
- int *queue;
-
- if (no_loops)
- SET_BIT (header, 0);
-
- /* Second travsersal:find reducible inner loops and topologically sort
- block of each region. */
-
- queue = (int *) 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++)
- {
- if (TEST_BIT (header, i) && TEST_BIT (inner, i))
- {
- edge e;
- int j;
-
- /* Now check that the loop is reducible. We do this separate
- from finding inner loops so that we do not find a reducible
- loop which contains an inner non-reducible loop.
-
- A simple way to find reducible/natural loops is to verify
- that each block in the loop is dominated by the loop
- header.
-
- 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++)
- {
- /* First identify blocks in the loop, except for the loop
- entry block. */
- if (i == max_hdr[j] && i != j)
- {
- /* Now verify that the block is dominated by the loop
- header. */
- if (!TEST_BIT (dom[j], i))
- break;
- }
- }
-
- /* 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)
- 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];
-
- /* Decrease degree of all I's successors for topological
- ordering. */
- for (e = BASIC_BLOCK (i)->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)));
-
- /* Find all loop latches (blocks with back edges to the loop
- header) or all the leaf blocks in the cfg has no loops.
-
- Place those blocks into the queue. */
- if (no_loops)
- {
- for (j = 0; j < n_basic_blocks; j++)
- /* 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)
- {
- queue[++tail] = j;
- SET_BIT (in_queue, j);
-
- if (too_large (j, &num_bbs, &num_insns))
- {
- too_large_failure = 1;
- break;
- }
- }
- }
- else
- {
- edge e;
-
- for (e = BASIC_BLOCK (i)->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)
- {
- /* This is a loop latch. */
- queue[++tail] = node;
- SET_BIT (in_queue, node);
-
- if (too_large (node, &num_bbs, &num_insns))
- {
- too_large_failure = 1;
- break;
- }
- }
- }
- }
-
- /* Now add all the blocks in the loop to the queue.
-
- We know the loop is a natural loop; however the algorithm
- above will not always mark certain blocks as being in the
- loop. Consider:
- node children
- a b,c
- b c
- c a,d
- d b
-
- The algorithm in the DFS traversal may not mark B & D as part
- of the loop (ie they will not have max_hdr set to A).
-
- We know they can not be loop latches (else they would have
- had max_hdr set since they'd have a backedge to a dominator
- block). So we don't need them on the initial queue.
-
- We know they are part of the loop because they are dominated
- by the loop header and can be reached by a backwards walk of
- the edges starting with nodes on the initial queue.
-
- It is safe and desirable to include those nodes in the
- loop/scheduling region. To do so we would need to decrease
- the degree of a node if it is the target of a backedge
- within the loop itself as the node is placed in the queue.
-
- We do not do this because I'm not sure that the actual
- scheduling code will properly handle this case. ?!? */
-
- while (head < tail && !too_large_failure)
- {
- edge e;
- child = queue[++head];
-
- for (e = BASIC_BLOCK (child)->pred; e; e = e->pred_next)
- {
- node = e->src->index;
-
- /* See discussion above about nodes not marked as in
- this loop during the initial DFS traversal. */
- if (e->src == ENTRY_BLOCK_PTR
- || max_hdr[node] != loop_head)
- {
- tail = -1;
- break;
- }
- else if (!TEST_BIT (in_queue, node) && node != i)
- {
- queue[++tail] = node;
- SET_BIT (in_queue, node);
-
- if (too_large (node, &num_bbs, &num_insns))
- {
- too_large_failure = 1;
- break;
- }
- }
- }
- }
-
- if (tail >= 0 && !too_large_failure)
- {
- /* Place the loop header into list of region blocks. */
- degree[i] = -1;
- rgn_bb_table[idx] = i;
- RGN_NR_BLOCKS (nr_regions) = num_bbs;
- RGN_BLOCKS (nr_regions) = idx++;
- CONTAINING_RGN (i) = nr_regions;
- BLOCK_TO_BB (i) = count = 0;
-
- /* Remove blocks from queue[] when their in degree
- becomes zero. Repeat until no blocks are left on the
- list. This produces a topological list of blocks in
- the region. */
- while (tail >= 0)
- {
- if (head < 0)
- head = tail;
- child = queue[head];
- if (degree[child] == 0)
- {
- edge e;
-
- degree[child] = -1;
- rgn_bb_table[idx++] = child;
- BLOCK_TO_BB (child) = ++count;
- CONTAINING_RGN (child) = nr_regions;
- queue[head] = queue[tail--];
-
- for (e = BASIC_BLOCK (child)->succ;
- e;
- e = e->succ_next)
- if (e->dest != EXIT_BLOCK_PTR)
- --degree[e->dest->index];
- }
- else
- --head;
- }
- ++nr_regions;
- }
- }
- }
- free (queue);
- }
-
- /* 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)
- {
- rgn_bb_table[idx] = i;
- RGN_NR_BLOCKS (nr_regions) = 1;
- RGN_BLOCKS (nr_regions) = idx++;
- CONTAINING_RGN (i) = nr_regions++;
- BLOCK_TO_BB (i) = 0;
- }
-
- free (max_hdr);
- free (dfs_nr);
- free (stack);
- free (passed);
- free (header);
- free (inner);
- free (in_queue);
- free (in_stack);
-}
-
-/* Functions for regions scheduling information. */
-
-/* Compute dominators, probability, and potential-split-edges of bb.
- Assume that these values were already computed for bb's predecessors. */
-
-static void
-compute_dom_prob_ps (bb)
- int bb;
-{
- int nxt_in_edge, fst_in_edge, pred;
- int fst_out_edge, nxt_out_edge, nr_out_edges, nr_rgn_out_edges;
-
- prob[bb] = 0.0;
- if (IS_RGN_ENTRY (bb))
- {
- BITSET_ADD (dom[bb], 0, bbset_size);
- 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);
-
- 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);
-
- BITSET_ADD (ancestor_edges[bb], EDGE_TO_BIT (nxt_in_edge), edgeset_size);
-
- 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);
-
- /* The successor doesn't belong in the region? */
- if (CONTAINING_RGN (TO_BLOCK (fst_out_edge)) !=
- CONTAINING_RGN (BB_TO_BLOCK (bb)))
- ++nr_rgn_out_edges;
-
- while (fst_out_edge != nxt_out_edge)
- {
- ++nr_out_edges;
- /* The successor doesn't belong in the region? */
- 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);
- nxt_out_edge = NEXT_OUT (nxt_out_edge);
-
- }
-
- /* Now nr_rgn_out_edges is the number of region-exit edges from
- pred, and nr_out_edges will be the number of pred out edges
- not leaving the region. */
- nr_out_edges -= nr_rgn_out_edges;
- if (nr_rgn_out_edges > 0)
- prob[bb] += 0.9 * prob[BLOCK_TO_BB (pred)] / nr_out_edges;
- else
- prob[bb] += prob[BLOCK_TO_BB (pred)] / nr_out_edges;
- nxt_in_edge = NEXT_IN (nxt_in_edge);
- }
- while (fst_in_edge != nxt_in_edge);
-
- BITSET_ADD (dom[bb], bb, bbset_size);
- BITSET_DIFFER (pot_split[bb], ancestor_edges[bb], edgeset_size);
-
- if (sched_verbose >= 2)
- fprintf (sched_dump, ";; bb_prob(%d, %d) = %3d\n", bb, BB_TO_BLOCK (bb),
- (int) (100.0 * prob[bb]));
-}
-
-/* Functions for target info. */
-
-/* Compute in BL the list of split-edges of bb_src relatively to bb_trg.
- Note that bb_trg dominates bb_src. */
-
-static void
-split_edges (bb_src, bb_trg, bl)
- 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);
-}
-
-/* Find the valid candidate-source-blocks for the target block TRG, compute
- their probability, and check if they are speculative or not.
- For speculative sources, compute their update-blocks and split-blocks. */
-
-static void
-compute_trg_info (trg)
- int trg;
-{
- register candidate *sp;
- edgelst el;
- int check_block, update_idx;
- int i, j, k, fst_edge, nxt_edge;
-
- /* Define some of the fields for the target bb as well. */
- sp = candidate_table + trg;
- sp->is_valid = 1;
- sp->is_speculative = 0;
- sp->src_prob = 100;
-
- for (i = trg + 1; i < current_nr_blocks; i++)
- {
- sp = candidate_table + i;
-
- sp->is_valid = IS_DOMINATED (i, trg);
- if (sp->is_valid)
- {
- sp->src_prob = GET_SRC_PROB (i, trg);
- sp->is_valid = (sp->src_prob >= MIN_PROBABILITY);
- }
-
- if (sp->is_valid)
- {
- split_edges (i, trg, &el);
- sp->is_speculative = (el.nr_members) ? 1 : 0;
- if (sp->is_speculative && !flag_schedule_speculative)
- sp->is_valid = 0;
- }
-
- if (sp->is_valid)
- {
- char *update_blocks;
-
- /* Compute split blocks and store them in bblst_table.
- The TO block of every split edge is a split block. */
- sp->split_bbs.first_member = &bblst_table[bblst_last];
- sp->split_bbs.nr_members = el.nr_members;
- for (j = 0; j < el.nr_members; bblst_last++, j++)
- bblst_table[bblst_last] =
- TO_BLOCK (rgn_edges[el.first_member[j]]);
- sp->update_bbs.first_member = &bblst_table[bblst_last];
-
- /* Compute update blocks and store them in bblst_table.
- For every split edge, look at the FROM block, and check
- all out edges. For each out edge that is not a split edge,
- 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_idx = 0;
- for (j = 0; j < el.nr_members; j++)
- {
- check_block = FROM_BLOCK (rgn_edges[el.first_member[j]]);
- fst_edge = nxt_edge = OUT_EDGES (check_block);
- do
- {
- if (! update_blocks[TO_BLOCK (nxt_edge)])
- {
- for (k = 0; k < el.nr_members; k++)
- if (EDGE_TO_BIT (nxt_edge) == el.first_member[k])
- break;
-
- if (k >= el.nr_members)
- {
- bblst_table[bblst_last++] = TO_BLOCK (nxt_edge);
- update_blocks[TO_BLOCK (nxt_edge)] = 1;
- update_idx++;
- }
- }
-
- nxt_edge = NEXT_OUT (nxt_edge);
- }
- while (fst_edge != nxt_edge);
- }
- sp->update_bbs.nr_members = update_idx;
-
- /* Make sure we didn't overrun the end of bblst_table. */
- if (bblst_last > bblst_size)
- abort ();
- }
- else
- {
- sp->split_bbs.nr_members = sp->update_bbs.nr_members = 0;
-
- sp->is_speculative = 0;
- sp->src_prob = 0;
- }
- }
-}
-
-/* Print candidates info, for debugging purposes. Callable from debugger. */
-
-void
-debug_candidate (i)
- int i;
-{
- if (!candidate_table[i].is_valid)
- return;
-
- if (candidate_table[i].is_speculative)
- {
- int j;
- fprintf (sched_dump, "src b %d bb %d speculative \n", BB_TO_BLOCK (i), i);
-
- fprintf (sched_dump, "split path: ");
- for (j = 0; j < candidate_table[i].split_bbs.nr_members; j++)
- {
- int b = candidate_table[i].split_bbs.first_member[j];
-
- fprintf (sched_dump, " %d ", b);
- }
- fprintf (sched_dump, "\n");
-
- fprintf (sched_dump, "update path: ");
- for (j = 0; j < candidate_table[i].update_bbs.nr_members; j++)
- {
- int b = candidate_table[i].update_bbs.first_member[j];
-
- fprintf (sched_dump, " %d ", b);
- }
- fprintf (sched_dump, "\n");
- }
- else
- {
- fprintf (sched_dump, " src %d equivalent\n", BB_TO_BLOCK (i));
- }
-}
-
-/* Print candidates info, for debugging purposes. Callable from debugger. */
-
-void
-debug_candidates (trg)
- int trg;
-{
- int i;
-
- fprintf (sched_dump, "----------- candidate table: target: b=%d bb=%d ---\n",
- BB_TO_BLOCK (trg), trg);
- for (i = trg + 1; i < current_nr_blocks; i++)
- debug_candidate (i);
-}
-
-/* Functions for speculative scheduing. */
-
-/* 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;
-{
- register int i;
- register int regno;
- register rtx reg = SET_DEST (x);
-
- if (reg == 0)
- return 1;
-
- while (GET_CODE (reg) == SUBREG || GET_CODE (reg) == ZERO_EXTRACT
- || GET_CODE (reg) == SIGN_EXTRACT
- || GET_CODE (reg) == STRICT_LOW_PART)
- reg = XEXP (reg, 0);
-
- if (GET_CODE (reg) == PARALLEL
- && GET_MODE (reg) == BLKmode)
- {
- register int i;
- for (i = XVECLEN (reg, 0) - 1; i >= 0; i--)
- if (check_live_1 (src, XVECEXP (reg, 0, i)))
- return 1;
- return 0;
- }
-
- if (GET_CODE (reg) != REG)
- return 1;
-
- regno = REGNO (reg);
-
- if (regno < FIRST_PSEUDO_REGISTER && global_regs[regno])
- {
- /* Global registers are assumed live. */
- return 0;
- }
- else
- {
- if (regno < FIRST_PSEUDO_REGISTER)
- {
- /* Check for hard registers. */
- int j = HARD_REGNO_NREGS (regno, GET_MODE (reg));
- while (--j >= 0)
- {
- for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++)
- {
- int b = candidate_table[src].split_bbs.first_member[i];
-
- if (REGNO_REG_SET_P (BASIC_BLOCK (b)->global_live_at_start,
- regno + j))
- {
- return 0;
- }
- }
- }
- }
- else
- {
- /* Check for psuedo registers. */
- for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++)
- {
- int b = candidate_table[src].split_bbs.first_member[i];
-
- if (REGNO_REG_SET_P (BASIC_BLOCK (b)->global_live_at_start, regno))
- {
- return 0;
- }
- }
- }
- }
-
- return 1;
-}
-
-/* If x is a set of a register R, mark that R is alive in the beginning
- of every update-block of src. */
-
-static void
-update_live_1 (src, x)
- int src;
- rtx x;
-{
- register int i;
- register int regno;
- register rtx reg = SET_DEST (x);
-
- if (reg == 0)
- return;
-
- while (GET_CODE (reg) == SUBREG || GET_CODE (reg) == ZERO_EXTRACT
- || GET_CODE (reg) == SIGN_EXTRACT
- || GET_CODE (reg) == STRICT_LOW_PART)
- reg = XEXP (reg, 0);
-
- if (GET_CODE (reg) == PARALLEL
- && GET_MODE (reg) == BLKmode)
- {
- register int i;
- for (i = XVECLEN (reg, 0) - 1; i >= 0; i--)
- update_live_1 (src, XVECEXP (reg, 0, i));
- return;
- }
-
- if (GET_CODE (reg) != REG)
- return;
-
- /* Global registers are always live, so the code below does not apply
- to them. */
-
- regno = REGNO (reg);
-
- if (regno >= FIRST_PSEUDO_REGISTER || !global_regs[regno])
- {
- if (regno < FIRST_PSEUDO_REGISTER)
- {
- int j = HARD_REGNO_NREGS (regno, GET_MODE (reg));
- while (--j >= 0)
- {
- for (i = 0; i < candidate_table[src].update_bbs.nr_members; i++)
- {
- int b = candidate_table[src].update_bbs.first_member[i];
-
- SET_REGNO_REG_SET (BASIC_BLOCK (b)->global_live_at_start,
- regno + j);
- }
- }
- }
- else
- {
- for (i = 0; i < candidate_table[src].update_bbs.nr_members; i++)
- {
- int b = candidate_table[src].update_bbs.first_member[i];
-
- SET_REGNO_REG_SET (BASIC_BLOCK (b)->global_live_at_start, regno);
- }
- }
- }
-}
-
-/* Return 1 if insn can be speculatively moved from block src to trg,
- otherwise return 0. Called before first insertion of insn to
- ready-list or before the scheduling. */
-
-static int
-check_live (insn, src)
- rtx insn;
- int src;
-{
- /* Find the registers set by instruction. */
- if (GET_CODE (PATTERN (insn)) == SET
- || GET_CODE (PATTERN (insn)) == CLOBBER)
- return check_live_1 (src, PATTERN (insn));
- else if (GET_CODE (PATTERN (insn)) == PARALLEL)
- {
- int j;
- for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
- if ((GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
- || GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
- && !check_live_1 (src, XVECEXP (PATTERN (insn), 0, j)))
- return 0;
-
- return 1;
- }
-
- return 1;
-}
-
-/* Update the live registers info after insn was moved speculatively from
- block src to trg. */
-
-static void
-update_live (insn, src)
- rtx insn;
- int src;
-{
- /* Find the registers set by instruction. */
- if (GET_CODE (PATTERN (insn)) == SET
- || GET_CODE (PATTERN (insn)) == CLOBBER)
- update_live_1 (src, PATTERN (insn));
- else if (GET_CODE (PATTERN (insn)) == PARALLEL)
- {
- int j;
- for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
- if (GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
- || GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
- update_live_1 (src, XVECEXP (PATTERN (insn), 0, j));
- }
-}
-
-/* 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. */
-#define IS_REACHABLE(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)))
-
-/* Turns on the fed_by_spec_load flag for insns fed by load_insn. */
-
-static void
-set_spec_fed (load_insn)
- rtx load_insn;
-{
- rtx link;
-
- for (link = INSN_DEPEND (load_insn); link; link = XEXP (link, 1))
- if (GET_MODE (link) == VOIDmode)
- FED_BY_SPEC_LOAD (XEXP (link, 0)) = 1;
-} /* set_spec_fed */
-
-/* On the path from the insn to load_insn_bb, find a conditional
-branch depending on insn, that guards the speculative load. */
-
-static int
-find_conditional_protection (insn, load_insn_bb)
- rtx insn;
- int load_insn_bb;
-{
- rtx link;
-
- /* Iterate through DEF-USE forward dependences. */
- for (link = INSN_DEPEND (insn); link; link = XEXP (link, 1))
- {
- rtx next = XEXP (link, 0);
- if ((CONTAINING_RGN (BLOCK_NUM (next)) ==
- CONTAINING_RGN (BB_TO_BLOCK (load_insn_bb)))
- && IS_REACHABLE (INSN_BB (next), load_insn_bb)
- && load_insn_bb != INSN_BB (next)
- && GET_MODE (link) == VOIDmode
- && (GET_CODE (next) == JUMP_INSN
- || find_conditional_protection (next, load_insn_bb)))
- return 1;
- }
- return 0;
-} /* find_conditional_protection */
-
-/* Returns 1 if the same insn1 that participates in the computation
- of load_insn's address is feeding a conditional branch that is
- guarding on load_insn. This is true if we find a the two DEF-USE
- chains:
- insn1 -> ... -> conditional-branch
- insn1 -> ... -> load_insn,
- and if a flow path exist:
- insn1 -> ... -> conditional-branch -> ... -> load_insn,
- and if insn1 is on the path
- region-entry -> ... -> bb_trg -> ... load_insn.
