ia32: do not change tarval output modes in init
[libfirm] / ir / be / ia32 / bearch_ia32.c
index 9fbf998..5f0be45 100644 (file)
  * @file
  * @brief       This is the main ia32 firm backend driver.
  * @author      Christian Wuerdig
- * @version     $Id$
  */
-#ifdef HAVE_CONFIG_H
 #include "config.h"
-#endif
 
 #include "lc_opts.h"
 #include "lc_opts_enum.h"
 
 #include <math.h>
 
-#include "pseudo_irg.h"
 #include "irarch.h"
 #include "irgwalk.h"
 #include "irprog.h"
@@ -42,8 +38,8 @@
 #include "irflag.h"
 #include "irgmod.h"
 #include "irgopt.h"
-#include "irbitset.h"
 #include "irgopt.h"
+#include "irdump.h"
 #include "pdeq.h"
 #include "pset.h"
 #include "debug.h"
 #include "irtools.h"
 #include "iroptimize.h"
 #include "instrument.h"
-
-#include "../beabi.h"
-#include "../beirg_t.h"
-#include "../benode_t.h"
-#include "../belower.h"
-#include "../besched_t.h"
+#include "iropt_t.h"
+#include "lower_dw.h"
+#include "lower_calls.h"
+#include "lower_mode_b.h"
+#include "lower_softfloat.h"
+
+#include "beabi.h"
+#include "beirg.h"
+#include "benode.h"
+#include "belower.h"
+#include "besched.h"
 #include "be.h"
-#include "../be_t.h"
-#include "../beirgmod.h"
-#include "../be_dbgout.h"
-#include "../beblocksched.h"
-#include "../bemachine.h"
-#include "../beilpsched.h"
-#include "../bespillslots.h"
-#include "../bemodule.h"
-#include "../begnuas.h"
-#include "../bestate.h"
-#include "../beflags.h"
+#include "be_t.h"
+#include "beirgmod.h"
+#include "be_dbgout.h"
+#include "beblocksched.h"
+#include "bespillutil.h"
+#include "bespillslots.h"
+#include "bemodule.h"
+#include "begnuas.h"
+#include "bestate.h"
+#include "beflags.h"
+#include "betranshlp.h"
+#include "belistsched.h"
+#include "beabihelper.h"
+#include "bestack.h"
 
 #include "bearch_ia32_t.h"
 
 #include "ia32_new_nodes.h"
 #include "gen_ia32_regalloc_if.h"
-#include "gen_ia32_machine.h"
 #include "ia32_common_transform.h"
 #include "ia32_transform.h"
 #include "ia32_emitter.h"
-#include "ia32_map_regs.h"
 #include "ia32_optimize.h"
 #include "ia32_x87.h"
 #include "ia32_dbg_stat.h"
 #include "ia32_finish.h"
-#include "ia32_util.h"
 #include "ia32_fpu.h"
 #include "ia32_architecture.h"
 
 #ifdef FIRM_GRGEN_BE
 #include "ia32_pbqp_transform.h"
-#endif
 
-DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
+transformer_t be_transformer = TRANSFORMER_DEFAULT;
+#endif
 
-/* TODO: ugly */
-static set *cur_reg_set = NULL;
+ir_mode *ia32_mode_fpcw;
+ir_mode *ia32_mode_E;
+ir_type *ia32_type_E;
 
-ir_mode         *mode_fpcw       = NULL;
-ia32_code_gen_t *ia32_current_cg = NULL;
+/** The current omit-fp state */
+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.
@@ -106,246 +112,82 @@ ia32_code_gen_t *ia32_current_cg = NULL;
 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 *dbgi, 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(ir_graph *irg, ir_node **place,
                                     create_const_node_func func,
                                     const arch_register_t* reg)
 {
        ir_node *block, *res;
 
-       if(*place != NULL)
+       if (*place != NULL)
                return *place;
 
-       block = get_irg_start_block(cg->irg);
-       res = func(NULL, cg->irg, block);
-       arch_set_irn_register(cg->arch_env, res, reg);
+       block = get_irg_start_block(irg);
+       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,
-                           &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,
-                           &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,
-                           &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,
-                           &ia32_gp_regs[REG_GP_UKNWN]);
+ir_node *ia32_new_NoReg_gp(ir_graph *irg)
+{
+       ia32_irg_data_t *irg_data = ia32_get_irg_data(irg);
+       return create_const(irg, &irg_data->noreg_gp, new_bd_ia32_NoReg_GP,
+                           &ia32_registers[REG_GP_NOREG]);
 }
 
-ir_node *ia32_new_Unknown_vfp(ia32_code_gen_t *cg) {
-       return create_const(cg, &cg->unknown_vfp, new_rd_ia32_Unknown_VFP,
-                           &ia32_vfp_regs[REG_VFP_UKNWN]);
+ir_node *ia32_new_NoReg_vfp(ir_graph *irg)
+{
+       ia32_irg_data_t *irg_data = ia32_get_irg_data(irg);
+       return create_const(irg, &irg_data->noreg_vfp, new_bd_ia32_NoReg_VFP,
+                           &ia32_registers[REG_VFP_NOREG]);
 }
 
-ir_node *ia32_new_Unknown_xmm(ia32_code_gen_t *cg) {
-       return create_const(cg, &cg->unknown_xmm, new_rd_ia32_Unknown_XMM,
-                           &ia32_xmm_regs[REG_XMM_UKNWN]);
+ir_node *ia32_new_NoReg_xmm(ir_graph *irg)
+{
+       ia32_irg_data_t *irg_data = ia32_get_irg_data(irg);
+       return create_const(irg, &irg_data->noreg_xmm, new_bd_ia32_NoReg_XMM,
+                           &ia32_registers[REG_XMM_NOREG]);
 }
 
-ir_node *ia32_new_Fpu_truncate(ia32_code_gen_t *cg) {
-       return create_const(cg, &cg->fpu_trunc_mode, new_rd_ia32_ChangeCW,
-                        &ia32_fp_cw_regs[REG_FPCW]);
+ir_node *ia32_new_Fpu_truncate(ir_graph *irg)
+{
+       ia32_irg_data_t *irg_data = ia32_get_irg_data(irg);
+       return create_const(irg, &irg_data->fpu_trunc_mode, new_bd_ia32_ChangeCW,
+                        &ia32_registers[REG_FPCW]);
 }
 
 
 /**
  * Returns the admissible noreg register node for input register pos of node irn.
  */
-ir_node *ia32_get_admissible_noreg(ia32_code_gen_t *cg, ir_node *irn, int pos) {
-       const arch_register_req_t *req;
+static ir_node *ia32_get_admissible_noreg(ir_node *irn, int pos)
+{
+       ir_graph                  *irg = get_irn_irg(irn);
+       const arch_register_req_t *req = arch_get_irn_register_req_in(irn, pos);
 
-       req = arch_get_register_req(cg->arch_env, irn, pos);
        assert(req != NULL && "Missing register requirements");
        if (req->cls == &ia32_reg_classes[CLASS_ia32_gp])
-               return ia32_new_NoReg_gp(cg);
+               return ia32_new_NoReg_gp(irg);
 
        if (ia32_cg_config.use_sse2) {
-               return ia32_new_NoReg_xmm(cg);
-       } else {
-               return ia32_new_NoReg_vfp(cg);
-       }
-}
-
-/**************************************************
- *                         _ _              _  __
- *                        | | |            (_)/ _|
- *  _ __ ___  __ _    __ _| | | ___   ___   _| |_
- * | '__/ _ \/ _` |  / _` | | |/ _ \ / __| | |  _|
- * | | |  __/ (_| | | (_| | | | (_) | (__  | | |
- * |_|  \___|\__, |  \__,_|_|_|\___/ \___| |_|_|
- *            __/ |
- *           |___/
- **************************************************/
-
-/**
- * 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)
-{
-       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));
-       return arch_no_register_req;
-}
-
-static void ia32_set_irn_reg(ir_node *irn, const arch_register_t *reg)
-{
-       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;
-
-               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];
+               return ia32_new_NoReg_xmm(irg);
        } 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;
-
-       if (is_ia32_need_stackent(irn))
-               classification |= arch_irn_class_reload;
-
-       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;
+               return ia32_new_NoReg_vfp(irg);
        }
-
-       if (is_ia32_irn(irn)) {
-               flags |= get_ia32_flags(irn);
-       }
-
-       return flags;
 }
 
 /**
@@ -353,16 +195,20 @@ static arch_irn_flags_t ia32_get_flags(const ir_node *irn) {
  */
 typedef struct {
        be_abi_call_flags_bits_t flags;  /**< The call flags. */
-       const arch_env_t *aenv;          /**< The architecture environment. */
        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) {
-       set_ia32_frame_ent(irn, ent);
+static void ia32_set_frame_entity(ir_node *node, ir_entity *entity)
+{
+       if (is_be_node(node))
+               be_node_set_frame_entity(node, entity);
+       else
+               set_ia32_frame_ent(node, entity);
 }
 
 static void ia32_set_frame_offset(ir_node *irn, int bias)
@@ -371,9 +217,9 @@ static void ia32_set_frame_offset(ir_node *irn, int bias)
                return;
 
        if (is_ia32_Pop(irn) || is_ia32_PopMem(irn)) {
-               ia32_code_gen_t *cg = ia32_current_cg;
-               int omit_fp = be_abi_omit_fp(cg->birg->abi);
-               if (omit_fp) {
+               ir_graph          *irg     = get_irn_irg(irn);
+               be_stack_layout_t *layout  = be_get_irg_stack_layout(irg);
+               if (layout->sp_relative) {
                        /* Pop nodes modify the stack pointer before calculating the
                         * destination address, so fix this here
                         */
@@ -385,195 +231,31 @@ static void ia32_set_frame_offset(ir_node *irn, int bias)
 
 static int ia32_get_sp_bias(const ir_node *node)
 {
+       if (is_ia32_Call(node))
+               return -(int)get_ia32_call_attr_const(node)->pop;
+
        if (is_ia32_Push(node))
                return 4;
 
        if (is_ia32_Pop(node) || is_ia32_PopMem(node))
                return -4;
 
-       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.
- *
- * @param self       The callback object.
- * @param mem        A pointer to the mem node. Update this if you define new memory.
- * @param reg_map    A map mapping all callee_save/ignore/parameter registers to their defining nodes.
- * @param stack_bias Points to the current stack bias, can be modified if needed.
- *
- * @return           The register which shall be used as a stack frame base.
- *
- * All nodes which define registers in @p reg_map must keep @p reg_map current.
- */
-static const arch_register_t *ia32_abi_prologue(void *self, ir_node **mem, pmap *reg_map, int *stack_bias)
-{
-       ia32_abi_env_t   *env      = self;
-       ia32_code_gen_t  *cg       = ia32_current_cg;
-       const arch_env_t *arch_env = env->aenv;
-
-       if (! env->flags.try_omit_fp) {
-               ir_graph *irg     =env->irg;
-               ir_node  *bl      = get_irg_start_block(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  *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);
-
-               /* 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);
-
-               /* 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);
-
-               /* 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);
-
-               /* 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);
-
-               be_abi_reg_map_set(reg_map, arch_env->sp, curr_sp);
-               be_abi_reg_map_set(reg_map, arch_env->bp, curr_bp);
-
-               return arch_env->bp;
-       }
-
-       return arch_env->sp;
-}
-
-/**
- * Generate the routine epilogue.
- * @param self    The callback object.
- * @param bl      The block for the epilog
- * @param mem     A pointer to the mem node. Update this if you define new memory.
- * @param reg_map A map mapping all callee_save/ignore/parameter registers to their defining nodes.
- * @return        The register which shall be used as a stack frame base.
- *
- * All nodes which define registers in @p reg_map must keep @p reg_map current.
- */
-static void ia32_abi_epilogue(void *self, ir_node *bl, ir_node **mem, pmap *reg_map)
-{
-       ia32_abi_env_t   *env      = self;
-       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);
-       } else {
-               ir_mode *mode_bp = arch_env->bp->reg_class->mode;
-
-               if (ia32_cg_config.use_leave) {
-                       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);
-               } else {
-                       ir_node *pop;
-
-                       /* the old SP is not needed anymore (kill the proj) */
-                       assert(is_Proj(curr_sp));
-                       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);
-
-                       /* 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);
-
-                       *mem = new_r_Proj(irg, 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);
+       if (is_ia32_Leave(node) || is_ia32_CopyEbpEsp(node)) {
+               return SP_BIAS_RESET;
        }
 
