Hands-on Introduction to the C Programming Language

HJ-82
Ontwerpen voor de optie multimedia en
signaalverwerking: seminaries

Hands-on Introduction to
the C Programming Language

Vincenzo De Florio
February 2003
V.ppt-1.3
2004-02-10

Introduction to the C Programming Language © V De Florio

Structure
2

•
•
•
•
•
•
•
•
•
•

Introduction
First examples
Variables
Fundamental data
types
Loops
Conditional statements
& other instructions
The C preprocessor
Functions
Variables again
Operators

•
•
•
•
•

•
•
•
•
•

Standard I/O functions
Derived types
Pointers and arrays
Use of pointers
Type casts &
conversions
Bitwise operators
Standard streams, files
Structures, bitfields,
unions
LLP
Data hiding

http://www.esat.kuleuven.ac.be/~deflorio/c/hj82c.prn
Introduction to the C Programming Language ( V De Florio)

Introduction
3

• C and UNIX : success stories.
• C: developed by Dennis Ritchie to port UNIX from a
PDP-7 to a PDP-11
• System programming language
• Most of the UNIX system is in C


Introduction
4

• C is simple:
A small number of simple and powerful
instructions…
…dealing with characters, numbers, addresses
No language facilities for handling complex
objects (strings, lists...)
No language facilities for the I/O, for memory
allocation...
…though those facilities are available as
standard (or user-defined) libraries of functions


Introduction
5

• C is small:
Easy to describe, easy to learn
The C compiler is considerably simple, compactsized, and easy to write
The C data types and control structures often
have a direct equivalent in the machine language
of many computers. This happens because C
was modelled after the machine language of the
PDP 11
Hence, while in general the translation from a
high level language instruction to machine
language is in general one-to-many, the
translation from C to machine language is oneto-few
A very simple run-time system: for instance
(PDP-11)

Introduction
6

• C is portable…
..though, to achieve this, we need to follow some
“rules of portability”
Types have no well-defined size
In OCCAM we have, e.g., int32,
On the contrary, in C we don’t have an a-priori
knowledge of the size of an integer

One can use symbolic constants for this
So called “#define” statements

A large set of standard (-> ported!) libraries exists
for I/O, string manipulation, memory allocation...


Introduction
7

• C provides:






the basic control-flow statements
statement grouping
selective statements
iterative statements
pointers + address arithmetic


First examples
8

Function with
no arguments
Function name

Program
starts here...

main ( )
{
printf(“hellon”);
}
Print string
“hello” and
character ‘n’

...and ends
here

First examples
9

• Log into the system
 login name
 password

• Type in the hello program in file first.c
 Activate an editor, e.g., xedit, pico, or vi
 for instance xedit first.c
 save the file at the end

• Compile the program
 cc -o first first.c (or gcc –o first first.c)

• Execute the program
 ./first

• Modify the program so that
 it prints another string / two strings / …
 jump to “Compile the program”

First Examples
10

printf
followed
by “(“
means
“Call
function
printf”


First examples
11

Loop
start,
loop
end

main() { int inf, sup, step;
float fahr, celsius;
inf = 0; sup = 300;
Integer and
step = 20;
real numbers
fahr = inf;
while (fahr <= sup) {
celsius = (5.0/9.0) * (fahr-32.0);
printf("%4.0f %6.1fn", fahr, celsius);
fahr = fahr + step;
Format
}
string
}

First examples
12


First examples
13

•
•
•
•

Type in the conversion program in file second.c
Compile the program
Execute the program
Modify the program:
 Change the format strings
 Change steps
 Downto vs. To
 (5 / 9.0)
 (5 / 9)
 fahr - 32 vs. fahr - 32.0
 fahr += fahr
 jump to “Compile the program”


Definition of Variables (1)
14

• To define a variable or a list of variables, one has to
mention:
TYPE LIST ;

• For instance:
int a, b_b, c2, A;
double f, F=1.3, dimX;
• With LIST one can also initialize variables, like it has
been done for F in the above example
• One can define variables only at the beginning of a
compound instruction such as
 { TYPE LIST; … ; INSTRUCTIONS }


First examples
15

main() { int inf, sup, step;
float fahr, celsius;
inf = 0; sup = 300;
init
step = 20;
test

}

fahr = inf;
while (fahr <= sup) {
celsius = (5.0/9.0) * (fahr-32.0);
printf("%4.0f %6.1fn", fahr, celsius);
fahr = fahr + step;
}
increment

for (fahr = inf; fahr <= sup; fahr += step) {...

First examples
16

• Fahr += step
==
Fahr = Fahr + step
Fahr++
==
Fahr = Fahr + 1
• Compound statements in C:
 { s1; s2; …; sn; }
 s1, s2, …, sn;
• Relational operators
 <=, <, !=, ==, >, >= ...
 Return 0 when false, 1 when true
• “test” is any operation
• 0 = false, non-0 = true
• Both these statements are valid:
while ( x == y ) { …
while ( x = y ) { …

First examples
17

1. Let a be 0
2. Repeat
the loop
while
“a = 0”
is “true”
3. So the
while loop
is simply
ignored!

First examples
18

1. Let a be 1
(returns 1)
2. Repeat
the loop
while
“a = 1”
returns “true”
3. So the
while loop
never ends!


First examples
19

The initial
value of a
is undefined!
Here is -85899…

Here is 0 (cygwin)

Fundamental data types
20

• Four types of integers:
char, short, int, long
• Two types of floating point numbers:
float, double
• Corresponding formats:
%c, %d, %d, %ld
%f, %lf
• Sizes:
sizeof(char) <= sizeof(short) <= sizeof(int) <=
sizeof(long)
sizeof(float) <= sizeof(double)
• Practically speaking:
char==1, short==2, int==2 or 4, long==4
float==4, double==8

21

• Check the actual size of type on your workstations
with function sizeof()


22

Same result on:
Linux / gcc
SunOS / (g)cc
HP-UX / cc
Win / ms visual c
But this is not
100% guaranteed
on all systems!

Type “char”
23

• A char in C is just an integer
• The only difference between a char and, e.g., an int
lies in the number of bytes that are used
• In C, we can write, e.g.
int i;
char c;
i = ‘A’;
c = 65;
if ( i != c ) printf(“not an “);
printf(“ASCII systemn”);
• A char is a small number that, when sent, e.g.,
to the screen, corresponds to a character
• In C there is no difference between
A character
The code of character

Type “char”
24

• Symbols such as ‘A’ are just symbolic constants
• Think of them as a #define statement:
#define
#define
...

