1. Instruction sets and data types

High-level languages define a number of data types, for both integers and floats. For integers, there are 8-bit, 16-bit, 32-bit and 64-bit signed and unsigned variants. For floats, there are 32-bit (single-precision) and 64-bit (double-precision) floats. This wide range of formats reflects the fact that applications have different requirements:

• image processing uses bytes (8 bits), with one byte for grayscale processing and three bytes for color processing;

• signal processing generally uses 16-bit data, etc.

With conventional, so-called scalar, instruction sets, arithmetic operations on integer data exist in just one format, corresponding to the number of bits in the instruction set's registers: 32-bit for 32-bit processors, 64-bit for 64-bit processors. When memory accesses...