Research Experience

Job/Internships:
Research Associate, Department of CSA, Indian Institute of Science Bangalore (June 2016-now)
    Working on Polyhedral Compilation Technique with Prof. Uday Reddy B.

Internship at School of Computing Sciences, University of Glasgow (Feb 2016 – March 2016)
    Developed techniques to share profile information among different virtual machines over cloud database.
    Developed techniques to use the profile information of a previous run of program for reducing warm up times.
    Implemented techniques using for JikesRVM and Java Hotspot 8 VM using MongoDB as the database.
    Advisor: Dr. Jeremy Singer

Internship at Institute for Software Research, Carnegie Mellon University (May 2015 - July 2015)
    Developed a Just-in-Time compiler for Wyvern using LLVM and C++.
    Developed the conversion of Wyvern AST to Wyvern Core IL to Wyvern Object IR to LLVM IR.
    Developed optimized techniques for Dynamic Method Dispatch for Delegation support in Wyvern.
    Advisors: Prof. Jonathan Aldrich and Prof. Alex Potanin

Research Projects:

Master’s Thesis: Logical Approach to Code Generation and Optimization
    Developed techniques to use SMT Solver for generating Optimal Code.
    Implementing proof of concept using LLVM IR for ARM Instruction Set Architecture.
    Advisors: Dr. Greta Yorsh, Queen Marry University of London and Dr. L. P. Singh, IIT (BHU), Varanasi. 

Block Parallelized Parser
    Developed an Incremental LR (1) Parser to parse different blocks of a source code in parallel.
    Block Parallelized Parser achieves a performance improvement of more than 50% over Shift Reduce LR (1) parser.
    Developed Block Parallelized Parser and an LR (1) Parser in C#.

Adaptive JIT Code Diversificationin JVM
    Proposed a method to regenerate code dynamically in a JIT compiler and to update the diversification frequently during the 
execution of the program. Diversity used as a mechanism to protect from code-reuse attacks.
    Implemented proof of concept in Jikes RVM and average performance overhead achieved was 5-6%.

Constant Blinding to prevent ROP attacks in Google V8
    Developed a technique to hide constants from machine code by replacing them with random addresses in JavaScript.
    More than 90% of gadgets in two different diversified versions are removed and performance overhead was 3%.
    Implemented proof of concept in Google V8 JavaScript VM.

Heap Randomization to prevent and detect Heap Spray Attacks 
    Developed a technique to randomize heap layout by randomizing object layout, randomizing array layout, and developing a randomized breadth first search.
    Implemented proof of concept in JikesRVM.
    Performance Overhead reported was less than 2% with memory requirements less than 4%.

Offloading Generational Garbage Collector on GPU
    Developed algorithm to execute parallel collection of a Generational Garbage Collector on GPU.
    Parallel algorithms for both Young and Mature Collections have been implemented.
    Algorithm is written in OpenCL and has been integrated with Java Hotspot VM using C++.
    Speed improvement achieved was many times as compared to Java Hotspot’s Parallel Garbage Collector and current techniques

Publications:
1. Abhinav, Mohit Mishra, and Benoit Baudry. libmask: Protecting Browser JIT Engines from the Devil in the Constants. 14th Annual Conference on Privacy, Security, and Trust, Auckland, New Zealand.

2. Abhinav and Rupesh Nasre. FastCollect: Offloading Generational Garbage Collection to GPUs. International Conference on Compilers, Architectures and Synthesis for Embedded Systems (CASES) 2016, ESWeek 2016, Pittsburgh, USA.

3. Abhinav Jangda, Mohit Mishra and Bjorn De Sutter. Adaptive Just-in-Time Code Diversification.2nd ACM Workshop on Moving Target Defense, ACM CCS’15, Colorado USA.