Releases

Download Sniper

You can download the latest version of Sniper here.

Recent Sniper Releases

Version 7.4 (April 15th, 2021)

• Add initial ARM64 port
• Automatically download Pin 3.18-98332
• Require GCC >= 5, support modern compilers (GCC 9), Ubuntu 20.04

Version 7.3 (August 8th, 2019)

• Fix compilation and version detection issues for modern Pin versions (3.10)
• Properly handle time during fast-forwarding and warmup by default
• Remove transitional Pin 3 SIFT recorder code

Version 7.2 (February 16th, 2019)

• Improve SIFT recorder error handling to reduce crashes caused by aborts
• Fix additional crashes caused by modern GCC behavior (NULL valued this pointer checks are removed by the compiler)
• Verify thread count in the sift recorder. Support larger thread counts with the new --maxthreads option
• Fix debug symbol parsing with modern Pin versions
• Numerous bug fixes and improvements

Version 7.1 (October 28th, 2018)

• Fix compilation errors for GCC 6 and 7
• Fix parallel make builds
• Add Dockerfile for Ubuntu 18.04
• Make zlib optional (Pin3 requires a PinCRT-compiled zlib which is found with Pinplay)
• Numerous bug fixes and improvements

Version 7.0 (September 6th, 2018)

• Added Pin 3, GCC 5 support by default, keeping Pin 2 support in the new pin2 branch
• Added initial RISC-V support
• Add a new utility to automatically compile RISC-V requirements and SPEC CPU2006 binaries for RISC-V
• Add a Docker image for Sniper and RISC-V compilation support
• Fixes SIFT recorder address mapping issue caused by physical page mapping access restrictions in Linux kernels 4.0+

Version 6.1 (March 24th, 2015)

• Support for newer Pin versions, up to Pin 2.14-71313 / PinPlay 2.1
• Support GCC 4.9
• Integrate the Cheetah cache models
• Remove legacy Graphite core models simple, magic and iocoom
• A new dumpstats.py option (-c|--config) prints configuration parameters one-per-line
• Numerous bug fixes and improvements

Version 6.0 (June 3rd, 2014)

• Add Instruction Window-Centric core model (internal name: ROB core model, use rob.cfg)
• Add SMT support (IWC/ROB core model only, use smt*.cfg)
• Add SniperLite, a fast cache-only model (use nehalem-lite.cfg)
• Add DRAM cache prefetcher
• Add support for non-standard SQLite paths through the SQLITE_PATH environment variable
• Add sniperdiff.py for easy diff-ing of Sniper results and configurations
• Branch predictor accuracy improvements
• New roi-icount script to more easily specify fast-forward/warmup/detailed lengths by instruction count
• Faster mode for running single-program / single-threaded PinBall (--pinball-non-sift)
• Support for newer Pin versions, up to Pin 2.13.65163 / PinPlay 1.3
• Numerous bug fixes and improvements

Version 5.3 (November 29th, 2013)

• Fast, cache-only one-IPC timing model (-c cacheonly)
• Add energystats script to provide runtime energy counters
• Add Python-based thread scheduler infrastructure and example
• Add support for Query-Based Selection (Jaleel, MICRO 2010)
• Add support for plotting Bottle graphs (Du Bois, OOPSLA 2013)
• Memory tracker infrastructure to measure cache hit rates by allocation site
• Support for McPAT 1.0
• Various bug fixes and improvements

Version 5.2 (September 17th, 2013)

• Configurable coherency protocol (MSI/MESI/MESIF), made MESI the default
• Add more cache statistics: LRU stack distance historgram, LLC miss latency breakdown
• Implement auxiliary tag directories (ATD) to track constructive/destructive interference in shared caches
• Implement 2-level TLB hierarchy with Nehalem configuration
• New hooks HOOK_APPLICATION_ROI_{BEGIN,END}, called even when ROI markers are not used directly; these hooks can be used to trigger ROI from a script
• Improved stop-by-icount script to support ROI-relative warmup and detailed lengths
• Ondemand routine stack printer: configure routine_tracer/type=ondemand, then send a SIGUSR1 to Sniper to get a per-thread application backtrace
• Emulate leaf 11 of the cpuid instruction to pass topology information to runtimes (used by Intel OpenMP)
• Emulation of sched_* system calls, gettimeofday replacement, cpuid in SIFT mode
• Improve handling of LD_LIBRARY_PATH: use SNIPER_SIM_LD_LIBRARY_PATH for the simulator, SNIPER_APP_LD_LIBRARY_PATH for the application
• Re-implemented BigSmall scheduler to use thread affinity calls rather than the low-level (and error prone) moveThread API
• sim.thread Python interface to interact with threads (get num threads, get appid, get/set affinity)
• Use newest Pin version 2.13.61206
• Numerous bug fixes and improvements

