X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;f=ir%2Fbe%2Fia32%2Fbearch_ia32.c;h=20e8a343dca7998fc276c8f3cf227cd10a332c1e;hb=c1290d60b0536494704dce14a479f26cbc075225;hp=d58728e5d761dda3aea2a0429e06b63b7a1b620b;hpb=3a9a24b6efa28c653f6e71564c402eda0e651e2e;p=libfirm

diff --git a/ir/be/ia32/bearch_ia32.c b/ir/be/ia32/bearch_ia32.c
index d58728e5d..20e8a343d 100644
--- a/ir/be/ia32/bearch_ia32.c
+++ b/ir/be/ia32/bearch_ia32.c
@@ -1,3 +1,9 @@
+/**
+ * This is the main ia32 firm backend driver.
+ *
+ * $Id$
+ */
+
 #ifdef HAVE_CONFIG_H
 #include "config.h"
 #endif
@@ -15,6 +21,7 @@
 #include "iredges_t.h"
 #include "ircons.h"
 #include "irgmod.h"
+#include "irgopt.h"
 
 #include "bitset.h"
 #include "debug.h"
@@ -67,6 +74,7 @@ static ir_node *my_skip_proj(const ir_node *n) {
 	return (ir_node *)n;
 }
 
+
 /**
  * Return register requirements for an ia32 node.
  * If the node returns a tuple (mode_T) then the proj's
@@ -75,12 +83,16 @@ static ir_node *my_skip_proj(const ir_node *n) {
 static const arch_register_req_t *ia32_get_irn_reg_req(const void *self, arch_register_req_t *req, const ir_node *irn, int pos) {
 	const ia32_register_req_t *irn_req;
 	long                       node_pos = pos == -1 ? 0 : pos;
-	ir_mode                   *mode     = get_irn_mode(irn);
+	ir_mode                   *mode     = is_Block(irn) ? NULL : get_irn_mode(irn);
 	firm_dbg_module_t         *mod      = firm_dbg_register(DEBUG_MODULE);
-	const ia32_irn_ops_t      *ops      = self;
 
-	if (mode == mode_T || mode == mode_M) {
-		DBG((mod, LEVEL_1, "ignoring mode_T, mode_M node %+F\n", irn));
+	if (is_Block(irn) || mode == mode_M || mode == mode_X) {
+		DBG((mod, LEVEL_1, "ignoring Block, mode_M, mode_X node %+F\n", irn));
+		return NULL;
+	}
+
+	if (mode == mode_T && pos < 0) {
+		DBG((mod, LEVEL_1, "ignoring request OUT requirements for node %+F\n", irn));
 		return NULL;
 	}
 
@@ -147,7 +159,14 @@ static const arch_register_req_t *ia32_get_irn_reg_req(const void *self, arch_re
 }
 
 static void ia32_set_irn_reg(const void *self, ir_node *irn, const arch_register_t *reg) {
-	int pos = 0;
+	int                   pos = 0;
+	const ia32_irn_ops_t *ops = self;
+
+	if (get_irn_mode(irn) == mode_X) {
+		return;
+	}
+
+	DBG((ops->cg->mod, LEVEL_1, "ia32 assigned register %s to node %+F\n", reg->name, irn));
 
 	if (is_Proj(irn)) {
 		pos = ia32_translate_proj_pos(irn);
@@ -170,6 +189,11 @@ static const arch_register_t *ia32_get_irn_reg(const void *self, const ir_node *
 	const arch_register_t *reg = NULL;
 
 	if (is_Proj(irn)) {
+
+		if (get_irn_mode(irn) == mode_X) {
+			return NULL;
+		}
+
 		pos = ia32_translate_proj_pos(irn);
 		irn = my_skip_proj(irn);
 	}
@@ -190,8 +214,6 @@ static arch_irn_class_t ia32_classify(const void *self, const ir_node *irn) {
 	irn = my_skip_proj(irn);
 	if (is_cfop(irn))
 		return arch_irn_class_branch;
-	else if (is_ia32_Call(irn))
-		return arch_irn_class_call;
 	else if (is_ia32_irn(irn))
 		return arch_irn_class_normal;
 	else
@@ -207,12 +229,154 @@ static arch_irn_flags_t ia32_get_flags(const void *self, const ir_node *irn) {
 	}
 }
 
