Parallelism and Computer Arithmetic

Parallelism and Computer Arithmetic

• As modern computers such as smartphones, tablets, and laptops began to include built‒in graphical displays, it became necessary for their processors to handle graphics efficiently. Over a time, as the number of available transistors on a chip increased, processor designers started adding special hardware support for graphics operations.

• In computer graphics, colors are represented using the three primary color components‒red, green, and blue (RGB). Typically, each of these colors is represented by 8 bits (1 byte), giving a total of 24 bits per pixel (8 bits × 3 colors). An additional 8 bits may be used for other information such as pixel location or transparency, making a 32‒bit color representation common.

• Similarly, when audio features such as speakers and microphones were added for video conferencing and games, support for sound data processing also became essential. Audio signals need higher precision than 8 bits, so each audio sample is often represented using 16 bits (2 bytes).

• Most traditional integer programs rarely perform arithmetic operations on small data types like bytes (8 bits) or halfwords (16 bits). However, multimedia applications‒such as image, video and sound processing‒perform many repetitive operations on these smaller data units. For example:

■ Adjusting the brightness of every pixel in an image, or

■ Changing the volume of every audio sample in a sound file.

• Performing such operations one data element at a time would be inefficient. To speed this up, architects introduced parallel processing within a single instruction, known as subword parallelism.

• A modern processor can operate on wide data words (For example, 128 bits) in a single instruction. By dividing this wide data path into smaller parts, the processor can perform the same operation on multiple smaller data elements simultaneously.

For instance:

■ Sixteen 8‒bit (byte) values, or

■ Eight 16‒bit (halfword) values, or

■ Four 32‒bit (word) values, or

■ Two 64‒bit (doubleword) values

can all be processed in parallel within one instruction.

• This means a 128‒bit adder can perform multiple additions at once‒a technique that provides large performance gains with very little additional hardware cost.

• Because this form of parallelism allows a single instruction to operate on multiple data elements, it is also known as SIMD (Single Instruction, Multiple Data). SIMD is a type of Data‒Level Parallelism (DLP), where the same computation is applied to several pieces of data simultaneously.

Review Questions

1. What is subword parallelism? Explain with an example.

2. Why is SIMD useful in multimedia processing ?