‘A’
‘B’

65
66

• Every time you encounter a
APOSTROPHE CHARACTER APOSTROPHE
you are dealing with a symbolic constant that
defines a small integer, that when printed can be
interpreted as a character


Type “char”
25

• Example
• A = 66;
printf(“ As a number, A is t%dn”, A);
printf(“ As a character, A is t%cn”, A);

%c ==
“print as a
character”


%d ==
“print as
an integer”

Types: some formatting characters
26

• %d : print as an integer
%c : print as a character
%ld : print as a long
%f : print as a float
%lf : print as a double


27


28


Types: an interpretation of a same
“substance” – bytes!
29

(To be explained later on)

The first four bytes of
A are interpreted in
different ways

Loops in C
30

While loops: while ( op1 ) op2;
• No initialisation

• Both op1 and op2 are operations
returning integers
• while op1 is non zero, do op2
• op2 can be a compound instruction, e.g.,
{ op21; op22; … ; op2N }
• relational operations return an integer!
• a <= b is 1 when a<=b and 0
otherwise

Loops in C
31

• Which of these loops are correct? And what do they
mean?
 while ( -25 ) a++ ;
 while ( read_next_character ( ) ) car = car + 1;
 while ( a + 1 ) { a = a + 1; b = 0; }
 while ( a ) a = b, b = tmp, tmp = a;


Loops in C
32

While loops: do op2 while ( op1 );
• No initialisation

• Both op1 and op2 are operations
returning integers
• Do op2; then, while op1 is not zero, do
op2 again
• op2 can again be a compound
instruction, e.g.,
{ op21; op22; … ; op2N }

Loops in C
33

•
•
•
•
•
•

For loops: for ( op1; op2; op3 ) op4;
Operation op1 is the initialisation
Operation op2 is the condition
Operation op3 is the conclusive part
Operation op4 is the body of the loop,
Both op1 and op3 can be compound (commabased!) instructions
• Usually used for iterations, e.g.
for (i=0; i<n; i++) { do_that(); }

• Exercise: modify second.c so to change the
while into a for.
• Exercise: modify the previous example so to
print the table in inverse order.

Loops in C
34

• Which of these loops are correct? And what do they
mean?
 for (;;) a++;
 for (; a<5; a++) ;
 for ( i=0; i<n; { i++; n--; } ) … /* do something */
 for ( i=0, j=n; i<n; i++, j-- ) … /* do sthg */


Loops in C
35

• Loops are an important source of performance for
the compiler and for optimizers
• Some computer architectures and compilers can
exploit them to achieve a very high speed-up
• This is called
“loop level parallelism (LLP) exploitation”


Conditional statements
36

• Statement “if”:
 if (condition) action1; else action2;
• “?:”
 (condition)? action1 : action2;
• Statement “switch”:
 switch(integer expression) {
case value1: op1; break;
case value2: op2; break;
…
case valuen: opn; break;
default: opn+1;
}

Other instructions
37

• break;
• continue;
• goto label;

goto y;
y:
while (a < b) {
a++;
…
break; continue;
b--;
}
goto x;
x:
…

The C Preprocessor
38

• The C compiler was originally composed of four
components:
 cpp | cc | as | ld
 .c -> .i -> .s -> .o

• Component cpp is the C preprocessor
 cpp looks for preprocessor commands (#cmd’s)
#cmd’s start with “#” on column 1
 cpp converts a text file f into a text file f’
 All the valid “#”-statements are removed and
substituted with C strings or text files


The C Preprocessor
39

•
•
•
•
•
•
•
•

Examples:
#define BEGIN
{
#define END
}
Dangerous
#define IF
if(
#define THEN
)
#define INT32
long
#define MAX(A, B) ((A) > (B)? (A):(B))
#define SQR(x) x * x

• IF SQR(x) == x THEN printf(“x==1n”);
is translated into
if( x * x ==x ) printf(“x==1n”);

The C Preprocessor
40

•
1.
2.
3.
4.

Inclusion of external files
#include <stdio.h>
#include “assoc.h”
#ifdef … [ #else … ] #endif
#ifndef … [ #else … ] #endif

•
•

Differences between 1. and 2.
Use of 3. and 4.


The C Preprocessor
41

•
•
•

#include <stdio.h> is equivalent to
#include “prefix/stdio.h”
On many UNIX systems, prefix == /usr/include

•

/usr/include contains the header files of the C
standard library
 stdio.h, string.h, time.h, ctype.h, math.h, …
A header file is the user interface of a library or a
part of a library
stdio.h defines the interface to a subset of the C
standard library
stdio.h interfaces the standard I/O functions and
defines symbols thereof

•
•
•


Functions
42

• Functions are names that points to some memory
location where code has been stored by the
compiler
• They take arguments and return a value of some
type
 E.g., double cos(double a);
 This is a function prototype, with which we declare a
function, called cos, that expects and returns a double

• To call a function, we just name it and pass an
argument to it
 cos (3.1415); /* cos (pi) */


Functions
43

• The name of a function is just a number – its
address in the code segment


Functions
44

• Functions can be
 Declared
 Defined

• A prototype declares a function
 “There’s a function that looks like that and that…”

• Defining a function means specifying what the
function shall do
 “The function does this and this…”
 Memory is allocated in the code segment


Functions
45

• To define a function one has to supply the
compound instruction that it has to execute:

double addd (double a, double b);
double addd (double a, double b)
{
return a + b;
}

Prototype (declaration)
Definition

Functions
46

• Recursion is allowed
• C supports a single-level of functions that
can be compiled separately


Functions
47

• return closes the function and returns an output
value (default: integer) to the caller.
• Arguments are passed by value: this means that the
arguments are copied in temporary variables.
• The only way to let a function modify an argument is
by passing the address of the object to be modified.
• Operator & returns the address of a variable.


Functions
48

• Functions have the following structure:
• [ type ] name ( [ args ] ) {
[ declarations ] [ instructions ]
}
• int main() {
int i; int power (int, int);
for (i=0; i<10; ++i)
printf("%d %dn", i, power(2,i));
}
int power (int x, int n) { int i, p;
for (i = p = 1; i <= n; ++i)
p=p*x;
return (p);
}

Again, on Variables
49

• Two classes of variables
 Dynamically allocated
 Statically allocated

• A variable is dynamically allocated when it is local to
a function or a compound instruction and is not
declared as “static”
• Examples
Automatic
main () {
variables
int i;
(hold undefined
for (i=0; i<n; i++) {
int j;
value)
j = i * i;
}
}

Again, on Variables
50

• If the variable is an automatic one, then the
initialisation is done each time the variable is
allocated (each time the function in which the
variable resides is called and the corresponding
compound instruction is executed).