+static entity *ia32_get_frame_entity(const void *self, const ir_node *irn) {
+	return is_ia32_irn(irn) ? get_ia32_frame_ent(irn) : NULL;
+}
+
 static void ia32_set_stack_bias(const void *self, ir_node *irn, int bias) {
-	if (get_ia32_use_frame(irn)) {
-		/* TODO: correct offset */
+	char buf[64];
+	const ia32_irn_ops_t *ops = self;
+
+	if (is_ia32_use_frame(irn) && bias != 0) {
+		ia32_am_flavour_t am_flav = get_ia32_am_flavour(irn);
+
+		DBG((ops->cg->mod, LEVEL_1, "stack biased %+F with %d\n", irn, bias));
+		snprintf(buf, sizeof(buf), "%d", bias);
+		add_ia32_am_offs(irn, buf);
+		am_flav |= ia32_O;
+		set_ia32_am_flavour(irn, am_flav);
+	}
+}
+
+typedef struct {
+	be_abi_call_flags_bits_t flags;
+	const arch_isa_t *isa;
+	ir_graph *irg;
+} ia32_abi_env_t;
+
+static void *ia32_abi_init(const be_abi_call_t *call, const arch_env_t *aenv, ir_graph *irg)
+{
+	ia32_abi_env_t *env    = xmalloc(sizeof(env[0]));
+	be_abi_call_flags_t fl = be_abi_call_get_flags(call);
+	env->flags = fl.bits;
+	env->irg   = irg;
+	env->isa   = aenv->isa;
+	return env;
+}
+
+static void ia32_abi_dont_save_regs(void *self, pset *s)
+{
+	ia32_abi_env_t *env = self;
+	if(env->flags.try_omit_fp)
+		pset_insert_ptr(s, env->isa->bp);
+}
+
+static const arch_register_t *ia32_abi_prologue(void *self, ir_node **mem, pmap *reg_map)
+{
+	ia32_abi_env_t *env              = self;
+	const arch_register_t *frame_reg = env->isa->sp;
+
+	if(!env->flags.try_omit_fp) {
+		int reg_size         = get_mode_size_bytes(env->isa->bp->reg_class->mode);
+		ir_node *bl          = get_irg_start_block(env->irg);
+		ir_node *curr_sp     = be_abi_reg_map_get(reg_map, env->isa->sp);
+		ir_node *curr_bp     = be_abi_reg_map_get(reg_map, env->isa->bp);
+		ir_node *curr_no_reg = be_abi_reg_map_get(reg_map, &ia32_gp_regs[REG_XXX]);
+		ir_node *store_bp;
+
+		curr_sp  = be_new_IncSP(env->isa->sp, env->irg, bl, curr_sp, *mem, reg_size, be_stack_dir_along);
+		store_bp = new_rd_ia32_Store(NULL, env->irg, bl, curr_sp, curr_no_reg, curr_bp, *mem, mode_T);
+		set_ia32_am_support(store_bp, ia32_am_Dest);
+		set_ia32_am_flavour(store_bp, ia32_B);
+		set_ia32_op_type(store_bp, ia32_AddrModeD);
+		*mem     = new_r_Proj(env->irg, bl, store_bp, mode_M, 0);
+		curr_bp  = be_new_Copy(env->isa->bp->reg_class, env->irg, bl, curr_sp);
+		be_set_constr_single_reg(curr_bp, BE_OUT_POS(0), env->isa->bp);
+		be_node_set_flags(curr_bp, BE_OUT_POS(0), arch_irn_flags_ignore);
+
+		be_abi_reg_map_set(reg_map, env->isa->sp, curr_sp);
+		be_abi_reg_map_set(reg_map, env->isa->bp, curr_bp);
+	}
+
+	return frame_reg;
+}
+
+static void ia32_abi_epilogue(void *self, ir_node *bl, ir_node **mem, pmap *reg_map)
+{
