[libfirm] / ir / be / ia32 / ia32_transform.c

/*
 * Copyright (C) 1995-2011 University of Karlsruhe.  All right reserved.
 *
 * This file is part of libFirm.
 *
 * This file may be distributed and/or modified under the terms of the
 * GNU General Public License version 2 as published by the Free Software
 * Foundation and appearing in the file LICENSE.GPL included in the
 * packaging of this file.
 *
 * Licensees holding valid libFirm Professional Edition licenses may use
 * this file in accordance with the libFirm Commercial License.
 * Agreement provided with the Software.
 *
 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE.
 */

/**
 * @file
 * @brief       This file implements the IR transformation from firm into
 *              ia32-Firm.
 * @author      Christian Wuerdig, Matthias Braun
 */
#include "config.h"

#include <limits.h>
#include <stdbool.h>

#include "irargs_t.h"
#include "irnode_t.h"
#include "irgraph_t.h"
#include "irmode_t.h"
#include "iropt_t.h"
#include "irop_t.h"
#include "irprog_t.h"
#include "iredges_t.h"
#include "irgmod.h"
#include "ircons.h"
#include "irgwalk.h"
#include "irprintf.h"
#include "debug.h"
#include "irdom.h"
#include "iropt.h"
#include "error.h"
#include "array_t.h"
#include "heights.h"

#include "benode.h"
#include "besched.h"
#include "beabi.h"
#include "beutil.h"
#include "beirg.h"
#include "betranshlp.h"
#include "be_t.h"

#include "bearch_ia32_t.h"
#include "ia32_common_transform.h"
#include "ia32_nodes_attr.h"
#include "ia32_transform.h"
#include "ia32_new_nodes.h"
#include "ia32_dbg_stat.h"
#include "ia32_optimize.h"
#include "ia32_address_mode.h"
#include "ia32_architecture.h"

#include "gen_ia32_regalloc_if.h"

/* define this to construct SSE constants instead of load them */
#undef CONSTRUCT_SSE_CONST

#define mode_vfp    (ia32_reg_classes[CLASS_ia32_vfp].mode)
#define mode_xmm    (ia32_reg_classes[CLASS_ia32_xmm].mode)

DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)

static ir_node         *old_initial_fpcw = NULL;
static ir_node         *initial_fpcw = NULL;
int                     ia32_no_pic_adjust;

typedef ir_node *construct_binop_func(dbg_info *db, ir_node *block,
        ir_node *base, ir_node *index, ir_node *mem, ir_node *op1,
        ir_node *op2);

typedef ir_node *construct_binop_flags_func(dbg_info *db, ir_node *block,
        ir_node *base, ir_node *index, ir_node *mem, ir_node *op1, ir_node *op2,
        ir_node *flags);

typedef ir_node *construct_shift_func(dbg_info *db, ir_node *block,
        ir_node *op1, ir_node *op2);

typedef ir_node *construct_binop_dest_func(dbg_info *db, ir_node *block,
        ir_node *base, ir_node *index, ir_node *mem, ir_node *op);

typedef ir_node *construct_unop_dest_func(dbg_info *db, ir_node *block,
        ir_node *base, ir_node *index, ir_node *mem);

typedef ir_node *construct_binop_float_func(dbg_info *db, ir_node *block,
        ir_node *base, ir_node *index, ir_node *mem, ir_node *op1, ir_node *op2,
        ir_node *fpcw);

typedef ir_node *construct_unop_func(dbg_info *db, ir_node *block, ir_node *op);

static ir_node *create_immediate_or_transform(ir_node *node,
                                              char immediate_constraint_type);

static ir_node *create_I2I_Conv(ir_mode *src_mode, ir_mode *tgt_mode,
                                dbg_info *dbgi, ir_node *block,
                                ir_node *op, ir_node *orig_node);

/* its enough to have those once */
static ir_node *nomem, *noreg_GP;

/** a list to postprocess all calls */
static ir_node **call_list;
static ir_type **call_types;

/** Return non-zero is a node represents the 0 constant. */
static bool is_Const_0(ir_node *node)
{
        return is_Const(node) && is_Const_null(node);
}

/** Return non-zero is a node represents the 1 constant. */
static bool is_Const_1(ir_node *node)
{
        return is_Const(node) && is_Const_one(node);
}

/** Return non-zero is a node represents the -1 constant. */
static bool is_Const_Minus_1(ir_node *node)
{
        return is_Const(node) && is_Const_all_one(node);
}

/**
 * returns true if constant can be created with a simple float command
 */
static bool is_simple_x87_Const(ir_node *node)
{
        ir_tarval *tv = get_Const_tarval(node);
        if (tarval_is_null(tv) || tarval_is_one(tv))
                return true;

        /* TODO: match all the other float constants */
        return false;
}

/**
 * returns true if constant can be created with a simple float command
 */
static bool is_simple_sse_Const(ir_node *node)
{
        ir_tarval *tv   = get_Const_tarval(node);
        ir_mode   *mode = get_tarval_mode(tv);

        if (mode == mode_F)
                return true;

        if (tarval_is_null(tv)
#ifdef CONSTRUCT_SSE_CONST
            || tarval_is_one(tv)
#endif
           )
                return true;
#ifdef CONSTRUCT_SSE_CONST
        if (mode == mode_D) {
                unsigned val = get_tarval_sub_bits(tv, 0) |
                        (get_tarval_sub_bits(tv, 1) << 8) |
                        (get_tarval_sub_bits(tv, 2) << 16) |
                        (get_tarval_sub_bits(tv, 3) << 24);
                if (val == 0)
                        /* lower 32bit are zero, really a 32bit constant */
                        return true;
        }
#endif /* CONSTRUCT_SSE_CONST */
        /* TODO: match all the other float constants */
        return false;
}

/**
 * return NoREG or pic_base in case of PIC.
 * This is necessary as base address for newly created symbols
 */
static ir_node *get_symconst_base(void)
{
        ir_graph *irg = current_ir_graph;

        if (be_options.pic) {
                const arch_env_t *arch_env = be_get_irg_arch_env(irg);
                return arch_env->impl->get_pic_base(irg);
        }

        return noreg_GP;
}

/**
 * Transforms a Const.
 */
static ir_node *gen_Const(ir_node *node)
{
        ir_node   *old_block = get_nodes_block(node);
        ir_node   *block     = be_transform_node(old_block);
        dbg_info  *dbgi      = get_irn_dbg_info(node);
        ir_mode   *mode      = get_irn_mode(node);
        ir_tarval *tv        = get_Const_tarval(node);

        assert(is_Const(node));

        if (mode_is_float(mode)) {
                ir_graph         *irg      = get_irn_irg(node);
                const arch_env_t *arch_env = be_get_irg_arch_env(irg);
                ia32_isa_t       *isa      = (ia32_isa_t*) arch_env;
                ir_node          *res      = NULL;
                ir_node          *load;
                ir_entity        *floatent;

                if (ia32_cg_config.use_sse2) {
                        if (tarval_is_null(tv)) {
                                load = new_bd_ia32_xZero(dbgi, block);
                                set_ia32_ls_mode(load, mode);
                                res  = load;
#ifdef CONSTRUCT_SSE_CONST
                        } else if (tarval_is_one(tv)) {
                                int     cnst  = mode == mode_F ? 26 : 55;
                                ir_node *imm1 = ia32_create_Immediate(NULL, 0, cnst);
                                ir_node *imm2 = ia32_create_Immediate(NULL, 0, 2);
                                ir_node *pslld, *psrld;

                                load = new_bd_ia32_xAllOnes(dbgi, block);
                                set_ia32_ls_mode(load, mode);
                                pslld = new_bd_ia32_xPslld(dbgi, block, load, imm1);
                                set_ia32_ls_mode(pslld, mode);
                                psrld = new_bd_ia32_xPsrld(dbgi, block, pslld, imm2);
                                set_ia32_ls_mode(psrld, mode);
                                res = psrld;
#endif /* CONSTRUCT_SSE_CONST */
                        } else if (mode == mode_F) {
                                /* we can place any 32bit constant by using a movd gp, sse */
                                unsigned val = get_tarval_sub_bits(tv, 0) |
                                               (get_tarval_sub_bits(tv, 1) << 8) |
                                               (get_tarval_sub_bits(tv, 2) << 16) |
                                               (get_tarval_sub_bits(tv, 3) << 24);
                                ir_node *cnst = new_bd_ia32_Const(dbgi, block, NULL, 0, 0, val);
                                load = new_bd_ia32_xMovd(dbgi, block, cnst);
                                set_ia32_ls_mode(load, mode);
                                res = load;
                        } else {
                                ir_node *base;
#ifdef CONSTRUCT_SSE_CONST
                                if (mode == mode_D) {
                                        unsigned val = get_tarval_sub_bits(tv, 0) |
                                                (get_tarval_sub_bits(tv, 1) << 8) |
                                                (get_tarval_sub_bits(tv, 2) << 16) |
                                                (get_tarval_sub_bits(tv, 3) << 24);
                                        if (val == 0) {
                                                ir_node *imm32 = ia32_create_Immediate(NULL, 0, 32);
                                                ir_node *cnst, *psllq;

                                                /* fine, lower 32bit are zero, produce 32bit value */
                                                val = get_tarval_sub_bits(tv, 4) |
                                                        (get_tarval_sub_bits(tv, 5) << 8) |
                                                        (get_tarval_sub_bits(tv, 6) << 16) |
                                                        (get_tarval_sub_bits(tv, 7) << 24);
                                                cnst = new_bd_ia32_Const(dbgi, block, NULL, 0, 0, val);
                                                load = new_bd_ia32_xMovd(dbgi, block, cnst);
                                                set_ia32_ls_mode(load, mode);
                                                psllq = new_bd_ia32_xPsllq(dbgi, block, load, imm32);
                                                set_ia32_ls_mode(psllq, mode);
                                                res = psllq;
                                                goto end;
                                        }
                                }
#endif /* CONSTRUCT_SSE_CONST */
                                floatent = ia32_create_float_const_entity(isa, tv, NULL);

                                base     = get_symconst_base();
                                load     = new_bd_ia32_xLoad(dbgi, block, base, noreg_GP, nomem,
                                                             mode);
                                set_ia32_op_type(load, ia32_AddrModeS);
                                set_ia32_am_sc(load, floatent);
                                arch_add_irn_flags(load, arch_irn_flags_rematerializable);
                                res = new_r_Proj(load, mode_xmm, pn_ia32_xLoad_res);
                        }
                } else {
                        if (tarval_is_null(tv)) {
                                load = new_bd_ia32_vfldz(dbgi, block);
                                res  = load;
                                set_ia32_ls_mode(load, mode);
                        } else if (tarval_is_one(tv)) {
                                load = new_bd_ia32_vfld1(dbgi, block);
                                res  = load;
                                set_ia32_ls_mode(load, mode);
                        } else {
                                ir_mode *ls_mode;
                                ir_node *base;

                                floatent = ia32_create_float_const_entity(isa, tv, NULL);
                                /* create_float_const_ent is smart and sometimes creates
                                   smaller entities */
                                ls_mode  = get_type_mode(get_entity_type(floatent));
                                base     = get_symconst_base();
                                load     = new_bd_ia32_vfld(dbgi, block, base, noreg_GP, nomem,
                                                            ls_mode);
                                set_ia32_op_type(load, ia32_AddrModeS);
                                set_ia32_am_sc(load, floatent);
                                arch_add_irn_flags(load, arch_irn_flags_rematerializable);
                                res = new_r_Proj(load, mode_vfp, pn_ia32_vfld_res);
                        }
                }
#ifdef CONSTRUCT_SSE_CONST
end:
#endif /* CONSTRUCT_SSE_CONST */
                SET_IA32_ORIG_NODE(load, node);
                return res;
        } else { /* non-float mode */
                ir_node *cnst;
                long     val;

                tv = tarval_convert_to(tv, mode_Iu);

                if (tv == get_tarval_bad() || tv == get_tarval_undefined() ||
                    tv == NULL) {
                        panic("couldn't convert constant tarval (%+F)", node);
                }
                val = get_tarval_long(tv);

                cnst = new_bd_ia32_Const(dbgi, block, NULL, 0, 0, val);
                SET_IA32_ORIG_NODE(cnst, node);

                return cnst;
        }
}

/**
 * Transforms a SymConst.
 */
static ir_node *gen_SymConst(ir_node *node)
{
        ir_node  *old_block = get_nodes_block(node);
        ir_node  *block = be_transform_node(old_block);
        dbg_info *dbgi  = get_irn_dbg_info(node);
        ir_mode  *mode  = get_irn_mode(node);
        ir_node  *cnst;

        if (mode_is_float(mode)) {
                if (ia32_cg_config.use_sse2)
                        cnst = new_bd_ia32_xLoad(dbgi, block, noreg_GP, noreg_GP, nomem, mode_D);
                else
                        cnst = new_bd_ia32_vfld(dbgi, block, noreg_GP, noreg_GP, nomem, ia32_mode_E);
                set_ia32_am_sc(cnst, get_SymConst_entity(node));
                set_ia32_use_frame(cnst);
        } else {
                ir_entity *entity;

                if (get_SymConst_kind(node) != symconst_addr_ent) {
                        panic("backend only support symconst_addr_ent (at %+F)", node);
                }
                entity = get_SymConst_entity(node);
                if (get_entity_owner(entity) == get_tls_type()) {
                        ir_node *tls_base = new_bd_ia32_LdTls(NULL, block);
                        ir_node *lea     = new_bd_ia32_Lea(dbgi, block, tls_base, noreg_GP);
                        set_ia32_am_sc(lea, entity);
                        cnst = lea;
                } else {
                        cnst = new_bd_ia32_Const(dbgi, block, entity, 0, 0, 0);
                }
        }

        SET_IA32_ORIG_NODE(cnst, node);

        return cnst;
}

static ir_type *make_array_type(ir_type *tp)
{
        unsigned alignment = get_type_alignment_bytes(tp);
        unsigned size      = get_type_size_bytes(tp);
        ir_type *res = new_type_array(1, tp);
        set_type_alignment_bytes(res, alignment);
        set_array_bounds_int(res, 0, 0, 2);
        if (alignment > size)
                size = alignment;
        set_type_size_bytes(res, 2 * size);
        set_type_state(res, layout_fixed);
        return res;
}

/**
 * Create a float[2] array type for the given atomic type.
 *
 * @param tp  the atomic type
 */
static ir_type *ia32_create_float_array(ir_type *tp)
{
        ir_mode  *mode = get_type_mode(tp);
        ir_type  *arr;

        if (mode == mode_F) {
                static ir_type *float_F;

                arr = float_F;
                if (arr == NULL)
                        arr = float_F = make_array_type(tp);
        } else if (mode == mode_D) {
                static ir_type *float_D;

                arr = float_D;
                if (arr == NULL)
                        arr = float_D = make_array_type(tp);
        } else {
                static ir_type *float_E;

                arr = float_E;
                if (arr == NULL)
                        arr = float_E = make_array_type(tp);
        }
        return arr;
}

/* Generates an entity for a known FP const (used for FP Neg + Abs) */
ir_entity *ia32_gen_fp_known_const(ia32_known_const_t kct)
{
        static const struct {
                const char *name;
                const char *cnst_str;
                char        mode;
        } names [ia32_known_const_max] = {
                { "C_sfp_sign", "0x80000000",          0 },
                { "C_dfp_sign", "0x8000000000000000",  1 },
                { "C_sfp_abs",  "0x7FFFFFFF",          0 },
                { "C_dfp_abs",  "0x7FFFFFFFFFFFFFFF",  1 },
                { "C_ull_bias", "0x10000000000000000", 2 }
        };
        static ir_entity *ent_cache[ia32_known_const_max];

        ir_entity *ent = ent_cache[kct];

        if (ent == NULL) {
                ir_graph         *irg      = current_ir_graph;
                const arch_env_t *arch_env = be_get_irg_arch_env(irg);
                ia32_isa_t       *isa      = (ia32_isa_t*) arch_env;
                const char       *cnst_str = names[kct].cnst_str;
                ident            *name     = new_id_from_str(names[kct].name);
                ir_mode          *mode;
                ir_tarval        *tv;
                switch (names[kct].mode) {
                case 0:  mode = mode_Iu; break;
                case 1:  mode = mode_Lu; break;
                case 2:  mode = mode_F;  break;
                default: panic("internal compiler error");
                }
                tv = new_tarval_from_str(cnst_str, strlen(cnst_str), mode);

                if (kct == ia32_ULLBIAS) {
                        ir_type          *type  = ia32_get_prim_type(mode_F);
                        ir_type          *atype = ia32_create_float_array(type);
                        ir_initializer_t *initializer;

                        ent = new_entity(get_glob_type(), name, atype);

                        set_entity_ld_ident(ent, name);
                        set_entity_visibility(ent, ir_visibility_private);
                        add_entity_linkage(ent, IR_LINKAGE_CONSTANT);

                        initializer = create_initializer_compound(2);
                        set_initializer_compound_value(initializer, 0,
                                create_initializer_tarval(get_mode_null(mode)));
                        set_initializer_compound_value(initializer, 1,
                                create_initializer_tarval(tv));
                        set_entity_initializer(ent, initializer);
                } else {
                        ent = ia32_create_float_const_entity(isa, tv, name);
                }
                /* cache the entry */
                ent_cache[kct] = ent;
        }

        return ent_cache[kct];
}

/**
 * return true if the node is a Proj(Load) and could be used in source address
 * mode for another node. Will return only true if the @p other node is not
 * dependent on the memory of the Load (for binary operations use the other
 * input here, for unary operations use NULL).
 */
static int ia32_use_source_address_mode(ir_node *block, ir_node *node,
                                        ir_node *other, ir_node *other2,
                                        match_flags_t flags)
{
        ir_node *load;
        ir_mode *mode;
        long     pn;

        /* float constants are always available */
        if (is_Const(node)) {
                mode = get_irn_mode(node);
                if (mode_is_float(mode)) {
                        ir_tarval *tv = get_Const_tarval(node);
                        if (!tarval_ieee754_can_conv_lossless(tv, mode_D))
                                return 0;
                        if (ia32_cg_config.use_sse2) {
                                if (is_simple_sse_Const(node))
                                        return 0;
                        } else {
                                if (is_simple_x87_Const(node))
                                        return 0;
                        }
                        if (get_irn_n_edges(node) > 1)
                                return 0;
                        return 1;
                }
                return 0;
        }

        if (!is_Proj(node))
                return 0;
        load = get_Proj_pred(node);
        pn   = get_Proj_proj(node);
        if (!is_Load(load) || pn != pn_Load_res)
                return 0;
        if (get_nodes_block(load) != block)
                return 0;
        mode = get_irn_mode(node);
        /* we can't fold mode_E AM */
        if (mode == ia32_mode_E)
                return 0;
        /* we only use address mode if we're the only user of the load */
        if (get_irn_n_edges(node) != (flags & match_two_users ? 2 : 1))
                return 0;
        /* in some edge cases with address mode we might reach the load normally
         * and through some AM sequence, if it is already materialized then we
         * can't create an AM node from it */
        if (be_is_transformed(node))
                return 0;

        /* don't do AM if other node inputs depend on the load (via mem-proj) */
        if (other != NULL && ia32_prevents_AM(block, load, other))
                return 0;

        if (other2 != NULL && ia32_prevents_AM(block, load, other2))
                return 0;

        return 1;
}

typedef struct ia32_address_mode_t ia32_address_mode_t;
struct ia32_address_mode_t {
        ia32_address_t  addr;
        ir_mode        *ls_mode;
        ir_node        *mem_proj;
        ir_node        *am_node;
        ia32_op_type_t  op_type;
        ir_node        *new_op1;
        ir_node        *new_op2;
        op_pin_state    pinned;
        unsigned        commutative  : 1;
        unsigned        ins_permuted : 1;
};

static void build_address_ptr(ia32_address_t *addr, ir_node *ptr, ir_node *mem)
{
        /* construct load address */
        memset(addr, 0, sizeof(addr[0]));
        ia32_create_address_mode(addr, ptr, ia32_create_am_normal);

        addr->base  = addr->base  ? be_transform_node(addr->base)  : noreg_GP;
        addr->index = addr->index ? be_transform_node(addr->index) : noreg_GP;
        addr->mem   = be_transform_node(mem);
}

static void build_address(ia32_address_mode_t *am, ir_node *node,
                          ia32_create_am_flags_t flags)
{
        ia32_address_t *addr = &am->addr;
        ir_node        *load;
        ir_node        *ptr;
        ir_node        *mem;
        ir_node        *new_mem;

        /* floating point immediates */
        if (is_Const(node)) {
                ir_graph         *irg      = get_irn_irg(node);
                const arch_env_t *arch_env = be_get_irg_arch_env(irg);
                ia32_isa_t       *isa      = (ia32_isa_t*) arch_env;
                ir_tarval        *tv       = get_Const_tarval(node);
                ir_entity *entity  = ia32_create_float_const_entity(isa, tv, NULL);
                addr->base         = get_symconst_base();
                addr->index        = noreg_GP;
                addr->mem          = nomem;
                addr->symconst_ent = entity;
                addr->tls_segment  = false;
                addr->use_frame    = 1;
                am->ls_mode        = get_type_mode(get_entity_type(entity));
                am->pinned         = op_pin_state_floats;
                return;
        }

        load         = get_Proj_pred(node);
        ptr          = get_Load_ptr(load);
        mem          = get_Load_mem(load);
        new_mem      = be_transform_node(mem);
        am->pinned   = get_irn_pinned(load);
        am->ls_mode  = get_Load_mode(load);
        am->mem_proj = be_get_Proj_for_pn(load, pn_Load_M);
        am->am_node  = node;

        /* construct load address */
        ia32_create_address_mode(addr, ptr, flags);

        addr->base  = addr->base  ? be_transform_node(addr->base)  : noreg_GP;
        addr->index = addr->index ? be_transform_node(addr->index) : noreg_GP;
        addr->mem   = new_mem;
}

static void set_address(ir_node *node, const ia32_address_t *addr)
{
        set_ia32_am_scale(node, addr->scale);
        set_ia32_am_sc(node, addr->symconst_ent);
        set_ia32_am_offs_int(node, addr->offset);
        set_ia32_am_tls_segment(node, addr->tls_segment);
        if (addr->symconst_sign)
                set_ia32_am_sc_sign(node);
        if (addr->use_frame)
                set_ia32_use_frame(node);
        set_ia32_frame_ent(node, addr->frame_entity);
}

/**
 * Apply attributes of a given address mode to a node.
 */
static void set_am_attributes(ir_node *node, const ia32_address_mode_t *am)
{
        set_address(node, &am->addr);

        set_ia32_op_type(node, am->op_type);
        set_ia32_ls_mode(node, am->ls_mode);
        if (am->pinned == op_pin_state_pinned) {
                /* beware: some nodes are already pinned and did not allow to change the state */
                if (get_irn_pinned(node) != op_pin_state_pinned)
                        set_irn_pinned(node, op_pin_state_pinned);
        }
        if (am->commutative)
                set_ia32_commutative(node);
}

/**
 * Check, if a given node is a Down-Conv, ie. a integer Conv
 * from a mode with a mode with more bits to a mode with lesser bits.
 * Moreover, we return only true if the node has not more than 1 user.
 *
 * @param node   the node
 * @return non-zero if node is a Down-Conv
 */
static int is_downconv(const ir_node *node)
{
        ir_mode *src_mode;
        ir_mode *dest_mode;

        if (!is_Conv(node))
                return 0;

        src_mode  = get_irn_mode(get_Conv_op(node));
        dest_mode = get_irn_mode(node);
        return
                ia32_mode_needs_gp_reg(src_mode)  &&
                ia32_mode_needs_gp_reg(dest_mode) &&
                get_mode_size_bits(dest_mode) <= get_mode_size_bits(src_mode);
}

/** Skip all Down-Conv's on a given node and return the resulting node. */
ir_node *ia32_skip_downconv(ir_node *node)
{
        while (is_downconv(node)) {
                /* we only want to skip the conv when we're the only user
                 * (because this test is used in the context of address-mode selection
                 *  and we don't want to use address mode for multiple users) */
                if (get_irn_n_edges(node) > 1)
                        break;

                node = get_Conv_op(node);
        }

        return node;
}

static bool is_sameconv(ir_node *node)
{
        ir_mode *src_mode;
        ir_mode *dest_mode;

        if (!is_Conv(node))
                return 0;

        /* we only want to skip the conv when we're the only user
         * (because this test is used in the context of address-mode selection
         *  and we don't want to use address mode for multiple users) */
        if (get_irn_n_edges(node) > 1)
                return 0;

        src_mode  = get_irn_mode(get_Conv_op(node));
        dest_mode = get_irn_mode(node);
        return
                ia32_mode_needs_gp_reg(src_mode)  &&
                ia32_mode_needs_gp_reg(dest_mode) &&
                get_mode_size_bits(dest_mode) == get_mode_size_bits(src_mode);
}

/** Skip all signedness convs */
static ir_node *ia32_skip_sameconv(ir_node *node)
{
        while (is_sameconv(node)) {
                node = get_Conv_op(node);
        }

        return node;
}

static ir_node *transform_sext(ir_node *node, ir_node *orig_node)
{
        ir_mode  *mode  = get_irn_mode(node);
        ir_node  *block = get_nodes_block(node);
        dbg_info *dbgi  = get_irn_dbg_info(node);
        return create_I2I_Conv(mode, mode_Is, dbgi, block, node, orig_node);
}

static ir_node *transform_zext(ir_node *node, ir_node *orig_node)
{
        ir_mode  *mode  = get_irn_mode(node);
        ir_node  *block = get_nodes_block(node);
        dbg_info *dbgi  = get_irn_dbg_info(node);
        /* normalize to an unsigned mode */
        switch (get_mode_size_bits(mode)) {
        case 8:  mode = mode_Bu; break;
        case 16: mode = mode_Hu; break;
        default:
                panic("ia32: invalid mode in zest: %+F", node);
        }
        return create_I2I_Conv(mode, mode_Iu, dbgi, block, node, orig_node);
}

static ir_node *transform_upconv(ir_node *node, ir_node *orig_node)
{
        ir_mode *mode = get_irn_mode(node);
        if (mode_is_signed(mode)) {
                return transform_sext(node, orig_node);
        } else {
                return transform_zext(node, orig_node);
        }
}

