Page 127 - Computer_Science_F5
P. 127
Computer Science This parallel execution significantly register concurrently. Vector registers can
accelerates computations compared to hold an array of data values of the same
processing elements one at a time.
type, like integers, floating-point numbers,
or booleans.
FOR ONLINE READING ONLY
Instruction Stream
Modern processors often provide
special instruction set extensions (ISA
Extensions) specifically designed for
Processing Unit
SIMD operations. These extensions
offer a rich set of instructions optimised
Data Stream Processing Unit efficient manipulation of data elements
Processing Unit
for vector processing, allowing for
within a vector register.
Processing Unit SIMD ISA extensions provide instructions
for performing arithmetic operations
Figure 2. 14: A 4-lane SIMD architecture (addition, subtraction, multiplication,
etc.), logical operations (AND, OR, NOT,
SIMD supports vectorization etc.), comparison operations (greater
SIMD is mainly dedicated to array than, less than, equal to, etc.), and data
movement operations (loading and storing
processing machines. However, vector data from memory to vector registers)
processors can also be seen as a part of on all elements of a vector register
this group. SIMD provides the hardware simultaneously. Examples of SIMD ISA
capability for vectorisation, which is Extensions:
the process of adapting algorithms to (a) SSE (Streaming SIMD Extensions): A
handle data in vectors (arrays) rather popular SIMD ISA extension for x86
than processing single data points one processors offering a wide range of
at a time. The data elements reside in instructions for various data types.
special registers called vector registers.
SIMD relies on specialised hardware (b) AVX (Advanced Vector
units called vector processing units Extensions): An extension to SSE
(VPUs). These VPUs contain multiple that provides wider vector registers
processing elements (PEs) that act and additional instructions for
like mini-processors. When a SIMD improved performance.
instruction is issued, each PE within the (c) AltiVec: A SIMD ISA extension for
VPU executes the same instruction on its PowerPC processors with similar
corresponding data element in the vector functionalities to SSE.
118
for Advanced Secondary Schools
Computer Science Form 5.indd 118 23/07/2024 12:33