-       be_abi_reg_map_set(reg_map, arch_env->sp, curr_sp);
-       be_abi_reg_map_set(reg_map, arch_env->bp, curr_bp);
-}
-
-/**
- * Initialize the callback object.
- * @param call The call object.
- * @param aenv The architecture environment.
- * @param irg  The graph with the method.
- * @return     Some pointer. This pointer is passed to all other callback functions as self object.
- */
-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->aenv  = aenv;
-       return env;
-}
-
-/**
- * Destroy the callback object.
- * @param self The callback object.
- */
-static void ia32_abi_done(void *self) {
-       free(self);
+       return 0;
 }
 
 /**
- * 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);
+               ir_type *old_bp_type   = new_type_primitive(mode_Iu);
+               ir_type *ret_addr_type = new_type_primitive(mode_Iu);
 
                between_type           = new_type_struct(IDENT("ia32_between_type"));
                old_bp_ent             = new_entity(between_type, IDENT("old_bp"), old_bp_type);
@@ -591,11 +273,41 @@ 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);
        }
-
-       return env->flags.try_omit_fp ? omit_fp_between_type : between_type;
 #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(ir_graph *irg)
+{
+       const be_stack_layout_t *layout = be_get_irg_stack_layout(irg);
+       ia32_build_between_type();
+       return layout->sp_relative ? omit_fp_between_type : between_type;
+}
+
+/**
+ * Return the stack entity that contains the return address.
+ */
+ir_entity *ia32_get_return_address_entity(ir_graph *irg)
+{
+       const be_stack_layout_t *layout = be_get_irg_stack_layout(irg);
+       ia32_build_between_type();
+       return layout->sp_relative ? omit_fp_ret_addr_ent : ret_addr_ent;
+}
+
+/**
+ * Return the stack entity that contains the frame address.
+ */
+ir_entity *ia32_get_frame_address_entity(ir_graph *irg)
+{
+       const be_stack_layout_t *layout = be_get_irg_stack_layout(irg);
+       ia32_build_between_type();
+       return layout->sp_relative ? NULL : old_bp_ent;
+}
+
 /**
  * Get the estimated cycle count for @p irn.
  *
@@ -634,8 +346,8 @@ static int ia32_get_op_estimated_cost(const ir_node *irn)
                        cycles.
                */
                if (is_ia32_use_frame(irn) || (
-                       is_ia32_NoReg_GP(get_irn_n(irn, n_ia32_base)) &&
-                       is_ia32_NoReg_GP(get_irn_n(irn, n_ia32_index))
+                   is_ia32_NoReg_GP(get_irn_n(irn, n_ia32_base)) &&
+                   is_ia32_NoReg_GP(get_irn_n(irn, n_ia32_index))
                    )) {
                        cost += 5;
                } else {
@@ -655,12 +367,19 @@ 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)
+{
+       (void) irn;
+       (void) i;
+       (void) inverse;
+       (void) obst;
+       return NULL;
+
+#if 0
        ir_mode  *mode;
        ir_mode  *irn_mode;
        ir_node  *block, *noreg, *nomem;
-       dbg_info *dbg;
+       dbg_info *dbgi;
 
        /* we cannot invert non-ia32 irns */
        if (! is_ia32_irn(irn))
@@ -679,13 +398,12 @@ 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);
-       dbg      = get_irn_dbg_info(irn);
+       nomem    = get_irg_no_mem(irg);
+       dbgi     = get_irn_dbg_info(irn);
 
        /* initialize structure */
        inverse->nodes = obstack_alloc(obst, 2 * sizeof(inverse->nodes[0]));
@@ -694,11 +412,10 @@ static arch_inverse_t *ia32_get_inverse(const ir_node *irn, int i, arch_inverse_
 
        switch (get_ia32_irn_opcode(irn)) {
                case iro_ia32_Add:
-#if 0
                        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(dbgi, 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)));
@@ -707,60 +424,55 @@ 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(dbgi, 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(dbgi, block, noreg, noreg, nomem, (ir_node*) irn, get_irn_n(irn, i ^ 1));
                                inverse->costs   += 2;
                        }
-#endif
                        break;
                case iro_ia32_Sub:
-#if 0
                        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(dbgi, 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(dbgi, 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(dbgi, block, noreg, noreg, nomem, get_irn_n(irn, n_ia32_binary_left), (ir_node*) irn);
                                }
                                inverse->costs += 1;
                        }
-#endif
                        break;
                case iro_ia32_Xor:
-#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(dbgi, 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(dbgi, 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(dbgi, 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(dbgi, block, (ir_node*) irn);
                        inverse->costs   += 1;
                        break;
                }
@@ -770,11 +482,12 @@ static arch_inverse_t *ia32_get_inverse(const ir_node *irn, int i, arch_inverse_
        }
 
        return inverse;
+#endif
 }
 
 static ir_mode *get_spill_mode_mode(const ir_mode *mode)
 {
-       if(mode_is_float(mode))
+       if (mode_is_float(mode))
                return mode_D;
 
        return mode_Iu;
@@ -795,48 +508,69 @@ static ir_mode *get_spill_mode(const ir_node *node)
  */
 static int ia32_is_spillmode_compatible(const ir_mode *mode, const ir_mode *spillmode)
 {
-       if(mode_is_float(mode)) {
-               return mode == spillmode;
-       } else {
-               return 1;
-       }
+       return !mode_is_float(mode) || mode == spillmode;
 }
 
 /**
  * Check if irn can load its operand at position i from memory (source addressmode).
- * @param self   Pointer to irn ops itself
  * @param irn    The irn to be checked
  * @param i      The operands position
  * @return Non-Zero if operand can be loaded
  */
-static int ia32_possible_memory_operand(const ir_node *irn, unsigned int i) {
-       ir_node *op = get_irn_n(irn, i);
-       const ir_mode *mode = get_irn_mode(op);
+static int ia32_possible_memory_operand(const ir_node *irn, unsigned int i)
+{
+       ir_node       *op        = get_irn_n(irn, i);
+       const ir_mode *mode      = get_irn_mode(op);
        const ir_mode *spillmode = get_spill_mode(op);
 
-       if (
-               (i != n_ia32_binary_left && i != n_ia32_binary_right) || /* a "real" operand position must be requested */
-               ! is_ia32_irn(irn)                                    ||  /* must be an ia32 irn */
-               get_ia32_am_arity(irn) != ia32_am_binary              ||  /* must be a binary operation TODO is this necessary? */
-               get_ia32_op_type(irn) != ia32_Normal                  ||  /* must not already be a addressmode irn */
-               ! (get_ia32_am_support(irn) & ia32_am_Source)         ||  /* must be capable of source addressmode */
-               ! ia32_is_spillmode_compatible(mode, spillmode)       ||
-               is_ia32_use_frame(irn))                                  /* must not already use frame */
+       if (!is_ia32_irn(irn)                              ||  /* must be an ia32 irn */
+           get_ia32_op_type(irn) != ia32_Normal           ||  /* must not already be a addressmode irn */
+           !ia32_is_spillmode_compatible(mode, spillmode) ||
+           is_ia32_use_frame(irn))                            /* must not already use frame */
                return 0;
 
-       if (i == n_ia32_binary_left) {
-               const arch_register_req_t *req;
-               if(!is_ia32_commutative(irn))
+       switch (get_ia32_am_support(irn)) {
+               case ia32_am_none:
                        return 0;
-               /* we can't swap left/right for limited registers
-                * (As this (currently) breaks constraint handling copies)
-                */
-               req = get_ia32_in_req(irn, n_ia32_binary_left);
-               if (req->type & arch_register_req_type_limited) {
-                       return 0;
-               }
+
+               case ia32_am_unary:
+                       if (i != n_ia32_unary_op)
+                               return 0;
+                       break;
+
+               case ia32_am_binary:
+                       switch (i) {
+                               case n_ia32_binary_left: {
+                                       const arch_register_req_t *req;
+                                       if (!is_ia32_commutative(irn))
+                                               return 0;
+
+                                       /* we can't swap left/right for limited registers
+                                        * (As this (currently) breaks constraint handling copies)
+                                        */
+                                       req = arch_get_irn_register_req_in(irn, n_ia32_binary_left);
+                                       if (req->type & arch_register_req_type_limited)
+                                               return 0;
+                                       break;
+                               }
+
+                               case n_ia32_binary_right:
+                                       break;
+
+                               default:
+                                       return 0;
+                       }
+                       break;
+
+               default:
+                       panic("Unknown AM type");
        }
 
+       /* HACK: must not already use "real" memory.
+        * This can happen for Call and Div */
+       if (!is_NoMem(get_irn_n(irn, n_ia32_mem)))
+               return 0;
+
        return 1;
 }
 
@@ -846,14 +580,8 @@ static void ia32_perform_memory_operand(ir_node *irn, ir_node *spill,
        ir_mode *load_mode;
        ir_mode *dest_op_mode;
 
-       ia32_code_gen_t *cg = ia32_current_cg;
-
        assert(ia32_possible_memory_operand(irn, i) && "Cannot perform memory operand change");
 
-       if (i == n_ia32_binary_left) {
-               ia32_swap_left_right(irn);
-       }
-
        set_ia32_op_type(irn, ia32_AddrModeS);
 
        load_mode    = get_irn_mode(get_irn_n(irn, i));
@@ -864,36 +592,29 @@ static void ia32_perform_memory_operand(ir_node *irn, ir_node *spill,
        set_ia32_use_frame(irn);
        set_ia32_need_stackent(irn);
 
-       set_irn_n(irn, n_ia32_base, get_irg_frame(get_irn_irg(irn)));
-       set_irn_n(irn, n_ia32_binary_right, ia32_get_admissible_noreg(cg, irn, n_ia32_binary_right));
-       set_irn_n(irn, n_ia32_mem, spill);
-       set_ia32_is_reload(irn);
-
-       /* immediates are only allowed on the right side */
-       if (i == n_ia32_binary_left && is_ia32_Immediate(get_irn_n(irn, n_ia32_binary_left))) {
+       if (i == n_ia32_binary_left                    &&
+           get_ia32_am_support(irn) == ia32_am_binary &&
+           /* immediates are only allowed on the right side */
+           !is_ia32_Immediate(get_irn_n(irn, n_ia32_binary_right))) {
                ia32_swap_left_right(irn);
+               i = n_ia32_binary_right;
        }
+
+       assert(is_NoMem(get_irn_n(irn, n_ia32_mem)));
+
+       set_irn_n(irn, n_ia32_base, get_irg_frame(get_irn_irg(irn)));
+       set_irn_n(irn, n_ia32_mem,  spill);
+       set_irn_n(irn, i,           ia32_get_admissible_noreg(irn, i));
+       set_ia32_is_reload(irn);
 }
 
 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,
-       ia32_classify,
-       ia32_get_flags,
        ia32_get_frame_entity,
-       ia32_set_frame_entity,
        ia32_set_frame_offset,
        ia32_get_sp_bias,
        ia32_get_inverse,
@@ -902,154 +623,108 @@ static const arch_irn_ops_t ia32_irn_ops = {
        ia32_perform_memory_operand,
 };
 
