belady spiller, initial version

ir/be/Makefile.in

@@ -26,7 +26,7 @@ SOURCES += 	Makefile.in besched.h belistsched.h belistsched.c \
 	becopyoptmain.c becopyopt.c becopyheur.c \
 	becopyilp.c becopystat.c bearch_firm.c bearch.c bechordal_draw.c \
 	bechordal_draw.h beirgmod.c beirgmod.h benode.c benode_t.h \
-	bessadestr.c beifg.c bespillilp.c
+	bessadestr.c beifg.c bespillilp.c bespillbelady.c beuses.c
 
 include $(topdir)/MakeRules
 

ir/be/becopystat.h

@@ -7,7 +7,7 @@
 #ifndef _BECOPYSTAT_H
 #define _BECOPYSTAT_H
 
-#define DO_STAT
+#undef DO_STAT
 
 #ifdef DO_STAT
 
@@ -36,7 +36,7 @@ void copystat_dump_pretty(ir_graph *irg);
 
 #else /* DO_STAT */
 
-#define copy_copystat_init();
+#define copystat_init();
 #define	copystat_reset();
 #define copystat_collect_cls(env);
 #define copystat_add_max_costs(costs);

ir/be/bemain.c

@@ -38,14 +38,15 @@
 #include "benode_t.h"
 #include "beirgmod.h"
 #include "bespillilp.h"
+#include "bespillbelady.h"
 
 #include "beasm_dump_globals.h"
 #include "beasm_asm_gnu.h"
 
 #undef DUMP_BEGIN
 #undef DUMP_PREPARED
-#undef DUMP_SPILL
-#undef DUMP_SCHED
+#define DUMP_SCHED
+#define DUMP_SPILL
 #undef DUMP_ALLOCATED
 #undef DUMP_COPYMIN
 #undef DUMP_SSADESTR
@@ -177,7 +178,8 @@ static void be_main_loop(void)
 			DBG((env.dbg, LEVEL_1, "----> Reg class: %s\n", cls->name));
 
 			/* spilling */
-			be_spill_ilp(&session, cls);
+			//be_spill_ilp(&session, cls);
+			be_spill_belady(&session, cls);
 #ifdef DUMP_SPILL
 			dump_ir_block_graph_sched(session.irg, "-spill");
 #endif

ir/be/besched.c

@@ -11,6 +11,7 @@
 #include "iredges_t.h"
 #include "debug.h"
 
+#include "bearch.h"
 #include "besched_t.h"
 #include "beutil.h"
 #include "belistsched.h"
@@ -186,6 +187,15 @@ int sched_verify_irg(ir_graph *irg)
   return res;
 }
 
+int sched_skip_cf_predicator(const ir_node *irn, void *data) {
+  arch_env_t *ae = data;
+  return arch_irn_classify(ae, irn) == arch_irn_class_branch;
+}
+
+int sched_skip_phi_predicator(const ir_node *irn, void *data) {
+	return is_Phi(irn);
+}
+
 extern ir_node *sched_skip(ir_node *from, int forward,
     sched_predicator_t *predicator, void *data)
 {

ir/be/besched_t.h

@@ -255,10 +255,14 @@ extern int sched_verify_irg(ir_graph *irg);
  * A predicate for a node.
  * @param irn The node.
  * @param data The custom data.
- * @return 1 or 0, depending on your criteria.
+ * @return 1 if irn should be skipped. Else 0.
  */
 typedef int (sched_predicator_t)(const ir_node *irn, void *data);
 
+
+int sched_skip_cf_predicator(const ir_node *irn, void *data);
+int sched_skip_phi_predicator(const ir_node *irn, void *data);
+
 /**
  * Skip nodes in a schedule.
  * @param from The node to start from.

