3 ms·
I'm still unconvinced by the modular nature of the ISA. Are we going to end up shipping fat binaries with a dozen binaries for all the different variants that a
by IshKebab 8y ago
I'm still unconvinced by the modular nature of the ISA. Are we going to end up shipping fat binaries with a dozen binaries for all the different variants that are allowed?
On x86-64 there's a reasonable minimum supported set of instructions, and you only really need runtime CPU detection for the newer vector extensions.
- audunw 8y agoI wouldn't expect server/desktop/phone variants of RISC-V to drop any of the current extensions of the ISA. Why would you, if gate count isn't what you're optimising for? There could be further, more advanced extension that may not be in every processor, but that's just the same as x86 The modularity in the ISA as it is now, is something that's more geared towards embedded devices or FPGAs, where you'll be compiling code specifically for that target anyway.
- kabacha 8y agoEasy solution: Just GPL it - every extension of RISC-V has to be open as well.
- IshKebab 8y agoI don't see how that is a solution. The problem is the variability in what the hardware implements, not the availability of extension specifications.