-/**************************************************
- *                _                         _  __
- *               | |                       (_)/ _|
- *   ___ ___   __| | ___  __ _  ___ _ __    _| |_
- *  / __/ _ \ / _` |/ _ \/ _` |/ _ \ '_ \  | |  _|
- * | (_| (_) | (_| |  __/ (_| |  __/ | | | | | |
- *  \___\___/ \__,_|\___|\__, |\___|_| |_| |_|_|
- *                        __/ |
- *                       |___/
- **************************************************/
-
 static ir_entity *mcount = NULL;
+static int gprof = 0;
 
-#define ID(s) new_id_from_chars(s, sizeof(s) - 1)
-
-static void ia32_before_abi(void *self) {
-       lower_mode_b_config_t lower_mode_b_config = {
-               mode_Iu,  /* lowered mode */
-               mode_Bu,  /* prefered mode for set */
-               0,        /* don't lower direct compares */
-       };
-       ia32_code_gen_t *cg = self;
-
-       ir_lower_mode_b(cg->irg, &lower_mode_b_config);
-       if (cg->dump)
-               be_dump(cg->irg, "-lower_modeb", dump_ir_block_graph_sched);
-       if (cg->gprof) {
+static void ia32_before_abi(ir_graph *irg)
+{
+       if (gprof) {
                if (mcount == NULL) {
-                       ir_type *tp = new_type_method(ID("FKT.mcount"), 0, 0);
-                       mcount = new_entity(get_glob_type(), ID("mcount"), tp);
+                       ir_type *tp = new_type_method(0, 0);
+                       ident   *id = new_id_from_str("mcount");
+                       mcount = new_entity(get_glob_type(), id, tp);
                        /* FIXME: enter the right ld_ident here */
                        set_entity_ld_ident(mcount, get_entity_ident(mcount));
-                       set_entity_visibility(mcount, visibility_external_allocated);
+                       set_entity_visibility(mcount, ir_visibility_external);
                }
-               instrument_initcall(cg->irg, mcount);
+               instrument_initcall(irg, mcount);
        }
 }
 
-transformer_t be_transformer = TRANSFORMER_DEFAULT;
-
 /**
  * Transforms the standard firm graph into
  * an ia32 firm graph
  */
-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;
+static void ia32_prepare_graph(ir_graph *irg)
+{
+       ia32_irg_data_t *irg_data = ia32_get_irg_data(irg);
 
 #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;
+       switch (be_transformer) {
+       case TRANSFORMER_DEFAULT:
+               /* transform remaining nodes into assembler instructions */
+               ia32_transform_graph(irg);
+               break;
+
+       case TRANSFORMER_PBQP:
+       case TRANSFORMER_RAND:
+               /* transform nodes into assembler instructions by PBQP magic */
+               ia32_transform_graph_by_pbqp(irg);
+               break;
+
+       default:
+               panic("invalid transformer");
+       }
+#else
+       ia32_transform_graph(irg);
 #endif
 
-               default: panic("invalid transformer");
-       }
-
        /* do local optimizations (mainly CSE) */
-       optimize_graph_df(cg->irg);
+       optimize_graph_df(irg);
+       /* backend code expects that outedges are always enabled */
+       edges_assure(irg);
 
-       if (cg->dump)
-               be_dump(cg->irg, "-transformed", dump_ir_block_graph_sched);
+       if (irg_data->dump)
+               dump_ir_graph(irg, "transformed");
 
        /* optimize address mode */
-       ia32_optimize_graph(cg);
-
-       if (cg->dump)
-               be_dump(cg->irg, "-am", dump_ir_block_graph_sched);
+       ia32_optimize_graph(irg);
 
        /* do code placement, to optimize the position of constants */
-       place_code(cg->irg);
-
-       if (cg->dump)
-               be_dump(cg->irg, "-place", dump_ir_block_graph_sched);
-}
+       place_code(irg);
+       /* backend code expects that outedges are always enabled */
+       edges_assure(irg);
 
-/**
- * Dummy functions for hooks we don't need but which must be filled.
- */
-static void ia32_before_sched(void *self) {
-       (void) self;
+       if (irg_data->dump)
+               dump_ir_graph(irg, "place");
 }
 
-static void turn_back_am(ir_node *node)
+ir_node *ia32_turn_back_am(ir_node *node)
 {
-       ir_graph *irg   = current_ir_graph;
        dbg_info *dbgi  = get_irn_dbg_info(node);
+       ir_graph *irg   = get_irn_irg(node);
        ir_node  *block = get_nodes_block(node);
        ir_node  *base  = get_irn_n(node, n_ia32_base);
-       ir_node  *index = get_irn_n(node, n_ia32_index);
+       ir_node  *idx   = get_irn_n(node, n_ia32_index);
        ir_node  *mem   = get_irn_n(node, n_ia32_mem);
-       ir_node  *noreg = ia32_new_NoReg_gp(ia32_current_cg);
-       ir_node  *load;
-       ir_node  *load_res;
-       ir_node  *mem_proj;
-       const ir_edge_t *edge;
+       ir_node  *noreg;
 
-       load     = new_rd_ia32_Load(dbgi, irg, block, base, index, mem);
-       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, idx, mem);
+       ir_node  *load_res = new_rd_Proj(dbgi, load, mode_Iu, pn_ia32_Load_res);
 
        ia32_copy_am_attrs(load, node);
        if (is_ia32_is_reload(node))
                set_ia32_is_reload(load);
-       set_irn_n(node, n_ia32_mem, new_NoMem());
+       set_irn_n(node, n_ia32_mem, get_irg_no_mem(irg));
 
-       switch (get_ia32_am_arity(node)) {
+       switch (get_ia32_am_support(node)) {
                case ia32_am_unary:
                        set_irn_n(node, n_ia32_unary_op, load_res);
                        break;
 
                case ia32_am_binary:
-                       if (is_ia32_Immediate(get_irn_n(node, n_ia32_Cmp_right))) {
-                               assert(is_ia32_Cmp(node)  || is_ia32_Cmp8Bit(node) ||
-                                      is_ia32_Test(node) || is_ia32_Test8Bit(node));
+                       if (is_ia32_Immediate(get_irn_n(node, n_ia32_binary_right))) {
                                set_irn_n(node, n_ia32_binary_left, load_res);
                        } else {
                                set_irn_n(node, n_ia32_binary_right, load_res);
                        }
                        break;
 
-               case ia32_am_ternary:
-                       set_irn_n(node, n_ia32_binary_right, load_res);
-                       break;
-
-               default: break;
+               default:
+                       panic("Unknown AM type");
        }
-       set_irn_n(node, n_ia32_base, noreg);
+       noreg = ia32_new_NoReg_gp(current_ir_graph);
+       set_irn_n(node, n_ia32_base,  noreg);
        set_irn_n(node, n_ia32_index, noreg);
        set_ia32_am_offs_int(node, 0);
        set_ia32_am_sc(node, NULL);
@@ -1058,24 +733,22 @@ static void turn_back_am(ir_node *node)
 
        /* rewire mem-proj */
        if (get_irn_mode(node) == mode_T) {
-               mem_proj = NULL;
+               const ir_edge_t *edge;
                foreach_out_edge(node, edge) {
                        ir_node *out = get_edge_src_irn(edge);
-                       if(get_irn_mode(out) == mode_M) {
-                               assert(mem_proj == NULL);
-                               mem_proj = out;
+                       if (get_irn_mode(out) == mode_M) {
+                               set_Proj_pred(out, load);
+                               set_Proj_proj(out, pn_ia32_Load_M);
+                               break;
                        }
                }
-
-               if(mem_proj != NULL) {
-                       set_Proj_pred(mem_proj, load);
-                       set_Proj_proj(mem_proj, pn_ia32_Load_M);
-               }
        }
 
        set_ia32_op_type(node, ia32_Normal);
        if (sched_is_scheduled(node))
                sched_add_before(node, load);
+
+       return load_res;
 }
 
 static ir_node *flags_remat(ir_node *node, ir_node *after)
@@ -1094,13 +767,12 @@ static ir_node *flags_remat(ir_node *node, ir_node *after)
        type = get_ia32_op_type(node);
        switch (type) {
                case ia32_AddrModeS:
-                       turn_back_am(node);
+                       ia32_turn_back_am(node);
                        break;
 
                case ia32_AddrModeD:
                        /* TODO implement this later... */
                        panic("found DestAM with flag user %+F this should not happen", node);
-                       break;
 
                default: assert(type == ia32_Normal); break;
        }
@@ -1115,34 +787,34 @@ static ir_node *flags_remat(ir_node *node, ir_node *after)
 /**
  * Called before the register allocator.
  */
-static void ia32_before_ra(void *self) {
-       ia32_code_gen_t *cg = self;
-
+static void ia32_before_ra(ir_graph *irg)
+{
        /* setup fpu rounding modes */
-       ia32_setup_fpu_mode(cg);
+       ia32_setup_fpu_mode(irg);
 
        /* fixup flags */
-       be_sched_fix_flags(cg->birg, &ia32_reg_classes[CLASS_ia32_flags],
-                          &flags_remat);
+       be_sched_fix_flags(irg, &ia32_reg_classes[CLASS_ia32_flags],
+                          &flags_remat, NULL);
 
-       ia32_add_missing_keeps(cg);
+       be_add_missing_keeps(irg);
 }
 
 
 /**
  * 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(ir_node *node)
+{
        ir_graph *irg        = get_irn_irg(node);
-       dbg_info *dbg        = get_irn_dbg_info(node);
+       dbg_info *dbgi       = get_irn_dbg_info(node);
        ir_node *block       = get_nodes_block(node);
        ir_entity *ent       = be_get_frame_entity(node);
        ir_mode *mode        = get_irn_mode(node);
        ir_mode *spillmode   = get_spill_mode(node);
-       ir_node *noreg       = ia32_new_NoReg_gp(cg);
+       ir_node *noreg       = ia32_new_NoReg_gp(irg);
        ir_node *sched_point = NULL;
        ir_node *ptr         = get_irg_frame(irg);
-       ir_node *mem         = get_irn_n(node, be_pos_Reload_mem);
+       ir_node *mem         = get_irn_n(node, n_be_Reload_mem);
        ir_node *new_op, *proj;
        const arch_register_t *reg;
 
@@ -1152,16 +824,16 @@ 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(dbgi, 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(dbgi, 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(dbgi, block, ptr, noreg, mem);
        }
        else
-               new_op = new_rd_ia32_Load(dbg, irg, block, ptr, noreg, mem);
+               new_op = new_bd_ia32_Load(dbgi, block, ptr, noreg, mem);
 
        set_ia32_op_type(new_op, ia32_AddrModeS);
        set_ia32_ls_mode(new_op, spillmode);
@@ -1171,7 +843,7 @@ 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(dbgi, new_op, mode, pn_ia32_Load_res);
 
        if (sched_point) {
                sched_add_after(sched_point, new_op);
@@ -1179,10 +851,10 @@ 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);
 }
@@ -1190,17 +862,19 @@ 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(ir_node *node)
+{
        ir_graph *irg  = get_irn_irg(node);
-       dbg_info *dbg  = get_irn_dbg_info(node);
+       dbg_info *dbgi = get_irn_dbg_info(node);
        ir_node *block = get_nodes_block(node);
        ir_entity *ent = be_get_frame_entity(node);
-       const ir_node *spillval = get_irn_n(node, be_pos_Spill_val);
+       const ir_node *spillval = get_irn_n(node, n_be_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 *noreg = ia32_new_NoReg_gp(irg);
+       ir_node *nomem = get_irg_no_mem(irg);
        ir_node *ptr   = get_irg_frame(irg);
-       ir_node *val   = get_irn_n(node, be_pos_Spill_val);
+       ir_node *val   = get_irn_n(node, n_be_Spill_val);
+       ir_node *res;
        ir_node *store;
        ir_node *sched_point = NULL;
 