ir/be/bespillbelady.c

+/**
+ * Author:      Daniel Grund
+ * Date:		20.09.2005
+ * Copyright:   (c) Universitaet Karlsruhe
+ * Licence:     This file protected by GPL -  GNU GENERAL PUBLIC LICENSE.
+ */
+
+#include <alloca.h>
+#include "obst.h"
+#include "set.h"
+#include "pset.h"
+#include "irgraph.h"
+#include "irnode.h"
+#include "irmode.h"
+#include "ircons.h"
+#include "irgwalk.h"
+#include "iredges_t.h"
+
+#include "beutil.h"
+#include "bearch.h"
+#include "bespillbelady.h"
+#include "beuses_t.h"
+#include "besched_t.h"
+#include "beirgmod.h"
+#include "belive_t.h"
+#include "benode_t.h"
+
+#define MIN(a,b) (((a)<(b))?(a):(b))
+
+#define DBG_DECIDE 1
+#define DBG_WSETS 2
+#define DBG_FIX 4
+#define DBG_SPILL 8
+#define DBG_TRACE 32
+#define DEBUG_LVL (DBG_DECIDE | DBG_WSETS | DBG_FIX | DBG_SPILL)
+static firm_dbg_module_t *dbg = NULL;
+
+
+typedef struct _workset_t workset_t;
+typedef struct _block_info_t block_info_t;
+typedef struct _reloader_t reloader_t;
+typedef struct _spill_info_t spill_info_t;
+typedef struct _spill_ctx_t spill_ctx_t;
+typedef struct _belady_env_t belady_env_t;
+
+struct _workset_t {
+	belady_env_t *bel;
+	int i;				/**< used for iteration */
+	int len;			/**< current length */
+	loc_t vals[1];		/**< inlined array of the values/distances in this working set */
+};
+
+struct _block_info_t {
+	workset_t *ws_start, *ws_end;
+};
+
+struct _reloader_t {
+	reloader_t *next;
+	ir_node *reloader;
+};
+
+struct _spill_info_t {
+	ir_node *spilled_node;
+	reloader_t *reloaders;
+};
+
+struct _spill_ctx_t {
+	ir_node *spilled;  /**< The spilled node. */
+	ir_node *user;     /**< The node this spill is for. */
+	ir_node *spill;    /**< The spill itself. */
+};
+
+struct _belady_env_t {
+	struct obstack ob;
+	const be_main_session_env_t *session;
+	const be_node_factory_t *factory;
+	const arch_env_t *arch;
+	const arch_register_class_t *cls;
+	int n_regs;			/** number of regs in this reg-class */
+
+	workset_t *ws;		/**< the main workset used while processing a block. ob-allocated */
+	be_uses_t *uses;	/**< env for the next-use magic */
+	ir_node *instr;		/**< current instruction */
+	unsigned instr_nr;	/**< current instruction number (relative to block start) */
+	pset *used;			/**< holds the values used (so far) in the current BB */
+	set *spills;		/**< all spill_info_t's, which must be placed */
+
+	set *spill_ctxs;	/**< all spill contexts (multiple spilling and stop recursion) */
+	pset *mem_phis;		/**< all phis which must be converted to memory phis */
+	pset *reloads;		/**< all reload nodes placed */
+};
+
+static int set_cmp_spillinfo(const void *x, const void *y, size_t size) {
+	const spill_info_t *xx = x;
+	const spill_info_t *yy = y;
+	return ! (xx->spilled_node == yy->spilled_node);
+}
+
+static int set_cmp_spillctx(const void *a, const void *b, size_t n) {
+	const spill_ctx_t *p = a;
+	const spill_ctx_t *q = b;
+	return !(p->user == q->user && p->spilled == q->spilled);
+}
+
+/**
+ * Alloc a new workset on obstack @p ob with maximum size @p max
+ */
+static INLINE workset_t *new_workset(struct obstack *ob, belady_env_t *bel) {