Again, on Variables
51

• A variable is defined at compile time when
 it is a global variable (I.e., a variable defined outside any
function)
 It is a variable defined as “static”

• Example
int fd;
int open(int tmp) {
int j;
static int first;
}

Global
Automatic
Static


Again, on Variables
52

• C is case sensitive: int J; float j; is valid
• The scope of variables is such that a new name
overrides an old one:
int h = 3;
{ int h = 4;
printf(“h == %dn”, h);
}
printf(“h == %dn”, h);


Operators
53

•
•
•
•

binary +, -, *, /, %
unary precedences:
(+,-)  (*,/,%)  (-)  (||)
 (&&)  (==, !=)  (> >= < <=)

• Note:pa  opb if opb has higher precedence than opa

54

• Memory model
• Memory is a long array of cells, each ofsize 1 byte
• When I say
{ int a;
what actually happens is
addr = please-os-allocate(4)
now a==(addr, [addr-addr+3])
• When I say
}
what actually happens is
os-please-free(4,addr)
• Automaticly!
• AUTOMATIC variables

Operators
55

• Expressions such as:
i < lim && (c=getchar()) != ’n’ && c != EOF
do not require extra parentheses:
=  !=, hence parentheses are required around
c=getchar().

• Logic clauses are evaluated left-to-right. Evaluation
stops when the truth value of a logic expression is
found.


Standard functions for input/output
56

• Defined in stdio.h
• Require the stdio.h file to be included
• This defines functions such as:
int getchar();
int putchar(int c);
• getchar reads the next character in the standard
input stream or an end-of-file value (EOF)
• getchar returns the character as one byte stored
into a 2-byte or 4-byte integer (an int value)
• Putchar puts on the standard output stream the
character we pass as an argument


57

void main()
{ char c;

c is declared
as a char

c = getchar();
while ( c != EOF ) {
putchar ( c );
c = getchar();
}

getchar either
returns a
char or EOF

}


Are any
faults
present?
Which
ones?

58

void main()
{ char c;

c is a char
getchar
returns
c = getchar();
an int!
while ( c != EOF ) {
putchar ( c );
This loop
c = getchar();
may never be
}
verified

}


59

void main()
{ int c;

OK

implicit
parentheses

while ( c = ( getchar() != EOF ) ) {
putchar ( c );
}
Are any
}
faults
present?
If stdin = ‘h’ ‘e’ ‘l’ ‘l’ ‘o’, EOF
Which
what goes to stdout?
ones?


Exercises
60

• Exercise: the just seen program can be easily
adapted to work, e.g., as a “word counter” (reports
the number of characters, words, and lines in the
input), or as a filter to remove blanks or tabs, and so
forth
• Input and output can be redirected with < and >.
Pipelines of programs can be built by chaining the
programs with |


A memory model for didactics
61

• Memory can be thought as finite, long array of cells,
each of size 1 byte
0

1

2

3

4

5

6

7

…

• Each cell has a label, called address, and
a content, i.e. the byte stored into it
• Think of a chest of drawers, with a label
on each drawer and possibly something
into it


62

4

Content

3
2

1


Address

63

• The character * has a special meaning
• It refers to the contents of a cell

• For instance:

*(1) ==

This means we’re inspecting the contents
of a cell (we open a drawer and see what’s in it)

64

• The character * has a special meaning
• It refers to the contents of a cell

• For instance:

*(1) =

This means we’re writing new contents
into a cell (we open a drawer and change its contents)

Derived types
65

• The following operators define complex types
derived from the basic types of C:
* operator pointer-to,
[] operator vector-of,
() operator function-pointer-to


Derived types
66

• char *p; : p is of type “pointer-to-chars”
• float v[10]; : v is a vector, i.e., a pointer to the
beginning of a memory area allocated by the system
and consisting of 10 floats, i.e., v[0], . . . , v[9].
• int getX(); : getX is a pointer to a function
returning an int.


Derived types
67

• Operator [] has higher priority with respect to
operator *. This means that

int *v[10]
means that v is a vector of ten pointers-to-int.
Parentheses are necessary to change the meaning:

int (*v)[10]
means that v is a pointer to a vector of ten integers
• What’s the difference in terms of sizes?


Derived types – examples
68

1.
2.
3.
4.

int *pi;
char **argv;
float *(fvp)[5];
long (*fp)(int);

pointer to integer;
pointer to pointer-to-char;
pointer to vectors-of-5-floats
pointer to function reading an
int and returning a long.


Derived types
69

• The address-of (&) operator returns the address of a
variable
char c; char *pc; pc = &c;
• Operator *, applied to a pointer, returns the pointed
object.
char c1 = ’a’; char *p = &c1;
char c2 = *p; /* c2 == ’a’ */
void func(char *s) { printf("bye %sn", s) }
main() { void (*pfunc)(char*) = func;
*(pfunc)("hello");
}


Derived types
70

void swap (int a, int b) { int t;
t = a; a = b; b = t;
}

These are
not the same
variables!

void main() { int a, b;
a = 5, b = 6;
swap (a, b);
printf(“a = %d, b = %dn”, a, b);
}


Are any
faults
present?
Which
ones?

Derived types
71

Step

Var Address Contents

int a, b;
a = 5,

SP

a
b
a
b

b = 6;

swap(a, b)
int a, int b
int t;
t = a;a = b;b = t

1024
1028
1024
1028

?
?
5
6

1028
1032

a
b
t
a
b
t

1032
1036
1040
1032
1036
1040

5
6
?
6
5
5

1036
1040
1044

}

1032

a
b

1024
1028


5
6

Derived types
72

• Pointers solve the problem of the lack of “call-byreference” in C functions. For instance, in order to
realize a function that swaps its argument, one may
use the following strategy:
int swap(int *a, int *b) { int t;
t = *a, *a = *b, *b = t;
}
• The caller then needs to specify the addresses of
the operands it wants to swap, as in
swap(&i, &j).


Derived types
73

void swap (int* a, int* b) { int t;
t = *a; *a = *b; *b = t;
}
void main() { int a, b;
a = 5, b = 6;
swap (&a, &b);
printf(“a = %d, b = %dn”, a, b);
}


Derived types
74

Step
a = 5,

Var Address Contents
b = 6;

SP

a
b

swap(&a, &b)
int *a, int *b
int t;
t=*a;*a=*b;*b=t

1024
1028

5
6

1028
1032

a
b
t
*a
*b
t

1032
1036
1040
1024
1028
1040

1024
1028
?
6
5
5

1036
1040
1044

}

1032
a
b

1024
1028


6
5

Working with the debugger
75

#include <stdio.h>
main()

Big, big mistake…

{
char *p = NULL;

printf("dereferencing a NULL pointer!n");
*p = 'A';

printf("done.n");
}

76


A useful tool!
77


78

Where did the fault take place?