-
- Locate insn1 by climbing on LOG_LINKS from load_insn.
- 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;
-{
- rtx link;
-
- for (link = LOG_LINKS (load_insn); link; link = XEXP (link, 1))
- {
- rtx insn1 = XEXP (link, 0);
-
- /* Must be a DEF-USE dependence upon non-branch. */
- if (GET_MODE (link) != VOIDmode
- || GET_CODE (insn1) == JUMP_INSN)
- continue;
-
- /* Must exist a path: region-entry -> ... -> bb_trg -> ... load_insn. */
- if (INSN_BB (insn1) == bb_src
- || (CONTAINING_RGN (BLOCK_NUM (insn1))
- != CONTAINING_RGN (BB_TO_BLOCK (bb_src)))
- || (!IS_REACHABLE (bb_trg, INSN_BB (insn1))
- && !IS_REACHABLE (INSN_BB (insn1), bb_trg)))
- continue;
-
- /* Now search for the conditional-branch. */
- if (find_conditional_protection (insn1, bb_src))
- return 1;
-
- /* Recursive step: search another insn1, "above" current insn1. */
- return is_conditionally_protected (insn1, bb_src, bb_trg);
- }
-
- /* The chain does not exist. */
- return 0;
-} /* is_conditionally_protected */
-
-/* Returns 1 if a clue for "similar load" 'insn2' is found, and hence
- load_insn can move speculatively from bb_src to bb_trg. All the
- following must hold:
-
- (1) both loads have 1 base register (PFREE_CANDIDATEs).
- (2) load_insn and load1 have a def-use dependence upon
- the same insn 'insn1'.
- (3) either load2 is in bb_trg, or:
- - there's only one split-block, and
- - load1 is on the escape path, and
-
- From all these we can conclude that the two loads access memory
- addresses that differ at most by a constant, and hence if moving
- load_insn would cause an exception, it would have been caused by
- load2 anyhow. */
-
-static int
-is_pfree (load_insn, bb_src, bb_trg)
- rtx load_insn;
- int bb_src, bb_trg;
-{
- rtx back_link;
- register candidate *candp = candidate_table + bb_src;
-
- if (candp->split_bbs.nr_members != 1)
- /* Must have exactly one escape block. */
- return 0;
-
- for (back_link = LOG_LINKS (load_insn);
- back_link; back_link = XEXP (back_link, 1))
- {
- rtx insn1 = XEXP (back_link, 0);
-
- if (GET_MODE (back_link) == VOIDmode)
- {
- /* Found a DEF-USE dependence (insn1, load_insn). */
- rtx fore_link;
-
- for (fore_link = INSN_DEPEND (insn1);
- fore_link; fore_link = XEXP (fore_link, 1))
- {
- rtx insn2 = XEXP (fore_link, 0);
- if (GET_MODE (fore_link) == VOIDmode)
- {
- /* Found a DEF-USE dependence (insn1, insn2). */
- if (haifa_classify_insn (insn2) != PFREE_CANDIDATE)
- /* insn2 not guaranteed to be a 1 base reg load. */
- continue;
-
- if (INSN_BB (insn2) == bb_trg)
- /* insn2 is the similar load, in the target block. */
- return 1;
-
- if (*(candp->split_bbs.first_member) == BLOCK_NUM (insn2))
- /* insn2 is a similar load, in a split-block. */
- return 1;
- }
- }
- }
- }
-
- /* Couldn't find a similar load. */
- 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;
-{
- if (FED_BY_SPEC_LOAD (load_insn))
- return 1;
-
- if (LOG_LINKS (load_insn) == NULL)
- /* Dependence may 'hide' out of the region. */
- return 1;
-
- if (is_conditionally_protected (load_insn, bb_src, bb_trg))
- return 1;
-
- return 0;
-}
-
-/* Insn is a candidate to be moved speculatively from bb_src to bb_trg.
- Return 1 if insn is exception-free (and the motion is valid)
- and 0 otherwise. */
-
-static int
-is_exception_free (insn, bb_src, bb_trg)
- rtx insn;
- int bb_src, bb_trg;
-{
- int insn_class = haifa_classify_insn (insn);
-
- /* Handle non-load insns. */
- switch (insn_class)
- {
- case TRAP_FREE:
- return 1;
- case TRAP_RISKY:
- return 0;
- default:;
- }
-
- /* Handle loads. */
- if (!flag_schedule_speculative_load)
- return 0;
- IS_LOAD_INSN (insn) = 1;
- switch (insn_class)
- {
- case IFREE:
- return (1);
- case IRISKY:
- return 0;
- case PFREE_CANDIDATE:
- if (is_pfree (insn, bb_src, bb_trg))
- return 1;
- /* Don't 'break' here: PFREE-candidate is also PRISKY-candidate. */
- case PRISKY_CANDIDATE:
- if (!flag_schedule_speculative_load_dangerous
- || is_prisky (insn, bb_src, bb_trg))
- return 0;
- break;
- default:;
- }
-
- return flag_schedule_speculative_load_dangerous;
-}
-\f
-/* Compute the function units used by INSN. This caches the value
- returned by function_units_used. A function unit is encoded as the
- unit number if the value is non-negative and the compliment of a
- mask if the value is negative. A function unit index is the
- non-negative encoding. */
-
-HAIFA_INLINE int
-insn_unit (insn)
- rtx insn;
-{
- register int unit = INSN_UNIT (insn);
-
- if (unit == 0)
- {
- recog_memoized (insn);
-
- /* A USE insn, or something else we don't need to understand.
- We can't pass these directly to function_units_used because it will
- trigger a fatal error for unrecognizable insns. */
- if (INSN_CODE (insn) < 0)
- unit = -1;
- else
- {
- unit = function_units_used (insn);
- /* Increment non-negative values so we can cache zero. */
- if (unit >= 0)
- unit++;
- }
- /* We only cache 16 bits of the result, so if the value is out of
- range, don't cache it. */
- if (FUNCTION_UNITS_SIZE < HOST_BITS_PER_SHORT
- || unit >= 0
- || (unit & ~((1 << (HOST_BITS_PER_SHORT - 1)) - 1)) == 0)
- INSN_UNIT (insn) = unit;
- }
- return (unit > 0 ? unit - 1 : unit);
-}
-
-/* Compute the blockage range for executing INSN on UNIT. This caches
- the value returned by the blockage_range_function for the unit.
- These values are encoded in an int where the upper half gives the
- minimum value and the lower half gives the maximum value. */
-
-HAIFA_INLINE static unsigned int
-blockage_range (unit, insn)
- int unit;
- rtx insn;
-{
- unsigned int blockage = INSN_BLOCKAGE (insn);
- unsigned int range;
-
- if ((int) UNIT_BLOCKED (blockage) != unit + 1)
- {
- range = function_units[unit].blockage_range_function (insn);
- /* We only cache the blockage range for one unit and then only if
- the values fit. */
- if (HOST_BITS_PER_INT >= UNIT_BITS + 2 * BLOCKAGE_BITS)
- INSN_BLOCKAGE (insn) = ENCODE_BLOCKAGE (unit + 1, range);
- }
- else
- range = BLOCKAGE_RANGE (blockage);
-
- return range;
-}
-
-/* A vector indexed by function unit instance giving the last insn to use
- the unit. The value of the function unit instance index for unit U
- instance I is (U + I * FUNCTION_UNITS_SIZE). */
-static rtx unit_last_insn[FUNCTION_UNITS_SIZE * MAX_MULTIPLICITY];
-
-/* A vector indexed by function unit instance giving the minimum time when
- the unit will unblock based on the maximum blockage cost. */
-static int unit_tick[FUNCTION_UNITS_SIZE * MAX_MULTIPLICITY];
-
-/* A vector indexed by function unit number giving the number of insns
- that remain to use the unit. */
-static int unit_n_insns[FUNCTION_UNITS_SIZE];
-
-/* Access the unit_last_insn array. Used by the visualization code. */
-
-rtx
-get_unit_last_insn (instance)
- int instance;
-{
- return unit_last_insn[instance];
-}
-
-/* Reset the function unit state to the null state. */
-
-static void
-clear_units ()
-{
- memset ((char *) unit_last_insn, 0, sizeof (unit_last_insn));
- memset ((char *) unit_tick, 0, sizeof (unit_tick));
- memset ((char *) unit_n_insns, 0, sizeof (unit_n_insns));
-}
-
-/* Return the issue-delay of an insn. */
-
-HAIFA_INLINE static int
-insn_issue_delay (insn)
- rtx insn;
-{
- int i, delay = 0;
- int unit = insn_unit (insn);
-
- /* Efficiency note: in fact, we are working 'hard' to compute a
- value that was available in md file, and is not available in
- function_units[] structure. It would be nice to have this
- value there, too. */
- if (unit >= 0)
- {
- if (function_units[unit].blockage_range_function &&
- function_units[unit].blockage_function)
- delay = function_units[unit].blockage_function (insn, insn);
- }
- else
- for (i = 0, unit = ~unit; unit; i++, unit >>= 1)
- if ((unit & 1) != 0 && function_units[i].blockage_range_function
- && function_units[i].blockage_function)
- delay = MAX (delay, function_units[i].blockage_function (insn, insn));
-
- return delay;
-}
-
-/* Return the actual hazard cost of executing INSN on the unit UNIT,
- instance INSTANCE at time CLOCK if the previous actual hazard cost
- was COST. */
-
-HAIFA_INLINE int
-actual_hazard_this_instance (unit, instance, insn, clock, cost)
- int unit, instance, clock, cost;
- rtx insn;
-{
- int tick = unit_tick[instance]; /* Issue time of the last issued insn. */
-
- if (tick - clock > cost)
- {
- /* The scheduler is operating forward, so unit's last insn is the
- executing insn and INSN is the candidate insn. We want a
- more exact measure of the blockage if we execute INSN at CLOCK
- given when we committed the execution of the unit's last insn.
-
- The blockage value is given by either the unit's max blockage
- constant, blockage range function, or blockage function. Use
- the most exact form for the given unit. */
-
- if (function_units[unit].blockage_range_function)
- {
- if (function_units[unit].blockage_function)
- tick += (function_units[unit].blockage_function
- (unit_last_insn[instance], insn)
- - function_units[unit].max_blockage);
- else
- tick += ((int) MAX_BLOCKAGE_COST (blockage_range (unit, insn))
- - function_units[unit].max_blockage);
- }
- if (tick - clock > cost)
- cost = tick - clock;
- }
- return cost;
-}
-
-/* Record INSN as having begun execution on the units encoded by UNIT at
- time CLOCK. */
+/* Record INSN as having begun execution on the units encoded by UNIT at
+ time CLOCK. */
HAIFA_INLINE static void
schedule_unit (unit, insn, clock)
This is the number of cycles between instruction issue and
instruction results. */
-HAIFA_INLINE static int
+HAIFA_INLINE int
insn_cost (insn, link, used)
rtx insn, link, used;
{
return this_priority;
}
\f
-/* Remove all INSN_LISTs and EXPR_LISTs from the pending lists and add
- them to the unused_*_list variables, so that they can be reused. */
-
-static void
-free_pending_lists ()
-{
- int bb;
-
- for (bb = 0; bb < current_nr_blocks; bb++)
- {
- free_INSN_LIST_list (&bb_deps[bb].pending_read_insns);
- free_INSN_LIST_list (&bb_deps[bb].pending_write_insns);
- free_EXPR_LIST_list (&bb_deps[bb].pending_read_mems);
- free_EXPR_LIST_list (&bb_deps[bb].pending_write_mems);
- }
-}
-\f
/* Macros and functions for keeping the priority queue sorted, and
dealing with queueing and dequeueing of instructions. */
/* Add an element INSN to the ready list so that it ends up with the lowest
priority. */
-HAIFA_INLINE static void
+HAIFA_INLINE void
ready_add (ready, insn)
struct ready_list *ready;
rtx insn;
/* Return the head and tail pointers of BB. */
-HAIFA_INLINE static void
+void
get_block_head_tail (b, headp, tailp)
int b;
rtx *headp;
*tailp = tail;
}
-HAIFA_INLINE static void
-get_bb_head_tail (bb, headp, tailp)
- int bb;
- rtx *headp;
- rtx *tailp;
-{
- get_block_head_tail (BB_TO_BLOCK (bb), headp, tailp);
-}
-
/* Return nonzero if there are no real insns in the range [ HEAD, TAIL ]. */
-static int
+int
no_real_insns_p (head, tail)
rtx head, tail;
{
/* Delete line notes from bb. Save them so they can be later restored
(in restore_line_notes ()). */
-static void
-rm_line_notes (bb)
- int bb;
+void
+rm_line_notes (b)
+ int b;
{
rtx next_tail;
rtx tail;
rtx head;
rtx insn;
- get_bb_head_tail (bb, &head, &tail);
+ get_block_head_tail (b, &head, &tail);
if (head == tail && (! INSN_P (head)))
return;
}
}
-/* Save line number notes for each insn in bb. */
+/* Save line number notes for each insn in block B. */
-static void
-save_line_notes (bb)
- int bb;
+void
+save_line_notes (b)
+ int b;
{
rtx head, tail;
rtx next_tail;
use the current line number, because scheduling of the previous
block may have changed the current line number. */
- rtx line = line_note_head[BB_TO_BLOCK (bb)];
+ rtx line = line_note_head[b];
rtx insn;
- get_bb_head_tail (bb, &head, &tail);
+ get_block_head_tail (b, &head, &tail);
next_tail = NEXT_INSN (tail);
- for (insn = BLOCK_HEAD (BB_TO_BLOCK (bb));
- insn != next_tail;
- insn = NEXT_INSN (insn))
+ for (insn = BLOCK_HEAD (b); insn != next_tail; insn = NEXT_INSN (insn))
if (GET_CODE (insn) == NOTE && NOTE_LINE_NUMBER (insn) > 0)
line = insn;
else
LINE_NOTE (insn) = line;
}
-/* After bb was scheduled, insert line notes into the insns list. */
+/* After block B was scheduled, insert line notes into the insns list. */
-static void
-restore_line_notes (bb)
- int bb;
+void
+restore_line_notes (b)
+ int b;
{
rtx line, note, prev, new;
int added_notes = 0;
- int b;
rtx head, next_tail, insn;
- b = BB_TO_BLOCK (bb);
-
head = BLOCK_HEAD (b);
next_tail = NEXT_INSN (BLOCK_END (b));
/* After scheduling the function, delete redundant line notes from the
insns list. */
-static void
+void
rm_redundant_line_notes ()
{
rtx line = 0;
/* Delete notes between head and tail and put them in the chain
of notes ended by NOTE_LIST. */
-static void
+void
rm_other_notes (head, tail)
rtx head;
rtx tail;
rtx next_tail;
rtx insn;
+ note_list = 0;
if (head == tail && (! INSN_P (head)))
return;
q_ptr = NEXT_Q (q_ptr);
/* Add all pending insns that can be scheduled without stalls to the
- ready list. */
- for (link = insn_queue[q_ptr]; link; link = XEXP (link, 1))
- {
- insn = XEXP (link, 0);
- q_size -= 1;
-
- if (sched_verbose >= 2)
- fprintf (sched_dump, ";;\t\tQ-->Ready: insn %s: ",
- (*current_sched_info->print_insn) (insn, 0));
-
- ready_add (ready, insn);
- if (sched_verbose >= 2)
- fprintf (sched_dump, "moving to ready without stalls\n");
- }
- insn_queue[q_ptr] = 0;
-
- /* If there are no ready insns, stall until one is ready and add all
- of the pending insns at that point to the ready list. */
- if (ready->n_ready == 0)
- {
- register int stalls;
-
- for (stalls = 1; stalls < INSN_QUEUE_SIZE; stalls++)
- {
- if ((link = insn_queue[NEXT_Q_AFTER (q_ptr, stalls)]))
- {
- for (; link; link = XEXP (link, 1))
- {
- insn = XEXP (link, 0);
- q_size -= 1;
-
- if (sched_verbose >= 2)
- fprintf (sched_dump, ";;\t\tQ-->Ready: insn %s: ",
- (*current_sched_info->print_insn) (insn, 0));
-
- ready_add (ready, insn);
- if (sched_verbose >= 2)
- fprintf (sched_dump, "moving to ready with %d stalls\n", stalls);
- }
- insn_queue[NEXT_Q_AFTER (q_ptr, stalls)] = 0;
-
- if (ready->n_ready)
- break;
- }
- }
-
- if (sched_verbose && stalls)
- visualize_stall_cycles (stalls);
- q_ptr = NEXT_Q_AFTER (q_ptr, stalls);
- clock_var += stalls;
- }
-}
-
-/* Print the ready list for debugging purposes. Callable from debugger. */
-
-static void
-debug_ready_list (ready)
- struct ready_list *ready;
-{
- rtx *p;
- int i;
-
- if (ready->n_ready == 0)
- return;
-
- p = ready_lastpos (ready);
- for (i = 0; i < ready->n_ready; i++)
- fprintf (sched_dump, " %s", (*current_sched_info->print_insn) (p[i], 0));
- fprintf (sched_dump, "\n");
-}
-
-/* The number of insns from the current block scheduled so far. */
-static int sched_target_n_insns;
-/* The number of insns from the current block to be scheduled in total. */
-static int target_n_insns;
-/* The number of insns from the entire region scheduled so far. */
-static int sched_n_insns;
-
-/* 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));
-
-/* Return nonzero if there are more insns that should be scheduled. */
-
-static int
-schedule_more_p ()
-{
- return sched_target_n_insns < target_n_insns;
-}
-
-/* Add all insns that are initially ready to the ready list READY. Called
- once before scheduling a set of insns. */
-
-static void
-init_ready_list (ready)
- struct ready_list *ready;
-{
- rtx prev_head = current_sched_info->prev_head;
- rtx next_tail = current_sched_info->next_tail;
- int bb_src;
- rtx insn;
-
- target_n_insns = 0;
- sched_target_n_insns = 0;
- sched_n_insns = 0;
-
- /* Print debugging information. */
- if (sched_verbose >= 5)
- debug_dependencies ();
-
- /* Prepare current target block info. */
- if (current_nr_blocks > 1)
- {
- candidate_table = (candidate *) xmalloc (current_nr_blocks
- * sizeof (candidate));
-
- bblst_last = 0;
- /* bblst_table holds split blocks and update blocks for each block after
- the current one in the region. split blocks and update blocks are
- 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));
-
- bitlst_table_last = 0;
- bitlst_table_size = rgn_nr_edges;
- bitlst_table = (int *) xmalloc (rgn_nr_edges * sizeof (int));
-
- compute_trg_info (target_bb);
- }
-
- /* Initialize ready list with all 'ready' insns in target block.
