X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;f=ir%2Fbe%2Fia32%2Fia32_emitter.c;h=192ecbe7178f576e190b0f6d1efe1703d21fb61b;hb=48071aea23fabc99044488d12757f274bc956fae;hp=084dc8ffe96095714ae7fcde8d089d74ce456c8c;hpb=87a6f0fa953a864bc2736ea292b6ec6c0b09c097;p=libfirm

diff --git a/ir/be/ia32/ia32_emitter.c b/ir/be/ia32/ia32_emitter.c
index 084dc8ffe..192ecbe71 100644
--- a/ir/be/ia32/ia32_emitter.c
+++ b/ir/be/ia32/ia32_emitter.c
@@ -1,6 +1,6 @@
 /**
  * This file implements the node emitter.
- *
+ * @author Christian Wuerdig
  * $Id$
  */
 
@@ -30,27 +30,77 @@
 #include "ia32_nodes_attr.h"
 #include "ia32_new_nodes.h"
 #include "ia32_map_regs.h"
-
-#ifdef obstack_chunk_alloc
-# undef obstack_chunk_alloc
-# define obstack_chunk_alloc xmalloc
-#else
-# define obstack_chunk_alloc xmalloc
-# define obstack_chunk_free free
-#endif
+#include "bearch_ia32_t.h"
 
 #define BLOCK_PREFIX(x) ".L" x
 
-extern int obstack_printf(struct obstack *obst, char *fmt, ...);
-
 #define SNPRINTF_BUF_LEN 128
 
 /* global arch_env for lc_printf functions */
 static const arch_env_t *arch_env = NULL;
 
-/* indicates whether blocks are scheduled or not
-   (this variable is set automatically) */
-static int have_block_sched       = 0;
+/** by default, we generate assembler code for the Linux gas */
+asm_flavour_t asm_flavour = ASM_LINUX_GAS;
+
+/**
+ * Switch to a new section
+ */
+void ia32_switch_section(FILE *F, section_t sec) {
+	static section_t curr_sec = NO_SECTION;
+	static const char *text[ASM_MAX][SECTION_MAX] = {
+		{
+			".section\t.text", ".section\t.data", ".section\t.rodata", ".section\t.text"
+		},
+		{
+			".section\t.text", ".section\t.data", ".section .rdata,\"dr\"", ".section\t.text"
+		}
+	};
+
+	if (curr_sec == sec)
+		return;
+
+	curr_sec = sec;
+	switch (sec) {
+
+	case NO_SECTION:
+		break;
+
+	case SECTION_TEXT:
+	case SECTION_DATA:
+	case SECTION_RODATA:
+	case SECTION_COMMON:
+		fprintf(F, "\t%s\n", text[asm_flavour][sec]);
+		break;
+
+	default:
+		break;
+	}
+}
+
+static void ia32_dump_function_object(FILE *F, const char *name)
+{
+	switch (asm_flavour) {
+	case ASM_LINUX_GAS:
+		fprintf(F, "\t.type\t%s, @function\n", name);
+		break;
+	case ASM_MINGW_GAS:
+		fprintf(F, "\t.def\t%s;\t.scl\t2;\t.type\t32;\t.endef\n", name);
+		break;
+	default:
+		break;
+	}
+}
+
+static void ia32_dump_function_size(FILE *F, const char *name)
+{
+	switch (asm_flavour) {
+	case ASM_LINUX_GAS:
+		fprintf(F, "\t.size\t%s, .-%s\n", name, name);
+		break;
+	default:
+		break;
+	}
+}
 
 /*************************************************************
  *             _       _    __   _          _
@@ -63,6 +113,20 @@ static int have_block_sched       = 0;
  * |_|                                       |_|
  *************************************************************/
 
+static INLINE int be_is_unknown_reg(const arch_register_t *reg) {
+	return \
+		REGS_ARE_EQUAL(reg, &ia32_gp_regs[REG_GP_UKNWN])   || \
+		REGS_ARE_EQUAL(reg, &ia32_xmm_regs[REG_XMM_UKNWN]) || \
+		REGS_ARE_EQUAL(reg, &ia32_vfp_regs[REG_VFP_UKNWN]);
+}
+
+/**
+ * returns true if a node has x87 registers
+ */
+static INLINE int has_x87_register(const ir_node *n) {
+	return is_irn_machine_user(n, 0);
+}
+
 /* We always pass the ir_node which is a pointer. */
 static int ia32_get_arg_type(const lc_arg_occ_t *occ) {
 	return lc_arg_type_ptr;
@@ -85,6 +149,15 @@ static const arch_register_t *get_in_reg(const ir_node *irn, int pos) {
 	reg = arch_get_irn_register(arch_env, op);
 
 	assert(reg && "no in register found");
+
+	/* in case of unknown: just return a register */
+	if (REGS_ARE_EQUAL(reg, &ia32_gp_regs[REG_GP_UKNWN]))
+		reg = &ia32_gp_regs[REG_EAX];
+	else if (REGS_ARE_EQUAL(reg, &ia32_xmm_regs[REG_XMM_UKNWN]))
+		reg = &ia32_xmm_regs[REG_XMM0];
+	else if (REGS_ARE_EQUAL(reg, &ia32_vfp_regs[REG_VFP_UKNWN]))
+		reg = &ia32_vfp_regs[REG_VF0];
+
 	return reg;
 }
 
@@ -133,12 +206,15 @@ enum io_direction {
  */
 static const char *get_ia32_reg_name(ir_node *irn, int pos, enum io_direction in_out) {
 	const arch_register_t *reg;
-	const char            *name;
-	static char           *buf = NULL;
-	int                    len;
 
 	if (in_out == IN_REG) {
 		reg = get_in_reg(irn, pos);
+
+		if (reg->reg_class == &ia32_reg_classes[CLASS_ia32_vfp]) {
+			/* FIXME: works for binop only */
+			assert(2 <= pos && pos <= 3);
+			reg = get_ia32_attr(irn)->x87[pos - 2];
+		}
 	}
 	else {
 		/* destination address mode nodes don't have outputs */
@@ -147,37 +223,49 @@ static const char *get_ia32_reg_name(ir_node *irn, int pos, enum io_direction in
 		}
 
 		reg = get_out_reg(irn, pos);
+		if (reg->reg_class == &ia32_reg_classes[CLASS_ia32_vfp])
+			reg = get_ia32_attr(irn)->x87[pos + 2];
 	}
+	return arch_register_get_name(reg);
+}
 
-	name = arch_register_get_name(reg);
-
-	if (buf) {
-		free(buf);
-	}
+/**
+ * Get the register name for a node.
+ */
+static int ia32_get_reg_name(lc_appendable_t *app,
+    const lc_arg_occ_t *occ, const lc_arg_value_t *arg)
+{
+	const char *buf;
+	ir_node    *irn = arg->v_ptr;
+	int         nr = occ->width - 1;
 
-	len = strlen(name) + 2;
-	buf = xcalloc(1, len);
+	if (! irn)
+		return lc_appendable_snadd(app, "(null)", 6);
 
-	snprintf(buf, len, "%%%s", name);
+	buf = get_ia32_reg_name(irn, nr, occ->conversion == 'S' ? IN_REG : OUT_REG);
 
-	return buf;
+	/* append the stupid % to register names */
+	lc_appendable_chadd(app, '%');
+	return lc_appendable_snadd(app, buf, strlen(buf));
 }
 
 /**
- * Get the register name for a node.
+ * Get the x87 register name for a node.
  */
-static int ia32_get_reg_name(lc_appendable_t *app,
+static int ia32_get_x87_name(lc_appendable_t *app,
     const lc_arg_occ_t *occ, const lc_arg_value_t *arg)
 {
 	const char *buf;
-	ir_node    *X  = arg->v_ptr;
+	ir_node     *irn = arg->v_ptr;
 	int         nr = occ->width - 1;
+	ia32_attr_t *attr;
 
-	if (!X)
+	if (! irn)
 		return lc_appendable_snadd(app, "(null)", 6);
 
-	buf = get_ia32_reg_name(X, nr, occ->conversion == 'S' ? IN_REG : OUT_REG);
-
+	attr = get_ia32_attr(irn);
+	buf = attr->x87[nr]->name;
+	lc_appendable_chadd(app, '%');
 	return lc_appendable_snadd(app, buf, strlen(buf));
 }
 
@@ -188,16 +276,16 @@ static int ia32_const_to_str(lc_appendable_t *app,
     const lc_arg_occ_t *occ, const lc_arg_value_t *arg)
 {
 	const char *buf;
-	ir_node    *X = arg->v_ptr;
+	ir_node    *irn = arg->v_ptr;
 
-	if (!X)
+	if (! irn)
 		return lc_arg_append(app, occ, "(null)", 6);
 
 	if (occ->conversion == 'C') {
-		buf = get_ia32_cnst(X);
+		buf = get_ia32_cnst(irn);
 	}
 	else { /* 'O' */
-		buf = get_ia32_am_offs(X);
+		buf = get_ia32_am_offs(irn);
 	}
 
 	return buf ? lc_appendable_snadd(app, buf, strlen(buf)) : 0;
@@ -209,14 +297,16 @@ static int ia32_const_to_str(lc_appendable_t *app,
 static int ia32_get_mode_suffix(lc_appendable_t *app,
     const lc_arg_occ_t *occ, const lc_arg_value_t *arg)
 {
-	ir_node *X    = arg->v_ptr;
-	ir_mode *mode = get_irn_mode(X);
+	ir_node *irn  = arg->v_ptr;
+	ir_mode *mode = get_irn_mode(irn);
 
 	if (mode == mode_T) {
-		mode = is_ia32_AddrModeS(X) || is_ia32_AddrModeD(X) ? get_ia32_ls_mode(X) : get_ia32_res_mode(X);
+		mode = get_ia32_res_mode(irn);
+		if (! mode)
+			mode = get_ia32_ls_mode(irn);
 	}
 
