X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;f=ir%2Fbe%2Fia32%2Fbearch_ia32.c;h=507663014ca6d431ff6d36521f73702178b58bc3;hb=7a84c10295219975b2cd73b28a168b18e13894ec;hp=d002ed22bba030291613164e6495e7f6d3fa232b;hpb=d64ea79ebc6d41bc4253f6b6fd0a09582dce20eb;p=libfirm

diff --git a/ir/be/ia32/bearch_ia32.c b/ir/be/ia32/bearch_ia32.c
index d002ed22b..507663014 100644
--- a/ir/be/ia32/bearch_ia32.c
+++ b/ir/be/ia32/bearch_ia32.c
@@ -23,9 +23,7 @@
  * @author      Christian Wuerdig
  * @version     $Id$
  */
-#ifdef HAVE_CONFIG_H
 #include "config.h"
-#endif
 
 #include "lc_opts.h"
 #include "lc_opts_enum.h"
@@ -54,10 +52,10 @@
 #include "instrument.h"
 
 #include "../beabi.h"
-#include "../beirg_t.h"
-#include "../benode_t.h"
+#include "../beirg.h"
+#include "../benode.h"
 #include "../belower.h"
-#include "../besched_t.h"
+#include "../besched.h"
 #include "be.h"
 #include "../be_t.h"
 #include "../beirgmod.h"
@@ -71,6 +69,7 @@
 #include "../bestate.h"
 #include "../beflags.h"
 #include "../betranshlp.h"
+#include "../belistsched.h"
 
 #include "bearch_ia32_t.h"
 
@@ -91,37 +90,42 @@
 
 #ifdef FIRM_GRGEN_BE
 #include "ia32_pbqp_transform.h"
+
+transformer_t be_transformer = TRANSFORMER_DEFAULT;
 #endif
 
 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
 
-/* TODO: ugly */
-static set *cur_reg_set = NULL;
-
 ir_mode         *mode_fpcw       = NULL;
 ia32_code_gen_t *ia32_current_cg = NULL;
 
+/** The current omit-fp state */
+static unsigned ia32_curr_fp_ommitted  = 0;
+static ir_type *omit_fp_between_type   = NULL;
+static ir_type *between_type           = NULL;
+static ir_entity *old_bp_ent           = NULL;
+static ir_entity *ret_addr_ent         = NULL;
+static ir_entity *omit_fp_ret_addr_ent = NULL;
+
 /**
  * The environment for the intrinsic mapping.
  */
 static ia32_intrinsic_env_t intrinsic_env = {
 	NULL,    /* the isa */
 	NULL,    /* the irg, these entities belong to */
-	NULL,    /* entity for first div operand (move into FPU) */
-	NULL,    /* entity for second div operand (move into FPU) */
-	NULL,    /* entity for converts ll -> d */
-	NULL,    /* entity for converts d -> ll */
 	NULL,    /* entity for __divdi3 library call */
 	NULL,    /* entity for __moddi3 library call */
 	NULL,    /* entity for __udivdi3 library call */
 	NULL,    /* entity for __umoddi3 library call */
-	NULL,    /* bias value for conversion from float to unsigned 64 */
 };
 
 
-typedef ir_node *(*create_const_node_func) (dbg_info *dbg, ir_graph *irg, ir_node *block);
+typedef ir_node *(*create_const_node_func) (dbg_info *dbg, ir_node *block);
 
-static INLINE ir_node *create_const(ia32_code_gen_t *cg, ir_node **place,
+/**
+ * Used to create per-graph unique pseudo nodes.
+ */
+static inline ir_node *create_const(ia32_code_gen_t *cg, ir_node **place,
                                     create_const_node_func func,
                                     const arch_register_t* reg)
 {
@@ -131,49 +135,53 @@ static INLINE ir_node *create_const(ia32_code_gen_t *cg, ir_node **place,
 		return *place;
 
 	block = get_irg_start_block(cg->irg);
-	res = func(NULL, cg->irg, block);
-	arch_set_irn_register(cg->arch_env, res, reg);
+	res = func(NULL, block);
+	arch_set_irn_register(res, reg);
 	*place = res;
 
-	add_irn_dep(get_irg_end(cg->irg), res);
-	/* add_irn_dep(get_irg_start(cg->irg), res); */
-
 	return res;
 }
 
 /* Creates the unique per irg GP NoReg node. */
-ir_node *ia32_new_NoReg_gp(ia32_code_gen_t *cg) {
-	return create_const(cg, &cg->noreg_gp, new_rd_ia32_NoReg_GP,
+ir_node *ia32_new_NoReg_gp(ia32_code_gen_t *cg)
+{
+	return create_const(cg, &cg->noreg_gp, new_bd_ia32_NoReg_GP,
 	                    &ia32_gp_regs[REG_GP_NOREG]);
 }
 
-ir_node *ia32_new_NoReg_vfp(ia32_code_gen_t *cg) {
-	return create_const(cg, &cg->noreg_vfp, new_rd_ia32_NoReg_VFP,
+ir_node *ia32_new_NoReg_vfp(ia32_code_gen_t *cg)
+{
+	return create_const(cg, &cg->noreg_vfp, new_bd_ia32_NoReg_VFP,
 	                    &ia32_vfp_regs[REG_VFP_NOREG]);
 }
 
-ir_node *ia32_new_NoReg_xmm(ia32_code_gen_t *cg) {
-	return create_const(cg, &cg->noreg_xmm, new_rd_ia32_NoReg_XMM,
+ir_node *ia32_new_NoReg_xmm(ia32_code_gen_t *cg)
+{
+	return create_const(cg, &cg->noreg_xmm, new_bd_ia32_NoReg_XMM,
 	                    &ia32_xmm_regs[REG_XMM_NOREG]);
 }
 
-ir_node *ia32_new_Unknown_gp(ia32_code_gen_t *cg) {
-	return create_const(cg, &cg->unknown_gp, new_rd_ia32_Unknown_GP,
+ir_node *ia32_new_Unknown_gp(ia32_code_gen_t *cg)
+{
+	return create_const(cg, &cg->unknown_gp, new_bd_ia32_Unknown_GP,
 	                    &ia32_gp_regs[REG_GP_UKNWN]);
 }
 
-ir_node *ia32_new_Unknown_vfp(ia32_code_gen_t *cg) {
-	return create_const(cg, &cg->unknown_vfp, new_rd_ia32_Unknown_VFP,
+ir_node *ia32_new_Unknown_vfp(ia32_code_gen_t *cg)
+{
+	return create_const(cg, &cg->unknown_vfp, new_bd_ia32_Unknown_VFP,
 	                    &ia32_vfp_regs[REG_VFP_UKNWN]);
 }
 
-ir_node *ia32_new_Unknown_xmm(ia32_code_gen_t *cg) {
-	return create_const(cg, &cg->unknown_xmm, new_rd_ia32_Unknown_XMM,
+ir_node *ia32_new_Unknown_xmm(ia32_code_gen_t *cg)
+{
+	return create_const(cg, &cg->unknown_xmm, new_bd_ia32_Unknown_XMM,
 	                    &ia32_xmm_regs[REG_XMM_UKNWN]);
 }
 
-ir_node *ia32_new_Fpu_truncate(ia32_code_gen_t *cg) {
-	return create_const(cg, &cg->fpu_trunc_mode, new_rd_ia32_ChangeCW,
+ir_node *ia32_new_Fpu_truncate(ia32_code_gen_t *cg)
+{
+	return create_const(cg, &cg->fpu_trunc_mode, new_bd_ia32_ChangeCW,
                         &ia32_fp_cw_regs[REG_FPCW]);
 }
 
@@ -207,118 +215,19 @@ static ir_node *ia32_get_admissible_noreg(ia32_code_gen_t *cg, ir_node *irn, int
  *           |___/
  **************************************************/
 
-/**
- * Return register requirements for an ia32 node.
- * If the node returns a tuple (mode_T) then the proj's
- * will be asked for this information.
- */
-static const arch_register_req_t *ia32_get_irn_reg_req(const ir_node *node,
-													   int pos)
+static const arch_register_req_t *get_ia32_SwitchJmp_out_req(
+		const ir_node *node, int pos)
 {
-	ir_mode *mode = get_irn_mode(node);
-	long    node_pos;
-
-	if (mode == mode_X || is_Block(node)) {
-		return arch_no_register_req;
-	}
-
-	if (mode == mode_T && pos < 0) {
-		return arch_no_register_req;
-	}
-
-	node_pos = pos == -1 ? 0 : pos;
-	if (is_Proj(node)) {
-		if (mode == mode_M || pos >= 0) {
-			return arch_no_register_req;
-		}
-
-		node_pos = (pos == -1) ? get_Proj_proj(node) : pos;
-		node     = skip_Proj_const(node);
-	}
-
-	if (is_ia32_irn(node)) {
-		const arch_register_req_t *req;
-		if (pos >= 0)
-			req = get_ia32_in_req(node, pos);
-		else
-			req = get_ia32_out_req(node, node_pos);
-
-		assert(req != NULL);
-
-		return req;
-	}
-
-	/* unknowns should be transformed already */
-	assert(!is_Unknown(node));
+	(void) node;
+	(void) pos;
 	return arch_no_register_req;
 }
 
-static void ia32_set_irn_reg(ir_node *irn, const arch_register_t *reg)
+static arch_irn_class_t ia32_classify(const ir_node *irn)
 {
-	int    pos = 0;
-
-	if (get_irn_mode(irn) == mode_X) {
-		return;
-	}
-
-	if (is_Proj(irn)) {
-		pos = get_Proj_proj(irn);
-		irn = skip_Proj(irn);
-	}
-
-	if (is_ia32_irn(irn)) {
-		const arch_register_t **slots;
+	arch_irn_class_t classification = 0;
 
-		slots      = get_ia32_slots(irn);
-		slots[pos] = reg;
-	} else {
-		ia32_set_firm_reg(irn, reg, cur_reg_set);
-	}
-}
-
-static const arch_register_t *ia32_get_irn_reg(const ir_node *irn)
-{
-	int pos = 0;
-	const arch_register_t *reg = NULL;
-
-	if (is_Proj(irn)) {
-
-		if (get_irn_mode(irn) == mode_X) {
-			return NULL;
-		}
-
-		pos = get_Proj_proj(irn);
-		irn = skip_Proj_const(irn);
-	}
-
-	if (is_ia32_irn(irn)) {
-		const arch_register_t **slots;
-		slots = get_ia32_slots(irn);
-		assert(pos < get_ia32_n_res(irn));
-		reg   = slots[pos];
-	} else {
-		reg = ia32_get_firm_reg(irn, cur_reg_set);
-	}
-
-	return reg;
-}
-
-static arch_irn_class_t ia32_classify(const ir_node *irn) {
-	arch_irn_class_t classification = arch_irn_class_normal;
-
-	irn = skip_Proj_const(irn);
-
-	if (is_cfop(irn))
-		classification |= arch_irn_class_branch;
-
-	if (! is_ia32_irn(irn))
-		return classification & ~arch_irn_class_normal;
-
-	if (is_ia32_Ld(irn))
-		classification |= arch_irn_class_load;
-
-	if (is_ia32_St(irn))
-		classification |= arch_irn_class_store;
+	assert(is_ia32_irn(irn));
 
 	if (is_ia32_is_reload(irn))
 		classification |= arch_irn_class_reload;
@@ -332,29 +241,6 @@ static arch_irn_class_t ia32_classify(const ir_node *irn) {
 	return classification;
 }
 
