Collective Knowledge Aggregator proof-of-concept
Crowdsourced experiments CK project Partners Open AI powered by CK Reusable AI artifacts Get CK

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

Scenario UID8289e0cf24346aa7 (experiment.tune.compiler.flags.gcc.e)
Data UIDb9a849608ec924c7
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:b9a849608ec924c7
CompilerGCC 4.9
CPUQualcomm Technologies, Inc MSM8996
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 da848015be3fb00f 2.09 1.00 -O3 -fbranch-target-load-optimize -fno-finite-math-only -fno-hoist-adjacent-loads -finline-small-functions -floop-nest-optimize -fno-function-cse -fno-peephole -fno-peephole2 -fno-sched-interblock -ftrapping-math -fno-peel-loops -fno-sched2-use-superblocks -fno-sched-pressure -fno-sched-spec-insn-heuristic -ftree-dse -fno-tree-loop-ivcanon -finline-limit=0 -fsched-stalled-insns-dep=0 --param large-function-growth=7 --param gcse-cost-distance-ratio=15 --param max-unroll-times=1 --param max-completely-peel-loop-nest-depth=8 --param hot-bb-frequency-fraction=1926 --param align-threshold=72 --param max-grow-copy-bb-insns=2 --param omega-max-geqs=246 --param max-sched-extend-regions-iters=0 --param simultaneous-prefetches=5 --param loop-max-datarefs-for-datadeps=1657 --param prefetch-min-insn-to-mem-ratio=1 --param ipa-cp-eval-threshold=679 --param asan-stack=1 --param asan-globals=0 --param asan-instrument-writes=1 --param asan-memintrin=1 -O3 1 0 2 1 milepost-codelet-mibench-office-rsynth-src-nsynth-codelet-5-1 default image-ppm-0001 data.ppm 1036.8, 1036.8, 1036.8, 1036.8 1 SAMSUNG SM-G930V Android 6.0.1
S2 cb2b44ec6b2fd12f 2.02 1.11 -O3 -fassociative-math -fbranch-target-load-optimize -fdelayed-branch -fdevirtualize -fhoist-adjacent-loads -fno-inline -freciprocal-math -freorder-blocks -freorder-blocks-and-partition -fno-rounding-math -fsched-last-insn-heuristic -fsplit-ivs-in-unroller -fno-tree-ch -fno-tree-loop-if-convert -ftree-loop-distribution -fno-vpt -falign-labels=0 -falign-loops=0 --param max-inline-insns-single=195 --param max-inline-insns-auto=66 --param max-inline-recursive-depth=15 --param max-inline-recursive-depth-auto=6 --param max-completely-peel-times=17 --param max-unswitch-insns=16 --param max-grow-copy-bb-insns=2 --param omega-max-vars=189 --param omega-max-eqs=133 --param selsched-max-sched-times=2 --param sccvn-max-alias-queries-per-access=1407 --param lra-max-considered-reload-pseudos=802 --param graphite-max-nb-scop-params=12 --param max-vartrack-reverse-op-size=2 --param allow-store-data-races=0 --param sched-pressure-algorithm=1 --param asan-memintrin=1 -O3 1 0 2 1 cbench-consumer-jpeg-d decode image-jpeg-fgg photo.jpg 1036.8, 1036.8, 1036.8, 1036.8 1 SAMSUNG SM-G930V Android 6.0.1
S3 4675b314967527a6 1.51 1.00 -O3 -fno-branch-target-load-optimize2 -fno-check-data-deps -fno-combine-stack-adjustments -fno-crossjumping -fgcse-after-reload -fif-conversion -fno-loop-block -fmerge-all-constants -fno-modulo-sched-allow-regmoves -fno-branch-count-reg -fno-peephole -fno-peel-loops -fno-shrink-wrap -fthread-jumps -ftree-loop-ivcanon -fno-tree-loop-vectorize -ftree-partial-pre -fno-tree-vectorize -fno-unsafe-math-optimizations --param max-inline-insns-auto=22 --param comdat-sharing-probability=17 --param inline-unit-growth=47 --param early-inlining-insns=3 --param tracer-min-branch-ratio=43 --param min-crossjump-insns=6 --param iv-consider-all-candidates-bound=57 --param scev-max-expr-size=117 --param omega-max-eqs=19 --param max-pipeline-region-insns=232 --param max-sched-extend-regions-iters=0 --param max-dse-active-local-stores=4498 --param prefetch-latency=346 --param l1-cache-size=58 --param ipa-cp-eval-threshold=535 --param ipa-cp-array-index-hint-bonus=56 -O3 1 0 2 1 milepost-codelet-mibench-consumer-tiffmedian-src-tiffmedian-codelet-4-1 default image-jpg-fgg-office-night-gray data_orig.jpg 1036.8, 1036.8, 1036.8, 1036.8 1 SAMSUNG SM-G930V Android 6.0.1
