/* Partial redundancy elimination / Hoisting for RTL. Copyright (C) 1997-2015 Free Software Foundation, Inc. This file is part of GCC. GCC is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3, or (at your option) any later version. GCC 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 GCC; see the file COPYING3. If not see . */ /* TODO - reordering of memory allocation and freeing to be more space efficient - calc rough register pressure information and use the info to drive all kinds of code motion (including code hoisting) in a unified way. */ /* References searched while implementing this. Compilers Principles, Techniques and Tools Aho, Sethi, Ullman Addison-Wesley, 1988 Global Optimization by Suppression of Partial Redundancies E. Morel, C. Renvoise communications of the acm, Vol. 22, Num. 2, Feb. 1979 A Portable Machine-Independent Global Optimizer - Design and Measurements Frederick Chow Stanford Ph.D. thesis, Dec. 1983 A Fast Algorithm for Code Movement Optimization D.M. Dhamdhere SIGPLAN Notices, Vol. 23, Num. 10, Oct. 1988 A Solution to a Problem with Morel and Renvoise's Global Optimization by Suppression of Partial Redundancies K-H Drechsler, M.P. Stadel ACM TOPLAS, Vol. 10, Num. 4, Oct. 1988 Practical Adaptation of the Global Optimization Algorithm of Morel and Renvoise D.M. Dhamdhere ACM TOPLAS, Vol. 13, Num. 2. Apr. 1991 Efficiently Computing Static Single Assignment Form and the Control Dependence Graph R. Cytron, J. Ferrante, B.K. Rosen, M.N. Wegman, and F.K. Zadeck ACM TOPLAS, Vol. 13, Num. 4, Oct. 1991 Lazy Code Motion J. Knoop, O. Ruthing, B. Steffen ACM SIGPLAN Notices Vol. 27, Num. 7, Jul. 1992, '92 Conference on PLDI What's In a Region? Or Computing Control Dependence Regions in Near-Linear Time for Reducible Flow Control Thomas Ball ACM Letters on Programming Languages and Systems, Vol. 2, Num. 1-4, Mar-Dec 1993 An Efficient Representation for Sparse Sets Preston Briggs, Linda Torczon ACM Letters on Programming Languages and Systems, Vol. 2, Num. 1-4, Mar-Dec 1993 A Variation of Knoop, Ruthing, and Steffen's Lazy Code Motion K-H Drechsler, M.P. Stadel ACM SIGPLAN Notices, Vol. 28, Num. 5, May 1993 Partial Dead Code Elimination J. Knoop, O. Ruthing, B. Steffen ACM SIGPLAN Notices, Vol. 29, Num. 6, Jun. 1994 Effective Partial Redundancy Elimination P. Briggs, K.D. Cooper ACM SIGPLAN Notices, Vol. 29, Num. 6, Jun. 1994 The Program Structure Tree: Computing Control Regions in Linear Time R. Johnson, D. Pearson, K. Pingali ACM SIGPLAN Notices, Vol. 29, Num. 6, Jun. 1994 Optimal Code Motion: Theory and Practice J. Knoop, O. Ruthing, B. Steffen ACM TOPLAS, Vol. 16, Num. 4, Jul. 1994 The power of assignment motion J. Knoop, O. Ruthing, B. Steffen ACM SIGPLAN Notices Vol. 30, Num. 6, Jun. 1995, '95 Conference on PLDI Global code motion / global value numbering C. Click ACM SIGPLAN Notices Vol. 30, Num. 6, Jun. 1995, '95 Conference on PLDI Value Driven Redundancy Elimination L.T. Simpson Rice University Ph.D. thesis, Apr. 1996 Value Numbering L.T. Simpson Massively Scalar Compiler Project, Rice University, Sep. 1996 High Performance Compilers for Parallel Computing Michael Wolfe Addison-Wesley, 1996 Advanced Compiler Design and Implementation Steven Muchnick Morgan Kaufmann, 1997 Building an Optimizing Compiler Robert Morgan Digital Press, 1998 People wishing to speed up the code here should read: Elimination Algorithms for Data Flow Analysis B.G. Ryder, M.C. Paull ACM Computing Surveys, Vol. 18, Num. 3, Sep. 1986 How to Analyze Large Programs Efficiently and Informatively D.M. Dhamdhere, B.K. Rosen, F.K. Zadeck ACM SIGPLAN Notices Vol. 27, Num. 7, Jul. 1992, '92 Conference on PLDI People wishing to do something different can find various possibilities in the above papers and elsewhere. */ #include "config.h" #include "system.h" #include "coretypes.h" #include "tm.h" #include "diagnostic-core.h" #include "toplev.h" #include "hard-reg-set.h" #include "rtl.h" #include "alias.h" #include "symtab.h" #include "tree.h" #include "tm_p.h" #include "regs.h" #include "ira.h" #include "flags.h" #include "insn-config.h" #include "recog.h" #include "predict.h" #include "function.h" #include "dominance.h" #include "cfg.h" #include "cfgrtl.h" #include "cfganal.h" #include "lcm.h" #include "cfgcleanup.h" #include "basic-block.h" #include "expmed.h" #include "dojump.h" #include "explow.h" #include "calls.h" #include "emit-rtl.h" #include "varasm.h" #include "stmt.h" #include "expr.h" #include "except.h" #include "params.h" #include "alloc-pool.h" #include "cselib.h" #include "intl.h" #include "obstack.h" #include "tree-pass.h" #include "df.h" #include "dbgcnt.h" #include "target.h" #include "gcse.h" #include "gcse-common.h" /* We support GCSE via Partial Redundancy Elimination. PRE optimizations are a superset of those done by classic GCSE. Two passes of copy/constant propagation are done around PRE or hoisting because the first one enables more GCSE and the second one helps to clean up the copies that PRE and HOIST create. This is needed more for PRE than for HOIST because code hoisting will try to use an existing register containing the common subexpression rather than create a new one. This is harder to do for PRE because of the code motion (which HOIST doesn't do). Expressions we are interested in GCSE-ing are of the form (set (pseudo-reg) (expression)). Function want_to_gcse_p says what these are. In addition, expressions in REG_EQUAL notes are candidates for GCSE-ing. This allows PRE to hoist expressions that are expressed in multiple insns, such as complex address calculations (e.g. for PIC code, or loads with a high part and a low part). PRE handles moving invariant expressions out of loops (by treating them as partially redundant). ********************** We used to support multiple passes but there are diminishing returns in doing so. The first pass usually makes 90% of the changes that are doable. A second pass can make a few more changes made possible by the first pass. Experiments show any further passes don't make enough changes to justify the expense. A study of spec92 using an unlimited number of passes: [1 pass] = 1208 substitutions, [2] = 577, [3] = 202, [4] = 192, [5] = 83, [6] = 34, [7] = 17, [8] = 9, [9] = 4, [10] = 4, [11] = 2, [12] = 2, [13] = 1, [15] = 1, [16] = 2, [41] = 1 It was found doing copy propagation between each pass enables further substitutions. This study was done before expressions in REG_EQUAL notes were added as candidate expressions for optimization, and before the GIMPLE optimizers were added. Probably, multiple passes is even less efficient now than at the time when the study was conducted. PRE is quite expensive in complicated functions because the DFA can take a while to converge. Hence we only perform one pass. ********************** The steps for PRE are: 1) Build the hash table of expressions we wish to GCSE (expr_hash_table). 2) Perform the data flow analysis for PRE. 3) Delete the redundant instructions 4) Insert the required copies [if any] that make the partially redundant instructions fully redundant. 5) For other reaching expressions, insert an instruction to copy the value to a newly created pseudo that will reach the redundant instruction. The deletion is done first so that when we do insertions we know which pseudo reg to use. Various papers have argued that PRE DFA is expensive (O(n^2)) and others argue it is not. The number of iterations for the algorithm to converge is typically 2-4 so I don't view it as that expensive (relatively speaking). PRE GCSE depends heavily on the second CPROP pass to clean up the copies we create. To make an expression reach the place where it's redundant, the result of the expression is copied to a new register, and the redundant expression is deleted by replacing it with this new register. Classic GCSE doesn't have this problem as much as it computes the reaching defs of each register in each block and thus can try to use an existing register. */ /* GCSE global vars. */ struct target_gcse default_target_gcse; #if SWITCHABLE_TARGET struct target_gcse *this_target_gcse = &default_target_gcse; #endif /* Set to non-zero if CSE should run after all GCSE optimizations are done. */ int flag_rerun_cse_after_global_opts; /* An obstack for our working variables. */ static struct obstack gcse_obstack; /* Hash table of expressions. */ struct gcse_expr { /* The expression. */ rtx expr; /* Index in the available expression bitmaps. */ int bitmap_index; /* Next entry with the same hash. */ struct gcse_expr *next_same_hash; /* List of anticipatable occurrences in basic blocks in the function. An "anticipatable occurrence" is one that is the first occurrence in the basic block, the operands are not modified in the basic block prior to the occurrence and the output is not used between the start of the block and the occurrence. */ struct gcse_occr *antic_occr; /* List of available occurrence in basic blocks in the function. An "available occurrence" is one that is the last occurrence in the basic block and the operands are not modified by following statements in the basic block [including this insn]. */ struct gcse_occr *avail_occr; /* Non-null if the computation is PRE redundant. The value is the newly created pseudo-reg to record a copy of the expression in all the places that reach the redundant copy. */ rtx reaching_reg; /* Maximum distance in instructions this expression can travel. We avoid moving simple expressions for more than a few instructions to keep register pressure under control. A value of "0" removes restrictions on how far the expression can travel. */ int max_distance; }; /* Occurrence of an expression. There is one per basic block. If a pattern appears more than once the last appearance is used [or first for anticipatable expressions]. */ struct gcse_occr { /* Next occurrence of this expression. */ struct gcse_occr *next; /* The insn that computes the expression. */ rtx_insn *insn; /* Nonzero if this [anticipatable] occurrence has been deleted. */ char deleted_p; /* Nonzero if this [available] occurrence has been copied to reaching_reg. */ /* ??? This is mutually exclusive with deleted_p, so they could share the same byte. */ char copied_p; }; typedef struct gcse_occr *occr_t; /* Expression hash tables. Each hash table is an array of buckets. ??? It is known that if it were an array of entries, structure elements `next_same_hash' and `bitmap_index' wouldn't be necessary. However, it is not clear whether in the final analysis a sufficient amount of memory would be saved as the size of the available expression bitmaps would be larger [one could build a mapping table without holes afterwards though]. Someday I'll perform the computation and figure it out. */ struct gcse_hash_table_d { /* The table itself. This is an array of `expr_hash_table_size' elements. */ struct gcse_expr **table; /* Size of the hash table, in elements. */ unsigned int size; /* Number of hash table elements. */ unsigned int n_elems; }; /* Expression hash table. */ static struct gcse_hash_table_d expr_hash_table; /* This is a list of expressions which are MEMs and will be used by load or store motion. Load motion tracks MEMs which aren't killed by anything except itself, i.e. loads and stores to a single location. We can then allow movement of these MEM refs with a little special allowance. (all stores copy the same value to the reaching reg used for the loads). This means all values used to store into memory must have no side effects so we can re-issue the setter value. */ struct ls_expr { struct gcse_expr * expr; /* Gcse expression reference for LM. */ rtx pattern; /* Pattern of this mem. */ rtx pattern_regs; /* List of registers mentioned by the mem. */ rtx_insn_list *loads; /* INSN list of loads seen. */ rtx_insn_list *stores; /* INSN list of stores seen. */ struct ls_expr * next; /* Next in the list. */ int invalid; /* Invalid for some reason. */ int index; /* If it maps to a bitmap index. */ unsigned int hash_index; /* Index when in a hash table. */ rtx reaching_reg; /* Register to use when re-writing. */ }; /* Head of the list of load/store memory refs. */ static struct ls_expr * pre_ldst_mems = NULL; struct pre_ldst_expr_hasher : nofree_ptr_hash { typedef value_type compare_type; static inline hashval_t hash (const ls_expr *); static inline bool equal (const ls_expr *, const ls_expr *); }; /* Hashtable helpers. */ inline hashval_t pre_ldst_expr_hasher::hash (const ls_expr *x) { int do_not_record_p = 0; return hash_rtx (x->pattern, GET_MODE (x->pattern), &do_not_record_p, NULL, false); } static int expr_equiv_p (const_rtx, const_rtx); inline bool pre_ldst_expr_hasher::equal (const ls_expr *ptr1, const ls_expr *ptr2) { return expr_equiv_p (ptr1->pattern, ptr2->pattern); } /* Hashtable for the load/store memory refs. */ static hash_table *pre_ldst_table; /* Bitmap containing one bit for each register in the program. Used when performing GCSE to track which registers have been set since the start of the basic block. */ static regset reg_set_bitmap; /* Array, indexed by basic block number for a list of insns which modify memory within that block. */ static vec *modify_mem_list; static bitmap modify_mem_list_set; /* This array parallels modify_mem_list, except that it stores MEMs being set and their canonicalized memory addresses. */ static vec *canon_modify_mem_list; /* Bitmap indexed by block numbers to record which blocks contain function calls. */ static bitmap blocks_with_calls; /* Various variables for statistics gathering. */ /* Memory used in a pass. This isn't intended to be absolutely precise. Its intent is only to keep an eye on memory usage. */ static int bytes_used; /* GCSE substitutions made. */ static int gcse_subst_count; /* Number of copy instructions created. */ static int gcse_create_count; /* Doing code hoisting. */ static bool doing_code_hoisting_p = false; /* For available exprs */ static sbitmap *ae_kill; /* Data stored for each basic block. */ struct