Index: doc/theses/mike_brooks_MMath/benchmarks/string/Makefile =================================================================== --- doc/theses/mike_brooks_MMath/benchmarks/string/Makefile (revision f85de474d7bae82407a8ea5f1551da282d77d9bd) +++ doc/theses/mike_brooks_MMath/benchmarks/string/Makefile (revision 7d02d35a7f30abbea5194b398c84870163fbe5dd) @@ -1,2 +1,12 @@ +# Usage +# $cfabuild/driver/cfa -nodebug -quiet -c -x c /dev/null && rm null.o || echo Need to fix \$cfabuild +# make CFABUILD=$cfabuild -j8 # compile +# make measurement CFA_APILEVELS=ll OPERATIONS='pta peq pbv' CORPORI='corpus-100-*-1.txt corpus-1-*-1.txt' # append-pbv +# make measurement2 CFA_APILEVELS=ll OPERATIONS='pall' PLATFORMS=cfa CFA_SHARINGS=share CORPORI='corpus-1-*-1.txt' # allocate-speed-cfa +# make measurement2 CFA_APILEVELS=ll OPERATIONS='pall' PLATFORMS=stl CFA_EXPANSIONS=-1.0 CORPORI='corpus-1-*-1.txt' # allocate-speed-stl +# make measurement3 CFA_APILEVELS=ll OPERATIONS='pall' PLATFORMS=cfa CFA_SHARINGS=share CORPORI='corpus-1-*-1.txt' # allocate-space-cfa +# make measurement3 CFA_APILEVELS=ll OPERATIONS='pall' PLATFORMS=stl CFA_EXPANSIONS=-1.0 CORPORI='corpus-1-*-1.txt' # allocate-space-stl +# make measurement4 CFA_APILEVELS=ll OPERATIONS='pall' PLATFORMS=cfa CFA_EXPANSIONS=0.2 CFA_SHARINGS=share CORPORI='corpus-1-*-1.txt' # allocate-attrib-cfa +# make measurement4 CFA_APILEVELS=ll OPERATIONS='pall' PLATFORMS=stl CFA_EXPANSIONS=-1.0 CORPORI='corpus-1-*-1.txt' # allocate-attrib-stl CFABUILD = ~/cfa2/build-perf @@ -37,9 +47,9 @@ endef -OPERATIONS=pta peq pbv pall pno +OPERATIONS=pta peq pbv pall #pno ALLOCS=reuse fresh CFA_APILEVELS=hl ll CFA_SHARINGS=share noshare -PLATFORMS=cfa stl buhr94 +PLATFORMS=cfa stl #buhr94 ifneq ($(filter cfa,$(PLATFORMS)),) @@ -142,4 +152,8 @@ CORPORI = corpus-100-*-1.txt +RUN_TASKSET_CPULIST=6 + +# General timing +# Output file gets concatenation of program outputs measurement: $(MEASURE) tofile=measurement-`date '+%F--%H-%M-%S'`.csv ; \ @@ -148,13 +162,14 @@ for corpus in $(CORPORI) ; do \ corpusbody=`cat $$corpus` ; \ - printed=`./$$prog 100 10 $$corpusbody` ; \ - echo $$prog,$$corpus,$$printed >> $$tofile ; \ - echo $$prog,$$corpus,$$printed ; \ - done ; \ - done -# printed=`./$$prog 10000 - 10 $$corpusbody` ; \ - -CFA_EXPANSIONS=0.02 0.05 0.1 0.2 0.5 0.9 - + resulttext=`taskset --cpu-list $(RUN_TASKSET_CPULIST) ./$$prog 100 10 $$corpusbody` ; \ + echo $$prog,$$corpus,$$resulttext >> $$tofile ; \ + echo $$prog,$$corpus,$$resulttext ; \ + done ; \ + done + +CFA_EXPANSIONS=0.02 0.05 0.1 0.2 0.4 0.5 0.9 0.98 + +# Special-case timing with extra IV for CFA_EXPANSIONS +# Output file gets concatenation of program outputs (one line per run) measurement2: $(MEASURE) tofile=measurement-`date '+%F--%H-%M-%S'`.csv ; \ @@ -179,12 +194,43 @@ done +# Space, with the IV for CFA_EXPANSIONS +# Runs Mubeen's malloc interceptor to get malloc-request state +# Output file gets concatenation of the interceptor's output (one line per run); program's output ignorred +# Expect and ignore crashes; they're during shutdown, after everything we care about is over measurement3: $(MEASURE) + tofile=measurement-`date '+%F--%H-%M-%S'`.ssv ; \ for prog in $(MEASURE) ; do \ for corpus in $(CORPORI) ; do \ for expansion in $(CFA_EXPANSIONS) ; do \ corpusbody=`cat $$corpus` ; \ - LD_PRELOAD=~/plg2/mubeen-stat-shim/malloc/mallocWrappers.so ./$$prog 1000 1.006 $$expansion 1 $$corpusbody ; \ - mv preload_dump.txt preload_dump--qrun1--$$corpus--expansion-$$expansion.txt ; \ - done ; \ - done ; \ - done + LD_PRELOAD=~/plg2/mubeen-stat-shim/malloc/mallocWrappers.so ./$$prog 1000 1.006 $$expansion 10 $$corpusbody ; \ + printed=`tail -n 1 preload_dump.txt` ; \ + echo $$prog $$corpus $$expansion $$printed >> $$tofile ; \ + echo $$prog $$corpus $$expansion $$printed ; \ + rm preload_dump.txt ; \ + done ; \ + done ; \ + done + +# Time attribution, with the IV for CFA_EXPANSIONS +# Runs the SUT under perf, then crunches the perf result +# Output file gets concatenation of perf summaries (several lines per run); program's output ignorred +# Expect and ignore output "addr2line: DWARF error: section .debug_info is larger than its filesize!" +measurement4: $(MEASURE) + RUNID=`date '+%F--%H-%M-%S'` ; \ + tofile=measurement-$$RUNID.ssv ; \ + echo $$tofile ; \ + for prog in $(MEASURE) ; do \ + for corpus in $(CORPORI) ; do \ + for expansion in $(CFA_EXPANSIONS) ; do \ + SLUG=measurement--$$prog--$$corpus--$$expansion--$$RUNID ; \ + corpusbody=`cat $$corpus` ; \ + perf record --call-graph dwarf -m16M ./$$prog 1000 1.006 $$expansion 10 $$corpusbody ; \ + mv perf.data $$SLUG.data ; \ + perf script -i $$SLUG.data > $$SLUG.perf ; \ + ~/flamegraph/FlameGraph/stackcollapse-perf.pl $$SLUG.perf > $$SLUG.folded ; \ + ~/flamegraph/FlameGraph/flamegraph.pl $$SLUG.folded > $$SLUG.svg ; \ + python3 process-allocn-attrib.py $$SLUG.folded | xargs -L1 echo $$prog $$corpus $$expansion >> $$tofile ; \ + done ; \ + done ; \ + done Index: doc/theses/mike_brooks_MMath/benchmarks/string/make-flamegraph.sh =================================================================== --- doc/theses/mike_brooks_MMath/benchmarks/string/make-flamegraph.sh (revision f85de474d7bae82407a8ea5f1551da282d77d9bd) +++ doc/theses/mike_brooks_MMath/benchmarks/string/make-flamegraph.sh (revision 7d02d35a7f30abbea5194b398c84870163fbe5dd) @@ -10,7 +10,9 @@ # $3 is coprus id : x-100-20-1 (goes with corpusx-100-20-1.txt) -SLUG=perf--$1--${2//[ .]/-}--$3 +SLUG=perf--$1--${2//[ .]/-}--$3--`date '+%F--%H-%M-%S'` +benchdir=$(dirname "$0") +echo benchdir = $benchdir -perf record --call-graph dwarf -m16M ~/plg2/cfa2/cfa-cc/tests/zombies/string-perf/perfexp-$1 $2 `cat ~/plg2/cfa2/cfa-cc/tests/zombies/string-perf/corpus$3.txt` +perf record --call-graph dwarf -m16M $benchdir/perfexp-$1 $2 `cat $benchdir/corpus$3.txt` mv perf.data $SLUG.data @@ -18,3 +20,3 @@ ../flamegraph/FlameGraph/stackcollapse-perf.pl $SLUG.perf > $SLUG.folded ../flamegraph/FlameGraph/flamegraph.pl $SLUG.folded > $SLUG.svg -cp $SLUG.svg ~/plg2/flames +#cp $SLUG.svg ~/plg2/flames Index: doc/theses/mike_brooks_MMath/benchmarks/string/process-allocn-attrib.py =================================================================== --- doc/theses/mike_brooks_MMath/benchmarks/string/process-allocn-attrib.py (revision 7d02d35a7f30abbea5194b398c84870163fbe5dd) +++ doc/theses/mike_brooks_MMath/benchmarks/string/process-allocn-attrib.py (revision 7d02d35a7f30abbea5194b398c84870163fbe5dd) @@ -0,0 +1,107 @@ +import sys +import re +import pandas as pd +from collections import defaultdict + +# Matched top to bottom, bailing on first match +# More general rules are not double-counted if they occur later +# Such cases are commented "overlap"; don't move those too far up +CATEGORY_RULES = { + "text-import": [ + "_X19eagerCopyCtorHelperFv_S10string_resPKcm__1;_X12_constructorFv_S10string_resPKcm__1;__memmove_ssse3", + "_X19eagerCopyCtorHelperFv_S10string_resPKcm__1;_X12_constructorFv_S10string_resPKcm__1;__memcpy_ssse3", + "helper;__memcpy_ssse3", +# "strlen" + ], + "gc": [ + "_X19eagerCopyCtorHelperFv_S10string_resPKcm__1;_X12_constructorFv_S10string_resPKcm__1;_X7garbageFv_S9VbyteHeapi__1" + ], + "malloc-free": [ + "operator new;_X8doMallocFPv_mj__1", + "operator new;malloc", + "_X6doFreeFv_Pv__1", + "free" + ], + "ctor-dtor": [ + "std::__cxx11::basic_string, std::allocator >::_M_create", + "operator new", # overlap stl malloc-free + "operator delete", + "_X12_constructorFv_S10string_resPKcm__1" # overlap cfa text import + ] +} + +DEFAULT_CATEGORY = "other" + +def classify_stack(stack): + for