- 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
- && (SCHED_GROUP_P (next) == 0 || ! INSN_P (next)))
- ready_add (ready, insn);
- if (!(SCHED_GROUP_P (insn)))
- target_n_insns++;
- }
-
- /* Add to ready list all 'ready' insns in valid source blocks.
- For speculative insns, check-live, exception-free, and
- issue-delay. */
- for (bb_src = target_bb + 1; bb_src < current_nr_blocks; bb_src++)
- if (IS_VALID (bb_src))
- {
- rtx src_head;
- rtx src_next_tail;
- rtx tail, head;
-
- get_bb_head_tail (bb_src, &head, &tail);
- src_next_tail = NEXT_INSN (tail);
- src_head = head;
-
- for (insn = src_head; insn != src_next_tail; insn = NEXT_INSN (insn))
- {
- if (! INSN_P (insn))
- continue;
-
- if (!CANT_MOVE (insn)
- && (!IS_SPECULATIVE_INSN (insn)
- || (insn_issue_delay (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);
- }
- }
- }
-}
-
-/* Called after taking INSN from the ready list. Returns nonzero if this
- insn can be scheduled, nonzero if we should silently discard it. */
-
-static int
-can_schedule_ready_p (insn)
- rtx insn;
-{
- /* An interblock motion? */
- if (INSN_BB (insn) != target_bb)
- {
- rtx temp;
- basic_block b1;
-
- if (IS_SPECULATIVE_INSN (insn))
- {
- if (!check_live (insn, INSN_BB (insn)))
- return 0;
- update_live (insn, INSN_BB (insn));
-
- /* For speculative load, mark insns fed by it. */
- if (IS_LOAD_INSN (insn) || FED_BY_SPEC_LOAD (insn))
- set_spec_fed (insn);
-
- nr_spec++;
- }
- 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. */
+ ready list. */
+ for (link = insn_queue[q_ptr]; link; link = XEXP (link, 1))
+ {
+ insn = XEXP (link, 0);
+ q_size -= 1;
- temp = insn;
- while (SCHED_GROUP_P (temp))
- temp = PREV_INSN (temp);
+ if (sched_verbose >= 2)
+ fprintf (sched_dump, ";;\t\tQ-->Ready: insn %s: ",
+ (*current_sched_info->print_insn) (insn, 0));
- /* Update source block boundaries. */
- b1 = BLOCK_FOR_INSN (temp);
- if (temp == b1->head && insn == b1->end)
- {
- /* 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;
- }
- else if (insn == b1->end)
- {
- /* 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);
- }
- else if (temp == b1->head)
- {
- /* 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);
- }
+ ready_add (ready, insn);
+ if (sched_verbose >= 2)
+ fprintf (sched_dump, "moving to ready without stalls\n");
}
- else
+ insn_queue[q_ptr] = 0;
+
+ /* If there are no ready insns, stall until one is ready and add all
+ of the pending insns at that point to the ready list. */
+ if (ready->n_ready == 0)
{
- /* In block motion. */
- sched_target_n_insns++;
- }
- sched_n_insns++;
+ register int stalls;
- return 1;
-}
+ for (stalls = 1; stalls < INSN_QUEUE_SIZE; stalls++)
+ {
+ if ((link = insn_queue[NEXT_Q_AFTER (q_ptr, stalls)]))
+ {
+ for (; link; link = XEXP (link, 1))
+ {
+ insn = XEXP (link, 0);
+ q_size -= 1;
-/* Called after INSN has all its dependencies resolved. Return nonzero
- 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;
-{
- /* For speculative insns, before inserting to ready/queue,
- check live, exception-free, and issue-delay. */
- if (INSN_BB (next) != target_bb
- && (!IS_VALID (INSN_BB (next))
- || CANT_MOVE (next)
- || (IS_SPECULATIVE_INSN (next)
- && (insn_issue_delay (next) > 3
- || !check_live (next, INSN_BB (next))
- || !is_exception_free (next, INSN_BB (next), target_bb)))))
- return 0;
- return 1;
-}
+ if (sched_verbose >= 2)
+ fprintf (sched_dump, ";;\t\tQ-->Ready: insn %s: ",
+ (*current_sched_info->print_insn) (insn, 0));
-/* Return a string that contains the insn uid and optionally anything else
- necessary to identify this insn in an output. It's valid to use a
- static buffer for this. The ALIGNED parameter should cause the string
- to be formatted so that multiple output lines will line up nicely. */
+ ready_add (ready, insn);
+ if (sched_verbose >= 2)
+ fprintf (sched_dump, "moving to ready with %d stalls\n", stalls);
+ }
+ insn_queue[NEXT_Q_AFTER (q_ptr, stalls)] = 0;
-static const char *
-rgn_print_insn (insn, aligned)
- rtx insn;
- int aligned;
-{
- static char tmp[80];
+ if (ready->n_ready)
+ break;
+ }
+ }
- if (aligned)
- sprintf (tmp, "b%3d: i%4d", INSN_BB (insn), INSN_UID (insn));
- else
- {
- sprintf (tmp, "%d", INSN_UID (insn));
- if (current_nr_blocks > 1 && INSN_BB (insn) != target_bb)
- sprintf (tmp, "/b%d ", INSN_BB (insn));
+ if (sched_verbose && stalls)
+ visualize_stall_cycles (stalls);
+ q_ptr = NEXT_Q_AFTER (q_ptr, stalls);
+ clock_var += stalls;
}
- return tmp;
}
-/* Compare priority of two insns. Return a positive number if the second
- insn is to be preferred for scheduling, and a negative one if the first
- is to be preferred. Zero if they are equally good. */
+/* Print the ready list for debugging purposes. Callable from debugger. */
-static int
-rgn_rank (insn1, insn2)
- rtx insn1, insn2;
+static void
+debug_ready_list (ready)
+ struct ready_list *ready;
{
- /* Some comparison make sense in interblock scheduling only. */
- if (INSN_BB (insn1) != INSN_BB (insn2))
- {
- int spec_val, prob_val;
-
- /* Prefer an inblock motion on an interblock motion. */
- if ((INSN_BB (insn2) == target_bb) && (INSN_BB (insn1) != target_bb))
- return 1;
- if ((INSN_BB (insn1) == target_bb) && (INSN_BB (insn2) != target_bb))
- return -1;
-
- /* Prefer a useful motion on a speculative one. */
- spec_val = IS_SPECULATIVE_INSN (insn1) - IS_SPECULATIVE_INSN (insn2);
- if (spec_val)
- return spec_val;
-
- /* Prefer a more probable (speculative) insn. */
- prob_val = INSN_PROBABILITY (insn2) - INSN_PROBABILITY (insn1);
- if (prob_val)
- return prob_val;
- }
- return 0;
-}
+ rtx *p;
+ int i;
-/* Used in schedule_insns to initialize current_sched_info for scheduling
- regions (or single basic blocks). */
+ if (ready->n_ready == 0)
+ return;
-static struct sched_info region_sched_info =
-{
- init_ready_list,
- can_schedule_ready_p,
- schedule_more_p,
- new_ready,
- rgn_rank,
- rgn_print_insn,
-
- NULL, NULL,
- NULL, NULL,
- 0
-};
+ p = ready_lastpos (ready);
+ for (i = 0; i < ready->n_ready; i++)
+ fprintf (sched_dump, " %s", (*current_sched_info->print_insn) (p[i], 0));
+ fprintf (sched_dump, "\n");
+}
/* move_insn1: Remove INSN from insn chain, and link it after LAST insn. */
return retval;
}
-/* Use forward list scheduling to rearrange insns of block BB in region RGN,
+/* Use forward list scheduling to rearrange insns of block B in region RGN,
possibly bringing insns from subsequent blocks in the same region. */
-static void
-schedule_block (bb, rgn_n_insns)
- int bb;
+void
+schedule_block (b, rgn_n_insns)
+ int b;
int rgn_n_insns;
{
rtx last;
struct ready_list ready;
int can_issue_more;
- /* Flow block of this bb. */
- int b = BB_TO_BLOCK (bb);
-
/* Head/tail info for this block. */
rtx prev_head = current_sched_info->prev_head;
rtx next_tail = current_sched_info->next_tail;
/* Sort the ready list based on priority. */
ready_sort (&ready);
- /* Allow the target to reorder the list, typically for
- better instruction bundling. */
-#ifdef MD_SCHED_REORDER
- MD_SCHED_REORDER (sched_dump, sched_verbose, ready_lastpos (&ready),
- ready.n_ready, clock_var, can_issue_more);
-#else
- can_issue_more = issue_rate;
-#endif
-
- if (sched_verbose)
- {
- fprintf (sched_dump, "\n;;\tReady list (t =%3d): ", clock_var);
- debug_ready_list (&ready);
- }
-
- /* Issue insns from ready list. */
- while (ready.n_ready != 0 && can_issue_more)
- {
- /* Select and remove the insn from the ready list. */
- rtx insn = ready_remove_first (&ready);
- int cost = actual_hazard (insn_unit (insn), insn, clock_var, 0);
-
- if (cost >= 1)
- {
- queue_insn (insn, cost);
- continue;
- }
-
- if (! (*current_sched_info->can_schedule_ready_p) (insn))
- goto next;
-
- last_scheduled_insn = insn;
- last = move_insn (insn, last);
-
-#ifdef MD_SCHED_VARIABLE_ISSUE
- MD_SCHED_VARIABLE_ISSUE (sched_dump, sched_verbose, insn,
- can_issue_more);
-#else
- can_issue_more--;
-#endif
-
- schedule_insn (insn, &ready, clock_var);
-
- next:
- /* Close this block after scheduling its jump. */
- if (GET_CODE (last_scheduled_insn) == JUMP_INSN)
- break;
- }
-
- /* Debug info. */
- if (sched_verbose)
- visualize_scheduled_insns (clock_var);
- }
-
- /* Debug info. */
- if (sched_verbose)
- {
- fprintf (sched_dump, ";;\tReady list (final): ");
- debug_ready_list (&ready);
- print_block_visualization ("");
- }
-
- /* Sanity check -- queue must be empty now. Meaningless if region has
- multiple bbs. */
- if (current_sched_info->queue_must_finish_empty && q_size != 0)
- abort ();
-
- /* Update head/tail boundaries. */
- head = NEXT_INSN (prev_head);
- tail = last;
-
- /* Restore-other-notes: NOTE_LIST is the end of a chain of notes
- previously found among the insns. Insert them at the beginning
- of the insns. */
- if (note_list != 0)
- {
- rtx note_head = note_list;
-
- while (PREV_INSN (note_head))
- {
- note_head = PREV_INSN (note_head);
- }
-
- PREV_INSN (note_head) = PREV_INSN (head);
- NEXT_INSN (PREV_INSN (head)) = note_head;
- PREV_INSN (head) = note_list;
- NEXT_INSN (note_list) = head;
- head = note_head;
- }
-
- /* Debugging. */
- if (sched_verbose)
- {
- fprintf (sched_dump, ";; total time = %d\n;; new head = %d\n",
- clock_var, INSN_UID (head));
- fprintf (sched_dump, ";; new tail = %d\n\n",
- INSN_UID (tail));
- visualize_free ();
- }
-
- current_sched_info->head = head;
- current_sched_info->tail = tail;
-
- free (ready.vec);
-}
-\f
-/* Print the bit-set of registers, S, callable from debugger. */
-
-extern void
-debug_reg_vector (s)
- regset s;
-{
- int regno;
-
- EXECUTE_IF_SET_IN_REG_SET (s, 0, regno,
- {
- fprintf (sched_dump, " %d", regno);
- });
-
- fprintf (sched_dump, "\n");
-}
-
-/* Add dependences so that branches are scheduled to run last in their
- block. */
-
-static void
-add_branch_dependences (head, tail)
- rtx head, 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.