bb_data { /* Maximal register pressure inside basic block for given register class (defined only for the pressure classes). */ int max_reg_pressure[N_REG_CLASSES]; /* Recorded register pressure of basic block before trying to hoist an expression. Will be used to restore the register pressure if the expression should not be hoisted. */ int old_pressure; /* Recorded register live_in info of basic block during code hoisting process. BACKUP is used to record live_in info before trying to hoist an expression, and will be used to restore LIVE_IN if the expression should not be hoisted. */ bitmap live_in, backup; }; #define BB_DATA(bb) ((struct bb_data *) (bb)->aux) static basic_block curr_bb; /* Current register pressure for each pressure class. */ static int curr_reg_pressure[N_REG_CLASSES]; static void compute_can_copy (void); static void *gmalloc (size_t) ATTRIBUTE_MALLOC; static void *gcalloc (size_t, size_t) ATTRIBUTE_MALLOC; static void *gcse_alloc (unsigned long); static void alloc_gcse_mem (void); static void free_gcse_mem (void); static void hash_scan_insn (rtx_insn *, struct gcse_hash_table_d *); static void hash_scan_set (rtx, rtx_insn *, struct gcse_hash_table_d *); static void hash_scan_clobber (rtx, rtx_insn *, struct gcse_hash_table_d *); static void hash_scan_call (rtx, rtx_insn *, struct gcse_hash_table_d *); static int want_to_gcse_p (rtx, int *); static int oprs_unchanged_p (const_rtx, const rtx_insn *, int); static int oprs_anticipatable_p (const_rtx, const rtx_insn *); static int oprs_available_p (const_rtx, const rtx_insn *); static void insert_expr_in_table (rtx, machine_mode, rtx_insn *, int, int, int, struct gcse_hash_table_d *); static unsigned int hash_expr (const_rtx, machine_mode, int *, int); static void record_last_reg_set_info (rtx_insn *, int); static void record_last_mem_set_info (rtx_insn *); static void record_last_set_info (rtx, const_rtx, void *); static void compute_hash_table (struct gcse_hash_table_d *); static void alloc_hash_table (struct gcse_hash_table_d *); static void free_hash_table (struct gcse_hash_table_d *); static void compute_hash_table_work (struct gcse_hash_table_d *); static void dump_hash_table (FILE *, const char *, struct gcse_hash_table_d *); static void compute_local_properties (sbitmap *, sbitmap *, sbitmap *, struct gcse_hash_table_d *); static void mems_conflict_for_gcse_p (rtx, const_rtx, void *); static int load_killed_in_block_p (const_basic_block, int, const_rtx, int); static void alloc_pre_mem (int, int); static void free_pre_mem (void); static struct edge_list *compute_pre_data (void); static int pre_expr_reaches_here_p (basic_block, struct gcse_expr *, basic_block); static void insert_insn_end_basic_block (struct gcse_expr *, basic_block); static void pre_insert_copy_insn (struct gcse_expr *, rtx_insn *); static void pre_insert_copies (void); static int pre_delete (void); static int pre_gcse (struct edge_list *); static int one_pre_gcse_pass (void); static void add_label_notes (rtx, rtx_insn *); static void alloc_code_hoist_mem (int, int); static void free_code_hoist_mem (void); static void compute_code_hoist_vbeinout (void); static void compute_code_hoist_data (void); static int should_hoist_expr_to_dom (basic_block, struct gcse_expr *, basic_block, sbitmap, int, int *, enum reg_class, int *, bitmap, rtx_insn *); static int hoist_code (void); static enum reg_class get_regno_pressure_class (int regno, int *nregs); static enum reg_class get_pressure_class_and_nregs (rtx_insn *insn, int *nregs); static int one_code_hoisting_pass (void); static rtx_insn *process_insert_insn (struct gcse_expr *); static int pre_edge_insert (struct edge_list *, struct gcse_expr **); static int pre_expr_reaches_here_p_work (basic_block, struct gcse_expr *, basic_block, char *); static struct ls_expr * ldst_entry (rtx); static void free_ldst_entry (struct ls_expr *); static void free_ld_motion_mems (void); static void print_ldst_list (FILE *); static struct ls_expr * find_rtx_in_ldst (rtx); static int simple_mem (const_rtx); static void invalidate_any_buried_refs (rtx); static void compute_ld_motion_mems (void); static void trim_ld_motion_mems (void); static void update_ld_motion_stores (struct gcse_expr *); static void clear_modify_mem_tables (void); static void free_modify_mem_tables (void); static bool is_too_expensive (const char *); #define GNEW(T) ((T *) gmalloc (sizeof (T))) #define GCNEW(T) ((T *) gcalloc (1, sizeof (T))) #define GNEWVEC(T, N) ((T *) gmalloc (sizeof (T) * (N))) #define GCNEWVEC(T, N) ((T *) gcalloc ((N), sizeof (T))) #define GNEWVAR(T, S) ((T *) gmalloc ((S))) #define GCNEWVAR(T, S) ((T *) gcalloc (1, (S))) #define GOBNEW(T) ((T *) gcse_alloc (sizeof (T))) #define GOBNEWVAR(T, S) ((T *) gcse_alloc ((S))) /* Misc. utilities. */ #define can_copy \ (this_target_gcse->x_can_copy) #define can_copy_init_p \ (this_target_gcse->x_can_copy_init_p) /* Compute which modes support reg/reg copy operations. */ static void compute_can_copy (void) { int i; #ifndef AVOID_CCMODE_COPIES rtx reg; rtx_insn *insn; #endif memset (can_copy, 0, NUM_MACHINE_MODES); start_sequence (); for (i = 0; i < NUM_MACHINE_MODES; i++) if (GET_MODE_CLASS (i) == MODE_CC) { #ifdef AVOID_CCMODE_COPIES can_copy[i] = 0; #else reg = gen_rtx_REG ((machine_mode) i, LAST_VIRTUAL_REGISTER + 1); insn = emit_insn (gen_rtx_SET (reg, reg)); if (recog (PATTERN (insn), insn, NULL) >= 0) can_copy[i] = 1; #endif } else can_copy[i] = 1; end_sequence (); } /* Returns whether the mode supports reg/reg copy operations. */ bool can_copy_p (machine_mode mode) { if (! can_copy_init_p) { compute_can_copy (); can_copy_init_p = true; } return can_copy[mode] != 0; } /* Cover function to xmalloc to record bytes allocated. */ static void * gmalloc (size_t size) { bytes_used += size; return xmalloc (size); } /* Cover function to xcalloc to record bytes allocated. */ static void * gcalloc (size_t nelem, size_t elsize) { bytes_used += nelem * elsize; return xcalloc (nelem, elsize); } /* Cover function to obstack_alloc. */ static void * gcse_alloc (unsigned long size) { bytes_used += size; return obstack_alloc (&gcse_obstack, size); } /* Allocate memory for the reg/memory set tracking tables. This is called at the start of each pass. */ static void alloc_gcse_mem (void) { /* Allocate vars to track sets of regs. */ reg_set_bitmap = ALLOC_REG_SET (NULL); /* Allocate array to keep a list of insns which modify memory in each basic block. The two typedefs are needed to work around the pre-processor limitation with template types in macro arguments. */ typedef vec vec_rtx_heap; typedef vec vec_modify_pair_heap; modify_mem_list = GCNEWVEC (vec_rtx_heap, last_basic_block_for_fn (cfun)); canon_modify_mem_list = GCNEWVEC (vec_modify_pair_heap, last_basic_block_for_fn (cfun)); modify_mem_list_set = BITMAP_ALLOC (NULL); blocks_with_calls = BITMAP_ALLOC (NULL); } /* Free memory allocated by alloc_gcse_mem. */ static void free_gcse_mem (void) { FREE_REG_SET (reg_set_bitmap); free_modify_mem_tables (); BITMAP_FREE (modify_mem_list_set); BITMAP_FREE (blocks_with_calls); } /* Compute the local properties of each recorded expression. Local properties are those that are defined by the block, irrespective of other blocks. An expression is transparent in a block if its operands are not modified in the block. An expression is computed (locally available) in a block if it is computed at least once and expression would contain the same value if the computation was moved to the end of the block. An expression is locally anticipatable in a block if it is computed at least once and expression would contain the same value if the computation was moved to the beginning of the block. We call this routine for pre and code hoisting. They all compute basically the same information and thus can easily share this code. TRANSP, COMP, and ANTLOC are destination sbitmaps for recording local properties. If NULL, then it is not necessary to compute or record that particular property. TABLE controls which hash table to look at. */ static void compute_local_properties (sbitmap *transp, sbitmap *comp, sbitmap *antloc, struct gcse_hash_table_d *table) { unsigned int i; /* Initialize any bitmaps that were passed in. */ if (transp) { bitmap_vector_ones (transp, last_basic_block_for_fn (cfun)); } if (comp) bitmap_vector_clear (comp, last_basic_block_for_fn (cfun)); if (antloc) bitmap_vector_clear (antloc, last_basic_block_for_fn (cfun)); for (i = 0; i < table->size; i++) { struct gcse_expr *expr; for (expr = table->table[i]; expr != NULL; expr = expr->next_same_hash) { int indx = expr->bitmap_index; struct gcse_occr *occr; /* The expression is transparent in this block if it is not killed. We start by assuming all are transparent [none are killed], and then reset the bits for those that are. */ if (transp) compute_transp (expr->expr, indx, transp, blocks_with_calls, modify_mem_list_set, canon_modify_mem_list); /* The occurrences recorded in antic_occr are exactly those that we want to set to nonzero in ANTLOC. */ if (antloc) for (occr = expr->antic_occr; occr != NULL; occr = occr->next) { bitmap_set_bit (antloc[BLOCK_FOR_INSN (occr->insn)->index], indx); /* While we're scanning the table, this is a good place to initialize this. */ occr->deleted_p = 0; } /* The occurrences recorded in avail_occr are exactly those that we want to set to nonzero in COMP. */ if (comp) for (occr = expr->avail_occr; occr != NULL; occr = occr->next) { bitmap_set_bit (comp[BLOCK_FOR_INSN (occr->insn)->index], indx); /* While we're scanning the table, this is a good place to initialize this. */ occr->copied_p = 0; } /* While we're scanning the table, this is a good place to initialize this. */ expr->reaching_reg = 0; } } } /* Hash table support. */ struct reg_avail_info { basic_block last_bb; int first_set; int last_set; }; static struct reg_avail_info *reg_avail_info; static basic_block current_bb; /* See whether X, the source of a set, is something we want to consider for GCSE. */ static int want_to_gcse_p (rtx x, int *max_distance_ptr) { #ifdef STACK_REGS /* On register stack architectures, don't GCSE constants from the constant pool, as the benefits are often swamped by the overhead of shuffling the register stack between basic blocks. */ if (IS_STACK_MODE (GET_MODE (x))) x = avoid_constant_pool_reference (x); #endif /* GCSE'ing constants: We do not specifically distinguish between constant and non-constant expressions in PRE and Hoist. We use set_src_cost below to limit the maximum distance simple expressions can travel. Nevertheless, constants are much easier to GCSE, and, hence, it is easy to overdo the optimizations. Usually, excessive PRE and Hoisting of constant leads to increased register pressure. RA can deal with this by rematerialing some of the constants. Therefore, it is important that the back-end generates sets of constants in a way that allows reload rematerialize them under high register pressure, i.e., a pseudo register with REG_EQUAL to constant is set only once. Failing to do so will result in IRA/reload spilling such constants under high register pressure instead of rematerializing them. */ switch (GET_CODE (x)) { case REG: case SUBREG: case CALL: return 0; CASE_CONST_ANY: if (!doing_code_hoisting_p) /* Do not PRE constants. */ return 0; /* FALLTHRU */ default: if (doing_code_hoisting_p) /* PRE doesn't implement max_distance restriction. */ { int cost; int max_distance; gcc_assert (!optimize_function_for_speed_p (cfun) && optimize_function_for_size_p (cfun)); cost = set_src_cost (x, 0); if (cost < COSTS_N_INSNS (GCSE_UNRESTRICTED_COST)) { max_distance = (GCSE_COST_DISTANCE_RATIO * cost) / 10; if (max_distance == 0) return 0; gcc_assert (max_distance > 0); } else max_distance = 0; if (max_distance_ptr) *max_distance_ptr = max_distance; } return can_assign_to_reg_without_clobbers_p (x); } } /* Used internally by can_assign_to_reg_without_clobbers_p. */ static GTY(()) rtx_insn *test_insn; /* Return true if we can assign X to a pseudo register such that the resulting insn does not result in clobbering a hard register as a side-effect. Additionally, if the target requires it, check that the resulting insn can be copied. If it cannot, this means that X is special and probably has hidden side-effects we don't want to mess with. This function is typically used by code motion passes, to verify that it is safe to insert an insn without worrying about clobbering maybe live hard regs. */ bool can_assign_to_reg_without_clobbers_p (rtx x) { int num_clobbers = 0; int icode; bool can_assign = false; /* If this is a valid operand, we are OK. If it's VOIDmode, we aren't. */ if (general_operand (x, GET_MODE (x))) return 1; else if (GET_MODE (x) == VOIDmode) return 0; /* Otherwise, check if we can make a valid insn from it. First initialize our test insn if we haven't already. */ if (test_insn == 0) { test_insn = make_insn_raw (gen_rtx_SET (gen_rtx_REG (word_mode, FIRST_PSEUDO_REGISTER * 2), const0_rtx)); SET_NEXT_INSN (test_insn) = SET_PREV_INSN (test_insn) = 0; INSN_LOCATION (test_insn) = UNKNOWN_LOCATION; } /* Now make an insn like the one we would make when GCSE'ing and see if valid. */ PUT_MODE (SET_DEST (PATTERN (test_insn)), GET_MODE (x)); SET_SRC (PATTERN (test_insn)) = x; icode = recog (PATTERN (test_insn), test_insn, &num_clobbers); /* If the test insn is valid and doesn't need clobbers, and the target also has no objections, we're good. */ if (icode >= 0 && (num_clobbers == 0 || !added_clobbers_hard_reg_p (icode)) && ! (targetm.cannot_copy_insn_p && targetm.cannot_copy_insn_p (test_insn))) can_assign = true; /* Make sure test_insn doesn't have any pointers into GC space. */ SET_SRC (PATTERN (test_insn)) = NULL_RTX; return can_assign; } /* Return nonzero if the operands of expression X are unchanged from the start of INSN's basic block up to but not including INSN (if AVAIL_P == 0), or from INSN to the end of INSN's basic block (if AVAIL_P != 0). */ static int oprs_unchanged_p (const_rtx x, const