renamed pmap_foreach to foreach_pmap, like other foreach_* macros
[libfirm] / ir / be / beabi.c
index d692cd9..d6c0bbd 100644 (file)
@@ -1,12 +1,5 @@
-/**
- * @file       becopyheur4.c
- * @brief      ABI lowering
- * @author     Sebastian Hack
- * @date       7.3.2005
- * @version    $Id$
- * @license
- *
- * Copyrigth (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.
  *
  * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
  * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  * PURPOSE.
- *
+ */
+
+/**
+ * @file
+ * @brief       Backend ABI implementation.
+ * @author      Sebastian Hack, Michael Beck
+ * @version     $Id$
  */
 #ifdef HAVE_CONFIG_H
 # include "config.h"
@@ -31,7 +30,6 @@
 #include "obst.h"
 #include "offset.h"
 
-#include "type.h"
 #include "irgopt.h"
 
 #include "irgraph_t.h"
@@ -47,6 +45,7 @@
 #include "pdeq.h"
 #include "irtools.h"
 #include "raw_bitset.h"
+#include "error.h"
 
 #include "be.h"
 #include "beabi.h"
@@ -62,16 +61,18 @@ typedef struct _be_abi_call_arg_t {
        unsigned in_reg   : 1;  /**< 1: this argument is transmitted in registers. */
        unsigned on_stack : 1;  /**< 1: this argument is transmitted on the stack. */
 
-       int pos;
+       int                    pos;
        const arch_register_t *reg;
-       ir_entity *stack_ent;
-       unsigned alignment;
-       unsigned space_before;
-       unsigned space_after;
+       ir_entity             *stack_ent;
+       ir_mode               *load_mode;
+       unsigned               alignment;    /**< stack alignment */
+       unsigned               space_before; /**< allocate space before */
+       unsigned               space_after;  /**< allocate space after */
 } be_abi_call_arg_t;
 
 struct _be_abi_call_t {
-       be_abi_call_flags_t         flags;
+       be_abi_call_flags_t          flags;
+       int                          pop;
        const be_abi_callbacks_t    *cb;
        ir_type                     *between_type;
        set                         *params;
@@ -89,15 +90,13 @@ struct _be_abi_irg_t {
        ir_type              *method_type;  /**< The type of the method of the IRG. */
 
        ir_node              *init_sp;      /**< The node representing the stack pointer
-                                                                            at the start of the function. */
-
-       ir_node              *start_barrier; /**< The barrier of the start block */
+                                                at the start of the function. */
 
        ir_node              *reg_params;   /**< The reg params node. */
        pmap                 *regs;         /**< A map of all callee-save and ignore regs to
-                                                                                       their Projs to the RegParams node. */
+                                                their Projs to the RegParams node. */
 
-       int                  start_block_bias;  /**< The stack bias at the end of the start block. */
+       int                  start_block_bias; /**< The stack bias at the end of the start block. */
 
        void                 *cb;           /**< ABI Callback self pointer. */
 
@@ -134,6 +133,7 @@ static int be_omit_fp = 1;
 static int cmp_call_arg(const void *a, const void *b, size_t n)
 {
        const be_abi_call_arg_t *p = a, *q = b;
+       (void) n;
        return !(p->is_res == q->is_res && p->pos == q->pos);
 }
 
@@ -180,6 +180,11 @@ void be_abi_call_set_flags(be_abi_call_t *call, be_abi_call_flags_t flags, const
        call->cb    = cb;
 }
 
+void be_abi_call_set_pop(be_abi_call_t *call, int pop)
+{
+       assert(pop >= 0);
+       call->pop = pop;
+}
 
 /* Set register class for call address */
 void be_abi_call_set_call_address_reg_class(be_abi_call_t *call, const arch_register_class_t *cls)
@@ -188,10 +193,11 @@ void be_abi_call_set_call_address_reg_class(be_abi_call_t *call, const arch_regi
 }
 
 
-void be_abi_call_param_stack(be_abi_call_t *call, int arg_pos, unsigned alignment, unsigned space_before, unsigned space_after)
+void be_abi_call_param_stack(be_abi_call_t *call, int arg_pos, ir_mode *load_mode, unsigned alignment, unsigned space_before, unsigned space_after)
 {
        be_abi_call_arg_t *arg = get_or_set_call_arg(call, 0, arg_pos, 1);
        arg->on_stack     = 1;
+       arg->load_mode    = load_mode;
        arg->alignment    = alignment;
        arg->space_before = space_before;
        arg->space_after  = space_after;
@@ -226,6 +232,7 @@ be_abi_call_flags_t be_abi_call_get_flags(const be_abi_call_t *call)
 static be_abi_call_t *be_abi_call_new(const arch_register_class_t *cls_addr)
 {
        be_abi_call_t *call = xmalloc(sizeof(call[0]));
+       memset(call, 0, sizeof(call[0]));
 
        call->flags.val  = 0;
        call->params     = new_set(cmp_call_arg, 16);
@@ -391,80 +398,82 @@ static INLINE int is_on_stack(be_abi_call_t *call, int pos)
 */
 
 /**
- * Transform a call node.
+ * Transform a call node into a be_Call node.
+ *
  * @param env The ABI environment for the current irg.
  * @param irn The call node.
  * @param curr_sp The stack pointer node to use.
  * @return The stack pointer after the call.
  */
-static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, ir_node *alloca_copy)
+static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp)
 {
-       ir_graph *irg             = env->birg->irg;
+       ir_graph *irg              = env->birg->irg;
        const arch_env_t *arch_env = env->birg->main_env->arch_env;
-       const arch_isa_t *isa     = arch_env->isa;
-       ir_type *mt               = get_Call_type(irn);
-       ir_node *call_ptr         = get_Call_ptr(irn);
-       int n_params              = get_method_n_params(mt);
-       ir_node *curr_mem         = get_Call_mem(irn);
-       ir_node *bl               = get_nodes_block(irn);
-       pset *results             = pset_new_ptr(8);
-       pset *caller_save         = pset_new_ptr(8);
-       pset *states              = pset_new_ptr(2);
-       int stack_size            = 0;
-       int stack_dir             = arch_isa_stack_dir(isa);
-       const arch_register_t *sp = arch_isa_sp(isa);
-       be_abi_call_t *call       = be_abi_call_new(sp->reg_class);
-       ir_mode *mach_mode        = sp->reg_class->mode;
-       struct obstack *obst      = &env->obst;
-       int no_alloc              = call->flags.bits.frame_is_setup_on_call;
-
-       ir_node *res_proj = NULL;
-       int curr_res_proj = pn_Call_max;
-       int n_low_args    = 0;
-       int n_pos         = 0;
-       int count;
+       const arch_isa_t *isa      = arch_env->isa;
+       ir_type *call_tp           = get_Call_type(irn);
+       ir_node *call_ptr          = get_Call_ptr(irn);
+       int n_params               = get_method_n_params(call_tp);
+       ir_node *curr_mem          = get_Call_mem(irn);
+       ir_node *bl                = get_nodes_block(irn);
+       pset *results              = pset_new_ptr(8);
+       pset *caller_save          = pset_new_ptr(8);
+       pset *states               = pset_new_ptr(2);
+       int stack_size             = 0;
+       int stack_dir              = arch_isa_stack_dir(isa);
+       const arch_register_t *sp  = arch_isa_sp(isa);
+       be_abi_call_t *call        = be_abi_call_new(sp->reg_class);
+       ir_mode *mach_mode         = sp->reg_class->mode;
+       struct obstack *obst       = &env->obst;
+       int no_alloc               = call->flags.bits.frame_is_setup_on_call;
+       int n_res                  = get_method_n_ress(call_tp);
+
+       ir_node *res_proj  = NULL;
+       int n_reg_params   = 0;
+       int n_stack_params = 0;
+       int n_ins;
 
        ir_node *low_call;
        ir_node **in;
        ir_node **res_projs;
+       int      n_reg_results = 0;
        const arch_register_t *reg;
        const ir_edge_t *edge;
-       int *low_args;
-       int *pos;
+       int *reg_param_idxs;
+       int *stack_param_idx;
        int i, n;
 
