Switch irg index to type size_t, making the API more consistent.
[libfirm] / ir / opt / ldstopt.c
index 9b6918d..01cf0fa 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 1995-2008 University of Karlsruhe.  All right reserved.
+ * Copyright (C) 1995-2011 University of Karlsruhe.  All right reserved.
  *
  * This file is part of libFirm.
  *
 #include "irphase_t.h"
 #include "irgopt.h"
 #include "set.h"
+#include "be.h"
 #include "debug.h"
 
 /** The debug handle. */
 DEBUG_ONLY(static firm_dbg_module_t *dbg;)
 
-#ifdef DO_CACHEOPT
-#include "cacheopt/cachesim.h"
-#endif
-
 #undef IMAX
 #define IMAX(a,b)   ((a) > (b) ? (a) : (b))
 
-#define MAX_PROJ    IMAX(IMAX(pn_Load_max, pn_Store_max), pn_Call_max)
+#define MAX_PROJ    IMAX(IMAX((long)pn_Load_max, (long)pn_Store_max), (long)pn_Call_max)
 
 enum changes_t {
        DF_CHANGED = 1,       /**< data flow changed */
@@ -110,7 +107,7 @@ static unsigned master_visited = 0;
  */
 static ldst_info_t *get_ldst_info(ir_node *node, struct obstack *obst)
 {
-       ldst_info_t *info = get_irn_link(node);
+       ldst_info_t *info = (ldst_info_t*)get_irn_link(node);
 
        if (! info) {
                info = OALLOCZ(obst, ldst_info_t);
@@ -124,7 +121,7 @@ static ldst_info_t *get_ldst_info(ir_node *node, struct obstack *obst)
  */
 static block_info_t *get_block_info(ir_node *node, struct obstack *obst)
 {
-       block_info_t *info = get_irn_link(node);
+       block_info_t *info = (block_info_t*)get_irn_link(node);
 
        if (! info) {
                info = OALLOCZ(obst, block_info_t);
@@ -179,10 +176,10 @@ static unsigned update_exc(ldst_info_t *info, ir_node *block, int pos)
  */
 static void collect_nodes(ir_node *node, void *env)
 {
-       ir_opcode   opcode = get_irn_opcode(node);
+       walk_env_t  *wenv   = (walk_env_t *)env;
+       unsigned     opcode = get_irn_opcode(node);
        ir_node     *pred, *blk, *pred_blk;
        ldst_info_t *ldst_info;
-       walk_env_t  *wenv = env;
 
        if (opcode == iro_Proj) {
                pred   = get_Proj_pred(node);
@@ -287,9 +284,9 @@ static ir_entity *find_constant_entity(ir_node *ptr)
                                        if (tlower == tarval_bad || tupper == tarval_bad)
                                                return NULL;
 
-                                       if (tarval_cmp(tv, tlower) & pn_Cmp_Lt)
+                                       if (tarval_cmp(tv, tlower) == ir_relation_less)
                                                return NULL;
-                                       if (tarval_cmp(tupper, tv) & pn_Cmp_Lt)
+                                       if (tarval_cmp(tupper, tv) == ir_relation_less)
                                                return NULL;
 
                                        /* ok, bounds check finished */
@@ -382,17 +379,19 @@ static compound_graph_path *rec_get_accessed_path(ir_node *ptr, int depth)
                        set_compound_graph_path_array_index(res, pos, get_Sel_array_index_long(ptr, 0));
                }
        } else if (is_Add(ptr)) {
-               ir_node   *l    = get_Add_left(ptr);
-               ir_node   *r    = get_Add_right(ptr);
-               ir_mode   *mode = get_irn_mode(ptr);
+               ir_mode   *mode;
                ir_tarval *tmp;
 
-               if (is_Const(r) && get_irn_mode(l) == mode) {
-                       ptr = l;
-                       tv  = get_Const_tarval(r);
-               } else {
-                       ptr = r;
-                       tv  = get_Const_tarval(l);
+               {
+                       ir_node   *l    = get_Add_left(ptr);
+                       ir_node   *r    = get_Add_right(ptr);
+                       if (is_Const(r) && get_irn_mode(l) == get_irn_mode(ptr)) {
+                               ptr = l;
+                               tv  = get_Const_tarval(r);
+                       } else {
+                               ptr = r;
+                               tv  = get_Const_tarval(l);
+                       }
                }
 ptr_arith:
                mode = get_tarval_mode(tv);
@@ -432,9 +431,9 @@ ptr_arith:
                        if (tlower == tarval_bad || tupper == tarval_bad)
                                return NULL;
 
