Jayvant Anantpur — Software Engineer

Software professional with experience in developing simulators for HDLs and related technologies. Research experience in GPGPU domain. I have done extensive research in the field of GPGPU, focusing mainly on improving performance of GPGPU applications. First research problem I worked on was to use CPU and GPU to execute MATLAB code, by automatically identifying the kernels to be executed on GPU and inserting the appropriate communication code between the two processors. Next, I developed a technique to parallelize and execute on GPUs, large loops that have dependencies among its iterations by dynamically identifying the dependencies and using the information to schedule the independent iterations. GPUs aren’t suitable for execution of code with branch statements where different threads can take different paths of the branches, effectively serializing the execution and duplicate execution in some cases. I developed a compiler transformation that avoids duplicate execution as well as helps extract more parallelism among nested branches. I also proposed enhancements to the warp scheduling algorithm to consider progress made by warps and thread blocks and prioritize the warps to reduce overall execution time. Continuing further in this direction, I worked on using the idea of persistent threads to define virtual warps and thread blocks. This involved source level code transformations and changes to the warp scheduling algorithm. I also explored the application of Reinforcement Learning techniques to design a warp scheduler that can learn to schedule warps. In addition, I also collaborated with other researchers to develop techniques to increase available parallelism by sharing registers and shared memory among thread blocks so that more thread blocks can be assigned to the SMs on the GPU. We also developed techniques to reduce energy consumption by identifying inactive registers and moving them to low power states. My research activities are listed at https://sites.google.com/site/jayvantanantpur Interested in Computer Architecture, Compilers, Parallel Computing, Heterogeneous Systems, EDA.

Stackforce AI infers this person is a GPGPU and Software Engineering expert in the semiconductor and EDA industries.

Location: Maharashtra, India

Experience: 27 yrs 9 mos