New AMD Zen 3 Fixes Published For The GCC 11 Compiler

[smitty3268]

Well, as you say: "AMD has had..." - the GCC or Clang guys didn't. How difficult is it to understand this chicken-and-egg problem?

Huh? Wasn't the whole topic about AMD providing patches to GCC/Clang ahead of time? Like Intel does?

Alternatively, if they don't think they can provide the patches themselves, they could provide an existing GCC/Clang developer an engineering sample under NDA just like they do others that need to support their processors ahead of time. Even just docs like instruction timings probably would have sufficed given that it's a fairly typical x86-64 architecture.

This isn't anything especially tricky, they do that all the time.

[lowflyer]

This isn't anything especially tricky, they do that all the time.

If they "do that all the time" - then why all the whining? (not you personally. I mainly address it to Michael and the other AMD lambasters)

As a software developer working with new hardware, I know how difficult it is to get the drivers right on the first shot. GCC and Clang are the second last dogs in a long chain of dogs that all get to eat first. That's the simple reason why the compiler support is always late.

Is it so difficult to just enjoy the new metal and the thrill of anticipation for the boosts that will come along in the future? Isn't "deferral of gratification" the hallmark of intelligence? And please appreciate AMD's efforts they put into AOCC.

[smitty3268]

If they "do that all the time" - then why all the whining? (not you personally. I mainly address it to Michael and the other AMD lambasters)

As a software developer working with new hardware, I know how difficult it is to get the drivers right on the first shot. GCC and Clang are the second last dogs in a long chain of dogs that all get to eat first. That's the simple reason why the compiler support is always late.

Is it so difficult to just enjoy the new metal and the thrill of anticipation for the boosts that will come along in the future? Isn't "deferral of gratification" the hallmark of intelligence? And please appreciate AMD's efforts they put into AOCC.

Personally I don't really care about it. Issues like the kernel not working are major showstoppers. Some little compiler tweaks for performance improvements are fairly unimportant in the grand scheme of things.

I just think the fact that they aren't bothered to go to the minimal level of effort it would require reflects poorly on AMD's professionalism, or else more likely it's indicative of an extremely low level of interest in providing proper linux support - something that builds on top of other issues AMD has had in recent memory.

I doubt that's the kind of impression AMD wants people to have of them. If I'm wrong about that then I stand corrected.

[lowflyer]

I just think the fact that they aren't bothered to go to the minimal level of effort

I think this is a stark misrepresentation of AMD

[Guest]

I think there's a correlation between how much we'll take as consumers and how well AMD is doing. The better they're doing, the less of this kind of bullshit are we willing to take. 7, 8 years ago I didn't mind letting them slide because they were the underdog. Now they have the best CPU architecture in the world, maybe the best GPU architecture (but we wouldn't know because cain't no one buy one), so they're not really the underdog and we're expecting them to put their money where their programmers are.

Since home users and their issues translates over to how the company committee is going to spend, I imagine that AMD wants us to stop complaining about their release timings so hopefully they'll get better in the upcoming year...I'm hoping that's what one or two of the Linux jobs AMD was hiring for go toward.

They had better CPU support 7-8 years ago when they were the underdog. They decided to let go of their whole open source team as a "cost cutting measure". Most idiotic move ever.

[skeevy420]

They had better CPU support 7-8 years ago when they were the underdog. They decided to let go of their whole open source team as a "cost cutting measure". Most idiotic move ever.

I was still an Intel CPU user and didn't keep up with AMD CPUs back then, but I vaguely remember something like that. Ah, but the end of my Core2Quad days that was.

That was when I made the switch to AMD GPUs. Got tired of Nvidia driver issues and wanted something as simple to set up as the Intel iGPU. Though I did get to experience the ass-end of the Catalyst days which was even worse than Nvidia. Catalyst + Arch was one of my worst Linux experiences ever.

The only AMD GPU problem I have these days is that the Windows version of DRI_PRIME doesn't work with my setup. It will not compute games on my RX 580 and output them with my Vega iGPU. It always sends shit to the iGPU. No amount of Graphics Settings or High Performance Power Plans or force use the 580 make a difference. I think my problem is Windows wants to use the newer GPU, but the newer Vega iGPU is worse than the older Polaris dGPU and the Windows override settings don't override that.

On Linux it works just fine; just set some env variables and I'm good to go. I played some almost maxed out THPS 1+2 in 4K like that the other night. Had around 3/4s of native Windows performance running on the 580 directly. Playable, but not as good as on the 580 directly.


EDIT: Wow. Really didn't mean to go off on a rant there. I suppose it isn't hard to figure out what I didn't succeed at last weekend

[ms178]

The only AMD GPU problem I have these days is that the Windows version of DRI_PRIME doesn't work with my setup. It will not compute games on my RX 580 and output them with my Vega iGPU. It always sends shit to the iGPU.

I faced the same issue on my old Sandybridge + 6770M laptop, what a diva that was when it came to drivers. If you do not need the iGPU you could try to disable it in the BIOS to save you the headaches.

[skeevy420]

I faced the same issue on my old Sandybridge + 6770M laptop, what a diva that was when it came to drivers. If you do not need the iGPU you could try to disable it in the BIOS to save you the headaches.

That's what I do with two UEFI profiles -- iGPU on and off which also doubles as Linux and Windows. My endgame is to setup a Windows VM on Linux running off the dGPU with the iGPU running Linux. I gotta do some rearranging and make a few purchases before I do that.