-                       if (tarval_cmp(tv_index, tlower) & pn_Cmp_Lt)
+                       if (tarval_cmp(tv_index, tlower) == ir_relation_less)
                                return NULL;
-                       if (tarval_cmp(tupper, tv_index) & pn_Cmp_Lt)
+                       if (tarval_cmp(tupper, tv_index) == ir_relation_less)
                                return NULL;
 
                        /* ok, bounds check finished */
@@ -585,17 +584,19 @@ static ir_node *rec_find_compound_ent_value(ir_node *ptr, path_entry *next)
                }
                return rec_find_compound_ent_value(get_Sel_ptr(ptr), &entry);
        }  else if (is_Add(ptr)) {
-               ir_node  *l = get_Add_left(ptr);
-               ir_node  *r = get_Add_right(ptr);
                ir_mode  *mode;
                unsigned pos;
 
-               if (is_Const(r)) {
-                       ptr = l;
-                       tv  = get_Const_tarval(r);
-               } else {
-                       ptr = r;
-                       tv  = get_Const_tarval(l);
+               {
+                       ir_node *l = get_Add_left(ptr);
+                       ir_node *r = get_Add_right(ptr);
+                       if (is_Const(r)) {
+                               ptr = l;
+                               tv  = get_Const_tarval(r);
+                       } else {
+                               ptr = r;
+                               tv  = get_Const_tarval(l);
+                       }
                }
 ptr_arith:
                mode = get_tarval_mode(tv);
@@ -656,9 +657,9 @@ ptr_arith:
                        if (tlower == tarval_bad || tupper == tarval_bad)
                                return NULL;
 
-                       if (tarval_cmp(tv_index, tlower) & pn_Cmp_Lt)
+                       if (tarval_cmp(tv_index, tlower) == ir_relation_less)
                                return NULL;
-                       if (tarval_cmp(tupper, tv_index) & pn_Cmp_Lt)
+                       if (tarval_cmp(tupper, tv_index) == ir_relation_less)
                                return NULL;
 
                        /* ok, bounds check finished */
@@ -697,7 +698,7 @@ static void reduce_adr_usage(ir_node *ptr);
  */
 static void handle_load_update(ir_node *load)
 {
-       ldst_info_t *info = get_irn_link(load);
+       ldst_info_t *info = (ldst_info_t*)get_irn_link(load);
 
        /* do NOT touch volatile loads for now */
        if (get_Load_volatility(load) == volatility_is_volatile)
@@ -731,7 +732,7 @@ static void reduce_adr_usage(ir_node *ptr)
        /* this Proj is dead now */
        pred = get_Proj_pred(ptr);
        if (is_Load(pred)) {
-               ldst_info_t *info = get_irn_link(pred);
+               ldst_info_t *info = (ldst_info_t*)get_irn_link(pred);
                info->projs[get_Proj_proj(ptr)] = NULL;
 
                /* this node lost its result proj, handle that */
@@ -745,17 +746,26 @@ static void reduce_adr_usage(ir_node *ptr)
  */
 static int can_use_stored_value(ir_mode *old_mode, ir_mode *new_mode)
 {
+       unsigned old_size;
+       unsigned new_size;
        if (old_mode == new_mode)
-               return 1;
+               return true;
+
+       old_size = get_mode_size_bits(old_mode);
+       new_size = get_mode_size_bits(new_mode);
 
        /* if both modes are two-complement ones, we can always convert the
-          Stored value into the needed one. */
-       if (get_mode_size_bits(old_mode) >= get_mode_size_bits(new_mode) &&
+          Stored value into the needed one. (on big endian machines we currently
+          only support this for modes of same size) */
+       if (old_size >= new_size &&
                  get_mode_arithmetic(old_mode) == irma_twos_complement &&
-                 get_mode_arithmetic(new_mode) == irma_twos_complement)
-               return 1;
-       return 0;
-}  /* can_use_stored_value */
+                 get_mode_arithmetic(new_mode) == irma_twos_complement &&
+                 (!be_get_backend_param()->byte_order_big_endian
+               || old_size == new_size)) {
+               return true;
+       }
+       return false;
+}
 