        /* Let the isa fill out the abi description for that call node. */
-       arch_isa_get_call_abi(isa, mt, call);
+       arch_isa_get_call_abi(isa, call_tp, call);
 
        /* Insert code to put the stack arguments on the stack. */
        assert(get_Call_n_params(irn) == n_params);
-       for(i = 0; i < n_params; ++i) {
+       for (i = 0; i < n_params; ++i) {
                be_abi_call_arg_t *arg = get_call_arg(call, 0, i);
                assert(arg);
                if (arg->on_stack) {
-                       int arg_size = get_type_size_bytes(get_method_param_type(mt, i));
+                       int arg_size = get_type_size_bytes(get_method_param_type(call_tp, i));
 
                        stack_size += round_up2(arg->space_before, arg->alignment);
                        stack_size += round_up2(arg_size, arg->alignment);
                        stack_size += round_up2(arg->space_after, arg->alignment);
                        obstack_int_grow(obst, i);
-                       n_pos++;
+                       ++n_stack_params;
                }
        }
-       pos = obstack_finish(obst);
+       stack_param_idx = obstack_finish(obst);
 
        /* Collect all arguments which are passed in registers. */
-       for(i = 0, n = get_Call_n_params(irn); i < n; ++i) {
+       for (i = 0; i < n_params; ++i) {
                be_abi_call_arg_t *arg = get_call_arg(call, 0, i);
-               if(arg && arg->in_reg) {
+               if (arg && arg->in_reg) {
                        obstack_int_grow(obst, i);
-                       n_low_args++;
+                       ++n_reg_params;
                }
        }
-       low_args = obstack_finish(obst);
+       reg_param_idxs = obstack_finish(obst);
 
        /* If there are some parameters which shall be passed on the stack. */
-       if(n_pos > 0) {
+       if (n_stack_params > 0) {
                int curr_ofs      = 0;
                int do_seq        = call->flags.bits.store_args_sequential && !no_alloc;
 
@@ -474,11 +483,11 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
                 * direction is downwards.
                 */
                if (call->flags.bits.left_to_right ^ (do_seq && stack_dir < 0)) {
-                       for (i = 0; i < n_pos >> 1; ++i) {
-                               int other  = n_pos - i - 1;
-                               int tmp    = pos[i];
-                               pos[i]     = pos[other];
-                               pos[other] = tmp;
+                       for (i = 0; i < n_stack_params >> 1; ++i) {
+                               int other  = n_stack_params - i - 1;
+                               int tmp    = stack_param_idx[i];
+                               stack_param_idx[i]     = stack_param_idx[other];
+                               stack_param_idx[other] = tmp;
                        }
                }
 
@@ -488,28 +497,22 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
                 * we allocate as much space on the stack all parameters need, by
                 * moving the stack pointer along the stack's direction.
                 */
-               if(stack_dir < 0 && !do_seq && !no_alloc) {
+               if (stack_dir < 0 && !do_seq && !no_alloc) {
                        curr_sp = be_new_IncSP(sp, irg, bl, curr_sp, stack_size);
-                       if(alloca_copy) {
-                               add_irn_dep(curr_sp, alloca_copy);
-                               alloca_copy = NULL;
-                       }
                }
 
-               if(!do_seq) {
-                       obstack_ptr_grow(obst, get_Call_mem(irn));
-                       curr_mem = new_NoMem();
-               } else {
-                       curr_mem = get_Call_mem(irn);
+               curr_mem = get_Call_mem(irn);
+               if (! do_seq) {
+                       obstack_ptr_grow(obst, curr_mem);
                }
 
-               for(i = 0; i < n_pos; ++i) {
-                       int p                  = pos[i];
+               for (i = 0; i < n_stack_params; ++i) {
+                       int p                  = stack_param_idx[i];
                        be_abi_call_arg_t *arg = get_call_arg(call, 0, p);
                        ir_node *param         = get_Call_param(irn, p);
                        ir_node *addr          = curr_sp;
                        ir_node *mem           = NULL;
-                       ir_type *param_type    = get_method_param_type(mt, p);
+                       ir_type *param_type    = get_method_param_type(call_tp, p);
                        int param_size         = get_type_size_bytes(param_type) + arg->space_after;
 
                        /*
@@ -520,10 +523,6 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
                        if (do_seq) {
                                curr_ofs = 0;
                                addr = curr_sp = be_new_IncSP(sp, irg, bl, curr_sp, param_size + arg->space_before);
-                               if(alloca_copy) {
-                                       add_irn_dep(curr_sp, alloca_copy);
-                                       alloca_copy = NULL;
-                               }
                                add_irn_dep(curr_sp, curr_mem);
                        }
                        else {
@@ -531,7 +530,7 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
                                curr_ofs =  round_up2(curr_ofs, arg->alignment);
 
                                /* Make the expression to compute the argument's offset. */
-                               if(curr_ofs > 0) {
+                               if (curr_ofs > 0) {
                                        ir_mode *constmode = mach_mode;
                                        if(mode_is_reference(mach_mode)) {
                                                constmode = mode_Is;
@@ -544,7 +543,8 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
                        /* Insert a store for primitive arguments. */
                        if (is_atomic_type(param_type)) {
                                ir_node *store;
-                               store = new_r_Store(irg, bl, curr_mem, addr, param);
+                               ir_node *mem_input = do_seq ? curr_mem : new_NoMem();
+                               store = new_r_Store(irg, bl, mem_input, addr, param);
                                mem = new_r_Proj(irg, bl, store, mode_M, pn_Store_M);
                        }
 
@@ -568,9 +568,9 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
                in = (ir_node **) obstack_finish(obst);
 
                /* We need the sync only, if we didn't build the stores sequentially. */
-               if(!do_seq) {
-                       if(n_pos >= 1) {
-                               curr_mem = new_r_Sync(irg, bl, n_pos + 1, in);
+               if (! do_seq) {
+                       if (n_stack_params >= 1) {
+                               curr_mem = new_r_Sync(irg, bl, n_stack_params + 1, in);
                        } else {
                                curr_mem = get_Call_mem(irn);
                        }
@@ -579,15 +579,15 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
        }
 
        /* Collect caller save registers */
-       for(i = 0, n = arch_isa_get_n_reg_class(isa); i < n; ++i) {
-               int j;
+       for (i = 0, n = arch_isa_get_n_reg_class(isa); i < n; ++i) {
+               unsigned j;
                const arch_register_class_t *cls = arch_isa_get_reg_class(isa, i);
-               for(j = 0; j < cls->n_regs; ++j) {
+               for (j = 0; j < cls->n_regs; ++j) {
                        const arch_register_t *reg = arch_register_for_index(cls, j);
-                       if(arch_register_type_is(reg, caller_save)) {
+                       if (arch_register_type_is(reg, caller_save)) {
                                pset_insert_ptr(caller_save, (void *) reg);
                        }
-                       if(arch_register_type_is(reg, state)) {
+                       if (arch_register_type_is(reg, state)) {
                                pset_insert_ptr(caller_save, (void*) reg);
                                pset_insert_ptr(states, (void*) reg);
                        }
@@ -596,56 +596,39 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
 
