primecount-8.6

This release introduces a new sieving algorithm for small sieving primes which increases Instruction-level parallelism (ILP) on today's wide out-of-order CPUs. The new method speeds up the D part of Gourdon’s prime counting algorithm by up to 45% on Apple Silicon CPUs and by up to 10% on AMD Zen5 CPUs.

ChangeLog

• Sieve.cpp: New sieving algorithm for small primes #122.
• count_simd.hpp: Prevent loop unrolling to reduce branch mispredictions.
• Sieve_count_simd.hpp: Fix undefined behavior.
• Sieve.hpp: Use Vector<uint64_t> sieve to fix undefined behavior.
• Sieve.cpp: Changes due to new sieve vector type.
• Sieve_pre_sieve.hpp: Changes due to new sieve vector type.
• Sieve_arrays.hpp: Changes due to new sieve vector type.
• Move the sieve source files into the new src/sieve directory.
• Fix LLVM/Clang -Wunused-template warning.
• Simplify ARM SVE runtime dispatching.