 /**
  * Check whether a Call is at least pure, ie. does only read memory.
@@ -866,7 +876,10 @@ static int try_load_after_store(ir_node *load,
        store_value    = get_Store_value(store);
 
        if (delta != 0 || store_mode != load_mode) {
-               if (delta < 0 || delta + load_mode_len > store_mode_len)
+               /* TODO: implement for big-endian */
+               if (delta < 0 || delta + load_mode_len > store_mode_len
+                               || (be_get_backend_param()->byte_order_big_endian
+                                   && load_mode_len != store_mode_len))
                        return 0;
 
                if (get_mode_arithmetic(store_mode) != irma_twos_complement ||
@@ -879,7 +892,6 @@ static int try_load_after_store(ir_node *load,
                        ir_node *cnst;
                        ir_graph *irg = get_irn_irg(load);
 
-                       /* FIXME: only true for little endian */
                        cnst        = new_r_Const_long(irg, mode_Iu, delta * 8);
                        store_value = new_r_Shr(get_nodes_block(load),
                                                                        store_value, cnst, store_mode);
@@ -893,7 +905,7 @@ static int try_load_after_store(ir_node *load,
 
        DBG_OPT_RAW(load, store_value);
 
-       info = get_irn_link(load);
+       info = (ldst_info_t*)get_irn_link(load);
        if (info->projs[pn_Load_M])
                exchange(info->projs[pn_Load_M], get_Load_mem(load));
 
@@ -931,14 +943,14 @@ static int try_load_after_store(ir_node *load,
 static unsigned follow_Mem_chain(ir_node *load, ir_node *curr)
 {
        unsigned    res = 0;
-       ldst_info_t *info = get_irn_link(load);
+       ldst_info_t *info = (ldst_info_t*)get_irn_link(load);
        ir_node     *pred;
        ir_node     *ptr       = get_Load_ptr(load);
        ir_node     *mem       = get_Load_mem(load);
        ir_mode     *load_mode = get_Load_mode(load);
 
        for (pred = curr; load != pred; ) {
-               ldst_info_t *pred_info = get_irn_link(pred);
+               ldst_info_t *pred_info = (ldst_info_t*)get_irn_link(pred);
 
                /*
                 * a Load immediately after a Store -- a read after write.
@@ -975,6 +987,9 @@ static unsigned follow_Mem_chain(ir_node *load, ir_node *curr)
                         * Here, there is no need to check if the previous Load has an
                         * exception hander because they would have exact the same
                         * exception...
+                        *
+                        * TODO: implement load-after-load with different mode for big
+                        *       endian
                         */
                        if (info->projs[pn_Load_X_except] == NULL
                                        || get_nodes_block(load) == get_nodes_block(pred)) {
@@ -1094,7 +1109,7 @@ ir_node *can_replace_load_by_const(const ir_node *load, ir_node *c)
  */
 static unsigned optimize_load(ir_node *load)
 {
-       ldst_info_t *info = get_irn_link(load);
+       ldst_info_t *info = (ldst_info_t*)get_irn_link(load);
        ir_node     *mem, *ptr, *value;
        ir_entity   *ent;
        long        dummy;
@@ -1247,7 +1262,7 @@ static int is_partially_same(ir_node *small, ir_node *large)
 static unsigned follow_Mem_chain_for_Store(ir_node *store, ir_node *curr)
 {
        unsigned res = 0;
-       ldst_info_t *info = get_irn_link(store);
+       ldst_info_t *info = (ldst_info_t*)get_irn_link(store);
        ir_node *pred;
        ir_node *ptr = get_Store_ptr(store);
        ir_node *mem = get_Store_mem(store);
@@ -1256,7 +1271,7 @@ static unsigned follow_Mem_chain_for_Store(ir_node *store, ir_node *curr)
        ir_node *block = get_nodes_block(store);
 
        for (pred = curr; pred != store;) {
-               ldst_info_t *pred_info = get_irn_link(pred);
+               ldst_info_t *pred_info = (ldst_info_t*)get_irn_link(pred);
 