+	ia32_abi_env_t *env = self;
+	ir_node *curr_sp     = be_abi_reg_map_get(reg_map, env->isa->sp);
+	ir_node *curr_bp     = be_abi_reg_map_get(reg_map, env->isa->bp);
+	ir_node *curr_no_reg = be_abi_reg_map_get(reg_map, &ia32_gp_regs[REG_XXX]);
+
+	if(env->flags.try_omit_fp) {
+		curr_sp = be_new_IncSP(env->isa->sp, env->irg, bl, curr_sp, *mem, BE_STACK_FRAME_SIZE, be_stack_dir_against);
+	}
+
+	else {
+		ir_node *load_bp;
+		ir_mode *mode_bp = env->isa->bp->reg_class->mode;
+
+		curr_sp = be_new_SetSP(env->isa->sp, env->irg, bl, curr_sp, curr_bp, *mem);
+		load_bp = new_rd_ia32_Load(NULL, env->irg, bl, curr_sp, curr_no_reg, *mem, mode_T);
+		set_ia32_am_support(load_bp, ia32_am_Source);
+		set_ia32_am_flavour(load_bp, ia32_B);
+		set_ia32_op_type(load_bp, ia32_AddrModeS);
+		set_ia32_ls_mode(load_bp, mode_bp);
+		curr_bp = new_r_Proj(env->irg, bl, load_bp, mode_bp, 0);
+		*mem    = new_r_Proj(env->irg, bl, load_bp, mode_M, 1);
+	}
+
+	be_abi_reg_map_set(reg_map, env->isa->sp, curr_sp);
+	be_abi_reg_map_set(reg_map, env->isa->bp, curr_bp);
+}
+
+/**
+ * Produces the type which sits between the stack args and the locals on the stack.
+ * it will contain the return address and space to store the old base pointer.
+ * @return The Firm type modeling the ABI between type.
+ */
+static ir_type *ia32_abi_get_between_type(void *self)
+{
+	static ir_type *omit_fp_between_type = NULL;
+	static ir_type *between_type         = NULL;
+
+	ia32_abi_env_t *env = self;
+
+	if(!between_type) {
+		entity *old_bp_ent;
+		entity *ret_addr_ent;
+		entity *omit_fp_ret_addr_ent;
+
+		ir_type *old_bp_type   = new_type_primitive(new_id_from_str("bp"), mode_P);
+		ir_type *ret_addr_type = new_type_primitive(new_id_from_str("return_addr"), mode_P);
+
+		between_type           = new_type_class(new_id_from_str("ia32_between_type"));
+		old_bp_ent             = new_entity(between_type, new_id_from_str("old_bp"), old_bp_type);
+		ret_addr_ent           = new_entity(between_type, new_id_from_str("ret_addr"), ret_addr_type);
+
+		set_entity_offset_bytes(old_bp_ent, 0);
+		set_entity_offset_bytes(ret_addr_ent, get_type_size_bytes(old_bp_type));
+		set_type_size_bytes(between_type, get_type_size_bytes(old_bp_type) + get_type_size_bytes(ret_addr_type));
+
+		omit_fp_between_type   = new_type_class(new_id_from_str("ia32_between_type_omit_fp"));
+		omit_fp_ret_addr_ent   = new_entity(omit_fp_between_type, new_id_from_str("ret_addr"), ret_addr_type);
+
+		set_entity_offset_bytes(omit_fp_ret_addr_ent, 0);
+		set_type_size_bytes(omit_fp_between_type, get_type_size_bytes(ret_addr_type));
 	}
+
+	return env->flags.try_omit_fp ? omit_fp_between_type : between_type;
 }
 