S4 41c11db844b8ba16 1.43 0.46 -O3 -fbranch-target-load-optimize -fcaller-saves -fcx-limited-range -fdata-sections -fdelete-null-pointer-checks -fno-expensive-optimizations -fno-gcse-lm -fno-indirect-inlining -fno-ipa-pta -floop-parallelize-all -flto -fno-function-cse -fno-trapping-math -fno-reschedule-modulo-scheduled-loops -fsched-rank-heuristic -fsched-dep-count-heuristic -fno-tree-loop-ivcanon -fira-algorithm=CB --param sms-dfa-history=0 --param align-loop-iterations=3 --param vect-max-peeling-for-alignment=33 --param sched-mem-true-dep-cost=1 --param max-dse-active-local-stores=3273 --param ira-loop-reserved-regs=1 --param lra-max-considered-reload-pseudos=883 --param switch-conversion-max-branch-ratio=1 --param graphite-max-bbs-per-function=57 --param ipa-cp-eval-threshold=637 --param cxx-max-namespaces-for-diagnostic-help=441 --param sched-pressure-algorithm=1 --param max-slsr-cand-scan=74169 --param asan-memintrin=0 -O3 1 0 2 1 milepost-codelet-mibench-telecomm-gsm-src-rpe-codelet-4-1 default au-0001 data.au 1632, 1632, 1632, 1632 1 SAMSUNG SM-G930V Android 6.0.1
S5 eb144329b448786d 1.35 1.05 -O3 -fdelete-null-pointer-checks -fno-function-sections -fno-ira-hoist-pressure -fira-share-save-slots -fno-isolate-erroneous-paths-attribute -floop-parallelize-all -fno-merge-all-constants -fno-modulo-sched-allow-regmoves -fmath-errno -fno-sched-interblock -freorder-blocks -fsched2-use-superblocks -fno-sched-critical-path-heuristic -fno-selective-scheduling -fthread-jumps -fno-tree-loop-linear -fno-tree-pre -ftree-vrp -ffp-contract=off --param max-inline-recursive-depth=4 --param max-delay-slot-insn-search=183 --param large-unit-insns=3487 --param max-completely-peeled-insns=187 --param max-completely-peel-loop-nest-depth=10 --param sms-max-ii-factor=158 --param sms-loop-average-count-threshold=0 --param hot-bb-frequency-fraction=787 --param builtin-expect-probability=74 --param max-cse-insns=1645 --param max-reload-search-insns=71 --param max-pipeline-region-blocks=27 --param selsched-insns-to-rename=2 --param use-canonical-types=0 --param max-vartrack-expr-depth=17 --param max-vartrack-reverse-op-size=81 --param ipa-max-agg-items=22 --param cxx-max-namespaces-for-diagnostic-help=1284 --param allow-packed-store-data-races=0 --param asan-instrument-reads=1 --param asan-use-after-return=0 -O3 1 0 2 1 milepost-codelet-mibench-telecomm-fft-fftmisc-codelet-5-1 default 307.2, 307.2, 307.2, 307.2 1 XIAOMI MI 5 Android 6.0.1
S6 ac744969da7d44e3 1.28 1.07 -O3 -fno-associative-math -fno-branch-probabilities -fno-devirtualize-speculatively -fno-finite-math-only -fno-function-sections -fgcse-lm -fno-inline-functions -fira-loop-pressure -fisolate-erroneous-paths-dereference -fno-merge-all-constants -fmove-loop-invariants -fno-trapping-math -fno-reorder-blocks -fno-reschedule-modulo-scheduled-loops -fno-single-precision-constant -fno-tree-dse -fno-tree-pre -fno-variable-expansion-in-unroller -fno-vect-cost-model --param max-inline-recursive-depth=6 --param comdat-sharing-probability=36 --param ipcp-unit-growth=11 --param large-stack-frame=357 --param gcse-after-reload-critical-fraction=19 --param gcse-unrestricted-cost=0 --param sms-dfa-history=0 --param max-predicted-iterations=64 --param max-goto-duplication-insns=14 --param vect-max-version-for-alias-checks=6 --param max-sched-insn-conflict-delay=7 --param graphite-max-nb-scop-params=1 --param lto-partitions=3 --param max-tracked-strlens=1062 --param asan-memintrin=1 -O3 1 0 2 1 milepost-codelet-mibench-telecomm-gsm-src-short-term-codelet-2-1 default image-ppm-0001 data.ppm 1036.8, 1036.8, 1036.8, 1036.8 1 SAMSUNG SM-G930V Android 6.0.1
