- moved the imul mem,imm32 splitting into peephole optimizations
[libfirm] / ir / be / ia32 / ia32_emitter.c
index 04eed6d..21bd11a 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.
  *
@@ -42,6 +42,7 @@
 #include "execfreq.h"
 #include "error.h"
 #include "raw_bitset.h"
+#include "dbginfo.h"
 
 #include "../besched_t.h"
 #include "../benode_t.h"
@@ -50,6 +51,7 @@
 #include "../beemitter.h"
 #include "../begnuas.h"
 #include "../beirg_t.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;)
@@ -68,6 +71,8 @@ DEBUG_ONLY(static firm_dbg_module_t *dbg = NULL;)
 static const arch_env_t *arch_env;
 static const ia32_isa_t *isa;
 static ia32_code_gen_t  *cg;
+static int               do_pic;
+static char              pic_base_label[128];
 
 /**
  * Returns the register at in position pos.
@@ -257,7 +262,8 @@ static void ia32_emit_mode_suffix_mode(const ir_mode *mode)
                switch(get_mode_size_bits(mode)) {
                case 32: be_emit_char('s'); return;
                case 64: be_emit_char('l'); return;
-               case 80: be_emit_char('t'); return;
+               case 80:
+               case 96: be_emit_char('t'); return;
                }
        } else {
                assert(mode_is_int(mode) || mode_is_reference(mode));
@@ -291,8 +297,7 @@ void ia32_emit_x87_mode_suffix(const ir_node *node)
                ia32_emit_mode_suffix_mode(mode);
 }
 
-static
-char get_xmm_mode_suffix(ir_mode *mode)
+static char get_xmm_mode_suffix(ir_mode *mode)
 {
        assert(mode_is_float(mode));
        switch(get_mode_size_bits(mode)) {
@@ -332,45 +337,6 @@ void ia32_emit_extend_suffix(const ir_mode *mode)
        }
 }
 
-static
-void ia32_emit_function_object(const char *name)
-{
-       switch (be_gas_flavour) {
-       case GAS_FLAVOUR_NORMAL:
-               be_emit_cstring("\t.type\t");
-               be_emit_string(name);
-               be_emit_cstring(", @function\n");
-               be_emit_write_line();
-               break;
-       case GAS_FLAVOUR_MINGW:
-               be_emit_cstring("\t.def\t");
-               be_emit_string(name);
-               be_emit_cstring(";\t.scl\t2;\t.type\t32;\t.endef\n");
-               be_emit_write_line();
-               break;
-       default:
-               break;
-       }
-}
-
-static
-void ia32_emit_function_size(const char *name)
-{
-       switch (be_gas_flavour) {
-       case GAS_FLAVOUR_NORMAL:
-               be_emit_cstring("\t.size\t");
-               be_emit_string(name);
-               be_emit_cstring(", .-");
-               be_emit_string(name);
-               be_emit_char('\n');
-               be_emit_write_line();
-               break;
-       default:
-               break;
-       }
-}
-
-
 void ia32_emit_source_register_or_immediate(const ir_node *node, int pos)
 {
        ir_node *in = get_irn_n(node, pos);
@@ -458,16 +424,6 @@ void ia32_emit_x87_binop(const ir_node *node) {
        }
 }
 
-void ia32_emit_am_or_dest_register(const ir_node *node,
-                                   int pos) {
-       if(get_ia32_op_type(node) == ia32_Normal) {
-               ia32_emit_dest_register(node, pos);
-       } else {
-               assert(get_ia32_op_type(node) == ia32_AddrModeD);
-               ia32_emit_am(node);
-       }
-}
-
 /**
  * Emits registers and/or address mode of a unary operation.
  */
@@ -492,6 +448,29 @@ void ia32_emit_unop(const ir_node *node, int pos) {
        }
 }
 
+static void ia32_emit_entity(ir_entity *entity, int no_pic_adjust)
+{
+       ident *id;
+
+       set_entity_backend_marked(entity, 1);
+       id = get_entity_ld_ident(entity);
+       be_emit_ident(id);
+
+       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");
+               }
+       }
+
+       if (!no_pic_adjust && do_pic) {
+               /* TODO: only do this when necessary */
+               be_emit_char('-');
+               be_emit_string(pic_base_label);
+       }
+}
+
 /**
  * Emits address mode.
  */
@@ -508,21 +487,9 @@ void ia32_emit_am(const ir_node *node) {
 