79


80

Commands:

l=list

b=breakpoint
r=run

s=step
(also
display var)


Command “print”
81


Command “print”
82


83


84

Command
“set”


Pointers and arrays
85

• Note: declaring an array means
1. declaring a pointer
2. allocating memory for the pointed objects.
• The name of the array is indeed a pointer to the first
element of the array. In C lingo, this is
written as vect == &vect[0].
• Exercise: write a program, called report, that reports
the occurrences of each digit char in the input
stream. Use an array of ten elements corresponding
to the ten decimal digits. Produce a histogram at
end-of-input.


Pointers and arrays
86

• Exercise: use two programs,
one that outputs the ten integer numbers that
count the occurrences of each digit char in the
input stream,
the other one that creates a histogram of its input
values.
• Then use a pipeline to hook the two programs:
report2 | histogram


Pointers and arrays
87

• Arrays and pointers are strictly related to each
other: In particular, if int a[10]; then
a == &a[0], a+1 == &a[1], … and so forth.
• In other words, *(a+i) == a[i]
• Any indexed array is equivalent to a pointer plus
some offset, and vice-versa
• Big difference: the array is a constant, i.e., it can
never appear on the left of the = sign, as in
a = pa;

or in
a ++;

Pointers and arrays
88

• When passing an array to a function we are indeed
passing the address of its first element; this address
is copied (call-by-value) in a temporary variable.
This variable may be a pointer.
• Example:
char s[7] = "hello!"; /* s is an array */
int i = strlen(s);
int strlen (char *x) { /* x is a pointer */
int n;
for (n=0; *x; x++) /* hence, can be modified */
n++;
return (n);
}

Pointers and arrays
89

• If p is a pointer, p++ lets p point to the next item. It is
the language that takes care of moving p of the right
amount of memory.
• For instance (let us assume an int is 2 bytes and a
double is 8 bytes):
int *p;
p == 1222
p+1 == 1224
double *p; p == 5644
p+1 == 5660
and so forth
• In other words: if p is a pointer to an object of type t,
then p+n points n objects further and p-n points n
objects before.
• The actual size of the object doesn’t matter.


Pointers to chars
90

• Strings are available in C as arrays of characters.
Any sentence enclosed between two quotes (“) is an
array of characters ending with character ’0’
(NULL).
• For instance, "hello" means ’h’, ’e’, ’l’, ’l’, ’o’, 0
• A very common mistake when learning C:
char s1[10] = "Hello ";
char s2[10] = "World";
s1=s2; /* ERROR */


Pointers to chars
91

• As strings are arrays, we can easily pass a string to
a function by passing its name, which points to its
characters:
char a[] = “Kennedy";
printf("Vote %s for president.n", a);
/* a = &a[0] */
• Variables defined within a function cannot be “seen”
from other functions.


Pointers to functions
92

• Exercise: write a program that uses an array of
pointers-to-function
Input char == number I -> call array[I]


Pointers and arrays (2)
93

• Function strcpy(char *a, char *b);
• Assumption: NULL-terminated strings. Note that
NULL is (int)0, i.e., “false”
• Function strcmp(char *s, char *t): returns a negative
number if s < t, 0 if s == t, a positive number if s > t:
strcmp(char *s, char *t) {
for ( ; *s == *t; s++, t++)
if (*s == ’0’) return (0);
return (*s - *t);
}


94

• Is this #define OK?
• #define STRCPY(a,b) while (*a++ = *b++) ;


95

• To declare multidimensional arrays one declares
arrays of arrays. For instance,

static int day_of_month[2][13] = {
{ 0, 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31},
{ 0, 31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31}
};
int day_of_year(int day, int month, int year) {
int i, leap =
year%4 == 0 && year%100 != 0 || year%400 == 0;
for (i=1; i<month; i++)
day += day_in_month[leap][i];
return (day);
}

96

• int v[i][j] vs. int v[i,j]
• Entries are stored “by row”: the rightmost index is
the one that varies the most when entries are
referenced in the order they are stored.
• Initialisation: using curly brackets.
• When passing a bidimensional array to a function, it
is mandatory that the number of columns be
specified. For instance:
f(int day_in_month[2][13]), or
f(int day_in_month[][13]), or
f(int (*day_in_month)[13]),
• i.e., pointer to array-of-13 integer entries.


97

• “Entries are stored by rows” means that, e.g.,
int v[100][200];
• is allocated the same way as a
int v[100 × 200];
i.e., as if it were a mono-dimensional array of
20000 int’s.


98

• Fortran stores objects the opposite way with respect
to C:
for (i..) for (j..) a[i][j]
• is equivalent to
DO I.. DO J.. A(J,I)
• Accessing element (i, j) in a n × m matrix means
accessing element k = i × m + j in the associated
mono-dimensional array.
• The same applies for N-dimensional
matrices, N > 2.


99

• Note how, in order to compute value k for an N
dimensional matrix whose dimensions are
(d1, d2, . . . , dN), it is necessary to know the values
d2, . . . , dN:
k = f(d2, . . . , dN).
• Note also that when we need to access sequentially
all the elements of a multidimensional matrix, it is
preferable to use a pointer initialised to the first
entry of the matrix.


100

• #define N 500
#define M 500
main() { int a[N][M];
int i, j, c;
int *pa = & (a[0][0]);
int *pl = & (a[N-1][M-1]);

while (pa < pl) {
*pa = 0;
c = *pa + 1;
*pa = c;
pa++;
}

for (i=0; i<N; i++)
for (j=0; j<M; j++)
{ a[i][j] = 0;
c = a[i][j] +1;
a[i][j] = c;
}

HP9K 0.7–0.8s

1.2–1.3 s

SUN3 1.1 s

2.4 s


101

• Even when the access is not sequential, it is
possible to exploit specific access patterns (e.g.,
constant stride access).
• An interesting alternative with respect to
multidimensional arrays is by using pointers. For
instance, the computational cost to access
entry (i, j) in a n × m matrix is the one for computing
multiplication (i * m) and addition (i * m + j).


102

• If we change from
int a[100][100];
to
int** a;
and if we properly allocate the 100 pointers in the
row and, for each of them, the memory required for
storing 100 int’s, then
accessing entry (i, j) means executing *((*(a+i)+j))
that has a computational cost of only two additions.
• Furthermore, no dimension information is required
anymore to access any element of the array.


