RPCS3 and Dolphin on macOS using gfx-portability(gfx-rs.github.io)
gfx-rs.github.io
RPCS3 and Dolphin on macOS using gfx-portability
https://gfx-rs.github.io/2018/09/03/rpcs3-dolphin.html
34 comments
This is exactly what Casey Muratori was ranting about in his "Thirty Million Line Problem" video. My interpretation (or extension) of the point he was making: if the same hardware (CPU + GPU) can boot Mac OSX or Windows, why can't the same game binary run on both? If hardware vendors created a good "ISA" and stuck to it, it should be possible to write one binary that talks to the hardware directly and runs the same anywhere as long as the hardware ISA is compatible.
> projects like Metal on Vulkan (MoltanVK)
It's not Metal on Vulkan, it's the other way around. Also, kind of strange to bring this up when the topic is about gfx-portability progress (which is competing with MoltenVK).
It's not Metal on Vulkan, it's the other way around. Also, kind of strange to bring this up when the topic is about gfx-portability progress (which is competing with MoltenVK).
That is correct, my morning brain mixed it up because of writing of MoltanVK started me off on Metal on Vulkan.
I bought it up because of the general topic of Vulkan and Metal, not because of gfx-rs itself. MoltanVK deserves a mention due to Khronos Group sponsoring it.
I bought it up because of the general topic of Vulkan and Metal, not because of gfx-rs itself. MoltanVK deserves a mention due to Khronos Group sponsoring it.
This is sadly a common misconception, partly due to Khronos messaging. MoltenVK is sponsored by Valve, not Khronos.
Both gfx-rs and MoltenVK are working in Vulkan Portability TSG, and both are promoted by the slides/talks if you read carefully. MoltenVK just gets more attention because it's packed by LunarG and used in Dota2.
Both gfx-rs and MoltenVK are working in Vulkan Portability TSG, and both are promoted by the slides/talks if you read carefully. MoltenVK just gets more attention because it's packed by LunarG and used in Dota2.
I saw MoltenVK under KhronosGroup control here, which led to my confusion once more: https://github.com/KhronosGroup/MoltenVK
And yea, the messaging is confusing because of this line on their site:
> As a first deliverable from the Vulkan Portability Initiative, Khronos members Valve, LunarG, and The Brenwill Workshop have released a collection of free and open source set of tools, SDKs, and runtime libraries to enable Vulkan development on macOS and deployment on macOS and iOS platforms
Source: https://www.khronos.org/vulkan/portability-initiative
That led me to think they’re sponsoring it.
It looks like MoltenVK was created by The Brenwill Workshop.
And yea, the messaging is confusing because of this line on their site:
> As a first deliverable from the Vulkan Portability Initiative, Khronos members Valve, LunarG, and The Brenwill Workshop have released a collection of free and open source set of tools, SDKs, and runtime libraries to enable Vulkan development on macOS and deployment on macOS and iOS platforms
Source: https://www.khronos.org/vulkan/portability-initiative
That led me to think they’re sponsoring it.
It looks like MoltenVK was created by The Brenwill Workshop.
KhronosGroup github organization may be hosting projects of Khronos members that align with Khronos mission, but they aren't necessary the deliverables of any working groups and thus can't be official/sponsored.
> Khronos members Valve, LunarG, and The Brenwill Workshop have released
It's like saying "W3C members Facebook, Instagram, WhatsApp have released ..." :D
Basically, Valve released something that they want to look like it's supported by Khronos. And this trick worked - now everybody believes it to be a Khronos product.
> Khronos members Valve, LunarG, and The Brenwill Workshop have released
It's like saying "W3C members Facebook, Instagram, WhatsApp have released ..." :D
Basically, Valve released something that they want to look like it's supported by Khronos. And this trick worked - now everybody believes it to be a Khronos product.
Cool, thanks for the details.
i still think osx killing opengl support is ridiculous.
Swift and Objective-C API with C++14 shaders vs plain old C API and manually compilation of shaders?
Obvious choice, plus all engines that matter already support Metal.
Same situation on PS4, XBox and UWP.
Obvious choice, plus all engines that matter already support Metal.