        /* search the greatest result proj number */
 
-       /* TODO: what if the result is NOT used? Currently there is
-        * no way to detect this later, especially there is no way to
-        * see this in the proj numbers.
-        * While this is ok for the register allocator, it is bad for
-        * backends which need to change the be_Call further (x87 simulator
-        * for instance. However for this particular case the call_type is
-        * sufficient.).
-        */
+       res_projs = alloca(n_res * sizeof(res_projs[0]));
+       memset(res_projs, 0, n_res * sizeof(res_projs[0]));
+
        foreach_out_edge(irn, edge) {
                const ir_edge_t *res_edge;
-               ir_node *irn = get_edge_src_irn(edge);
+               ir_node         *irn = get_edge_src_irn(edge);
 
-               if(is_Proj(irn) && get_Proj_proj(irn) == pn_Call_T_result) {
-                       res_proj = irn;
-                       foreach_out_edge(irn, res_edge) {
-                               int proj;
-                               be_abi_call_arg_t *arg;
-                               ir_node *res = get_edge_src_irn(res_edge);
-
-                               assert(is_Proj(res));
-
-                               proj = get_Proj_proj(res);
-                               arg = get_call_arg(call, 1, proj);
-
-                               /*
-                                       shift the proj number to the right, since we will drop the
-                                       unspeakable Proj_T from the Call. Therefore, all real argument
-                                       Proj numbers must be increased by pn_be_Call_first_res
-                               */
-                               proj += pn_be_Call_first_res;
-                               set_Proj_proj(res, proj);
-                               obstack_ptr_grow(obst, res);
-
-                               if(proj > curr_res_proj)
-                                       curr_res_proj = proj;
-                               if(arg->in_reg) {
-                                       pset_remove_ptr(caller_save, arg->reg);
-                                       //pmap_insert(arg_regs, arg->reg, INT_TO_PTR(proj + 1))
-                               }
-                       }
+               if(!is_Proj(irn) || get_Proj_proj(irn) != pn_Call_T_result)
+                       continue;
+
+               foreach_out_edge(irn, res_edge) {
+                       int proj;
+                       ir_node *res = get_edge_src_irn(res_edge);
+
+                       assert(is_Proj(res));
+
+                       proj = get_Proj_proj(res);
+                       assert(proj < n_res);
+                       assert(res_projs[proj] == NULL);
+                       res_projs[proj] = res;
                }
+               res_proj = irn;
+               break;
        }
 
-       curr_res_proj++;
-       obstack_ptr_grow(obst, NULL);
-       res_projs = obstack_finish(obst);
+       /** TODO: this is not correct for cases where return values are passed
+        * on the stack, but no known ABI does this currentl...
+        */
+       n_reg_results = n_res;
 
        /* make the back end call node and set its register requirements. */
-       for(i = 0; i < n_low_args; ++i) {
-               obstack_ptr_grow(obst, get_Call_param(irn, low_args[i]));
+       for (i = 0; i < n_reg_params; ++i) {
+               obstack_ptr_grow(obst, get_Call_param(irn, reg_param_idxs[i]));
        }
        foreach_pset(states, reg) {
                const arch_register_class_t *cls = arch_register_get_class(reg);
@@ -656,23 +639,64 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
                ir_node *regnode = new_rd_Unknown(irg, arch_register_class_mode(cls));
                obstack_ptr_grow(obst, regnode);
        }
-       count = n_low_args + pset_count(states);
+       n_ins = n_reg_params + pset_count(states);
 
        in = obstack_finish(obst);
 
-       if(env->call->flags.bits.call_has_imm && get_irn_opcode(call_ptr) == iro_SymConst) {
+       if (env->call->flags.bits.call_has_imm && is_SymConst(call_ptr)) {
+               /* direct call */
                low_call = be_new_Call(get_irn_dbg_info(irn), irg, bl, curr_mem,
                                       curr_sp, curr_sp,
-                                      curr_res_proj + pset_count(caller_save), count,
-                                      in, get_Call_type(irn));
+                                      n_reg_results + pn_be_Call_first_res + pset_count(caller_save),
+                                      n_ins, in, get_Call_type(irn));
                be_Call_set_entity(low_call, get_SymConst_entity(call_ptr));
        } else {
+               /* indirect call */
                low_call = be_new_Call(get_irn_dbg_info(irn), irg, bl, curr_mem,
                                       curr_sp, call_ptr,
-                                      curr_res_proj + pset_count(caller_save),
-                                      count, in, get_Call_type(irn));
+                                      n_reg_results + pn_be_Call_first_res + pset_count(caller_save),
+                                      n_ins, in, get_Call_type(irn));
+       }
+       be_Call_set_pop(low_call, call->pop);
+       ARR_APP1(ir_node *, env->calls, low_call);
+
+       /* create new stack pointer */
+       curr_sp = new_r_Proj(irg, bl, low_call, get_irn_mode(curr_sp),
+                            pn_be_Call_sp);
+       be_set_constr_single_reg(low_call, BE_OUT_POS(pn_be_Call_sp), sp);
+       arch_set_irn_register(arch_env, curr_sp, sp);
+       be_node_set_flags(low_call, BE_OUT_POS(pn_be_Call_sp),
+                       arch_irn_flags_ignore | arch_irn_flags_modify_sp);
+
+       for(i = 0; i < n_res; ++i) {
+               int pn;
+               ir_node           *proj = res_projs[i];
+               be_abi_call_arg_t *arg  = get_call_arg(call, 1, i);
+
+               /* returns values on stack not supported yet */
+               assert(arg->in_reg);
+
+               /*
+                       shift the proj number to the right, since we will drop the
+                       unspeakable Proj_T from the Call. Therefore, all real argument
+                       Proj numbers must be increased by pn_be_Call_first_res
+               */
+               pn = i + pn_be_Call_first_res;
+
+               if(proj == NULL) {
+                       ir_type *res_type = get_method_res_type(call_tp, i);
+                       ir_mode *mode     = get_type_mode(res_type);
+                       proj              = new_r_Proj(irg, bl, low_call, mode, pn);
+                       res_projs[i]      = proj;
+               } else {
+                       set_Proj_pred(proj, low_call);
+                       set_Proj_proj(proj, pn);
+               }
+
+               if (arg->in_reg) {
+                       pset_remove_ptr(caller_save, arg->reg);
+               }
        }
-       ARR_APP1(ir_node*, env->calls, low_call);
 
        /*
                Set the register class of the call address to
@@ -683,71 +707,74 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
        DBG((env->dbg, LEVEL_3, "\tcreated backend call %+F\n", low_call));
 
        /* Set the register classes and constraints of the Call parameters. */
-       for(i = 0; i < n_low_args; ++i) {
-               int index = low_args[i];
+       for (i = 0; i < n_reg_params; ++i) {
+               int index = reg_param_idxs[i];
                be_abi_call_arg_t *arg = get_call_arg(call, 0, index);
                assert(arg->reg != NULL);
 