@@ -1208,30 +882,24 @@ static void transform_to_Store(ia32_code_gen_t *cg, ir_node *node) {
                sched_point = sched_prev(node);
        }
 
-       /* No need to spill unknown values... */
-       if(is_ia32_Unknown_GP(val) ||
-               is_ia32_Unknown_VFP(val) ||
-               is_ia32_Unknown_XMM(val)) {
-               store = nomem;
-               if(sched_point)
-                       sched_remove(node);
-
-               exchange(node, store);
-               return;
-       }
-
        if (mode_is_float(mode)) {
-               if (ia32_cg_config.use_sse2)
-                       store = new_rd_ia32_xStore(dbg, irg, block, ptr, noreg, nomem, val);
-               else
-                       store = new_rd_ia32_vfst(dbg, irg, block, ptr, noreg, nomem, val, mode);
+               if (ia32_cg_config.use_sse2) {
+                       store = new_bd_ia32_xStore(dbgi, block, ptr, noreg, nomem, val);
+                       res   = new_r_Proj(store, mode_M, pn_ia32_xStore_M);
+               } else {
+                       store = new_bd_ia32_vfst(dbgi, block, ptr, noreg, nomem, val, mode);
+                       res   = new_r_Proj(store, mode_M, pn_ia32_vfst_M);
+               }
        } 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(dbgi, block, ptr, noreg, nomem, val);
+               res   = new_r_Proj(store, mode_M, pn_ia32_xxStore_M);
        } 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(dbgi, block, ptr, noreg, nomem, val);
+               res   = new_r_Proj(store, mode_M, pn_ia32_Store8Bit_M);
        } else {
-               store = new_rd_ia32_Store(dbg, irg, block, ptr, noreg, nomem, val);
+               store = new_bd_ia32_Store(dbgi, block, ptr, noreg, nomem, val);
+               res   = new_r_Proj(store, mode_M, pn_ia32_Store_M);
        }
 
        set_ia32_op_type(store, ia32_AddrModeD);
@@ -1239,7 +907,7 @@ 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) {
@@ -1247,56 +915,60 @@ static void transform_to_Store(ia32_code_gen_t *cg, ir_node *node) {
                sched_remove(node);
        }
 
-       exchange(node, store);
+       exchange(node, res);
 }
 
-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);
-       dbg_info *dbg = get_irn_dbg_info(node);
-       ir_node *block = get_nodes_block(node);
-       ir_node *noreg = ia32_new_NoReg_gp(cg);
-       ir_node *frame = get_irg_frame(irg);
+static ir_node *create_push(ir_node *node, ir_node *schedpoint, ir_node *sp, ir_node *mem, ir_entity *ent)
+{
+       dbg_info *dbgi  = get_irn_dbg_info(node);
+       ir_node  *block = get_nodes_block(node);
+       ir_graph *irg   = get_irn_irg(node);
+       ir_node  *noreg = ia32_new_NoReg_gp(irg);
+       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(dbgi, block, frame, noreg, mem, noreg, sp);
 
        set_ia32_frame_ent(push, ent);
        set_ia32_use_frame(push);
        set_ia32_op_type(push, ia32_AddrModeS);
        set_ia32_ls_mode(push, mode_Is);
+       set_ia32_is_spill(push);
 
        sched_add_before(schedpoint, push);
        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);
-       dbg_info *dbg = get_irn_dbg_info(node);
-       ir_node *block = get_nodes_block(node);
-       ir_node *noreg = ia32_new_NoReg_gp(cg);
-       ir_node *frame = get_irg_frame(irg);
+static ir_node *create_pop(ir_node *node, ir_node *schedpoint, ir_node *sp, ir_entity *ent)
+{
+       dbg_info *dbgi  = get_irn_dbg_info(node);
+       ir_node  *block = get_nodes_block(node);
+       ir_graph *irg   = get_irn_irg(node);
+       ir_node  *noreg = ia32_new_NoReg_gp(irg);
+       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(dbgi, block, frame, noreg,
+                                         get_irg_no_mem(irg), sp);
 
        set_ia32_frame_ent(pop, ent);
        set_ia32_use_frame(pop);
        set_ia32_op_type(pop, ia32_AddrModeD);
        set_ia32_ls_mode(pop, mode_Is);
+       set_ia32_is_reload(pop);
 
        sched_add_before(schedpoint, pop);
 
        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);
-       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];
+static ir_node* create_spproj(ir_node *node, ir_node *pred, int pos)
+{
+       dbg_info *dbgi   = get_irn_dbg_info(node);
+       ir_mode  *spmode = mode_Iu;
+       const arch_register_t *spreg = &ia32_registers[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(dbgi, pred, spmode, pos);
+       arch_set_irn_register(sp, spreg);
 
        return sp;
 }
@@ -1306,22 +978,21 @@ 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(ir_node *node)
+{
+       ir_node         *block = get_nodes_block(node);
+       ir_graph        *irg   = get_irn_irg(node);
+       ir_node         *sp    = be_get_initial_reg_value(irg, &ia32_registers[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) {
+       for (i = 0; i < arity; ++i) {
                ir_entity *inent = be_get_MemPerm_in_entity(node, i);
                ir_entity *outent = be_get_MemPerm_out_entity(node, i);
                ir_type *enttype = get_entity_type(inent);
@@ -1331,24 +1002,24 @@ static void transform_MemPerm(ia32_code_gen_t *cg, ir_node *node) {
                ir_node *push;
 
                /* work around cases where entities have different sizes */
-               if(entsize2 < entsize)
+               if (entsize2 < entsize)
                        entsize = entsize2;
                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);
-               if(entsize == 8) {
+               push = create_push(node, node, sp, mem, inent);
+               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);
+                       push = create_push(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());
+               set_irn_n(node, i, new_r_Bad(irg, mode_X));
        }
 
        /* create pops */
-       for(i = arity - 1; i >= 0; --i) {
+       for (i = arity - 1; i >= 0; --i) {
                ir_entity *inent = be_get_MemPerm_in_entity(node, i);
                ir_entity *outent = be_get_MemPerm_out_entity(node, i);
                ir_type *enttype = get_entity_type(outent);
@@ -1357,25 +1028,25 @@ static void transform_MemPerm(ia32_code_gen_t *cg, ir_node *node) {
                ir_node *pop;
 
                /* work around cases where entities have different sizes */
-               if(entsize2 < entsize)
+               if (entsize2 < entsize)
                        entsize = entsize2;
                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);
-               if(entsize == 8) {
+               pop = create_pop(node, node, sp, outent);
+               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);
+                       pop = create_pop(node, node, sp, outent);
+                       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 */
@@ -1390,30 +1061,28 @@ static void transform_MemPerm(ia32_code_gen_t *cg, ir_node *node) {
        }
 
        /* remove memperm */
-       arity = get_irn_arity(node);
-       for(i = 0; i < arity; ++i) {
-               set_irn_n(node, i, new_Bad());
-       }
        sched_remove(node);
+       kill_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;
+       (void) env;
 
        /* beware: the schedule is changed here */
        for (node = sched_last(block); !sched_is_begin(node); node = prev) {
                prev = sched_prev(node);
 
                if (be_is_Reload(node)) {
-                       transform_to_Load(cg, node);
+                       transform_to_Load(node);
                } else if (be_is_Spill(node)) {
-                       transform_to_Store(cg, node);
+                       transform_to_Store(node);
                } else if (be_is_MemPerm(node)) {
-                       transform_MemPerm(cg, node);
+                       transform_MemPerm(node);
                }
        }
 }
@@ -1423,69 +1092,217 @@ 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 = (be_fec_env_t*)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);
+}
+
+static int determine_ebp_input(ir_node *ret)
+{
+       const arch_register_t *bp = &ia32_registers[REG_EBP];
+       int   arity               = get_irn_arity(ret);
+       int   i;
+
+       for (i = 0; i < arity; ++i) {
+               ir_node *input = get_irn_n(ret, i);
+               if (arch_get_irn_register(input) == bp)
+                       return i;
+       }
+       panic("no ebp input found at %+F", ret);
+}
+
+static void introduce_epilog(ir_node *ret)
+{
+       const arch_register_t *sp         = &ia32_registers[REG_ESP];
+       const arch_register_t *bp         = &ia32_registers[REG_EBP];
+       ir_graph              *irg        = get_irn_irg(ret);
+       ir_type               *frame_type = get_irg_frame_type(irg);
+       unsigned               frame_size = get_type_size_bytes(frame_type);
+       be_stack_layout_t     *layout     = be_get_irg_stack_layout(irg);
+       ir_node               *block      = get_nodes_block(ret);
+       ir_node               *first_sp   = get_irn_n(ret, n_be_Return_sp);
+       ir_node               *curr_sp    = first_sp;
+       ir_mode               *mode_gp    = mode_Iu;
+
+       if (!layout->sp_relative) {
+               int      n_ebp   = determine_ebp_input(ret);
+               ir_node *curr_bp = get_irn_n(ret, n_ebp);
+               if (ia32_cg_config.use_leave) {
+                       ir_node *leave = new_bd_ia32_Leave(NULL, block, curr_bp);
+                       curr_bp        = new_r_Proj(leave, mode_gp, pn_ia32_Leave_frame);
+                       curr_sp        = new_r_Proj(leave, mode_gp, pn_ia32_Leave_stack);
+                       arch_set_irn_register(curr_bp, bp);
+                       arch_set_irn_register(curr_sp, sp);
+                       sched_add_before(ret, leave);
+               } else {
+                       ir_node *pop;
+                       ir_node *curr_mem = get_irn_n(ret, n_be_Return_mem);
+                       /* copy ebp to esp */
+                       curr_sp = new_bd_ia32_CopyEbpEsp(NULL, block, curr_bp);
+                       arch_set_irn_register(curr_sp, sp);
+                       sched_add_before(ret, curr_sp);
+
+                       /* pop ebp */
+                       pop      = new_bd_ia32_PopEbp(NULL, block, curr_mem, curr_sp);
+                       curr_bp  = new_r_Proj(pop, mode_gp, pn_ia32_PopEbp_res);
+                       curr_sp  = new_r_Proj(pop, mode_gp, pn_ia32_PopEbp_stack);
+                       curr_mem = new_r_Proj(pop, mode_M, pn_ia32_Pop_M);
+                       arch_set_irn_register(curr_bp, bp);
+                       arch_set_irn_register(curr_sp, sp);
+                       sched_add_before(ret, pop);
+
+                       set_irn_n(ret, n_be_Return_mem, curr_mem);
+               }
+               set_irn_n(ret, n_ebp, curr_bp);
+       } else {
+               ir_node *incsp = be_new_IncSP(sp, block, curr_sp, -(int)frame_size, 0);
+               sched_add_before(ret, incsp);
+               curr_sp = incsp;
+       }
+       set_irn_n(ret, n_be_Return_sp, curr_sp);
+
+       /* keep verifier happy... */
+       if (get_irn_n_edges(first_sp) == 0 && is_Proj(first_sp)) {
+               kill_node(first_sp);
        }
 }
 
