ARM Community: The Mali Ecosystem makes GPU Computing a reality! - ARM Community

I have just attended the Mobile World Congress in Barcelona, and for another year survived an intense week of tapas, paella and customer meetings overindulgence. I was overwhelmed by all the positive feedback received from ARM partners and booth visitors on the great progress and industry leadership position of ARM Mali™ GPU Computing technologies. This is not surprising, we worked very hard to get here, but what makes this special is seeing all the great work from our partners and collaborations starting to flourish.

Fast and continued progress
Little over a year ago I wrote with enthusiasm about how the Mali-T600 series of GPUs is uniquely placed to accelerate GPU Computing on Android thought Renderscript Compute. Since then, Mali-T604 has been certified Khronos Conformant for OpenCL 1.1 Full Profile (yes FULL profile) on Linux and Android. Mali-T604 has been proven in silicon (Samsung Exynos 5 Dual) and is shipping in chart topping consumer products (Google Chromebook, Google Nexus 10) as well as mass production low-cost Developer Boards (InSignal Arndale Community Board).

The ecosystem delivers
Our ecosystem partners have now started making good use of this technology, and at the Mobile World Congress we were proud to showcase and see a variety of demos that validate this.

Kishonti Informatics is a provider of standards-based cross-architecture benchmarking tools. On our booth we welcomed Kishonti’s CLBenchmark running on a Mali-T604 based InSignal Arndale platform as well as their RSBenchmark (the new Renderscript compute benchmark) running on a Google Nexus 10 device. To date I have not seen either of these benchmarks running on any other mobile class GPU.



MulticoreWare is a software vendor and provider of professional services to optimize software and algorithms to leverage heterogeneous computing, including technologies such as Renderscript and OpenCL. In the interview below, Curtis Davies, COO and VP Engineering, illustrates the performance benefits achieved for computational photography applications showcasing a variety of image filters accelerated using Renderscript compute on Mali-T604 GPU on the Google Nexus 10 device, as well as a GPU accelerated video image stabilization application. It is great to see how applications that would typically be done on desktop or even server platforms can now be migrated directly onto the mobile device (and this is another reason why supporting FULL Profile OpenCL capabilities makes sense).



There were many more demos previewed at the show from our partners.

Aptina, the industry leader in CMOS sensor technologies, demonstrated GPU accelerated computational photography. They used their AR0833 iHDR sensor (this is a high performance 8MP BSI sensor that can run full resolution at 30fps), to feed raw data to the Samsung Exynos 5250 on an Arndale platform, with the whole image processing pipeline offloaded to the GPU using OpenCL (noise reduction, iHDR reconstruction, tone mapping, colour-conversion, gamma correction, and de-mosaicing).



Apical, a leader in advanced image and video processing technology for embedded applications, is working with ARM on the optimization of their IRIDIX software solution for Mali-T604 using OpenGL ES and OpenCL, and was able to preview their Assertive Display technology on an InSignal Arndale platform. The screenshots below show the benefits of using Assertive Display to improve the user visual experience in a common setting (reasonably bright day, outdoors, with indirect sunlight).




Ittiam Systems, a market leading Media Processing and Communications IP Company, which specializes in embedded multimedia Software and Systems, has been previewing an OpenCL enabled HEVC decoder on Mali-T604 GPU. Some screenshots can be seen below.



Synthesis Corporation, a pioneering Japanese industry-academic cooperative ventures provider of semiconductor and software IP with recent focus around image and video processing as been collaborating with ARM on the optimization of some proprietary technologies using GPU compute on Mali. Synthesis’ Super-resolution Scaler has been implemented in OpenGL ES and Renderscript compute for the Mali-T604 GPU, and is capable of achieving up to five times performance increase in comparison to a conventional CPU-only implementation. Synthesis Corporation’s proprietary adaptive luminance/dynamic-range enhancement software library optimized for Mali GPU using OpenCL.



Morpho is a Tokyo based R&D company with strong expertise in image processing, who licenses software libraries and SDKs for mobile/consumer devices. Morpho and ARM are collaborating on GPU Compute enablement including port and optimization of movie effect application and their industry leading video stabilizer technology MovieSolid to Mali based platform and acceleration using GPU Compute on Mali-T604.



This is just the start, and we expect a large amount of application and technologies been enabled and optimized by partners using Mali GPU Computing through the rest of 2013 and beyond. GPU Computing is now here for mobile computing, and its benefits are available for the whole ARM Mali partnership to take advantage of.


Would you like to be part of this?
ARM is investing heavily on enabling and promoting a strong ecosystem around GPU Compute.

This is done through directly engaging with leaders and experts across the whole industry, targeting a variety of markets segments (Consumer electronics and mobile computing, DTV, entertainment, security, safety, HPC/Server) and expertise and skills (Heterogeneous Computing, GPU Computing, Computational Photography, Sensors, Compilation technologies, Gaming, UIs, Tools, Libraries, Consultancy and Services, Imaging and Computer Vision).

Our primary objective in building a strong GPU Compute Ecosystem around Mali is to help OEMs and SiPs extract the benefits of GPU Compute for their applications and use-cases.

If you would like to be part of this success story, please get in touch at gpucompute-info@arm.com.

Roberto Mijat, GPU Computing Marketing Manager, ARM. Since joining ARM in 2004, Roberto worked as a software engineer in a multitude of projects: enabling SMP operating systems, benchmarking, 3D graphics, hardware validation and software optimization. Subsequently, as a Senior Software Architect, he spent most of his time driving evangelical and enabling activities around advanced features of the ARM application processor roadmap: multi-core software, virtualization and power management. In his current role, Roberto is responsible for GPU Computing products and technologies such as OpenCL and RenderScript. Roberto holds a degree in Artificial Intelligence, Natural Language Processing and Virtual Reality, and his research focused on parallelization of multimedia codecs.