-static arch_irn_flags_t ia32_get_flags(const ir_node *irn) {
-	arch_irn_flags_t flags = arch_irn_flags_none;
-
-	if (is_Unknown(irn))
-		return arch_irn_flags_ignore;
-
-	if(is_Proj(irn) && mode_is_datab(get_irn_mode(irn))) {
-		ir_node *pred = get_Proj_pred(irn);
-
-		if(is_ia32_irn(pred)) {
-			flags = get_ia32_out_flags(pred, get_Proj_proj(irn));
-		}
-
-		irn = pred;
-	}
-
-	if (is_ia32_irn(irn)) {
-		flags |= get_ia32_flags(irn);
-	}
-
-	return flags;
-}
-
 /**
  * The IA32 ABI callback object.
  */
@@ -364,11 +250,13 @@ typedef struct {
 	ir_graph *irg;                   /**< The associated graph. */
 } ia32_abi_env_t;
 
-static ir_entity *ia32_get_frame_entity(const ir_node *irn) {
+static ir_entity *ia32_get_frame_entity(const ir_node *irn)
+{
 	return is_ia32_irn(irn) ? get_ia32_frame_ent(irn) : NULL;
 }
 
-static void ia32_set_frame_entity(ir_node *irn, ir_entity *ent) {
+static void ia32_set_frame_entity(ir_node *irn, ir_entity *ent)
+{
 	set_ia32_frame_ent(irn, ent);
 }
 
@@ -404,19 +292,6 @@ static int ia32_get_sp_bias(const ir_node *node)
 	return 0;
 }
 
-/**
- * Put all registers which are saved by the prologue/epilogue in a set.
- *
- * @param self  The callback object.
- * @param s     The result set.
- */
-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->aenv->bp);
-}
-
 /**
  * Generate the routine prologue.
  *
@@ -435,40 +310,38 @@ static const arch_register_t *ia32_abi_prologue(void *self, ir_node **mem, pmap
 	ia32_code_gen_t  *cg       = ia32_current_cg;
 	const arch_env_t *arch_env = env->aenv;
 
+	ia32_curr_fp_ommitted = env->flags.try_omit_fp;
 	if (! env->flags.try_omit_fp) {
-		ir_graph *irg     =env->irg;
-		ir_node  *bl      = get_irg_start_block(irg);
+		ir_node  *bl      = get_irg_start_block(env->irg);
 		ir_node  *curr_sp = be_abi_reg_map_get(reg_map, arch_env->sp);
 		ir_node  *curr_bp = be_abi_reg_map_get(reg_map, arch_env->bp);
-		ir_node  *noreg = ia32_new_NoReg_gp(cg);
+		ir_node  *noreg   = ia32_new_NoReg_gp(cg);
 		ir_node  *push;
 
-		/* ALL nodes representing bp must be set to ignore. */
-		be_node_set_flags(get_Proj_pred(curr_bp), BE_OUT_POS(get_Proj_proj(curr_bp)), arch_irn_flags_ignore);
+		/* mark bp register as ignore */
+		be_set_constr_single_reg_out(get_Proj_pred(curr_bp),
+				get_Proj_proj(curr_bp), arch_env->bp, arch_register_req_type_ignore);
 
 		/* push ebp */
-		push    = new_rd_ia32_Push(NULL, irg, bl, noreg, noreg, *mem, curr_bp, curr_sp);
-		curr_sp = new_r_Proj(irg, bl, push, get_irn_mode(curr_sp), pn_ia32_Push_stack);
-		*mem    = new_r_Proj(irg, bl, push, mode_M, pn_ia32_Push_M);
+		push    = new_bd_ia32_Push(NULL, bl, noreg, noreg, *mem, curr_bp, curr_sp);
+		curr_sp = new_r_Proj(bl, push, get_irn_mode(curr_sp), pn_ia32_Push_stack);
+		*mem    = new_r_Proj(bl, push, mode_M, pn_ia32_Push_M);
 
 		/* the push must have SP out register */
-		arch_set_irn_register(arch_env, curr_sp, arch_env->sp);
-		set_ia32_flags(push, arch_irn_flags_ignore);
+		arch_set_irn_register(curr_sp, arch_env->sp);
 
 		/* this modifies the stack bias, because we pushed 32bit */
 		*stack_bias -= 4;
 
 		/* move esp to ebp */
-		curr_bp  = be_new_Copy(arch_env->bp->reg_class, irg, bl, curr_sp);
-		be_set_constr_single_reg(curr_bp, BE_OUT_POS(0), arch_env->bp);
-		arch_set_irn_register(arch_env, curr_bp, arch_env->bp);
-		be_node_set_flags(curr_bp, BE_OUT_POS(0), arch_irn_flags_ignore);
+		curr_bp = be_new_Copy(arch_env->bp->reg_class, bl, curr_sp);
+		be_set_constr_single_reg_out(curr_bp, 0, arch_env->bp,
+		                             arch_register_req_type_ignore);
 
 		/* beware: the copy must be done before any other sp use */
-		curr_sp = be_new_CopyKeep_single(arch_env->sp->reg_class, irg, bl, curr_sp, curr_bp, get_irn_mode(curr_sp));
-		be_set_constr_single_reg(curr_sp, BE_OUT_POS(0), arch_env->sp);
-		arch_set_irn_register(arch_env, curr_sp, arch_env->sp);
-		be_node_set_flags(curr_sp, BE_OUT_POS(0), arch_irn_flags_ignore);
+		curr_sp = be_new_CopyKeep_single(arch_env->sp->reg_class, bl, curr_sp, curr_bp, get_irn_mode(curr_sp));
+		be_set_constr_single_reg_out(curr_sp, 0, arch_env->sp,
+				                     arch_register_req_type_produces_sp);
 
 		be_abi_reg_map_set(reg_map, arch_env->sp, curr_sp);
 		be_abi_reg_map_set(reg_map, arch_env->bp, curr_bp);
@@ -495,11 +368,10 @@ static void ia32_abi_epilogue(void *self, ir_node *bl, ir_node **mem, pmap *reg_
 	const arch_env_t *arch_env = env->aenv;
 	ir_node          *curr_sp  = be_abi_reg_map_get(reg_map, arch_env->sp);
 	ir_node          *curr_bp  = be_abi_reg_map_get(reg_map, arch_env->bp);
-	ir_graph         *irg      = env->irg;
 
 	if (env->flags.try_omit_fp) {
 		/* simply remove the stack frame here */
-		curr_sp = be_new_IncSP(arch_env->sp, irg, bl, curr_sp, BE_STACK_FRAME_SIZE_SHRINK, 0);
+		curr_sp = be_new_IncSP(arch_env->sp, bl, curr_sp, BE_STACK_FRAME_SIZE_SHRINK, 0);
 	} else {
 		ir_mode *mode_bp = arch_env->bp->reg_class->mode;
 
@@ -507,10 +379,9 @@ static void ia32_abi_epilogue(void *self, ir_node *bl, ir_node **mem, pmap *reg_
 			ir_node *leave;
 
 			/* leave */
-			leave   = new_rd_ia32_Leave(NULL, irg, bl, curr_bp);
-			set_ia32_flags(leave, arch_irn_flags_ignore);
-			curr_bp = new_r_Proj(irg, bl, leave, mode_bp, pn_ia32_Leave_frame);
-			curr_sp = new_r_Proj(irg, bl, leave, get_irn_mode(curr_sp), pn_ia32_Leave_stack);
+			leave   = new_bd_ia32_Leave(NULL, bl, curr_bp);
+			curr_bp = new_r_Proj(bl, leave, mode_bp, pn_ia32_Leave_frame);
+			curr_sp = new_r_Proj(bl, leave, get_irn_mode(curr_sp), pn_ia32_Leave_stack);
 		} else {
 			ir_node *pop;
 
@@ -519,20 +390,20 @@ static void ia32_abi_epilogue(void *self, ir_node *bl, ir_node **mem, pmap *reg_
 			kill_node(curr_sp);
 
 			/* copy ebp to esp */
-			curr_sp = be_new_Copy(&ia32_reg_classes[CLASS_ia32_gp], irg, bl, curr_bp);
-			arch_set_irn_register(arch_env, curr_sp, arch_env->sp);
-			be_node_set_flags(curr_sp, BE_OUT_POS(0), arch_irn_flags_ignore);
+			curr_sp = be_new_Copy(&ia32_reg_classes[CLASS_ia32_gp], bl, curr_bp);
+			arch_set_irn_register(curr_sp, arch_env->sp);
+			be_set_constr_single_reg_out(curr_sp, 0, arch_env->sp,
+				                         arch_register_req_type_ignore);
 
 			/* pop ebp */
-			pop     = new_rd_ia32_Pop(NULL, env->irg, bl, *mem, curr_sp);
-			set_ia32_flags(pop, arch_irn_flags_ignore);
-			curr_bp = new_r_Proj(irg, bl, pop, mode_bp, pn_ia32_Pop_res);
-			curr_sp = new_r_Proj(irg, bl, pop, get_irn_mode(curr_sp), pn_ia32_Pop_stack);
+			pop     = new_bd_ia32_PopEbp(NULL, bl, *mem, curr_sp);
+			curr_bp = new_r_Proj(bl, pop, mode_bp, pn_ia32_Pop_res);
+			curr_sp = new_r_Proj(bl, pop, get_irn_mode(curr_sp), pn_ia32_Pop_stack);
 
-			*mem = new_r_Proj(irg, bl, pop, mode_M, pn_ia32_Pop_M);
+			*mem = new_r_Proj(bl, pop, mode_M, pn_ia32_Pop_M);
 		}
-		arch_set_irn_register(arch_env, curr_sp, arch_env->sp);
-		arch_set_irn_register(arch_env, curr_bp, arch_env->bp);
+		arch_set_irn_register(curr_sp, arch_env->sp);
+		arch_set_irn_register(curr_bp, arch_env->bp);
 	}
 
 	be_abi_reg_map_set(reg_map, arch_env->sp, curr_sp);
@@ -560,28 +431,18 @@ static void *ia32_abi_init(const be_abi_call_t *call, const arch_env_t *aenv, ir
  * Destroy the callback object.
  * @param self The callback object.
  */
-static void ia32_abi_done(void *self) {
+static void ia32_abi_done(void *self)
+{
 	free(self);
 }
 
