In bemit_mod_am() we must check for the ebp special case, if there is an index, too.
[libfirm] / ir / be / ia32 / ia32_emitter.c
index 9ed4d55..3a825fc 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 1995-2007 University of Karlsruhe.  All right reserved.
+ * Copyright (C) 1995-2008 University of Karlsruhe.  All right reserved.
  *
  * This file is part of libFirm.
  *
@@ -23,9 +23,7 @@
  * @author      Christian Wuerdig, Matthias Braun
  * @version     $Id$
  */
-#ifdef HAVE_CONFIG_H
 #include "config.h"
-#endif
 
 #include <limits.h>
 
 #include "irargs_t.h"
 #include "irprog_t.h"
 #include "iredges_t.h"
+#include "irtools.h"
 #include "execfreq.h"
 #include "error.h"
 #include "raw_bitset.h"
+#include "dbginfo.h"
+#include "lc_opts.h"
 
-#include "../besched_t.h"
-#include "../benode_t.h"
+#include "../besched.h"
+#include "../benode.h"
 #include "../beabi.h"
 #include "../be_dbgout.h"
 #include "../beemitter.h"
 #include "../begnuas.h"
-#include "../beirg_t.h"
+#include "../beirg.h"
+#include "../be_dbgout.h"
 
 #include "ia32_emitter.h"
 #include "gen_ia32_emitter.h"
@@ -57,6 +59,7 @@
 #include "ia32_nodes_attr.h"
 #include "ia32_new_nodes.h"
 #include "ia32_map_regs.h"
+#include "ia32_architecture.h"
 #include "bearch_ia32_t.h"
 
 DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
@@ -65,16 +68,67 @@ DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
 
 #define SNPRINTF_BUF_LEN 128
 
+static const ia32_isa_t *isa;
+static ia32_code_gen_t  *cg;
+static char              pic_base_label[128];
+static ir_label_t        exc_label_id;
+static int               mark_spill_reload = 0;
+static int               do_pic;
+
+/** Return the next block in Block schedule */
+static ir_node *get_prev_block_sched(const ir_node *block)
+{
+       return get_irn_link(block);
+}
+
+/** Checks if the current block is a fall-through target. */
+static int is_fallthrough(const ir_node *cfgpred)
+{
+       ir_node *pred;
+
+       if (!is_Proj(cfgpred))
+               return 1;
+       pred = get_Proj_pred(cfgpred);
+       if (is_ia32_SwitchJmp(pred))
+               return 0;
+
+       return 1;
+}
+
+/**
+ * returns non-zero if the given block needs a label
+ * because of being a jump-target (and not a fall-through)
+ */
+static int block_needs_label(const ir_node *block)
+{
+       int need_label = 1;
+       int  n_cfgpreds = get_Block_n_cfgpreds(block);
+
+       if (has_Block_entity(block))
+               return 1;
+
+       if (n_cfgpreds == 0) {
+               need_label = 0;
+       } else if (n_cfgpreds == 1) {
+               ir_node *cfgpred       = get_Block_cfgpred(block, 0);
+               ir_node *cfgpred_block = get_nodes_block(cfgpred);
+
+               if (get_prev_block_sched(block) == cfgpred_block
+                               && is_fallthrough(cfgpred)) {
+                       need_label = 0;
+               }
+       }
+
+       return need_label;
+}
+
 /**
  * Returns the register at in position pos.
  */
-static
-const arch_register_t *get_in_reg(ia32_emit_env_t *env, const ir_node *irn,
-                                  int pos)
+static const arch_register_t *get_in_reg(const ir_node *irn, int pos)
 {
-       const arch_env_t       *arch_env = env->arch_env;
-       ir_node                *op;
-       const arch_register_t  *reg = NULL;
+       ir_node               *op;
+       const arch_register_t *reg = NULL;
 
        assert(get_irn_arity(irn) > pos && "Invalid IN position");
 
@@ -82,19 +136,16 @@ const arch_register_t *get_in_reg(ia32_emit_env_t *env, const ir_node *irn,
           in register we need. */
        op = get_irn_n(irn, pos);
 
-       reg = arch_get_irn_register(arch_env, op);
+       reg = arch_get_irn_register(op);
 
        assert(reg && "no in register found");
 
-       if(reg == &ia32_gp_regs[REG_GP_NOREG])
-               panic("trying to emit noreg");
+       if (reg == &ia32_gp_regs[REG_GP_NOREG])
+               panic("trying to emit noreg for %+F input %d", irn, pos);
 
        /* in case of unknown register: just return a valid register */
        if (reg == &ia32_gp_regs[REG_GP_UKNWN]) {
-               const arch_register_req_t *req;
-
-               /* ask for the requirements */
-               req = arch_get_register_req(arch_env, irn, pos);
+               const arch_register_req_t *req = arch_get_register_req(irn, pos);
 
                if (arch_register_req_is(req, limited)) {
                        /* in case of limited requirements: get the first allowed register */
@@ -112,11 +163,8 @@ const arch_register_t *get_in_reg(ia32_emit_env_t *env, const ir_node *irn,
 /**
  * Returns the register at out position pos.
  */
-static
-const arch_register_t *get_out_reg(ia32_emit_env_t *env, const ir_node *irn,
-                                   int pos)
+static const arch_register_t *get_out_reg(const ir_node *irn, int pos)
 {
-       const arch_env_t      *arch_env = env->arch_env;
        ir_node               *proj;
        const arch_register_t *reg = NULL;
 
@@ -126,9 +174,10 @@ const arch_register_t *get_out_reg(ia32_emit_env_t *env, const ir_node *irn,
        /*           Proj with the corresponding projnum for the register */
 
        if (get_irn_mode(irn) != mode_T) {
-               reg = arch_get_irn_register(arch_env, irn);
+               assert(pos == 0);
+               reg = arch_get_irn_register(irn);
        } else if (is_ia32_irn(irn)) {
-               reg = get_ia32_out_reg(irn, pos);
+               reg = arch_irn_get_register(irn, pos);
        } else {
                const ir_edge_t *edge;
 
@@ -136,7 +185,7 @@ const arch_register_t *get_out_reg(ia32_emit_env_t *env, const ir_node *irn,
                        proj = get_edge_src_irn(edge);
                        assert(is_Proj(proj) && "non-Proj from mode_T node");
                        if (get_Proj_proj(proj) == pos) {
-                               reg = arch_get_irn_register(arch_env, proj);
+                               reg = arch_get_irn_register(proj);
                                break;
                        }
                }
@@ -146,76 +195,11 @@ const arch_register_t *get_out_reg(ia32_emit_env_t *env, const ir_node *irn,
        return reg;
 }
 
-/**
- * Determine the gnu assembler suffix that indicates a mode
- */
-static
-char get_mode_suffix(const ir_mode *mode) {
-       if(mode_is_float(mode)) {
-               switch(get_mode_size_bits(mode)) {
-               case 32:
-                       return 's';
-               case 64:
-                       return 'l';
-               case 80:
-               case 96:
-                       return 't';
-               }
-       } else {
-               assert(mode_is_int(mode) || mode_is_reference(mode));
-               switch(get_mode_size_bits(mode)) {
-               case 64:
-                       return 'q';
-               case 32:
-                       return 'l';
-               case 16:
-                       return 'w';
-               case 8:
-                       return 'b';
-               }
-       }
-       panic("Can't output mode_suffix for %+F\n", mode);
-}
-
-static
-int produces_result(const ir_node *node) {
-       return
-               !is_ia32_CmpSet(node)    &&
-               !is_ia32_CmpSet8Bit(node)    &&
-               !is_ia32_CmpJmp(node)   &&
-               !is_ia32_CmpJmp8Bit(node)   &&
-               !is_ia32_St(node)        &&
-               !is_ia32_SwitchJmp(node) &&
-               !is_ia32_TestJmp(node)   &&
-               !is_ia32_TestJmp8Bit(node)   &&
-               !is_ia32_xCmpSet(node)   &&
-               !is_ia32_xCmpJmp(node)  &&
-               !is_ia32_CmpCMov(node)   &&
-               !is_ia32_CmpCMov8Bit(node)   &&
-               !is_ia32_TestCMov(node)  &&
-               !is_ia32_TestCMov8Bit(node)  &&
-               !is_ia32_CmpSet(node)    && /* this is correct, the Cmp has no result */
-               !is_ia32_TestSet(node);
-}
-
-static
-const char *ia32_get_reg_name_for_mode(ia32_emit_env_t *env, ir_mode *mode,
-                                       const arch_register_t *reg) {
-       switch(get_mode_size_bits(mode)) {
-               case 8:
-                       return ia32_get_mapped_reg_name(env->isa->regs_8bit, reg);
-               case 16:
-                       return ia32_get_mapped_reg_name(env->isa->regs_16bit, reg);
-               default:
-                       return (char *)arch_register_get_name(reg);
-       }
-}
-
 /**
  * Add a number to a prefix. This number will not be used a second time.
  */
-static
-char *get_unique_label(char *buf, size_t buflen, const char *prefix) {
+static char *get_unique_label(char *buf, size_t buflen, const char *prefix)
+{
        static unsigned long id = 0;
        snprintf(buf, buflen, "%s%lu", prefix, ++id);
        return buf;
@@ -232,342 +216,371 @@ char *get_unique_label(char *buf, size_t buflen, const char *prefix) {
  * |_|                                       |_|
  *************************************************************/
 
-// we have no C++ and can't define an implicit ia32_emit_env_t* cast to
-// be_emit_env_t* so we cheat a bit...
-#define be_emit_char(env,c)             be_emit_char(env->emit,c)
-#define be_emit_string(env,s)           be_emit_string(env->emit,s)
-#undef be_emit_cstring
-#define be_emit_cstring(env,x)          { be_emit_string_len(env->emit, x, sizeof(x)-1); }
-#define be_emit_ident(env,i)            be_emit_ident(env->emit,i)
-#define be_emit_tarval(env,tv)          be_emit_tarval(env->emit,tv)
-#define be_emit_write_line(env)         be_emit_write_line(env->emit)
-#define be_emit_finish_line_gas(env,n)  be_emit_finish_line_gas(env->emit,n)
-#define be_emit_pad_comment(env)        be_emit_pad_comment(env->emit)
-
-void ia32_emit_source_register(ia32_emit_env_t *env, const ir_node *node, int pos)
-{
-       const arch_register_t *reg = get_in_reg(env, node, pos);
+/**
+ * Emit the name of the 8bit low register
+ */
+static void emit_8bit_register(const arch_register_t *reg)
+{
        const char *reg_name = arch_register_get_name(reg);
 
-       assert(pos < get_irn_arity(node));
-
-       be_emit_char(env, '%');
-       be_emit_string(env, reg_name);
+       be_emit_char('%');
+       be_emit_char(reg_name[1]);
+       be_emit_char('l');
 }
 
-void ia32_emit_dest_register(ia32_emit_env_t *env, const ir_node *node, int pos) {
-       const arch_register_t *reg = get_out_reg(env, node, pos);
+/**
+ * Emit the name of the 8bit high register
+ */
+static void emit_8bit_register_high(const arch_register_t *reg)
+{
        const char *reg_name = arch_register_get_name(reg);
 
-       be_emit_char(env, '%');
-       be_emit_string(env, reg_name);
+       be_emit_char('%');
+       be_emit_char(reg_name[1]);
+       be_emit_char('h');
 }
 
-static void ia32_emit_register(ia32_emit_env_t *env, const arch_register_t *reg)
+static void emit_16bit_register(const arch_register_t *reg)
 {
-       const char *reg_name = arch_register_get_name(reg);
+       const char *reg_name = ia32_get_mapped_reg_name(isa->regs_16bit, reg);
 
-       be_emit_char(env, '%');
-       be_emit_string(env, reg_name);
+       be_emit_char('%');
+       be_emit_string(reg_name);
 }
 
-void ia32_emit_x87_name(ia32_emit_env_t *env, const ir_node *node, int pos)
+/**
+ * emit a register, possible shortened by a mode
+ *
+ * @param reg   the register
+ * @param mode  the mode of the register or NULL for full register
+ */
+static void emit_register(const arch_register_t *reg, const ir_mode *mode)
 {
-       const ia32_x87_attr_t *attr = get_ia32_x87_attr_const(node);
+       const char *reg_name;
 
-       assert(pos < 3);
-       be_emit_char(env, '%');
-       be_emit_string(env, attr->x87[pos]->name);
+       if (mode != NULL) {
+               int size = get_mode_size_bits(mode);
+               switch (size) {
+                       case  8: emit_8bit_register(reg);  return;
+                       case 16: emit_16bit_register(reg); return;
+               }
+               assert(mode_is_float(mode) || size == 32);
+       }
+
+       reg_name = arch_register_get_name(reg);
+
+       be_emit_char('%');
+       be_emit_string(reg_name);
 }
 
-static
-void ia32_emit_mode_suffix_mode(ia32_emit_env_t *env, const ir_mode *mode)
+void ia32_emit_source_register(const ir_node *node, int pos)
 {
-       be_emit_char(env, get_mode_suffix(mode));
+       const arch_register_t *reg = get_in_reg(node, pos);
+
+       emit_register(reg, NULL);
 }
 
-void ia32_emit_mode_suffix(ia32_emit_env_t *env, const ir_node *node)
+static void ia32_emit_entity(ir_entity *entity, int no_pic_adjust)
 {
-       ir_mode *mode = get_ia32_ls_mode(node);
-       if(mode == NULL)
-               mode = mode_Iu;
+       set_entity_backend_marked(entity, 1);
+       be_gas_emit_entity(entity);
+
+       if (get_entity_owner(entity) == get_tls_type()) {
+               if (get_entity_visibility(entity) == visibility_external_allocated) {
+                       be_emit_cstring("@INDNTPOFF");
+               } else {
+                       be_emit_cstring("@NTPOFF");
+               }
+       }
 
-       ia32_emit_mode_suffix_mode(env, mode);
+       if (do_pic && !no_pic_adjust) {
+               be_emit_char('-');
+               be_emit_string(pic_base_label);
+       }
 }
 
-void ia32_emit_x87_mode_suffix(ia32_emit_env_t *env, const ir_node *node)
+static void emit_ia32_Immediate_no_prefix(const ir_node *node)
 {
-       ir_mode *mode = get_ia32_ls_mode(node);
-       if(mode != NULL)
-               ia32_emit_mode_suffix_mode(env, mode);
+       const ia32_immediate_attr_t *attr = get_ia32_immediate_attr_const(node);
+
+       if (attr->symconst != NULL) {
+               if (attr->sc_sign)
+                       be_emit_char('-');
+               ia32_emit_entity(attr->symconst, attr->no_pic_adjust);
+       }
+       if (attr->symconst == NULL || attr->offset != 0) {
+               if (attr->symconst != NULL) {
+                       be_emit_irprintf("%+d", attr->offset);
+               } else {
+                       be_emit_irprintf("0x%X", attr->offset);
+               }
+       }
 }
 
-static
-char get_xmm_mode_suffix(ir_mode *mode)
+static void emit_ia32_Immediate(const ir_node *node)
 {
-       assert(mode_is_float(mode));
-       switch(get_mode_size_bits(mode)) {
-       case 32:
-               return 's';
-       case 64:
-               return 'd';
-       default:
-               assert(0);
-       }
-       return '%';
+       be_emit_char('$');
+       emit_ia32_Immediate_no_prefix(node);
 }
 
-void ia32_emit_xmm_mode_suffix(ia32_emit_env_t *env, const ir_node *node)
+void ia32_emit_8bit_source_register_or_immediate(const ir_node *node, int pos)
 {
-       ir_mode *mode = get_ia32_ls_mode(node);
-       assert(mode != NULL);
-       be_emit_char(env, 's');
-       be_emit_char(env, get_xmm_mode_suffix(mode));
+       const arch_register_t *reg;
+       const ir_node         *in = get_irn_n(node, pos);
+       if (is_ia32_Immediate(in)) {
+               emit_ia32_Immediate(in);
+               return;
+       }
+
+       reg = get_in_reg(node, pos);
+       emit_8bit_register(reg);
 }
 
-void ia32_emit_xmm_mode_suffix_s(ia32_emit_env_t *env, const ir_node *node)
+void ia32_emit_8bit_high_source_register(const ir_node *node, int pos)
 {
-       ir_mode *mode = get_ia32_ls_mode(node);
-       assert(mode != NULL);
-       be_emit_char(env, get_xmm_mode_suffix(mode));
+       const arch_register_t *reg = get_in_reg(node, pos);
+       emit_8bit_register_high(reg);
 }
 
-void ia32_emit_extend_suffix(ia32_emit_env_t *env, const ir_mode *mode)
+void ia32_emit_16bit_source_register_or_immediate(const ir_node *node, int pos)
 {
-       if(get_mode_size_bits(mode) == 32)
+       const arch_register_t *reg;
+       const ir_node         *in = get_irn_n(node, pos);
+       if (is_ia32_Immediate(in)) {
+               emit_ia32_Immediate(in);
                return;
-       if(mode_is_signed(mode)) {
-               be_emit_char(env, 's');
-       } else {
-               be_emit_char(env, 'z');
        }
+
+       reg = get_in_reg(node, pos);
+       emit_16bit_register(reg);
 }
 
-static
-void ia32_emit_function_object(ia32_emit_env_t *env, const char *name)
+void ia32_emit_dest_register(const ir_node *node, int pos)
 {
-       switch (be_gas_flavour) {
-       case GAS_FLAVOUR_NORMAL:
-               be_emit_cstring(env, "\t.type\t");
-               be_emit_string(env, name);
-               be_emit_cstring(env, ", @function\n");
-               be_emit_write_line(env);
-               break;
-       case GAS_FLAVOUR_MINGW:
-               be_emit_cstring(env, "\t.def\t");
-               be_emit_string(env, name);
-               be_emit_cstring(env, ";\t.scl\t2;\t.type\t32;\t.endef\n");
-               be_emit_write_line(env);
-               break;
-       default:
-               break;
-       }
+       const arch_register_t *reg  = get_out_reg(node, pos);
+
+       emit_register(reg, NULL);
 }
 
-static
-void ia32_emit_function_size(ia32_emit_env_t *env, const char *name)
+void ia32_emit_dest_register_size(const ir_node *node, int pos)
 {
-       switch (be_gas_flavour) {
-       case GAS_FLAVOUR_NORMAL:
-               be_emit_cstring(env, "\t.size\t");
-               be_emit_string(env, name);
-               be_emit_cstring(env, ", .-");
-               be_emit_string(env, name);
-               be_emit_char(env, '\n');
-               be_emit_write_line(env);
-               break;
-       default:
-               break;
-       }
+       const arch_register_t *reg  = get_out_reg(node, pos);
+
+       emit_register(reg, get_ia32_ls_mode(node));
 }
 
+void ia32_emit_8bit_dest_register(const ir_node *node, int pos)
+{
+       const arch_register_t *reg  = get_out_reg(node, pos);
 
-static
-void emit_ia32_Immediate(ia32_emit_env_t *env, const ir_node *node);
+       emit_register(reg, mode_Bu);
+}
 
-void ia32_emit_8bit_source_register(ia32_emit_env_t *env, const ir_node *node,
-                                    int pos)
+void ia32_emit_x87_register(const ir_node *node, int pos)
 {
-       const arch_register_t *reg;
-       const char            *reg_name;
-       ir_node               *in;
+       const ia32_x87_attr_t *attr = get_ia32_x87_attr_const(node);
 
-       in = get_irn_n(node, pos);
-       if(is_ia32_Immediate(in)) {
-               emit_ia32_Immediate(env, in);
-               return;
+       assert(pos < 3);
+       be_emit_char('%');
+       be_emit_string(attr->x87[pos]->name);
+}
+
+static void ia32_emit_mode_suffix_mode(const ir_mode *mode)
+{
+       assert(mode_is_int(mode) || mode_is_reference(mode));
+       switch (get_mode_size_bits(mode)) {
+               case 8:  be_emit_char('b');     return;
+               case 16: be_emit_char('w');     return;
+               case 32: be_emit_char('l');     return;
+               /* gas docu says q is the suffix but gcc, objdump and icc use ll
+                * apparently */
+               case 64: be_emit_cstring("ll"); return;
        }
+       panic("Can't output mode_suffix for %+F", mode);
+}
 
-       reg      = get_in_reg(env, node, pos);
-       reg_name = arch_register_get_name(reg);
+void ia32_emit_mode_suffix(const ir_node *node)
+{
+       ir_mode *mode = get_ia32_ls_mode(node);
+       if (mode == NULL)
+               mode = mode_Iu;
 
-       be_emit_char(env, '%');
-       be_emit_char(env, reg_name[1]);
-       be_emit_char(env, 'l');
+       ia32_emit_mode_suffix_mode(mode);
 }
 
-void ia32_emit_16bit_source_register(ia32_emit_env_t *env, const ir_node *node,
-                                     int pos)
+void ia32_emit_x87_mode_suffix(const ir_node *node)
 {
-       const arch_register_t *reg;
-       const char            *reg_name;
-       ir_node               *in;
+       ir_mode *mode;
 
