At this time Media SDK became part of Intel Media Server Studio, intended for cloud transcoding, broadcasting and live streaming. Enterprise segment brought new targets. From the beginning SDK had user friendly API and worked right out of the box without significant effort from developers. And now we had to provide direct access to HW capability and enable integration of customer algorithms into existent HW pipeline. I concentrated on designing and implementing such interface in close collaboration with HW architect team. I gathered customer requirements and incorporated their feedbacks, managed development and validation team. But at the beginning we can’t achieve required targets due to high complexity of the interface. There were no ways to validate all exposed controls and parameters. So I designed unique validation approach based on generation of artificial video sequences. After about a year of development we at last closed major validation gaps and got enterprise-quality solution. We have not had any full featured releases yet, but several top tier broadcasting companies are working on this low level interface integration into their products.

Together with more experienced colleagues I started design of the Intel Media SDK library aka Intel Quick Sync Video. It was an ambitious project, involving more than a hundred people, to create a complete set of video tools in one library - for decoding, encoding and video processing. All tools HW accelerated and with easy to use, user friendly interface. We reused existent UMC code and interfaces for decoding acceleration and together with HW architect team designed additional interfaces for encoding and video processing. The HW implementation itself was an ingenious mix of dedicated HW and general purpose processing units shared with 3D graphics. That gave powerful but still cost efficient solution. During these year my responsibility gradually changed and in 2011 I took full responsibility of the project as software architect. I designed numerous new features and test methods to support ever changing silicon and to accommodate customer requirements. Coordinated work between QA, driver and library development teams. Provided technical support for high touch customers in sophisticated use cases. Media SDK became de facto standard for video acceleration on all Intel silicon and OSes, including Android. It is used in hundreds of products from Microsoft Lync to HandBrake. It is distributed as part of video driver and installed on most of the windows systems, just search for libmfxhw.

PMP project demonstrated that CPU based video playback is not optimal for mobile applications and dedicated HW should be used instead. New project was started based on Menlow MID platform with Atom CPU and PowerVR graphics. I designed and implemented HW abstraction layer between driver and video codec, so called UMC Video Acceleration (VA) interface. It was my first experience in designing really complex interface and it was done surprisingly well. Even now, ten years later, this HAL is still used in Media SDK. It supports DXVA on Windows and just emerging at the time of creation VA API on Linux. I am proud to note, that I contributed to VA API definition and wrote its implementation on UMC side. I also developed cross platform video player that provided unified testing environment for HW accelerated playback on different OSes, including HW emulation mode. The project was definitely a success, we got several design wins on both Linus and Windows, but what is more important, it became the solid foundation for Media SDK.

I joined Integrated Performance Primitives (IPP) team and at the beginning optimized XScale version of IPP video library (ippvc). This library as part of a broader Unified Media Classes (UMC) package was used in Personal Media Player (PMP) project on both Linux and WinCE platforms. My task was to improve performance of MPEG2, MPEG4 and H.264 decoders using XScale assembler and Wireless MMX instructions. It soon became obvious that project was in really bad shape - none of the performance targets were met, no formal testing, constant schedule slip. So I took ownership of the whole PMP project. I quickly dealt with performance, brought it to required real time targets. Establishing proper testing took more time and get back to schedule was the most difficult task, but eventually the codec package was successfully released. I was rated outstanding at annual performance evaluation and got my first significant promotion.

It was quite unusual for Intel project. We developed full featured PSTN SW modem in silicon oriented company and what is even more remarkable we used third party chips in this modem. I worked with Silicon Labs Si3052 and Si3054 chipsets, wrote HW abstraction layer for PCI and AC’97 interfaces, including power management support and lot of HW specific stuff like surge protection. It was the closest to HW work I ever had in Intel, often used logic analyzer for debugging and even couple of times soldering station. The project was a great success, we had several design wins with top tiers OEMs like Dell and millions of our modems had been sold at the moment I left the team.