/**
 * matches operands of a node into ia32 addressing/operand modes. This covers
 * usage of source address mode, immediates, operations with non 32-bit modes,
 * ...
 * The resulting data is filled into the @p am struct. block is the block
 * of the node whose arguments are matched. op1, op2 are the first and second
 * input that are matched (op1 may be NULL). other_op is another unrelated
 * input that is not matched! but which is needed sometimes to check if AM
 * for op1/op2 is legal.
 * @p flags describes the supported modes of the operation in detail.
 */
static void match_arguments(ia32_address_mode_t *am, ir_node *block,
                            ir_node *op1, ir_node *op2, ir_node *other_op,
                            match_flags_t flags)
{
        ia32_address_t *addr      = &am->addr;
        ir_mode        *mode      = get_irn_mode(op2);
        int             mode_bits = get_mode_size_bits(mode);
        ir_node        *new_op1, *new_op2;
        int             use_am;
        unsigned        commutative;
        int             use_am_and_immediates;
        int             use_immediate;

        memset(am, 0, sizeof(am[0]));

        commutative           = (flags & match_commutative) != 0;
        use_am_and_immediates = (flags & match_am_and_immediates) != 0;
        use_am                = (flags & match_am) != 0;
        use_immediate         = (flags & match_immediate) != 0;
        assert(!use_am_and_immediates || use_immediate);

        assert(op2 != NULL);
        assert(!commutative || op1 != NULL);
        assert(use_am || !(flags & match_8bit_am));
        assert(use_am || !(flags & match_16bit_am));

        if ((mode_bits ==  8 && !(flags & match_8bit_am)) ||
            (mode_bits == 16 && !(flags & match_16bit_am))) {
                use_am = 0;
        }

        /* we can simply skip downconvs for mode neutral nodes: the upper bits
         * can be random for these operations */
        if (flags & match_mode_neutral) {
                op2 = ia32_skip_downconv(op2);
                if (op1 != NULL) {
                        op1 = ia32_skip_downconv(op1);
                }
        } else {
                op2 = ia32_skip_sameconv(op2);
                if (op1 != NULL) {
                        op1 = ia32_skip_sameconv(op1);
                }
        }

        /* match immediates. firm nodes are normalized: constants are always on the
         * op2 input */
        new_op2 = NULL;
        if (!(flags & match_try_am) && use_immediate) {
                new_op2 = ia32_try_create_Immediate(op2, 0);
        }

        if (new_op2 == NULL &&
            use_am && ia32_use_source_address_mode(block, op2, op1, other_op, flags)) {
                build_address(am, op2, ia32_create_am_normal);
                new_op1     = (op1 == NULL ? NULL : be_transform_node(op1));
                if (mode_is_float(mode)) {
                        new_op2 = ia32_new_NoReg_vfp(current_ir_graph);
                } else {
                        new_op2 = noreg_GP;
                }
                am->op_type = ia32_AddrModeS;
        } else if (commutative && (new_op2 == NULL || use_am_and_immediates) &&
                       use_am &&
                       ia32_use_source_address_mode(block, op1, op2, other_op, flags)) {
                ir_node *noreg;
                build_address(am, op1, ia32_create_am_normal);

                if (mode_is_float(mode)) {
                        noreg = ia32_new_NoReg_vfp(current_ir_graph);
                } else {
                        noreg = noreg_GP;
                }

                if (new_op2 != NULL) {
                        new_op1 = noreg;
                } else {
                        new_op1 = be_transform_node(op2);
                        new_op2 = noreg;
                        am->ins_permuted = true;
                }
                am->op_type = ia32_AddrModeS;
        } else {
                am->op_type = ia32_Normal;

                if (flags & match_try_am) {
                        am->new_op1 = NULL;
                        am->new_op2 = NULL;
                        return;
                }

                mode = get_irn_mode(op2);
                if (get_mode_size_bits(mode) != 32
                        && (flags & (match_mode_neutral | match_upconv | match_zero_ext))) {
                        if (flags & match_upconv) {
                                new_op1 = (op1 == NULL ? NULL : transform_upconv(op1, op1));
                                if (new_op2 == NULL)
                                        new_op2 = transform_upconv(op2, op2);
                        } else if (flags & match_zero_ext) {
                                new_op1 = (op1 == NULL ? NULL : transform_zext(op1, op1));
                                if (new_op2 == NULL)
                                        new_op2 = transform_zext(op2, op2);
                        } else {
                                new_op1 = (op1 == NULL ? NULL : be_transform_node(op1));
                                if (new_op2 == NULL)
                                        new_op2 = be_transform_node(op2);
                                assert(flags & match_mode_neutral);
                        }
                        mode = mode_Iu;
                } else {
                        new_op1 = (op1 == NULL ? NULL : be_transform_node(op1));
                        if (new_op2 == NULL)
                                new_op2 = be_transform_node(op2);
                }
                am->ls_mode = mode;
        }
        if (addr->base == NULL)
                addr->base = noreg_GP;
        if (addr->index == NULL)
                addr->index = noreg_GP;
        if (addr->mem == NULL)
                addr->mem = nomem;

        am->new_op1     = new_op1;
        am->new_op2     = new_op2;
        am->commutative = commutative;
}

/**
 * "Fixes" a node that uses address mode by turning it into mode_T
 * and returning a pn_ia32_res Proj.
 *
 * @param node  the node
 * @param am    its address mode
 *
 * @return a Proj(pn_ia32_res) if a memory address mode is used,
 *         node else
 */
static ir_node *fix_mem_proj(ir_node *node, ia32_address_mode_t *am)
{
        ir_mode  *mode;
        ir_node  *load;

        if (am->mem_proj == NULL)
                return node;

        /* we have to create a mode_T so the old MemProj can attach to us */
        mode = get_irn_mode(node);
        load = get_Proj_pred(am->mem_proj);

        be_set_transformed_node(load, node);

        if (mode != mode_T) {
                set_irn_mode(node, mode_T);
                return new_rd_Proj(NULL, node, mode, pn_ia32_res);
        } else {
                return node;
        }
}

/**
 * Construct a standard binary operation, set AM and immediate if required.
 *
 * @param node  The original node for which the binop is created
 * @param op1   The first operand
 * @param op2   The second operand
 * @param func  The node constructor function
 * @return The constructed ia32 node.
 */
static ir_node *gen_binop(ir_node *node, ir_node *op1, ir_node *op2,
                          construct_binop_func *func, match_flags_t flags)
{
        dbg_info            *dbgi;
        ir_node             *block, *new_block, *new_node;
        ia32_address_mode_t  am;
        ia32_address_t      *addr = &am.addr;

        block = get_nodes_block(node);
        match_arguments(&am, block, op1, op2, NULL, flags);

        dbgi      = get_irn_dbg_info(node);
        new_block = be_transform_node(block);
        new_node  = func(dbgi, new_block, addr->base, addr->index, addr->mem,
                        am.new_op1, am.new_op2);
        set_am_attributes(new_node, &am);
        /* we can't use source address mode anymore when using immediates */
        if (!(flags & match_am_and_immediates) &&
            (is_ia32_Immediate(am.new_op1) || is_ia32_Immediate(am.new_op2)))
                set_ia32_am_support(new_node, ia32_am_none);
        SET_IA32_ORIG_NODE(new_node, node);

        new_node = fix_mem_proj(new_node, &am);

        return new_node;
}

/**
 * Generic names for the inputs of an ia32 binary op.
 */
enum {
        n_ia32_l_binop_left,  /**< ia32 left input */
        n_ia32_l_binop_right, /**< ia32 right input */
        n_ia32_l_binop_eflags /**< ia32 eflags input */
};
COMPILETIME_ASSERT((int)n_ia32_l_binop_left   == (int)n_ia32_l_Adc_left,       n_Adc_left)
COMPILETIME_ASSERT((int)n_ia32_l_binop_right  == (int)n_ia32_l_Adc_right,      n_Adc_right)
COMPILETIME_ASSERT((int)n_ia32_l_binop_eflags == (int)n_ia32_l_Adc_eflags,     n_Adc_eflags)
COMPILETIME_ASSERT((int)n_ia32_l_binop_left   == (int)n_ia32_l_Sbb_minuend,    n_Sbb_minuend)
COMPILETIME_ASSERT((int)n_ia32_l_binop_right  == (int)n_ia32_l_Sbb_subtrahend, n_Sbb_subtrahend)
COMPILETIME_ASSERT((int)n_ia32_l_binop_eflags == (int)n_ia32_l_Sbb_eflags,     n_Sbb_eflags)

/**
 * Construct a binary operation which also consumes the eflags.
 *
 * @param node  The node to transform
 * @param func  The node constructor function
 * @param flags The match flags
 * @return      The constructor ia32 node
 */
static ir_node *gen_binop_flags(ir_node *node, construct_binop_flags_func *func,
                                match_flags_t flags)
{
        ir_node             *src_block  = get_nodes_block(node);
        ir_node             *op1        = get_irn_n(node, n_ia32_l_binop_left);
        ir_node             *op2        = get_irn_n(node, n_ia32_l_binop_right);
        ir_node             *eflags     = get_irn_n(node, n_ia32_l_binop_eflags);
        dbg_info            *dbgi;
        ir_node             *block, *new_node, *new_eflags;
        ia32_address_mode_t  am;
        ia32_address_t      *addr       = &am.addr;

        match_arguments(&am, src_block, op1, op2, eflags, flags);

        dbgi       = get_irn_dbg_info(node);
        block      = be_transform_node(src_block);
        new_eflags = be_transform_node(eflags);
        new_node   = func(dbgi, block, addr->base, addr->index, addr->mem,
                          am.new_op1, am.new_op2, new_eflags);
        set_am_attributes(new_node, &am);
        /* we can't use source address mode anymore when using immediates */
        if (!(flags & match_am_and_immediates) &&
            (is_ia32_Immediate(am.new_op1) || is_ia32_Immediate(am.new_op2)))
                set_ia32_am_support(new_node, ia32_am_none);
        SET_IA32_ORIG_NODE(new_node, node);

        new_node = fix_mem_proj(new_node, &am);

        return new_node;
}

static ir_node *get_fpcw(void)
{
        if (initial_fpcw != NULL)
                return initial_fpcw;

        initial_fpcw = be_transform_node(old_initial_fpcw);
        return initial_fpcw;
}

static ir_node *skip_float_upconv(ir_node *node)
{
        ir_mode *mode = get_irn_mode(node);
        assert(mode_is_float(mode));

        while (is_Conv(node)) {
                ir_node *pred      = get_Conv_op(node);
                ir_mode *pred_mode = get_irn_mode(pred);

                /**
                 * suboptimal, but without this check the address mode matcher
                 * can incorrectly think that something has only 1 user
                 */
                if (get_irn_n_edges(node) > 1)
                        break;

                if (!mode_is_float(pred_mode)
                        || get_mode_size_bits(pred_mode) > get_mode_size_bits(mode))
                        break;
                node = pred;
                mode = pred_mode;
        }
        return node;
}

static void check_x87_floatmode(ir_mode *mode)
{
        if (mode != ia32_mode_E) {
                panic("ia32: x87 only supports x86 extended float mode");
        }
}

/**
 * Construct a standard binary operation, set AM and immediate if required.
 *
 * @param op1   The first operand
 * @param op2   The second operand
 * @param func  The node constructor function
 * @return The constructed ia32 node.
 */
static ir_node *gen_binop_x87_float(ir_node *node, ir_node *op1, ir_node *op2,
                                    construct_binop_float_func *func)
{
        dbg_info            *dbgi;
        ir_node             *block;
        ir_node             *new_block;
        ir_node             *new_node;
        ia32_address_mode_t  am;
        ia32_address_t      *addr = &am.addr;
        ia32_x87_attr_t     *attr;
        /* All operations are considered commutative, because there are reverse
         * variants */
        match_flags_t        flags = match_commutative | match_am;
        ir_mode             *mode
                = is_Div(node) ? get_Div_resmode(node) : get_irn_mode(node);
        check_x87_floatmode(mode);

        op1 = skip_float_upconv(op1);
        op2 = skip_float_upconv(op2);

        block = get_nodes_block(node);
        match_arguments(&am, block, op1, op2, NULL, flags);

        dbgi      = get_irn_dbg_info(node);
        new_block = be_transform_node(block);
        new_node  = func(dbgi, new_block, addr->base, addr->index, addr->mem,
                         am.new_op1, am.new_op2, get_fpcw());
        set_am_attributes(new_node, &am);

        attr = get_ia32_x87_attr(new_node);
        attr->attr.data.ins_permuted = am.ins_permuted;

        SET_IA32_ORIG_NODE(new_node, node);

        new_node = fix_mem_proj(new_node, &am);

        return new_node;
}

/**
 * Construct a shift/rotate binary operation, sets AM and immediate if required.
 *
 * @param op1   The first operand
 * @param op2   The second operand
 * @param func  The node constructor function
 * @return The constructed ia32 node.
 */
static ir_node *gen_shift_binop(ir_node *node, ir_node *op1, ir_node *op2,
                                construct_shift_func *func,
                                match_flags_t flags)
{
        ir_mode *mode = get_irn_mode(node);

        assert(! mode_is_float(mode));
        assert(flags & match_immediate);
        assert((flags & ~(match_mode_neutral | match_zero_ext | match_upconv | match_immediate)) == 0);

        if (get_mode_modulo_shift(mode) != 32) {
                /* TODO: implement special cases for non-modulo shifts */
                panic("modulo shift!=32 not supported by ia32 backend");
        }

        ir_node *new_op1;
        ir_node *new_op2;
        if (flags & match_mode_neutral) {
                op1     = ia32_skip_downconv(op1);
                new_op1 = be_transform_node(op1);
        } else {
                op1 = ia32_skip_sameconv(op1);
                if (get_mode_size_bits(mode) != 32) {
                        if (flags & match_upconv) {
                                new_op1 = transform_upconv(op1, node);
                        } else if (flags & match_zero_ext) {
                                new_op1 = transform_zext(op1, node);
                        } else {
                                /* match_mode_neutral not handled here because it makes no
                                 * sense for shift operations */
                                panic("ia32 code selection failed for %+F", node);
                        }
                } else {
                        new_op1 = be_transform_node(op1);
                }
        }

        /* the shift amount can be any mode that is bigger than 5 bits, since all
         * other bits are ignored anyway */
        while (is_Conv(op2) && get_irn_n_edges(op2) == 1) {
                ir_node *const op = get_Conv_op(op2);
                if (mode_is_float(get_irn_mode(op)))
                        break;
                op2 = op;
                assert(get_mode_size_bits(get_irn_mode(op2)) >= 5);
        }
        new_op2 = create_immediate_or_transform(op2, 0);

        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_node  *block     = get_nodes_block(node);
        ir_node  *new_block = be_transform_node(block);
        ir_node  *new_node  = func(dbgi, new_block, new_op1, new_op2);
        SET_IA32_ORIG_NODE(new_node, node);

        /* lowered shift instruction may have a dependency operand, handle it here */
        if (get_irn_arity(node) == 3) {
                /* we have a dependency */
                ir_node* dep = get_irn_n(node, 2);
                if (get_irn_n_edges(dep) > 1) {
                        /* ... which has at least one user other than 'node' */
                        ir_node *new_dep = be_transform_node(dep);
                        add_irn_dep(new_node, new_dep);
                }
        }

        return new_node;
}


/**
 * Construct a standard unary operation, set AM and immediate if required.
 *
 * @param op    The operand
 * @param func  The node constructor function
 * @return The constructed ia32 node.
 */
static ir_node *gen_unop(ir_node *node, ir_node *op, construct_unop_func *func,
                         match_flags_t flags)
{
        dbg_info *dbgi;
        ir_node  *block, *new_block, *new_op, *new_node;

        assert(flags == 0 || flags == match_mode_neutral);
        if (flags & match_mode_neutral) {
                op = ia32_skip_downconv(op);
        }

        new_op    = be_transform_node(op);
        dbgi      = get_irn_dbg_info(node);
        block     = get_nodes_block(node);
        new_block = be_transform_node(block);
        new_node  = func(dbgi, new_block, new_op);

        SET_IA32_ORIG_NODE(new_node, node);

        return new_node;
}

static ir_node *create_lea_from_address(dbg_info *dbgi, ir_node *block,
                                        ia32_address_t *addr)
{
        ir_node *base;
        ir_node *idx;
        ir_node *res;

        base = addr->base;
        if (base == NULL) {
                base = noreg_GP;
        } else {
                base = be_transform_node(base);
        }

        idx = addr->index;
        if (idx == NULL) {
                idx = noreg_GP;
        } else {
                idx = be_transform_node(idx);
        }

        /* segment overrides are ineffective for Leas :-( so we have to patch
         * around... */
        if (addr->tls_segment) {
                ir_node *tls_base = new_bd_ia32_LdTls(NULL, block);
                assert(addr->symconst_ent != NULL);
                if (base == noreg_GP)
                        base = tls_base;
                else
                        base = new_bd_ia32_Lea(dbgi, block, tls_base, base);
                addr->tls_segment = false;
        }

        res = new_bd_ia32_Lea(dbgi, block, base, idx);
        set_address(res, addr);

        return res;
}

/**
 * Returns non-zero if a given address mode has a symbolic or
 * numerical offset != 0.
 */
static int am_has_immediates(const ia32_address_t *addr)
{
        return addr->offset != 0 || addr->symconst_ent != NULL
                || addr->frame_entity || addr->use_frame;
}

typedef ir_node* (*new_shiftd_func)(dbg_info *dbgi, ir_node *block,
                                    ir_node *high, ir_node *low,
                                    ir_node *count);

/**
 * Transforms a l_ShlD/l_ShrD into a ShlD/ShrD. Those nodes have 3 data inputs:
 * op1 - target to be shifted
 * op2 - contains bits to be shifted into target
 * op3 - shift count
 * Only op3 can be an immediate.
 */
static ir_node *gen_64bit_shifts(dbg_info *dbgi, ir_node *block,
                                 ir_node *high, ir_node *low, ir_node *count,
                                 new_shiftd_func func)
{
        ir_node  *new_block = be_transform_node(block);
        ir_node  *new_high  = be_transform_node(high);
        ir_node  *new_low   = be_transform_node(low);
        ir_node  *new_count;
        ir_node  *new_node;

        /* the shift amount can be any mode that is bigger than 5 bits, since all
         * other bits are ignored anyway */
        while (is_Conv(count)              &&
               get_irn_n_edges(count) == 1 &&
               mode_is_int(get_irn_mode(count))) {
                assert(get_mode_size_bits(get_irn_mode(count)) >= 5);
                count = get_Conv_op(count);
        }
        new_count = create_immediate_or_transform(count, 0);

        new_node = func(dbgi, new_block, new_high, new_low, new_count);
        return new_node;
}

/**
 * test wether 2 values result in 'x' and '32-x' when interpreted as a shift
 * value.
 */
static bool is_complementary_shifts(ir_node *value1, ir_node *value2)
{
        if (is_Const(value1) && is_Const(value2)) {
                ir_tarval *tv1 = get_Const_tarval(value1);
                ir_tarval *tv2 = get_Const_tarval(value2);
                if (tarval_is_long(tv1) && tarval_is_long(tv2)) {
                        long v1 = get_tarval_long(tv1);
                        long v2 = get_tarval_long(tv2);
                        return v1 <= v2 && v2 == 32-v1;
                }
        }
        return false;
}

static ir_node *match_64bit_shift(ir_node *node)
{
        ir_node *op1 = get_binop_left(node);
        ir_node *op2 = get_binop_right(node);
        assert(is_Or(node) || is_Add(node));

        if (is_Shr(op1)) {
                ir_node *tmp = op1;
                op1 = op2;
                op2 = tmp;
        }

        /* match ShlD operation */
        if (is_Shl(op1) && is_Shr(op2)) {
                ir_node *shl_right = get_Shl_right(op1);
                ir_node *shl_left  = get_Shl_left(op1);
                ir_node *shr_right = get_Shr_right(op2);
                ir_node *shr_left  = get_Shr_left(op2);
                /* constant ShlD operation */
                if (is_complementary_shifts(shl_right, shr_right)) {
                        dbg_info *dbgi  = get_irn_dbg_info(node);
                        ir_node  *block = get_nodes_block(node);
                        return gen_64bit_shifts(dbgi, block, shl_left, shr_left, shl_right,
                                                new_bd_ia32_ShlD);
                }
                /* constant ShrD operation */
                if (is_complementary_shifts(shr_right, shl_right)) {
                        dbg_info *dbgi  = get_irn_dbg_info(node);
                        ir_node  *block = get_nodes_block(node);
                        return gen_64bit_shifts(dbgi, block, shr_left, shl_left, shr_right,
                                                new_bd_ia32_ShrD);
                }
                /* lower_dw produces the following for ShlD:
                 * Or(Shr(Shr(high,1),Not(c)),Shl(low,c)) */
                if (is_Shr(shr_left) && is_Not(shr_right)
                        && is_Const_1(get_Shr_right(shr_left))
                    && get_Not_op(shr_right) == shl_right) {
                        dbg_info *dbgi  = get_irn_dbg_info(node);
                        ir_node  *block = get_nodes_block(node);
                        ir_node  *val_h = get_Shr_left(shr_left);
                        return gen_64bit_shifts(dbgi, block, shl_left, val_h, shl_right,
                                                new_bd_ia32_ShlD);
                }
                /* lower_dw produces the following for ShrD:
                 * Or(Shl(Shl(high,1),Not(c)), Shr(low,c)) */
                if (is_Shl(shl_left) && is_Not(shl_right)
                    && is_Const_1(get_Shl_right(shl_left))
                    && get_Not_op(shl_right) == shr_right) {
                        dbg_info *dbgi  = get_irn_dbg_info(node);
                        ir_node  *block = get_nodes_block(node);
                        ir_node  *val_h = get_Shl_left(shl_left);
                    return gen_64bit_shifts(dbgi, block, shr_left, val_h, shr_right,
                                            new_bd_ia32_ShrD);
                }
        }

        return NULL;
}

/**
 * Creates an ia32 Add.
 *
 * @return the created ia32 Add node
 */
static ir_node *gen_Add(ir_node *node)
{
        ir_mode  *mode = get_irn_mode(node);
        ir_node  *op1  = get_Add_left(node);
        ir_node  *op2  = get_Add_right(node);
        dbg_info *dbgi;
        ir_node  *block, *new_block, *new_node, *add_immediate_op;
        ia32_address_t       addr;
        ia32_address_mode_t  am;

        new_node = match_64bit_shift(node);
        if (new_node != NULL)
                return new_node;

        if (mode_is_float(mode)) {
                if (ia32_cg_config.use_sse2)
                        return gen_binop(node, op1, op2, new_bd_ia32_xAdd,
                                         match_commutative | match_am);
                else
                        return gen_binop_x87_float(node, op1, op2, new_bd_ia32_vfadd);
        }

        ia32_mark_non_am(node);

        /**
         * Rules for an Add:
         *   0. Immediate Trees (example Add(Symconst, Const) -> Const)
         *   1. Add with immediate -> Lea
         *   2. Add with possible source address mode -> Add
         *   3. Otherwise -> Lea
         */
        memset(&addr, 0, sizeof(addr));
        ia32_create_address_mode(&addr, node, ia32_create_am_force);
        add_immediate_op = NULL;

        dbgi      = get_irn_dbg_info(node);
        block     = get_nodes_block(node);
        new_block = be_transform_node(block);

        /* a constant? */
        if (addr.base == NULL && addr.index == NULL) {
                new_node = new_bd_ia32_Const(dbgi, new_block, addr.symconst_ent,
                                             addr.symconst_sign, 0, addr.offset);
                SET_IA32_ORIG_NODE(new_node, node);
                return new_node;
        }
        /* add with immediate? */
        if (addr.index == NULL) {
                add_immediate_op = addr.base;
        } else if (addr.base == NULL && addr.scale == 0) {
                add_immediate_op = addr.index;
        }

        if (add_immediate_op != NULL) {
                if (!am_has_immediates(&addr)) {
#ifdef DEBUG_libfirm
                        ir_fprintf(stderr, "Optimisation warning Add x,0 (%+F) found\n",
                                           node);
#endif
                        return be_transform_node(add_immediate_op);
                }

                new_node = create_lea_from_address(dbgi, new_block, &addr);
                SET_IA32_ORIG_NODE(new_node, node);
                return new_node;
        }

        /* test if we can use source address mode */
        match_arguments(&am, block, op1, op2, NULL, match_commutative
                        | match_mode_neutral | match_am | match_immediate | match_try_am);

        /* construct an Add with source address mode */
        if (am.op_type == ia32_AddrModeS) {
                ia32_address_t *am_addr = &am.addr;
                new_node = new_bd_ia32_Add(dbgi, new_block, am_addr->base,
                                         am_addr->index, am_addr->mem, am.new_op1,
                                         am.new_op2);
                set_am_attributes(new_node, &am);
                SET_IA32_ORIG_NODE(new_node, node);

                new_node = fix_mem_proj(new_node, &am);

                return new_node;
        }

        /* otherwise construct a lea */
        new_node = create_lea_from_address(dbgi, new_block, &addr);
        SET_IA32_ORIG_NODE(new_node, node);
        return new_node;
}

/**
 * Creates an ia32 Mul.
 *
 * @return the created ia32 Mul node
 */
static ir_node *gen_Mul(ir_node *node)
{
        ir_node *op1  = get_Mul_left(node);
        ir_node *op2  = get_Mul_right(node);
        ir_mode *mode = get_irn_mode(node);

        if (mode_is_float(mode)) {
                if (ia32_cg_config.use_sse2)
                        return gen_binop(node, op1, op2, new_bd_ia32_xMul,
                                         match_commutative | match_am);
                else
                        return gen_binop_x87_float(node, op1, op2, new_bd_ia32_vfmul);
        }
        return gen_binop(node, op1, op2, new_bd_ia32_IMul,
                         match_commutative | match_am | match_mode_neutral |
                         match_immediate | match_am_and_immediates);
}

/**
 * Creates an ia32 Mulh.
 * Note: Mul produces a 64Bit result and Mulh returns the upper 32 bit of
 * this result while Mul returns the lower 32 bit.
 *
 * @return the created ia32 Mulh node
 */
static ir_node *gen_Mulh(ir_node *node)
{
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_node  *op1       = get_Mulh_left(node);
        ir_node  *op2       = get_Mulh_right(node);
        ir_mode  *mode      = get_irn_mode(node);
        ir_node  *new_node;
        ir_node  *proj_res_high;

        if (get_mode_size_bits(mode) != 32) {
                panic("Mulh without 32bit size not supported in ia32 backend (%+F)", node);
        }

        if (mode_is_signed(mode)) {
                new_node = gen_binop(node, op1, op2, new_bd_ia32_IMul1OP, match_commutative | match_am);
                proj_res_high = new_rd_Proj(dbgi, new_node, mode_Iu, pn_ia32_IMul1OP_res_high);
        } else {
                new_node = gen_binop(node, op1, op2, new_bd_ia32_Mul, match_commutative | match_am);
                proj_res_high = new_rd_Proj(dbgi, new_node, mode_Iu, pn_ia32_Mul_res_high);
        }
        return proj_res_high;
}

