We are not comparing the absolute numbers. We are comparing the ratios, i.e., the relative performance of copying contiguous data and of copying non-contiguous data. Why would the PCI bandwidth have an effect?
The code used is straight-forward. Here is an example with more details:
- For copying data which is contiguous in memory: An array A of size 512 x 512 was used to copy 512 x 512 elements, and the entire array was copied using OpenCL rectangular-copy; i.e., (0,0) to (511,511) was the rectangle within A which was copied. In this case, the 512 x 512 elements to be copied are contiguous in memory.
- For copying data which is not contiguous in memory: An array A of size 1024 x 1024 was used to copy 512 x 512 elements, and the first 512 column elements in the first 512 rows were copied using OpenCL rectangular-copy; i.e., (0,0) to (511,511) was the rectangle within A which was copied. In this case, the 512 x 512 elements to be copied are not contiguous in memory.
The performance of these two on the same GPU are compared.
We are using only 1 discrete GPU, and the other discrete GPU as well as the APU is idle.
