Performance drop in matrix multiplication for certain sizes on AMD Polaris...

What is the best practice for memory access in this N-body problem solved on AMD Radeon RX580?...

SIMD-16 and SIMD-32 advantage/disadvantage?...