May 2016 – Nov 2019 · 3 yrs 6 mos · Bengaluru Area, India · On-site

• Accelerating Deep Learning Libraries for embedded platforms. Design and Implementation of "DNN engine" for Nvidia TX1 platforms. The engine runs Deep learning models on the input video stream on a CUDA(NVIDIA)/SNPE(Qualcomm) device. Keep the CPU free from any deep learning process and offload all inference computation to the GPU. The inference Framework was also easily portable across different GPU architectures.

Dec 2014 – May 2016 · 1 yr 5 mos · Suwon, Gyeonggi-do, Korea

• Optimized various algorithms for the Samsung Exynos Platforms.
• Optimized the OpenCV library for the android pltforms, enabling the OpenCL backend. Further optimizing key routines.
• Designed a load distribution framework for CPU and GPU, for compute heavy workloads.

Sep 2008 – Dec 2014 · 6 yrs 3 mos · Bangalore, India

• I worked as an OpenCL Developer in AMD and do code optimizations for AMD's OpenCL Platforms. One of my project is linked below.
• http://github.com/HSA-Libraries/Bolt
• BOLT is an OpenCL accelerated C++ Template Library.
• I have also worked on Other projects like Framewave, ACML and AMDLIBM.
• http://developer.amd.com/tools-and-sdks/cpu-development/libm
• http://developer.amd.com/tools-and-sdks/cpu-development/amd-core-math-library-acml/
• My professional experience primarily invloves library developement in C, C++ and x86 Assembly programming. Have excellent experience in X86 Assembly Language programming with the following SIMD Instruction set families.
• SSE, FMA, XOP, AVX
• Worked to develop new algorithms for the math library routines. Have good experience in Parallel programming and vector input processing in x86 Assembly.
• Presently my key interest lies in OpenCL and accelerating applications using OpenCL. I have developed small applications like JPEG decoder, SPEC LBM, Spec Leslie as proof points of using the heterogeneous (CPU/GPU) platforms. The platform involved the AMD x86 CPU and the ATI Radeon HD Graphics GPU. General Purpose GPU computing is currently used for code speed ups in various fields like Multimedia/High Performance Computing etc. My work involves writing and developing data parallel programs and algorithms, to exploit the highly parallel compute architecture of the GPU.
• Expertise gained:
• C, C++ and X86 Assembly(SSE/XOP/FMA/AVX) - Excellent
• Code Optimizations, GPU and CPU - Excellent
• Architecture CPU and GPU. - Good
• OpenCL - Excellent
• OpenGL - Beginner

Dec 2005 – Sep 2008 · 2 yrs 9 mos · Bangalore, India

• Worked in the Settop Box Domain. Initially worked with test automation of the Motorola OCAP HAL. Then moved to firmware development for the Motorola DCT settop Boxes. Developed the Audio/Video Splicing API for the DCT settop platforms. The Splicing API was provided for both across the TS and within the TS splicing. Main use of this API is Targetted Advertising. Work involved from Requirements collection to development of the feature.
• Knowledge Gained:
• Embedded C programming Excellent.
• MPEG 2 - Transport layer.
• Cable headend and Settop box interface. - Minimal Experience

Jul 2005 – Dec 2005 · 5 mos · Noida Area, India

• This was my first job after my post graduation. My work was to develop XPATH interpreter for Samsung's NAVi Browser.
• This browser was to be used for Blue Ray Disc Players.