-
- 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. */
- insn = tail;
- last = 0;
- while (GET_CODE (insn) == CALL_INSN
- || GET_CODE (insn) == JUMP_INSN
- || (GET_CODE (insn) == INSN
- && (GET_CODE (PATTERN (insn)) == USE
- || GET_CODE (PATTERN (insn)) == CLOBBER
-#ifdef HAVE_cc0
- || sets_cc0_p (PATTERN (insn))
-#endif
- ))
- || GET_CODE (insn) == NOTE)
- {
- if (GET_CODE (insn) != NOTE)
- {
- if (last != 0
- && !find_insn_list (insn, LOG_LINKS (last)))
- {
- add_dependence (last, insn, REG_DEP_ANTI);
- INSN_REF_COUNT (insn)++;
- }
-
- 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. */
- if (insn == head)
- break;
-
- insn = PREV_INSN (insn);
- }
-
- /* Make sure these insns are scheduled last in their block. */
- insn = last;
- if (insn != 0)
- while (insn != head)
- {
- insn = prev_nonnote_insn (insn);
-
- if (INSN_REF_COUNT (insn) != 0)
- continue;
-
- 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);
- }
-}
-
-/* After computing the dependencies for block BB, propagate the dependencies
- found in TMP_DEPS to the successors of the block. MAX_REG is the number
- of registers. */
-static void
-propagate_deps (bb, tmp_deps, max_reg)
- int bb;
- struct deps *tmp_deps;
- int max_reg;
-{
- 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;
- }
-
- for (reg = 0; reg < max_reg; reg++)
- {
- /* reg-last-uses lists are inherited by bb_succ. */
- for (u = tmp_deps->reg_last_uses[reg]; u; u = XEXP (u, 1))
- {
- if (find_insn_list (XEXP (u, 0),
- succ_deps->reg_last_uses[reg]))
- continue;
-
- succ_deps->reg_last_uses[reg]
- = alloc_INSN_LIST (XEXP (u, 0),
- succ_deps->reg_last_uses[reg]);
- }
-
- /* reg-last-defs lists are inherited by bb_succ. */
- for (u = tmp_deps->reg_last_sets[reg]; u; u = XEXP (u, 1))
- {
- if (find_insn_list (XEXP (u, 0),
- succ_deps->reg_last_sets[reg]))
- continue;
-
- succ_deps->reg_last_sets[reg]
- = alloc_INSN_LIST (XEXP (u, 0),
- succ_deps->reg_last_sets[reg]);
- }
-
- for (u = tmp_deps->reg_last_clobbers[reg]; u; u = XEXP (u, 1))
- {
- if (find_insn_list (XEXP (u, 0),
- succ_deps->reg_last_clobbers[reg]))
- continue;
-
- succ_deps->reg_last_clobbers[reg]
- = alloc_INSN_LIST (XEXP (u, 0),
- succ_deps->reg_last_clobbers[reg]);
- }
- }
-
- /* 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);
- }
-
- 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);
- }
-
- /* 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))
- continue;
-
- succ_deps->last_function_call
- = alloc_INSN_LIST (XEXP (u, 0),
- succ_deps->last_function_call);
- }
+ /* Allow the target to reorder the list, typically for
+ better instruction bundling. */
+#ifdef MD_SCHED_REORDER
+ MD_SCHED_REORDER (sched_dump, sched_verbose, ready_lastpos (&ready),
+ ready.n_ready, clock_var, can_issue_more);
+#else
+ can_issue_more = issue_rate;
+#endif
- /* last_pending_memory_flush is inherited by bb_succ. */
- for (u = tmp_deps->last_pending_memory_flush; u; u = XEXP (u, 1))
+ if (sched_verbose)
{
- if (find_insn_list (XEXP (u, 0),
- succ_deps->last_pending_memory_flush))
- continue;
-
- succ_deps->last_pending_memory_flush
- = alloc_INSN_LIST (XEXP (u, 0),
- succ_deps->last_pending_memory_flush);
+ fprintf (sched_dump, "\n;;\tReady list (t =%3d): ", clock_var);
+ debug_ready_list (&ready);
}
- /* 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);
+ /* Issue insns from ready list. */
+ while (ready.n_ready != 0 && can_issue_more)
+ {
+ /* Select and remove the insn from the ready list. */
+ rtx insn = ready_remove_first (&ready);
+ int cost = actual_hazard (insn_unit (insn), insn, clock_var, 0);
- e = NEXT_OUT (e);
- }
- while (e != first_edge);
-}
+ if (cost >= 1)
+ {
+ queue_insn (insn, cost);
+ continue;
+ }
-/* 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.
+ if (! (*current_sched_info->can_schedule_ready_p) (insn))
+ goto next;
- Specifically for reg-reg data dependences, the block insns are
- scanned by sched_analyze () top-to-bottom. Two lists are
- maintained by sched_analyze (): reg_last_sets[] for register DEFs,
- and reg_last_uses[] for register USEs.
+ last_scheduled_insn = insn;
+ last = move_insn (insn, last);
- When analysis is completed for bb, we update for its successors:
- ; - DEFS[succ] = Union (DEFS [succ], DEFS [bb])
- ; - USES[succ] = Union (USES [succ], DEFS [bb])
+#ifdef MD_SCHED_VARIABLE_ISSUE
+ MD_SCHED_VARIABLE_ISSUE (sched_dump, sched_verbose, insn,
+ can_issue_more);
+#else
+ can_issue_more--;
+#endif
- The mechanism for computing mem-mem data dependence is very
- similar, and the result is interblock dependences in the region. */
+ schedule_insn (insn, &ready, clock_var);
-static void
-compute_block_backward_dependences (bb)
- int bb;
-{
- rtx head, tail;
- int max_reg = max_reg_num ();
- struct deps tmp_deps;
+ next:
+ /* Close this block after scheduling its jump. */
+ if (GET_CODE (last_scheduled_insn) == JUMP_INSN)
+ break;
+ }
- tmp_deps = bb_deps[bb];
+ /* Debug info. */
+ if (sched_verbose)
+ visualize_scheduled_insns (clock_var);
+ }
- /* Do the analysis for this block. */
- get_bb_head_tail (bb, &head, &tail);
- sched_analyze (&tmp_deps, head, tail);
- add_branch_dependences (head, tail);
+ /* Debug info. */
+ if (sched_verbose)
+ {
+ fprintf (sched_dump, ";;\tReady list (final): ");
+ debug_ready_list (&ready);
+ print_block_visualization ("");
+ }
- if (current_nr_blocks > 1)
- propagate_deps (bb, &tmp_deps, max_reg);
+ /* Sanity check -- queue must be empty now. Meaningless if region has
+ multiple bbs. */
+ if (current_sched_info->queue_must_finish_empty && q_size != 0)
+ abort ();
- /* Free up the INSN_LISTs. */
- free_deps (&tmp_deps);
+ /* Update head/tail boundaries. */
+ head = NEXT_INSN (prev_head);
+ tail = last;
- /* Assert that we won't need bb_reg_last_* for this block anymore. */
- free (bb_deps[bb].reg_last_uses);
- free (bb_deps[bb].reg_last_sets);
- free (bb_deps[bb].reg_last_clobbers);
- bb_deps[bb].reg_last_uses = 0;
- bb_deps[bb].reg_last_sets = 0;
- bb_deps[bb].reg_last_clobbers = 0;
-}
+ /* Restore-other-notes: NOTE_LIST is the end of a chain of notes
+ previously found among the insns. Insert them at the beginning
+ of the insns. */
+ if (note_list != 0)
+ {
+ rtx note_head = note_list;
-/* Print dependences for debugging, callable from debugger. */
+ while (PREV_INSN (note_head))
+ {
+ note_head = PREV_INSN (note_head);
+ }
-void
-debug_dependencies ()
-{
- int bb;
+ PREV_INSN (note_head) = PREV_INSN (head);
+ NEXT_INSN (PREV_INSN (head)) = note_head;
+ PREV_INSN (head) = note_list;
+ NEXT_INSN (note_list) = head;
+ head = note_head;
+ }
- fprintf (sched_dump, ";; --------------- forward dependences: ------------ \n");
- for (bb = 0; bb < current_nr_blocks; bb++)
+ /* Debugging. */
+ if (sched_verbose)
{
- if (1)
- {
- rtx head, tail;
- rtx next_tail;
- rtx insn;
-
- get_bb_head_tail (bb, &head, &tail);
- next_tail = NEXT_INSN (tail);
- 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",
- "----", "----", "--", "---", "----", "----", "--------", "-----");
- for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
- {
- rtx link;
- int unit, range;
+ fprintf (sched_dump, ";; total time = %d\n;; new head = %d\n",
+ clock_var, INSN_UID (head));
+ fprintf (sched_dump, ";; new tail = %d\n\n",
+ INSN_UID (tail));
+ visualize_free ();
+ }
- if (! INSN_P (insn))
- {
- int n;
- fprintf (sched_dump, ";; %6d ", INSN_UID (insn));
- if (GET_CODE (insn) == NOTE)
- {
- n = NOTE_LINE_NUMBER (insn);
- if (n < 0)
- fprintf (sched_dump, "%s\n", GET_NOTE_INSN_NAME (n));
- else
- fprintf (sched_dump, "line %d, file %s\n", n,
- NOTE_SOURCE_FILE (insn));
- }
- else
- fprintf (sched_dump, " {%s}\n", GET_RTX_NAME (GET_CODE (insn)));
- continue;
- }
+ current_sched_info->head = head;
+ current_sched_info->tail = tail;
- 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);
- fprintf (sched_dump, "\t: ");
- for (link = INSN_DEPEND (insn); link; link = XEXP (link, 1))
- fprintf (sched_dump, "%d ", INSN_UID (XEXP (link, 0)));
- fprintf (sched_dump, "\n");
- }
- }
- }
- fprintf (sched_dump, "\n");
+ free (ready.vec);
}
-
+\f
/* Set_priorities: compute priority of each insn in the block. */
-static int
-set_priorities (bb)
- int bb;
+int
+set_priorities (b)
+ int b;
{
rtx insn;
int n_insn;
rtx prev_head;
rtx head;
- get_bb_head_tail (bb, &head, &tail);
+ get_block_head_tail (b, &head, &tail);
prev_head = PREV_INSN (head);
if (head == tail && (! INSN_P (head)))
return n_insn;
}
-/* Schedule a region. A region is either an inner loop, a loop-free
- subroutine, or a single basic block. Each bb in the region is
- scheduled after its flow predecessors. */
-
-static void
-schedule_region (rgn)
- int rgn;
-{
- int bb;
- int rgn_n_insns = 0;
- int sched_rgn_n_insns = 0;
-
- /* Set variables for the current region. */
- current_nr_blocks = RGN_NR_BLOCKS (rgn);
- current_blocks = RGN_BLOCKS (rgn);
-
- init_deps_global ();
-
- /* Initializations for region data dependence analyisis. */
- bb_deps = (struct deps *) xmalloc (sizeof (struct deps) * current_nr_blocks);
- for (bb = 0; bb < current_nr_blocks; bb++)
- init_deps (bb_deps + bb);
-
- /* Compute LOG_LINKS. */
- for (bb = 0; bb < current_nr_blocks; bb++)
- compute_block_backward_dependences (bb);
-
- /* Compute INSN_DEPEND. */
- for (bb = current_nr_blocks - 1; bb >= 0; bb--)
- {
- rtx head, tail;
- get_bb_head_tail (bb, &head, &tail);
-
- compute_forward_dependences (head, tail);
- }
-
- /* Set priorities. */
- for (bb = 0; bb < current_nr_blocks; bb++)
- rgn_n_insns += set_priorities (bb);
-
- /* Compute interblock info: probabilities, split-edges, dominators, etc. */
- if (current_nr_blocks > 1)
- {
- 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));
-
- /* Edge to bit. */
- rgn_nr_edges = 0;
- edge_to_bit = (int *) 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_nr_edges = 0;
- for (i = 1; i < nr_edges; i++)
- if (CONTAINING_RGN (FROM_BLOCK (i)) == (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));
- }
-
- /* Compute probabilities, dominators, split_edges. */
- for (bb = 0; bb < current_nr_blocks; bb++)
- compute_dom_prob_ps (bb);
- }
-
- /* Now we can schedule all blocks. */
- for (bb = 0; bb < current_nr_blocks; bb++)
- {
- rtx head, tail;
- int b = BB_TO_BLOCK (bb);
-
- get_block_head_tail (b, &head, &tail);
-
- if (no_real_insns_p (head, tail))
- continue;
-
- current_sched_info->prev_head = PREV_INSN (head);
- current_sched_info->next_tail = NEXT_INSN (tail);
-
- if (write_symbols != NO_DEBUG)
- {
- save_line_notes (bb);
- rm_line_notes (bb);
- }
-
- /* 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
- 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.
-
- FIXME: Probably the same thing should be done with REG_SAVE_NOTEs
- referencing NOTE_INSN_SETJMP at the end of the block. */
- if (INSN_P (head))
- {
- rtx note;
-
- for (note = REG_NOTES (head); note; note = XEXP (note, 1))
- if (REG_NOTE_KIND (note) == REG_SAVE_NOTE)
- {
- if (INTVAL (XEXP (note, 0)) != NOTE_INSN_SETJMP)
- {
- remove_note (head, note);
- note = XEXP (note, 1);
- remove_note (head, note);
- }
- else
- note = XEXP (note, 1);
- }
- }
-
- /* Remove remaining note insns from the block, save them in
- note_list. These notes are restored at the end of
- schedule_block (). */
- note_list = 0;
- rm_other_notes (head, tail);
-
- target_bb = bb;
-
- current_sched_info->queue_must_finish_empty
- = current_nr_blocks > 1 && !flag_schedule_interblock;
-
- schedule_block (bb, rgn_n_insns);
- 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;
-
- /* Clean up. */
- if (current_nr_blocks > 1)
- {
- free (candidate_table);
- free (bblst_table);
- free (bitlst_table);
- }
- }
-
- /* Sanity check: verify that all region insns were scheduled. */
- if (sched_rgn_n_insns != rgn_n_insns)
- abort ();
-
- /* Restore line notes. */
- if (write_symbols != NO_DEBUG)
- {
- for (bb = 0; bb < current_nr_blocks; bb++)
- restore_line_notes (bb);
- }
-
- /* Done with this region. */
- free_pending_lists ();
-
- finish_deps_global ();
-
- free (bb_deps);
-
- 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);
- free (edge_to_bit);
- free (rgn_edges);
- free (pot_split);
- free (ancestor_edges);
- }
-}
-
/* Initialize some global state for the scheduler. DUMP_FILE is to be used
for debugging output. */
-static void
+void
sched_init (dump_file)
FILE *dump_file;
{
find_insn_reg_weight (b);
}
-/* Indexed by region, holds the number of death notes found in that region.
- Used for consistency checks. */
-static int *deaths_in_region;
-
-/* Initialize data structures for region scheduling. */
-
-static void
-init_regions ()
-{
- 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);
-
- /* Compute regions for scheduling. */
- if (reload_completed
- || n_basic_blocks == 1
- || !flag_schedule_interblock)
- {
- find_single_block_region ();
- }
- else
- {
- /* Verify that a 'good' control flow graph can be built. */
- if (is_cfg_nonregular ())
- {
- find_single_block_region ();
- }
- 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. */
- edge_list = create_edge_list ();
-
- /* Compute the dominators and post dominators. */
- calculate_dominance_info (NULL, dom, CDI_DOMINATORS);
-
- /* build_control_flow will return nonzero if it detects unreachable
- blocks or any other irregularity with the cfg which prevents
- cross block scheduling. */
- if (build_control_flow (edge_list) != 0)
- find_single_block_region ();
- else
- find_rgns (edge_list, dom);
-
- if (sched_verbose >= 3)
- debug_regions ();
-
- /* We are done with flow's edge list. */
- free_edge_list (edge_list);
-
- /* For now. This will move as more and more of haifa is converted
- to using the cfg code in flow.c. */
- free (dom);
- }
- }
-
- deaths_in_region = (int *) 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]);
-
- deaths_in_region[rgn] = count_or_remove_death_notes (blocks, 1);
- }
-
- sbitmap_free (blocks);
-}
-
-/* The one entry point in this file. DUMP_FILE is the dump file for
- this pass. */
+/* Free global data used during insn scheduling. */
void
-schedule_insns (dump_file)
- FILE *dump_file;
+sched_finish ()
{
- sbitmap large_region_blocks, blocks;
- int rgn;
- int any_large_regions;
-
- /* Taking care of this degenerate case makes the rest of
- this code simpler. */
- if (n_basic_blocks == 0)
- return;
-
- nr_inter = 0;
- nr_spec = 0;
-
- sched_init (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. */
- /* ??? There is an outside possibility that update_life_info, or more
- to the point propagate_block, could get called with non-zero 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.
-
- I'm fairly certain that this _shouldn't_ happen, since I don't think
- that live_at_start should change at region heads. Not sure what the
- best way to test for this kind of thing... */
-
- allocate_reg_life_data ();
- compute_bb_for_insn (old_max_uid);
-
- any_large_regions = 0;
- large_region_blocks = sbitmap_alloc (n_basic_blocks);
- sbitmap_ones (large_region_blocks);
-
- blocks = sbitmap_alloc (n_basic_blocks);
-
- 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
- }
-
- if (any_large_regions)
- {
- update_life_info (large_region_blocks, UPDATE_LIFE_GLOBAL,
- PROP_DEATH_NOTES | PROP_REG_INFO);
- }
-
- /* Reposition the prologue and epilogue notes in case we moved the
- prologue/epilogue insns. */
- if (reload_completed)
- reposition_prologue_and_epilogue_notes (get_insns ());
-
- /* Delete redundant line notes. */
- if (write_symbols != NO_DEBUG)
- rm_redundant_line_notes ();
-
- if (sched_verbose)
- {
- if (reload_completed == 0 && flag_schedule_interblock)
- {
- fprintf (sched_dump,
- "\n;; Procedure interblock/speculative motions == %d/%d \n",
- nr_inter, nr_spec);
- }
- else
- {
- if (nr_inter > 0)
- abort ();
- }
- fprintf (sched_dump, "\n\n");
- }
-
- /* Clean up. */
- end_alias_analysis ();
-
- free_dependency_caches ();
- free (rgn_table);
- free (rgn_bb_table);
- free (block_to_bb);
- free (containing_rgn);
-
free (h_i_d);
-
+ free_dependency_caches ();
+ end_alias_analysis ();
if (write_symbols != NO_DEBUG)
free (line_note_head);
-
- if (edge_table)
- {
- free (edge_table);
- edge_table = NULL;
- }
-
- if (in_edges)
- {
- free (in_edges);
- in_edges = NULL;
- }
- if (out_edges)
- {
- free (out_edges);
- out_edges = NULL;
- }
-
- sbitmap_free (blocks);
- sbitmap_free (large_region_blocks);
-
- free (deaths_in_region);
}
-
#endif /* INSN_SCHEDULING */
--- /dev/null
+/* Instruction scheduling pass.
+ Copyright (C) 1992, 1993, 1994, 1995, 1996, 1997, 1998,
+ 1999, 2000 Free Software Foundation, Inc.
+ Contributed by Michael Tiemann (tiemann@cygnus.com) Enhanced by,
+ and currently maintained by, Jim Wilson (wilson@cygnus.com)
+
+This file is part of GNU CC.
