Marek Continues Baking More Mesa Optimizations

[smitty3268]

What's stopping you?

Lack of interest, presumably.

[Awesomeness]

I believe it is more important to go the way of optimizing existing things, than adding something new just for the sake of it.

By that logic we'd optimize the heck out of OpenGL 1.0. No, no, adding new OpenGL extensions for the sake of having new OpenGL extensions is totally valid for a project like Mesa.

[eydee]

By that logic we'd optimize the heck out of OpenGL 1.0. No, no, adding new OpenGL extensions for the sake of having new OpenGL extensions is totally valid for a project like Mesa.

It never hurts, really. Just consider the official AMD windows OpenGL driver, that can't launch basic stuff like Neverwinter Nights. Embarrassing. Good thing Linux and Mesa aren't affected.

[dungeon]

It never hurts, really. Just consider the official AMD windows OpenGL driver, that can't launch basic stuff like Neverwinter Nights. Embarrassing. Good thing Linux and Mesa aren't affected.

I would not call Neverwinter Nights basic, as that trying to use EXT_vertex_shader, NV_vertex_program but not ARB_vertex_program. So it is kind of special or out of spec shit

[pewspewpew]

By that logic we'd optimize the heck out of OpenGL 1.0. No, no, adding new OpenGL extensions for the sake of having new OpenGL extensions is totally valid for a project like Mesa.

In moderation, of course. I think there is not much good in adding something new to unfinished code and jumping to next interesting to write code. It may come a bit later, but come out splendid as a system, or it may come together and maybe shine in that part, while others get ignored, which would you choose? I'd vote for quality over quantity.

[starcrossed]

I believe it is more important to go the way of optimizing existing things

New hardware optimizes existing things for us.

I know that sounds harsh but imagine if you could choose how 10,000 hours was spent a decade ago: optimizing 3dfx drivers + OpenGL 1.2 + GLQuake vs implementing OpenGL 2.0 and shaders? One still benefits us today, one does not. Optimizations are important but they frequently depreciate in value over time. A 30% speed increase today in GLQuake is irrelevant when it's already running at 1000 fps. Standards compliant GL4, Vulkan and good foundational tools for the ages (like LLVM and NIR) are investments that will be with us for decades.

I'm as happy as the next guy that optimizations are being done, I just prefer time being spent on projects that will still be relevant when the lowest end card in use is Volta/Vega era and all new software is Vulkan/DX12+.

[pewspewpew]

New hardware optimizes existing things for us.

I know that sounds harsh but imagine if you could choose how 10,000 hours was spent a decade ago: optimizing 3dfx drivers + OpenGL 1.2 + GLQuake vs implementing OpenGL 2.0 and shaders? One still benefits us today, one does not. Optimizations are important but they frequently depreciate in value over time. A 30% speed increase today in GLQuake is irrelevant when it's already running at 1000 fps. Standards compliant GL4, Vulkan and good foundational tools for the ages (like LLVM and NIR) are investments that will be with us for decades.

I'm as happy as the next guy that optimizations are being done, I just prefer time being spent on projects that will still be relevant when the lowest end card in use is Volta/Vega era and all new software is Vulkan/DX12+.

Well, I am noone to direct anyone in development, as I do not participate myself. As a consumer though, i see nvidia winning now and considering time needed for market to switch to vulkan i see it only increasing it's lead unless amd can catch up now. So for mesa itself this strategy is correct, but for amd I imagine it means nvidia totally and finally owning linux (especially when they decide to be more os). I mean, I love the idea of os drivers and open minded companies, but I would lie if i said I do not regret not choosing nvidia as current device, that bad the current performance is for me. And thus in my eyes it is not optimizing - it's fixing subpar functionality. Just imo.

[dungeon]

It never hurts, really. Just consider the official AMD windows OpenGL driver, that can't launch basic stuff like Neverwinter Nights. Embarrassing. Good thing Linux and Mesa aren't affected.

I actually took a time to investigate this...so it can't launch it cos of GL_ATI_fragment_shader extension... neither Catalyst or Radeon Software on WIndows, neither Catalyst Linux, but amdgpu-pro can just in slow and buggy mode ... They should just disable extension via profile (on nwmain.exe and nwmain), as even if you rename binary to make it start some effects like "Environment Mapping on Creatures" does not work, etc... disabling fixes everything.

Newer mesa drivers aren't affected as they don't implement extension... i actually think only r200 driver expose that

[dungeon]

BTW in addition to dasabling extension culprit, enabling MCCaps make it flyby too