                /*
                 * BEWARE: one might think that checking the modes is useless, because
@@ -1420,7 +1435,7 @@ static unsigned optimize_store(ir_node *store)
 
        /* a store to an entity which is never read is unnecessary */
        if (entity != NULL && !(get_entity_usage(entity) & ir_usage_read)) {
-               ldst_info_t *info = get_irn_link(store);
+               ldst_info_t *info = (ldst_info_t*)get_irn_link(store);
                if (info->projs[pn_Store_X_except] == NULL) {
                        DB((dbg, LEVEL_1, "  Killing useless %+F to never read entity %+F\n", store, entity));
                        exchange(info->projs[pn_Store_M], get_Store_mem(store));
@@ -1500,7 +1515,7 @@ static unsigned optimize_phi(ir_node *phi, walk_env_t *wenv)
 
        /* check if the block is post dominated by Phi-block
           and has no exception exit */
-       bl_info = get_irn_link(block);
+       bl_info = (block_info_t*)get_irn_link(block);
        if (bl_info->flags & BLOCK_HAS_EXC)
                return 0;
 
@@ -1511,7 +1526,7 @@ static unsigned optimize_phi(ir_node *phi, walk_env_t *wenv)
        /* this is the address of the store */
        ptr  = get_Store_ptr(store);
        mode = get_irn_mode(get_Store_value(store));
-       info = get_irn_link(store);
+       info = (ldst_info_t*)get_irn_link(store);
        exc  = info->exc_block;
 
        for (i = 1; i < n; ++i) {
@@ -1527,7 +1542,7 @@ static unsigned optimize_phi(ir_node *phi, walk_env_t *wenv)
                if (ptr != get_Store_ptr(pred) || mode != get_irn_mode(get_Store_value(pred)))
                        return 0;
 
-               info = get_irn_link(pred);
+               info = (ldst_info_t*)get_irn_link(pred);
 
                /* check, if all stores have the same exception flow */
                if (exc != info->exc_block)
@@ -1542,7 +1557,7 @@ static unsigned optimize_phi(ir_node *phi, walk_env_t *wenv)
                   and has no exception exit. Note that block must be different from
                   Phi-block, else we would move a Store from end End of a block to its
                   Start... */
-               bl_info = get_irn_link(block);
+               bl_info = (block_info_t*)get_irn_link(block);
                if (bl_info->flags & BLOCK_HAS_EXC)
                        return 0;
                if (block == phi_block || ! block_postdominates(phi_block, block))
@@ -1582,7 +1597,7 @@ static unsigned optimize_phi(ir_node *phi, walk_env_t *wenv)
                assert(is_Proj(projMs[i]));
 
                store = get_Proj_pred(projMs[i]);
-               info  = get_irn_link(store);
+               info  = (ldst_info_t*)get_irn_link(store);
 
                inM[i] = get_Store_mem(store);
                inD[i] = get_Store_value(store);
@@ -1608,7 +1623,7 @@ static unsigned optimize_phi(ir_node *phi, walk_env_t *wenv)
        }
 
        /* fourth step: create the Store */
-       store = new_rd_Store(db, block, phiM, ptr, phiD, 0);
+       store = new_rd_Store(db, block, phiM, ptr, phiD, cons_none);
 #ifdef DO_CACHEOPT
        co_set_irn_name(store, co_get_irn_ident(old_store));
 #endif
@@ -1648,7 +1663,7 @@ static unsigned optimize_phi(ir_node *phi, walk_env_t *wenv)
  */
 static void do_load_store_optimize(ir_node *n, void *env)
 {
-       walk_env_t *wenv = env;
+       walk_env_t *wenv = (walk_env_t*)env;
 
        switch (get_irn_opcode(n)) {
 
@@ -1688,7 +1703,7 @@ typedef struct node_entry {
 typedef struct loop_env {
        ir_phase ph;           /**< the phase object */
        ir_node  **stack;      /**< the node stack */
-       int      tos;          /**< tos index */
+       size_t   tos;          /**< tos index */
        unsigned nextDFSnum;   /**< the current DFS number */
        unsigned POnum;        /**< current post order number */
 