rtx_insn *insn, int avail_p) { int i, j; enum rtx_code code; const char *fmt; if (x == 0) return 1; code = GET_CODE (x); switch (code) { case REG: { struct reg_avail_info *info = ®_avail_info[REGNO (x)]; if (info->last_bb != current_bb) return 1; if (avail_p) return info->last_set < DF_INSN_LUID (insn); else return info->first_set >= DF_INSN_LUID (insn); } case MEM: if (! flag_gcse_lm || load_killed_in_block_p (current_bb, DF_INSN_LUID (insn), x, avail_p)) return 0; else return oprs_unchanged_p (XEXP (x, 0), insn, avail_p); case PRE_DEC: case PRE_INC: case POST_DEC: case POST_INC: case PRE_MODIFY: case POST_MODIFY: return 0; case PC: case CC0: /*FIXME*/ case CONST: CASE_CONST_ANY: case SYMBOL_REF: case LABEL_REF: case ADDR_VEC: case ADDR_DIFF_VEC: return 1; default: break; } for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--) { if (fmt[i] == 'e') { /* If we are about to do the last recursive call needed at this level, change it into iteration. This function is called enough to be worth it. */ if (i == 0) return oprs_unchanged_p (XEXP (x, i), insn, avail_p); else if (! oprs_unchanged_p (XEXP (x, i), insn, avail_p)) return 0; } else if (fmt[i] == 'E') for (j = 0; j < XVECLEN (x, i); j++) if (! oprs_unchanged_p (XVECEXP (x, i, j), insn, avail_p)) return 0; } return 1; } /* Info passed from load_killed_in_block_p to mems_conflict_for_gcse_p. */ struct mem_conflict_info { /* A memory reference for a load instruction, mems_conflict_for_gcse_p will see if a memory store conflicts with this memory load. */ const_rtx mem; /* True if mems_conflict_for_gcse_p finds a conflict between two memory references. */ bool conflict; }; /* DEST is the output of an instruction. If it is a memory reference and possibly conflicts with the load found in DATA, then communicate this information back through DATA. */ static void mems_conflict_for_gcse_p (rtx dest, const_rtx setter ATTRIBUTE_UNUSED, void *data) { struct mem_conflict_info *mci = (struct mem_conflict_info *) data; while (GET_CODE (dest) == SUBREG || GET_CODE (dest) == ZERO_EXTRACT || GET_CODE (dest) == STRICT_LOW_PART) dest = XEXP (dest, 0); /* If DEST is not a MEM, then it will not conflict with the load. Note that function calls are assumed to clobber memory, but are handled elsewhere. */ if (! MEM_P (dest)) return; /* If we are setting a MEM in our list of specially recognized MEMs, don't mark as killed this time. */ if (pre_ldst_mems != NULL && expr_equiv_p (dest, mci->mem)) { if (!find_rtx_in_ldst (dest)) mci->conflict = true; return; } if (true_dependence (dest, GET_MODE (dest), mci->mem)) mci->conflict = true; } /* Return nonzero if the expression in X (a memory reference) is killed in block BB before or after the insn with the LUID in UID_LIMIT. AVAIL_P is nonzero for kills after UID_LIMIT, and zero for kills before UID_LIMIT. To check the entire block, set UID_LIMIT to max_uid + 1 and AVAIL_P to 0. */ static int load_killed_in_block_p (const_basic_block bb, int uid_limit, const_rtx x, int avail_p) { vec list = modify_mem_list[bb->index]; rtx_insn *setter; unsigned ix; /* If this is a readonly then we aren't going to be changing it. */ if (MEM_READONLY_P (x)) return 0; FOR_EACH_VEC_ELT_REVERSE (list, ix, setter) { struct mem_conflict_info mci; /* Ignore entries in the list that do not apply. */ if ((avail_p && DF_INSN_LUID (setter) < uid_limit) || (! avail_p && DF_INSN_LUID (setter) > uid_limit)) continue; /* If SETTER is a call everything is clobbered. Note that calls to pure functions are never put on the list, so we need not worry about them. */ if (CALL_P (setter)) return 1; /* SETTER must be an INSN of some kind that sets memory. Call note_stores to examine each hunk of memory that is modified. */ mci.mem = x; mci.conflict = false; note_stores (PATTERN (setter), mems_conflict_for_gcse_p, &mci); if (mci.conflict) return 1; } return 0; } /* Return nonzero if the operands of expression X are unchanged from the start of INSN's basic block up to but not including INSN. */ static int oprs_anticipatable_p (const_rtx x, const rtx_insn *insn) { return oprs_unchanged_p (x, insn, 0); } /* Return nonzero if the operands of expression X are unchanged from INSN to the end of INSN's basic block. */ static int oprs_available_p (const_rtx x, const rtx_insn *insn) { return oprs_unchanged_p (x, insn, 1); } /* Hash expression X. MODE is only used if X is a CONST_INT. DO_NOT_RECORD_P is a boolean indicating if a volatile operand is found or if the expression contains something we don't want to insert in the table. HASH_TABLE_SIZE is the current size of the hash table to be probed. */ static unsigned int hash_expr (const_rtx x, machine_mode mode, int *do_not_record_p, int hash_table_size) { unsigned int hash; *do_not_record_p = 0; hash = hash_rtx (x, mode, do_not_record_p, NULL, /*have_reg_qty=*/false); return hash % hash_table_size; } /* Return nonzero if exp1 is equivalent to exp2. */ static int expr_equiv_p (const_rtx x, const_rtx y) { return exp_equiv_p (x, y, 0, true); } /* Insert expression X in INSN in the hash TABLE. If it is already present, record it as the last occurrence in INSN's basic block. MODE is the mode of the value X is being stored into. It is only used if X is a CONST_INT. ANTIC_P is nonzero if X is an anticipatable expression. AVAIL_P is nonzero if X is an available expression. MAX_DISTANCE is the maximum distance in instructions this expression can be moved. */ static void insert_expr_in_table (rtx x, machine_mode mode, rtx_insn *insn, int antic_p, int avail_p, int max_distance, struct gcse_hash_table_d *table) { int found, do_not_record_p; unsigned int hash; struct gcse_expr *cur_expr, *last_expr = NULL; struct gcse_occr *antic_occr, *avail_occr; hash = hash_expr (x, mode, &do_not_record_p, table->size); /* Do not insert expression in table if it contains volatile operands, or if hash_expr determines the expression is something we don't want to or can't handle. */ if (do_not_record_p) return; cur_expr = table->table[hash]; found = 0; while (cur_expr && 0 == (found = expr_equiv_p (cur_expr->expr, x))) { /* If the expression isn't found, save a pointer to the end of the list. */ last_expr = cur_expr; cur_expr = cur_expr->next_same_hash; } if (! found) { cur_expr = GOBNEW (struct gcse_expr); bytes_used += sizeof (struct gcse_expr); if (table->table[hash] == NULL) /* This is the first pattern that hashed to this index. */ table->table[hash] = cur_expr; else /* Add EXPR to end of this hash chain. */ last_expr->next_same_hash = cur_expr; /* Set the fields of the expr element. */ cur_expr->expr = x; cur_expr->bitmap_index = table->n_elems++; cur_expr->next_same_hash = NULL; cur_expr->antic_occr = NULL; cur_expr->avail_occr = NULL; gcc_assert (max_distance >= 0); cur_expr->max_distance = max_distance; } else gcc_assert (cur_expr->max_distance == max_distance); /* Now record the occurrence(s). */ if (antic_p) { antic_occr = cur_expr->antic_occr; if (antic_occr && BLOCK_FOR_INSN (antic_occr->insn) != BLOCK_FOR_INSN (insn)) antic_occr = NULL; if (antic_occr) /* Found another instance of the expression in the same basic block. Prefer the currently recorded one. We want the first one in the block and the block is scanned from start to end. */ ; /* nothing to do */ else { /* First occurrence of this expression in this basic block. */ antic_occr = GOBNEW (struct gcse_occr); bytes_used += sizeof (struct gcse_occr); antic_occr->insn = insn; antic_occr->next = cur_expr->antic_occr; antic_occr->deleted_p = 0; cur_expr->antic_occr = antic_occr; } } if (avail_p) { avail_occr = cur_expr->avail_occr; if (avail_occr && BLOCK_FOR_INSN (avail_occr->insn) == BLOCK_FOR_INSN (insn)) { /* Found another instance of the expression in the same basic block. Prefer this occurrence to the currently recorded one. We want the last one in the block and the block is scanned from start to end. */ avail_occr->insn = insn; } else { /* First occurrence of this expression in this basic block. */ avail_occr = GOBNEW (struct gcse_occr); bytes_used += sizeof (struct gcse_occr); avail_occr->insn = insn; avail_occr->next = cur_expr->avail_occr; avail_occr->deleted_p = 0; cur_expr->avail_occr = avail_occr; } } } /* Scan SET present in INSN and add an entry to the hash TABLE. */ static void hash_scan_set (rtx set, rtx_insn *insn, struct gcse_hash_table_d *table) { rtx src = SET_SRC (set); rtx dest = SET_DEST (set); rtx note; if (GET_CODE (src) == CALL) hash_scan_call (src, insn, table); else if (REG_P (dest)) { unsigned int regno = REGNO (dest); int max_distance = 0; /* See if a REG_EQUAL note shows this equivalent to a simpler expression. This allows us to do a single GCSE pass and still eliminate redundant constants, addresses or other expressions that are constructed with multiple instructions. However, keep the original SRC if INSN is a simple reg-reg move. In this case, there will almost always be a REG_EQUAL note on the insn that sets SRC. By recording the REG_EQUAL value here as SRC for INSN, we miss copy propagation opportunities and we perform the same PRE GCSE operation repeatedly on the same REG_EQUAL value if we do more than one PRE GCSE pass. Note that this does not impede profitable constant propagations. We "look through" reg-reg sets in lookup_avail_set. */ note = find_reg_equal_equiv_note (insn); if (note != 0 && REG_NOTE_KIND (note) == REG_EQUAL && !REG_P (src) && want_to_gcse_p (XEXP (note, 0), NULL)) src = XEXP (note, 0), set = gen_rtx_SET (dest, src); /* Only record sets of pseudo-regs in the hash table. */ if (regno >= FIRST_PSEUDO_REGISTER /* Don't GCSE something if we can't do a reg/reg copy. */ && can_copy_p (GET_MODE (dest)) /* GCSE commonly inserts instruction after the insn. We can't do that easily for EH edges so disable GCSE on these for now. */ /* ??? We can now easily create new EH landing pads at the gimple level, for splitting edges; there's no reason we can't do the same thing at the rtl level. */ && !can_throw_internal (insn) /* Is SET_SRC something we want to gcse? */ && want_to_gcse_p (src, &max_distance) /* Don't CSE a nop. */ && ! set_noop_p (set) /* Don't GCSE if it has attached REG_EQUIV note. At this point this only function parameters should have REG_EQUIV notes and if the argument slot is used somewhere explicitly, it means address of parameter has been taken, so we should not extend the lifetime of the pseudo. */ && (note == NULL_RTX || ! MEM_P (XEXP (note, 0)))) { /* An expression is not anticipatable if its operands are modified before this insn or if this is not the only SET in this insn. The latter condition does not have to mean that SRC itself is not anticipatable, but we just will not be able to handle code motion of insns with multiple sets. */ int antic_p = oprs_anticipatable_p (src, insn) && !multiple_sets (insn); /* An expression is not available if its operands are subsequently modified, including this insn. It's also not available if this is a branch, because we can't insert a set after the branch. */ int avail_p = (oprs_available_p (src, insn) && ! JUMP_P (insn)); insert_expr_in_table (src, GET_MODE (dest), insn, antic_p, avail_p, max_distance, table); } } /* In case of store we want to consider the memory value as available in the REG stored in that memory. This makes it possible to remove redundant loads from due to stores to the same location. */ else if (flag_gcse_las && REG_P (src) && MEM_P (dest)) { unsigned int regno = REGNO (src); int max_distance = 0; /* Only record sets of pseudo-regs in the hash table. */ if (regno >= FIRST_PSEUDO_REGISTER /* Don't GCSE something if we can't do a reg/reg copy. */ && can_copy_p (GET_MODE (src)) /* GCSE commonly inserts instruction after the insn. We can't do that easily for EH edges so disable GCSE on these for now. */ && !can_throw_internal (insn) /* Is SET_DEST something we want to gcse? */ && want_to_gcse_p (dest, &max_distance) /* Don't CSE a nop. */ && ! set_noop_p (set) /* Don't GCSE if it has attached REG_EQUIV note. At this point this only function parameters should have REG_EQUIV notes and if the argument slot is used somewhere explicitly, it means address of parameter has been taken, so we should not extend the lifetime of the pseudo. */ && ((note = find_reg_note (insn, REG_EQUIV, NULL_RTX)) == 0 || ! MEM_P (XEXP (note, 0)))) { /* Stores are never anticipatable. */ int antic_p = 0; /* An expression is not available if its operands are subsequently modified, including this insn. It's also not available if this is a branch, because we can't insert a set after the branch. */ int avail_p = oprs_available_p (dest, insn) && ! JUMP_P (insn); /* Record the memory expression (DEST) in the hash table. */ insert_expr_in_table (dest, GET_MODE (dest), insn, antic_p, avail_p, max_distance, table); } } } static void hash_scan_clobber (rtx x ATTRIBUTE_UNUSED, rtx_insn *insn ATTRIBUTE_UNUSED, struct gcse_hash_table_d *table ATTRIBUTE_UNUSED) { /* Currently nothing to do. */ } static void hash_scan_call (rtx x ATTRIBUTE_UNUSED, rtx_insn *insn ATTRIBUTE_UNUSED, struct gcse_hash_table_d *table ATTRIBUTE_UNUSED) { /* Currently nothing to do. */ } /* Process INSN and add hash table entries as appropriate. */ static void hash_scan_insn (rtx_insn *insn, struct gcse_hash_table_d *table) { rtx pat = PATTERN (insn); int i; /* Pick out the sets of INSN and for other forms of instructions record what's been modified. */ if (GET_CODE (pat) == SET) hash_scan_set (pat, insn, table); else if (GET_CODE (pat) == CLOBBER) hash_scan_clobber (pat, insn, table); else if (GET_CODE (pat) == CALL) hash_scan_call (pat, insn, table); else if (GET_CODE (pat) == PARALLEL) for (i = 0; i < XVECLEN (pat, 0); i++) { rtx x = XVECEXP (pat, 0, i); if (GET_CODE (x) == SET) hash_scan_set (x, insn, table); else if (GET_CODE (x) == CLOBBER) hash_scan_clobber (x, insn, table); else if (GET_CODE (x) == CALL) hash_scan_call (x, insn, table); } } /* Dump the hash table TABLE to file FILE under the name NAME. */ static void dump_hash_table (FILE *file, const char *name, struct