/**
 * Creates an ia32 And.
 *
 * @return The created ia32 And node
 */
static ir_node *gen_And(ir_node *node)
{
        ir_node *op1 = get_And_left(node);
        ir_node *op2 = get_And_right(node);
        assert(! mode_is_float(get_irn_mode(node)));

        /* is it a zero extension? */
        if (is_Const(op2)) {
                ir_tarval *tv = get_Const_tarval(op2);
                long       v  = get_tarval_long(tv);

                if (v == 0xFF || v == 0xFFFF) {
                        dbg_info *dbgi   = get_irn_dbg_info(node);
                        ir_node  *block  = get_nodes_block(node);
                        ir_mode  *src_mode;
                        ir_node  *res;

                        if (v == 0xFF) {
                                src_mode = mode_Bu;
                        } else {
                                assert(v == 0xFFFF);
                                src_mode = mode_Hu;
                        }
                        res = create_I2I_Conv(src_mode, mode_Iu, dbgi, block, op1, node);

                        return res;
                }
        }
        return gen_binop(node, op1, op2, new_bd_ia32_And,
                        match_commutative | match_mode_neutral | match_am | match_immediate);
}

/**
 * Creates an ia32 Or.
 *
 * @return The created ia32 Or node
 */
static ir_node *gen_Or(ir_node *node)
{
        ir_node *op1 = get_Or_left(node);
        ir_node *op2 = get_Or_right(node);
        ir_node *res;

        res = match_64bit_shift(node);
        if (res != NULL)
                return res;

        assert (! mode_is_float(get_irn_mode(node)));
        return gen_binop(node, op1, op2, new_bd_ia32_Or, match_commutative
                        | match_mode_neutral | match_am | match_immediate);
}



/**
 * Creates an ia32 Eor.
 *
 * @return The created ia32 Eor node
 */
static ir_node *gen_Eor(ir_node *node)
{
        ir_node *op1 = get_Eor_left(node);
        ir_node *op2 = get_Eor_right(node);

        assert(! mode_is_float(get_irn_mode(node)));
        return gen_binop(node, op1, op2, new_bd_ia32_Xor, match_commutative
                        | match_mode_neutral | match_am | match_immediate);
}


/**
 * Creates an ia32 Sub.
 *
 * @return The created ia32 Sub node
 */
static ir_node *gen_Sub(ir_node *node)
{
        ir_node  *op1  = get_Sub_left(node);
        ir_node  *op2  = get_Sub_right(node);
        ir_mode  *mode = get_irn_mode(node);

        if (mode_is_float(mode)) {
                if (ia32_cg_config.use_sse2)
                        return gen_binop(node, op1, op2, new_bd_ia32_xSub, match_am);
                else
                        return gen_binop_x87_float(node, op1, op2, new_bd_ia32_vfsub);
        }

        if (is_Const(op2)) {
                ir_fprintf(stderr, "Optimisation warning: found sub with const (%+F)\n",
                           node);
        }

        return gen_binop(node, op1, op2, new_bd_ia32_Sub, match_mode_neutral
                        | match_am | match_immediate);
}

static ir_node *transform_AM_mem(ir_node *const block,
                                 ir_node  *const src_val,
                                 ir_node  *const src_mem,
                                 ir_node  *const am_mem)
{
        if (is_NoMem(am_mem)) {
                return be_transform_node(src_mem);
        } else if (is_Proj(src_val) &&
                   is_Proj(src_mem) &&
                   get_Proj_pred(src_val) == get_Proj_pred(src_mem)) {
                /* avoid memory loop */
                return am_mem;
        } else if (is_Proj(src_val) && is_Sync(src_mem)) {
                ir_node  *const ptr_pred = get_Proj_pred(src_val);
                int       const arity    = get_Sync_n_preds(src_mem);
                int             n        = 0;
                ir_node **      ins;
                int             i;

                NEW_ARR_A(ir_node*, ins, arity + 1);

                /* NOTE: This sometimes produces dead-code because the old sync in
                 * src_mem might not be used anymore, we should detect this case
                 * and kill the sync... */
                for (i = arity - 1; i >= 0; --i) {
                        ir_node *const pred = get_Sync_pred(src_mem, i);

                        /* avoid memory loop */
                        if (is_Proj(pred) && get_Proj_pred(pred) == ptr_pred)
                                continue;

                        ins[n++] = be_transform_node(pred);
                }

                if (n==1 && ins[0] == am_mem) {
                        return am_mem;
                        /* creating a new Sync and relying on CSE may fail,
                         * if am_mem is a ProjM, which does not yet verify. */
                }

                ins[n++] = am_mem;
                return new_r_Sync(block, n, ins);
        } else {
                ir_node *ins[2];

                ins[0] = be_transform_node(src_mem);
                ins[1] = am_mem;
                return new_r_Sync(block, 2, ins);
        }
}

/**
 * Create a 32bit to 64bit signed extension.
 *
 * @param dbgi   debug info
 * @param block  the block where node nodes should be placed
 * @param val    the value to extend
 * @param orig   the original node
 */
static ir_node *create_sex_32_64(dbg_info *dbgi, ir_node *block,
                                 ir_node *val, const ir_node *orig)
{
        ir_node *res;

        (void)orig;
        if (ia32_cg_config.use_short_sex_eax) {
                ir_node *pval = new_bd_ia32_ProduceVal(dbgi, block);
                res = new_bd_ia32_Cltd(dbgi, block, val, pval);
        } else {
                ir_node *imm31 = ia32_create_Immediate(NULL, 0, 31);
                res = new_bd_ia32_Sar(dbgi, block, val, imm31);
        }
        SET_IA32_ORIG_NODE(res, orig);
        return res;
}

/**
 * Generates an ia32 Div with additional infrastructure for the
 * register allocator if needed.
 */
static ir_node *create_Div(ir_node *node)
{
        dbg_info *dbgi             = get_irn_dbg_info(node);
        ir_node  *block            = get_nodes_block(node);
        ir_node  *new_block        = be_transform_node(block);
        int       throws_exception = ir_throws_exception(node);
        ir_node  *mem;
        ir_node  *new_mem;
        ir_node  *op1;
        ir_node  *op2;
        ir_node  *new_node;
        ir_mode  *mode;
        ir_node  *sign_extension;
        ia32_address_mode_t  am;
        ia32_address_t      *addr = &am.addr;

        /* the upper bits have random contents for smaller modes */
        switch (get_irn_opcode(node)) {
        case iro_Div:
                op1     = get_Div_left(node);
                op2     = get_Div_right(node);
                mem     = get_Div_mem(node);
                mode    = get_Div_resmode(node);
                break;
        case iro_Mod:
                op1     = get_Mod_left(node);
                op2     = get_Mod_right(node);
                mem     = get_Mod_mem(node);
                mode    = get_Mod_resmode(node);
                break;
        default:
                panic("invalid divmod node %+F", node);
        }

        match_arguments(&am, block, op1, op2, NULL, match_am | match_upconv);

        /* Beware: We don't need a Sync, if the memory predecessor of the Div node
           is the memory of the consumed address. We can have only the second op as address
           in Div nodes, so check only op2. */
        new_mem = transform_AM_mem(block, op2, mem, addr->mem);

        if (mode_is_signed(mode)) {
                sign_extension = create_sex_32_64(dbgi, new_block, am.new_op1, node);
                new_node       = new_bd_ia32_IDiv(dbgi, new_block, addr->base,
                                addr->index, new_mem, am.new_op2, am.new_op1, sign_extension);
        } else {
                sign_extension = new_bd_ia32_Const(dbgi, new_block, NULL, 0, 0, 0);

                new_node = new_bd_ia32_Div(dbgi, new_block, addr->base,
                                           addr->index, new_mem, am.new_op2,
                                           am.new_op1, sign_extension);
        }
        ir_set_throws_exception(new_node, throws_exception);

        set_irn_pinned(new_node, get_irn_pinned(node));

        set_am_attributes(new_node, &am);
        SET_IA32_ORIG_NODE(new_node, node);

        new_node = fix_mem_proj(new_node, &am);

        return new_node;
}

/**
 * Generates an ia32 Mod.
 */
static ir_node *gen_Mod(ir_node *node)
{
        return create_Div(node);
}

/**
 * Generates an ia32 Div.
 */
static ir_node *gen_Div(ir_node *node)
{
        ir_mode *mode = get_Div_resmode(node);
        if (mode_is_float(mode)) {
                ir_node *op1 = get_Div_left(node);
                ir_node *op2 = get_Div_right(node);

                if (ia32_cg_config.use_sse2) {
                        return gen_binop(node, op1, op2, new_bd_ia32_xDiv, match_am);
                } else {
                        return gen_binop_x87_float(node, op1, op2, new_bd_ia32_vfdiv);
                }
        }

        return create_Div(node);
}

/**
 * Creates an ia32 Shl.
 *
 * @return The created ia32 Shl node
 */
static ir_node *gen_Shl(ir_node *node)
{
        ir_node *left  = get_Shl_left(node);
        ir_node *right = get_Shl_right(node);

        return gen_shift_binop(node, left, right, new_bd_ia32_Shl,
                               match_mode_neutral | match_immediate);
}

/**
 * Creates an ia32 Shr.
 *
 * @return The created ia32 Shr node
 */
static ir_node *gen_Shr(ir_node *node)
{
        ir_node *left  = get_Shr_left(node);
        ir_node *right = get_Shr_right(node);

        return gen_shift_binop(node, left, right, new_bd_ia32_Shr,
                               match_immediate | match_zero_ext);
}

/**
 * Creates an ia32 Sar.
 *
 * @return The created ia32 Shrs node
 */
static ir_node *gen_Shrs(ir_node *node)
{
        ir_node *left  = get_Shrs_left(node);
        ir_node *right = get_Shrs_right(node);

        if (is_Const(right)) {
                ir_tarval *tv  = get_Const_tarval(right);
                long       val = get_tarval_long(tv);
                if (val == 31) {
                        /* this is a sign extension */
                        dbg_info *dbgi   = get_irn_dbg_info(node);
                        ir_node  *block  = be_transform_node(get_nodes_block(node));
                        ir_node  *new_op = be_transform_node(left);

                        return create_sex_32_64(dbgi, block, new_op, node);
                }
        }

        /* 8 or 16 bit sign extension? */
        if (is_Const(right) && is_Shl(left)) {
                ir_node *shl_left  = get_Shl_left(left);
                ir_node *shl_right = get_Shl_right(left);
                if (is_Const(shl_right)) {
                        ir_tarval *tv1 = get_Const_tarval(right);
                        ir_tarval *tv2 = get_Const_tarval(shl_right);
                        if (tv1 == tv2 && tarval_is_long(tv1)) {
                                long val = get_tarval_long(tv1);
                                if (val == 16 || val == 24) {
                                        dbg_info *dbgi   = get_irn_dbg_info(node);
                                        ir_node  *block  = get_nodes_block(node);
                                        ir_mode  *src_mode;
                                        ir_node  *res;

                                        if (val == 24) {
                                                src_mode = mode_Bs;
                                        } else {
                                                assert(val == 16);
                                                src_mode = mode_Hs;
                                        }
                                        res = create_I2I_Conv(src_mode, mode_Is, dbgi, block,
                                                              shl_left, node);

                                        return res;
                                }
                        }
                }
        }

        return gen_shift_binop(node, left, right, new_bd_ia32_Sar,
                               match_immediate | match_upconv);
}



/**
 * Creates an ia32 Rol.
 *
 * @param op1   The first operator
 * @param op2   The second operator
 * @return The created ia32 RotL node
 */
static ir_node *gen_Rol(ir_node *node, ir_node *op1, ir_node *op2)
{
        return gen_shift_binop(node, op1, op2, new_bd_ia32_Rol, match_immediate);
}



/**
 * Creates an ia32 Ror.
 * NOTE: There is no RotR with immediate because this would always be a RotL
 *       "imm-mode_size_bits" which can be pre-calculated.
 *
 * @param op1   The first operator
 * @param op2   The second operator
 * @return The created ia32 RotR node
 */
static ir_node *gen_Ror(ir_node *node, ir_node *op1, ir_node *op2)
{
        return gen_shift_binop(node, op1, op2, new_bd_ia32_Ror, match_immediate);
}



/**
 * Creates an ia32 RotR or RotL (depending on the found pattern).
 *
 * @return The created ia32 RotL or RotR node
 */
static ir_node *gen_Rotl(ir_node *node)
{
        ir_node *op1    = get_Rotl_left(node);
        ir_node *op2    = get_Rotl_right(node);

        if (is_Minus(op2)) {
                return gen_Ror(node, op1, get_Minus_op(op2));
        }

        return gen_Rol(node, op1, op2);
}



/**
 * Transforms a Minus node.
 *
 * @return The created ia32 Minus node
 */
static ir_node *gen_Minus(ir_node *node)
{
        ir_node   *op    = get_Minus_op(node);
        ir_node   *block = be_transform_node(get_nodes_block(node));
        dbg_info  *dbgi  = get_irn_dbg_info(node);
        ir_mode   *mode  = get_irn_mode(node);
        ir_entity *ent;
        ir_node   *new_node;
        int        size;

        if (mode_is_float(mode)) {
                ir_node *new_op = be_transform_node(op);
                if (ia32_cg_config.use_sse2) {
                        /* TODO: non-optimal... if we have many xXors, then we should
                         * rather create a load for the const and use that instead of
                         * several AM nodes... */
                        ir_node *noreg_xmm = ia32_new_NoReg_xmm(current_ir_graph);

                        new_node = new_bd_ia32_xXor(dbgi, block, get_symconst_base(),
                                                    noreg_GP, nomem, new_op, noreg_xmm);

                        size = get_mode_size_bits(mode);
                        ent  = ia32_gen_fp_known_const(size == 32 ? ia32_SSIGN : ia32_DSIGN);

                        set_ia32_am_sc(new_node, ent);
                        set_ia32_op_type(new_node, ia32_AddrModeS);
                        set_ia32_ls_mode(new_node, mode);
                } else {
                        check_x87_floatmode(mode);
                        new_node = new_bd_ia32_vfchs(dbgi, block, new_op);
                }
        } else {
                new_node = gen_unop(node, op, new_bd_ia32_Neg, match_mode_neutral);
        }

        SET_IA32_ORIG_NODE(new_node, node);

        return new_node;
}

/**
 * Transforms a Not node.
 *
 * @return The created ia32 Not node
 */
static ir_node *gen_Not(ir_node *node)
{
        ir_node *op = get_Not_op(node);

        assert(get_irn_mode(node) != mode_b); /* should be lowered already */
        assert (! mode_is_float(get_irn_mode(node)));

        return gen_unop(node, op, new_bd_ia32_Not, match_mode_neutral);
}

static ir_node *create_float_abs(dbg_info *dbgi, ir_node *block, ir_node *op,
                                 bool negate, ir_node *node)
{
        ir_node   *new_block = be_transform_node(block);
        ir_mode   *mode      = get_irn_mode(op);
        ir_node   *new_op    = be_transform_node(op);
        ir_node   *new_node;
        int        size;
        ir_entity *ent;

        assert(mode_is_float(mode));

        if (ia32_cg_config.use_sse2) {
                ir_node *noreg_fp = ia32_new_NoReg_xmm(current_ir_graph);
                new_node = new_bd_ia32_xAnd(dbgi, new_block, get_symconst_base(),
                                                                        noreg_GP, nomem, new_op, noreg_fp);

                size = get_mode_size_bits(mode);
                ent  = ia32_gen_fp_known_const(size == 32 ? ia32_SABS : ia32_DABS);

                set_ia32_am_sc(new_node, ent);

                SET_IA32_ORIG_NODE(new_node, node);

                set_ia32_op_type(new_node, ia32_AddrModeS);
                set_ia32_ls_mode(new_node, mode);

                /* TODO, implement -Abs case */
                assert(!negate);
        } else {
                check_x87_floatmode(mode);
                new_node = new_bd_ia32_vfabs(dbgi, new_block, new_op);
                SET_IA32_ORIG_NODE(new_node, node);
                if (negate) {
                        new_node = new_bd_ia32_vfchs(dbgi, new_block, new_node);
                        SET_IA32_ORIG_NODE(new_node, node);
                }
        }

        return new_node;
}

/**
 * Create a bt instruction for x & (1 << n) and place it into the block of cmp.
 */
static ir_node *gen_bt(ir_node *cmp, ir_node *x, ir_node *n)
{
        dbg_info *dbgi      = get_irn_dbg_info(cmp);
        ir_node  *block     = get_nodes_block(cmp);
        ir_node  *new_block = be_transform_node(block);
        ir_node  *op1       = be_transform_node(x);
        ir_node  *op2       = be_transform_node(n);

        return new_bd_ia32_Bt(dbgi, new_block, op1, op2);
}

static ia32_condition_code_t relation_to_condition_code(ir_relation relation,
                                                        ir_mode *mode,
                                                        bool overflow_possible)
{
        if (mode_is_float(mode)) {
                switch (relation) {
                case ir_relation_equal:              return ia32_cc_float_equal;
                case ir_relation_less:               return ia32_cc_float_below;
                case ir_relation_less_equal:         return ia32_cc_float_below_equal;
                case ir_relation_greater:            return ia32_cc_float_above;
                case ir_relation_greater_equal:      return ia32_cc_float_above_equal;
                case ir_relation_less_greater:       return ia32_cc_not_equal;
                case ir_relation_less_equal_greater: return ia32_cc_not_parity;
                case ir_relation_unordered:          return ia32_cc_parity;
                case ir_relation_unordered_equal:    return ia32_cc_equal;
                case ir_relation_unordered_less:   return ia32_cc_float_unordered_below;
                case ir_relation_unordered_less_equal:
                                             return ia32_cc_float_unordered_below_equal;
                case ir_relation_unordered_greater:
                                             return ia32_cc_float_unordered_above;
                case ir_relation_unordered_greater_equal:
                                             return ia32_cc_float_unordered_above_equal;
                case ir_relation_unordered_less_greater:
                                             return ia32_cc_float_not_equal;
                case ir_relation_false:
                case ir_relation_true:
                        /* should we introduce a jump always/jump never? */
                        break;
                }
                panic("Unexpected float pnc");
        } else if (mode_is_signed(mode)) {
                switch (relation) {
                case ir_relation_unordered_equal:
                case ir_relation_equal:                return ia32_cc_equal;
                case ir_relation_unordered_less:
                case ir_relation_less:
                        return overflow_possible ? ia32_cc_less : ia32_cc_sign;
                case ir_relation_unordered_less_equal:
                case ir_relation_less_equal:           return ia32_cc_less_equal;
                case ir_relation_unordered_greater:
                case ir_relation_greater:              return ia32_cc_greater;
                case ir_relation_unordered_greater_equal:
                case ir_relation_greater_equal:
                        return overflow_possible ? ia32_cc_greater_equal : ia32_cc_not_sign;
                case ir_relation_unordered_less_greater:
                case ir_relation_less_greater:         return ia32_cc_not_equal;
                case ir_relation_less_equal_greater:
                case ir_relation_unordered:
                case ir_relation_false:
                case ir_relation_true:
                        /* introduce jump always/jump never? */
                        break;
                }
                panic("Unexpected pnc");
        } else {
                switch (relation) {
                case ir_relation_unordered_equal:
                case ir_relation_equal:         return ia32_cc_equal;
                case ir_relation_unordered_less:
                case ir_relation_less:          return ia32_cc_below;
                case ir_relation_unordered_less_equal:
                case ir_relation_less_equal:    return ia32_cc_below_equal;
                case ir_relation_unordered_greater:
                case ir_relation_greater:       return ia32_cc_above;
                case ir_relation_unordered_greater_equal:
                case ir_relation_greater_equal: return ia32_cc_above_equal;
                case ir_relation_unordered_less_greater:
                case ir_relation_less_greater:  return ia32_cc_not_equal;
                case ir_relation_less_equal_greater:
                case ir_relation_unordered:
                case ir_relation_false:
                case ir_relation_true:
                        /* introduce jump always/jump never? */
                        break;
                }
                panic("Unexpected pnc");
        }
}

static ir_node *get_flags_node(ir_node *cmp, ia32_condition_code_t *cc_out)
{
        /* must have a Cmp as input */
        ir_relation relation = get_Cmp_relation(cmp);
        ir_node    *l        = get_Cmp_left(cmp);
        ir_node    *r        = get_Cmp_right(cmp);
        ir_mode    *mode     = get_irn_mode(l);
        bool        overflow_possible;
        ir_node    *flags;

        /* check for bit-test */
        if (ia32_cg_config.use_bt
            && (relation == ir_relation_equal
                || (mode_is_signed(mode) && relation == ir_relation_less_greater)
                || (!mode_is_signed(mode) && ((relation & ir_relation_greater_equal) == ir_relation_greater)))
            && is_And(l)) {
                ir_node *la = get_And_left(l);
                ir_node *ra = get_And_right(l);
                if (is_Shl(ra)) {
                        ir_node *tmp = la;
                        la = ra;
                        ra = tmp;
                }
                if (is_Shl(la)) {
                        ir_node *c = get_Shl_left(la);
                        if (is_Const_1(c) && is_Const_0(r)) {
                                /* (1 << n) & ra) */
                                ir_node *n = get_Shl_right(la);
                                flags = gen_bt(cmp, ra, n);
                                /* the bit is copied into the CF flag */
                                if (relation & ir_relation_equal)
                                        *cc_out = ia32_cc_above_equal; /* test for CF=0 */
                                else
                                        *cc_out = ia32_cc_below;       /* test for CF=1 */
                                return flags;
                        }
                }
        }

        /* the middle-end tries to eliminate impossible relations, so a ptr <> 0
         * test becomes ptr > 0. But for x86 an equal comparison is preferable to
         * a >0 (we can sometimes eliminate the cmp in favor of flags produced by
         * a predecessor node). So add the < bit.
         * (Note that we do not want to produce <=> (which can happen for
         * unoptimized code), because no x86 flag can represent that */
        if (!(relation & ir_relation_equal) && relation & ir_relation_less_greater)
                relation |= get_negated_relation(ir_get_possible_cmp_relations(l, r)) & ir_relation_less_greater;

        overflow_possible = true;
        if (is_Const(r) && is_Const_null(r))
                overflow_possible = false;

        /* just do a normal transformation of the Cmp */
        *cc_out = relation_to_condition_code(relation, mode, overflow_possible);
        flags   = be_transform_node(cmp);
        return flags;
}

/**
 * Transforms a Load.
 *
 * @return the created ia32 Load node
 */
static ir_node *gen_Load(ir_node *node)
{
        ir_node  *old_block = get_nodes_block(node);
        ir_node  *block     = be_transform_node(old_block);
        ir_node  *ptr       = get_Load_ptr(node);
        ir_node  *mem       = get_Load_mem(node);
        ir_node  *new_mem   = be_transform_node(mem);
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_mode  *mode      = get_Load_mode(node);
        int       throws_exception = ir_throws_exception(node);
        ir_node  *base;
        ir_node  *idx;
        ir_node  *new_node;
        ia32_address_t addr;

        /* construct load address */
        memset(&addr, 0, sizeof(addr));
        ia32_create_address_mode(&addr, ptr, ia32_create_am_normal);
        base = addr.base;
        idx  = addr.index;

        if (base == NULL) {
                base = noreg_GP;
        } else {
                base = be_transform_node(base);
        }

        if (idx == NULL) {
                idx = noreg_GP;
        } else {
                idx = be_transform_node(idx);
        }

        if (mode_is_float(mode)) {
                if (ia32_cg_config.use_sse2) {
                        new_node = new_bd_ia32_xLoad(dbgi, block, base, idx, new_mem,
                                                     mode);
                } else {
                        new_node = new_bd_ia32_vfld(dbgi, block, base, idx, new_mem,
                                                    mode);
                }
        } else {
                assert(mode != mode_b);

                /* create a conv node with address mode for smaller modes */
                if (get_mode_size_bits(mode) < 32) {
                        new_node = new_bd_ia32_Conv_I2I(dbgi, block, base, idx,
                                                        new_mem, noreg_GP, mode);
                } else {
                        new_node = new_bd_ia32_Load(dbgi, block, base, idx, new_mem);
                }
        }
        ir_set_throws_exception(new_node, throws_exception);

        set_irn_pinned(new_node, get_irn_pinned(node));
        set_ia32_op_type(new_node, ia32_AddrModeS);
        set_ia32_ls_mode(new_node, mode);
        set_address(new_node, &addr);

        if (get_irn_pinned(node) == op_pin_state_floats) {
                assert((int)pn_ia32_xLoad_res == (int)pn_ia32_vfld_res
                                && (int)pn_ia32_vfld_res == (int)pn_ia32_Load_res
                                && (int)pn_ia32_Load_res == (int)pn_ia32_res);
                arch_add_irn_flags(new_node, arch_irn_flags_rematerializable);
        }

        SET_IA32_ORIG_NODE(new_node, node);

        return new_node;
}

static int use_dest_am(ir_node *block, ir_node *node, ir_node *mem,
                       ir_node *ptr, ir_node *other)
{
        ir_node *load;

        if (!is_Proj(node))
                return 0;

        /* we only use address mode if we're the only user of the load */
        if (get_irn_n_edges(node) > 1)
                return 0;

        load = get_Proj_pred(node);
        if (!is_Load(load))
                return 0;
        if (get_nodes_block(load) != block)
                return 0;

        /* store should have the same pointer as the load */
        if (get_Load_ptr(load) != ptr)
                return 0;

