- Uses scratchpad interleaving to access only the closest L3 slice from each CPU core.
- Also activates MSR mod for cn-heavy because CPU prefetchers get confused with interleaving
- 7-8% speedup on Zen3
This patch updated SSE2NEON [1], which contains more functions
provided by Intel intrinsics, only implemented with NEON-based
counterparts to produce the exact semantics of the intrinsics.
Consequently, ARM-specific changes against CryptoNight_arm can
be reduced as well.
[1] https://github.com/DLTcollab/sse2neon/