By: Foo_ (foo.delete@this.nomail.com), March 20, 2021 10:50 am
Room: Moderated Discussions
Hugo Décharnes (hdecharn.delete@this.outlook.fr) on March 20, 2021 8:55 am wrote:
> I haven’t said “invent an IR” but “deliver programs in IR”.
You mean execute IR on the CPU? The problem is either the IR is close enough to the machine, and you're just inventing another instruction set (with no obvious reason why it would be better than e.g. Aarch64 or x86-64), or it's significantly higher level than a traditional instruction set, and you'll end up spending tons of transistors on hardware translation, again with no obvious reason why spending silicon and power doing that instead of actually *executing* instructions would be a good idea.
> I haven’t said “invent an IR” but “deliver programs in IR”.
You mean execute IR on the CPU? The problem is either the IR is close enough to the machine, and you're just inventing another instruction set (with no obvious reason why it would be better than e.g. Aarch64 or x86-64), or it's significantly higher level than a traditional instruction set, and you'll end up spending tons of transistors on hardware translation, again with no obvious reason why spending silicon and power doing that instead of actually *executing* instructions would be a good idea.