juanrga (nospam.delete@this.juanrga.com) on August 10, 2014 4:32 am wrote:

> The data will not change if you don't believe it. I know both software and hardware engineers that will say
> you that the ISA matters. The last one was Keller, who noted during AMD core conference that "ARMv8 doesn't
> require the same instruction decoding hardware as an x86 processor, leaving more room to concentrate on performance".
> He even mentioned that his K12 core will have a "wider engine" than its x86 sister core.
>
> Moreover, I am convinced that Filippo (not "Fillipo") knows that the A57 is about 10% faster when running
> the new ISA. Same arch. same program and 10% faster due only to change of ISAon 64bit more than

ISA certainly matters, but the interesting thing is there are only minor differences between ARM and Thumb-2 on one hand and AAarch64 on the other. If you look at recent GCC's the generated code for either is pretty similar (eg. both use load/store of 2 registers for function entry/exit).

So given the ISAs are very similar it is hard to believe there will be a big performance difference between 32 and 64-bit (the main differences are due to the extra registers and extra memory bandwidth). Neither do I think there will be a huge gain in removing the ARM and Thumb-2 decoders apart from saving some area. There won't be much difference after decode, if at all. That's because pretty much all instructions in ARM/Thumb-2 are in Aarch64 (any complex instructions will be cracked during decode).

Wilco