        /* emit offset */
        if (ent != NULL) {
-               ident *id;
-
-               set_entity_backend_marked(ent, 1);
-               id = get_entity_ld_ident(ent);
                if (is_ia32_am_sc_sign(node))
                        be_emit_char('-');
-               be_emit_ident(id);
-
-               if(get_entity_owner(ent) == get_tls_type()) {
-                       if (get_entity_visibility(ent) == visibility_external_allocated) {
-                               be_emit_cstring("@INDNTPOFF");
-                       } else {
-                               be_emit_cstring("@NTPOFF");
-                       }
-               }
+               ia32_emit_entity(ent, 0);
        }
 
        if(offs != 0) {
@@ -563,6 +530,26 @@ void ia32_emit_am(const ir_node *node) {
        }
 }
 
+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);
+
+       be_emit_cstring("\timul");
+       ia32_emit_mode_suffix(node);
+       be_emit_char(' ');
+
+       ia32_emit_binop(node);
+
+       /* do we need the 3-address form? */
+       if(is_ia32_NoReg_GP(left) ||
+                       get_in_reg(node, n_ia32_IMul_left) != out_reg) {
+               be_emit_cstring(", ");
+               emit_register(out_reg, get_ia32_ls_mode(node));
+       }
+       be_emit_finish_line_gas(node);
+}
+
 /*************************************************
  *                 _ _                         _
  *                (_) |                       | |
@@ -677,22 +664,24 @@ static int determine_final_pnc(const ir_node *node, int flags_pos,
                }
 
                flags_attr = get_ia32_attr_const(cmp);
-               if(flags_attr->data.cmp_flipped)
+               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.cmp_flipped)
+               if(flags_attr->data.ins_permuted)
                        pnc = get_mirrored_pnc(pnc);
                pnc |= ia32_pn_Cmp_float;
        } else {
+#if 0
                assert(is_ia32_Cmp(flags) || is_ia32_Test(flags)
                                || is_ia32_Cmp8Bit(flags) || is_ia32_Test8Bit(flags));
+#endif
                flags_attr = get_ia32_attr_const(flags);
 
-               if(flags_attr->data.cmp_flipped)
+               if(flags_attr->data.ins_permuted)
                        pnc = get_mirrored_pnc(pnc);
                if(flags_attr->data.cmp_unsigned)
                        pnc |= ia32_pn_Cmp_unsigned;
@@ -721,25 +710,33 @@ static void ia32_emit_cmp_suffix(int pnc)
 void ia32_emit_cmp_suffix_node(const ir_node *node,
                                int flags_pos)
 {
-       pn_Cmp pnc = get_ia32_pncode(node);
+       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) {
+               if(pnc & ia32_pn_Cmp_float) {
+                       pnc = get_negated_pnc(pnc, mode_F);
+               } else {
+                       pnc = get_negated_pnc(pnc, mode_Iu);
+               }
+       }
+
        ia32_emit_cmp_suffix(pnc);
 }
 
 /**
  * Returns the target block for a control flow node.
  */
-static
-ir_node *get_cfop_target_block(const ir_node *irn) {
+static ir_node *get_cfop_target_block(const ir_node *irn) {
        return get_irn_link(irn);
 }
 
 /**
  * Emits a block label for the given block.
  */
-static
-void ia32_emit_block_name(const ir_node *block)
+static void ia32_emit_block_name(const ir_node *block)
 {
        if (has_Block_label(block)) {
                be_emit_string(be_gas_label_prefix());
@@ -793,11 +790,12 @@ static ir_node *get_proj(const ir_node *node, long proj) {
  */
 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;
-       pn_Cmp         pnc = get_ia32_pncode(node);
+       pn_Cmp         pnc = get_ia32_condcode(node);
 
        pnc = determine_final_pnc(node, 0, pnc);
 
@@ -828,11 +826,12 @@ static void emit_ia32_Jcc(const ir_node *node)
                /* 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:
+                       case pn_Cmp_Uo: {
                                be_emit_cstring("\tjp ");
                                ia32_emit_cfop_target(proj_true);
                                be_emit_finish_line_gas(proj_true);
                                break;
+                       }
 