        /* don't do AM if other node inputs depend on the load (via mem-proj) */
        if (other != NULL                   &&
            get_nodes_block(other) == block &&
            heights_reachable_in_block(ia32_heights, other, load)) {
                return 0;
        }

        if (ia32_prevents_AM(block, load, mem))
                return 0;
        /* Store should be attached to the load via mem */
        assert(heights_reachable_in_block(ia32_heights, mem, load));

        return 1;
}

static ir_node *dest_am_binop(ir_node *node, ir_node *op1, ir_node *op2,
                              ir_node *mem, ir_node *ptr, ir_mode *mode,
                              construct_binop_dest_func *func,
                              construct_binop_dest_func *func8bit,
                                                          match_flags_t flags)
{
        ir_node  *src_block = get_nodes_block(node);
        ir_node  *block;
        dbg_info *dbgi;
        ir_node  *new_mem;
        ir_node  *new_node;
        ir_node  *new_op;
        ir_node  *mem_proj;
        int       commutative;
        ia32_address_mode_t  am;
        ia32_address_t      *addr = &am.addr;
        memset(&am, 0, sizeof(am));

        assert(flags & match_immediate); /* there is no destam node without... */
        commutative = (flags & match_commutative) != 0;

        if (use_dest_am(src_block, op1, mem, ptr, op2)) {
                build_address(&am, op1, ia32_create_am_double_use);
                new_op = create_immediate_or_transform(op2, 0);
        } else if (commutative && use_dest_am(src_block, op2, mem, ptr, op1)) {
                build_address(&am, op2, ia32_create_am_double_use);
                new_op = create_immediate_or_transform(op1, 0);
        } else {
                return NULL;
        }

        if (addr->base == NULL)
                addr->base = noreg_GP;
        if (addr->index == NULL)
                addr->index = noreg_GP;
        if (addr->mem == NULL)
                addr->mem = nomem;

        dbgi    = get_irn_dbg_info(node);
        block   = be_transform_node(src_block);
        new_mem = transform_AM_mem(block, am.am_node, mem, addr->mem);

        if (get_mode_size_bits(mode) == 8) {
                new_node = func8bit(dbgi, block, addr->base, addr->index, new_mem, new_op);
        } else {
                new_node = func(dbgi, block, addr->base, addr->index, new_mem, new_op);
        }
        set_address(new_node, addr);
        set_ia32_op_type(new_node, ia32_AddrModeD);
        set_ia32_ls_mode(new_node, mode);
        SET_IA32_ORIG_NODE(new_node, node);

        be_set_transformed_node(get_Proj_pred(am.mem_proj), new_node);
        mem_proj = be_transform_node(am.mem_proj);
        be_set_transformed_node(am.mem_proj, new_node);
        be_set_transformed_node(mem_proj, new_node);

        return new_node;
}

static ir_node *dest_am_unop(ir_node *node, ir_node *op, ir_node *mem,
                             ir_node *ptr, ir_mode *mode,
                             construct_unop_dest_func *func)
{
        ir_node  *src_block = get_nodes_block(node);
        ir_node  *block;
        dbg_info *dbgi;
        ir_node  *new_mem;
        ir_node  *new_node;
        ir_node  *mem_proj;
        ia32_address_mode_t  am;
        ia32_address_t *addr = &am.addr;

        if (!use_dest_am(src_block, op, mem, ptr, NULL))
                return NULL;

        memset(&am, 0, sizeof(am));
        build_address(&am, op, ia32_create_am_double_use);

        dbgi     = get_irn_dbg_info(node);
        block    = be_transform_node(src_block);
        new_mem  = transform_AM_mem(block, am.am_node, mem, addr->mem);
        new_node = func(dbgi, block, addr->base, addr->index, new_mem);
        set_address(new_node, addr);
        set_ia32_op_type(new_node, ia32_AddrModeD);
        set_ia32_ls_mode(new_node, mode);
        SET_IA32_ORIG_NODE(new_node, node);

        be_set_transformed_node(get_Proj_pred(am.mem_proj), new_node);
        mem_proj = be_transform_node(am.mem_proj);
        be_set_transformed_node(am.mem_proj, new_node);
        be_set_transformed_node(mem_proj, new_node);

        return new_node;
}

static ir_node *try_create_SetMem(ir_node *node, ir_node *ptr, ir_node *mem)
{
        ir_mode              *mode      = get_irn_mode(node);
        ir_node              *mux_true  = get_Mux_true(node);
        ir_node              *mux_false = get_Mux_false(node);
        ir_node              *cond;
        dbg_info             *dbgi;
        ir_node              *block;
        ir_node              *new_block;
        ir_node              *flags;
        ir_node              *new_node;
        bool                  negated;
        ia32_condition_code_t cc;
        ia32_address_t        addr;

        if (get_mode_size_bits(mode) != 8)
                return NULL;

        if (is_Const_1(mux_true) && is_Const_0(mux_false)) {
                negated = false;
        } else if (is_Const_0(mux_true) && is_Const_1(mux_false)) {
                negated = true;
        } else {
                return NULL;
        }

        cond  = get_Mux_sel(node);
        flags = get_flags_node(cond, &cc);
        /* we can't handle the float special cases with SetM */
        if (cc & ia32_cc_additional_float_cases)
                return NULL;
        if (negated)
                cc = ia32_negate_condition_code(cc);

        build_address_ptr(&addr, ptr, mem);

        dbgi      = get_irn_dbg_info(node);
        block     = get_nodes_block(node);
        new_block = be_transform_node(block);
        new_node  = new_bd_ia32_SetccMem(dbgi, new_block, addr.base,
                                         addr.index, addr.mem, flags, cc);
        set_address(new_node, &addr);
        set_ia32_op_type(new_node, ia32_AddrModeD);
        set_ia32_ls_mode(new_node, mode);
        SET_IA32_ORIG_NODE(new_node, node);

        return new_node;
}

static ir_node *try_create_dest_am(ir_node *node)
{
        ir_node  *val  = get_Store_value(node);
        ir_node  *mem  = get_Store_mem(node);
        ir_node  *ptr  = get_Store_ptr(node);
        ir_mode  *mode = get_irn_mode(val);
        unsigned  bits = get_mode_size_bits(mode);
        ir_node  *op1;
        ir_node  *op2;
        ir_node  *new_node;

        /* handle only GP modes for now... */
        if (!ia32_mode_needs_gp_reg(mode))
                return NULL;

        for (;;) {
                /* store must be the only user of the val node */
                if (get_irn_n_edges(val) > 1)
                        return NULL;
                /* skip pointless convs */
                if (is_Conv(val)) {
                        ir_node *conv_op   = get_Conv_op(val);
                        ir_mode *pred_mode = get_irn_mode(conv_op);
                        if (!ia32_mode_needs_gp_reg(pred_mode))
                                break;
                        if (pred_mode == mode_b || bits <= get_mode_size_bits(pred_mode)) {
                                val = conv_op;
                                continue;
                        }
                }
                break;
        }

        /* value must be in the same block */
        if (get_nodes_block(node) != get_nodes_block(val))
                return NULL;

        switch (get_irn_opcode(val)) {
        case iro_Add:
                op1      = get_Add_left(val);
                op2      = get_Add_right(val);
                if (ia32_cg_config.use_incdec) {
                        if (is_Const_1(op2)) {
                                new_node = dest_am_unop(val, op1, mem, ptr, mode, new_bd_ia32_IncMem);
                                break;
                        } else if (is_Const_Minus_1(op2)) {
                                new_node = dest_am_unop(val, op1, mem, ptr, mode, new_bd_ia32_DecMem);
                                break;
                        }
                }
                new_node = dest_am_binop(val, op1, op2, mem, ptr, mode,
                                         new_bd_ia32_AddMem, new_bd_ia32_AddMem8Bit,
                                         match_commutative | match_immediate);
                break;
        case iro_Sub:
                op1      = get_Sub_left(val);
                op2      = get_Sub_right(val);
                if (is_Const(op2)) {
                        ir_fprintf(stderr, "Optimisation warning: not-normalized sub ,C found\n");
                }
                new_node = dest_am_binop(val, op1, op2, mem, ptr, mode,
                                         new_bd_ia32_SubMem, new_bd_ia32_SubMem8Bit,
                                         match_immediate);
                break;
        case iro_And:
                op1      = get_And_left(val);
                op2      = get_And_right(val);
                new_node = dest_am_binop(val, op1, op2, mem, ptr, mode,
                                         new_bd_ia32_AndMem, new_bd_ia32_AndMem8Bit,
                                         match_commutative | match_immediate);
                break;
        case iro_Or:
                op1      = get_Or_left(val);
                op2      = get_Or_right(val);
                new_node = dest_am_binop(val, op1, op2, mem, ptr, mode,
                                         new_bd_ia32_OrMem, new_bd_ia32_OrMem8Bit,
                                         match_commutative | match_immediate);
                break;
        case iro_Eor:
                op1      = get_Eor_left(val);
                op2      = get_Eor_right(val);
                new_node = dest_am_binop(val, op1, op2, mem, ptr, mode,
                                         new_bd_ia32_XorMem, new_bd_ia32_XorMem8Bit,
                                         match_commutative | match_immediate);
                break;
        case iro_Shl:
                op1      = get_Shl_left(val);
                op2      = get_Shl_right(val);
                new_node = dest_am_binop(val, op1, op2, mem, ptr, mode,
                                         new_bd_ia32_ShlMem, new_bd_ia32_ShlMem,
                                         match_immediate);
                break;
        case iro_Shr:
                op1      = get_Shr_left(val);
                op2      = get_Shr_right(val);
                new_node = dest_am_binop(val, op1, op2, mem, ptr, mode,
                                         new_bd_ia32_ShrMem, new_bd_ia32_ShrMem,
                                         match_immediate);
                break;
        case iro_Shrs:
                op1      = get_Shrs_left(val);
                op2      = get_Shrs_right(val);
                new_node = dest_am_binop(val, op1, op2, mem, ptr, mode,
                                         new_bd_ia32_SarMem, new_bd_ia32_SarMem,
                                         match_immediate);
                break;
        case iro_Rotl:
                op1      = get_Rotl_left(val);
                op2      = get_Rotl_right(val);
                new_node = dest_am_binop(val, op1, op2, mem, ptr, mode,
                                         new_bd_ia32_RolMem, new_bd_ia32_RolMem,
                                         match_immediate);
                break;
        /* TODO: match ROR patterns... */
        case iro_Mux:
                new_node = try_create_SetMem(val, ptr, mem);
                break;

        case iro_Minus:
                op1      = get_Minus_op(val);
                new_node = dest_am_unop(val, op1, mem, ptr, mode, new_bd_ia32_NegMem);
                break;
        case iro_Not:
                /* should be lowered already */
                assert(mode != mode_b);
                op1      = get_Not_op(val);
                new_node = dest_am_unop(val, op1, mem, ptr, mode, new_bd_ia32_NotMem);
                break;
        default:
                return NULL;
        }

        if (new_node != NULL) {
                if (get_irn_pinned(new_node) != op_pin_state_pinned &&
                                get_irn_pinned(node) == op_pin_state_pinned) {
                        set_irn_pinned(new_node, op_pin_state_pinned);
                }
        }

        return new_node;
}

static bool possible_int_mode_for_fp(ir_mode *mode)
{
        unsigned size;

        if (!mode_is_signed(mode))
                return false;
        size = get_mode_size_bits(mode);
        if (size != 16 && size != 32)
                return false;
        return true;
}

static int is_float_to_int_conv(const ir_node *node)
{
        ir_mode  *mode = get_irn_mode(node);
        ir_node  *conv_op;
        ir_mode  *conv_mode;

        if (!possible_int_mode_for_fp(mode))
                return 0;

        if (!is_Conv(node))
                return 0;
        conv_op   = get_Conv_op(node);
        conv_mode = get_irn_mode(conv_op);

        if (!mode_is_float(conv_mode))
                return 0;

        return 1;
}

/**
 * Transform a Store(floatConst) into a sequence of
 * integer stores.
 *
 * @return the created ia32 Store node
 */
static ir_node *gen_float_const_Store(ir_node *node, ir_node *cns)
{
        ir_mode        *mode      = get_irn_mode(cns);
        unsigned        size      = get_mode_size_bytes(mode);
        ir_tarval      *tv        = get_Const_tarval(cns);
        ir_node        *block     = get_nodes_block(node);
        ir_node        *new_block = be_transform_node(block);
        ir_node        *ptr       = get_Store_ptr(node);
        ir_node        *mem       = get_Store_mem(node);
        dbg_info       *dbgi      = get_irn_dbg_info(node);
        int             ofs       = 0;
        int             i         = 0;
        int             throws_exception = ir_throws_exception(node);
        ir_node        *ins[4];
        ia32_address_t  addr;

        build_address_ptr(&addr, ptr, mem);

        do {
                unsigned val;
                unsigned delta;
                ir_mode *mode;
                if (size >= 4) {
                        val= get_tarval_sub_bits(tv, ofs)            |
                            (get_tarval_sub_bits(tv, ofs + 1) <<  8) |
                            (get_tarval_sub_bits(tv, ofs + 2) << 16) |
                            (get_tarval_sub_bits(tv, ofs + 3) << 24);
                        delta = 4;
                        mode  = mode_Iu;
                } else if (size >= 2) {
                        val= get_tarval_sub_bits(tv, ofs)            |
                            (get_tarval_sub_bits(tv, ofs + 1) <<  8);
                        delta = 2;
                        mode  = mode_Hu;
                } else {
                        panic("invalid size of Store float to mem (%+F)", node);
                }
                ir_node *imm = ia32_create_Immediate(NULL, 0, val);

                ir_node *new_node = new_bd_ia32_Store(dbgi, new_block, addr.base,
                        addr.index, addr.mem, imm);
                ir_node *new_mem  = new_r_Proj(new_node, mode_M, pn_ia32_Store_M);

                ir_set_throws_exception(new_node, throws_exception);
                set_irn_pinned(new_node, get_irn_pinned(node));
                set_ia32_op_type(new_node, ia32_AddrModeD);
                set_ia32_ls_mode(new_node, mode);
                set_address(new_node, &addr);
                SET_IA32_ORIG_NODE(new_node, node);

                assert(i < 4);
                ins[i++] = new_mem;

                size -= delta;
                ofs  += delta;
                addr.offset += delta;
        } while (size != 0);

        if (i > 1) {
                return new_rd_Sync(dbgi, new_block, i, ins);
        } else {
                return get_Proj_pred(ins[0]);
        }
}

/**
 * Generate a vfist or vfisttp instruction.
 */
static ir_node *gen_vfist(dbg_info *dbgi, ir_node *block, ir_node *base,
                          ir_node *index, ir_node *mem,  ir_node *val)
{
        if (ia32_cg_config.use_fisttp) {
                /* Note: fisttp ALWAYS pop the tos. We have to ensure here that the value is copied
                if other users exists */
                ir_node *vfisttp = new_bd_ia32_vfisttp(dbgi, block, base, index, mem, val);
                ir_node *value   = new_r_Proj(vfisttp, ia32_mode_E, pn_ia32_vfisttp_res);
                be_new_Keep(block, 1, &value);

                return vfisttp;
        } else {
                ir_node *trunc_mode = ia32_new_Fpu_truncate(current_ir_graph);

                /* do a fist */
                ir_node *vfist = new_bd_ia32_vfist(dbgi, block, base, index, mem, val, trunc_mode);
                return vfist;
        }
}

/**
 * Transforms a general (no special case) Store.
 *
 * @return the created ia32 Store node
 */
static ir_node *gen_general_Store(ir_node *node)
{
        ir_node  *val       = get_Store_value(node);
        ir_mode  *mode      = get_irn_mode(val);
        ir_node  *block     = get_nodes_block(node);
        ir_node  *new_block = be_transform_node(block);
        ir_node  *ptr       = get_Store_ptr(node);
        ir_node  *mem       = get_Store_mem(node);
        dbg_info *dbgi      = get_irn_dbg_info(node);
        int       throws_exception = ir_throws_exception(node);
        ir_node  *new_val;
        ir_node  *new_node;
        ia32_address_t addr;

        /* check for destination address mode */
        new_node = try_create_dest_am(node);
        if (new_node != NULL)
                return new_node;

        /* construct store address */
        memset(&addr, 0, sizeof(addr));
        ia32_create_address_mode(&addr, ptr, ia32_create_am_normal);

        if (addr.base == NULL) {
                addr.base = noreg_GP;
        } else {
                addr.base = be_transform_node(addr.base);
        }

        if (addr.index == NULL) {
                addr.index = noreg_GP;
        } else {
                addr.index = be_transform_node(addr.index);
        }
        addr.mem = be_transform_node(mem);

        if (mode_is_float(mode)) {
                new_val = be_transform_node(val);
                if (ia32_cg_config.use_sse2) {
                        new_node = new_bd_ia32_xStore(dbgi, new_block, addr.base,
                                                      addr.index, addr.mem, new_val);
                } else {
                        new_node = new_bd_ia32_vfst(dbgi, new_block, addr.base,
                                                    addr.index, addr.mem, new_val, mode);
                }
        } else if (!ia32_cg_config.use_sse2 && is_float_to_int_conv(val)) {
                val      = get_Conv_op(val);
                new_val  = be_transform_node(val);
                new_node = gen_vfist(dbgi, new_block, addr.base, addr.index, addr.mem, new_val);
        } else {
                unsigned dest_bits = get_mode_size_bits(mode);
                while (is_downconv(val)
                       && get_mode_size_bits(get_irn_mode(val)) >= dest_bits) {
                    val = get_Conv_op(val);
                }
                new_val = create_immediate_or_transform(val, 0);
                assert(mode != mode_b);

                if (dest_bits == 8) {
                        new_node = new_bd_ia32_Store8Bit(dbgi, new_block, addr.base,
                                                         addr.index, addr.mem, new_val);
                } else {
                        new_node = new_bd_ia32_Store(dbgi, new_block, addr.base,
                                                     addr.index, addr.mem, new_val);
                }
        }
        ir_set_throws_exception(new_node, throws_exception);

        set_irn_pinned(new_node, get_irn_pinned(node));
        set_ia32_op_type(new_node, ia32_AddrModeD);
        set_ia32_ls_mode(new_node, mode);

        set_address(new_node, &addr);
        SET_IA32_ORIG_NODE(new_node, node);

        return new_node;
}

/**
 * Transforms a Store.
 *
 * @return the created ia32 Store node
 */
static ir_node *gen_Store(ir_node *node)
{
        ir_node *val  = get_Store_value(node);
        ir_mode *mode = get_irn_mode(val);

        if (mode_is_float(mode) && is_Const(val)) {
                /* We can transform every floating const store
                   into a sequence of integer stores.
                   If the constant is already in a register,
                   it would be better to use it, but we don't
                   have this information here. */
                return gen_float_const_Store(node, val);
        }
        return gen_general_Store(node);
}

/**
 * Transforms a Switch.
 *
 * @return the created ia32 SwitchJmp node
 */
static ir_node *gen_Switch(ir_node *node)
{
        dbg_info              *dbgi     = get_irn_dbg_info(node);
        ir_graph              *irg      = get_irn_irg(node);
        ir_node               *block    = be_transform_node(get_nodes_block(node));
        ir_node               *sel      = get_Switch_selector(node);
        ir_node               *new_sel  = be_transform_node(sel);
        ir_mode               *sel_mode = get_irn_mode(sel);
        const ir_switch_table *table    = get_Switch_table(node);
        unsigned               n_outs   = get_Switch_n_outs(node);
        ir_node               *new_node;
        ir_entity             *entity;

        assert(get_mode_size_bits(sel_mode) <= 32);
        assert(!mode_is_float(sel_mode));
        sel = ia32_skip_sameconv(sel);
        if (get_mode_size_bits(sel_mode) < 32)
                new_sel = transform_upconv(sel, node);

        entity = new_entity(NULL, id_unique("TBL%u"), get_unknown_type());
        set_entity_visibility(entity, ir_visibility_private);
        add_entity_linkage(entity, IR_LINKAGE_CONSTANT);

        table = ir_switch_table_duplicate(irg, table);

        new_node = new_bd_ia32_SwitchJmp(dbgi, block, noreg_GP, new_sel, n_outs, table);
        set_ia32_am_scale(new_node, 2);
        set_ia32_am_sc(new_node, entity);
        set_ia32_op_type(new_node, ia32_AddrModeS);
        set_ia32_ls_mode(new_node, mode_Iu);
        SET_IA32_ORIG_NODE(new_node, node);
        // FIXME This seems wrong. GCC uses PIC for switch on OS X.
        get_ia32_attr(new_node)->data.am_sc_no_pic_adjust = true;

        return new_node;
}

/**
 * Transform a Cond node.
 */
static ir_node *gen_Cond(ir_node *node)
{
        ir_node              *block     = get_nodes_block(node);
        ir_node              *new_block = be_transform_node(block);
        dbg_info             *dbgi      = get_irn_dbg_info(node);
        ir_node              *sel       = get_Cond_selector(node);
        ir_node              *flags     = NULL;
        ir_node              *new_node;
        ia32_condition_code_t cc;

        /* we get flags from a Cmp */
        flags = get_flags_node(sel, &cc);

        new_node = new_bd_ia32_Jcc(dbgi, new_block, flags, cc);
        SET_IA32_ORIG_NODE(new_node, node);

        return new_node;
}

/**
 * Transform a be_Copy.
 */
static ir_node *gen_be_Copy(ir_node *node)
{
        ir_node *new_node = be_duplicate_node(node);
        ir_mode *mode     = get_irn_mode(new_node);

        if (ia32_mode_needs_gp_reg(mode)) {
                set_irn_mode(new_node, mode_Iu);
        }

        return new_node;
}

static ir_node *create_Fucom(ir_node *node)
{
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_node  *block     = get_nodes_block(node);
        ir_node  *new_block = be_transform_node(block);
        ir_node  *left      = get_Cmp_left(node);
        ir_node  *new_left  = be_transform_node(left);
        ir_node  *right     = get_Cmp_right(node);
        ir_mode  *cmp_mode  = get_irn_mode(left);
        ir_node  *new_right;
        ir_node  *new_node;
        check_x87_floatmode(cmp_mode);

        if (ia32_cg_config.use_fucomi) {
                new_right = be_transform_node(right);
                new_node  = new_bd_ia32_vFucomi(dbgi, new_block, new_left,
                                                new_right, 0);
                set_ia32_commutative(new_node);
                SET_IA32_ORIG_NODE(new_node, node);
        } else {
                if (is_Const_0(right)) {
                        new_node = new_bd_ia32_vFtstFnstsw(dbgi, new_block, new_left, 0);
                } else {
                        new_right = be_transform_node(right);
                        new_node  = new_bd_ia32_vFucomFnstsw(dbgi, new_block, new_left, new_right, 0);
                        set_ia32_commutative(new_node);
                }

                SET_IA32_ORIG_NODE(new_node, node);

                new_node = new_bd_ia32_Sahf(dbgi, new_block, new_node);
                SET_IA32_ORIG_NODE(new_node, node);
        }

        return new_node;
}

static ir_node *create_Ucomi(ir_node *node)
{
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_node  *src_block = get_nodes_block(node);
        ir_node  *new_block = be_transform_node(src_block);
        ir_node  *left      = get_Cmp_left(node);
        ir_node  *right     = get_Cmp_right(node);
        ir_node  *new_node;
        ia32_address_mode_t  am;
        ia32_address_t      *addr = &am.addr;

        match_arguments(&am, src_block, left, right, NULL,
                        match_commutative | match_am);

        new_node = new_bd_ia32_Ucomi(dbgi, new_block, addr->base, addr->index,
                                     addr->mem, am.new_op1, am.new_op2,
                                     am.ins_permuted);
        set_am_attributes(new_node, &am);

        SET_IA32_ORIG_NODE(new_node, node);

        new_node = fix_mem_proj(new_node, &am);

        return new_node;
}

static bool ia32_mux_upper_bits_clean(const ir_node *node, ir_mode *mode)
{
        ir_node *mux_true  = get_Mux_true(node);
        ir_node *mux_false = get_Mux_false(node);
        ir_mode *mux_mode  = get_irn_mode(node);
        /* mux nodes which get transformed to the set instruction are not clean */
        if (is_Const(mux_true) && is_Const(mux_false)
                && get_mode_size_bits(mux_mode) == 8) {
                return false;
        }
        return be_upper_bits_clean(mux_true, mode)
                && be_upper_bits_clean(mux_false, mode);
}

/**
 * Generate code for a Cmp.
 */
static ir_node *gen_Cmp(ir_node *node)
{
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_node  *block     = get_nodes_block(node);
        ir_node  *new_block = be_transform_node(block);
        ir_node  *left      = get_Cmp_left(node);
        ir_node  *right     = get_Cmp_right(node);
        ir_mode  *cmp_mode  = get_irn_mode(left);
        ir_node  *new_node;
        ia32_address_mode_t  am;
        ia32_address_t      *addr = &am.addr;

        if (mode_is_float(cmp_mode)) {
                if (ia32_cg_config.use_sse2) {
                        return create_Ucomi(node);
                } else {
                        return create_Fucom(node);
                }
        }

        assert(ia32_mode_needs_gp_reg(cmp_mode));

        /* Prefer the Test instruction, when encountering (x & y) ==/!= 0 */
        if (is_Const_0(right)          &&
            is_And(left)               &&
            get_irn_n_edges(left) == 1) {
                /* Test(and_left, and_right) */
                ir_node *and_left  = get_And_left(left);
                ir_node *and_right = get_And_right(left);

                /* matze: code here used mode instead of cmd_mode, I think it is always
                 * the same as cmp_mode, but I leave this here to see if this is really
                 * true...
                 */
                assert(get_irn_mode(and_left) == cmp_mode);

                match_arguments(&am, block, and_left, and_right, NULL,
                                match_commutative |
                                match_am | match_8bit_am | match_16bit_am |
                                match_am_and_immediates | match_immediate);

                /* use 32bit compare mode if possible since the opcode is smaller */
                if (am.op_type == ia32_Normal &&
                        be_upper_bits_clean(and_left, cmp_mode) &&
                    be_upper_bits_clean(and_right, cmp_mode)) {
                        cmp_mode = mode_is_signed(cmp_mode) ? mode_Is : mode_Iu;
                }

                if (get_mode_size_bits(cmp_mode) == 8) {
                        new_node = new_bd_ia32_Test8Bit(dbgi, new_block, addr->base,
                                                        addr->index, addr->mem,
                                                        am.new_op1, am.new_op2,
                                                        am.ins_permuted);
                } else {
                        new_node = new_bd_ia32_Test(dbgi, new_block, addr->base,
                                                    addr->index, addr->mem, am.new_op1,
                                                    am.new_op2, am.ins_permuted);
                }
        } else {
                /* Cmp(left, right) */
                match_arguments(&am, block, left, right, NULL,
                                match_commutative |
                                match_am | match_8bit_am | match_16bit_am |
                                match_am_and_immediates | match_immediate);
                /* use 32bit compare mode if possible since the opcode is smaller */
                if (am.op_type == ia32_Normal &&
                        be_upper_bits_clean(left, cmp_mode) &&
                    be_upper_bits_clean(right, cmp_mode)) {
                        cmp_mode = mode_is_signed(cmp_mode) ? mode_Is : mode_Iu;
                }

                if (get_mode_size_bits(cmp_mode) == 8) {
                        new_node = new_bd_ia32_Cmp8Bit(dbgi, new_block, addr->base,
                                                       addr->index, addr->mem, am.new_op1,
                                                       am.new_op2, am.ins_permuted);
                } else {
                        new_node = new_bd_ia32_Cmp(dbgi, new_block, addr->base, addr->index,
                                                   addr->mem, am.new_op1, am.new_op2,
                                                   am.ins_permuted);
                }
        }
        set_am_attributes(new_node, &am);
        set_ia32_ls_mode(new_node, cmp_mode);

