Mesa Considers Raising CPU Support Baseline

[Zan Lynx]

Oddly enough I remember when Linux was for enthusiasts with weird, wonderful and "ancient" hardware.

Yes, and those enthusiasts did not think it was too difficult to move the pin jumpers around on their Ethernet card and edit and recompile the driver.

Now these "enthusiasts" are upset they have to change default settings in a Makefile somewhere?

[unis_torvalds]

A lot of comments in this thread are variations on the sentiment "who in their right mind would still be running 10+ year old hardware in the first place?"

Obviously nobody in a G7 country. But consider that there are computer users in poor countries where hardware typically enjoys MUCH longer useful lifespans (same with clothing and cars, as compared to rich countries).
Furthermore, there are some countries subject to trade embargoes (like Venezuela and Iran) where even if you have the money, you simply can't order the latest GTX or Epyc because they won't ship to you.

[TemplarGR]

A lot of comments in this thread are variations on the sentiment "who in their right mind would still be running 10+ year old hardware in the first place?"

Obviously nobody in a G7 country. But consider that there are computer users in poor countries where hardware typically enjoys MUCH longer useful lifespans (same with clothing and cars, as compared to rich countries).
Furthermore, there are some countries subject to trade embargoes (like Venezuela and Iran) where even if you have the money, you simply can't order the latest GTX or Epyc because they won't ship to you.

Yes, and there are many linux distros that can accomondate these people. You people don't seem to understand: Legacy code does not DISSAPPEAR FROM THE INTERNETS! It is still there, you can use it. If you have a legacy machine, use code that can support it. There is no reason to force people who use hardware made during the last decade to be held back by people who use obsolete hardware.

[Teggs]

Would it make sense to match the current (non-SSE2) compiler target to the proposed Mesa-LTS? The idea being that ancient CPUs are probably matched to ancient GPUs, so Mesa-LTS would still serve such hardware, while Mesa-current could reasonably assume SSE2. Or is there too much mismatched hardware in use for that idea to pan out?

[reba]

x86 (_64) is an architecture that has a long history; over 20 years and more. It spans across many technological advancements and conceptual changes. Of cource something with such a legacy cannot be a "one size fits all".
There are old-ish machines and new-ish machines and they all are thrown into the same bin for x86. For other architectures, that are either older and now not anymore actively used or newer and still on the rise, it's easier but the x86-heritage bring its own burden.
Regardless where you cut, it's both too early and too late.
I think the v1-v4 levels are a good idea and deperately needed to cut x86 into smaller pieces. Older hardware uses v1-v2, newer v2-v4. No harm done. But it really can't come soon enough.
And it would be great if the different distributions and Unixoids could come up with a consistent plan for this. The multitude of different distributions and package managers are sadly already fragmenting the ecosystem enough.

[unis_torvalds]

Yes, and there are many linux distros that can accomondate these people. You people don't seem to understand: Legacy code does not DISSAPPEAR FROM THE INTERNETS! It is still there, you can use it. If you have a legacy machine, use code that can support it. There is no reason to force people who use hardware made during the last decade to be held back by people who use obsolete hardware.

Just curious: what do you mean by "you people"?

[flashmozzg]

Why would people with 20+ years old CPU would want a new MESA driver? Their CPU is likely matched with similarly outdated GPU which is unlikely to receive any improvements from a new driver.

[microcode]

What hardware from before 2003 is even currently supported by Mesa?

[Raka555]

If people played around with the compiler optimizations like -march=native , they will know that it does not make a spectacular difference.
In reality newer machines were not held back that much.

Big gains can be made by rewriting certain functions in inline assembly using newer instructions.
That can still be done in a way where a generic version is available for architectures which don't have the said instructions.
gcc had support for FMV since about version 4.8
Compilers and linkers are smart enough to do dead code elimination to remove the bloat.

In essence the best way to gain from newer instructions is to use a distro like Gentoo that builds everything from source.

[mangeek]

Simply stated, there are lots of industrial systems... meant to last decades or longer and use some sort of display for interaction.

Those systems aren't ever going to run Mesa 21.2. They're going to run old versions of LTS distros, and likely never be upgraded. When they can't be patched, the vendor will recommend sequestering them off with network rules and host firewalls.

I deal with a LOT of these types of systems, and I don't think I can recall any of them making a change like upgrading from Ubuntu 16.04 to 18.04 for already-deployed hardware.