Same situation on PS4, XBox and UWP.
There is nothing wrong with saying something new could be better but is Metal actually so much better than Vulkan to be the only supported API?
Metal isn't new, it's older than Vulkan, and by some metrics I would consider it better.
For one, the learning curve is much smoother: you can get something up on screen quickly, but only later discover the advanced features like argument buffers, manual hazard tracking, etc.
For one, the learning curve is much smoother: you can get something up on screen quickly, but only later discover the advanced features like argument buffers, manual hazard tracking, etc.
We are talking 4 years old (3 for desktop) years old vs 2 years old, you're kidding yourself if you think either is established by virtue of being old vs new. That comparison was only fair with OpenGl.
Sure! I'm just pointing out that your earlier statement was off the mark, and it posed Metal as "something new that could be better than Vulkan".
The fact that Vulkan is a C API and requires searching for extensions even for a plain 3D Hello World, makes Metal a better option on the Mac.
Erm, not really. I have implemented a reasonably serious PBR renderer in Vulkan on Mac on top of MoltenVK, and the C API is not at all an obstacle. Swift/C interop is fairly straightforward, and Objective C is a superset of C so that is a total non-issue.
The ability to easily integrate the wealth of C libraries directly into your high-level code (as opposed to, say, the verbosity of implementing a JNI) is one of the strengths of Swift.
The ability to easily integrate the wealth of C libraries directly into your high-level code (as opposed to, say, the verbosity of implementing a JNI) is one of the strengths of Swift.
C is an obstacle to safe and modern IT stacks.
Why are you using graphics engines in your IT stack?
I guess because I am not using a VT 100 terminal, rather a desktop computer using a GPU to provide a nice modern user experience.
The biggest theme on Linux Security Summit 2018 was taming security exploits on Linux kernel due to C's lack of security.
I count at least 10 talks about this subject.
https://www.youtube.com/user/TheLinuxFoundation/videos
So apparently the review process, approvals and current static validation is still not enough.
The biggest theme on Linux Security Summit 2018 was taming security exploits on Linux kernel due to C's lack of security.
I count at least 10 talks about this subject.
https://www.youtube.com/user/TheLinuxFoundation/videos
So apparently the review process, approvals and current static validation is still not enough.
> C++14 shaders
What's a C++14 shader? Metal uses MSL last time I checked...
> manually compilation of shaders
The toolchain for this in Vulkan is pretty solid. It's actually nice to be able to write shaders in a high-level language, and then compile them down to SPIR-V which is more likely to be interpreted consistently across drivers than a high-level language.
What's a C++14 shader? Metal uses MSL last time I checked...
> manually compilation of shaders
The toolchain for this in Vulkan is pretty solid. It's actually nice to be able to write shaders in a high-level language, and then compile them down to SPIR-V which is more likely to be interpreted consistently across drivers than a high-level language.
So apparently you missed the fact that MSL is based on C++14.
https://developer.apple.com/documentation/metal/hello_triang...
https://developer.apple.com/documentation/metal/hello_triang...
It is. Also not supporting Vulkan natively is a mistake.
Yes but there's nothing surprising about the Walled Garden Company promoting their own snowflake version of Vulkan and insisting everyone who wants to run software must do things their way. This is the same company who has totally locked down the tooling to create and publish iOS apps to only be possible on their hardware.
To be fair, Metal is more than a "snowflake version of Vulkan". For one thing, it was in production before vulkan, largely to address the issue of OpenGL's overhead on mobile devices, and basically it achieves that goal. Also there are certain things which are exposed in Metal which are not in Vulkan, for instance more control over the tile memory which mobile GPUs use. So it's not as if there is no reason for Metal to exist (of course I would be happier if Apple contributed to Khronos to get those features in Vulkan instead).
But the walled garden approach seems counterproductive here. In the days of Microsoft's dominance Apple made an effort to make windows formats usable on Apple because that meant it would be easier for consumers to switch to Apple without worrying about losing all their stuff from Windows. With regard to applications in which graphics API's are relevant (i.e. games) Apple still has a minuscule market-share, and it seems like they would be well served by adopting the dominant technology. Lock-in only works if you're already winning.