-       in = get_irn_n(node, pos);
-       if(is_ia32_Immediate(in)) {
-               emit_ia32_Immediate(env, in);
+       /* we only need to emit the mode on address mode */
+       if (get_ia32_op_type(node) == ia32_Normal)
                return;
+
+       mode = get_ia32_ls_mode(node);
+       assert(mode != NULL);
+
+       if (mode_is_float(mode)) {
+               switch (get_mode_size_bits(mode)) {
+                       case 32: be_emit_char('s'); return;
+                       case 64: be_emit_char('l'); return;
+                       case 80:
+                       case 96: be_emit_char('t'); return;
+               }
+       } else {
+               assert(mode_is_int(mode));
+               switch (get_mode_size_bits(mode)) {
+                       case 16: be_emit_char('s');     return;
+                       case 32: be_emit_char('l');     return;
+                       /* gas docu says q is the suffix but gcc, objdump and icc use ll
+                        * apparently */
+                       case 64: be_emit_cstring("ll"); return;
+               }
        }
+       panic("Can't output mode_suffix for %+F", mode);
+}
 
-       reg      = get_in_reg(env, node, pos);
-       reg_name = arch_register_get_name(reg);
+static char get_xmm_mode_suffix(ir_mode *mode)
+{
+       assert(mode_is_float(mode));
+       switch(get_mode_size_bits(mode)) {
+       case 32: return 's';
+       case 64: return 'd';
+       default: panic("Invalid XMM mode");
+       }
+}
+
+void ia32_emit_xmm_mode_suffix(const ir_node *node)
+{
+       ir_mode *mode = get_ia32_ls_mode(node);
+       assert(mode != NULL);
+       be_emit_char('s');
+       be_emit_char(get_xmm_mode_suffix(mode));
+}
+
+void ia32_emit_xmm_mode_suffix_s(const ir_node *node)
+{
+       ir_mode *mode = get_ia32_ls_mode(node);
+       assert(mode != NULL);
+       be_emit_char(get_xmm_mode_suffix(mode));
+}
 
-       be_emit_char(env, '%');
-       be_emit_string(env, &reg_name[1]);
-       be_emit_char(env, 'x');
+void ia32_emit_extend_suffix(const ir_node *node)
+{
+       ir_mode *mode = get_ia32_ls_mode(node);
+       if (get_mode_size_bits(mode) == 32)
+               return;
+       be_emit_char(mode_is_signed(mode) ? 's' : 'z');
+       ia32_emit_mode_suffix_mode(mode);
 }
 
-void ia32_emit_source_register_or_immediate(ia32_emit_env_t *env,
-                                            const ir_node *node, int pos)
+void ia32_emit_source_register_or_immediate(const ir_node *node, int pos)
 {
        ir_node *in = get_irn_n(node, pos);
-       if(is_ia32_Immediate(in)) {
-               emit_ia32_Immediate(env, in);
+       if (is_ia32_Immediate(in)) {
+               emit_ia32_Immediate(in);
        } else {
-               ia32_emit_source_register(env, node, pos);
+               const ir_mode         *mode = get_ia32_ls_mode(node);
+               const arch_register_t *reg  = get_in_reg(node, pos);
+               emit_register(reg, mode);
        }
 }
 
 /**
- * Emits registers and/or address mode of a binary operation.
+ * Returns the target block for a control flow node.
  */
-void ia32_emit_binop(ia32_emit_env_t *env, const ir_node *node) {
-       const ir_node *right_op = get_irn_n(node, 3);
-
-       switch(get_ia32_op_type(node)) {
-       case ia32_Normal:
-               if(is_ia32_Immediate(right_op)) {
-                       emit_ia32_Immediate(env, right_op);
-                       be_emit_cstring(env, ", ");
-                       ia32_emit_source_register(env, node, 2);
-                       break;
-               } else {
-                       const arch_register_t *in1 = get_in_reg(env, node, 2);
-                       const arch_register_t *in2 = get_in_reg(env, node, 3);
-                       const arch_register_t *out = produces_result(node) ? get_out_reg(env, node, 0) : NULL;
-                       const arch_register_t *in;
-                       const char            *in_name;
-
-                       in      = out ? ((out == in2) ? in1 : in2) : in2;
-                       out     = out ? out : in1;
-                       in_name = arch_register_get_name(in);
-
-                       be_emit_char(env, '%');
-                       be_emit_string(env, in_name);
-                       be_emit_cstring(env, ", %");
-                       be_emit_string(env, arch_register_get_name(out));
-               }
-               break;
-       case ia32_AddrModeS:
-               if(is_ia32_Immediate(right_op)) {
-                       assert(!produces_result(node) &&
-                                       "Source AM with Const must not produce result");
-
-                       emit_ia32_Immediate(env, right_op);
-                       be_emit_cstring(env, ", ");
-                       ia32_emit_am(env, node);
-               } else if (produces_result(node)) {
-                       ia32_emit_am(env, node);
-                       be_emit_cstring(env, ", ");
-                       ia32_emit_dest_register(env, node, 0);
-               } else {
-                       ia32_emit_am(env, node);
-                       be_emit_cstring(env, ", ");
-                       ia32_emit_source_register(env, node, 2);
-               }
-               break;
-       case ia32_AddrModeD:
-               panic("DestMode can't be output by %%binop anymore");
-               break;
-       default:
-               assert(0 && "unsupported op type");
-       }
+static ir_node *get_cfop_target_block(const ir_node *irn)
+{
+       assert(get_irn_mode(irn) == mode_X);
+       return get_irn_link(irn);
 }
 
 /**
- * Emits registers and/or address mode of a binary operation.
+ * Emits a block label for the given block.
  */
-void ia32_emit_x87_binop(ia32_emit_env_t *env, const ir_node *node) {
-       switch(get_ia32_op_type(node)) {
-               case ia32_Normal:
-                       {
-                               const ia32_x87_attr_t *x87_attr = get_ia32_x87_attr_const(node);
-                               const arch_register_t *in1      = x87_attr->x87[0];
-                               const arch_register_t *in2      = x87_attr->x87[1];
-                               const arch_register_t *out      = x87_attr->x87[2];
-                               const arch_register_t *in;
-
-                               in  = out ? ((out == in2) ? in1 : in2) : in2;
-                               out = out ? out : in1;
-
-                               be_emit_char(env, '%');
-                               be_emit_string(env, arch_register_get_name(in));
-                               be_emit_cstring(env, ", %");
-                               be_emit_string(env, arch_register_get_name(out));
-                       }
-                       break;
-               case ia32_AddrModeS:
-               case ia32_AddrModeD:
-                       ia32_emit_am(env, node);
-                       break;
-               default:
-                       assert(0 && "unsupported op type");
+static void ia32_emit_block_name(const ir_node *block)
+{
+       if (has_Block_entity(block)) {
+               ir_entity *entity = get_Block_entity(block);
+               be_gas_emit_entity(entity);
+       } else {
+               be_emit_cstring(BLOCK_PREFIX);
+               be_emit_irprintf("%ld", get_irn_node_nr(block));
        }
 }
 
-void ia32_emit_am_or_dest_register(ia32_emit_env_t *env, const ir_node *node,
-                                   int pos) {
-       if(get_ia32_op_type(node) == ia32_Normal) {
-               ia32_emit_dest_register(env, node, pos);
-       } else {
-               assert(get_ia32_op_type(node) == ia32_AddrModeD);
-               ia32_emit_am(env, node);
-       }
+/**
+ * Emits the target label for a control flow node.
+ */
+static void ia32_emit_cfop_target(const ir_node *node)
+{
+       ir_node *block = get_cfop_target_block(node);
+       ia32_emit_block_name(block);
 }
 
+/*
+ * positive conditions for signed compares
+ */
+static const char *const cmp2condition_s[] = {
+       NULL, /* always false */
+       "e",  /* == */
+       "l",  /* <  */
+       "le", /* <= */
+       "g",  /* >  */
+       "ge", /* >= */
+       "ne", /* != */
+       NULL  /* always true */
+};
+
+/*
+ * positive conditions for unsigned compares
+ */
+static const char *const cmp2condition_u[] = {
+       NULL, /* always false */
+       "e",  /* == */
+       "b",  /* <  */
+       "be", /* <= */
+       "a",  /* >  */
+       "ae", /* >= */
+       "ne", /* != */
+       NULL  /* always true */
+};
+
 /**
- * Emits registers and/or address mode of a unary operation.
+ * Emit the suffix for a compare instruction.
  */
-void ia32_emit_unop(ia32_emit_env_t *env, const ir_node *node, int pos) {
-       const ir_node *op;
+static void ia32_emit_cmp_suffix(int pnc)
+{
+       const char *str;
 
-       switch(get_ia32_op_type(node)) {
-       case ia32_Normal:
-               op = get_irn_n(node, pos);
-               if (is_ia32_Immediate(op)) {
-                       emit_ia32_Immediate(env, op);
-               } else {
-                       ia32_emit_source_register(env, node, pos);
-               }
-               break;
-       case ia32_AddrModeS:
-       case ia32_AddrModeD:
-               ia32_emit_am(env, node);
-               break;
-       default:
-               assert(0 && "unsupported op type");
+       if (pnc == ia32_pn_Cmp_parity) {
+               be_emit_char('p');
+               return;
        }
+       if (pnc & ia32_pn_Cmp_float || pnc & ia32_pn_Cmp_unsigned) {
+               str = cmp2condition_u[pnc & 7];
+       } else {
+               str = cmp2condition_s[pnc & 7];
+       }
+
+       be_emit_string(str);
 }
 
+typedef enum ia32_emit_mod_t {
+       EMIT_RESPECT_LS   = 1U << 0,
+       EMIT_ALTERNATE_AM = 1U << 1,
+       EMIT_LONG         = 1U << 2
+} ia32_emit_mod_t;
+
 /**
  * Emits address mode.
  */
-void ia32_emit_am(ia32_emit_env_t *env, const ir_node *node) {
+void ia32_emit_am(const ir_node *node)
+{
        ir_entity *ent       = get_ia32_am_sc(node);
        int        offs      = get_ia32_am_offs_int(node);
-       ir_node   *base      = get_irn_n(node, 0);
+       ir_node   *base      = get_irn_n(node, n_ia32_base);
        int        has_base  = !is_ia32_NoReg_GP(base);
-       ir_node   *index     = get_irn_n(node, 1);
+       ir_node   *index     = get_irn_n(node, n_ia32_index);
        int        has_index = !is_ia32_NoReg_GP(index);
 
        /* just to be sure... */
@@ -575,175 +588,442 @@ void ia32_emit_am(ia32_emit_env_t *env, const ir_node *node) {
 
        /* emit offset */
        if (ent != NULL) {
-               ident *id;
-
-               set_entity_backend_marked(ent, 1);
-               id = get_entity_ld_ident(ent);
+               const ia32_attr_t *attr = get_ia32_attr_const(node);
                if (is_ia32_am_sc_sign(node))
-                       be_emit_char(env, '-');
-               be_emit_ident(env, id);
-
-               if(get_entity_owner(ent) == get_tls_type()) {
-                       if (get_entity_visibility(ent) == visibility_external_allocated) {
-                               be_emit_cstring(env, "@INDNTPOFF");
-                       } else {
-                               be_emit_cstring(env, "@NTPOFF");
-                       }
-               }
+                       be_emit_char('-');
+               ia32_emit_entity(ent, attr->data.am_sc_no_pic_adjust);
        }
 
-       if(offs != 0) {
-               if(ent != NULL) {
-                       be_emit_irprintf(env->emit, "%+d", offs);
+       /* also handle special case if nothing is set */
+       if (offs != 0 || (ent == NULL && !has_base && !has_index)) {
+               if (ent != NULL) {
+                       be_emit_irprintf("%+d", offs);
                } else {
-                       be_emit_irprintf(env->emit, "%d", offs);
+                       be_emit_irprintf("%d", offs);
                }
        }
 
        if (has_base || has_index) {
-               be_emit_char(env, '(');
+               be_emit_char('(');
 
                /* emit base */
                if (has_base) {
-                       ia32_emit_source_register(env, node, 0);
+                       const arch_register_t *reg = get_in_reg(node, n_ia32_base);
+                       emit_register(reg, NULL);
                }
 
                /* emit index + scale */
                if (has_index) {
+                       const arch_register_t *reg = get_in_reg(node, n_ia32_index);
                        int scale;
-                       be_emit_char(env, ',');
-                       ia32_emit_source_register(env, node, 1);
+                       be_emit_char(',');
+                       emit_register(reg, NULL);
 
                        scale = get_ia32_am_scale(node);
                        if (scale > 0) {
-                               be_emit_irprintf(env->emit, ",%d", 1 << get_ia32_am_scale(node));
+                               be_emit_irprintf(",%d", 1 << scale);
                        }
                }
-               be_emit_char(env, ')');
-       }
-
-       /* special case if nothing is set */
-       if(ent == NULL && offs == 0 && !has_base && !has_index) {
-               be_emit_char(env, '0');
+               be_emit_char(')');
        }
 }
 
-/*************************************************
- *                 _ _                         _
- *                (_) |                       | |
- *   ___ _ __ ___  _| |_    ___ ___  _ __   __| |
- *  / _ \ '_ ` _ \| | __|  / __/ _ \| '_ \ / _` |
- * |  __/ | | | | | | |_  | (_| (_) | | | | (_| |
- *  \___|_| |_| |_|_|\__|  \___\___/|_| |_|\__,_|
+/**
+ * fmt  parameter               output
+ * ---- ----------------------  ---------------------------------------------
+ * %%                           %
+ * %AM  <node>                  address mode of the node
+ * %AR  const arch_register_t*  address mode of the node or register
+ * %ASx <node>                  address mode of the node or source register x
+ * %Dx  <node>                  destination register x
+ * %I   <node>                  immediate of the node
+ * %L   <node>                  control flow target of the node
+ * %M   <node>                  mode suffix of the node
+ * %P   int                     condition code
+ * %R   const arch_register_t*  register
+ * %Sx  <node>                  source register x
+ * %s   const char*             string
+ * %u   unsigned int            unsigned int
+ * %d   signed int              signed int
  *
- *************************************************/
+ * x starts at 0
+ * # modifier for %ASx, %D and %S uses ls mode of node to alter register width
+ * * modifier does not prefix immediates with $, but AM with *
+ * l modifier for %lu and %ld
+ */
+static void ia32_emitf(const ir_node *node, const char *fmt, ...)
+{
+       va_list ap;
+       va_start(ap, fmt);
 
-#undef IA32_DO_EMIT
-#define IA32_DO_EMIT(irn) ia32_fprintf_format(F, irn, cmd_buf, cmnt_buf)
+       for (;;) {
+               const char      *start = fmt;
+               ia32_emit_mod_t  mod   = 0;
 
-/*
- * coding of conditions
- */
-struct cmp2conditon_t {
-       const char *name;
-       pn_Cmp      num;
-};
+               while (*fmt != '%' && *fmt != '\n' && *fmt != '\0')
+                       ++fmt;
+               if (fmt != start) {
+                       be_emit_string_len(start, fmt - start);
+               }
 
-/*
- * positive conditions for signed compares
- */
-static
-const struct cmp2conditon_t cmp2condition_s[] = {
-       { NULL,              pn_Cmp_False },  /* always false */
-       { "e",               pn_Cmp_Eq },     /* == */
-       { "l",               pn_Cmp_Lt },     /* < */
-       { "le",              pn_Cmp_Le },     /* <= */
-       { "g",               pn_Cmp_Gt },     /* > */
-       { "ge",              pn_Cmp_Ge },     /* >= */
-       { "ne",              pn_Cmp_Lg },     /* != */
-       { NULL,              pn_Cmp_Leg},     /* Floating point: ordered */
-};
+               if (*fmt == '\n') {
+                       be_emit_finish_line_gas(node);
+                       ++fmt;
+                       if (*fmt == '\0')
+                               break;
+                       continue;
+               }
 
-/*
- * positive conditions for unsigned compares
- */
-static
-const struct cmp2conditon_t cmp2condition_u[] = {
-       { NULL,              pn_Cmp_False },  /* always false */
-       { "e",               pn_Cmp_Eq },     /* == */
-       { "b",               pn_Cmp_Lt },     /* < */
-       { "be",              pn_Cmp_Le },     /* <= */
-       { "a",               pn_Cmp_Gt },     /* > */
-       { "ae",              pn_Cmp_Ge },     /* >= */
-       { "ne",              pn_Cmp_Lg },     /* != */
-       { NULL,              pn_Cmp_True },   /* always true */
-};
+               if (*fmt == '\0')
+                       break;
 
-/*
- * returns the condition code
- */
-static
-const char *get_cmp_suffix(pn_Cmp cmp_code)
-{
-       assert( (cmp2condition_s[cmp_code & 7].num) == (cmp_code & 7));
-       assert( (cmp2condition_u[cmp_code & 7].num) == (cmp_code & 7));
+               ++fmt;
+               if (*fmt == '*') {
+                       mod |= EMIT_ALTERNATE_AM;
+                       ++fmt;
+               }
 
-       if((cmp_code & ia32_pn_Cmp_Unsigned)) {
-               return cmp2condition_u[cmp_code & 7].name;
-       } else {
-               return cmp2condition_s[cmp_code & 7].name;
-       }
-}
+               if (*fmt == '#') {
+                       mod |= EMIT_RESPECT_LS;
+                       ++fmt;
+               }
 
-void ia32_emit_cmp_suffix(ia32_emit_env_t *env, long pnc)
-{
-       be_emit_string(env, get_cmp_suffix(pnc));
-}
+               if (*fmt == 'l') {
+                       mod |= EMIT_LONG;
+                       ++fmt;
+               }
 
+               switch (*fmt++) {
+                       case '%':
+                               be_emit_char('%');
+                               break;
 
-/**
- * Returns the target block for a control flow node.
+                       case 'A': {
+                               switch (*fmt++) {
+                                       case 'M':
+                                               if (mod & EMIT_ALTERNATE_AM)
+                                                       be_emit_char('*');
+
+                                               ia32_emit_am(node);
+                                               break;
+
+                                       case 'R': {
+                                               const arch_register_t *reg = va_arg(ap, const arch_register_t*);
+                                               if (mod & EMIT_ALTERNATE_AM)
+                                                       be_emit_char('*');
+                                               if (get_ia32_op_type(node) == ia32_AddrModeS) {
+                                                       ia32_emit_am(node);
+                                               } else {
+                                                       emit_register(reg, NULL);
+                                               }
+                                               break;
+                                       }
+
+                                       case 'S':
+                                               if (get_ia32_op_type(node) == ia32_AddrModeS) {
+                                                       if (mod & EMIT_ALTERNATE_AM)
+                                                               be_emit_char('*');
+                                                       ia32_emit_am(node);
+                                                       ++fmt;
+                                               } else {
+                                                       assert(get_ia32_op_type(node) == ia32_Normal);
+                                                       goto emit_S;
+                                               }
+                                               break;
+
+                                       default: goto unknown;
+                               }
+                               break;
+                       }
+
+                       case 'D': {
+                               unsigned               pos;
+                               const arch_register_t *reg;
+
+                               if (*fmt < '0' || '9' <= *fmt)
+                                       goto unknown;
+
+                               pos = *fmt++ - '0';
+                               reg = get_out_reg(node, pos);
+                               emit_register(reg, mod & EMIT_RESPECT_LS ? get_ia32_ls_mode(node) : NULL);
+                               break;
+                       }
+
+                       case 'I':
+                               if (!(mod & EMIT_ALTERNATE_AM))
+                                       be_emit_char('$');
+                               emit_ia32_Immediate_no_prefix(node);
+                               break;
+
+                       case 'L':
+                               ia32_emit_cfop_target(node);
+                               break;
+
+                       case 'M': {
+                               ia32_emit_mode_suffix_mode(get_ia32_ls_mode(node));
+                               break;
+                       }
+
+                       case 'P': {
+                               int pnc = va_arg(ap, int);
+                               ia32_emit_cmp_suffix(pnc);
+                               break;
+                       }
+
+                       case 'R': {
+                               const arch_register_t *reg = va_arg(ap, const arch_register_t*);
+                               emit_register(reg, NULL);
+                               break;
+                       }
+
+emit_S:
+                       case 'S': {
+                               unsigned       pos;
+                               const ir_node *in;
+
+                               if (*fmt < '0' || '9' <= *fmt)
+                                       goto unknown;
+
+                               pos = *fmt++ - '0';
+                               in  = get_irn_n(node, pos);
+                               if (is_ia32_Immediate(in)) {
+                                       if (!(mod & EMIT_ALTERNATE_AM))
+                                               be_emit_char('$');
+                                       emit_ia32_Immediate_no_prefix(in);
+                               } else {
+                                       const arch_register_t *reg;
+
+                                       if (mod & EMIT_ALTERNATE_AM)
+                                               be_emit_char('*');
+                                       reg = get_in_reg(node, pos);
+                                       emit_register(reg, mod & EMIT_RESPECT_LS ? get_ia32_ls_mode(node) : NULL);
+                               }
+                               break;
+                       }
+
+                       case 's': {
+                               const char *str = va_arg(ap, const char*);
+                               be_emit_string(str);
+                               break;
+                       }
+
+                       case 'u':
+                               if (mod & EMIT_LONG) {
+                                       unsigned long num = va_arg(ap, unsigned long);
+                                       be_emit_irprintf("%lu", num);
+                               } else {
+                                       unsigned num = va_arg(ap, unsigned);
+                                       be_emit_irprintf("%u", num);
+                               }
+                               break;
+
+                       case 'd':
+                               if (mod & EMIT_LONG) {
+                                       long num = va_arg(ap, long);
+                                       be_emit_irprintf("%ld", num);
+                               } else {
+                                       int num = va_arg(ap, int);
+                                       be_emit_irprintf("%d", num);
+                               }
+                               break;
+
+                       default:
+unknown:
+                               panic("unknown format conversion in ia32_emitf()");
+               }
+       }
+
+       va_end(ap);
+}
+
+/**
+ * Emits registers and/or address mode of a binary operation.
  */
-static
-ir_node *get_cfop_target_block(const ir_node *irn) {
-       return get_irn_link(irn);
+void ia32_emit_binop(const ir_node *node)
+{
+       if (is_ia32_Immediate(get_irn_n(node, n_ia32_binary_right))) {
+               ia32_emitf(node, "%#S4, %#AS3");
+       } else {
+               ia32_emitf(node, "%#AS4, %#S3");
+       }
 }
 