+	workset_t *res;
+	size_t size = sizeof(*res) + (bel->n_regs-1)*sizeof(res->vals[0]);
+	res = obstack_alloc(ob, size);
+	memset(res, 0, size);
+	res->bel = bel;
+	return res;
+}
+
+static INLINE workset_t *workset_clone(struct obstack *ob, workset_t *ws) {
+	workset_t *res;
+	size_t size = sizeof(*res) + (ws->bel->n_regs-1)*sizeof(res->vals[0]);
+	res = obstack_alloc(ob, size);
+	memcpy(res, ws, size);
+	return res;
+}
+
+/**
+ * Inserts the value @p val into the workset, iff it is not
+ * already contained. The workset must not be full.
+ */
+static INLINE void workset_insert(workset_t *ws, ir_node *val) {
+	int i;
+	assert(ws->len < ws->bel->n_regs && "Workset already full!");
+	/* check for current regclass */
+	if (arch_get_irn_reg_class(ws->bel->arch, val, 0) != ws->bel->cls) {
+		DBG((dbg, 0, "Dropped %+F\n", val));
+		return;
+	}
+
+	/* check if val is already contained */
+	for(i=0; i<ws->len; ++i)
+		if (ws->vals[i].irn == val)
+			return;
+
+	/* insert val */
+	ws->vals[ws->len++].irn = val;
+}
+
+/**
+ * Inserts all values in array @p vals of length @p cnt
+ * into the workset. There must be enough space for the
+ * entries.
+ */
+static INLINE void workset_bulk_insert(workset_t *ws, int cnt, ir_node **vals) {
+	int i, o;
+	assert(ws->len + cnt <= ws->bel->n_regs && "Workset does not have enough room!");
+
+	for(o=0; o<cnt; ++o) {
+		ir_node *val = vals[o];
+		DBG((dbg, DBG_TRACE, "Bulk insert %+F\n", val));
+		/* check for current regclass */
+		if (arch_get_irn_reg_class(ws->bel->arch, val, 0) != ws->bel->cls) {
+			DBG((dbg, DBG_TRACE, "Wrong reg class\n"));
+			goto no_insert;
+		}
+
+		/* check if val is already contained */
+		for(i=0; i<ws->len; ++i)
+			if (ws->vals[i].irn == val) {
+				DBG((dbg, DBG_TRACE, "Already contained\n"));
+				goto no_insert;
+			}
+
+		/* insert val */
+		ws->vals[ws->len++].irn = val;
+		DBG((dbg, DBG_TRACE, "Inserted\n"));
+
+no_insert:
+		/*epsilon statement :)*/;
+	}
+}
+
+/**
+ * Overwrites the current contents of @p ws with the
+ * locations given in @p locs
+ */
+#define workset_bulk_fill(ws, count, locs) memcpy(&(ws)->vals[0], locs, ((ws)->len=count)*sizeof(locs[0]));
+
+/**
+ * Removes all entries from this workset
+ */
+#define workset_clear(ws) (ws)->len = 0;
+
+/**
+ * Removes the value @p val from the workset if present.
+ */
+static INLINE void workset_remove(workset_t *ws, ir_node *val) {
+	int i;
+	for(i=0; i<ws->len; ++i)
+		if (ws->vals[i].irn == val) {
+			ws->vals[i] = ws->vals[--ws->len];
+			return;
+		}
+}
+
+static INLINE int workset_contains(const workset_t *ws, ir_node *val) {
+	int i;
+	for(i=0; i<ws->len; ++i)
+		if (ws->vals[i].irn == val)
+			return 1;
+	return 0;
+}
+
+#define workset_foreach(ws, v)	for(ws->i=0; \
+									v=(ws->i < ws->len) ? ws->vals[ws->i].irn : NULL, ws->i < ws->len; \
+									ws->i++)
+
+#define workset_set_time(ws, i, t) (ws)->vals[i].time=t
+#define workset_set_length(ws, length) (ws)->len = length
+#define workset_get_length(ws) ((ws)->len)
+#define workset_get_val(ws, i) ((ws)->vals[i].irn)