 /**
- * 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.
+ * Build the between type and entities if not already build.
  */
-static ir_type *ia32_abi_get_between_type(void *self)
+static void ia32_build_between_type(void)
 {
 #define IDENT(s) new_id_from_chars(s, sizeof(s)-1)
-	static ir_type *omit_fp_between_type = NULL;
-	static ir_type *between_type         = NULL;
-
-	ia32_abi_env_t *env = self;
-
 	if (! between_type) {
-		ir_entity *old_bp_ent;
-		ir_entity *ret_addr_ent;
-		ir_entity *omit_fp_ret_addr_ent;
-
 		ir_type *old_bp_type   = new_type_primitive(IDENT("bp"), mode_Iu);
 		ir_type *ret_addr_type = new_type_primitive(IDENT("return_addr"), mode_Iu);
 
@@ -601,9 +462,38 @@ static ir_type *ia32_abi_get_between_type(void *self)
 		set_type_size_bytes(omit_fp_between_type, get_type_size_bytes(ret_addr_type));
 		set_type_state(omit_fp_between_type, layout_fixed);
 	}
+#undef IDENT
+}
 
+/**
+ * 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)
+{
+	ia32_abi_env_t *env = self;
+
+	ia32_build_between_type();
 	return env->flags.try_omit_fp ? omit_fp_between_type : between_type;
-#undef IDENT
+}
+
+/**
+ * Return the stack entity that contains the return address.
+ */
+ir_entity *ia32_get_return_address_entity(void)
+{
+	ia32_build_between_type();
+	return ia32_curr_fp_ommitted ? omit_fp_ret_addr_ent : ret_addr_ent;
+}
+
+/**
+ * Return the stack entity that contains the frame address.
+ */
+ir_entity *ia32_get_frame_address_entity(void)
+{
+	ia32_build_between_type();
+	return ia32_curr_fp_ommitted ? NULL : old_bp_ent;
 }
 
 /**
@@ -665,8 +555,8 @@ static int ia32_get_op_estimated_cost(const ir_node *irn)
  * @param obstack   The obstack to use for allocation of the returned nodes array
  * @return          The inverse operation or NULL if operation invertible
  */
-static arch_inverse_t *ia32_get_inverse(const ir_node *irn, int i, arch_inverse_t *inverse, struct obstack *obst) {
-	ir_graph *irg;
+static arch_inverse_t *ia32_get_inverse(const ir_node *irn, int i, arch_inverse_t *inverse, struct obstack *obst)
+{
 	ir_mode  *mode;
 	ir_mode  *irn_mode;
 	ir_node  *block, *noreg, *nomem;
@@ -689,12 +579,11 @@ static arch_inverse_t *ia32_get_inverse(const ir_node *irn, int i, arch_inverse_
 	           irn);
 	return NULL;
 
-	irg      = get_irn_irg(irn);
 	block    = get_nodes_block(irn);
 	mode     = get_irn_mode(irn);
 	irn_mode = get_irn_mode(irn);
 	noreg    = get_irn_n(irn, 0);
-	nomem    = new_r_NoMem(irg);
+	nomem    = new_NoMem();
 	dbg      = get_irn_dbg_info(irn);
 
 	/* initialize structure */
@@ -708,7 +597,7 @@ static arch_inverse_t *ia32_get_inverse(const ir_node *irn, int i, arch_inverse_
 			if (get_ia32_immop_type(irn) == ia32_ImmConst) {
 				/* we have an add with a const here */
 				/* invers == add with negated const */
-				inverse->nodes[0] = new_rd_ia32_Add(dbg, irg, block, noreg, noreg, nomem, get_irn_n(irn, i), noreg);
+				inverse->nodes[0] = new_bd_ia32_Add(dbg, block, noreg, noreg, nomem, get_irn_n(irn, i), noreg);
 				inverse->costs   += 1;
 				copy_ia32_Immop_attr(inverse->nodes[0], (ir_node *)irn);
 				set_ia32_Immop_tarval(inverse->nodes[0], tarval_neg(get_ia32_Immop_tarval(irn)));
@@ -717,13 +606,13 @@ static arch_inverse_t *ia32_get_inverse(const ir_node *irn, int i, arch_inverse_
 			else if (get_ia32_immop_type(irn) == ia32_ImmSymConst) {
 				/* we have an add with a symconst here */
 				/* invers == sub with const */
-				inverse->nodes[0] = new_rd_ia32_Sub(dbg, irg, block, noreg, noreg, nomem, get_irn_n(irn, i), noreg);
+				inverse->nodes[0] = new_bd_ia32_Sub(dbg, block, noreg, noreg, nomem, get_irn_n(irn, i), noreg);
 				inverse->costs   += 2;
 				copy_ia32_Immop_attr(inverse->nodes[0], (ir_node *)irn);
 			}
 			else {
 				/* normal add: inverse == sub */
-				inverse->nodes[0] = new_rd_ia32_Sub(dbg, irg, block, noreg, noreg, nomem, (ir_node*) irn, get_irn_n(irn, i ^ 1));
+				inverse->nodes[0] = new_bd_ia32_Sub(dbg, block, noreg, noreg, nomem, (ir_node*) irn, get_irn_n(irn, i ^ 1));
 				inverse->costs   += 2;
 			}
 #endif
@@ -733,17 +622,17 @@ static arch_inverse_t *ia32_get_inverse(const ir_node *irn, int i, arch_inverse_
 			if (get_ia32_immop_type(irn) != ia32_ImmNone) {
 				/* we have a sub with a const/symconst here */
 				/* invers == add with this const */
-				inverse->nodes[0] = new_rd_ia32_Add(dbg, irg, block, noreg, noreg, nomem, get_irn_n(irn, i), noreg);
+				inverse->nodes[0] = new_bd_ia32_Add(dbg, block, noreg, noreg, nomem, get_irn_n(irn, i), noreg);
 				inverse->costs   += (get_ia32_immop_type(irn) == ia32_ImmSymConst) ? 5 : 1;
 				copy_ia32_Immop_attr(inverse->nodes[0], (ir_node *)irn);
 			}
 			else {
 				/* normal sub */
 				if (i == n_ia32_binary_left) {
-					inverse->nodes[0] = new_rd_ia32_Add(dbg, irg, block, noreg, noreg, nomem, (ir_node*) irn, get_irn_n(irn, 3));
+					inverse->nodes[0] = new_bd_ia32_Add(dbg, block, noreg, noreg, nomem, (ir_node*) irn, get_irn_n(irn, 3));
 				}
 				else {
-					inverse->nodes[0] = new_rd_ia32_Sub(dbg, irg, block, noreg, noreg, nomem, get_irn_n(irn, n_ia32_binary_left), (ir_node*) irn);
+					inverse->nodes[0] = new_bd_ia32_Sub(dbg, block, noreg, noreg, nomem, get_irn_n(irn, n_ia32_binary_left), (ir_node*) irn);
 				}
 				inverse->costs += 1;
 			}
@@ -753,24 +642,24 @@ static arch_inverse_t *ia32_get_inverse(const ir_node *irn, int i, arch_inverse_
 #if 0
 			if (get_ia32_immop_type(irn) != ia32_ImmNone) {
 				/* xor with const: inverse = xor */
-				inverse->nodes[0] = new_rd_ia32_Xor(dbg, irg, block, noreg, noreg, nomem, get_irn_n(irn, i), noreg);
+				inverse->nodes[0] = new_bd_ia32_Xor(dbg, block, noreg, noreg, nomem, get_irn_n(irn, i), noreg);
 				inverse->costs   += (get_ia32_immop_type(irn) == ia32_ImmSymConst) ? 5 : 1;
 				copy_ia32_Immop_attr(inverse->nodes[0], (ir_node *)irn);
 			}
 			else {
 				/* normal xor */
-				inverse->nodes[0] = new_rd_ia32_Xor(dbg, irg, block, noreg, noreg, nomem, (ir_node *) irn, get_irn_n(irn, i));
+				inverse->nodes[0] = new_bd_ia32_Xor(dbg, block, noreg, noreg, nomem, (ir_node *) irn, get_irn_n(irn, i));
 				inverse->costs   += 1;
 			}
 #endif
 			break;
 		case iro_ia32_Not: {
-			inverse->nodes[0] = new_rd_ia32_Not(dbg, irg, block, (ir_node*) irn);
+			inverse->nodes[0] = new_bd_ia32_Not(dbg, block, (ir_node*) irn);
 			inverse->costs   += 1;
 			break;
 		}
 		case iro_ia32_Neg: {
-			inverse->nodes[0] = new_rd_ia32_Neg(dbg, irg, block, (ir_node*) irn);
+			inverse->nodes[0] = new_bd_ia32_Neg(dbg, block, (ir_node*) irn);
 			inverse->costs   += 1;
 			break;
 		}
@@ -909,19 +798,35 @@ static const be_abi_callbacks_t ia32_abi_callbacks = {
 	ia32_abi_init,
 	ia32_abi_done,
 	ia32_abi_get_between_type,
-	ia32_abi_dont_save_regs,
 	ia32_abi_prologue,
 	ia32_abi_epilogue
 };
 
-/* fill register allocator interface */
-
+/* register allocator interface */
 static const arch_irn_ops_t ia32_irn_ops = {
-	ia32_get_irn_reg_req,
-	ia32_set_irn_reg,
-	ia32_get_irn_reg,
+	get_ia32_in_req,
+	ia32_classify,
+	ia32_get_frame_entity,
+	ia32_set_frame_entity,
+	ia32_set_frame_offset,
+	ia32_get_sp_bias,
+	ia32_get_inverse,
+	ia32_get_op_estimated_cost,
+	ia32_possible_memory_operand,
+	ia32_perform_memory_operand,
+};
+
+/* special register allocator interface for SwitchJmp
+   as it possibly has a WIDE range of Proj numbers.
+   We don't want to allocate output for register constraints for
+   all these. */
+static const arch_irn_ops_t ia32_SwitchJmp_irn_ops = {
+	/* Note: we also use SwitchJmp_out_req for the inputs too:
+	   This is because the bearch API has a conceptual problem at the moment.
+	   Querying for negative proj numbers which can happen for switchs
+	   isn't possible and will result in inputs getting queried */
+	get_ia32_SwitchJmp_out_req,
 	ia32_classify,
-	ia32_get_flags,
 	ia32_get_frame_entity,
 	ia32_set_frame_entity,
 	ia32_set_frame_offset,
@@ -947,7 +852,8 @@ static ir_entity *mcount = NULL;
 
 #define ID(s) new_id_from_chars(s, sizeof(s) - 1)
 
-static void ia32_before_abi(void *self) {
+static void ia32_before_abi(void *self)
+{
 	lower_mode_b_config_t lower_mode_b_config = {
 		mode_Iu,  /* lowered mode */
 		mode_Bu,  /* preferred mode for set */
@@ -970,44 +876,30 @@ static void ia32_before_abi(void *self) {
 	}
 }
 
-transformer_t be_transformer = TRANSFORMER_DEFAULT;
-
 /**
  * Transforms the standard firm graph into
  * an ia32 firm graph
  */
-static void ia32_prepare_graph(void *self) {
+static void ia32_prepare_graph(void *self)
+{
 	ia32_code_gen_t *cg = self;
 