-               be_set_constr_single_reg(low_call, be_pos_Call_first_arg + index, arg->reg);
+               be_set_constr_single_reg(low_call, be_pos_Call_first_arg + i, arg->reg);
        }
 
        /* Set the register constraints of the results. */
-       for (i = 0; res_projs[i]; ++i) {
-               int pn = get_Proj_proj(res_projs[i]);
-
-               /* Correct Proj number since it has been adjusted! (see above) */
-               const be_abi_call_arg_t *arg = get_call_arg(call, 1, pn - pn_Call_max);
-
-               /* Matze: we need the information about the real mode for later
-                * transforms (signed/unsigend compares, stores...), so leave the fixup
-                * for the backend transform phase... */
-#if 0
-               /* correct mode */
-               const arch_register_class_t *cls = arch_register_get_class(arg->reg);
-               ir_mode *mode = arch_register_class_mode(cls);
-               set_irn_mode(irn, mode);
-#endif
+       for (i = 0; i < n_res; ++i) {
+               ir_node                 *proj = res_projs[i];
+               const be_abi_call_arg_t *arg  = get_call_arg(call, 1, i);
+               int                      pn   = get_Proj_proj(proj);
 
                assert(arg->in_reg);
                be_set_constr_single_reg(low_call, BE_OUT_POS(pn), arg->reg);
-               arch_set_irn_register(arch_env, res_projs[i], arg->reg);
+               arch_set_irn_register(arch_env, proj, arg->reg);
        }
        obstack_free(obst, in);
        exchange(irn, low_call);
 
-       /* redirect the result projs to the lowered call instead of the Proj_T */
-       for (i = 0; res_projs[i]; ++i)
-               set_Proj_pred(res_projs[i], low_call);
-
-       /* set the now unnecessary projT to bad */
-       if(res_proj != NULL) {
+       /* kill the ProjT node */
+       if (res_proj != NULL) {
                be_kill_node(res_proj);
        }
 
        /* Make additional projs for the caller save registers
           and the Keep node which keeps them alive. */
-       if (pset_count(caller_save) > 0) {
+       if (1 || pset_count(caller_save) + n_reg_results > 0) {
                const arch_register_t *reg;
                ir_node               **in, *keep;
-               int                   i, n;
+               int                   i;
+               int                   n = 0;
+               int                   curr_res_proj
+                       = pn_be_Call_first_res + n_reg_results;
+
+               /* also keep the stack pointer */
+               ++n;
+               set_irn_link(curr_sp, (void*) sp);
+               obstack_ptr_grow(obst, curr_sp);
 
-               for (reg = pset_first(caller_save), n = 0; reg; reg = pset_next(caller_save), ++n) {
-                       ir_node *proj = new_r_Proj(irg, bl, low_call, reg->reg_class->mode, curr_res_proj);
+               for (reg = pset_first(caller_save); reg; reg = pset_next(caller_save), ++n) {
+                       ir_node *proj = new_r_Proj(irg, bl, low_call, reg->reg_class->mode,
+                                                  curr_res_proj);
 
                        /* memorize the register in the link field. we need afterwards to set the register class of the keep correctly. */
                        be_set_constr_single_reg(low_call, BE_OUT_POS(curr_res_proj), reg);
+                       arch_set_irn_register(arch_env, proj, reg);
 
                        /* a call can produce ignore registers, in this case set the flag and register for the Proj */
                        if (arch_register_type_is(reg, ignore)) {
-                               arch_set_irn_register(env->birg->main_env->arch_env, proj, reg);
-                               be_node_set_flags(low_call, BE_OUT_POS(curr_res_proj), arch_irn_flags_ignore);
+                               be_node_set_flags(low_call, BE_OUT_POS(curr_res_proj),
+                                                 arch_irn_flags_ignore);
                        }
 
-                       set_irn_link(proj, (void *) reg);
+                       set_irn_link(proj, (void*) reg);
                        obstack_ptr_grow(obst, proj);
                        curr_res_proj++;
                }
 
+               for(i = 0; i < n_reg_results; ++i) {
+                       ir_node *proj = res_projs[i];
+                       const arch_register_t *reg = arch_get_irn_register(arch_env, proj);
+                       set_irn_link(proj, (void*) reg);
+                       obstack_ptr_grow(obst, proj);
+               }
+               n += n_reg_results;
+
                /* create the Keep for the caller save registers */
                in   = (ir_node **) obstack_finish(obst);
                keep = be_new_Keep(NULL, irg, bl, n, in);
@@ -759,7 +786,10 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
        }
 
        /* Clean up the stack. */
-       if(stack_size > 0) {
+       assert(stack_size >= call->pop);
+       stack_size -= call->pop;
+
+       if (stack_size > 0) {
                ir_node *mem_proj = NULL;
 
                foreach_out_edge(low_call, edge) {
@@ -770,24 +800,19 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
                        }
                }
 
-               if(!mem_proj) {
-                       mem_proj = new_r_Proj(irg, bl, low_call, mode_M, pn_Call_M);
+               if (! mem_proj) {
+                       mem_proj = new_r_Proj(irg, bl, low_call, mode_M, pn_be_Call_M_regular);
                        keep_alive(mem_proj);
                }
 
                 /* Clean up the stack frame if we allocated it */
-               if(!no_alloc) {
+               if (! no_alloc) {
                        curr_sp = be_new_IncSP(sp, irg, bl, curr_sp, -stack_size);
-                       add_irn_dep(curr_sp, mem_proj);
-                       if(alloca_copy) {
-                               add_irn_dep(curr_sp, alloca_copy);
-                               alloca_copy = NULL;
-                       }
                }
        }
 
        be_abi_call_free(call);
-       obstack_free(obst, pos);
+       obstack_free(obst, stack_param_idx);
        del_pset(results);
        del_pset(states);
        del_pset(caller_save);
@@ -795,24 +820,49 @@ static ir_node *adjust_call(be_abi_irg_t *env, ir_node *irn, ir_node *curr_sp, i
        return curr_sp;
 }
 
+/**
+ * Adjust the size of a node representing a stack alloc or free for the minimum stack alignment.
+ *
+ * @param alignment  the minimum stack alignment
+ * @param size       the node containing the non-aligned size
+ * @param irg        the irg where new nodes are allocated on
+ * @param irg        the block where new nodes are allocated on
+ * @param dbg        debug info for new nodes
+ *
+ * @return a node representing the aligned size
+ */
+static ir_node *adjust_alloc_size(unsigned stack_alignment, ir_node *size,
+                                  ir_graph *irg, ir_node *block, dbg_info *dbg)
+{
+       if (stack_alignment > 1) {
+               ir_mode *mode = get_irn_mode(size);
+               tarval  *tv   = new_tarval_from_long(stack_alignment-1, mode);
+               ir_node *mask = new_r_Const(irg, block, mode, tv);
+
+               size = new_rd_Add(dbg, irg, block, size, mask, mode);
+
+               tv   = new_tarval_from_long(-(long)stack_alignment, mode);
+               mask = new_r_Const(irg, block, mode, tv);
+               size = new_rd_And(dbg, irg, block, size, mask, mode);
+       }
+       return size;
+}
 /**
  * Adjust an alloca.
  * The alloca is transformed into a back end alloca node and connected to the stack nodes.
  */
-static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp, ir_node **result_copy)
+static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp)
 {
        ir_node *block;
        ir_graph *irg;
        ir_node *alloc_mem;
        ir_node *alloc_res;
        ir_type *type;
+       dbg_info *dbg;
 
        const ir_edge_t *edge;
-       ir_node *new_alloc;
-       ir_node *size;
-       ir_node *addr;
-       ir_node *copy;
-       ir_node *ins[2];
+       ir_node *new_alloc, *size, *addr, *ins[2];
+       unsigned stack_alignment;
 
        if (get_Alloc_where(alloc) != stack_alloc) {
                assert(0);
@@ -849,12 +899,15 @@ static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp
                return curr_sp;
        }
 