-	if (!X)
+	if (! irn)
 		return lc_arg_append(app, occ, "(null)", 6);
 
 	if (mode_is_float(mode)) {
@@ -237,6 +327,7 @@ const lc_arg_env_t *ia32_get_arg_env(void) {
 	static const lc_arg_handler_t ia32_reg_handler   = { ia32_get_arg_type, ia32_get_reg_name };
 	static const lc_arg_handler_t ia32_const_handler = { ia32_get_arg_type, ia32_const_to_str };
 	static const lc_arg_handler_t ia32_mode_handler  = { ia32_get_arg_type, ia32_get_mode_suffix };
+	static const lc_arg_handler_t ia32_x87_handler   = { ia32_get_arg_type, ia32_get_x87_name };
 
 	if(env == NULL) {
 		/* extend the firm printer */
@@ -247,12 +338,13 @@ const lc_arg_env_t *ia32_get_arg_env(void) {
 		lc_arg_register(env, "ia32:cnst", 'C', &ia32_const_handler);
 		lc_arg_register(env, "ia32:offs", 'O', &ia32_const_handler);
 		lc_arg_register(env, "ia32:mode", 'M', &ia32_mode_handler);
+		lc_arg_register(env, "ia32:x87",  'X', &ia32_x87_handler);
 	}
 
 	return env;
 }
 
-static char *ia32_get_reg_name_for_mode(ia32_emit_env_t *env, ir_mode *mode, const arch_register_t *reg) {
+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);
@@ -266,7 +358,7 @@ static char *ia32_get_reg_name_for_mode(ia32_emit_env_t *env, ir_mode *mode, con
 /**
  * Emits registers and/or address mode of a binary operation.
  */
-char *ia32_emit_binop(const ir_node *n, ia32_emit_env_t *env) {
+const char *ia32_emit_binop(const ir_node *n, ia32_emit_env_t *env) {
 	static char *buf = NULL;
 
 	/* verify that this function is never called on non-AM supporting operations */
@@ -276,7 +368,9 @@ char *ia32_emit_binop(const ir_node *n, ia32_emit_env_t *env) {
 	(!(is_ia32_St(n)      || \
 	is_ia32_Store8Bit(n)  || \
 	is_ia32_CondJmp(n)    || \
-	is_ia32_fCondJmp(n)   || \
+	is_ia32_xCondJmp(n)   || \
+	is_ia32_CmpSet(n)     || \
+	is_ia32_xCmpSet(n)    || \
 	is_ia32_SwitchJmp(n)))
 
 	if (! buf) {
@@ -288,24 +382,44 @@ char *ia32_emit_binop(const ir_node *n, ia32_emit_env_t *env) {
 
 	switch(get_ia32_op_type(n)) {
 		case ia32_Normal:
-			if (is_ia32_ImmConst(n) || is_ia32_ImmSymConst(n)) {
+			if (is_ia32_ImmConst(n)) {
 				lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%3S, %s", n, get_ia32_cnst(n));
 			}
+			else if (is_ia32_ImmSymConst(n)) {
+				lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%3S, OFFSET FLAT:%s", n, get_ia32_cnst(n));
+			}
 			else {
 				const arch_register_t *in1 = get_in_reg(n, 2);
 				const arch_register_t *in2 = get_in_reg(n, 3);
 				const arch_register_t *out = PRODUCES_RESULT(n) ? get_out_reg(n, 0) : NULL;
 				const arch_register_t *in;
+				const char            *in_name;
+
+				in      = out ? (REGS_ARE_EQUAL(out, in2) ? in1 : in2) : in2;
+				out     = out ? out : in1;
+				in_name = arch_register_get_name(in);
 
-				in  = out ? (REGS_ARE_EQUAL(out, in2) ? in1 : in2) : in2;
-				out = out ? out : in1;
+				if (is_ia32_emit_cl(n)) {
+					assert(REGS_ARE_EQUAL(&ia32_gp_regs[REG_ECX], in) && "shift operation needs ecx");
+					in_name = "cl";
+				}
 
-				snprintf(buf, SNPRINTF_BUF_LEN, "%%%s, %%%s", \
-					arch_register_get_name(out), arch_register_get_name(in));
+				snprintf(buf, SNPRINTF_BUF_LEN, "%%%s, %%%s", arch_register_get_name(out), in_name);
 			}
 			break;
 		case ia32_AddrModeS:
-			lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%4S, %s", n, ia32_emit_am(n, env));
+			if (is_ia32_ImmConst(n) || is_ia32_ImmSymConst(n)) {
+				assert(! PRODUCES_RESULT(n) && "Source AM with Const must not produce result");
+				snprintf(buf, SNPRINTF_BUF_LEN, "%s, %s", get_ia32_cnst(n), ia32_emit_am(n, env));
+			}
+			else {
+				if (PRODUCES_RESULT(n)) {
+					lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%1D, %s", n, ia32_emit_am(n, env));
+				}
+				else {
+					lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%3S, %s", n, ia32_emit_am(n, env));
+				}
+			}
 			break;
 		case ia32_AddrModeD:
 			if (is_ia32_ImmConst(n) || is_ia32_ImmSymConst(n)) {
@@ -315,12 +429,19 @@ char *ia32_emit_binop(const ir_node *n, ia32_emit_env_t *env) {
 					get_ia32_cnst(n));                               /* tell the assembler to store it's address.   */
 			}
 			else {
-				const arch_register_t *in1 = get_in_reg(n, 2);
-				ir_mode              *mode = get_ia32_res_mode(n);
+				const arch_register_t *in1 = get_in_reg(n, get_irn_arity(n) == 5 ? 3 : 2);
+				ir_mode               *mode = get_ia32_res_mode(n);
+				const char            *in_name;
+
+				mode    = mode ? mode : get_ia32_ls_mode(n);
+				in_name = ia32_get_reg_name_for_mode(env, mode, in1);
+
+				if (is_ia32_emit_cl(n)) {
+					assert(REGS_ARE_EQUAL(&ia32_gp_regs[REG_ECX], in1) && "shift operation needs ecx");
+					in_name = "cl";
+				}
 
-				mode = mode ? mode : get_ia32_ls_mode(n);
-				lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%s, %%%s",
-					ia32_emit_am(n, env), ia32_get_reg_name_for_mode(env, mode, in1));
+				lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%s, %%%s", ia32_emit_am(n, env), in_name);
 			}
 			break;
 		default:
@@ -332,10 +453,85 @@ char *ia32_emit_binop(const ir_node *n, ia32_emit_env_t *env) {
 	return buf;
 }
 
+/**
+ * Returns the xxx PTR string for a given mode
+ *
+ * @param mode      the mode
+ * @param x87_insn  if non-zero returns the string for a x87 instruction
+ *                  else for a SSE instruction
+ */
+static const char *pointer_size(ir_mode *mode, int x87_insn)
+{
+	if (mode) {
+		switch (get_mode_size_bits(mode)) {
+		case 8:  return "BYTE PTR";
+		case 16: return "WORD PTR";
+		case 32: return "DWORD PTR";
+		case 64:
+			if (x87_insn)
+				return "QWORD PTR";
+			return NULL;
+		case 80:
+		case 96: return "XWORD PTR";
+		default: return NULL;
+		}
+	}
+	return NULL;
+}
+
+/**
+ * Emits registers and/or address mode of a binary operation.
+ */
+const char *ia32_emit_x87_binop(const ir_node *n, ia32_emit_env_t *env) {
+	static char *buf = NULL;
+
+	/* verify that this function is never called on non-AM supporting operations */
+	//assert(get_ia32_am_support(n) != ia32_am_None && "emit binop expects addressmode support");
+
+	if (! buf) {
+		buf = xcalloc(1, SNPRINTF_BUF_LEN);
+	}
+	else {
+		memset(buf, 0, SNPRINTF_BUF_LEN);
+	}
+
+	switch(get_ia32_op_type(n)) {
+		case ia32_Normal:
+			if (is_ia32_ImmConst(n) || is_ia32_ImmSymConst(n)) {
+				ir_mode *mode = get_ia32_ls_mode(n);
+				const char *p = pointer_size(mode, 1);
+				lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%s %s", p, get_ia32_cnst(n));
+			}
+			else {
+				ia32_attr_t *attr = get_ia32_attr(n);
+				const arch_register_t *in1 = attr->x87[0];
+				const arch_register_t *in2 = attr->x87[1];
+				const arch_register_t *out = attr->x87[2];
+				const arch_register_t *in;
+				const char            *in_name;
+
+				in      = out ? (REGS_ARE_EQUAL(out, in2) ? in1 : in2) : in2;
+				out     = out ? out : in1;
+				in_name = arch_register_get_name(in);
+
+				snprintf(buf, SNPRINTF_BUF_LEN, "%%%s, %%%s", arch_register_get_name(out), in_name);
+			}
+			break;
+		case ia32_AddrModeS:
+		case ia32_AddrModeD:
+			lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%s", ia32_emit_am(n, env));
+			break;
+		default:
+			assert(0 && "unsupported op type");
+	}
+
+	return buf;
+}
+
 /**
  * Emits registers and/or address mode of a unary operation.
  */
-char *ia32_emit_unop(const ir_node *n, ia32_emit_env_t *env) {
+const char *ia32_emit_unop(const ir_node *n, ia32_emit_env_t *env) {
 	static char *buf = NULL;
 
 	if (! buf) {
@@ -347,10 +543,28 @@ char *ia32_emit_unop(const ir_node *n, ia32_emit_env_t *env) {
 
 	switch(get_ia32_op_type(n)) {
 		case ia32_Normal:
-			lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%1D", n);
+			if (is_ia32_ImmConst(n) || is_ia32_ImmSymConst(n)) {
+				lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%C", n);
+			}
+			else {
+				if (is_ia32_MulS(n) || is_ia32_Mulh(n)) {
+					/* MulS and Mulh implicitly multiply by EAX */
+					lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%4S", n);
+				}
+				else
+					lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%1D", n);
+			}
+			break;
+		case ia32_AddrModeD:
+			snprintf(buf, SNPRINTF_BUF_LEN, "%s", ia32_emit_am(n, env));
 			break;
-		case ia32_am_Dest:
-			snprintf(buf, SNPRINTF_BUF_LEN, ia32_emit_am(n, env));
+		case ia32_AddrModeS:
+			/*
+				Mulh is emitted via emit_unop
+				imul [MEM]  means EDX:EAX <- EAX * [MEM]
+			*/
+			assert((is_ia32_Mulh(n) || is_ia32_MulS(n)) && "Only MulS and Mulh can have AM source as unop");
+			lc_esnprintf(ia32_get_arg_env(), buf, SNPRINTF_BUF_LEN, "%s", ia32_emit_am(n, env));
 			break;
 		default:
 			assert(0 && "unsupported op type");
@@ -362,11 +576,11 @@ char *ia32_emit_unop(const ir_node *n, ia32_emit_env_t *env) {
 /**
  * Emits address mode.
  */
-char *ia32_emit_am(const ir_node *n, ia32_emit_env_t *env) {
+const char *ia32_emit_am(const ir_node *n, ia32_emit_env_t *env) {
 	ia32_am_flavour_t am_flav    = get_ia32_am_flavour(n);
 	int               had_output = 0;
-	char             *s;
-	int               size;
+	char              *s;
+	const char        *p;
 	static struct obstack *obst  = NULL;
 	ir_mode *mode = get_ia32_ls_mode(n);
 