+static const be_abi_callbacks_t ia32_abi_callbacks = {
+	ia32_abi_init,
+	free,
+	ia32_abi_get_between_type,
+	ia32_abi_dont_save_regs,
+	ia32_abi_prologue,
+	ia32_abi_epilogue,
+};
+
 /* fill register allocator interface */
 
 static const arch_irn_ops_if_t ia32_irn_ops_if = {
@@ -221,6 +385,7 @@ static const arch_irn_ops_if_t ia32_irn_ops_if = {
 	ia32_get_irn_reg,
 	ia32_classify,
 	ia32_get_flags,
+	ia32_get_frame_entity,
 	ia32_set_stack_bias
 };
 
@@ -248,22 +413,87 @@ ia32_irn_ops_t ia32_irn_ops = {
  */
 static void ia32_prepare_graph(void *self) {
 	ia32_code_gen_t *cg = self;
+	firm_dbg_module_t *old_mod = cg->mod;
+
+	cg->mod = firm_dbg_register("firm.be.ia32.transform");
+	irg_walk_blkwise_graph(cg->irg, ia32_place_consts_set_modes, ia32_transform_node, cg);
+	be_dump(cg->irg, "-transformed", dump_ir_block_graph_sched);
 
-	irg_walk_blkwise_graph(cg->irg, ia32_place_consts, ia32_transform_node, cg);
-	dump_ir_block_graph_sched(cg->irg, "-transformed");
 	edges_deactivate(cg->irg);
+	dead_node_elimination(cg->irg);
 	edges_activate(cg->irg);
-	irg_walk_blkwise_graph(cg->irg, NULL, ia32_optimize_am, cg);
-	dump_ir_block_graph_sched(cg->irg, "-am");
+
+	cg->mod = old_mod;
+
+	if (cg->opt.doam) {
+		irg_walk_blkwise_graph(cg->irg, NULL, ia32_optimize_am, cg);
+		be_dump(cg->irg, "-am", dump_ir_block_graph_sched);
+	}
 }
 
 
+/**
+ * Insert copies for all ia32 nodes where the should_be_same requirement
+ * is not fulfilled.
+ * Transform Sub into Neg -- Add if IN2 == OUT
+ */
+static void ia32_finish_irg_walker(ir_node *irn, void *env) {
+	ia32_code_gen_t            *cg = env;
+	const ia32_register_req_t **reqs;
+	const arch_register_t      *out_reg, *in_reg;
+	int                         n_res, i;
+	ir_node                    *copy, *in_node, *block;
+
+	if (! is_ia32_irn(irn))
+		return;
+
+	/* nodes with destination address mode don't produce values */
+	if (get_ia32_op_type(irn) == ia32_AddrModeD)
+		return;
+
+	reqs  = get_ia32_out_req_all(irn);
+	n_res = get_ia32_n_res(irn);
+	block = get_nodes_block(irn);
+
+	/* check all OUT requirements, if there is a should_be_same */
+	for (i = 0; i < n_res; i++) {
+		if (arch_register_req_is(&(reqs[i]->req), should_be_same)) {
+			/* get in and out register */
+			out_reg = get_ia32_out_reg(irn, i);
+			in_node = get_irn_n(irn, reqs[i]->same_pos);
+			in_reg  = arch_get_irn_register(cg->arch_env, in_node);
+
+			/* check if in and out register are equal */
+			if (arch_register_get_index(out_reg) != arch_register_get_index(in_reg)) {
+				DBG((cg->mod, LEVEL_1, "inserting copy for %+F in_pos %d\n", irn, reqs[i]->same_pos));
+
+				/* create copy from in register */
+				copy = be_new_Copy(arch_register_get_class(in_reg), cg->irg, block, in_node);
+
+				/* destination is the out register */
+				arch_set_irn_register(cg->arch_env, copy, out_reg);
+
+				/* insert copy before the node into the schedule */
+				sched_add_before(irn, copy);
+
+				/* set copy as in */
+				set_irn_n(irn, reqs[i]->same_pos, copy);
+			}
+		}
+	}
+
+	/* check if there is a sub which need to be transformed */
+	ia32_transform_sub_to_neg_add(irn, cg);
+
+	/* transform a LEA into an Add if possible */
+	ia32_transform_lea_to_add(irn, cg);
+}
 
 /**
- * Stack reservation and StackParam lowering.
+ * Add Copy nodes for not fulfilled should_be_equal constraints
  */
 static void ia32_finish_irg(ir_graph *irg, ia32_code_gen_t *cg) {
-
+	irg_walk_blkwise_graph(irg, NULL, ia32_finish_irg_walker, cg);
 }
 