 /**
- * Emits a block label for the given block.
+ * Emits registers and/or address mode of a binary operation.
+ */
+void ia32_emit_x87_binop(const ir_node *node)
+{
+       switch(get_ia32_op_type(node)) {
+               case ia32_Normal:
+                       {
+                               const ia32_x87_attr_t *x87_attr = get_ia32_x87_attr_const(node);
+                               const arch_register_t *in1      = x87_attr->x87[0];
+                               const arch_register_t *in       = x87_attr->x87[1];
+                               const arch_register_t *out      = x87_attr->x87[2];
+
+                               if (out == NULL) {
+                                       out = in1;
+                               } else if (out == in) {
+                                       in = in1;
+                               }
+
+                               be_emit_char('%');
+                               be_emit_string(arch_register_get_name(in));
+                               be_emit_cstring(", %");
+                               be_emit_string(arch_register_get_name(out));
+                       }
+                       break;
+               case ia32_AddrModeS:
+                       ia32_emit_am(node);
+                       break;
+               case ia32_AddrModeD:
+               default:
+                       assert(0 && "unsupported op type");
+       }
+}
+
+/**
+ * Emits registers and/or address mode of a unary operation.
  */
-static
-void ia32_emit_block_name(ia32_emit_env_t *env, const ir_node *block)
+void ia32_emit_unop(const ir_node *node, int pos)
 {
-       if (has_Block_label(block)) {
-               be_emit_string(env, be_gas_label_prefix());
-               be_emit_irprintf(env->emit, "%u", (unsigned)get_Block_label(block));
+       char fmt[] = "%ASx";
+       fmt[3] = '0' + pos;
+       ia32_emitf(node, fmt);
+}
+
+static void emit_ia32_IMul(const ir_node *node)
+{
+       ir_node               *left    = get_irn_n(node, n_ia32_IMul_left);
+       const arch_register_t *out_reg = get_out_reg(node, pn_ia32_IMul_res);
+
+       /* do we need the 3-address form? */
+       if (is_ia32_NoReg_GP(left) ||
+                       get_in_reg(node, n_ia32_IMul_left) != out_reg) {
+               ia32_emitf(node, "\timul%M %#S4, %#AS3, %#D0\n");
        } else {
-               be_emit_cstring(env, BLOCK_PREFIX);
-               be_emit_irprintf(env->emit, "%d", get_irn_node_nr(block));
+               ia32_emitf(node, "\timul%M %#AS4, %#S3\n");
        }
 }
 
 /**
- * Emits the target label for a control flow node.
+ * walks up a tree of copies/perms/spills/reloads to find the original value
+ * that is moved around
  */
-static
-void ia32_emit_cfop_target(ia32_emit_env_t * env, const ir_node *node) {
-       ir_node *block = get_cfop_target_block(node);
+static ir_node *find_original_value(ir_node *node)
+{
+       if (irn_visited(node))
+               return NULL;
+
+       mark_irn_visited(node);
+       if (be_is_Copy(node)) {
+               return find_original_value(be_get_Copy_op(node));
+       } else if (be_is_CopyKeep(node)) {
+               return find_original_value(be_get_CopyKeep_op(node));
+       } else if (is_Proj(node)) {
+               ir_node *pred = get_Proj_pred(node);
+               if (be_is_Perm(pred)) {
+                       return find_original_value(get_irn_n(pred, get_Proj_proj(node)));
+               } else if (be_is_MemPerm(pred)) {
+                       return find_original_value(get_irn_n(pred, get_Proj_proj(node) + 1));
+               } else if (is_ia32_Load(pred)) {
+                       return find_original_value(get_irn_n(pred, n_ia32_Load_mem));
+               } else {
+                       return node;
+               }
+       } else if (is_ia32_Store(node)) {
+               return find_original_value(get_irn_n(node, n_ia32_Store_val));
+       } else if (is_Phi(node)) {
+               int i, arity;
+               arity = get_irn_arity(node);
+               for (i = 0; i < arity; ++i) {
+                       ir_node *in  = get_irn_n(node, i);
+                       ir_node *res = find_original_value(in);
+
+                       if (res != NULL)
+                               return res;
+               }
+               return NULL;
+       } else {
+               return node;
+       }
+}
+
+static int determine_final_pnc(const ir_node *node, int flags_pos,
+                               int pnc)
+{
+       ir_node           *flags = get_irn_n(node, flags_pos);
+       const ia32_attr_t *flags_attr;
+       flags = skip_Proj(flags);
+
+       if (is_ia32_Sahf(flags)) {
+               ir_node *cmp = get_irn_n(flags, n_ia32_Sahf_val);
+               if (!(is_ia32_FucomFnstsw(cmp) || is_ia32_FucompFnstsw(cmp)
+                               || is_ia32_FucomppFnstsw(cmp) || is_ia32_FtstFnstsw(cmp))) {
+                       inc_irg_visited(current_ir_graph);
+                       cmp = find_original_value(cmp);
+                       assert(cmp != NULL);
+                       assert(is_ia32_FucomFnstsw(cmp) || is_ia32_FucompFnstsw(cmp)
+                              || is_ia32_FucomppFnstsw(cmp) || is_ia32_FtstFnstsw(cmp));
+               }
+
+               flags_attr = get_ia32_attr_const(cmp);
+               if (flags_attr->data.ins_permuted)
+                       pnc = get_mirrored_pnc(pnc);
+               pnc |= ia32_pn_Cmp_float;
+       } else if (is_ia32_Ucomi(flags) || is_ia32_Fucomi(flags)
+                       || is_ia32_Fucompi(flags)) {
+               flags_attr = get_ia32_attr_const(flags);
+
+               if (flags_attr->data.ins_permuted)
+                       pnc = get_mirrored_pnc(pnc);
+               pnc |= ia32_pn_Cmp_float;
+       } else {
+               flags_attr = get_ia32_attr_const(flags);
 
-       ia32_emit_block_name(env, block);
+               if (flags_attr->data.ins_permuted)
+                       pnc = get_mirrored_pnc(pnc);
+               if (flags_attr->data.cmp_unsigned)
+                       pnc |= ia32_pn_Cmp_unsigned;
+       }
+
+       return pnc;
 }
 
-/** Return the next block in Block schedule */
-static ir_node *next_blk_sched(const ir_node *block) {
-       return get_irn_link(block);
+static pn_Cmp ia32_get_negated_pnc(pn_Cmp pnc)
+{
+       ir_mode *mode = pnc & ia32_pn_Cmp_float ? mode_F : mode_Iu;
+       return get_negated_pnc(pnc, mode);
+}
+
+void ia32_emit_cmp_suffix_node(const ir_node *node,
+                               int flags_pos)
+{
+       const ia32_attr_t *attr = get_ia32_attr_const(node);
+
+       pn_Cmp pnc = get_ia32_condcode(node);
+
+       pnc = determine_final_pnc(node, flags_pos, pnc);
+       if (attr->data.ins_permuted)
+               pnc = ia32_get_negated_pnc(pnc);
+
+       ia32_emit_cmp_suffix(pnc);
+}
+
+/**
+ * Emits an exception label for a given node.
+ */
+static void ia32_emit_exc_label(const ir_node *node)
+{
+       be_emit_string(be_gas_insn_label_prefix());
+       be_emit_irprintf("%lu", get_ia32_exc_label_id(node));
 }
 
 /**
  * Returns the Proj with projection number proj and NOT mode_M
  */
-static
-ir_node *get_proj(const ir_node *node, long proj) {
+static ir_node *get_proj(const ir_node *node, long proj)
+{
        const ir_edge_t *edge;
        ir_node         *src;
 
@@ -762,419 +1042,134 @@ ir_node *get_proj(const ir_node *node, long proj) {
        return NULL;
 }
 
+static int can_be_fallthrough(const ir_node *node)
+{
+       ir_node *target_block = get_cfop_target_block(node);
+       ir_node *block        = get_nodes_block(node);
+       return get_prev_block_sched(target_block) == block;
+}
+
 /**
  * Emits the jump sequence for a conditional jump (cmp + jmp_true + jmp_false)
  */
-static
-void finish_CondJmp(ia32_emit_env_t *env, const ir_node *node, ir_mode *mode,
-                    long pnc) {
+static void emit_ia32_Jcc(const ir_node *node)
+{
+       int            need_parity_label = 0;
        const ir_node *proj_true;
        const ir_node *proj_false;
        const ir_node *block;
-       const ir_node *next_block;
-       int flipped = 0;
+       pn_Cmp         pnc = get_ia32_condcode(node);
 
-       /* get both Proj's */
-       proj_true = get_proj(node, pn_Cond_true);
-       assert(proj_true && "CondJmp without true Proj");
+       pnc = determine_final_pnc(node, 0, pnc);
 
-       proj_false = get_proj(node, pn_Cond_false);
-       assert(proj_false && "CondJmp without false Proj");
+       /* get both Projs */
+       proj_true = get_proj(node, pn_ia32_Jcc_true);
+       assert(proj_true && "Jcc without true Proj");
 
-       /* for now, the code works for scheduled and non-schedules blocks */
-       block = get_nodes_block(node);
+       proj_false = get_proj(node, pn_ia32_Jcc_false);
+       assert(proj_false && "Jcc without false Proj");
 
-       /* we have a block schedule */
-       next_block = next_blk_sched(block);
+       block      = get_nodes_block(node);
 
-       if (get_cfop_target_block(proj_true) == next_block) {
+       if (can_be_fallthrough(proj_true)) {
                /* exchange both proj's so the second one can be omitted */
                const ir_node *t = proj_true;
 
                proj_true  = proj_false;
                proj_false = t;
-               flipped    = 1;
-               pnc        = get_negated_pnc(pnc, mode);
+               pnc        = ia32_get_negated_pnc(pnc);
        }
 
-       if (mode_is_float(mode)) {
-               /* Some floating point comparisons require a test of the parity flag, which
-                * indicates that the result is unordered */
-               switch (pnc) {
-                       case pn_Cmp_Uo:
-                               be_emit_cstring(env, "\tjp ");
-                               ia32_emit_cfop_target(env, proj_true);
-                               be_emit_finish_line_gas(env, proj_true);
+       if (pnc & ia32_pn_Cmp_float) {
+               /* Some floating point comparisons require a test of the parity flag,
+                * which indicates that the result is unordered */
+               switch (pnc & 15) {
+                       case pn_Cmp_Uo: {
+                               ia32_emitf(proj_true, "\tjp %L\n");
                                break;
+                       }
 
                        case pn_Cmp_Leg:
-                               be_emit_cstring(env, "\tjnp ");
-                               ia32_emit_cfop_target(env, proj_true);
-                               be_emit_finish_line_gas(env, proj_true);
+                               ia32_emitf(proj_true, "\tjnp %L\n");
                                break;
 
                        case pn_Cmp_Eq:
                        case pn_Cmp_Lt:
                        case pn_Cmp_Le:
-                               be_emit_cstring(env, "\tjp ");
-                               ia32_emit_cfop_target(env, proj_false);
-                               be_emit_finish_line_gas(env, proj_false);
-                               goto float_jcc;
+                               /* we need a local label if the false proj is a fallthrough
+                                * as the falseblock might have no label emitted then */
+                               if (can_be_fallthrough(proj_false)) {
+                                       need_parity_label = 1;
+                                       ia32_emitf(proj_false, "\tjp 1f\n");
+                               } else {
+                                       ia32_emitf(proj_false, "\tjp %L\n");
+                               }
+                               goto emit_jcc;
 
                        case pn_Cmp_Ug:
                        case pn_Cmp_Uge:
                        case pn_Cmp_Ne:
-                               be_emit_cstring(env, "\tjp ");
-                               ia32_emit_cfop_target(env, proj_true);
-                               be_emit_finish_line_gas(env, proj_true);
-                               goto float_jcc;
+                               ia32_emitf(proj_true, "\tjp %L\n");
+                               goto emit_jcc;
 
                        default:
-                       float_jcc:
-                               /* The bits set by floating point compares correspond to unsigned
-                                * comparisons */
-                               pnc |= ia32_pn_Cmp_Unsigned;
                                goto emit_jcc;
                }
        } else {
 emit_jcc:
-               be_emit_cstring(env, "\tj");
-               ia32_emit_cmp_suffix(env, pnc);
-               be_emit_char(env, ' ');
-               ia32_emit_cfop_target(env, proj_true);
-               be_emit_finish_line_gas(env, proj_true);
-       }
-
-       /* the second Proj might be a fallthrough */
-       if (get_cfop_target_block(proj_false) != next_block) {
-               be_emit_cstring(env, "\tjmp ");
-               ia32_emit_cfop_target(env, proj_false);
-               be_emit_finish_line_gas(env, proj_false);
-       } else {
-               be_emit_cstring(env, "\t/* fallthrough to ");
-               ia32_emit_cfop_target(env, proj_false);
-               be_emit_cstring(env, " */");
-               be_emit_finish_line_gas(env, proj_false);
-       }
-}
-
-/**
- * Emits code for conditional jump.
- */
-static
-void CondJmp_emitter(ia32_emit_env_t *env, const ir_node *node) {
-       be_emit_cstring(env, "\tcmp");
-       ia32_emit_mode_suffix(env, node);
-       be_emit_char(env, ' ');
-       ia32_emit_binop(env, node);
-       be_emit_finish_line_gas(env, node);
-
-       finish_CondJmp(env, node, mode_Iu, get_ia32_pncode(node));
-}
-
-/**
- * Emits code for conditional jump with two variables.
- */
-static
-void emit_ia32_CmpJmp(ia32_emit_env_t *env, const ir_node *node) {
-       CondJmp_emitter(env, node);
-}
-
-/**
- * Emits code for conditional test and jump.
- */
-static
-void TestJmp_emitter(ia32_emit_env_t *env, const ir_node *node) {
-       be_emit_cstring(env, "\ttest");
-       ia32_emit_mode_suffix(env, node);
-       be_emit_char(env, ' ');
-
-       ia32_emit_binop(env, node);
-       be_emit_finish_line_gas(env, node);
-
-       finish_CondJmp(env, node, mode_Iu, get_ia32_pncode(node));
-}
-
-/**
- * Emits code for conditional test and jump with two variables.
- */
-static
-void emit_ia32_TestJmp(ia32_emit_env_t *env, const ir_node *node) {
-       TestJmp_emitter(env, node);
-}
-
-/**
- * Emits code for conditional SSE floating point jump with two variables.
- */
-static
-void emit_ia32_xCmpJmp(ia32_emit_env_t *env, const ir_node *node) {
-       be_emit_cstring(env, "\tucomi");
-       ia32_emit_xmm_mode_suffix(env, node);
-       be_emit_char(env, ' ');
-       ia32_emit_binop(env, node);
-       be_emit_finish_line_gas(env, node);
-
-       finish_CondJmp(env, node, mode_F, get_ia32_pncode(node));
-}
-
-/**
- * Emits code for conditional x87 floating point jump with two variables.
- */
-static
-void emit_ia32_x87CmpJmp(ia32_emit_env_t *env, const ir_node *node) {
-       const ia32_x87_attr_t *x87_attr = get_ia32_x87_attr_const(node);
-       const char            *reg      = x87_attr->x87[1]->name;
-       long                   pnc      = get_ia32_pncode(node);
-
-       switch (get_ia32_irn_opcode(node)) {
-       case iro_ia32_fcomrJmp:
-               pnc = get_inversed_pnc(pnc);
-               reg = x87_attr->x87[0]->name;
-       case iro_ia32_fcomJmp:
-       default:
-               be_emit_cstring(env, "\tfucom ");
-               break;
-       case iro_ia32_fcomrpJmp:
-               pnc = get_inversed_pnc(pnc);
-               reg = x87_attr->x87[0]->name;
-       case iro_ia32_fcompJmp:
-               be_emit_cstring(env, "\tfucomp ");
-               break;
-       case iro_ia32_fcomrppJmp:
-               pnc = get_inversed_pnc(pnc);
-       case iro_ia32_fcomppJmp:
-               be_emit_cstring(env, "\tfucompp ");
-               reg = "";
-               break;
+               ia32_emitf(proj_true, "\tj%P %L\n", pnc);
        }
 
-       if(reg[0] != '\0') {
-               be_emit_char(env, '%');
-               be_emit_string(env, reg);
+       if (need_parity_label) {
+               ia32_emitf(NULL, "1:\n");
        }
-       be_emit_finish_line_gas(env, node);
-
-       be_emit_cstring(env, "\tfnstsw %ax");
-       be_emit_finish_line_gas(env, node);
-       be_emit_cstring(env, "\tsahf");
-       be_emit_finish_line_gas(env, node);
-
-       finish_CondJmp(env, node, mode_E, pnc);
-}
-
-static
-void CMov_emitter(ia32_emit_env_t *env, const ir_node *node)
-{
-       const arch_register_t *in1, *in2, *out;
-       long  pnc = get_ia32_pncode(node);
-
-       out = arch_get_irn_register(env->arch_env, node);
 
-       /* we have to emit the cmp first, because the destination register */
-       /* could be one of the compare registers                           */
-       if (is_ia32_xCmpCMov(node)) {
-               be_emit_cstring(env, "\tucomis");
-               ia32_emit_mode_suffix_mode(env, get_irn_mode(node));
-               be_emit_char(env, ' ');
-               ia32_emit_source_register(env, node, 1);
-               be_emit_cstring(env, ", ");
-               ia32_emit_source_register(env, node, 0);
-
-               in1 = arch_get_irn_register(env->arch_env, get_irn_n(node, 2));
-               in2 = arch_get_irn_register(env->arch_env, get_irn_n(node, 3));
-       } else {
-               if (is_ia32_CmpCMov(node) || is_ia32_CmpCMov8Bit(node)) {
-                       be_emit_cstring(env, "\tcmp ");
-               } else {
-                       assert(is_ia32_TestCMov(node) || is_ia32_TestCMov8Bit(node));
-                       be_emit_cstring(env, "\ttest ");
-               }
-               ia32_emit_binop(env, node);
-
-               in1 = arch_get_irn_register(env->arch_env, get_irn_n(node, 5));
-               in2 = arch_get_irn_register(env->arch_env, get_irn_n(node, 6));
-       }
-       be_emit_finish_line_gas(env, node);
-
-       if (out == in2) {
-               /* best case: default in == out -> do nothing */
-       } else if(in2 == &ia32_gp_regs[REG_GP_UKNWN]) {
-               /* also nothign to do for unknown regs */
-       } else if (out == in1) {
-               const arch_register_t *t;
-               /* true in == out -> need complement compare and exchange true and
-                * default in */
-               t   = in1;
-               in1 = in2;
-               in2 = t;
-               pnc = get_negated_pnc(pnc, get_irn_mode(node));
+       /* the second Proj might be a fallthrough */
+       if (can_be_fallthrough(proj_false)) {
+               ia32_emitf(proj_false, "\t/* fallthrough to %L */\n");
        } else {
-               /* out is different from both ins: need copy default -> out */
-               be_emit_cstring(env, "\tmovl ");
-               ia32_emit_register(env, in2);
-               be_emit_cstring(env, ", ");
-               ia32_emit_register(env, out);
-               be_emit_finish_line_gas(env, node);
+               ia32_emitf(proj_false, "\tjmp %L\n");
        }
-
-       be_emit_cstring(env, "\tcmov");
-       ia32_emit_cmp_suffix(env, pnc );
-       be_emit_cstring(env, "l ");
-       ia32_emit_register(env, in1);
-       be_emit_cstring(env, ", ");
-       ia32_emit_register(env, out);
-
-       be_emit_finish_line_gas(env, node);
-}
-
-static
-void emit_ia32_CmpCMov(ia32_emit_env_t *env, const ir_node *node)
-{
-       CMov_emitter(env, node);
-}
-
-static
-void emit_ia32_TestCMov(ia32_emit_env_t *env, const ir_node *node)
-{
-       CMov_emitter(env, node);
-}
-
-static
-void emit_ia32_xCmpCMov(ia32_emit_env_t *env, const ir_node *node)
-{
-       CMov_emitter(env, node);
 }
 