But the walled garden approach seems counterproductive here. In the days of Microsoft's dominance Apple made an effort to make windows formats usable on Apple because that meant it would be easier for consumers to switch to Apple without worrying about losing all their stuff from Windows. With regard to applications in which graphics API's are relevant (i.e. games) Apple still has a minuscule market-share, and it seems like they would be well served by adopting the dominant technology. Lock-in only works if you're already winning.
Some devs are so obsessed complaining about Metal that most aren't aware that UWP and Windows Store are DirectX only.
An advantage of all proprietary 3D APIs is the amount of out-of-box infrastructure code, debugging tools and having progressed beyond C, while Khronos APIs are still mostly C, and requiring creating mini-engine from scratch after fishing for libs.
An advantage of all proprietary 3D APIs is the amount of out-of-box infrastructure code, debugging tools and having progressed beyond C, while Khronos APIs are still mostly C, and requiring creating mini-engine from scratch after fishing for libs.
> having progressed beyond C
I don't really see the issue with C APIs. IMO C is a fantastic integration point for a library which wants to have the widest reach possible. Interoperability with C is a solved problem in a wide variety of languages, so C APIs tend to be easy to integrate. Also since C is a fairly thin abstraction on top of what a computer actually does, good C APIs tend not to be opinionated about how they should be integrated.
I would consider working with C APIs to be a basic programming skill.
I don't really see the issue with C APIs. IMO C is a fantastic integration point for a library which wants to have the widest reach possible. Interoperability with C is a solved problem in a wide variety of languages, so C APIs tend to be easy to integrate. Also since C is a fairly thin abstraction on top of what a computer actually does, good C APIs tend not to be opinionated about how they should be integrated.
I would consider working with C APIs to be a basic programming skill.
I see a big one.
https://www.hpe.com/us/en/insights/articles/making-c-less-da...
Plus it is about time to move from 70s style APIs.
https://www.hpe.com/us/en/insights/articles/making-c-less-da...
Plus it is about time to move from 70s style APIs.
> With regard to applications in which graphics API's are relevant (i.e. games) Apple still has a minuscule market-share, and it seems like they would be well served by adopting the dominant technology. Lock-in only works if you're already winning.
Except Metal is the dominant API on almost all Apple platforms. Metal supported-iOS devices are at 700 million (last year's WWDC stat) and majority of games are using Metal on iOS. You can't ignore that market, iOS games are very profitable.
Fortnite made $3M in sales in the first three days on iOS: https://www.tweaktown.com/news/61266/fortnite-mobile-makes-1...
Except Metal is the dominant API on almost all Apple platforms. Metal supported-iOS devices are at 700 million (last year's WWDC stat) and majority of games are using Metal on iOS. You can't ignore that market, iOS games are very profitable.
Fortnite made $3M in sales in the first three days on iOS: https://www.tweaktown.com/news/61266/fortnite-mobile-makes-1...
Well as you say, Metal is really for iOS. And I think the evidence points to Apple caring much more about iOS than the Mac. So it makes sense from that perspective.
And not only do you need a Mac to write iOS apps, you need to pay them $100/yr to run the apps on anything other than your own devices.
And even on your own devices, you need to reinstall every few (days? weeks?) because the certificate expired.
And even on your own devices, you need to reinstall every few (days? weeks?) because the certificate expired.
That is peanuts compared with games consoles and you don't see game studios crying, on the contrary, they prefer that model to desktop computers.
This is super cool!
Awesome!
Imagine if Apple did support Vulkan natively, all Windows games via Steam Play Beta can be run on macOS easily and with support of eGPU, it would help a lot to make them run better.
All I can realistically hope for is that Apple continues to close the gap between Metal and Vulkan, so projects like Metal on Vulkan (MoltanVK), Proton (Steam Play + dxvk + Wine) can benefit with less work to do.
It'll be interesting to see what Apple does with Metal 3 in the next few years. If they're smart (they haven't been in the last few years with Mac Pro, rMBP keyboard+touchbar, etc), they'd work with the Kronos group.