        SET_IA32_ORIG_NODE(new_node, node);

        new_node = fix_mem_proj(new_node, &am);

        return new_node;
}

static ir_node *create_CMov(ir_node *node, ir_node *flags, ir_node *new_flags,
                            ia32_condition_code_t cc)
{
        dbg_info            *dbgi          = get_irn_dbg_info(node);
        ir_node             *block         = get_nodes_block(node);
        ir_node             *new_block     = be_transform_node(block);
        ir_node             *val_true      = get_Mux_true(node);
        ir_node             *val_false     = get_Mux_false(node);
        ir_node             *new_node;
        ia32_address_mode_t  am;
        ia32_address_t      *addr;

        assert(ia32_cg_config.use_cmov);
        assert(ia32_mode_needs_gp_reg(get_irn_mode(val_true)));

        addr = &am.addr;

        match_arguments(&am, block, val_false, val_true, flags,
                        match_commutative | match_am | match_16bit_am | match_mode_neutral);

        if (am.ins_permuted)
                cc = ia32_negate_condition_code(cc);

        new_node = new_bd_ia32_CMovcc(dbgi, new_block, addr->base, addr->index,
                                      addr->mem, am.new_op1, am.new_op2, new_flags,
                                      cc);
        set_am_attributes(new_node, &am);

        SET_IA32_ORIG_NODE(new_node, node);

        new_node = fix_mem_proj(new_node, &am);

        return new_node;
}

/**
 * Creates a ia32 Setcc instruction.
 */
static ir_node *create_set_32bit(dbg_info *dbgi, ir_node *new_block,
                                 ir_node *flags, ia32_condition_code_t cc,
                                 ir_node *orig_node)
{
        ir_mode *mode  = get_irn_mode(orig_node);
        ir_node *new_node;

        new_node = new_bd_ia32_Setcc(dbgi, new_block, flags, cc);
        SET_IA32_ORIG_NODE(new_node, orig_node);

        /* we might need to conv the result up */
        if (get_mode_size_bits(mode) > 8) {
                new_node = new_bd_ia32_Conv_I2I8Bit(dbgi, new_block, noreg_GP, noreg_GP,
                                                    nomem, new_node, mode_Bu);
                SET_IA32_ORIG_NODE(new_node, orig_node);
        }

        return new_node;
}

/**
 * Create instruction for an unsigned Difference or Zero.
 */
static ir_node *create_doz(ir_node *psi, ir_node *a, ir_node *b)
{
        ir_mode *mode  = get_irn_mode(psi);
        ir_node *new_node;
        ir_node *sub;
        ir_node *sbb;
        ir_node *notn;
        ir_node *eflags;
        ir_node *block;

        dbg_info *dbgi;

        new_node = gen_binop(psi, a, b, new_bd_ia32_Sub,
                match_mode_neutral | match_am | match_immediate | match_two_users);

        block = get_nodes_block(new_node);

        if (is_Proj(new_node)) {
                sub = get_Proj_pred(new_node);
        } else {
                sub = new_node;
                set_irn_mode(sub, mode_T);
                new_node = new_rd_Proj(NULL, sub, mode, pn_ia32_res);
        }
        assert(is_ia32_Sub(sub));
        eflags = new_rd_Proj(NULL, sub, mode_Iu, pn_ia32_Sub_flags);

        dbgi = get_irn_dbg_info(psi);
        sbb  = new_bd_ia32_Sbb0(dbgi, block, eflags);
        set_ia32_ls_mode(sbb, mode_Iu);
        notn = new_bd_ia32_Not(dbgi, block, sbb);

        new_node = new_bd_ia32_And(dbgi, block, noreg_GP, noreg_GP, nomem, new_node, notn);
        set_ia32_ls_mode(new_node, mode_Iu);
        set_ia32_commutative(new_node);
        return new_node;
}

/**
 * Create an const array of two float consts.
 *
 * @param c0        the first constant
 * @param c1        the second constant
 * @param new_mode  IN/OUT for the mode of the constants, if NULL
 *                  smallest possible mode will be used
 */
static ir_entity *ia32_create_const_array(ir_node *c0, ir_node *c1, ir_mode **new_mode)
{
        ir_entity        *ent;
        ir_mode          *mode = *new_mode;
        ir_type          *tp;
        ir_initializer_t *initializer;
        ir_tarval        *tv0 = get_Const_tarval(c0);
        ir_tarval        *tv1 = get_Const_tarval(c1);

        if (mode == NULL) {
                /* detect the best mode for the constants */
                mode = get_tarval_mode(tv0);

                if (mode != mode_F) {
                        if (tarval_ieee754_can_conv_lossless(tv0, mode_F) &&
                            tarval_ieee754_can_conv_lossless(tv1, mode_F)) {
                                mode = mode_F;
                                tv0 = tarval_convert_to(tv0, mode);
                                tv1 = tarval_convert_to(tv1, mode);
                        } else if (mode != mode_D) {
                                if (tarval_ieee754_can_conv_lossless(tv0, mode_D) &&
                                    tarval_ieee754_can_conv_lossless(tv1, mode_D)) {
                                        mode = mode_D;
                                        tv0 = tarval_convert_to(tv0, mode);
                                        tv1 = tarval_convert_to(tv1, mode);
                                }
                        }
                }

        }

        tp = ia32_get_prim_type(mode);
        tp = ia32_create_float_array(tp);

        ent = new_entity(get_glob_type(), id_unique("C%u"), tp);

        set_entity_ld_ident(ent, get_entity_ident(ent));
        set_entity_visibility(ent, ir_visibility_private);
        add_entity_linkage(ent, IR_LINKAGE_CONSTANT);

        initializer = create_initializer_compound(2);

        set_initializer_compound_value(initializer, 0, create_initializer_tarval(tv0));
        set_initializer_compound_value(initializer, 1, create_initializer_tarval(tv1));

        set_entity_initializer(ent, initializer);

        *new_mode = mode;
        return ent;
}

/**
 * Possible transformations for creating a Setcc.
 */
enum setcc_transform_insn {
        SETCC_TR_ADD,
        SETCC_TR_ADDxx,
        SETCC_TR_LEA,
        SETCC_TR_LEAxx,
        SETCC_TR_SHL,
        SETCC_TR_NEG,
        SETCC_TR_NOT,
        SETCC_TR_AND,
        SETCC_TR_SET,
};

typedef struct setcc_transform {
        unsigned              num_steps;
        ia32_condition_code_t cc;
        struct {
                enum setcc_transform_insn  transform;
                long val;
                int  scale;
        } steps[4];
} setcc_transform_t;

/**
 * Setcc can only handle 0 and 1 result.
 * Find a transformation that creates 0 and 1 from
 * tv_t and tv_f.
 */
static void find_const_transform(ia32_condition_code_t cc,
                                 ir_tarval *t, ir_tarval *f,
                                 setcc_transform_t *res)
{
        unsigned step = 0;

        res->num_steps = 0;

        if (tarval_is_null(t)) {
                ir_tarval *tmp = t;
                t = f;
                f = tmp;
                cc = ia32_negate_condition_code(cc);
        } else if (tarval_cmp(t, f) == ir_relation_less) {
                // now, t is the bigger one
                ir_tarval *tmp = t;
                t = f;
                f = tmp;
                cc = ia32_negate_condition_code(cc);
        }
        res->cc = cc;

        if (! tarval_is_null(f)) {
                ir_tarval *t_sub = tarval_sub(t, f, NULL);

                t = t_sub;
                res->steps[step].transform = SETCC_TR_ADD;

                if (t == tarval_bad)
                        panic("constant subtract failed");
                if (! tarval_is_long(f))
                        panic("tarval is not long");

                res->steps[step].val = get_tarval_long(f);
                ++step;
                f = tarval_sub(f, f, NULL);
                assert(tarval_is_null(f));
        }

        if (tarval_is_one(t)) {
                res->steps[step].transform = SETCC_TR_SET;
                res->num_steps = ++step;
                return;
        }

        if (tarval_is_minus_one(t)) {
                res->steps[step].transform = SETCC_TR_NEG;
                ++step;
                res->steps[step].transform = SETCC_TR_SET;
                res->num_steps = ++step;
                return;
        }
        if (tarval_is_long(t)) {
                long v = get_tarval_long(t);

                res->steps[step].val = 0;
                switch (v) {
                case 9:
                        if (step > 0 && res->steps[step - 1].transform == SETCC_TR_ADD)
                                --step;
                        res->steps[step].transform = SETCC_TR_LEAxx;
                        res->steps[step].scale     = 3; /* (a << 3) + a */
                        break;
                case 8:
                        if (step > 0 && res->steps[step - 1].transform == SETCC_TR_ADD)
                                --step;
                        res->steps[step].transform = res->steps[step].val == 0 ? SETCC_TR_SHL : SETCC_TR_LEA;
                        res->steps[step].scale     = 3; /* (a << 3) */
                        break;
                case 5:
                        if (step > 0 && res->steps[step - 1].transform == SETCC_TR_ADD)
                                --step;
                        res->steps[step].transform = SETCC_TR_LEAxx;
                        res->steps[step].scale     = 2; /* (a << 2) + a */
                        break;
                case 4:
                        if (step > 0 && res->steps[step - 1].transform == SETCC_TR_ADD)
                                --step;
                        res->steps[step].transform = res->steps[step].val == 0 ? SETCC_TR_SHL : SETCC_TR_LEA;
                        res->steps[step].scale     = 2; /* (a << 2) */
                        break;
                case 3:
                        if (step > 0 && res->steps[step - 1].transform == SETCC_TR_ADD)
                                --step;
                        res->steps[step].transform = SETCC_TR_LEAxx;
                        res->steps[step].scale     = 1; /* (a << 1) + a */
                        break;
                case 2:
                        if (step > 0 && res->steps[step - 1].transform == SETCC_TR_ADD)
                                --step;
                        res->steps[step].transform = res->steps[step].val == 0 ? SETCC_TR_SHL : SETCC_TR_LEA;
                        res->steps[step].scale     = 1; /* (a << 1) */
                        break;
                case 1:
                        res->num_steps = step;
                        return;
                default:
                        if (! tarval_is_single_bit(t)) {
                                res->steps[step].transform = SETCC_TR_AND;
                                res->steps[step].val       = v;
                                ++step;
                                res->steps[step].transform = SETCC_TR_NEG;
                        } else {
                                int val = get_tarval_lowest_bit(t);
                                assert(val >= 0);

                                res->steps[step].transform = SETCC_TR_SHL;
                                res->steps[step].scale     = val;
                        }
                }
                ++step;
                res->steps[step].transform = SETCC_TR_SET;
                res->num_steps = ++step;
                return;
        }
        panic("tarval is not long");
}

/**
 * Transforms a Mux node into some code sequence.
 *
 * @return The transformed node.
 */
static ir_node *gen_Mux(ir_node *node)
{
        dbg_info             *dbgi      = get_irn_dbg_info(node);
        ir_node              *block     = get_nodes_block(node);
        ir_node              *new_block = be_transform_node(block);
        ir_node              *mux_true  = get_Mux_true(node);
        ir_node              *mux_false = get_Mux_false(node);
        ir_node              *sel       = get_Mux_sel(node);
        ir_mode              *mode      = get_irn_mode(node);
        ir_node              *flags;
        ir_node              *new_node;
        int                   is_abs;
        ia32_condition_code_t cc;

        assert(get_irn_mode(sel) == mode_b);

        is_abs = ir_mux_is_abs(sel, mux_false, mux_true);
        if (is_abs != 0) {
                if (ia32_mode_needs_gp_reg(mode)) {
                        ir_fprintf(stderr, "Optimisation warning: Integer abs %+F not transformed\n",
                                   node);
                } else {
                        ir_node *op = ir_get_abs_op(sel, mux_false, mux_true);
                        return create_float_abs(dbgi, block, op, is_abs < 0, node);
                }
        }

        /* Note: a Mux node uses a Load two times IFF it's used in the compare AND in the result */
        if (mode_is_float(mode)) {
                ir_node    *cmp_left  = get_Cmp_left(sel);
                ir_node    *cmp_right = get_Cmp_right(sel);
                ir_relation relation  = get_Cmp_relation(sel);

                if (ia32_cg_config.use_sse2) {
                        if (relation == ir_relation_less || relation == ir_relation_less_equal) {
                                if (cmp_left == mux_true && cmp_right == mux_false) {
                                        /* Mux(a <= b, a, b) => MIN */
                                        return gen_binop(node, cmp_left, cmp_right, new_bd_ia32_xMin,
                                         match_commutative | match_am | match_two_users);
                                } else if (cmp_left == mux_false && cmp_right == mux_true) {
                                        /* Mux(a <= b, b, a) => MAX */
                                        return gen_binop(node, cmp_left, cmp_right, new_bd_ia32_xMax,
                                         match_commutative | match_am | match_two_users);
                                }
                        } else if (relation == ir_relation_greater || relation == ir_relation_greater_equal) {
                                if (cmp_left == mux_true && cmp_right == mux_false) {
                                        /* Mux(a >= b, a, b) => MAX */
                                        return gen_binop(node, cmp_left, cmp_right, new_bd_ia32_xMax,
                                         match_commutative | match_am | match_two_users);
                                } else if (cmp_left == mux_false && cmp_right == mux_true) {
                                        /* Mux(a >= b, b, a) => MIN */
                                        return gen_binop(node, cmp_left, cmp_right, new_bd_ia32_xMin,
                                         match_commutative | match_am | match_two_users);
                                }
                        }
                }

                if (is_Const(mux_true) && is_Const(mux_false)) {
                        ia32_address_mode_t am;
                        ir_node             *load;
                        ir_mode             *new_mode;
                        unsigned            scale;

                        flags    = get_flags_node(sel, &cc);
                        new_node = create_set_32bit(dbgi, new_block, flags, cc, node);

                        if (ia32_cg_config.use_sse2) {
                                /* cannot load from different mode on SSE */
                                new_mode = mode;
                        } else {
                                /* x87 can load any mode */
                                new_mode = NULL;
                        }

                        am.addr.symconst_ent = ia32_create_const_array(mux_false, mux_true, &new_mode);

                        if (new_mode == mode_F) {
                                scale = 2;
                        } else if (new_mode == mode_D) {
                                scale = 3;
                        } else if (new_mode == ia32_mode_E) {
                                /* arg, shift 16 NOT supported */
                                scale = 3;
                                new_node = new_bd_ia32_Lea(dbgi, new_block, new_node, new_node);
                        } else {
                                panic("Unsupported constant size");
                        }

                        am.ls_mode            = new_mode;
                        am.addr.base          = get_symconst_base();
                        am.addr.index         = new_node;
                        am.addr.mem           = nomem;
                        am.addr.offset        = 0;
                        am.addr.scale         = scale;
                        am.addr.use_frame     = 0;
                        am.addr.tls_segment   = false;
                        am.addr.frame_entity  = NULL;
                        am.addr.symconst_sign = 0;
                        am.mem_proj           = am.addr.mem;
                        am.op_type            = ia32_AddrModeS;
                        am.new_op1            = NULL;
                        am.new_op2            = NULL;
                        am.pinned             = op_pin_state_floats;
                        am.commutative        = 1;
                        am.ins_permuted       = false;

                        if (ia32_cg_config.use_sse2)
                                load = new_bd_ia32_xLoad(dbgi, block, am.addr.base, am.addr.index, am.addr.mem, new_mode);
                        else
                                load = new_bd_ia32_vfld(dbgi, block, am.addr.base, am.addr.index, am.addr.mem, new_mode);
                        set_am_attributes(load, &am);

                        return new_rd_Proj(NULL, load, mode_vfp, pn_ia32_res);
                }
                panic("cannot transform floating point Mux");

        } else {
                assert(ia32_mode_needs_gp_reg(mode));

                if (is_Cmp(sel)) {
                        ir_node    *cmp_left  = get_Cmp_left(sel);
                        ir_node    *cmp_right = get_Cmp_right(sel);
                        ir_relation relation  = get_Cmp_relation(sel);
                        ir_node    *val_true  = mux_true;
                        ir_node    *val_false = mux_false;

                        if (is_Const(val_true) && is_Const_null(val_true)) {
                                ir_node *tmp = val_false;
                                val_false = val_true;
                                val_true  = tmp;
                                relation  = get_negated_relation(relation);
                        }
                        if (is_Const_0(val_false) && is_Sub(val_true)) {
                                if ((relation & ir_relation_greater)
                                        && get_Sub_left(val_true) == cmp_left
                                        && get_Sub_right(val_true) == cmp_right) {
                                        return create_doz(node, cmp_left, cmp_right);
                                }
                                if ((relation & ir_relation_less)
                                        && get_Sub_left(val_true) == cmp_right
                                        && get_Sub_right(val_true) == cmp_left) {
                                        return create_doz(node, cmp_right, cmp_left);
                                }
                        }
                }

                flags = get_flags_node(sel, &cc);

                if (is_Const(mux_true) && is_Const(mux_false)) {
                        /* both are const, good */
                        ir_tarval *tv_true  = get_Const_tarval(mux_true);
                        ir_tarval *tv_false = get_Const_tarval(mux_false);
                        setcc_transform_t res;
                        int step;

                        find_const_transform(cc, tv_true, tv_false, &res);
                        new_node = node;
                        for (step = (int)res.num_steps - 1; step >= 0; --step) {
                                ir_node *imm;

                                switch (res.steps[step].transform) {
                                case SETCC_TR_ADD:
                                        new_node = new_bd_ia32_Lea(dbgi, new_block, new_node, noreg_GP);
                                        add_ia32_am_offs_int(new_node, res.steps[step].val);
                                        break;
                                case SETCC_TR_ADDxx:
                                        new_node = new_bd_ia32_Lea(dbgi, new_block, new_node, new_node);
                                        break;
                                case SETCC_TR_LEA:
                                        new_node = new_bd_ia32_Lea(dbgi, new_block, noreg_GP, new_node);
                                        set_ia32_am_scale(new_node, res.steps[step].scale);
                                        set_ia32_am_offs_int(new_node, res.steps[step].val);
                                        break;
                                case SETCC_TR_LEAxx:
                                        new_node = new_bd_ia32_Lea(dbgi, new_block, new_node, new_node);
                                        set_ia32_am_scale(new_node, res.steps[step].scale);
                                        set_ia32_am_offs_int(new_node, res.steps[step].val);
                                        break;
                                case SETCC_TR_SHL:
                                        imm = ia32_immediate_from_long(res.steps[step].scale);
                                        new_node = new_bd_ia32_Shl(dbgi, new_block, new_node, imm);
                                        break;
                                case SETCC_TR_NEG:
                                        new_node = new_bd_ia32_Neg(dbgi, new_block, new_node);
                                        break;
                                case SETCC_TR_NOT:
                                        new_node = new_bd_ia32_Not(dbgi, new_block, new_node);
                                        break;
                                case SETCC_TR_AND:
                                        imm = ia32_immediate_from_long(res.steps[step].val);
                                        new_node = new_bd_ia32_And(dbgi, new_block, noreg_GP, noreg_GP, nomem, new_node, imm);
                                        break;
                                case SETCC_TR_SET:
                                        new_node = create_set_32bit(dbgi, new_block, flags, res.cc, node);
                                        break;
                                default:
                                        panic("unknown setcc transform");
                                }
                        }
                } else {
                        new_node = create_CMov(node, sel, flags, cc);
                }
                return new_node;
        }
}

/**
 * Create a conversion from x87 state register to general purpose.
 */
static ir_node *gen_x87_fp_to_gp(ir_node *node)
{
        ir_node         *block      = be_transform_node(get_nodes_block(node));
        ir_node         *op         = get_Conv_op(node);
        ir_node         *new_op     = be_transform_node(op);
        ir_graph        *irg        = current_ir_graph;
        dbg_info        *dbgi       = get_irn_dbg_info(node);
        ir_mode         *mode       = get_irn_mode(node);
        ir_node         *frame      = get_irg_frame(irg);
        ir_node         *fist, *load, *mem;

        fist = gen_vfist(dbgi, block, frame, noreg_GP, nomem, new_op);
        set_irn_pinned(fist, op_pin_state_floats);
        set_ia32_use_frame(fist);
        set_ia32_op_type(fist, ia32_AddrModeD);

        assert((long)pn_ia32_vfist_M == (long) pn_ia32_vfisttp_M);
        mem = new_r_Proj(fist, mode_M, pn_ia32_vfist_M);

        assert(get_mode_size_bits(mode) <= 32);
        /* exception we can only store signed 32 bit integers, so for unsigned
           we store a 64bit (signed) integer and load the lower bits */
        if (get_mode_size_bits(mode) == 32 && !mode_is_signed(mode)) {
                set_ia32_ls_mode(fist, mode_Ls);
        } else {
                set_ia32_ls_mode(fist, mode_Is);
        }
        SET_IA32_ORIG_NODE(fist, node);

        /* do a Load */
        load = new_bd_ia32_Load(dbgi, block, get_irg_frame(irg), noreg_GP, mem);

        set_irn_pinned(load, op_pin_state_floats);
        set_ia32_use_frame(load);
        set_ia32_op_type(load, ia32_AddrModeS);
        set_ia32_ls_mode(load, mode_Is);
        if (get_ia32_ls_mode(fist) == mode_Ls) {
                ia32_attr_t *attr = get_ia32_attr(load);
                attr->data.need_64bit_stackent = 1;
        } else {
                ia32_attr_t *attr = get_ia32_attr(load);
                attr->data.need_32bit_stackent = 1;
        }
        SET_IA32_ORIG_NODE(load, node);

        return new_r_Proj(load, mode_Iu, pn_ia32_Load_res);
}

/**
 * Creates a x87 Conv by placing a Store and a Load
 */
static ir_node *gen_x87_conv(ir_mode *tgt_mode, ir_node *node)
{
        ir_node  *block    = get_nodes_block(node);
        ir_graph *irg      = get_Block_irg(block);
        dbg_info *dbgi     = get_irn_dbg_info(node);
        ir_node  *frame    = get_irg_frame(irg);
        ir_node  *store_mem;
        ir_node  *store, *load;
        ir_node  *new_node;

        store = new_bd_ia32_vfst(dbgi, block, frame, noreg_GP, nomem, node, tgt_mode);
        set_ia32_use_frame(store);
        set_ia32_op_type(store, ia32_AddrModeD);
        SET_IA32_ORIG_NODE(store, node);

        store_mem = new_r_Proj(store, mode_M, pn_ia32_vfst_M);

        load = new_bd_ia32_vfld(dbgi, block, frame, noreg_GP, store_mem, tgt_mode);
        set_ia32_use_frame(load);
        set_ia32_op_type(load, ia32_AddrModeS);
        SET_IA32_ORIG_NODE(load, node);

        new_node = new_r_Proj(load, ia32_mode_E, pn_ia32_vfld_res);
        return new_node;
}

static ir_node *create_Conv_I2I(dbg_info *dbgi, ir_node *block, ir_node *base,
                ir_node *index, ir_node *mem, ir_node *val, ir_mode *mode)
{
        ir_node *(*func)(dbg_info*, ir_node*, ir_node*, ir_node*, ir_node*, ir_node*, ir_mode*);

        func = get_mode_size_bits(mode) == 8 ?
                new_bd_ia32_Conv_I2I8Bit : new_bd_ia32_Conv_I2I;
        return func(dbgi, block, base, index, mem, val, mode);
}

/**
 * Create a conversion from general purpose to x87 register
 */
static ir_node *gen_x87_gp_to_fp(ir_node *node, ir_mode *src_mode)
{
        ir_node  *src_block = get_nodes_block(node);
        ir_node  *block     = be_transform_node(src_block);
        ir_graph *irg       = get_Block_irg(block);
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_node  *op        = get_Conv_op(node);
        ir_node  *new_op    = NULL;
        ir_mode  *mode;
        ir_mode  *store_mode;
        ir_node  *fild;
        ir_node  *store;
        ir_node  *store_mem;
        ir_node  *new_node;

        /* fild can use source AM if the operand is a signed 16bit or 32bit integer */
        if (possible_int_mode_for_fp(src_mode)) {
                ia32_address_mode_t am;

                match_arguments(&am, src_block, NULL, op, NULL, match_am | match_try_am | match_16bit_am | match_upconv);
                if (am.op_type == ia32_AddrModeS) {
                        ia32_address_t *addr = &am.addr;

                        fild     = new_bd_ia32_vfild(dbgi, block, addr->base, addr->index, addr->mem);
                        new_node = new_r_Proj(fild, mode_vfp, pn_ia32_vfild_res);

                        set_am_attributes(fild, &am);
                        SET_IA32_ORIG_NODE(fild, node);

                        fix_mem_proj(fild, &am);

                        return new_node;
                }
        }
        if (new_op == NULL) {
                new_op = be_transform_node(op);
        }

        mode = get_irn_mode(op);

        /* first convert to 32 bit signed if necessary */
        if (get_mode_size_bits(src_mode) < 32) {
                if (!be_upper_bits_clean(op, src_mode)) {
                        new_op = create_Conv_I2I(dbgi, block, noreg_GP, noreg_GP, nomem, new_op, src_mode);
                        SET_IA32_ORIG_NODE(new_op, node);
                }
                mode = mode_Is;
        }

        assert(get_mode_size_bits(mode) == 32);

        /* do a store */
        store = new_bd_ia32_Store(dbgi, block, get_irg_frame(irg), noreg_GP, nomem, new_op);

        set_ia32_use_frame(store);
        set_ia32_op_type(store, ia32_AddrModeD);
        set_ia32_ls_mode(store, mode_Iu);

        store_mem = new_r_Proj(store, mode_M, pn_ia32_Store_M);

        /* exception for 32bit unsigned, do a 64bit spill+load */
        if (!mode_is_signed(mode)) {
                ir_node *in[2];
                /* store a zero */
                ir_node *zero_const = ia32_create_Immediate(NULL, 0, 0);

                ir_node *zero_store = new_bd_ia32_Store(dbgi, block, get_irg_frame(irg),
                                                        noreg_GP, nomem, zero_const);
                ir_node *zero_store_mem = new_r_Proj(zero_store, mode_M, pn_ia32_Store_M);

                set_ia32_use_frame(zero_store);
                set_ia32_op_type(zero_store, ia32_AddrModeD);
                add_ia32_am_offs_int(zero_store, 4);
                set_ia32_ls_mode(zero_store, mode_Iu);

                in[0] = zero_store_mem;
                in[1] = store_mem;

                store_mem  = new_rd_Sync(dbgi, block, 2, in);
                store_mode = mode_Ls;
        } else {
                store_mode = mode_Is;
        }

