If we run something like
https://gridwise-webgpu.github.io/gridwise/examples/scan_sort_perf.html?primitive=sort_keys&inputLengthStart=8388608&inputLengthEnd=33554432
then the plotted sort bandwidth is pretty modest. I want to check if it is reasonable. I need to go look at how I am calculating sort data volume and probably do some lower-level benchmarking.
If we run something like
https://gridwise-webgpu.github.io/gridwise/examples/scan_sort_perf.html?primitive=sort_keys&inputLengthStart=8388608&inputLengthEnd=33554432
then the plotted sort bandwidth is pretty modest. I want to check if it is reasonable. I need to go look at how I am calculating sort data volume and probably do some lower-level benchmarking.