@@ -383,25 +597,19 @@ char *ia32_emit_am(const ir_node *n, ia32_emit_env_t *env) {
 	/* obstack_free with NULL results in an uninitialized obstack */
 	obstack_init(obst);
 
-	if (mode) {
-		switch (get_mode_size_bits(mode)) {
-			case 8:
-				obstack_printf(obst, "BYTE PTR ");
-				break;
-			case 16:
-				obstack_printf(obst, "WORD PTR ");
-				break;
-			case 32:
-				obstack_printf(obst, "DWORD PTR ");
-				break;
-			default:
-				break;
-		}
-	}
+	p = pointer_size(mode, has_x87_register(n) || is_ia32_GetST0(n) || is_ia32_SetST0(n));
+	if (p)
+		obstack_printf(obst, "%s ", p);
 
-	obstack_printf(obst, "[");
+	/* emit address mode symconst */
+	if (get_ia32_am_sc(n)) {
+		if (is_ia32_am_sc_sign(n))
+			obstack_printf(obst, "-");
+		obstack_printf(obst, "%s", get_id_str(get_ia32_am_sc(n)));
+	}
 
 	if (am_flav & ia32_B) {
+		obstack_printf(obst, "[");
 		lc_eoprintf(ia32_get_arg_env(), obst, "%1S", n);
 		had_output = 1;
 	}
@@ -410,6 +618,9 @@ char *ia32_emit_am(const ir_node *n, ia32_emit_env_t *env) {
 		if (had_output) {
 			obstack_printf(obst, "+");
 		}
+		else {
+			obstack_printf(obst, "[");
+		}
 
 		lc_eoprintf(ia32_get_arg_env(), obst, "%2S", n);
 
@@ -421,19 +632,43 @@ char *ia32_emit_am(const ir_node *n, ia32_emit_env_t *env) {
 	}
 
 	if (am_flav & ia32_O) {
-		obstack_printf(obst, get_ia32_am_offs(n));
+		s = get_ia32_am_offs(n);
+
+		if (s) {
+			/* omit explicit + if there was no base or index */
+			if (! had_output) {
+				obstack_printf(obst, "[");
+				if (s[0] == '+')
+					s++;
+			}
+
+			obstack_printf(obst, s);
+			had_output = 1;
+		}
 	}
 
-	obstack_printf(obst, "] ");
+	if (had_output)
+		obstack_printf(obst, "] ");
 
-	size        = obstack_object_size(obst);
-	s           = obstack_finish(obst);
-	s[size - 1] = '\0';
+	obstack_1grow(obst, '\0');
+	s = obstack_finish(obst);
 
 	return s;
 }
 
+/**
+ * emit an address
+ */
+const char *ia32_emit_adr(const ir_node *irn, ia32_emit_env_t *env)
+{
+	static char buf[SNPRINTF_BUF_LEN];
+	ir_mode    *mode = get_ia32_ls_mode(irn);
+	const char *adr  = get_ia32_cnst(irn);
+	const char *pref = pointer_size(mode, has_x87_register(irn));
 
+	snprintf(buf, SNPRINTF_BUF_LEN, "%s %s", pref ? pref : "", adr);
+	return buf;
+}
 
 /**
  * Formated print of commands and comments.
@@ -495,12 +730,12 @@ static const struct cmp2conditon_t cmp2condition_s[] = {
   { "ne",              pn_Cmp_Lg },     /* != */
   { "ordered",         pn_Cmp_Leg },    /* Floating point: ordered */
   { "unordered",       pn_Cmp_Uo },     /* FLoting point: unordered */
-  { "unordered or ==", pn_Cmp_Ue },     /* Floating point: unordered or == */
-  { "unordered or <",  pn_Cmp_Ul },     /* Floating point: unordered or < */
-  { "unordered or <=", pn_Cmp_Ule },    /* Floating point: unordered or <= */
-  { "unordered or >",  pn_Cmp_Ug },     /* Floating point: unordered or > */
-  { "unordered or >=", pn_Cmp_Uge },    /* Floating point: unordered or >= */
-  { "unordered or !=", pn_Cmp_Ne },     /* Floating point: unordered or != */
+  { "e",               pn_Cmp_Ue },     /* Floating point: unordered or == */
+  { "b",               pn_Cmp_Ul },     /* Floating point: unordered or < */
+  { "be",              pn_Cmp_Ule },    /* Floating point: unordered or <= */
+  { "a",               pn_Cmp_Ug },     /* Floating point: unordered or > */
+  { "ae",              pn_Cmp_Uge },    /* Floating point: unordered or >= */
+  { "ne",              pn_Cmp_Ne },     /* Floating point: unordered or != */
   { NULL,              pn_Cmp_True },   /* always true */
 };
 
@@ -511,18 +746,18 @@ 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 },     /* <= */
+	{ "be",              pn_Cmp_Le },     /* <= */
 	{ "a",               pn_Cmp_Gt },     /* > */
 	{ "ae",              pn_Cmp_Ge },     /* >= */
 	{ "ne",              pn_Cmp_Lg },     /* != */
 	{ "ordered",         pn_Cmp_Leg },    /* Floating point: ordered */
 	{ "unordered",       pn_Cmp_Uo },     /* FLoting point: unordered */
-	{ "unordered or ==", pn_Cmp_Ue },     /* Floating point: unordered or == */
-	{ "unordered or <",  pn_Cmp_Ul },     /* Floating point: unordered or < */
-	{ "unordered or <=", pn_Cmp_Ule },    /* Floating point: unordered or <= */
-	{ "unordered or >",  pn_Cmp_Ug },     /* Floating point: unordered or > */
-	{ "unordered or >=", pn_Cmp_Uge },    /* Floating point: unordered or >= */
-	{ "unordered or !=", pn_Cmp_Ne },     /* Floating point: unordered or != */
+	{ "e",               pn_Cmp_Ue },     /* Floating point: unordered or == */
+	{ "b",               pn_Cmp_Ul },     /* Floating point: unordered or < */
+	{ "be",              pn_Cmp_Ule },    /* Floating point: unordered or <= */
+	{ "a",               pn_Cmp_Ug },     /* Floating point: unordered or > */
+	{ "ae",              pn_Cmp_Uge },    /* Floating point: unordered or >= */
+	{ "ne",              pn_Cmp_Ne },     /* Floating point: unordered or != */
 	{ NULL,              pn_Cmp_True },   /* always true */
 };
 
@@ -556,7 +791,29 @@ static char *get_cfop_target(const ir_node *irn, char *buf) {
 
 /** Return the next block in Block schedule */
 static ir_node *next_blk_sched(const ir_node *block) {
-	return have_block_sched ? get_irn_link(block) : NULL;
+	return get_irn_link(block);
+}
+
+/**
+ * Returns the Proj with projection number proj and NOT mode_M
+ */
+static ir_node *get_proj(const ir_node *irn, long proj) {
+	const ir_edge_t *edge;
+	ir_node         *src;
+
+	assert(get_irn_mode(irn) == mode_T && "expected mode_T node");
+
+	foreach_out_edge(irn, edge) {
+		src = get_edge_src_irn(edge);
+
+		assert(is_Proj(src) && "Proj expected");
+		if (get_irn_mode(src) == mode_M)
+			continue;
+
+		if (get_Proj_proj(src) == proj)
+			return src;
+	}
+	return NULL;
 }
 
 /**
@@ -565,64 +822,57 @@ static ir_node *next_blk_sched(const ir_node *block) {
 static void finish_CondJmp(FILE *F, const ir_node *irn, ir_mode *mode) {
 	const ir_node   *proj1, *proj2 = NULL;
 	const ir_node   *block, *next_bl = NULL;
-	const ir_edge_t *edge;
 	char buf[SNPRINTF_BUF_LEN];
 	char cmd_buf[SNPRINTF_BUF_LEN];
 	char cmnt_buf[SNPRINTF_BUF_LEN];
+	int is_unsigned;
 
 	/* get both Proj's */
-	edge = get_irn_out_edge_first(irn);
-	proj1 = get_edge_src_irn(edge);
-	assert(is_Proj(proj1) && "CondJmp with a non-Proj");
+	proj1 = get_proj(irn, pn_Cond_true);
+	assert(proj1 && "CondJmp without true Proj");
 
-	edge = get_irn_out_edge_next(irn, edge);
-	if (edge) {
-		proj2 = get_edge_src_irn(edge);
-		assert(is_Proj(proj2) && "CondJmp with a non-Proj");
-	}
+	proj2 = get_proj(irn, pn_Cond_false);
+	assert(proj2 && "CondJmp without false Proj");
 
 	/* for now, the code works for scheduled and non-schedules blocks */
 	block = get_nodes_block(irn);
-	if (proj2) {
-		/* we have a block schedule */
-		next_bl = next_blk_sched(block);
-
-		if (get_cfop_target_block(proj1) == next_bl) {
-			/* exchange both proj's so the second one can be omitted */
-			const ir_node *t = proj1;
-			proj1 = proj2;
-			proj2 = t;
-		}
+
+	/* we have a block schedule */
+	next_bl = next_blk_sched(block);
+
+	if (get_cfop_target_block(proj1) == next_bl) {
+		/* exchange both proj's so the second one can be omitted */
+		const ir_node *t = proj1;
+		proj1 = proj2;
+		proj2 = t;
 	}
 
