Khronos has recently announced the conformance program for OpenGL 4.6 and I am very happy to say that Intel has submitted successful conformance applications for various of its GPU models for the Mesa Linux driver. For specifics on the conformant hardware you can check the list of conformant OpenGL products at the Khronos webstite.
Being conformant on day one, which the Intel Mesa Vulkan driver also obtained back in the day, is a significant achievement. Besides Intel Mesa, only NVIDIA managed to do this, which I think speaks of the amount of work and effort that one needs to put to achieve it. The days where Linux implementations lagged behind are long gone, we should all celebrate this and acknowledge the important efforts that companies like Intel have put into making this a reality.
Over the last 8-9 months or so, I have been working together with some of my Igalian friends to keep the Intel drivers (for both OpenGL and Vulkan) conformant, so I am very proud that we have reached this milestone. Kudos to all my work mates who have worked with me on this, to our friends at Intel, who have been providing reviews for our patches, feedback and additional driver fixes, and to many other members in the Mesa community who have contributed to make this possible in one way or another.
Of course, OpenGL 4.6 conformance requires that we have an implementation of GL_ARB_gl_spirv, which allows OpenGL applications to consume SPIR-V shaders. If you have been following Igalia’s work, you have already seen some of my colleagues sending patches for this over the last months, but the feature is not completely upstreamed yet. We are working hard on this, but the scope of the implementation that we want for upstream is rather ambitious, since it involves to (finally) have a full shader linker in NIR. Getting that to be as complete as the current GLSL linker and in a shape that is good enough for review and upstreaming is going to take some time, but it is surely a worthwhile effort that will pay off in the future, so please look forward to it and be patient with us as we upstream more of it in the coming months.
It is also important to remark that OpenGL 4.6 conformance doesn’t just validate new features in OpenGL 4.6, it is a full conformance program for OpenGL drivers that includes OpenGL 4.6 functionality, and as such, it is a super set of the OpenGL 4.5 conformance. The OpenGL 4.6 CTS does, in fact, incorporate a whole lot of bugfixes and expanded coverage for OpenGL features that were already present in OpenGL 4.5 and prior.
What is the conformance process and why is it important?
It is a well known issue with standards that different implementations are not always consistent. This can happen for a number of reasons. For example, implementations have bugs which can make something work on one platform but not on another (which will then require applications to implement work arounds). Another reason for this is that some times implementators just have different interpretations of the standard.
The Khronos conformance program is intended to ensure that products that implement Khronos standards (such as OpenGL or Vulkan drivers) do what they are supposed to do and they do it consistently across implementations from the same or different vendors. This is achieved by producing an extensive test suite, the Conformance Test Suite (or CTS for short), which aims to verify that the semantics of the standard are properly implemented by as many vendors as possible.
Why is CTS different to other test suites available?
One reason is that CTS development includes Khronos members who are involved in the definition of the API specifications. This means there are people well versed in the spec language who can review and provide feedback to test developers to ensure that the tests are good.
Another reason is that before new tests go in, it is required that there are at least a number of implementation (from different vendors) that pass them. This means that various vendors have implemented the related specifications and these different implementations agree on the expected result, which is usually a good sign that the tests and the implementations are good (although this is not always enough!).
How does CTS and the Khronos conformance process help API implementators and users?
First, it makes it so that existing and new functionality covered in the API specifications is tested before granting the conformance status. This means that implementations have to run all these tests and pass them, producing the same results as other implementations, so as far as the test coverage goes, the implementations are correct and consistent, which is the whole point of this process: it wont’ matter if you’re running your application on Intel, NVIDIA, AMD or a different GPU vendor, if your application is correct, it should run the
same no matter the driver you are running on.
Now, this doesn’t mean that your application will run smoothly on all conformant platforms out of the box. Application developers still need to be aware that certain aspects or features in the specifications are optional, or that different hardware implementations may have different limits for certain things. Writing software that can run on multiple platforms is always a challenge and some of that will always need to be addressed on the application side, but at least the conformance process attempts to ensure that for applications that do their part of the work, things will work as intended.
There is another interesting implication of conformance that has to do with correct API specification. Designing APIs that can work across hardware from different vendors is a challenging process. With the CTS, Khronos has an opportunity to validate the specifications against actual implementations. In other words, the CTS allows Khronos to verify that vendors can implement the specifications as intended and revisit the specification if they can’t before releasing them. This ensures that API specifications are reasonable and a good match for existing hardware implementations.