+#define workset_sort(ws) qsort((ws)->vals, (ws)->len, sizeof((ws)->vals[0]), loc_compare);
+
+
+/**
+ * Collects all values live-in at block @p blk and all phi results in this block.
+ * Then it adds the best values (at most n_regs) to the ws.
+ */
+static void build_start_set(belady_env_t *bel, ir_node *blk) {
+	workset_t *ws = bel->ws;
+	ir_node *irn, *first;
+	irn_live_t *li;
+	int count;
+	loc_t loc, *starters;
+	struct obstack ob;
+
+	obstack_init(&ob);
+
+	/* get all values */
+	first = sched_first(blk);
+	count = 0;
+	sched_foreach(blk, irn)
+		if (is_Phi(irn) && arch_get_irn_reg_class(bel->arch, irn, 0) == bel->cls) {
+			loc.irn = irn;
+			loc.time = be_get_next_use(bel->uses, first, 0, irn);
+			obstack_grow(&ob, &loc, sizeof(loc));
+			count++;
+		} else
+			break;
+
+	live_foreach(blk, li)
+		if (live_is_in(li) && arch_get_irn_reg_class(bel->arch, li->irn, 0) == bel->cls) {
+			loc.irn = (ir_node *)li->irn;
+			loc.time = be_get_next_use(bel->uses, first, 0, li->irn);
+			obstack_grow(&ob, &loc, sizeof(loc));
+			count++;
+		}
+	starters = obstack_finish(&ob);
+
+	/* sort all values */
+	qsort(starters, count, sizeof(starters[0]), loc_compare);
+
+	/* copy the best ones to the ws */
+	count = MIN(count, ws->bel->n_regs);
+	workset_bulk_fill(ws, count, starters);
+
+	obstack_free(&ob, NULL);
+}
+
+/**
+ * Performs the actions neccessary to grant the request that:
+ * - new_vals can be held in registers
+ * - as few as possible other values are disposed
+ * - the worst values get disposed
+ *
+ * @p is_usage indicates that the values in new_vals are used (not defined)
+ * In this case reloads must be performed
+ */
+static void displace(belady_env_t *bel, workset_t *new_vals, int is_usage) {
+	ir_node *val;
+	int i, len, max_allowed, demand;
+	workset_t *ws = bel->ws;
+	ir_node **to_insert = alloca(bel->n_regs * sizeof(*to_insert));
+
+	/*
+	 * 1. Identify the number of needed slots and the values to reload
+	 */
+	demand = 0;
+	workset_foreach(new_vals, val) {
+		/* mark value as used */
+		if (is_usage)
+			pset_insert_ptr(bel->used, val);
+
+		if (!workset_contains(ws, val)) {
+			DBG((dbg, DBG_DECIDE, "    insert %+F\n", val));
+			to_insert[demand++] = val;
+
+			if (is_usage) {
+				spill_info_t si, *found;
+				reloader_t *rld;
+
+				/* find the spill info or create it */
+				si.spilled_node = val;
+				si.reloaders = NULL;
+				found = set_insert(bel->spills, &si, sizeof(si), HASH_PTR(si.spilled_node));
+
+				/* insert the reloader into the linked list */
+				rld = obstack_alloc(&bel->ob, sizeof(*rld));
+				rld->reloader = bel->instr;
+				rld->next = found->reloaders;
+				found->reloaders = rld;
+			}
+		} else
+			DBG((dbg, DBG_DECIDE, "    skip %+F\n", val));
+	}
+	DBG((dbg, DBG_DECIDE, "    demand = %d\n", demand));
+
+
+	/*
+	 * 2. Make room for at least 'demand' slots
+	 */
+	len = workset_get_length(ws);
+	max_allowed = bel->n_regs - demand;
+
+	/* Only make more free room if we do not have enough */
+	if (len > max_allowed) {
+		/* get current next-use distance */
+		for (i=0; i<ws->len; ++i)
+			workset_set_time(ws, i, be_get_next_use(bel->uses, bel->instr, bel->instr_nr, workset_get_val(ws, i)));