gcse_hash_table_d *table) { int i; /* Flattened out table, so it's printed in proper order. */ struct gcse_expr **flat_table; unsigned int *hash_val; struct gcse_expr *expr; flat_table = XCNEWVEC (struct gcse_expr *, table->n_elems); hash_val = XNEWVEC (unsigned int, table->n_elems); for (i = 0; i < (int) table->size; i++) for (expr = table->table[i]; expr != NULL; expr = expr->next_same_hash) { flat_table[expr->bitmap_index] = expr; hash_val[expr->bitmap_index] = i; } fprintf (file, "%s hash table (%d buckets, %d entries)\n", name, table->size, table->n_elems); for (i = 0; i < (int) table->n_elems; i++) if (flat_table[i] != 0) { expr = flat_table[i]; fprintf (file, "Index %d (hash value %d; max distance %d)\n ", expr->bitmap_index, hash_val[i], expr->max_distance); print_rtl (file, expr->expr); fprintf (file, "\n"); } fprintf (file, "\n"); free (flat_table); free (hash_val); } /* Record register first/last/block set information for REGNO in INSN. first_set records the first place in the block where the register is set and is used to compute "anticipatability". last_set records the last place in the block where the register is set and is used to compute "availability". last_bb records the block for which first_set and last_set are valid, as a quick test to invalidate them. */ static void record_last_reg_set_info (rtx_insn *insn, int regno) { struct reg_avail_info *info = ®_avail_info[regno]; int luid = DF_INSN_LUID (insn); info->last_set = luid; if (info->last_bb != current_bb) { info->last_bb = current_bb; info->first_set = luid; } } /* Record memory modification information for INSN. We do not actually care about the memory location(s) that are set, or even how they are set (consider a CALL_INSN). We merely need to record which insns modify memory. */ static void record_last_mem_set_info (rtx_insn *insn) { if (! flag_gcse_lm) return; record_last_mem_set_info_common (insn, modify_mem_list, canon_modify_mem_list, modify_mem_list_set, blocks_with_calls); } /* Called from compute_hash_table via note_stores to handle one SET or CLOBBER in an insn. DATA is really the instruction in which the SET is taking place. */ static void record_last_set_info (rtx dest, const_rtx setter ATTRIBUTE_UNUSED, void *data) { rtx_insn *last_set_insn = (rtx_insn *) data; if (GET_CODE (dest) == SUBREG) dest = SUBREG_REG (dest); if (REG_P (dest)) record_last_reg_set_info (last_set_insn, REGNO (dest)); else if (MEM_P (dest) /* Ignore pushes, they clobber nothing. */ && ! push_operand (dest, GET_MODE (dest))) record_last_mem_set_info (last_set_insn); } /* Top level function to create an expression hash table. Expression entries are placed in the hash table if - they are of the form (set (pseudo-reg) src), - src is something we want to perform GCSE on, - none of the operands are subsequently modified in the block Currently src must be a pseudo-reg or a const_int. TABLE is the table computed. */ static void compute_hash_table_work (struct gcse_hash_table_d *table) { int i; /* re-Cache any INSN_LIST nodes we have allocated. */ clear_modify_mem_tables (); /* Some working arrays used to track first and last set in each block. */ reg_avail_info = GNEWVEC (struct reg_avail_info, max_reg_num ()); for (i = 0; i < max_reg_num (); ++i) reg_avail_info[i].last_bb = NULL; FOR_EACH_BB_FN (current_bb, cfun) { rtx_insn *insn; unsigned int regno; /* First pass over the instructions records information used to determine when registers and memory are first and last set. */ FOR_BB_INSNS (current_bb, insn) { if (!NONDEBUG_INSN_P (insn)) continue; if (CALL_P (insn)) { hard_reg_set_iterator hrsi; EXECUTE_IF_SET_IN_HARD_REG_SET (regs_invalidated_by_call, 0, regno, hrsi) record_last_reg_set_info (insn, regno); if (! RTL_CONST_OR_PURE_CALL_P (insn)) record_last_mem_set_info (insn); } note_stores (PATTERN (insn), record_last_set_info, insn); } /* The next pass builds the hash table. */ FOR_BB_INSNS (current_bb, insn) if (NONDEBUG_INSN_P (insn)) hash_scan_insn (insn, table); } free (reg_avail_info); reg_avail_info = NULL; } /* Allocate space for the set/expr hash TABLE. It is used to determine the number of buckets to use. */ static void alloc_hash_table (struct gcse_hash_table_d *table) { int n; n = get_max_insn_count (); table->size = n / 4; if (table->size < 11) table->size = 11; /* Attempt to maintain efficient use of hash table. Making it an odd number is simplest for now. ??? Later take some measurements. */ table->size |= 1; n = table->size * sizeof (struct gcse_expr *); table->table = GNEWVAR (struct gcse_expr *, n); } /* Free things allocated by alloc_hash_table. */ static void free_hash_table (struct gcse_hash_table_d *table) { free (table->table); } /* Compute the expression hash table TABLE. */ static void compute_hash_table (struct gcse_hash_table_d *table) { /* Initialize count of number of entries in hash table. */ table->n_elems = 0; memset (table->table, 0, table->size * sizeof (struct gcse_expr *)); compute_hash_table_work (table); } /* Expression tracking support. */ /* Clear canon_modify_mem_list and modify_mem_list tables. */ static void clear_modify_mem_tables (void) { unsigned i; bitmap_iterator bi; EXECUTE_IF_SET_IN_BITMAP (modify_mem_list_set, 0, i, bi) { modify_mem_list[i].release (); canon_modify_mem_list[i].release (); } bitmap_clear (modify_mem_list_set); bitmap_clear (blocks_with_calls); } /* Release memory used by modify_mem_list_set. */ static void free_modify_mem_tables (void) { clear_modify_mem_tables (); free (modify_mem_list); free (canon_modify_mem_list); modify_mem_list = 0; canon_modify_mem_list = 0; } /* Compute PRE+LCM working variables. */ /* Local properties of expressions. */ /* Nonzero for expressions that are transparent in the block. */ static sbitmap *transp; /* Nonzero for expressions that are computed (available) in the block. */ static sbitmap *comp; /* Nonzero for expressions that are locally anticipatable in the block. */ static sbitmap *antloc; /* Nonzero for expressions where this block is an optimal computation point. */ static sbitmap *pre_optimal; /* Nonzero for expressions which are redundant in a particular block. */ static sbitmap *pre_redundant; /* Nonzero for expressions which should be inserted on a specific edge. */ static sbitmap *pre_insert_map; /* Nonzero for expressions which should be deleted in a specific block. */ static sbitmap *pre_delete_map; /* Allocate vars used for PRE analysis. */ static void alloc_pre_mem (int n_blocks, int n_exprs) { transp = sbitmap_vector_alloc (n_blocks, n_exprs); comp = sbitmap_vector_alloc (n_blocks, n_exprs); antloc = sbitmap_vector_alloc (n_blocks, n_exprs); pre_optimal = NULL; pre_redundant = NULL; pre_insert_map = NULL; pre_delete_map = NULL; ae_kill = sbitmap_vector_alloc (n_blocks, n_exprs); /* pre_insert and pre_delete are allocated later. */ } /* Free vars used for PRE analysis. */ static void free_pre_mem (void) { sbitmap_vector_free (transp); sbitmap_vector_free (comp); /* ANTLOC and AE_KILL are freed just after pre_lcm finishes. */ if (pre_optimal) sbitmap_vector_free (pre_optimal); if (pre_redundant) sbitmap_vector_free (pre_redundant); if (pre_insert_map) sbitmap_vector_free (pre_insert_map); if (pre_delete_map) sbitmap_vector_free (pre_delete_map); transp = comp = NULL; pre_optimal = pre_redundant = pre_insert_map = pre_delete_map = NULL; } /* Remove certain expressions from anticipatable and transparent sets of basic blocks that have incoming abnormal edge. For PRE remove potentially trapping expressions to avoid placing them on abnormal edges. For hoisting remove memory references that can be clobbered by calls. */ static void prune_expressions (bool pre_p) { sbitmap prune_exprs; struct gcse_expr *expr; unsigned int ui; basic_block bb; prune_exprs = sbitmap_alloc (expr_hash_table.n_elems); bitmap_clear (prune_exprs); for (ui = 0; ui < expr_hash_table.size; ui++) { for (expr = expr_hash_table.table[ui]; expr; expr = expr->next_same_hash) { /* Note potentially trapping expressions. */ if (may_trap_p (expr->expr)) { bitmap_set_bit (prune_exprs, expr->bitmap_index); continue; } if (!pre_p && MEM_P (expr->expr)) /* Note memory references that can be clobbered by a call. We do not split abnormal edges in hoisting, so would a memory reference get hoisted along an abnormal edge, it would be placed /before/ the call. Therefore, only constant memory references can be hoisted along abnormal edges. */ { if (GET_CODE (XEXP (expr->expr, 0)) == SYMBOL_REF && CONSTANT_POOL_ADDRESS_P (XEXP (expr->expr, 0))) continue; if (MEM_READONLY_P (expr->expr) && !MEM_VOLATILE_P (expr->expr) && MEM_NOTRAP_P (expr->expr)) /* Constant memory reference, e.g., a PIC address. */ continue; /* ??? Optimally, we would use interprocedural alias analysis to determine if this mem is actually killed by this call. */ bitmap_set_bit (prune_exprs, expr->bitmap_index); } } } FOR_EACH_BB_FN (bb, cfun) { edge e; edge_iterator ei; /* If the current block is the destination of an abnormal edge, we kill all trapping (for PRE) and memory (for hoist) expressions because we won't be able to properly place the instruction on the edge. So make them neither anticipatable nor transparent. This is fairly conservative. ??? For hoisting it may be necessary to check for set-and-jump instructions here, not just for abnormal edges. The general problem is that when an expression cannot not be placed right at the end of a basic block we should account for any side-effects of a subsequent jump instructions that could clobber the expression. It would be best to implement this check along the lines of should_hoist_expr_to_dom where the target block is already known and, hence, there's no need to conservatively prune expressions on "intermediate" set-and-jump instructions. */ FOR_EACH_EDGE (e, ei, bb->preds) if ((e->flags & EDGE_ABNORMAL) && (pre_p || CALL_P (BB_END (e->src)))) { bitmap_and_compl (antloc[bb->index], antloc[bb->index], prune_exprs); bitmap_and_compl (transp[bb->index], transp[bb->index], prune_exprs); break; } } sbitmap_free (prune_exprs); } /* It may be necessary to insert a large number of insns on edges to make the existing occurrences of expressions fully redundant. This routine examines the set of insertions and deletions and if the ratio of insertions to deletions is too high for a particular expression, then the expression is removed from the insertion/deletion sets. N_ELEMS is the number of elements in the hash table. */ static void prune_insertions_deletions (int n_elems) { sbitmap_iterator sbi; sbitmap prune_exprs; /* We always use I to iterate over blocks/edges and J to iterate over expressions. */ unsigned int i, j; /* Counts for the number of times an expression needs to be inserted and number of times an expression can be removed as a result. */ int *insertions = GCNEWVEC (int, n_elems); int *deletions = GCNEWVEC (int, n_elems); /* Set of expressions which require too many insertions relative to the number of deletions achieved. We will prune these out of the insertion/deletion sets. */ prune_exprs = sbitmap_alloc (n_elems); bitmap_clear (prune_exprs); /* Iterate over the edges counting the number of times each expression needs to be inserted. */ for (i = 0; i < (unsigned) n_edges_for_fn (cfun); i++) { EXECUTE_IF_SET_IN_BITMAP (pre_insert_map[i], 0, j, sbi) insertions[j]++; } /* Similarly for deletions, but those occur in blocks rather than on edges. */ for (i = 0; i < (unsigned) last_basic_block_for_fn (cfun); i++) { EXECUTE_IF_SET_IN_BITMAP (pre_delete_map[i], 0, j, sbi) deletions[j]++; } /* Now that we have accurate counts, iterate over the elements in the hash table and see if any need too many insertions relative to the number of evaluations that can be removed. If so, mark them in PRUNE_EXPRS. */ for (j = 0; j < (unsigned) n_elems; j++) if (deletions[j] && ((unsigned) insertions[j] / deletions[j]) > MAX_GCSE_INSERTION_RATIO) bitmap_set_bit (prune_exprs, j); /* Now prune PRE_INSERT_MAP and PRE_DELETE_MAP based on PRUNE_EXPRS. */ EXECUTE_IF_SET_IN_BITMAP (prune_exprs, 0, j, sbi) { for (i = 0; i < (unsigned) n_edges_for_fn (cfun); i++) bitmap_clear_bit (pre_insert_map[i], j); for (i = 0; i < (unsigned) last_basic_block_for_fn (cfun); i++) bitmap_clear_bit (pre_delete_map[i], j); } sbitmap_free (prune_exprs); free (insertions); free (deletions); } /* Top level routine to do the dataflow analysis needed by PRE. */ static struct edge_list * compute_pre_data (void) { struct edge_list *edge_list; basic_block bb; compute_local_properties (transp, comp, antloc, &expr_hash_table); prune_expressions (true); bitmap_vector_clear (ae_kill, last_basic_block_for_fn (cfun)); /* Compute ae_kill for each basic block using: ~(TRANSP | COMP) */ FOR_EACH_BB_FN (bb, cfun) { bitmap_ior (ae_kill[bb->index], transp[bb->index], comp[bb->index]); bitmap_not (ae_kill[bb->index], ae_kill[bb->index]); } edge_list = pre_edge_lcm (expr_hash_table.n_elems, transp, comp, antloc, ae_kill, &pre_insert_map, &pre_delete_map); sbitmap_vector_free (antloc); antloc = NULL; sbitmap_vector_free (ae_kill); ae_kill = NULL; prune_insertions_deletions (expr_hash_table.n_elems); return edge_list; } /* PRE utilities */ /* Return nonzero if an occurrence of expression EXPR in OCCR_BB would reach block BB. VISITED is a pointer to a working buffer for tracking which BB's have been visited. It is NULL for the top-level call. We treat reaching expressions that go through blocks containing the same reaching expression as "not reaching". E.g. if EXPR is generated in blocks 2 and 3, INSN is in block 4, and 2->3->4, we treat the expression in block 2 as not reaching. The intent is to improve the probability of finding only one reaching expression and to reduce register lifetimes by picking the closest such expression. */ static int pre_expr_reaches_here_p_work (basic_block occr_bb, struct gcse_expr *expr, basic_block bb, char *visited) { edge pred; edge_iterator ei; FOR_EACH_EDGE (pred, ei, bb->preds) { basic_block pred_bb = pred->src; if (pred->src == ENTRY_BLOCK_PTR_FOR_FN (cfun) /* Has predecessor has already been visited? */ || visited[pred_bb->index]) ;/* Nothing to do. */ /* Does this predecessor generate this expression? */ else if (bitmap_bit_p (comp[pred_bb->index], expr->bitmap_index)) { /* Is this the occurrence we're looking for? Note that there's only one generating occurrence per block so we just need to check the block number. */ if (occr_bb == pred_bb) return 1; visited[pred_bb->index] = 1; } /* Ignore this predecessor if it kills the expression. */ else if (! bitmap_bit_p (transp[pred_bb->index], expr->bitmap_index)) visited[pred_bb->index] = 1; /* Neither gen nor kill. */ else { visited[pred_bb->index] = 1; if (pre_expr_reaches_here_p_work (occr_bb, expr, pred_bb, visited)) return 1; } } /* All paths have been checked. */ return 0; } /* The wrapper for pre_expr_reaches_here_work that ensures that any memory allocated for that function is returned. */ static int pre_expr_reaches_here_p (basic_block occr_bb, struct gcse_expr *expr, basic_block bb) { int rval; char *visited = XCNEWVEC (char, last_basic_block_for_fn (cfun)); rval = pre_expr_reaches_here_p_work (occr_bb, expr, bb, visited); free (visited); return rval; } /* Generate RTL to copy an EXPR to its `reaching_reg' and return it. */ static rtx_insn * process_insert_insn (struct gcse_expr *expr) { rtx reg = expr->reaching_reg; /* Copy the expression to make sure we don't have any sharing issues. */ rtx exp = copy_rtx (expr->expr); rtx_insn *pat; start_sequence (); /* If the expression is something that's an operand, like a constant, just copy it to a register. */ if (general_operand (exp, GET_MODE (reg))) emit_move_insn (reg, exp); /* Otherwise, make a new insn to compute this expression and make sure the insn will be recognized (this also adds any needed CLOBBERs). */ else { rtx_insn *insn = emit_insn (gen_rtx_SET (reg, exp)); if (insn_invalid_p (insn, false)) gcc_unreachable (); } pat = get_insns (); end_sequence (); return pat; } /* Add EXPR to the end of basic block BB. This is used by both the PRE and code hoisting. */ static void insert_insn_end_basic_block (struct gcse_expr *expr, basic_block bb) { rtx_insn *insn = BB_END (bb); rtx_insn *new_insn; rtx reg = expr->reaching_reg; int regno = REGNO (reg); rtx_insn *pat, *pat_end; pat = process_insert_insn (expr); gcc_assert (pat && INSN_P (pat)); pat_end = pat; while (NEXT_INSN (pat_end) != NULL_RTX) pat_end = NEXT_INSN (pat_end); /* If the last insn is a jump, insert EXPR in front [taking care to handle cc0, etc. properly]. Similarly we need to care trapping instructions in presence of non-call exceptions. */ if (JUMP_P (insn) || (NONJUMP_INSN_P (insn) && (!single_succ_p (bb) || single_succ_edge (bb)->flags & EDGE_ABNORMAL))) { /* FIXME: 'twould be nice to call prev_cc0_setter here but it aborts if cc0 isn't set. */ if (HAVE_cc0) { rtx note = find_reg_note (insn, REG_CC_SETTER, NULL_RTX); if (note) insn = safe_as_a (XEXP (note, 0)); else { rtx_insn *maybe_cc0_setter = prev_nonnote_insn (insn); if (maybe_cc0_setter && INSN_P (maybe_cc0_setter) && sets_cc0_p (PATTERN (maybe_cc0_setter))) insn = maybe_cc0_setter; } } /* FIXME: What if something in cc0/jump uses value set in new insn? */ new_insn = emit_insn_before_noloc (pat, insn, bb); } /* Likewise if the last insn is a call, as will happen in the presence of exception handling. */ else if (CALL_P (insn) && (!single_succ_p (bb) || single_succ_edge (bb)->flags & EDGE_ABNORMAL)) { /* Keeping in mind targets with small register classes and parameters in registers, we search backward and place the instructions before the first parameter is loaded. Do this for everyone for consistency and a presumption that we'll get better code elsewhere as well. */ /* Since different machines initialize their parameter registers in different orders, assume nothing. Collect the set of all parameter registers. */ insn = find_first_parameter_load (insn, BB_HEAD (bb)); /* If we found all the parameter loads, then we want to insert before the first parameter load. If we did not find all the parameter loads, then we might have stopped on the head of the block, which could be a CODE_LABEL. If we inserted before the CODE_LABEL, then we would be putting the insn in the wrong basic block. In that case, put the insn after the CODE_LABEL. Also, respect NOTE_INSN_BASIC_BLOCK. */ while (LABEL_P (insn) || NOTE_INSN_BASIC_BLOCK_P (insn)) insn = NEXT_INSN (insn); new_insn = emit_insn_before_noloc (pat, insn, bb); } else new_insn = emit_insn_after_noloc (pat, insn, bb); while (1) { if (INSN_P (pat)) add_label_notes (PATTERN (pat), new_insn); if (pat == pat_end) break; pat = NEXT_INSN (pat); } gcse_create_count++; if (dump_file) { fprintf (dump_file, "PRE/HOIST: end of bb %d, insn %d, ", bb->index, INSN_UID (new_insn)); fprintf (dump_file, "copying expression %d to reg %d\n", expr->bitmap_index, regno); } } /* Insert partially redundant expressions on edges in the CFG to make the expressions fully redundant. */ static int pre_edge_insert (struct edge_list *edge_list, struct gcse_expr **index_map) { int e, i, j, num_edges, set_size, did_insert = 0; sbitmap *inserted; /* Where PRE_INSERT_MAP is nonzero, we add the expression on that edge if it reaches any of the deleted expressions. */ set_size = pre_insert_map[0]->size; num_edges = NUM_EDGES (edge_list); inserted = sbitmap_vector_alloc (num_edges, expr_hash_table.n_elems); bitmap_vector_clear (inserted, num_edges); for (e = 0; e < num_edges; e++) { int indx; basic_block bb = INDEX_EDGE_PRED_BB (edge_list, e); for (i = indx = 0; i < set_size; i++, indx += SBITMAP_ELT_BITS) { SBITMAP_ELT_TYPE insert = pre_insert_map[e]->elms[i]; for (j = indx; insert && j < (int) expr_hash_table.n_elems; j++, insert >>= 1) if ((insert & 1) != 0 && index_map[j]->reaching_reg != NULL_RTX) { struct gcse_expr *expr = index_map[j]; struct gcse_occr *occr; /* Now look at each deleted occurrence of this expression. */ for (occr = expr->antic_occr; occr != NULL; occr = occr->next) { if (! occr->deleted_p) continue; /* Insert this expression on this edge if it would reach the deleted occurrence in BB. */ if (!bitmap_bit_p (inserted[e], j)) { rtx_insn *insn; edge eg = INDEX_EDGE (edge_list, e); /* We can't insert anything on an abnormal and critical edge, so we insert the insn at the end of the previous block. There are several alternatives detailed in Morgans book P277 (sec 10.5) for handling this situation. This one is easiest for now. */ if (eg->flags & EDGE_ABNORMAL) insert_insn_end_basic_block (index_map[j], bb); else { insn = process_insert_insn (index_map[j]); insert_insn_on_edge (insn, eg); } if (dump_file) { fprintf (dump_file, "PRE: edge (%d,%d), ", bb->index, INDEX_EDGE_SUCC_BB (edge_list, e)->index); fprintf (dump_file, "copy expression %d\n", expr->bitmap_index); } update_ld_motion_stores (expr); bitmap_set_bit (inserted[e], j); did_insert = 1; gcse_create_count++; } } } } } sbitmap_vector_free (inserted); return did_insert; } /* Copy the result of EXPR->EXPR generated by INSN to EXPR->REACHING_REG. Given "old_reg <- expr" (INSN), instead of adding after it reaching_reg <- old_reg it's better to do the following: reaching_reg <- expr old_reg <- reaching_reg because this way copy propagation can discover additional PRE opportunities. But if this fails, we try the old way. When "expr" is a store, i.e. given "MEM <- old_reg", instead of adding after it reaching_reg <- old_reg it's better to add it before as follows: reaching_reg <- old_reg MEM <- reaching_reg. */ static void pre_insert_copy_insn (struct gcse_expr *expr, rtx_insn *insn) { rtx reg = expr->reaching_reg; int regno = REGNO (reg); int indx = expr->bitmap_index; rtx pat = PATTERN (insn); rtx set, first_set; rtx_insn *new_insn; rtx old_reg; int i; /* This block matches the logic in hash_scan_insn. */ switch (GET_CODE (pat)) { case SET: set = pat; break; case PARALLEL: /* Search through the parallel looking for the set whose source was the expression that we're interested in. */ first_set = NULL_RTX; set = NULL_RTX; for (i = 0; i < XVECLEN (pat, 0); i++) { rtx x = XVECEXP (pat, 0, i); if (GET_CODE (x) == SET) { /* If the source was a REG_EQUAL or REG_EQUIV note, we may not find an equivalent expression, but in this case the PARALLEL will have a single set. */ if (first_set == NULL_RTX) first_set = x; if (expr_equiv_p (SET_SRC (x), expr->expr)) { set = x; break; } } } gcc_assert (first_set); if (set == NULL_RTX) set = first_set; break; default: gcc_unreachable (); } if (REG_P (SET_DEST (set))) { old_reg = SET_DEST (set); /* Check if we can modify the set destination in the original insn. */ if (validate_change (insn, &SET_DEST (set), reg, 0)) { new_insn = gen_move_insn (old_reg, reg); new_insn = emit_insn_after (new_insn, insn); } else { new_insn = gen_move_insn (reg, old_reg); new_insn = emit_insn_after (new_insn, insn); } } else /* This is possible only in case of a store to memory. */ { old_reg = SET_SRC (set); new_insn = gen_move_insn (reg, old_reg); /* Check if we can modify the set source in the original insn. */ if (validate_change (insn, &SET_SRC (set), reg, 0)) new_insn = emit_insn_before (new_insn, insn); else new_insn = emit_insn_after (new_insn, insn); } gcse_create_count++; if (dump_file) fprintf (dump_file, "PRE: bb %d, insn %d, copy expression %d in insn %d to reg %d\n", BLOCK_FOR_INSN (insn)->index, INSN_UID (new_insn), indx, INSN_UID (insn), regno); } /* Copy available expressions that reach the redundant expression to `reaching_reg'. */ static void pre_insert_copies (void) { unsigned int i, added_copy; struct gcse_expr *expr; struct gcse_occr *occr; struct gcse_occr *avail; /* For each available expression in the table, copy the result to `reaching_reg' if the expression reaches a deleted one. ??? The current algorithm is rather brute force. Need to do some profiling. */ for (i = 0; i < expr_hash_table.size; i++) for (expr = expr_hash_table.table[i]; expr; expr = expr->next_same_hash) { /* If the basic block isn't reachable, PPOUT will be TRUE. However, we don't want to insert a copy here because the expression may not really be redundant. So only insert an insn if the expression was deleted. This test also avoids further processing if the expression wasn't deleted anywhere. */ if (expr->reaching_reg == NULL) continue; /* Set when we add a copy for that expression. */ added_copy = 0; for (occr = expr->antic_occr; occr != NULL; occr = occr->next) { if (! occr->deleted_p) continue; for (avail = expr->avail_occr; avail != NULL; avail = avail->next) { rtx_insn *insn = avail->insn; /* No need to handle this one if handled already. */ if (avail->copied_p) continue; /* Don't handle this one if it's a redundant one. */ if (insn->deleted ()) continue; /* Or if the expression doesn't reach the deleted one. */ if (! pre_expr_reaches_here_p (BLOCK_FOR_INSN (avail->insn), expr, BLOCK_FOR_INSN (occr->insn))) continue; added_copy = 1; /* Copy the result of avail to reaching_reg. */ pre_insert_copy_insn (expr, insn); avail->copied_p = 1; } } if (added_copy) update_ld_motion_stores (expr); } } struct set_data { rtx_insn *insn; const_rtx set; int nsets; }; /* Increment number of sets and record set in DATA. */ static void record_set_data (rtx dest, const_rtx set, void *data) { struct set_data *s = (struct set_data *)data; if (GET_CODE (set) == SET) { /* We allow insns having multiple sets, where all but one are dead as single set insns. In the common case only a single set is present, so we want to avoid checking for REG_UNUSED notes unless necessary. */ if (s->nsets == 1 && find_reg_note (s->insn, REG_UNUSED, SET_DEST (s->set)) && !side_effects_p (s->set)) s->nsets = 0; if (!s->nsets) { /* Record this set. */ s->nsets += 1; s->set = set; } else if (!find_reg_note (s->insn, REG_UNUSED, dest) || side_effects_p (set)) s->nsets += 1; } } static const_rtx single_set_gcse (rtx_insn *insn) { struct set_data s; rtx pattern; gcc_assert (INSN_P (insn)); /* Optimize common case. */ pattern = PATTERN (insn); if (GET_CODE (pattern) == SET) return pattern; s.insn = insn; s.nsets = 0; note_stores (pattern, record_set_data, &s); /* Considered invariant insns have exactly one set. */ gcc_assert (s.nsets == 1); return s.set; } /* Emit move from SRC to DEST noting the equivalence with expression computed in INSN. */ static rtx_insn * gcse_emit_move_after (rtx dest, rtx src, rtx_insn *insn) { rtx_insn *new_rtx; const_rtx set = single_set_gcse (insn); rtx set2; rtx note; rtx eqv = NULL_RTX; /* This should never fail since we're creating a reg->reg copy we've verified to be valid. */ new_rtx = emit_insn_after (gen_move_insn (dest, src), insn); /* Note the equivalence for local CSE pass. Take the note from the old set if there was one. Otherwise record the SET_SRC from the old set unless DEST is also an operand of the SET_SRC. */ set2 = single_set (new_rtx); if (!set2 || !rtx_equal_p (SET_DEST (set2), dest)) return new_rtx; if ((note = find_reg_equal_equiv_note (insn))) eqv = XEXP (note, 0); else if (! REG_P (dest) || ! reg_mentioned_p (dest, SET_SRC (set))) eqv = SET_SRC (set); if (eqv != NULL_RTX) set_unique_reg_note (new_rtx, REG_EQUAL, copy_insn_1 (eqv)); return new_rtx; } /* Delete redundant computations. Deletion is done by changing the insn to copy the `reaching_reg' of the expression into the result of the SET. It is left to later passes to propagate the copy or eliminate it. Return nonzero if a change is made. */ static int pre_delete (void) { unsigned int i; int changed; struct gcse_expr *expr; struct gcse_occr *occr; changed = 0; for (i = 0; i < expr_hash_table.size; i++) for (expr = expr_hash_table.table[i]; expr; expr = expr->next_same_hash) { int indx = expr->bitmap_index; /* We only need to search antic_occr since we require ANTLOC != 0. */ for (occr = expr->antic_occr; occr != NULL; occr = occr->next) { rtx_insn *insn = occr->insn; rtx set; basic_block bb = BLOCK_FOR_INSN (insn); /* We only delete insns that have a single_set. */ if (bitmap_bit_p (pre_delete_map[bb->index], indx) && (set = single_set (insn)) != 0 && dbg_cnt (pre_insn)) { /* Create a pseudo-reg to store the result of reaching expressions into. Get the mode for the new pseudo from the mode of the original destination pseudo. */ if (expr->reaching_reg == NULL) expr->reaching_reg = gen_reg_rtx_and_attrs (SET_DEST (set)); gcse_emit_move_after (SET_DEST (set), expr->reaching_reg, insn); delete_insn (insn); occr->deleted_p = 1; changed = 1; gcse_subst_count++; if (dump_file) { fprintf (dump_file, "PRE: redundant insn %d (expression %d) in ", INSN_UID (insn), indx); fprintf (dump_file, "bb %d, reaching reg is %d\n", bb->index, REGNO (expr->reaching_reg)); } } } } return changed; } /* Perform GCSE optimizations using PRE. This is called by one_pre_gcse_pass after all the dataflow analysis has been done. This is based on the original Morel-Renvoise paper Fred Chow's thesis, and lazy code motion from Knoop, Ruthing and Steffen as described in Advanced Compiler Design and Implementation. ??? A new pseudo reg is created to hold the reaching expression. The nice thing about the classical approach is that it would try to use an existing reg. If the register can't be adequately optimized [i.e. we introduce reload problems], one could add a pass here to propagate the new register through the block. ??? We don't handle single sets in PARALLELs because we're [currently] not able to copy the rest of the parallel when we insert copies to create full redundancies from partial redundancies. However, there's no reason why we can't handle PARALLELs in the cases where there are no partial redundancies. */ static int pre_gcse (struct edge_list *edge_list) { unsigned int i; int did_insert, changed; struct gcse_expr **index_map; struct gcse_expr *expr; /* Compute a mapping from expression number (`bitmap_index') to hash table entry. */ index_map = XCNEWVEC (struct gcse_expr *, expr_hash_table.n_elems); for (i = 0; i < expr_hash_table.size; i++) for (expr = expr_hash_table.table[i]; expr; expr = expr->next_same_hash) index_map[expr->bitmap_index] = expr; /* Delete the redundant insns first so that - we know what register to use for the new insns and for the other ones with reaching expressions - we know which insns are redundant when we go to create copies */ changed = pre_delete (); did_insert = pre_edge_insert (edge_list, index_map); /* In other places with reaching expressions, copy the expression to the specially allocated pseudo-reg that reaches the redundant expr. */ pre_insert_copies (); if (did_insert) { commit_edge_insertions (); changed = 1; } free (index_map); return changed; } /* Top level routine to perform one PRE GCSE pass. Return nonzero if a change was made. */ static int one_pre_gcse_pass (void) { int changed = 0; gcse_subst_count = 0; gcse_create_count = 0; /* Return if there's nothing to do, or it is too expensive. */ if (n_basic_blocks_for_fn (cfun) <= NUM_FIXED_BLOCKS + 1 || is_too_expensive (_("PRE disabled"))) return 0; /* We need alias. */ init_alias_analysis (); bytes_used = 0; gcc_obstack_init (&gcse_obstack); alloc_gcse_mem (); alloc_hash_table (&expr_hash_table); add_noreturn_fake_exit_edges (); if (flag_gcse_lm) compute_ld_motion_mems (); compute_hash_table (&expr_hash_table); if (flag_gcse_lm) trim_ld_motion_mems (); if (dump_file) dump_hash_table (dump_file, "Expression", &expr_hash_table); if (expr_hash_table.n_elems > 0) { struct edge_list *edge_list; alloc_pre_mem (last_basic_block_for_fn (cfun), expr_hash_table.n_elems); edge_list = compute_pre_data (); changed |= pre_gcse (edge_list); free_edge_list (edge_list); free_pre_mem (); } if (flag_gcse_lm) free_ld_motion_mems (); remove_fake_exit_edges (); free_hash_table (&expr_hash_table); free_gcse_mem (); obstack_free (&gcse_obstack, NULL); /* We are finished with alias. */ end_alias_analysis (); if (dump_file) { fprintf (dump_file, "PRE GCSE of %s, %d basic blocks, %d bytes needed, ", current_function_name (), n_basic_blocks_for_fn (cfun), bytes_used); fprintf (dump_file, "%d substs, %d insns created\n", gcse_subst_count, gcse_create_count); } return changed; } /* If X contains any LABEL_REF's, add REG_LABEL_OPERAND notes for them to INSN. If such notes are added to an insn which references a CODE_LABEL, the LABEL_NUSES count is incremented. We have to add that note, because the following loop optimization pass requires them. */ /* ??? If there was a jump optimization pass after gcse and before loop, then we would not need to do this here, because jump would add the necessary REG_LABEL_OPERAND and REG_LABEL_TARGET notes. */ static void add_label_notes (rtx x, rtx_insn *insn) { enum rtx_code code = GET_CODE (x); int i, j; const char *fmt; if (code == LABEL_REF && !LABEL_REF_NONLOCAL_P (x)) { /* This code used to ignore labels that referred to dispatch tables to avoid flow generating (slightly) worse code. We no longer ignore such label references (see LABEL_REF handling in mark_jump_label for additional information). */ /* There's no reason for current users to emit jump-insns with such a LABEL_REF, so we don't have to handle REG_LABEL_TARGET notes. */ gcc_assert (!JUMP_P (insn)); add_reg_note (insn, REG_LABEL_OPERAND, LABEL_REF_LABEL (x)); if (LABEL_P (LABEL_REF_LABEL (x))) LABEL_NUSES (LABEL_REF_LABEL (x))++; return; } for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--) { if (fmt[i] == 'e') add_label_notes (XEXP (x, i), insn); else if (fmt[i] == 'E') for (j = XVECLEN (x, i) - 1; j >= 0; j--) add_label_notes (XVECEXP (x, i, j), insn); } } /* Code Hoisting variables and subroutines. */ /* Very busy expressions. */ static sbitmap *hoist_vbein; static sbitmap *hoist_vbeout; /* ??? We could compute post dominators and run this algorithm in reverse to perform tail merging, doing so would probably be more effective than the tail merging code in jump.c. It's unclear if tail merging could be run in parallel with code hoisting. It would be nice. */ /* Allocate vars used for code hoisting analysis. */ static void alloc_code_hoist_mem (int n_blocks, int n_exprs) { antloc = sbitmap_vector_alloc (n_blocks, n_exprs); transp = sbitmap_vector_alloc (n_blocks, n_exprs); comp = sbitmap_vector_alloc (n_blocks, n_exprs); hoist_vbein = sbitmap_vector_alloc (n_blocks, n_exprs); hoist_vbeout = sbitmap_vector_alloc (n_blocks, n_exprs); } /* Free vars used for code hoisting analysis. */ static void free_code_hoist_mem (void) { sbitmap_vector_free (antloc); sbitmap_vector_free (transp); sbitmap_vector_free (comp); sbitmap_vector_free (hoist_vbein); sbitmap_vector_free (hoist_vbeout); free_dominance_info (CDI_DOMINATORS); } /* Compute the very busy expressions at entry/exit from each block. An expression is very busy if all paths from a given point compute the expression. */ static void compute_code_hoist_vbeinout (void) { int changed, passes; basic_block bb; bitmap_vector_clear (hoist_vbeout, last_basic_block_for_fn (cfun)); bitmap_vector_clear (hoist_vbein, last_basic_block_for_fn (cfun)); passes = 0; changed = 1; while (changed) { changed = 0; /* We scan the blocks in the reverse order to speed up the convergence. */ FOR_EACH_BB_REVERSE_FN (bb, cfun) { if (bb->next_bb != EXIT_BLOCK_PTR_FOR_FN (cfun)) { bitmap_intersection_of_succs (hoist_vbeout[bb->index], hoist_vbein, bb); /* Include expressions in VBEout that are calculated in BB and available at its end. */ bitmap_ior (hoist_vbeout[bb->index], hoist_vbeout[bb->index], comp[bb->index]); } changed |= bitmap_or_and (hoist_vbein[bb->index], antloc[bb->index], hoist_vbeout[bb->index], transp[bb->index]); } passes++; } if (dump_file) { fprintf (dump_file, "hoisting vbeinout computation: %d passes\n", passes); FOR_EACH_BB_FN (bb, cfun) { fprintf (dump_file, "vbein (%d): ", bb->index); dump_bitmap_file (dump_file, hoist_vbein[bb->index]); fprintf (dump_file, "vbeout(%d): ", bb->index); dump_bitmap_file (dump_file, hoist_vbeout[bb->index]); } } } /* Top level routine to do the dataflow analysis needed by code hoisting. */ static void compute_code_hoist_data (void) { compute_local_properties (transp, comp, antloc, &expr_hash_table); prune_expressions (false); compute_code_hoist_vbeinout (); calculate_dominance_info (CDI_DOMINATORS); if (dump_file) fprintf (dump_file, "\n"); } /* Update register pressure for BB when hoisting an expression from instruction FROM, if live ranges of inputs are shrunk. Also maintain live_in information if live range of register referred in FROM is shrunk. Return 0 if register pressure doesn't change, otherwise return the number by which register pressure is decreased. NOTE: Register pressure won't be increased in this function. */ static int update_bb_reg_pressure (basic_block bb, rtx_insn *from) { rtx dreg; rtx_insn *insn; basic_block succ_bb; df_ref use, op_ref; edge succ; edge_iterator ei; int decreased_pressure = 0; int nregs; enum reg_class pressure_class; FOR_EACH_INSN_USE (use, from) { dreg = DF_REF_REAL_REG (use); /* The live range of register is shrunk only if it isn't: 1. referred on any path from the end of this block to EXIT, or 2. referred by insns other than FROM in this block. */ FOR_EACH_EDGE (succ, ei, bb->succs) { succ_bb = succ->dest; if (succ_bb == EXIT_BLOCK_PTR_FOR_FN (cfun)) continue; if (bitmap_bit_p (BB_DATA (succ_bb)->live_in, REGNO (dreg))) break; } if (succ != NULL) continue; op_ref = DF_REG_USE_CHAIN (REGNO (dreg)); for (; op_ref; op_ref = DF_REF_NEXT_REG (op_ref)) { if (!DF_REF_INSN_INFO (op_ref)) continue; insn = DF_REF_INSN (op_ref); if (BLOCK_FOR_INSN (insn) == bb && NONDEBUG_INSN_P (insn) && insn != from) break; } pressure_class = get_regno_pressure_class (REGNO (dreg), &nregs); /* Decrease register pressure and update live_in information for this block. */ if (!op_ref && pressure_class != NO_REGS) { decreased_pressure += nregs; BB_DATA (bb)->max_reg_pressure[pressure_class] -= nregs; bitmap_clear_bit (BB_DATA (bb)->live_in, REGNO (dreg)); } } return decreased_pressure; } /* Determine if the expression EXPR should be hoisted to EXPR_BB up in flow graph, if it can reach BB unimpared. Stop the search if the expression would need to be moved more than DISTANCE instructions. DISTANCE is the number of instructions through which EXPR can be hoisted up in flow graph. BB_SIZE points to an array which contains the number of instructions for each basic block. PRESSURE_CLASS and NREGS are register class and number of hard registers for storing EXPR. HOISTED_BBS points to a bitmap indicating basic blocks through which EXPR is hoisted. FROM is the instruction from which EXPR is hoisted. It's unclear exactly what Muchnick meant by "unimpared". It seems to me that the expression must either be computed or transparent in *every* block in the path(s) from EXPR_BB to BB. Any other definition would allow the expression to be hoisted out of loops, even if the expression wasn't a loop invariant. Contrast this to reachability for PRE where an expression is considered reachable if *any* path reaches instead of *all* paths. */ static int should_hoist_expr_to_dom (basic_block expr_bb, struct gcse_expr *expr, basic_block bb, sbitmap visited, int distance, int *bb_size, enum reg_class pressure_class, int *nregs, bitmap hoisted_bbs, rtx_insn *from) { unsigned int i; edge pred; edge_iterator ei; sbitmap_iterator sbi; int visited_allocated_locally = 0; int decreased_pressure = 0; if (flag_ira_hoist_pressure) { /* Record old information of basic block BB when it is visited at the first time. */ if (!bitmap_bit_p (hoisted_bbs, bb->index)) { struct bb_data *data = BB_DATA (bb); bitmap_copy (data->backup, data->live_in); data->old_pressure = data->max_reg_pressure[pressure_class]; } decreased_pressure = update_bb_reg_pressure (bb, from); } /* Terminate the search if distance, for which EXPR is allowed to move, is exhausted. */ if (distance > 0) { if (flag_ira_hoist_pressure) { /* Prefer to hoist EXPR if register pressure is decreased. */ if (decreased_pressure > *nregs) distance += bb_size[bb->index]; /* Let EXPR be hoisted through basic block at no cost if one of following conditions is satisfied: 1. The basic block has low register pressure. 2. Register pressure won't be increases after hoisting EXPR. Constant expressions is handled conservatively, because hoisting constant expression aggressively results in worse code. This decision is made by the observation of CSiBE on ARM target, while it has no obvious effect on other targets like x86, x86_64, mips and powerpc. */ else if (CONST_INT_P (expr->expr) || (BB_DATA (bb)->max_reg_pressure[pressure_class] >= ira_class_hard_regs_num[pressure_class] && decreased_pressure < *nregs)) distance -= bb_size[bb->index]; } else distance -= bb_size[bb->index]; if (distance <= 0) return 0; } else gcc_assert (distance == 0); if (visited == NULL) { visited_allocated_locally = 1; visited = sbitmap_alloc (last_basic_block_for_fn (cfun)); bitmap_clear (visited); } FOR_EACH_EDGE (pred, ei, bb->preds) { basic_block pred_bb = pred->src; if (pred->src == ENTRY_BLOCK_PTR_FOR_FN (cfun)) break; else if (pred_bb == expr_bb) continue; else if (bitmap_bit_p (visited, pred_bb->index)) continue; else if (! bitmap_bit_p (transp[pred_bb->index], expr->bitmap_index)) break; /* Not killed. */ else { bitmap_set_bit (visited, pred_bb->index); if (! should_hoist_expr_to_dom (expr_bb, expr, pred_bb, visited, distance, bb_size, pressure_class, nregs, hoisted_bbs, from)) break; } } if (visited_allocated_locally) { /* If EXPR can be hoisted to expr_bb, record basic blocks through which EXPR is hoisted in hoisted_bbs. */ if (flag_ira_hoist_pressure && !pred) { /* Record the basic block from which EXPR is hoisted. */ bitmap_set_bit (visited, bb->index); EXECUTE_IF_SET_IN_BITMAP (visited, 0, i, sbi) bitmap_set_bit (hoisted_bbs, i); } sbitmap_free (visited); } return (pred == NULL); } /* Find occurrence in BB. */ static struct gcse_occr * find_occr_in_bb (struct gcse_occr *occr, basic_block bb) { /* Find the right occurrence of this expression. */ while (occr && BLOCK_FOR_INSN (occr->insn) != bb) occr = occr->next; return occr; } /* Actually perform code hoisting. The code hoisting pass can hoist multiple computations of the same expression along dominated path to a dominating basic block, like from b2/b3 to b1 as depicted below: b1 ------ /\ | / \ | bx by distance / \ | / \ | b2 b3 ------ Unfortunately code hoisting generally extends the live range of an output pseudo register, which increases register pressure and hurts register allocation. To address this issue, an attribute MAX_DISTANCE is computed and attached to each expression. The attribute is computed from rtx cost of the corresponding expression and it's used to control how long the expression can be hoisted up in flow graph. As the expression is hoisted up in flow graph, GCC decreases its DISTANCE and stops the hoist if DISTANCE reaches 0. Code hoisting can decrease register pressure if live ranges of inputs are shrunk. Option "-fira-hoist-pressure" implements register pressure directed hoist based on upper method. The rationale is: 1. Calculate register pressure for each basic block by reusing IRA facility. 