S7 ec9e9b9555ad0034 1.25 1.00 -O3 -fno-ipa-sra -fno-hoist-adjacent-loads -fno-ipa-reference -fno-isolate-erroneous-paths-attribute -fno-inline -fno-optimize-sibling-calls -fpredictive-commoning -fsched-rank-heuristic -fschedule-insns2 -fno-tree-builtin-call-dce -ftree-ch -fno-tree-coalesce-vars -fno-tree-dce -fno-tree-dominator-opts -ftree-loop-if-convert -ftree-loop-im -ftree-phiprop -ftree-vectorize -fno-unsafe-loop-optimizations -fno-vect-cost-model -ffp-contract=fast --param comdat-sharing-probability=17 --param partial-inlining-entry-probability=10 --param max-modulo-backtrack-attempts=38 --param large-unit-insns=2473 --param max-peeled-insns=127 --param max-peel-times=17 --param max-once-peeled-insns=185 --param max-iterations-to-track=333 --param tracer-min-branch-ratio=18 --param iv-max-considered-uses=482 --param vect-max-version-for-alignment-checks=3 --param vect-max-version-for-alias-checks=1 --param max-pipeline-region-insns=170 --param sched-mem-true-dep-cost=1 --param max-dse-active-local-stores=4268 --param use-canonical-types=0 --param sccvn-max-alias-queries-per-access=722 --param slp-max-insns-in-bb=648 --param prefetch-min-insn-to-mem-ratio=2 --param tm-max-aggregate-size=8 --param allow-load-data-races=1 --param max-slsr-cand-scan=899814 --param asan-globals=0 -O3 1 0 2 1 milepost-codelet-mibench-telecomm-fft-fourierf-codelet-3-1 default 1248, 1248, 1248, 1248 1 SAMSUNG SM-G935V Android 6.0.1
S8 52c92d7469011abd 1.14 0.31 -O3 -ffunction-sections -fhoist-adjacent-loads -fno-ipa-reference -fno-keep-inline-functions -floop-nest-optimize -flto -fbranch-count-reg -fno-reorder-functions -fno-sched2-use-superblocks -fno-schedule-insns -ftree-builtin-call-dce -fno-tree-dominator-opts -fno-tree-pta -fno-unroll-loops -fno-whole-program -fno-wpa -fira-algorithm=CB --param max-inline-recursive-depth-auto=10 --param comdat-sharing-probability=4 --param max-pending-list-length=47 --param max-gcse-insertion-ratio=35 --param gcse-cost-distance-ratio=3 --param max-unswitch-level=0 --param tracer-max-code-growth=67 --param max-goto-duplication-insns=2 --param iv-consider-all-candidates-bound=51 --param sched-spec-prob-cutoff=45 --param integer-share-limit=89 --param graphite-max-nb-scop-params=17 --param max-vartrack-size=39774157 --param tm-max-aggregate-size=18 --param cxx-max-namespaces-for-diagnostic-help=1109 -O3 1 0 2 1 milepost-codelet-mibench-automotive-susan-e-src-susan-codelet-2-1 default DGEMM_NN DGEMM_NN_2x1.json 1190.4, 1190.4, 1190.4, 1190.4 1 SAMSUNG SM-G930V Android 6.0.1
S9 464284073a4d2670 1.13 0.29 -O3 -fno-compare-elim -fdelayed-branch -fdevirtualize-speculatively -fno-float-store -fno-hoist-adjacent-loads -fno-isolate-erroneous-paths-dereference -fno-ivopts -flto -fbranch-count-reg -fno-signed-zeros -freorder-blocks-and-partition -fsched-group-heuristic -fno-selective-scheduling -fno-sel-sched-pipelining -fstrict-aliasing -ftree-copyrename -fno-tree-dce -fno-tree-dse -ftree-forwprop -ftree-loop-im -falign-labels=0 -ffp-contract=fast -finline-limit=0 --param predictable-branch-outcome=37 --param inline-min-speedup=0 --param max-inline-recursive-depth-auto=2 --param max-delay-slot-live-search=281 --param max-unrolled-insns=284 --param max-peel-times=0 --param max-once-peeled-insns=149 --param max-unswitch-level=1 --param sms-max-ii-factor=182 --param align-loop-iterations=3 --param max-goto-duplication-insns=1 --param omega-max-vars=24 --param max-sched-extend-regions-iters=0 --param max-sched-insn-conflict-delay=1 --param sched-state-edge-prob-cutoff=12 --param integer-share-limit=430 --param min-size-for-stack-sharing=63 --param l2-cache-size=619 --param sccvn-max-alias-queries-per-access=572 --param ira-max-conflict-table-size=1252 --param switch-conversion-max-branch-ratio=13 --param ipa-cp-eval-threshold=514 --param max-tracked-strlens=1975 --param asan-use-after-return=0 -O3 1 0 2 1 milepost-codelet-mibench-network-dijkstra-src-dijkstra-large-codelet-5-1 default txt-0001 data.s.txt 1785.6, 1785.6, 1785.6, 1785.6 1 SAMSUNG SM-G930V Android 6.0.1



[ Participated users, platforms, OS, CPU, GPU, GPGPU, NN ] [ How to participate ] [ Slides ] [ Paper ] [ Android app ] [ dividiti ] [ Collective training set ] [ Unified AI ]
View entry in raw format

Developed by dividiti,
cTuning foundation,
and the community
          
Implemented as a CK workflow
                     
   
   
                      Hosted at