-	/* do local optimizations */
-	optimize_graph_df(cg->irg);
-
-	/* TODO: we often have dead code reachable through out-edges here. So for
-	 * now we rebuild edges (as we need correct user count for code selection)
-	 */
-#if 1
-	edges_deactivate(cg->irg);
-	edges_activate(cg->irg);
-#endif
-
-	if (cg->dump)
-		be_dump(cg->irg, "-pre_transform", dump_ir_block_graph_sched);
-
 	switch (be_transformer) {
-		case TRANSFORMER_DEFAULT:
-			/* transform remaining nodes into assembler instructions */
-			ia32_transform_graph(cg);
-			break;
+	case TRANSFORMER_DEFAULT:
+		/* transform remaining nodes into assembler instructions */
+		ia32_transform_graph(cg);
+		break;
 
 #ifdef FIRM_GRGEN_BE
-		case TRANSFORMER_PBQP:
-		case TRANSFORMER_RAND:
-			/* transform nodes into assembler instructions by PBQP magic */
-			ia32_transform_graph_by_pbqp(cg);
-			break;
+	case TRANSFORMER_PBQP:
+	case TRANSFORMER_RAND:
+		/* transform nodes into assembler instructions by PBQP magic */
+		ia32_transform_graph_by_pbqp(cg);
+		break;
 #endif
 
-		default: panic("invalid transformer");
+	default:
+		panic("invalid transformer");
 	}
 
 	/* do local optimizations (mainly CSE) */
@@ -1019,9 +911,6 @@ static void ia32_prepare_graph(void *self) {
 	/* optimize address mode */
 	ia32_optimize_graph(cg);
 
-	if (cg->dump)
-		be_dump(cg->irg, "-am", dump_ir_block_graph_sched);
-
 	/* do code placement, to optimize the position of constants */
 	place_code(cg->irg);
 
@@ -1029,16 +918,8 @@ static void ia32_prepare_graph(void *self) {
 		be_dump(cg->irg, "-place", dump_ir_block_graph_sched);
 }
 
-/**
- * Dummy functions for hooks we don't need but which must be filled.
- */
-static void ia32_before_sched(void *self) {
-	(void) self;
-}
-
 ir_node *turn_back_am(ir_node *node)
 {
-	ir_graph *irg   = current_ir_graph;
 	dbg_info *dbgi  = get_irn_dbg_info(node);
 	ir_node  *block = get_nodes_block(node);
 	ir_node  *base  = get_irn_n(node, n_ia32_base);
@@ -1046,8 +927,8 @@ ir_node *turn_back_am(ir_node *node)
 	ir_node  *mem   = get_irn_n(node, n_ia32_mem);
 	ir_node  *noreg;
 
-	ir_node  *load     = new_rd_ia32_Load(dbgi, irg, block, base, index, mem);
-	ir_node  *load_res = new_rd_Proj(dbgi, irg, block, load, mode_Iu, pn_ia32_Load_res);
+	ir_node  *load     = new_bd_ia32_Load(dbgi, block, base, index, mem);
+	ir_node  *load_res = new_rd_Proj(dbgi, block, load, mode_Iu, pn_ia32_Load_res);
 
 	ia32_copy_am_attrs(load, node);
 	if (is_ia32_is_reload(node))
@@ -1135,7 +1016,8 @@ static ir_node *flags_remat(ir_node *node, ir_node *after)
 /**
  * Called before the register allocator.
  */
-static void ia32_before_ra(void *self) {
+static void ia32_before_ra(void *self)
+{
 	ia32_code_gen_t *cg = self;
 
 	/* setup fpu rounding modes */
@@ -1152,7 +1034,8 @@ static void ia32_before_ra(void *self) {
 /**
  * Transforms a be_Reload into a ia32 Load.
  */
-static void transform_to_Load(ia32_code_gen_t *cg, ir_node *node) {
+static void transform_to_Load(ia32_code_gen_t *cg, ir_node *node)
+{
 	ir_graph *irg        = get_irn_irg(node);
 	dbg_info *dbg        = get_irn_dbg_info(node);
 	ir_node *block       = get_nodes_block(node);
@@ -1172,16 +1055,16 @@ static void transform_to_Load(ia32_code_gen_t *cg, ir_node *node) {
 
 	if (mode_is_float(spillmode)) {
 		if (ia32_cg_config.use_sse2)
-			new_op = new_rd_ia32_xLoad(dbg, irg, block, ptr, noreg, mem, spillmode);
+			new_op = new_bd_ia32_xLoad(dbg, block, ptr, noreg, mem, spillmode);
 		else
-			new_op = new_rd_ia32_vfld(dbg, irg, block, ptr, noreg, mem, spillmode);
+			new_op = new_bd_ia32_vfld(dbg, block, ptr, noreg, mem, spillmode);
 	}
 	else if (get_mode_size_bits(spillmode) == 128) {
 		/* Reload 128 bit SSE registers */
-		new_op = new_rd_ia32_xxLoad(dbg, irg, block, ptr, noreg, mem);
+		new_op = new_bd_ia32_xxLoad(dbg, block, ptr, noreg, mem);
 	}
 	else
-		new_op = new_rd_ia32_Load(dbg, irg, block, ptr, noreg, mem);
+		new_op = new_bd_ia32_Load(dbg, block, ptr, noreg, mem);
 
 	set_ia32_op_type(new_op, ia32_AddrModeS);
 	set_ia32_ls_mode(new_op, spillmode);
@@ -1191,7 +1074,7 @@ static void transform_to_Load(ia32_code_gen_t *cg, ir_node *node) {
 
 	DBG_OPT_RELOAD2LD(node, new_op);
 
-	proj = new_rd_Proj(dbg, irg, block, new_op, mode, pn_ia32_Load_res);
+	proj = new_rd_Proj(dbg, block, new_op, mode, pn_ia32_Load_res);
 
 	if (sched_point) {
 		sched_add_after(sched_point, new_op);
@@ -1199,10 +1082,10 @@ static void transform_to_Load(ia32_code_gen_t *cg, ir_node *node) {
 	}
 
 	/* copy the register from the old node to the new Load */
-	reg = arch_get_irn_register(cg->arch_env, node);
-	arch_set_irn_register(cg->arch_env, new_op, reg);
+	reg = arch_get_irn_register(node);
+	arch_set_irn_register(proj, reg);
 
-	SET_IA32_ORIG_NODE(new_op, ia32_get_old_node_name(cg, node));
+	SET_IA32_ORIG_NODE(new_op, node);
 
 	exchange(node, proj);
 }
@@ -1210,7 +1093,8 @@ static void transform_to_Load(ia32_code_gen_t *cg, ir_node *node) {
 /**
  * Transforms a be_Spill node into a ia32 Store.
  */
-static void transform_to_Store(ia32_code_gen_t *cg, ir_node *node) {
+static void transform_to_Store(ia32_code_gen_t *cg, ir_node *node)
+{
 	ir_graph *irg  = get_irn_irg(node);
 	dbg_info *dbg  = get_irn_dbg_info(node);
 	ir_node *block = get_nodes_block(node);
@@ -1218,7 +1102,7 @@ static void transform_to_Store(ia32_code_gen_t *cg, ir_node *node) {
 	const ir_node *spillval = get_irn_n(node, be_pos_Spill_val);
 	ir_mode *mode  = get_spill_mode(spillval);
 	ir_node *noreg = ia32_new_NoReg_gp(cg);
-	ir_node *nomem = new_rd_NoMem(irg);
+	ir_node *nomem = new_NoMem();
 	ir_node *ptr   = get_irg_frame(irg);
 	ir_node *val   = get_irn_n(node, be_pos_Spill_val);
 	ir_node *store;
@@ -1242,16 +1126,16 @@ static void transform_to_Store(ia32_code_gen_t *cg, ir_node *node) {
 
 	if (mode_is_float(mode)) {
 		if (ia32_cg_config.use_sse2)
-			store = new_rd_ia32_xStore(dbg, irg, block, ptr, noreg, nomem, val);
+			store = new_bd_ia32_xStore(dbg, block, ptr, noreg, nomem, val);
 		else
-			store = new_rd_ia32_vfst(dbg, irg, block, ptr, noreg, nomem, val, mode);
+			store = new_bd_ia32_vfst(dbg, block, ptr, noreg, nomem, val, mode);
 	} else if (get_mode_size_bits(mode) == 128) {
 		/* Spill 128 bit SSE registers */
-		store = new_rd_ia32_xxStore(dbg, irg, block, ptr, noreg, nomem, val);
+		store = new_bd_ia32_xxStore(dbg, block, ptr, noreg, nomem, val);
 	} else if (get_mode_size_bits(mode) == 8) {
-		store = new_rd_ia32_Store8Bit(dbg, irg, block, ptr, noreg, nomem, val);
+		store = new_bd_ia32_Store8Bit(dbg, block, ptr, noreg, nomem, val);
 	} else {
-		store = new_rd_ia32_Store(dbg, irg, block, ptr, noreg, nomem, val);
+		store = new_bd_ia32_Store(dbg, block, ptr, noreg, nomem, val);
 	}
 
 	set_ia32_op_type(store, ia32_AddrModeD);
@@ -1259,7 +1143,7 @@ static void transform_to_Store(ia32_code_gen_t *cg, ir_node *node) {
 	set_ia32_frame_ent(store, ent);
 	set_ia32_use_frame(store);
 	set_ia32_is_spill(store);
-	SET_IA32_ORIG_NODE(store, ia32_get_old_node_name(cg, node));
+	SET_IA32_ORIG_NODE(store, node);
 	DBG_OPT_SPILL2ST(node, store);
 
 	if (sched_point) {
@@ -1270,14 +1154,15 @@ static void transform_to_Store(ia32_code_gen_t *cg, ir_node *node) {
 	exchange(node, store);
 }
 
-static ir_node *create_push(ia32_code_gen_t *cg, ir_node *node, ir_node *schedpoint, ir_node *sp, ir_node *mem, ir_entity *ent) {
-	ir_graph *irg = get_irn_irg(node);
+static ir_node *create_push(ia32_code_gen_t *cg, ir_node *node, ir_node *schedpoint, ir_node *sp, ir_node *mem, ir_entity *ent)
+{
 	dbg_info *dbg = get_irn_dbg_info(node);
 	ir_node *block = get_nodes_block(node);
 	ir_node *noreg = ia32_new_NoReg_gp(cg);
+	ir_graph *irg = get_irn_irg(node);
 	ir_node *frame = get_irg_frame(irg);
 
-	ir_node *push = new_rd_ia32_Push(dbg, irg, block, frame, noreg, mem, noreg, sp);
+	ir_node *push = new_bd_ia32_Push(dbg, block, frame, noreg, mem, noreg, sp);
 
 	set_ia32_frame_ent(push, ent);
 	set_ia32_use_frame(push);
@@ -1289,14 +1174,15 @@ static ir_node *create_push(ia32_code_gen_t *cg, ir_node *node, ir_node *schedpo
 	return push;
 }
 
