Slashdot Mirror
OpenGL 4.4 and OpenCL 2.0 Specs Released
Via Ars comes news that the OpenGL 4.4 and OpenCL 2.0 were released yesterday. OpenGL 4.4 features a few new extensions, perhaps most importantly a few to ease porting applications from Direct3D. New bindless shaders have access to the entire virtual address space of the card, and new sparse textures allow streaming tiles of textures too large for the graphics card memory. Finally, the ARB has announced the first set of conformance tests since OpenGL 2.0, so going forward anything calling itself OpenGL must pass certification. The OpenCL 2.0 spec is still provisional, but now features a memory model that is a subset of C11, allowing sharing of complex data between the host and GPU and avoiding the overhead of copying data to and from the GPU (which can often make using OpenCL a losing proposition). There is also a new spec for an intermediate language: "'SPIR' stands for Standard Portable Intermediate Representation and is a portable non-source representation for OpenCL 1.2 device programs. It enables application developers to avoid shipping kernel source and to manage the proliferation of devices and drivers from multiple vendors. OpenCL SPIR will enable consumption of code from third party compiler front-ends for alternative languages, such as C++, and is based on LLVM 3.2. Khronos has contributed open source patches for Clang 3.2 to enable SPIR code generation." For full details see Khronos's OpenGL 4.4 announcement, and their OpenCL 2.0 announcement.
Update: 07/23 20:17 GMT by U L : edxwelch notes that Anandtech published notes and slides from the SIGGRAPH announcement.
There's a better article here:
http://www.anandtech.com/show/7161/khronos-siggraph-2013-opengl-44-opencl-20-opencl-12-spir-announced
If you can't understand C, you have no business touching the GPU or even calling yourself a programmer.
Nothing wrong with C, but you don't really need to limit your self to it just because the code is running on the GPU. Have a look at C++ AMP for example.
Of course I understand C. I just DONT_WANT_TO_DEAL_WITH_INT_CONSTANTS, when modern type safe enums exist, and other pointless annoyances that are conventional in C.
>When i started programming 20 years ago i was told the following:
If you can't understand Assembly, you have no business touching the CPU or even calling yourself a programmer.
Just Because something is hard doesn't mean it's good.
10 print "Basic For The Win!!!"
20 End
Nothing wrong with C, but you don't really need to limit your self to it just because the code is running on the GPU.
The portion of the code running on the GPU in OpenCL isn't C.
It's true that it's not C, but the way it's designed make it look like C.
Why can I run a kernel for CUDA in 3 lines of code but I need 2000 lines for running an opencl one?
Why at one point you had to have your kernel in comments ffs? http://developer.apple.com/library/mac/#samplecode/OpenCL_Hello_World_Example/Listings/hello_c.html
OpenCl is retarded.
Of course I understand C. I just DONT_WANT_TO_DEAL_WITH_INT_CONSTANTS, when modern type safe enums exist, and other pointless annoyances that are conventional in C.
There is nothing magical about type safety in a programming language. If you want type safety in C compile with type checking warning turned on. Any C compiler can tell you when you are using different types.
comment/string my bad
There's nothing magical about type safety, you know, except for the confidence that the setting your assigning relates at all the value you're using.
In almost every Slashdot article for the past century I have seen this mysterious message which includes random words with "BSD" there somewhere. Who is the stubborn party behind these spams and how can they still get through?
OpenCL requires a lot of boilerplate code to initialize resources and whatever. But the code is a lot more machine portable than CUDA.
Except for the fact that CUDA only works on nvidia devices, and OpenCL works on everything...
Hamsters are at least as feathery as penguins. HamLix
OpenCL requires a lot of boilerplate code to initialize resources and whatever.
Not really. You just have to find the device(s) you want to use, then use them per their specs that they make available.
But the code is a lot more machine portable than CUDA.
Yes. It runs on AMD and even Intel, though Intel has some terrible drivers. (I'm looking at you, whoever thought adding install directory to PATH with their internal Qt version DLLs was a Good Idea!)
And who cares what is best if you can't use it?
Hamsters are at least as feathery as penguins. HamLix
Until you want to debug or run things fast.
You want portability? Use a framework.
