University

 of  Washington  

Sec.on  2:  Integer  &  Floa.ng  Point  Numbers  

¢ Representa.on  of  integers:  unsigned  and  signed  
¢ Unsigned  and  signed  integers  in  C  
¢ Arithme.c  and  shiBing  
¢ Sign  extension  

¢ Background:  frac.onal  binary  numbers  

¢ IEEE  floa.ng-­‐point  standard  
¢ Floa.ng-­‐point  opera.ons  and  rounding  
¢ Floa.ng-­‐point  in  C  
 

Integers  
 University  of  Washington  

Unsigned  Integers  
¢ Unsigned  values  are  just  what  you  expect  
§ b7b6b5b4b3b2b1b0  =  b727  +  b626  +  b525  +  …  +  b121  +  b020  
§ Useful  formula:    1+2+4+8+...+2N-­‐1  =  2N  -­‐1  
 
¢ You  add/subtract  them  using  the  normal    
“carry/borrow”  rules,  just  in  binary  
 
63 00111111
+ 8 +00001000
71 01000111

Integers  
 University  of  Washington  

Signed  Integers  
¢ Let's  do  the  natural  thing  for  the  posi.ves  
§ They  correspond  to  the  unsigned  integers  of  the  same  value  
§ Example  (8  bits):  0x00  =  0,  0x01  =  1,  …,  0x7F  =  127  
¢ But,  we  need  to  let  about  half  of  them  be  nega.ve  
§ Use  the  high  order  bit  to  indicate  nega%ve:  call  it  the  “sign  bit”  
Call  this  a  “sign-­‐and-­‐magnitude”  representaQon  
§
§ Examples  (8  bits):    
§ 0x00  =  000000002  is  non-­‐negaQve,  because  the  sign  bit  is  0  
§ 0x7F  =  011111112  is  non-­‐negaQve  
§ 0x85  =  100001012  is  negaQve  
§ 0x80  =  100000002  is  negaQve…  

Integers  
 University  of  Washington  

Sign-­‐and-­‐Magnitude  Nega.ves  
¢ How  should  we  represent  -­‐1  in  binary?  
§ Sign-­‐and-­‐magnitude:    100000012  
Use  the  MSB  for  +  or  -­‐,  and  the  other  bits  to  give  magnitude  
 

–7 +0
–6 1111 0000 +1

–5 1110 0001 +2
1101 0010
–4 +3
1100 0011

– 3 1011 0100 + 4
1010 0101
–2 1001 0110 +5

–1 1000 0111 +6
–0 +7
Integers  
 University  of  Washington  

Sign-­‐and-­‐Magnitude  Nega.ves  
¢ How  should  we  represent  -­‐1  in  binary?  
§ Sign-­‐and-­‐magnitude:    100000012  
Use  the  MSB  for  +  or  -­‐,  and  the  other  bits  to  give  magnitude  
(Unfortunate  side  effect:  there  are  two  representaQons  of  0!)  

–7 +0
–6 1111 0000 +1

–5 1110 0001 +2
1101 0010
–4 +3
1100 0011

– 3 1011 0100 + 4
1010 0101
–2 1001 0110 +5

–1 1000 0111 +6
–0 +7
Integers  
 University  of  Washington  

Sign-­‐and-­‐Magnitude  Nega.ves  
¢ How  should  we  represent  -­‐1  in  binary?  
§ Sign-­‐and-­‐magnitude:    100000012  
Use  the  MSB  for  +  or  -­‐,  and  the  other  bits  to  give  magnitude  
(Unfortunate  side  effect:  there  are  two  representaQons  of  0!)  
§ Another  problem:  math  is  cumbersome  
§ Example:   –7 +0
–6 1111 0000 +1
4  -­‐  3  !=  4  +  (-­‐3)  
–5 1110 0001 +2
1101 0010
0100 –4 +3
+1011 1100 0011
1111
– 3 1011 0100 + 4
1010 0101
–2 1001 0110 +5

–1 1000 0111 +6
–0 +7
Integers  
 University  of  Washington  

Two’s  Complement  Nega.ves  

¢ How  should  we  represent  -­‐1  in  binary?  
§ Rather  than  a  sign  bit,  let  MSB  have  same  value,  but  nega%ve  weight  
§ W-­‐bit  word:  Bits  0,  1,  …,  W-­‐2  add  20,  21,  …,  2W-­‐2  to  value  of  integer  
when  set,  but  bit  W-­‐1  adds  -­‐2W-­‐1  when  set  
§ e.g.  unsigned  10102:    1*23  +  0*22  +  1*21  +  0*20  =  1010  
 2’s  comp.  10102:  -­‐1*23  +  0*22  +  1*21  +  0*20  =  -­‐610  
§ So  -­‐1  represented  as  11112;  all   –1 0
negaQve  integers  sQll  have  MSB  =  1   –2 1111 0000 +1
§ Advantages  of  two’s  complement:   –3 1110 0001 +2
only  one  zero,  simple  arithmeQc   1101 0010
–4 +3
§ To  get  negaQve  representaQon  of   1100 0011
any  integer,  take  bitwise  complement  
– 5 1011 0100 + 4
and  then  add  one!  
1010 0101
~x + 1 = -x –6 1001 0110 +5