103

• Array of pointers – an example
• char *name_of_month (int n) {
static char *names[] = {
"illegal",
"January",
"February",
...
"December“
};
return ((n < 1 || n > 12)? names[0] : names[n] ;
}


104

• Argc and argv: a mechanism that allows a program
to inspect the strings on the command
• line.
• For instance, command echo:
main(int argc, char *argv[]) {
while (--argc > 0)
printf("%s%c", *++argv, (argc>1)?’ ’:’n’ );
}
• Exercise: compute 2 3 4 + * (in inverse Polish
notation).
• Exercise: write a function that associates user
functions to the command options.


105

• Exercise: write a function that sorts the entries of an
array of integers. Modify the function so that it
requires a pointer-to-function,
int (*confr)(int,int),
to realize sortings in increasing vs. decreasing
order.
• Exercise: “opaque” function, working with pointers to
object of unknown size.
• Exercise (array of functions): design a scanner of a
simple grammar. Tokens must correspond to the
entries in an array of functions.


Using pointers
106

• Using pointers in C may be described as a
connection-oriented service
• One has to open a connection to the memory
service, then use the service (read/write mode),
then disconnect (close) the service
• Example:
int *pi;
pi = malloc(4); /* open:memory-alloc:4 bytes */
*pi = 5; /* use:write-memory:store-in(p,5) */
j = 1 + *pi; /* use:read-memory:get-from(p) */
free(pi); /* close:free-memory-referred-in(p) */

Using pointers
107

1) Definition of a pointer
int *pi;
Note: by doing so, one allocates memory for the
pointer, not for the pointed
2) Memory allocation
pi = malloc(4); or better malloc(sizeof(int));
Note: this is a request for memory allocation.
malloc returns NULL (def:stdio.h) when the
request cannot be fulfilled, a value different from
NULL when the request can be fulfilled
3) Memory access
*pi = …
or
… = … *pi …
Note: the valid address returned by malloc points to
some memory location where the requested
memory resides

Using pointers
108

4) Memory release
free(pi);
Note: the memory pointed to by pi is freed
Note: pi is not “erased”! Still it holds the stale
reference (be careful…)


Using pointers
109

• A common mistake:
char *p;
*p = …
• This is faulty (use-before-connect fault)
• Weird consequences…


Using pointers
110

• A common mistake:
free(p);
*p = *p + 1;
• This is faulty (use-after-disconnect fault)
• Weird consequences…


Using pointers
111

void main() {
int *a, *b;
int c[10], d[10];

No check
on return
values

a = malloc(10*sizeof(int));
The area
b = calloc(10, sizeof(int));
pointed by
a = b;
a is lost
a = c;
Illegal:
d = a;
arrays are
}
const


Real-life
experiences
112

Courtesy
Siddika
Berna Ors
(COSIC)

• Commenting a
real life code
• What does this
excerpt do?

• Comment
on the
following:


Real-life
experiences
113

• Check out this
code
• Check what is
clear and what
is not
• Try to figure
out the
meaning of “?”
and “%”
• What does
mp_add do?


Real-life experiences
114

• Check out this code
• Check what is clear and what is not
• Try to figure out the meaning of the program


Constants
115

• scientific notation, e.g., 1.5e3 -> double
• postfix notation, e.g., 145L -> long
• prefix notation:
’0x44’ -> unsigned int, hexadecimal,
’044’ -> unsigned int, octal
• constant char: ’x’ = character x -> char
• special constants, e.g., n, t, 0, , " -> char
• “bit patterns”: ddd, ddd being an octal number.
• string constants, e.g., "string" or "".


Type casts and conversions
116

• Implicit type casts occur when, in expressions,
operands belong to different types.
• Conversions obey the following rules:
char and short  int , float  double
if an operand is double , the other becomes
double and the result is double
otherwise, if an operand is long , the other
becomes long and the result is a long
otherwise, if an operand is unsigned, the other
one becomes unsigned and the result is
unsigned.
otherwise, operands and result are int .


117

• Converting a string of digits into a number, and viceversa, requires specific support. Functions are
available for this, e.g., atoi():
int atoi(char s[]) { int i, n;
n = 0;
for (i=0; s[i]>=’0’ && s[i]<=’9’; ++i)
n=10*n + s[i] -’0’;
return (n);
}

• Note how expression s[i] - ’0’ converts numeric
character s[i] into the digit it represents.

118

• The following function can be used to convert an
uppercase character into its lowercase counterpart
int lower( int c)
{
if (c >=’A’ && c <= ’Z’)
return (c + ’a’ - ’A’);
else
return (c);
}
• Note: this program only works for code tables in
which ’a’ follows ’A’. This is true with ASCII and
false with, e.g., EBCDIC.


119

• Explicit cast:
• The cast operator changes the type of an object.
For instance, the following expression:
celsius = ( (double)5 /9) * (fahr-32.0);
is equivalent to
celsius = (5.0/9.0) * (fahr-32.0);
• Casting is invoked by specifying a type between
parentheses.


Increment/decrement operators
120

• IN C:

int i = 0;
i++;

in Assembly:
DATA segment
i DB 0
...
INC i
• Operators such as ++ or -- may have a direct
counterpart in the machine language.
• Operator ++ increments the contents of a variable.
x = n++ is not equivalent to x = ++n.
• Operator -- decrements the contents of a variable.
x = n-- is not equivalent to x = --n.


Exercises
121

• Exercise: function purge(char s[], int c) removes all
occurrences of c from s[].
• Exercise: functions strcpy() and strcat().


Exercises
122

• int strlen (char *p) { int i=0;
while (*p++) i++;
return i;
}
• *p returns the char pointed to by p. *p++ does the
same, but increments the pointer afterwards.
• When does the while loop ends?
• This is an alternative way to write function strlen:
int strlen (char *p) { char *q = p;
while (*p++) ;
return q - p - 1;
}


Bitwise operators
123

• The following six operands can be applied to any
integer expression:
& : bitwise AND
| : bitwise OR
ˆ : bitwise exclusive OR
<< : left shift
>> : right shift
˜ : unary complement.


Bitwise operators
124

• Bitwise AND can be used to set up “masks”, e.g.,
c = n & 0177;
which zeroes all the bits from bit 8 onward.
• Bitwise OR sets bits:
x = x | MASK
sets to 1 the bits that are set in MASK.
• << and >> are respectively arithmetical shifts to the
left and to the right (multiplication re: division by 2).
• Operator ˜ turns each 1 into 0 and vice-versa; it is
used in expressions like, e.g.,
x & ~ 077,
which zeroes the last bits of x. Note that this is
independent of the actual size of x, and hence it is
“more portable” than, e.g., x & 0177700.

