Romain Dolbeau (romain.delete@this.dolbeau.org) on July 14, 2020 11:27 am wrote:
> Obviously SVE doesn't have that particular issue, that's true. Whether this will translate into
> a market advantage is less obvious; why would anyone rewrite 128-bits NEON code to SVE scalable
> code if most SVE implementations are 128-bits and offers next to no benefits performance-wise?


That depends on what ARMv9 ends up being, and whether NEON is a part of it. If NEON is not included as a part of the ARMv9 ISA, then ARMv9 code will use SVE. Obviously ARMv9 implementations would be forced to execute both ARMv8 and ARMv9 code, much like they used to (and most still do) execute both ARMv7 and ARMv8, and thus support NEON. But someday there would be ARMv9 only implementations that no longer support NEON, just as there are already ARMv8 only implementations today.

There's also the 'line in the sand' mentality for specific implementations. If the first ARM Macs support SVE, there is no reason for anyone to use NEON for Mac code. They could if they really wanted to since the CPUs would obviously execute it as part of the ARMv8 ISA, but Apple would probably list it as deprecated on ARM Macs from day one to discourage its use and serve notice it will go away someday (even if 'someday' isn't for 5-10 years)