+       dbg = get_irn_dbg_info(alloc);
+
        /* we might need to multiply the size with the element size */
        if(type != get_unknown_type() && get_type_size_bytes(type) != 1) {
-               tarval *tv = new_tarval_from_long(get_type_size_bytes(type), mode_Iu);
-               ir_node *cnst = new_rd_Const(NULL, irg, block, mode_Iu, tv);
-               ir_node *mul = new_rd_Mul(NULL, irg, block, get_Alloc_size(alloc),
-                                         cnst, mode_Iu);
+               tarval *tv    = new_tarval_from_long(get_type_size_bytes(type),
+                                                    mode_Iu);
+               ir_node *cnst = new_rd_Const(dbg, irg, block, mode_Iu, tv);
+               ir_node *mul  = new_rd_Mul(dbg, irg, block, get_Alloc_size(alloc),
+                                          cnst, mode_Iu);
                size = mul;
        } else {
                size = get_Alloc_size(alloc);
@@ -863,7 +916,13 @@ static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp
        /* The stack pointer will be modified in an unknown manner.
           We cannot omit it. */
        env->call->flags.bits.try_omit_fp = 0;
-       new_alloc = be_new_AddSP(env->isa->sp, irg, block, curr_sp, size);
+
+       /* FIXME: size must be here round up for the stack alignment, but
+          this must be transmitted from the backend. */
+       stack_alignment = 4;
+       size            = adjust_alloc_size(stack_alignment, size, irg, block, dbg);
+       new_alloc       = be_new_AddSP(env->isa->sp, irg, block, curr_sp, size);
+       set_irn_dbg_info(new_alloc, dbg);
 
        if(alloc_mem != NULL) {
                ir_node *addsp_mem;
@@ -871,8 +930,8 @@ static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp
 
                addsp_mem = new_r_Proj(irg, block, new_alloc, mode_M, pn_be_AddSP_M);
 
-               // We need to sync the output mem of the AddSP with the input mem
-               // edge into the alloc node
+               /* We need to sync the output mem of the AddSP with the input mem
+                  edge into the alloc node. */
                ins[0] = get_Alloc_mem(alloc);
                ins[1] = addsp_mem;
                sync = new_r_Sync(irg, block, 2, ins);
@@ -885,19 +944,9 @@ static ir_node *adjust_alloc(be_abi_irg_t *env, ir_node *alloc, ir_node *curr_sp
        /* fix projnum of alloca res */
        set_Proj_proj(alloc_res, pn_be_AddSP_res);
 
-       addr = env->isa->stack_dir < 0 ? alloc_res : curr_sp;
-
-       /* copy the address away, since it could be used after further stack pointer modifications. */
-       /* Let it point curr_sp just for the moment, I'll reroute it in a second. */
-       *result_copy = copy = be_new_Copy(env->isa->sp->reg_class, irg, block, curr_sp);
-
-       /* Let all users of the Alloc() result now point to the copy. */
-       edges_reroute(alloc_res, copy, irg);
-
-       /* Rewire the copy appropriately. */
-       set_irn_n(copy, be_pos_Copy_op, addr);
-
-       curr_sp = alloc_res;
+       addr    = alloc_res;
+       curr_sp = new_r_Proj(irg, block, new_alloc,  get_irn_mode(curr_sp),
+                            pn_be_AddSP_sp);
 
        return curr_sp;
 }  /* adjust_alloc */
@@ -914,6 +963,8 @@ static ir_node *adjust_free(be_abi_irg_t *env, ir_node *free, ir_node *curr_sp)
        ir_type *type;
        ir_node *in[2];
        ir_mode *sp_mode;
+       unsigned stack_alignment;
+       dbg_info *dbg;
 
        if (get_Free_where(free) != stack_alloc) {
                assert(0);
@@ -924,25 +975,32 @@ static ir_node *adjust_free(be_abi_irg_t *env, ir_node *free, ir_node *curr_sp)
        irg = get_irn_irg(block);
        type = get_Free_type(free);
        sp_mode = env->isa->sp->reg_class->mode;
+       dbg = get_irn_dbg_info(free);
 
        /* we might need to multiply the size with the element size */
        if(type != get_unknown_type() && get_type_size_bytes(type) != 1) {
                tarval *tv = new_tarval_from_long(get_type_size_bytes(type), mode_Iu);
-               ir_node *cnst = new_rd_Const(NULL, irg, block, mode_Iu, tv);
-               ir_node *mul = new_rd_Mul(NULL, irg, block, get_Free_size(free),
+               ir_node *cnst = new_rd_Const(dbg, irg, block, mode_Iu, tv);
+               ir_node *mul = new_rd_Mul(dbg, irg, block, get_Free_size(free),
                                          cnst, mode_Iu);
                size = mul;
        } else {
                size = get_Free_size(free);
        }
 
+       /* FIXME: size must be here round up for the stack alignment, but
+          this must be transmitted from the backend. */
+       stack_alignment = 4;
+       size = adjust_alloc_size(stack_alignment, size, irg, block, dbg);
+
        /* The stack pointer will be modified in an unknown manner.
           We cannot omit it. */
        env->call->flags.bits.try_omit_fp = 0;
        subsp = be_new_SubSP(env->isa->sp, irg, block, curr_sp, size);
+       set_irn_dbg_info(subsp, dbg);
 
        mem = new_r_Proj(irg, block, subsp, mode_M, pn_be_SubSP_M);
-       res = new_r_Proj(irg, block, subsp, sp_mode, pn_be_SubSP_res);
+       res = new_r_Proj(irg, block, subsp, sp_mode, pn_be_SubSP_sp);
 
        /* we need to sync the memory */
        in[0] = get_Free_mem(free);
@@ -1003,7 +1061,7 @@ static int dependent_on(ir_node *n1, ir_node *n2)
        return heights_reachable_in_block(ir_heights, n1, n2);
 }
 
-static int cmp_call_dependecy(const void *c1, const void *c2)
+static int cmp_call_dependency(const void *c1, const void *c2)
 {
        ir_node *n1 = *(ir_node **) c1;
        ir_node *n2 = *(ir_node **) c2;
@@ -1066,13 +1124,12 @@ static void process_calls_in_block(ir_node *bl, void *data)
        if(n > 0) {
                ir_node *keep;
                ir_node **nodes;
-               ir_node *copy = NULL;
                int i;
 
                nodes = obstack_finish(&env->obst);
 