 
@@ -278,89 +508,142 @@ static void ia32_before_ra(void *self) {
 }
 
 
+
 /**
- * Creates a Store for a Spill
+ * Transforms a be node into a Load.
  */
-static ir_node *ia32_lower_spill(void *self, ir_node *spill) {
-	ia32_code_gen_t *cg    = self;
-	ir_graph        *irg   = cg->irg;
-	dbg_info        *dbg   = get_irn_dbg_info(spill);
-	ir_node         *block = get_nodes_block(spill);
-	ir_node         *ptr   = get_irg_frame(irg);
-	ir_node         *val   = be_get_Spill_context(spill);
-	ir_node         *mem   = new_rd_NoMem(irg);
-	ir_mode         *mode  = get_irn_mode(spill);
-	entity          *ent   = be_get_spill_entity(spill);
-	unsigned         offs  = get_entity_offset_bytes(ent);
-	ir_node         *noreg, *res;
-	char             buf[64];
-
-	DB((cg->mod, LEVEL_1, "lower_spill: got offset %d for %+F\n", offs, ent));
+static void transform_to_Load(ia32_transform_env_t *env) {
+	ir_node *irn         = env->irn;
+	entity  *ent         = be_get_frame_entity(irn);
+	ir_mode *mode        = env->mode;
+	ir_node *noreg       = ia32_new_NoReg_gp(env->cg);
+	ir_node *nomem       = new_rd_NoMem(env->irg);
+	ir_node *sched_point = NULL;
+	ir_node *ptr         = get_irn_n(irn, 0);
+	ir_node *mem         = be_is_Reload(irn) ? get_irn_n(irn, 1) : nomem;
+	ir_node *new_op, *proj;
+	const arch_register_t *reg;
+
+	if (sched_is_scheduled(irn)) {
+		sched_point = sched_prev(irn);
+	}
 
 	if (mode_is_float(mode)) {
-		noreg = ia32_new_NoReg_fp(cg);
-		res   = new_rd_ia32_fStore(dbg, irg, block, ptr, noreg, val, mem, mode);
+		new_op = new_rd_ia32_fLoad(env->dbg, env->irg, env->block, ptr, noreg, mem, mode_T);
 	}
 	else {
-		noreg = ia32_new_NoReg_gp(cg);
-		res   = new_rd_ia32_Store(dbg, irg, block, ptr, noreg, val, mem, mode);
+		new_op = new_rd_ia32_Load(env->dbg, env->irg, env->block, ptr, noreg, mem, mode_T);
+	}
+
+	set_ia32_am_support(new_op, ia32_am_Source);
+	set_ia32_op_type(new_op, ia32_AddrModeS);
+	set_ia32_am_flavour(new_op, ia32_B);
+	set_ia32_ls_mode(new_op, mode);
+	set_ia32_frame_ent(new_op, ent);
+	set_ia32_use_frame(new_op);
+
+	proj = new_rd_Proj(env->dbg, env->irg, env->block, new_op, mode, pn_Load_res);
+
+	if (sched_point) {
+		sched_add_after(sched_point, new_op);
+		sched_add_after(new_op, proj);
+
+		sched_remove(irn);
 	}
 
-	snprintf(buf, sizeof(buf), "%d", offs);
-	add_ia32_am_offs(res, buf);
+	/* copy the register from the old node to the new Load */
+	reg = arch_get_irn_register(env->cg->arch_env, irn);
+	arch_set_irn_register(env->cg->arch_env, new_op, reg);
+
+	exchange(irn, proj);
 
