Prashant Patil

Experience

• 

  Conquest Integrated Technologies

  Feb 2005 - Nov 2005

  DSP - Embedded Engineer

  1. Image processing algorithms like JPEG.2. Video codec development (H.264) and optimizations for TI C64x+ DSP

• 

  Trinity Convergence

  Dec 2005 - May 2010

  Senior Software Engineer

  1. H.264 video encoder/decoder developmentand optimization on TI DSP platforms such asTI C64x+ DSP.2. Video codecs development (MPEG-2/4, H.264,H.263) and streaming containers likeMPEG2TS, MP4 etc.3. Media Playback System Design andimplementation.

• 

  Persistent Systems

  Jun 2010 - Jan 2012

  Module Lead

  1. Developing skype v2ip framework stack backend.2. Integration for video and audio codecs intoskype backend.3. Video conferencing features implementationand testing4. RTP packetizer implementation foraudio/video streaming in skype5. AEC integration and testing into skypebackend

• 

  Imagination Technologies

  Dec 2011 - Oct 2023

  Staff Software Engineer

  1. Spearheaded the development of AI/ML CNN compilers for dedicated Neural Network Accelerator hardware IP, resulting in improved performance and efficiency. - Design and development of graph manipulation features of the compiler - Layers fusion for better optimization. - Layer splitting (Eltwise/Convolutiom/Pooling) to fit into hardware pass. - Optimizer enhancements to decide the best splits - Network segmentation (hardware/software) - IO/Internal formats decisions for layer groups - Hierarchical CnnModel design and implementation - Fixing accuracy issues for classification/ssd networks - Netowrk Performance (Inferences Per second) enhancements - End-to-End network mapping to the NNA hardware - Porting different networks from mutiple framworks such as Tensorflow, TFLite, Caffe etc. - Graph optimization for performance and bandwidth - Quantized networks support in NNA compiler - LSTM/WLM networks support in NNA compiler - Harware adapter to create command stream to run on NNA hardware - Customer support to fix accuracy & Performance2. Worked on the design and development of V2IP Framework.3. Developed and optimized the C reference code for H.264/H.265 video codecs, achieving 30 FPS performance improvement on multiple platforms like ARM (NEON), Intel SSE3 and MIPS MSA.4. Contributed to the development of OpenMax plugins for in-house developed video codecs, enhancing the functionality and compatibility of the software5. Played a key role in implementing MIPS MSA (SIMD) optimizations for AOSP, improving the performance of the software on MIPS-based platforms Show less

• 

  Lattice Semiconductor

  Apr 2024 - now

  Staff Machine Learn Engineer