On Wednesday, March 22, 2023, Linux User #330250 <linuxuser330...@gmx.net> wrote:
> On the other hand, I don't really see performant and cheap (for desktop > systems) systems on the market that aren't mainstream x86 (Intel and > AMD). Letting those now older alternatives go (that were mostly servers > to begin with) is probably a reasonable decision. But also sad... thus we (RED, Libre-SOC) are "on the clock" with a limited window of opportunity to stop that from happening. there are additional complications: IBM signed a non-compete agreement over 12 years ago with Motorola to stay out of the desktop market and then the *entire Motorola team* was killed when that Malay Air passenger plane was shot down. an additional complication is that VSX is so insane (950 instructions) that to even attempt adding it to a very first SoC by a team that has never done HDL before would be, frankly, flat-out stupid. thus we made the very easy decision to cut it, and i would dearly love VSX, more specifically its PackedSIMD instructions, to be buried and a small tactical nuke to be dropped on them. https://www.sigarch.org/simd-instructions-considered-harmful/ that leaves us with "only" 214 instructions as part of the SFFS Compliancy Level to implement https://en.m.wikipedia.org/wiki/Power_ISA#Compliancy now, as part of the effort to fix the issues caused by developers (even those in IBM!) assuming that IBM's products are the world's only relevant Power ISA implementations, we've begun an initiative funded by NLnet to create a debian and gentoo SFFS port. very interestingly it will also run (with some v2.06->3.0 instruction trap emulation) on the E5500 which is a very unusual 64-bit Power ISA CPU from Motorola (Quorl family) https://bugs.libre-soc.org/show_bug.cgi?id=999 SFFS will also be ideally suited to Microwatt, A2O, A2I, as well as Libre-SOC, all of which run in FPGAs and there are qemu options to drop down to SFFS (killing altivec and vsx). all of this needs active interest, active help, and people to take responsibility for solving these issues instead of leaving it to large Corporates whom we know, from long and bitter experience, are just never going to get it: they always always take the easiest route to market. License PowerVR. license an HDMI solution from your competitor (Intel) and forget to obtain a license for the firmware source code such that all your products (AMD) since 2013 require a proprietary blob just to start the HDMI interface. easy, right? if there's interest in helping out we do have NLnet and other NGI funding. l. -- --- crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68