I would appreciate clarity around hardware acceleration supported for ML Kit. Some applications make an explicit mention that models are run on CPU, implying that there can be other modes of acceleration. Using GPU via something like OpenCL seems like a natural way of doing so.
I wish to know if Google is capable and willing of mastering OpenCL for Machine Learning applications.
Currently, ML Kit features are all running in CPU to be compatible with all devices. We are adding GPU / NNAPI support for features, and will update them in future releases.
Related
I'm about to convert some GPU kernels of my project from OpenCL/Cuda to Metal in order to run my application on Apple devices. Currently, my project was written completely in C/C++. After doing some research, I think I need to get my hand dirty with Swift or Objective-C. But to be honest, I'm not sure about this stuff because Metal language for computation and deep learning is quite new.
I know there's a library called "CoreML", but my app requires some custom kernels. My question: what is the best way to deal with low-level API of Apple devices in my situation?
The Metal Shading Language is a version of C++. I haven't had too much trouble with porting OpenCL or CUDA kernels to Metal.
Core ML only supports a limited set of layers. You can write your own custom layers, which involves writing a CPU version and optionally a GPU version (in the Metal language).
I wrote a blog post about this: http://machinethink.net/blog/coreml-custom-layers/
A just curious question.
I'm aware that:
Skia, Cairo: library
Direct2D: API
But without any additional components, exclusively on Windows platform, which bears the most features? I'm not talking about performance or programming simplicity.
Is there any possibility that Cairo could use D2D as the rendering backend on Windows platform?
I am using all three in my Sciter as graphics backends so you can compare them in action.
As of today (Nov 20, 2017):
Feature set:
Skia and Direct2D are on par by feature set.
And both are feature wise supersets of Cairo.
From these three, only Direct2D (Direct2D v 1.1) supports printing. Skia and Cairo instead are capable of PDF generation so for real printing you will need 3rd party tools.
3D integration:
Direct2D has good integration with DirectX - they can work on the same DXGI device.
But it is far not that trivial to integrate Skia with existing 3D OpenGL code. Even Skia can work with OpenGL backend it changes state of OpenGL device and that creates problems.
Performance:
With Direct2D you can reach 60-250 FPS.
On Windows, Skia when used with OpenGL backend may surprisingly work better than Direct2D. But YMMV.
Platforms:
On Windows I'd suggest to use Direct2D or Skia.
On MacOS - definitely Skia with OpenGL backend. It is significantly better than Mac's native CoreGraphics.
On Linux/GTK: use Cairo as this is the default rendering API on GTK/Gnome3. And Cairo works there pretty well.
Is it possible to run OpenCL on a system designed by a user on a SoC prototyping board? To be more specific, I have a ZedBoard (Xilinx Zynq) that has Dual ARM cores and a Programmable Logic (PL) Area. If I design a simple system of my own that has a video processing accelerator implemented in the logic area, an ARM core and an AXI interconnect, what do I have to do to provide OpenCL support for this simple system? (In this simple system, the ARM core could be the "Host" and the video processing accelerator could be the "device").
I am a student and I have only some basic knowledge about OpenCL. I have researched about my question and have only ended up confusing myself. What are the things that have to be done to provide OpenCL support for a SoC? I understand that this may be a big project, but I need a guideline where to start and how to proceed.
what do I have to do to provide OpenCL support for this simple system?
Implement a OpenCL platform that makes either use of your ARM CPU or the FPGA (or both). I'd say that is pretty much impossible for you; ARM would surely offer one for the CPU if it was easy (and they definitely have the financial means to employ capable engineers/computer scientists), and implementing accelerators on an FPGA requires in-depth knowledge of FPGA development, as well as compiler theory and experience in systems design. I don't want to sound mean, but you seem to have none of these three.
You asked where to get started; I recommend just writing a first accelerator that e.g. adds up a vector of numbers; as soon as you have that, you will have a clearer idea of your task.
If you want to have a look at a reference: The Ettus USRP E310 is a zynq-based SDR device. Ettus has a technology called RFNoC, which allows users to write their own blocks to push data through. Notice that this took quite a few engineers and quite some time to get started. Notice further that it's much easier than implementing something that converts OpenCL to FPGA implementations.
If you have access to the Xilinx tools: Vivado HLS 15.1 System Edition should compile OpenCL kernels. This will also be included in the SDAccel tool suite.
Source: UG973: Vivado Design Suite User Guide Release Notes, Installation,and Licensing
An alternative might be switching to Altera. They provide some good examples for the Altera Cyclone V SoC which is comparable to Xilinx Zynq devices (also includes ARM Cortex-A9) :
AlteraSDK for OpenCL
I am also a student and my current project is also going on a similar direction, i have successfully installed a version of opencl called POCL on the zedboard, it successfully detects the arm cpu of the zedboard. To install pocl, you need llvm and a horde of other things as well. but basic steps to get pocl up on the zedboard are given below:-
Installing pocl:
http://www.hosseinabady.com/install-pocl-opencl
running example:
http://www.hosseinabady.com/embedded-system-by-examples/opencl_embedded_system/opencl-vector-addition
Lots of dependency: can resolved easily
but LLVM make sure you install 3.4 version for pocl 0.9
Steps to install llvm
https://github.com/pacs-course/pacs/wiki/Instructions-to-install-clang-3.1-on-ubuntu-12.04.1-and-12.10
POCL 0.9 is successfully working for me, as you do the installation you will face many other missing dependencies like hwloc, mesa libraries, open gl/cl headers icd loaders i hope you can resolve them as its a very big list to put up in stack overflow.