2. When expression is hoisted through one basic block, GCC checks the change of live ranges for inputs/output. The basic block's register pressure will be increased because of extended live range of output. However, register pressure will be decreased if the live ranges of inputs are shrunk. 3. After knowing how hoisting affects register pressure, GCC prefers to hoist the expression if it can decrease register pressure, by increasing DISTANCE of the corresponding expression. 4. If hoisting the expression increases register pressure, GCC checks register pressure of the basic block and decrease DISTANCE only if the register pressure is high. In other words, expression will be hoisted through at no cost if the basic block has low register pressure. 5. Update register pressure information for basic blocks through which expression is hoisted. */ static int hoist_code (void) { basic_block bb, dominated; vec dom_tree_walk; unsigned int dom_tree_walk_index; vec domby; unsigned int i, j, k; struct gcse_expr **index_map; struct gcse_expr *expr; int *to_bb_head; int *bb_size; int changed = 0; struct bb_data *data; /* Basic blocks that have occurrences reachable from BB. */ bitmap from_bbs; /* Basic blocks through which expr is hoisted. */ bitmap hoisted_bbs = NULL; bitmap_iterator bi; /* Compute a mapping from expression number (`bitmap_index') to hash table entry. */ index_map = XCNEWVEC (struct gcse_expr *, expr_hash_table.n_elems); for (i = 0; i < expr_hash_table.size; i++) for (expr = expr_hash_table.table[i]; expr; expr = expr->next_same_hash) index_map[expr->bitmap_index] = expr; /* Calculate sizes of basic blocks and note how far each instruction is from the start of its block. We then use this data to restrict distance an expression can travel. */ to_bb_head = XCNEWVEC (int, get_max_uid ()); bb_size = XCNEWVEC (int, last_basic_block_for_fn (cfun)); FOR_EACH_BB_FN (bb, cfun) { rtx_insn *insn; int to_head; to_head = 0; FOR_BB_INSNS (bb, insn) { /* Don't count debug instructions to avoid them affecting decision choices. */ if (NONDEBUG_INSN_P (insn)) to_bb_head[INSN_UID (insn)] = to_head++; } bb_size[bb->index] = to_head; } gcc_assert (EDGE_COUNT (ENTRY_BLOCK_PTR_FOR_FN (cfun)->succs) == 1 && (EDGE_SUCC (ENTRY_BLOCK_PTR_FOR_FN (cfun), 0)->dest == ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb)); from_bbs = BITMAP_ALLOC (NULL); if (flag_ira_hoist_pressure) hoisted_bbs = BITMAP_ALLOC (NULL); dom_tree_walk = get_all_dominated_blocks (CDI_DOMINATORS, ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb); /* Walk over each basic block looking for potentially hoistable expressions, nothing gets hoisted from the entry block. */ FOR_EACH_VEC_ELT (dom_tree_walk, dom_tree_walk_index, bb) { domby = get_dominated_to_depth (CDI_DOMINATORS, bb, MAX_HOIST_DEPTH); if (domby.length () == 0) continue; /* Examine each expression that is very busy at the exit of this block. These are the potentially hoistable expressions. */ for (i = 0; i < SBITMAP_SIZE (hoist_vbeout[bb->index]); i++) { if (bitmap_bit_p (hoist_vbeout[bb->index], i)) { int nregs = 0; enum reg_class pressure_class = NO_REGS; /* Current expression. */ struct gcse_expr *expr = index_map[i]; /* Number of occurrences of EXPR that can be hoisted to BB. */ int hoistable = 0; /* Occurrences reachable from BB. */ vec occrs_to_hoist = vNULL; /* We want to insert the expression into BB only once, so note when we've inserted it. */ int insn_inserted_p; occr_t occr; /* If an expression is computed in BB and is available at end of BB, hoist all occurrences dominated by BB to BB. */ if (bitmap_bit_p (comp[bb->index], i)) { occr = find_occr_in_bb (expr->antic_occr, bb); if (occr) { /* An occurrence might've been already deleted while processing a dominator of BB. */ if (!occr->deleted_p) { gcc_assert (NONDEBUG_INSN_P (occr->insn)); hoistable++; } } else hoistable++; } /* We've found a potentially hoistable expression, now we look at every block BB dominates to see if it computes the expression. */ FOR_EACH_VEC_ELT (domby, j, dominated) { int max_distance; /* Ignore self dominance. */ if (bb == dominated) continue; /* We've found a dominated block, now see if it computes the busy expression and whether or not moving that expression to the "beginning" of that block is safe. */ if (!bitmap_bit_p (antloc[dominated->index], i)) continue; occr = find_occr_in_bb (expr->antic_occr, dominated); gcc_assert (occr); /* An occurrence might've been already deleted while processing a dominator of BB. */ if (occr->deleted_p) continue; gcc_assert (NONDEBUG_INSN_P (occr->insn)); max_distance = expr->max_distance; if (max_distance > 0) /* Adjust MAX_DISTANCE to account for the fact that OCCR won't have to travel all of DOMINATED, but only part of it. */ max_distance += (bb_size[dominated->index] - to_bb_head[INSN_UID (occr->insn)]); pressure_class = get_pressure_class_and_nregs (occr->insn, &nregs); /* Note if the expression should be hoisted from the dominated block to BB if it can reach DOMINATED unimpared. Keep track of how many times this expression is hoistable from a dominated block into BB. */ if (should_hoist_expr_to_dom (bb, expr, dominated, NULL, max_distance, bb_size, pressure_class, &nregs, hoisted_bbs, occr->insn)) { hoistable++; occrs_to_hoist.safe_push (occr); bitmap_set_bit (from_bbs, dominated->index); } } /* If we found more than one hoistable occurrence of this expression, then note it in the vector of expressions to hoist. It makes no sense to hoist things which are computed in only one BB, and doing so tends to pessimize register allocation. One could increase this value to try harder to avoid any possible code expansion due to register allocation issues; however experiments have shown that the vast majority of hoistable expressions are only movable from two successors, so raising this threshold is likely to nullify any benefit we get from code hoisting. */ if (hoistable > 1 && dbg_cnt (hoist_insn)) { /* If (hoistable != vec::length), then there is an occurrence of EXPR in BB itself. Don't waste time looking for LCA in this case. */ if ((unsigned) hoistable == occrs_to_hoist.length ()) { basic_block lca; lca = nearest_common_dominator_for_set (CDI_DOMINATORS, from_bbs); if (lca != bb) /* Punt, it's better to hoist these occurrences to LCA. */ occrs_to_hoist.release (); } } else /* Punt, no point hoisting a single occurrence. */ occrs_to_hoist.release (); if (flag_ira_hoist_pressure && !occrs_to_hoist.is_empty ()) { /* Increase register pressure of basic blocks to which expr is hoisted because of extended live range of output. */ data = BB_DATA (bb); data->max_reg_pressure[pressure_class] += nregs; EXECUTE_IF_SET_IN_BITMAP (hoisted_bbs, 0, k, bi) { data = BB_DATA (BASIC_BLOCK_FOR_FN (cfun, k)); data->max_reg_pressure[pressure_class] += nregs; } } else if (flag_ira_hoist_pressure) { /* Restore register pressure and live_in info for basic blocks recorded in hoisted_bbs when expr will not be hoisted. */ EXECUTE_IF_SET_IN_BITMAP (hoisted_bbs, 0, k, bi) { data = BB_DATA (BASIC_BLOCK_FOR_FN (cfun, k)); bitmap_copy (data->live_in, data->backup); data->max_reg_pressure[pressure_class] = data->old_pressure; } } if (flag_ira_hoist_pressure) bitmap_clear (hoisted_bbs); insn_inserted_p = 0; /* Walk through occurrences of I'th expressions we want to hoist to BB and make the transformations. */ FOR_EACH_VEC_ELT (occrs_to_hoist, j, occr) { rtx_insn *insn; const_rtx set; gcc_assert (!occr->deleted_p); insn = occr->insn; set = single_set_gcse (insn); /* Create a pseudo-reg to store the result of reaching expressions into. Get the mode for the new pseudo from the mode of the original destination pseudo. It is important to use new pseudos whenever we emit a set. This will allow reload to use rematerialization for such registers. */ if (!insn_inserted_p) expr->reaching_reg = gen_reg_rtx_and_attrs (SET_DEST (set)); gcse_emit_move_after (SET_DEST (set), expr->reaching_reg, insn); delete_insn (insn); occr->deleted_p = 1; changed = 1; gcse_subst_count++; if (!insn_inserted_p) { insert_insn_end_basic_block (expr, bb); insn_inserted_p = 1; } } occrs_to_hoist.release (); bitmap_clear (from_bbs); } } domby.release (); } dom_tree_walk.release (); BITMAP_FREE (from_bbs); if (flag_ira_hoist_pressure) BITMAP_FREE (hoisted_bbs); free (bb_size); free (to_bb_head); free (index_map); return changed; } /* Return pressure class and number of needed hard registers (through *NREGS) of register REGNO. */ static enum reg_class get_regno_pressure_class (int regno, int *nregs) { if (regno >= FIRST_PSEUDO_REGISTER) { enum reg_class pressure_class; pressure_class = reg_allocno_class (regno); pressure_class = ira_pressure_class_translate[pressure_class]; *nregs = ira_reg_class_max_nregs[pressure_class][PSEUDO_REGNO_MODE (regno)]; return pressure_class; } else if (! TEST_HARD_REG_BIT (ira_no_alloc_regs, regno) && ! TEST_HARD_REG_BIT (eliminable_regset, regno)) { *nregs = 1; return ira_pressure_class_translate[REGNO_REG_CLASS (regno)]; } else { *nregs = 0; return NO_REGS; } } /* Return pressure class and number of hard registers (through *NREGS) for destination of INSN. */ static enum reg_class get_pressure_class_and_nregs (rtx_insn *insn, int *nregs) { rtx reg; enum reg_class pressure_class; const_rtx set = single_set_gcse (insn); reg = SET_DEST (set); if (GET_CODE (reg) == SUBREG) reg = SUBREG_REG (reg); if (MEM_P (reg)) { *nregs = 0; pressure_class = NO_REGS; } else { gcc_assert (REG_P (reg)); pressure_class = reg_allocno_class (REGNO (reg)); pressure_class = ira_pressure_class_translate[pressure_class]; *nregs = ira_reg_class_max_nregs[pressure_class][GET_MODE (SET_SRC (set))]; } return pressure_class; } /* Increase (if INCR_P) or decrease current register pressure for register REGNO. */ static void change_pressure (int regno, bool incr_p) { int nregs; enum reg_class pressure_class; pressure_class = get_regno_pressure_class (regno, &nregs); if (! incr_p) curr_reg_pressure[pressure_class] -= nregs; else { curr_reg_pressure[pressure_class] += nregs; if (BB_DATA (curr_bb)->max_reg_pressure[pressure_class] < curr_reg_pressure[pressure_class]) BB_DATA (curr_bb)->max_reg_pressure[pressure_class] = curr_reg_pressure[pressure_class]; } } /* Calculate register pressure for each basic block by walking insns from last to first. */ static void calculate_bb_reg_pressure (void) { int i; unsigned int j; rtx_insn *insn; basic_block bb; bitmap curr_regs_live; bitmap_iterator bi; ira_setup_eliminable_regset (); curr_regs_live = BITMAP_ALLOC (®_obstack); FOR_EACH_BB_FN (bb, cfun) { curr_bb = bb; BB_DATA (bb)->live_in = BITMAP_ALLOC (NULL); BB_DATA (bb)->backup = BITMAP_ALLOC (NULL); bitmap_copy (BB_DATA (bb)->live_in, df_get_live_in (bb)); bitmap_copy (curr_regs_live, df_get_live_out (bb)); for (i = 0; i < ira_pressure_classes_num; i++) curr_reg_pressure[ira_pressure_classes[i]] = 0; EXECUTE_IF_SET_IN_BITMAP (curr_regs_live, 0, j, bi) change_pressure (j, true); FOR_BB_INSNS_REVERSE (bb, insn) { rtx dreg; int regno; df_ref def, use; if (! NONDEBUG_INSN_P (insn)) continue; FOR_EACH_INSN_DEF (def, insn) { dreg = DF_REF_REAL_REG (def); gcc_assert (REG_P (dreg)); regno = REGNO (dreg); if (!