 	/* the first Proj must always be created */
+	is_unsigned = mode_is_float(mode) || ! mode_is_signed(mode);
 	if (get_Proj_proj(proj1) == pn_Cond_true) {
 		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "j%s %s",
-					get_cmp_suffix(get_ia32_pncode(irn), !mode_is_signed(get_irn_mode(get_irn_n(irn, 0)))),
+					get_cmp_suffix(get_ia32_pncode(irn), is_unsigned),
 					get_cfop_target(proj1, buf));
 		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* cmp(a, b) == TRUE */");
 	}
 	else  {
 		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "j%s %s",
-					get_cmp_suffix(get_negated_pnc(get_ia32_pncode(irn), mode),
-					!mode_is_signed(get_irn_mode(get_irn_n(irn, 0)))),
+					get_cmp_suffix(get_negated_pnc(get_ia32_pncode(irn), mode), is_unsigned),
 					get_cfop_target(proj1, buf));
 		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* cmp(a, b) == FALSE */");
 	}
 	IA32_DO_EMIT(irn);
 
 	/* the second Proj might be a fallthrough */
-	if (proj2) {
-		if (get_cfop_target_block(proj2) != next_bl) {
-			snprintf(cmd_buf, SNPRINTF_BUF_LEN, "jmp %s", get_cfop_target(proj2, buf));
-			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* otherwise */");
-		}
-		else {
-			cmd_buf[0] = '\0';
-			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* fallthrogh %s */", get_cfop_target(proj2, buf));
-		}
-		IA32_DO_EMIT(irn);
+	if (get_cfop_target_block(proj2) != next_bl) {
+		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "jmp %s", get_cfop_target(proj2, buf));
+		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* otherwise */");
+	}
+	else {
+		cmd_buf[0] = '\0';
+		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* fallthrough %s */", get_cfop_target(proj2, buf));
 	}
+	IA32_DO_EMIT(irn);
 }
 
 /**
@@ -636,7 +886,7 @@ static void CondJmp_emitter(const ir_node *irn, ia32_emit_env_t *env) {
 	snprintf(cmd_buf, SNPRINTF_BUF_LEN, "cmp %s", ia32_emit_binop(irn, env));
 	lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F */", irn);
 	IA32_DO_EMIT(irn);
-	finish_CondJmp(F, irn, get_irn_mode(get_irn_n(irn, 2)));
+	finish_CondJmp(F, irn, get_ia32_res_mode(irn));
 }
 
 /**
@@ -646,13 +896,6 @@ static void emit_ia32_CondJmp(const ir_node *irn, ia32_emit_env_t *env) {
 	CondJmp_emitter(irn, env);
 }
 
-/**
- * Emits code for conditional jump with immediate.
- */
-void emit_ia32_CondJmp_i(const ir_node *irn, ia32_emit_env_t *env) {
-	CondJmp_emitter(irn, env);
-}
-
 /**
  * Emits code for conditional test and jump.
  */
@@ -673,7 +916,7 @@ static void TestJmp_emitter(const ir_node *irn, ia32_emit_env_t *env) {
 	lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F */", irn);
 
 	IA32_DO_EMIT(irn);
-	finish_CondJmp(F, irn, get_irn_mode(get_irn_n(irn, 0)));
+	finish_CondJmp(F, irn, get_ia32_res_mode(irn));
 
 #undef IA32_IS_IMMOP
 }
@@ -691,9 +934,9 @@ static void emit_ia32_CJmp(const ir_node *irn, ia32_emit_env_t *env) {
 	char cmnt_buf[SNPRINTF_BUF_LEN];
 
 	snprintf(cmd_buf, SNPRINTF_BUF_LEN, " ");
-	lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F omitted redundant test/cmp */", irn);
+	lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F omitted redundant test */", irn);
 	IA32_DO_EMIT(irn);
-	finish_CondJmp(F, irn, get_irn_mode(get_irn_n(irn, 0)));
+	finish_CondJmp(F, irn, get_ia32_res_mode(irn));
 }
 
 static void emit_ia32_CJmpAM(const ir_node *irn, ia32_emit_env_t *env) {
@@ -704,7 +947,272 @@ static void emit_ia32_CJmpAM(const ir_node *irn, ia32_emit_env_t *env) {
 	snprintf(cmd_buf, SNPRINTF_BUF_LEN, " ");
 	lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F omitted redundant test/cmp */", irn);
 	IA32_DO_EMIT(irn);
-	finish_CondJmp(F, irn, get_irn_mode(get_irn_n(irn, 2)));
+	finish_CondJmp(F, irn, get_ia32_res_mode(irn));
+}
+
+/**
+ * Emits code for conditional SSE floating point jump with two variables.
+ */
+static void emit_ia32_xCondJmp(ir_node *irn, ia32_emit_env_t *env) {
+	FILE *F = env->out;
+	char cmd_buf[SNPRINTF_BUF_LEN];
+	char cmnt_buf[SNPRINTF_BUF_LEN];
+
+	lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "ucomis%M %s", irn, ia32_emit_binop(irn, env));
+	lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F */", irn);
+	IA32_DO_EMIT(irn);
+	finish_CondJmp(F, irn, mode_F);
+
+}
+
+/**
+ * Emits code for conditional x87 floating point jump with two variables.
+ */
+static void emit_ia32_x87CondJmp(ir_node *irn, ia32_emit_env_t *env) {
+	FILE *F = env->out;
+	char cmd_buf[SNPRINTF_BUF_LEN];
+	char cmnt_buf[SNPRINTF_BUF_LEN];
+	ia32_attr_t *attr = get_ia32_attr(irn);
+	const char *reg = attr->x87[1]->name;
+	const char *instr = "fcom";
+	int reverse = 0;
+
+	switch (get_ia32_pncode(irn)) {
+	case iro_ia32_fcomrJmp:
+		reverse = 1;
+	case iro_ia32_fcomJmp:
+	default:
+		instr = "fucom";
+		break;
+	case iro_ia32_fcomrpJmp:
+		reverse = 1;
+	case iro_ia32_fcompJmp:
+		instr = "fucomp";
+		break;
+	case iro_ia32_fcomrppJmp:
+		reverse = 1;
+	case iro_ia32_fcomppJmp:
+		instr = "fucompp";
+		reg = "";
+		break;
+	}
+
+	if (reverse)
+		set_ia32_pncode(irn, (long)get_negated_pnc(get_ia32_pncode(irn), mode_Is));
+
+	snprintf(cmd_buf, SNPRINTF_BUF_LEN, "%s %%%s", instr, reg);
+	lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F */", irn);
+	IA32_DO_EMIT(irn);
+	lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "fnstsw %%ax", irn);
+	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* Store x87 FPU Control Word */");
+	IA32_DO_EMIT(irn);
+	snprintf(cmd_buf, SNPRINTF_BUF_LEN, "sahf");
+	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* Store ah into flags */");
+	IA32_DO_EMIT(irn);
+
+	finish_CondJmp(F, irn, mode_Is);
+}
+
+static void CMov_emitter(ir_node *irn, ia32_emit_env_t *env) {
+	FILE               *F       = env->out;
+	const lc_arg_env_t *arg_env = ia32_get_arg_env();
+	ir_mode    *mode       = get_irn_mode(get_irn_n(irn, 0));
+	int        is_unsigned = mode_is_float(mode) || ! mode_is_signed(mode);
+	const char *cmp_suffix = get_cmp_suffix(get_ia32_pncode(irn), is_unsigned);
+	int is_PsiCondCMov     = is_ia32_PsiCondCMov(irn);
+
+	char cmd_buf[SNPRINTF_BUF_LEN];
+	char cmnt_buf[SNPRINTF_BUF_LEN];
+	const arch_register_t *in1, *in2, *out;
+
+	out = arch_get_irn_register(env->arch_env, irn);
+	in1 = arch_get_irn_register(env->arch_env, get_irn_n(irn, 2 - is_PsiCondCMov));
+	in2 = arch_get_irn_register(env->arch_env, get_irn_n(irn, 3 - is_PsiCondCMov));
+
+	/* we have to emit the cmp first, because the destination register */
+	/* could be one of the compare registers                           */
+	if (is_ia32_CmpCMov(irn)) {
+		lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "cmp %1S, %2S", irn, irn);
+	}
+	else if (is_ia32_xCmpCMov(irn)) {
+		lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "ucomis%M %1S, %2S", get_irn_n(irn, 0), irn, irn);
+	}
+	else if (is_PsiCondCMov) {
+		/* omit compare because flags are already set by And/Or */
+		snprintf(cmd_buf, SNPRINTF_BUF_LEN, " ");
+	}
+	else {
+		assert(0 && "unsupported CMov");
+	}
+	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* Psi condition */" );
+	IA32_DO_EMIT(irn);
+
+	if (REGS_ARE_EQUAL(out, in2)) {
+		/* best case: default in == out -> do nothing */
+	}
+	else if (REGS_ARE_EQUAL(out, in1)) {
+		/* true in == out -> need complement compare and exchange true and default in */
+		ir_node *t = get_irn_n(irn, 2);
+		set_irn_n(irn, 2, get_irn_n(irn, 3));
+		set_irn_n(irn, 3, t);
+
+		cmp_suffix  = get_cmp_suffix(get_inversed_pnc(get_ia32_pncode(irn)), is_unsigned);
+
+	}
+	else {
+		/* out is different from in: need copy default -> out */
+		lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "mov %1D, %4S", irn, irn);
+		lc_esnprintf(arg_env, cmnt_buf, SNPRINTF_BUF_LEN, "/* copy default -> out */" );
+		IA32_DO_EMIT(irn);
+	}
+
+	lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "cmov%s %1D, %3S", cmp_suffix, irn, irn);
+	lc_esnprintf(arg_env, cmnt_buf, SNPRINTF_BUF_LEN, "/* condition is true case */" );
+	IA32_DO_EMIT(irn);
+}
+
+static void emit_ia32_CmpCMov(ir_node *irn, ia32_emit_env_t *env) {
+	CMov_emitter(irn, env);
+}
+
+static void emit_ia32_PsiCondCMov(ir_node *irn, ia32_emit_env_t *env) {
+	CMov_emitter(irn, env);
+}
+
+static void emit_ia32_xCmpCMov(ir_node *irn, ia32_emit_env_t *env) {
+	CMov_emitter(irn, env);
+}
+
+static void Set_emitter(ir_node *irn, ir_mode *mode, ia32_emit_env_t *env) {
+	FILE               *F       = env->out;
+	const lc_arg_env_t *arg_env = ia32_get_arg_env();
+	int        is_unsigned = mode_is_float(mode) || ! mode_is_signed(mode);
+	const char *cmp_suffix = get_cmp_suffix(get_ia32_pncode(irn), is_unsigned);
+	const char *reg8bit;
+
+	char cmd_buf[SNPRINTF_BUF_LEN];
+	char cmnt_buf[SNPRINTF_BUF_LEN];
+	const arch_register_t *out;
+
+	out     = arch_get_irn_register(env->arch_env, irn);
+	reg8bit = ia32_get_mapped_reg_name(env->isa->regs_8bit, out);
+
+	if (is_ia32_CmpSet(irn)) {
+		lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "cmp %s", ia32_emit_binop(irn, env));
+	}
+	else if (is_ia32_xCmpSet(irn)) {
+		lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "ucomis%M %s", get_irn_n(irn, 2), ia32_emit_binop(irn, env));
+	}
+	else if (is_ia32_PsiCondSet(irn)) {
+		/* omit compare because flags are already set by And/Or */
+		snprintf(cmd_buf, SNPRINTF_BUF_LEN, " ");
+	}
+	else {
+		assert(0 && "unsupported Set");
+	}
+	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* calculate Psi condition */" );
+	IA32_DO_EMIT(irn);
+
+	/* use mov to clear target because it doesn't affect the eflags */
+	snprintf(cmd_buf, SNPRINTF_BUF_LEN, "mov %%%s, 0", arch_register_get_name(out));
+	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* clear target as set modifies only lower 8 bit */");
+	IA32_DO_EMIT(irn);
+
+	snprintf(cmd_buf, SNPRINTF_BUF_LEN, "set%s %%%s", cmp_suffix, reg8bit);
+	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* set 1 iff true, 0 otherweise */" );
+	IA32_DO_EMIT(irn);
+}
+
+static void emit_ia32_CmpSet(ir_node *irn, ia32_emit_env_t *env) {
+	Set_emitter(irn, get_irn_mode(get_irn_n(irn, 2)), env);
+}
+
+static void emit_ia32_PsiCondSet(ir_node *irn, ia32_emit_env_t *env) {
+	Set_emitter(irn, get_irn_mode(get_irn_n(irn, 0)), env);
+}
+
+static void emit_ia32_xCmpSet(ir_node *irn, ia32_emit_env_t *env) {
+	Set_emitter(irn, get_irn_mode(get_irn_n(irn, 2)), env);
+}
+
+static void emit_ia32_xCmp(ir_node *irn, ia32_emit_env_t *env) {
+	FILE               *F       = env->out;
+	const lc_arg_env_t *arg_env = ia32_get_arg_env();
+	int                sse_pnc  = -1;
+	long               pnc      = get_ia32_pncode(irn);
+	long               unord    = pnc & pn_Cmp_Uo;
+	char cmd_buf[SNPRINTF_BUF_LEN];
+	char cmnt_buf[SNPRINTF_BUF_LEN];
+
+	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;
+	}
+
+	assert(sse_pnc >= 0 && "unsupported compare");
+
+	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
+		*/
+		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "sub %%esp, 8");
+		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* reserve some space for unordered compare result */");
+		IA32_DO_EMIT(NULL);
+		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "cmpsd %s, 3", ia32_emit_binop(irn, env));
+		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* SSE compare: unordered */");
+		IA32_DO_EMIT(NULL);
+		lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "movsd [%%esp], %1D", irn);
+		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* store compare result */");
+		IA32_DO_EMIT(NULL);
+	}
+
+	snprintf(cmd_buf, SNPRINTF_BUF_LEN, "cmpsd %s, %d", ia32_emit_binop(irn, env), sse_pnc);
+	lc_esnprintf(arg_env, cmnt_buf, SNPRINTF_BUF_LEN, "/* SSE compare (%+F) with result in %1D */", irn, irn);
+	IA32_DO_EMIT(irn);
+
+	if (unord && sse_pnc != 3) {
+		lc_esnprintf(arg_env, cmd_buf, SNPRINTF_BUF_LEN, "andpd %1D, [%%esp]", irn);
+		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* build the final result */");
+		IA32_DO_EMIT(NULL);
+		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "add %%esp, 8");
+		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* free allocated space */");
+		IA32_DO_EMIT(NULL);
+	}
 }
 