-static ir_node *create_pop(ia32_code_gen_t *cg, ir_node *node, ir_node *schedpoint, ir_node *sp, ir_entity *ent) {
-	ir_graph *irg = get_irn_irg(node);
+static ir_node *create_pop(ia32_code_gen_t *cg, ir_node *node, ir_node *schedpoint, ir_node *sp, ir_entity *ent)
+{
 	dbg_info *dbg = get_irn_dbg_info(node);
 	ir_node *block = get_nodes_block(node);
 	ir_node *noreg = ia32_new_NoReg_gp(cg);
+	ir_graph *irg  = get_irn_irg(node);
 	ir_node *frame = get_irg_frame(irg);
 
-	ir_node *pop = new_rd_ia32_PopMem(dbg, irg, block, frame, noreg, new_NoMem(), sp);
+	ir_node *pop = new_bd_ia32_PopMem(dbg, block, frame, noreg, new_NoMem(), sp);
 
 	set_ia32_frame_ent(pop, ent);
 	set_ia32_use_frame(pop);
@@ -1309,16 +1195,16 @@ static ir_node *create_pop(ia32_code_gen_t *cg, ir_node *node, ir_node *schedpoi
 	return pop;
 }
 
-static ir_node* create_spproj(ia32_code_gen_t *cg, ir_node *node, ir_node *pred, int pos) {
-	ir_graph *irg = get_irn_irg(node);
+static ir_node* create_spproj(ir_node *node, ir_node *pred, int pos)
+{
 	dbg_info *dbg = get_irn_dbg_info(node);
 	ir_node *block = get_nodes_block(node);
 	ir_mode *spmode = mode_Iu;
 	const arch_register_t *spreg = &ia32_gp_regs[REG_ESP];
 	ir_node *sp;
 
-	sp = new_rd_Proj(dbg, irg, block, pred, spmode, pos);
-	arch_set_irn_register(cg->arch_env, sp, spreg);
+	sp = new_rd_Proj(dbg, block, pred, spmode, pos);
+	arch_set_irn_register(sp, spreg);
 
 	return sp;
 }
@@ -1328,19 +1214,17 @@ static ir_node* create_spproj(ia32_code_gen_t *cg, ir_node *node, ir_node *pred,
  * push/pop into/from memory cascades. This is possible without using
  * any registers.
  */
-static void transform_MemPerm(ia32_code_gen_t *cg, ir_node *node) {
-	ir_graph *irg = get_irn_irg(node);
-	ir_node *block = get_nodes_block(node);
-	ir_node *in[1];
-	ir_node *keep;
-	int i, arity;
-	ir_node *sp = be_abi_get_ignore_irn(cg->birg->abi, &ia32_gp_regs[REG_ESP]);
+static void transform_MemPerm(ia32_code_gen_t *cg, ir_node *node)
+{
+	ir_node         *block = get_nodes_block(node);
+	ir_node         *sp    = be_abi_get_ignore_irn(cg->birg->abi, &ia32_gp_regs[REG_ESP]);
+	int              arity = be_get_MemPerm_entity_arity(node);
+	ir_node        **pops  = ALLOCAN(ir_node*, arity);
+	ir_node         *in[1];
+	ir_node         *keep;
+	int              i;
 	const ir_edge_t *edge;
 	const ir_edge_t *next;
-	ir_node **pops;
-
-	arity = be_get_MemPerm_entity_arity(node);
-	pops = alloca(arity * sizeof(pops[0]));
 
 	/* create Pushs */
 	for(i = 0; i < arity; ++i) {
@@ -1358,12 +1242,12 @@ static void transform_MemPerm(ia32_code_gen_t *cg, ir_node *node) {
 		assert( (entsize == 4 || entsize == 8) && "spillslot on x86 should be 32 or 64 bit");
 
 		push = create_push(cg, node, node, sp, mem, inent);
-		sp = create_spproj(cg, node, push, pn_ia32_Push_stack);
+		sp = create_spproj(node, push, pn_ia32_Push_stack);
 		if(entsize == 8) {
 			/* add another push after the first one */
 			push = create_push(cg, node, node, sp, mem, inent);
 			add_ia32_am_offs_int(push, 4);
-			sp = create_spproj(cg, node, push, pn_ia32_Push_stack);
+			sp = create_spproj(node, push, pn_ia32_Push_stack);
 		}
 
 		set_irn_n(node, i, new_Bad());
@@ -1384,20 +1268,20 @@ static void transform_MemPerm(ia32_code_gen_t *cg, ir_node *node) {
 		assert( (entsize == 4 || entsize == 8) && "spillslot on x86 should be 32 or 64 bit");
 
 		pop = create_pop(cg, node, node, sp, outent);
-		sp = create_spproj(cg, node, pop, pn_ia32_Pop_stack);
+		sp = create_spproj(node, pop, pn_ia32_Pop_stack);
 		if(entsize == 8) {
 			add_ia32_am_offs_int(pop, 4);
 
 			/* add another pop after the first one */
 			pop = create_pop(cg, node, node, sp, outent);
-			sp = create_spproj(cg, node, pop, pn_ia32_Pop_stack);
+			sp = create_spproj(node, pop, pn_ia32_Pop_stack);
 		}
 
 		pops[i] = pop;
 	}
 
 	in[0] = sp;
-	keep  = be_new_Keep(&ia32_reg_classes[CLASS_ia32_gp], irg, block, 1, in);
+	keep  = be_new_Keep(block, 1, in);
 	sched_add_before(node, keep);
 
 	/* exchange memprojs */
@@ -1422,7 +1306,8 @@ static void transform_MemPerm(ia32_code_gen_t *cg, ir_node *node) {
 /**
  * Block-Walker: Calls the transform functions Spill and Reload.
  */
-static void ia32_after_ra_walker(ir_node *block, void *env) {
+static void ia32_after_ra_walker(ir_node *block, void *env)
+{
 	ir_node *node, *prev;
 	ia32_code_gen_t *cg = env;
 
@@ -1445,59 +1330,81 @@ static void ia32_after_ra_walker(ir_node *block, void *env) {
  */
 static void ia32_collect_frame_entity_nodes(ir_node *node, void *data)
 {
-	be_fec_env_t *env = data;
+	be_fec_env_t  *env = data;
+	const ir_mode *mode;
+	int            align;
 
 	if (be_is_Reload(node) && be_get_frame_entity(node) == NULL) {
-		const ir_mode *mode = get_spill_mode_mode(get_irn_mode(node));
-		int align = get_mode_size_bytes(mode);
-		be_node_needs_frame_entity(env, node, mode, align);
-	} else if(is_ia32_irn(node) && get_ia32_frame_ent(node) == NULL
-	          && is_ia32_use_frame(node)) {
-		if (is_ia32_need_stackent(node) || is_ia32_Load(node)) {
-			const ir_mode     *mode  = get_ia32_ls_mode(node);
-			const ia32_attr_t *attr  = get_ia32_attr_const(node);
-			int                align;
-
-			if (is_ia32_is_reload(node)) {
-				mode = get_spill_mode_mode(mode);
+		mode  = get_spill_mode_mode(get_irn_mode(node));
+		align = get_mode_size_bytes(mode);
+	} else if (is_ia32_irn(node)         &&
+			get_ia32_frame_ent(node) == NULL &&
+			is_ia32_use_frame(node)) {
+		if (is_ia32_need_stackent(node))
+			goto need_stackent;
+
+		switch (get_ia32_irn_opcode(node)) {
+need_stackent:
+			case iro_ia32_Load: {
+				const ia32_attr_t *attr = get_ia32_attr_const(node);
+
+				if (attr->data.need_32bit_stackent) {
+					mode = mode_Is;
+				} else if (attr->data.need_64bit_stackent) {
+					mode = mode_Ls;
+				} else {
+					mode = get_ia32_ls_mode(node);
+					if (is_ia32_is_reload(node))
+						mode = get_spill_mode_mode(mode);
+				}
+				align = get_mode_size_bytes(mode);
+				break;
 			}
 
-			if(attr->data.need_64bit_stackent) {
-				mode = mode_Ls;
+			case iro_ia32_vfild:
+			case iro_ia32_vfld:
+			case iro_ia32_xLoad: {
+				mode  = get_ia32_ls_mode(node);
+				align = 4;
+				break;
 			}
-			if(attr->data.need_32bit_stackent) {
-				mode = mode_Is;
+
+			case iro_ia32_FldCW: {
+				/* although 2 byte would be enough 4 byte performs best */
+				mode  = mode_Iu;
+				align = 4;
+				break;
 			}
-			align = get_mode_size_bytes(mode);
-			be_node_needs_frame_entity(env, node, mode, align);
-		} else if (is_ia32_vfild(node) || is_ia32_xLoad(node)
-		           || is_ia32_vfld(node)) {
-			const ir_mode *mode  = get_ia32_ls_mode(node);
-			int            align = 4;
-			be_node_needs_frame_entity(env, node, mode, align);
-		} else if(is_ia32_FldCW(node)) {
-			/* although 2 byte would be enough 4 byte performs best */
-			const ir_mode *mode  = mode_Iu;
-			int            align = 4;
-			be_node_needs_frame_entity(env, node, mode, align);
-		} else {
+
+			default:
 #ifndef NDEBUG
-			assert(is_ia32_St(node) ||
- 				   is_ia32_xStoreSimple(node) ||
-				   is_ia32_vfst(node) ||
-				   is_ia32_vfist(node) ||
-				   is_ia32_vfisttp(node) ||
-			       is_ia32_FnstCW(node));
+				panic("unexpected frame user while collection frame entity nodes");
+
+			case iro_ia32_FnstCW:
+			case iro_ia32_Store8Bit:
+			case iro_ia32_Store:
+			case iro_ia32_fst:
+			case iro_ia32_fstp:
+			case iro_ia32_vfist:
+			case iro_ia32_vfisttp:
+			case iro_ia32_vfst:
+			case iro_ia32_xStore:
+			case iro_ia32_xStoreSimple:
 #endif
+				return;
 		}
+	} else {
+		return;
 	}
+	be_node_needs_frame_entity(env, node, mode, align);
 }
 
 /**
  * We transform 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) {
+static void ia32_after_ra(void *self)
+{
 	ia32_code_gen_t *cg = self;
 	ir_graph *irg = cg->irg;
 	be_fec_env_t *fec_env = be_new_frame_entity_coalescer(cg->birg);
@@ -1515,7 +1422,8 @@ static void ia32_after_ra(void *self) {
  * virtual with real x87 instructions, creating a block schedule and peephole
  * optimisations.
  */
-static void ia32_finish(void *self) {
+static void ia32_finish(void *self)
+{
 	ia32_code_gen_t *cg = self;
 	ir_graph        *irg = cg->irg;
 
@@ -1523,7 +1431,7 @@ static void ia32_finish(void *self) {
 
 	/* we might have to rewrite x87 virtual registers */
 	if (cg->do_x87_sim) {
-		x87_simulate_graph(cg->arch_env, cg->birg);
+		x87_simulate_graph(cg->birg);
 	}
 
 	/* do peephole optimisations */
@@ -1538,13 +1446,16 @@ static void ia32_finish(void *self) {
  * Emits the code, closes the output file and frees
  * the code generator interface.
  */
-static void ia32_codegen(void *self) {
+static void ia32_codegen(void *self)
+{
 	ia32_code_gen_t *cg = self;
 	ir_graph        *irg = cg->irg;
 