-	return res;
 }
 
 /**
- * Create a Load for a Spill
+ * Transforms a be node into a Store.
  */
-static ir_node *ia32_lower_reload(void *self, ir_node *reload) {
-	ia32_code_gen_t *cg    = self;
-	ir_graph        *irg   = cg->irg;
-	dbg_info        *dbg   = get_irn_dbg_info(reload);
-	ir_node         *block = get_nodes_block(reload);
-	ir_node         *ptr   = get_irg_frame(irg);
-	ir_mode         *mode  = get_irn_mode(reload);
-	ir_node         *pred  = get_irn_n(reload, 0);
-	char             buf[64];
-	char            *ofs;
-	ir_node         *noreg, *res;
-
-	/* Get the offset to Load from. It can either be a Spill or a Store. */
-	if (be_is_Spill(pred)) {
-		entity   *ent  = be_get_spill_entity(pred);
-		unsigned  offs = get_entity_offset_bytes(ent);
-		DB((cg->mod, LEVEL_1, "lower_reload: got offset %d for %+F\n", offs, ent));
-
-		snprintf(buf, sizeof(buf), "%d", offs);
-	}
-	else if (is_ia32_Store(pred) || is_ia32_fStore(pred)) {
-		ofs = get_ia32_am_offs(pred);
-		strncpy(buf, ofs, sizeof(buf));
-		free(ofs);
-	}
-	else {
-		assert(0 && "unsupported Reload predecessor");
+static void transform_to_Store(ia32_transform_env_t *env) {
+	ir_node *irn   = env->irn;
+	entity  *ent   = be_get_frame_entity(irn);
+	ir_mode *mode  = env->mode;
+	ir_node *noreg = ia32_new_NoReg_gp(env->cg);
+	ir_node *nomem = new_rd_NoMem(env->irg);
+	ir_node *ptr   = get_irn_n(irn, 0);
+	ir_node *val   = get_irn_n(irn, 1);
+	ir_node *new_op, *proj;
+	ir_node *sched_point = NULL;
+
+	if (sched_is_scheduled(irn)) {
+		sched_point = sched_prev(irn);
 	}
 
-	/* Create the Load */
 	if (mode_is_float(mode)) {
-		noreg = ia32_new_NoReg_fp(cg);
-		res   = new_rd_ia32_fLoad(dbg, irg, block, ptr, noreg, pred, mode_T);
+		new_op = new_rd_ia32_fStore(env->dbg, env->irg, env->block, ptr, noreg, val, nomem, mode_T);
+	}
+	else if (get_mode_size_bits(mode) == 8) {
+		new_op = new_rd_ia32_Store8Bit(env->dbg, env->irg, env->block, ptr, noreg, val, nomem, mode_T);
 	}
 	else {
-		noreg = ia32_new_NoReg_gp(cg);
-		res   = new_rd_ia32_Load(dbg, irg, block, ptr, noreg, pred, mode_T);
+		new_op = new_rd_ia32_Store(env->dbg, env->irg, env->block, ptr, noreg, val, nomem, mode_T);
+	}
+
+	set_ia32_am_support(new_op, ia32_am_Dest);
+	set_ia32_op_type(new_op, ia32_AddrModeD);
+	set_ia32_am_flavour(new_op, ia32_B);
+	set_ia32_ls_mode(new_op, get_irn_mode(val));
+	set_ia32_frame_ent(new_op, ent);
+	set_ia32_use_frame(new_op);
+
+	proj = new_rd_Proj(env->dbg, env->irg, env->block, new_op, mode, 0);
+
+	if (sched_point) {
+		sched_add_after(sched_point, new_op);
+		sched_add_after(new_op, proj);
+
+		sched_remove(irn);
 	}
 
-	/* Set offset */
-	add_ia32_am_offs(res, buf);
+	exchange(irn, proj);
+
+}
+
+/**
+ * Calls the transform functions for StackParam, Spill and Reload.
+ */
+static void ia32_after_ra_walker(ir_node *node, void *env) {
+	ia32_code_gen_t *cg = env;
+	ia32_transform_env_t tenv;
+
+	if (is_Block(node))
+		return;
+
+	tenv.block = get_nodes_block(node);
+	tenv.dbg   = get_irn_dbg_info(node);
+	tenv.irg   = current_ir_graph;
+	tenv.irn   = node;
+	tenv.mod   = cg->mod;
+	tenv.mode  = get_irn_mode(node);
+	tenv.cg    = cg;
+
+	/* be_is_StackParam(node) || */
+	if (be_is_Reload(node)) {
+		transform_to_Load(&tenv);
+	}
+	else if (be_is_Spill(node)) {
+		transform_to_Store(&tenv);
+	}
+}
 