@@ -1701,10 +1716,10 @@ typedef struct loop_env {
 static node_entry *get_irn_ne(ir_node *irn, loop_env *env)
 {
        ir_phase   *ph = &env->ph;
-       node_entry *e  = phase_get_irn_data(&env->ph, irn);
+       node_entry *e  = (node_entry*)phase_get_irn_data(&env->ph, irn);
 
        if (! e) {
-               e = phase_alloc(ph, sizeof(*e));
+               e = (node_entry*)phase_alloc(ph, sizeof(*e));
                memset(e, 0, sizeof(*e));
                phase_set_irn_data(ph, irn, e);
        }
@@ -1722,7 +1737,7 @@ static void push(loop_env *env, ir_node *n)
        node_entry *e;
 
        if (env->tos == ARR_LEN(env->stack)) {
-               int nlen = ARR_LEN(env->stack) * 2;
+               size_t nlen = ARR_LEN(env->stack) * 2;
                ARR_RESIZE(ir_node *, env->stack, nlen);
        }
        env->stack[env->tos++] = n;
@@ -1782,8 +1797,8 @@ typedef struct avail_entry_t {
  */
 static int cmp_avail_entry(const void *elt, const void *key, size_t size)
 {
-       const avail_entry_t *a = elt;
-       const avail_entry_t *b = key;
+       const avail_entry_t *a = (const avail_entry_t*)elt;
+       const avail_entry_t *b = (const avail_entry_t*)key;
        (void) size;
 
        return a->ptr != b->ptr || a->mode != b->mode;
@@ -1830,7 +1845,7 @@ static void move_loads_out_of_loops(scc *pscc, loop_env *env)
 
                        if (pe->pscc != ne->pscc) {
                                /* not in the same SCC, is region const */
-                               phi_entry *pe = phase_alloc(&env->ph, sizeof(*pe));
+                               phi_entry *pe = (phi_entry*)phase_alloc(&env->ph, sizeof(*pe));
 
                                pe->phi  = phi;
                                pe->pos  = j;
@@ -1852,7 +1867,7 @@ static void move_loads_out_of_loops(scc *pscc, loop_env *env)
                next = ne->next;
 
                if (is_Load(load)) {
-                       ldst_info_t *info = get_irn_link(load);
+                       ldst_info_t *info = (ldst_info_t*)get_irn_link(load);
                        ir_node     *ptr = get_Load_ptr(load);
 
                        /* for now, we cannot handle Loads with exceptions */
@@ -1898,11 +1913,11 @@ static void move_loads_out_of_loops(scc *pscc, loop_env *env)
 
                                        entry.ptr  = ptr;
                                        entry.mode = load_mode;
-                                       res = set_find(avail, &entry, sizeof(entry), hash_cache_entry(&entry));
+                                       res = (avail_entry_t*)set_find(avail, &entry, sizeof(entry), hash_cache_entry(&entry));
                                        if (res != NULL) {
                                                irn = res->load;
                                        } else {
-                                               irn = new_rd_Load(db, pred, get_Phi_pred(phi, pos), ptr, load_mode, 0);
+                                               irn = new_rd_Load(db, pred, get_Phi_pred(phi, pos), ptr, load_mode, cons_none);
                                                entry.load = irn;
                                                set_insert(avail, &entry, sizeof(entry), hash_cache_entry(&entry));
                                                DB((dbg, LEVEL_1, "  Created %+F in %+F\n", irn, pred));
@@ -2143,7 +2158,7 @@ static void dfs(ir_node *irn, loop_env *env)
        }
 
        if (node->low == node->DFSnum) {
-               scc *pscc = phase_alloc(&env->ph, sizeof(*pscc));
+               scc *pscc = (scc*)phase_alloc(&env->ph, sizeof(*pscc));
                ir_node *x;
 
                pscc->head = NULL;
@@ -2186,6 +2201,8 @@ static void do_dfs(ir_graph *irg, loop_env *env)
                        dfs(get_Raise_mem(pred), env);
                else if (is_fragile_op(pred))
                        dfs(get_fragile_op_mem(pred), env);
+               else if (is_Bad(pred))
+                       /* ignore non-optimized block predecessor */;
                else {
                        assert(0 && "Unknown EndBlock predecessor");
                }