                        case pn_Cmp_Leg:
                                be_emit_cstring("\tjnp ");
@@ -843,8 +842,15 @@ static void emit_ia32_Jcc(const ir_node *node)
                        case pn_Cmp_Eq:
                        case pn_Cmp_Lt:
                        case pn_Cmp_Le:
-                               be_emit_cstring("\tjp ");
-                               ia32_emit_cfop_target(proj_false);
+                               /* we need a local label if the false proj is a fallthrough
+                                * as the falseblock might have no label emitted then */
+                               if (get_cfop_target_block(proj_false) == next_block) {
+                                       need_parity_label = 1;
+                                       be_emit_cstring("\tjp 1f");
+                               } else {
+                                       be_emit_cstring("\tjp ");
+                                       ia32_emit_cfop_target(proj_false);
+                               }
                                be_emit_finish_line_gas(proj_false);
                                goto emit_jcc;
 
@@ -868,6 +874,11 @@ emit_jcc:
                be_emit_finish_line_gas(proj_true);
        }
 
+       if(need_parity_label) {
+               be_emit_cstring("1:");
+               be_emit_write_line();
+       }
+
        /* the second Proj might be a fallthrough */
        if (get_cfop_target_block(proj_false) != next_block) {
                be_emit_cstring("\tjmp ");
@@ -883,8 +894,10 @@ emit_jcc:
 
 static void emit_ia32_CMov(const ir_node *node)
 {
-       const arch_register_t *out     = arch_get_irn_register(arch_env, node);
-       pn_Cmp                 pnc     = get_ia32_pncode(node);
+       const ia32_attr_t     *attr         = get_ia32_attr_const(node);
+       int                    ins_permuted = attr->data.ins_permuted;
+       const arch_register_t *out          = arch_get_irn_register(arch_env, node);
+       pn_Cmp                 pnc          = get_ia32_condcode(node);
        const arch_register_t *in_true;
        const arch_register_t *in_false;
 
@@ -901,11 +914,9 @@ static void emit_ia32_CMov(const ir_node *node)
        } else if(out == in_true) {
                const arch_register_t *tmp;
 
-               if(pnc & ia32_pn_Cmp_float) {
-                       pnc = get_negated_pnc(pnc, mode_F);
-               } else {
-                       pnc = get_negated_pnc(pnc, mode_Iu);
-               }
+               assert(get_ia32_op_type(node) == ia32_Normal);
+
+               ins_permuted = !ins_permuted;
 
                tmp      = in_true;
                in_true  = in_false;
@@ -919,6 +930,14 @@ static void emit_ia32_CMov(const ir_node *node)
                be_emit_finish_line_gas(node);
        }
 
+       if(ins_permuted) {
+               if(pnc & ia32_pn_Cmp_float) {
+                       pnc = get_negated_pnc(pnc, mode_F);
+               } else {
+                       pnc = get_negated_pnc(pnc, mode_Iu);
+               }
+       }
+
        /* TODO: handling of Nans isn't correct yet */
 
        be_emit_cstring("\tcmov");
@@ -964,8 +983,7 @@ 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;
 
@@ -980,11 +998,12 @@ 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(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;
@@ -993,11 +1012,12 @@ void emit_ia32_SwitchJmp(const ir_node *node) {
        tbl.label        = xmalloc(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.num_branches = get_irn_n_edges(node) - 1;
        tbl.branches     = xcalloc(tbl.num_branches, sizeof(tbl.branches[0]));
        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) {
@@ -1006,21 +1026,22 @@ void emit_ia32_SwitchJmp(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);
@@ -1113,11 +1134,9 @@ static void emit_ia32_Immediate(const ir_node *node)
 
        be_emit_char('$');
        if(attr->symconst != NULL) {
-               ident *id = get_entity_ld_ident(attr->symconst);
-
-               if(attr->attr.data.am_sc_sign)
+               if(attr->sc_sign)
                        be_emit_char('-');
-               be_emit_ident(id);
+               ia32_emit_entity(attr->symconst, 0);
        }
        if(attr->symconst == NULL || attr->offset != 0) {
                if(attr->symconst != NULL) {
@@ -1128,15 +1147,26 @@ static void emit_ia32_Immediate(const ir_node *node)
        }
 }
 
+/**
+ * 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 == '%');
@@ -1185,26 +1215,26 @@ static const char* emit_asm_operand(const ir_node *node, const char *s)
                s += p;
        }
 
+       if(num < 0 || num >= ARR_LEN(asm_regs)) {
+               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(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(pred);
                        return s;
                }
-               reg = get_in_reg(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 "
@@ -1212,25 +1242,34 @@ static const char* emit_asm_operand(const ir_node *node, const char *s)
                return s;
        }
 
