/* Copyright (C) 1998 - 2000 by Universitaet Karlsruhe
-** All rights reserved.
-**
-** Authors: Goetz Lindenmaier
-**
-** testprogram.
+* All rights reserved.
+*
+* Authors: Goetz Lindenmaier
+*
+* testprogram.
*/
+# include <string.h>
+# include <stdio.h>
+
+# include "irvrfy.h"
# include "irdump.h"
# include "firm.h"
-/** This example describes a possible representation of heap allocated
-*** variables of imperative programs.
-*** It constructs the IR for the following program:
-***
-***
-*** main(): int
-*** int *a[10];
-***
-*** a = malloc(sizeof(a[10]));
-*** return (a[3]);
-*** end;
-***
-*** The array is placed on the heap. The pointer to the array that
-*** is a local variable is represented as a dataflow edge.
-*** There are two ways to model allocation to the heap in programs with
-*** explicit memory allocation:
-*** 1. Model the calls to malloc and free as simple procedure (of compiler
-*** known procedures returning a pointer. This is the simpler way of
-*** generating FIRM, but restricts the information that can be deduced
-*** for the call.
-*** 2. Insert an Alloc node. A later pass can lower this to the compiler
-*** known function. This makes the allocation explicit in FIRM, supporting
-*** optimization.
-*** A problem is modeling free. There is no free node in FIRM. Is this
-*** a necessary extension?
-*** This example shows the second alternative, where the size of the array
-*** is explicitly computed.
+/**
+* variables of imperative programs.
+* It constructs the IR for the following program:
+*
+*
+* main(): int
+* int *a[10];
+*
+* a = malloc(sizeof(a[10]));
+* return (a[3]);
+* end;
+*
+* The array is placed on the heap. The pointer to the array that
+* is a local variable is represented as a dataflow edge.
+* There are two ways to model allocation to the heap in programs with
+* explicit memory allocation:
+* 1. Model the calls to malloc and free as simple procedure (of compiler
+* known procedures returning a pointer. This is the simpler way of
+* generating FIRM, but restricts the information that can be deduced
+* for the call.
+* 2. Insert an Alloc node. A later pass can lower this to the compiler
+* known function. This makes the allocation explicit in FIRM, supporting
+* optimization.
+* A problem is modeling free. There is no free node in FIRM. Is this
+* a necessary extension?
+* This example shows the second alternative, where the size of the array
+* is explicitly computed.
**/
#define OPTIMIZE_NODE 0
This is the modeling appropriate for other languages.
Mode_i says that all integers shall be implemented as a
32 bit integer value. */
- prim_t_int = new_type_primitive(id_from_str ("int", 3), mode_i);
+ prim_t_int = new_type_primitive(id_from_str ("int", 3), mode_Is);
printf("\nCreating an IR graph: ARRAY-HEAP_EXAMPLE...\n");
# define U_BOUND 9
array_type = new_type_array(id_from_str("a", 1), N_DIMS, prim_t_int);
set_array_bounds(array_type, 1,
- new_Const(mode_I, tarval_from_long (mode_I, L_BOUND)),
- new_Const(mode_I, tarval_from_long (mode_I, U_BOUND)));
+ new_Const(mode_Iu, tarval_from_long (mode_Iu, L_BOUND)),
+ new_Const(mode_Iu, tarval_from_long (mode_Iu, U_BOUND)));
/* As the array is accessed by Sel nodes, we need information about
the entity the node selects. Entities of an array are it's elements
which are, in this case, integers. */
/* better: read bounds out of array type information */
size = (U_BOUND - L_BOUND + 1) * get_mode_size(elt_type_mode);
/* make constant representing the size */
- arr_size = new_Const(mode_I, tarval_from_long (mode_I, size));
+ arr_size = new_Const(mode_Iu, tarval_from_long (mode_Iu, size));
/* allocate and generate the Proj nodes. */
array = new_Alloc(get_store(), arr_size, (type*)array_type, stack_alloc);
set_store(new_Proj(array, mode_M, 0)); /* make the changed memory visible */
- array_ptr = new_Proj(array, mode_p, 2); /* remember the pointer to the array */
+ array_ptr = new_Proj(array, mode_P, 2); /* remember the pointer to the array */
/* Now the "real" program: */
/* Load element 3 of the array. For this first generate the pointer to this
by (three * elt_size), but this complicates some optimizations. The
type information accessible via the entity allows to generate the
pointer increment later. */
- c3 = new_Const (mode_I, tarval_from_long (mode_I, 3));
+ c3 = new_Const (mode_Iu, tarval_from_long (mode_Iu, 3));
{
ir_node *in[1];
in[0] = c3;
}
val = new_Load(get_store(), elt);
set_store(new_Proj(val, mode_M, 0));
- val = new_Proj(val, mode_i, 2);
+ val = new_Proj(val, mode_Is, 2);
/* return the result of procedure main */
{
add_in_edge (get_irg_end_block(main_irg), x);
mature_block (get_irg_end_block(main_irg));
+ finalize_cons (main_irg);
+
printf("Optimizing ...\n");
dead_node_elimination(main_irg);
printf("use xvcg to view these graphs:\n");
printf("/ben/goetz/bin/xvcg GRAPHNAME\n\n");
- return (1);
+ return (0);
}