-	ia32_gen_routine(cg, irg);
-
-	cur_reg_set = NULL;
+	if (ia32_cg_config.emit_machcode) {
+		ia32_gen_binary_routine(cg, irg);
+	} else {
+		ia32_gen_routine(cg, irg);
+	}
 
 	/* remove it from the isa */
 	cg->isa->cg = NULL;
@@ -1553,14 +1464,14 @@ static void ia32_codegen(void *self) {
 	ia32_current_cg = NULL;
 
 	/* de-allocate code generator */
-	del_set(cg->reg_set);
 	free(cg);
 }
 
 /**
  * Returns the node representing the PIC base.
  */
-static ir_node *ia32_get_pic_base(void *self) {
+static ir_node *ia32_get_pic_base(void *self)
+{
 	ir_node         *block;
 	ia32_code_gen_t *cg      = self;
 	ir_node         *get_eip = cg->get_eip;
@@ -1568,7 +1479,7 @@ static ir_node *ia32_get_pic_base(void *self) {
 		return get_eip;
 
 	block       = get_irg_start_block(cg->irg);
-	get_eip     = new_rd_ia32_GetEIP(NULL, cg->irg, block);
+	get_eip     = new_bd_ia32_GetEIP(NULL, block);
 	cg->get_eip = get_eip;
 
 	be_dep_on_frame(get_eip);
@@ -1583,7 +1494,6 @@ static const arch_code_generator_if_t ia32_code_gen_if = {
 	ia32_before_abi,     /* before abi introduce hook */
 	ia32_prepare_graph,
 	NULL,                /* spill */
-	ia32_before_sched,   /* before scheduling hook */
 	ia32_before_ra,      /* before register allocation hook */
 	ia32_after_ra,       /* after register allocation hook */
 	ia32_finish,         /* called before codegen */
@@ -1593,15 +1503,14 @@ static const arch_code_generator_if_t ia32_code_gen_if = {
 /**
  * Initializes a IA32 code generator.
  */
-static void *ia32_cg_init(be_irg_t *birg) {
+static void *ia32_cg_init(be_irg_t *birg)
+{
 	ia32_isa_t      *isa = (ia32_isa_t *)birg->main_env->arch_env;
 	ia32_code_gen_t *cg  = XMALLOCZ(ia32_code_gen_t);
 
 	cg->impl      = &ia32_code_gen_if;
 	cg->irg       = birg->irg;
-	cg->reg_set   = new_set(ia32_cmp_irn_reg_assoc, 1024);
 	cg->isa       = isa;
-	cg->arch_env  = birg->main_env->arch_env;
 	cg->birg      = birg;
 	cg->blk_sched = NULL;
 	cg->dump      = (birg->main_env->options->dump_flags & DUMP_BE) ? 1 : 0;
@@ -1622,8 +1531,6 @@ static void *ia32_cg_init(be_irg_t *birg) {
 	}
 #endif /* NDEBUG */
 
-	cur_reg_set = cg->reg_set;
-
 	assert(ia32_current_cg == NULL);
 	ia32_current_cg = cg;
 
@@ -1678,6 +1585,7 @@ static ia32_isa_t ia32_isa_template = {
 		&ia32_isa_if,            /* isa interface implementation */
 		&ia32_gp_regs[REG_ESP],  /* stack pointer register */
 		&ia32_gp_regs[REG_EBP],  /* base pointer register */
+		&ia32_reg_classes[CLASS_ia32_gp],  /* static link pointer register class */
 		-1,                      /* stack direction */
 		2,                       /* power of two stack alignment, 2^2 == 4 */
 		NULL,                    /* main environment */
@@ -1746,7 +1654,8 @@ static void init_asm_constraints(void)
 /**
  * Initializes the backend ISA.
  */
-static arch_env_t *ia32_init(FILE *file_handle) {
+static arch_env_t *ia32_init(FILE *file_handle)
+{
 	static int inited = 0;
 	ia32_isa_t *isa;
 	int        i, n;
@@ -1766,6 +1675,8 @@ static arch_env_t *ia32_init(FILE *file_handle) {
 
 	ia32_register_init();
 	ia32_create_opcodes(&ia32_irn_ops);
+	/* special handling for SwitchJmp */
+	op_ia32_SwitchJmp->ops.be_ops = &ia32_SwitchJmp_irn_ops;
 
 	be_emit_init(file_handle);
 	isa->regs_16bit     = pmap_create();
@@ -1786,7 +1697,6 @@ static arch_env_t *ia32_init(FILE *file_handle) {
 
 	/* enter the ISA object into the intrinsic environment */
 	intrinsic_env.isa = isa;
-	ia32_handle_intrinsics();
 
 	/* emit asm includes */
 	n = get_irp_n_asms();
@@ -1815,7 +1725,8 @@ static arch_env_t *ia32_init(FILE *file_handle) {
 /**
  * Closes the output file and frees the ISA structure.
  */
-static void ia32_done(void *self) {
+static void ia32_done(void *self)
+{
 	ia32_isa_t *isa = self;
 
 	/* emit now all global declarations */
@@ -1845,18 +1756,16 @@ static void ia32_done(void *self) {
  *  - the virtual floating point registers
  *  - the SSE vector register set
  */
-static unsigned ia32_get_n_reg_class(const void *self) {
-	(void) self;
+static unsigned ia32_get_n_reg_class(void)
+{
 	return N_CLASSES;
 }
 
 /**
  * Return the register class for index i.
  */
-static const arch_register_class_t *ia32_get_reg_class(const void *self,
-                                                       unsigned i)
+static const arch_register_class_t *ia32_get_reg_class(unsigned i)
 {
-	(void) self;
 	assert(i < N_CLASSES);
 	return &ia32_reg_classes[i];
 }
@@ -1867,11 +1776,8 @@ static const arch_register_class_t *ia32_get_reg_class(const void *self,
  * @param mode The mode in question.
  * @return A register class which can hold values of the given mode.
  */
-const arch_register_class_t *ia32_get_reg_class_for_mode(const void *self,
-		const ir_mode *mode)
+const arch_register_class_t *ia32_get_reg_class_for_mode(const ir_mode *mode)
 {
-	(void) self;
-
 	if (mode_is_float(mode)) {
 		return ia32_cg_config.use_sse2 ? &ia32_reg_classes[CLASS_ia32_xmm] : &ia32_reg_classes[CLASS_ia32_vfp];
 	}
@@ -1879,6 +1785,88 @@ const arch_register_class_t *ia32_get_reg_class_for_mode(const void *self,
 		return &ia32_reg_classes[CLASS_ia32_gp];
 }
 
+/**
+ * Returns the register for parameter nr.
+ */
+static const arch_register_t *ia32_get_RegParam_reg(unsigned cc, unsigned nr,
+                                                    const ir_mode *mode)
+{
+	static const arch_register_t *gpreg_param_reg_fastcall[] = {
+		&ia32_gp_regs[REG_ECX],
+		&ia32_gp_regs[REG_EDX],
+		NULL
+	};
+	static const unsigned MAXNUM_GPREG_ARGS = 3;
+
+	static const arch_register_t *gpreg_param_reg_regparam[] = {
+		&ia32_gp_regs[REG_EAX],
+		&ia32_gp_regs[REG_EDX],
+		&ia32_gp_regs[REG_ECX]
+	};
+
+	static const arch_register_t *gpreg_param_reg_this[] = {
+		&ia32_gp_regs[REG_ECX],
+		NULL,
+		NULL
+	};
+
+	static const arch_register_t *fpreg_sse_param_reg_std[] = {
+		&ia32_xmm_regs[REG_XMM0],
+		&ia32_xmm_regs[REG_XMM1],
+		&ia32_xmm_regs[REG_XMM2],
+		&ia32_xmm_regs[REG_XMM3],
+		&ia32_xmm_regs[REG_XMM4],
+		&ia32_xmm_regs[REG_XMM5],
+		&ia32_xmm_regs[REG_XMM6],
+		&ia32_xmm_regs[REG_XMM7]
+	};
+
+	static const arch_register_t *fpreg_sse_param_reg_this[] = {
+		NULL,  /* in case of a "this" pointer, the first parameter must not be a float */
+	};
+	static const unsigned MAXNUM_SSE_ARGS = 8;
+
+	if ((cc & cc_this_call) && nr == 0)
+		return gpreg_param_reg_this[0];
+
+	if (! (cc & cc_reg_param))
+		return NULL;
+
+	if (mode_is_float(mode)) {
+		if (!ia32_cg_config.use_sse2 || (cc & cc_fpreg_param) == 0)
+			return NULL;
+		if (nr >= MAXNUM_SSE_ARGS)
+			return NULL;
+
+		if (cc & cc_this_call) {
+			return fpreg_sse_param_reg_this[nr];
+		}
+		return fpreg_sse_param_reg_std[nr];
+	} else if (mode_is_int(mode) || mode_is_reference(mode)) {
+		unsigned num_regparam;
+
+		if (get_mode_size_bits(mode) > 32)
+			return NULL;
+
+		if (nr >= MAXNUM_GPREG_ARGS)
+			return NULL;
+
+		if (cc & cc_this_call) {
+			return gpreg_param_reg_this[nr];
+		}
+		num_regparam = cc & ~cc_bits;
+		if (num_regparam == 0) {
+			/* default fastcall */
+			return gpreg_param_reg_fastcall[nr];
+		}
+		if (nr < num_regparam)
+			return gpreg_param_reg_regparam[nr];
+		return NULL;
+	}
+
+	panic("unknown argument mode");
+}
+
 /**
  * Get the ABI restrictions for procedure calls.
  * @param self        The this pointer.
@@ -1902,25 +1890,25 @@ static void ia32_get_call_abi(const void *self, ir_type *method_type,
 	call_flags.bits.store_args_sequential = 0;
 	/* call_flags.bits.try_omit_fp                 not changed: can handle both settings */
 	call_flags.bits.fp_free               = 0;  /* the frame pointer is fixed in IA32 */
-	call_flags.bits.call_has_imm          = 0;  /* No call immediates, we handle this by ourselves */
+	call_flags.bits.call_has_imm          = 0;  /* No call immediate, we handle this by ourselves */
 
 	/* set parameter passing style */
 	be_abi_call_set_flags(abi, call_flags, &ia32_abi_callbacks);
 
+	cc = get_method_calling_convention(method_type);
 	if (get_method_variadicity(method_type) == variadicity_variadic) {
 		/* pass all parameters of a variadic function on the stack */
-		cc = cc_cdecl_set;
+		cc = cc_cdecl_set | (cc & cc_this_call);
 	} else {
-		cc = get_method_calling_convention(method_type);
 		if (get_method_additional_properties(method_type) & mtp_property_private &&
 		    ia32_cg_config.optimize_cc) {
-			/* set the calling conventions to register parameter */
-			cc = (cc & ~cc_bits) | cc_reg_param;
+			/* set the fast calling conventions (allowing up to 3) */
+			cc = SET_FASTCALL(cc) | 3;
 		}
 	}
 