 /*********************************************************
@@ -752,10 +1260,10 @@ static int ia32_cmp_branch_t(const void *a, const void *b) {
  * possible otherwise a cmp-jmp cascade). Port from
  * cggg ia32 backend
  */
-void emit_ia32_SwitchJmp(const ir_node *irn, ia32_emit_env_t *emit_env) {
+static void emit_ia32_SwitchJmp(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	unsigned long       interval;
 	char                buf[SNPRINTF_BUF_LEN];
-	int                 last_value, i, pn, do_jmp_tbl = 1;
+	int                 last_value, i, pn;
 	jmp_tbl_t           tbl;
 	ir_node            *proj;
 	const ir_edge_t    *edge;
@@ -765,7 +1273,7 @@ void emit_ia32_SwitchJmp(const ir_node *irn, ia32_emit_env_t *emit_env) {
 
 	/* fill the table structure */
 	tbl.label        = xmalloc(SNPRINTF_BUF_LEN);
-	tbl.label        = get_unique_label(tbl.label, SNPRINTF_BUF_LEN, "JMPTBL_");
+	tbl.label        = get_unique_label(tbl.label, SNPRINTF_BUF_LEN, ".TBL_");
 	tbl.defProj      = NULL;
 	tbl.num_branches = get_irn_n_edges(irn);
 	tbl.branches     = xcalloc(tbl.num_branches, sizeof(tbl.branches[0]));
@@ -802,83 +1310,48 @@ void emit_ia32_SwitchJmp(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	/* two-complement's magic make this work without overflow */
 	interval = tbl.max_value - tbl.min_value;
 
-	/* check value interval */
-	if (interval > 16 * 1024) {
-		do_jmp_tbl = 0;
-	}
-
-	/* check ratio of value interval to number of branches */
-	if ((float)(interval + 1) / (float)tbl.num_branches > 8.0) {
-		do_jmp_tbl = 0;
-	}
+	/* emit the table */
+	lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "cmp %1S, %u", irn, interval);
+	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* compare for switch */");
+	IA32_DO_EMIT(irn);
 
-	if (do_jmp_tbl) {
-		/* emit the table */
-		if (tbl.min_value != 0) {
-			lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "cmpl %lu, -%d(%1S)",
-				interval, tbl.min_value, irn);
-			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* first switch value is not 0 */");
+	snprintf(cmd_buf, SNPRINTF_BUF_LEN, "ja %s", get_cfop_target(tbl.defProj, buf));
+	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* default jump if out of range  */");
+	IA32_DO_EMIT(irn);
 
-			IA32_DO_EMIT(irn);
-		}
-		else {
-			lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "cmpl %lu, %1S", interval, irn);
-			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* compare for switch */");
+	if (tbl.num_branches > 1) {
+		/* create table */
 
-			IA32_DO_EMIT(irn);
-		}
-
-		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "ja %s", get_cfop_target(tbl.defProj, buf));
-		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* default jump if out of range  */");
+		lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "jmp %s[%1S*4]", tbl.label, irn);
+		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* get jump table entry as target */");
 		IA32_DO_EMIT(irn);
 
-		if (tbl.num_branches > 1) {
-			/* create table */
-
-			lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "jmp [%1S*4+%s]", irn, tbl.label);
-			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* get jump table entry as target */");
-			IA32_DO_EMIT(irn);
-
-			fprintf(F, "\t.section\t.rodata\n");
-			fprintf(F, "\t.align 4\n");
+		ia32_switch_section(F, SECTION_RODATA);
+		fprintf(F, "\t.align 4\n");
 
-			fprintf(F, "%s:\n", tbl.label);
+		fprintf(F, "%s:\n", tbl.label);
 
-			snprintf(cmd_buf, SNPRINTF_BUF_LEN, ".long %s", get_cfop_target(tbl.branches[0].target, buf));
-			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* case %d */\n",  tbl.branches[0].value);
-			IA32_DO_EMIT(irn);
+		snprintf(cmd_buf, SNPRINTF_BUF_LEN, ".long %s", get_cfop_target(tbl.branches[0].target, buf));
+		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* case %d */",  tbl.branches[0].value);
+		IA32_DO_EMIT(irn);
 
-			last_value = tbl.branches[0].value;
-			for (i = 1; i < tbl.num_branches; ++i) {
-				while (++last_value < tbl.branches[i].value) {
-					snprintf(cmd_buf, SNPRINTF_BUF_LEN, ".long %s", get_cfop_target(tbl.defProj, buf));
-					snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* default case */");
-					IA32_DO_EMIT(irn);
-				}
-				snprintf(cmd_buf, SNPRINTF_BUF_LEN, ".long %s", get_cfop_target(tbl.branches[i].target, buf));
-				snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* case %d */", last_value);
+		last_value = tbl.branches[0].value;
+		for (i = 1; i < tbl.num_branches; ++i) {
+			while (++last_value < tbl.branches[i].value) {
+				snprintf(cmd_buf, SNPRINTF_BUF_LEN, ".long %s", get_cfop_target(tbl.defProj, buf));
+				snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* default case */");
 				IA32_DO_EMIT(irn);
 			}
-
-			fprintf(F, "\t.text");
-		}
-		else {
-			/* one jump is enough */
-			snprintf(cmd_buf, SNPRINTF_BUF_LEN, "jmp %s", get_cfop_target(tbl.branches[0].target, buf));
-			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* only one case given */");
+			snprintf(cmd_buf, SNPRINTF_BUF_LEN, ".long %s", get_cfop_target(tbl.branches[i].target, buf));
+			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* case %d */", last_value);
 			IA32_DO_EMIT(irn);
 		}
+		ia32_switch_section(F, SECTION_TEXT);
 	}
-	else { // no jump table
-		for (i = 0; i < tbl.num_branches; ++i) {
-			lc_esnprintf(env, cmd_buf, SNPRINTF_BUF_LEN, "cmpl %d, %1S", tbl.branches[i].value, irn);
-			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* case %d */", i);
-			IA32_DO_EMIT(irn);
-			fprintf(F, "\tje %s\n", get_cfop_target(tbl.branches[i].target, buf));
-		}
-
-		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "jmp %s", get_cfop_target(tbl.defProj, buf));
-		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* default case */");
+	else {
+		/* one jump is enough */
+		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "jmp %s", get_cfop_target(tbl.branches[0].target, buf));
+		snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* only one case given */");
 		IA32_DO_EMIT(irn);
 	}
 
