load feature placements into shared memory before iterating

[AdityaGupta1]

for block size of 128, load 128 features into shared memory and iterate over those, then load next 128 features and repeat