                /* order the call nodes according to data dependency */
-               qsort(nodes, n, sizeof(nodes[0]), cmp_call_dependecy);
+               qsort(nodes, n, sizeof(nodes[0]), cmp_call_dependency);
 
                for(i = n - 1; i >= 0; --i) {
                        ir_node *irn = nodes[i];
@@ -1080,25 +1137,31 @@ static void process_calls_in_block(ir_node *bl, void *data)
                        DBG((env->dbg, LEVEL_3, "\tprocessing call %+F\n", irn));
                        switch(get_irn_opcode(irn)) {
                        case iro_Call:
-                               curr_sp = adjust_call(env, irn, curr_sp, copy);
+                               curr_sp = adjust_call(env, irn, curr_sp);
                                break;
                        case iro_Alloc:
-                               curr_sp = adjust_alloc(env, irn, curr_sp, &copy);
+                               curr_sp = adjust_alloc(env, irn, curr_sp);
                                break;
                        case iro_Free:
                                curr_sp = adjust_free(env, irn, curr_sp);
                                break;
                        default:
+                               panic("invalid call");
                                break;
                        }
                }
 
                obstack_free(&env->obst, nodes);
 
-               /* Keep the last stack state in the block by tying it to Keep node */
-               nodes[0] = curr_sp;
-               keep     = be_new_Keep(env->isa->sp->reg_class, get_irn_irg(bl), bl, 1, nodes);
-               pmap_insert(env->keep_map, bl, keep);
+               /* Keep the last stack state in the block by tying it to Keep node,
+                * the proj from calls is already kept */
+               if(curr_sp != env->init_sp
+                               && !(is_Proj(curr_sp) && be_is_Call(get_Proj_pred(curr_sp)))) {
+                       nodes[0] = curr_sp;
+                       keep     = be_new_Keep(env->isa->sp->reg_class, get_irn_irg(bl),
+                                              bl, 1, nodes);
+                       pmap_insert(env->keep_map, bl, keep);
+               }
        }
 
        set_irn_link(bl, curr_sp);
@@ -1119,86 +1182,6 @@ static void process_calls(be_abi_irg_t *env)
        heights_free(ir_heights);
 }
 
-#if 0 /*
-static ir_node *setup_frame(be_abi_irg_t *env)
-{
-       const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
-       const arch_register_t *sp = isa->sp;
-       const arch_register_t *bp = isa->bp;
-       be_abi_call_flags_bits_t flags = env->call->flags.bits;
-       ir_graph *irg      = env->birg->irg;
-       ir_node *bl        = get_irg_start_block(irg);
-       ir_node *no_mem    = get_irg_no_mem(irg);
-       ir_node *old_frame = get_irg_frame(irg);
-       ir_node *stack     = pmap_get(env->regs, (void *) sp);
-       ir_node *frame     = pmap_get(env->regs, (void *) bp);
-
-       int stack_nr       = get_Proj_proj(stack);
-
-       if(flags.try_omit_fp) {
-               stack = be_new_IncSP(sp, irg, bl, stack, no_mem, BE_STACK_FRAME_SIZE_EXPAND);
-               frame = stack;
-       }
-
-       else {
-               frame = be_new_Copy(bp->reg_class, irg, bl, stack);
-
-               be_node_set_flags(frame, -1, arch_irn_flags_dont_spill);
-               if(!flags.fp_free) {
-                       be_set_constr_single_reg(frame, -1, bp);
-                       be_node_set_flags(frame, -1, arch_irn_flags_ignore);
-                       arch_set_irn_register(env->birg->main_env->arch_env, frame, bp);
-               }
-
-               stack = be_new_IncSP(sp, irg, bl, stack, frame, BE_STACK_FRAME_SIZE_EXPAND);
-       }
-
-       be_node_set_flags(env->reg_params, -(stack_nr + 1), arch_irn_flags_ignore);
-       env->init_sp = stack;
-       set_irg_frame(irg, frame);
-       edges_reroute(old_frame, frame, irg);
-
-       return frame;
-}
-
-static void clearup_frame(be_abi_irg_t *env, ir_node *ret, pmap *reg_map, struct obstack *obst)
-{
-       const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
-       const arch_register_t *sp = isa->sp;
-       const arch_register_t *bp = isa->bp;
-       ir_graph *irg      = env->birg->irg;
-       ir_node *ret_mem   = get_Return_mem(ret);
-       ir_node *frame     = get_irg_frame(irg);
-       ir_node *bl        = get_nodes_block(ret);
-       ir_node *stack     = get_irn_link(bl);
-
-       pmap_entry *ent;
-
-       if(env->call->flags.bits.try_omit_fp) {
-               stack = be_new_IncSP(sp, irg, bl, stack, ret_mem, -BE_STACK_FRAME_SIZE_SHRINK);
-       }
-
-       else {
-               stack = be_new_SetSP(sp, irg, bl, stack, frame, ret_mem);
-               be_set_constr_single_reg(stack, -1, sp);
-               be_node_set_flags(stack, -1, arch_irn_flags_ignore);
-       }
-
-       pmap_foreach(env->regs, ent) {
-               const arch_register_t *reg = ent->key;
-               ir_node *irn               = ent->value;
-
-               if(reg == sp)
-                       obstack_ptr_grow(&env->obst, stack);
-               else if(reg == bp)
-                       obstack_ptr_grow(&env->obst, frame);
-               else if(arch_register_type_is(reg, callee_save) || arch_register_type_is(reg, ignore))
-                       obstack_ptr_grow(obst, irn);
-       }
-}
-*/
-#endif
-
 /**
  * Computes the stack argument layout type.
  * Changes a possibly allocated value param type by moving
@@ -1327,7 +1310,7 @@ static reg_node_map_t *reg_map_to_arr(struct obstack *obst, pmap *reg_map)
        int i = 0;
        reg_node_map_t *res = obstack_alloc(obst, n * sizeof(res[0]));
 
-       pmap_foreach(reg_map, ent) {
+       foreach_pmap(reg_map, ent) {
                res[i].reg = ent->key;
                res[i].irn = ent->value;
                i++;
@@ -1406,15 +1389,18 @@ static ir_node *create_barrier(be_abi_irg_t *env, ir_node *bl, ir_node **mem, pm
  * @param mem     the current memory
  * @param n_res   number of return results
  */
-static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl, ir_node *mem, int n_res) {
-       be_abi_call_t *call = env->call;
+static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl,
+               ir_node *mem, int n_res)
+{
+       be_abi_call_t *call        = env->call;
        const arch_isa_t *isa = env->birg->main_env->arch_env->isa;
-
+       dbg_info *dbgi;
        pmap *reg_map  = pmap_create();
        ir_node *keep  = pmap_get(env->keep_map, bl);
        int in_max;
        ir_node *ret;
        int i, n;
+       unsigned pop;
        ir_node **in;
        ir_node *stack;
        const arch_register_t **regs;
@@ -1429,11 +1415,9 @@ static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl, i
        */
        stack = be_abi_reg_map_get(env->regs, isa->sp);
        if (keep) {
-               ir_node *bad = new_r_Bad(env->birg->irg);
                stack = get_irn_n(keep, 0);
-               set_nodes_block(keep, bad);
-               set_irn_n(keep, 0, bad);
-               // exchange(keep, new_r_Bad(env->birg->irg));
+               be_kill_node(keep);
+               remove_End_keepalive(get_irg_end(env->birg->irg), keep);
        }
 
        /* Insert results for Return into the register map. */
@@ -1445,7 +1429,7 @@ static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl, i
        }
 
        /* Add uses of the callee save registers. */
-       pmap_foreach(env->regs, ent) {
+       foreach_pmap(env->regs, ent) {
                const arch_register_t *reg = ent->key;
                if(arch_register_type_is(reg, callee_save) || arch_register_type_is(reg, ignore))
                        pmap_insert(reg_map, ent->key, ent->value);
@@ -1485,7 +1469,7 @@ static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl, i
        }
 