+
+GNU CC is free software; you can redistribute it and/or modify it
+under the terms of the GNU General Public License as published by the
+Free Software Foundation; either version 2, or (at your option) any
+later version.
+
+GNU CC is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+for more details.
+
+You should have received a copy of the GNU General Public License
+along with GNU CC; see the file COPYING. If not, write to the Free
+the Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA
+02111-1307, USA. */
+
+/* This pass implements list scheduling within basic blocks. It is
+ run twice: (1) after flow analysis, but before register allocation,
+ and (2) after register allocation.
+
+ The first run performs interblock scheduling, moving insns between
+ different blocks in the same "region", and the second runs only
+ basic block scheduling.
+
+ Interblock motions performed are useful motions and speculative
+ motions, including speculative loads. Motions requiring code
+ duplication are not supported. The identification of motion type
+ and the check for validity of speculative motions requires
+ construction and analysis of the function's control flow graph.
+
+ The main entry point for this pass is schedule_insns(), called for
+ each function. The work of the scheduler is organized in three
+ levels: (1) function level: insns are subject to splitting,
+ control-flow-graph is constructed, regions are computed (after
+ reload, each region is of one block), (2) region level: control
+ flow graph attributes required for interblock scheduling are
+ computed (dominators, reachability, etc.), data dependences and
+ priorities are computed, and (3) block level: insns in the block
+ are actually scheduled. */
+\f
+#include "config.h"
+#include "system.h"
+#include "toplev.h"
+#include "rtl.h"
+#include "tm_p.h"
+#include "hard-reg-set.h"
+#include "basic-block.h"
+#include "regs.h"
+#include "function.h"
+#include "flags.h"
+#include "insn-config.h"
+#include "insn-attr.h"
+#include "except.h"
+#include "toplev.h"
+#include "recog.h"
+#include "sched-int.h"
+
+/* Some accessor macros for h_i_d members only used within this file. */
+#define INSN_REF_COUNT(INSN) (h_i_d[INSN_UID (INSN)].ref_count)
+#define FED_BY_SPEC_LOAD(insn) (h_i_d[INSN_UID (insn)].fed_by_spec_load)
+#define IS_LOAD_INSN(insn) (h_i_d[INSN_UID (insn)].is_load_insn)
+
+#define MAX_RGN_BLOCKS 10
+#define MAX_RGN_INSNS 100
+
+/* nr_inter/spec counts interblock/speculative motion for the function. */
+static int nr_inter, nr_spec;
+
+/* Control flow graph edges are kept in circular lists. */
+typedef struct
+{
+ int from_block;
+ int to_block;
+ int next_in;
+ int next_out;
+}
+haifa_edge;
+static haifa_edge *edge_table;
+
+#define NEXT_IN(edge) (edge_table[edge].next_in)
+#define NEXT_OUT(edge) (edge_table[edge].next_out)
+#define FROM_BLOCK(edge) (edge_table[edge].from_block)
+#define TO_BLOCK(edge) (edge_table[edge].to_block)
+
+/* Number of edges in the control flow graph. (In fact, larger than
+ that by 1, since edge 0 is unused.) */
+static int nr_edges;
+
+/* Circular list of incoming/outgoing edges of a block. */
+static int *in_edges;
+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));
+
+/* A region is the main entity for interblock scheduling: insns
+ are allowed to move between blocks in the same region, along
+ control flow graph edges, in the 'up' direction. */
+typedef struct
+{
+ int rgn_nr_blocks; /* Number of blocks in region. */
+ int rgn_blocks; /* cblocks in the region (actually index in rgn_bb_table). */
+}
+region;
+
+/* Number of regions in the procedure. */
+static int nr_regions;
+
+/* Table of region descriptions. */
+static region *rgn_table;
+
+/* Array of lists of regions' blocks. */
+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. */
+static int *block_to_bb;
+
+/* The number of the region containing a block. */
+static int *containing_rgn;
+
+#define RGN_NR_BLOCKS(rgn) (rgn_table[rgn].rgn_nr_blocks)
+#define RGN_BLOCKS(rgn) (rgn_table[rgn].rgn_blocks)
+#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 *));
+
+extern void debug_live PARAMS ((int, int));
+
+/* Blocks of the current region being scheduled. */
+static int current_nr_blocks;
+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. */
+ int nr_members; /* The number of members of the bit list. */
+}
+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 *));
+
+/* Target info declarations.
+
+ The block currently being scheduled is referred to as the "target" block,
+ while other blocks in the region from which insns can be moved to the
+ target are called "source" blocks. The candidate structure holds info
+ about such sources: are they valid? Speculative? Etc. */
+typedef bitlst bblst;
+typedef struct
+{
+ char is_valid;
+ char is_speculative;
+ int src_prob;
+ bblst split_bbs;
+ bblst update_bbs;
+}
+candidate;
+
+static candidate *candidate_table;
+
+/* A speculative motion requires checking live information on the path
+ from 'source' to 'target'. The split blocks are those to be checked.
+ After a speculative motion, live information should be modified in
+ the 'update' blocks.
+
+ Lists of split and update blocks for each candidate of the current
+ target are in array bblst_table. */
+static int *bblst_table, bblst_size, bblst_last;
+
+#define IS_VALID(src) ( candidate_table[src].is_valid )
+#define IS_SPECULATIVE(src) ( candidate_table[src].is_speculative )
+#define SRC_PROB(src) ( candidate_table[src].src_prob )
+
+/* The bb being currently scheduled. */
+static int target_bb;
+
+/* List of edges. */
+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;
+
+/* Dominators array: dom[i] contains the bbset of dominators of
+ bb i in the region. */
+static bbset *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) )
+
+/* Probability: Prob[i] is a float in [0, 1] which is the probability
+ of bb i relative to the region entry. */
+static float *prob;
+
+/* The probability of bb_src, relative to bb_trg. Note, that while the
+ 'prob[bb]' is a float in [0, 1], this macro returns an integer
+ in [0, 100]. */
+#define GET_SRC_PROB(bb_src, bb_trg) ((int) (100.0 * (prob[bb_src] / \
+ prob[bb_trg])))
+
+/* Bit-set of edges, where bit i stands for edge i. */
+typedef bitset edgeset;
+
+/* Number of edges in the region. */
+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;
+#define EDGE_TO_BIT(edge) (edge_to_bit[edge])
+
+/* The split edges of a source bb is different for each target
+ bb. In order to compute this efficiently, the 'potential-split edges'
+ are computed for each bb prior to scheduling a region. This is actually
+ the split edges of each bb relative to the region entry.
+
+ pot_split[bb] is the set of potential split edges of bb. */
+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));
+
+#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
+#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 *, int));
+static void free_pending_lists PARAMS ((void));
+
+/* Functions for construction of the control flow graph. */
+
+/* Return 1 if control flow graph should not be constructed, 0 otherwise.
+
+ We decide not to build the control flow graph if there is possibly more
+ than one entry to the function, if computed branches exist, of if we
+ have nonlocal gotos. */
+
+static int
+is_cfg_nonregular ()
+{
+ int b;
+ rtx insn;
+ RTX_CODE code;
+
+ /* If we have a label that could be the target of a nonlocal goto, then
+ the cfg is not well structured. */
+ if (nonlocal_goto_handler_labels)
+ return 1;
+
+ /* If we have any forced labels, then the cfg is not well structured. */
+ if (forced_labels)
+ return 1;
+
+ /* If this function has a computed jump, then we consider the cfg
+ not well structured. */
+ if (current_function_has_computed_jump)
+ return 1;
+
+ /* 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)
+ 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))
+ {
+ code = GET_CODE (insn);
+ if (GET_RTX_CLASS (code) == 'i')
+ {
+ rtx note;
+
+ for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
+ if (REG_NOTE_KIND (note) == REG_LABEL)
+ return 1;
+ }
+
+ if (insn == BLOCK_END (b))
+ break;
+ }
+
+ /* All the tests passed. Consider the cfg well structured. */
+ return 0;
+}
+
+/* Build the control flow graph and set nr_edges.
+
+ Instead of trying to build a cfg ourselves, we rely on flow to
+ do it for us. Stamp out useless code (and bug) duplication.
+
+ Return nonzero if an irregularity in the cfg is found which would
+ prevent cross block scheduling. */
+
+static int
+build_control_flow (edge_list)
+ struct edge_list *edge_list;
+{
+ int i, unreachable, num_edges;
+
+ /* This already accounts for entry/exit edges. */
+ num_edges = NUM_EDGES (edge_list);
+
+ /* Unreachable loops with more than one basic block are detected
+ during the DFS traversal in find_rgns.
+
+ Unreachable loops with a single block are detected here. This
+ 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++)
+ {
+ basic_block b = BASIC_BLOCK (i);
+
+ if (b->pred == NULL
+ || (b->pred->src == b
+ && b->pred->pred_next == NULL))
+ unreachable = 1;
+ }
+
+ /* ??? 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));
+
+ nr_edges = 0;
+ for (i = 0; i < num_edges; i++)
+ {
+ edge e = INDEX_EDGE (edge_list, i);
+
+ if (e->dest != EXIT_BLOCK_PTR
+ && e->src != ENTRY_BLOCK_PTR)
+ new_edge (e->src->index, e->dest->index);
+ }
+
+ /* Increment by 1, since edge 0 is unused. */
+ nr_edges++;
+
+ return unreachable;
+}
+
+/* Record an edge in the control flow graph from SOURCE to TARGET.
+
+ In theory, this is redundant with the s_succs computed above, but
+ we have not converted all of haifa to use information from the
+ integer lists. */
+
+static void
+new_edge (source, target)
+ int source, target;
+{
+ int e, next_edge;
+ int curr_edge, fst_edge;
+
+ /* Check for duplicates. */
+ fst_edge = curr_edge = OUT_EDGES (source);
+ while (curr_edge)
+ {
+ if (FROM_BLOCK (curr_edge) == source
+ && TO_BLOCK (curr_edge) == target)
+ {
+ return;
+ }
+
+ curr_edge = NEXT_OUT (curr_edge);
+
+ if (fst_edge == curr_edge)
+ break;
+ }
+
+ e = ++nr_edges;
+
+ FROM_BLOCK (e) = source;
+ TO_BLOCK (e) = target;
+
+ if (OUT_EDGES (source))
+ {
+ next_edge = NEXT_OUT (OUT_EDGES (source));
+ NEXT_OUT (OUT_EDGES (source)) = e;
+ NEXT_OUT (e) = next_edge;
+ }
+ else
+ {
+ OUT_EDGES (source) = e;
+ NEXT_OUT (e) = e;
+ }
+
+ if (IN_EDGES (target))
+ {
+ next_edge = NEXT_IN (IN_EDGES (target));
+ NEXT_IN (IN_EDGES (target)) = e;
+ NEXT_IN (e) = next_edge;
+ }
+ else
+ {
+ IN_EDGES (target) = e;
+ NEXT_IN (e) = e;
+ }
+}
+
+/* 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] |= \
+ 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] &= \
+ ~(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] &
+ 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;
+{
+ int i, j, offset;
+ unsigned HOST_WIDE_INT word;
+
+ /* bblst table space is reused in each call to extract_bitlst. */
+ bitlst_table_last = 0;
+
+ bl->first_member = &bitlst_table[bitlst_table_last];
+ 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;
+ }
+ }
+
+}
+
+/* Functions for the construction of regions. */
+
+/* Print the regions, for debugging purposes. Callable from debugger. */
+
+void
+debug_regions ()
+{
+ int rgn, bb;
+
+ fprintf (sched_dump, "\n;; ------------ REGIONS ----------\n\n");
+ for (rgn = 0; rgn < nr_regions; rgn++)
+ {
+ fprintf (sched_dump, ";;\trgn %d nr_blocks %d:\n", rgn,
+ rgn_table[rgn].rgn_nr_blocks);
+ fprintf (sched_dump, ";;\tbb/block: ");
+
+ for (bb = 0; bb < rgn_table[rgn].rgn_nr_blocks; bb++)
+ {
+ current_blocks = RGN_BLOCKS (rgn);
+
+ if (bb != BLOCK_TO_BB (BB_TO_BLOCK (bb)))
+ abort ();
+
+ fprintf (sched_dump, " %d/%d ", bb, BB_TO_BLOCK (bb));
+ }
+
+ fprintf (sched_dump, "\n\n");
+ }
+}
+
+/* Build a single block region for each basic block in the function.
+ This allows for using the same code for interblock and basic block
+ scheduling. */
+
+static void
+find_single_block_region ()
+{
+ int i;
+
+ for (i = 0; i < n_basic_blocks; i++)
+ {
+ rgn_bb_table[i] = i;
+ RGN_NR_BLOCKS (i) = 1;
+ RGN_BLOCKS (i) = i;
+ CONTAINING_RGN (i) = i;
+ BLOCK_TO_BB (i) = 0;
+ }
+ nr_regions = n_basic_blocks;
+}
+
+/* Update number of blocks and the estimate for number of insns
+ in the region. Return 1 if the region is "too large" for interblock
+ scheduling (compile time considerations), otherwise return 0. */
+
+static int
+too_large (block, num_bbs, num_insns)
+ int block, *num_bbs, *num_insns;
+{
+ (*num_bbs)++;
+ (*num_insns) += (INSN_LUID (BLOCK_END (block)) -
+ INSN_LUID (BLOCK_HEAD (block)));
+ if ((*num_bbs > MAX_RGN_BLOCKS) || (*num_insns > MAX_RGN_INSNS))
+ return 1;
+ else
+ return 0;
+}
+
+/* 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; \
+ } \
+}
+
+/* Find regions for interblock scheduling.
+
+ A region for scheduling can be:
+
+ * A loop-free procedure, or
+
+ * A reducible inner loop, or
+
+ * A basic block not contained in any other region.
+
+ ?!? In theory we could build other regions based on extended basic
+ blocks or reverse extended basic blocks. Is it worth the trouble?
+
+ Loop blocks that form a region are put into the region's block list
+ in topological order.
+
+ This procedure stores its results into the following global (ick) variables
+
+ * rgn_nr
+ * rgn_table
+ * rgn_bb_table
+ * block_to_bb
+ * containing region
+
+ We use dominator relationships to avoid making regions out of non-reducible
+ loops.
+
+ This procedure needs to be converted to work on pred/succ lists instead
+ of edge tables. That would simplify it somewhat. */
+
+static void
+find_rgns (edge_list, dom)
+ struct edge_list *edge_list;
+ sbitmap *dom;
+{
+ 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 num_bbs, num_insns, unreachable;
+ int too_large_failure;
+
+ /* Note if an edge has been passed. */
+ sbitmap passed;
+
+ /* Note if a block is a natural loop header. */
+ sbitmap header;
+
+ /* Note if a block is an natural inner loop header. */
+ sbitmap inner;
+
+ /* Note if a block is in the block queue. */
+ sbitmap in_queue;
+
+ /* Note if a block is in the block queue. */
+ sbitmap in_stack;
+
+ int num_edges = NUM_EDGES (edge_list);
+
+ /* Perform a DFS traversal of the cfg. Identify loop headers, inner loops
+ and a mapping from block to its loop header (if the block is contained
+ in a loop, else -1).
+
+ Store results in HEADER, INNER, and MAX_HDR respectively, these will
+ be used as inputs to the second traversal.
+
+ 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));
+
+ inner = sbitmap_alloc (n_basic_blocks);
+ sbitmap_ones (inner);
+
+ header = sbitmap_alloc (n_basic_blocks);
+ sbitmap_zero (header);
+
+ passed = sbitmap_alloc (nr_edges);
+ sbitmap_zero (passed);
+
+ in_queue = sbitmap_alloc (n_basic_blocks);
+ sbitmap_zero (in_queue);
+
+ in_stack = sbitmap_alloc (n_basic_blocks);
+ sbitmap_zero (in_stack);
+
+ for (i = 0; i < n_basic_blocks; i++)
+ max_hdr[i] = -1;
+
+ /* DFS traversal to find inner loops in the cfg. */
+
+ sp = -1;
+ while (1)
+ {
+ if (current_edge == 0 || TEST_BIT (passed, current_edge))
+ {
+ /* We have reached a leaf node or a node that was already
+ processed. Pop edges off the stack until we find
+ an edge that has not yet been processed. */
+ while (sp >= 0
+ && (current_edge == 0 || TEST_BIT (passed, current_edge)))
+ {
+ /* Pop entry off the stack. */
+ current_edge = stack[sp--];
+ node = FROM_BLOCK (current_edge);
+ child = TO_BLOCK (current_edge);
+ RESET_BIT (in_stack, child);
+ if (max_hdr[child] >= 0 && TEST_BIT (in_stack, max_hdr[child]))
+ UPDATE_LOOP_RELATIONS (node, max_hdr[child]);
+ current_edge = NEXT_OUT (current_edge);
+ }
+
+ /* See if have finished the DFS tree traversal. */
+ if (sp < 0 && TEST_BIT (passed, current_edge))
+ break;
+
+ /* Nope, continue the traversal with the popped node. */
+ continue;
+ }
+
+ /* Process a node. */
+ node = FROM_BLOCK (current_edge);
+ child = TO_BLOCK (current_edge);
+ SET_BIT (in_stack, node);
+ dfs_nr[node] = ++count;
+
+ /* If the successor is in the stack, then we've found a loop.