category, patterns in CATEGORY_RULES.items(): + for pattern in patterns: + if pattern in stack: + return category + if re.search(r"_X6helperFv_i__1$", stack): + return "harness-leaf" + if re.search(r"helper$", stack): + return "harness-leaf" + return DEFAULT_CATEGORY + +# def parse_sut_and_size(filename): +# # Extract SUT after "perfexp-" and before the next hyphen +# sut_match = re.search(r"perfexp-([a-zA-Z0-9]+)", filename) +# # Extract SIZE from "corpus-A-B-C.txt", capturing B +# size_match = re.search(r"corpus-\d+-(\d+)-\d+\.txt", filename) + +# if not sut_match or not size_match: +# print("Error: Could not parse sut or size from filename.") +# sys.exit(1) + +# return sut_match.group(1), size_match.group(1) + +def read_and_aggregate(input_file): + category_map = defaultdict(lambda: defaultdict(int)) # category -> lineno -> sample_count + total_samples = 0 + + with open(input_file) as f: + for lineno, line in enumerate(f, 1): + line = line.strip() + if not line: + continue + *stack_parts, count_str = line.split() + count = int(count_str) + stack = ' '.join(stack_parts) + category = classify_stack(stack) + category_map[category][lineno] += count + total_samples += count + + return category_map, total_samples + +def flatten(category_map, total_samples): #, sut, size): + rows = [] + for category, source_map in category_map.items(): + samples_in_category = sum(source_map.values()) + sources = "|".join(f"{lineno}:{count}" for lineno, count in source_map.items()) + fraction = samples_in_category / total_samples if total_samples else 0.0 + rows.append({ +# "sut": sut, +# "size": size, + "category": category, + "samples_in_category": samples_in_category, + "total_samples": total_samples, + "fraction": fraction, + "sources": sources + }) + return pd.DataFrame(rows) + +def main(): + if len(sys.argv) != 2: + print("Usage: python3 process-allocn-attrib.py ") + sys.exit(1) + + input_file = sys.argv[1] + # sut, size = parse_sut_and_size(input_file) + category_map, total_samples = read_and_aggregate(input_file) + df = flatten(category_map, total_samples) #, sut, size) + + # Print the result to stdout in tab-separated format + df.to_csv(sys.stdout, sep="\t", index=False, header=False) + +if __name__ == "__main__": + main() Index: doc/theses/mike_brooks_MMath/benchmarks/string/prog-allocn.cfa =================================================================== --- doc/theses/mike_brooks_MMath/benchmarks/string/prog-allocn.cfa (revision f85de474d7bae82407a8ea5f1551da282d77d9bd) +++ doc/theses/mike_brooks_MMath/benchmarks/string/prog-allocn.cfa (revision 7d02d35a7f30abbea5194b398c84870163fbe5dd) @@ -42,10 +42,11 @@ #endif - #if defined IMPL_CFA_HL #include + #include extern void TUNING_set_string_heap_liveness_threshold(double); // in string_res.hfa #elif defined IMPL_CFA_LL #include + #include #endif Index: doc/theses/mike_brooks_MMath/benchmarks/string/prog-passbyval.cfa =================================================================== --- doc/theses/mike_brooks_MMath/benchmarks/string/prog-passbyval.cfa (revision f85de474d7bae82407a8ea5f1551da282d77d9bd) +++ doc/theses/mike_brooks_MMath/benchmarks/string/prog-passbyval.cfa (revision 7d02d35a7f30abbea5194b398c84870163fbe5dd) @@ -45,6 +45,8 @@ #if defined IMPL_CFA_HL #include + #include #elif defined IMPL_CFA_LL #include + #include #endif Index: doc/theses/mike_brooks_MMath/benchmarks/string/prog.cfa =================================================================== --- doc/theses/mike_brooks_MMath/benchmarks/string/prog.cfa (revision f85de474d7bae82407a8ea5f1551da282d77d9bd) +++ doc/theses/mike_brooks_MMath/benchmarks/string/prog.cfa (revision 7d02d35a7f30abbea5194b398c84870163fbe5dd) @@ -45,6 +45,8 @@ #if defined IMPL_CFA_HL #include + #include #elif defined IMPL_CFA_LL #include + #include #endif