 	/* we have to pop the shadow parameter ourself for compound calls */
-	if( (get_method_calling_convention(method_type) & cc_compound_ret)
+	if ( (get_method_calling_convention(method_type) & cc_compound_ret)
 			&& !(cc & cc_reg_param)) {
 		pop_amount += get_mode_size_bytes(mode_P_data);
 	}
@@ -2021,7 +2009,8 @@ static const arch_code_generator_if_t *ia32_get_code_generator_if(void *self)
 /**
  * Returns the estimated execution time of an ia32 irn.
  */
-static sched_timestep_t ia32_sched_exectime(void *env, const ir_node *irn) {
+static sched_timestep_t ia32_sched_exectime(void *env, const ir_node *irn)
+{
 	(void) env;
 	return is_ia32_irn(irn) ? ia32_get_op_estimated_cost(irn) : 1;
 }
@@ -2050,12 +2039,10 @@ static const ilp_sched_selector_t *ia32_get_ilp_sched_selector(const void *self)
 /**
  * Returns the necessary byte alignment for storing a register of given class.
  */
-static int ia32_get_reg_class_alignment(const void *self,
-                                        const arch_register_class_t *cls)
+static int ia32_get_reg_class_alignment(const arch_register_class_t *cls)
 {
 	ir_mode *mode = arch_register_class_mode(cls);
 	int bytes     = get_mode_size_bytes(mode);
-	(void) self;
 
 	if (mode_is_float(mode) && bytes > 8)
 		return 16;
@@ -2063,7 +2050,7 @@ static int ia32_get_reg_class_alignment(const void *self,
 }
 
 static const be_execution_unit_t ***ia32_get_allowed_execution_units(
-		const void *self, const ir_node *irn)
+		const ir_node *irn)
 {
 	static const be_execution_unit_t *_allowed_units_BRANCH[] = {
 		&ia32_execution_units_BRANCH[IA32_EXECUNIT_TP_BRANCH_BRANCH1],
@@ -2097,7 +2084,6 @@ static const be_execution_unit_t ***ia32_get_allowed_execution_units(
 		NULL
 	};
 	const be_execution_unit_t ***ret;
-	(void) self;
 
 	if (is_ia32_irn(irn)) {
 		ret = get_ia32_exec_units(irn);
@@ -2120,7 +2106,8 @@ static const be_execution_unit_t ***ia32_get_allowed_execution_units(
 /**
  * Return the abstract ia32 machine.
  */
-static const be_machine_t *ia32_get_machine(const void *self) {
+static const be_machine_t *ia32_get_machine(const void *self)
+{
 	const ia32_isa_t *isa = self;
 	return isa->cpu;
 }
@@ -2135,41 +2122,133 @@ static ir_graph **ia32_get_irg_list(const void *self, ir_graph ***irg_list)
 	return NULL;
 }
 
-static void ia32_mark_remat(const void *self, ir_node *node) {
-	(void) self;
+static void ia32_mark_remat(ir_node *node)
+{
 	if (is_ia32_irn(node)) {
 		set_ia32_is_remat(node);
 	}
 }
 
 /**
- * Allows or disallows the creation of Psi nodes for the given Phi nodes.
+ * Check for Abs or -Abs.
+ */
+static int psi_is_Abs_or_Nabs(ir_node *cmp, ir_node *sel, ir_node *t, ir_node *f)
+{
+	ir_node *l, *r;
+	pn_Cmp  pnc;
+
+	if (cmp == NULL)
+		return 0;
+
+	/* must be <, <=, >=, > */
+	pnc = get_Proj_proj(sel);
+	if (pnc != pn_Cmp_Ge && pnc != pn_Cmp_Gt &&
+		pnc != pn_Cmp_Le && pnc != pn_Cmp_Lt)
+		return 0;
+
+	l = get_Cmp_left(cmp);
+	r = get_Cmp_right(cmp);
+
+	/* must be x cmp 0 */
+	if ((l != t && l != f) || !is_Const(r) || !is_Const_null(r))
+		return 0;
+
+	if ((!is_Minus(t) || get_Minus_op(t) != f) &&
+		(!is_Minus(f) || get_Minus_op(f) != t))
+		return 0;
+	return 1;
+}
+
+/**
+ * Check for Abs only
+ */
+static int psi_is_Abs(ir_node *cmp, ir_node *sel, ir_node *t, ir_node *f)
+{
+	ir_node *l, *r;
+	pn_Cmp  pnc;
+
+	if (cmp == NULL)
+		return 0;
+
+	/* must be <, <=, >=, > */
+	pnc = get_Proj_proj(sel);
+	if (pnc != pn_Cmp_Ge && pnc != pn_Cmp_Gt &&
+		pnc != pn_Cmp_Le && pnc != pn_Cmp_Lt)
+		return 0;
+
+	l = get_Cmp_left(cmp);
+	r = get_Cmp_right(cmp);
+
+	/* must be x cmp 0 */
+	if ((l != t && l != f) || !is_Const(r) || !is_Const_null(r))
+		return 0;
+
+	if ((!is_Minus(t) || get_Minus_op(t) != f) &&
+		(!is_Minus(f) || get_Minus_op(f) != t))
+		return 0;
+
+	if (pnc & pn_Cmp_Gt) {
+		/* x >= 0 ? -x : x is NABS */
+		if (is_Minus(t))
+			return 0;
+	} else {
+		/* x < 0 ? x : -x is NABS */
+		if (is_Minus(f))
+			return 0;
+	}
+	return 1;
+}
+
+
+/**
+ * Allows or disallows the creation of Mux nodes for the given Phi nodes.
+ *
+ * @param sel        A selector of a Cond.
+ * @param phi_list   List of Phi nodes about to be converted (linked via get_Phi_next() field)
+ * @param i          First data predecessor involved in if conversion
+ * @param j          Second data predecessor involved in if conversion
+ *
  * @return 1 if allowed, 0 otherwise
  */
-static int ia32_is_psi_allowed(ir_node *sel, ir_node *phi_list, int i, int j)
+static int ia32_is_mux_allowed(ir_node *sel, ir_node *phi_list, int i, int j)
 {
 	ir_node *phi;
-	ir_node *cmp = NULL;
+	ir_node *cmp;
+	pn_Cmp  pn;
+	ir_node *cl, *cr;
 
-	/* we can't handle psis with 64bit compares yet */
+	/* we can't handle Muxs with 64bit compares yet */
 	if (is_Proj(sel)) {
 		cmp = get_Proj_pred(sel);
 		if (is_Cmp(cmp)) {
 			ir_node *left     = get_Cmp_left(cmp);
 			ir_mode *cmp_mode = get_irn_mode(left);
-			if (!mode_is_float(cmp_mode) && get_mode_size_bits(cmp_mode) > 32)
-				return 0;
+			if (!mode_is_float(cmp_mode) && get_mode_size_bits(cmp_mode) > 32) {
+				/* 64bit Abs IS supported */
+				for (phi = phi_list; phi; phi = get_Phi_next(phi)) {
+					ir_node *t = get_Phi_pred(phi, i);
+					ir_node *f = get_Phi_pred(phi, j);
+
+					if (! psi_is_Abs(cmp, sel, t, f))
+						return 0;
+				}
+				return 1;
+			}
 		} else {
-			cmp = NULL;
+			/* we do not support nodes without Cmp yet */
+			return 0;
 		}
+	} else {
+		/* we do not support nodes without Cmp yet */
+		return 0;
 	}
 
-	if (ia32_cg_config.use_cmov) {
-		if (ia32_cg_config.use_sse2 && cmp != NULL) {
-			pn_Cmp pn   = get_Proj_proj(sel);
-			ir_node *cl = get_Cmp_left(cmp);
-			ir_node *cr = get_Cmp_right(cmp);
+	pn = get_Proj_proj(sel);
+	cl = get_Cmp_left(cmp);
+	cr = get_Cmp_right(cmp);
 
+	if (ia32_cg_config.use_cmov) {
+		if (ia32_cg_config.use_sse2) {
 			/* check the Phi nodes: no 64bit and no floating point cmov */
 			for (phi = phi_list; phi; phi = get_Phi_next(phi)) {
 				ir_mode *mode = get_irn_mode(phi);
@@ -2178,97 +2257,132 @@ static int ia32_is_psi_allowed(ir_node *sel, ir_node *phi_list, int i, int j)
 					/* check for Min, Max */
 					ir_node *t = get_Phi_pred(phi, i);
 					ir_node *f = get_Phi_pred(phi, j);
-					int res    = 0;
 
 					/* SSE2 supports Min & Max */
 					if (pn == pn_Cmp_Lt || pn == pn_Cmp_Le || pn == pn_Cmp_Ge || pn == pn_Cmp_Gt) {
 						if (cl == t && cr == f) {
-							/* Psi(a <=/>= b, a, b) => MIN, MAX */
-							res = 1;
+							/* Mux(a <=/>= b, a, b) => MIN, MAX */
+							continue;
 						} else if (cl == f && cr == t) {
-							/* Psi(a <=/>= b, b, a) => MAX, MIN */
-							res = 1;
+							/* Mux(a <=/>= b, b, a) => MAX, MIN */
+							continue;
 						}
 					}
-					if (! res)
-						return 0;
-
-				} else if (get_mode_size_bits(mode) > 32)
 					return 0;
+				} else if (get_mode_size_bits(mode) > 32) {
+					/* no 64bit cmov */
+					return 0;
+				}
 			}
 		} else {
 			/* check the Phi nodes: no 64bit and no floating point cmov */
 			for (phi = phi_list; phi; phi = get_Phi_next(phi)) {
 				ir_mode *mode = get_irn_mode(phi);
 
-				if (mode_is_float(mode) || get_mode_size_bits(mode) > 32)
+				if (mode_is_float(mode)) {
+					ir_node *t = get_Phi_pred(phi, i);
+					ir_node *f = get_Phi_pred(phi, j);
+
+					/* always support Mux(!float, C1, C2) */
+					if (is_Const(t) && is_Const(f) && !mode_is_float(get_irn_mode(cl))) {
+						switch (be_transformer) {
+						case TRANSFORMER_DEFAULT:
+							/* always support Mux(!float, C1, C2) */
+							continue;
+#ifdef FIRM_GRGEN_BE
+						case TRANSFORMER_PBQP:
+						case TRANSFORMER_RAND:
+							/* no support for Mux(*, C1, C2) */
+							return 0;
+#endif
+						default:
+							panic("invalid transformer");
+						}
+					}
+					/* only abs or nabs supported */
+					if (! psi_is_Abs_or_Nabs(cmp, sel, t, f))
+						return 0;
+				} else if (get_mode_size_bits(mode) > 32)
 					return 0;
 			}
 		}
 
 		return 1;
-	} else {
-		ir_node *cl, *cr;
-		pn_Cmp  pn;
-
-		/* No cmov, only some special cases */
-		if (cmp == NULL)
-			return 0;
+	} else { /* No Cmov, only some special cases */
 