@@ -891,7 +1364,7 @@ void emit_ia32_SwitchJmp(const ir_node *irn, ia32_emit_env_t *emit_env) {
 /**
  * Emits code for a unconditional jump.
  */
-void emit_Jmp(const ir_node *irn, ia32_emit_env_t *env) {
+static void emit_Jmp(const ir_node *irn, ia32_emit_env_t *env) {
 	ir_node *block, *next_bl;
 	FILE *F = env->out;
 	char buf[SNPRINTF_BUF_LEN], cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
@@ -926,7 +1399,7 @@ void emit_Jmp(const ir_node *irn, ia32_emit_env_t *env) {
 /**
  * Emits code for a proj -> node
  */
-void emit_Proj(const ir_node *irn, ia32_emit_env_t *env) {
+static void emit_Proj(const ir_node *irn, ia32_emit_env_t *env) {
 	ir_node *pred = get_Proj_pred(irn);
 
 	if (get_irn_op(pred) == op_Start) {
@@ -954,25 +1427,27 @@ void emit_Proj(const ir_node *irn, ia32_emit_env_t *env) {
 /**
  * Emit movsb/w instructions to make mov count divideable by 4
  */
-static void emit_CopyB_prolog(FILE *F, int rem, int size) {
+static void emit_CopyB_prolog(FILE *F, const ir_node *irn, int rem) {
 	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 
-	fprintf(F, "\t/* memcopy %d bytes*/\n", size);
+	ir_fprintf(F, "\t/* memcopy prolog %+F */\n", irn);
 
 	snprintf(cmd_buf, SNPRINTF_BUF_LEN, "cld");
-	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* copy direction forward*/");
-	IA32_DO_EMIT(NULL);
+	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* copy direction forward */");
 
 	switch(rem) {
 		case 1:
+			IA32_DO_EMIT(NULL);
 			snprintf(cmd_buf, SNPRINTF_BUF_LEN, "movsb");
 			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* memcopy remainder 1 */");
 			break;
 		case 2:
+			IA32_DO_EMIT(NULL);
 			snprintf(cmd_buf, SNPRINTF_BUF_LEN, "movsw");
 			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* memcopy remainder 2 */");
 			break;
 		case 3:
+			IA32_DO_EMIT(NULL);
 			snprintf(cmd_buf, SNPRINTF_BUF_LEN, "movsb");
 			snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* memcopy remainder 3 */");
 			IA32_DO_EMIT(NULL);
@@ -987,14 +1462,13 @@ static void emit_CopyB_prolog(FILE *F, int rem, int size) {
 /**
  * Emit rep movsd instruction for memcopy.
  */
-void emit_ia32_CopyB(const ir_node *irn, ia32_emit_env_t *emit_env) {
-	FILE   *F    = emit_env->out;
-	tarval *tv   = get_ia32_Immop_tarval(irn);
-	int     rem  = get_tarval_long(tv);
-	int     size = get_tarval_long(get_ia32_Immop_tarval(get_irn_n(irn, 2)));
+static void emit_ia32_CopyB(const ir_node *irn, ia32_emit_env_t *emit_env) {
+	FILE   *F  = emit_env->out;
+	tarval *tv = get_ia32_Immop_tarval(irn);
+	int    rem = get_tarval_long(tv);
 	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 
-	emit_CopyB_prolog(F, rem, size);
+	emit_CopyB_prolog(F, irn, rem);
 
 	snprintf(cmd_buf, SNPRINTF_BUF_LEN, "rep movsd");
 	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* memcopy */");
@@ -1004,13 +1478,13 @@ void emit_ia32_CopyB(const ir_node *irn, ia32_emit_env_t *emit_env) {
 /**
  * Emits unrolled memcopy.
  */
-void emit_ia32_CopyB_i(const ir_node *irn, ia32_emit_env_t *emit_env) {
+static void emit_ia32_CopyB_i(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	tarval *tv   = get_ia32_Immop_tarval(irn);
 	int     size = get_tarval_long(tv);
 	FILE   *F    = emit_env->out;
 	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 
-	emit_CopyB_prolog(F, size & 0x3, size);
+	emit_CopyB_prolog(F, irn, size & 0x3);
 
 	size >>= 2;
 	while (size--) {
@@ -1036,15 +1510,13 @@ void emit_ia32_CopyB_i(const ir_node *irn, ia32_emit_env_t *emit_env) {
  * Emit code for conversions (I, FP), (FP, I) and (FP, FP).
  */
 static void emit_ia32_Conv_with_FP(const ir_node *irn, ia32_emit_env_t *emit_env) {
-	FILE               *F    = emit_env->out;
-	const lc_arg_env_t *env  = ia32_get_arg_env();
+	FILE               *F        = emit_env->out;
+	const lc_arg_env_t *env      = ia32_get_arg_env();
+	ir_mode	           *src_mode = get_ia32_src_mode(irn);
+	ir_mode            *tgt_mode = get_ia32_tgt_mode(irn);
 	char               *from, *to, buf[64];
-	ir_mode *src_mode, *tgt_mode;
 	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 
-	src_mode = is_ia32_AddrModeS(irn) ? get_ia32_ls_mode(irn) : get_irn_mode(get_irn_n(irn, 2));
-	tgt_mode = get_ia32_res_mode(irn);
-
 	from = mode_is_float(src_mode) ? (get_mode_size_bits(src_mode) == 32 ? "ss" : "sd") : "si";
 	to   = mode_is_float(tgt_mode) ? (get_mode_size_bits(tgt_mode) == 32 ? "ss" : "sd") : "si";
 
@@ -1064,33 +1536,32 @@ static void emit_ia32_Conv_with_FP(const ir_node *irn, ia32_emit_env_t *emit_env
 	IA32_DO_EMIT(irn);
 }
 
-void emit_ia32_Conv_I2FP(const ir_node *irn, ia32_emit_env_t *emit_env) {
+static void emit_ia32_Conv_I2FP(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	emit_ia32_Conv_with_FP(irn, emit_env);
 }
 
-void emit_ia32_Conv_FP2I(const ir_node *irn, ia32_emit_env_t *emit_env) {
+static void emit_ia32_Conv_FP2I(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	emit_ia32_Conv_with_FP(irn, emit_env);
 }
 
-void emit_ia32_Conv_FP2FP(const ir_node *irn, ia32_emit_env_t *emit_env) {
+static void emit_ia32_Conv_FP2FP(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	emit_ia32_Conv_with_FP(irn, emit_env);
 }
 
 /**
  * Emits code for an Int conversion.
  */
-void emit_ia32_Conv_I2I(const ir_node *irn, ia32_emit_env_t *emit_env) {
-	FILE               *F    = emit_env->out;
-	const lc_arg_env_t *env  = ia32_get_arg_env();
-	char *move_cmd, *conv_cmd;
-	ir_mode *src_mode, *tgt_mode;
+static void emit_ia32_Conv_I2I(const ir_node *irn, ia32_emit_env_t *emit_env) {
+	FILE               *F        = emit_env->out;
+	const lc_arg_env_t *env      = ia32_get_arg_env();
+	char               *move_cmd = "movzx";
+	char               *conv_cmd = NULL;
+	ir_mode	           *src_mode = get_ia32_src_mode(irn);
+	ir_mode            *tgt_mode = get_ia32_tgt_mode(irn);
 	int n, m;
 	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 	const arch_register_t *in_reg, *out_reg;
 
-	src_mode = is_ia32_AddrModeS(irn) ? get_ia32_ls_mode(irn) : get_irn_mode(get_irn_n(irn, 2));
-	tgt_mode = get_ia32_res_mode(irn);
-
 	n = get_mode_size_bits(src_mode);
 	m = get_mode_size_bits(tgt_mode);
 
@@ -1103,12 +1574,8 @@ void emit_ia32_Conv_I2I(const ir_node *irn, ia32_emit_env_t *emit_env) {
 		else
 			assert(0 && "unsupported Conv_I2I");
 	}
-	else {
-		move_cmd = "movzx";
-		conv_cmd = NULL;
-	}
 
-	switch(get_ia32_op_type(irn)) {
+	 switch(get_ia32_op_type(irn)) {
 		case ia32_Normal:
 			in_reg  = get_in_reg(irn, 2);
 			out_reg = get_out_reg(irn, 0);
@@ -1150,6 +1617,14 @@ void emit_ia32_Conv_I2I(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	IA32_DO_EMIT(irn);
 }
 
+/**
+ * Emits code for an 8Bit Int conversion.
+ */
+void emit_ia32_Conv_I2I8Bit(const ir_node *irn, ia32_emit_env_t *emit_env) {
+	emit_ia32_Conv_I2I(irn, emit_env);
+}
+
+
 /*******************************************
  *  _                          _
  * | |                        | |
@@ -1163,16 +1638,16 @@ void emit_ia32_Conv_I2I(const ir_node *irn, ia32_emit_env_t *emit_env) {
 /**
  * Emits a backend call
  */
-void emit_be_Call(const ir_node *irn, ia32_emit_env_t *emit_env) {
+static void emit_be_Call(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	FILE *F = emit_env->out;
 	entity *ent = be_Call_get_entity(irn);
 	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 
 	if (ent) {
-		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "call %s", get_entity_name(ent));
+		snprintf(cmd_buf, SNPRINTF_BUF_LEN, "call %s", get_entity_ld_name(ent));
 	}
 	else {
-		lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "%1D", get_irn_n(irn, be_pos_Call_ptr));
+		lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "call %1D", get_irn_n(irn, be_pos_Call_ptr));
 	}
 
 	lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F (be_Call) */", irn);
@@ -1183,15 +1658,17 @@ void emit_be_Call(const ir_node *irn, ia32_emit_env_t *emit_env) {
 /**
  * Emits code to increase stack pointer.
  */
-void emit_be_IncSP(const ir_node *irn, ia32_emit_env_t *emit_env) {
+static void emit_be_IncSP(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	FILE          *F    = emit_env->out;
 	unsigned       offs = be_get_IncSP_offset(irn);
 	be_stack_dir_t dir  = be_get_IncSP_direction(irn);
 	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 
 	if (offs) {
-		lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "add %1S,%s%u", irn,
-			(dir == be_stack_dir_along) ? " -" : " ", offs);
+		if (dir == be_stack_dir_expand)
+			lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "sub %1S, %u", irn, offs);
+		else
+			lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "add %1S, %u", irn, offs);
 		lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F (IncSP) */", irn);
 	}
 	else {
@@ -1205,7 +1682,7 @@ void emit_be_IncSP(const ir_node *irn, ia32_emit_env_t *emit_env) {
 /**
  * Emits code to set stack pointer.
  */
-void emit_be_SetSP(const ir_node *irn, ia32_emit_env_t *emit_env) {
+static void emit_be_SetSP(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	FILE *F = emit_env->out;
 	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 
@@ -1215,29 +1692,136 @@ void emit_be_SetSP(const ir_node *irn, ia32_emit_env_t *emit_env) {
 }
 