+       if(asm_reg->memory) {
+               be_emit_char('(');
+       }
+
        /* emit it */
-       be_emit_char('%');
-       switch(modifier) {
-       case 0:
-               reg_name = arch_register_get_name(reg);
-               break;
-       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");
+       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(reg_name);
 
        return s;
 }
@@ -1283,11 +1322,11 @@ static void emit_ia32_Asm(const ir_node *node)
 /**
  * Emit movsb/w instructions to make mov count divideable by 4
  */
-static void emit_CopyB_prolog(int rem) {
+static void emit_CopyB_prolog(unsigned size) {
        be_emit_cstring("\tcld");
        be_emit_finish_line_gas(NULL);
 
-       switch(rem) {
+       switch (size) {
        case 1:
                be_emit_cstring("\tmovsb");
                be_emit_finish_line_gas(NULL);
@@ -1310,9 +1349,9 @@ static void emit_CopyB_prolog(int rem) {
  */
 static void emit_ia32_CopyB(const ir_node *node)
 {
-       int rem = get_ia32_pncode(node);
+       unsigned size = get_ia32_copyb_size(node);
 
-       emit_CopyB_prolog(rem);
+       emit_CopyB_prolog(size);
 
        be_emit_cstring("\trep movsd");
        be_emit_finish_line_gas(node);
@@ -1323,7 +1362,7 @@ static void emit_ia32_CopyB(const ir_node *node)
  */
 static void emit_ia32_CopyB_i(const ir_node *node)
 {
-       int size = get_ia32_pncode(node);
+       unsigned size = get_ia32_copyb_size(node);
 
        emit_CopyB_prolog(size & 0x3);
 
@@ -1471,14 +1510,6 @@ static void emit_ia32_Conv_I2I(const ir_node *node)
        be_emit_finish_line_gas(node);
 }
 
-/**
- * Emits code for an 8Bit Int conversion.
- */
-static void emit_ia32_Conv_I2I8Bit(const ir_node *node)
-{
-       emit_ia32_Conv_I2I(node);
-}
-
 
 /*******************************************
  *  _                          _
@@ -1499,8 +1530,7 @@ static void emit_be_Call(const ir_node *node)
 
        be_emit_cstring("\tcall ");
        if (ent) {
-               set_entity_backend_marked(ent, 1);
-               be_emit_string(get_entity_ld_name(ent));
+               ia32_emit_entity(ent, 1);
        } else {
                const arch_register_t *reg = get_in_reg(node, be_pos_Call_ptr);
                be_emit_char('*');
@@ -1757,13 +1787,29 @@ zero_neg:
        emit_sbb( node, in_hi, out_hi);
 }
 
+static void emit_ia32_GetEIP(const ir_node *node)
+{
+       be_emit_cstring("\tcall ");
+       be_emit_string(pic_base_label);
+       be_emit_finish_line_gas(node);
+
+       be_emit_string(pic_base_label);
+       be_emit_cstring(":\n");
+       be_emit_write_line();
+
+       be_emit_cstring("\tpopl ");
+       ia32_emit_dest_register(node, 0);
+       be_emit_char('\n');
+       be_emit_write_line();
+}
+
 static void emit_be_Return(const ir_node *node)
 {
        unsigned pop;
        be_emit_cstring("\tret");
 
        pop = be_Return_get_pop(node);
-       if(pop > 0) {
+       if (pop > 0 || be_Return_get_emit_pop(node)) {
                be_emit_irprintf(" $%d", pop);
        }
        be_emit_finish_line_gas(node);
@@ -1789,8 +1835,7 @@ static void emit_Nothing(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)
@@ -1808,6 +1853,7 @@ void ia32_register_emitters(void) {
        /* other ia32 emitter functions */
        IA32_EMIT(Asm);
        IA32_EMIT(CMov);
+       IA32_EMIT(IMul);
        IA32_EMIT(SwitchJmp);
        IA32_EMIT(CopyB);
        IA32_EMIT(CopyB_i);
@@ -1815,11 +1861,12 @@ void ia32_register_emitters(void) {
        IA32_EMIT(Conv_FP2I);
        IA32_EMIT(Conv_FP2FP);
        IA32_EMIT(Conv_I2I);
-       IA32_EMIT(Conv_I2I8Bit);
+       IA32_EMIT2(Conv_I2I8Bit, Conv_I2I);
        IA32_EMIT(Const);
        IA32_EMIT(LdTls);
        IA32_EMIT(Minus64Bit);
        IA32_EMIT(Jcc);
+       IA32_EMIT(GetEIP);
 
