-
Notifications
You must be signed in to change notification settings - Fork 1
/
CHANGELOG
214 lines (184 loc) · 9.1 KB
/
CHANGELOG
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
Version 7.2
* 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
* 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
* 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
* 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
* 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
* 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
* 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
* 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
* 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
* 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
* Visualization support (--viz)
* Minor cleanups and bug fixes
Version 4.0
* 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
* 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
* Fix modeled size of network messages
* Build fixes for 32-bit, compiler overrides
Version 3.05
* 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
* 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
* Bugfixes in configuration parser, starting of multi-program workloads
Version 3.02
* 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
* Add heterogeneous cache configuration support
* Emulate pause, sleep system calls
* Improve support for 32-bit applications
* Pin 2.11 support
Version 3.0
* 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
Version 2.04
* Fix trace playback for non-predicated instructions
Version 2.03
* Fix record-trace -d 0
* Fix keyboard interrupt behavior
Version 2.02
* [sift_recorder] Add support for -f (number of instructions to fast-forward) and -d (number of instructions to trace in detail) command-line options
Version 2.01
* Compilation fixes for GCC 4.6
Version 2.0
* Multi-program mode
* Instruction trace collection tool
* New ROI-aware sim.out file generation
* CPI components for various SYS_futex system calls
* Improve accuracy of the core interval model
* GCC 4.6 support
Version 1.06
* Fix queue overflow when many TLB misses occur in a single basic block
Version 1.05
* [cpistack] Add --abstime command-line parameter to scale Y axis according to absolute time in seconds
* [interval] Fix branch resolution latency calculation
Version 1.04
* Fix instruction dependencies for LEA and stores
* Support synchronous signal handling (when using the
-g --general/enable_signals=true command-line parameter)
Version 1.03
* Build changes to improve compatibility with Fedora 16
Version 1.02
* Remove unneeded warnings from CPI stack script
Version 1.0
* Initial public release