Quiz submission record for quiz3-2-1 at Tue Jul  6 20:56:50 2004:

Your Answer for Question 1:
Because of the fact that bit patterns have no inherent meaning and can represent integers, floats, characters, instructions, etc, much emphasis is placed on the instruction that operates on the bits in each word. The constraints, of course, are that computer numbers are finite. Thus, sacrifices must be made somewhere, specifically in the case of floating-point operations. One technique probably would be to add additional hardware to increase accuracy/range, but the cost to this would obviously be slower processing rate. (One of the key concepts to hardware design is to keep things small and simple.) 

Your Answer for Question 2:
A normalized number is a number in scientific notation that has no leading 0s. A subnormal is otherwise known as a denormalized number. Subnormals allow a number to degrade in significance until it becomes 0 (known as gradual underflow). Essentially, this enables greater precision from a floating-point operation. 

Your Answer for Question 3:
In the old days, many programmers wrote (0.5 – x) + 0.5 instead of 1.0 –x probably because of the fact that floating-point addition is not associative. Due to the nature of how the addition is done and to rules of how many decimal digits can exist, programmers must try and maintain the most accurate and precise solution possible. After all, computer arithmetic is finite. Hence, by doing it the first way, it is probably easier to achieve greater accuracy if say x was a small decimal number less than 0.5.

Your Answer for Question 4:
The advantages of IEEE 754 are that it is a widespread standard with a common floating-point format, requiring minimum accuracy to one-half ulp in the least significant bit, and that the operations must be commutative. However, at the same time, during IEEE 754’s creation, some of the features did not balance language and compiler issues, such has humane exception handling and directed rounding. Such features, in my opinion, ought to be integrated in the next generation of math co-processors to get greater accuracy when dealing with floating-point arithmetic. 


Your unique submission ID is quiz3-2-1-cs61c-ei-1089172610-664.