        /* benode emitter */
        BE_EMIT(Call);
@@ -1846,41 +1893,6 @@ void ia32_register_emitters(void) {
 #undef IA32_EMIT
 }
 
-static const char *last_name = NULL;
-static unsigned last_line = -1;
-static unsigned num = -1;
-
-/**
- * Emit the debug support for node node.
- */
-static void ia32_emit_dbg(const ir_node *node)
-{
-       dbg_info *db = get_irn_dbg_info(node);
-       unsigned lineno;
-       const char *fname = be_retrieve_dbg_info(db, &lineno);
-
-       if (! cg->birg->main_env->options->stabs_debug_support)
-               return;
-
-       if (fname) {
-               if (last_name != fname) {
-                       last_line = -1;
-                       be_dbg_include_begin(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(cg->birg->main_env->db_handle, lineno, name);
-                       be_emit_string(name);
-                       be_emit_cstring(":\n");
-                       be_emit_write_line();
-               }
-       }
-}
-
 typedef void (*emit_func_ptr) (const ir_node *);
 
 /**
@@ -1894,7 +1906,9 @@ static void ia32_emit_node(const ir_node *node)
 
        if (op->ops.generic) {
                emit_func_ptr func = (emit_func_ptr) op->ops.generic;
-               ia32_emit_dbg(node);
+
+               be_dbg_set_dbg_info(get_irn_dbg_info(node));
+
                (*func) (node);
        } else {
                emit_Nothing(node);
@@ -1913,58 +1927,19 @@ static void ia32_emit_alignment(unsigned align, unsigned skip)
        be_emit_write_line();
 }
 
-/**
- * Emits gas alignment directives for Functions depended on cpu architecture.
- */
-static void ia32_emit_align_func(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(align, maximum_skip);
-}
-
 /**
  * Emits gas alignment directives for Labels depended on cpu architecture.
  */
-static void ia32_emit_align_label(cpu_support cpu)
+static void ia32_emit_align_label(void)
 {
-       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;
+       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.
  */
@@ -1975,12 +1950,11 @@ static int should_align_block(ir_node *block, ir_node *prev)
        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       = isa->opt_arch;
        int           i, n_cfgpreds;
 
        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);
@@ -2004,49 +1978,99 @@ static int should_align_block(ir_node *block, ir_node *prev)
 
        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(ir_node *block, ir_node *prev)
+/**
+ * Return non-zero, if a instruction in a fall-through.
+ */
+static int is_fallthrough(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;
+}
+
+/**
+ * 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, ir_node *prev_block)
+{
+       ir_graph     *irg = current_ir_graph;
        int           n_cfgpreds;
-       int           need_label;
+       int           need_label = 1;
        int           i, arity;
-       ir_exec_freq  *exec_freq = cg->birg->exec_freq;
+       ir_exec_freq *exec_freq = cg->birg->exec_freq;
+
+       if(block == get_irg_end_block(irg) || block == get_irg_start_block(irg))
+               return;
 
        n_cfgpreds = get_Block_n_cfgpreds(block);
-       need_label = (n_cfgpreds != 0);
 
-       if (should_align_block(block, prev)) {
-               assert(need_label);
-               ia32_emit_align_label(isa->opt_arch);
+       if(n_cfgpreds == 0) {
+               need_label = 0;
+       } else if(n_cfgpreds == 1) {
+               ir_node *cfgpred = get_Block_cfgpred(block, 0);
+               if(get_nodes_block(cfgpred) == prev_block && is_fallthrough(cfgpred)) {
+                       need_label = 0;
+               }
        }
 
