AVX-512 is the sweet spot: a cache line as 64 bytes/512bits (event RISC-V RV22+ mandates a 64 bytes cache line).
Hopefully, they will start to unroll nasm macros in ffmpeg assembly, because it is vendor locking that assembly code on nasm, and HARD: "exit cost" towards gas or fasm2 or even a simpler assembler is made stellar because of nasm pre-processor abuse. It start to feel like c++ and similar with their 2 compilers, lol.
Hopefully, they will start to unroll nasm macros in ffmpeg assembly, because it is vendor locking that assembly code on nasm, and HARD: "exit cost" towards gas or fasm2 or even a simpler assembler is made stellar because of nasm pre-processor abuse. It start to feel like c++ and similar with their 2 compilers, lol.