The gpuocelot project has been able to run CUDA in non-NVIDIA hardware for some time now, including x86 CPUs and AMD GPUs.
Too bad the CUDA compiler often segfaults on ordinary C++ libraries even when they are host-only (in which case nvcc is supposed to just forward it to GCC). Hopefully the LLVM-based compiler for OpenCL 2.0 won't be as buggy.
Why can I run a kernel for CUDA in 3 lines of code but I need 2000 lines for running an opencl one?
How well do those 3 lines of code work when you are on an AMD or Intel GPU?
CUDA and the people that use it are retarded.
In my experience CUDA is not any faster than OpenCL. Frameworks don't solve the problem properly. There are a lot of debugging tools for OpenCL my guess is you did not look hard enough. You can run OpenCL programs without installing all the cruft required to do CUDA development since the driver will compile and run code by itself. This means a lot of people don't bother looking for tools but they are out there.
There are also several mobile devices (smartphones, tablets) running ARM which have OpenCL support and zero CUDA support. Not to mention that it is also a web standard namely WebCL.
Indeed. A functional language like Haskell would be so much better suited, it's not even funny. No side effects is *key* in being able to parallelize things. Because you can trust that the same input will *always* give the exact same output.
And the whole architecture is designed for a function-based model anyway. Look at shaders. They are basically huge expressions. Just like functional code.
the important question is... when we will see it delivered on Linux drivers?
You think that defining an IR is a sign of being stuck with C? The goal is clearly to open up the standard to a wide range of languages. C++ AMP is just a compiler front on top of DX IL It's a tool (which happens to implement a standard library) on top of a well defined intermediate layer.
There's nothing magical about type safety, you know, except for the confidence that the setting your assigning relates at all the value you're using.
If you want type safety in C compile with type checking warning turned on.
Constants aren't type safe.
John
That's surprisingly uncommon among standardization organizations. I wish IETF could do the same for RFCs...
If you can't understand C, you have no business ... even calling yourself a programmer.
So I guess the hundreds of other programming languages don't count?
NVidia, who own the 50% of the GPU market and have the most advanced GPU architectures ( K20 ) are still on OpenCL 1.1. They haven't released a version of OpenCL 1.2. It's a shame as a good OpenCL 2.0 or 1.2 release will grow the overall category of GPU's. If NVidia actively support OpenCL it gives the market a sign of security that GPUs are useful beyond games and graphics.
So when will NVidia implement OpenCL 2.0 ??
I can use it. I have Nvidia GPUs.
NVidia, who own the 50% of the GPU market
Not even close NVidia has 18% of the GPU market with Intel at 61.8% and AMD at 20.2%. NVidia is less prolific than you think. Basically 80% of the market can implement it without Nvidia. I don't think they want to do that.
NVidia, who own the 50% of the GPU market...
Just where did you get that data point!? Most everything I've seen retail wise has Intel graphics inside while few come with NVidia's stuff, and those that do come at a premium price.
Time is what keeps everything from happening all at once.
Dunno, and the motivation? strange try to demote BSD in search engine rankings? If they index slashdot and they mind modding...
Because CUDA does that for you?
Any specific reason why you couldn't hide all that behind a library for opencl?
A quick Wikipedia search shows e.g. boost.compute, asgard, etc.
Give him a break. He misspelled "assembly".
Understanding C typically means understanding memory layout. Most other languages use managed memory, so it does make a lot of sense.
Anyway, optimal memory access patterns for GPUs are completely different, so you'd better change your loops.
Well, let's look at the use cases for OpenCL right now:
* Scientific computing, at levels from workstations to supercomputers
* Games that need to offload stuff too parallel for the CPU to handle, or for code that needs to run on the GPU as the output will be used by other GPU code (streaming texture decompression is a common task).
* Video transcoders, encoders and decoders
* Bitcoin miners (obligatory Bitcoin reference: check!)
All of those are fields where performance is a very high priority - in some cases, above even correctness. They're also fields for experts - if you don't know how to program at essentially the assembly level, you won't make it in the field. So is it harder? Sure. But this is stuff where you can't just wave a magic wand and make it easy - it's tough because massively multi-threaded programming is intrinsically difficult.
