Collective Knowledge Aggregator proof-of-concept
Crowd results Raw CK browser Graphs Reports Datasets Models Home

This page is outdated! New version is available here.


Distinct solutions after online classification (auto/crowd-tune GCC compiler flags (minimize execution time))

Scenario UID8289e0cf24346aa7 (experiment.tune.compiler.flags.gcc.e)
Data UID2a0abf405aa42206
Discuss (optimizations to improve compilers,
semantic/data set/hardware features
to improve predictions
, etc):
GitHub wiki, Google group
Download:[ All solutions in JSON ], [ Solutions' classification in JSON ]
Reproduce all (with reactions):ck replay 8289e0cf24346aa7:2a0abf405aa42206
CompilerGCC 4.9
CPUQualcomm Technologies, Inc MSM8917
Objectivemin
Improvement key IK1Main kernel execution time speedup [min]
Improvement key IK2Code size improvement

Improvements (<4% variation) Distinct workload for highest improvement
# Solution UID IK1 IK2 New distinct optimization choices Ref Best species Worst species Touched Iters Program CMD Dataset Dataset file CPU freq (MHz) Cores Platform OS Replay
S1 458cb284951e59d7 1.39 1.00 -O3 -fno-indirect-inlining -fno-inline-small-functions -fno-keep-static-consts -fzero-initialized-in-bss -fno-rerun-cse-after-loop -fsched2-use-superblocks -fno-tree-copy-prop -fno-tree-loop-im -fno-tree-switch-conversion -ffp-contract=fast --param max-pending-list-length=40 --param gcse-after-reload-critical-fraction=8 --param builtin-expect-probability=86 --param max-cselib-memory-locations=583 --param max-sched-extend-regions-iters=0 --param max-fields-for-field-sensitive=0 --param prefetch-latency=340 --param tm-max-aggregate-size=10 -O3 1 0 2 1 milepost-codelet-mibench-consumer-jpeg-c-src-jchuff-codelet-9-1 default 1401, 1401, 1401, 1401 1 VIVO PD1628F_EX Android 6.0.1
S2 4b8b9f3fffff1e26 1.37 0.26 -O3 -fno-dse -fipa-sra -fno-expensive-optimizations -fgcse-las -fipa-cp -fipa-pure-const -fno-keep-static-consts -flto -fno-merge-all-constants -fno-function-cse -fno-guess-branch-probability -fno-inline -fpeephole -fthread-jumps -fno-tree-bit-ccp -ftree-forwprop -ftree-pta -fno-tree-reassoc -funit-at-a-time -fno-vect-cost-model --param max-inline-insns-recursive=668 --param max-early-inliner-iterations=2 --param comdat-sharing-probability=5 --param max-delay-slot-insn-search=112 --param large-function-growth=13 --param inline-unit-growth=21 --param sms-loop-average-count-threshold=0 --param hot-bb-count-ws-permille=209 --param min-crossjump-insns=10 --param max-cse-insns=1969 --param max-sched-extend-regions-iters=0 --param sched-state-edge-prob-cutoff=72 --param l1-cache-size=1 --param sccvn-max-alias-queries-per-access=417 --param max-stores-to-sink=1 --param asan-use-after-return=0 -O3 1 0 2 1 milepost-codelet-mibench-automotive-susan-e-src-susan-codelet-10-1 default 960, 960, 960, 960 1 SAMSUNG SM-S727VL Android 6.0.1
S3 e80409bd9c14bdeb 1.14 1.01 -O3 -fassociative-math -fno-crossjumping -fcx-limited-range -fgcse-las -findirect-inlining -fipa-pta -fkeep-static-consts -floop-strip-mine -floop-parallelize-all -fno-lto -fsigned-zeros -fsched2-use-superblocks -fsched-spec-load -fno-sched-rank-heuristic -fno-section-anchors -fsignaling-nans -fno-strict-overflow -fno-tree-bit-ccp -ftree-dse -fno-tree-loop-distribute-patterns -fno-tree-pre -ftree-sra --param partial-inlining-entry-probability=72 --param max-delay-slot-insn-search=37 --param max-modulo-backtrack-attempts=42 --param max-peeled-insns=18 --param max-completely-peel-loop-nest-depth=1 --param hot-bb-count-ws-permille=857 --param max-cse-path-length=11 --param max-sched-insn-conflict-delay=6 --param max-fields-for-field-sensitive=0 --param l1-cache-line-size=15 --param loop-invariant-max-bbs-in-loop=7734 --param max-vartrack-size=15666785 --param asan-instrument-reads=0 --param asan-memintrin=1 -O3 1 0 2 1 milepost-codelet-mibench-telecomm-adpcm-c-src-adpcm-codelet-1-1 default 1401, 1401, 1401, 1401 1 VIVO 1610 Android 6.0.1
S4 3f6339f435b9e22e 1.12 1.01 -O3 -fno-conserve-stack -fno-delayed-branch -fno-forward-propagate -fgcse-las -fif-conversion2 -fno-indirect-inlining -fno-ipa-pta -floop-parallelize-all -fno-modulo-sched -fno-modulo-sched-allow-regmoves -fsched-interblock -fsched-dep-count-heuristic -ftree-loop-linear -ftree-loop-vectorize -fno-tree-pre -funswitch-loops --param max-inline-insns-auto=12 --param early-inlining-insns=8 --param max-average-unrolled-insns=124 --param max-completely-peeled-insns=184 --param max-once-peeled-insns=783 --param max-completely-peel-loop-nest-depth=13 --param max-unswitch-insns=52 --param max-iterations-to-track=861 --param max-cse-path-length=20 --param scev-max-expr-complexity=13 --param omega-max-eqs=174 --param omega-max-wild-cards=9 --param vect-max-peeling-for-alignment=49 --param l1-cache-line-size=6 --param ira-max-conflict-table-size=395 --param ipa-max-agg-items=8 --param lto-min-partition=1113 --param max-tracked-strlens=388 --param asan-use-after-return=1 --param uninit-control-dep-attempts=329 -O3 1 0 2 1 milepost-codelet-mibench-telecomm-gsm-src-short-term-codelet-2-1 default 1401, 1401, 1401, 1401 1 VIVO PD1628F_EX Android 6.0.1



[ Participated users, platforms, OS, CPU, GPU, GPGPU, NN, NPU ] [ How to participate ] [ Motivation (PPT) (PDF) ] [ Papers 1 , 2 , 3] [ Android app ] [ Collective training set ] [ Unified AI ]
View entry in raw format

Developed by Grigori Fursin           
Implemented as a CK workflow
                         
   
                      Hosted at