–7 1000 0111 +6
Integers   –8 +7
 University  of  Washington  

Two’s  Complement  Arithme.c  

¢ The  same  addi.on  procedure  works  for  both  unsigned  and  
two’s  complement  integers  
§ Simplifies  hardware:  only  one  adder  needed  
§ Algorithm:  simple  addiQon,  discard  the  highest  carry  bit  
§ Called  “modular”  addiQon:  result  is  sum  modulo  2W  
¢ Examples:  

Integers  
 University  of  Washington  

Two’s  Complement  
¢ Why  does  it  work?  
§ Put  another  way:  given  the  bit  representaQon  of  a  posiQve  integer,  we  
want  the  negaQve  bit  representaQon  to  always  sum  to  0  (ignoring  the  
carry-­‐out  bit)  when  added  to  the  posiQve  representaQon  
§ This  turns  out  to  be  the  bitwise  complement  plus  one  
§ What  should  the  8-­‐bit  representaQon  of  -­‐1  be?  
00000001
+????????    (we  want  whichever  bit  string  gives  the  right  result)  
00000000

00000010 00000011
+???????? +????????
00000000 00000000

Integers  
 University  of  Washington  

Two’s  Complement  
¢ Why  does  it  work?  
§ Put  another  way:  given  the  bit  representaQon  of  a  posiQve  integer,  we  
want  the  negaQve  bit  representaQon  to  always  sum  to  0  (ignoring  the  
carry-­‐out  bit)  when  added  to  the  posiQve  representaQon  
§ This  turns  out  to  be  the  bitwise  complement  plus  one  
§ What  should  the  8-­‐bit  representaQon  of  -­‐1  be?  
00000001
+11111111    (we  want  whichever  bit  string  gives  the  right  result)  
100000000

00000010 00000011
+???????? +????????
00000000 00000000

Integers  
 University  of  Washington  

Two’s  Complement  
¢ Why  does  it  work?  
§ Put  another  way:  given  the  bit  representaQon  of  a  posiQve  integer,  we  
want  the  negaQve  bit  representaQon  to  always  sum  to  0  (ignoring  the  
carry-­‐out  bit)  when  added  to  the  posiQve  representaQon  
§ This  turns  out  to  be  the  bitwise  complement  plus  one  
§ What  should  the  8-­‐bit  representaQon  of  -­‐1  be?  
00000001
+11111111    (we  want  whichever  bit  string  gives  the  right  result)  
100000000

00000010 00000011
+11111110 +11111101
100000000 100000000

Integers  
 University  of  Washington  

Unsigned  &  Signed  Numeric  Values  

X   Unsigned   Signed  
l Both  signed  and  unsigned  integers    
0000   0   0  
have  limits  
0001   1   1   l If  you  compute  a  number  that  is  too  big,  
0010   2   2   you  wrap:    6  +  4  =  ?    15U  +  2U  =  ?  
0011   3   3  
If  you  compute  a  number  that  is  too  small,  
0100   4   4  
l

you  wrap:    -­‐7  -­‐  3  =  ?    0U  -­‐  2U  =  ?  

0101   5   5  
 
0110   6   6  
0111   7   7   l The  CPU  may  be  capable  of  “throwing  an  
1000   8   –8   excepQon”  for  overflow  on  signed  values  
1001   9   –7  
l But  it  won't  for  unsigned  
1010   10   –6  
 
1011   11   –5  
1100   12   –4   l C  and  Java  just  cruise  along  silently  when  
1101   13   –3   overflow  occurs...  
1110   14   –2  
1111   15   –1  

Integers  
 University  of  Washington  

Visualiza.ons  
¢ Same  W  bits  interpreted  as  signed  vs.  unsigned:  
  Two’s" 2w" 2w" Two’s"
 
complement" complement"
+2w–1" 2w–1" 2w–1" +2w–1"
 
  0" 0" 0" 0"
  –2w–1" Unsigned" Unsigned"
–2w–1"
¢ Two’s  complement  (signed)  addi.on:  x  and  y  are  W  bits  wide  
x + y!
+2w" Positive overflow"

+2w –1" +2w –1"

0" Normal" 0"

–2w –1" –2w –1"

Negative overflow"
–2w"
Integers  
 University  of  Washington  

Values  To  Remember  

¢ Unsigned  Values   ¢ Two’s  Complement  Values  
§ UMin  =  0   § TMin  =    –2w–1  
§ 000…0   § 100…0  
§ UMax    =    2w  –  1   § TMax    =    2w–1  –  1  
§ 111…1   § 011…1  
§ NegaQve  1  
§ 111…1        0xFFFFFFFF  (32  bits)  

Values  for  W  =  16  

Decimal Hex Binary
UMax 65535 FF FF 11111111 11111111
TMax 32767 7F FF 01111111 11111111
TMin -32768 80 00 10000000 00000000
-1 -1 FF FF 11111111 11111111
0 0 00 00 00000000 00000000
Integers