Is the leveling of growth in ILP due more to which of the following:
1) The fact that most code can't benefit from parallelism across more than 3-5 assembly instructions as mentioned in class and thus not much work was invested into trying any harder
or
2) real hardware limitations in improving the ILP
kayvonf
For traditional applications, it's definitely much more of (1). There's only so much parallelism to be found in a single instruction stream.
However, data-parallel applications like machine learning, graphics/image processing, etc. offer huge amounts of parallelism.
mlakshmi
I had to miss this lecture; could someone help me understand what we mean by a 'single instruction stream'?
Is the leveling of growth in ILP due more to which of the following:
1) The fact that most code can't benefit from parallelism across more than 3-5 assembly instructions as mentioned in class and thus not much work was invested into trying any harder
or
2) real hardware limitations in improving the ILP
For traditional applications, it's definitely much more of (1). There's only so much parallelism to be found in a single instruction stream.
However, data-parallel applications like machine learning, graphics/image processing, etc. offer huge amounts of parallelism.
I had to miss this lecture; could someone help me understand what we mean by a 'single instruction stream'?