+ Mark the loop, if it is not a natural loop, then it will
+ be rejected during the second traversal. */
+ if (TEST_BIT (in_stack, child))
+ {
+ no_loops = 0;
+ SET_BIT (header, child);
+ UPDATE_LOOP_RELATIONS (node, child);
+ SET_BIT (passed, current_edge);
+ current_edge = NEXT_OUT (current_edge);
+ continue;
+ }
+
+ /* If the child was already visited, then there is no need to visit
+ it again. Just update the loop relationships and restart
+ with a new edge. */
+ if (dfs_nr[child])
+ {
+ if (max_hdr[child] >= 0 && TEST_BIT (in_stack, max_hdr[child]))
+ UPDATE_LOOP_RELATIONS (node, max_hdr[child]);
+ SET_BIT (passed, current_edge);
+ current_edge = NEXT_OUT (current_edge);
+ continue;
+ }
+
+ /* Push an entry on the stack and continue DFS traversal. */
+ stack[++sp] = current_edge;
+ SET_BIT (passed, current_edge);
+ current_edge = OUT_EDGES (child);
+
+ /* This is temporary until haifa is converted to use rth's new
+ cfg routines which have true entry/exit blocks and the
+ appropriate edges from/to those blocks.
+
+ Generally we update dfs_nr for a node when we process its
+ out edge. However, if the node has no out edge then we will
+ not set dfs_nr for that node. This can confuse the scheduler
+ into thinking that we have unreachable blocks, which in turn
+ disables cross block scheduling.
+
+ So, if we have a node with no out edges, go ahead and mark it
+ as reachable now. */
+ if (current_edge == 0)
+ dfs_nr[child] = ++count;
+ }
+
+ /* Another check for unreachable blocks. The earlier test in
+ is_cfg_nonregular only finds unreachable blocks that do not
+ form a loop.
+
+ The DFS traversal will mark every block that is reachable from
+ 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)
+ {
+ unreachable = 1;
+ break;
+ }
+
+ /* Gross. To avoid wasting memory, the second pass uses the dfs_nr array
+ to hold degree counts. */
+ degree = dfs_nr;
+
+ for (i = 0; i < n_basic_blocks; i++)
+ degree[i] = 0;
+ for (i = 0; i < num_edges; i++)
+ {
+ edge e = INDEX_EDGE (edge_list, i);
+
+ if (e->dest != EXIT_BLOCK_PTR)
+ degree[e->dest->index]++;
+ }
+
+ /* Do not perform region scheduling if there are any unreachable
+ blocks. */
+ if (!unreachable)
+ {
+ int *queue;
+
+ if (no_loops)
+ SET_BIT (header, 0);
+
+ /* Second travsersal:find reducible inner loops and topologically sort
+ block of each region. */
+
+ queue = (int *) 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++)
+ {
+ if (TEST_BIT (header, i) && TEST_BIT (inner, i))
+ {
+ edge e;
+ int j;
+
+ /* Now check that the loop is reducible. We do this separate
+ from finding inner loops so that we do not find a reducible
+ loop which contains an inner non-reducible loop.
+
+ A simple way to find reducible/natural loops is to verify
+ that each block in the loop is dominated by the loop
+ header.
+
+ 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++)
+ {
+ /* First identify blocks in the loop, except for the loop
+ entry block. */
+ if (i == max_hdr[j] && i != j)
+ {
+ /* Now verify that the block is dominated by the loop
+ header. */
+ if (!TEST_BIT (dom[j], i))
+ break;
+ }
+ }
+
+ /* 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)
+ 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];
+
+ /* Decrease degree of all I's successors for topological
+ ordering. */
+ for (e = BASIC_BLOCK (i)->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)));
+
+ /* Find all loop latches (blocks with back edges to the loop
+ header) or all the leaf blocks in the cfg has no loops.
+
+ Place those blocks into the queue. */
+ if (no_loops)
+ {
+ for (j = 0; j < n_basic_blocks; j++)
+ /* 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)
+ {
+ queue[++tail] = j;
+ SET_BIT (in_queue, j);
+
+ if (too_large (j, &num_bbs, &num_insns))
+ {
+ too_large_failure = 1;
+ break;
+ }
+ }
+ }
+ else
+ {
+ edge e;
+
+ for (e = BASIC_BLOCK (i)->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)
+ {
+ /* This is a loop latch. */
+ queue[++tail] = node;
+ SET_BIT (in_queue, node);
+
+ if (too_large (node, &num_bbs, &num_insns))
+ {
+ too_large_failure = 1;
+ break;
+ }
+ }
+ }
+ }
+
+ /* Now add all the blocks in the loop to the queue.
+
+ We know the loop is a natural loop; however the algorithm
+ above will not always mark certain blocks as being in the
+ loop. Consider:
+ node children
+ a b,c
+ b c
+ c a,d
+ d b
+
+ The algorithm in the DFS traversal may not mark B & D as part
+ of the loop (ie they will not have max_hdr set to A).
+
+ We know they can not be loop latches (else they would have
+ had max_hdr set since they'd have a backedge to a dominator
+ block). So we don't need them on the initial queue.
+
+ We know they are part of the loop because they are dominated
+ by the loop header and can be reached by a backwards walk of
+ the edges starting with nodes on the initial queue.
+
+ It is safe and desirable to include those nodes in the
+ loop/scheduling region. To do so we would need to decrease
+ the degree of a node if it is the target of a backedge
+ within the loop itself as the node is placed in the queue.
+
+ We do not do this because I'm not sure that the actual
+ scheduling code will properly handle this case. ?!? */
+
+ while (head < tail && !too_large_failure)
+ {
+ edge e;
+ child = queue[++head];
+
+ for (e = BASIC_BLOCK (child)->pred; e; e = e->pred_next)
+ {
+ node = e->src->index;
+
+ /* See discussion above about nodes not marked as in
+ this loop during the initial DFS traversal. */
+ if (e->src == ENTRY_BLOCK_PTR
+ || max_hdr[node] != loop_head)
+ {
+ tail = -1;
+ break;
+ }
+ else if (!TEST_BIT (in_queue, node) && node != i)
+ {
+ queue[++tail] = node;
+ SET_BIT (in_queue, node);
+
+ if (too_large (node, &num_bbs, &num_insns))
+ {
+ too_large_failure = 1;
+ break;
+ }
+ }
+ }
+ }
+
+ if (tail >= 0 && !too_large_failure)
+ {
+ /* Place the loop header into list of region blocks. */
+ degree[i] = -1;
+ rgn_bb_table[idx] = i;
+ RGN_NR_BLOCKS (nr_regions) = num_bbs;
+ RGN_BLOCKS (nr_regions) = idx++;
+ CONTAINING_RGN (i) = nr_regions;
+ BLOCK_TO_BB (i) = count = 0;
+
+ /* Remove blocks from queue[] when their in degree
+ becomes zero. Repeat until no blocks are left on the
+ list. This produces a topological list of blocks in
+ the region. */
+ while (tail >= 0)
+ {
+ if (head < 0)
+ head = tail;
+ child = queue[head];
+ if (degree[child] == 0)
+ {
+ edge e;
+
+ degree[child] = -1;
+ rgn_bb_table[idx++] = child;
+ BLOCK_TO_BB (child) = ++count;
+ CONTAINING_RGN (child) = nr_regions;
+ queue[head] = queue[tail--];
+
+ for (e = BASIC_BLOCK (child)->succ;
+ e;
+ e = e->succ_next)
+ if (e->dest != EXIT_BLOCK_PTR)
+ --degree[e->dest->index];
+ }
+ else
+ --head;
+ }
+ ++nr_regions;
+ }
+ }
+ }
+ free (queue);
+ }
+
+ /* 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)
+ {
+ rgn_bb_table[idx] = i;
+ RGN_NR_BLOCKS (nr_regions) = 1;
+ RGN_BLOCKS (nr_regions) = idx++;
+ CONTAINING_RGN (i) = nr_regions++;
+ BLOCK_TO_BB (i) = 0;
+ }
+
+ free (max_hdr);
+ free (dfs_nr);
+ free (stack);
+ free (passed);
+ free (header);
+ free (inner);
+ free (in_queue);
+ free (in_stack);
+}
+
+/* Functions for regions scheduling information. */
+
+/* Compute dominators, probability, and potential-split-edges of bb.
+ Assume that these values were already computed for bb's predecessors. */
+
+static void
+compute_dom_prob_ps (bb)
+ int bb;
+{
+ int nxt_in_edge, fst_in_edge, pred;
+ int fst_out_edge, nxt_out_edge, nr_out_edges, nr_rgn_out_edges;
+
+ prob[bb] = 0.0;
+ if (IS_RGN_ENTRY (bb))
+ {
+ BITSET_ADD (dom[bb], 0, bbset_size);
+ 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);
+
+ 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);
+
+ BITSET_ADD (ancestor_edges[bb], EDGE_TO_BIT (nxt_in_edge), edgeset_size);
+
+ 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);
+
+ /* The successor doesn't belong in the region? */
+ if (CONTAINING_RGN (TO_BLOCK (fst_out_edge)) !=
+ CONTAINING_RGN (BB_TO_BLOCK (bb)))
+ ++nr_rgn_out_edges;
+
+ while (fst_out_edge != nxt_out_edge)
+ {
+ ++nr_out_edges;
+ /* The successor doesn't belong in the region? */
+ 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);
+ nxt_out_edge = NEXT_OUT (nxt_out_edge);
+
+ }
+
+ /* Now nr_rgn_out_edges is the number of region-exit edges from
+ pred, and nr_out_edges will be the number of pred out edges
+ not leaving the region. */
+ nr_out_edges -= nr_rgn_out_edges;
+ if (nr_rgn_out_edges > 0)
+ prob[bb] += 0.9 * prob[BLOCK_TO_BB (pred)] / nr_out_edges;
+ else
+ prob[bb] += prob[BLOCK_TO_BB (pred)] / nr_out_edges;
+ nxt_in_edge = NEXT_IN (nxt_in_edge);
+ }
+ while (fst_in_edge != nxt_in_edge);
+
+ BITSET_ADD (dom[bb], bb, bbset_size);
+ BITSET_DIFFER (pot_split[bb], ancestor_edges[bb], edgeset_size);
+
+ if (sched_verbose >= 2)
+ fprintf (sched_dump, ";; bb_prob(%d, %d) = %3d\n", bb, BB_TO_BLOCK (bb),
+ (int) (100.0 * prob[bb]));
+}
+
+/* Functions for target info. */
+
+/* Compute in BL the list of split-edges of bb_src relatively to bb_trg.
+ Note that bb_trg dominates bb_src. */
+
+static void
+split_edges (bb_src, bb_trg, bl)
+ 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);
+}
+
+/* Find the valid candidate-source-blocks for the target block TRG, compute
+ their probability, and check if they are speculative or not.
+ For speculative sources, compute their update-blocks and split-blocks. */
+
+static void
+compute_trg_info (trg)
+ int trg;
+{
+ register candidate *sp;
+ edgelst el;
+ int check_block, update_idx;
+ int i, j, k, fst_edge, nxt_edge;
+
+ /* Define some of the fields for the target bb as well. */
+ sp = candidate_table + trg;
+ sp->is_valid = 1;
+ sp->is_speculative = 0;
+ sp->src_prob = 100;
+
+ for (i = trg + 1; i < current_nr_blocks; i++)
+ {
+ sp = candidate_table + i;
+
+ sp->is_valid = IS_DOMINATED (i, trg);
+ if (sp->is_valid)
+ {
+ sp->src_prob = GET_SRC_PROB (i, trg);
+ sp->is_valid = (sp->src_prob >= MIN_PROBABILITY);
+ }
+
+ if (sp->is_valid)
+ {
+ split_edges (i, trg, &el);
+ sp->is_speculative = (el.nr_members) ? 1 : 0;
+ if (sp->is_speculative && !flag_schedule_speculative)
+ sp->is_valid = 0;
+ }
+
+ if (sp->is_valid)
+ {
+ char *update_blocks;
+
+ /* Compute split blocks and store them in bblst_table.
+ The TO block of every split edge is a split block. */
+ sp->split_bbs.first_member = &bblst_table[bblst_last];
+ sp->split_bbs.nr_members = el.nr_members;
+ for (j = 0; j < el.nr_members; bblst_last++, j++)
+ bblst_table[bblst_last] =
+ TO_BLOCK (rgn_edges[el.first_member[j]]);
+ sp->update_bbs.first_member = &bblst_table[bblst_last];
+
+ /* Compute update blocks and store them in bblst_table.
