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This page is outdated!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 UIDdcb3f442651f1046
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:dcb3f442651f1046
CompilerGCC 4.9
CPUQualcomm Technologies, Inc MSM8952
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 d32126f187e4a1c8 1.79 1.00 -O3 -fno-branch-probabilities -fno-devirtualize -fgcse -fgcse-las -fno-ipa-pure-const -floop-interchange -fpeephole -fno-sched-interblock -fno-split-ivs-in-unroller -ftracer -ftree-copyrename -fno-tree-loop-distribute-patterns -ftree-slsr -fno-tree-vrp -ffp-contract=off -fsched-stalled-insns=0 --param inline-unit-growth=38 --param large-stack-frame=79 --param gcse-after-reload-partial-fraction=2 --param gcse-cost-distance-ratio=12 --param max-unroll-times=9 --param max-unswitch-level=5 --param hot-bb-frequency-fraction=1904 --param max-predicted-iterations=23 --param omega-max-wild-cards=25 --param omega-hash-table-size=267 --param max-pipeline-region-insns=323 --param max-last-value-rtl=18963 --param min-size-for-stack-sharing=2 --param ipa-cp-loop-hint-bonus=39 --param allow-packed-load-data-races=1 -O3 1 0 2 1 milepost-codelet-mibench-consumer-mad-src-layer3-codelet-6-1 default 806.4, 806.4, 806.4, 806.4, 806.4, 806.4, 806.4, 806.4 1 TCL 6055B Android 6.0.1
S2 cabd54ca0b97b0d4 1.25 0.76 -O3 -fno-cse-follow-jumps -fdata-sections -fno-gcse-sm -fno-if-conversion -fdefer-pop -fno-math-errno -fno-optimize-sibling-calls -fno-rerun-cse-after-loop -fno-sched-critical-path-heuristic -fno-selective-scheduling -fsplit-ivs-in-unroller -ftree-loop-optimize -ftree-loop-vectorize -ftree-partial-pre -fno-tree-ter -funroll-all-loops -fno-unroll-loops -fno-web --param max-inline-insns-recursive-auto=540 --param large-unit-insns=1068 --param large-stack-frame=438 --param max-gcse-insertion-ratio=13 --param max-completely-peeled-insns=144 --param sms-max-ii-factor=11 --param hot-bb-frequency-fraction=202 --param max-predicted-iterations=120 --param max-crossjump-edges=102 --param lim-expensive=13 --param iv-max-considered-uses=396 --param omega-max-vars=241 --param sink-frequency-threshold=3 --param ssp-buffer-size=11 --param max-fields-for-field-sensitive=0 --param max-sched-ready-insns=183 --param prefetch-latency=399 --param sccvn-max-scc-size=4691 --param loop-block-tile-size=51 --param min-insn-to-prefetch-ratio=17 --param max-vartrack-expr-depth=9 --param lto-min-partition=222 --param allow-load-data-races=1 --param asan-globals=1 --param asan-memintrin=0 -O3 1 0 2 1 milepost-codelet-mibench-consumer-jpeg-c-src-jchuff-codelet-9-1 default 806.4, 806.4, 806.4, 806.4, 806.4, 806.4, 806.4, 806.4 1 TCL 6055B Android 6.0.1
S3 58f4f345dca48612 1.21 0.97 -O3 -fno-associative-math -fcheck-data-deps -fno-cx-limited-range -fno-expensive-optimizations -fno-gcse-after-reload -fno-if-conversion -fipa-cp-clone -floop-block -floop-strip-mine -ftoplevel-reorder -foptimize-sibling-calls -fno-rounding-math -fsingle-precision-constant -ftree-fre -fuse-linker-plugin -falign-loops=0 --param max-inline-insns-recursive=478 --param max-inline-insns-recursive-auto=215 --param min-vect-loop-bound=2 --param max-gcse-insertion-ratio=7 --param max-completely-peeled-insns=152 --param max-completely-peel-times=5 --param max-completely-peel-loop-nest-depth=0 --param sms-dfa-history=0 --param sms-loop-average-count-threshold=0 --param iv-max-considered-uses=283 --param vect-max-version-for-alignment-checks=2 --param max-sched-insn-conflict-delay=2 --param max-vartrack-reverse-op-size=30 --param ipa-cp-loop-hint-bonus=14 --param allow-packed-load-data-races=1 -O3 1 0 2 1 milepost-codelet-mibench-consumer-lame-src-newmdct-codelet-10-1 default 806.4, 806.4, 806.4, 806.4, 806.4, 806.4, 806.4, 806.4 1 TCL 6055B Android 6.0.1
S4 3a86832864c54b4b 1.12 0.88 -O3 -fno-branch-target-load-optimize -fdata-sections -fno-inline-functions -fipa-pure-const -fmodulo-sched -fno-toplevel-reorder -fno-rename-registers -fno-sched-dep-count-heuristic -fno-selective-scheduling2 -fsignaling-nans -fno-split-wide-types -ftree-dce -fno-tree-loop-optimize -fno-tree-sink -fno-tree-vectorize -funit-at-a-time -funroll-loops --param min-inline-recursive-probability=4 --param max-delay-slot-live-search=238 --param large-stack-frame-growth=588 --param gcse-after-reload-critical-fraction=2 --param max-peeled-insns=131 --param sms-max-ii-factor=80 --param sms-loop-average-count-threshold=0 --param builtin-expect-probability=52 --param lim-expensive=23 --param max-sched-insn-conflict-delay=1 --param ssp-buffer-size=4 --param max-sched-ready-insns=168 --param sccvn-max-scc-size=9447 --param switch-conversion-max-branch-ratio=5 --param graphite-max-nb-scop-params=1 --param loop-invariant-max-bbs-in-loop=2618 --param slp-max-insns-in-bb=711 --param tm-max-aggregate-size=11 --param ipa-cp-eval-threshold=247 --param asan-memintrin=1 -O3 1 0 2 1 milepost-codelet-mibench-consumer-tiff2rgba-src-tif-predict-codelet-4-1 default 806.4, 806.4, 806.4, 806.4, 806.4, 806.4, 806.4, 806.4 1 TCL 6055B 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