-static
-void Set_emitter(ia32_emit_env_t *env, const ir_node *node)
+static void emit_ia32_CMov(const ir_node *node)
 {
-       long pnc = get_ia32_pncode(node);
-       const char *reg8bit;
-       const arch_register_t *out;
+       const ia32_attr_t     *attr         = get_ia32_attr_const(node);
+       int                    ins_permuted = attr->data.ins_permuted;
+       const arch_register_t *out          = arch_irn_get_register(node, pn_ia32_res);
+       pn_Cmp                 pnc          = get_ia32_condcode(node);
+       const arch_register_t *in_true;
+       const arch_register_t *in_false;
 
-       out     = arch_get_irn_register(env->arch_env, node);
-       reg8bit = ia32_get_mapped_reg_name(env->isa->regs_8bit, out);
+       pnc = determine_final_pnc(node, n_ia32_CMov_eflags, pnc);
 
-       if(is_ia32_xCmpSet(node)) {
-               be_emit_cstring(env, "\tucomis");
-               ia32_emit_mode_suffix_mode(env, get_irn_mode(get_irn_n(node, 2)));
-               be_emit_char(env, ' ');
-               ia32_emit_binop(env, node);
-       } else {
-               if (is_ia32_CmpSet(node) || is_ia32_CmpSet8Bit(node)) {
-                       be_emit_cstring(env, "\tcmp");
-               } else {
-                       assert(is_ia32_TestSet(node) || is_ia32_TestSet8Bit(node));
-                       be_emit_cstring(env, "\ttest");
-               }
-               ia32_emit_mode_suffix(env, node);
-               be_emit_char(env, ' ');
-               ia32_emit_binop(env, node);
-       }
-       be_emit_finish_line_gas(env, node);
-
-       be_emit_cstring(env, "\tset");
-       ia32_emit_cmp_suffix(env, pnc);
-       be_emit_cstring(env, " %");
-       be_emit_string(env, reg8bit);
-       be_emit_finish_line_gas(env, node);
-}
-
-static
-void emit_ia32_CmpSet(ia32_emit_env_t *env, const ir_node *node) {
-       Set_emitter(env, node);
-}
-
-static
-void emit_ia32_TestSet(ia32_emit_env_t *env, const ir_node *node) {
-       Set_emitter(env, node);
-}
-
-static
-void emit_ia32_xCmpSet(ia32_emit_env_t *env, const ir_node *node) {
-       Set_emitter(env, node);
-}
-
-static
-void emit_ia32_xCmp(ia32_emit_env_t *env, const ir_node *node) {
-       int  sse_pnc  = -1;
-       long pnc      = get_ia32_pncode(node);
-       long unord    = pnc & pn_Cmp_Uo;
-
-       assert( (pnc & ia32_pn_Cmp_Unsigned) == 0);
-
-       switch (pnc) {
-               case pn_Cmp_Leg: /* odered */
-                       sse_pnc = 7;
-                       break;
-               case pn_Cmp_Uo:  /* unordered */
-                       sse_pnc = 3;
-                       break;
-               case pn_Cmp_Ue:
-               case pn_Cmp_Eq:  /* == */
-                       sse_pnc = 0;
-                       break;
-               case pn_Cmp_Ul:
-               case pn_Cmp_Lt:  /* < */
-                       sse_pnc = 1;
-                       break;
-               case pn_Cmp_Ule:
-               case pn_Cmp_Le: /* <= */
-                       sse_pnc = 2;
-                       break;
-               case pn_Cmp_Ug:
-               case pn_Cmp_Gt:  /* > */
-                       sse_pnc = 6;
-                       break;
-               case pn_Cmp_Uge:
-               case pn_Cmp_Ge: /* >= */
-                       sse_pnc = 5;
-                       break;
-               case pn_Cmp_Ne:
-               case pn_Cmp_Lg:  /* != */
-                       sse_pnc = 4;
-                       break;
-       }
+       in_true  = arch_get_irn_register(get_irn_n(node, n_ia32_CMov_val_true));
+       in_false = arch_get_irn_register(get_irn_n(node, n_ia32_CMov_val_false));
 
-       assert(sse_pnc >= 0 && "unsupported compare");
+       /* should be same constraint fullfilled? */
+       if (out == in_false) {
+               /* yes -> nothing to do */
+       } else if (out == in_true) {
+               const arch_register_t *tmp;
 
-       if (unord && sse_pnc != 3) {
-               /*
-                       We need a separate compare against unordered.
-                       Quick and Dirty solution:
-                       - get some memory on stack
-                       - compare
-                       - store result
-                       - compare
-                       - and result and stored result
-                   - cleanup stack
-               */
-               be_emit_cstring(env, "\tsubl $8, %esp");
-               be_emit_finish_line_gas(env, node);
+               assert(get_ia32_op_type(node) == ia32_Normal);
 
-               be_emit_cstring(env, "\tcmpsd $3, ");
-               ia32_emit_binop(env, node);
-               be_emit_finish_line_gas(env, node);
+               ins_permuted = !ins_permuted;
 
-               be_emit_cstring(env, "\tmovsd ");
-               ia32_emit_dest_register(env, node, 0);
-               be_emit_cstring(env, ", (%esp)");
-               be_emit_finish_line_gas(env, node);
+               tmp      = in_true;
+               in_true  = in_false;
+               in_false = tmp;
+       } else {
+               /* we need a mov */
+               ia32_emitf(node, "\tmovl %R, %R\n", in_false, out);
        }
 
-       be_emit_cstring(env, "\tcmpsd ");
-       be_emit_irprintf(env->emit, "%d, ", sse_pnc);
-       ia32_emit_binop(env, node);
-       be_emit_finish_line_gas(env, node);
+       if (ins_permuted)
+               pnc = ia32_get_negated_pnc(pnc);
 
-       if (unord && sse_pnc != 3) {
-               be_emit_cstring(env, "\tandpd (%esp), ");
-               ia32_emit_dest_register(env, node, 0);
-               be_emit_finish_line_gas(env, node);
+       /* TODO: handling of Nans isn't correct yet */
 
-               be_emit_cstring(env, "\taddl $8, %esp");
-               be_emit_finish_line_gas(env, node);
-       }
+       ia32_emitf(node, "\tcmov%P %#AR, %#R\n", pnc, in_true, out);
 }
 
 /*********************************************************
@@ -1207,8 +1202,8 @@ typedef struct _jmp_tbl_t {
 /**
  * Compare two variables of type branch_t. Used to sort all switch cases
  */
-static
-int ia32_cmp_branch_t(const void *a, const void *b) {
+static int ia32_cmp_branch_t(const void *a, const void *b)
+{
        branch_t *b1 = (branch_t *)a;
        branch_t *b2 = (branch_t *)b;
 
@@ -1223,24 +1218,26 @@ int ia32_cmp_branch_t(const void *a, const void *b) {
  * possible otherwise a cmp-jmp cascade). Port from
  * cggg ia32 backend
  */
-static
-void emit_ia32_SwitchJmp(ia32_emit_env_t *env, const ir_node *node) {
+static void emit_ia32_SwitchJmp(const ir_node *node)
+{
        unsigned long       interval;
        int                 last_value, i;
        long                pnc;
+       long                default_pn;
        jmp_tbl_t           tbl;
        ir_node            *proj;
        const ir_edge_t    *edge;
 
        /* fill the table structure */
-       tbl.label        = xmalloc(SNPRINTF_BUF_LEN);
+       tbl.label        = XMALLOCN(char, SNPRINTF_BUF_LEN);
        tbl.label        = get_unique_label(tbl.label, SNPRINTF_BUF_LEN, ".TBL_");
        tbl.defProj      = NULL;
-       tbl.num_branches = get_irn_n_edges(node);
-       tbl.branches     = xcalloc(tbl.num_branches, sizeof(tbl.branches[0]));
+       tbl.num_branches = get_irn_n_edges(node) - 1;
+       tbl.branches     = XMALLOCNZ(branch_t, tbl.num_branches);
        tbl.min_value    = INT_MAX;
        tbl.max_value    = INT_MIN;
 
+       default_pn = get_ia32_condcode(node);
        i = 0;
        /* go over all proj's and collect them */
        foreach_out_edge(node, edge) {
@@ -1249,21 +1246,22 @@ void emit_ia32_SwitchJmp(ia32_emit_env_t *env, const ir_node *node) {
 
                pnc = get_Proj_proj(proj);
 
-               /* create branch entry */
-               tbl.branches[i].target = proj;
-               tbl.branches[i].value  = pnc;
-
-               tbl.min_value = pnc < tbl.min_value ? pnc : tbl.min_value;
-               tbl.max_value = pnc > tbl.max_value ? pnc : tbl.max_value;
-
                /* check for default proj */
-               if (pnc == get_ia32_pncode(node)) {
-                       assert(tbl.defProj == NULL && "found two defProjs at SwitchJmp");
+               if (pnc == default_pn) {
+                       assert(tbl.defProj == NULL && "found two default Projs at SwitchJmp");
                        tbl.defProj = proj;
+               } else {
+                       tbl.min_value = pnc < tbl.min_value ? pnc : tbl.min_value;
+                       tbl.max_value = pnc > tbl.max_value ? pnc : tbl.max_value;
+
+                       /* create branch entry */
+                       tbl.branches[i].target = proj;
+                       tbl.branches[i].value  = pnc;
+                       ++i;
                }
 
-               i++;
        }
+       assert(i == tbl.num_branches);
 
        /* sort the branches by their number */
        qsort(tbl.branches, tbl.num_branches, sizeof(tbl.branches[0]), ia32_cmp_branch_t);
@@ -1272,53 +1270,30 @@ void emit_ia32_SwitchJmp(ia32_emit_env_t *env, const ir_node *node) {
        interval = tbl.max_value - tbl.min_value;
 
        /* emit the table */
-       be_emit_cstring(env, "\tcmpl $");
-       be_emit_irprintf(env->emit, "%u, ", interval);
-       ia32_emit_source_register(env, node, 0);
-       be_emit_finish_line_gas(env, node);
-
-       be_emit_cstring(env, "\tja ");
-       ia32_emit_cfop_target(env, tbl.defProj);
-       be_emit_finish_line_gas(env, node);
+       ia32_emitf(node,        "\tcmpl $%u, %S0\n", interval);
+       ia32_emitf(tbl.defProj, "\tja %L\n");
 
        if (tbl.num_branches > 1) {
                /* create table */
-               be_emit_cstring(env, "\tjmp *");
-               be_emit_string(env, tbl.label);
-               be_emit_cstring(env, "(,");
-               ia32_emit_source_register(env, node, 0);
-               be_emit_cstring(env, ",4)");
-               be_emit_finish_line_gas(env, node);
-
-               be_gas_emit_switch_section(env->emit, GAS_SECTION_RODATA);
-               be_emit_cstring(env, "\t.align 4\n");
-               be_emit_write_line(env);
+               ia32_emitf(node, "\tjmp *%s(,%S0,4)\n", tbl.label);
 
-               be_emit_string(env, tbl.label);
-               be_emit_cstring(env, ":\n");
-               be_emit_write_line(env);
-
-               be_emit_cstring(env, ".long ");
-               ia32_emit_cfop_target(env, tbl.branches[0].target);
-               be_emit_finish_line_gas(env, NULL);
+               be_gas_emit_switch_section(GAS_SECTION_RODATA);
+               ia32_emitf(NULL, "\t.align 4\n");
+               ia32_emitf(NULL, "%s:\n", tbl.label);
 
                last_value = tbl.branches[0].value;
-               for (i = 1; i < tbl.num_branches; ++i) {
-                       while (++last_value < tbl.branches[i].value) {
-                               be_emit_cstring(env, ".long ");
-                               ia32_emit_cfop_target(env, tbl.defProj);
-                               be_emit_finish_line_gas(env, NULL);
+               for (i = 0; i != tbl.num_branches; ++i) {
+                       while (last_value != tbl.branches[i].value) {
+                               ia32_emitf(tbl.defProj, ".long %L\n");
+                               ++last_value;
                        }
-                       be_emit_cstring(env, ".long ");
-                       ia32_emit_cfop_target(env, tbl.branches[i].target);
-                       be_emit_finish_line_gas(env, NULL);
+                       ia32_emitf(tbl.branches[i].target, ".long %L\n");
+                       ++last_value;
                }
-               be_gas_emit_switch_section(env->emit, GAS_SECTION_TEXT);
+               be_gas_emit_switch_section(GAS_SECTION_TEXT);
        } else {
                /* one jump is enough */
-               be_emit_cstring(env, "\tjmp ");
-               ia32_emit_cfop_target(env, tbl.branches[0].target);
-               be_emit_finish_line_gas(env, node);
+               ia32_emitf(tbl.branches[0].target, "\tjmp %L\n");
        }
 
        if (tbl.label)
@@ -1330,57 +1305,41 @@ void emit_ia32_SwitchJmp(ia32_emit_env_t *env, const ir_node *node) {
 /**
  * Emits code for a unconditional jump.
  */
-static
-void emit_Jmp(ia32_emit_env_t *env, const ir_node *node) {
-       ir_node *block, *next_block;
+static void emit_ia32_Jmp(const ir_node *node)
+{
+       ir_node *block;
 
        /* for now, the code works for scheduled and non-schedules blocks */
        block = get_nodes_block(node);
 
        /* we have a block schedule */
-       next_block = next_blk_sched(block);
-       if (get_cfop_target_block(node) != next_block) {
-               be_emit_cstring(env, "\tjmp ");
-               ia32_emit_cfop_target(env, node);
+       if (can_be_fallthrough(node)) {
+               ia32_emitf(node, "\t/* fallthrough to %L */\n");
        } else {
-               be_emit_cstring(env, "\t/* fallthrough to ");
-               ia32_emit_cfop_target(env, node);
-               be_emit_cstring(env, " */");
-       }
-       be_emit_finish_line_gas(env, node);
-}
-
-static
-void emit_ia32_Immediate(ia32_emit_env_t *env, const ir_node *node)
-{
-       const ia32_immediate_attr_t *attr = get_ia32_immediate_attr_const(node);
-
-       be_emit_char(env, '$');
-       if(attr->symconst != NULL) {
-               ident *id = get_entity_ld_ident(attr->symconst);
-
-               if(attr->attr.data.am_sc_sign)
-                       be_emit_char(env, '-');
-               be_emit_ident(env, id);
-       }
-       if(attr->symconst == NULL || attr->offset != 0) {
-               if(attr->symconst != NULL)
-                       be_emit_char(env, '+');
-               be_emit_irprintf(env->emit, "0x%X", attr->offset);
+               ia32_emitf(node, "\tjmp %L\n");
        }
 }
 
-static
-const char* emit_asm_operand(ia32_emit_env_t *env, const ir_node *node,
-                             const char *s)
+/**
+ * Emit an inline assembler operand.
+ *
+ * @param node  the ia32_ASM node
+ * @param s     points to the operand (a %c)
+ *
+ * @return  pointer to the first char in s NOT in the current operand
+ */
+static const char* emit_asm_operand(const ir_node *node, const char *s)
 {
+       const ia32_attr_t     *ia32_attr = get_ia32_attr_const(node);
+       const ia32_asm_attr_t *attr      = CONST_CAST_IA32_ATTR(ia32_asm_attr_t,
+                                                            ia32_attr);
        const arch_register_t *reg;
+       const ia32_asm_reg_t  *asm_regs = attr->register_map;
+       const ia32_asm_reg_t  *asm_reg;
        const char            *reg_name;
        char                   c;
        char                   modifier = 0;
        int                    num      = -1;
-       const ia32_attr_t     *attr;
-       int                    n_outs;
        int                    p;
 
        assert(*s == '%');
@@ -1389,11 +1348,11 @@ const char* emit_asm_operand(ia32_emit_env_t *env, const ir_node *node,
        /* parse modifiers */
        switch(c) {
        case 0:
-               ir_fprintf(stderr, "Warning: asm text (%+F) ends with %\n", node);
-               be_emit_char(env, '%');
+               ir_fprintf(stderr, "Warning: asm text (%+F) ends with %%\n", node);
+               be_emit_char('%');
                return s + 1;
        case '%':
-               be_emit_char(env, '%');
+               be_emit_char('%');
                return s + 1;
        case 'w':
        case 'b':
@@ -1413,15 +1372,16 @@ const char* emit_asm_operand(ia32_emit_env_t *env, const ir_node *node,
        case '9':
                break;
        default:
-               ir_fprintf(stderr, "Warning: asm text (%+F) contains unknown modifier "
-                          "'%c' for asm op\n", node, c);
+               ir_fprintf(stderr,
+                               "Warning: asm text (%+F) contains unknown modifier '%c' for asm op\n",
+                               node, c);
                ++s;
                break;
        }
 
        /* parse number */
        sscanf(s, "%d%n", &num, &p);
-       if(num < 0) {
+       if (num < 0) {
                ir_fprintf(stderr, "Warning: Couldn't parse assembler operand (%+F)\n",
                           node);
                return s;
@@ -1429,52 +1389,63 @@ const char* emit_asm_operand(ia32_emit_env_t *env, const ir_node *node,
                s += p;
        }
 
+       if (num < 0 || ARR_LEN(asm_regs) <= num) {
+               ir_fprintf(stderr,
+                               "Error: Custom assembler references invalid input/output (%+F)\n",
+                               node);
+               return s;
+       }
+       asm_reg = & asm_regs[num];
+       assert(asm_reg->valid);
+
        /* get register */
-       attr   = get_ia32_attr_const(node);
-       n_outs = ARR_LEN(attr->slots);
-       if(num < n_outs) {
-               reg = get_out_reg(env, node, num);
+       if (asm_reg->use_input == 0) {
+               reg = get_out_reg(node, asm_reg->inout_pos);
        } else {
-               ir_node *pred;
-               int      in = num - n_outs;
-               if(in >= get_irn_arity(node)) {
-                       ir_fprintf(stderr, "Warning: Invalid input %d specified in asm "
-                                  "op (%+F)\n", num, node);
-                       return s;
-               }
-               pred = get_irn_n(node, in);
+               ir_node *pred = get_irn_n(node, asm_reg->inout_pos);
+
                /* might be an immediate value */
-               if(is_ia32_Immediate(pred)) {
-                       emit_ia32_Immediate(env, pred);
+               if (is_ia32_Immediate(pred)) {
+                       emit_ia32_Immediate(pred);
                        return s;
                }
-               reg = get_in_reg(env, node, in);
+               reg = get_in_reg(node, asm_reg->inout_pos);
        }
-       if(reg == NULL) {
-               ir_fprintf(stderr, "Warning: no register assigned for %d asm op "
-                          "(%+F)\n", num, node);
+       if (reg == NULL) {
+               ir_fprintf(stderr,
+                               "Warning: no register assigned for %d asm op (%+F)\n",
+                               num, node);
                return s;
        }
 
+       if (asm_reg->memory) {
+               be_emit_char('(');
+       }
+
        /* emit it */
-       be_emit_char(env, '%');
-       switch(modifier) {
-       case 0:
-               reg_name = arch_register_get_name(reg);
-               break;
-       case 'b':
-               reg_name = ia32_get_mapped_reg_name(env->isa->regs_8bit, reg);
-               break;
-       case 'h':
-               reg_name = ia32_get_mapped_reg_name(env->isa->regs_8bit_high, reg);
-               break;
-       case 'w':
-               reg_name = ia32_get_mapped_reg_name(env->isa->regs_16bit, reg);
-               break;
-       default:
-               panic("Invalid asm op modifier");
+       if (modifier != 0) {
+               be_emit_char('%');
+               switch(modifier) {
+               case 'b':
+                       reg_name = ia32_get_mapped_reg_name(isa->regs_8bit, reg);
+                       break;
+               case 'h':
+                       reg_name = ia32_get_mapped_reg_name(isa->regs_8bit_high, reg);
+                       break;
+               case 'w':
+                       reg_name = ia32_get_mapped_reg_name(isa->regs_16bit, reg);
+                       break;
+               default:
+                       panic("Invalid asm op modifier");
+               }
+               be_emit_string(reg_name);
+       } else {
+               emit_register(reg, asm_reg->mode);
+       }
+
+       if (asm_reg->memory) {
+               be_emit_char(')');
        }
-       be_emit_string(env, reg_name);
 
        return s;
 }
@@ -1482,8 +1453,7 @@ const char* emit_asm_operand(ia32_emit_env_t *env, const ir_node *node,
 /**
  * Emits code for an ASM pseudo op.
  */
-static
-void emit_ia32_Asm(ia32_emit_env_t *env, const ir_node *node)
+static void emit_ia32_Asm(const ir_node *node)
 {
        const void            *gen_attr = get_irn_generic_attr_const(node);
        const ia32_asm_attr_t *attr
@@ -1491,27 +1461,20 @@ void emit_ia32_Asm(ia32_emit_env_t *env, const ir_node *node)
        ident                 *asm_text = attr->asm_text;
        const char            *s        = get_id_str(asm_text);
 
