Here, parallelization can be thought of as 'vectorizing' loop instructions so cut iteration counts, but one must be mindful that A: doing so uses more resources, and B: highly-parallel programs will most likely hit a bandwidth wall, and after that extra parallelization will not improve performance (see later slides for a demo of this)
Here, parallelization can be thought of as 'vectorizing' loop instructions so cut iteration counts, but one must be mindful that A: doing so uses more resources, and B: highly-parallel programs will most likely hit a bandwidth wall, and after that extra parallelization will not improve performance (see later slides for a demo of this)