 /**
- * We transform Spill and Reload here. This needs to be done before
- * stack biasing otherwise we would miss the corrected offset for these nodes.
+ * put the Prolog code at the beginning, epilog code before each return
  */
-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);
+static void introduce_prolog_epilog(ir_graph *irg)
+{
+       const arch_register_t *sp         = &ia32_registers[REG_ESP];
+       const arch_register_t *bp         = &ia32_registers[REG_EBP];
+       ir_node               *start      = get_irg_start(irg);
+       ir_node               *block      = get_nodes_block(start);
+       ir_type               *frame_type = get_irg_frame_type(irg);
+       unsigned               frame_size = get_type_size_bytes(frame_type);
+       be_stack_layout_t     *layout     = be_get_irg_stack_layout(irg);
+       ir_node               *initial_sp = be_get_initial_reg_value(irg, sp);
+       ir_node               *curr_sp    = initial_sp;
+       ir_mode               *mode_gp    = mode_Iu;
+
+       if (!layout->sp_relative) {
+               /* push ebp */
+               ir_node *mem        = get_irg_initial_mem(irg);
+               ir_node *noreg      = ia32_new_NoReg_gp(irg);
+               ir_node *initial_bp = be_get_initial_reg_value(irg, bp);
+               ir_node *curr_bp    = initial_bp;
+               ir_node *push       = new_bd_ia32_Push(NULL, block, noreg, noreg, mem, curr_bp, curr_sp);
+               ir_node *incsp;
+
+               curr_sp = new_r_Proj(push, mode_gp, pn_ia32_Push_stack);
+               mem     = new_r_Proj(push, mode_M, pn_ia32_Push_M);
+               arch_set_irn_register(curr_sp, sp);
+               sched_add_after(start, push);
 
-       /* create and coalesce frame entities */
-       irg_walk_graph(irg, NULL, ia32_collect_frame_entity_nodes, fec_env);
-       be_assign_entities(fec_env);
-       be_free_frame_entity_coalescer(fec_env);
+               /* move esp to ebp */
+               curr_bp = be_new_Copy(block, curr_sp);
+               sched_add_after(push, curr_bp);
+               be_set_constr_single_reg_out(curr_bp, 0, bp, arch_register_req_type_ignore);
+               curr_sp = be_new_CopyKeep_single(block, curr_sp, curr_bp);
+               sched_add_after(curr_bp, curr_sp);
+               be_set_constr_single_reg_out(curr_sp, 0, sp, arch_register_req_type_produces_sp);
+               edges_reroute(initial_bp, curr_bp);
+               set_irn_n(push, n_ia32_Push_val, initial_bp);
+
+               incsp = be_new_IncSP(sp, block, curr_sp, frame_size, 0);
+               edges_reroute(initial_sp, incsp);
+               set_irn_n(push, n_ia32_Push_stack, initial_sp);
+               sched_add_after(curr_sp, incsp);
+
+               /* make sure the initial IncSP is really used by someone */
+               if (get_irn_n_edges(incsp) <= 1) {
+                       ir_node *in[] = { incsp };
+                       ir_node *keep = be_new_Keep(block, 1, in);
+                       sched_add_after(incsp, keep);
+               }
 
-       irg_block_walk_graph(irg, NULL, ia32_after_ra_walker, cg);
+               layout->initial_bias = -4;
+       } else {
+               ir_node *incsp = be_new_IncSP(sp, block, curr_sp, frame_size, 0);
+               edges_reroute(initial_sp, incsp);
+               be_set_IncSP_pred(incsp, curr_sp);
+               sched_add_after(start, incsp);
+       }
+
+       /* introduce epilog for every return node */
+       {
+               ir_node *end_block = get_irg_end_block(irg);
+               int      arity     = get_irn_arity(end_block);
+               int      i;
+
+               for (i = 0; i < arity; ++i) {
+                       ir_node *ret = get_irn_n(end_block, i);
+                       assert(be_is_Return(ret));
+                       introduce_epilog(ret);
+               }
+       }
 }
 
 /**
@@ -1493,137 +1310,93 @@ 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) {
-       ia32_code_gen_t *cg = self;
-       ir_graph        *irg = cg->irg;
+static void ia32_finish(ir_graph *irg)
+{
+       ia32_irg_data_t   *irg_data     = ia32_get_irg_data(irg);
+       be_stack_layout_t *stack_layout = be_get_irg_stack_layout(irg);
+       bool               at_begin     = stack_layout->sp_relative ? true : false;
+       be_fec_env_t      *fec_env      = be_new_frame_entity_coalescer(irg);
+
+       /* create and coalesce frame entities */
+       irg_walk_graph(irg, NULL, ia32_collect_frame_entity_nodes, fec_env);
+       be_assign_entities(fec_env, ia32_set_frame_entity, at_begin);
+       be_free_frame_entity_coalescer(fec_env);
+
+       irg_block_walk_graph(irg, NULL, ia32_after_ra_walker, NULL);
 
-       ia32_finish_irg(irg, cg);
+       introduce_prolog_epilog(irg);
+
+       /* fix stack entity offsets */
+       be_abi_fix_stack_nodes(irg);
+       be_abi_fix_stack_bias(irg);
+
+       /* fix 2-address code constraints */
+       ia32_finish_irg(irg);
 
        /* we might have to rewrite x87 virtual registers */
-       if (cg->do_x87_sim) {
-               x87_simulate_graph(cg->arch_env, cg->birg);
+       if (irg_data->do_x87_sim) {
+               ia32_x87_simulate_graph(irg);
        }
 
        /* do peephole optimisations */
-       ia32_peephole_optimization(cg);
+       ia32_peephole_optimization(irg);
+
+       be_remove_dead_nodes_from_schedule(irg);
 
        /* create block schedule, this also removes empty blocks which might
         * produce critical edges */
-       cg->blk_sched = be_create_block_schedule(irg, cg->birg->exec_freq);
+       irg_data->blk_sched = be_create_block_schedule(irg);
 }
 
 /**
  * Emits the code, closes the output file and frees
  * the code generator interface.
  */
-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;
-
-       /* remove it from the isa */
-       cg->isa->cg = NULL;
-
-       assert(ia32_current_cg == cg);
-       ia32_current_cg = NULL;
-
-       /* de-allocate code generator */
-       del_set(cg->reg_set);
-       free(cg);
+static void ia32_emit(ir_graph *irg)
+{
+       if (ia32_cg_config.emit_machcode) {
+               ia32_gen_binary_routine(irg);
+       } else {
+               ia32_gen_routine(irg);
+       }
 }
 
 /**
  * Returns the node representing the PIC base.
  */
-static ir_node *ia32_get_pic_base(void *self) {
+static ir_node *ia32_get_pic_base(ir_graph *irg)
+{
+       ia32_irg_data_t *irg_data = ia32_get_irg_data(irg);
        ir_node         *block;
-       ia32_code_gen_t *cg      = self;
-       ir_node         *get_eip = cg->get_eip;
+       ir_node         *get_eip = irg_data->get_eip;
        if (get_eip != NULL)
                return get_eip;
 
-       block       = get_irg_start_block(cg->irg);
-       get_eip     = new_rd_ia32_GetEIP(NULL, cg->irg, block);
-       cg->get_eip = get_eip;
-
-       add_irn_dep(get_eip, get_irg_frame(cg->irg));
+       block             = get_irg_start_block(irg);
+       get_eip           = new_bd_ia32_GetEIP(NULL, block);
+       irg_data->get_eip = get_eip;
 
        return get_eip;
 }
 
-static void *ia32_cg_init(be_irg_t *birg);
-
-static const arch_code_generator_if_t ia32_code_gen_if = {
-       ia32_cg_init,
-       ia32_get_pic_base,   /* return node used as base in pic code addresses */
-       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 */
-       ia32_codegen         /* emit && done */
-};
-
 /**
  * Initializes a IA32 code generator.
  */
-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  = xcalloc(1, sizeof(*cg));
-
-       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;
-       cg->gprof     = (birg->main_env->options->gprof) ? 1 : 0;
-
-       if (cg->gprof) {
-               /* Linux gprof implementation needs base pointer */
-               birg->main_env->options->omit_fp = 0;
-       }
+static void ia32_init_graph(ir_graph *irg)
+{
+       struct obstack  *obst     = be_get_be_obst(irg);
+       ia32_irg_data_t *irg_data = OALLOCZ(obst, ia32_irg_data_t);
 
-       /* enter it */
-       isa->cg = cg;
+       irg_data->dump = (be_get_irg_options(irg)->dump_flags & DUMP_BE) ? 1 : 0;
 
-#ifndef NDEBUG
-       if (isa->name_obst) {
-               obstack_free(isa->name_obst, NULL);
-               obstack_init(isa->name_obst);
+       if (gprof) {
+               /* Linux gprof implementation needs base pointer */
+               be_get_irg_options(irg)->omit_fp = 0;
        }
-#endif /* NDEBUG */
 
-       cur_reg_set = cg->reg_set;
-
-       assert(ia32_current_cg == NULL);
-       ia32_current_cg = cg;
-
-       return (arch_code_generator_t *)cg;
+       be_birg_from_irg(irg)->isa_link = irg_data;
 }
 
-
-
-/*****************************************************************
- *  ____             _                  _   _____  _____
- * |  _ \           | |                | | |_   _|/ ____|  /\
- * | |_) | __ _  ___| | _____ _ __   __| |   | | | (___   /  \
- * |  _ < / _` |/ __| |/ / _ \ '_ \ / _` |   | |  \___ \ / /\ \
- * | |_) | (_| | (__|   <  __/ | | | (_| |  _| |_ ____) / ____ \
- * |____/ \__,_|\___|_|\_\___|_| |_|\__,_| |_____|_____/_/    \_\
- *
- *****************************************************************/
-
-/**
- * Set output modes for GCC
- */
 static const tarval_mode_info mo_integer = {
        TVO_HEX,
        "0x",
@@ -1635,45 +1408,17 @@ static const tarval_mode_info mo_integer = {
  */
 static void set_tarval_output_modes(void)
 {
-       int i;
+       size_t i;
 
-       for (i = get_irp_n_modes() - 1; i >= 0; --i) {
-               ir_mode *mode = get_irp_mode(i);
+       for (i = get_irp_n_modes(); i > 0;) {
+               ir_mode *mode = get_irp_mode(--i);
 
                if (mode_is_int(mode))
                        set_tarval_mode_output_option(mode, &mo_integer);
        }
 }
 
-const arch_isa_if_t ia32_isa_if;
-
-/**
- * The template that generates a new ISA object.
- * Note that this template can be changed by command line
- * arguments.
- */
-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 */
-               -1,                      /* stack direction */
-               2,                       /* power of two stack alignment, 2^2 == 4 */
-               NULL,                    /* main environment */
-               7,                       /* costs for a spill instruction */
-               5,                       /* costs for a reload instruction */
-       },
-       NULL,                    /* 16bit register names */
-       NULL,                    /* 8bit register names */
-       NULL,                    /* 8bit register names high */
-       NULL,                    /* types */
-       NULL,                    /* tv_ents */
-       NULL,                    /* current code generator */
-       NULL,                    /* abstract machine */
-#ifndef NDEBUG
-       NULL,                    /* name obstack */
-#endif
-};
+extern const arch_isa_if_t ia32_isa_if;
 
 static void init_asm_constraints(void)
 {
@@ -1723,199 +1468,493 @@ static void init_asm_constraints(void)
 }
 
 /**
- * Initializes the backend ISA.
+ * Check if Mux(sel, mux_true, mux_false) would represent a Max or Min operation
  */
-static arch_env_t *ia32_init(FILE *file_handle) {
-       static int inited = 0;
-       ia32_isa_t *isa;
-       int        i, n;
+static bool mux_is_float_min_max(ir_node *sel, ir_node *mux_true,
+                                 ir_node *mux_false)
+{
+       ir_node    *cmp_l;
+       ir_node    *cmp_r;
+       ir_relation relation;
+
+       if (!is_Cmp(sel))
+               return false;
+
+       cmp_l = get_Cmp_left(sel);
+       cmp_r = get_Cmp_right(sel);
+       if (!mode_is_float(get_irn_mode(cmp_l)))
+               return false;
+
+       /* check for min/max. They're defined as (C-Semantik):
+        *  min(a, b) = a < b ? a : b
+        *  or min(a, b) = a <= b ? a : b
+        *  max(a, b) = a > b ? a : b
+        *  or max(a, b) = a >= b ? a : b
+        * (Note we only handle float min/max here)
+        */
+       relation = get_Cmp_relation(sel);
+       switch (relation) {
+       case ir_relation_greater_equal:
+       case ir_relation_greater:
+               /* this is a max */
+               if (cmp_l == mux_true && cmp_r == mux_false)
+                       return true;
+               break;
+       case ir_relation_less_equal:
+       case ir_relation_less:
+               /* this is a min */
+               if (cmp_l == mux_true && cmp_r == mux_false)
+                       return true;
+               break;
+       case ir_relation_unordered_greater_equal:
+       case ir_relation_unordered_greater:
+               /* this is a min */
+               if (cmp_l == mux_false && cmp_r == mux_true)
+                       return true;
+               break;
+       case ir_relation_unordered_less_equal:
+       case ir_relation_unordered_less:
+               /* this is a max */
+               if (cmp_l == mux_false && cmp_r == mux_true)
+                       return true;
+               break;
+
+       default:
+               break;
+       }
+
+       return false;
+}
+
+static bool mux_is_set(ir_node *sel, ir_node *mux_true, ir_node *mux_false)
+{
+       ir_mode *mode = get_irn_mode(mux_true);
+       (void) sel;
 