Version 5.1 (June 27th, 2013)

• New Suggestions for Optimization visualization (--viz-aso)
• KCacheGrind-compatible output for profiling simulated applications (--profile)
• Roaming (equal-time) scheduler allowing for thread migrations (scheduler/type=roaming)
• Support for newest Pin version 2.12.58423
• Various bugfixes and improvements

Version 5.0 (April 15th, 2013)

• Periodic sampling infrastructure
• Extensible per-thread statistics infrastructure
• Routine tracing infrastructure and per-function statistics
• NUCA cache model
• Distributed tag directories
• sim.mem Python module for reading application memory
• Various other improvements and bugfixes

Version 4.2 (February 9th, 2013)

• Various accuracy fixes for Nehalem core model
• Add cache replacement policies: NRU, MRU, NMRU, PLRU, S-RRIP, Random
• Add statistical DRAM performance model
• Add syscall enter/exit hooks
• Add topology view to visualization
• Speed up McPAT by caching architecture-specific CACTI results
• Fixes to running multiple multi-threaded workloads
• Multi-programmed mode: end simulation at first/last program end, optional trace/application restart
• PinPlay support

Version 4.1 (December 11th, 2012)

• Visualization support (--viz)
• Minor cleanups and bug fixes

Version 4.0 (November 28th, 2012)

• Thread migration and scheduler support
• Pinned (round-robin), static, random thread schedulers
• Heterogeneous configuration files with tags
• Configurable address2set hash functions for non-power of two sized caches
• Various prefetcher improvements
• DRAM cache model
• One-IPC fast-forward model
• Fault injection framework
• New SQLite3-based statistics format
• ROI support for SIFT
• Support for MPI applications (shared-memory backend)
• Limited support for Jikes/DaCapo benchmarks
• Use newest Pin 2.12.53271
• Add script for generating topology images
• Preserve history in Git repository
• Many cleanups and bugfixes

Version 3.07 (September 7th, 2012)

• Prefetcher improvements, add global history buffer-based prefetcher
• HOOK_PERIODIC_INS: Instruction-based periodic callback
• Implement CLONE_CHILD_CLEARTID syscall interface
• Add example scripts for periodic statistics, periodic McPAT, simulating limited iteration counts
• Support for Pin 2.12
• Fixes to Python environment
• Various bugfixes

Version 3.06 (August 7th, 2012)

• Fix modeled size of network messages
• Build fixes for 32-bit, compiler overrides

Version 3.05 (August 7th, 2012)

• Scheduler: expose application ID
• Add example script roi-iter.py to dynamically select ROI based on SimMarkers
• CPI stacks: --aggregate and --partial support, fixes for heterogeneous configurations
• Traces: support for 32-bit executables
• Build fixes for older Linux versions

Version 3.04 (July 18th, 2012)

• Support for running multiple multi-threaded workloads in a single simulation
• McPAT fixes for heterogeneous configurations
• Build system fixes for newer Linux versions

Version 3.03 (June 28th, 2012)

• Bugfixes in configuration parser, starting of multi-program workloads

Version 3.02 (June 25th, 2012)

• Fixes for specifying heterogeneous configurations
• L2 prefetcher improvements
• Perfect cache modeling
• Self-modifying code support
• PyControl scripting interface
• GCC 4.7 support
• McPAT integration for area, power and energy predictions

Version 3.01 (June 12th, 2012)

• Add heterogeneous cache configuration support
• Emulate pause, sleep system calls
• Improve support for 32-bit applications
• Pin 2.11 support

Version 3.0 (May 16th, 2012)

• Support for heterogeneous core types
• Separate core microarchitectural characteristics into CoreModel class
• Improve CPI stack detail
• Add initial implementation for basic L2 prefetcher
• Optionally access DRAM directly in configurations with a single LLC
• Deprecate replacement of pthread_* synchronization calls
• Support more SYS_futex options
• Remove unused code for Graphite FULL mode
• Fixes to the build system, including parallel builds (make -j)
• Support for building on 32-bit hosts
• Remove configuration defaults from code, require everything to be specified in a configuration file