 /**
- * Emits code for Copy.
+ * Emits code for Copy/CopyKeep.
  */
-void emit_be_Copy(const ir_node *irn, ia32_emit_env_t *emit_env) {
-	FILE *F = emit_env->out;
+static void Copy_emitter(const ir_node *irn, ir_node *op, ia32_emit_env_t *emit_env) {
+	FILE             *F    = emit_env->out;
+	const arch_env_t *aenv = emit_env->arch_env;
 	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
 
-	lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "mov %1D, %1S", irn, irn);
+	if (REGS_ARE_EQUAL(arch_get_irn_register(aenv, irn), arch_get_irn_register(aenv, op)) ||
+		be_is_unknown_reg(arch_get_irn_register(aenv, op)))
+		return;
+
+	if (mode_is_float(get_irn_mode(irn)))
+		lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "movs%M %1D, %1S", irn, irn, irn);
+	else
+		lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "mov %1D, %1S", irn, irn);
 	lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F */", irn);
 	IA32_DO_EMIT(irn);
 }
 
+static void emit_be_Copy(const ir_node *irn, ia32_emit_env_t *emit_env) {
+	Copy_emitter(irn, be_get_Copy_op(irn), emit_env);
+}
+
+static void emit_be_CopyKeep(const ir_node *irn, ia32_emit_env_t *emit_env) {
+	Copy_emitter(irn, be_get_CopyKeep_op(irn), emit_env);
+}
+
 /**
  * Emits code for exchange.
  */
-void emit_be_Perm(const ir_node *irn, ia32_emit_env_t *emit_env) {
+static void emit_be_Perm(const ir_node *irn, ia32_emit_env_t *emit_env) {
 	FILE *F = emit_env->out;
 	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
+	const arch_register_t *in1, *in2;
+	const arch_register_class_t *cls1, *cls2;
+
+	in1 = arch_get_irn_register(emit_env->arch_env, get_irn_n(irn, 0));
+	in2 = arch_get_irn_register(emit_env->arch_env, get_irn_n(irn, 1));
+
+	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]) {
+		lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "xchg %1S, %2S", irn, irn);
+	}
+	else if (cls1 == &ia32_reg_classes[CLASS_ia32_xmm]) {
+		lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN,
+			"pxor %1S, %2S\n\tpxor %2S, %1S\n\tpxor %1S, %2S", irn, irn, irn, irn, irn, irn);
+	}
+	else if (cls1 == &ia32_reg_classes[CLASS_ia32_vfp]) {
+		assert(0 && "Perm with vfp should not happen");
+	}
+	else if (cls1 == &ia32_reg_classes[CLASS_ia32_st]) {
+		assert(0 && "Perm with st(X) should not happen");
+	}
 
-	lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "xchg %1S, %2S", irn, irn);
 	lc_esnprintf(ia32_get_arg_env(), cmnt_buf, SNPRINTF_BUF_LEN, "/* %+F(%1A, %2A) */", irn, irn, irn);
 	IA32_DO_EMIT(irn);
 }
 
+/**
+ * Emits code for Constant loading.
+ */
+static void emit_ia32_Const(const ir_node *n, ia32_emit_env_t *env) {
+	FILE *F = env->out;
+	char cmd_buf[256], cmnt_buf[256];
+	const lc_arg_env_t *arg_env = ia32_get_arg_env();
+
+	if (get_ia32_Immop_tarval(n) == get_tarval_null(get_irn_mode(n))) {
+		const char *instr = "xor";
+		if (env->isa->opt_arch == arch_pentium_4) {
+			/* P4 prefers sub r, r, others xor r, r */
+			instr = "sub";
+		}
+		lc_esnprintf(arg_env, cmd_buf, 256, "%s %1D, %1D ", instr, n, n);
+		lc_esnprintf(arg_env, cmnt_buf, 256, "/* optimized mov 0 to register */");
+	}
+	else {
+		if (get_ia32_op_type(n) == ia32_SymConst) {
+			lc_esnprintf(arg_env, cmd_buf, 256, "mov %1D, OFFSET FLAT:%C ", n, n);
+			lc_esnprintf(arg_env, cmnt_buf, 256, "/* Move address of SymConst into register */");
+		}
+		else {
+			lc_esnprintf(arg_env, cmd_buf, 256, "mov %1D, %C ", n, n);
+			lc_esnprintf(arg_env, cmnt_buf, 256, "/* Mov Const into register */");
+		}
+	}
+	lc_efprintf(arg_env, F, "\t%-35s %-60s /* %+F (%+G) */\n", cmd_buf, cmnt_buf, n, n);
+}
+
+/**
+ * Emits code to increase stack pointer.
+ */
+static void emit_ia32_AddSP(const ir_node *irn, ia32_emit_env_t *emit_env) {
+	FILE          *F = emit_env->out;
+	char cmd_buf[SNPRINTF_BUF_LEN], cmnt_buf[SNPRINTF_BUF_LEN];
+
+	if (is_ia32_ImmConst(irn)) {
+		lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "sub %1D, %C", irn, irn);
+	}
+	else if (is_ia32_ImmSymConst(irn)) {
+		if (get_ia32_op_type(irn) == ia32_Normal)
+			lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "sub %1D, OFFSET_FLAT:%C", irn, irn);
+		else /* source address mode */
+			lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "sub %1D, [%s%s]", irn, get_id_str(get_ia32_am_sc(irn)), get_ia32_am_offs(irn));
+	}
+	else {
+		lc_esnprintf(ia32_get_arg_env(), cmd_buf, SNPRINTF_BUF_LEN, "sub %1D, %2S", irn, irn);
+	}
+	snprintf(cmnt_buf, SNPRINTF_BUF_LEN, "/* reserve space on stack */");
+
+	IA32_DO_EMIT(irn);
+}
+
+static void emit_be_Return(const ir_node *n, ia32_emit_env_t *env) {
+	FILE *F = env->out;
+	const lc_arg_env_t *arg_env = ia32_get_arg_env();
+
+	lc_efprintf(arg_env, F, "\t%-35s %-60s /* %+F (%+G) */\n", "ret", "/* be_Return */", n, n);
+}
+
+static void emit_Nothing(const ir_node *n, ia32_emit_env_t *env) {
+	FILE *F = env->out;
+
+	ir_fprintf(F, "\t%35s /* %+F (%+G) */\n", " ", n, n);
+}
+
+
 /***********************************************************************************
  *                  _          __                                             _
  *                 (_)        / _|                                           | |
@@ -1254,9 +1838,12 @@ void emit_be_Perm(const ir_node *irn, ia32_emit_env_t *emit_env) {
  */
 static void ia32_register_emitters(void) {
 
-#define IA32_EMIT(a) op_ia32_##a->ops.generic = (op_func)emit_ia32_##a
-#define EMIT(a)      op_##a->ops.generic = (op_func)emit_##a
-#define BE_EMIT(a)   op_be_##a->ops.generic = (op_func)emit_be_##a
+#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
+#define IGN(a)			op_##a->ops.generic = (op_func)emit_Nothing
+#define BE_EMIT(a)      op_be_##a->ops.generic = (op_func)emit_be_##a
+#define BE_IGN(a)		op_be_##a->ops.generic = (op_func)emit_Nothing
 
 	/* first clear the generic function pointer for all ops */
 	clear_irp_opcodes_generic_func();
@@ -1269,6 +1856,10 @@ static void ia32_register_emitters(void) {
 	IA32_EMIT(TestJmp);
 	IA32_EMIT(CJmp);
 	IA32_EMIT(CJmpAM);
+	IA32_EMIT(CmpCMov);
+	IA32_EMIT(PsiCondCMov);
+	IA32_EMIT(CmpSet);
+	IA32_EMIT(PsiCondSet);
 	IA32_EMIT(SwitchJmp);
 	IA32_EMIT(CopyB);
 	IA32_EMIT(CopyB_i);
@@ -1276,31 +1867,53 @@ static void ia32_register_emitters(void) {
 	IA32_EMIT(Conv_FP2I);
 	IA32_EMIT(Conv_FP2FP);
 	IA32_EMIT(Conv_I2I);
+	IA32_EMIT(Conv_I2I8Bit);
+	IA32_EMIT(Const);
+	IA32_EMIT(AddSP);
+	IA32_EMIT(xCmp);
+	IA32_EMIT(xCmpSet);
+	IA32_EMIT(xCmpCMov);
+	IA32_EMIT(xCondJmp);
+	IA32_EMIT2(fcomJmp, x87CondJmp);
+	IA32_EMIT2(fcompJmp, x87CondJmp);
+	IA32_EMIT2(fcomppJmp, x87CondJmp);
+	IA32_EMIT2(fcomrJmp, x87CondJmp);
+	IA32_EMIT2(fcomrpJmp, x87CondJmp);
+	IA32_EMIT2(fcomrppJmp, x87CondJmp);
 
 	/* benode emitter */
 	BE_EMIT(Call);
 	BE_EMIT(IncSP);
 	BE_EMIT(SetSP);
 	BE_EMIT(Copy);
+	BE_EMIT(CopyKeep);
 	BE_EMIT(Perm);
+	BE_EMIT(Return);
+
+	BE_IGN(RegParams);
+	BE_IGN(Barrier);
+	BE_IGN(Keep);
 