Yes really. After finding the device you want to use, you still have to create buffers, manage kernel strings, build program objects, set kernel arguments, manage the queue, events, errors, reads, writes and so on.
It's like having to build a vehicle every time you want to go somewhere. Faster than walking, once it's built, but it is a lot of unnecessary bother to have to build something every time.
APIs like C++AMP and Bolt can reduce that boilerplate requirement, though, while building on all of OpenCL's good stuff.
So I guess the hundreds of other programming languages don't count?
Those hundreds on other programming languages are the reason we have been saddled with slow, bloated and inefficient programs for the last 30 years.
then TFA goes on to say
So give it a few months.
You can completely circumvent this problem by simply writing correct code.
Ezekiel 23:20
Don't care. CUDA is better.
Define "better".
Ezekiel 23:20
Why can I run a kernel for CUDA in 3 lines of code but I need 2000 lines for running an opencl one?
You need to use clut, just like you would use glut with OpenGL.
Ezekiel 23:20
If you can't understand C, you have no business ... even calling yourself a programmer.
So I guess the hundreds of other programming languages don't count?
It's like calling yourself an international businessman and not knowing English. Yes, those hundreds of other programming languages actually don't count in this case.
Ezekiel 23:20
No side effects is *key* in being able to parallelize things. Because you can trust that the same input will *always* give the exact same output.
Actually, that's mostly irrelevant. That could be useful for memoization, but it's not a sufficient condition for parallelization - if you take it to the logical conclusion, you're asking for nothing more than a computer that is reliable, which is an assumption you do for most computer programs, so you're asking for a very weak property. The key to parallel computing is the associativity of individual operations. Other properties that are of lesser help are commutativity, idempotency (basically the thing you've mentioned), and the existence of zeros and identities, but it's associativity that is vital. If you can do (((1+2)+3)+(4+5))+((6+7)+(8+(9+10))) instead of ((((((((1+2)+3)+4)+5)+6)+7)+8)+9)+10, you win big. If you can't, you lose.
Ezekiel 23:20
it's tough because massively multi-threaded programming is intrinsically difficult.
Only in weak programming paradigms.
Ezekiel 23:20
You can completely circumvent this problem by simply writing correct code.
Well, I'm sure super human awesome coders like yourself can. The rest of us fallible mortals enjoy having tools to automatically catch mistakes without adding any overhead or verbosity. I'll bet you "simply write correct code" so there's no need to test before pushing it to the production server, right?
SJW n. One who posts facts.
They can be in C++11: strongly typed enums.
SJW n. One who posts facts.
You *routinely* program in C things that you "push to production servers"? Exactly *how much* free time do you have on your hands?
Ezekiel 23:20
Does anyone understand why Khronos keeps releasing new versions of OpenGL? If there was adoption of the previous versions by hardware vendors and software developers that would be one thing, but from what I can see there isn't. Only Nvidia seems to support the version released last year (4.3). Apple and Intel seem to be stuck using versions from many years ago (3.x). As for software developers I don't know of any apps that use OpenGL 4.x, and very few that use OpenGL 3.x (we plan to continue using OpenGL ES 2.0 for years since that has the broadest adoption).
My understanding is in standard bodies like Khronos all the involved companies get a vote. It would be interesting to understand how they keep getting a majority of the members to approve these new specs because it just doesn't seem to make sense to me.
Exactly *how much* free time do you have on your hands?
I have literally no idea what you are talking about. You were decrying type safety because you could "simply write correct code".
If that's possible, then there is no reason to ever test anything because it will "simply be correct".
And yeah, I do test stuff before pushing it to production servers, though no, I do not use C these days.
SJW n. One who posts facts.
Sorry, what I first meant was that #defined macros in C are not at all typesafe (which is the claim the GGG[G*]P was ignoring), but also that not all "ints" should be considered compatible. While it may be syntactically correct and safe to add "int milesPerGallon" and "int averageAgeOfRockStars", semantically it's nonsense. Defining these as distinct types of "miles" and "years" would enable the type system to save a developer from making that mistake.
John
Or CLU, or simply the OpenCL C++ bindings which cover all of the boilerplate except stringification. Wrapping boiler plate in library code is hardly rocket science.