        /* do a fild */
        fild = new_bd_ia32_vfild(dbgi, block, get_irg_frame(irg), noreg_GP, store_mem);

        set_ia32_use_frame(fild);
        set_ia32_op_type(fild, ia32_AddrModeS);
        set_ia32_ls_mode(fild, store_mode);

        new_node = new_r_Proj(fild, mode_vfp, pn_ia32_vfild_res);

        return new_node;
}

/**
 * Create a conversion from one integer mode into another one
 */
static ir_node *create_I2I_Conv(ir_mode *src_mode, ir_mode *tgt_mode,
                                dbg_info *dbgi, ir_node *block, ir_node *op,
                                ir_node *node)
{
        ir_node             *new_block = be_transform_node(block);
        ir_node             *new_node;
        ia32_address_mode_t  am;
        ia32_address_t      *addr = &am.addr;

        (void) node;
        assert(get_mode_size_bits(src_mode) < get_mode_size_bits(tgt_mode));

#ifdef DEBUG_libfirm
        if (is_Const(op)) {
                ir_fprintf(stderr, "Optimisation warning: conv after constant %+F\n",
                           op);
        }
#endif

        if (be_upper_bits_clean(op, src_mode)) {
                return be_transform_node(op);
        }

        match_arguments(&am, block, NULL, op, NULL,
                        match_am | match_8bit_am | match_16bit_am);

        new_node = create_Conv_I2I(dbgi, new_block, addr->base, addr->index,
                                   addr->mem, am.new_op2, src_mode);
        set_am_attributes(new_node, &am);
        /* match_arguments assume that out-mode = in-mode, this isn't true here
         * so fix it */
        set_ia32_ls_mode(new_node, src_mode);
        SET_IA32_ORIG_NODE(new_node, node);
        new_node = fix_mem_proj(new_node, &am);
        return new_node;
}

/**
 * Transforms a Conv node.
 *
 * @return The created ia32 Conv node
 */
static ir_node *gen_Conv(ir_node *node)
{
        ir_node  *block     = get_nodes_block(node);
        ir_node  *new_block = be_transform_node(block);
        ir_node  *op        = get_Conv_op(node);
        ir_node  *new_op    = NULL;
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_mode  *src_mode  = get_irn_mode(op);
        ir_mode  *tgt_mode  = get_irn_mode(node);
        int       src_bits  = get_mode_size_bits(src_mode);
        int       tgt_bits  = get_mode_size_bits(tgt_mode);
        ir_node  *res       = NULL;

        assert(!mode_is_int(src_mode) || src_bits <= 32);
        assert(!mode_is_int(tgt_mode) || tgt_bits <= 32);

        /* modeB -> X should already be lowered by the lower_mode_b pass */
        if (src_mode == mode_b) {
                panic("ConvB not lowered %+F", node);
        }

        if (src_mode == tgt_mode) {
                /* this should be optimized already, but who knows... */
                DEBUG_ONLY(ir_fprintf(stderr, "Debug warning: conv %+F is pointless\n", node);)
                        DB((dbg, LEVEL_1, "killed Conv(mode, mode) ..."));
                return be_transform_node(op);
        }

        if (mode_is_float(src_mode)) {
                new_op = be_transform_node(op);
                /* we convert from float ... */
                if (mode_is_float(tgt_mode)) {
                        /* ... to float */
                        if (ia32_cg_config.use_sse2) {
                                DB((dbg, LEVEL_1, "create Conv(float, float) ..."));
                                res = new_bd_ia32_Conv_FP2FP(dbgi, new_block, noreg_GP, noreg_GP,
                                                             nomem, new_op);
                                set_ia32_ls_mode(res, tgt_mode);
                        } else {
                                if (src_bits < tgt_bits) {
                                        DB((dbg, LEVEL_1, "killed Conv(float, float) ..."));
                                        return new_op;
                                } else {
                                        res = gen_x87_conv(tgt_mode, new_op);
                                        SET_IA32_ORIG_NODE(get_Proj_pred(res), node);
                                        return res;
                                }
                        }
                } else {
                        /* ... to int */
                        DB((dbg, LEVEL_1, "create Conv(float, int) ..."));
                        if (ia32_cg_config.use_sse2) {
                                res = new_bd_ia32_Conv_FP2I(dbgi, new_block, noreg_GP, noreg_GP,
                                                            nomem, new_op);
                                set_ia32_ls_mode(res, src_mode);
                        } else {
                                return gen_x87_fp_to_gp(node);
                        }
                }
        } else {
                /* we convert from int ... */
                if (mode_is_float(tgt_mode)) {
                        /* ... to float */
                        DB((dbg, LEVEL_1, "create Conv(int, float) ..."));
                        if (ia32_cg_config.use_sse2) {
                                new_op = be_transform_node(op);
                                res = new_bd_ia32_Conv_I2FP(dbgi, new_block, noreg_GP, noreg_GP,
                                                            nomem, new_op);
                                set_ia32_ls_mode(res, tgt_mode);
                        } else {
                                unsigned int_mantissa   = get_mode_size_bits(src_mode) - (mode_is_signed(src_mode) ? 1 : 0);
                                unsigned float_mantissa = get_mode_mantissa_size(tgt_mode);
                                res = gen_x87_gp_to_fp(node, src_mode);

                                /* we need a float-conv, if the int mode has more bits than the
                                 * float mantissa */
                                if (float_mantissa < int_mantissa) {
                                        res = gen_x87_conv(tgt_mode, res);
                                        SET_IA32_ORIG_NODE(get_Proj_pred(res), node);
                                }
                                return res;
                        }
                } else if (tgt_mode == mode_b) {
                        /* mode_b lowering already took care that we only have 0/1 values */
                        DB((dbg, LEVEL_1, "omitting unnecessary Conv(%+F, %+F) ...",
                            src_mode, tgt_mode));
                        return be_transform_node(op);
                } else {
                        /* to int */
                        if (src_bits >= tgt_bits) {
                                DB((dbg, LEVEL_1, "omitting unnecessary Conv(%+F, %+F) ...",
                                    src_mode, tgt_mode));
                                return be_transform_node(op);
                        }

                        res = create_I2I_Conv(src_mode, tgt_mode, dbgi, block, op, node);
                        return res;
                }
        }

        return res;
}

static ir_node *create_immediate_or_transform(ir_node *node,
                                              char immediate_constraint_type)
{
        ir_node *new_node = ia32_try_create_Immediate(node, immediate_constraint_type);
        if (new_node == NULL) {
                new_node = be_transform_node(node);
        }
        return new_node;
}

/**
 * Transforms a FrameAddr into an ia32 Add.
 */
static ir_node *gen_be_FrameAddr(ir_node *node)
{
        ir_node  *block  = be_transform_node(get_nodes_block(node));
        ir_node  *op     = be_get_FrameAddr_frame(node);
        ir_node  *new_op = be_transform_node(op);
        dbg_info *dbgi   = get_irn_dbg_info(node);
        ir_node  *new_node;

        new_node = new_bd_ia32_Lea(dbgi, block, new_op, noreg_GP);
        set_ia32_frame_ent(new_node, arch_get_frame_entity(node));
        set_ia32_use_frame(new_node);

        SET_IA32_ORIG_NODE(new_node, node);

        return new_node;
}

/**
 * In case SSE is used we need to copy the result from XMM0 to FPU TOS before return.
 */
static ir_node *gen_be_Return(ir_node *node)
{
        ir_graph  *irg         = current_ir_graph;
        ir_node   *ret_val     = get_irn_n(node, n_be_Return_val);
        ir_node   *ret_mem     = get_irn_n(node, n_be_Return_mem);
        ir_node   *new_ret_val = be_transform_node(ret_val);
        ir_node   *new_ret_mem = be_transform_node(ret_mem);
        ir_entity *ent         = get_irg_entity(irg);
        ir_type   *tp          = get_entity_type(ent);
        dbg_info  *dbgi        = get_irn_dbg_info(node);
        ir_node   *block       = be_transform_node(get_nodes_block(node));
        ir_type   *res_type;
        ir_mode   *mode;
        ir_node   *frame;
        ir_node   *sse_store;
        ir_node   *store_mem;
        ir_node   *fld;
        ir_node   *mproj;
        int        i;
        int        arity;
        unsigned   pop;
        ir_node  **in;
        ir_node   *new_node;

        assert(ret_val != NULL);
        if (be_Return_get_n_rets(node) < 1 || ! ia32_cg_config.use_sse2) {
                return be_duplicate_node(node);
        }

        res_type = get_method_res_type(tp, 0);

        if (! is_Primitive_type(res_type)) {
                return be_duplicate_node(node);
        }

        mode = get_type_mode(res_type);
        if (! mode_is_float(mode)) {
                return be_duplicate_node(node);
        }

        assert(get_method_n_ress(tp) == 1);

        frame = get_irg_frame(irg);

        /* store xmm0 onto stack */
        sse_store = new_bd_ia32_xStoreSimple(dbgi, block, frame, noreg_GP,
                                             new_ret_mem, new_ret_val);
        set_ia32_ls_mode(sse_store, mode);
        set_ia32_op_type(sse_store, ia32_AddrModeD);
        set_ia32_use_frame(sse_store);
        store_mem = new_r_Proj(sse_store, mode_M, pn_ia32_xStoreSimple_M);

        /* load into x87 register */
        fld = new_bd_ia32_vfld(dbgi, block, frame, noreg_GP, store_mem, mode);
        set_ia32_op_type(fld, ia32_AddrModeS);
        set_ia32_use_frame(fld);

        mproj = new_r_Proj(fld, mode_M, pn_ia32_vfld_M);
        fld   = new_r_Proj(fld, mode_vfp, pn_ia32_vfld_res);

        /* create a new return */
        arity = get_irn_arity(node);
        in    = ALLOCAN(ir_node*, arity);
        pop   = be_Return_get_pop(node);
        for (i = 0; i < arity; ++i) {
                ir_node *op = get_irn_n(node, i);
                if (op == ret_val) {
                        in[i] = fld;
                } else if (op == ret_mem) {
                        in[i] = mproj;
                } else {
                        in[i] = be_transform_node(op);
                }
        }
        new_node = be_new_Return(dbgi, irg, block, arity, pop, arity, in);
        copy_node_attr(irg, node, new_node);

        return new_node;
}

/**
 * Transform a be_AddSP into an ia32_SubSP.
 */
static ir_node *gen_be_AddSP(ir_node *node)
{
        ir_node *sz = get_irn_n(node, n_be_AddSP_size);
        ir_node *sp = get_irn_n(node, n_be_AddSP_old_sp);

        ir_node *new_node = gen_binop(node, sp, sz, new_bd_ia32_SubSP,
                                      match_am | match_immediate);
        assert(is_ia32_SubSP(new_node));
        arch_set_irn_register_out(new_node, pn_ia32_SubSP_stack,
                                  &ia32_registers[REG_ESP]);
        return new_node;
}

/**
 * Transform a be_SubSP into an ia32_AddSP
 */
static ir_node *gen_be_SubSP(ir_node *node)
{
        ir_node *sz = get_irn_n(node, n_be_SubSP_size);
        ir_node *sp = get_irn_n(node, n_be_SubSP_old_sp);

        ir_node *new_node = gen_binop(node, sp, sz, new_bd_ia32_AddSP,
                                      match_am | match_immediate);
        assert(is_ia32_AddSP(new_node));
        arch_set_irn_register_out(new_node, pn_ia32_AddSP_stack,
                                  &ia32_registers[REG_ESP]);
        return new_node;
}

/**
 * Change some phi modes
 */
static ir_node *gen_Phi(ir_node *node)
{
        const arch_register_req_t *req;
        ir_node  *block = be_transform_node(get_nodes_block(node));
        ir_graph *irg   = current_ir_graph;
        dbg_info *dbgi  = get_irn_dbg_info(node);
        ir_mode  *mode  = get_irn_mode(node);
        ir_node  *phi;

        if (ia32_mode_needs_gp_reg(mode)) {
                /* we shouldn't have any 64bit stuff around anymore */
                assert(get_mode_size_bits(mode) <= 32);
                /* all integer operations are on 32bit registers now */
                mode = mode_Iu;
                req  = ia32_reg_classes[CLASS_ia32_gp].class_req;
        } else if (mode_is_float(mode)) {
                if (ia32_cg_config.use_sse2) {
                        mode = mode_xmm;
                        req  = ia32_reg_classes[CLASS_ia32_xmm].class_req;
                } else {
                        mode = mode_vfp;
                        req  = ia32_reg_classes[CLASS_ia32_vfp].class_req;
                }
        } else {
                req = arch_no_register_req;
        }

        /* phi nodes allow loops, so we use the old arguments for now
         * and fix this later */
        phi = new_ir_node(dbgi, irg, block, op_Phi, mode, get_irn_arity(node),
                          get_irn_in(node) + 1);
        copy_node_attr(irg, node, phi);
        be_duplicate_deps(node, phi);

        arch_set_irn_register_req_out(phi, 0, req);

        be_enqueue_preds(node);

        return phi;
}

static ir_node *gen_Jmp(ir_node *node)
{
        ir_node  *block     = get_nodes_block(node);
        ir_node  *new_block = be_transform_node(block);
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_node  *new_node;

        new_node = new_bd_ia32_Jmp(dbgi, new_block);
        SET_IA32_ORIG_NODE(new_node, node);

        return new_node;
}

/**
 * Transform IJmp
 */
static ir_node *gen_IJmp(ir_node *node)
{
        ir_node  *block     = get_nodes_block(node);
        ir_node  *new_block = be_transform_node(block);
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_node  *op        = get_IJmp_target(node);
        ir_node  *new_node;
        ia32_address_mode_t  am;
        ia32_address_t      *addr = &am.addr;

        assert(get_irn_mode(op) == mode_P);

        match_arguments(&am, block, NULL, op, NULL,
                        match_am | match_immediate | match_upconv);

        new_node = new_bd_ia32_IJmp(dbgi, new_block, addr->base, addr->index,
                        addr->mem, am.new_op2);
        set_am_attributes(new_node, &am);
        SET_IA32_ORIG_NODE(new_node, node);

        new_node = fix_mem_proj(new_node, &am);

        return new_node;
}

static ir_node *gen_ia32_l_Add(ir_node *node)
{
        ir_node *left    = get_irn_n(node, n_ia32_l_Add_left);
        ir_node *right   = get_irn_n(node, n_ia32_l_Add_right);
        ir_node *lowered = gen_binop(node, left, right, new_bd_ia32_Add,
                        match_commutative | match_am | match_immediate |
                        match_mode_neutral);

        if (is_Proj(lowered)) {
                lowered = get_Proj_pred(lowered);
        } else {
                assert(is_ia32_Add(lowered));
                set_irn_mode(lowered, mode_T);
        }

        return lowered;
}

static ir_node *gen_ia32_l_Adc(ir_node *node)
{
        return gen_binop_flags(node, new_bd_ia32_Adc,
                        match_commutative | match_am | match_immediate |
                        match_mode_neutral);
}

/**
 * Transforms a l_MulS into a "real" MulS node.
 *
 * @return the created ia32 Mul node
 */
static ir_node *gen_ia32_l_Mul(ir_node *node)
{
        ir_node *left  = get_binop_left(node);
        ir_node *right = get_binop_right(node);

        return gen_binop(node, left, right, new_bd_ia32_Mul,
                         match_commutative | match_am | match_mode_neutral);
}

/**
 * Transforms a l_IMulS into a "real" IMul1OPS node.
 *
 * @return the created ia32 IMul1OP node
 */
static ir_node *gen_ia32_l_IMul(ir_node *node)
{
        ir_node  *left  = get_binop_left(node);
        ir_node  *right = get_binop_right(node);

        return gen_binop(node, left, right, new_bd_ia32_IMul1OP,
                         match_commutative | match_am | match_mode_neutral);
}

static ir_node *gen_ia32_l_Sub(ir_node *node)
{
        ir_node *left    = get_irn_n(node, n_ia32_l_Sub_minuend);
        ir_node *right   = get_irn_n(node, n_ia32_l_Sub_subtrahend);
        ir_node *lowered = gen_binop(node, left, right, new_bd_ia32_Sub,
                        match_am | match_immediate | match_mode_neutral);

        if (is_Proj(lowered)) {
                lowered = get_Proj_pred(lowered);
        } else {
                assert(is_ia32_Sub(lowered));
                set_irn_mode(lowered, mode_T);
        }

        return lowered;
}

static ir_node *gen_ia32_l_Sbb(ir_node *node)
{
        return gen_binop_flags(node, new_bd_ia32_Sbb,
                        match_am | match_immediate | match_mode_neutral);
}

static ir_node *gen_ia32_l_LLtoFloat(ir_node *node)
{
        ir_node  *src_block    = get_nodes_block(node);
        ir_node  *block        = be_transform_node(src_block);
        ir_graph *irg          = current_ir_graph;
        dbg_info *dbgi         = get_irn_dbg_info(node);
        ir_node  *frame        = get_irg_frame(irg);
        ir_node  *val_low      = get_irn_n(node, n_ia32_l_LLtoFloat_val_low);
        ir_node  *val_high     = get_irn_n(node, n_ia32_l_LLtoFloat_val_high);
        ir_node  *new_val_low  = be_transform_node(val_low);
        ir_node  *new_val_high = be_transform_node(val_high);
        ir_node  *in[2];
        ir_node  *sync, *fild, *res;
        ir_node  *store_low;
        ir_node  *store_high;
        ir_node  *mem_low;
        ir_node  *mem_high;

        if (ia32_cg_config.use_sse2) {
                panic("not implemented for SSE2");
        }

        /* do a store */
        store_low = new_bd_ia32_Store(dbgi, block, frame, noreg_GP, nomem,
                                      new_val_low);
        store_high = new_bd_ia32_Store(dbgi, block, frame, noreg_GP, nomem,
                                       new_val_high);
        SET_IA32_ORIG_NODE(store_low,  node);
        SET_IA32_ORIG_NODE(store_high, node);

        mem_low  = new_r_Proj(store_low, mode_M, pn_ia32_Store_M);
        mem_high = new_r_Proj(store_high, mode_M, pn_ia32_Store_M);

        set_ia32_use_frame(store_low);
        set_ia32_use_frame(store_high);
        set_ia32_op_type(store_low, ia32_AddrModeD);
        set_ia32_op_type(store_high, ia32_AddrModeD);
        set_ia32_ls_mode(store_low, mode_Iu);
        set_ia32_ls_mode(store_high, mode_Is);
        add_ia32_am_offs_int(store_high, 4);

        in[0] = mem_low;
        in[1] = mem_high;
        sync  = new_rd_Sync(dbgi, block, 2, in);

        /* do a fild */
        fild = new_bd_ia32_vfild(dbgi, block, frame, noreg_GP, sync);

        set_ia32_use_frame(fild);
        set_ia32_op_type(fild, ia32_AddrModeS);
        set_ia32_ls_mode(fild, mode_Ls);

        SET_IA32_ORIG_NODE(fild, node);

        res = new_r_Proj(fild, mode_vfp, pn_ia32_vfild_res);

        if (! mode_is_signed(get_irn_mode(val_high))) {
                ia32_address_mode_t  am;

                ir_node *count = ia32_create_Immediate(NULL, 0, 31);
                ir_node *fadd;

                am.addr.base          = get_symconst_base();
                am.addr.index         = new_bd_ia32_Shr(dbgi, block, new_val_high, count);
                am.addr.mem           = nomem;
                am.addr.offset        = 0;
                am.addr.scale         = 2;
                am.addr.symconst_ent  = ia32_gen_fp_known_const(ia32_ULLBIAS);
                am.addr.tls_segment   = false;
                am.addr.use_frame     = 0;
                am.addr.frame_entity  = NULL;
                am.addr.symconst_sign = 0;
                am.ls_mode            = mode_F;
                am.mem_proj           = nomem;
                am.op_type            = ia32_AddrModeS;
                am.new_op1            = res;
                am.new_op2            = ia32_new_NoReg_vfp(current_ir_graph);
                am.pinned             = op_pin_state_floats;
                am.commutative        = 1;
                am.ins_permuted       = false;

                fadd  = new_bd_ia32_vfadd(dbgi, block, am.addr.base, am.addr.index, am.addr.mem,
                        am.new_op1, am.new_op2, get_fpcw());
                set_am_attributes(fadd, &am);

                set_irn_mode(fadd, mode_T);
                res = new_rd_Proj(NULL, fadd, mode_vfp, pn_ia32_res);
        }
        return res;
}

static ir_node *gen_ia32_l_FloattoLL(ir_node *node)
{
        ir_node  *src_block  = get_nodes_block(node);
        ir_node  *block      = be_transform_node(src_block);
        ir_graph *irg        = get_Block_irg(block);
        dbg_info *dbgi       = get_irn_dbg_info(node);
        ir_node  *frame      = get_irg_frame(irg);
        ir_node  *val        = get_irn_n(node, n_ia32_l_FloattoLL_val);
        ir_node  *new_val    = be_transform_node(val);
        ir_node  *fist;

        fist = gen_vfist(dbgi, block, frame, noreg_GP, nomem, new_val);
        SET_IA32_ORIG_NODE(fist, node);
        set_ia32_use_frame(fist);
        set_ia32_op_type(fist, ia32_AddrModeD);
        set_ia32_ls_mode(fist, mode_Ls);

        assert((long)pn_ia32_vfist_M == (long) pn_ia32_vfisttp_M);
        return new_r_Proj(fist, mode_M, pn_ia32_vfist_M);
}

static ir_node *gen_Proj_l_FloattoLL(ir_node *node)
{
        ir_node  *block    = be_transform_node(get_nodes_block(node));
        ir_graph *irg      = get_Block_irg(block);
        ir_node  *pred     = get_Proj_pred(node);
        ir_node  *new_pred = be_transform_node(pred);
        ir_node  *frame    = get_irg_frame(irg);
        dbg_info *dbgi     = get_irn_dbg_info(node);
        long      pn       = get_Proj_proj(node);
        ir_node  *load;
        ir_node  *proj;
        ia32_attr_t *attr;

        load = new_bd_ia32_Load(dbgi, block, frame, noreg_GP, new_pred);
        SET_IA32_ORIG_NODE(load, node);
        set_ia32_use_frame(load);
        set_ia32_op_type(load, ia32_AddrModeS);
        set_ia32_ls_mode(load, mode_Iu);
        /* we need a 64bit stackslot (fist stores 64bit) even though we only load
         * 32 bit from it with this particular load */
        attr = get_ia32_attr(load);
        attr->data.need_64bit_stackent = 1;

        if (pn == pn_ia32_l_FloattoLL_res_high) {
                add_ia32_am_offs_int(load, 4);
        } else {
                assert(pn == pn_ia32_l_FloattoLL_res_low);
        }

        proj = new_r_Proj(load, mode_Iu, pn_ia32_Load_res);

        return proj;
}

/**
 * Transform the Projs of an AddSP.
 */
static ir_node *gen_Proj_be_AddSP(ir_node *node)
{
        ir_node  *pred     = get_Proj_pred(node);
        ir_node  *new_pred = be_transform_node(pred);
        dbg_info *dbgi     = get_irn_dbg_info(node);
        long     proj      = get_Proj_proj(node);

        if (proj == pn_be_AddSP_sp) {
                ir_node *res = new_rd_Proj(dbgi, new_pred, mode_Iu,
                                           pn_ia32_SubSP_stack);
                arch_set_irn_register(res, &ia32_registers[REG_ESP]);
                return res;
        } else if (proj == pn_be_AddSP_res) {
                return new_rd_Proj(dbgi, new_pred, mode_Iu,
                                   pn_ia32_SubSP_addr);
        } else if (proj == pn_be_AddSP_M) {
                return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_SubSP_M);
        }

        panic("No idea how to transform proj->AddSP");
}

/**
 * Transform the Projs of a SubSP.
 */
static ir_node *gen_Proj_be_SubSP(ir_node *node)
{
        ir_node  *pred     = get_Proj_pred(node);
        ir_node  *new_pred = be_transform_node(pred);
        dbg_info *dbgi     = get_irn_dbg_info(node);
        long     proj      = get_Proj_proj(node);

        if (proj == pn_be_SubSP_sp) {
                ir_node *res = new_rd_Proj(dbgi, new_pred, mode_Iu,
                                           pn_ia32_AddSP_stack);
                arch_set_irn_register(res, &ia32_registers[REG_ESP]);
                return res;
        } else if (proj == pn_be_SubSP_M) {
                return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_AddSP_M);
        }

        panic("No idea how to transform proj->SubSP");
}

/**
 * Transform and renumber the Projs from a Load.
 */
static ir_node *gen_Proj_Load(ir_node *node)
{
        ir_node  *new_pred;
        ir_node  *pred = get_Proj_pred(node);
        dbg_info *dbgi = get_irn_dbg_info(node);
        long      proj = get_Proj_proj(node);

        /* loads might be part of source address mode matches, so we don't
         * transform the ProjMs yet (with the exception of loads whose result is
         * not used)
         */
        if (is_Load(pred) && proj == pn_Load_M && get_irn_n_edges(pred) > 1) {
                ir_node *res;

                /* this is needed, because sometimes we have loops that are only
                   reachable through the ProjM */
                be_enqueue_preds(node);
                /* do it in 2 steps, to silence firm verifier */
                res = new_rd_Proj(dbgi, pred, mode_M, pn_Load_M);
                set_Proj_proj(res, pn_ia32_mem);
                return res;
        }

        /* renumber the proj */
        new_pred = be_transform_node(pred);
        if (is_ia32_Load(new_pred)) {
                switch ((pn_Load)proj) {
                case pn_Load_res:
                        return new_rd_Proj(dbgi, new_pred, mode_Iu, pn_ia32_Load_res);
                case pn_Load_M:
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_Load_M);
                case pn_Load_X_except:
                        /* This Load might raise an exception. Mark it. */
                        set_ia32_exc_label(new_pred, 1);
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_Load_X_except);
                case pn_Load_X_regular:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_Load_X_regular);
                }
        } else if (is_ia32_Conv_I2I(new_pred) ||
                   is_ia32_Conv_I2I8Bit(new_pred)) {
                set_irn_mode(new_pred, mode_T);
                switch ((pn_Load)proj) {
                case pn_Load_res:
                        return new_rd_Proj(dbgi, new_pred, mode_Iu, pn_ia32_res);
                case pn_Load_M:
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_mem);
                case pn_Load_X_except:
                        /* This Load might raise an exception. Mark it. */
                        set_ia32_exc_label(new_pred, 1);
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_Conv_I2I_X_except);
                case pn_Load_X_regular:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_Conv_I2I_X_regular);
                }
        } else if (is_ia32_xLoad(new_pred)) {
                switch ((pn_Load)proj) {
                case pn_Load_res:
                        return new_rd_Proj(dbgi, new_pred, mode_xmm, pn_ia32_xLoad_res);
                case pn_Load_M:
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_xLoad_M);
                case pn_Load_X_except:
                        /* This Load might raise an exception. Mark it. */
                        set_ia32_exc_label(new_pred, 1);
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_xLoad_X_except);
                case pn_Load_X_regular:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_xLoad_X_regular);
                }
        } else if (is_ia32_vfld(new_pred)) {
                switch ((pn_Load)proj) {
                case pn_Load_res:
                        return new_rd_Proj(dbgi, new_pred, mode_vfp, pn_ia32_vfld_res);
                case pn_Load_M:
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_vfld_M);
                case pn_Load_X_except:
                        /* This Load might raise an exception. Mark it. */
                        set_ia32_exc_label(new_pred, 1);
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_vfld_X_except);
                case pn_Load_X_regular:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_vfld_X_regular);
                }
        } else {
                /* can happen for ProJMs when source address mode happened for the
                   node */