Bitwise operators
125

• Let’s consider the following function:
unsigned int
MIDbit (unsigned x, unsigned start, unsigned num)
{
return((x >> (start+1-num)) & ~(~0 << num));
}
• For instance, MIDbit(x, 4, 3) returns the three bits at
position 4, 3, and 2.
• x >> (p+1-n) shifts the bits of interest on the
rightmost position in the word.
• ~ 0 is 11...1;
n zeroes
• n shifts to left lead to 11...1 00..0
n ones
• complementing this value we reach 00...0 11..1

Bitwise operators
126

• Binary operators
+ . / % << >> & ˆ and |
can use notation
e1 op= e2
instead of
e1 = (e1) op (e2).
• Note that x *= y+1 is equivalent to x = x*(y+1).
• An example based on botwise operators follows:
int bitcount(unsigned n) { int b;
for (b=0; n != 0; n >>= 1)
if (n & 01)
b++;
return (b);
}

The FILE class
127

• Using files in C may be described as a connectionoriented service
• One has to open a connection to the file service,
then use the service (read/write mode), then
disconnect (close) the service
• Example:
FILE *pf;
pf = fopen(“readme”, “r”); /* openfile:(readme,rm) */
fprintf(pf, “hin”); /* usefile:store-chars(pf,”hin”) */
fread(buffer,1,1,pf); /*usefile:read-chars(pf,1,buf) */
fclose (pf); /* closefile(pf) */

Structures
128

• Keyword struct defines a “record.” An example
follows:
struct cd {
char author[30];
int year;
char producer[20];
};
• This is a declaration of a type: no element of this
type has been defined so far. No memory has been
allocated.
• It is now possible to declare objects of type struct cd
struct cd x, y, z;


Structures
129

• The members of a struct can be accessed via the
“dot-operator” (“.”). For instance,

struct cd d;
leap = d.year%4 == 0 &&
d.year%100 != 0 ||
d.year%400 == 0;
• Nesting of structures is allowed:
struct a { struct disco c; } d;
• Access: d.c.year.
• Typedefs.


Structures
130

• A pointer to a structure can be declared, e.g., as
follows:
struct disco *a;
• Access:
(*a).year;
• a->year is equivalent to (*a).year


Typedef
131

• What does, e.g., this statement do?
float complex[2];
• Consider now
typedef float complex[2];
• We have defined a new type
• Example:
complex a;
a[0] = 0.0, a[1] = 1.0;

Typedef
132

• General rule:
consider the definition of a complex object, called x
write

“typedef x”

Now x is no more an identifier. It’s a type


Typedef
133

• Example:
int func_t (int, int, float);
This defines func_t, a pointer-to-function
typedef int func_t (int, int, float);

This defines a new type, called func_t. Now
func_t myF;

is equivalent to
int myF(int int, float);

Typedef
134

• There are two valid reasons for encouraging the use
of typedef:
parametrizing a program with its types may turn
into semplifying the solution of portability
problems: for instance,
typedef short integer;
Moving the code to a machine where the role of
short is played by int we only need to change
one line of code:
typedef int integer;
enhancing a program’s readability: a type called
LENGTH brings with its name also a hint at its
usage and meaning within the program.

Structures
135

• Arrays of structures:
• struct key {
char *keyword;
int keycount;
} keytab[100];
• Or
typedef struct key { … } key_t;
and then
key_t keytab[100];


Structures
136

• key_t keytab[100];
• Initialisation:
key_t Keywords[] = {
"break", 0,
"case", 0,
"char", 0,
…
"while", 0
};
• Exercise: Write a program that writes a static arrays
of struct’s to be used as look-up table for another
program.

Structures
137

• Structures can reference themselves:
struct tnode {
char *word; int count;
struct tnode *left;
struct tnode *right;
};
• Another example:
struct nlist {
char *name; char *def;
struct nlist *next;
};

Structures
138

• Exercise:
Write a set of functions dealing with linked lists
Structure the set of functions as a service
Open a connection to the file service, then use
the service (read/write/delete…), then disconnect
(close) the service


Bitfield structures
139

• Flag registers:
#define KEYWORD 01
#define EXTERN 02
#define STATIC 04
#define AUTOMT 010
…
flags |= EXTERN | STATIC;
…
if ( flags & (EXTERN | STATIC) ) ...
• It is possible to store in a same variable, flags, a
series of conditions that can be “turned on” through
a bitwise OR (|) and can be tested via the bitwise
AND (&).
• Clearly the definitions must be powers of 2. Octal
implies unsigned.

Bitfield structures
140

• Structures can be used to specify ag registers via
so-called “bitfields”:
struct {
unsigned is_keyword : 1;
unsigned is_extern : 1;
unsigned is_static : 1;
unsigned is_regist : 1;
} flags;
• Accessing a field: standard way (“.” operator). For
instance: flags.is_extern.


Bitfield structures
141

• Bitfields can be used as lvalues and rvalues as any
other integer variable.
• This means that bitwise OR’s and AND’s are not
necessary:
flags.is extern = 0
if (flags.is extern == 0) ...


Bitfield structures
142

•
•
•
•

Bitfields can only be unsigned or int
Asking the address of a bitfield is illegal
Bitfields can also be “unnamed,” e.g., for padding
The standard doesn’t specify if MSB-first or LSBfirst is used.


Unions
143

• An union is a variable having many identifiers
associated to it.
• These identifiers may be of different type and hence
different sizeof’s. Each identifier refer to the same
amount of memory, i.e., as many bytes as the
“largest” type requires. For instance:

union point_x {
char *pc;
float *pf;
double *pd;
long regarded_as_an_int;
} object;


Unions
144

• Variable object has many “aliases”: it can be
regarded as a pointer-to-char, if referred to as
object.pc, or as pointer-to-double (object.pd), or as
a long (object.regarded as an int).
• The amount of memory is always the same, the one
that is enough in order to store the “largest” among
a (char*), a (float*), a (double*), or a (long).


Unions
145

• Typical use of unions:
struct {
int its_type;
union {
int ival;
float fval;
char *pval;
} uval;
} symbol_table[500];

• If we store in symbol table[i].its type a code that
represents the type of object i, then we can set up,
e.g., arrays of objects of different types, or linked
lists of different objects, and so forth.