        /* grow the rest of the stuff. */
-       pmap_foreach(reg_map, ent) {
+       foreach_pmap(reg_map, ent) {
                if(ent->value) {
                        in[n]     = ent->value;
                        regs[n++] = ent->key;
@@ -1493,7 +1477,14 @@ static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl, i
        }
 
        /* The in array for the new back end return is now ready. */
-       ret = be_new_Return(irn ? get_irn_dbg_info(irn) : NULL, env->birg->irg, bl, n_res, n, in);
+       if(irn != NULL) {
+               dbgi = get_irn_dbg_info(irn);
+       } else {
+               dbgi = NULL;
+       }
+       /* we have to pop the shadow parameter in in case of struct returns */
+       pop = call->pop;
+       ret = be_new_Return(dbgi, env->birg->irg, bl, n_res, pop, n, in);
 
        /* Set the register classes of the return's parameter accordingly. */
        for(i = 0; i < n; ++i)
@@ -1510,14 +1501,15 @@ static ir_node *create_be_return(be_abi_irg_t *env, ir_node *irn, ir_node *bl, i
 typedef struct lower_frame_sels_env_t {
        be_abi_irg_t *env;
        ir_entity    *value_param_list;  /**< the list of all value param entities */
+       ir_entity    *value_param_tail;  /**< the tail of the list of all value param entities */
 } lower_frame_sels_env_t;
 
 /**
  * Walker: Replaces Sels of frame type and
  * value param type entities by FrameAddress.
+ * Links all used entities.
  */
-static void lower_frame_sels_walker(ir_node *irn, void *data)
-{
+static void lower_frame_sels_walker(ir_node *irn, void *data) {
        lower_frame_sels_env_t *ctx = data;
 
        if (is_Sel(irn)) {
@@ -1535,10 +1527,13 @@ static void lower_frame_sels_walker(ir_node *irn, void *data)
                        nw = be_new_FrameAddr(env->isa->sp->reg_class, irg, bl, frame, ent);
                        exchange(irn, nw);
 
-                       /* check, if it's a param sel and if have not seen this entity immediatly before */
-                       if (ptr == param_base && ctx->value_param_list != ent) {
+                       /* check, if it's a param sel and if have not seen this entity before */
+                       if (ptr == param_base &&
+                                       ent != ctx->value_param_tail &&
+                                       get_entity_link(ent) == NULL) {
                                set_entity_link(ent, ctx->value_param_list);
                                ctx->value_param_list = ent;
+                               if (ctx->value_param_tail == NULL) ctx->value_param_tail = ent;
                        }
                }
        }
@@ -1581,7 +1576,7 @@ static void fix_address_of_parameter_access(be_abi_irg_t *env, ir_entity *value_
                ir_node *frame, *imem, *nmem, *store, *mem, *args, *args_bl;
                const ir_edge_t *edge;
                optimization_state_t state;
-               int offset;
+               unsigned offset;
 
                foreach_block_succ(start_bl, edge) {
                        ir_node *succ = get_edge_src_irn(edge);
@@ -1635,11 +1630,11 @@ static void fix_address_of_parameter_access(be_abi_irg_t *env, ir_entity *value_
                frame_tp = get_irg_frame_type(irg);
                offset   = get_type_size_bytes(frame_tp);
                for (ent = new_list; ent; ent = get_entity_link(ent)) {
-                       ir_type *tp = get_entity_type(ent);
-                       int align = get_type_alignment_bytes(tp);
+                       ir_type  *tp   = get_entity_type(ent);
+                       unsigned align = get_type_alignment_bytes(tp);
 
                        offset += align - 1;
-                       offset &= -align;
+                       offset &= ~(align - 1);
                        set_entity_owner(ent, frame_tp);
                        add_class_member(frame_tp, ent);
                        /* must be automatic to set a fixed layout */
@@ -1651,6 +1646,7 @@ static void fix_address_of_parameter_access(be_abi_irg_t *env, ir_entity *value_
        }
 }
 
+#if 1
 /**
  * The start block has no jump, instead it has an initial exec Proj.
  * The backend wants to handle all blocks the same way, so we replace
@@ -1681,6 +1677,7 @@ static void fix_start_block(ir_node *block, void *env) {
                }
        }
 }
+#endif
 
 /**
  * Modify the irg itself and the frame type.
@@ -1700,18 +1697,18 @@ static void modify_irg(be_abi_irg_t *env)
        pset *dont_save           = pset_new_ptr(8);
 
        int n_params;
-       int i, j, n, temp;
+       int i, n;
+       unsigned j;
 
        reg_node_map_t *rm;
        const arch_register_t *fp_reg;
        ir_node *frame_pointer;
-       ir_node *barrier;
        ir_node *reg_params_bl;
        ir_node **args;
        ir_node *arg_tuple;
        ir_node *value_param_base;
        const ir_edge_t *edge;
-       ir_type *arg_type, *bet_type;
+       ir_type *arg_type, *bet_type, *tp;
        lower_frame_sels_env_t ctx;
        ir_entity **param_map;
 
@@ -1720,9 +1717,18 @@ static void modify_irg(be_abi_irg_t *env)
 
        DBG((dbg, LEVEL_1, "introducing abi on %+F\n", irg));
 
+       /* set the links of all frame entities to NULL, we use it
+          to detect if an entity is already linked in the value_param_list */
+       tp = get_method_value_param_type(method_type);
+       if (tp != NULL) {
+               for (i = get_struct_n_members(tp) - 1; i >= 0; --i)
+                       set_entity_link(get_struct_member(tp, i), NULL);
+       }
+
        /* Convert the Sel nodes in the irg to frame load/store/addr nodes. */
        ctx.env              = env;
        ctx.value_param_list = NULL;
+       ctx.value_param_tail = NULL;
        irg_walk_graph(irg, lower_frame_sels_walker, NULL, &ctx);
 
        /* value_param_base anchor is not needed anymore now */
@@ -1754,9 +1760,11 @@ static void modify_irg(be_abi_irg_t *env)
        arg_tuple = get_irg_args(irg);
        foreach_out_edge(arg_tuple, edge) {
                ir_node *irn = get_edge_src_irn(edge);
-               int nr       = get_Proj_proj(irn);
-               args[nr]     = irn;
-               DBG((dbg, LEVEL_2, "\treading arg: %d -> %+F\n", nr, irn));
+               if (! is_Anchor(irn)) {
+                       int nr       = get_Proj_proj(irn);
+                       args[nr]     = irn;
+                       DBG((dbg, LEVEL_2, "\treading arg: %d -> %+F\n", nr, irn));
+               }
        }
 
        arg_type = compute_arg_type(env, call, method_type, &param_map);
@@ -1852,7 +1860,7 @@ static void modify_irg(be_abi_irg_t *env)
        env->init_sp = be_new_IncSP(sp, irg, bl, env->init_sp, BE_STACK_FRAME_SIZE_EXPAND);
        be_abi_reg_map_set(env->regs, sp, env->init_sp);
 
-       env->start_barrier = barrier = create_barrier(env, bl, &mem, env->regs, 0);
+       create_barrier(env, bl, &mem, env->regs, 0);
 
        env->init_sp = be_abi_reg_map_get(env->regs, sp);
        arch_set_irn_register(env->birg->main_env->arch_env, env->init_sp, sp);
@@ -1875,39 +1883,52 @@ static void modify_irg(be_abi_irg_t *env)
                if(arg_proj != NULL) {
                        be_abi_call_arg_t *arg;
                        ir_type *param_type;
-                       int nr = get_Proj_proj(arg_proj);
+                       int     nr = get_Proj_proj(arg_proj);
+                       ir_mode *mode;
 