                /* however it should not be the result proj, as that would mean the
                   load had multiple users and should not have been used for
                   SourceAM */
                if (proj != pn_Load_M) {
                        panic("internal error: transformed node not a Load");
                }
                return new_rd_Proj(dbgi, new_pred, mode_M, 1);
        }

        panic("No idea how to transform Proj(Load) %+F", node);
}

static ir_node *gen_Proj_Store(ir_node *node)
{
        ir_node  *pred     = get_Proj_pred(node);
        ir_node  *new_pred = be_transform_node(pred);
        dbg_info *dbgi     = get_irn_dbg_info(node);
        long      pn       = get_Proj_proj(node);

        if (is_ia32_Store(new_pred) || is_ia32_Store8Bit(new_pred)) {
                switch ((pn_Store)pn) {
                case pn_Store_M:
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_Store_M);
                case pn_Store_X_except:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_Store_X_except);
                case pn_Store_X_regular:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_Store_X_regular);
                }
        } else if (is_ia32_vfist(new_pred)) {
                switch ((pn_Store)pn) {
                case pn_Store_M:
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_vfist_M);
                case pn_Store_X_except:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_vfist_X_except);
                case pn_Store_X_regular:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_vfist_X_regular);
                }
        } else if (is_ia32_vfisttp(new_pred)) {
                switch ((pn_Store)pn) {
                case pn_Store_M:
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_vfisttp_M);
                case pn_Store_X_except:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_vfisttp_X_except);
                case pn_Store_X_regular:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_vfisttp_X_regular);
                }
        } else if (is_ia32_vfst(new_pred)) {
                switch ((pn_Store)pn) {
                case pn_Store_M:
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_vfst_M);
                case pn_Store_X_except:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_vfst_X_except);
                case pn_Store_X_regular:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_vfst_X_regular);
                }
        } else if (is_ia32_xStore(new_pred)) {
                switch ((pn_Store)pn) {
                case pn_Store_M:
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_xStore_M);
                case pn_Store_X_except:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_xStore_X_except);
                case pn_Store_X_regular:
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_xStore_X_regular);
                }
        } else if (is_Sync(new_pred)) {
                /* hack for the case that gen_float_const_Store produced a Sync */
                if (pn == pn_Store_M) {
                        return new_pred;
                }
                panic("exception control flow not implemented yet");
        } else if (get_ia32_op_type(new_pred) == ia32_AddrModeD) {
                /* destination address mode */
                if (pn == pn_Store_M) {
                        return new_pred;
                }
                panic("exception control flow for destination AM not implemented yet");
        }

        panic("No idea how to transform Proj(Store) %+F", node);
}

/**
 * Transform and renumber the Projs from a Div or Mod instruction.
 */
static ir_node *gen_Proj_Div(ir_node *node)
{
        ir_node  *pred     = get_Proj_pred(node);
        ir_node  *new_pred = be_transform_node(pred);
        dbg_info *dbgi     = get_irn_dbg_info(node);
        long      proj     = get_Proj_proj(node);

        assert((long)pn_ia32_Div_M == (long)pn_ia32_IDiv_M);
        assert((long)pn_ia32_Div_div_res == (long)pn_ia32_IDiv_div_res);

        switch ((pn_Div)proj) {
        case pn_Div_M:
                if (is_ia32_Div(new_pred) || is_ia32_IDiv(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_Div_M);
                } else if (is_ia32_xDiv(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_xDiv_M);
                } else if (is_ia32_vfdiv(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_vfdiv_M);
                } else {
                        panic("Div transformed to unexpected thing %+F", new_pred);
                }
        case pn_Div_res:
                if (is_ia32_Div(new_pred) || is_ia32_IDiv(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_Iu, pn_ia32_Div_div_res);
                } else if (is_ia32_xDiv(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_xmm, pn_ia32_xDiv_res);
                } else if (is_ia32_vfdiv(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_vfp, pn_ia32_vfdiv_res);
                } else {
                        panic("Div transformed to unexpected thing %+F", new_pred);
                }
        case pn_Div_X_except:
                set_ia32_exc_label(new_pred, 1);
                return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_Div_X_except);
        case pn_Div_X_regular:
                return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_Div_X_regular);
        }

        panic("No idea how to transform proj->Div");
}

/**
 * Transform and renumber the Projs from a Div or Mod instruction.
 */
static ir_node *gen_Proj_Mod(ir_node *node)
{
        ir_node  *pred     = get_Proj_pred(node);
        ir_node  *new_pred = be_transform_node(pred);
        dbg_info *dbgi     = get_irn_dbg_info(node);
        long     proj      = get_Proj_proj(node);

        assert(is_ia32_Div(new_pred) || is_ia32_IDiv(new_pred));
        assert((long)pn_ia32_Div_M == (long)pn_ia32_IDiv_M);
        assert((long)pn_ia32_Div_mod_res == (long)pn_ia32_IDiv_mod_res);

        switch ((pn_Mod)proj) {
        case pn_Mod_M:
                return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_Div_M);
        case pn_Mod_res:
                return new_rd_Proj(dbgi, new_pred, mode_Iu, pn_ia32_Div_mod_res);
        case pn_Mod_X_except:
                set_ia32_exc_label(new_pred, 1);
                return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_Div_X_except);
        case pn_Mod_X_regular:
                return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_Div_X_regular);
        }
        panic("No idea how to transform proj->Mod");
}

/**
 * Transform and renumber the Projs from a CopyB.
 */
static ir_node *gen_Proj_CopyB(ir_node *node)
{
        ir_node  *pred     = get_Proj_pred(node);
        ir_node  *new_pred = be_transform_node(pred);
        dbg_info *dbgi     = get_irn_dbg_info(node);
        long     proj      = get_Proj_proj(node);

        switch ((pn_CopyB)proj) {
        case pn_CopyB_M:
                if (is_ia32_CopyB_i(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_CopyB_i_M);
                } else if (is_ia32_CopyB(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_M, pn_ia32_CopyB_M);
                }
                break;
        case pn_CopyB_X_regular:
                if (is_ia32_CopyB_i(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_CopyB_i_X_regular);
                } else if (is_ia32_CopyB(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_CopyB_X_regular);
                }
                break;
        case pn_CopyB_X_except:
                if (is_ia32_CopyB_i(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_CopyB_i_X_except);
                } else if (is_ia32_CopyB(new_pred)) {
                        return new_rd_Proj(dbgi, new_pred, mode_X, pn_ia32_CopyB_X_except);
                }
                break;
        }

        panic("No idea how to transform proj->CopyB");
}

static ir_node *gen_be_Call(ir_node *node)
{
        dbg_info       *const dbgi      = get_irn_dbg_info(node);
        ir_node        *const src_block = get_nodes_block(node);
        ir_node        *const block     = be_transform_node(src_block);
        ir_node        *const src_mem   = get_irn_n(node, n_be_Call_mem);
        ir_node        *const src_sp    = get_irn_n(node, n_be_Call_sp);
        ir_node        *const sp        = be_transform_node(src_sp);
        ir_node        *const src_ptr   = get_irn_n(node, n_be_Call_ptr);
        ia32_address_mode_t   am;
        ia32_address_t *const addr      = &am.addr;
        ir_node        *      mem;
        ir_node        *      call;
        int                   i;
        ir_node        *      fpcw;
        ir_node        *      eax       = noreg_GP;
        ir_node        *      ecx       = noreg_GP;
        ir_node        *      edx       = noreg_GP;
        unsigned        const pop       = be_Call_get_pop(node);
        ir_type        *const call_tp   = be_Call_get_type(node);
        int                   old_no_pic_adjust;
        int                   throws_exception = ir_throws_exception(node);

        /* Run the x87 simulator if the call returns a float value */
        if (get_method_n_ress(call_tp) > 0) {
                ir_type *const res_type = get_method_res_type(call_tp, 0);
                ir_mode *const res_mode = get_type_mode(res_type);

                if (res_mode != NULL && mode_is_float(res_mode)) {
                        ir_graph        *irg      = current_ir_graph;
                        ia32_irg_data_t *irg_data = ia32_get_irg_data(irg);
                        irg_data->do_x87_sim = 1;
                }
        }

        /* We do not want be_Call direct calls */
        assert(be_Call_get_entity(node) == NULL);

        /* special case for PIC trampoline calls */
        old_no_pic_adjust  = ia32_no_pic_adjust;
        ia32_no_pic_adjust = be_options.pic;

        match_arguments(&am, src_block, NULL, src_ptr, src_mem,
                        match_am | match_immediate | match_upconv);

        ia32_no_pic_adjust = old_no_pic_adjust;

        i    = get_irn_arity(node) - 1;
        fpcw = be_transform_node(get_irn_n(node, i--));
        for (; i >= n_be_Call_first_arg; --i) {
                arch_register_req_t const *const req
                        = arch_get_irn_register_req_in(node, i);
                ir_node *const reg_parm = be_transform_node(get_irn_n(node, i));

                assert(req->type == arch_register_req_type_limited);
                assert(req->cls == &ia32_reg_classes[CLASS_ia32_gp]);

                switch (*req->limited) {
                        case 1 << REG_GP_EAX: assert(eax == noreg_GP); eax = reg_parm; break;
                        case 1 << REG_GP_ECX: assert(ecx == noreg_GP); ecx = reg_parm; break;
                        case 1 << REG_GP_EDX: assert(edx == noreg_GP); edx = reg_parm; break;
                        default: panic("Invalid GP register for register parameter");
                }
        }

        mem  = transform_AM_mem(block, src_ptr, src_mem, addr->mem);
        call = new_bd_ia32_Call(dbgi, block, addr->base, addr->index, mem,
                                am.new_op2, sp, fpcw, eax, ecx, edx, pop, call_tp);
        ir_set_throws_exception(call, throws_exception);
        set_am_attributes(call, &am);
        call = fix_mem_proj(call, &am);

        if (get_irn_pinned(node) == op_pin_state_pinned)
                set_irn_pinned(call, op_pin_state_pinned);

        SET_IA32_ORIG_NODE(call, node);

        if (ia32_cg_config.use_sse2) {
                /* remember this call for post-processing */
                ARR_APP1(ir_node *, call_list, call);
                ARR_APP1(ir_type *, call_types, be_Call_get_type(node));
        }

        return call;
}

/**
 * Transform Builtin trap
 */
static ir_node *gen_trap(ir_node *node)
{
        dbg_info *dbgi  = get_irn_dbg_info(node);
        ir_node *block  = be_transform_node(get_nodes_block(node));
        ir_node *mem    = be_transform_node(get_Builtin_mem(node));

        return new_bd_ia32_UD2(dbgi, block, mem);
}

/**
 * Transform Builtin debugbreak
 */
static ir_node *gen_debugbreak(ir_node *node)
{
        dbg_info *dbgi  = get_irn_dbg_info(node);
        ir_node *block  = be_transform_node(get_nodes_block(node));
        ir_node *mem    = be_transform_node(get_Builtin_mem(node));

        return new_bd_ia32_Breakpoint(dbgi, block, mem);
}

/**
 * Transform Builtin return_address
 */
static ir_node *gen_return_address(ir_node *node)
{
        ir_node   *param    = get_Builtin_param(node, 0);
        ir_node   *frame    = get_Builtin_param(node, 1);
        dbg_info  *dbgi     = get_irn_dbg_info(node);
        ir_tarval *tv       = get_Const_tarval(param);
        ir_graph  *irg      = get_irn_irg(node);
        unsigned long value = get_tarval_long(tv);

        ir_node *block  = be_transform_node(get_nodes_block(node));
        ir_node *ptr    = be_transform_node(frame);
        ir_node *load;

        if (value > 0) {
                ir_node *cnt = new_bd_ia32_ProduceVal(dbgi, block);
                ir_node *res = new_bd_ia32_ProduceVal(dbgi, block);
                ptr = new_bd_ia32_ClimbFrame(dbgi, block, ptr, cnt, res, value);
        }

        /* load the return address from this frame */
        load = new_bd_ia32_Load(dbgi, block, ptr, noreg_GP, nomem);

        set_irn_pinned(load, get_irn_pinned(node));
        set_ia32_op_type(load, ia32_AddrModeS);
        set_ia32_ls_mode(load, mode_Iu);

        set_ia32_am_offs_int(load, 0);
        set_ia32_use_frame(load);
        set_ia32_frame_ent(load, ia32_get_return_address_entity(irg));

        if (get_irn_pinned(node) == op_pin_state_floats) {
                assert((int)pn_ia32_xLoad_res == (int)pn_ia32_vfld_res
                                && (int)pn_ia32_vfld_res == (int)pn_ia32_Load_res
                                && (int)pn_ia32_Load_res == (int)pn_ia32_res);
                arch_add_irn_flags(load, arch_irn_flags_rematerializable);
        }

        SET_IA32_ORIG_NODE(load, node);
        return new_r_Proj(load, mode_Iu, pn_ia32_Load_res);
}

/**
 * Transform Builtin frame_address
 */
static ir_node *gen_frame_address(ir_node *node)
{
        ir_node   *param    = get_Builtin_param(node, 0);
        ir_node   *frame    = get_Builtin_param(node, 1);
        dbg_info  *dbgi     = get_irn_dbg_info(node);
        ir_tarval *tv       = get_Const_tarval(param);
        ir_graph  *irg      = get_irn_irg(node);
        unsigned long value = get_tarval_long(tv);

        ir_node *block  = be_transform_node(get_nodes_block(node));
        ir_node *ptr    = be_transform_node(frame);
        ir_node *load;
        ir_entity *ent;

        if (value > 0) {
                ir_node *cnt = new_bd_ia32_ProduceVal(dbgi, block);
                ir_node *res = new_bd_ia32_ProduceVal(dbgi, block);
                ptr = new_bd_ia32_ClimbFrame(dbgi, block, ptr, cnt, res, value);
        }

        /* load the frame address from this frame */
        load = new_bd_ia32_Load(dbgi, block, ptr, noreg_GP, nomem);

        set_irn_pinned(load, get_irn_pinned(node));
        set_ia32_op_type(load, ia32_AddrModeS);
        set_ia32_ls_mode(load, mode_Iu);

        ent = ia32_get_frame_address_entity(irg);
        if (ent != NULL) {
                set_ia32_am_offs_int(load, 0);
                set_ia32_use_frame(load);
                set_ia32_frame_ent(load, ent);
        } else {
                /* will fail anyway, but gcc does this: */
                set_ia32_am_offs_int(load, 0);
        }

        if (get_irn_pinned(node) == op_pin_state_floats) {
                assert((int)pn_ia32_xLoad_res == (int)pn_ia32_vfld_res
                                && (int)pn_ia32_vfld_res == (int)pn_ia32_Load_res
                                && (int)pn_ia32_Load_res == (int)pn_ia32_res);
                arch_add_irn_flags(load, arch_irn_flags_rematerializable);
        }

        SET_IA32_ORIG_NODE(load, node);
        return new_r_Proj(load, mode_Iu, pn_ia32_Load_res);
}

/**
 * Transform Builtin frame_address
 */
static ir_node *gen_prefetch(ir_node *node)
{
        dbg_info       *dbgi;
        ir_node        *ptr, *block, *mem, *base, *idx;
        ir_node        *param,  *new_node;
        long           rw, locality;
        ir_tarval      *tv;
        ia32_address_t addr;

        if (!ia32_cg_config.use_sse_prefetch && !ia32_cg_config.use_3dnow_prefetch) {
                /* no prefetch at all, route memory */
                return be_transform_node(get_Builtin_mem(node));
        }

        param = get_Builtin_param(node, 1);
        tv    = get_Const_tarval(param);
        rw    = get_tarval_long(tv);

        /* construct load address */
        memset(&addr, 0, sizeof(addr));
        ptr = get_Builtin_param(node, 0);
        ia32_create_address_mode(&addr, ptr, ia32_create_am_normal);
        base = addr.base;
        idx  = addr.index;

        if (base == NULL) {
                base = noreg_GP;
        } else {
                base = be_transform_node(base);
        }

        if (idx == NULL) {
                idx = noreg_GP;
        } else {
                idx = be_transform_node(idx);
        }

        dbgi     = get_irn_dbg_info(node);
        block    = be_transform_node(get_nodes_block(node));
        mem      = be_transform_node(get_Builtin_mem(node));

        if (rw == 1 && ia32_cg_config.use_3dnow_prefetch) {
                /* we have 3DNow!, this was already checked above */
                new_node = new_bd_ia32_PrefetchW(dbgi, block, base, idx, mem);
        } else if (ia32_cg_config.use_sse_prefetch) {
                /* note: rw == 1 is IGNORED in that case */
                param    = get_Builtin_param(node, 2);
                tv       = get_Const_tarval(param);
                locality = get_tarval_long(tv);

                /* SSE style prefetch */
                switch (locality) {
                case 0:
                        new_node = new_bd_ia32_PrefetchNTA(dbgi, block, base, idx, mem);
                        break;
                case 1:
                        new_node = new_bd_ia32_Prefetch2(dbgi, block, base, idx, mem);
                        break;
                case 2:
                        new_node = new_bd_ia32_Prefetch1(dbgi, block, base, idx, mem);
                        break;
                default:
                        new_node = new_bd_ia32_Prefetch0(dbgi, block, base, idx, mem);
                        break;
                }
        } else {
                assert(ia32_cg_config.use_3dnow_prefetch);
                /* 3DNow! style prefetch */
                new_node = new_bd_ia32_Prefetch(dbgi, block, base, idx, mem);
        }

        set_irn_pinned(new_node, get_irn_pinned(node));
        set_ia32_op_type(new_node, ia32_AddrModeS);
        set_ia32_ls_mode(new_node, mode_Bu);
        set_address(new_node, &addr);

        SET_IA32_ORIG_NODE(new_node, node);

        return new_r_Proj(new_node, mode_M, pn_ia32_Prefetch_M);
}

/**
 * Transform bsf like node
 */
static ir_node *gen_unop_AM(ir_node *node, construct_binop_dest_func *func)
{
        ir_node *param     = get_Builtin_param(node, 0);
        dbg_info *dbgi     = get_irn_dbg_info(node);

        ir_node *block     = get_nodes_block(node);
        ir_node *new_block = be_transform_node(block);

        ia32_address_mode_t  am;
        ia32_address_t      *addr = &am.addr;
        ir_node             *cnt;

        match_arguments(&am, block, NULL, param, NULL, match_am);

        cnt = func(dbgi, new_block, addr->base, addr->index, addr->mem, am.new_op2);
        set_am_attributes(cnt, &am);
        set_ia32_ls_mode(cnt, get_irn_mode(param));

        SET_IA32_ORIG_NODE(cnt, node);
        return fix_mem_proj(cnt, &am);
}

/**
 * Transform builtin ffs.
 */
static ir_node *gen_ffs(ir_node *node)
{
        ir_node  *bsf   = gen_unop_AM(node, new_bd_ia32_Bsf);
        ir_node  *real  = skip_Proj(bsf);
        dbg_info *dbgi  = get_irn_dbg_info(real);
        ir_node  *block = get_nodes_block(real);
        ir_node  *flag, *set, *conv, *neg, *orn, *add;

        /* bsf x */
        if (get_irn_mode(real) != mode_T) {
                set_irn_mode(real, mode_T);
                bsf = new_r_Proj(real, mode_Iu, pn_ia32_res);
        }

        flag = new_r_Proj(real, mode_b, pn_ia32_flags);

        /* sete */
        set = new_bd_ia32_Setcc(dbgi, block, flag, ia32_cc_equal);
        SET_IA32_ORIG_NODE(set, node);

        /* conv to 32bit */
        conv = new_bd_ia32_Conv_I2I8Bit(dbgi, block, noreg_GP, noreg_GP, nomem, set, mode_Bu);
        SET_IA32_ORIG_NODE(conv, node);

        /* neg */
        neg = new_bd_ia32_Neg(dbgi, block, conv);

        /* or */
        orn = new_bd_ia32_Or(dbgi, block, noreg_GP, noreg_GP, nomem, bsf, neg);
        set_ia32_ls_mode(orn, mode_Iu);
        set_ia32_commutative(orn);

        /* add 1 */
        add = new_bd_ia32_Lea(dbgi, block, orn, noreg_GP);
        add_ia32_am_offs_int(add, 1);
        return add;
}

/**
 * Transform builtin clz.
 */
static ir_node *gen_clz(ir_node *node)
{
        ir_node  *bsr   = gen_unop_AM(node, new_bd_ia32_Bsr);
        ir_node  *real  = skip_Proj(bsr);
        dbg_info *dbgi  = get_irn_dbg_info(real);
        ir_node  *block = get_nodes_block(real);
        ir_node  *imm   = ia32_create_Immediate(NULL, 0, 31);

        return new_bd_ia32_Xor(dbgi, block, noreg_GP, noreg_GP, nomem, bsr, imm);
}

/**
 * Transform builtin ctz.
 */
static ir_node *gen_ctz(ir_node *node)
{
        return gen_unop_AM(node, new_bd_ia32_Bsf);
}

/**
 * Transform builtin parity.
 */
static ir_node *gen_parity(ir_node *node)
{
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ir_node  *block     = get_nodes_block(node);
        ir_node  *new_block = be_transform_node(block);
        ir_node  *param     = get_Builtin_param(node, 0);
        ir_node  *new_param = be_transform_node(param);
        ir_node  *new_node;

        /* the x86 parity bit is stupid: it only looks at the lowest byte,
         * so we have to do complicated xoring first.
         * (we should also better lower this before the backend so we still have a
         * chance for CSE, constant folding and other goodies for some of these
         * operations)
         */
        ir_node *count = ia32_create_Immediate(NULL, 0, 16);
        ir_node *shr   = new_bd_ia32_Shr(dbgi, new_block, new_param, count);
        ir_node *xorn  = new_bd_ia32_Xor(dbgi, new_block, noreg_GP, noreg_GP, nomem,
                                         shr, new_param);
        ir_node *xor2  = new_bd_ia32_XorHighLow(dbgi, new_block, xorn);
        ir_node *flags;

        set_ia32_ls_mode(xorn, mode_Iu);
        set_ia32_commutative(xorn);

        set_irn_mode(xor2, mode_T);
        flags = new_r_Proj(xor2, mode_Iu, pn_ia32_XorHighLow_flags);

        /* setp */
        new_node = new_bd_ia32_Setcc(dbgi, new_block, flags, ia32_cc_not_parity);
        SET_IA32_ORIG_NODE(new_node, node);

        /* conv to 32bit */
        new_node = new_bd_ia32_Conv_I2I8Bit(dbgi, new_block, noreg_GP, noreg_GP,
                                            nomem, new_node, mode_Bu);
        SET_IA32_ORIG_NODE(new_node, node);
        return new_node;
}

/**
 * Transform builtin popcount
 */
static ir_node *gen_popcount(ir_node *node)
{
        ir_node *param     = get_Builtin_param(node, 0);
        dbg_info *dbgi     = get_irn_dbg_info(node);

        ir_node *block     = get_nodes_block(node);
        ir_node *new_block = be_transform_node(block);

        ir_node *new_param;
        ir_node *imm, *simm, *m1, *s1, *s2, *s3, *s4, *s5, *m2, *m3, *m4, *m5, *m6, *m7, *m8, *m9, *m10, *m11, *m12, *m13;

        /* check for SSE4.2 or SSE4a and use the popcnt instruction */
        if (ia32_cg_config.use_popcnt) {
                ia32_address_mode_t am;
                ia32_address_t      *addr = &am.addr;
                ir_node             *cnt;

                match_arguments(&am, block, NULL, param, NULL, match_am | match_16bit_am | match_upconv);

                cnt = new_bd_ia32_Popcnt(dbgi, new_block, addr->base, addr->index, addr->mem, am.new_op2);
                set_am_attributes(cnt, &am);
                set_ia32_ls_mode(cnt, get_irn_mode(param));

                SET_IA32_ORIG_NODE(cnt, node);
                return fix_mem_proj(cnt, &am);
        }

        new_param = be_transform_node(param);

        /* do the standard popcount algo */
        /* TODO: This is stupid, we should transform this before the backend,
         * to get CSE, localopts, etc. for the operations
         * TODO: This is also not the optimal algorithm (it is just the starting
         * example in hackers delight, they optimize it more on the following page)
         * But I'm too lazy to fix this now, as the code should get lowered before
         * the backend anyway.
         */

        /* m1 = x & 0x55555555 */
        imm = ia32_create_Immediate(NULL, 0, 0x55555555);
        m1 = new_bd_ia32_And(dbgi, new_block, noreg_GP, noreg_GP, nomem, new_param, imm);

        /* s1 = x >> 1 */
        simm = ia32_create_Immediate(NULL, 0, 1);
        s1 = new_bd_ia32_Shr(dbgi, new_block, new_param, simm);

        /* m2 = s1 & 0x55555555 */
        m2 = new_bd_ia32_And(dbgi, new_block, noreg_GP, noreg_GP, nomem, s1, imm);

        /* m3 = m1 + m2 */
        m3 = new_bd_ia32_Lea(dbgi, new_block, m2, m1);

        /* m4 = m3 & 0x33333333 */
        imm = ia32_create_Immediate(NULL, 0, 0x33333333);
        m4 = new_bd_ia32_And(dbgi, new_block, noreg_GP, noreg_GP, nomem, m3, imm);

        /* s2 = m3 >> 2 */
        simm = ia32_create_Immediate(NULL, 0, 2);
        s2 = new_bd_ia32_Shr(dbgi, new_block, m3, simm);

        /* m5 = s2 & 0x33333333 */
        m5 = new_bd_ia32_And(dbgi, new_block, noreg_GP, noreg_GP, nomem, s2, imm);

        /* m6 = m4 + m5 */
        m6 = new_bd_ia32_Lea(dbgi, new_block, m4, m5);

        /* m7 = m6 & 0x0F0F0F0F */
        imm = ia32_create_Immediate(NULL, 0, 0x0F0F0F0F);
        m7 = new_bd_ia32_And(dbgi, new_block, noreg_GP, noreg_GP, nomem, m6, imm);

        /* s3 = m6 >> 4 */
        simm = ia32_create_Immediate(NULL, 0, 4);
        s3 = new_bd_ia32_Shr(dbgi, new_block, m6, simm);