-       if (inited)
-               return NULL;
-       inited = 1;
+       if (!mode_is_int(mode) && !mode_is_reference(mode)
+                       && mode != mode_b)
+               return false;
 
-       set_tarval_output_modes();
+       if (is_Const(mux_true) && is_Const(mux_false)) {
+               /* we can create a set plus up two 3 instructions for any combination
+                * of constants */
+               return true;
+       }
+
+       return false;
+}
 
-       isa = xmalloc(sizeof(*isa));
-       memcpy(isa, &ia32_isa_template, sizeof(*isa));
+static bool mux_is_float_const_const(ir_node *sel, ir_node *mux_true,
+                                     ir_node *mux_false)
+{
+       (void) sel;
+
+       if (!mode_is_float(get_irn_mode(mux_true)))
+               return false;
 
-       if(mode_fpcw == NULL) {
-               mode_fpcw = new_ir_mode("Fpcw", irms_int_number, 16, 0, irma_none, 0);
+       return is_Const(mux_true) && is_Const(mux_false);
+}
+
+static bool mux_is_doz(ir_node *sel, ir_node *mux_true, ir_node *mux_false)
+{
+       ir_node    *cmp_left;
+       ir_node    *cmp_right;
+       ir_node    *sub_left;
+       ir_node    *sub_right;
+       ir_mode    *mode;
+       ir_relation relation;
+
+       if (!is_Cmp(sel))
+               return false;
+
+       mode = get_irn_mode(mux_true);
+       if (mode_is_signed(mode) || mode_is_float(mode))
+               return false;
+
+       relation  = get_Cmp_relation(sel);
+       cmp_left  = get_Cmp_left(sel);
+       cmp_right = get_Cmp_right(sel);
+
+       /* "move" zero constant to false input */
+       if (is_Const(mux_true) && is_Const_null(mux_true)) {
+               ir_node *tmp = mux_false;
+               mux_false = mux_true;
+               mux_true  = tmp;
+               relation = get_negated_relation(relation);
+       }
+       if (!is_Const(mux_false) || !is_Const_null(mux_false))
+               return false;
+       if (!is_Sub(mux_true))
+               return false;
+       sub_left  = get_Sub_left(mux_true);
+       sub_right = get_Sub_right(mux_true);
+
+       /* Mux(a >=u b, 0, a-b) */
+       if ((relation & ir_relation_greater)
+                       && sub_left == cmp_left && sub_right == cmp_right)
+               return true;
+       /* Mux(a <=u b, 0, b-a) */
+       if ((relation & ir_relation_less)
+                       && sub_left == cmp_right && sub_right == cmp_left)
+               return true;
+
+       return false;
+}
+
+static int ia32_is_mux_allowed(ir_node *sel, ir_node *mux_false,
+                               ir_node *mux_true)
+{
+       ir_mode *mode;
+
+       /* middleend can handle some things */
+       if (ir_is_optimizable_mux(sel, mux_false, mux_true))
+               return true;
+       /* we can handle Set for all modes and compares */
+       if (mux_is_set(sel, mux_true, mux_false))
+               return true;
+       /* SSE has own min/max operations */
+       if (ia32_cg_config.use_sse2
+                       && mux_is_float_min_max(sel, mux_true, mux_false))
+               return true;
+       /* we can handle Mux(?, Const[f], Const[f]) */
+       if (mux_is_float_const_const(sel, mux_true, mux_false)) {
+#ifdef FIRM_GRGEN_BE
+               /* well, some code selectors can't handle it */
+               if (be_transformer != TRANSFORMER_PBQP
+                               || be_transformer != TRANSFORMER_RAND)
+                       return true;
+#else
+               return true;
+#endif
        }
 
-       ia32_register_init();
-       ia32_create_opcodes(&ia32_irn_ops);
+       /* no support for 64bit inputs to cmov */
+       mode = get_irn_mode(mux_true);
+       if (get_mode_size_bits(mode) > 32)
+               return false;
+       /* we can handle Abs for all modes and compares (except 64bit) */
+       if (ir_mux_is_abs(sel, mux_false, mux_true) != 0)
+               return true;
+       /* we can't handle MuxF yet */
+       if (mode_is_float(mode))
+               return false;
 
-       be_emit_init(file_handle);
-       isa->regs_16bit     = pmap_create();
-       isa->regs_8bit      = pmap_create();
-       isa->regs_8bit_high = pmap_create();
-       isa->types          = pmap_create();
-       isa->tv_ent         = pmap_create();
-       isa->cpu            = ia32_init_machine_description();
+       if (mux_is_doz(sel, mux_true, mux_false))
+               return true;
 
-       ia32_build_16bit_reg_map(isa->regs_16bit);
-       ia32_build_8bit_reg_map(isa->regs_8bit);
-       ia32_build_8bit_reg_map_high(isa->regs_8bit_high);
+       /* Check Cmp before the node */
+       if (is_Cmp(sel)) {
+               ir_mode *cmp_mode = get_irn_mode(get_Cmp_left(sel));
 
-#ifndef NDEBUG
-       isa->name_obst = xmalloc(sizeof(*isa->name_obst));
-       obstack_init(isa->name_obst);
-#endif /* NDEBUG */
+               /* we can't handle 64bit compares */
+               if (get_mode_size_bits(cmp_mode) > 32)
+                       return false;
 
-       /* enter the ISA object into the intrinsic environment */
-       intrinsic_env.isa = isa;
-       ia32_handle_intrinsics();
-
-       /* emit asm includes */
-       n = get_irp_n_asms();
-       for (i = 0; i < n; ++i) {
-               be_emit_cstring("#APP\n");
-               be_emit_ident(get_irp_asm(i));
-               be_emit_cstring("\n#NO_APP\n");
+               /* we can't handle float compares */
+               if (mode_is_float(cmp_mode))
+                       return false;
        }
 
-       /* needed for the debug support */
-       be_gas_emit_switch_section(GAS_SECTION_TEXT);
-       be_emit_cstring(".Ltext0:\n");
-       be_emit_write_line();
-
-       /* we mark referenced global entities, so we can only emit those which
-        * are actually referenced. (Note: you mustn't use the type visited flag
-        * elsewhere in the backend)
-        */
-       inc_master_type_visited();
+       /* did we disable cmov generation? */
+       if (!ia32_cg_config.use_cmov)
+               return false;
 
-       return &isa->arch_env;
+       /* we can use a cmov */
+       return true;
 }
 
-
+/**
+ * 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_graph *const irg  = get_irn_irg(block);
+       ir_node  *      p    = trampoline;
+       ir_mode  *const mode = get_irn_mode(p);
+       ir_node  *const one  = new_r_Const(irg, get_mode_one(mode_Iu));
+       ir_node  *const four = new_r_Const_long(irg, mode_Iu, 4);
+       ir_node  *      st;
+
+       /* mov  ecx,<env> */
+       st  = new_r_Store(block, mem, p, new_r_Const_long(irg, mode_Bu, 0xb9), cons_none);
+       mem = new_r_Proj(st, mode_M, pn_Store_M);
+       p   = new_r_Add(block, p, one, mode);
+       st  = new_r_Store(block, mem, p, env, cons_none);
+       mem = new_r_Proj(st, mode_M, pn_Store_M);
+       p   = new_r_Add(block, p, four, mode);
+       /* jmp  <callee> */
+       st  = new_r_Store(block, mem, p, new_r_Const_long(irg, mode_Bu, 0xe9), cons_none);
+       mem = new_r_Proj(st, mode_M, pn_Store_M);
+       p   = new_r_Add(block, p, one, mode);
+       st  = new_r_Store(block, mem, p, callee, cons_none);
+       mem = new_r_Proj(st, mode_M, pn_Store_M);
+       p   = new_r_Add(block, p, four, mode);
+
+       return mem;
+}
+
+static const ir_settings_arch_dep_t ia32_arch_dep = {
+       1,                   /* also use subs */
+       4,                   /* maximum shifts */
+       63,                  /* maximum shift amount */
+       ia32_evaluate_insn,  /* evaluate the instruction sequence */
+
+       1,  /* allow Mulhs */
+       1,  /* allow Mulus */
+       32, /* Mulh allowed up to 32 bit */
+};
+static backend_params ia32_backend_params = {
+       1,     /* support inline assembly */
+       1,     /* support Rotl nodes */
+       0,     /* little endian */
+       1,     /* modulo shift efficient */
+       0,     /* non-modulo shift not efficient */
+       &ia32_arch_dep, /* will be set later */
+       ia32_is_mux_allowed,
+       32,    /* machine_size */
+       NULL,  /* float arithmetic mode, will be set below */
+       NULL,  /* long long type */
+       NULL,  /* unsigned long long type */
+       NULL,  /* long double type */
+       12,    /* size of trampoline code */
+       4,     /* alignment of trampoline code */
+       ia32_create_trampoline_fkt,
+       4      /* alignment of stack parameter */
+};
 
 /**
- * Closes the output file and frees the ISA structure.
+ * Initializes the backend ISA.
  */
-static void ia32_done(void *self) {
-       ia32_isa_t *isa = self;
+static void ia32_init(void)
+{
+       ir_mode    *mode_long_long;
+       ir_mode    *mode_unsigned_long_long;
+       ir_type    *type_long_long;
+       ir_type    *type_unsigned_long_long;
 
-       /* emit now all global declarations */
-       be_gas_emit_decls(isa->arch_env.main_env, 1);
+       ia32_setup_cg_config();
 
-       pmap_destroy(isa->regs_16bit);
-       pmap_destroy(isa->regs_8bit);
-       pmap_destroy(isa->regs_8bit_high);
-       pmap_destroy(isa->tv_ent);
-       pmap_destroy(isa->types);
+       init_asm_constraints();
 
-#ifndef NDEBUG
-       obstack_free(isa->name_obst, NULL);
-#endif /* NDEBUG */
+       ia32_register_init();
+       ia32_create_opcodes(&ia32_irn_ops);
 
-       be_emit_exit();
+       ia32_mode_fpcw = new_int_mode("Fpcw", irma_twos_complement, 16, 0, 0);
 
-       free(self);
-}
+       /* note mantissa is 64bit but with explicitely encoded 1 so the really
+        * usable part as counted by firm is only 63 bits */
+       ia32_mode_E = new_float_mode("E", irma_x86_extended_float, 15, 63);
+       ia32_type_E = new_type_primitive(ia32_mode_E);
+       set_type_size_bytes(ia32_type_E, 12);
+       set_type_alignment_bytes(ia32_type_E, 16);
 