Unions
146

• A switch on its type is the typical way to execute
diverse actions depending on the nature of the
current object, as it’s done in the following example:
for (i=0;i<500;++i)
switch(symbol_table[i].its_type) {
case ’i’: printf("%d",
symbol_table[i].uval.ival);
break;
…
}


Standard libraries
147

• In C many functions are defined in the so-called
“standard library”, libc.a. A set of headers refer to
the functions and definitions in the standard library.
The header file stdio.h contains the basic functions
and definitions for I/O:
#include <stdio.h>


Standard streams
148

• C and UNIX define three standard I/O streams:
stdin, i.e., the standard input stream, normally
given by the characters typed at the keyboard.
As in DOS, the redirection character < allows to
specify an alternative standard input stream as
follows:
prog < inputfile
stdout, the standard output stream, normally
given by the characters typed onto our display.
Character > redirects stdout on an other file, as
in
prog > outputfile.
stderr, the standard error stream, is again by
default our display. Is the stream where the
programmer does (should) send the error
messages. Can be redirected via 2 >, e.g.,
prog 2> /dev/null.

Pipes
149

• Besides the stream redirection facilities, UNIX
provides the concept of pipe: if we type

prog1 | prog2
the stdout of prog1 becomes the stdin of prog2.


Pipes
150

• pipe() creates an I/O mechanism called “pipe” and returns
two file descriptors, fildes[0] and fildes[1]. The files
associated with fildes[0] and fildes[1] are streams and
are both opened for reading and writing.
• A read from pipeline[0] accesses the data written to
pipeline[1] and a read from pipeline[1] accesses the data
written to pipeline[0] (both on a FIFO basis.)
int pipeline[2];
pipein = pipeline[STDIN], pipeout = pipeline[STDOUT];
pipe(pipeline);
stdin

TASK 2

pip e

ut

TASK 1

stdout

eo

stdin

in

pi p

stdout

Pipes
151

• dup2() duplicates an open file descriptor
• dup2(int fildes, int fildes2)
• The dup2() function causes the file descriptor fildes2 to
refer to the same file as fildes. The fildes argument is a
file descriptor referring to an open file.

void main() {
dup2(1,2);
printf(“hello”);
fprintf(stderr, “ worldn”);
}
$ ./a.out
hello world

Pipes
152

• STDOUT == 1, STDIN ==0
• Task 1’s stdout is redirected to task 2’s stdin
dup2(pipeout, STDOUT);

TASK 1

stdout

stdin

pip
h e l l o 0

e ou

p ip e

in

puts(”hello”);

TASK 2

stdout

t

stdin

dup2(pipein, STDIN);

gets(msg); // msg is “hello”

Pipes
153

• Pipes are also available as FILE*: for instance,
FILE *popen(char *command, const char *type);
int pclose (FILE *stream);
• the popen() function creates a pipe between the
calling program and the command to be executed.
command consists of a shell command line. type is
an I/O mode, either r for reading or w for writing
• The value returned is a stream pointer such that one
can write to the standard input of the command, if
the I/O mode is w, by writing to the file stream; and
one can read from the standard output of the
command, if the I/O mode is r, by reading from the
file stream.

Pipes
154

1. FILE *f = popen(”date”, ”r”);
2. FILE *g=popen(”/bin/sort”, ”w”);
1. with the first one we can, e.g., read the output of
command date.
2. The second one connects to service /bin/sort so to
ask that external service (in this case, to sort a
stream)


The UNIX man
155

• The UNIX command “man” is an important tool
• If you don’t remember how a certain functions is to
be invoked, or what is does, just type
man function
• Try for instance
 man printf
 man putc
 man strcpy
 man tolower


Makefiles
156

• A makefile is a script that tells how to compile an
application
• It specifies the dependencies among a number of
resources (header and C files)
• An example follows:
all:

myfile.exe

myfile.exe: myfile.o myobj.o
cc –o myfile.exe myfile.o
myobj.o

myfile.o:
myfile.c common.h
cc –c myfile.c
myobj.o: myobj.c common.h myobj.h
cc –c myobj.c

VLIW

Main instruction
memory
157

128 bit

Instruction Cache
128 bit

Instruction Register
32 bit each

Dec

Dec

Dec

Dec
256 decoded bits each

EU

EU

EU

Register file

EU
32 bit each; 8 read ports, 4 write ports

32 bit each; 2 read ports, 1 write port

Cache/
RAM

32 bit;
1 bi-directional port

Main data
memory

VLIW
158

• Properties
 Multiple Execution Units: multiple instructions issued in one
clock cycle
 Every EU requires 2 operands and delivers one result
every clock cycle: high data memory bandwidth needed


VLIW
159

• Properties
 Every EU requires an instruction every clock cycle
 Waste of memory and performance when not enough parallel
instructions are available to keep all EUs concurrently busy
 Many instruction slots will contain a NOP


VLIW
160

• Properties
 Compiler should determine which instructions can be
issued in a single cycle without control dependency conflict
nor data dependency conflict
 Deterministic utilization of parallelism: good for hard-real-time
 Compile-time analysis of source code: worst case analysis
instead of actual case
 Very sophisticated compilers, especially when the EUs are
pipelined! Perform well since early 2000


VLIW
161

• Properties
 Compiler should determine which instructions can be
issued in a single cycle without control dependency conflict
nor data dependency conflict
 Very difficult to write assembly:





programmer should resolve all control flow conflicts
all data flow conflicts
all pipelining conflicts
and at the same time fit data accesses into the available data
memory bandwidth
 and all program accesses into the available program memory
bandwidth
 e.g. 2 weeks for a sum-of-products (3 lines of C-code)

 All high end DSP processors since 1999 are VLIW
processors (examples: Philips Trimedia -- high end TV, TI
TMS320C6x -- GSM base stations and ISP modem arrays)


Loop Level Parallelism
162

We assume to work with a simple loop such as
for (I=1; I<=1000; I++)
x[I] = X[I] + s;
•




Note: each iteration is independent of the others
Very simple case
We can “unroll the loop”
This provides the compiler with more chances to fill
in more slots and execute more instructions in the
same cycle!