 	/* firm emitter */
 	EMIT(Jmp);
 	EMIT(Proj);
+	IGN(Phi);
+	IGN(Start);
 
-#undef IA32_EMIT
 #undef BE_EMIT
 #undef EMIT
+#undef IGN
+#undef IA32_EMIT2
+#undef IA32_EMIT
 }
 
 /**
  * Emits code for a node.
  */
 static void ia32_emit_node(const ir_node *irn, void *env) {
-	ia32_emit_env_t        *emit_env = env;
-	firm_dbg_module_t *mod      = emit_env->mod;
-	FILE              *F        = emit_env->out;
+	ia32_emit_env_t   *emit_env = env;
 	ir_op             *op       = get_irn_op(irn);
+	DEBUG_ONLY(firm_dbg_module_t *mod = emit_env->mod;)
 
 	DBG((mod, LEVEL_1, "emitting code for %+F\n", irn));
 
@@ -1309,8 +1922,68 @@ static void ia32_emit_node(const ir_node *irn, void *env) {
 		(*emit)(irn, env);
 	}
 	else {
-		ir_fprintf(F, "\t%35s /* %+F */\n", " ", irn);
+		emit_Nothing(irn, env);
+		ir_fprintf(stderr, "Warning: No emit handler for node %+F (%+G)\n", irn, irn);
+	}
+}
+
+/**
+ * Emits gas alignment directives
+ */
+static void ia32_emit_alignment(FILE *F, unsigned align, unsigned skip) {
+	fprintf(F, "\t.p2align %u,,%u\n", align, skip);
+}
+
+/**
+ * Emits gas alignment directives for Functions depended on cpu architecture.
+ */
+static void ia32_emit_align_func(FILE *F, cpu_support cpu) {
+	unsigned align; unsigned maximum_skip;
+
+	/* gcc doesn't emit alignment for p4 ?*/
+    if (cpu == arch_pentium_4)
+		return;
+
+	switch (cpu) {
+		case arch_i386:
+			align = 2; maximum_skip = 3;
+			break;
+		case arch_i486:
+			align = 4; maximum_skip = 15;
+			break;
+		case arch_k6:
+			align = 5; maximum_skip = 31;
+			break;
+		default:
+			align = 4; maximum_skip = 15;
 	}
+	ia32_emit_alignment(F, align, maximum_skip);
+}
+
+/**
+ * Emits gas alignment directives for Labels depended on cpu architecture.
+ */
+static void ia32_emit_align_label(FILE *F, cpu_support cpu) {
+	unsigned align; unsigned maximum_skip;
+
+	/* gcc doesn't emit alignment for p4 ?*/
+	if (cpu == arch_pentium_4)
+		return;
+
+	switch (cpu) {
+		case arch_i386:
+			align = 2; maximum_skip = 3;
+			break;
+		case arch_i486:
+			align = 4; maximum_skip = 15;
+			break;
+		case arch_k6:
+			align = 5; maximum_skip = 7;
+			break;
+		default:
+			align = 4; maximum_skip = 7;
+	}
+	ia32_emit_alignment(F, align, maximum_skip);
 }
 
 /**
@@ -1318,12 +1991,57 @@ static void ia32_emit_node(const ir_node *irn, void *env) {
  * and emits code for each node.
  */
 static void ia32_gen_block(ir_node *block, void *env) {
+	ia32_emit_env_t *emit_env = env;
 	const ir_node *irn;
+	char cmd_buf[SNPRINTF_BUF_LEN];
+	char cmnt_buf[SNPRINTF_BUF_LEN];
+	int need_label = block != get_irg_start_block(get_irn_irg(block));
+	FILE *F = emit_env->out;
 
 	if (! is_Block(block))
 		return;
 
-	fprintf(((ia32_emit_env_t *)env)->out, BLOCK_PREFIX("%ld:\n"), get_irn_node_nr(block));
+	if (need_label && (emit_env->cg->opt & IA32_OPT_EXTBB)) {
+		/* if the extended block scheduler is used, only leader blocks need
+		   labels. */
+		need_label = (block == get_extbb_leader(get_nodes_extbb(block)));
+	}
+
+	if (need_label) {
+		int i, arity;
+		char* predstring = cmnt_buf;
+		size_t cmntsize;
+		int res;
+
+		ia32_emit_align_label(emit_env->out, emit_env->isa->opt_arch);
+
+		ir_snprintf(cmd_buf, sizeof(cmd_buf), BLOCK_PREFIX("%d:"),
+		            get_irn_node_nr(block));
+
+		/* emit list of pred blocks in comment */
+		cmntsize = sizeof(cmnt_buf);
+		res = snprintf(predstring, cmntsize, "/* preds: ");
+		cmntsize -= res;
+		predstring += res;
+
+		arity = get_irn_arity(block);
+		for(i = 0; i < arity; ++i) {
+			ir_node *predblock = get_Block_cfgpred_block(block, i);
+			res = snprintf(predstring, cmntsize, " %ld", get_irn_node_nr(predblock));
+			cmntsize -= res;
+			predstring += res;
+			if(cmntsize <= 3)
+				break;
+		}
+		if(cmntsize < 3) {
+			predstring = cmnt_buf + sizeof(cmnt_buf) - 3;
+			cmntsize = 3;
+		}
+		snprintf(predstring, cmntsize, "*/");
+		fprintf(F, "%-43s %-60s\n", cmd_buf, cmnt_buf);
+	}
+
+	/* emit the contents of the block */
 	sched_foreach(block, irn) {
 		ia32_emit_node(irn, env);
 	}
@@ -1332,15 +2050,17 @@ static void ia32_gen_block(ir_node *block, void *env) {
 /**
  * Emits code for function start.
  */
-static void ia32_emit_func_prolog(FILE *F, ir_graph *irg) {
+static void ia32_emit_func_prolog(FILE *F, ir_graph *irg, cpu_support cpu) {
 	entity     *irg_ent  = get_irg_entity(irg);
-	const char *irg_name = get_entity_name(irg_ent);
+	const char *irg_name = get_entity_ld_name(irg_ent);
 
-	fprintf(F, "\t.text\n");
+	fprintf(F, "\n");
+	ia32_switch_section(F, SECTION_TEXT);
+	ia32_emit_align_func(F, cpu);
 	if (get_entity_visibility(irg_ent) == visibility_external_visible) {
 		fprintf(F, ".globl %s\n", irg_name);
 	}
-	fprintf(F, "\t.type\t%s, @function\n", irg_name);
+	ia32_dump_function_object(F, irg_name);
 	fprintf(F, "%s:\n", irg_name);
 }
 
@@ -1348,10 +2068,10 @@ static void ia32_emit_func_prolog(FILE *F, ir_graph *irg) {
  * Emits code for function end
  */
 static void ia32_emit_func_epilog(FILE *F, ir_graph *irg) {
-	const char *irg_name = get_entity_name(get_irg_entity(irg));
+	const char *irg_name = get_entity_ld_name(get_irg_entity(irg));
 
-	fprintf(F, "\tret\n");
-	fprintf(F, "\t.size\t%s, .-%s\n\n", irg_name, irg_name);
+	ia32_dump_function_size(F, irg_name);
+	fprintf(F, "\n");
 }
 
 /**
@@ -1369,70 +2089,46 @@ static void ia32_gen_labels(ir_node *block, void *env) {
 	}
 }
 
-typedef struct {
-	ir_node *start;
-	ir_node *end;
-} anchor;
-
-/**
- * Ext-Block walker: create a block schedule
- */
-static void create_block_list(ir_extblk *blk, void *env) {
-	anchor *list = env;
-	int i, n;
-
-	for (i = 0, n = get_extbb_n_blocks(blk); i < n; ++i) {
-		ir_node *block = get_extbb_block(blk, i);
-
-		set_irn_link(block, NULL);
-		if (list->start)
-			set_irn_link(list->end, block);
-		else
-			list->start = block;
-
-		list->end = block;
-	}
-}
-
 /**
  * Main driver. Emits the code for one routine.
  */
 void ia32_gen_routine(FILE *F, ir_graph *irg, const ia32_code_gen_t *cg) {
 	ia32_emit_env_t emit_env;
-	anchor list;
 	ir_node *block;
 
-	emit_env.mod      = firm_dbg_register("firm.be.ia32.emitter");
 	emit_env.out      = F;
 	emit_env.arch_env = cg->arch_env;
 	emit_env.cg       = cg;
 	emit_env.isa      = (ia32_isa_t *)cg->arch_env->isa;
+	FIRM_DBG_REGISTER(emit_env.mod, "firm.be.ia32.emitter");
 
 	/* set the global arch_env (needed by print hooks) */
 	arch_env = cg->arch_env;
 
 	ia32_register_emitters();
 
-	ia32_emit_func_prolog(F, irg);
+	ia32_emit_func_prolog(F, irg, emit_env.isa->opt_arch);
 	irg_block_walk_graph(irg, ia32_gen_labels, NULL, &emit_env);
 
-	if (cg->opt.extbb) {
-		/* schedule extended basic blocks */
+	if ((cg->opt & IA32_OPT_EXTBB) && cg->blk_sched) {
+		int i, n = ARR_LEN(cg->blk_sched);
 
-		compute_extbb(irg);
+		for (i = 0; i < n;) {
+			ir_node *next_bl;
 
-		list.start = NULL;
-		list.end   = NULL;
-		irg_extblock_walk_graph(irg, NULL, create_block_list, &list);
+			block   = cg->blk_sched[i];
+			++i;
+			next_bl = i < n ? cg->blk_sched[i] : NULL;
 
-		have_block_sched = 1;
-		for (block = list.start; block; block = get_irn_link(block))
+			/* set here the link. the emitter expects to find the next block here */
+			set_irn_link(block, next_bl);
 			ia32_gen_block(block, &emit_env);
+		}
 	}
 	else {
-		/* "normal" block schedule */
-
-		have_block_sched = 0;
+		/* "normal" block schedule: Note the get_next_block() returns the NUMBER of the block
+		   in the block schedule. As this number should NEVER be equal the next block,
+		   we does not need a clear block link here. */
 		irg_walk_blkwise_graph(irg, NULL, ia32_gen_block, &emit_env);
 	}