+       mode_long_long = new_int_mode("long long", irma_twos_complement, 64, 1, 64);
+       type_long_long = new_type_primitive(mode_long_long);
+       mode_unsigned_long_long
+               = new_int_mode("unsigned long long", irma_twos_complement, 64, 0, 64);
+       type_unsigned_long_long = new_type_primitive(mode_unsigned_long_long);
+
+       ia32_backend_params.type_long_long          = type_long_long;
+       ia32_backend_params.type_unsigned_long_long = type_unsigned_long_long;
+
+       if (ia32_cg_config.use_sse2 || ia32_cg_config.use_softfloat) {
+               ia32_backend_params.mode_float_arithmetic = NULL;
+               ia32_backend_params.type_long_double = NULL;
+       } else {
+               ia32_backend_params.mode_float_arithmetic = ia32_mode_E;
+               ia32_backend_params.type_long_double      = ia32_type_E;
+       }
+}
 
 /**
- * Return the number of register classes for this architecture.
- * We report always these:
- *  - the general purpose registers
- *  - the SSE floating point register set
- *  - the virtual floating point registers
- *  - the SSE vector register set
+ * The template that generates a new ISA object.
+ * Note that this template can be changed by command line
+ * arguments.
  */
-static unsigned ia32_get_n_reg_class(const void *self) {
-       (void) self;
-       return N_CLASSES;
+static ia32_isa_t ia32_isa_template = {
+       {
+               &ia32_isa_if,            /* isa interface implementation */
+               N_IA32_REGISTERS,
+               ia32_registers,
+               N_IA32_CLASSES,
+               ia32_reg_classes,
+               &ia32_registers[REG_ESP],  /* stack pointer register */
+               &ia32_registers[REG_EBP],  /* base pointer register */
+               &ia32_reg_classes[CLASS_ia32_gp],  /* static link pointer register class */
+               2,                       /* power of two stack alignment, 2^2 == 4 */
+               NULL,                    /* main environment */
+               7,                       /* costs for a spill instruction */
+               5,                       /* costs for a reload instruction */
+               false,                   /* no custom abi handling */
+       },
+       NULL,                    /* tv_ents */
+       IA32_FPU_ARCH_X87,       /* FPU architecture */
+};
+
+static arch_env_t *ia32_begin_codegeneration(const be_main_env_t *env)
+{
+       ia32_isa_t *isa = XMALLOC(ia32_isa_t);
+
+       set_tarval_output_modes();
+
+       *isa        = ia32_isa_template;
+       isa->tv_ent = pmap_create();
+
+       /* enter the ISA object into the intrinsic environment */
+       intrinsic_env.isa = isa;
+
+       be_emit_init(env->file_handle);
+       be_gas_begin_compilation_unit(env);
+
+       return &isa->base;
 }
 
 /**
- * Return the register class for index i.
+ * Closes the output file and frees the ISA structure.
  */
-static const arch_register_class_t *ia32_get_reg_class(const void *self,
-                                                       unsigned i)
+static void ia32_end_codegeneration(void *self)
 {
-       (void) self;
-       assert(i < N_CLASSES);
-       return &ia32_reg_classes[i];
+       ia32_isa_t *isa = (ia32_isa_t*)self;
+
+       /* emit now all global declarations */
+       be_gas_end_compilation_unit(isa->base.main_env);
+
+       be_emit_exit();
+
+       pmap_destroy(isa->tv_ent);
+       free(self);
 }
 
 /**
- * Get the register class which shall be used to store a value of a given mode.
- * @param self The this pointer.
- * @param mode The mode in question.
- * @return A register class which can hold values of the given mode.
+ * Returns the register for parameter nr.
  */
-const arch_register_class_t *ia32_get_reg_class_for_mode(const void *self,
-               const ir_mode *mode)
+static const arch_register_t *ia32_get_RegParam_reg(unsigned cc, unsigned nr,
+                                                    const ir_mode *mode)
 {
-       (void) self;
+       static const arch_register_t *gpreg_param_reg_fastcall[] = {
+               &ia32_registers[REG_ECX],
+               &ia32_registers[REG_EDX],
+               NULL
+       };
+       static const unsigned MAXNUM_GPREG_ARGS = 3;
+
+       static const arch_register_t *gpreg_param_reg_regparam[] = {
+               &ia32_registers[REG_EAX],
+               &ia32_registers[REG_EDX],
+               &ia32_registers[REG_ECX]
+       };
+
+       static const arch_register_t *gpreg_param_reg_this[] = {
+               &ia32_registers[REG_ECX],
+               NULL,
+               NULL
+       };
+
+       static const arch_register_t *fpreg_sse_param_reg_std[] = {
+               &ia32_registers[REG_XMM0],
+               &ia32_registers[REG_XMM1],
+               &ia32_registers[REG_XMM2],
+               &ia32_registers[REG_XMM3],
+               &ia32_registers[REG_XMM4],
+               &ia32_registers[REG_XMM5],
+               &ia32_registers[REG_XMM6],
+               &ia32_registers[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)) {
-               return ia32_cg_config.use_sse2 ? &ia32_reg_classes[CLASS_ia32_xmm] : &ia32_reg_classes[CLASS_ia32_vfp];
+               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;
        }
-       else
-               return &ia32_reg_classes[CLASS_ia32_gp];
+
+       panic("unknown argument mode");
 }
 
 /**
  * Get the ABI restrictions for procedure calls.
- * @param self        The this pointer.
- * @param method_type The type of the method (procedure) in question.
- * @param abi         The abi object to be modified
  */
-static void ia32_get_call_abi(const void *self, ir_type *method_type,
-                              be_abi_call_t *abi)
+static void ia32_get_call_abi(ir_type *method_type, be_abi_call_t *abi)
 {
-       ir_type  *tp;
-       ir_mode  *mode;
        unsigned  cc;
        int       n, i, regnum;
        int                 pop_amount = 0;
        be_abi_call_flags_t call_flags = be_abi_call_get_flags(abi);
 
-       (void) self;
-
        /* set abi flags for calls */
-       call_flags.bits.left_to_right         = 0;  /* always last arg first on stack */
        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          = 1;  /* 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);
        }
 
        n = get_method_n_params(method_type);
        for (i = regnum = 0; i < n; i++) {
-               ir_mode               *mode;
-               const arch_register_t *reg = NULL;
+               const arch_register_t *reg  = NULL;
+               ir_type               *tp   = get_method_param_type(method_type, i);
+               ir_mode               *mode = get_type_mode(tp);
 
-               tp   = get_method_param_type(method_type, i);
-               mode = get_type_mode(tp);
                if (mode != NULL) {
                        reg  = ia32_get_RegParam_reg(cc, regnum, mode);
                }
                if (reg != NULL) {
-                       be_abi_call_param_reg(abi, i, reg);
+                       be_abi_call_param_reg(abi, i, reg, ABI_CONTEXT_BOTH);
                        ++regnum;
                } else {
                        /* Micro optimisation: if the mode is shorter than 4 bytes, load 4 bytes.
@@ -1932,7 +1971,7 @@ static void ia32_get_call_abi(const void *self, ir_type *method_type,
                                if (size < 4) load_mode = mode_Iu;
                        }
 
-                       be_abi_call_param_stack(abi, i, load_mode, 4, 0, 0);
+                       be_abi_call_param_stack(abi, i, load_mode, 4, 0, 0, ABI_CONTEXT_BOTH);
                }
        }
 
@@ -1945,8 +1984,8 @@ static void ia32_get_call_abi(const void *self, ir_type *method_type,
 
        /* In case of 64bit returns, we will have two 32bit values */
        if (n == 2) {
-               tp   = get_method_res_type(method_type, 0);
-               mode = get_type_mode(tp);
+               ir_type *tp   = get_method_res_type(method_type, 0);
+               ir_mode *mode = get_type_mode(tp);
 
                assert(!mode_is_float(mode) && "two FP results not supported");
 
@@ -1955,426 +1994,199 @@ static void ia32_get_call_abi(const void *self, ir_type *method_type,
 
                assert(!mode_is_float(mode) && "mixed INT, FP results not supported");
 
-               be_abi_call_res_reg(abi, 0, &ia32_gp_regs[REG_EAX]);
-               be_abi_call_res_reg(abi, 1, &ia32_gp_regs[REG_EDX]);
+               be_abi_call_res_reg(abi, 0, &ia32_registers[REG_EAX], ABI_CONTEXT_BOTH);
+               be_abi_call_res_reg(abi, 1, &ia32_registers[REG_EDX], ABI_CONTEXT_BOTH);
        }
        else if (n == 1) {
+               ir_type *tp   = get_method_res_type(method_type, 0);
+               ir_mode *mode = get_type_mode(tp);
                const arch_register_t *reg;
-
-               tp   = get_method_res_type(method_type, 0);
                assert(is_atomic_type(tp));
-               mode = get_type_mode(tp);
 
-               reg = mode_is_float(mode) ? &ia32_vfp_regs[REG_VF0] : &ia32_gp_regs[REG_EAX];
+               reg = mode_is_float(mode) ? &ia32_registers[REG_VF0] : &ia32_registers[REG_EAX];
 
-               be_abi_call_res_reg(abi, 0, reg);
+               be_abi_call_res_reg(abi, 0, reg, ABI_CONTEXT_BOTH);
        }
 }
 
-int ia32_to_appear_in_schedule(void *block_env, const ir_node *irn)
+static void ia32_mark_remat(ir_node *node)
 {
-       (void) block_env;
-
-       if(!is_ia32_irn(irn)) {
-               return -1;
+       if (is_ia32_irn(node)) {
+               set_ia32_is_remat(node);
        }
-
-       if(is_ia32_NoReg_GP(irn) || is_ia32_NoReg_VFP(irn) || is_ia32_NoReg_XMM(irn)
-               || is_ia32_Unknown_GP(irn) || is_ia32_Unknown_XMM(irn)
-               || is_ia32_Unknown_VFP(irn) || is_ia32_ChangeCW(irn)
-               || is_ia32_Immediate(irn))
-               return 0;
-
-       return 1;
 }
 
-/**
- * Initializes the code generator interface.
- */
-static const arch_code_generator_if_t *ia32_get_code_generator_if(void *self)
+static asm_constraint_flags_t ia32_parse_asm_constraint(const char **c)
 {
-       (void) self;
-       return &ia32_code_gen_if;
-}
+       (void) c;
 
-/**
- * Returns the estimated execution time of an ia32 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;
+       /* we already added all our simple flags to the flags modifier list in
+        * init, so this flag we don't know. */
+       return ASM_CONSTRAINT_FLAG_INVALID;
 }
 
-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, list_sched_selector_t *selector)
+static int ia32_is_valid_clobber(const char *clobber)
 {
-       (void) self;
-       memcpy(&ia32_sched_selector, selector, sizeof(ia32_sched_selector));
-       ia32_sched_selector.exectime              = ia32_sched_exectime;
-       ia32_sched_selector.to_appear_in_schedule = ia32_to_appear_in_schedule;
-       return &ia32_sched_selector;
+       return ia32_get_clobber_register(clobber) != NULL;
 }
 
-static const ilp_sched_selector_t *ia32_get_ilp_sched_selector(const void *self)
+static void ia32_lower_for_target(void)
 {
-       (void) self;
-       return NULL;
-}
+       size_t i, n_irgs = get_irp_n_irgs();
 
-/**
- * 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)
-{
-       ir_mode *mode = arch_register_class_mode(cls);
-       int bytes     = get_mode_size_bytes(mode);
-       (void) self;
+       /* perform doubleword lowering */
+       lwrdw_param_t lower_dw_params = {
+               1,  /* little endian */
+               64, /* doubleword size */
+               ia32_create_intrinsic_fkt,
+               &intrinsic_env,
+       };
 
-       if (mode_is_float(mode) && bytes > 8)
-               return 16;
-       return bytes;
-}
+       ia32_create_opcodes(&ia32_irn_ops);
 