In order to detect your fpga as an open cl device, thats not going to be a trivial thing to do, you can refer to this link question i posted on github
https://github.com/pocl/pocl/issues/285
and also a research paper published by hosseinbady found on the publications link on the pocl website
http://pocl.sourceforge.net/publications.html
hope this helps you
Try the ARM OpenCL SDK. The Zedboard has an ARM A9 CPU, this should have a NEON SIMD vector unit http://www.arm.com/products/processors/technologies/neon.php which can run OpenCL. See http://www.arm.com/products/multimedia/mali-technologies/opencl-for-neon.php.
The Zedboard isn't listed as an OpenCL conformant platform https://www.khronos.org/conformance/adopters/conformant-products#opencl.
So there is a chance the ARM driver will not work.
Good luck!
If still relevant, try this paper OpenCL on ZYNQ [PDF]
Also note that Zynq-7000 is listed on https://www.khronos.org/conformance/adopters/conformant-products#opencl ( OpenCL_1_0 ), hence the compatibility.
I am starting to create a QT application with sqlite for a hand held device. My Project Manager asks me to select an operating system (embedded linux) for the device (we are not considering android).
As in Desktop, are there many embedded-Linux distributions for devices?
If so, Which embedded linux I should consider?
You have multiple choices, but I will suggest the easier and - in my opinion - better two.
Buildroot - is a set of makefiles that lets you create your custom embedded distribution. Can take care of building the Linux
kernel, the toolchain and a barebox or U-Boot bootloader, too. Easily expandable and
with a practically zero learning curve. You have a fully working
system in a matter of hours.
Yocto - a fully fledged (and complicated) build system. Suggested over Buildroot when you need a LOT of packages/components
and may need flexibility in expanding the system directly on
premises. What you can do substantially depends on the "layers" (sets of rules for building things) available: you combine layers together to obtain your system. Has a steep learning curve but is used and directly
supported by multiple vendors (e.g.: Atmel, TI).
Anyway, unless you have more than good reasons, I strongly suggest the former.
There are several Linux distros to be used with ARM. Maybe you should consider Fedora ARM https://fedoraproject.org/wiki/Architectures/ARM
This is a difficult question to answer not knowing more about the project requirements (not just software requirements, but also non-functional ones as well) and capabilities of the platform.
Angstrom (based on OpenEmbedded) is another possibility for Linux.
I would challenge the assumption that the operating system must be Linux. Why? If time-to-market or having commercial support are important, you might be better off with commercial embedded or RT operations systems such as VxWorks or QNX.
There are also professionally supported Linux distros such as Montavista
Whilst free linux distros are, well, free, you are generally on your own and your team's time isn't free.
You can use Qt for embedded device , it’s fast and compatible with many hardwares and if your hardware is not supported, porting it to a new hardware is not so hard
plus it has special rendering system
I have a system with an NVidia graphics card and I'm looking at using openCL to replace openMP for some small on CPU tasks (thanks to VS2010 making openMP useless)
Since I have NVidia's opencl SDK installed clGetPlatformIDs() only returns a single platform (NVidia's) and so only a single device (the GPU).
Do I need to also install Intel's openCL sdk to get access to the CPU platform?
Shouldn't the CPU platform always be available - I mean, how do you NOT have a cpu?
How do you manage to build against two openCL SDKs simultaneously?
You need to have a SDK which provides interface to CPU. nVidia does not, AMD and Intel's SDKs do; in my case the one from Intel is significantly (something like 10x) faster, which might due to bad programming on my part however.
You don't need the SDK for programs to run, just the runtime. In Linux, each vendor installs a file in /etc/OpenCL/vendors/*.icd, which contains path of the runtime library to use. That is scanned by the OpenCL runtime you link to (libOpenCL.so), which then calls each of the vendor's libs when querying for devices on that particular platform.
In Linux, the GPU drivers install OpenCL runtime automatically, the Intel runtime is likely to be downloadable separately from the SDK, but is part of the SDK as well, of course.
Today i finally got around to trying to start doing openCl development and wow... it is not straight forward at all.
There's an AMD sdk, there's an intel sdk, there's an nvidia sdk, each with their own properties (CPU only vs GPU only vs specific video card support only perhaps?)
There may be valid technical reasons for it having to be this way but i really wish there was just one sdk, and that when programming perhaps you could specify GPU / CPU tasks, or that maybe it would use whatever resources made most sense / preformed best or SOMETHING.
Time to dive in though I guess... trying to decide though if i go CPU or GPU. I have a pretty new 4000$ alienware laptop with SLI video cards, but then also an 8 core cpu so yeah... guess ill have to try a couple sdk's and see which preforms best for my needs?
Not sure what end users of my applications would do though... it doesnt seem like they can flip a switch to make it run on cpu or gpu instead.
The OpenCL landscape really needs some help...