Skip to content

Latest commit

 

History

History
84 lines (53 loc) · 4.07 KB

README.md

File metadata and controls

84 lines (53 loc) · 4.07 KB
Raw

Intel-TSX-test-suite

This project includes a suite of simple programs to test Intels' TSX extension. The suite now includes the following tests

  • counting.cpp

    • increment a shared counter using pthread mutex and TSX instruction respetively

  • more to come!

Intel TSX

Intel TSX provides two software interfaces to specify regions of code for transactional execution

  • Hardware Lock Elision (HLE)

    • legacy compatible instruction set extension (comprising the XACQUIRE and XRELEASE prefixes) to specify transactional regions

  • Restricted Transactional Memory (RTM)

    • a new instruction set interface (comprising the XBEGIN, XEND, and XABORT instructions)
    • more flexible than HLE

Since a successful transactional execution ensures an atomic commit, the processor executes the code region optimistically without explicit synchronization. If synchronization was unnecessary for that specific execution, execution can commit without any cross-thread serialization. If the processor cannot commit atomically, the optimistic execution fails. When this happens, the processor will roll back the execution, a process referred to as a transactional abort. On a transactional abort, the processor will discard all updates performed in the region, restore architectural state to appear as if the optimistic execution never occurred, and resume execution nontransactionally.

Transactional Abort

A primary abort cause is due to conflicting accesses between the transactionally executing logical processor and another logical processor. Memory addresses read from within a transactional region constitute the read-set of the transactional region and addresses written to within the transactional region constitute the write-set of the transactional region. Intel TSX maintains the read- and write-sets at the granularity of a cache line. A conflicting access occurs if another logical processor

  • either reads a location that is part of the transactional region's write-set
  • writes a location that is a part of either the read-set or write-set of the transactional region.

A conflicting access typically means serialization is indeed required for this code region.

Check TSX Support

gcc checktsx.c -o check; ./check

The script checks all ISA extensions introduced in Haswell, only TSX support is relevant for this project.

If your machine does not support TSX extensions, you can use Intel SDE, which has the support but may not really represent a raw peformance.

Build & Test

Make sure your platform and your gcc supports -mrtm -mhle and c++11 std::chrono.

make -f makefile-counting

./benchmark --help

Option -c N specifies the counter to be incremented to 2^N for each thread. A good N should be 10-25.

Performance

My test machine only supports HLE, and the results were strange.

For the single-threaded case, pthread mutex always has better performance by a roughly 200%. Even though pthread mutex lock/unlock is high optimized, this is still a surprising result. Command: ./benchmark -v -t 1 -i 1 -c 24 -b 3.

For multi-threaded case, HLE could slightly outperform mutex. Command: ./benchmark -v -t 8 -i 1 -c 20 -b 3.

Resources

Existing TSX-enabled Locking Libraries (non-exhaustive)

  • On Linux, GNU glibc 2.18 added support for lock elision of pthread mutexes of PTHREAD_MUTEX_DEFAULT type. Glibc 2.19 added support for elision of read/write mutexes. Whether elision is enabled. depends whether the --enable-lock-elision=yes parameter was set at compilation time of the library.
  • Java JDK 8u20 or later support adaptive elision for synchronized sections when the -XX:+UseRTMLocking option is enabled.
  • Intel Composer XE 2013 SP1 or later supports lock elision for OpenMP omp_lock_t.

Tutorials