                        nr         = MIN(nr, n_params);
                        arg        = get_call_arg(call, 0, nr);
                        param_type = get_method_param_type(method_type, nr);
 
-                       if(arg->in_reg) {
+                       if (arg->in_reg) {
                                repl = pmap_get(env->regs, (void *) arg->reg);
-                       }
+                       } else if(arg->on_stack) {
+                               ir_node *addr = be_new_FrameAddr(sp->reg_class, irg, reg_params_bl, frame_pointer, arg->stack_ent);
 
-                       else if(arg->on_stack) {
-                               /* For atomic parameters which are actually used, we create a StackParam node. */
+                               /* For atomic parameters which are actually used, we create a Load node. */
                                if(is_atomic_type(param_type) && get_irn_n_edges(args[i]) > 0) {
-                                       ir_mode *mode                    = get_type_mode(param_type);
-                                       const arch_register_class_t *cls = arch_isa_get_reg_class_for_mode(isa, mode);
-                                       repl = be_new_StackParam(cls, isa->bp->reg_class, irg, reg_params_bl, mode, frame_pointer, arg->stack_ent);
-                               }
+                                       ir_mode *mode      = get_type_mode(param_type);
+                                       ir_mode *load_mode = arg->load_mode;
+
+                                       ir_node *load = new_r_Load(irg, reg_params_bl, new_NoMem(), addr, load_mode);
+                                       set_irn_pinned(load, op_pin_state_floats);
+                                       repl = new_r_Proj(irg, reg_params_bl, load, load_mode, pn_Load_res);
 
-                               /* The stack parameter is not primitive (it is a struct or array),
-                               we thus will create a node representing the parameter's address
-                               on the stack. */
-                               else {
-                                       repl = be_new_FrameAddr(sp->reg_class, irg, reg_params_bl, frame_pointer, arg->stack_ent);
+                                       if (mode != load_mode) {
+                                               repl = new_r_Conv(irg, reg_params_bl, repl, mode);
+                                       }
+                               } else {
+                                       /* The stack parameter is not primitive (it is a struct or array),
+                                        * we thus will create a node representing the parameter's address
+                                        * on the stack. */
+                                       repl = addr;
                                }
                        }
 
                        assert(repl != NULL);
+
+                       /* Beware: the mode of the register parameters is always the mode of the register class
+                          which may be wrong. Add Conv's then. */
+                       mode = get_irn_mode(args[i]);
+                       if (mode != get_irn_mode(repl)) {
+                               repl = new_r_Conv(irg, get_irn_n(repl, -1), repl, mode);
+                       }
                        exchange(args[i], repl);
                }
        }
 
-       /* the arg proj is not needed anymore now */
-       assert(get_irn_n_edges(arg_tuple) == 0);
+       /* the arg proj is not needed anymore now and should be only used by the anchor */
+       assert(get_irn_n_edges(arg_tuple) == 1);
        be_kill_node(arg_tuple);
        set_irg_args(irg, new_rd_Bad(irg));
 
@@ -1916,7 +1937,9 @@ static void modify_irg(be_abi_irg_t *env)
                ir_node *irn = get_Block_cfgpred(end, i);
 
                if (is_Return(irn)) {
-                       ir_node *ret = create_be_return(env, irn, get_nodes_block(irn), get_Return_mem(irn), get_Return_n_ress(irn));
+                       ir_node *blk = get_nodes_block(irn);
+                       ir_node *mem = get_Return_mem(irn);
+                       ir_node *ret = create_be_return(env, irn, blk, mem, get_Return_n_ress(irn));
                        exchange(irn, ret);
                }
        }
@@ -1927,8 +1950,8 @@ static void modify_irg(be_abi_irg_t *env)
        obstack_free(&env->obst, args);
 
        /* handle start block here (place a jump in the block) */
-       temp = 0;
-       irg_block_walk_graph(irg, fix_start_block, NULL, &temp);
+       i = 0;
+       irg_block_walk_graph(irg, fix_start_block, NULL, &i);
 }
 
 /** Fix the state inputs of calls that still hang on unknowns */
@@ -1942,7 +1965,7 @@ void fix_call_state_inputs(be_abi_irg_t *env)
        /* Collect caller save registers */
        n = arch_isa_get_n_reg_class(isa);
        for(i = 0; i < n; ++i) {
-               int j;
+               unsigned j;
                const arch_register_class_t *cls = arch_isa_get_reg_class(isa, i);
                for(j = 0; j < cls->n_regs; ++j) {
                        const arch_register_t *reg = arch_register_for_index(cls, j);
@@ -2042,7 +2065,7 @@ be_abi_irg_t *be_abi_introduce(be_irg_t *birg)
 
        /* Make some important node pointers survive the dead node elimination. */
        survive_dce_register_irn(env->dce_survivor, &env->init_sp);
-       pmap_foreach(env->regs, ent) {
+       foreach_pmap(env->regs, ent) {
                survive_dce_register_irn(env->dce_survivor, (ir_node **) &ent->value);
        }
 
@@ -2167,6 +2190,7 @@ static int process_stack_bias(be_abi_irg_t *env, ir_node *bl, int bias)
        ir_node *irn;
 
        sched_foreach(bl, irn) {
+               int ofs;
 
                /*
                   Check, if the node relates to an entity on the stack frame.
@@ -2180,19 +2204,19 @@ static int process_stack_bias(be_abi_irg_t *env, ir_node *bl, int bias)
                        DBG((env->dbg, LEVEL_2, "%F has offset %d (including bias %d)\n", ent, offset, bias));
                }
 
-               /*
-                  If the node modifies the stack pointer by a constant offset,
-                  record that in the bias.
-                */
-               if(arch_irn_is(arch_env, irn, modify_sp)) {
-                       int ofs = arch_get_sp_bias(arch_env, irn);
+               if(omit_fp || be_is_IncSP(irn)) {
+                       /*
+                        * If the node modifies the stack pointer by a constant offset,
+                        * record that in the bias.
+                        */
+                       ofs = arch_get_sp_bias(arch_env, irn);
 
                        if(be_is_IncSP(irn)) {
                                if(ofs == BE_STACK_FRAME_SIZE_EXPAND) {
-                                       ofs = get_type_size_bytes(get_irg_frame_type(env->birg->irg));
+                                       ofs = (int)get_type_size_bytes(get_irg_frame_type(env->birg->irg));
                                        be_set_IncSP_offset(irn, ofs);
                                } else if(ofs == BE_STACK_FRAME_SIZE_SHRINK) {
-                                       ofs = - get_type_size_bytes(get_irg_frame_type(env->birg->irg));
+                                       ofs = - (int)get_type_size_bytes(get_irg_frame_type(env->birg->irg));
                                        be_set_IncSP_offset(irn, ofs);
                                }
                        }
@@ -2256,11 +2280,6 @@ ir_node *be_abi_get_ignore_irn(be_abi_irg_t *abi, const arch_register_t *reg)
        return pmap_get(abi->regs, (void *) reg);
 }
 
-ir_node *be_abi_get_start_barrier(be_abi_irg_t *abi)
-{
-       return abi->start_barrier;
-}
-
 /**
  * Returns non-zero if the ABI has omitted the frame pointer in
  * the current graph.