-	/* Return the result Proj */
-	return new_rd_Proj(dbg, irg, block, res, mode, 0);
+/**
+ * We transform StackParam, Spill and Reload here. This needs to be done before
+ * stack biasing otherwise we would miss the corrected offset for these nodes.
+ */
+static void ia32_after_ra(void *self) {
+	ia32_code_gen_t *cg = self;
+	irg_walk_blkwise_graph(cg->irg, NULL, ia32_after_ra_walker, self);
 }
 
+
 /**
  * Emits the code, closes the output file and frees
  * the code generator interface.
@@ -376,7 +659,7 @@ static void ia32_codegen(void *self) {
 	}
 
 	ia32_finish_irg(irg, cg);
-	//dump_ir_block_graph_sched(irg, "-finished");
+	be_dump(irg, "-finished", dump_ir_block_graph_sched);
 	ia32_gen_routine(out, irg, cg);
 
 	cur_reg_set = NULL;
@@ -393,11 +676,11 @@ static void *ia32_cg_init(FILE *F, const be_irg_t *birg);
 
 static const arch_code_generator_if_t ia32_code_gen_if = {
 	ia32_cg_init,
+	NULL,                /* before abi introduce hook */
 	ia32_prepare_graph,
 	ia32_before_sched,   /* before scheduling hook */
 	ia32_before_ra,      /* before register allocation hook */
-	ia32_lower_spill,
-	ia32_lower_reload,
+	ia32_after_ra,       /* after register allocation hook */
 	ia32_codegen         /* emit && done */
 };
 
@@ -418,6 +701,22 @@ static void *ia32_cg_init(FILE *F, const be_irg_t *birg) {
 	cg->tv_ent   = pmap_create();
 	cg->birg     = birg;
 
+	/* set optimizations */
+	cg->opt.incdec    = 0;
+	cg->opt.doam      = 1;
+	cg->opt.placecnst = 1;
+	cg->opt.immops    = 1;
+	cg->opt.extbb     = 1;
+
+#ifndef NDEBUG
+	if (isa->name_obst_size) {
+		//printf("freed %d bytes from name obst\n", isa->name_obst_size);
+		isa->name_obst_size = 0;
+		obstack_free(isa->name_obst, NULL);
+		obstack_init(isa->name_obst);
+	}
+#endif /* NDEBUG */
+
 	isa->num_codegens++;
 
 	if (isa->num_codegens > 1)
@@ -445,11 +744,17 @@ static void *ia32_cg_init(FILE *F, const be_irg_t *birg) {
  *****************************************************************/
 
 static ia32_isa_t ia32_isa_template = {
-	&ia32_isa_if,
-	&ia32_gp_regs[REG_ESP],
-	&ia32_gp_regs[REG_EBP],
-	-1,
-	0
+	&ia32_isa_if,            /* isa interface implementation */
+	&ia32_gp_regs[REG_ESP],  /* stack pointer register */
+	&ia32_gp_regs[REG_EBP],  /* base pointer register */
+	-1,                      /* stack direction */
+	0,                       /* number of code generator objects so far */
+	NULL,                    /* 16bit register names */
+	NULL,                    /* 8bit register names */
+#ifndef NDEBUG
+	NULL,                    /* name obstack */
+	0                        /* name obst size */
+#endif
 };
 
 /**
@@ -468,6 +773,18 @@ static void *ia32_init(void) {
 	ia32_register_init(isa);
 	ia32_create_opcodes();
 