 		/* Now some supported cases here */
-		pn = get_Proj_proj(sel);
-		cl = get_Cmp_left(cmp);
-		cr = get_Cmp_right(cmp);
-
 		for (phi = phi_list; phi; phi = get_Phi_next(phi)) {
 			ir_mode *mode = get_irn_mode(phi);
-			int res = 0;
 			ir_node *t, *f;
 
 			t = get_Phi_pred(phi, i);
 			f = get_Phi_pred(phi, j);
 
-			/* no floating point and no 64bit yet */
-			if (mode_is_float(mode) || get_mode_size_bits(mode) > 32)
+			if (mode_is_float(mode)) {
+				/* always support Mux(!float, C1, C2) */
+				if (is_Const(t) && is_Const(f) &&
+						!mode_is_float(get_irn_mode(cl))) {
+					switch (be_transformer) {
+						case TRANSFORMER_DEFAULT:
+							/* always support Mux(!float, C1, C2) */
+							continue;
+#ifdef FIRM_GRGEN_BE
+						case TRANSFORMER_PBQP:
+						case TRANSFORMER_RAND:
+							/* no support for Mux(*, C1, C2) */
+							return 0;
+#endif
+						default:
+							panic("invalid transformer");
+					}
+				}
+				/* only abs or nabs supported */
+				if (! psi_is_Abs_or_Nabs(cmp, sel, t, f))
+					return 0;
+			} else if (get_mode_size_bits(mode) > 32) {
+				/* no 64bit yet */
 				return 0;
+			}
 
 			if (is_Const(t) && is_Const(f)) {
 				if ((is_Const_null(t) && is_Const_one(f)) || (is_Const_one(t) && is_Const_null(f))) {
-					/* always support Psi(x, C1, C2) */
-					res = 1;
+					/* always support Mux(x, C1, C2) */
+					continue;
 				}
 			} else if (pn == pn_Cmp_Lt || pn == pn_Cmp_Le || pn == pn_Cmp_Ge || pn == pn_Cmp_Gt) {
-				if (0) {
 #if 0
-				} else if (cl == t && cr == f) {
-					/* Psi(a <=/>= b, a, b) => Min, Max */
-					res = 1;
-				} else if (cl == f && cr == t) {
-					/* Psi(a <=/>= b, b, a) => Max, Min */
-					res = 1;
+				if (cl == t && cr == f) {
+					/* Mux(a <=/>= b, a, b) => Min, Max */
+					continue;
+				}
+				if (cl == f && cr == t) {
+					/* Mux(a <=/>= b, b, a) => Max, Min */
+					continue;
+				}
 #endif
-				} else if ((pn & pn_Cmp_Gt) && !mode_is_signed(mode) &&
-				           is_Const(f) && is_Const_null(f) && is_Sub(t) &&
-				           get_Sub_left(t) == cl && get_Sub_right(t) == cr) {
-					/* Psi(a >=u b, a - b, 0) unsigned Doz */
-					res = 1;
-				} else if ((pn & pn_Cmp_Lt) && !mode_is_signed(mode) &&
-				           is_Const(t) && is_Const_null(t) && is_Sub(f) &&
-				           get_Sub_left(f) == cl && get_Sub_right(f) == cr) {
-					/* Psi(a <=u b, 0, a - b) unsigned Doz */
-					res = 1;
-				} else if (is_Const(cr) && is_Const_null(cr)) {
+				if ((pn & pn_Cmp_Gt) && !mode_is_signed(mode) &&
+				    is_Const(f) && is_Const_null(f) && is_Sub(t) &&
+				    get_Sub_left(t) == cl && get_Sub_right(t) == cr) {
+					/* Mux(a >=u b, a - b, 0) unsigned Doz */
+					continue;
+				}
+				if ((pn & pn_Cmp_Lt) && !mode_is_signed(mode) &&
+				    is_Const(t) && is_Const_null(t) && is_Sub(f) &&
+				    get_Sub_left(f) == cl && get_Sub_right(f) == cr) {
+					/* Mux(a <=u b, 0, a - b) unsigned Doz */
+					continue;
+				}
+				if (is_Const(cr) && is_Const_null(cr)) {
 					if (cl == t && is_Minus(f) && get_Minus_op(f) == cl) {
-						/* Psi(a <=/>= 0 ? a : -a) Nabs/Abs */
-						res = 1;
+						/* Mux(a <=/>= 0 ? a : -a) Nabs/Abs */
+						continue;
 					} else if (cl == f && is_Minus(t) && get_Minus_op(t) == cl) {
-						/* Psi(a <=/>= 0 ? -a : a) Abs/Nabs */
-						res = 1;
+						/* Mux(a <=/>= 0 ? -a : a) Abs/Nabs */
+						continue;
 					}
 				}
 			}
-			if (! res)
-				return 0;
+			return 0;
 		}
 		/* all checks passed */
 		return 1;
@@ -2276,9 +2390,8 @@ static int ia32_is_psi_allowed(ir_node *sel, ir_node *phi_list, int i, int j)
 	return 0;
 }
 
-static asm_constraint_flags_t ia32_parse_asm_constraint(const void *self, const char **c)
+static asm_constraint_flags_t ia32_parse_asm_constraint(const char **c)
 {
-	(void) self;
 	(void) c;
 
 	/* we already added all our simple flags to the flags modifier list in
@@ -2286,20 +2399,45 @@ static asm_constraint_flags_t ia32_parse_asm_constraint(const void *self, const
 	return ASM_CONSTRAINT_FLAG_INVALID;
 }
 
-static int ia32_is_valid_clobber(const void *self, const char *clobber)
+static int ia32_is_valid_clobber(const char *clobber)
 {
-	(void) self;
-
 	return ia32_get_clobber_register(clobber) != NULL;
 }
 
+/**
+ * Create the trampoline code.
+ */
+static ir_node *ia32_create_trampoline_fkt(ir_node *block, ir_node *mem, ir_node *trampoline, ir_node *env, ir_node *callee)
+{
+	ir_node  *st, *p = trampoline;
+	ir_mode *mode    = get_irn_mode(p);
+
+	/* mov  ecx,<env> */
+	st  = new_r_Store(block, mem, p, new_Const_long(mode_Bu, 0xb9), 0);
+	mem = new_r_Proj(block, st, mode_M, pn_Store_M);
+	p   = new_r_Add(block, p, new_Const_long(mode_Iu, 1), mode);
+	st  = new_r_Store(block, mem, p, env, 0);
+	mem = new_r_Proj(block, st, mode_M, pn_Store_M);
+	p   = new_r_Add(block, p, new_Const_long(mode_Iu, 4), mode);
+	/* jmp  <callee> */
+	st  = new_r_Store(block, mem, p, new_Const_long(mode_Bu, 0xe9), 0);
+	mem = new_r_Proj(block, st, mode_M, pn_Store_M);
+	p   = new_r_Add(block, p, new_Const_long(mode_Iu, 1), mode);
+	st  = new_r_Store(block, mem, p, callee, 0);
+	mem = new_r_Proj(block, st, mode_M, pn_Store_M);
+	p   = new_r_Add(block, p, new_Const_long(mode_Iu, 4), mode);
+
+	return mem;
+}
+
 /**
  * Returns the libFirm configuration parameter for this backend.
  */
-static const backend_params *ia32_get_libfirm_params(void) {
+static const backend_params *ia32_get_libfirm_params(void)
+{
 	static const ir_settings_if_conv_t ifconv = {
-		4,                    /* maxdepth, doesn't matter for Psi-conversion */
-		ia32_is_psi_allowed   /* allows or disallows Psi creation for given selector */
+		4,                    /* maxdepth, doesn't matter for Mux-conversion */
+		ia32_is_mux_allowed   /* allows or disallows Mux creation for given selector */
 	};
 	static const ir_settings_arch_dep_t ad = {
 		1,                   /* also use subs */
@@ -2309,18 +2447,20 @@ static const backend_params *ia32_get_libfirm_params(void) {
 
 		1,  /* allow Mulhs */
 		1,  /* allow Mulus */
-		32  /* Mulh allowed up to 32 bit */
+		32, /* Mulh allowed up to 32 bit */
 	};
 	static backend_params p = {
 		1,     /* need dword lowering */
 		1,     /* support inline assembly */
-		0,     /* no immediate floating point mode. */
-		NULL,  /* no additional opcodes */
 		NULL,  /* will be set later */
 		ia32_create_intrinsic_fkt,
 		&intrinsic_env,  /* context for ia32_create_intrinsic_fkt */
-		NULL,  /* will be set below */
-		NULL   /* will be set below */
+		NULL,  /* ifconv info will be set below */
+		NULL,  /* float arithmetic mode, will be set below */
+		12,    /* size of trampoline code */
+		4,     /* alignment of trampoline code */
+		ia32_create_trampoline_fkt,
+		4      /* alignment of stack parameter */
 	};
 
 	ia32_setup_cg_config();
@@ -2331,6 +2471,8 @@ static const backend_params *ia32_get_libfirm_params(void) {
 
 	p.dep_param    = &ad;
 	p.if_conv_info = &ifconv;
+	if (! ia32_cg_config.use_sse2)
+		p.mode_float_arithmetic = mode_E;
 	return &p;
 }
 
@@ -2346,22 +2488,24 @@ static lc_opt_enum_int_var_t gas_var = {
 	(int*) &be_gas_flavour, gas_items
 };
 
+#ifdef FIRM_GRGEN_BE
 static const lc_opt_enum_int_items_t transformer_items[] = {
 	{ "default", TRANSFORMER_DEFAULT },
-#ifdef FIRM_GRGEN_BE
 	{ "pbqp",    TRANSFORMER_PBQP    },
 	{ "random",  TRANSFORMER_RAND    },
-#endif
 	{ NULL,      0                   }
 };
 
 static lc_opt_enum_int_var_t transformer_var = {
 	(int*)&be_transformer, transformer_items
 };
+#endif
 
 static const lc_opt_table_entry_t ia32_options[] = {
 	LC_OPT_ENT_ENUM_INT("gasmode", "set the GAS compatibility mode", &gas_var),
+#ifdef FIRM_GRGEN_BE
 	LC_OPT_ENT_ENUM_INT("transformer", "the transformer used for code selection", &transformer_var),
+#endif
 	LC_OPT_ENT_INT("stackalign", "set power of two stack alignment for calls",
 	               &ia32_isa_template.arch_env.stack_alignment),
 	LC_OPT_LAST
@@ -2370,6 +2514,7 @@ static const lc_opt_table_entry_t ia32_options[] = {
 const arch_isa_if_t ia32_isa_if = {
 	ia32_init,
 	ia32_done,
+	ia32_handle_intrinsics,
 	ia32_get_n_reg_class,
 	ia32_get_reg_class,
 	ia32_get_reg_class_for_mode,
@@ -2387,12 +2532,6 @@ const arch_isa_if_t ia32_isa_if = {
 	ia32_is_valid_clobber
 };
 
-void ia32_init_emitter(void);
-void ia32_init_finish(void);
-void ia32_init_optimize(void);
-void ia32_init_transform(void);
-void ia32_init_x87(void);
-
 void be_init_arch_ia32(void)
 {
 	lc_opt_entry_t *be_grp   = lc_opt_get_grp(firm_opt_get_root(), "be");