(DF_REF_FLAGS (def) & (DF_REF_PARTIAL | DF_REF_CONDITIONAL))) { if (bitmap_clear_bit (curr_regs_live, regno)) change_pressure (regno, false); } } FOR_EACH_INSN_USE (use, insn) { dreg = DF_REF_REAL_REG (use); gcc_assert (REG_P (dreg)); regno = REGNO (dreg); if (bitmap_set_bit (curr_regs_live, regno)) change_pressure (regno, true); } } } BITMAP_FREE (curr_regs_live); if (dump_file == NULL) return; fprintf (dump_file, "\nRegister Pressure: \n"); FOR_EACH_BB_FN (bb, cfun) { fprintf (dump_file, " Basic block %d: \n", bb->index); for (i = 0; (int) i < ira_pressure_classes_num; i++) { enum reg_class pressure_class; pressure_class = ira_pressure_classes[i]; if (BB_DATA (bb)->max_reg_pressure[pressure_class] == 0) continue; fprintf (dump_file, " %s=%d\n", reg_class_names[pressure_class], BB_DATA (bb)->max_reg_pressure[pressure_class]); } } fprintf (dump_file, "\n"); } /* Top level routine to perform one code hoisting (aka unification) pass Return nonzero if a change was made. */ static int one_code_hoisting_pass (void) { int changed = 0; gcse_subst_count = 0; gcse_create_count = 0; /* Return if there's nothing to do, or it is too expensive. */ if (n_basic_blocks_for_fn (cfun) <= NUM_FIXED_BLOCKS + 1 || is_too_expensive (_("GCSE disabled"))) return 0; doing_code_hoisting_p = true; /* Calculate register pressure for each basic block. */ if (flag_ira_hoist_pressure) { regstat_init_n_sets_and_refs (); ira_set_pseudo_classes (false, dump_file); alloc_aux_for_blocks (sizeof (struct bb_data)); calculate_bb_reg_pressure (); regstat_free_n_sets_and_refs (); } /* We need alias. */ init_alias_analysis (); bytes_used = 0; gcc_obstack_init (&gcse_obstack); alloc_gcse_mem (); alloc_hash_table (&expr_hash_table); compute_hash_table (&expr_hash_table); if (dump_file) dump_hash_table (dump_file, "Code Hosting Expressions", &expr_hash_table); if (expr_hash_table.n_elems > 0) { alloc_code_hoist_mem (last_basic_block_for_fn (cfun), expr_hash_table.n_elems); compute_code_hoist_data (); changed = hoist_code (); free_code_hoist_mem (); } if (flag_ira_hoist_pressure) { free_aux_for_blocks (); free_reg_info (); } free_hash_table (&expr_hash_table); free_gcse_mem (); obstack_free (&gcse_obstack, NULL); /* We are finished with alias. */ end_alias_analysis (); if (dump_file) { fprintf (dump_file, "HOIST of %s, %d basic blocks, %d bytes needed, ", current_function_name (), n_basic_blocks_for_fn (cfun), bytes_used); fprintf (dump_file, "%d substs, %d insns created\n", gcse_subst_count, gcse_create_count); } doing_code_hoisting_p = false; return changed; } /* Here we provide the things required to do store motion towards the exit. In order for this to be effective, gcse also needed to be taught how to move a load when it is killed only by a store to itself. int i; float a[10]; void foo(float scale) { for (i=0; i<10; i++) a[i] *= scale; } 'i' is both loaded and stored to in the loop. Normally, gcse cannot move the load out since its live around the loop, and stored at the bottom of the loop. The 'Load Motion' referred to and implemented in this file is an enhancement to gcse which when using edge based LCM, recognizes this situation and allows gcse to move the load out of the loop. Once gcse has hoisted the load, store motion can then push this load towards the exit, and we end up with no loads or stores of 'i' in the loop. */ /* This will search the ldst list for a matching expression. If it doesn't find one, we create one and initialize it. */ static struct ls_expr * ldst_entry (rtx x) { int do_not_record_p = 0; struct ls_expr * ptr; unsigned int hash; ls_expr **slot; struct ls_expr e; hash = hash_rtx (x, GET_MODE (x), &do_not_record_p, NULL, /*have_reg_qty=*/false); e.pattern = x; slot = pre_ldst_table->find_slot_with_hash (&e, hash, INSERT); if (*slot) return *slot; ptr = XNEW (struct ls_expr); ptr->next = pre_ldst_mems; ptr->expr = NULL; ptr->pattern = x; ptr->pattern_regs = NULL_RTX; ptr->loads = NULL; ptr->stores = NULL; ptr->reaching_reg = NULL_RTX; ptr->invalid = 0; ptr->index = 0; ptr->hash_index = hash; pre_ldst_mems = ptr; *slot = ptr; return ptr; } /* Free up an individual ldst entry. */ static void free_ldst_entry (struct ls_expr * ptr) { free_INSN_LIST_list (& ptr->loads); free_INSN_LIST_list (& ptr->stores); free (ptr); } /* Free up all memory associated with the ldst list. */ static void free_ld_motion_mems (void) { delete pre_ldst_table; pre_ldst_table = NULL; while (pre_ldst_mems) { struct ls_expr * tmp = pre_ldst_mems; pre_ldst_mems = pre_ldst_mems->next; free_ldst_entry (tmp); } pre_ldst_mems = NULL; } /* Dump debugging info about the ldst list. */ static void print_ldst_list (FILE * file) { struct ls_expr * ptr; fprintf (file, "LDST list: \n"); for (ptr = pre_ldst_mems; ptr != NULL; ptr = ptr->next) { fprintf (file, " Pattern (%3d): ", ptr->index); print_rtl (file, ptr->pattern); fprintf (file, "\n Loads : "); if (ptr->loads) print_rtl (file, ptr->loads); else fprintf (file, "(nil)"); fprintf (file, "\n Stores : "); if (ptr->stores) print_rtl (file, ptr->stores); else fprintf (file, "(nil)"); fprintf (file, "\n\n"); } fprintf (file, "\n"); } /* Returns 1 if X is in the list of ldst only expressions. */ static struct ls_expr * find_rtx_in_ldst (rtx x) { struct ls_expr e; ls_expr **slot; if (!pre_ldst_table) return NULL; e.pattern = x; slot = pre_ldst_table->find_slot (&e, NO_INSERT); if (!slot || (*slot)->invalid) return NULL; return *slot; } /* Load Motion for loads which only kill themselves. */ /* Return true if x, a MEM, is a simple access with no side effects. These are the types of loads we consider for the ld_motion list, otherwise we let the usual aliasing take care of it. */ static int simple_mem (const_rtx x) { if (MEM_VOLATILE_P (x)) return 0; if (GET_MODE (x) == BLKmode) return 0; /* If we are handling exceptions, we must be careful with memory references that may trap. If we are not, the behavior is undefined, so we may just continue. */ if (cfun->can_throw_non_call_exceptions && may_trap_p (x)) return 0; if (side_effects_p (x)) return 0; /* Do not consider function arguments passed on stack. */ if (reg_mentioned_p (stack_pointer_rtx, x)) return 0; if (flag_float_store && FLOAT_MODE_P (GET_MODE (x))) return 0; return 1; } /* Make sure there isn't a buried reference in this pattern anywhere. If there is, invalidate the entry for it since we're not capable of fixing it up just yet.. We have to be sure we know about ALL loads since the aliasing code will allow all entries in the ld_motion list to not-alias itself. If we miss a load, we will get the wrong value since gcse might common it and we won't know to fix it up. */ static void invalidate_any_buried_refs (rtx x) { const char * fmt; int i, j; struct ls_expr * ptr; /* Invalidate it in the list. */ if (MEM_P (x) && simple_mem (x)) { ptr = ldst_entry (x); ptr->invalid = 1; } /* Recursively process the insn. */ fmt = GET_RTX_FORMAT (GET_CODE (x)); for (i = GET_RTX_LENGTH (GET_CODE (x)) - 1; i >= 0; i--) { if (fmt[i] == 'e') invalidate_any_buried_refs (XEXP (x, i)); else if (fmt[i] == 'E') for (j = XVECLEN (x, i) - 1; j >= 0; j--) invalidate_any_buried_refs (XVECEXP (x, i, j)); } } /* Find all the 'simple' MEMs which are used in LOADs and STORES. Simple being defined as MEM loads and stores to symbols, with no side effects and no registers in the expression. For a MEM destination, we also check that the insn is still valid if we replace the destination with a REG, as is done in update_ld_motion_stores. If there are any uses/defs which don't match this criteria, they are invalidated and trimmed out later. */ static void compute_ld_motion_mems (void) { struct ls_expr * ptr; basic_block bb; rtx_insn *insn; pre_ldst_mems = NULL; pre_ldst_table = new hash_table (13); FOR_EACH_BB_FN (bb, cfun) { FOR_BB_INSNS (bb, insn) { if (NONDEBUG_INSN_P (insn)) { if (GET_CODE (PATTERN (insn)) == SET) { rtx src = SET_SRC (PATTERN (insn)); rtx dest = SET_DEST (PATTERN (insn)); rtx note = find_reg_equal_equiv_note (insn); rtx src_eq; /* Check for a simple LOAD... */ if (MEM_P (src) && simple_mem (src)) { ptr = ldst_entry (src); if (REG_P (dest)) ptr->loads = alloc_INSN_LIST (insn, ptr->loads); else ptr->invalid = 1; } else { /* Make sure there isn't a buried load somewhere. */ invalidate_any_buried_refs (src); } if (note != 0 && REG_NOTE_KIND (note) == REG_EQUAL) src_eq = XEXP (note, 0); else src_eq = NULL_RTX; if (src_eq != NULL_RTX && !(MEM_P (src_eq) && simple_mem (src_eq))) invalidate_any_buried_refs (src_eq); /* Check for stores. Don't worry about aliased ones, they will block any movement we might do later. We only care about this exact pattern since those are the only circumstance that we will ignore the aliasing info. */ if (MEM_P (dest) && simple_mem (dest)) { ptr = ldst_entry (dest); if (! MEM_P (src) && GET_CODE (src) != ASM_OPERANDS /* Check for REG manually since want_to_gcse_p returns 0 for all REGs. */ && can_assign_to_reg_without_clobbers_p (src)) ptr->stores = alloc_INSN_LIST (insn, ptr->stores); else ptr->invalid = 1; } } else invalidate_any_buried_refs (PATTERN (insn)); } } } } /* Remove any references that have been either invalidated or are not in the expression list for pre gcse. */ static void trim_ld_motion_mems (void) { struct ls_expr * * last = & pre_ldst_mems; struct ls_expr * ptr = pre_ldst_mems; while (ptr != NULL) { struct gcse_expr * expr; /* Delete if entry has been made invalid. */ if (! ptr->invalid) { /* Delete if we cannot find this mem in the expression list. */ unsigned int hash = ptr->hash_index % expr_hash_table.size; for (expr = expr_hash_table.table[hash]; expr != NULL; expr = expr->next_same_hash) if (expr_equiv_p (expr->expr, ptr->pattern)) break; } else expr = (struct gcse_expr *) 0; if (expr) { /* Set the expression field if we are keeping it. */ ptr->expr = expr; last = & ptr->next; ptr = ptr->next; } else { *last = ptr->next; pre_ldst_table->remove_elt_with_hash (ptr, ptr->hash_index); free_ldst_entry (ptr); ptr = * last; } } /* Show the world what we've found. */ if (dump_file && pre_ldst_mems != NULL) print_ldst_list (dump_file); } /* This routine will take an expression which we are replacing with a reaching register, and update any stores that are needed if that expression is in the ld_motion list. Stores are updated by copying their SRC to the reaching register, and then storing the reaching register into the store location. These keeps the correct value in the reaching register for the loads. */ static void update_ld_motion_stores (struct gcse_expr * expr) { struct ls_expr * mem_ptr; if ((mem_ptr = find_rtx_in_ldst (expr->expr))) { /* We can try to find just the REACHED stores, but is shouldn't matter to set the reaching reg everywhere... some might be dead and should be eliminated later. */ /* We replace (set mem expr) with (set reg expr) (set mem reg) where reg is the reaching reg used in the load. We checked in compute_ld_motion_mems that we can replace (set mem expr) with (set reg expr) in that insn. */ rtx list = mem_ptr->stores; for ( ; list != NULL_RTX; list = XEXP (list, 1)) { rtx_insn *insn = as_a (XEXP (list, 0)); rtx pat = PATTERN (insn); rtx src = SET_SRC (pat); rtx reg = expr->reaching_reg; /* If we've already copied it, continue. */ if (expr->reaching_reg == src) continue; if (dump_file) { fprintf (dump_file, "PRE: store updated with reaching reg "); print_rtl (dump_file, reg); fprintf (dump_file, ":\n "); print_inline_rtx (dump_file, insn, 8); fprintf (dump_file, "\n"); } rtx_insn *copy = gen_move_insn (reg, copy_rtx (SET_SRC (pat))); emit_insn_before (copy, insn); SET_SRC (pat) = reg; df_insn_rescan (insn); /* un-recognize this pattern since it's probably different now. */ INSN_CODE (insn) = -1; gcse_create_count++; } } } /* Return true if the graph is too expensive to optimize. PASS is the optimization about to be performed. */ static bool is_too_expensive (const char *pass) { /* Trying to perform global optimizations on flow graphs which have a high connectivity will take a long time and is unlikely to be particularly useful. In normal circumstances a cfg should have about twice as many edges as blocks. But we do not want to punish small functions which have a couple switch statements. Rather than simply threshold the number of blocks, uses something with a more graceful degradation. */ if (n_edges_for_fn (cfun) > 20000 + n_basic_blocks_for_fn (cfun) * 4) { warning (OPT_Wdisabled_optimization, "%s: %d basic blocks and %d edges/basic block", pass, n_basic_blocks_for_fn (cfun), n_edges_for_fn (cfun) / n_basic_blocks_for_fn (cfun)); return true; } /* If allocating memory for the dataflow bitmaps would take up too much storage it's better just to disable the optimization. */ if ((n_basic_blocks_for_fn (cfun) * SBITMAP_SET_SIZE (max_reg_num ()) * sizeof (SBITMAP_ELT_TYPE)) > MAX_GCSE_MEMORY) { warning (OPT_Wdisabled_optimization, "%s: %d basic blocks and %d registers", pass, n_basic_blocks_for_fn (cfun), max_reg_num ()); return true; } return false; } static unsigned int execute_rtl_pre (void) { int changed; delete_unreachable_blocks (); df_analyze (); changed = one_pre_gcse_pass (); flag_rerun_cse_after_global_opts |= changed; if (changed) cleanup_cfg (0); return 0; } static unsigned int execute_rtl_hoist (void) { int changed; delete_unreachable_blocks (); df_analyze (); changed = one_code_hoisting_pass (); flag_rerun_cse_after_global_opts |= changed; if (changed) cleanup_cfg (0); return 0; } namespace { const pass_data pass_data_rtl_pre = { RTL_PASS, /* type */ "rtl pre", /* name */ OPTGROUP_NONE, /* optinfo_flags */ TV_PRE, /* tv_id */ PROP_cfglayout, /* properties_required */ 0, /* properties_provided */ 0, /* properties_destroyed */ 0, /* todo_flags_start */ TODO_df_finish, /* todo_flags_finish */ }; class pass_rtl_pre : public rtl_opt_pass { public: pass_rtl_pre (gcc::context *ctxt) : rtl_opt_pass (pass_data_rtl_pre, ctxt) {} /* opt_pass methods: */ virtual bool gate (function *); virtual unsigned int execute (function *) { return execute_rtl_pre (); } }; // class pass_rtl_pre /* We do not construct an accurate cfg in functions which call setjmp, so none of these passes runs if the function calls setjmp. FIXME: Should just handle setjmp via REG_SETJMP notes. */ bool pass_rtl_pre::gate (function *fun) { return optimize > 0 && flag_gcse && !fun->calls_setjmp && optimize_function_for_speed_p (fun) && dbg_cnt (pre); } } // anon namespace rtl_opt_pass * make_pass_rtl_pre (gcc::context *ctxt) { return new pass_rtl_pre (ctxt); } namespace { const pass_data pass_data_rtl_hoist = { RTL_PASS, /* type */ "hoist", /* name */ OPTGROUP_NONE, /* optinfo_flags */ TV_HOIST, /* tv_id */ PROP_cfglayout, /* properties_required */ 0, /* properties_provided */ 0, /* properties_destroyed */ 0, /* todo_flags_start */ TODO_df_finish, /* todo_flags_finish */ }; class pass_rtl_hoist : public rtl_opt_pass { public: pass_rtl_hoist (gcc::context *ctxt) : rtl_opt_pass (pass_data_rtl_hoist, ctxt) {} /* opt_pass methods: */ virtual bool gate (function *); virtual unsigned int execute (function *) { return execute_rtl_hoist (); } }; // class pass_rtl_hoist bool pass_rtl_hoist::gate (function *) { return optimize > 0 && flag_gcse && !cfun->calls_setjmp /* It does not make sense to run code hoisting unless we are optimizing for code size -- it rarely makes programs faster, and can make then bigger if we did PRE (when optimizing for space, we don't run PRE). */ && optimize_function_for_size_p (cfun) && dbg_cnt (hoist); } } // anon namespace rtl_opt_pass * make_pass_rtl_hoist (gcc::context *ctxt) { return new pass_rtl_hoist (ctxt); } /* Reset all state within gcse.c so that we can rerun the compiler within the same process. For use by toplev::finalize. */ void gcse_c_finalize (void) { test_insn = NULL; } #include "gt-gcse.h"