+	isa->regs_16bit = pmap_create();
+	isa->regs_8bit  = pmap_create();
+
+	ia32_build_16bit_reg_map(isa->regs_16bit);
+	ia32_build_8bit_reg_map(isa->regs_8bit);
+
+#ifndef NDEBUG
+	isa->name_obst = xcalloc(1, sizeof(*(isa->name_obst)));
+	obstack_init(isa->name_obst);
+	isa->name_obst_size = 0;
+#endif /* NDEBUG */
+
 	inited = 1;
 
 	return isa;
@@ -479,6 +796,16 @@ static void *ia32_init(void) {
  * Closes the output file and frees the ISA structure.
  */
 static void ia32_done(void *self) {
+	ia32_isa_t *isa = self;
+
+	pmap_destroy(isa->regs_16bit);
+	pmap_destroy(isa->regs_8bit);
+
+#ifndef NDEBUG
+	//printf("name obst size = %d bytes\n", isa->name_obst_size);
+	obstack_free(isa->name_obst, NULL);
+#endif /* NDEBUG */
+
 	free(self);
 }
 
@@ -519,12 +846,20 @@ void ia32_get_call_abi(const void *self, ir_type *method_type, be_abi_call_t *ab
 	int       n         = get_method_n_params(method_type);
 	int       biggest_n = -1;
 	int       stack_idx = 0;
-	int       i, ignore;
+	int       i, ignore_1, ignore_2;
 	ir_mode **modes;
 	const arch_register_t *reg;
+	be_abi_call_flags_t call_flags;
+
+	/* set abi flags for calls */
+	call_flags.bits.left_to_right         = 0;
+	call_flags.bits.store_args_sequential = 0;
+	call_flags.bits.try_omit_fp           = 1;
+	call_flags.bits.fp_free               = 0;
+	call_flags.bits.call_has_imm          = 1;
 
 	/* set stack parameter passing style */
-	be_abi_call_set_flags(abi, BE_ABI_FRAME_POINTER_DEDICATED, 4);
+	be_abi_call_set_flags(abi, call_flags, &ia32_abi_callbacks);
 
 	/* collect the mode for each type */
 	modes = alloca(n * sizeof(modes[0]));
@@ -537,7 +872,7 @@ void ia32_get_call_abi(const void *self, ir_type *method_type, be_abi_call_t *ab
 	/* set register parameters  */
 	if (cc & cc_reg_param) {
 		/* determine the number of parameters passed via registers */
-		biggest_n = ia32_get_n_regparam_class(n, modes, &ignore, &ignore);
+		biggest_n = ia32_get_n_regparam_class(n, modes, &ignore_1, &ignore_2);
 
 		/* loop over all parameters and set the register requirements */
 		for (i = 0; i <= biggest_n; i++) {
@@ -552,7 +887,7 @@ void ia32_get_call_abi(const void *self, ir_type *method_type, be_abi_call_t *ab
 
 	/* set stack parameters */
 	for (i = stack_idx; i < n; i++) {
-		be_abi_call_param_stack(abi, i);
+		be_abi_call_param_stack(abi, i, 1);
 	}
 
 
@@ -578,14 +913,10 @@ void ia32_get_call_abi(const void *self, ir_type *method_type, be_abi_call_t *ab
 	}
 	else if (n == 1) {
 		tp   = get_method_res_type(method_type, 0);
+		assert(is_atomic_type(tp));
 		mode = get_type_mode(tp);
 
-		if (mode_is_float(mode)) {
-			be_abi_call_res_reg(abi, 1, &ia32_fp_regs[REG_XMM0]);
-		}
-		else {
-			be_abi_call_res_reg(abi, 1, &ia32_gp_regs[REG_EAX]);
-		}
+		be_abi_call_res_reg(abi, 0, mode_is_float(mode) ? &ia32_fp_regs[REG_XMM0] : &ia32_gp_regs[REG_EAX]);
 	}
 }
 
@@ -619,7 +950,7 @@ list_sched_selector_t ia32_sched_selector;
  * Returns the reg_pressure scheduler with to_appear_in_schedule() overloaded
  */
 static const list_sched_selector_t *ia32_get_list_sched_selector(const void *self) {
-	memcpy(&ia32_sched_selector, trivial_selector, sizeof(list_sched_selector_t));
+	memcpy(&ia32_sched_selector, reg_pressure_selector, sizeof(list_sched_selector_t));
 	ia32_sched_selector.to_appear_in_schedule = ia32_to_appear_in_schedule;
 	return &ia32_sched_selector;
 }