-       be_emit_cstring(env, "# Begin ASM \t");
-       be_emit_finish_line_gas(env, node);
+       ia32_emitf(node, "#APP\t\n");
 
        if (s[0] != '\t')
-               be_emit_char(env, '\t');
+               be_emit_char('\t');
 
        while(*s != 0) {
-               if(*s == '%') {
-                       s = emit_asm_operand(env, node, s);
-                       continue;
+               if (*s == '%') {
+                       s = emit_asm_operand(node, s);
                } else {
-                       be_emit_char(env, *s);
+                       be_emit_char(*s++);
                }
-               ++s;
        }
 
-       be_emit_char(env, '\n');
-       be_emit_write_line(env);
-
-       be_emit_cstring(env, "# End ASM\n");
-       be_emit_write_line(env);
+       ia32_emitf(NULL, "\n#NO_APP\n");
 }
 
 /**********************************
@@ -1528,55 +1491,37 @@ void emit_ia32_Asm(ia32_emit_env_t *env, const ir_node *node)
 /**
  * Emit movsb/w instructions to make mov count divideable by 4
  */
-static
-void emit_CopyB_prolog(ia32_emit_env_t *env, int rem) {
-       be_emit_cstring(env, "\tcld");
-       be_emit_finish_line_gas(env, NULL);
-
-       switch(rem) {
-       case 1:
-               be_emit_cstring(env, "\tmovsb");
-               be_emit_finish_line_gas(env, NULL);
-               break;
-       case 2:
-               be_emit_cstring(env, "\tmovsw");
-               be_emit_finish_line_gas(env, NULL);
-               break;
-       case 3:
-               be_emit_cstring(env, "\tmovsb");
-               be_emit_finish_line_gas(env, NULL);
-               be_emit_cstring(env, "\tmovsw");
-               be_emit_finish_line_gas(env, NULL);
-               break;
-       }
-}
+static void emit_CopyB_prolog(unsigned size)
+{
+       if (size & 1)
+               ia32_emitf(NULL, "\tmovsb\n");
+       if (size & 2)
+               ia32_emitf(NULL, "\tmovsw\n");
+}
 
 /**
  * Emit rep movsd instruction for memcopy.
  */
-static
-void emit_ia32_CopyB(ia32_emit_env_t *env, const ir_node *node) {
-       int rem = get_ia32_pncode(node);
-
-       emit_CopyB_prolog(env, rem);
+static void emit_ia32_CopyB(const ir_node *node)
+{
+       unsigned size = get_ia32_copyb_size(node);
 
-       be_emit_cstring(env, "\trep movsd");
-       be_emit_finish_line_gas(env, node);
+       emit_CopyB_prolog(size);
+       ia32_emitf(node, "\trep movsd\n");
 }
 
 /**
  * Emits unrolled memcopy.
  */
-static
-void emit_ia32_CopyB_i(ia32_emit_env_t *env, const ir_node *node) {
-       int size = get_ia32_pncode(node);
+static void emit_ia32_CopyB_i(const ir_node *node)
+{
+       unsigned size = get_ia32_copyb_size(node);
 
-       emit_CopyB_prolog(env, size & 0x3);
+       emit_CopyB_prolog(size);
 
        size >>= 2;
        while (size--) {
-               be_emit_cstring(env, "\tmovsd");
-               be_emit_finish_line_gas(env, NULL);
+               ia32_emitf(NULL, "\tmovsd\n");
        }
 }
 
@@ -1595,136 +1540,54 @@ void emit_ia32_CopyB_i(ia32_emit_env_t *env, const ir_node *node) {
 /**
  * Emit code for conversions (I, FP), (FP, I) and (FP, FP).
  */
-static
-void emit_ia32_Conv_with_FP(ia32_emit_env_t *env, const ir_node *node) {
+static void emit_ia32_Conv_with_FP(const ir_node *node, const char* conv_f,
+               const char* conv_d)
+{
        ir_mode            *ls_mode = get_ia32_ls_mode(node);
        int                 ls_bits = get_mode_size_bits(ls_mode);
+       const char         *conv    = ls_bits == 32 ? conv_f : conv_d;
 
-       be_emit_cstring(env, "\tcvt");
-
-       if(is_ia32_Conv_I2FP(node)) {
-               if(ls_bits == 32) {
-                       be_emit_cstring(env, "si2ss");
-               } else {
-                       be_emit_cstring(env, "si2sd");
-               }
-       } else if(is_ia32_Conv_FP2I(node)) {
-               if(ls_bits == 32) {
-                       be_emit_cstring(env, "ss2si");
-               } else {
-                       be_emit_cstring(env, "sd2si");
-               }
-       } else {
-               assert(is_ia32_Conv_FP2FP(node));
-               if(ls_bits == 32) {
-                       be_emit_cstring(env, "sd2ss");
-               } else {
-                       be_emit_cstring(env, "ss2sd");
-               }
-       }
-       be_emit_char(env, ' ');
-
-       switch(get_ia32_op_type(node)) {
-               case ia32_Normal:
-                       ia32_emit_source_register(env, node, 2);
-                       be_emit_cstring(env, ", ");
-                       ia32_emit_dest_register(env, node, 0);
-                       break;
-               case ia32_AddrModeS:
-                       ia32_emit_dest_register(env, node, 0);
-                       be_emit_cstring(env, ", ");
-                       ia32_emit_am(env, node);
-                       break;
-               default:
-                       assert(0 && "unsupported op type for Conv");
-       }
-       be_emit_finish_line_gas(env, node);
+       ia32_emitf(node, "\tcvt%s %AS3, %D0\n", conv);
 }
 
-static
-void emit_ia32_Conv_I2FP(ia32_emit_env_t *env, const ir_node *node) {
-       emit_ia32_Conv_with_FP(env, node);
+static void emit_ia32_Conv_I2FP(const ir_node *node)
+{
+       emit_ia32_Conv_with_FP(node, "si2ss", "si2sd");
 }
 
-static
-void emit_ia32_Conv_FP2I(ia32_emit_env_t *env, const ir_node *node) {
-       emit_ia32_Conv_with_FP(env, node);
+static void emit_ia32_Conv_FP2I(const ir_node *node)
+{
+       emit_ia32_Conv_with_FP(node, "ss2si", "sd2si");
 }
 
-static
-void emit_ia32_Conv_FP2FP(ia32_emit_env_t *env, const ir_node *node) {
-       emit_ia32_Conv_with_FP(env, node);
+static void emit_ia32_Conv_FP2FP(const ir_node *node)
+{
+       emit_ia32_Conv_with_FP(node, "sd2ss", "ss2sd");
 }
 
 /**
  * Emits code for an Int conversion.
  */
-static
-void emit_ia32_Conv_I2I(ia32_emit_env_t *env, const ir_node *node) {
-       const char *sign_suffix;
+static void emit_ia32_Conv_I2I(const ir_node *node)
+{
        ir_mode *smaller_mode = get_ia32_ls_mode(node);
-       int smaller_bits = get_mode_size_bits(smaller_mode);
-       int signed_mode;
-       const arch_register_t *in_reg, *out_reg;
+       int      signed_mode  = mode_is_signed(smaller_mode);
+       const char *sign_suffix;
 
        assert(!mode_is_float(smaller_mode));
-       assert(smaller_bits == 8 || smaller_bits == 16 || smaller_bits == 32);
-
-       signed_mode = mode_is_signed(smaller_mode);
-       if(smaller_bits == 32) {
-               // this should not happen as it's no convert
-               assert(0);
-               sign_suffix = "";
-       } else {
-               sign_suffix = signed_mode ? "s" : "z";
-       }
-
-       switch(get_ia32_op_type(node)) {
-               case ia32_Normal:
-                       in_reg  = get_in_reg(env, node, 2);
-                       out_reg = get_out_reg(env, node, 0);
 
-                       if (in_reg  == &ia32_gp_regs[REG_EAX] &&
-                               out_reg == &ia32_gp_regs[REG_EAX] &&
-                               signed_mode &&
-                               smaller_bits == 16)
-                       {
-                               /* argument and result are both in EAX and */
-                               /* signedness is ok: -> use the smaller cwtl opcode */
-                               be_emit_cstring(env, "\tcwtl");
-                       } else {
-                               const char *sreg = ia32_get_reg_name_for_mode(env, smaller_mode, in_reg);
-
-                               be_emit_cstring(env, "\tmov");
-                               be_emit_string(env, sign_suffix);
-                               ia32_emit_mode_suffix_mode(env, smaller_mode);
-                               be_emit_cstring(env, "l %");
-                               be_emit_string(env, sreg);
-                               be_emit_cstring(env, ", ");
-                               ia32_emit_dest_register(env, node, 0);
-                       }
-                       break;
-               case ia32_AddrModeS: {
-                       be_emit_cstring(env, "\tmov");
-                       be_emit_string(env, sign_suffix);
-                       ia32_emit_mode_suffix_mode(env, smaller_mode);
-                       be_emit_cstring(env, "l ");
-                       ia32_emit_am(env, node);
-                       be_emit_cstring(env, ", ");
-                       ia32_emit_dest_register(env, node, 0);
-                       break;
-               }
-               default:
-                       assert(0 && "unsupported op type for Conv");
-       }
-       be_emit_finish_line_gas(env, node);
+       sign_suffix = signed_mode ? "s" : "z";
+       ia32_emitf(node, "\tmov%s%Ml %#AS3, %D0\n", sign_suffix);
 }
 
 /**
- * Emits code for an 8Bit Int conversion.
+ * Emits a call
  */
-void emit_ia32_Conv_I2I8Bit(ia32_emit_env_t *env, const ir_node *node) {
-       emit_ia32_Conv_I2I(env, node);
+static void emit_ia32_Call(const ir_node *node)
+{
+       /* Special case: Call must not have its immediates prefixed by $, instead
+        * address mode is prefixed by *. */
+       ia32_emitf(node, "\tcall %*AS3\n");
 }
 
 
@@ -1738,199 +1601,236 @@ void emit_ia32_Conv_I2I8Bit(ia32_emit_env_t *env, const ir_node *node) {
  *
  *******************************************/
 
-/**
- * Emits a backend call
- */
-static
-void emit_be_Call(ia32_emit_env_t *env, const ir_node *node) {
-       ir_entity *ent = be_Call_get_entity(node);
-
-       be_emit_cstring(env, "\tcall ");
-       if (ent) {
-               set_entity_backend_marked(ent, 1);
-               be_emit_string(env, get_entity_ld_name(ent));
-       } else {
-               be_emit_char(env, '*');
-               ia32_emit_dest_register(env, get_irn_n(node, be_pos_Call_ptr), 0);
-       }
-       be_emit_finish_line_gas(env, node);
-}
-
 /**
  * Emits code to increase stack pointer.
  */
-static
-void emit_be_IncSP(ia32_emit_env_t *env, const ir_node *node) {
+static void emit_be_IncSP(const ir_node *node)
+{
        int offs = be_get_IncSP_offset(node);
 
        if (offs == 0)
                return;
 
        if (offs > 0) {
-               be_emit_cstring(env, "\tsubl $");
-               be_emit_irprintf(env->emit, "%u, ", offs);
-               ia32_emit_source_register(env, node, 0);
+               ia32_emitf(node, "\tsubl $%u, %D0\n", offs);
        } else {
-               be_emit_cstring(env, "\taddl $");
-               be_emit_irprintf(env->emit, "%u, ", -offs);
-               ia32_emit_source_register(env, node, 0);
+               ia32_emitf(node, "\taddl $%u, %D0\n", -offs);
        }
-       be_emit_finish_line_gas(env, node);
 }
 
-/**
- * Emits code to set stack pointer.
- */
-static
-void emit_be_SetSP(ia32_emit_env_t *env, const ir_node *node) {
-       be_emit_cstring(env, "\tmovl ");
-       ia32_emit_source_register(env, node, 2);
-       be_emit_cstring(env, ", ");
-       ia32_emit_dest_register(env, node, 0);
-       be_emit_finish_line_gas(env, node);
+static inline bool is_unknown_reg(const arch_register_t *reg)
+{
+       if(reg == &ia32_gp_regs[REG_GP_UKNWN]
+                       || reg == &ia32_xmm_regs[REG_XMM_UKNWN]
+                       || reg == &ia32_vfp_regs[REG_VFP_UKNWN])
+               return true;
+
+       return false;
 }
 
 /**
  * Emits code for Copy/CopyKeep.
  */
-static
-void Copy_emitter(ia32_emit_env_t *env, const ir_node *node, const ir_node *op)
+static void Copy_emitter(const ir_node *node, const ir_node *op)
 {
-       const arch_env_t      *arch_env = env->arch_env;
-       const arch_register_t *in       = arch_get_irn_register(arch_env, op);
-       const arch_register_t *out      = arch_get_irn_register(arch_env, node);
-       ir_mode *mode;
+       const arch_register_t *in  = arch_get_irn_register(op);
+       const arch_register_t *out = arch_get_irn_register(node);
 
-       if(in == out) {
+       if (in == out) {
                return;
        }
-       if(is_unknown_reg(in))
+       if (is_unknown_reg(in))
                return;
        /* copies of vf nodes aren't real... */
-       if(arch_register_get_class(in) == &ia32_reg_classes[CLASS_ia32_vfp])
+       if (arch_register_get_class(in) == &ia32_reg_classes[CLASS_ia32_vfp])
                return;
 
-       mode = get_irn_mode(node);
-       if (mode == mode_E) {
-               be_emit_cstring(env, "\tmovsd ");
-               ia32_emit_register(env, in);
-               be_emit_cstring(env, ", ");
-               ia32_emit_register(env, out);
+       if (get_irn_mode(node) == mode_E) {
+               ia32_emitf(node, "\tmovsd %R, %R\n", in, out);
        } else {
-               be_emit_cstring(env, "\tmovl ");
-               ia32_emit_register(env, in);
-               be_emit_cstring(env, ", ");
-               ia32_emit_register(env, out);
+               ia32_emitf(node, "\tmovl %R, %R\n", in, out);
        }
-       be_emit_finish_line_gas(env, node);
 }
 
-static
-void emit_be_Copy(ia32_emit_env_t *env, const ir_node *node) {
-       Copy_emitter(env, node, be_get_Copy_op(node));
+static void emit_be_Copy(const ir_node *node)
+{
+       Copy_emitter(node, be_get_Copy_op(node));
 }
 
-static
-void emit_be_CopyKeep(ia32_emit_env_t *env, const ir_node *node) {
-       Copy_emitter(env, node, be_get_CopyKeep_op(node));
+static void emit_be_CopyKeep(const ir_node *node)
+{
+       Copy_emitter(node, be_get_CopyKeep_op(node));
 }
 
 /**
  * Emits code for exchange.
  */
-static
-void emit_be_Perm(ia32_emit_env_t *env, const ir_node *node) {
-       const arch_register_t *in1, *in2;
-       const arch_register_class_t *cls1, *cls2;
+static void emit_be_Perm(const ir_node *node)
+{
+       const arch_register_t *in0, *in1;
+       const arch_register_class_t *cls0, *cls1;
 
-       in1 = arch_get_irn_register(env->arch_env, get_irn_n(node, 0));
-       in2 = arch_get_irn_register(env->arch_env, get_irn_n(node, 1));
+       in0 = arch_get_irn_register(get_irn_n(node, 0));
+       in1 = arch_get_irn_register(get_irn_n(node, 1));
 
+       cls0 = arch_register_get_class(in0);
        cls1 = arch_register_get_class(in1);
-       cls2 = arch_register_get_class(in2);
-
-       assert(cls1 == cls2 && "Register class mismatch at Perm");
-
-       if (cls1 == &ia32_reg_classes[CLASS_ia32_gp]) {
-               be_emit_cstring(env, "\txchg ");
-               ia32_emit_source_register(env, node, 1);
-               be_emit_cstring(env, ", ");
-               ia32_emit_source_register(env, node, 0);
-               be_emit_finish_line_gas(env, node);
-       } else if (cls1 == &ia32_reg_classes[CLASS_ia32_xmm]) {
-               be_emit_cstring(env, "\txorpd ");
-               ia32_emit_source_register(env, node, 1);
-               be_emit_cstring(env, ", ");
-               ia32_emit_source_register(env, node, 0);
-               be_emit_finish_line_gas(env, NULL);
-
-               be_emit_cstring(env, "\txorpd ");
-               ia32_emit_source_register(env, node, 0);
-               be_emit_cstring(env, ", ");
-               ia32_emit_source_register(env, node, 1);
-               be_emit_finish_line_gas(env, NULL);
-
-               be_emit_cstring(env, "\txorpd ");
-               ia32_emit_source_register(env, node, 1);
-               be_emit_cstring(env, ", ");
-               ia32_emit_source_register(env, node, 0);
-               be_emit_finish_line_gas(env, node);
-       } else if (cls1 == &ia32_reg_classes[CLASS_ia32_vfp]) {
+
+       assert(cls0 == cls1 && "Register class mismatch at Perm");
+
+       if (cls0 == &ia32_reg_classes[CLASS_ia32_gp]) {
+               ia32_emitf(node, "\txchg %R, %R\n", in1, in0);
+       } else if (cls0 == &ia32_reg_classes[CLASS_ia32_xmm]) {
+               ia32_emitf(NULL, "\txorpd %R, %R\n", in1, in0);
+               ia32_emitf(NULL, "\txorpd %R, %R\n", in0, in1);
+               ia32_emitf(node, "\txorpd %R, %R\n", in1, in0);
+       } else if (cls0 == &ia32_reg_classes[CLASS_ia32_vfp]) {
                /* is a NOP */
-       } else if (cls1 == &ia32_reg_classes[CLASS_ia32_st]) {
+       } else if (cls0 == &ia32_reg_classes[CLASS_ia32_st]) {
                /* is a NOP */
+       } else {
+               panic("unexpected register class in be_Perm (%+F)", node);
        }
 }
 
 /**
  * Emits code for Constant loading.
  */
-static
-void emit_ia32_Const(ia32_emit_env_t *env, const ir_node *node) {
-       const ia32_immediate_attr_t *attr = get_ia32_immediate_attr_const(node);
+static void emit_ia32_Const(const ir_node *node)
+{
+       ia32_emitf(node, "\tmovl %I, %D0\n");
+}
+
+/**
+ * Emits code to load the TLS base
+ */
+static void emit_ia32_LdTls(const ir_node *node)
+{
+       ia32_emitf(node, "\tmovl %%gs:0, %D0\n");
+}
+
+/* helper function for emit_ia32_Minus64Bit */
+static void emit_mov(const ir_node* node, const arch_register_t *src, const arch_register_t *dst)
+{
+       ia32_emitf(node, "\tmovl %R, %R\n", src, dst);
+}
 
-       /* a zero? */
-       if(attr->symconst == NULL && attr->offset == 0) {
-               if (env->isa->opt_arch == arch_pentium_4) {
-                       /* P4 prefers sub r, r, others xor r, r */
-                       be_emit_cstring(env, "\tsubl ");
+/* helper function for emit_ia32_Minus64Bit */
+static void emit_neg(const ir_node* node, const arch_register_t *reg)
+{
+       ia32_emitf(node, "\tnegl %R\n", reg);
+}
+
+/* helper function for emit_ia32_Minus64Bit */
+static void emit_sbb0(const ir_node* node, const arch_register_t *reg)
+{
+       ia32_emitf(node, "\tsbbl $0, %R\n", reg);
+}
+
+/* helper function for emit_ia32_Minus64Bit */
+static void emit_sbb(const ir_node* node, const arch_register_t *src, const arch_register_t *dst)
+{
+       ia32_emitf(node, "\tsbbl %R, %R\n", src, dst);
+}
+
+/* helper function for emit_ia32_Minus64Bit */
+static void emit_xchg(const ir_node* node, const arch_register_t *src, const arch_register_t *dst)
+{
+       ia32_emitf(node, "\txchgl %R, %R\n", src, dst);
+}
+
+/* helper function for emit_ia32_Minus64Bit */
+static void emit_zero(const ir_node* node, const arch_register_t *reg)
+{
+       ia32_emitf(node, "\txorl %R, %R\n", reg, reg);
+}
+
+static void emit_ia32_Minus64Bit(const ir_node *node)
+{
+       const arch_register_t *in_lo  = get_in_reg(node, 0);
+       const arch_register_t *in_hi  = get_in_reg(node, 1);
+       const arch_register_t *out_lo = get_out_reg(node, 0);
+       const arch_register_t *out_hi = get_out_reg(node, 1);
+
+       if (out_lo == in_lo) {
+               if (out_hi != in_hi) {
+                       /* a -> a, b -> d */
+                       goto zero_neg;
                } else {
-                       be_emit_cstring(env, "\txorl ");
+                       /* a -> a, b -> b */
+                       goto normal_neg;
+               }
+       } else if (out_lo == in_hi) {
+               if (out_hi == in_lo) {
+                       /* a -> b, b -> a */
+                       emit_xchg(node, in_lo, in_hi);
+                       goto normal_neg;
+               } else {
+                       /* a -> b, b -> d */
+                       emit_mov(node, in_hi, out_hi);
+                       emit_mov(node, in_lo, out_lo);
+                       goto normal_neg;
                }
-               ia32_emit_dest_register(env, node, 0);
-               be_emit_cstring(env, ", ");
-               ia32_emit_dest_register(env, node, 0);
        } else {
-               be_emit_cstring(env, "\tmovl ");
-               emit_ia32_Immediate(env, node);
-               be_emit_cstring(env, ", ");
-               ia32_emit_dest_register(env, node, 0);
+               if (out_hi == in_lo) {
+                       /* a -> c, b -> a */
+                       emit_mov(node, in_lo, out_lo);
+                       goto zero_neg;
+               } else if (out_hi == in_hi) {
+                       /* a -> c, b -> b */
+                       emit_mov(node, in_lo, out_lo);
+                       goto normal_neg;
+               } else {
+                       /* a -> c, b -> d */
+                       emit_mov(node, in_lo, out_lo);
+                       goto zero_neg;
+               }
        }
 