Another benefit of CTS is that vendors working on any API implementation will always need some level of testing to verify their code during development. Without CTS, they would have to write their own tests (which would be biased towards their own interpretations of the spec anyway), but with CTS, they can leave that to Khronos and focus on the implementation instead, cutting down development times and sharing testing code with other vendors.
What about Piglit or other testing frameworks?
CTS doesn’t make Piglit obsolete or redundant at all. While CTS coverage will improve over the years it is nearly impossible to have 100% coverage, so having other testing frameworks around that can provide extra coverage is always good.
My experience working on the Mesa drivers is that it is surprisingly easy to break stuff, specially on OpenGL which has a lot of legacy stuff in it. I have seen way too many instances of patches that seemed correct and in fact fixed actual problems only to later see Piglit, CTS and/or dEQP report regressions on existing tests. The (not so surprising) thing is that many times, the regressions were reported on just one of these testing frameworks, which means they all provide some level of coverage that is not included in the others.
It is for this reason that the continuous integration system for Mesa provided by Intel runs all of these testing frameworks (and then some others). You just never get enough testing. And even then, some regressions slip into releases despite all the testing!
Also, for the case of Piglit in particular, I have to say that it is very easy to write new tests, specially shader runner tests, which is always a bonus. Writing tests for CTS or dEQP, for example, requires more work in general.
So what have I been doing exactly?
For the last 9 months or so, I have been working on ensuring that the Intel Mesa drivers for both Vulkan and OpenGL are conformant. If you have followed any of my work in Mesa over the last year or so, you have probably already guessed this, since most of the patches I have been sending to Mesa reference the conformance tests they fix.
To be more thorough, my work included:
- Reviewing and testing patches submitted for inclusion in CTS that either fixed test bugs, extended coverage for existing features or added new tests for new API specifications. CTS is a fairly active project with numerous changes submitted for review pretty much every day, for OpenGL, OpenGL ES and Vulkan, so staying on top of things requires a significant dedication.
-
Ensuring that the Intel Mesa drivers passed all the CTS tests for both Vulkan and OpenGL. This requires to run the conformance tests, identifying test failures, identifying the cause for the failures and providing proper fixes. The fixes would go to CTS when the cause for the issue was a bogus test, to the driver, when it was a bug in our implementation or the fact that the driver was simply missing some functionality, or they could even go to the OpenGL or Vulkan specs, when the source of the problem was incomplete, ambiguous or incorrect spec language that was used to drive the test development. I have found instances of all these situations.
Where can I get the CTS code?
Good news, it is open source and available at GitHub.
This is a very important and welcomed change by Khronos. When I started helping Intel with OpenGL conformance, specifically for OpenGL 4.5, the CTS code was only available to specific Khronos members. Since then, Khronos has done a significant effort in working towards having an open source testing framework where anyone can contribute, so kudos to Khronos for doing this!
Going open source not only leverages larger collaboration and further development of the CTS. It also puts in the hands of API users a handful of small test samples that people can use to learn how some of the new Vulkan and OpenGL APIs released to the public are to be used, which is always nice.
What is next?
As I said above, CTS development is always ongoing, there is always testing coverage to expand for existing features, bugfixes to provide for existing tests, more tests that need to be adapted or changed to match corrections in in the spec language, new extensions and APIs that need testing coverage, etc.
And on the driver side, there are always new features to implement that come with their potential bugs that need to be fixed, occasional regressions that need to be addressed promptly, new bugs uncovered by new tests that need fixes, etc
So the fun never really stops 🙂
Final words
In this post, besides bringing the good news to everyone, I hope that I have made a good case for why the Khronos CTS is important for the industry and why we should care about it. I hope that I also managed to give a sense for the enormous amount of work that goes into making all of this possible, both on the side of Khronos and the side of the driver developer teams. I think all this effort means better drivers for everyone and I hope that we all, as users, come to appreciate it for that.
Finally, big thanks to Intel for sponsoring our work on Mesa and CTS, and also to Igalia, for having me work on this wonderful project.
OpenGL® and the oval logo are trademarks or registered trademarks of Silicon Graphics, Inc. in the United States and/or other countries worldwide. Additional license details are available on the SGI website.
Nice post! Indeed, a lot of work was done to make driver compliant, both in the driver itself and/or the testsuite itself.
Glad to be humbly helping on this.