+ For every split edge, look at the FROM block, and check
+ all out edges. For each out edge that is not a split edge,
+ 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_idx = 0;
+ for (j = 0; j < el.nr_members; j++)
+ {
+ check_block = FROM_BLOCK (rgn_edges[el.first_member[j]]);
+ fst_edge = nxt_edge = OUT_EDGES (check_block);
+ do
+ {
+ if (! update_blocks[TO_BLOCK (nxt_edge)])
+ {
+ for (k = 0; k < el.nr_members; k++)
+ if (EDGE_TO_BIT (nxt_edge) == el.first_member[k])
+ break;
+
+ if (k >= el.nr_members)
+ {
+ bblst_table[bblst_last++] = TO_BLOCK (nxt_edge);
+ update_blocks[TO_BLOCK (nxt_edge)] = 1;
+ update_idx++;
+ }
+ }
+
+ nxt_edge = NEXT_OUT (nxt_edge);
+ }
+ while (fst_edge != nxt_edge);
+ }
+ sp->update_bbs.nr_members = update_idx;
+
+ /* Make sure we didn't overrun the end of bblst_table. */
+ if (bblst_last > bblst_size)
+ abort ();
+ }
+ else
+ {
+ sp->split_bbs.nr_members = sp->update_bbs.nr_members = 0;
+
+ sp->is_speculative = 0;
+ sp->src_prob = 0;
+ }
+ }
+}
+
+/* Print candidates info, for debugging purposes. Callable from debugger. */
+
+void
+debug_candidate (i)
+ int i;
+{
+ if (!candidate_table[i].is_valid)
+ return;
+
+ if (candidate_table[i].is_speculative)
+ {
+ int j;
+ fprintf (sched_dump, "src b %d bb %d speculative \n", BB_TO_BLOCK (i), i);
+
+ fprintf (sched_dump, "split path: ");
+ for (j = 0; j < candidate_table[i].split_bbs.nr_members; j++)
+ {
+ int b = candidate_table[i].split_bbs.first_member[j];
+
+ fprintf (sched_dump, " %d ", b);
+ }
+ fprintf (sched_dump, "\n");
+
+ fprintf (sched_dump, "update path: ");
+ for (j = 0; j < candidate_table[i].update_bbs.nr_members; j++)
+ {
+ int b = candidate_table[i].update_bbs.first_member[j];
+
+ fprintf (sched_dump, " %d ", b);
+ }
+ fprintf (sched_dump, "\n");
+ }
+ else
+ {
+ fprintf (sched_dump, " src %d equivalent\n", BB_TO_BLOCK (i));
+ }
+}
+
+/* Print candidates info, for debugging purposes. Callable from debugger. */
+
+void
+debug_candidates (trg)
+ int trg;
+{
+ int i;
+
+ fprintf (sched_dump, "----------- candidate table: target: b=%d bb=%d ---\n",
+ BB_TO_BLOCK (trg), trg);
+ for (i = trg + 1; i < current_nr_blocks; i++)
+ debug_candidate (i);
+}
+
+/* Functions for speculative scheduing. */
+
+/* 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;
+{
+ register int i;
+ register int regno;
+ register rtx reg = SET_DEST (x);
+
+ if (reg == 0)
+ return 1;
+
+ while (GET_CODE (reg) == SUBREG || GET_CODE (reg) == ZERO_EXTRACT
+ || GET_CODE (reg) == SIGN_EXTRACT
+ || GET_CODE (reg) == STRICT_LOW_PART)
+ reg = XEXP (reg, 0);
+
+ if (GET_CODE (reg) == PARALLEL
+ && GET_MODE (reg) == BLKmode)
+ {
+ register int i;
+ for (i = XVECLEN (reg, 0) - 1; i >= 0; i--)
+ if (check_live_1 (src, XVECEXP (reg, 0, i)))
+ return 1;
+ return 0;
+ }
+
+ if (GET_CODE (reg) != REG)
+ return 1;
+
+ regno = REGNO (reg);
+
+ if (regno < FIRST_PSEUDO_REGISTER && global_regs[regno])
+ {
+ /* Global registers are assumed live. */
+ return 0;
+ }
+ else
+ {
+ if (regno < FIRST_PSEUDO_REGISTER)
+ {
+ /* Check for hard registers. */
+ int j = HARD_REGNO_NREGS (regno, GET_MODE (reg));
+ while (--j >= 0)
+ {
+ for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++)
+ {
+ int b = candidate_table[src].split_bbs.first_member[i];
+
+ if (REGNO_REG_SET_P (BASIC_BLOCK (b)->global_live_at_start,
+ regno + j))
+ {
+ return 0;
+ }
+ }
+ }
+ }
+ else
+ {
+ /* Check for psuedo registers. */
+ for (i = 0; i < candidate_table[src].split_bbs.nr_members; i++)
+ {
+ int b = candidate_table[src].split_bbs.first_member[i];
+
+ if (REGNO_REG_SET_P (BASIC_BLOCK (b)->global_live_at_start, regno))
+ {
+ return 0;
+ }
+ }
+ }
+ }
+
+ return 1;
+}
+
+/* If x is a set of a register R, mark that R is alive in the beginning
+ of every update-block of src. */
+
+static void
+update_live_1 (src, x)
+ int src;
+ rtx x;
+{
+ register int i;
+ register int regno;
+ register rtx reg = SET_DEST (x);
+
+ if (reg == 0)
+ return;
+
+ while (GET_CODE (reg) == SUBREG || GET_CODE (reg) == ZERO_EXTRACT
+ || GET_CODE (reg) == SIGN_EXTRACT
+ || GET_CODE (reg) == STRICT_LOW_PART)
+ reg = XEXP (reg, 0);
+
+ if (GET_CODE (reg) == PARALLEL
+ && GET_MODE (reg) == BLKmode)
+ {
+ register int i;
+ for (i = XVECLEN (reg, 0) - 1; i >= 0; i--)
+ update_live_1 (src, XVECEXP (reg, 0, i));
+ return;
+ }
+
+ if (GET_CODE (reg) != REG)
+ return;
+
+ /* Global registers are always live, so the code below does not apply
+ to them. */
+
+ regno = REGNO (reg);
+
+ if (regno >= FIRST_PSEUDO_REGISTER || !global_regs[regno])
+ {
+ if (regno < FIRST_PSEUDO_REGISTER)
+ {
+ int j = HARD_REGNO_NREGS (regno, GET_MODE (reg));
+ while (--j >= 0)
+ {
+ for (i = 0; i < candidate_table[src].update_bbs.nr_members; i++)
+ {
+ int b = candidate_table[src].update_bbs.first_member[i];
+
+ SET_REGNO_REG_SET (BASIC_BLOCK (b)->global_live_at_start,
+ regno + j);
+ }
+ }
+ }
+ else
+ {
+ for (i = 0; i < candidate_table[src].update_bbs.nr_members; i++)
+ {
+ int b = candidate_table[src].update_bbs.first_member[i];
+
+ SET_REGNO_REG_SET (BASIC_BLOCK (b)->global_live_at_start, regno);
+ }
+ }
+ }
+}
+
+/* Return 1 if insn can be speculatively moved from block src to trg,
+ otherwise return 0. Called before first insertion of insn to
+ ready-list or before the scheduling. */
+
+static int
+check_live (insn, src)
+ rtx insn;
+ int src;
+{
+ /* Find the registers set by instruction. */
+ if (GET_CODE (PATTERN (insn)) == SET
+ || GET_CODE (PATTERN (insn)) == CLOBBER)
+ return check_live_1 (src, PATTERN (insn));
+ else if (GET_CODE (PATTERN (insn)) == PARALLEL)
+ {
+ int j;
+ for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
+ if ((GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
+ || GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
+ && !check_live_1 (src, XVECEXP (PATTERN (insn), 0, j)))
+ return 0;
+
+ return 1;
+ }
+
+ return 1;
+}
+
+/* Update the live registers info after insn was moved speculatively from
+ block src to trg. */
+
+static void
+update_live (insn, src)
+ rtx insn;
+ int src;
+{
+ /* Find the registers set by instruction. */
+ if (GET_CODE (PATTERN (insn)) == SET
+ || GET_CODE (PATTERN (insn)) == CLOBBER)
+ update_live_1 (src, PATTERN (insn));
+ else if (GET_CODE (PATTERN (insn)) == PARALLEL)
+ {
+ int j;
+ for (j = XVECLEN (PATTERN (insn), 0) - 1; j >= 0; j--)
+ if (GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == SET
+ || GET_CODE (XVECEXP (PATTERN (insn), 0, j)) == CLOBBER)
+ update_live_1 (src, XVECEXP (PATTERN (insn), 0, j));
+ }
+}
+
+/* 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. */
+#define IS_REACHABLE(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)))
+
+/* Turns on the fed_by_spec_load flag for insns fed by load_insn. */
+
+static void
+set_spec_fed (load_insn)
+ rtx load_insn;
+{
+ rtx link;
+
+ for (link = INSN_DEPEND (load_insn); link; link = XEXP (link, 1))
+ if (GET_MODE (link) == VOIDmode)
+ FED_BY_SPEC_LOAD (XEXP (link, 0)) = 1;
+} /* set_spec_fed */
+
+/* On the path from the insn to load_insn_bb, find a conditional
+branch depending on insn, that guards the speculative load. */
+
+static int
+find_conditional_protection (insn, load_insn_bb)
+ rtx insn;
+ int load_insn_bb;
+{
+ rtx link;
+
+ /* Iterate through DEF-USE forward dependences. */
+ for (link = INSN_DEPEND (insn); link; link = XEXP (link, 1))
+ {
+ rtx next = XEXP (link, 0);
+ if ((CONTAINING_RGN (BLOCK_NUM (next)) ==
+ CONTAINING_RGN (BB_TO_BLOCK (load_insn_bb)))
+ && IS_REACHABLE (INSN_BB (next), load_insn_bb)
+ && load_insn_bb != INSN_BB (next)
+ && GET_MODE (link) == VOIDmode
+ && (GET_CODE (next) == JUMP_INSN
+ || find_conditional_protection (next, load_insn_bb)))
+ return 1;
+ }
+ return 0;
+} /* find_conditional_protection */
+
+/* Returns 1 if the same insn1 that participates in the computation
+ of load_insn's address is feeding a conditional branch that is
+ guarding on load_insn. This is true if we find a the two DEF-USE
+ chains:
+ insn1 -> ... -> conditional-branch
+ insn1 -> ... -> load_insn,
+ and if a flow path exist:
+ insn1 -> ... -> conditional-branch -> ... -> load_insn,
+ and if insn1 is on the path
+ region-entry -> ... -> bb_trg -> ... load_insn.
+
+ Locate insn1 by climbing on LOG_LINKS from load_insn.
+ 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;
+{
+ rtx link;
+
+ for (link = LOG_LINKS (load_insn); link; link = XEXP (link, 1))
+ {
+ rtx insn1 = XEXP (link, 0);
+
+ /* Must be a DEF-USE dependence upon non-branch. */
+ if (GET_MODE (link) != VOIDmode
+ || GET_CODE (insn1) == JUMP_INSN)
+ continue;
+
+ /* Must exist a path: region-entry -> ... -> bb_trg -> ... load_insn. */
+ if (INSN_BB (insn1) == bb_src
+ || (CONTAINING_RGN (BLOCK_NUM (insn1))
+ != CONTAINING_RGN (BB_TO_BLOCK (bb_src)))
+ || (!IS_REACHABLE (bb_trg, INSN_BB (insn1))
+ && !IS_REACHABLE (INSN_BB (insn1), bb_trg)))
+ continue;
+
+ /* Now search for the conditional-branch. */
+ if (find_conditional_protection (insn1, bb_src))
+ return 1;
+
+ /* Recursive step: search another insn1, "above" current insn1. */
+ return is_conditionally_protected (insn1, bb_src, bb_trg);
+ }
+
+ /* The chain does not exist. */
+ return 0;
+} /* is_conditionally_protected */
+
+/* Returns 1 if a clue for "similar load" 'insn2' is found, and hence
+ load_insn can move speculatively from bb_src to bb_trg. All the
+ following must hold:
+
+ (1) both loads have 1 base register (PFREE_CANDIDATEs).
+ (2) load_insn and load1 have a def-use dependence upon
+ the same insn 'insn1'.
+ (3) either load2 is in bb_trg, or:
+ - there's only one split-block, and
+ - load1 is on the escape path, and
+
+ From all these we can conclude that the two loads access memory
+ addresses that differ at most by a constant, and hence if moving
+ load_insn would cause an exception, it would have been caused by
+ load2 anyhow. */
+
+static int
+is_pfree (load_insn, bb_src, bb_trg)
+ rtx load_insn;
+ int bb_src, bb_trg;
+{
+ rtx back_link;
+ register candidate *candp = candidate_table + bb_src;
+
+ if (candp->split_bbs.nr_members != 1)
+ /* Must have exactly one escape block. */
+ return 0;
+
+ for (back_link = LOG_LINKS (load_insn);
+ back_link; back_link = XEXP (back_link, 1))
+ {
+ rtx insn1 = XEXP (back_link, 0);
+
+ if (GET_MODE (back_link) == VOIDmode)
+ {
+ /* Found a DEF-USE dependence (insn1, load_insn). */
+ rtx fore_link;
+
+ for (fore_link = INSN_DEPEND (insn1);
+ fore_link; fore_link = XEXP (fore_link, 1))
+ {
+ rtx insn2 = XEXP (fore_link, 0);
+ if (GET_MODE (fore_link) == VOIDmode)
+ {
+ /* Found a DEF-USE dependence (insn1, insn2). */
+ if (haifa_classify_insn (insn2) != PFREE_CANDIDATE)
+ /* insn2 not guaranteed to be a 1 base reg load. */
+ continue;
+
+ if (INSN_BB (insn2) == bb_trg)
+ /* insn2 is the similar load, in the target block. */
+ return 1;
+
+ if (*(candp->split_bbs.first_member) == BLOCK_NUM (insn2))
+ /* insn2 is a similar load, in a split-block. */
+ return 1;
+ }
+ }
+ }
+ }
+
+ /* Couldn't find a similar load. */
+ 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;
+{
+ if (FED_BY_SPEC_LOAD (load_insn))
+ return 1;
+
+ if (LOG_LINKS (load_insn) == NULL)
+ /* Dependence may 'hide' out of the region. */
+ return 1;
+
+ if (is_conditionally_protected (load_insn, bb_src, bb_trg))
+ return 1;
+
+ return 0;
+}
+
+/* Insn is a candidate to be moved speculatively from bb_src to bb_trg.
+ Return 1 if insn is exception-free (and the motion is valid)
+ and 0 otherwise. */
+
+static int
+is_exception_free (insn, bb_src, bb_trg)
+ rtx insn;
+ int bb_src, bb_trg;
+{
+ int insn_class = haifa_classify_insn (insn);
+
+ /* Handle non-load insns. */
+ switch (insn_class)
+ {
+ case TRAP_FREE:
+ return 1;
+ case TRAP_RISKY:
+ return 0;
+ default:;
+ }
+
+ /* Handle loads. */
+ if (!flag_schedule_speculative_load)
+ return 0;
+ IS_LOAD_INSN (insn) = 1;
+ switch (insn_class)
+ {
+ case IFREE:
+ return (1);
+ case IRISKY:
+ return 0;
+ case PFREE_CANDIDATE:
+ if (is_pfree (insn, bb_src, bb_trg))
+ return 1;
+ /* Don't 'break' here: PFREE-candidate is also PRISKY-candidate. */
+ case PRISKY_CANDIDATE:
+ if (!flag_schedule_speculative_load_dangerous
+ || is_prisky (insn, bb_src, bb_trg))
+ return 0;
+ break;
+ default:;
+ }
+
+ return flag_schedule_speculative_load_dangerous;
+}
+\f
+/* The number of insns from the current block scheduled so far. */
+static int sched_target_n_insns;
+/* The number of insns from the current block to be scheduled in total. */
+static int target_n_insns;
+/* The number of insns from the entire region scheduled so far. */
+static int sched_n_insns;
+
+/* 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));
+
+/* Return nonzero if there are more insns that should be scheduled. */
+
+static int
+schedule_more_p ()
+{
+ return sched_target_n_insns < target_n_insns;
+}
+
+/* Add all insns that are initially ready to the ready list READY. Called
+ once before scheduling a set of insns. */
+
+static void
+init_ready_list (ready)
+ struct ready_list *ready;
+{
+ rtx prev_head = current_sched_info->prev_head;
+ rtx next_tail = current_sched_info->next_tail;
+ int bb_src;
+ rtx insn;
+
+ target_n_insns = 0;
+ sched_target_n_insns = 0;
+ sched_n_insns = 0;
+
+ /* Print debugging information. */
+ if (sched_verbose >= 5)
+ debug_dependencies ();
+
+ /* Prepare current target block info. */
+ if (current_nr_blocks > 1)
+ {
+ candidate_table = (candidate *) xmalloc (current_nr_blocks
+ * sizeof (candidate));
+
+ bblst_last = 0;
+ /* bblst_table holds split blocks and update blocks for each block after
+ the current one in the region. split blocks and update blocks are
+ 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));
+
+ bitlst_table_last = 0;
+ bitlst_table_size = rgn_nr_edges;
+ bitlst_table = (int *) xmalloc (rgn_nr_edges * sizeof (int));
+
+ compute_trg_info (target_bb);
+ }
+
+ /* Initialize ready list with all 'ready' insns in target block.
+ 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
+ && (SCHED_GROUP_P (next) == 0 || ! INSN_P (next)))
+ ready_add (ready, insn);
+ if (!(SCHED_GROUP_P (insn)))
+ target_n_insns++;
+ }
+
+ /* Add to ready list all 'ready' insns in valid source blocks.
+ For speculative insns, check-live, exception-free, and
+ issue-delay. */
+ for (bb_src = target_bb + 1; bb_src < current_nr_blocks; bb_src++)
+ if (IS_VALID (bb_src))
+ {
+ rtx src_head;
+ rtx src_next_tail;
+ rtx tail, head;
+
+ get_block_head_tail (BB_TO_BLOCK (bb_src), &head, &tail);
+ src_next_tail = NEXT_INSN (tail);
+ src_head = head;
+
+ for (insn = src_head; insn != src_next_tail; insn = NEXT_INSN (insn))
+ {
+ if (! INSN_P (insn))
+ continue;
+
+ if (!CANT_MOVE (insn)
+ && (!IS_SPECULATIVE_INSN (insn)
+ || (insn_issue_delay (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);
+ }
+ }
+ }
+}
+
+/* Called after taking INSN from the ready list. Returns nonzero if this
+ insn can be scheduled, nonzero if we should silently discard it. */
+
+static int
+can_schedule_ready_p (insn)
+ rtx insn;
+{
+ /* An interblock motion? */
+ if (INSN_BB (insn) != target_bb)
+ {
+ rtx temp;
+ basic_block b1;
+
+ if (IS_SPECULATIVE_INSN (insn))
+ {
+ if (!check_live (insn, INSN_BB (insn)))
+ return 0;
+ update_live (insn, INSN_BB (insn));
+
+ /* For speculative load, mark insns fed by it. */
+ if (IS_LOAD_INSN (insn) || FED_BY_SPEC_LOAD (insn))
+ set_spec_fed (insn);
+
+ nr_spec++;
+ }
+ 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)
+ {
+ /* 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;
+ }
+ else if (insn == b1->end)
+ {
+ /* 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);
+ }
+ else if (temp == b1->head)
+ {
+ /* 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);
+ }
+ }
+ else
+ {
+ /* In block motion. */
+ sched_target_n_insns++;
+ }
+ sched_n_insns++;
+
+ return 1;
+}
+
+/* Called after INSN has all its dependencies resolved. Return nonzero
+ 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;
+{
+ /* For speculative insns, before inserting to ready/queue,
+ check live, exception-free, and issue-delay. */
+ if (INSN_BB (next) != target_bb
+ && (!IS_VALID (INSN_BB (next))
+ || CANT_MOVE (next)
+ || (IS_SPECULATIVE_INSN (next)
+ && (insn_issue_delay (next) > 3
+ || !check_live (next, INSN_BB (next))
+ || !is_exception_free (next, INSN_BB (next), target_bb)))))
+ return 0;
+ return 1;
+}
+
+/* Return a string that contains the insn uid and optionally anything else
+ necessary to identify this insn in an output. It's valid to use a
+ static buffer for this. The ALIGNED parameter should cause the string
+ to be formatted so that multiple output lines will line up nicely. */
+
+static const char *
+rgn_print_insn (insn, aligned)
+ rtx insn;
+ int aligned;
+{
+ static char tmp[80];
+
+ if (aligned)
+ sprintf (tmp, "b%3d: i%4d", INSN_BB (insn), INSN_UID (insn));
+ else
+ {
+ sprintf (tmp, "%d", INSN_UID (insn));
+ if (current_nr_blocks > 1 && INSN_BB (insn) != target_bb)
+ sprintf (tmp, "/b%d ", INSN_BB (insn));
+ }
+ return tmp;
+}
+
+/* Compare priority of two insns. Return a positive number if the second
+ insn is to be preferred for scheduling, and a negative one if the first
+ is to be preferred. Zero if they are equally good. */
+
+static int
+rgn_rank (insn1, insn2)
+ rtx insn1, insn2;
+{
+ /* Some comparison make sense in interblock scheduling only. */
+ if (INSN_BB (insn1) != INSN_BB (insn2))
+ {
+ int spec_val, prob_val;
+
+ /* Prefer an inblock motion on an interblock motion. */
+ if ((INSN_BB (insn2) == target_bb) && (INSN_BB (insn1) != target_bb))
+ return 1;
+ if ((INSN_BB (insn1) == target_bb) && (INSN_BB (insn2) != target_bb))
+ return -1;
+
+ /* Prefer a useful motion on a speculative one. */
+ spec_val = IS_SPECULATIVE_INSN (insn1) - IS_SPECULATIVE_INSN (insn2);
+ if (spec_val)
+ return spec_val;
+
+ /* Prefer a more probable (speculative) insn. */
+ prob_val = INSN_PROBABILITY (insn2) - INSN_PROBABILITY (insn1);
+ if (prob_val)
+ return prob_val;
+ }
+ return 0;
+}
+
+/* Used in schedule_insns to initialize current_sched_info for scheduling
+ regions (or single basic blocks). */
+
+static struct sched_info region_sched_info =
+{
+ init_ready_list,
+ can_schedule_ready_p,
+ schedule_more_p,
+ new_ready,
+ rgn_rank,
+ rgn_print_insn,
+
+ NULL, NULL,
+ NULL, NULL,
+ 0
+};
+
+/* Add dependences so that branches are scheduled to run last in their
+ block. */
+
+static void
+add_branch_dependences (head, tail)
+ rtx head, 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.