-       be_emit_finish_line_gas(env, node);
+normal_neg:
+       emit_neg( node, out_hi);
+       emit_neg( node, out_lo);
+       emit_sbb0(node, out_hi);
+       return;
+
+zero_neg:
+       emit_zero(node, out_hi);
+       emit_neg( node, out_lo);
+       emit_sbb( node, in_hi, out_hi);
 }
 
-/**
- * Emits code to load the TLS base
- */
-static
-void emit_ia32_LdTls(ia32_emit_env_t *env, const ir_node *node) {
-       be_emit_cstring(env, "\tmovl %gs:0, ");
-       ia32_emit_dest_register(env, node, 0);
-       be_emit_finish_line_gas(env, node);
+static void emit_ia32_GetEIP(const ir_node *node)
+{
+       ia32_emitf(node, "\tcall %s\n", pic_base_label);
+       ia32_emitf(NULL, "%s:\n", pic_base_label);
+       ia32_emitf(node, "\tpopl %D0\n");
+}
+
+static void emit_ia32_ClimbFrame(const ir_node *node)
+{
+       const ia32_climbframe_attr_t *attr = get_ia32_climbframe_attr_const(node);
+
+       ia32_emitf(node, "\tmovl %S0, %D0\n");
+       ia32_emitf(node, "\tmovl $%u, %S1\n", attr->count);
+       ia32_emitf(NULL, BLOCK_PREFIX "%ld:\n", get_irn_node_nr(node));
+       ia32_emitf(node, "\tmovl (%D0), %D0\n");
+       ia32_emitf(node, "\tdec %S1\n");
+       ia32_emitf(node, "\tjnz " BLOCK_PREFIX "%ld\n", get_irn_node_nr(node));
 }
 
-static
-void emit_be_Return(ia32_emit_env_t *env, const ir_node *node)
+static void emit_be_Return(const ir_node *node)
 {
-       be_emit_cstring(env, "\tret");
-       be_emit_finish_line_gas(env, node);
+       unsigned pop = be_Return_get_pop(node);
+
+       if (pop > 0 || be_Return_get_emit_pop(node)) {
+               ia32_emitf(node, "\tret $%u\n", pop);
+       } else {
+               ia32_emitf(node, "\tret\n");
+       }
 }
 
-static
-void emit_Nothing(ia32_emit_env_t *env, const ir_node *node)
+static void emit_Nothing(const ir_node *node)
 {
-       (void) env;
        (void) node;
 }
 
@@ -1949,9 +1849,8 @@ void emit_Nothing(ia32_emit_env_t *env, const ir_node *node)
  * Enters the emitter functions for handled nodes into the generic
  * pointer of an opcode.
  */
-static
-void ia32_register_emitters(void) {
-
+static void ia32_register_emitters(void)
+{
 #define IA32_EMIT2(a,b) op_ia32_##a->ops.generic = (op_func)emit_ia32_##b
 #define IA32_EMIT(a)    IA32_EMIT2(a,a)
 #define EMIT(a)         op_##a->ops.generic = (op_func)emit_##a
@@ -1966,58 +1865,39 @@ void ia32_register_emitters(void) {
        ia32_register_spec_emitters();
 
        /* other ia32 emitter functions */
+       IA32_EMIT2(Conv_I2I8Bit, Conv_I2I);
        IA32_EMIT(Asm);
-       IA32_EMIT(CmpJmp);
-       IA32_EMIT2(CmpJmp8Bit, CmpJmp);
-       IA32_EMIT(TestJmp);
-       IA32_EMIT2(TestJmp8Bit, TestJmp);
-       IA32_EMIT(CmpCMov);
-       IA32_EMIT2(CmpCMov8Bit, CmpCMov);
-       IA32_EMIT(TestCMov);
-       IA32_EMIT2(TestCMov8Bit, TestCMov);
-       IA32_EMIT(CmpSet);
-       IA32_EMIT2(CmpSet8Bit, CmpSet);
-       IA32_EMIT(TestSet);
-       IA32_EMIT2(TestSet8Bit, TestSet);
-       IA32_EMIT(SwitchJmp);
-       IA32_EMIT(CopyB);
-       IA32_EMIT(CopyB_i);
-       IA32_EMIT(Conv_I2FP);
-       IA32_EMIT(Conv_FP2I);
+       IA32_EMIT(CMov);
+       IA32_EMIT(Call);
+       IA32_EMIT(Const);
        IA32_EMIT(Conv_FP2FP);
+       IA32_EMIT(Conv_FP2I);
+       IA32_EMIT(Conv_I2FP);
        IA32_EMIT(Conv_I2I);
-       IA32_EMIT(Conv_I2I8Bit);
-       IA32_EMIT(Const);
+       IA32_EMIT(CopyB);
+       IA32_EMIT(CopyB_i);
+       IA32_EMIT(GetEIP);
+       IA32_EMIT(IMul);
+       IA32_EMIT(Jcc);
        IA32_EMIT(LdTls);
-       IA32_EMIT(xCmp);
-       IA32_EMIT(xCmpSet);
-       IA32_EMIT(xCmpCMov);
-       IA32_EMIT(xCmpJmp);
-       IA32_EMIT2(fcomJmp, x87CmpJmp);
-       IA32_EMIT2(fcompJmp, x87CmpJmp);
-       IA32_EMIT2(fcomppJmp, x87CmpJmp);
-       IA32_EMIT2(fcomrJmp, x87CmpJmp);
-       IA32_EMIT2(fcomrpJmp, x87CmpJmp);
-       IA32_EMIT2(fcomrppJmp, x87CmpJmp);
+       IA32_EMIT(Minus64Bit);
+       IA32_EMIT(SwitchJmp);
+       IA32_EMIT(ClimbFrame);
+       IA32_EMIT(Jmp);
 
        /* benode emitter */
-       BE_EMIT(Call);
-       BE_EMIT(IncSP);
-       BE_EMIT(SetSP);
        BE_EMIT(Copy);
        BE_EMIT(CopyKeep);
+       BE_EMIT(IncSP);
        BE_EMIT(Perm);
        BE_EMIT(Return);
 
-       BE_IGN(RegParams);
        BE_IGN(Barrier);
        BE_IGN(Keep);
+       BE_IGN(Start);
 
        /* firm emitter */
-       EMIT(Jmp);
-       IGN(Proj);
        IGN(Phi);
-       IGN(Start);
 
 #undef BE_EMIT
 #undef EMIT
@@ -2026,58 +1906,59 @@ void ia32_register_emitters(void) {
 #undef IA32_EMIT
 }
 
-static const char *last_name = NULL;
-static unsigned last_line = -1;
-static unsigned num = -1;
+typedef void (*emit_func_ptr) (const ir_node *);
 
 /**
- * Emit the debug support for node node.
+ * Assign and emit an exception label if the current instruction can fail.
  */
-static
-void ia32_emit_dbg(ia32_emit_env_t *env, const ir_node *node) {
-       dbg_info *db = get_irn_dbg_info(node);
-       unsigned lineno;
-       const char *fname = be_retrieve_dbg_info(db, &lineno);
-
-       if (! env->cg->birg->main_env->options->stabs_debug_support)
-               return;
-
-       if (fname) {
-               if (last_name != fname) {
-                       last_line = -1;
-                       be_dbg_include_begin(env->cg->birg->main_env->db_handle, fname);
-                       last_name = fname;
-               }
-               if (last_line != lineno) {
-                       char name[64];
-
-                       snprintf(name, sizeof(name), ".LM%u", ++num);
-                       last_line = lineno;
-                       be_dbg_line(env->cg->birg->main_env->db_handle, lineno, name);
-                       be_emit_string(env, name);
-                       be_emit_cstring(env, ":\n");
-                       be_emit_write_line(env);
-               }
-       }
+static void ia32_assign_exc_label(ir_node *node)
+{
+       /* assign a new ID to the instruction */
+       set_ia32_exc_label_id(node, ++exc_label_id);
+       /* print it */
+       ia32_emit_exc_label(node);
+       be_emit_char(':');
+       be_emit_pad_comment();
+       be_emit_cstring("/* exception to Block ");
+       ia32_emit_cfop_target(node);
+       be_emit_cstring(" */\n");
+       be_emit_write_line();
 }
 
-typedef void (*emit_func_ptr) (ia32_emit_env_t *, const ir_node *);
-
 /**
  * Emits code for a node.
  */
-static
-void ia32_emit_node(ia32_emit_env_t *env, const ir_node *node) {
+static void ia32_emit_node(ir_node *node)
+{
        ir_op *op = get_irn_op(node);
 
        DBG((dbg, LEVEL_1, "emitting code for %+F\n", node));
 
+       if (is_ia32_irn(node)) {
+               if (get_ia32_exc_label(node)) {
+                       /* emit the exception label of this instruction */
+                       ia32_assign_exc_label(node);
+               }
+               if (mark_spill_reload) {
+                       if (is_ia32_is_spill(node)) {
+                               ia32_emitf(NULL, "\txchg %ebx, %ebx        /* spill mark */\n");
+                       }
+                       if (is_ia32_is_reload(node)) {
+                               ia32_emitf(NULL, "\txchg %edx, %edx        /* reload mark */\n");
+                       }
+                       if (is_ia32_is_remat(node)) {
+                               ia32_emitf(NULL, "\txchg %ecx, %ecx        /* remat mark */\n");
+                       }
+               }
+       }
        if (op->ops.generic) {
                emit_func_ptr func = (emit_func_ptr) op->ops.generic;
-               ia32_emit_dbg(env, node);
-               (*func) (env, node);
+
+               be_dbg_set_dbg_info(get_irn_dbg_info(node));
+
+               (*func) (node);
        } else {
-               emit_Nothing(env, node);
+               emit_Nothing(node);
                ir_fprintf(stderr, "Error: No emit handler for node %+F (%+G, graph %+F)\n", node, node, current_ir_graph);
                abort();
        }
@@ -2086,278 +1967,1284 @@ void ia32_emit_node(ia32_emit_env_t *env, const ir_node *node) {
 /**
  * Emits gas alignment directives
  */
-static
-void ia32_emit_alignment(ia32_emit_env_t *env, unsigned align, unsigned skip) {
-       be_emit_cstring(env, "\t.p2align ");
-       be_emit_irprintf(env->emit, "%u,,%u\n", align, skip);
-       be_emit_write_line(env);
-}
-
-/**
- * Emits gas alignment directives for Functions depended on cpu architecture.
- */
-static
-void ia32_emit_align_func(ia32_emit_env_t *env, cpu_support cpu) {
-       unsigned align;
-       unsigned maximum_skip;
-
-       switch (cpu) {
-               case arch_i386:
-                       align = 2;
-                       break;
-               case arch_i486:
-                       align = 4;
-                       break;
-               case arch_k6:
-                       align = 5;
-                       break;
-               default:
-                       align = 4;
-       }
-       maximum_skip = (1 << align) - 1;
-       ia32_emit_alignment(env, align, maximum_skip);
+static void ia32_emit_alignment(unsigned align, unsigned skip)
+{
+       ia32_emitf(NULL, "\t.p2align %u,,%u\n", align, skip);
 }
 
 /**
  * Emits gas alignment directives for Labels depended on cpu architecture.
  */
-static
-void ia32_emit_align_label(ia32_emit_env_t *env, cpu_support cpu) {
-       unsigned align; unsigned maximum_skip;
-
-       switch (cpu) {
-               case arch_i386:
-                       align = 2;
-                       break;
-               case arch_i486:
-                       align = 4;
-                       break;
-               case arch_k6:
-                       align = 5;
-                       break;
-               default:
-                       align = 4;
-       }
-       maximum_skip = (1 << align) - 1;
-       ia32_emit_alignment(env, align, maximum_skip);
+static void ia32_emit_align_label(void)
+{
+       unsigned align        = ia32_cg_config.label_alignment;
+       unsigned maximum_skip = ia32_cg_config.label_alignment_max_skip;
+       ia32_emit_alignment(align, maximum_skip);
 }
 
 /**
- * Test wether a block should be aligned.
- * For cpus in the P4/Athlon class it is usefull to align jump labels to
+ * Test whether a block should be aligned.
+ * For cpus in the P4/Athlon class it is useful to align jump labels to
  * 16 bytes. However we should only do that if the alignment nops before the
  * label aren't executed more often than we have jumps to the label.
  */
-static
-int should_align_block(ia32_emit_env_t *env, ir_node *block, ir_node *prev) {
+static int should_align_block(const ir_node *block)
+{
        static const double DELTA = .0001;
-       ir_exec_freq *exec_freq = env->cg->birg->exec_freq;
+       ir_exec_freq *exec_freq   = cg->birg->exec_freq;
+       ir_node      *prev        = get_prev_block_sched(block);
        double        block_freq;
        double        prev_freq = 0;  /**< execfreq of the fallthrough block */
        double        jmp_freq  = 0;  /**< execfreq of all non-fallthrough blocks */
-       cpu_support   cpu       = env->isa->opt_arch;
        int           i, n_cfgpreds;
 
-       if(exec_freq == NULL)
+       if (exec_freq == NULL)
                return 0;
-       if(cpu == arch_i386 || cpu == arch_i486)
+       if (ia32_cg_config.label_alignment_factor <= 0)
                return 0;
 
        block_freq = get_block_execfreq(exec_freq, block);
-       if(block_freq < DELTA)
+       if (block_freq < DELTA)
                return 0;
 
        n_cfgpreds = get_Block_n_cfgpreds(block);
        for(i = 0; i < n_cfgpreds; ++i) {
-               ir_node *pred      = get_Block_cfgpred_block(block, i);
-               double   pred_freq = get_block_execfreq(exec_freq, pred);
+               const ir_node *pred      = get_Block_cfgpred_block(block, i);
+               double         pred_freq = get_block_execfreq(exec_freq, pred);
 
-               if(pred == prev) {
+               if (pred == prev) {
                        prev_freq += pred_freq;
                } else {
                        jmp_freq  += pred_freq;
                }
        }
 
-       if(prev_freq < DELTA && !(jmp_freq < DELTA))
+       if (prev_freq < DELTA && !(jmp_freq < DELTA))
                return 1;
 
        jmp_freq /= prev_freq;
 
-       switch (cpu) {
-               case arch_athlon:
-               case arch_athlon_64:
-               case arch_k6:
-                       return jmp_freq > 3;
-               default:
-                       return jmp_freq > 2;
-       }
+       return jmp_freq > ia32_cg_config.label_alignment_factor;
 }
 
-static
-void ia32_emit_block_header(ia32_emit_env_t *env, ir_node *block, ir_node *prev)
+/**
+ * Emit the block header for a block.
+ *
+ * @param block       the block
+ * @param prev_block  the previous block
+ */
+static void ia32_emit_block_header(ir_node *block)
 {
-       int           n_cfgpreds;
-       int           need_label;
+       ir_graph     *irg = current_ir_graph;
+       int           need_label = block_needs_label(block);
        int           i, arity;
-       ir_exec_freq  *exec_freq = env->cg->birg->exec_freq;
+       ir_exec_freq *exec_freq = cg->birg->exec_freq;
 
-       n_cfgpreds = get_Block_n_cfgpreds(block);
-       need_label = (n_cfgpreds != 0);
+       if (block == get_irg_end_block(irg))
+               return;
+
+       if (ia32_cg_config.label_alignment > 0) {
+               /* align the current block if:
+                * a) if should be aligned due to its execution frequency
+                * b) there is no fall-through here
+                */
+               if (should_align_block(block)) {
+                       ia32_emit_align_label();
+               } else {
+                       /* if the predecessor block has no fall-through,
+                          we can always align the label. */
+                       int i;
+                       int has_fallthrough = 0;
+
+                       for (i = get_Block_n_cfgpreds(block) - 1; i >= 0; --i) {
+                               ir_node *cfg_pred = get_Block_cfgpred(block, i);
+                               if (can_be_fallthrough(cfg_pred)) {
+                                       has_fallthrough = 1;
+                                       break;
+                               }
+                       }
 
-       if (should_align_block(env, block, prev)) {
-               assert(need_label);
-               ia32_emit_align_label(env, env->isa->opt_arch);
+                       if (!has_fallthrough)
+                               ia32_emit_align_label();
+               }
        }
 
-       if(need_label) {
-               ia32_emit_block_name(env, block);
-               be_emit_char(env, ':');
+       if (need_label) {
+               ia32_emit_block_name(block);
+               be_emit_char(':');
+
+               be_emit_pad_comment();
+               be_emit_cstring("   /* ");
+       } else {
+               be_emit_cstring("\t/* ");
+               ia32_emit_block_name(block);
+               be_emit_cstring(": ");
+       }
 
-               be_emit_pad_comment(env);
-               be_emit_cstring(env, "   /* preds:");
+       be_emit_cstring("preds:");
 
-               /* emit list of pred blocks in comment */
-               arity = get_irn_arity(block);
+       /* emit list of pred blocks in comment */
+       arity = get_irn_arity(block);
+       if (arity <= 0) {
+               be_emit_cstring(" none");
+       } else {
                for (i = 0; i < arity; ++i) {
                        ir_node *predblock = get_Block_cfgpred_block(block, i);
-                       be_emit_irprintf(env->emit, " %d", get_irn_node_nr(predblock));
+                       be_emit_irprintf(" %d", get_irn_node_nr(predblock));
                }
-       } else {
-               be_emit_cstring(env, "\t/* ");
-               ia32_emit_block_name(env, block);
-               be_emit_cstring(env, ": ");
        }
        if (exec_freq != NULL) {
-               be_emit_irprintf(env->emit, " freq: %f",
+               be_emit_irprintf(", freq: %f",
                                 get_block_execfreq(exec_freq, block));
        }
-       be_emit_cstring(env, " */\n");
-       be_emit_write_line(env);
+       be_emit_cstring(" */\n");
+       be_emit_write_line();
 }
 
 /**
  * Walks over the nodes in a block connected by scheduling edges
  * and emits code for each node.
  */
-static
-void ia32_gen_block(ia32_emit_env_t *env, ir_node *block, ir_node *last_block)
+static void ia32_gen_block(ir_node *block)
 {
-       const ir_node *node;
+       ir_node *node;
 
-       ia32_emit_block_header(env, block, last_block);
+       ia32_emit_block_header(block);
 
        /* emit the contents of the block */
-       ia32_emit_dbg(env, block);
+       be_dbg_set_dbg_info(get_irn_dbg_info(block));
        sched_foreach(block, node) {
-               ia32_emit_node(env, node);
-       }
-}
-
-/**
- * Emits code for function start.
- */
-static
-void ia32_emit_func_prolog(ia32_emit_env_t *env, ir_graph *irg) {
-       ir_entity  *irg_ent  = get_irg_entity(irg);
-       const char *irg_name = get_entity_ld_name(irg_ent);
-       cpu_support cpu      = env->isa->opt_arch;
-       const be_irg_t *birg = env->cg->birg;
-
-       be_emit_write_line(env);
-       be_gas_emit_switch_section(env->emit, GAS_SECTION_TEXT);
-       be_dbg_method_begin(birg->main_env->db_handle, irg_ent, be_abi_get_stack_layout(birg->abi));
-       ia32_emit_align_func(env, cpu);
-       if (get_entity_visibility(irg_ent) == visibility_external_visible) {
-               be_emit_cstring(env, ".global ");
-               be_emit_string(env, irg_name);
-               be_emit_char(env, '\n');
-               be_emit_write_line(env);
+               ia32_emit_node(node);
        }
-       ia32_emit_function_object(env, irg_name);
-       be_emit_string(env, irg_name);
-       be_emit_cstring(env, ":\n");
-       be_emit_write_line(env);
 }
 
-/**
- * Emits code for function end
- */
-static
-void ia32_emit_func_epilog(ia32_emit_env_t *env, ir_graph *irg) {
-       const char *irg_name = get_entity_ld_name(get_irg_entity(irg));
-       const be_irg_t *birg = env->cg->birg;
-
-       ia32_emit_function_size(env, irg_name);
-       be_dbg_method_end(birg->main_env->db_handle);
-       be_emit_char(env, '\n');
-       be_emit_write_line(env);
-}
+typedef struct exc_entry {
+       ir_node *exc_instr;  /** The instruction that can issue an exception. */
+       ir_node *block;      /** The block to call then. */
+} exc_entry;
 
