DV-STM

A dual-versioned software transactional memory library with over $3 \times$ speedup against the reference implementation using a coarse-grained lock.

TODO

• To understand the difference between inline … and static inline …
  and why

  cc -Wall -Wextra -Wfatal-errors -O2 -std=c11 -fPIC -I../include -c -o batcher.c.o batcher.c
  batcher.c:268:20: error: static declaration of ‘acquire’ follows non-static declaration
    268 | static inline void acquire(atomic_flag* lock) {
        |                    ^~~~~~~
  compilation terminated due to -Wfatal-errors.
  

• To understand the behavior of posix_memalign(…) regarding virtual address and alignment
• To understand dynamic allocation of struct with union
• To optimize "access sets" update
  Currently, per-word "access sets" are always updated during each read/write operation even though it is not necessary to do so.
• To understand synchronization examples in sync-examples/
• To understand examples in playground/
• To understand the workload and how an implementation is graded in grading/

Introduction

The project description includes

• An introduction to STM;
• The specifications of the STM implemented;
  • Sufficient properties for an STM to be deemed correct
  • The DV-STM algorithm
  • A thorough description of the STM interface
• Practical information.
  • How to test an implementation locally
  • The performance metric

Directory organization

Directory	Description
dv-stm/	DV-STM implementation
grading/	Workload and grader
include/	STM API
playground/	Unknown
reference/	A reference implementation using a coarse-grained lock
sync-examples/	Examples on synchronization primitives
submit.py	Autograding submission script

Acknowledgement

This project is the coursework of CS-453 Concurrenct Computing by École Polytechnique Fédérale de Lausanne (EPFL).