-       if(need_label) {
+       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, prev_block)) {
+                       ia32_emit_align_label();
+               } else {
+                       /* if the predecessor block has no fall-through,
+                          we can always align the label. */
+                       int i;
+                       ir_node *check_node = NULL;
+
+                       for (i = n_cfgpreds - 1; i >= 0; --i) {
+                               ir_node *cfg_pred = get_Block_cfgpred(block, i);
+
+                               if (get_nodes_block(skip_Proj(cfg_pred)) == prev_block) {
+                                       check_node = cfg_pred;
+                                       break;
+                               }
+                       }
+                       if (check_node == NULL || !is_fallthrough(check_node))
+                               ia32_emit_align_label();
+               }
+       }
+
+       if (need_label) {
                ia32_emit_block_name(block);
                be_emit_char(':');
 
                be_emit_pad_comment();
-               be_emit_cstring("   /* preds:");
-
-               /* emit list of pred blocks in comment */
-               arity = get_irn_arity(block);
-               for (i = 0; i < arity; ++i) {
-                       ir_node *predblock = get_Block_cfgpred_block(block, i);
-                       be_emit_irprintf(" %d", get_irn_node_nr(predblock));
-               }
+               be_emit_cstring("   /* ");
        } else {
                be_emit_cstring("\t/* ");
                ia32_emit_block_name(block);
                be_emit_cstring(": ");
        }
+
+       be_emit_cstring("preds:");
+
+       /* emit list of pred blocks in comment */
+       arity = get_irn_arity(block);
+       for (i = 0; i < arity; ++i) {
+               ir_node *predblock = get_Block_cfgpred_block(block, i);
+               be_emit_irprintf(" %d", get_irn_node_nr(predblock));
+       }
        if (exec_freq != NULL) {
                be_emit_irprintf(" freq: %f",
                                 get_block_execfreq(exec_freq, block));
@@ -2066,52 +2090,12 @@ static void ia32_gen_block(ir_node *block, ir_node *last_block)
        ia32_emit_block_header(block, last_block);
 
        /* emit the contents of the block */
-       ia32_emit_dbg(block);
+       be_dbg_set_dbg_info(get_irn_dbg_info(block));
        sched_foreach(block, node) {
                ia32_emit_node(node);
        }
 }
 
-/**
- * Emits code for function start.
- */
-static void ia32_emit_func_prolog(ir_graph *irg)
-{
-       ir_entity  *irg_ent  = get_irg_entity(irg);
-       const char *irg_name = get_entity_ld_name(irg_ent);
-       cpu_support cpu      = isa->opt_arch;
-       const be_irg_t *birg = cg->birg;
-
-       be_emit_write_line();
-       be_gas_emit_switch_section(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(cpu);
-       if (get_entity_visibility(irg_ent) == visibility_external_visible) {
-               be_emit_cstring(".global ");
-               be_emit_string(irg_name);
-               be_emit_char('\n');
-               be_emit_write_line();
-       }
-       ia32_emit_function_object(irg_name);
-       be_emit_string(irg_name);
-       be_emit_cstring(":\n");
-       be_emit_write_line();
-}
-
-/**
- * Emits code for function end
- */
-static void ia32_emit_func_epilog(ir_graph *irg)
-{
-       const char     *irg_name = get_entity_ld_name(get_irg_entity(irg));
-       const be_irg_t *birg     = cg->birg;
-
-       ia32_emit_function_size(irg_name);
-       be_dbg_method_end(birg->main_env->db_handle);
-       be_emit_char('\n');
-       be_emit_write_line();
-}
-
 /**
  * Block-walker:
  * Sets labels for control flow nodes (jump target)
@@ -2144,17 +2128,23 @@ void ia32_emit_exc_label(const ir_node *node)
  */
 void ia32_gen_routine(ia32_code_gen_t *ia32_cg, ir_graph *irg)
 {
-       ir_node *block;
-       ir_node *last_block = NULL;
+       ir_node   *block;
+       ir_node   *last_block = NULL;
+       ir_entity *entity     = get_irg_entity(irg);
        int i, n;
 
        cg       = ia32_cg;
        isa      = (const ia32_isa_t*) cg->arch_env->isa;
        arch_env = cg->arch_env;
+       do_pic   = cg->birg->main_env->options->pic;
 
        ia32_register_emitters();
 
-       ia32_emit_func_prolog(irg);
+       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);
+
        irg_block_walk_graph(irg, ia32_gen_labels, NULL, NULL);
 
        n = ARR_LEN(cg->blk_sched);
@@ -2171,7 +2161,10 @@ void ia32_gen_routine(ia32_code_gen_t *ia32_cg, ir_graph *irg)
                last_block = block;
        }
 
-       ia32_emit_func_epilog(irg);
+       be_gas_emit_function_epilog(entity);
+       be_dbg_method_end();
+       be_emit_char('\n');
+       be_emit_write_line();
 }
 
 void ia32_init_emitter(void)