1 / 35

Representing fractions – Fixed point

Representing fractions – Fixed point. The problem: How to represent fractions with finite number of bits ?. Representing fractions – Fixed point. A number with 10 bits. a 1 a 2 a 3 a 4 a 5 a 6 a 7 a 8 a 9 a 10. Representing fractions – Fixed point. A number with 10 bits.

luisa
Download Presentation

Representing fractions – Fixed point

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Representing fractions – Fixed point • The problem: • How to represent fractions with finite number of bits ?

  2. Representing fractions – Fixed point A number with 10 bits a1a2a3a4a5a6a7a8a9a10

  3. Representing fractions – Fixed point A number with 10 bits a1a2a3a4a5a6a7a8a9a10 a1a2a3a4a5a6a7a8.a9a10 Fixing the point

  4. Representing fractions – Fixed point Range of representation:

  5. Fixed point : the problem • Cannot represent wide ranges of numbers. • In scientific applications.

  6. Representing Fractions – Floating point 1 * 101 10 Base (radix) - r -1.23 * 10-2 -0.123

  7. Representing Fractions – Floating point 1 * 101 10 exponent -1.23 * 10-1 -0.123

  8. Representing Fractions – Floating point 1 * 101 10 Number (Mantissa) -1.23 * 10-1 -0.123

  9. Representing Fractions – Floating point (-1)0*1 * 101 10 Sign bit (-1)1*1.23 * 10-1 -0.123

  10. Problem of uniqueness 100*10-4 0.1 Representation is not Unique 0.001*102

  11. Problem of uniqueness - Normalization 610*10-4 0.61 6.1*10-1 Standardization One digit to the Left of the point 0.0061*102

  12. Normalized Binary Floating point D = (-1)a0 * (1.a1a2a3…)*2b1b2b3… a0b1b2…bna1a2a3…am String of bits

  13. Floating point - Questions • Representing the (signed) exponent • How to represent zero? • And Nan, infinity ? • How to add, subtract and multiply? • Rounding Errors.

  14. Floating point – Representing the exponent How to represent singed number ? Sign bit 2-Complement

  15. Floating point – Representing the exponent How to represent singed number ? Sign bit Neither 2-Complement

  16. Floating point – Representing the exponent • We want the exponent to be binary ordered: 0000 < 0001 < …. < 1000 < … < 1111

  17. Floating point – Representing the exponent Number = Number - B Usually B = 2n-1-1 We define the following sizes like this: emin 000…0001 emax 111…1110

  18. Floating point – Representing zero,NAN, ± IEEE754 special values Denormalized number normalized number

  19. IEEE 754 (Including the sign Bit)

  20. What is NaN (not a number) Partial list

  21. Infinity • Provide a safe was to continue calculation when overflow is encountered.

  22. Calculations with Floating Point numbers • Addition: • Equalize the exponents (smallerlarger exponent) • Sum the mantissa • Renormalize if necessary

  23. Calculations with Floating Point numbers • Example (in base 10): |E| = 1 , |M| = 3 91  9.10*101 9.7  9.70*100

  24. Calculations with Floating Point numbers 9.10*101 + 9.70*100 Not The same Order.

  25. Calculations with Floating Point numbers 9.10*101 + 0.97*101 9.10*101 + 9.70*100 10.7*101 renormalize 1.07*102

  26. Calculations with Floating Point numbers • Example II (in base 10): |E| = 1 , |M| = 3 91  9.10*101 9.75  9.75*100

  27. Calculations with Floating Point numbers 9.10*101 + 9.75*100 Not The same Order.

  28. Calculations with Floating Point numbers 9.10 *101 + 0.975*101 9.10*101 + 9.75*100 10.75*101 renormalize 5 (rounding error) 1.07*102

  29. Rounding Errors The Problem: Squeezing infinite many real numbers into a finite number of bits

  30. Measuring Rounding Errors • Units in last place (Ulps) • Relative Error

  31. Measuring Rounding Errors – ULP p digits If d.dddd*re represent z error = |d.dddd – (z/re)|*rp-1

  32. Measuring Rounding Errors – ULP Example I: r = 10 , p = 3 The number 3.14*10-2 represents 0.0314159 Error = 0.159

  33. Measuring Rounding Errors – ULP • What is the maximum ULP if the rounding is toward the nearest number? 0.5 ULP

  34. Measuring Rounding Errors – Relative Error p digits If d.dddd*re represent z Relative error = |d.dddd*re – z|/z

  35. Measuring Rounding Errors – Relative errors Example I: r = 10 , p = 3 The number 3.14*10-2 represents 0.0314159 Relative Error ~ 0.0005

More Related