What is vector processing and how does it accelerate computer code?

  • Thread starter Thread starter Cerxes
  • Start date Start date
  • Tags Tags
    Code Computer
Click For Summary
SUMMARY

Vector processing accelerates computer code by allowing multiple operands to be processed simultaneously through pipelined math processors. This parallel operation enables faster execution of mathematical operations, such as matrix multiplication, by processing each row and column combination concurrently. Scoreboarding, a technique used to manage outputs from vector operations, was implemented in the CDC 6000 series, which predates the IBM 360. Understanding these concepts is crucial for optimizing performance in computational tasks.

PREREQUISITES
  • Understanding of vector processors and their architecture
  • Familiarity with pipelining in computer architecture
  • Knowledge of matrix multiplication algorithms
  • Awareness of scoreboarding techniques in CPU design
NEXT STEPS
  • Research the architecture and functionality of CDC 6000 series vector processors
  • Explore advanced matrix multiplication techniques and optimizations
  • Learn about pipelining and its impact on CPU performance
  • Investigate modern implementations of scoreboarding in contemporary processors
USEFUL FOR

Computer architects, software engineers, and anyone involved in optimizing computational performance will benefit from this discussion on vector processing and its applications.

Cerxes
Messages
1
Reaction score
0
What is "vectorized computer code"?
 
Technology news on Phys.org
Wikipedia has a very elementary description of it here.
 
Also refer to:

http://en.wikipedia.org/wiki/Vector_processor

The artilcles mention picking up multiple operands at the same time, but this depends on the width (how many bits) of the bus holding the operands. On vector processors the operands are pipelined into separate math processors, so the math processors can operate in parallel. If there are enough math processors, then math operations effectively occur as fast as the operands can be read from memory.

As noted in the vector processing article, instead of storing results, additional math operations can be peformed, such as summation of all the products produced in a vector multiply, a key component of matrix multiplaction. Each row / column combination are processed in parallel.

Note that scoreboarding can be used to coordinate the outputs from vector operations. The CDC 6000 series included scoreboarding, and date back to 1964, predating the IBM 360 by almost a year. (IBM announced their 360's in April of 1964, but never shippped any models until mid 1965, essentially the 1964 360's were "paper machines", they never existed).

http://en.wikipedia.org/wiki/Scoreboarding
 

Similar threads

Replies
7
Views
4K
  • · Replies 40 ·
2
Replies
40
Views
2K
Replies
3
Views
3K
Replies
11
Views
4K
  • · Replies 0 ·
Replies
0
Views
3K
  • · Replies 22 ·
Replies
22
Views
4K
  • · Replies 18 ·
Replies
18
Views
4K
Replies
1
Views
7K
  • · Replies 12 ·
Replies
12
Views
2K
  • · Replies 10 ·
Replies
10
Views
2K