+
+		/* sort entries by increasing nextuse-distance*/
+		workset_sort(ws);
+
+		/* Logic for not needed live-ins: If a value is disposed
+		   before its first usage, remove it from start workset */
+		for (i=max_allowed; i<ws->len; ++i) {
+			ir_node *irn = ws->vals[i].irn;
+			if (!pset_find_ptr(bel->used, irn)) {
+				ir_node *curr_bb = get_nodes_block(bel->instr);
+				workset_t *ws_start = ((block_info_t *) get_irn_link(curr_bb))->ws_start;
+				workset_remove(ws_start, irn);
+
+				DBG((dbg, DBG_DECIDE, "    dispose %+F dumb\n", irn));
+			} else
+				DBG((dbg, DBG_DECIDE, "    dispose %+F\n", irn));
+		}
+
+		/* kill the last 'demand' entries in the array */
+		workset_set_length(ws, max_allowed);
+	}
+
+	/*
+	 * 3. Insert the new values into the workset
+	 */
+	workset_bulk_insert(bel->ws, demand, to_insert);
+}
+
+/**
+ * For the given block @p blk, decide for each values
+ * whether it is used from a register or is reloaded
+ * before the use.
+ */
+static void decide(ir_node *blk, void *env) {
+	belady_env_t *bel = env;
+	workset_t *new_vals;
+	ir_node *irn;
+	block_info_t *blk_info = obstack_alloc(&bel->ob, sizeof(*blk_info));
+	set_irn_link(blk, blk_info);
+
+	DBG((dbg, DBG_DECIDE, "\n"));
+	DBG((dbg, DBG_DECIDE, "Decide for %+F\n", blk));
+
+	/* build starting-workset for this block */
+	build_start_set(bel, blk);
+	blk_info->ws_start = workset_clone(&bel->ob, bel->ws);
+	DBG((dbg, DBG_WSETS, "Initial start workset for %+F:\n", blk));
+	workset_foreach(blk_info->ws_start, irn)
+		DBG((dbg, DBG_WSETS, "  %+F\n", irn));
+
+	/* process the block from start to end */
+	DBG((dbg, DBG_WSETS, "Processing...\n"));
+	bel->used = pset_new_ptr(32);
+	bel->instr_nr = 0;
+	new_vals = new_workset(&bel->ob, bel);
+	sched_foreach(blk, irn) {
+		DBG((dbg, DBG_DECIDE, "  %+F\n", irn));
+
+		/* projs are handled with the tuple value.
+		 * Phis are no real instr (see insert_starters)
+		 * instr_nr does not increase */
+		if (is_Proj(irn) || is_Phi(irn))
+			continue;
+
+		/* set instruction in the workset */
+		bel->instr = irn;
+
+		/* allocate all values _used_ by this instruction */
+		workset_clear(new_vals);
+		workset_bulk_insert(new_vals, get_irn_arity(irn)+1, get_irn_in(irn));
+		displace(bel, new_vals, 1);
+
+		/* allocate all values _defined_ by this instruction */
+		workset_clear(new_vals);
+		if (get_irn_mode(irn) == mode_T) { /* special handling for tuples and projs */
+			ir_node *proj;
+			for(proj=sched_next(irn); is_Proj(proj); proj=sched_next(proj))
+				workset_insert(new_vals, proj);
+		} else {
+			workset_insert(new_vals, irn);
+		}
+		displace(bel, new_vals, 0);
+
+		bel->instr_nr++;
+	}
+	del_pset(bel->used);
+
+	/* Remember end-workset for this block */
+	blk_info->ws_end = workset_clone(&bel->ob, bel->ws);
+	DBG((dbg, DBG_WSETS, "Start workset for %+F:\n", blk));
+	workset_foreach(blk_info->ws_start, irn)
+		DBG((dbg, DBG_WSETS, "  %+F\n", irn));
+	DBG((dbg, DBG_WSETS, "End workset for %+F:\n", blk));
+	workset_foreach(blk_info->ws_end, irn)
+		DBG((dbg, DBG_WSETS, "  %+F\n", irn));
+}