+
+ 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. */
+ insn = tail;
+ last = 0;
+ while (GET_CODE (insn) == CALL_INSN
+ || GET_CODE (insn) == JUMP_INSN
+ || (GET_CODE (insn) == INSN
+ && (GET_CODE (PATTERN (insn)) == USE
+ || GET_CODE (PATTERN (insn)) == CLOBBER
+#ifdef HAVE_cc0
+ || sets_cc0_p (PATTERN (insn))
+#endif
+ ))
+ || GET_CODE (insn) == NOTE)
+ {
+ if (GET_CODE (insn) != NOTE)
+ {
+ if (last != 0
+ && !find_insn_list (insn, LOG_LINKS (last)))
+ {
+ add_dependence (last, insn, REG_DEP_ANTI);
+ INSN_REF_COUNT (insn)++;
+ }
+
+ 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. */
+ if (insn == head)
+ break;
+
+ insn = PREV_INSN (insn);
+ }
+
+ /* Make sure these insns are scheduled last in their block. */
+ insn = last;
+ if (insn != 0)
+ while (insn != head)
+ {
+ insn = prev_nonnote_insn (insn);
+
+ if (INSN_REF_COUNT (insn) != 0)
+ continue;
+
+ 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);
+ }
+}
+
+/* Data structures for the computation of data dependences in a regions. We
+ keep one `deps' structure for every basic block. Before analyzing the
+ data dependences for a bb, its variables are initialized as a function of
+ the variables of its predecessors. When the analysis for a bb completes,
+ we save the contents to the corresponding bb_deps[bb] variable. */
+
+static struct deps *bb_deps;
+
+/* After computing the dependencies for block BB, propagate the dependencies
+ found in TMP_DEPS to the successors of the block. MAX_REG is the number
+ of registers. */
+static void
+propagate_deps (bb, tmp_deps, max_reg)
+ int bb;
+ struct deps *tmp_deps;
+ int max_reg;
+{
+ 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;
+ }
+
+ for (reg = 0; reg < max_reg; reg++)
+ {
+ /* reg-last-uses lists are inherited by bb_succ. */
+ for (u = tmp_deps->reg_last_uses[reg]; u; u = XEXP (u, 1))
+ {
+ if (find_insn_list (XEXP (u, 0),
+ succ_deps->reg_last_uses[reg]))
+ continue;
+
+ succ_deps->reg_last_uses[reg]
+ = alloc_INSN_LIST (XEXP (u, 0),
+ succ_deps->reg_last_uses[reg]);
+ }
+
+ /* reg-last-defs lists are inherited by bb_succ. */
+ for (u = tmp_deps->reg_last_sets[reg]; u; u = XEXP (u, 1))
+ {
+ if (find_insn_list (XEXP (u, 0),
+ succ_deps->reg_last_sets[reg]))
+ continue;
+
+ succ_deps->reg_last_sets[reg]
+ = alloc_INSN_LIST (XEXP (u, 0),
+ succ_deps->reg_last_sets[reg]);
+ }
+
+ for (u = tmp_deps->reg_last_clobbers[reg]; u; u = XEXP (u, 1))
+ {
+ if (find_insn_list (XEXP (u, 0),
+ succ_deps->reg_last_clobbers[reg]))
+ continue;
+
+ succ_deps->reg_last_clobbers[reg]
+ = alloc_INSN_LIST (XEXP (u, 0),
+ succ_deps->reg_last_clobbers[reg]);
+ }
+ }
+
+ /* 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);
+ }
+
+ 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);
+ }
+
+ /* 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))
+ continue;
+
+ 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))
+ continue;
+
+ 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);
+}
+
+/* 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.
+
+ Specifically for reg-reg data dependences, the block insns are
+ scanned by sched_analyze () top-to-bottom. Two lists are
+ maintained by sched_analyze (): reg_last_sets[] for register DEFs,
+ and reg_last_uses[] for register USEs.
+
+ When analysis is completed for bb, we update for its successors:
+ ; - DEFS[succ] = Union (DEFS [succ], DEFS [bb])
+ ; - USES[succ] = Union (USES [succ], DEFS [bb])
+
+ The mechanism for computing mem-mem data dependence is very
+ similar, and the result is interblock dependences in the region. */
+
+static void
+compute_block_backward_dependences (bb)
+ int bb;
+{
+ rtx head, tail;
+ int max_reg = max_reg_num ();
+ struct deps tmp_deps;
+
+ tmp_deps = bb_deps[bb];
+
+ /* Do the analysis for this block. */
+ get_block_head_tail (BB_TO_BLOCK (bb), &head, &tail);
+ sched_analyze (&tmp_deps, head, tail);
+ add_branch_dependences (head, tail);
+
+ if (current_nr_blocks > 1)
+ propagate_deps (bb, &tmp_deps, max_reg);
+
+ /* Free up the INSN_LISTs. */
+ free_deps (&tmp_deps);
+
+ /* Assert that we won't need bb_reg_last_* for this block anymore. */
+ free (bb_deps[bb].reg_last_uses);
+ free (bb_deps[bb].reg_last_sets);
+ free (bb_deps[bb].reg_last_clobbers);
+ bb_deps[bb].reg_last_uses = 0;
+ bb_deps[bb].reg_last_sets = 0;
+ bb_deps[bb].reg_last_clobbers = 0;
+}
+/* Remove all INSN_LISTs and EXPR_LISTs from the pending lists and add
+ them to the unused_*_list variables, so that they can be reused. */
+
+static void
+free_pending_lists ()
+{
+ int bb;
+
+ for (bb = 0; bb < current_nr_blocks; bb++)
+ {
+ free_INSN_LIST_list (&bb_deps[bb].pending_read_insns);
+ free_INSN_LIST_list (&bb_deps[bb].pending_write_insns);
+ free_EXPR_LIST_list (&bb_deps[bb].pending_read_mems);
+ free_EXPR_LIST_list (&bb_deps[bb].pending_write_mems);
+ }
+}
+\f
+/* Print dependences for debugging, callable from debugger. */
+
+void
+debug_dependencies ()
+{
+ int bb;
+
+ fprintf (sched_dump, ";; --------------- forward dependences: ------------ \n");
+ for (bb = 0; bb < current_nr_blocks; bb++)
+ {
+ if (1)
+ {
+ rtx head, tail;
+ rtx next_tail;
+ rtx insn;
+
+ get_block_head_tail (BB_TO_BLOCK (bb), &head, &tail);
+ next_tail = NEXT_INSN (tail);
+ 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",
+ "----", "----", "--", "---", "----", "----", "--------", "-----");
+ for (insn = head; insn != next_tail; insn = NEXT_INSN (insn))
+ {
+ rtx link;
+ int unit, range;
+
+ if (! INSN_P (insn))
+ {
+ int n;
+ fprintf (sched_dump, ";; %6d ", INSN_UID (insn));
+ if (GET_CODE (insn) == NOTE)
+ {
+ n = NOTE_LINE_NUMBER (insn);
+ if (n < 0)
+ fprintf (sched_dump, "%s\n", GET_NOTE_INSN_NAME (n));
+ else
+ fprintf (sched_dump, "line %d, file %s\n", n,
+ NOTE_SOURCE_FILE (insn));
+ }
+ else
+ fprintf (sched_dump, " {%s}\n", GET_RTX_NAME (GET_CODE (insn)));
+ 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);
+ fprintf (sched_dump, "\t: ");
+ for (link = INSN_DEPEND (insn); link; link = XEXP (link, 1))
+ fprintf (sched_dump, "%d ", INSN_UID (XEXP (link, 0)));
+ fprintf (sched_dump, "\n");
+ }
+ }
+ }
+ fprintf (sched_dump, "\n");
+}
+\f
+/* Schedule a region. A region is either an inner loop, a loop-free
+ subroutine, or a single basic block. Each bb in the region is
+ scheduled after its flow predecessors. */
+
+static void
+schedule_region (rgn)
+ int rgn;
+{
+ int bb;
+ int rgn_n_insns = 0;
+ int sched_rgn_n_insns = 0;
+
+ /* Set variables for the current region. */
+ current_nr_blocks = RGN_NR_BLOCKS (rgn);
+ current_blocks = RGN_BLOCKS (rgn);
+
+ init_deps_global ();
+
+ /* Initializations for region data dependence analyisis. */
+ bb_deps = (struct deps *) xmalloc (sizeof (struct deps) * current_nr_blocks);
+ for (bb = 0; bb < current_nr_blocks; bb++)
+ init_deps (bb_deps + bb);
+
+ /* Compute LOG_LINKS. */
+ for (bb = 0; bb < current_nr_blocks; bb++)
+ compute_block_backward_dependences (bb);
+
+ /* Compute INSN_DEPEND. */
+ for (bb = current_nr_blocks - 1; bb >= 0; bb--)
+ {
+ rtx head, tail;
+ get_block_head_tail (BB_TO_BLOCK (bb), &head, &tail);
+
+ compute_forward_dependences (head, tail);
+ }
+
+ /* Set priorities. */
+ for (bb = 0; bb < current_nr_blocks; bb++)
+ rgn_n_insns += set_priorities (BB_TO_BLOCK (bb));
+
+ /* Compute interblock info: probabilities, split-edges, dominators, etc. */
+ if (current_nr_blocks > 1)
+ {
+ 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));
+
+ /* Edge to bit. */
+ rgn_nr_edges = 0;
+ edge_to_bit = (int *) 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_nr_edges = 0;
+ for (i = 1; i < nr_edges; i++)
+ if (CONTAINING_RGN (FROM_BLOCK (i)) == (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));
+ }
+
+ /* Compute probabilities, dominators, split_edges. */
+ for (bb = 0; bb < current_nr_blocks; bb++)
+ compute_dom_prob_ps (bb);
+ }
+
+ /* Now we can schedule all blocks. */
+ for (bb = 0; bb < current_nr_blocks; bb++)
+ {
+ rtx head, tail;
+ int b = BB_TO_BLOCK (bb);
+
+ get_block_head_tail (b, &head, &tail);
+
+ if (no_real_insns_p (head, tail))
+ continue;
+
+ current_sched_info->prev_head = PREV_INSN (head);
+ current_sched_info->next_tail = NEXT_INSN (tail);
+
+ if (write_symbols != NO_DEBUG)
+ {
+ save_line_notes (b);
+ rm_line_notes (b);
+ }
+
+ /* 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
+ 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.
+
+ FIXME: Probably the same thing should be done with REG_SAVE_NOTEs
+ referencing NOTE_INSN_SETJMP at the end of the block. */
+ if (INSN_P (head))
+ {
+ rtx note;
+
+ for (note = REG_NOTES (head); note; note = XEXP (note, 1))
+ if (REG_NOTE_KIND (note) == REG_SAVE_NOTE)
+ {
+ if (INTVAL (XEXP (note, 0)) != NOTE_INSN_SETJMP)
+ {
+ remove_note (head, note);
+ note = XEXP (note, 1);
+ remove_note (head, note);
+ }
+ else
+ note = XEXP (note, 1);
+ }
+ }
+
+ /* Remove remaining note insns from the block, save them in
+ note_list. These notes are restored at the end of
+ schedule_block (). */
+ rm_other_notes (head, tail);
+
+ target_bb = bb;
+
+ current_sched_info->queue_must_finish_empty
+ = current_nr_blocks > 1 && !flag_schedule_interblock;
+
+ schedule_block (b, rgn_n_insns);
+ 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;
+
+ /* Clean up. */
+ if (current_nr_blocks > 1)
+ {
+ free (candidate_table);
+ free (bblst_table);
+ free (bitlst_table);
+ }
+ }
+
+ /* Sanity check: verify that all region insns were scheduled. */
+ if (sched_rgn_n_insns != rgn_n_insns)
+ abort ();
+
+ /* Restore line notes. */
+ if (write_symbols != NO_DEBUG)
+ {
+ for (bb = 0; bb < current_nr_blocks; bb++)
+ restore_line_notes (BB_TO_BLOCK (bb));
+ }
+
+ /* Done with this region. */
+ free_pending_lists ();
+
+ finish_deps_global ();
+
+ free (bb_deps);
+
+ 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);
+ free (edge_to_bit);
+ free (rgn_edges);
+ free (pot_split);
+ free (ancestor_edges);
+ }
+}
+
+/* Indexed by region, holds the number of death notes found in that region.
+ Used for consistency checks. */
+static int *deaths_in_region;
+
+/* Initialize data structures for region scheduling. */
+
+static void
+init_regions ()
+{
+ 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);
+
+ /* Compute regions for scheduling. */
+ if (reload_completed
+ || n_basic_blocks == 1
+ || !flag_schedule_interblock)
+ {
+ find_single_block_region ();
+ }
+ else
+ {
+ /* Verify that a 'good' control flow graph can be built. */
+ if (is_cfg_nonregular ())
+ {
+ find_single_block_region ();
+ }
+ 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. */
+ edge_list = create_edge_list ();
+
+ /* Compute the dominators and post dominators. */
+ calculate_dominance_info (NULL, dom, CDI_DOMINATORS);
+
+ /* build_control_flow will return nonzero if it detects unreachable
+ blocks or any other irregularity with the cfg which prevents
+ cross block scheduling. */
+ if (build_control_flow (edge_list) != 0)
+ find_single_block_region ();
+ else
+ find_rgns (edge_list, dom);
+
+ if (sched_verbose >= 3)
+ debug_regions ();
+
+ /* We are done with flow's edge list. */
+ free_edge_list (edge_list);
+
+ /* For now. This will move as more and more of haifa is converted
+ to using the cfg code in flow.c. */
+ free (dom);
+ }
+ }
+
+ deaths_in_region = (int *) 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]);
+
+ deaths_in_region[rgn] = count_or_remove_death_notes (blocks, 1);
+ }
+
+ sbitmap_free (blocks);
+}
+
+/* The one entry point in this file. DUMP_FILE is the dump file for
+ this pass. */
+
+void
+schedule_insns (dump_file)
+ FILE *dump_file;
+{
+ sbitmap large_region_blocks, blocks;
+ int rgn;
+ int any_large_regions;
+
+ /* Taking care of this degenerate case makes the rest of
+ this code simpler. */
+ if (n_basic_blocks == 0)
+ return;
+
+ nr_inter = 0;
+ nr_spec = 0;
+
+ sched_init (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. */
+ /* ??? There is an outside possibility that update_life_info, or more
+ to the point propagate_block, could get called with non-zero 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.
+
+ I'm fairly certain that this _shouldn't_ happen, since I don't think
+ that live_at_start should change at region heads. Not sure what the
+ best way to test for this kind of thing... */
+
+ allocate_reg_life_data ();
+ compute_bb_for_insn (get_max_uid ());
+
+ any_large_regions = 0;
+ large_region_blocks = sbitmap_alloc (n_basic_blocks);
+ sbitmap_ones (large_region_blocks);
+
+ blocks = sbitmap_alloc (n_basic_blocks);
+
+ 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
+ }
+
+ if (any_large_regions)
+ {
+ update_life_info (large_region_blocks, UPDATE_LIFE_GLOBAL,
+ PROP_DEATH_NOTES | PROP_REG_INFO);
+ }
+
+ /* Reposition the prologue and epilogue notes in case we moved the
+ prologue/epilogue insns. */
+ if (reload_completed)
+ reposition_prologue_and_epilogue_notes (get_insns ());
+
+ /* Delete redundant line notes. */
+ if (write_symbols != NO_DEBUG)
+ rm_redundant_line_notes ();
+
+ if (sched_verbose)
+ {
+ if (reload_completed == 0 && flag_schedule_interblock)
+ {
+ fprintf (sched_dump,
+ "\n;; Procedure interblock/speculative motions == %d/%d \n",
+ nr_inter, nr_spec);
+ }
+ else
+ {
+ if (nr_inter > 0)
+ abort ();
+ }
+ fprintf (sched_dump, "\n\n");
+ }
+
+ /* Clean up. */
+ free (rgn_table);
+ free (rgn_bb_table);
+ free (block_to_bb);
+ free (containing_rgn);
+
+ sched_finish ();
+
+ if (edge_table)
+ {
+ free (edge_table);
+ edge_table = NULL;
+ }
+
+ if (in_edges)
+ {
+ free (in_edges);
+ in_edges = NULL;
+ }
+ if (out_edges)
+ {
+ free (out_edges);
+ out_edges = NULL;
+ }
+
+ sbitmap_free (blocks);
+ sbitmap_free (large_region_blocks);
+
+ free (deaths_in_region);
+}