Changes between Version 27 and Version 28 of SoC2013

Timestamp:

Mar 27, 2013, 9:16:11 PM (10 years ago)

Author:

christopher_kormanyos

Comment:

Add improved description for multiprecision

SoC2013

@@ -158 +158 @@
 * First of all, we will closely monitor the progress of this project and decide what precision ranges we would like to support.
 * The easiest precision level involves floating-point numbers with less than about 1,000 decimal digits of precision.
-* Further levels climb to thousands or even millions of digits, but these require algorithms of increasing complexity.
-* One of the first steps is to providing I/O handling for character-based strings and C++ <iostream> support.
-* The next important step involves implementing the basic algebraic operations (+, +, *, /) using a variety of techniques.
-* Multiplication is the performance bottleneck in multiprecision, and depending on the range goals, we may optionally include high-performance with Karatsuba multiplication and potentially ultra high performance and high digit counts with FFT-based multiplication.
-* Optionally we will be providing support for the elementary transcendental functions listed in <cmath>, as is now done for the existing back-ends.
+* Further precision levels climb to thousands or even millions of decimal digits. These require algorithms of increasing complexity.
+* One of the first steps is to provide I/O handling for character-based strings and C++ <iostream> and <iomanip> support.
+* The next important step involves implementing the basic algebraic operations (+, -, *, /), in other words addition, subtraction, multiplication, and division, using a variety of techniques.
+* Multiplication is the performance bottleneck in multiprecision and we will be spending a lot of time here. Depending on the digit-range goals, we may optionally include high-performance with Karatsuba multiplication and potentially ultra-high performance with FFT-based multiplication.
+* Optionally, we will be providing support for the elementary transcendental functions listed in <cmath>, as is now done for the existing back-ends.
 * Finally, we need to ensure that the back-end seamlessly interoperates with Boost.Math.
 * Testing is essential in order to verify the correctness of the implementation. We will, therefore, be writing a variety of tests to validate the algebra as well as the functions.
+
+The existing radix-10 implementation in <boost/multiprecision/cpp_dec_float.hpp> can serve as a guide for candidates interested in visualizing the level of complexity involved in this project. In addition, a sketch consisting of mainly a feasibility check for radix-2 can be found in the sandbox.
 
 Heavy use will be made of
@@ -174 +176 @@
 * Fast Fourier transforms
 * Conversion to and from radix-2 and radix-10
-
-The existing radix-10 implementation in <boost/multiprecision/cpp_dec_float.hpp> can serve as a guide for candidates interested in visualizing the level of complexity involved in this project. In addition, a sketch consisting of mainly a feasibility check for radix-2 can be found in the sandbox.
 
 The new reference book "Modern Computer Arithmetic" is a valuable source for the algorithms in this project. A draft version of the book is available free of charge and the printed copy can be found in a library or at a book seller.