May 14, 2026

Created in May 14, 2026

2026   ·   gpu-performance   raytracing   gpu-security   rowhammer   ·   daily-log

Goals

  • Find out where exactly is the matcherTable stored in triton and how to access it.

What I worked on

  • Dumped the matcherTable using the libtriton.so in order to find out adjacent instructions (instructions at a 1-bit distance from the target instructions).
  • Developed a small PoC to and corrupted the a matmul triton program.

Key Insights

  • Learnt about the concept of catchpoints in gdb and how they can used to catch events like loading of a shared library.
  • Also learnt how to use pipe and grep inside gdb.

Questions / Confusions

  • I still do not understand how exactly everything is mapped inside the .so file will have to look at the code for it.

Ideas / Hypotheses

  • The manual process of figuring out flippable location is very tiresome and fragile, I would need to develop an automated framework for this.

Papers / Resources

  • None

Next Steps

  • Create an end-to-end automated framework for finding out the exploitable flips.

Introspection

  • Today was a nice day, my idea got somewhat validated, now after significant engineering effort I will be able to turn it into a good research work.
  • I have not had the chance to work on the raytracing idea, I will work on it after this is done.