VMW Group Research Projects

Performance Counter Related

• Performance Counter Determinism Analysis

Performance API (PAPI) Related

• PAPI overhead on various kernel interfaces, including perf_event rdpmc
• PAPI papers and documentation
• Simple Walkthrough of How to Instrumnt Code with PAPI

Linux perf_events subsystem

• Unofficial Linux Perf Events Performance Counter web-page
• PAPI and perf_event for Knights Corner / Xeon Phi

Security / Fuzzing

• The perf_fuzzer

RAPL and Power Measurement

• Reading Intel RAPL Energy Counters on Linux
• DRAM RAPL Validation
• Notes on using WattsUpPro meter on Linux
• Summary of Power Measurement research

Computer Architecture

• Program to disable hardware prefetch on Intel CPUs

Raspberry Pi Related

• Raspberry Pi Cluster
• VMW Bare-metal Operating System

Code Density

• ll_asm code density comparisons

Silly Projects

• Insert umlauts into your LaTeX document

VMW Software Projects

• More projects can be found at my personal webpage: VMW Software Productions

Older Projects, Infrequently Updated

Performance Counter Related

• Performance Counter Event Validation Discussion
• Performance Counter Validation Tests
• Perf Events from inside of Qemu
• Floating Point Ops Validation

PAPI Related

• PAPI component to count interrupts
• PAPI libpfm4 migration info
• PAPI-V (virtualization) results

Old Class Projects

• A simple implementation of Steengaard's algorithm (May 2006)
• Using machine learning techniques to optimize ATLAS (May 2006)

Basic Block Vectors (BBV) / SimPoint

• BBV Research - Various methods for generating Basic Block Vectors as needed by the SimPoint analysis tool.
• alpha-simpoints - At one point I had actual Alpha hardware so I could easily generate modern SPEC2000/2006 binaries as well as simpoints for use with simplescalar.
• html2simpoint-0.1.tar.gz (2kB) Code that converts the Alpha/atom BBV results on the SimPoint website back into discrete bbv files.
• Phase Plots -- SPEC CPU Phase Behavior Plots
• SimSnap (page at Cornell now defunct) - This project used checkpoint/restart in conjunction with Simpoint to allow full-speed fastforwarding to the interval of interest. My work was writing a custom File I/O checkpoint/restore implementation with enough features to work for the SPEC2000 benchmarks.

Cache Conscious Data Placement

• Cache Tool - This tool gathers statistics on cache behavior. The eventual goal was cache-conscious data placement but as you will eventually discover this is a thankless task.

Simulator Related

• m5 -- For a while I maintained a set of patches needed to get SPEC CPU running on x86 gem5
• nursim - new version of ursim that runs Irix MIPS binaries. Lots of hacking trying to get simsnap binaries to run.
• Qemu - Investigated using qemu to generate memory traces (project info here).
• SimpleScalar Alpha - various projects
• SimpleScalar x86 - trying to get simsnap binaries to run.
• TAXI - I created a valgrind plugin (rather than the original BOCHS frontend) for TAXI and used it to simulate x86 binaries.

Instrumentation Tools

• Atom - (for the Alpha) used for various projects
• FIT - used for a cache performance tool. Contributed code to the FIT developers which has been merged into the main codebase.
• Valgrind
  • cachetool - made a plugin for Valgrind that generates memory traces for use with my cache tool.
  • exp-bbv - made a Valgrind plugin that generates SimPoint info, enabling Basic Block Vectors (BBVs).
  • valtaxi - made a Valgrind plugin that produces TAXI-compatible input streams (TAXI decodes pre-decoded x86 instruction streams into PISA uops and runs through a modified version of the PISA SimpleScalar).
  • Valgrind - Ruby - an attempt to generate multi-core memory simulations by using Valgrind and the Ruby multi-core cache simulator from GEMS/m5.
• Qemu
  • qemusim - a patch to Qemu to generate SimPoint Basic Block Vectors (BBVs).
  • qemu-trace - a patch to Qemu that generates memory traces for cache simulators.

Example Code

• convolve Simple image convolution example. Gives your cache a workout.