 /**
  * Block-walker:
- * Sets labels for control flow nodes (jump target)
+ * Sets labels for control flow nodes (jump target).
+ * Links control predecessors to there destination blocks.
  */
-static
-void ia32_gen_labels(ir_node *block, void *data)
+static void ia32_gen_labels(ir_node *block, void *data)
 {
+       exc_entry **exc_list = data;
        ir_node *pred;
-       int n = get_Block_n_cfgpreds(block);
-       (void) data;
+       int     n;
 
-       for (n--; n >= 0; n--) {
+       for (n = get_Block_n_cfgpreds(block) - 1; n >= 0; --n) {
                pred = get_Block_cfgpred(block, n);
                set_irn_link(pred, block);
+
+               pred = skip_Proj(pred);
+               if (is_ia32_irn(pred) && get_ia32_exc_label(pred)) {
+                       exc_entry e;
+
+                       e.exc_instr = pred;
+                       e.block     = block;
+                       ARR_APP1(exc_entry, *exc_list, e);
+                       set_irn_link(pred, block);
+               }
        }
 }
 
 /**
- * Emit an exception label if the current instruction can fail.
+ * Compare two exception_entries.
  */
-void ia32_emit_exc_label(ia32_emit_env_t *env, const ir_node *node) {
-       if (get_ia32_exc_label(node)) {
-               be_emit_irprintf(env->emit, ".EXL%u\n", 0);
-               be_emit_write_line(env);
-       }
+static int cmp_exc_entry(const void *a, const void *b)
+{
+       const exc_entry *ea = a;
+       const exc_entry *eb = b;
+
+       if (get_ia32_exc_label_id(ea->exc_instr) < get_ia32_exc_label_id(eb->exc_instr))
+               return -1;
+       return +1;
 }
 
 /**
  * Main driver. Emits the code for one routine.
  */
-void ia32_gen_routine(ia32_code_gen_t *cg, ir_graph *irg) {
-       ia32_emit_env_t env;
-       ir_node *block;
-       ir_node *last_block = NULL;
+void ia32_gen_routine(ia32_code_gen_t *ia32_cg, ir_graph *irg)
+{
+       ir_entity *entity     = get_irg_entity(irg);
+       exc_entry *exc_list   = NEW_ARR_F(exc_entry, 0);
        int i, n;
 
-       env.isa      = (ia32_isa_t *)cg->arch_env->isa;
-       env.emit     = &env.isa->emit;
-       env.arch_env = cg->arch_env;
-       env.cg       = cg;
+       cg       = ia32_cg;
+       isa      = cg->isa;
+       do_pic   = cg->birg->main_env->options->pic;
 
        ia32_register_emitters();
 
-       ia32_emit_func_prolog(&env, irg);
-       irg_block_walk_graph(irg, ia32_gen_labels, NULL, &env);
+       get_unique_label(pic_base_label, sizeof(pic_base_label), ".PIC_BASE");
+
+       be_dbg_method_begin(entity, be_abi_get_stack_layout(cg->birg->abi));
+       be_gas_emit_function_prolog(entity, ia32_cg_config.function_alignment);
 
+       /* we use links to point to target blocks */
+       ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK);
+       irg_block_walk_graph(irg, ia32_gen_labels, NULL, &exc_list);
+
+       /* initialize next block links */
        n = ARR_LEN(cg->blk_sched);
-       for (i = 0; i < n;) {
-               ir_node *next_bl;
+       for (i = 0; i < n; ++i) {
+               ir_node *block = cg->blk_sched[i];
+               ir_node *prev  = i > 0 ? cg->blk_sched[i-1] : NULL;
+
+               set_irn_link(block, prev);
+       }
+
+       for (i = 0; i < n; ++i) {
+               ir_node *block = cg->blk_sched[i];
+
+               ia32_gen_block(block);
+       }
+
+       be_gas_emit_function_epilog(entity);
+       be_dbg_method_end();
+       be_emit_char('\n');
+       be_emit_write_line();
+
+       ir_free_resources(irg, IR_RESOURCE_IRN_LINK);
 
-               block   = cg->blk_sched[i];
-               ++i;
-               next_bl = i < n ? cg->blk_sched[i] : NULL;
+       /* Sort the exception table using the exception label id's.
+          Those are ascending with ascending addresses. */
+       qsort(exc_list, ARR_LEN(exc_list), sizeof(exc_list[0]), cmp_exc_entry);
+       {
+               int i;
 
-               /* set here the link. the emitter expects to find the next block here */
-               set_irn_link(block, next_bl);
-               ia32_gen_block(&env, block, last_block);
-               last_block = block;
+               for (i = 0; i < ARR_LEN(exc_list); ++i) {
+                       be_emit_cstring("\t.long ");
+                       ia32_emit_exc_label(exc_list[i].exc_instr);
+                       be_emit_char('\n');
+                       be_emit_cstring("\t.long ");
+                       ia32_emit_block_name(exc_list[i].block);
+                       be_emit_char('\n');
+               }
        }
+       DEL_ARR_F(exc_list);
+}
+
+static const lc_opt_table_entry_t ia32_emitter_options[] = {
+       LC_OPT_ENT_BOOL("mark_spill_reload",   "mark spills and reloads with ud opcodes", &mark_spill_reload),
+       LC_OPT_LAST
+};
+
+/* ==== Experimental binary emitter ==== */
+
+static unsigned char reg_gp_map[N_ia32_gp_REGS];
+static unsigned char reg_mmx_map[N_ia32_mmx_REGS];
+static unsigned char reg_sse_map[N_ia32_xmm_REGS];
+static unsigned char pnc_map_signed[8];
+static unsigned char pnc_map_unsigned[8];
+
+static void build_reg_map(void)
+{
+       reg_gp_map[REG_EAX] = 0x0;
+       reg_gp_map[REG_ECX] = 0x1;
+       reg_gp_map[REG_EDX] = 0x2;
+       reg_gp_map[REG_EBX] = 0x3;
+       reg_gp_map[REG_ESP] = 0x4;
+       reg_gp_map[REG_EBP] = 0x5;
+       reg_gp_map[REG_ESI] = 0x6;
+       reg_gp_map[REG_EDI] = 0x7;
+
+       pnc_map_signed[pn_Cmp_False] = 0xFF;
+       pnc_map_signed[pn_Cmp_Eq]    = 0x04;
+       pnc_map_signed[pn_Cmp_Lt]    = 0x0C;
+       pnc_map_signed[pn_Cmp_Le]    = 0x0E;
+       pnc_map_signed[pn_Cmp_Gt]    = 0x0F;
+       pnc_map_signed[pn_Cmp_Ge]    = 0x0D;
+       pnc_map_signed[pn_Cmp_Lg]    = 0x05;
+
+       pnc_map_unsigned[pn_Cmp_False] = 0xFF;
+       pnc_map_unsigned[pn_Cmp_Eq]    = 0x04;
+       pnc_map_unsigned[pn_Cmp_Lt]    = 0x02;
+       pnc_map_unsigned[pn_Cmp_Le]    = 0x06;
+       pnc_map_unsigned[pn_Cmp_Gt]    = 0x07;
+       pnc_map_unsigned[pn_Cmp_Ge]    = 0x03;
+       pnc_map_unsigned[pn_Cmp_Lg]    = 0x05;
+}
+
+#define GET_MODE(code) ((code) & 0xC0)
+
+/** Sign extension bit values for binops */
+enum SignExt {
+       UNSIGNED_IMM = 0,  /**< unsigned immediate */
+       SIGNEXT_IMM  = 2,  /**< sign extended immediate */
+};
+
+/** The mod encoding of the ModR/M */
+enum Mod {
+       MOD_IND          = 0x00, /**< [reg1] */
+       MOD_IND_BYTE_OFS = 0x40, /**< [reg1 + byte ofs] */
+       MOD_IND_WORD_OFS = 0x80, /**< [reg1 + word ofs] */
+       MOD_REG          = 0xC0  /**< reg1 */
+};
+
+/** create R/M encoding for ModR/M */
+#define ENC_RM(x) (x)
+/** create REG encoding for ModR/M */
+#define ENC_REG(x) ((x) << 3)
 
-       ia32_emit_func_epilog(&env, irg);
+/** create encoding for a SIB byte */
+#define ENC_SIB(scale, index, base) ((scale) << 6 | (index) << 3 | (base))
+
+/* Node: The following routines are supposed to append bytes, words, dwords
+   to the output stream.
+   Currently the implementation is stupid in that it still creates output
+   for an "assembler" in the form of .byte, .long
+   We will change this when enough infrastructure is there to create complete
+   machine code in memory/object files */
+
+static void bemit8(const unsigned char byte)
+{
+       be_emit_irprintf("\t.byte 0x%x\n", byte);
+       be_emit_write_line();
 }
 