+
+/**
+ * 'decide' is block-local and makes assumtions
+ * about the set of live-ins. Thus we must adapt the
+ * live-outs to the live-ins at each block-border.
+ */
+static void fix_block_borders(ir_node *blk, void *env) {
+	belady_env_t *bel = env;
+	int i, max;
+
+	DBG((dbg, DBG_FIX, "\n"));
+	DBG((dbg, DBG_FIX, "Fixing %+F\n", blk));
+
+	workset_t *wsb = ((block_info_t *)get_irn_link(blk))->ws_start;
+
+	/* process all pred blocks */
+	for (i=0, max=get_irn_arity(blk); i<max; ++i) {
+		ir_node *irnb, *irnp, *pred = get_Block_cfgpred_block(blk, i);
+		workset_t *wsp = ((block_info_t *)get_irn_link(pred))->ws_end;
+
+		DBG((dbg, DBG_FIX, "  Pred %+F\n", pred));
+
+		workset_foreach(wsb, irnb) {
+			spill_info_t si, *found;
+			reloader_t *rld;
+
+			/* if irnb is a phi of the current block we reload
+			 * the corresponding argument, else irnb itself */
+			if(is_Phi(irnb) && blk == get_nodes_block(irnb))
+				irnb = get_irn_n(irnb, i);
+
+			/* check if irnb is in a register at end of pred */
+			workset_foreach(wsp, irnp)
+				if (irnb == irnp)
+					goto next_value;
+
+			/* irnb is in memory at the end of pred, so we have to reload it */
+
+			/* find the spill info or create it */
+			si.spilled_node = irnb;
+			si.reloaders = NULL;
+			found = set_insert(bel->spills, &si, sizeof(si), HASH_PTR(si.spilled_node));
+
+			/* insert the reloader into the linked list.
+			 * the schedule position depends on the cf-situation of the block */
+			rld = obstack_alloc(&bel->ob, sizeof(*rld));
+			rld->reloader = (max==1) ? sched_skip(sched_first(blk), 1, sched_skip_phi_predicator, NULL) : pred;
+			rld->next = found->reloaders;
+			found->reloaders = rld;
+
+			DBG((dbg, DBG_FIX, "    reload %+F before %+F\n", irnb, rld->reloader));
+
+next_value:
+			/*epsilon statement :)*/;
+		}
+	}
+}
+
+static INLINE spill_ctx_t *get_spill_ctx(set *sc, ir_node *to_spill, ir_node *ctx_irn) {
+	spill_ctx_t templ;
+
+	templ.spilled = to_spill;
+	templ.user    = ctx_irn;
+	templ.spill   = NULL;
+
+	return set_insert(sc, &templ, sizeof(templ), HASH_COMBINE(HASH_PTR(to_spill), HASH_PTR(ctx_irn)));
+}
+
+static INLINE ir_node *spill_irn(belady_env_t *bel, ir_node *irn, ir_node *ctx_irn) {
+	spill_ctx_t *ctx;
+	DBG((dbg, DBG_SPILL, "spill_irn %+F\n", irn));
+
+	ctx = get_spill_ctx(bel->spill_ctxs, irn, ctx_irn);
+	if(!ctx->spill)
+		ctx->spill = be_spill(bel->factory, bel->arch, irn);
+
+	return ctx->spill;
+}
+
+/**
+ * If the first usage of a phi result would be out of memory
+ * there is no sense in allocating a register for it.
+ * Thus we spill it and all its operands to the same spill slot.
+ * Therefore the phi/dataB becomes a phi/Memory
+ */
+static ir_node *spill_phi(belady_env_t *bel, ir_node *phi, ir_node *ctx_irn) {
+	int i, n = get_irn_arity(phi);
+	ir_node **ins, *bl = get_nodes_block(phi);
+	ir_graph *irg = get_irn_irg(bl);
+	spill_ctx_t *ctx;
+
+	assert(is_Phi(phi));
+	DBG((dbg, DBG_SPILL, "spill_phi %+F\n", phi));
+
+	/* search an existing spill for this context */
+	ctx = get_spill_ctx(bel->spill_ctxs, phi, ctx_irn);
+
+	/* if not found spill the phi */