Loop unrolling: dependencies
163

• Dealing with data dependencies
• Two classes of methods:
1. Keeping the dependence though avoiding the
hazard (via scheduling)
2. Eliminating a dependence by transforming the code


Loop unrolling: 1. dependencies
164

• Class 2 implies more work
• These are optimization methods used by the
compilers
• Detecting dependencies when only using registers
is easy; the difficulties come from detecting
dependencies in memory:
• For instance 100(R4) and 20(R6) may point to the
same memory location
• Also the opposite situation may take place:
LD 20(R4), R2
…
ADD R3, R1, 20(R4)
• If R4 changes, this is no dependency

165

• Let us consider the following loop:
for (I=1; I<=100; I++) {
A[I+1] = A[I] + C[I]; /* S1 */
B[I+1] = B[I] + A[I+1]; /* S2 */ }
• S1 is a loop-carried dependency (LCD):
iteration I+1 is dependent on iteration I:
A’ = f(A)
• S2 is
B’ = f(B,A’)
• If a loop has only non-LCD’s, then it is possible to
execute more than one loop iteration in parallel – as
long as the dependencies within each iteration are
not violated


166

• What to do in the presence of LCD’s?
• Loop transformations. Example:
for (I=1; I<=100; I++) {
A[I+1] = A[I] + B[I]; /* S1 */
B[I+1] = C[I] + D[I]; /* S2 */ }
•
A’ = f(A, B)
B’ = f(C, D)
• Note: no dependencies except LCD’s
Instructions can be swapped!


167

• What to do in the presence of LCD’s?
• Loop transformations. Example:
for (I=1; I<=100; I++) {
A[I+1] = A[I] + B[I]; /* S1 */
B[I+1] = C[I] + D[I]; /* S2 */ }
• Note: the flow, i.e.,
A0 B0
A0 B0
C0 D0
C0 D0
A1 B1 can be
A1 B1
C1 D1 changed
into
C1 D1
A2 B2
A2 B2
C2 D2
...
...

168

for (i=1; i <= 100; i=i+1) {
A[i] = A[i] + B[i]; /* S1 */
B[i+1] = C[i] + D[i];
/* S2 */
}

becomes
A[1] = A[1] + B[1];
for (i=1; i <= 99; i=i+1) {
B[i+1] = C[i] + D[i];
A[i+1] = A[i+1] + B[i+1];
}
B[101] = C[100] + D[100];


Loop Level Parallelim
169

• A’ = f(A, B)
B’ = f(C, D)
B’ = f(C, D)
A’ = f(A’, B’)
• Now we have dependencies but no more LCD’s!
It is possible to execute more than one loop iteration
in parallel – as long as the dependencies within
each iteration are not violated


170

• The original code:
for (x=GB; x<=N-1-GB; ++x)
for (y=GB; y<=M-1-GB; ++y) {
gauss_x_compute[x][y][0]=0;
for (k=-GB; k<=GB; ++k)
gauss_x_compute[x][y][GB+k+1] =
gauss_x_compute[x][y][GB+k] +
image_in[x+k][y]*Gauss[abs(k)];
gauss_x_image[x][y] =
gauss_x_compute[x][y][(2*GB)+1]/tot;
}
is transformed in 3 intermediate substeps.

171

• for (x=0; x<N; ++x)
for (y=0; y<M; ++y)
if (x>=GB && x<=N-1-GB &&
y>=GB && y<=M-1-GB) {
for (k=-GB; k<=GB; ++k) ...
gauss_x_image[x][y]=
gauss_x_compute[x][y][(2*GB)+1]/tot;
}
• In version 1, the loop bounds of the second loop
nest with the horizontal filtering are modified to
match the bounds of the initialisation loop preceding
it


172

• In version 2, the initialisation loop is split into two
pieces to match the condition of the second loop
nest with the horizontal filter.
for (x=0; x<N; ++x)
for (y=0; y<M; ++y)
if (x>=GB && x<=N-1-GB && y>=GB && y<=M-1-GB)
gauss_x_image[x][y]=0;
else
for (x=0; x<N; ++x)
for (y=0; y<M; ++y)
if (x>=GB && x<=N-1-GB && y>=GB && y<=M-1-GB) {
for (k=-GB; k<=GB; ++k) ...
gauss_x_image[x][y]= gauss_x_compute[x][y][(2*GB)+1]/tot;
}

173

• In version 3, the initialisation loop is then fully merged with the
second loop nest:
for (x=0; x<N; ++x)
for (y=0; y<M; ++y)
if (x>=GB && x<=N-1-GB && y>=GB && y<=M-1-GB) {
for (k=-GB; k<=GB; ++k) ...
gauss_x_image[x][y]= gauss_x_compute[x][y][(2*GB)+1]/tot;
}
else
}
• This gauss_x_compute[x][y][0]=0; statement can then be
worked into the k loop to remove it fully.

Performance issues
174

• How much can the choice of a data structure
influence a property such as locality in sequential
data accesses? In order to evaluate this we run a
simple experiment on a Win2000/Pentium3 PC


Performance issues
175

• Experiment: the following structure:
typedef struct {
int a;
char b[STRIDE];
} stru_t;
and the following access scheme:
for (i=0; i<itera-1; i++)
v[i].a = v[i+1].a +
1;
are compared with the following two data structures:
typedef int stru_ta;
typedef char stru_tb[STRIDE];
and access scheme:
for (i=0; i<itera-1; i++)

a[i] = a[i+1] + 1;


176

• Parameters: STRIDE, itera, optimisation level (compiler:
cygwin/gcc 2.95.3-5)
•
• The following script is used, where itera goes from 100000 to
2000000 by 100000, and the actual value is an average of 10
run per value of itera:
•
• gcc -DSTRU –O3 stru.c # -O1, -O2, -O3
• #gcc -DSTRU stru.c
# or no optimisation at all
• for i in 100000 200000 300000 400000 500000 600000
700000 800000 900000 1000000 1100000 1200000 1300000
1400000 1500000 1600000 1700000 1800000 1900000 2000000
#i==itera
• do
•
rm -f $i.stru
•
for j in 0 1 2 3 4 5 6 7 8 9
•
do
•
( a.exe $i ) >> stru/$i.stru 2>&1

177

STRIDE == 4 (y axis: usecs; x axis: (x+1)*100000)

178

STRIDE == 8 (y axis: usecs; x axis: (x+1)*100000)

179

STRIDE == 12

180

STRIDE == 16

181

The following pictures plot vertical that
represent the gain in performance between the best “-O3”
cases for different values of STRIDE:

Stride = 4


182

Stride = 16

183

Stride = 32


184

Stride = 64

185

Stride = 128

References
186

• Kernighan & Ritchie,The C programming Language,
Addison-Wesley


Hands-on Introduction to the C Programming Language

More Related Content

What's hot (20)

Viewers also liked (18)

Similar to Hands-on Introduction to the C Programming Language (20)

More from Vincenzo De Florio (20)

Recently uploaded (20)

Hands-on Introduction to the C Programming Language