        /* m8 = s3 & 0x0F0F0F0F */
        m8 = new_bd_ia32_And(dbgi, new_block, noreg_GP, noreg_GP, nomem, s3, imm);

        /* m9 = m7 + m8 */
        m9 = new_bd_ia32_Lea(dbgi, new_block, m7, m8);

        /* m10 = m9 & 0x00FF00FF */
        imm = ia32_create_Immediate(NULL, 0, 0x00FF00FF);
        m10 = new_bd_ia32_And(dbgi, new_block, noreg_GP, noreg_GP, nomem, m9, imm);

        /* s4 = m9 >> 8 */
        simm = ia32_create_Immediate(NULL, 0, 8);
        s4 = new_bd_ia32_Shr(dbgi, new_block, m9, simm);

        /* m11 = s4 & 0x00FF00FF */
        m11 = new_bd_ia32_And(dbgi, new_block, noreg_GP, noreg_GP, nomem, s4, imm);

        /* m12 = m10 + m11 */
        m12 = new_bd_ia32_Lea(dbgi, new_block, m10, m11);

        /* m13 = m12 & 0x0000FFFF */
        imm = ia32_create_Immediate(NULL, 0, 0x0000FFFF);
        m13 = new_bd_ia32_And(dbgi, new_block, noreg_GP, noreg_GP, nomem, m12, imm);

        /* s5 = m12 >> 16 */
        simm = ia32_create_Immediate(NULL, 0, 16);
        s5 = new_bd_ia32_Shr(dbgi, new_block, m12, simm);

        /* res = m13 + s5 */
        return new_bd_ia32_Lea(dbgi, new_block, m13, s5);
}

/**
 * Transform builtin byte swap.
 */
static ir_node *gen_bswap(ir_node *node)
{
        ir_node *param     = be_transform_node(get_Builtin_param(node, 0));
        dbg_info *dbgi     = get_irn_dbg_info(node);

        ir_node *block     = get_nodes_block(node);
        ir_node *new_block = be_transform_node(block);
        ir_mode *mode      = get_irn_mode(param);
        unsigned size      = get_mode_size_bits(mode);

        switch (size) {
        case 32:
                if (ia32_cg_config.use_bswap) {
                        /* swap available */
                        return new_bd_ia32_Bswap(dbgi, new_block, param);
                } else {
                        ir_node *i8 = ia32_create_Immediate(NULL, 0, 8);
                        ir_node *rol1 = new_bd_ia32_Rol(dbgi, new_block, param, i8);
                        ir_node *i16 = ia32_create_Immediate(NULL, 0, 16);
                        ir_node *rol2 = new_bd_ia32_Rol(dbgi, new_block, rol1, i16);
                        ir_node *rol3 = new_bd_ia32_Rol(dbgi, new_block, rol2, i8);
                        set_ia32_ls_mode(rol1, mode_Hu);
                        set_ia32_ls_mode(rol2, mode_Iu);
                        set_ia32_ls_mode(rol3, mode_Hu);
                        return rol3;
                }

        case 16:
                /* swap16 always available */
                return new_bd_ia32_Bswap16(dbgi, new_block, param);

        default:
                panic("Invalid bswap size (%d)", size);
        }
}

/**
 * Transform builtin outport.
 */
static ir_node *gen_outport(ir_node *node)
{
        ir_node *port  = create_immediate_or_transform(get_Builtin_param(node, 0), 0);
        ir_node *oldv  = get_Builtin_param(node, 1);
        ir_mode *mode  = get_irn_mode(oldv);
        ir_node *value = be_transform_node(oldv);
        ir_node *block = be_transform_node(get_nodes_block(node));
        ir_node *mem   = be_transform_node(get_Builtin_mem(node));
        dbg_info *dbgi = get_irn_dbg_info(node);

        ir_node *res = new_bd_ia32_Outport(dbgi, block, port, value, mem);
        set_ia32_ls_mode(res, mode);
        return res;
}

/**
 * Transform builtin inport.
 */
static ir_node *gen_inport(ir_node *node)
{
        ir_type *tp    = get_Builtin_type(node);
        ir_type *rstp  = get_method_res_type(tp, 0);
        ir_mode *mode  = get_type_mode(rstp);
        ir_node *port  = create_immediate_or_transform(get_Builtin_param(node, 0), 0);
        ir_node *block = be_transform_node(get_nodes_block(node));
        ir_node *mem   = be_transform_node(get_Builtin_mem(node));
        dbg_info *dbgi = get_irn_dbg_info(node);

        ir_node *res = new_bd_ia32_Inport(dbgi, block, port, mem);
        set_ia32_ls_mode(res, mode);

        /* check for missing Result Proj */
        return res;
}

/**
 * Transform a builtin inner trampoline
 */
static ir_node *gen_inner_trampoline(ir_node *node)
{
        ir_node  *ptr       = get_Builtin_param(node, 0);
        ir_node  *callee    = get_Builtin_param(node, 1);
        ir_node  *env       = be_transform_node(get_Builtin_param(node, 2));
        ir_node  *mem       = get_Builtin_mem(node);
        ir_node  *block     = get_nodes_block(node);
        ir_node  *new_block = be_transform_node(block);
        ir_node  *val;
        ir_node  *store;
        ir_node  *rel;
        ir_node  *trampoline;
        ir_node  *in[2];
        dbg_info *dbgi      = get_irn_dbg_info(node);
        ia32_address_t addr;

        /* construct store address */
        memset(&addr, 0, sizeof(addr));
        ia32_create_address_mode(&addr, ptr, ia32_create_am_normal);

        if (addr.base == NULL) {
                addr.base = noreg_GP;
        } else {
                addr.base = be_transform_node(addr.base);
        }

        if (addr.index == NULL) {
                addr.index = noreg_GP;
        } else {
                addr.index = be_transform_node(addr.index);
        }
        addr.mem = be_transform_node(mem);

        /* mov  ecx, <env> */
        val   = ia32_create_Immediate(NULL, 0, 0xB9);
        store = new_bd_ia32_Store8Bit(dbgi, new_block, addr.base,
                                      addr.index, addr.mem, val);
        set_irn_pinned(store, get_irn_pinned(node));
        set_ia32_op_type(store, ia32_AddrModeD);
        set_ia32_ls_mode(store, mode_Bu);
        set_address(store, &addr);
        addr.mem = store;
        addr.offset += 1;

        store = new_bd_ia32_Store(dbgi, new_block, addr.base,
                                  addr.index, addr.mem, env);
        set_irn_pinned(store, get_irn_pinned(node));
        set_ia32_op_type(store, ia32_AddrModeD);
        set_ia32_ls_mode(store, mode_Iu);
        set_address(store, &addr);
        addr.mem = store;
        addr.offset += 4;

        /* jmp rel <callee> */
        val   = ia32_create_Immediate(NULL, 0, 0xE9);
        store = new_bd_ia32_Store8Bit(dbgi, new_block, addr.base,
                                     addr.index, addr.mem, val);
        set_irn_pinned(store, get_irn_pinned(node));
        set_ia32_op_type(store, ia32_AddrModeD);
        set_ia32_ls_mode(store, mode_Bu);
        set_address(store, &addr);
        addr.mem = store;
        addr.offset += 1;

        trampoline = be_transform_node(ptr);

        /* the callee is typically an immediate */
        if (is_SymConst(callee)) {
                rel = new_bd_ia32_Const(dbgi, new_block, get_SymConst_entity(callee), 0, 0, -10);
        } else {
                rel = new_bd_ia32_Lea(dbgi, new_block, be_transform_node(callee), noreg_GP);
                add_ia32_am_offs_int(rel, -10);
        }
        rel = new_bd_ia32_Sub(dbgi, new_block, noreg_GP, noreg_GP, nomem, rel, trampoline);

        store = new_bd_ia32_Store(dbgi, new_block, addr.base,
                                  addr.index, addr.mem, rel);
        set_irn_pinned(store, get_irn_pinned(node));
        set_ia32_op_type(store, ia32_AddrModeD);
        set_ia32_ls_mode(store, mode_Iu);
        set_address(store, &addr);

        in[0] = store;
        in[1] = trampoline;

        return new_r_Tuple(new_block, 2, in);
}

/**
 * Transform Builtin node.
 */
static ir_node *gen_Builtin(ir_node *node)
{
        ir_builtin_kind kind = get_Builtin_kind(node);

        switch (kind) {
        case ir_bk_trap:
                return gen_trap(node);
        case ir_bk_debugbreak:
                return gen_debugbreak(node);
        case ir_bk_return_address:
                return gen_return_address(node);
        case ir_bk_frame_address:
                return gen_frame_address(node);
        case ir_bk_prefetch:
                return gen_prefetch(node);
        case ir_bk_ffs:
                return gen_ffs(node);
        case ir_bk_clz:
                return gen_clz(node);
        case ir_bk_ctz:
                return gen_ctz(node);
        case ir_bk_parity:
                return gen_parity(node);
        case ir_bk_popcount:
                return gen_popcount(node);
        case ir_bk_bswap:
                return gen_bswap(node);
        case ir_bk_outport:
                return gen_outport(node);
        case ir_bk_inport:
                return gen_inport(node);
        case ir_bk_inner_trampoline:
                return gen_inner_trampoline(node);
        }
        panic("Builtin %s not implemented", get_builtin_kind_name(kind));
}

/**
 * Transform Proj(Builtin) node.
 */
static ir_node *gen_Proj_Builtin(ir_node *proj)
{
        ir_node         *node     = get_Proj_pred(proj);
        ir_node         *new_node = be_transform_node(node);
        ir_builtin_kind kind      = get_Builtin_kind(node);

        switch (kind) {
        case ir_bk_return_address:
        case ir_bk_frame_address:
        case ir_bk_ffs:
        case ir_bk_clz:
        case ir_bk_ctz:
        case ir_bk_parity:
        case ir_bk_popcount:
        case ir_bk_bswap:
                assert(get_Proj_proj(proj) == pn_Builtin_max+1);
                return new_node;
        case ir_bk_trap:
        case ir_bk_debugbreak:
        case ir_bk_prefetch:
        case ir_bk_outport:
                assert(get_Proj_proj(proj) == pn_Builtin_M);
                return new_node;
        case ir_bk_inport:
                if (get_Proj_proj(proj) == pn_Builtin_max+1) {
                        return new_r_Proj(new_node, get_irn_mode(proj), pn_ia32_Inport_res);
                } else {
                        assert(get_Proj_proj(proj) == pn_Builtin_M);
                        return new_r_Proj(new_node, mode_M, pn_ia32_Inport_M);
                }
        case ir_bk_inner_trampoline:
                if (get_Proj_proj(proj) == pn_Builtin_max+1) {
                        return get_Tuple_pred(new_node, 1);
                } else {
                        assert(get_Proj_proj(proj) == pn_Builtin_M);
                        return get_Tuple_pred(new_node, 0);
                }
        }
        panic("Builtin %s not implemented", get_builtin_kind_name(kind));
}

static ir_node *gen_be_IncSP(ir_node *node)
{
        ir_node *res = be_duplicate_node(node);
        arch_add_irn_flags(res, arch_irn_flags_modify_flags);

        return res;
}

/**
 * Transform the Projs from a be_Call.
 */
static ir_node *gen_Proj_be_Call(ir_node *node)
{
        ir_node  *call        = get_Proj_pred(node);
        ir_node  *new_call    = be_transform_node(call);
        dbg_info *dbgi        = get_irn_dbg_info(node);
        long      proj        = get_Proj_proj(node);
        ir_mode  *mode        = get_irn_mode(node);
        ir_node  *res;

        if (proj == pn_be_Call_M) {
                return new_rd_Proj(dbgi, new_call, mode_M, n_ia32_Call_mem);
        }
        /* transform call modes */
        if (mode_is_data(mode)) {
                const arch_register_class_t *cls = arch_get_irn_reg_class(node);
                mode = cls->mode;
        }

        /* Map from be_Call to ia32_Call proj number */
        if (proj == pn_be_Call_sp) {
                proj = pn_ia32_Call_stack;
        } else if (proj == pn_be_Call_M) {
                proj = pn_ia32_Call_M;
        } else if (proj == pn_be_Call_X_except) {
                proj = pn_ia32_Call_X_except;
        } else if (proj == pn_be_Call_X_regular) {
                proj = pn_ia32_Call_X_regular;
        } else {
                arch_register_req_t const *const req    = arch_get_irn_register_req(node);
                int                        const n_outs = arch_get_irn_n_outs(new_call);
                int                              i;

                assert(proj      >= pn_be_Call_first_res);
                assert(req->type & arch_register_req_type_limited);

                for (i = 0; i < n_outs; ++i) {
                        arch_register_req_t const *const new_req
                                = arch_get_irn_register_req_out(new_call, i);

                        if (!(new_req->type & arch_register_req_type_limited) ||
                            new_req->cls      != req->cls                     ||
                            *new_req->limited != *req->limited)
                                continue;

                        proj = i;
                        break;
                }
                assert(i < n_outs);
        }

        res = new_rd_Proj(dbgi, new_call, mode, proj);

        /* TODO arch_set_irn_register() only operates on Projs, need variant with index */
        switch (proj) {
        case pn_ia32_Call_stack:
                arch_set_irn_register(res, &ia32_registers[REG_ESP]);
                break;

        case pn_ia32_Call_fpcw:
                arch_set_irn_register(res, &ia32_registers[REG_FPCW]);
                break;
        }

        return res;
}

static ir_node *gen_Proj_ASM(ir_node *node)
{
        ir_mode *mode     = get_irn_mode(node);
        ir_node *pred     = get_Proj_pred(node);
        ir_node *new_pred = be_transform_node(pred);
        long     pos      = get_Proj_proj(node);

        if (mode == mode_M) {
                pos = arch_get_irn_n_outs(new_pred)-1;
        } else if (mode_is_int(mode) || mode_is_reference(mode)) {
                mode = mode_Iu;
        } else if (mode_is_float(mode)) {
                mode = ia32_mode_E;
        } else {
                panic("unexpected proj mode at ASM");
        }

        return new_r_Proj(new_pred, mode, pos);
}

/**
 * Transform and potentially renumber Proj nodes.
 */
static ir_node *gen_Proj(ir_node *node)
{
        ir_node *pred = get_Proj_pred(node);
        long    proj;

        switch (get_irn_opcode(pred)) {
        case iro_Load:
                return gen_Proj_Load(node);
        case iro_Store:
                return gen_Proj_Store(node);
        case iro_ASM:
                return gen_Proj_ASM(node);
        case iro_Builtin:
                return gen_Proj_Builtin(node);
        case iro_Div:
                return gen_Proj_Div(node);
        case iro_Mod:
                return gen_Proj_Mod(node);
        case iro_CopyB:
                return gen_Proj_CopyB(node);
        case beo_SubSP:
                return gen_Proj_be_SubSP(node);
        case beo_AddSP:
                return gen_Proj_be_AddSP(node);
        case beo_Call:
                return gen_Proj_be_Call(node);
        case iro_Start:
                proj = get_Proj_proj(node);
                switch (proj) {
                        case pn_Start_X_initial_exec: {
                                ir_node  *block     = get_nodes_block(pred);
                                ir_node  *new_block = be_transform_node(block);
                                dbg_info *dbgi      = get_irn_dbg_info(node);
                                /* we exchange the ProjX with a jump */
                                ir_node  *jump      = new_rd_Jmp(dbgi, new_block);

                                return jump;
                        }
                }
                break;

        default:
                if (is_ia32_l_FloattoLL(pred)) {
                        return gen_Proj_l_FloattoLL(node);
#ifdef FIRM_EXT_GRS
                } else if (!is_ia32_irn(pred)) { // Quick hack for SIMD optimization
#else
                } else {
#endif
                        ir_mode *mode = get_irn_mode(node);
                        if (ia32_mode_needs_gp_reg(mode)) {
                                ir_node *new_pred = be_transform_node(pred);
                                ir_node *new_proj = new_r_Proj(new_pred, mode_Iu,
                                                               get_Proj_proj(node));
                                new_proj->node_nr = node->node_nr;
                                return new_proj;
                        }
                }
        }
        return be_duplicate_node(node);
}

/**
 * Enters all transform functions into the generic pointer
 */
static void register_transformers(void)
{
        /* first clear the generic function pointer for all ops */
        be_start_transform_setup();

        be_set_transform_function(op_Add,              gen_Add);
        be_set_transform_function(op_And,              gen_And);
        be_set_transform_function(op_ASM,              ia32_gen_ASM);
        be_set_transform_function(op_be_AddSP,         gen_be_AddSP);
        be_set_transform_function(op_be_Call,          gen_be_Call);
        be_set_transform_function(op_be_Copy,          gen_be_Copy);
        be_set_transform_function(op_be_FrameAddr,     gen_be_FrameAddr);
        be_set_transform_function(op_be_IncSP,         gen_be_IncSP);
        be_set_transform_function(op_be_Return,        gen_be_Return);
        be_set_transform_function(op_be_SubSP,         gen_be_SubSP);
        be_set_transform_function(op_Builtin,          gen_Builtin);
        be_set_transform_function(op_Cmp,              gen_Cmp);
        be_set_transform_function(op_Cond,             gen_Cond);
        be_set_transform_function(op_Const,            gen_Const);
        be_set_transform_function(op_Conv,             gen_Conv);
        be_set_transform_function(op_CopyB,            ia32_gen_CopyB);
        be_set_transform_function(op_Div,              gen_Div);
        be_set_transform_function(op_Eor,              gen_Eor);
        be_set_transform_function(op_ia32_l_Adc,       gen_ia32_l_Adc);
        be_set_transform_function(op_ia32_l_Add,       gen_ia32_l_Add);
        be_set_transform_function(op_ia32_Leave,       be_duplicate_node);
        be_set_transform_function(op_ia32_l_FloattoLL, gen_ia32_l_FloattoLL);
        be_set_transform_function(op_ia32_l_IMul,      gen_ia32_l_IMul);
        be_set_transform_function(op_ia32_l_LLtoFloat, gen_ia32_l_LLtoFloat);
        be_set_transform_function(op_ia32_l_Mul,       gen_ia32_l_Mul);
        be_set_transform_function(op_ia32_l_Sbb,       gen_ia32_l_Sbb);
        be_set_transform_function(op_ia32_l_Sub,       gen_ia32_l_Sub);
        be_set_transform_function(op_ia32_GetEIP,      be_duplicate_node);
        be_set_transform_function(op_ia32_Minus64Bit,  be_duplicate_node);
        be_set_transform_function(op_ia32_NoReg_GP,    be_duplicate_node);
        be_set_transform_function(op_ia32_NoReg_VFP,   be_duplicate_node);
        be_set_transform_function(op_ia32_NoReg_XMM,   be_duplicate_node);
        be_set_transform_function(op_ia32_PopEbp,      be_duplicate_node);
        be_set_transform_function(op_ia32_Push,        be_duplicate_node);
        be_set_transform_function(op_IJmp,             gen_IJmp);
        be_set_transform_function(op_Jmp,              gen_Jmp);
        be_set_transform_function(op_Load,             gen_Load);
        be_set_transform_function(op_Minus,            gen_Minus);
        be_set_transform_function(op_Mod,              gen_Mod);
        be_set_transform_function(op_Mul,              gen_Mul);
        be_set_transform_function(op_Mulh,             gen_Mulh);
        be_set_transform_function(op_Mux,              gen_Mux);
        be_set_transform_function(op_Not,              gen_Not);
        be_set_transform_function(op_Or,               gen_Or);
        be_set_transform_function(op_Phi,              gen_Phi);
        be_set_transform_function(op_Proj,             gen_Proj);
        be_set_transform_function(op_Rotl,             gen_Rotl);
        be_set_transform_function(op_Shl,              gen_Shl);
        be_set_transform_function(op_Shr,              gen_Shr);
        be_set_transform_function(op_Shrs,             gen_Shrs);
        be_set_transform_function(op_Store,            gen_Store);
        be_set_transform_function(op_Sub,              gen_Sub);
        be_set_transform_function(op_Switch,           gen_Switch);
        be_set_transform_function(op_SymConst,         gen_SymConst);
        be_set_transform_function(op_Unknown,          ia32_gen_Unknown);

        be_set_upper_bits_clean_function(op_Mux, ia32_mux_upper_bits_clean);
}

/**
 * Pre-transform all unknown and noreg nodes.
 */
static void ia32_pretransform_node(void)
{
        ir_graph        *irg      = current_ir_graph;
        ia32_irg_data_t *irg_data = ia32_get_irg_data(current_ir_graph);

        irg_data->noreg_gp       = be_pre_transform_node(irg_data->noreg_gp);
        irg_data->noreg_vfp      = be_pre_transform_node(irg_data->noreg_vfp);
        irg_data->noreg_xmm      = be_pre_transform_node(irg_data->noreg_xmm);
        irg_data->get_eip        = be_pre_transform_node(irg_data->get_eip);
        irg_data->fpu_trunc_mode = be_pre_transform_node(irg_data->fpu_trunc_mode);

        nomem    = get_irg_no_mem(irg);
        noreg_GP = ia32_new_NoReg_gp(irg);
}

/**
 * Post-process all calls if we are in SSE mode.
 * The ABI requires that the results are in st0, copy them
 * to a xmm register.
 */
static void postprocess_fp_call_results(void)
{
        size_t i, n;

        for (i = 0, n = ARR_LEN(call_list); i < n; ++i) {
                ir_node *call = call_list[i];
                ir_type *mtp  = call_types[i];
                int     j;

                for (j = get_method_n_ress(mtp) - 1; j >= 0; --j) {
                        ir_type *res_tp = get_method_res_type(mtp, j);
                        ir_node *res, *new_res;
                        ir_mode *res_mode;

                        if (! is_atomic_type(res_tp)) {
                                /* no floating point return */
                                continue;
                        }
                        res_mode = get_type_mode(res_tp);
                        if (! mode_is_float(res_mode)) {
                                /* no floating point return */
                                continue;
                        }

                        res     = be_get_Proj_for_pn(call, pn_ia32_Call_vf0 + j);
                        new_res = NULL;

                        /* now patch the users */
                        foreach_out_edge_safe(res, edge) {
                                ir_node *succ = get_edge_src_irn(edge);

                                /* ignore Keeps */
                                if (be_is_Keep(succ))
                                        continue;

                                if (is_ia32_xStore(succ)) {
                                        /* an xStore can be patched into an vfst */
                                        dbg_info *db    = get_irn_dbg_info(succ);
                                        ir_node  *block = get_nodes_block(succ);
                                        ir_node  *base  = get_irn_n(succ, n_ia32_xStore_base);
                                        ir_node  *idx   = get_irn_n(succ, n_ia32_xStore_index);
                                        ir_node  *mem   = get_irn_n(succ, n_ia32_xStore_mem);
                                        ir_node  *value = get_irn_n(succ, n_ia32_xStore_val);
                                        ir_mode  *mode  = get_ia32_ls_mode(succ);

                                        ir_node  *st = new_bd_ia32_vfst(db, block, base, idx, mem, value, mode);
                                        //ir_node  *mem = new_r_Proj(st, mode_M, pn_ia32_vfst_M);
                                        set_ia32_am_offs_int(st, get_ia32_am_offs_int(succ));
                                        if (is_ia32_use_frame(succ))
                                                set_ia32_use_frame(st);
                                        set_ia32_frame_ent(st, get_ia32_frame_ent(succ));
                                        set_irn_pinned(st, get_irn_pinned(succ));
                                        set_ia32_op_type(st, ia32_AddrModeD);

                                        assert((long)pn_ia32_xStore_M == (long)pn_ia32_vfst_M);
                                        assert((long)pn_ia32_xStore_X_regular == (long)pn_ia32_vfst_X_regular);
                                        assert((long)pn_ia32_xStore_X_except == (long)pn_ia32_vfst_X_except);

                                        exchange(succ, st);

                                        continue;
                                }

                                if (new_res == NULL) {
                                        dbg_info *db       = get_irn_dbg_info(call);
                                        ir_node  *block    = get_nodes_block(call);
                                        ir_node  *frame    = get_irg_frame(current_ir_graph);
                                        ir_node  *old_mem  = be_get_Proj_for_pn(call, pn_ia32_Call_M);
                                        ir_node  *call_mem = new_r_Proj(call, mode_M, pn_ia32_Call_M);
                                        ir_node  *vfst, *xld, *new_mem;
                                        ir_node  *vfst_mem;

                                        /* store st(0) on stack */
                                        vfst = new_bd_ia32_vfst(db, block, frame, noreg_GP, call_mem,
                                                                res, res_mode);
                                        set_ia32_op_type(vfst, ia32_AddrModeD);
                                        set_ia32_use_frame(vfst);

                                        vfst_mem = new_r_Proj(vfst, mode_M, pn_ia32_vfst_M);

                                        /* load into SSE register */
                                        xld = new_bd_ia32_xLoad(db, block, frame, noreg_GP, vfst_mem,
                                                                res_mode);
                                        set_ia32_op_type(xld, ia32_AddrModeS);
                                        set_ia32_use_frame(xld);

                                        new_res = new_r_Proj(xld, res_mode, pn_ia32_xLoad_res);
                                        new_mem = new_r_Proj(xld, mode_M, pn_ia32_xLoad_M);

                                        if (old_mem != NULL) {
                                                edges_reroute(old_mem, new_mem);
                                                kill_node(old_mem);
                                        }
                                }
                                set_irn_n(succ, get_edge_src_pos(edge), new_res);
                        }
                }
        }
}

/* do the transformation */
void ia32_transform_graph(ir_graph *irg)
{
        int cse_last;

        register_transformers();
        initial_fpcw       = NULL;
        ia32_no_pic_adjust = 0;

        old_initial_fpcw = be_get_initial_reg_value(irg, &ia32_registers[REG_FPCW]);

        be_timer_push(T_HEIGHTS);
        ia32_heights = heights_new(irg);
        be_timer_pop(T_HEIGHTS);
        ia32_calculate_non_address_mode_nodes(irg);

        /* the transform phase is not safe for CSE (yet) because several nodes get
         * attributes set after their creation */
        cse_last = get_opt_cse();
        set_opt_cse(0);

        call_list  = NEW_ARR_F(ir_node *, 0);
        call_types = NEW_ARR_F(ir_type *, 0);
        be_transform_graph(irg, ia32_pretransform_node);

        if (ia32_cg_config.use_sse2)
                postprocess_fp_call_results();
        DEL_ARR_F(call_types);
        DEL_ARR_F(call_list);

        set_opt_cse(cse_last);

        ia32_free_non_address_mode_nodes();
        heights_free(ia32_heights);
        ia32_heights = NULL;
}

void ia32_init_transform(void)
{
        FIRM_DBG_REGISTER(dbg, "firm.be.ia32.transform");
}