i have wasted several weeks worth of evenings on vectorizing heaps (4ary heaps: with SIMD, you’re not limited to binary heaps). It did not provide any speedup. I’d expect that halving heap depth would help but no. Still don’t know why.