-void ia32_init_emitter(void)
+static void bemit16(const unsigned u16)
 {
+       be_emit_irprintf("\t.word 0x%x\n", u16);
+       be_emit_write_line();
+}
+
+static void bemit32(const unsigned u32)
+{
+       be_emit_irprintf("\t.long 0x%x\n", u32);
+       be_emit_write_line();
+}
+
+/**
+ * Emit address of an entity. If @p is_relative is true then a relative
+ * offset from behind the address to the entity is created.
+ */
+static void bemit_entity(ir_entity *entity, bool entity_sign, int offset,
+                         bool is_relative)
+{
+       if (entity == NULL) {
+               bemit32(offset);
+               return;
+       }
+
+       if (is_relative) {
+               offset -= 4;
+       }
+
+       /* the final version should remember the position in the bytestream
+          and patch it with the correct address at linktime... */
+       be_emit_cstring("\t.long ");
+       if (entity_sign)
+               be_emit_char('-');
+       set_entity_backend_marked(entity, 1);
+       be_gas_emit_entity(entity);
+
+       if (is_relative) {
+               be_emit_cstring("-.");
+       }
+
+       if (offset != 0) {
+               be_emit_irprintf("%+d", offset);
+       }
+       be_emit_char('\n');
+       be_emit_write_line();
+}
+
+static void bemit_jmp_destination(const ir_node *dest_block)
+{
+       be_emit_cstring("\t.long ");
+       ia32_emit_block_name(dest_block);
+       be_emit_cstring(" - . - 4\n");
+       be_emit_write_line();
+}
+
+/* end emit routines, all emitters following here should only use the functions
+   above. */
+
+/** Create a ModR/M byte for src1,src2 registers */
+static void bemit_modrr(const arch_register_t *src1,
+                        const arch_register_t *src2)
+{
+       unsigned char modrm = MOD_REG;
+       modrm |= ENC_RM(reg_gp_map[src1->index]);
+       modrm |= ENC_REG(reg_gp_map[src2->index]);
+       bemit8(modrm);
+}
+
+/** Create a ModR/M byte for one register and extension */
+static void bemit_modru(const arch_register_t *reg, unsigned ext)
+{
+       unsigned char modrm = MOD_REG;
+       assert(ext <= 7);
+       modrm |= ENC_RM(reg_gp_map[reg->index]);
+       modrm |= ENC_REG(ext);
+       bemit8(modrm);
+}
+
+/**
+ * Calculate the size of an (unsigned) immediate in bytes.
+ *
+ * @param offset  an offset
+ */
+static unsigned get_unsigned_imm_size(unsigned offset)
+{
+       if (offset < 256) {
+               return 1;
+       } else if (offset < 65536) {
+               return 2;
+       } else {
+               return 4;
+       }
+}
+
+/**
+ * Calculate the size of an signed immediate in bytes.
+ *
+ * @param offset  an offset
+ */
+static unsigned get_signed_imm_size(int offset)
+{
+       if (-128 <= offset && offset < 128) {
+               return 1;
+       } else if (-32768 <= offset && offset < 32768) {
+               return 2;
+       } else {
+               return 4;
+       }
+}
+
+/**
+ * Emit an address mode.
+ *
+ * @param reg   content of the reg field: either a register index or an opcode extension
+ * @param node  the node
+ */
+static void bemit_mod_am(unsigned reg, const ir_node *node)
+{
+       ir_entity *ent       = get_ia32_am_sc(node);
+       int        offs      = get_ia32_am_offs_int(node);
+       ir_node   *base      = get_irn_n(node, n_ia32_base);
+       int        has_base  = !is_ia32_NoReg_GP(base);
+       ir_node   *index     = get_irn_n(node, n_ia32_index);
+       int        has_index = !is_ia32_NoReg_GP(index);
+       unsigned   modrm     = 0;
+       unsigned   sib       = 0;
+       unsigned   emitoffs  = 0;
+       bool       emitsib   = false;
+       unsigned   base_enc;
+
+       /* set the mod part depending on displacement */
+       if (ent != NULL) {
+               modrm |= MOD_IND_WORD_OFS;
+               emitoffs = 32;
+       } else if (offs == 0) {
+               modrm |= MOD_IND;
+               emitoffs = 0;
+       } else if (-128 <= offs && offs < 128) {
+               modrm |= MOD_IND_BYTE_OFS;
+               emitoffs = 8;
+       } else {
+               modrm |= MOD_IND_WORD_OFS;
+               emitoffs = 32;
+       }
+
+       if (has_base) {
+               const arch_register_t *base_reg = arch_get_irn_register(base);
+               base_enc = reg_gp_map[base_reg->index];
+       } else {
+               /* Use the EBP encoding + MOD_IND if NO base register. There is
+                * always a 32bit offset present in this case. */
+               modrm    = MOD_IND;
+               base_enc = 0x05;
+               emitoffs = 32;
+       }
+
+       /* Determine if we need a SIB byte. */
+       if (has_index) {
+               const arch_register_t *reg_index = arch_get_irn_register(index);
+               int                    scale     = get_ia32_am_scale(node);
+               assert(scale < 4);
+               /* R/M set to ESP means SIB in 32bit mode. */
+               modrm   |= ENC_RM(0x04);
+               sib      = ENC_SIB(scale, reg_gp_map[reg_index->index], base_enc);
+               emitsib = true;
+       } else if (base_enc == 0x04) {
+               /* for the above reason we are forced to emit a SIB when base is ESP.
+                * Only the base is used, index must be ESP too, which means no index.
+                */
+               modrm   |= ENC_RM(0x04);
+               sib      = ENC_SIB(0, 0x04, 0x04);
+               emitsib  = true;
+       } else {
+               modrm |= ENC_RM(base_enc);
+       }
+
+       /* We are forced to emit an 8bit offset as EBP base without offset is a
+        * special case for SIB without base register. */
+       if (base_enc == 0x05 && emitoffs == 0) {
+               modrm    |= MOD_IND_BYTE_OFS;
+               emitoffs  = 8;
+       }
+
+       modrm |= ENC_REG(reg);
+
+       bemit8(modrm);
+       if (emitsib)
+               bemit8(sib);
+
+       /* emit displacement */
+       if (emitoffs == 8) {
+               bemit8((unsigned) offs);
+       } else if (emitoffs == 32) {
+               bemit_entity(ent, is_ia32_am_sc_sign(node), offs, false);
+       }
+}
+
+/**
+ * Emit a binop with a immediate operand.
+ *
+ * @param node        the node to emit
+ * @param opcode_eax  the opcode for the op eax, imm variant
+ * @param opcode      the opcode for the reg, imm variant
+ * @param ruval       the opcode extension for opcode
+ */
+static void bemit_binop_with_imm(
+       const ir_node *node,
+       unsigned char opcode_ax,
+       unsigned char opcode, unsigned char ruval)
+{
+       /* Use in-reg, because some instructions (cmp, test) have no out-reg. */
+       const ir_node               *op   = get_irn_n(node, n_ia32_binary_right);
+       const ia32_immediate_attr_t *attr = get_ia32_immediate_attr_const(op);
+       unsigned                     size;
+
+       /* Some instructions (test) have no short form with 32bit value + 8bit
+        * immediate. */
+       if (attr->symconst != NULL || opcode & SIGNEXT_IMM) {
+               size = 4;
+       } else {
+               /* check for sign extension */
+               size = get_signed_imm_size(attr->offset);
+       }
+
+       switch (size) {
+       case 1:
+               bemit8(opcode | SIGNEXT_IMM);
+               /* cmp has this special mode */
+               if (get_ia32_op_type(node) == ia32_AddrModeS) {
+                       bemit_mod_am(ruval, node);
+               } else {
+                       const arch_register_t *reg = get_in_reg(node, n_ia32_binary_left);
+                       bemit_modru(reg, ruval);
+               }
+               bemit8((unsigned char)attr->offset);
+               return;
+       case 2:
+       case 4:
+               /* check for eax variant: this variant is shorter for 32bit immediates only */
+               if (get_ia32_op_type(node) == ia32_AddrModeS) {
+                       bemit8(opcode);
+                       bemit_mod_am(ruval, node);
+               } else {
+                       const arch_register_t *reg = get_in_reg(node, n_ia32_binary_left);
+                       if (reg->index == REG_EAX) {
+                               bemit8(opcode_ax);
+                       } else {
+                               bemit8(opcode);
+                               bemit_modru(reg, ruval);
+                       }
+               }
+               bemit_entity(attr->symconst, attr->sc_sign, attr->offset, false);
+               return;
+       }
+       panic("invalid imm size?!?");
+}
+
+/**
+ * Emits a binop.
+ */
+static void bemit_binop_2(const ir_node *node, unsigned code)
+{
+       const arch_register_t *out    = get_in_reg(node, n_ia32_binary_left);
+       ia32_op_type_t        am_type = get_ia32_op_type(node);
+       unsigned char         d       = 0;
+       const arch_register_t *op2;
+
+       switch (am_type) {
+       case ia32_AddrModeS:
+               d = 2;
+               /* FALLTHROUGH */
+       case ia32_AddrModeD:
+               bemit8(code | d);
+               bemit_mod_am(reg_gp_map[out->index], node);
+               return;
+       case ia32_Normal:
+               bemit8(code);
+               op2 = get_in_reg(node, n_ia32_binary_right);
+               bemit_modrr(out, op2);
+               return;
+       }
+       panic("invalid address mode");
+}
+
+/**
+ * Emit a binop.
+ */
+static void bemit_binop(const ir_node *node, const unsigned char opcodes[4])
+{
+       ir_node *right = get_irn_n(node, n_ia32_binary_right);
+       if (is_ia32_Immediate(right)) {
+               bemit_binop_with_imm(node, opcodes[1], opcodes[2], opcodes[3]);
+       } else {
+               bemit_binop_2(node, opcodes[0]);
+       }
+}
+
+/**
+ * Emit an unop.
+ */
+static void bemit_unop(const ir_node *node, unsigned char code, unsigned char ext, int input)
+{
+       bemit8(code);
+       if (get_ia32_op_type(node) == ia32_Normal) {
+               const arch_register_t *in = get_in_reg(node, input);
+               bemit_modru(in, ext);
+       } else {
+               bemit_mod_am(ext, node);
+       }
+}
+
+static void bemit_unop_reg(const ir_node *node, unsigned char code, int input)
+{
+       const arch_register_t *out = get_out_reg(node, 0);
+       bemit_unop(node, code, reg_gp_map[out->index], input);
+}
+
+static void bemit_immediate(const ir_node *node, bool relative)
+{
+       const ia32_immediate_attr_t *attr = get_ia32_immediate_attr_const(node);
+       bemit_entity(attr->symconst, attr->sc_sign, attr->offset, relative);
+}
+
+static void bemit_copy(const ir_node *copy)
+{
+       const arch_register_t *in  = get_in_reg(copy, 0);
+       const arch_register_t *out = get_out_reg(copy, 0);
+
+       if (in == out || is_unknown_reg(in))
+               return;
+       /* copies of vf nodes aren't real... */
+       if (arch_register_get_class(in) == &ia32_reg_classes[CLASS_ia32_vfp])
+               return;
+
+       if (get_irn_mode(copy) == mode_E) {
+               panic("NIY");
+       } else {
+               assert(arch_register_get_class(in) == &ia32_reg_classes[CLASS_ia32_gp]);
+               bemit8(0x89);
+               bemit_modrr(out, in);
+       }
+}
+
+static void bemit_xor0(const ir_node *node)
+{
+       const arch_register_t *out = get_out_reg(node, 0);
+       bemit8(0x31);
+       bemit_modrr(out, out);
+}
+
+static void bemit_mov_const(const ir_node *node)
+{
+       const arch_register_t *out = get_out_reg(node, 0);
+       bemit8(0xB8 + reg_gp_map[out->index]);
+       bemit_immediate(node, false);
+}
+
+/**
+ * Creates a function for a Binop with 3 possible encodings.
+ */
+#define BINOP(op, op0, op1, op2, op2_ext)                                 \
+static void bemit_ ## op(const ir_node *node) {                           \
+       static const unsigned char op ## _codes[] = {op0, op1, op2, op2_ext}; \
+       bemit_binop(node, op ## _codes);                                      \
+}
+
+/*    insn  def  eax,imm   imm */
+BINOP(add,  0x01, 0x05, 0x81, 0)
+BINOP(or,   0x09, 0x0D, 0x81, 1)
+BINOP(adc,  0x11, 0x15, 0x81, 2)
+BINOP(sbb,  0x19, 0x1D, 0x81, 3)
+BINOP(and,  0x21, 0x25, 0x81, 4)
+BINOP(sub,  0x29, 0x2D, 0x81, 5)
+BINOP(xor,  0x31, 0x35, 0x81, 6)
+BINOP(cmp,  0x39, 0x3D, 0x81, 7)
+BINOP(test, 0x85, 0xA9, 0xF7, 0)
+
+/**
+ * Creates a function for an Unop with code /ext encoding.
+ */
+#define UNOP(op, code, ext, input)              \
+static void bemit_ ## op(const ir_node *node) { \
+       bemit_unop(node, code, ext, input);         \
+}
+
+UNOP(not,     0xF7, 2, n_ia32_Not_val)
+UNOP(neg,     0xF7, 3, n_ia32_Neg_val)
+UNOP(mul,     0xF7, 4, n_ia32_Mul_right)
+UNOP(imul1op, 0xF7, 5, n_ia32_IMul1OP_right)
+UNOP(div,     0xF7, 6, n_ia32_Div_divisor)
+UNOP(idiv,    0xF7, 7, n_ia32_IDiv_divisor)
+
+/* TODO: am support for IJmp */
+UNOP(ijmp,    0xFF, 4, n_ia32_IJmp_target)
+
+#define SHIFT(op, ext) \
+static void bemit_##op(const ir_node *node) \
+{ \
+       const arch_register_t *out   = get_out_reg(node, 0); \
+       ir_node               *count = get_irn_n(node, 1); \
+       if (is_ia32_Immediate(count)) { \
+               int offset = get_ia32_immediate_attr_const(count)->offset; \
+               if (offset == 1) { \
+                       bemit8(0xD1); \
+                       bemit_modru(out, ext); \
+               } else { \
+                       bemit8(0xC1); \
+                       bemit_modru(out, ext); \
+                       bemit8(offset); \
+               } \
+       } else { \
+               bemit8(0xD3); \
+               bemit_modru(out, ext); \
+       } \
+}
+
+SHIFT(rol, 0)
+SHIFT(ror, 1)
+SHIFT(shl, 4)
+SHIFT(shr, 5)
+SHIFT(sar, 7)
+
+static void bemit_imul(const ir_node *node)
+{
+       ir_node *right = get_irn_n(node, n_ia32_IMul_right);
+       /* Do we need the immediate form? */
+       if (is_ia32_Immediate(right)) {
+               int imm = get_ia32_immediate_attr_const(right)->offset;
+               if (get_signed_imm_size(imm) == 1) {
+                       bemit_unop_reg(node, 0x6B, n_ia32_IMul_left);
+                       bemit8(imm);
+               } else {
+                       bemit_unop_reg(node, 0x69, n_ia32_IMul_left);
+                       bemit32(imm);
+               }
+       } else {
+               bemit8(0x0F);
+               bemit_unop_reg(node, 0xAF, n_ia32_IMul_right);
+       }
+}
+
+static void bemit_dec(const ir_node *node)
+{
+       const arch_register_t *out = get_out_reg(node, pn_ia32_Dec_res);
+       bemit8(0x48 + reg_gp_map[out->index]);
+}
+
+static void bemit_inc(const ir_node *node)
+{
+       const arch_register_t *out = get_out_reg(node, pn_ia32_Inc_res);
+       bemit8(0x40 + reg_gp_map[out->index]);
+}
+
+/**
+ * Emit a Lea.
+ */
+static void bemit_lea(const ir_node *node)
+{
+       const arch_register_t *out = get_out_reg(node, 0);
+       bemit8(0x8D);
+       bemit_mod_am(reg_gp_map[out->index], node);
+}
+
+/**
+ * Emit a single optcode.
+ */
+#define EMIT_SINGLEOP(op, code)                 \
+static void bemit_ ## op(const ir_node *node) { \
+       (void) node;                                \
+       bemit8(code);                               \
+}
+
+//EMIT_SINGLEOP(daa,  0x27)
+//EMIT_SINGLEOP(das,  0x2F)
+//EMIT_SINGLEOP(aaa,  0x37)
+//EMIT_SINGLEOP(aas,  0x3F)
+//EMIT_SINGLEOP(nop,  0x90)
+EMIT_SINGLEOP(cwtl, 0x98)
+EMIT_SINGLEOP(cltd, 0x99)
+//EMIT_SINGLEOP(fwait, 0x9B)
+EMIT_SINGLEOP(sahf, 0x9E)
+//EMIT_SINGLEOP(popf, 0x9D)
+EMIT_SINGLEOP(int3, 0xCC)
+//EMIT_SINGLEOP(iret, 0xCF)
+//EMIT_SINGLEOP(xlat, 0xD7)
+//EMIT_SINGLEOP(lock, 0xF0)
+EMIT_SINGLEOP(rep,  0xF3)
+//EMIT_SINGLEOP(halt, 0xF4)
+EMIT_SINGLEOP(cmc,  0xF5)
+EMIT_SINGLEOP(stc,  0xF9)
+//EMIT_SINGLEOP(cli,  0xFA)
+//EMIT_SINGLEOP(sti,  0xFB)
+//EMIT_SINGLEOP(std,  0xFD)
+
+/**
+ * Emits a MOV out, [MEM].
+ */
+static void bemit_load(const ir_node *node)
+{
+       const arch_register_t *out = get_out_reg(node, 0);
+
+       if (out->index == REG_EAX) {
+               ir_entity *ent       = get_ia32_am_sc(node);
+               int        offs      = get_ia32_am_offs_int(node);
+               ir_node   *base      = get_irn_n(node, n_ia32_base);
+               int        has_base  = !is_ia32_NoReg_GP(base);
+               ir_node   *index     = get_irn_n(node, n_ia32_index);
+               int        has_index = !is_ia32_NoReg_GP(index);
+
+               if (ent == NULL && !has_base && !has_index) {
+                       /* load from constant address to EAX can be encoded
+                          as 0xA1 [offset] */
+                       bemit8(0xA1);
+                       bemit_entity(NULL, 0, offs, false);
+                       return;
+               }
+       }
+       bemit8(0x8B);
+       bemit_mod_am(reg_gp_map[out->index], node);
+}
+
+/**
+ * Emits a MOV [mem], in.
+ */
+static void bemit_store(const ir_node *node)
+{
+       const ir_node *value = get_irn_n(node, n_ia32_Store_val);
+
+       if (is_ia32_Immediate(value)) {
+               bemit8(0xC7);
+               bemit_mod_am(0, node);
+               bemit_immediate(value, false);
+       } else {
+               const arch_register_t *in = get_in_reg(node, n_ia32_Store_val);
+
+               if (in->index == REG_EAX) {
+                       ir_entity *ent       = get_ia32_am_sc(node);
+                       int        offs      = get_ia32_am_offs_int(node);
+                       ir_node   *base      = get_irn_n(node, n_ia32_base);
+                       int        has_base  = !is_ia32_NoReg_GP(base);
+                       ir_node   *index     = get_irn_n(node, n_ia32_index);
+                       int        has_index = !is_ia32_NoReg_GP(index);
+
+                       if (ent == NULL && !has_base && !has_index) {
+                               /* store to constant address from EAX can be encoded as
+                                  0xA3 [offset]*/
+                               bemit8(0xA3);
+                               bemit_entity(NULL, 0, offs, false);
+                               return;
+                       }
+               }
+               bemit8(0x89);
+               bemit_mod_am(reg_gp_map[in->index], node);
+       }
+}
+
+static void bemit_conv_i2i(const ir_node *node)
+{
+       ir_mode  *smaller_mode = get_ia32_ls_mode(node);
+       unsigned  opcode;
+
+       bemit8(0x0F);
+       /*        8 16 bit source
+        * movzx B6 B7
+        * movsx BE BF
+        */
+       opcode = 0xB6;
+       if (mode_is_signed(smaller_mode))           opcode |= 0x08;
+       if (get_mode_size_bits(smaller_mode) == 16) opcode |= 0x01;
+       bemit_unop_reg(node, opcode, n_ia32_Conv_I2I_val);
+}
+
+/**
+ * Emit a Push.
+ */
+static void bemit_push(const ir_node *node)
+{
+       const ir_node *value = get_irn_n(node, n_ia32_Push_val);
+
+       if (is_ia32_Immediate(value)) {
+               const ia32_immediate_attr_t *attr
+                       = get_ia32_immediate_attr_const(value);
+               unsigned size = get_unsigned_imm_size(attr->offset);
+               if (attr->symconst)
+                       size = 4;
+               switch (size) {
+               case 1:
+                       bemit8(0x6A);
+                       bemit8((unsigned char)attr->offset);
+                       break;
+               case 2:
+               case 4:
+                       bemit8(0x68);
+                       bemit_immediate(value, false);
+                       break;
+               }
+       } else if (is_ia32_NoReg_GP(value)) {
+               bemit8(0xFF);
+               bemit_mod_am(6, node);
+       } else {
+               const arch_register_t *reg = get_in_reg(node, n_ia32_Push_val);
+               bemit8(0x50 + reg_gp_map[reg->index]);
+       }
+}
+
+/**
+ * Emit a Pop.
+ */
+static void bemit_pop(const ir_node *node)
+{
+       const arch_register_t *reg = get_out_reg(node, pn_ia32_Pop_res);
+       bemit8(0x58 + reg_gp_map[reg->index]);
+}
+
+static void bemit_popmem(const ir_node *node)
+{
+       bemit8(0x8F);
+       bemit_mod_am(0, node);
+}
+
+static void bemit_call(const ir_node *node)
+{
+       ir_node *proc = get_irn_n(node, n_ia32_Call_addr);
+
+       if (is_ia32_Immediate(proc)) {
+               bemit8(0xE8);
+               bemit_immediate(proc, true);
+       } else {
+               bemit_unop(node, 0xFF, 2, n_ia32_Call_addr);
+       }
+}
+
+static void bemit_jmp(const ir_node *dest_block)
+{
+       bemit8(0xE9);
+       bemit_jmp_destination(dest_block);
+}
+
+static void bemit_jump(const ir_node *node)
+{
+       if (can_be_fallthrough(node))
+               return;
+
+       bemit_jmp(get_cfop_target_block(node));
+}
+
+static void bemit_jcc(int pnc, const ir_node *dest_block)
+{
+       unsigned char cc;
+
+       if (pnc == ia32_pn_Cmp_parity) {
+               cc = 0x0A;
+       } else {
+               if (pnc & ia32_pn_Cmp_float || pnc & ia32_pn_Cmp_unsigned) {
+                       cc = pnc_map_unsigned[pnc & 0x07];
+               } else {
+                       cc = pnc_map_signed[pnc & 0x07];
+               }
+       }
+       assert(cc != 0xFF);
+
+       bemit8(0x0F);
+       bemit8(0x80 + cc);
+       bemit_jmp_destination(dest_block);
+}
+
+static void bemit_ia32_jcc(const ir_node *node)
+{
+       int            pnc = get_ia32_condcode(node);
+       int            need_parity_label = 0;
+       const ir_node *proj_true;
+       const ir_node *proj_false;
+       const ir_node *dest_true;
+       const ir_node *dest_false;
+       const ir_node *block;
+
+       pnc = determine_final_pnc(node, 0, pnc);
+
+       /* get both Projs */
+       proj_true = get_proj(node, pn_ia32_Jcc_true);
+       assert(proj_true && "Jcc without true Proj");
+
+       proj_false = get_proj(node, pn_ia32_Jcc_false);
+       assert(proj_false && "Jcc without false Proj");
+
+       block = get_nodes_block(node);
+
+       if (can_be_fallthrough(proj_true)) {
+               /* exchange both proj's so the second one can be omitted */
+               const ir_node *t = proj_true;
+
+               proj_true  = proj_false;
+               proj_false = t;
+               pnc        = ia32_get_negated_pnc(pnc);
+       }
+
+       dest_true  = get_cfop_target_block(proj_true);
+       dest_false = get_cfop_target_block(proj_false);
+
+       if (pnc & ia32_pn_Cmp_float) {
+               panic("Float jump NIY");
+               /* Some floating point comparisons require a test of the parity flag,
+                * which indicates that the result is unordered */
+               switch (pnc & 15) {
+                       case pn_Cmp_Uo: {
+                               ia32_emitf(proj_true, "\tjp %L\n");
+                               break;
+                       }
+
+                       case pn_Cmp_Leg:
+                               ia32_emitf(proj_true, "\tjnp %L\n");
+                               break;
+
+                       case pn_Cmp_Eq:
+                       case pn_Cmp_Lt:
+                       case pn_Cmp_Le:
+                               /* we need a local label if the false proj is a fallthrough
+                                * as the falseblock might have no label emitted then */
+                               if (can_be_fallthrough(proj_false)) {
+                                       need_parity_label = 1;
+                                       ia32_emitf(proj_false, "\tjp 1f\n");
+                               } else {
+                                       ia32_emitf(proj_false, "\tjp %L\n");
+                               }
+                               goto emit_jcc;
+
+                       case pn_Cmp_Ug:
+                       case pn_Cmp_Uge:
+                       case pn_Cmp_Ne:
+                               ia32_emitf(proj_true, "\tjp %L\n");
+                               goto emit_jcc;
+
+                       default:
+                               goto emit_jcc;
+               }
+       } else {
+emit_jcc:
+               bemit_jcc(pnc, dest_true);
+       }
+
+       if (need_parity_label) {
+               panic("parity label NIY");
+       }
+
+       /* the second Proj might be a fallthrough */
+       if (can_be_fallthrough(proj_false)) {
+               /* it's a fallthrough */
+       } else {
+               bemit_jmp(dest_false);
+       }
+}
+
+/**
+ * Emits a return.
+ */
+static void bemit_return(const ir_node *node)
+{
+       unsigned pop = be_Return_get_pop(node);
+       if (pop > 0 || be_Return_get_emit_pop(node)) {
+               bemit8(0xC2);
+               assert(pop <= 0xffff);
+               bemit16(pop);
+       } else {
+               bemit8(0xC3);
+       }
+}
+
+static void bemit_incsp(const ir_node *node)
+{
+       int                    offs;
+       const arch_register_t *reg;
+       unsigned               size;
+       unsigned               ext;
+
+       offs = be_get_IncSP_offset(node);
+       if (offs == 0)
+               return;
+
+       if (offs > 0) {
+               ext = 5; /* sub */
+       } else {
+               ext = 0; /* add */
+               offs = -offs;
+       }
+
+       size = get_signed_imm_size(offs);
+       bemit8(size == 1 ? 0x83 : 0x81);
+
+       reg  = get_out_reg(node, 0);
+       bemit_modru(reg, ext);
+
+       if (size == 1) {
+               bemit8(offs);
+       } else {
+               bemit32(offs);
+       }
+}
+
+/**
+ * The type of a emitter function.
+ */
+typedef void (*emit_func) (const ir_node *);
+
+/**
+ * Set a node emitter. Make it a bit more type safe.
+ */
+static void register_emitter(ir_op *op, emit_func func)
+{
+       op->ops.generic = (op_func) func;
+}
+
+static void ia32_register_binary_emitters(void)
+{
+       /* first clear the generic function pointer for all ops */
+       clear_irp_opcodes_generic_func();
+
+       /* benode emitter */
+       register_emitter(op_be_Copy,           bemit_copy);
+       register_emitter(op_be_CopyKeep,       bemit_copy);
+       register_emitter(op_be_IncSP,          bemit_incsp);
+       register_emitter(op_be_Return,         bemit_return);
+       register_emitter(op_ia32_Adc,          bemit_adc);
+       register_emitter(op_ia32_Add,          bemit_add);
+       register_emitter(op_ia32_And,          bemit_and);
+       register_emitter(op_ia32_Breakpoint,   bemit_int3);
+       register_emitter(op_ia32_Call,         bemit_call);
+       register_emitter(op_ia32_Cltd,         bemit_cltd);
+       register_emitter(op_ia32_Cmc,          bemit_cmc);
+       register_emitter(op_ia32_Cmp,          bemit_cmp);
+       register_emitter(op_ia32_Const,        bemit_mov_const);
+       register_emitter(op_ia32_Conv_I2I,     bemit_conv_i2i);
+       register_emitter(op_ia32_Conv_I2I8Bit, bemit_conv_i2i);
+       register_emitter(op_ia32_Cwtl,         bemit_cwtl);
+       register_emitter(op_ia32_Dec,          bemit_dec);
+       register_emitter(op_ia32_Div,          bemit_div);
+       register_emitter(op_ia32_IDiv,         bemit_idiv);
+       register_emitter(op_ia32_IJmp,         bemit_ijmp);
+       register_emitter(op_ia32_IMul1OP,      bemit_imul1op);
+       register_emitter(op_ia32_IMul,         bemit_imul);
+       register_emitter(op_ia32_Inc,          bemit_inc);
+       register_emitter(op_ia32_Jcc,          bemit_ia32_jcc);
+       register_emitter(op_ia32_Jmp,          bemit_jump);
+       register_emitter(op_ia32_Lea,          bemit_lea);
+       register_emitter(op_ia32_Load,         bemit_load);
+       register_emitter(op_ia32_Mul,          bemit_mul);
+       register_emitter(op_ia32_Neg,          bemit_neg);
+       register_emitter(op_ia32_Not,          bemit_not);
+       register_emitter(op_ia32_Or,           bemit_or);
+       register_emitter(op_ia32_Pop,          bemit_pop);
+       register_emitter(op_ia32_PopEbp,       bemit_pop);
+       register_emitter(op_ia32_PopMem,       bemit_popmem);
+       register_emitter(op_ia32_Push,         bemit_push);
+       register_emitter(op_ia32_RepPrefix,    bemit_rep);
+       register_emitter(op_ia32_Rol,          bemit_rol);
+       register_emitter(op_ia32_Ror,          bemit_ror);
+       register_emitter(op_ia32_Sahf,         bemit_sahf);
+       register_emitter(op_ia32_Sar,          bemit_sar);
+       register_emitter(op_ia32_Sbb,          bemit_sbb);
+       register_emitter(op_ia32_Shl,          bemit_shl);
+       register_emitter(op_ia32_Shr,          bemit_shr);
+       register_emitter(op_ia32_Stc,          bemit_stc);
+       register_emitter(op_ia32_Store,        bemit_store);
+       register_emitter(op_ia32_Sub,          bemit_sub);
+       register_emitter(op_ia32_Test,         bemit_test);
+       register_emitter(op_ia32_Xor,          bemit_xor);
+       register_emitter(op_ia32_Xor0,         bemit_xor0);
+
+       /* ignore the following nodes */
+       register_emitter(op_ia32_ProduceVal,   emit_Nothing);
+       register_emitter(op_be_Barrier,        emit_Nothing);
+       register_emitter(op_be_Keep,           emit_Nothing);
+       register_emitter(op_be_Start,          emit_Nothing);
+       register_emitter(op_Phi,               emit_Nothing);
+       register_emitter(op_Start,             emit_Nothing);
+}
+
+static void gen_binary_block(ir_node *block)
+{
+       ir_node *node;
+
+       ia32_emit_block_header(block);
+
+       /* emit the contents of the block */
+       sched_foreach(block, node) {
+               ia32_emit_node(node);
+       }
+}
+
+void ia32_gen_binary_routine(ia32_code_gen_t *ia32_cg, ir_graph *irg)
+{
+       ir_entity *entity     = get_irg_entity(irg);
+       int i, n;
+
+       cg  = ia32_cg;
+       isa = cg->isa;
+
+       ia32_register_binary_emitters();
+
+       be_gas_emit_function_prolog(entity, ia32_cg_config.function_alignment);
+
+       /* we use links to point to target blocks */
+       ir_reserve_resources(irg, IR_RESOURCE_IRN_LINK);
+       irg_block_walk_graph(irg, ia32_gen_labels, NULL, NULL);
+
+       /* initialize next block links */
+       n = ARR_LEN(cg->blk_sched);
+       for (i = 0; i < n; ++i) {
+               ir_node *block = cg->blk_sched[i];
+               ir_node *prev  = i > 0 ? cg->blk_sched[i-1] : NULL;
+
+               set_irn_link(block, prev);
+       }
+
+       for (i = 0; i < n; ++i) {
+               ir_node *block = cg->blk_sched[i];
+               gen_binary_block(block);
+       }
+
+       be_gas_emit_function_epilog(entity);
+       be_dbg_method_end();
+       be_emit_char('\n');
+       be_emit_write_line();
+
+       ir_free_resources(irg, IR_RESOURCE_IRN_LINK);
+}
+
+
+
+
+void ia32_init_emitter(void)
+{
+       lc_opt_entry_t *be_grp;
+       lc_opt_entry_t *ia32_grp;
+
+       be_grp   = lc_opt_get_grp(firm_opt_get_root(), "be");
+       ia32_grp = lc_opt_get_grp(be_grp, "ia32");
+
+       lc_opt_add_table(ia32_grp, ia32_emitter_options);
+
+       build_reg_map();
+
        FIRM_DBG_REGISTER(dbg, "firm.be.ia32.emitter");
 }