-static const be_execution_unit_t ***ia32_get_allowed_execution_units(
-               const void *self, const ir_node *irn)
-{
-       static const be_execution_unit_t *_allowed_units_BRANCH[] = {
-               &ia32_execution_units_BRANCH[IA32_EXECUNIT_TP_BRANCH_BRANCH1],
-               &ia32_execution_units_BRANCH[IA32_EXECUNIT_TP_BRANCH_BRANCH2],
-               NULL,
-       };
-       static const be_execution_unit_t *_allowed_units_GP[] = {
-               &ia32_execution_units_GP[IA32_EXECUNIT_TP_GP_GP_EAX],
-               &ia32_execution_units_GP[IA32_EXECUNIT_TP_GP_GP_EBX],
-               &ia32_execution_units_GP[IA32_EXECUNIT_TP_GP_GP_ECX],
-               &ia32_execution_units_GP[IA32_EXECUNIT_TP_GP_GP_EDX],
-               &ia32_execution_units_GP[IA32_EXECUNIT_TP_GP_GP_ESI],
-               &ia32_execution_units_GP[IA32_EXECUNIT_TP_GP_GP_EDI],
-               &ia32_execution_units_GP[IA32_EXECUNIT_TP_GP_GP_EBP],
-               NULL,
-       };
-       static const be_execution_unit_t *_allowed_units_DUMMY[] = {
-               &be_machine_execution_units_DUMMY[0],
-               NULL,
-       };
-       static const be_execution_unit_t **_units_callret[] = {
-               _allowed_units_BRANCH,
-               NULL
-       };
-       static const be_execution_unit_t **_units_other[] = {
-               _allowed_units_GP,
-               NULL
-       };
-       static const be_execution_unit_t **_units_dummy[] = {
-               _allowed_units_DUMMY,
-               NULL
-       };
-       const be_execution_unit_t ***ret;
-       (void) self;
+       /* lower compound param handling
+        * Note: we lower compound arguments ourself, since on ia32 we don't
+        * have hidden parameters but know where to find the structs on the stack.
+        * (This also forces us to always allocate space for the compound arguments
+        *  on the callframe and we can't just use an arbitrary position on the
+        *  stackframe)
+        */
+       lower_calls_with_compounds(LF_RETURN_HIDDEN | LF_DONT_LOWER_ARGUMENTS);
 
-       if (is_ia32_irn(irn)) {
-               ret = get_ia32_exec_units(irn);
-       }
-       else if (is_be_node(irn)) {
-               if (be_is_Call(irn) || be_is_Return(irn)) {
-                       ret = _units_callret;
-               }
-               else if (be_is_Barrier(irn)) {
-                       ret = _units_dummy;
-               }
-               else {
-                        ret = _units_other;
-               }
-       }
-       else {
-               ret = _units_dummy;
+       /* replace floating point operations by function calls */
+       if (ia32_cg_config.use_softfloat) {
+               lower_floating_point();
        }
 
-       return ret;
-}
+       ir_prepare_dw_lowering(&lower_dw_params);
+       ir_lower_dw_ops();
 
-/**
- * Return the abstract ia32 machine.
- */
-static const be_machine_t *ia32_get_machine(const void *self) {
-       const ia32_isa_t *isa = self;
-       return isa->cpu;
+       for (i = 0; i < n_irgs; ++i) {
+               ir_graph *irg = get_irp_irg(i);
+               /* lower for mode_b stuff */
+               ir_lower_mode_b(irg, mode_Iu);
+               /* break up switches with wide ranges */
+               lower_switch(irg, 4, 256, false);
+       }
+
+       for (i = 0; i < n_irgs; ++i) {
+               ir_graph *irg = get_irp_irg(i);
+               /* Turn all small CopyBs into loads/stores, keep medium-sized CopyBs,
+                * so we can generate rep movs later, and turn all big CopyBs into
+                * memcpy calls. */
+               lower_CopyB(irg, 64, 8193, true);
+       }
 }
 
 /**
- * Return irp irgs in the desired order.
+ * Returns the libFirm configuration parameter for this backend.
  */
-static ir_graph **ia32_get_irg_list(const void *self, ir_graph ***irg_list)
+static const backend_params *ia32_get_libfirm_params(void)
 {
-       (void) self;
-       (void) irg_list;
-       return NULL;
-}
-
-static void ia32_mark_remat(const void *self, ir_node *node) {
-       (void) self;
-       if (is_ia32_irn(node)) {
-               set_ia32_is_remat(node);
-       }
+       return &ia32_backend_params;
 }
 
 /**
- * Allows or disallows the creation of Psi nodes for the given Phi nodes.
- * @return 1 if allowed, 0 otherwise
+ * Check if the given register is callee or caller save.
  */
-static int ia32_is_psi_allowed(ir_node *sel, ir_node *phi_list, int i, int j)
+static int ia32_register_saved_by(const arch_register_t *reg, int callee)
 {
-       ir_node *phi;
-       ir_node *cmp = NULL;
-
-       /* we can't handle psis 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)
+       if (callee) {
+               /* check for callee saved */
+               if (reg->reg_class == &ia32_reg_classes[CLASS_ia32_gp]) {
+                       switch (reg->index) {
+                       case REG_GP_EBX:
+                       case REG_GP_ESI:
+                       case REG_GP_EDI:
+                       case REG_GP_EBP:
+                               return 1;
+                       default:
                                return 0;
-               } else {
-                       cmp = NULL;
-               }
-       }
-
-       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);
-
-                       /* 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)) {
-                                       /* 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;
-                                               } else if (cl == f && cr == t) {
-                                                       /* Psi(a <=/>= b, b, a) => MAX, MIN */
-                                                       res = 1;
-                                               }
-                                       }
-                                       if (! res)
-                                               return 0;
-
-                               } else if (get_mode_size_bits(mode) > 32)
-                                       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)
-                                       return 0;
                        }
                }
-
-               return 1;
        } else {
-               ir_node *cl, *cr;
-               pn_Cmp  pn;
-
-               /* No cmov, only some special cases */
-               if (cmp == NULL)
-                       return 0;
-
-               /* 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)
+               /* check for caller saved */
+               if (reg->reg_class == &ia32_reg_classes[CLASS_ia32_gp]) {
+                       switch (reg->index) {
+                       case REG_GP_EDX:
+                       case REG_GP_ECX:
+                       case REG_GP_EAX:
+                               return 1;
+                       default:
                                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;
-                               }
-                       } 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;
-#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 (cl == t && is_Minus(f) && get_Minus_op(f) == cl) {
-                                               /* Psi(a <=/>= 0 ? a : -a) Nabs/Abs */
-                                               res = 1;
-                                       } else if (cl == f && is_Minus(t) && get_Minus_op(t) == cl) {
-                                               /* Psi(a <=/>= 0 ? -a : a) Abs/Nabs */
-                                               res = 1;
-                                       }
-                               }
                        }
-                       if (! res)
-                               return 0;
+               } else if (reg->reg_class == &ia32_reg_classes[CLASS_ia32_xmm]) {
+                       /* all XMM registers are caller save */
+                       return reg->index != REG_XMM_NOREG;
+               } else if (reg->reg_class == &ia32_reg_classes[CLASS_ia32_vfp]) {
+                       /* all VFP registers are caller save */
+                       return reg->index != REG_VFP_NOREG;
                }
-               /* all checks passed */
-               return 1;
        }
        return 0;
 }
 
-static asm_constraint_flags_t ia32_parse_asm_constraint(const void *self, const char **c)
-{
-       (void) self;
-       (void) c;
-
-       /* we already added all our simple flags to the flags modifier list in
-        * init, so this flag we don't know. */
-       return ASM_CONSTRAINT_FLAG_INVALID;
-}
-
-static int ia32_is_valid_clobber(const void *self, const char *clobber)
-{
-       (void) self;
-
-       return ia32_get_clobber_register(clobber) != NULL;
-}
-
-/**
- * Returns the libFirm configuration parameter for this backend.
- */
-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 */
-       };
-       static const ir_settings_arch_dep_t ad = {
-               1,                   /* also use subs */
-               4,                   /* maximum shifts */
-               31,                  /* maximum shift amount */
-               ia32_evaluate_insn,  /* evaluate the instruction sequence */
-
-               1,  /* allow Mulhs */
-               1,  /* allow Mulus */
-               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 */
-       };
-
-       ia32_setup_cg_config();
-
-       /* doesn't really belong here, but this is the earliest place the backend
-        * is called... */
-       init_asm_constraints();
-
-       p.dep_param    = &ad;
-       p.if_conv_info = &ifconv;
-       return &p;
-}
-
 static const lc_opt_enum_int_items_t gas_items[] = {
-       { "elf",     GAS_FLAVOUR_ELF },
-       { "mingw",   GAS_FLAVOUR_MINGW  },
-       { "yasm",    GAS_FLAVOUR_YASM   },
-       { "macho",   GAS_FLAVOUR_MACH_O },
-       { NULL,      0 }
+       { "elf",   OBJECT_FILE_FORMAT_ELF    },
+       { "mingw", OBJECT_FILE_FORMAT_COFF   },
+       { "macho", OBJECT_FILE_FORMAT_MACH_O },
+       { NULL,    0 }
 };
 
 static lc_opt_enum_int_var_t gas_var = {
-       (int*) &be_gas_flavour, gas_items
+       (int*) &be_gas_object_file_format, 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),
-       LC_OPT_ENT_INT("stackalign", "set power of two stack alignment for calls",
-                      &ia32_isa_template.arch_env.stack_alignment),
+#endif
+       LC_OPT_ENT_INT ("stackalign", "set power of two stack alignment for calls",
+                       &ia32_isa_template.base.stack_alignment),
+       LC_OPT_ENT_BOOL("gprof",      "create gprof profiling code",                                    &gprof),
        LC_OPT_LAST
 };
 
 const arch_isa_if_t ia32_isa_if = {
        ia32_init,
-       ia32_done,
-       ia32_get_n_reg_class,
-       ia32_get_reg_class,
-       ia32_get_reg_class_for_mode,
-       ia32_get_call_abi,
-       ia32_get_code_generator_if,
-       ia32_get_list_sched_selector,
-       ia32_get_ilp_sched_selector,
-       ia32_get_reg_class_alignment,
        ia32_get_libfirm_params,
-       ia32_get_allowed_execution_units,
-       ia32_get_machine,
-       ia32_get_irg_list,
-       ia32_mark_remat,
+       ia32_lower_for_target,
        ia32_parse_asm_constraint,
-       ia32_is_valid_clobber
-};
+       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);
+       ia32_begin_codegeneration,
+       ia32_end_codegeneration,
+       ia32_init_graph,
+       ia32_get_call_abi,
+       ia32_mark_remat,
+       ia32_get_pic_base,   /* return node used as base in pic code addresses */
+       be_new_spill,
+       be_new_reload,
+       ia32_register_saved_by,
 
+       ia32_handle_intrinsics,
+       ia32_before_abi,     /* before abi introduce hook */
+       ia32_prepare_graph,
+       ia32_before_ra,      /* before register allocation hook */
+       ia32_finish,         /* called before codegen */
+       ia32_emit,           /* emit && done */
+};
+
+BE_REGISTER_MODULE_CONSTRUCTOR(be_init_arch_ia32)
 void be_init_arch_ia32(void)
 {
        lc_opt_entry_t *be_grp   = lc_opt_get_grp(firm_opt_get_root(), "be");
@@ -2383,8 +2195,6 @@ void be_init_arch_ia32(void)
        lc_opt_add_table(ia32_grp, ia32_options);
        be_register_isa_if("ia32", &ia32_isa_if);
 
-       FIRM_DBG_REGISTER(dbg, "firm.be.ia32.cg");
-
        ia32_init_emitter();
        ia32_init_finish();
        ia32_init_optimize();
@@ -2392,5 +2202,3 @@ void be_init_arch_ia32(void)
        ia32_init_x87();
        ia32_init_architecture();
 }
-
-BE_REGISTER_MODULE_CONSTRUCTOR(be_init_arch_ia32);